kvm2: gofmt volume.go

Making baseVolumeName multi-line in the H4 commit collapsed the alignment gofmt gives a run of single-line funcs, and the commit went out with the old spacing. Caught by make test, whose fmt target rewrites the file and leaves the tree dirty. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
kvm2: record the review validation, enable the provider entry, changelog
2026-09-17 21:07:10 +00:00 · 2026-09-17 20:50:45 +00:00 · 2026-09-17 20:24:37 +00:00 · 2026-09-17 20:05:02 +00:00 · 2026-09-17 19:23:44 +00:00 · 2026-09-17 17:57:02 +00:00
74 changed files with 15739 additions and 101 deletions
@@ -0,0 +1,11 @@
+---
+name: deep-reviewer
+description: Deep-dive verification of a single escalated hypothesis from the sweep.
+model: fable
+effort: high
+tools: Read, Grep, Glob, Bash
+---
+You get ONE hypothesis with entry points. Follow it across the whole repo: callers,
+callees, tests, configs. Verdict: CONFIRMED / REFUTED / PLAUSIBLE, with the exact
+path/interleaving that breaks, or the reason it can't. Propose a minimal fix.
+Don't re-review the package generally. Stay on the hypothesis.
@@ -0,0 +1,23 @@
+---
+name: sweep-reviewer
+description: First-pass broad code review of one package/module. Use for fan-out sweeps.
+model: opus
+effort: high
+tools: Read, Grep, Glob, Bash
+---
+You review ONE package/module given in the prompt. Read-only, never edit.
+
+Output two sections:
+
+## Findings
+Concrete defects you verified. Per item: file:line, severity, one-line defect,
+failure scenario (inputs/state -> wrong behavior). No style nits unless they hide a bug.
+
+## Escalate
+Things you could NOT confirm or refute within this package, where being wrong is costly.
+Escalate when: shared mutable state across goroutines, lock ordering, context/cancel
+propagation, retry+timeout+idempotency interactions, reconcile races, authn/authz
+boundaries, invariants that depend on code outside this package, state machines.
+Per item: entry-point files, the specific hypothesis ("X may be read without mu
+on path A->B->C"), what you already checked, what's still unknown.
+Do NOT escalate things you can resolve yourself. Go resolve them.
@@ -1,5 +1,45 @@
 # Changelog

+## 2026-09-17 20:36 UTC — kvm2: the four high findings from the code review, and its first run on a real hypervisor
+
+A scoped review of the kvm2 provider (`review/REPORT.md`: 52 findings, 4 high, 20 medium,
+28 low) produced four fixes, one commit each, then the provider was deployed in-cluster and
+validated against beelink — the first time any of this code had run against real libvirt.
+
+- **H1 — a stalled write no longer delays a connection teardown.** `sshSocket.Close` ran the
+  blocking close before the unblocking one, so a transfer stuck in a TCP write pinned the
+  teardown for as long as the kernel retransmitted (~15 min). `link.markDead` compounded it by
+  killing inside a `sync.Once` before signalling dead, which made even cancelling the reconcile
+  useless. With `MaxConcurrentReconciles: 3`, one silent host could stall every Proxy on every
+  provider. `internal/provider/kvm2/{ssh,backend}.go`.
+- **H2 — a source error no longer reaches go-libvirt's stream sender.** A failing HTTP body
+  deadlocked `StorageVolUpload` outright and leaked two goroutines plus 4 MiB per occurrence,
+  against a 128Mi pod limit. The defect is upstream (`requestStream` never drains an unbuffered
+  `outErr`); the workaround wraps the reader so it can only ever end the stream cleanly, and
+  moves the transfer deadline onto the HTTP source so a stalled mirror unwinds instead of
+  leaking. A daemon-side failure mid-upload still leaks — documented in *Known gaps* with the
+  two-character upstream patch. `internal/provider/kvm2/{backend,image}.go`.
+- **H3 — an unreadable address is no longer reported as a missing one.** Any error from both
+  address sources read as "no address yet", so a pooled connection killed by an unrelated
+  reconcile demoted a healthy proxy to Provisioning, cleared its IP and made the health engine
+  drop its counters. One source failing stays "no address yet"; every source failing, or a dead
+  connection, is now Transient. `internal/provider/kvm2/get.go`.
+- **H4 — the base image cache is no longer shared between provider entries.** kvm2 built
+  byte-identical volume names to the `kvm` provider in the same pool, so two entries could share
+  one file, a concurrent fetch could boot a guest from a half-written base, and a `sha256` edit
+  deleted the base under live overlays. Names now carry the entry
+  (`base-<entry>-<key>.<format>`), and a digest mismatch is refused rather than resolved by
+  deleting. `internal/provider/kvm2/{volume,image}.go`, `internal/provider/config_kvm.go`.
+- **klog and `net/http` errors now join the structured log stream.** client-go's leader election
+  wrote `I0917 ...` lines straight to stderr because nothing routed klog into logr, and the
+  discovery server's `http.Server` had no `ErrorLog`. The deployment also gained
+  `--zap-encoder=json` and `--zap-time-encoding=rfc3339`, so `kubectl logs | jq` works.
+  `cmd/main.go`, `internal/discovery/server.go`, `config/manager/manager.yaml`.
+- **The `kvm2-home` provider entry is enabled** in `config/manager/providers_config.yaml`.
+  Validated end to end: health check Ready, a 433 MB image cached in 8.3 s under the new name
+  beside kvm's untouched one, a second Proxy reusing the cache, Squid answering a real CONNECT
+  probe, and Delete leaving the base image while removing the domain and its volumes.
+
 ## 2026-09-16 11:46 UTC — Provider blueprint split into contract, spec template and appendix

 - **`docs/provider-contract.md`** is the generic provider specification: MUST/SHOULD/MAY
@@ -70,6 +70,10 @@ test: manifests generate fmt vet setup-envtest ## Run tests.
 test-integration-kvm: ## Run the KVM provider test against a real hypervisor. Needs KVM_TEST_HOST/USER/SSH_KEY/BRIDGE/IMAGE_URL; see docs/testing.md.
 	go test -race -tags=integration -count=1 -v -timeout 40m -run TestKVM ./internal/provider/kvm/

+.PHONY: test-integration-kvm2
+test-integration-kvm2: ## Run the kvm2 provider test against a real hypervisor. Needs KVM2_TEST_HOST/USER/SSH_KEY/BRIDGE/IMAGE_URL; see docs/testing.md.
+	go test -race -tags=integration -count=1 -v -timeout 40m -run TestKVM2 ./internal/provider/kvm2/
+
 # TODO(user): To use a different vendor for e2e tests, modify the setup under 'tests/e2e'.
 # The default setup assumes Kind is pre-installed and builds/loads the Manager Docker image locally.
 # kubectl kuberc is disabled by default for test isolation; enable with:
@@ -160,6 +160,10 @@ Managed proxies as VMs on libvirt/KVM hosts you own. Full guide:
   size (`placement.machineType`) and an image (`placement.image`) — see
   [config/samples/proxy_kvm.yaml](config/samples/proxy_kvm.yaml).

+The `kvm2` provider uses the same Secret and the same block shape (`type: kvm2`,
+block `kvm2:`), so the steps above apply to it unchanged; its guide is
+[docs/kvm2-provider.md](docs/kvm2-provider.md).
+
 Cloud-init from a Secret: the Secret **must** carry the label
 `crawl.example.com/cloud-init: "true"` — the operator's cache only holds
 labelled Secrets, so an unlabelled one is invisible (the Proxy reports
@@ -187,6 +191,7 @@ can be two differently-configured instances of the same `gcp` type (see
 | `kubernetes` | A real Squid pod (`ubuntu/squid:6.6-24.04_edge` by default) in the same cluster the operator runs in | `image` (optional) | Needs no cloud account — local dev, CI, and the kind quickstart. Pods share the cluster's egress IP, so it exercises the full lifecycle but not distinct egress paths. |
 | `gcp` | A Compute Engine VM with an ephemeral external IP | `project` (required), `network`, `networkTag`, `diskSizeGb` | The real egress fleet. Deliberately uses only four API calls (Insert / Get / Delete / AggregatedList), all fire-and-forget: `Create` returns as soon as the operation is submitted and the reconciler discovers progress by polling `Get`. Auth is Application Default Credentials — workload identity in-cluster, `gcloud` ADC locally; no key-file plumbing. |
 | `kvm` | A libvirt VM on one of your own hypervisors, reached over SSH | `sshKeyFile`, `knownHostsFile` (unless `insecureIgnoreHostKey`), `machineTypes`, `hosts` (required); `images`, `diskSizeGb`, `vendorData`, `installGuestAgent` | Home-lab or on-prem egress. The base image is downloaded into each host's pool on first use; `Create` waits for the VM to start. See [docs/kvm-provider.md](docs/kvm-provider.md). |
+| `kvm2` | The same: a libvirt VM on one of your own hypervisors, reached over SSH | the same block shape as `kvm`, under `kvm2:` | The second-generation libvirt provider, built from the provider contract. Adds a child span per libvirt RPC, a pool-space precheck reported as quota (the reconciler waits instead of failing), SSH credentials re-read on every dial, and image downloads and health checks on connections of their own. It does not adopt `kvm`'s VMs. See [docs/kvm2-provider.md](docs/kvm2-provider.md). |

 ### Adding a provider

@@ -68,7 +68,7 @@ type ProviderTarget struct {
 // ProviderHealthStatus is the result of the latest check of one provider.
 type ProviderHealthStatus struct {
 	// type is the provider's type from the providers config: kubernetes,
-	// gcp or kvm.
+	// gcp, kvm or kvm2.
 	// +optional
 	Type string `json:"type,omitempty"`

@@ -38,6 +38,7 @@ import (
 	"k8s.io/apimachinery/pkg/runtime"
 	utilruntime "k8s.io/apimachinery/pkg/util/runtime"
 	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
+	"k8s.io/klog/v2"
 	ctrl "sigs.k8s.io/controller-runtime"
 	"sigs.k8s.io/controller-runtime/pkg/cache"
 	"sigs.k8s.io/controller-runtime/pkg/client"
@@ -59,6 +60,7 @@ import (
 	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider/gcp"
 	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider/kubernetes"
 	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider/kvm"
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider/kvm2"
 	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider/registry"
 	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/providerhealth"
 	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/tracing"
@@ -161,7 +163,15 @@ func main() {
 		os.Exit(0)
 	}

-	ctrl.SetLogger(zap.New(zap.UseFlagOptions(&opts)))
+	logger := zap.New(zap.UseFlagOptions(&opts))
+	ctrl.SetLogger(logger)
+	// client-go logs through klog, and controller-runtime hands the leader
+	// elector a context with no logr in it, so klog.FromContext falls back
+	// to klog's own stderr writer and those lines ignore the encoder
+	// entirely. Plain SetLogger rather than the contextual form: it keeps
+	// klog's verbosity check in front, so this changes the format of what
+	// klog already emits without letting client-go's V(1) chatter through.
+	klog.SetLogger(logger)
 	setupLog.Info("Starting egress-proxies-operator",
 		"commit", version.Resolve(), "goVersion", goruntime.Version())

@@ -202,6 +212,7 @@ func main() {
 			return gcp.NewWithWireOptions(ctx, pc, gcp.WireLogOptions{FullPayloads: gcpWireFullPayloads})
 		},
 		"kvm":  kvm.New,
+		"kvm2": kvm2.New,
 	})
 	if err != nil {
 		setupLog.Error(err, "Failed to build providers")
@@ -183,7 +183,7 @@ spec:
              type:
                description: |-
                  type is the provider's type from the providers config: kubernetes,
-                  gcp or kvm.
+                  gcp, kvm or kvm2.
                type: string
            type: object
        type: object
@@ -64,6 +64,13 @@ spec:
          - --leader-elect
          - --health-probe-bind-address=:8081
          - --providers-config=/etc/proxy-operator/providers.yaml
+          # Structured logs: every line is one JSON object, so `kubectl logs
+          # | jq` works. The time encoding is not optional — the JSON encoder
+          # defaults to epoch seconds. Verbosity stays at the binary's
+          # default (debug, one line per provider API call); add
+          # --zap-log-level=info for outcomes only.
+          - --zap-encoder=json
+          - --zap-time-encoding=rfc3339
          # Add --trace-health-probes to emit one span per health probe
          # (high volume: ~1/s per proxy).
          # Provider health checks (kubectl get providerhealth) run every
@@ -149,9 +156,10 @@ spec:
        - name: providers-config
          mountPath: /etc/proxy-operator
          readOnly: true
-        # SSH material for kvm providers (sshKeyFile / knownHostsFile in the
-        # providers config point here). Optional: the mount is empty when the
-        # Secret does not exist, and only kvm entries need it. Create it with:
+        # SSH material for the kvm and kvm2 providers (sshKeyFile /
+        # knownHostsFile in the providers config point here). One Secret
+        # serves both types. Optional: the mount is empty when the Secret
+        # does not exist, and only kvm and kvm2 entries need it. Create it with:
        #   kubectl -n egress-proxies-operator-system create secret generic kvm-ssh \
        #     --from-file=id_ed25519=<private key> \
        #     --from-file=known_hosts=<known_hosts with every host's key>
@@ -167,6 +175,7 @@ spec:
      - name: providers-config
        configMap:
          name: providers-config
+      # Shared by the kvm and kvm2 providers; see the mount above.
      - name: kvm-ssh
        secret:
          secretName: kvm-ssh
@@ -51,4 +51,37 @@ data:
              bridge: br0                 # to a LAN with DHCP; the VM's IP comes from there
              # socket: /var/run/libvirt/libvirt-sock
              # pool: default
+      # kvm2 takes the same block shape as kvm: an entry ports by changing
+      # type to kvm2 and the block name to kvm2. kvm2 does not adopt domains
+      # made by kvm, so delete those Proxies first. It caches its base images
+      # under base-<entry>-<key>.<format>, so it downloads its own copy rather
+      # than sharing kvm's; budget the pool space. See docs/kvm2-provider.md.
+      - name: kvm2-home
+        type: kvm2
+        kvm2:
+          sshKeyFile: /etc/proxy-operator/kvm/id_ed25519
+          knownHostsFile: /etc/proxy-operator/kvm/known_hosts
+          # insecureIgnoreHostKey: true
+          # diskSizeGb: 10
+          # installGuestAgent: false
+          # vendorData: |
+          #   #cloud-config
+          #   ...
+          images:
+            debian-13:
+              url: https://cloud.debian.org/images/cloud/trixie/latest/debian-13-generic-amd64.qcow2
+              # sha256: "<64 hex>"
+              # firmware: uefi
+            ubuntu-24.04:
+              url: https://cloud-images.ubuntu.com/noble/current/noble-server-cloudimg-amd64.img
+          machineTypes:
+            small:  {vcpus: 1, memoryMiB: 1024}
+            medium: {vcpus: 2, memoryMiB: 2048}
+          hosts:
+            - name: beelink
+              address: beelink.tail8c69c.ts.net:22
+              user: novakj
+              bridge: br0
+              # socket: /var/run/libvirt/libvirt-sock
+              # pool: default

@@ -5,5 +5,6 @@ resources:
 - proxy_kubernetes.yaml
 - proxy_gcp.yaml
 - proxy_kvm.yaml
+- proxy_kvm2.yaml
 - proxy_external.yaml
 # +kubebuilder:scaffold:manifestskustomizesamples
@@ -55,3 +55,36 @@ providers:
          bridge: br0                 # to a LAN with DHCP; the VM's IP comes from there
          # socket: /var/run/libvirt/libvirt-sock
          # pool: default
+  # kvm2 takes the same block shape as kvm: an entry ports by changing type
+  # to kvm2 and the block name to kvm2. kvm2 does not adopt domains made by
+  # kvm, so delete those Proxies first. It caches its base images under
+  # base-<entry>-<key>.<format>, so it downloads its own copy rather than
+  # sharing kvm's; budget the pool space. See docs/kvm2-provider.md.
+  # - name: kvm2-home
+  #   type: kvm2
+  #   kvm2:
+  #     sshKeyFile: /etc/proxy-operator/kvm/id_ed25519
+  #     knownHostsFile: /etc/proxy-operator/kvm/known_hosts
+  #     # insecureIgnoreHostKey: true
+  #     # diskSizeGb: 10
+  #     # installGuestAgent: false
+  #     # vendorData: |
+  #     #   #cloud-config
+  #     #   ...
+  #     images:
+  #       debian-13:
+  #         url: https://cloud.debian.org/images/cloud/trixie/latest/debian-13-generic-amd64.qcow2
+  #         # sha256: "<64 hex>"
+  #         # firmware: uefi
+  #       ubuntu-24.04:
+  #         url: https://cloud-images.ubuntu.com/noble/current/noble-server-cloudimg-amd64.img
+  #     machineTypes:
+  #       small:  {vcpus: 1, memoryMiB: 1024}
+  #       medium: {vcpus: 2, memoryMiB: 2048}
+  #     hosts:
+  #       - name: beelink
+  #         address: beelink:22
+  #         user: novakj
+  #         bridge: br0
+  #         # socket: /var/run/libvirt/libvirt-sock
+  #         # pool: default
@@ -0,0 +1,44 @@
+# A Managed proxy backed by a libvirt/KVM hypervisor through the kvm2
+# provider: the operator boots a stock Debian cloud image as a VM on the
+# named host, with a NIC on the host's bridge (IP from the LAN's DHCP), and
+# installs Squid via cloud-init. Requires a providers-config entry named
+# "kvm2-home" (see providers-config.yaml) and the kvm-ssh Secret, which the
+# kvm and kvm2 providers share (docs/kvm2-provider.md).
+#
+# placement.zone is the host name from kvm2.hosts (optional when only one
+# host is configured); machineType and image are keys in the provider's
+# machineTypes and images maps. The first Proxy for an image on a host
+# waits for the image download; later ones reuse the cached copy.
+apiVersion: crawl.example.com/v1alpha1
+kind: Proxy
+metadata:
+  name: proxy-kvm2-sample
+spec:
+  mode: Managed
+  provider: kvm2-home
+  placement:
+    zone: beelink
+    machineType: small
+    image: debian-13
+  port: 3128
+  cloudInit:
+    inline: |
+      #cloud-config
+      packages:
+        - squid
+      write_files:
+        - path: /etc/squid/conf.d/proxy-operator.conf
+          content: |
+            http_port 3128
+            http_access allow all
+            via off
+            forwarded_for off
+      runcmd:
+        - systemctl restart squid
+      # Optional: log in with `ssh debian@<ip>` (the Debian image's default
+      # user) to look around; `virsh console <name>` on the host works too.
+      # ssh_authorized_keys:
+      #   - ssh-ed25519 AAAA... you@laptop
+  attributes:
+    geo: home
+    purpose: crawl
@@ -10,6 +10,7 @@
 | kubernetes provider | `internal/provider/kubernetes` | library | — | real Squid pods in this cluster (local dev/CI) |
 | gcp provider | `internal/provider/gcp` | library | — | Compute Engine VMs, four API calls, fire-and-forget ops |
 | kvm provider | `internal/provider/kvm` | library | — | VMs on libvirt hosts over SSH-forwarded sockets; image cache, overlay disks, NoCloud seeds |
+| kvm2 provider | `internal/provider/kvm2` | library | — | the same backend, built clean-room from the provider contract as a second type: per-RPC child spans, pool-space quota, per-call budgets, its own metadata namespace, credentials re-read on every dial |
 | Health engine | `internal/health` | Runnable | yes | through-the-proxy probes, thresholds, transition events |
 | Lease store | `internal/lease` | Runnable (expiry sweep) | no | in-memory leases + cooldowns, single mutex |
 | Discovery API | `internal/discovery` | Runnable | no | HTTP list/lease/release/report on `:8090` |
@@ -168,6 +169,40 @@ providerID = hosts/<host>/domains/<name>; GC identity lives in the domain's
 <metadata> element (libvirt has no labels). All RPCs run under a context-
 bounded call: go-libvirt never times out on its own, so a cancelled reconcile
 closes the SSH connection to unblock it.
+
+kvm2 provider (internal/provider/kvm2/) - the same transport, a second type built from the contract
+prov.Create ──► placement checked against the catalogue (Permanent before any dial)
+                DomainLookupByName: exists → mark+uid guard (no mark or foreign   ─┐ every step
+                  uid = Permanent; paused = Permanent),                            │
+                  start if needed, wait for running, done                          │ idempotent,
+                else: StoragePoolGetInfo precheck: image (if uncached) + diskSizeGb │ under a
+                        + 16 MiB must fit, else ErrQuotaExceeded (reconciler       │ per-name lock
+                        waits 5m)                                                  │ shared with
+                      ensure base image: marker present = cached; otherwise GET    │ Delete
+                        streamed into StorageVolUpload on a DEDICATED connection,  │
+                        StoragePoolRefresh, marker written last                    │
+                      overlay volume (exists = reuse)                              │
+                      seed ISO (always rebuilt)                                    │
+                      DomainDefineXML (<metadata> in kvm2's own namespace:         │
+                        managed, uid, createdAt) + DomainCreate, poll until        │
+                        running (15s cap; a domain that dies at once = Transient) ─┘
+prov.Get    ──► DomainLookupByName + DomainGetMetadata + DomainGetState
+                  + DomainInterfaceAddresses (guest agent, then ARP; first routable IPv4)
+                  running without IP = Provisioning; nostate = Provisioning;
+                  only ErrNoDomain is NotFound
+prov.Delete ──► destroy (not running = fine) → undefine (nvram, managed save, snapshots)
+                  → delete overlay + seed by name; unconfigured host = logged no-op
+prov.ListByTag ─► ConnectListAllDomains on every host, keep domains whose kvm2
+                    <metadata> says managed=true; one host down fails the sweep,
+                    one unreadable domain is skipped
+Check       ──► per host, on a FRESH connection: version, hostname, pool lookup,
+                  pool info, close; never the pooled one
+providerID = hosts/<host>/domains/<name>. Every RPC has its own budget (dial 10s,
+RPC 30s, DomainCreate 90s, seed upload 60s, image transfer 20m): a call that
+outlives its ctx or budget kills the pooled connection (the only way to unblock
+go-libvirt), so other calls in flight on that host fail once with Transient and
+retry; the next call redials. Each RPC is a child span kvm2.<Method> under
+provider.*, plus kvm2.dial and kvm2.download.
 ```

 The reconciler never watches provider-side resources (Pods or GCP VMs).
@@ -343,6 +378,15 @@ incoming trace context):
 - `provider.create|get|delete|list` (client spans, `provider.WithTracing`
  decorator wired outermost around `WithMetrics`); the GCP SDK's own
  otelhttp transport contributes HTTP child spans automatically.
+- kvm2 child spans: under `provider.*`, one `kvm2.<Method>` client span per
+  libvirt RPC (for example `kvm2.DomainDefineXML`, attributes `kvm2.host`,
+  `kvm2.domain`, `kvm2.pool`, `kvm2.volume`, `kvm2.libvirt_code` on
+  failure), `kvm2.dial` around the SSH handshake and libvirt connect, and
+  `kvm2.download` around an image fetch; under `providerhealth.check`,
+  `kvm2.dial` and the four `kvm2.<Method>` spans of a check. They come
+  from a span decorator over the seam, applied once in the dialer, so no
+  RPC is unspanned. The `libvirt call` log line of an RPC is written
+  inside its span. The `kvm` provider has no child spans.
 - Kubernetes API calls become client child spans via a wrapped
  `rest.Config` transport — gated on an existing parent span, so informer
  list/watch long-polls and leader-election renewals never create root
@@ -368,11 +412,12 @@ the slog handler's ctx.
 Answers "can this cluster reach each provider's backend right now?"
 without waiting for a Proxy to fail. A provider opts in by implementing
 `provider.Checker`: `Check(ctx) []TargetStatus` returns one verdict per
-target and never fails as a whole. All three providers do:
+target and never fails as a whole. All four providers do:

 | Provider | Target | What one check does |
 |---|---|---|
 | kvm | each `kvm.hosts` entry | over the pooled SSH+libvirt connection: libvirt version and hostname, then the configured pool must exist and be running; reports pool capacity and free space ([details](kvm-provider.md#health-check)) |
+| kvm2 | each `kvm2.hosts` entry | on a fresh SSH+libvirt connection dialled for the check and closed after it: libvirt version and hostname, then the configured pool must exist and be running; `details` carry `user`, `pool`, `hostname`, `libvirtVersion`, `poolState`, `poolCapacityGiB` and `poolAvailableGiB` ([details](kvm2-provider.md#health-check)) |
 | kubernetes | `apiserver` | lists at most one managed Pod cluster-wide: the verb and scope `ListByTag` and GC rely on |
 | gcp | the project | one page of an aggregated instance list with `maxResults=1`: credentials, Compute API enabled, `compute.instances.list` |

@@ -613,3 +658,48 @@ they read as choices, not accidents. Chronological by build step.
 - **A provider name that is not a valid object name** gets metrics but no
  `ProviderHealth` object, with a warning at start. Tightening config
  validation instead would crash-loop existing deployments on upgrade.
+- **kvm2 is a second provider type built clean-room from the contract,**
+  not a rewrite of `kvm` in place. Its spec
+  (`docs/kvm2-provider-spec.md`) and code were written from
+  [provider-contract.md](provider-contract.md), the kvm user guide and
+  go-libvirt alone, never from `internal/provider/kvm/`, as a test of
+  whether the contract is enough to build a provider. Both types ship,
+  share the `KVMConfig` block shape and the `kvm-ssh` Secret, and may
+  share a host.
+- **kvm2 image downloads run on a dedicated connection.** A cancelled or
+  overlong call kills the connection it ran on (the only way to unblock
+  go-libvirt), so a 435 MB upload on the pooled connection could die
+  because a poll for another Proxy was cancelled. The transfer dials its
+  own connection and closes it afterwards; the cheap checks and the marker
+  stay on the pooled one.
+- **kvm2 health checks dial fresh connections**, the opposite of the kvm
+  decision above. With credentials re-read on every dial a fresh
+  connection does pick up a rotated Secret, and a check that runs out of
+  time then closes only its own connection, so it cannot disrupt a
+  reconcile. The cost is one sshd session per host per round.
+- **A cancelled kvm2 call kills the pooled connection.** Every RPC has a
+  budget; when a call outlives its ctx or budget the adapter closes the
+  socket and the SSH client, returns Transient, and marks the connection
+  dead so the next call redials. Other calls in flight on that host fail
+  once with Transient and their callers retry. Documented in the guide
+  rather than avoided: the alternative, leaving a hung RPC blocked
+  forever, pins a worker.
+- **kvm2 marks domains in its own metadata namespace**, so it never lists
+  or adopts domains the `kvm` provider made. Switching an entry from `kvm`
+  to `kvm2` means deleting its Proxies first; the two types can share a
+  host without one collecting the other's VMs.
+- **kvm2 turns pool free space into the quota signal.** libvirt reports a
+  full pool as an ordinary error, which would retry on the fast backoff.
+  Before creating any volume, `Create` compares `StoragePoolGetInfo`'s
+  available bytes with the image (if uncached) plus `diskSizeGb` plus 16
+  MiB for the seed, and returns `ErrQuotaExceeded` below that, so the
+  reconciler waits 5 minutes instead.
+- **kvm2 classifies SSH authentication and host-key failures as
+  Permanent.** A key missing from `authorized_keys` or a host key missing
+  from or different in known_hosts needs an admin; kvm retried these
+  forever. A missing key or known_hosts file stays Transient, because the
+  optional Secret may land later.
+- **kvm2 re-reads the SSH key and known_hosts on every dial**, so rotating
+  the Secret needs no restart; nothing is cached. The constructor only
+  checks that the files exist, to warn early without failing the provider
+  set.
@@ -0,0 +1,665 @@
+# Provider spec: kvm2
+
+Filled from [provider-spec-template.md](provider-spec-template.md) against
+[provider-contract.md](provider-contract.md). This is the design the `kvm2`
+package is built from. Where an entry names an axis in backticks, the answer is
+one of that axis's profiles in contract §4. Nothing in this spec was taken from
+the code of the existing `kvm` provider; backend facts come from
+[kvm-provider.md](kvm-provider.md), the execution log of the first build, the
+go-libvirt module in the module cache, and the beelink host.
+
+## Header
+
+- **Type**: `kvm2`
+- **Example configured name**: `kvm2-home`
+- **Backend and API version this spec was written against**: libvirt 11.6.0 and
+  QEMU 10.1.0 on Ubuntu (host `beelink`), reached with go-libvirt
+  `github.com/digitalocean/go-libvirt v0.0.0-20260814190004-1a83157e1858` over an
+  SSH-forwarded unix socket.
+- **Client**: `rpc-over-ssh`. go-libvirt's RPC client (`libvirt.New(conn)`
+  then `Connect()`) over a `direct-streamlocal` channel opened with
+  `golang.org/x/crypto/ssh` to the host's libvirt socket. Both modules are
+  already direct requires in `go.mod`; no new dependency.
+- **Author and date**: Claude (orchestrator session), 2026-09-16.
+
+## 1. Identity and providerID
+
+**Answer:** `hosts/<host>/domains/<name>`, where `<host>` is `hosts[].name`
+from the config block and `<name>` is `req.Name`. Parsing is strict: exactly
+four segments, literal `hosts` and `domains`, non-empty host and name, no
+further slashes; anything else is `ErrPermanent` from `Get` and `Delete`. Host
+names are validated at config load by the shared `kvmIdentifier` regexp in
+`internal/provider/config_kvm.go` (letters, digits, dots, dashes, underscores),
+so a host name can never contain the separator. Instances do not migrate
+between hosts (libvirt domains are per host), so the location belongs in the
+ID.
+
+**Evidence:** contract §6; `kvmIdentifier` in `internal/provider/config_kvm.go`.
+
+## 2. Primary key (`primary-key`)
+
+**Answer:** `name-is-key`. libvirt addresses domains by name
+(`DomainLookupByName`); `req.Name` is used verbatim as the domain name and as
+the base of the per-VM volume names. Create looks the name up first and adopts
+an existing domain only if it carries the kvm2 mark (`managed` true, entry 8)
+and its metadata `uid` is empty or equals `req.UID`. A domain without the
+mark returns `ErrPermanent` saying it was not created by this provider: kvm2
+never starts or takes over a VM someone else defined under that name. A
+different non-empty `uid` returns `ErrPermanent` naming both UIDs. This
+closes the contract's "adoption is not verified" limit for this provider.
+
+**Evidence:** `DomainLookupByName(Name string) (Domain, error)` in the
+go-libvirt module; contract §2.1 idempotency and §13.
+
+## 3. Placement mapping
+
+**Answer:** three fields are catalogue keys checked before any backend call, one is ignored:
+
+| Field | Meaning | Required | Unknown value |
+|---|---|---|---|
+| `Zone` | a `hosts[].name` | only when more than one host is configured; with exactly one host an empty zone selects it | `ErrPermanent`: `placement.zone "x" is not in kvm2.hosts` |
+| `MachineType` | a key of `machineTypes` (`vcpus`, `memoryMiB`) | yes | `ErrPermanent`: `placement.machineType "x" is not in kvm2.machineTypes` |
+| `Image` | a key of `images`, or an absolute http(s) URL used directly | yes | `ErrPermanent`: `placement.image "x" is neither a key in kvm2.images nor an http(s) URL` |
+| `Region` | ignored | | still part of the spec hash: changing it replaces the VM |
+
+All three checks run before any backend call. The guide documents the ignored
+field and that catalogue edits affect only new VMs.
+
+**Evidence:** contract §2.1 (reject early, MAY ignore with documentation);
+`IsHTTPURL` in `internal/provider/config_kvm.go`.
+
+## 4. Image source
+
+**Answer:** a catalogue in the config block (`images.<key>.url`, optional
+`sha256`, `format` qcow2 or raw, `firmware` uefi or bios) or a URL given
+directly in `placement.image`. The provider downloads the image into the
+host's storage pool on first use and caches it there as a base volume named
+`base-<entry>-<key>.<format>` (for a URL: `<key>` is the first 16 hex of
+sha256(url)), where `<entry>` is the provider entry's name, with a sidecar
+marker volume `base-<entry>-<key>.<format>.sha256` holding the digest of
+the verified upload, written last. The entry is in the name because the `kvm`
+provider builds `base-<key>.<format>` in the same pool from the same catalogue
+shape, and two entries pointing one key at different URLs would otherwise share
+a file; the cost is one copy per entry per host, and the entry's name is checked
+against `kvmIdentifier` at construction. A base volume without a marker is
+incomplete and is deleted and redone. When `sha256` is configured and differs
+from the marker's content, the image is **not** replaced: that is `ErrPermanent`
+naming both digests, because overlays back onto the base file and libvirt
+unlinks it under running domains without complaint. The download streams
+from the URL through the operator into `StorageVolUpload` with a running
+sha256; nothing is buffered. `Content-Length` is required because the volume's
+capacity must be declared before the upload; a response without it is
+`ErrPermanent` naming the URL. Per `(host, volume)` a keyed mutex makes
+concurrent Creates download once. The download runs on a dedicated
+connection to the host, not the pooled one, so a cancelled poll on another
+Proxy cannot cut it off. After a successful upload `StoragePoolRefresh` makes
+the file usable as a backing image.
+
+Catalogue edit warning for the guide: overlays back onto the base file, so
+evicting a base volume while Proxies use it corrupts those VMs. A `sha256`
+change is refused for that reason; a `url` change is a silent no-op, since the
+cache is keyed by entry and key. Add a new key for either.
+
+**Evidence:** kvm-provider.md "The image cache" and "Known gaps"; execution log
+of 2026-09-04 (raw qcow2 upload plus pool refresh accepted by libvirt 11.6;
+435 MB trixie image); contract §1.4 (stream, no buffering) and §2.1
+(catalogue changes).
+
+## 5. Cloud-init delivery
+
+**Answer:** a NoCloud seed ISO built in-process with
+`github.com/kdomanski/iso9660`, volume label `cidata`, files `user-data` (the
+Proxy's `CloudInit`, byte for byte, never parsed), `meta-data`
+(`instance-id: <name>` and `local-hostname: <name>`, so a reboot is never taken
+for a new instance) and `vendor-data`. The ISO is written into a temp
+directory created under the default temp dir (`/tmp`, a 64Mi emptyDir in the
+manager), uploaded as a raw volume `<name>-seed.iso` and attached as a CD-ROM.
+The seed is rebuilt on every Create attempt: a previous attempt may have been
+cut off mid-upload, and a truncated seed would boot a VM without cloud-init.
+
+`vendor-data` carries a shell script that installs and starts
+`qemu-guest-agent` with whichever of apt-get, dnf, yum, zypper or apk the image
+has. It is a script and not a cloud-config on purpose: cloud-init merges a
+vendor-data cloud-config under the user-data with mergers that keep the
+user-data's value for every top-level key both define and never append lists,
+so a Proxy with its own `packages:` would silently lose the agent. Scripts
+bypass the merge. When the config's `vendorData` is set, the two travel as
+separate parts of a MIME multipart (`text/plain` for the configured document,
+`text/x-shellscript` for the agent script), headed by `MIME-Version` within
+the first bytes. With `installGuestAgent: false` and no `vendorData` the file
+is empty. Size: cloud-init has no practical limit for NoCloud; the ISO is a
+few hundred KB.
+
+**Evidence:** kvm-provider.md "What lands on the host"; execution log run 7
+(vendor-data as YAML silently dropped) and run 8 (script survives); contract
+§2.1 (byte for byte, never logged) and §11 (sentinel test).
+
+## 6. Machine type mapping
+
+**Answer:** a catalogue: `machineTypes.<key>` is `{vcpus, memoryMiB}`,
+validated at load (both greater than zero, at least one profile). Disk size is
+provider-wide: `diskSizeGb` (default 10) is the capacity of every VM's overlay
+volume, thin-provisioned; cloud-init's growpart expands the root filesystem
+into it.
+
+**Evidence:** `KVMMachineType` and `validateKVM` in
+`internal/provider/config_kvm.go`.
+
+## 7. Network and IP discovery (`readiness`, `ip-source`)
+
+**Answer:** `readiness`: `provisioning-window` (boot plus cloud-init, plus the
+image download on a cold pool). `ip-source`: `discovered`. The VM gets one
+virtio NIC on `hosts[].bridge`, which must lead to a LAN with DHCP that the
+cluster can route to. The address comes from `DomainInterfaceAddresses`, guest
+agent source first, then the ARP table (on beelink the agent answered about
+20 s after start where ARP never did, because the host itself does not talk to
+the guest). The first IPv4 that is not loopback, link-local or unspecified is
+reported. One source failing means "no address yet" and `Get` returns
+`StateProvisioning`, which is the ordinary boot on beelink: the agent is not up
+while ARP answers with nothing. A connection already gone, or every source
+failing, is `ErrTransient` instead — a running VM whose address cannot be read
+is not a VM without an address, and reporting it as one would make the
+reconciler clear its IP and the health engine drop its counters. The guest needs the agent: it is installed by the
+vendor-data script (entry 5) unless `installGuestAgent: false`, which needs an
+image that ships it or a network where ARP sees guest traffic. `Port` is not
+used by the provider; nothing on the host filters it. Reachability is a
+prerequisite in the guide: the operator's pod network and the clients must
+route to the bridge's LAN.
+
+**Evidence:** kvm-provider.md "Hypervisor prerequisites" and "Known gaps"
+(IP discovery); execution log run 6 (agent 20 s, ARP never); contract §1.3
+and §2.2 (Running only with IP).
+
+## 8. GC marks and CreatedAt (`gc-mark`, `created-at`)
+
+**Answer:** `gc-mark`: `written-metadata`. `created-at`: `written-at-create`.
+The domain XML passed to `DomainDefineXML` carries, under `<metadata>`, an
+element `<instance xmlns="https://gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/kvm2">`
+with children `<managed>true</managed>`, `<uid>req.UID</uid>` and
+`<createdAt>RFC3339 UTC</createdAt>`. The mark lands in the same call that
+defines the domain, so there is no mark window. Read back with
+`DomainGetMetadata(dom, DomainMetadataElement, uri, DomainAffectCurrent)`;
+libvirt 11.6 returns the element without its namespace declaration, so the
+parser matches on the local name `instance` only (the URI was already used to
+select it). A domain without the element is not managed and reports an empty
+UID. A `createdAt` that does not parse gives the zero time, which GC treats as
+old; that is accepted. The volumes carry no mark (documented GC blind spot,
+entry 24). The namespace is kvm2's own, so kvm2 never lists or adopts domains
+made by the `kvm` provider; the two types may share a host.
+
+**Evidence:** execution log run 3 (namespace asymmetry);
+`DomainGetMetadata(Dom Domain, Type int32, Uri OptString, Flags DomainModificationImpact)`
+in the go-libvirt module; contract §2.1 (mark in the create call) and §4.5.
+
+## 9. State mapping (`provisioning-state`)
+
+**Answer:** `provisioning-state`: `create-blocks`. Create defines and starts the
+domain in one call and returns only once `DomainGetState` reports running, so
+`Get` never sees a domain kvm2 defined but did not start. A domain that is
+shut off later stopped on its own or was stopped by someone, and Stopped is
+the right answer.
+
+| Backend state (`DomainGetState`) | Lock or task | Provider state |
+|---|---|---|
+| nostate (0) | none | Provisioning (libvirt could not determine it; next poll decides) |
+| running (1), blocked (2) | no IP found | Provisioning |
+| running (1), blocked (2) | IP found | Running |
+| paused (3), shutdown (4), shutoff (5), crashed (6), pmsuspended (7) | none | Stopped |
+| any other code | none | Stopped |
+
+libvirt has no locks or tasks; every RPC is synchronous.
+
+**Evidence:** go-libvirt `DomainState` constants; contract §2.2 and §4.11.
+
+## 10. Create: blocking and task tracking (`create-return`, `async-ops`)
+
+**Answer:** `create-return`: `blocking`. `async-ops`: `none`. Create returns
+when the domain is running, holding one reconcile worker for the duration,
+including the first image download on a host (minutes on a cold pool; the
+guide says so). The wait after `DomainCreate` is capped at 15 seconds:
+`DomainCreate` is synchronous in libvirt, so the first `DomainGetState`
+normally answers running and the cap only covers state settling. Two rules
+keep the wait from pinning a worker: a state after the start that is neither
+active nor `nostate` (shutoff, crashed, paused, pmsuspended, shutdown) is
+`ErrTransient` at once, since polling will not bring that guest up; and a
+domain found paused or pmsuspended before the start is `ErrPermanent`
+(`cannot be started by the operator; resume or destroy it on the host`), an
+admin's doing that `DomainCreate` would refuse and the reconciler must not
+retry. Every RPC has its own deadline (entry 18).
+
+Steps, each idempotent, in order, all under the per-name lock (entry 19):
+
+1. Validate placement (entry 3). No backend call yet.
+2. `DomainLookupByName`. Found: adoption guard (entry 2), then start it if it
+   is not running (paused or pmsuspended is `ErrPermanent`; `DomainCreate`;
+   `ErrOperationInvalid` means already running), poll as in step 9, return
+   the ID.
+3. Pool space precheck (entry 13, quota).
+4. Ensure the base image (entry 4): marker present and matching means done.
+5. Overlay volume `<name>.qcow2`, qcow2 backed by the base file with
+   `diskSizeGb` capacity; `ErrStorageVolExist` means reuse (creation is
+   atomic, so an existing overlay is complete).
+6. Seed volume `<name>-seed.iso`: delete if it exists, create, upload
+   (entry 5). Always rebuilt.
+7. `StorageVolGetPath` for the overlay and the seed.
+8. `DomainDefineXML` with file-type disks (entry 15's AppArmor note), the
+   metadata element, firmware per image.
+9. `DomainCreate`, then poll `DomainGetState` until running: `nostate` is
+   polled again, any other inactive state is `ErrTransient` at once, and the
+   15-second cap is `ErrTransient` too.
+10. Return `hosts/<host>/domains/<name>`.
+
+A retry after a crash finds whatever the previous attempt got to: an existing
+overlay is reused, a seed is rebuilt, a defined domain is started, a running
+domain is returned. `ErrNoDomain` and `ErrNoStorageVol` inside Create (a
+volume vanishing between steps) are `ErrTransient`, never `ErrNotFound`. The
+log line for a Create records `cloudInitBytes`, never the content.
+
+**Evidence:** contract §2.1, §4.2, §5 ("do not return NotFound from Create");
+execution log run 1 (AppArmor).
+
+## 11. Delete and secondary resources (`delete-return`, `secondary-resources`)
+
+**Answer:** `delete-return`: `waits`. `secondary-resources`: `explicit`. Under
+the per-name lock:
+
+1. Parse the ID (`ErrPermanent` if malformed). A host no longer in the config
+   is a logged no-op success: the provider cannot reach it and refusing would
+   pin the finalizer; the guide tells admins to delete Proxies before retiring
+   a host.
+2. `DomainLookupByName`; `ErrNoDomain` means the domain is already gone, skip
+   to step 5.
+3. `DomainDestroy`; `ErrOperationInvalid` (not running) and `ErrNoDomain` are
+   success.
+4. `DomainUndefineFlags` with `DomainUndefineNvram`,
+   `DomainUndefineManagedSave` and `DomainUndefineSnapshotsMetadata`
+   (UEFI domains keep NVRAM; undefine refuses without the flag); `ErrNoDomain`
+   is success.
+5. Delete `<name>.qcow2` and `<name>-seed.iso` by name from the host's pool;
+   `ErrNoStorageVol` is success, and `ErrNoStoragePool` means nothing is left
+   to clean.
+6. Return nil. The base image and its marker stay (shared cache).
+
+Everything is gone when Delete returns nil, so `Get` afterwards is NotFound.
+A second Delete during the first waits on the lock and then finds nothing.
+Volumes that exist without a domain (a crash between step 5 and 8 of Create)
+are removed by the same Proxy's next Delete but are invisible to GC (entry 24).
+
+**Evidence:** kvm-provider.md "Known gaps"; execution log run 4 (NVRAM);
+contract §2.3.
+
+## 12. ListByTag scope and partial failure (`list-partial`)
+
+**Answer:** `fail-whole`, per host, with single unreadable domains skipped.
+Scope: every host in `hosts`, in config order, `ConnectListAllDomains` with
+active and inactive domains. For each domain, read the metadata element; keep
+those with `managed` true and build the `Instance` from the domain name and
+the element alone: `ID`, `UID` and `CreatedAt`, which is all GC reads
+(contract §2.4). `State` and `IP` are left empty; no state or address RPC is
+made. A domain whose metadata cannot be read (mid teardown, corrupt element)
+is skipped with a V(1) line, unless ctx ended, which fails the call; a host
+that cannot be dialled or whose listing fails fails the whole call, so GC
+skips this provider for the sweep rather than act on a partial view.
+Filtering is client-side; libvirt has no label index. Cost: exactly one
+`DomainGetMetadata` per domain per sweep, about half a millisecond each over
+the tunnel.
+
+**Evidence:** execution log run 6 (per-domain metadata cost); contract §2.4
+and §4.7.
+
+## 13. Error classification (`not-found`, `quota`)
+
+**Answer:** `not-found`: `error-code`. `quota`: `detected`, from a pool-space
+precheck rather than a backend error: before creating any volume, Create reads
+`StoragePoolGetInfo` and compares `available` with the bytes it is about to
+commit: the image's `Content-Length` when the base is not cached, plus
+`diskSizeGb` in bytes, plus 16 MiB for the seed. Below that it returns
+`ErrQuotaExceeded` naming the pool, available and needed, and the reconciler
+waits `QuotaRetry`.
+
+One `classify(op string, err error)`:
+
+| Backend response | Class | Why |
+|---|---|---|
+| `libvirt.Error` code `ErrNoDomain`, `ErrNoStorageVol` | NotFound (Transient when `op == "create"`) | the object is gone; inside Create it vanished mid-call |
+| `ErrNoStoragePool` | Permanent | config names a pool that does not exist |
+| `ErrXMLError`, `ErrNoSupport`, `ErrInvalidArg` | Permanent | the request itself is wrong |
+| `ErrAuthFailed`, `ErrAccessDenied` | Permanent | the SSH user may not use the socket |
+| `ErrOperationInvalid` | Transient (Delete and Create treat the specific cases as success before classify) | state race; the next poll decides |
+| any other `libvirt.Error` | Transient | unknown means retry |
+| SSH: `ssh: unable to authenticate` (from `x/crypto/ssh`) | Permanent | the key is not in `authorized_keys` |
+| SSH: `knownhosts.KeyError` with a known key that differs, or with no known key | Permanent | host key changed or missing from known_hosts; the admin must act |
+| SSH key or known_hosts file missing (`fs.ErrNotExist`) | Transient | the Secret has not landed yet |
+| SSH key present but does not parse | Permanent | wrong file in the Secret |
+| `context.Canceled`, `context.DeadlineExceeded`, `net.OpError`, `io.EOF`, `io.ErrUnexpectedEOF` | Transient | network or cancellation |
+| HTTP 404, 403, 410 on the image URL; missing `Content-Length`; sha256 mismatch | Permanent | the catalogue entry is wrong |
+| HTTP 5xx, 429, network errors on the image URL | Transient | retry |
+| an error already wrapped with a class (`errors.Is` any sentinel) | that class | validation and builders classify at the source |
+| anything else | Transient | contract default |
+
+The unit test table uses a `libvirt.Error` per code, the real SSH error
+strings produced by `x/crypto/ssh` against the in-process server, and a real
+`knownhosts.KeyError`.
+
+**Evidence:** go-libvirt `ErrorNumber` constants and `Error{Code uint32, Message}`
+(value receiver, so `errors.As` uses a `libvirt.Error` value); contract §5;
+kvm-provider.md "Known gaps" (SSH auth retried forever in v1: closed here).
+
+## 14. Credentials and TLS (`credentials`)
+
+**Answer:** `mounted-files`. The config block names `sshKeyFile` (private key,
+OpenSSH or PEM, parsed with `ssh.ParsePrivateKey`) and `knownHostsFile`
+(OpenSSH `known_hosts`, parsed with `knownhosts.New`), both expected under the
+existing optional `kvm-ssh` Secret mount at `/etc/proxy-operator/kvm`
+(`config/manager/manager.yaml`; the mount is shared with the `kvm` provider and
+its comment says so). `insecureIgnoreHostKey: true` replaces known_hosts with
+`ssh.InsecureIgnoreHostKey()` and is logged as a warning by `New`. Both files
+are read on every dial, so rotating the Secret takes effect on the next
+connection without a restart. `New` tries to read them once to log a warning
+when they are unusable, but never fails on them and never dials.
+
+Minimum privilege: the SSH user must be able to open the libvirt socket
+(member of the `libvirt` group); libvirt itself is used as `qemu:///system`
+through the socket, so no further host permissions.
+
+TLS: none (SSH provides transport security); `system-roots` for the image
+download's HTTPS, which uses the distroless image's CA bundle.
+
+Hygiene: the private key bytes and the known_hosts content never appear in
+logs, spans or error strings; errors from parsing say "parsing ssh key file
+<path>" plus the parser's message. Tested with a sentinel placed in a bad key
+file.
+
+**Evidence:** kvm-provider.md "Cluster-side setup"; execution log run 8
+(lazy load: a missing Secret must not abort the provider set); contract §7.
+
+## 15. Config block
+
+**Answer:** the block is `kvm2:` and reuses the Go type `provider.KVMConfig`
+from `internal/provider/config_kvm.go` unchanged, so an existing `kvm-home`
+entry ports by changing `type: kvm` to `type: kvm2` and `kvm:` to `kvm2:`.
+Validation calls the shared validator with the block name `kvm2` so messages
+read `kvm2.hosts` and not `kvm.hosts`.
+
+| Field | Required | Default | Validation | Ends up in |
+|---|---|---|---|---|
+| `sshKeyFile` | yes | | non-empty | dial |
+| `knownHostsFile` | unless `insecureIgnoreHostKey` | | non-empty | dial |
+| `insecureIgnoreHostKey` | | false | | dial, warning at `New` |
+| `diskSizeGb` | | 10 | not negative | overlay capacity, quota precheck |
+| `vendorData` | | | | seed `vendor-data` |
+| `installGuestAgent` | | true | | seed `vendor-data` |
+| `images.<key>` | | | key matches `kvmIdentifier` | volume name `base-<entry>-<key>.<format>` |
+| `images.<key>.url` | yes per entry | | http(s) URL | download |
+| `images.<key>.sha256` | | | 64 hex | marker comparison |
+| `images.<key>.format` | | qcow2 | qcow2 or raw | base volume XML, overlay backing format |
+| `images.<key>.firmware` | | uefi | uefi or bios | domain XML `<os>` |
+| `machineTypes.<key>` | at least one | | vcpus and memoryMiB greater than zero | domain XML |
+| `hosts[].name` | yes | | unique, matches `kvmIdentifier` | providerID segment |
+| `hosts[].address` | yes | | non-empty `host:port` | TCP dial |
+| `hosts[].user` | yes | | non-empty | SSH |
+| `hosts[].socket` | | `/var/run/libvirt/libvirt-sock` | | SSH channel target |
+| `hosts[].pool` | | `default` | | every storage RPC |
+| `hosts[].bridge` | yes | | non-empty | domain XML NIC |
+
+Wiring: `ProviderConfig.KVM2 *KVMConfig` with tag `kvm2,omitempty`;
+`setBlocks` reports `kvm2`; the allow-list in `Config.validate` and the
+constructor map in `cmd/main.go` both gain `kvm2` in the same commit.
+
+**Evidence:** `KVMConfig`, `validateKVM`, `kvmIdentifier` in
+`internal/provider/config_kvm.go`; contract §10.
+
+## 16. Checker
+
+**Answer:** one target per `hosts[]` entry, `Name` the host name, `Address`
+the SSH address, checked concurrently in config order on a fresh dedicated
+connection each time (dial, `ConnectGetLibVersion`, `ConnectGetHostname`,
+`StoragePoolLookupByName`, `StoragePoolGetInfo`, close). It never touches the
+pooled connection, so a check that runs out of time closes only its own
+connection and cannot disrupt a reconcile. Cost: one SSH handshake and four
+cheap RPCs per host per round.
+
+Reasons: `Reachable`; `Timeout` (the round's ctx ended before the check
+finished, whichever stage was running, the dial included, or an RPC's own
+budget ran out once the connection was up); `Unreachable` (TCP connect or the
+SSH transport failed, including on the dial's own 10 s budget while the round
+still had time, or the link died mid-check); `Unauthorized` (SSH authentication refused, host key unknown or
+changed); `Forbidden` (`ErrAccessDenied` or `ErrAuthFailed` from libvirt: the
+socket refused the user); `APIError` (any other libvirt error);
+`CredentialsUnavailable` (the key or known_hosts file is missing or
+unparsable, i.e. the Secret is absent or wrong: fixed in the cluster, not on
+the host); `PoolNotFound`; `PoolNotActive`.
+
+Details on success: `user`, `hostname`, `libvirtVersion` (decoded
+major.minor.micro), `pool`, `poolState`, `poolCapacityGiB`, `poolAvailableGiB`
+(the closest thing to a capacity signal, and the same number the quota
+precheck reads). The `status.type` comment in
+`api/v1alpha1/providerhealth_types.go` gains `kvm2` and the CRD is
+regenerated.
+
+**Evidence:** contract §8; `StoragePoolGetInfo` returning state, capacity,
+allocation, available; `ConnectGetLibVersion` returning major*1e6 + minor*1e3
+ micro.
+
+## 17. Client and child spans (`client-kind`, `child-spans`)
+
+**Answer:** `client-kind`: `rpc-over-ssh`. `child-spans`: `span-decorator`.
+
+The seam is an interface over exactly the go-libvirt methods the provider
+uses, with `ctx` as the first parameter of every method (go-libvirt's own
+methods take none and block): `DomainLookupByName`, `DomainDefineXML`,
+`DomainCreate`, `DomainDestroy`, `DomainUndefineFlags`, `DomainGetState`,
+`DomainGetMetadata`, `DomainInterfaceAddresses`, `ConnectListAllDomains`,
+`StoragePoolLookupByName`, `StoragePoolRefresh`, `StoragePoolGetInfo`,
+`StorageVolLookupByName`, `StorageVolGetPath`, `StorageVolCreateXML`,
+`StorageVolUpload`, `StorageVolDownload`, `StorageVolDelete`,
+`ConnectGetLibVersion`, `ConnectGetHostname`, plus `Close`. One production
+adapter wraps `*libvirt.Libvirt` and applies the per-call deadline
+(entry 18); one in-memory fake implements the seam for unit tests. A dialer
+seam (`dial(ctx, host) (backend, error)`) separates the SSH transport so tests
+inject fakes and an in-process SSH server test covers the real dialer.
+
+The decorator is applied once, inside the real dialer, at the point where a
+connection is created, so provider code never sees an undecorated backend. Per
+call it starts a span `kvm2.<Method>` (for example `kvm2.DomainDefineXML`) with
+attributes `kvm2.host` and the identifiers passed (`kvm2.domain`, `kvm2.pool`,
+`kvm2.volume`), records the libvirt error code on failure, and inside the span
+writes the V(1) log line, so the line carries that span's ID. Two more spans:
+`kvm2.dial` around the SSH handshake and libvirt connect (attributes address
+and user) and `kvm2.download` around an image fetch (attributes URL host,
+bytes, whether a sha256 was configured). All spans are children of the
+decorator's `provider.create`, `provider.get`, `provider.delete`,
+`provider.list` or the monitor's `providerhealth.check`. No CloudInit, key
+material or upload bodies in attributes. A reflection test walks the seam
+interface and asserts every method produces exactly one span and one log
+line.
+
+**Evidence:** contract §4.15, §4.18, §9; `tracing.Start` in
+`internal/tracing/logger.go`.
+
+## 18. Timeouts (`timeouts`)
+
+**Answer:** `per-call`. The adapter runs each blocking go-libvirt call in a
+goroutine and waits on the result, `ctx.Done()` and a timer:
+
+| Call | Budget |
+|---|---|
+| TCP connect plus SSH handshake plus libvirt connect | 10 s |
+| ordinary RPC | 30 s |
+| `DomainCreate` | 90 s |
+| `StorageVolUpload` (seed) | 60 s |
+| `StorageVolUpload` (image), HTTP download | ctx, capped at 20 m |
+| wait for running after `DomainCreate` | 15 s (entry 10) |
+| HTTP client for images | no client timeout; the per-request ctx carries the cap |
+
+When a call is abandoned (ctx cancelled or budget exceeded), the adapter
+closes the underlying connection, because that is the only way to unblock
+go-libvirt, and marks it dead so the next call redials; the abandoned call
+returns `ErrTransient` wrapping the ctx or deadline error. Consequence,
+documented in the guide: other RPCs in flight on that host's pooled connection
+fail once with Transient and their callers retry. The image download uses a
+dedicated connection so the expensive case (a cancelled poll killing a
+435 MB upload) cannot happen. The keepalive (`keepalive@openssh.com` request
+every 15 s) marks the connection dead on failure.
+
+**Evidence:** contract §1.5 (no caller deadline; a hung ListByTag stalls GC),
+§4.16; go-libvirt methods take no ctx.
+
+## 19. Concurrency
+
+**Answer:** shared state is the host connection table (one pooled connection
+per host, dialled lazily, under a mutex with a per-host dial mutex so
+concurrent callers dial once and share the result), the per-`(host, volume)`
+download locks, and a lock per instance name. Both lock sets are one instance
+each of `keyedLocks`: a one-slot semaphore per key, so a waiter gives up when
+its ctx ends, reference-counted so an entry is removed from the map once no
+caller holds or waits for it and the map does not grow with every name ever
+seen, while two waiters on one key always share one semaphore. Create and
+Delete for the same name run under that name's lock, so two Deletes (GC and
+the reconciler) never race on destroy and undefine, and a Delete cannot
+interleave with a Create's steps. Get and ListByTag take no name lock. go-libvirt multiplexes
+RPCs on one connection, so calls for different names proceed concurrently
+on the shared connection. Three reconciles, a GC sweep and a check round at
+once are safe; the check uses its own connections.
+
+**Evidence:** contract §1.2 (concurrency), §4.15.
+
+## 20. Logging (`logging`)
+
+**Answer:** `tiered`. Logger from `logf.FromContext(ctx)` on every call,
+named `kvm2`, with `provider` on every line and `host` and `name` on every
+line about an instance.
+
+- V(1): one line per operation outcome (`kvm2 domain ensured running`,
+  `kvm2 domain fetched` with state, ip and uid, `kvm2 domain deleted`,
+  `kvm2 domains listed` with counts), one `libvirt call` line per RPC from the
+  decorator (`rpc`, identifiers, `duration`, on failure `error`, `libvirtCode`
+  and `libvirtError` name), connection events (dialled, dropped with reason),
+  and a line per failed backend call with the raw error.
+- V(2): the same lines also carry domain and volume XML, the metadata element,
+  interface addresses, listed domain names, and the resolved Create parameters
+  (`cloudInitBytes`, never the content).
+- No V(5) wire tier; go-libvirt exposes no hook for it.
+- Never logged at any level: `CloudInit`, key material, known_hosts content,
+  upload or download bodies.
+
+**Evidence:** contract §4.17, §9.
+
+## 21. Tests (`backend-test`)
+
+**Answer:** `backend-test`: `integration-tag`. Unit tests run over the
+in-memory fake behind the seam (and an in-process `x/crypto/ssh` server for
+the dialer); the integration test is `kvm2_integration_test.go` behind
+`//go:build integration`, skipped unless `KVM2_TEST_HOST`, `KVM2_TEST_USER`,
+`KVM2_TEST_SSH_KEY`, `KVM2_TEST_BRIDGE` and `KVM2_TEST_IMAGE_URL` are set,
+optional `KVM2_TEST_KNOWN_HOSTS` (without it host keys are not verified and the
+log says so), `KVM2_TEST_IMAGE_SHA256`, `KVM2_TEST_POOL`, `KVM2_TEST_SOCKET`,
+`KVM2_TEST_TIMEOUT` (default 20m), `KVM2_TEST_KEEP`, `KVM2_TEST_LOG_LEVEL`;
+run with `make test-integration-kvm2`. It asserts, in order: `New` returns
+without dialling; `Check` reports the host Ready with details; `Create`
+returns `hosts/<host>/domains/<name>`; a second `Create` returns the same ID
+and the host's domain count is unchanged; `Get` reaches Running with an IPv4
+within the budget and its `UID` and `CreatedAt` equal what Create wrote;
+`ListByTag` contains it; `Delete` returns nil; `Get` is `ErrNotFound`;
+`Delete` again returns nil; the pool holds the base image and its marker and
+no `<name>.qcow2` or `<name>-seed.iso`; with an in-memory span recorder,
+`kvm2.DomainDefineXML` and `kvm2.dial` spans are children of
+`provider.create`.
+
+Contract §11 list, with the test that covers each item:
+
+- [x] repeat `Create`: `TestCreate_repeatReturnsSameIDAndCreatesNothing`
+- [x] `Get` after `Delete`: `TestGet_afterDeleteReturnsNotFound`
+- [x] `Delete` twice: `TestDelete_twiceSucceeds`
+- [x] malformed ID: `TestGet_malformedIDIsPermanent`, `TestDelete_malformedIDIsPermanent`, `TestParseProviderID_malformed`
+- [x] unknown state maps to `StateStopped`: `TestMapState_table`
+- [x] `StateRunning` requires an IP: `TestGet_runningWithoutIPIsProvisioning`, `TestGet_runningWithIPIsRunning`
+- [x] `UID` and `CreatedAt` read back: `TestGet_readsUIDAndCreatedAtFromBackend`
+- [x] `ListByTag` excludes unmarked instances: `TestListByTag_excludesUnmarkedDomains`, `TestListByTag_emptyUIDStillListed`
+- [x] `Get` during an in-progress delete: `TestGet_duringDeleteStillReturnsInstance`
+- [x] `Get` during an asynchronous operation or lock: `TestGet_backendBusyIsTransientNeverNotFound`
+- [x] never started is never `StateStopped`: `TestCreate_returnsOnlyWhenRunning`
+- [x] cancelled `ctx` returns promptly: `TestProvider_cancelledContextReturnsPromptly`, `TestCall_cancelUnblocksRPC`
+- [x] `classify` table with captured responses: `TestClassify_table`
+- [x] Checker reasons, expired `ctx`, stable order: `TestCheck_reasonsTable`, `TestCheck_expiredContextReturnsPromptly`, `TestCheck_oneEntryPerHostInConfigOrder`
+- [x] sentinel tests for `CloudInit` and the credential: `TestCreate_cloudInitSentinelNeverLoggedOrInErrors`, `TestSSH_keyMaterialNeverInErrorsOrLogs`
+- [x] child spans under the operation span: `TestObserve_childSpansUnderParent`, `TestObserve_everySeamMethodInstrumented`
+- [x] real-backend lifecycle test: `TestKVM2_lifecycle` and `TestKVM2Check_realHost` in `kvm2_integration_test.go`
+
+Further tests per file are listed in the plan
+(`docs/plans/2026-09-16-1333-kvm2-provider.md`).
+
+**Evidence:** contract §11; docs/testing.md conventions.
+
+## 22. Deployment
+
+**Answer:** the existing `kvm-ssh` mount is reused; the rest is samples and a Makefile target:
+
+- `config/manager/manager.yaml`: no new mount; the existing optional `kvm-ssh`
+  Secret at `/etc/proxy-operator/kvm` serves both `kvm` and `kvm2`, and its
+  comment says so. `/tmp` is already an emptyDir.
+- `config/manager/providers_config.yaml` and
+  `config/samples/providers-config.yaml`: a commented `kvm2-home` entry, the
+  same block as `kvm-home` under `kvm2:`.
+- `config/samples/proxy_kvm2.yaml`, listed in `config/samples/kustomization.yaml`:
+  provider `kvm2-home`, zone `beelink` (matching the shipped config, unlike the
+  `kvm` sample's `nas1`), machine type `small`, image `debian-13`, a cloud-init
+  that installs Squid.
+- Makefile: `test-integration-kvm2` (`go test -race -tags=integration -count=1
+  -v -timeout 40m -run TestKVM2 ./internal/provider/kvm2/`). The Secret helper
+  `make kvm-ssh-secret` already exists and is reused.
+- RBAC: none; the provider does not talk to the cluster's API.
+- `hack/providers-dev.yaml`: no entry (needs credentials).
+
+**Evidence:** contract §12; `config/manager/manager.yaml`.
+
+## 23. Docs
+
+**Answer:** `docs/kvm2-provider.md` (hypervisor prerequisites including OVMF
+and the `libvirt` group and a bridge with DHCP; cluster-side setup reusing the
+`kvm-ssh` Secret; config reference; placement mapping; what lands on the host
+including the metadata namespace; the image cache and the quota precheck; the
+health check reason table; logging tiers and span names; migrating an entry
+from `kvm`, which requires deleting its Proxies first because kvm2 does not
+adopt v1 domains; known gaps). A fourth column in the three tables of
+`docs/providers-existing.md` and a kvm2 section. `docs/architecture.md`:
+components row, a `kvm2` block in §5, the span topology in §10, a §11 row, and
+Decisions bullets (dedicated download connection, checks on fresh connections,
+own metadata namespace, quota precheck, SSH auth failures Permanent). A README
+row. `docs/testing.md` rows and a kvm2 integration section. A `CHANGELOG.md`
+entry after the user confirms.
+
+**Evidence:** contract §12.
+
+## 24. Known limits accepted at design time
+
+**Answer:** nine limits, each with its workaround:
+
+- **Create blocks**, including the first image download per host, holding one
+  of three reconcile workers. Workaround: warm the cache with one Proxy per
+  image per host before creating many.
+- **Volumes without a domain are invisible to GC** (crash between volume
+  creation and define). The same Proxy's next reconcile removes them; a Proxy
+  deleted for good leaves them for `virsh vol-list`.
+- **A host removed from the config orphans its VMs**: Delete is a no-op
+  success for it and GC does not sweep it. Delete the Proxies first.
+- **No adoption of `kvm` (v1) domains**: switching an entry's type requires
+  deleting its Proxies first, or the old domains stay behind.
+- **Cancelling one call drops the host's pooled connection**, so other RPCs
+  in flight on that host fail once with Transient and retry. The download is
+  isolated on its own connection.
+- **`Content-Length` is required** for image downloads; a chunked mirror is
+  rejected as Permanent.
+- **A paused or pmsuspended domain is not started.** A repeat `Create` for
+  a domain in that state is `ErrPermanent` (entry 10), so the Proxy latches
+  as Failed until its spec changes, rather than retrying a `DomainCreate`
+  that libvirt refuses. Resume the domain on the host (`virsh resume`) and
+  touch the spec, or delete the Proxy.
+- **No v1 log-format compatibility**: the `libvirt call` lines have the same
+  keys where the same facts exist, but dashboards keyed on `kvm` logger names
+  need `kvm2`.
+- Contract §13 limits that apply unchanged: config read once per process
+  (adding a host needs a restart; credentials do not), no installation identity
+  in GC, an unconfigured provider latches the Proxy.
+
+**Evidence:** contract §13; entries 4, 8, 11, 18 above.
@@ -0,0 +1,679 @@
+# The kvm2 provider
+
+Runs Managed proxies as VMs on your own libvirt/KVM hypervisors. It is the
+second-generation libvirt provider, built clean-room from
+[provider-contract.md](provider-contract.md) and the design in
+[kvm2-provider-spec.md](kvm2-provider-spec.md), and it ships side by side
+with the first-generation `kvm` provider: both types can be configured at
+once, both take the same config block shape, and both read the same
+`kvm-ssh` Secret. What kvm2 adds over kvm: a child span per libvirt RPC,
+a pool-space precheck that maps to the reconciler's quota wait, SSH
+credentials re-read on every dial, image downloads and health checks on
+connections of their own, an adoption guard on the owning UID, and SSH
+authentication failures that stop retrying.
+
+Like kvm, the operator reaches each host over SSH, forwards the libvirt
+unix socket through that connection and speaks libvirt's own RPC over it,
+so the hosts need nothing beyond `sshd` and `libvirtd` and the operator
+image carries no libvirt libraries. VMs boot a stock distro cloud image
+that the operator downloads into the host's storage pool on first use,
+with the Proxy's cloud-init delivered on a NoCloud seed ISO built
+in-process.
+
+## Hypervisor prerequisites
+
+Per host, once:
+
+1. **libvirt and KVM** (`libvirtd` running, `/dev/kvm` present).
+2. **An SSH user in the `libvirt` group**, so it may open
+   `/var/run/libvirt/libvirt-sock`. The operator authenticates with a
+   private key; nothing else on the host is touched. The socket is used
+   as `qemu:///system`, so no further host permissions are needed.
+3. **A bridge to a network with DHCP** (for example `br0` bridged onto
+   the LAN). Each VM gets one virtio NIC on it and takes its IP from that
+   network's DHCP. The operator's pod network and the crawler clients
+   must route to those IPs: the health engine probes `http://<ip>:<port>`
+   through the proxy and clients get the same address from the discovery
+   API. Nothing on the host filters the port; the provider does not use
+   `spec.port` at all.
+4. **A storage pool** (default `default`, usually `/var/lib/libvirt/images`)
+   with room for one base image per distro and provider entry (a few
+   hundred MB each) plus the
+   per-VM overlays (`diskSizeGb` each, thin-provisioned). The pool must be
+   running: a stopped pool fails the health check with `PoolNotActive` and
+   fails `Create` permanently.
+5. **UEFI firmware for guests**: the `ovmf` package on Debian/Ubuntu,
+   `edk2-ovmf` on Fedora/RHEL. VMs boot UEFI by default with secure boot
+   off (libvirt picks the OVMF binary itself); a domain start fails with
+   "no firmware found" without it. Only `firmware: bios` images skip this.
+6. **Outbound internet from the VMs**, in practice: a vendor-data script
+   installs `qemu-guest-agent` from the distro mirrors on first boot, and
+   the agent is how the operator learns the VM's address. Without it the
+   operator falls back to the host's ARP table, which only sees traffic
+   addressed to the host itself, so on a plain bridge a VM without the
+   agent stays `Provisioning`.
+
+Check from the operator's point of view before wiring it in:
+
+```sh
+ssh novakj@beelink virsh -c qemu:///system list --all      # user can drive libvirt
+ssh novakj@beelink virsh -c qemu:///system pool-info default
+ssh novakj@beelink ip link show br0
+```
+
+## Cluster-side setup
+
+1. Create the SSH Secret from a key pair dedicated to the operator and a
+   `known_hosts` file holding every host's key. It is the same `kvm-ssh`
+   Secret the `kvm` provider uses; one Secret serves both types, and if
+   you already have it there is nothing to do here.
+
+   ```sh
+   # once: a key pair for the operator; the .pub goes on every host
+   ssh-keygen -t ed25519 -f ~/.ssh/proxy-operator_ed25519 -N '' -C proxy-operator
+   ssh-copy-id -i ~/.ssh/proxy-operator_ed25519.pub novakj@beelink
+
+   # once per host: its host key, under the address the *operator* will
+   # dial (hosts[].address), which may differ from what your laptop uses
+   ssh-keyscan -H beelink.tail8c69c.ts.net >> ~/.ssh/kvm_known_hosts
+
+   # the Secret: the PRIVATE key (not the .pub) plus known_hosts
+   kubectl -n egress-proxies-operator-system create secret generic kvm-ssh \
+     --from-file=id_ed25519=$HOME/.ssh/proxy-operator_ed25519 \
+     --from-file=known_hosts=$HOME/.ssh/kvm_known_hosts
+   ```
+
+   `make kvm-ssh-secret` runs that last command as an apply, so it also
+   updates an existing Secret; override the paths with
+   `KVM_SSH_KEY=... KVM_KNOWN_HOSTS=...`. Putting the `.pub` in the
+   Secret is the classic mistake: the target's `kvm-ssh-secret` check
+   refuses a file without `PRIVATE KEY` in it, and the provider itself
+   would fail every call permanently with a "parsing ssh key file" error.
+   Without `known_hosts`, set `insecureIgnoreHostKey: true` and accept
+   that a MITM could hand you a fake hypervisor.
+
+   `config/manager/manager.yaml` mounts the Secret (optional) at
+   `/etc/proxy-operator/kvm` as two files, `id_ed25519` and
+   `known_hosts`, which is what `sshKeyFile` and `knownHostsFile` in the
+   provider config point at. The deployment starts fine without the
+   Secret, even with a kvm2 provider configured: `New` only checks that
+   the files exist and logs a `WARNING:` line when one does not; the
+   provider's proxies fail transiently and recover once the Secret lands.
+
+   **Rotating the key or a host key needs no restart.** Both files are
+   read and parsed on every dial (`loadCredentials` in
+   `internal/provider/kvm2/ssh.go`); nothing is cached. Re-run the
+   command or the make target, and the next connection, health check
+   included, uses the new material. Connections that are already open
+   stay open until they drop.
+
+2. Add a provider entry to the providers config
+   (`config/manager/providers_config.yaml` in-cluster; full sample in
+   `config/samples/providers-config.yaml`). The type is `kvm2` and the
+   block is `kvm2:`; its fields are those of the `kvm` block:
+
+   ```yaml
+   - name: kvm2-home
+     type: kvm2
+     kvm2:
+       sshKeyFile: /etc/proxy-operator/kvm/id_ed25519
+       knownHostsFile: /etc/proxy-operator/kvm/known_hosts
+       images:
+         debian-13:
+           url: https://cloud.debian.org/images/cloud/trixie/latest/debian-13-generic-amd64.qcow2
+       machineTypes:
+         small: {vcpus: 1, memoryMiB: 1024}
+       hosts:
+         - name: beelink
+           address: beelink.tail8c69c.ts.net:22
+           user: novakj
+           bridge: br0
+   ```
+
+   Validation messages name the block, so a mistake reads
+   `kvm2.hosts[0]: bridge is required`, not `kvm.hosts`. The providers
+   config is read once per process: adding a host or an image entry needs
+   a manager restart; credentials do not.
+
+3. Apply a Proxy (`config/samples/proxy_kvm2.yaml`) and watch it:
+
+   ```sh
+   kubectl get px -w
+   # NAME                MODE      PROVIDER    PHASE          IP             HEALTHY
+   # proxy-kvm2-sample   Managed   kvm2-home   Provisioning
+   # proxy-kvm2-sample   Managed   kvm2-home   Running        192.168.1.73
+   # proxy-kvm2-sample   Managed   kvm2-home   Ready          192.168.1.73   True
+   ```
+
+   The first Proxy for a given image on a given host sits in
+   `Provisioning` for the length of the image download; the operator log
+   shows `fetching base image into host pool` with the URL and size and
+   `base image cached` with the size, duration and digest.
+
+## Config reference
+
+The block is `kvm2:` and its Go type is the shared `KVMConfig` in
+`internal/provider/config_kvm.go`, so the fields, defaults and validation
+are the same as for `kvm`. Defaults are applied by `withDefaults` in
+`internal/provider/kvm2/kvm2.go`.
+
+| Field | Required | Default | Meaning |
+|---|---|---|---|
+| `sshKeyFile` | yes | | private key used for every host; OpenSSH or PEM, parsed on every dial |
+| `knownHostsFile` | unless `insecureIgnoreHostKey` | | OpenSSH `known_hosts` to verify host keys against, read on every dial |
+| `insecureIgnoreHostKey` | | `false` | skip host-key verification entirely; `New` logs a `WARNING:` line when it is on |
+| `diskSizeGb` | | `10` | capacity of each VM's overlay disk; cloud-init's growpart expands the root fs into it. Also what the pool-space precheck reserves per VM |
+| `vendorData` | | | extra cloud-init vendor-data placed on every seed next to the Proxy's user-data. Beware that cloud-init drops any top-level key of a vendor-data **cloud-config** that the user-data also defines. Shipped as one part of a MIME multipart alongside the agent script when `installGuestAgent` is on |
+| `installGuestAgent` | | `true` | add the vendor-data script that installs and starts `qemu-guest-agent`. Turn off for images that ship it already or cannot reach a package mirror; IP discovery then depends on the host's ARP table |
+| `images.<key>` | | | the key names the image in `placement.image`. Letters, digits, dots, dashes and underscores only: it becomes the cached volume name `base-<key>.<format>` |
+| `images.<key>.url` | yes per entry | | http(s) URL of a cloud image (qcow2 or raw). The server must send `Content-Length` |
+| `images.<key>.sha256` | | | expected digest, 64 hex characters; verified after download, a change triggers a re-download |
+| `images.<key>.format` | | `qcow2` | `qcow2` or `raw` |
+| `images.<key>.firmware` | | `uefi` | `uefi` (OVMF, secure boot off) or `bios` (SeaBIOS) for images without an EFI system partition |
+| `machineTypes.<key>` | at least one | | `{vcpus, memoryMiB}` size profile, both greater than zero |
+| `hosts[].name` | yes | | the `placement.zone` value that selects this host; unique. Letters, digits, dots, dashes and underscores only: it becomes part of the providerID `hosts/<name>/domains/<vm>` |
+| `hosts[].address` | yes | | SSH endpoint, `host:port` |
+| `hosts[].user` | yes | | SSH user, member of the `libvirt` group |
+| `hosts[].socket` | | `/var/run/libvirt/libvirt-sock` | libvirt socket path on the host |
+| `hosts[].pool` | | `default` | storage pool for base images, overlays and seeds |
+| `hosts[].bridge` | yes | | host bridge for the VM's NIC |
+
+## Placement
+
+All three fields are checked before any host is dialled. An unknown value
+is `ErrPermanent`, which the reconciler latches as `Failed` with the
+message below until the spec changes.
+
+| Proxy field | Meaning for kvm2 | Error when unknown |
+|---|---|---|
+| `placement.zone` | host name from `hosts`. May be omitted when exactly one host is configured; otherwise required | `placement.zone "x" is not in kvm2.hosts`; with several hosts and no zone, `placement.zone is required when kvm2.hosts lists more than one host` |
+| `placement.machineType` | key in `machineTypes` | `placement.machineType "x" is not in kvm2.machineTypes` |
+| `placement.image` | key in `images`, or an `http(s)://` URL used directly with the defaults (`qcow2`, `uefi`) | `placement.image "x" is neither a key in kvm2.images nor an http(s) URL` |
+| `placement.region` | ignored. It is still part of the spec hash, so changing it replaces the VM and does nothing else | |
+
+The messages come from `Create` in `internal/provider/kvm2/create.go`,
+`host` in `kvm2.go` and `resolveImage` in `image.go`. Fix the spec (which
+replaces the VM) or the config. Catalogue edits affect only new VMs: the
+spec hash covers the placement values, not what the config maps them to.
+
+## What lands on the host
+
+For a Proxy whose deterministic name is `proxy-abc…`, on the chosen
+host's pool:
+
+| Object | Name | Lifetime |
+|---|---|---|
+| base image | `base-<entry>-<key>.<format>`, where `<entry>` is the provider entry's name (for a URL image, `<key>` is the first 16 hex characters of the URL's sha256) | cached; survives Proxy deletion |
+| image marker | `base-<entry>-<key>.<format>.sha256`, a 64-byte raw volume holding the digest | with the base image |
+| overlay disk | `proxy-abc….qcow2`, qcow2 backed by the base file, `diskSizeGb` capacity | with the Proxy |
+| seed ISO | `proxy-abc…-seed.iso`, raw | with the Proxy |
+| domain | `proxy-abc…` | with the Proxy |
+
+The names are built in `internal/provider/kvm2/volume.go`. The base image
+carries the **provider entry's name** because the file is shared by every
+Proxy on that host but must not be shared by two entries: the `kvm`
+provider builds `base-<key>.<format>` in the same pool from the same
+catalogue shape, and two kvm2 entries pointing one key at different URLs
+would otherwise agree on the name and disagree on the contents. The cost
+is one copy of each image per entry per host. For that reason the entry's
+name must itself be a usable file name — letters, digits, dots, dashes and
+underscores — which `New` checks at startup, since provider names are
+otherwise only required to be non-empty and unique.
+
+The domain (`buildDomainXML` in `xml.go`) is a `q35` machine with a
+host-passthrough CPU, ACPI and APIC, memory in MiB and vcpus from the
+machine type, the overlay on `vda` (virtio) and the seed as a read-only
+SATA cdrom on `sda`, one virtio NIC on the bridge, a serial console on a
+pty, the guest-agent channel `org.qemu.guest_agent.0`, a virtio RNG fed
+from `/dev/urandom`, and no graphics. Both disks are `type="file"` with
+the paths `StorageVolGetPath` returned: AppArmor derives QEMU's profile
+from those paths, and volume-type disks fail with permission denied on
+stock Debian and Ubuntu hosts. UEFI images get `<os firmware="efi">` with
+the `secure-boot` feature disabled; BIOS images get neither.
+
+Its `<metadata>` carries one element in kvm2's own namespace,
+`https://gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/kvm2`
+(`metadataURI` in `metadata.go`):
+
+```xml
+<instance xmlns="https://gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/kvm2">
+  <managed>true</managed>
+  <uid>0f3a…</uid>
+  <createdAt>2026-09-16T17:42:11Z</createdAt>
+</instance>
+```
+
+That is how orphan GC recognises its own VMs, since libvirt has no
+labels, and how a repeat `Create` knows whether an existing domain of the
+same name belongs to this Proxy: a domain without the element is a
+permanent error (`carries no kvm2 mark; it was not created by this
+provider`), a different non-empty `uid` is a permanent error naming both
+UIDs, and a marked domain with a matching or empty `uid` is adopted. The
+namespace is kvm2's own, so the `kvm` provider's domains are invisible to
+it (see *Migrating from kvm*).
+
+The seed (`buildSeedISO` in `seed.go`) is an ISO 9660 image labelled
+`cidata` with three files. `user-data` is the Proxy's `cloudInit`, **byte
+for byte**: any format cloud-init accepts works, including a Jinja
+template, a MIME archive, a shell script or a gzipped payload; the
+operator never parses or re-renders it. `meta-data` holds
+`instance-id: <name>` and `local-hostname: <name>`, so a reboot is never
+taken for a new instance. `vendor-data` carries the guest-agent
+bootstrap. The ISO is staged in a temp directory under `/tmp` (the
+manager's 64Mi emptyDir), assembled in memory (tens of KiB) and uploaded
+as a raw volume; the staging directory is removed before the call
+returns. The seed is rebuilt on every `Create` attempt, because a
+previous attempt may have been cut off mid-upload and a truncated seed
+would boot a VM without cloud-init.
+
+The guest agent rides along in `vendor-data` as a **shell script**
+(`guestAgentScript` in `seed.go`), not a cloud-config, on purpose:
+cloud-init merges a vendor-data cloud-config under the user-data with
+mergers that keep the user-data's value for every top-level key both
+define and never append lists, so a Proxy carrying its own `packages:`
+would silently lose the agent. Scripts bypass the merge. The script is
+idempotent, picks whichever of `apt-get`, `dnf`, `yum`, `zypper` or `apk`
+the image has, enables the service with `systemctl` or OpenRC, and logs
+into `/var/log/cloud-init-output.log`. With `vendorData` also configured,
+the two travel as separate parts of a MIME multipart (`text/plain` for
+your document, `text/x-shellscript` for the script), with the
+`MIME-Version` header first so cloud-init detects it. With
+`installGuestAgent: false` the file is your `vendorData` alone, or empty.
+
+## The image cache
+
+The base image is downloaded **by the operator, through libvirt**, into
+the host's pool the first time a Proxy asks for it on that host. Nothing
+is pre-placed on the hypervisor and the operator needs no filesystem
+access there. The bytes stream from the URL through the operator into
+`StorageVolUpload` with a running sha256; nothing is buffered
+(`fetchImage` and `uploadImage` in `internal/provider/kvm2/image.go`).
+
+The upload runs on a **dedicated connection** to the host, dialled for
+the transfer and closed after it, not on the pooled connection the other
+operations share. A cancelled or overlong call kills the connection it
+ran on (see *Timeouts and connections*), and a poll for some other Proxy
+being cancelled must not cut off a 20-minute upload.
+
+A body that fails part way through is reported to the caller **out of
+band**: the reader handed to `StorageVolUpload` turns any source error
+into a clean end of stream and keeps the real error for afterwards, so
+the short upload is caught by the length and digest checks rather than
+by the stream. This is not tidiness. go-libvirt answers a reader error
+by parking its stream sender forever on a channel it no longer reads, so
+handing it one wedges the call and leaks the sender with its 4 MiB
+buffer; see *Known gaps*. For the same reason the download carries its
+own deadline, a minute inside the transfer budget, so a stalled mirror
+ends through its body rather than through the killed socket the budget
+would otherwise leave behind.
+
+The download needs a `Content-Length`: the volume's capacity is declared
+before the upload starts. The request sends `Accept-Encoding: identity`
+so a transparently gzipped body cannot hide the length. A response
+without it, or with length 0, is a permanent error naming the URL, and so
+is any 4xx status; 429 and 5xx are transient and retried.
+
+The marker volume is written **last**, on the pooled connection, after
+`StoragePoolRefresh` has made libvirt read the file's real format. Its
+presence is what marks a base complete: a crash mid-download leaves a
+base volume without a marker, which the next attempt deletes and redoes
+(`cachedImage`). A re-download also happens when the marker's content is
+not a 64-character hex digest, or when `sha256` is configured and differs
+from it. Otherwise the cached copy is used for as long as it exists, even
+if the URL's content changed upstream: a `latest` URL is pinned at the
+moment of first fetch, per host. Concurrent Creates for the same image
+on the same host queue on a per-(host, volume) lock; the first fetches,
+the rest find the marker.
+
+To evict manually:
+
+```sh
+ssh novakj@beelink virsh -c qemu:///system vol-list --pool default | grep base-kvm2-home-
+ssh novakj@beelink virsh -c qemu:///system vol-delete --pool default base-kvm2-home-debian-13.qcow2.sha256
+ssh novakj@beelink virsh -c qemu:///system vol-delete --pool default base-kvm2-home-debian-13.qcow2
+```
+
+Marker first: a base without its marker is refetched, a marker without its
+base is cleaned up, and neither leaves a Proxy pointing at nothing.
+
+**Warning: do not evict a base volume while Proxies using that image exist
+on the host.** The overlays back onto the base file, and libvirt will
+unlink it under running domains without complaint; the guests keep running
+on the open file descriptor and fail at their next start. Delete those
+Proxies first.
+
+Changing `sha256` on an entry whose base is already cached is **refused**,
+permanently and by name, rather than resolved by replacing the file — that
+replacement was the same corruption, just automated. Add the image under a
+new key and move Proxies over by changing their `placement.image` (which
+replaces the VM), or delete the Proxies and evict the base by hand.
+Changing `url` alone does nothing at all: the cache is keyed by entry and
+key, so the new URL is never fetched. Add a new key for that too.
+
+## Quota: the pool-space precheck
+
+Before creating any volume, `Create` reads `StoragePoolGetInfo` and
+compares the pool's free space with what it is about to commit
+(`checkPoolSpace` and `neededBytes` in `image.go`): the image's
+`Content-Length` when the base is not cached yet (found with a `HEAD`
+request, or a `GET` whose body is closed at once when `HEAD` answers
+anything but 2xx, since mirrors refuse it with 405, 403 or 501 alike; the
+`GET`'s status is then what counts), plus `diskSizeGb` in bytes for the
+overlay, plus 16 MiB for the
+seed. Below that it returns `ErrQuotaExceeded` with a message naming the
+pool, the bytes available and the bytes needed. The reconciler answers
+quota with a wait, not a failure: the Proxy shows `Provisioned=False`
+with reason `QuotaExceeded` and is retried after `QuotaRetry` (5
+minutes), off the error backoff curve and never `Failed`. Free up the
+pool or lower `diskSizeGb` and the next attempt goes through.
+
+The precheck is the only quota signal; libvirt itself reports a full pool
+as an ordinary error, which would be transient. The health check exposes
+the same number as `poolAvailableGiB`, so a pool that is filling up shows
+there before Creates start waiting.
+
+A pool that is not running fails the precheck permanently (`storage pool
+<pool> is not running`): start it with `virsh pool-start` and
+`virsh pool-autostart`.
+
+## Migrating from `kvm`
+
+The `kvm2` block is the `kvm` block under another name, so an entry ports
+by changing `type: kvm` to `type: kvm2` and `kvm:` to `kvm2:`. The
+`kvm-ssh` Secret is shared; nothing changes on the hosts.
+
+What does not port is the VMs. kvm2 marks its domains in its own metadata
+namespace and neither lists nor adopts domains the `kvm` provider made:
+`ListByTag` skips them, so orphan GC never sees them, and `Create` refuses
+a domain of the same name that lacks the kvm2 mark with a permanent error
+rather than start it (a Proxy recreated under the new entry gets the same
+deterministic name, so this is what happens when step 1 below is skipped).
+A Proxy's `spec.provider` is immutable, so an existing Proxy cannot be
+moved to the new entry anyway. Do it in this order:
+
+1. Delete the Proxies on the `kvm` entry and wait for them to go; the
+   `kvm` provider removes their domains and volumes.
+2. Change the entry (or add a second one under a new name and keep the
+   old one for a while; both types may share a host).
+3. Restart the manager, then recreate the Proxies with the new
+   `spec.provider`.
+
+A `kvm` domain left behind on a host is invisible to kvm2 forever; clean
+it up on the host with `virsh undefine --nvram <name>` and
+`virsh vol-delete` for its `<name>.qcow2` and `<name>-seed.iso`.
+
+The base image cache is **not** shared between the two types, and that is
+deliberate: kvm names its copy `base-<key>.<format>` while kvm2 names its
+own `base-<entry>-<key>.<format>`, so kvm2 downloads the image once more
+per host rather than adopting a file another provider maintains. Nothing
+needs deleting; the `kvm` base stays where it is and keeps serving the
+overlays that back onto it. Budget the extra few hundred MB in the pool.
+
+## Logging
+
+The provider logs through controller-runtime's logger, named `kvm2`, with
+`provider` (the configured name) on every line, and `host` and `name`
+(the domain) on every line about an instance. Every line inside a traced
+operation also carries `traceID` and `spanID`.
+
+- `--zap-log-level=debug` (logr V(1)):
+  - one line per operation outcome: `kvm2 domain ensured running` with the
+    `id`, `kvm2 domain fetched` with `state`, `ip` and `uid`,
+    `kvm2 domain not found`, `kvm2 domain deleted`, `kvm2 domains listed`
+    with `hosts` and `managed` counts, `kvm2 host checked` with `ready`,
+    `reason` and `latency`; plus the steps inside a Create
+    (`kvm2 domain already defined`, `overlay volume created`,
+    `seed volume uploaded` with `bytes`, `domain defined` with `uid` and
+    `firmware`, `domain started`, `fetching base image into host pool`,
+    `base image cached`);
+  - **one `libvirt call` line per libvirt RPC**, written by the decorator
+    in `internal/provider/kvm2/observe.go` inside that RPC's span: `rpc`
+    (the go-libvirt method name, for example `DomainLookupByName`), the
+    identifiers it was given (`domain`, `pool`, `volume`), `duration`, and
+    on failure `error` plus `libvirtCode` and `libvirtError` (the code's
+    name, for example `ErrNoDomain`). A Get is four or five RPCs; a
+    Create with a cached image is around twenty;
+  - connection events from `internal/provider/kvm2/hosts.go`:
+    `libvirt host connected` with `address`, `user` and `duration`, and
+    `libvirt host connect failed` with the same plus `error` and `stage`
+    (`credentials`, `tcp`, `ssh` or `socket`);
+  - `kvm2 backend call failed` with `op` and the raw `error` once per
+    failed operation, before the error is returned;
+  - a line per domain that `ListByTag` skips, and a line when the image
+    marker forces a refetch.
+- `--zap-log-level=2`: the `libvirt call` lines also carry the domain and
+  volume `xml`, the `metadata` element, the interface `addresses`, the
+  listed `domains`, the pool `state` and sizes and the volume `path`; the
+  `domain defined` line carries the `xml`; `kvm2 create resolved` logs the
+  resolved parameters (pool, bridge, machine type, sizes, image volume,
+  firmware) and `cloudInitBytes`; `kvm2 listed instance` logs each listed
+  domain.
+- Never logged at any level: the cloud-init content, the SSH key, the
+  known_hosts content, and upload or download bodies. The seed and image
+  uploads report `offset` and `length` only. There is no raw wire tier
+  like the GCP provider's V(5); go-libvirt exposes no hook for it.
+
+## Tracing
+
+With an OTLP exporter configured ([architecture.md
+§10](architecture.md#10-tracing-internaltracing)), every backend call is
+a child span of the operation span the decorators open
+(`provider.create`, `provider.get`, `provider.delete`, `provider.list`)
+or of the monitor's `providerhealth.check`:
+
+| Span | Around | Attributes |
+|---|---|---|
+| `kvm2.<Method>`, for example `kvm2.DomainDefineXML` | one libvirt RPC (`observe.go`) | `kvm2.host`, and the identifiers passed: `kvm2.domain`, `kvm2.pool`, `kvm2.volume`; on failure the status is Error and `kvm2.libvirt_code` carries the libvirt error number |
+| `kvm2.dial` | the SSH handshake and the libvirt connect (`hosts.go`) | `kvm2.host`, `kvm2.address`, `kvm2.user` |
+| `kvm2.download` | one image fetch, from the HTTP request to the marker (`image.go`) | `kvm2.host`, `kvm2.volume`, `kvm2.url_host`, `kvm2.sha256_configured`, and `kvm2.bytes` once the response arrived |
+
+All spans are client spans. No attribute carries cloud-init, key material
+or a URL beyond its host. The `libvirt call` log line for an RPC is
+written inside that RPC's span, so its `spanID` points at the span.
+
+## Timeouts and connections
+
+No caller puts a deadline on the context, so every backend call carries
+its own budget (`defaultBudgets` in `internal/provider/kvm2/backend.go`,
+applied by `call`):
+
+| Call | Budget |
+|---|---|
+| TCP connect plus SSH handshake plus libvirt connect | 10 s |
+| ordinary RPC | 30 s |
+| `DomainCreate` | 90 s |
+| `StorageVolUpload` of at most 16 MiB (the seed, the marker) | 60 s |
+| `StorageVolUpload` of an image, `StorageVolDownload`, and the whole image fetch from HTTP request to marker | 20 m; the caller's ctx may end it earlier |
+| wait for running after `DomainCreate` | 15 s, polled every 500 ms (`defaultRunningWait` and `defaultRunningPoll` in `create.go`); `DomainCreate` is synchronous, so the first poll normally answers running, and a domain that is shut off, crashed or paused right after the start fails at once rather than waiting out the cap |
+| the `HEAD` request of the pool-space precheck | 30 s |
+| the courtesy `ConnectClose` when a connection is closed | 5 s |
+
+go-libvirt's calls take no context and block, so the only way to abandon
+one is to close the connection under it. That is what a cancelled or
+overlong call does: the adapter marks the connection dead so the next call
+redials, kills it by closing the SSH client's TCP connection, which takes
+the forwarded socket with it, and returns `ErrTransient` wrapping the
+context or deadline error. It marks dead before it kills, and closes the
+TCP connection rather than the forwarded channel, because both of those are
+lock-free: closing the channel writes a close message under the channel's
+write lock, so it would queue behind a transfer stuck in a TCP write to a
+host that has gone away — for as long as the kernel retransmits, which is
+the one case the kill exists to survive. The consequence:
+**other RPCs in flight on that host's pooled connection fail once with
+`ErrTransient`** and their callers retry on a fresh connection. In the
+reconciler that is one requeue with backoff; in GC it is one skipped
+sweep.
+
+Two callers are kept off the pooled connection for that reason
+(`dialFresh` in `hosts.go`): the image upload, because a 435 MB transfer
+must not die because a poll for another Proxy was cancelled, and the
+health check, because a check that runs out of the monitor's timeout must
+close only its own connection.
+
+Otherwise there is one pooled SSH plus libvirt connection per host,
+dialled lazily on the first call that needs it and shared by every
+operation for that host (go-libvirt multiplexes RPCs on one connection).
+Concurrent callers wait for one dial and share the result; waiting ends
+with the caller's context. A `keepalive@openssh.com` request every 15 s
+marks the connection dead when it fails or gets no reply within another
+15 s, as does go-libvirt noticing the socket is gone; the next call
+redials. Create and Delete for the same domain name are serialised on a
+per-name lock, so two Deletes (GC and the reconciler) never race on
+destroy and undefine, and a Delete cannot interleave with a Create's
+steps; Get and ListByTag take no lock.
+
+## Health check
+
+The operator checks every configured host every `--provider-check-interval`
+(default 1m) and publishes the result in the `ProviderHealth` object named
+after the provider (`kubectl get ph kvm2-home -o yaml`) and in the
+`proxy_operator_provider_target_up{provider,target}` gauge. One target per
+`hosts[]` entry, in config order, all checked concurrently.
+
+Each check dials a **fresh connection** to the host and closes it when
+done (`probeHost` in `internal/provider/kvm2/check.go`); the pooled
+connection is never touched, so a check that runs out of time cannot fail
+a reconcile's call in flight, and a fresh dial picks up a rotated Secret
+at once. The sequence is the SSH and libvirt handshakes, then four RPCs:
+`ConnectGetLibVersion`, `ConnectGetHostname`, `StoragePoolLookupByName`
+and `StoragePoolGetInfo` on the host's `pool`. The cost per host per
+round is one sshd session and those four calls.
+
+Every target's `details` carry `user` and `pool`. As facts arrive they
+gain `libvirtVersion` (decoded `major.minor.micro`), `hostname`,
+`poolState` (virsh's word: `running`, `inactive`, ...),
+`poolCapacityGiB` and `poolAvailableGiB`, so a target that fails late in
+the sequence still says what the host answered before. The free space is
+the number the quota precheck compares against.
+
+A failing target says where the check broke (`checkReason` in `check.go`):
+
+| Reason | Meaning | Look at |
+|---|---|---|
+| `Reachable` | ready | |
+| `CredentialsUnavailable` | the key or known_hosts file is missing or does not parse | the `kvm-ssh` Secret (see *Cluster-side setup*); fixed in the cluster, not on the host |
+| `Unreachable` | the TCP connect failed or ran out of the 10 s dial budget while the round still had time, the SSH transport failed for a reason other than authentication, the forwarded libvirt socket was refused, or the link died mid-check | DNS, Tailscale or a firewall between the cluster and `address`; `libvirtd` running and the `socket` path |
+| `Unauthorized` | the SSH server rejected the key, or the host key is missing from known_hosts or has changed | `user`'s `authorized_keys`; the known_hosts entry (re-run `ssh-keyscan` and update the Secret) |
+| `Forbidden` | SSH worked, but libvirt refused the user (`ErrAccessDenied` or `ErrAuthFailed`) | `user` in the `libvirt` group |
+| `Timeout` | the monitor's `--provider-check-timeout` (default 15s) ran out before the check finished, whichever stage was running, the dial included; or the connection was up and an RPC's own budget ran out | a hung `libvirtd`, a saturated link, or a round timeout shorter than the dial takes |
+| `PoolNotFound` | there is no pool named `pool` | `virsh pool-list --all` |
+| `PoolNotActive` | the pool exists but is not running | `virsh pool-start <pool>` and `virsh pool-autostart <pool>` |
+| `APIError` | any other libvirt error | the message carries libvirt's own text |
+
+## Error classes
+
+Every error is classified once, in `classify` in
+`internal/provider/kvm2/errors.go`, with errors that carry a class from
+their source (validation, the image fetch, the dial) keeping it. The
+reconciler latches `ErrPermanent` as `Failed` until the spec changes,
+waits 5 minutes on `ErrQuotaExceeded`, recreates on `ErrNotFound` from
+`Get`, and retries everything else with backoff.
+
+| Class | What lands there |
+|---|---|
+| `ErrPermanent` | placement values not in the catalogue; a malformed providerID; a host no longer in the config (from `Get`; `Delete` treats it as done); libvirt `ErrXMLError`, `ErrNoSupport`, `ErrInvalidArg`, `ErrAuthFailed`, `ErrAccessDenied` and `ErrNoStoragePool`; a pool that is not running; SSH authentication refused (`unable to authenticate`); a host key missing from known_hosts or changed (`knownhosts.KeyError`); a key or known_hosts file that exists but does not parse; a domain of the same name owned by another UID or carrying no kvm2 mark; a domain of the same name that is paused or pmsuspended; corrupt metadata on a domain; an image URL answering 4xx, a response without `Content-Length` or with length 0, a sha256 that does not match the configured one |
+| `ErrQuotaExceeded` | the pool-space precheck |
+| `ErrNotFound` | libvirt `ErrNoDomain` or `ErrNoStorageVol` from `Get`, `Delete` or `ListByTag`; inside `Create` the same codes are `ErrTransient`, because the contract forbids NotFound from Create |
+| `ErrTransient` | everything else: a missing key or known_hosts file (the Secret has not landed yet), TCP and network errors, context cancellation, an abandoned or budget-exceeded call, a connection killed by another caller, libvirt `ErrOperationInvalid` and every unrecognised libvirt code, an image URL answering 429 or 5xx or ending the body early, every address source failing on an active domain, the wait for running expiring or the domain going shut off or crashed right after its start |
+
+Error strings name paths, hosts, volumes and libvirt's own text. They
+never contain key bytes, the known_hosts content or cloud-init.
+
+## Known gaps
+
+- **An upload that libvirtd itself fails leaks a goroutine, upstream.**
+  go-libvirt's `requestStream` hands its stream sender an unbuffered
+  channel and then, on the error path, returns without ever reading it:
+
+  ```go
+  abort  := make(chan bool)    // rpc.go
+  outErr := make(chan error)
+  go func() { outErr <- l.socket.SendStream(...) }()
+  resp, err = l.processIncomingStream(c, in)
+  if err != nil {
+      abort <- true
+      return resp, err          // outErr is never drained
+  }
+  ```
+
+  Whenever the daemon errors mid-upload — a pool that fills, a volume
+  removed underneath — the sender parks on `outErr` for good, taking its
+  4 MiB buffer with it. Closing the socket does not help: both ends are
+  blocked on channel sends, not on I/O. kvm2 avoids the two cases it
+  controls (a failing body, and its own transfer deadline; see *The image
+  cache*), so what is left needs a server-side failure. The fix is two
+  characters upstream, `make(chan bool, 1)` and `make(chan error, 1)`; if
+  it ever bites hard enough, a `replace` carrying that patch is the
+  escape hatch.
+- **Create blocks**, including the first image download on each host,
+  holding one of the three reconcile workers for its duration (LAN
+  speed, a few hundred MB, once per image per host). Other Proxies keep
+  reconciling on the remaining workers. Warm the cache with one Proxy per
+  image per host before creating many.
+- **Volumes without a domain are invisible to GC.** A crash between
+  creating the overlay or seed and defining the domain leaves volumes
+  that `ListByTag` cannot see (it lists domains). The same Proxy's next
+  Create reuses the overlay and rebuilds the seed, and its Delete removes
+  both by name even with no domain, but a Proxy CR that is gone for good
+  leaves them for `virsh vol-list` to find.
+- **A host removed from the config orphans its VMs.** `Delete` for a
+  providerID naming an unconfigured host is a logged no-op success, so
+  the Proxy finishes deleting instead of pinning its finalizer, but the
+  domain and its volumes stay on the hypervisor and orphan GC cannot see
+  them either (it only sweeps configured hosts). Delete the Proxies before
+  retiring a host, or clean up afterwards with `virsh undefine --nvram`
+  plus `virsh vol-delete`. `Get` fails permanently for such a providerID
+  on purpose: a typo in `hosts` must not look like a missing VM and
+  trigger a rebuild elsewhere.
+- **No adoption of `kvm` domains.** See *Migrating from kvm*.
+- **A cancelled call drops the host's pooled connection once**, so other
+  RPCs in flight on that host fail once with `ErrTransient` and retry.
+  The image upload and the health check are isolated on their own
+  connections.
+- **`Content-Length` is required** for image downloads; a mirror that
+  only chunks is rejected permanently. Point the catalogue at one that
+  sends the length.
+- **A paused or pmsuspended domain is not started.** A repeat `Create`
+  for a domain in that state fails permanently (`is DomainPaused and
+  cannot be started by the operator; resume or destroy it on the host`),
+  so the Proxy shows `Failed` until its spec changes; the operator never
+  resumes a VM an admin paused. Resume it on the host
+  (`virsh resume <name>`) and touch the spec, or delete the Proxy.
+- **Parser text in credential errors.** A key or known_hosts file that
+  does not parse fails with the parser's own message, which can quote a
+  PEM header type or a malformed known_hosts token. Key bytes are never
+  quoted.
+- **The sample cloud-init is an open proxy.**
+  `config/samples/proxy_kvm2.yaml` installs Squid with
+  `http_access allow all`; restrict it before exposing the bridge network
+  to anything but the cluster.
+- **IP discovery needs the agent or guest traffic to the host.** A VM on
+  a bridge with no DHCP never gets an address and stays `Provisioning`
+  forever; check `virsh console <name>` on the host. The same symptom
+  with a booted, addressed VM means the guest agent never came up: check
+  `systemctl status qemu-guest-agent`, `cloud-init status --long` and the
+  vendor script's output in `/var/log/cloud-init-output.log` (a VM without
+  outbound internet cannot install it). A user-data that disables
+  vendor-data (`vendor_data: {enabled: false}`) also removes the agent, by
+  design. A host where the ARP source *errors* rather than answering with
+  nothing is the one case this reads differently: with no source left to
+  answer, a booting VM retries with `ErrTransient` instead of resting in
+  `Provisioning`. It still converges; the giveaway is two "no addresses
+  from source" lines per poll at `-v 2`.
+  `virsh domifaddr <name> --source agent` and `--source arp` show
+  what the operator sees, in that order.
+- **Debian 13 cloud images hang in GRUB under BIOS on a headless VM**
+  ([Debian bug #1111240](https://bugs.debian.org/1111240)), which is why
+  `firmware` defaults to `uefi`; the same image boots under UEFI with the
+  serial console intact. Images that need `firmware: bios` are the
+  exception.
+
+## Testing
+
+Unit tests run over an in-memory fake of the libvirt seam and an
+in-process SSH server. The optional integration test against a real host
+is described in
+[testing.md](testing.md#the-kvm2-integration-test-real-hypervisor); run it
+with `make test-integration-kvm2`. `TestKVM2Check_realHost` runs the
+health check against the same host and creates nothing:
+`go test -tags=integration -count=1 -v -run TestKVM2Check ./internal/provider/kvm2/`.
@@ -0,0 +1,382 @@
+# Execution: kvm2, a clean-room libvirt/KVM provider built from the provider contract
+
+Plan: [docs/plans/2026-09-16-1333-kvm2-provider.md](../plans/2026-09-16-1333-kvm2-provider.md)
+
+- [x] Step 0 — Plan file and branch
+- [x] Step 1 — The spec, `docs/kvm2-provider-spec.md`
+- [x] Step 2 — Config wiring, IDs, state mapping, classify
+- [x] Step 3 — Pure builders: domain and volume XML, metadata, seed
+- [x] Step 4 — Transport, seam, decorator, fakes
+- [x] Step 5 — Image cache and pool-space precheck
+- [x] Step 6 — The four contract methods
+- [x] Step 7 — Checker and the CRD comment
+- [x] Step 8 — Deployment files, Makefile, integration test
+- [x] Step 9 — Documentation
+- [x] Step 10 — Milestone review
+- [ ] Step 11 — Real-host validation, changelog, MR
+
+## Step 0 — Plan file and branch
+
+`main` was fast-forwarded to the merge of PR #7 (`0dc25a1`), which carries the
+three provider documents, and `feat/kvm2-provider` was cut from it. The plan
+was committed alone as `854856c`. The pre-existing local modification of
+`.claude/settings.json` stays uncommitted throughout.
+
+## Step 1 — The spec, `docs/kvm2-provider-spec.md`
+
+2026-09-16 13:37. Written by the orchestrator, not a subagent, since the spec
+is the "specifics" a user hands to the implementer. All 24 entries answered;
+every Evidence line cites the v1 user guide, the 2026-09-04 execution log, a
+go-libvirt symbol, or the contract. Nothing was taken from
+`internal/provider/kvm/`.
+
+Design choices that go beyond the user's profile answers, all recorded in the
+spec: an adoption guard on `Create` (foreign UID is Permanent), a pool-space
+precheck mapped to `ErrQuotaExceeded`, a dedicated connection for image
+downloads so a cancelled poll cannot cut one off, the Checker on fresh
+connections rather than the pool, SSH authentication and host-key failures
+classified Permanent (v1 retried them forever), SSH material re-read on every
+dial so rotation needs no restart, kvm2's own metadata namespace (no adoption
+of v1 domains), and the shared `validateKVM` refactored to take the block name
+so messages say `kvm2.`.
+
+Checks run on the spec:
+
+```sh
+grep -c '^\*\*Answer:\*\*$' docs/kvm2-provider-spec.md      # 0 blank answers
+grep -n 'internal/provider/kvm/' docs/kvm2-provider-spec.md   # empty
+```
+
+plus the axis-code check against contract §4 headings and the path-existence
+loop from the docs-split log. All clean.
+
+Worth noting: the spec's entry 21 lists the concrete test name for every
+contract §11 item, so the later steps' briefs can quote it rather than
+re-derive the mapping.
+
+## Step 2 — Config wiring, IDs, state mapping, classify
+
+2026-09-16 16:38. First subagent step. The brief named the allowed files,
+the spec entries, the test names and the prohibition on
+`internal/provider/kvm/`; the agent reported nothing opened there, with one
+honest caveat: a `golangci-lint` run printed seven single source lines from
+v1 files among its findings, which it did not act on. The review greps
+(`grep -rn 'provider/kvm"' internal/provider/kvm2/`, the v1 identifier
+smell list) are clean.
+
+What landed: `ProviderConfig.KVM2` reusing `KVMConfig`, `setBlocks`, the
+allow-list and dispatch in `Config.validate`, `validateKVM` refactored into
+`validateKVMBlock(prefix, block, c)` with `validateKVM` kept as a wrapper so
+v1 messages and tests are untouched, `"kvm2": kvm2.New` in `cmd/main.go`
+(both type lists in one commit), and the package skeleton: `kvm2.go`
+(`New` never dials, warns on unreadable files, `withDefaults`, `host`,
+`wrapErr`, stubbed methods), `id.go`, `state.go`, `errors.go` with 34
+`classify` rows.
+
+Decisions the agent made that the spec left open, all accepted: `New`
+checks file existence only (parsing belongs to the dialer); `withDefaults`
+copies hosts and images; `firstIPv4` also drops an address whose type tag and
+literal disagree; there is no explicit `fs.ErrNotExist` branch in `classify`
+because it falls to the Transient default anyway.
+
+Commands as run by the agent and re-run by the orchestrator:
+
+```sh
+go vet ./... && go build ./... && go test -race -count=1 ./internal/provider/... ./cmd/...
+go test -short ./...
+```
+
+All packages `ok`.
+
+Worth noting: `bin/golangci-lint` panics inside staticcheck's IR builder on
+this machine (`unexpected expr: *ast.KeyValueExpr` on stdlib `internal/poll`)
+because the local toolchain is Go 1.27.0 while `go.mod` says 1.26 and the
+bundled staticcheck predates it; it reproduces on untouched packages, so lint
+is deferred to Step 10 with a toolchain note. The `.claude/settings.json`
+modification is still local and uncommitted.
+
+## Step 3 — Pure builders: domain and volume XML, metadata, seed
+
+2026-09-16 16:55. Second subagent step, new files only: `xml.go`,
+`metadata.go`, `volume.go`, `seed.go` and their tests (26 tests). The
+domain XML is built from `encoding/xml` structs, both disks are
+`type="file"` with resolved paths (the AppArmor lesson), UEFI is the default
+with the secure-boot feature disabled, the metadata element parses with or
+without its namespace declaration, and the seed ISO is staged under
+`os.TempDir()` (the library uses `os.MkdirTemp("", "")`, so `TMPDIR` is
+honoured; the test points it at a `t.TempDir()` and asserts nothing is left
+behind).
+
+Decisions the agent made, all accepted: `buildDomainXML` defaults an empty
+disk format to qcow2 and rejects anything but qcow2 or raw;
+`buildVolumeXML` rejects a non-positive capacity; `buildSeedISO` rejects an
+empty name; the guest-agent script still runs the enable-and-start step when
+the agent is already installed; the multipart boundary is random, so
+vendor-data is not byte-deterministic across calls (the seed is rebuilt every
+Create anyway).
+
+Commands:
+
+```sh
+gofmt -l ./internal/provider/kvm2 ; go vet ./internal/provider/kvm2/ && go test -race -count=1 ./internal/provider/kvm2/
+go build ./... && go test -short ./...
+```
+
+All `ok`. Review greps for a v1 import or v1 identifiers are empty; the only
+`type="volume"` in the package is the comment explaining why it is not used.
+The agent reported "files read outside the allowed set: none".
+
+## Step 4 — Transport, seam, decorator, fakes
+
+2026-09-16 17:42. Third subagent step, the largest: `backend.go` (the
+ctx-first seam `backend`, the production adapter `libvirtBackend`, a
+generic `call[T]` helper with per-call budgets and the abandon policy),
+`ssh.go` (credentials read on every dial, stage-tagged `dialError`, the
+forwarded socket, keepalive), `hosts.go` (`hostTable` with a per-host
+channel semaphore so waiting for a dial ends with the caller's ctx,
+`dialFresh` for dedicated connections, the production `libvirtDialer`
+starting the `kvm2.dial` span and applying the decorator once),
+`observe.go` (one span `kvm2.<Method>` and one `libvirt call` line per
+RPC), plus an in-memory `fakeBackend`, an in-process SSH server and 41
+tests. The agent also drove the real adapter over go-libvirt's own
+`libvirttest` mock through a `net.Pipe`, which the brief did not ask for.
+
+Abandon policy as built: a call that outlives its ctx or budget kills the
+link (closes the socket and the SSH client), returns Transient wrapping the
+ctx or deadline error, and other in-flight calls on that connection fail once
+with go-libvirt's `ErrInterrupted`; the next `get` redials. Results travel
+through a channel so a late-finishing abandoned call never writes into its
+caller's frame; the agent reported that the first `-race` run caught exactly
+that bug in an earlier version.
+
+Decisions the agent made, accepted: liveness is a separate unexported
+interface so the reflection test over `backend` stays at the 20 RPCs plus
+`Close`; `Close` sends a bounded `ConnectClose` before killing the link;
+`observe` and the dialer take `...tracing.Option` so tests can inject a
+recording tracer provider, as `provider.WithTracing` does.
+
+Clean-room note: three type names coincide with names v1 uses
+(`dialError`, `dialStage`, `hostEntry`). The brief described each concept
+("an error type that records which stage failed", "a per-host entry") so the
+names follow from the description; the greps for a v1 import and for the
+listed v1 identifiers are empty and the agent attested to reading nothing
+under `internal/provider/kvm/`.
+
+Commands (agent and orchestrator):
+
+```sh
+gofmt -l ./internal/provider/kvm2 ; go vet ./internal/provider/kvm2/ && go test -race -count=2 ./internal/provider/kvm2/
+go build ./... && go test -short ./...
+grep -n 'context.Background()' internal/provider/kvm2/*.go | grep -v _test.go   # empty
+```
+
+All `ok`; the agent also ran the timing-sensitive tests at `-count=8`.
+
+## Step 5 — Image cache and pool-space precheck
+
+2026-09-16 18:11. Fourth subagent step: `image.go` and `image_test.go`
+(27 tests), plus additive changes to `fake_test.go` (a `twin()` sharing
+state with its own call log, so a test can tell the pooled connection from
+the dedicated one) and `kvm2.go` (`http` client, budgets and tracing
+options behind functional options, so `New` is unchanged). The pipeline:
+marker lookup and bounded read on the pooled connection, then a GET bound to
+ctx plus the 20-minute cap under a `kvm2.download` span, `StorageVolCreateXML`
+and a streamed `StorageVolUpload` through sha256 on a fresh connection,
+`StoragePoolRefresh`, and the marker written last; every failure after the
+volume exists discards it with the cancellation stripped from ctx so the
+cleanup RPC still runs.
+
+Streaming is asserted by counting: a hook on the dedicated fake's upload
+snapshots how many bytes the 64 MiB test server had written when the upload
+began, and asserts under 16 MiB. The agent chose that over a process-wide
+`MemStats` delta because the tests run in parallel.
+
+Decisions the agent made, accepted: the base volume is created on the
+dedicated connection (nothing exists if the dial fails); `Accept-Encoding:
+identity` is sent so transparent gunzip cannot hide the Content-Length; a
+marker whose content is not 64 hex characters counts as debris;
+`Content-Length: 0` is its own Permanent error; the image lock is a
+one-slot channel so a waiter gives up with its ctx instead of sitting out a
+transfer; the fetch logger is derived after the span starts so its lines
+carry the span ID (the agent's first test run caught that).
+
+Commands (agent and orchestrator):
+
+```sh
+gofmt -l ./internal/provider/kvm2 ; go vet ./internal/provider/kvm2/ && go test -race -count=2 ./internal/provider/kvm2/
+go build ./... && go test -short ./...
+grep -n 'io.ReadAll\|bytes.Buffer' internal/provider/kvm2/image.go            # empty
+grep -n 'context.Background()' internal/provider/kvm2/*.go | grep -v _test.go  # empty
+```
+
+All `ok`. Attestation: "files read outside the allowed set: none".
+
+## Step 6 — The four contract methods
+
+2026-09-16 18:36. Fifth subagent step: `create.go`, `get.go`, `delete.go`,
+`list.go` and five test files (the package now has 150 top-level tests).
+Create follows spec entry 10 step for step under a per-name lock, with the
+adoption guard, the quota precheck, the image cache, an overlay that is
+reused and a seed that is always rebuilt, file-type disks from resolved
+paths, the metadata element in the define call, and a wait for running
+capped at two minutes. Get maps only `ErrNoDomain` to NotFound, prefers the
+guest agent over ARP, and reads UID and CreatedAt back every time. Delete is
+idempotent at every step and a no-op success for an unconfigured host.
+ListByTag fails whole per host and skips single unreadable domains.
+
+Decisions the agent made, accepted: the adoption path also waits for
+running, so a repeat Create on a shut-off domain blocks like a first one;
+`address` and `ListByTag` propagate ctx cancellation instead of reading it as
+"no address" or "skip", so a cancelled Get is Transient rather than a false
+Provisioning; NotFound from Get gets its own V(1) line because polling to
+NotFound is expected; the domain XML rides on the `domain defined` line at
+V(2). Flagged, not solved: adopting a paused domain fails Transient on every
+retry (`DomainCreate` refuses and the wait expires), which the spec does not
+cover; recorded as a known gap for the docs step.
+
+The agent found a one-word conflict between the brief and the spec
+(`DomainAffectCurrent` versus `DomainAffectConfig` for `DomainGetMetadata`);
+the code uses `DomainAffectCurrent`, and spec entry 8 was corrected to
+match in this commit.
+
+Commands (agent and orchestrator):
+
+```sh
+gofmt -l ./internal/provider/kvm2 ; go vet ./internal/provider/kvm2/ && go test -race -count=2 ./internal/provider/kvm2/
+go build ./... && go test -short ./...
+grep -n 'CloudInit' internal/provider/kvm2/create.go          # only len(req.CloudInit) and the seed builder call
+grep -n 'not implemented' internal/provider/kvm2/*.go          # empty
+grep -n 'provider.Wrap(' internal/provider/kvm2/*.go | grep -v _test.go | grep -v 'p.name'   # empty
+```
+
+All `ok`. Attestation: "files read outside the allowed set: none".
+
+## Step 7 — Checker and the CRD comment
+
+2026-09-16 18:52. Sixth subagent step: `check.go` and `check_test.go`
+(ten tests, the reasons table has 19 rows). Every host is checked
+concurrently on a fresh dedicated connection that is closed when the probe
+ends; the pooled table is never consulted, which the test proves with a
+counting dialer. Details fill in as facts arrive, so a `PoolNotFound`
+verdict still says which libvirt version answered. The `status.type` comment
+in `api/v1alpha1/providerhealth_types.go` gained `kvm2` and `make manifests`
+changed only the CRD's description line:
+
+```sh
+make manifests && git diff --stat config/ api/
+#  api/v1alpha1/providerhealth_types.go                    | 2 +-
+#  config/crd/bases/crawl.example.com_providerhealths.yaml | 2 +-
+```
+
+Decisions the agent made, accepted: a dial that runs out of time is
+`Unreachable` (the host did not answer), and `Timeout` is reserved for ctx or
+budget expiry on an established connection; `Close` runs synchronously in
+the probe so the close-count assertion is deterministic, at the cost of up
+to five seconds past ctx if a disconnect hangs on a live link.
+
+The agent also reported that the `captureContext` helper in `kvm2_test.go`
+appends to an unlocked slice, which the race detector flagged once `Check`
+logged from several goroutines. The helper was outside its allowed file set,
+so it built its own locked capture; the orchestrator fixed `captureContext`
+with a mutex in this commit and re-ran the package at `-race -count=3`.
+
+Attestation: "files read outside the allowed set: none".
+
+## Step 8 — Deployment files, Makefile, integration test
+
+2026-09-16 19:39. Seventh subagent step. `kvm2_integration_test.go`
+(`//go:build integration`, `KVM2_TEST_*` variables, the eleven ordered
+assertions of spec entry 21 plus `TestKVM2Check_realHost`),
+`config/samples/proxy_kvm2.yaml` (zone `beelink`, so the sample matches the
+shipped config), a commented `kvm2-home` entry in both providers configs
+with the porting note, comment-only edits on the shared `kvm-ssh` mount, the
+`test-integration-kvm2` Makefile target and the kvm2 rows and section in
+`docs/testing.md`. The integration test was not run; that is Step 11 with
+the user's go-ahead.
+
+The agent found that `New` builds the production dialer without tracing
+options, so in production the dialer and decorator use the global tracer
+provider (correct: `cmd/main.go` sets it up), while a test that wants every
+span family in one recorder sets the dialer's option itself. The test does
+so and asserts that `kvm2.DomainDefineXML` and `kvm2.dial` are children of
+`provider.create`.
+
+Decisions the agent made, accepted: the UID is generated from `crypto/rand`
+rather than promoting `google/uuid` to a direct dependency; the sentinel
+check looks for the full cloud-init string because the provider name also
+appears in attributes; `-run TestKVM2` matches both tests, as the kvm
+target's pattern does. Forward links to `docs/kvm2-provider.md` dangle until
+Step 9.
+
+Commands:
+
+```sh
+go vet ./... && go vet -tags=integration ./... && go vet -tags=e2e ./...
+bin/kustomize build config/default > /dev/null && bin/kustomize build config/samples > /dev/null
+make -n test-integration-kvm2
+# go test -race -tags=integration -count=1 -v -timeout 40m -run TestKVM2 ./internal/provider/kvm2/
+```
+
+All clean. Attestation: "files read outside the allowed set: none".
+
+## Step 9 — Documentation
+
+2026-09-16 19:54. Eighth subagent step: `docs/kvm2-provider.md` (sixteen
+sections modelled on the kvm guide), a `kvm2` column in the three tables of
+`docs/providers-existing.md` plus a kvm2 section and renumbering, the
+components row, §5 block, §10 span bullet, §11 row and eight Decisions
+bullets in `docs/architecture.md`, a README row and paragraph, and the
+guide in the `docs/providers.md` index. The agent verified each claim
+against the package rather than the spec and dropped a draft sentence
+claiming kvm and kvm2 share the base-image cache, because checking it would
+have meant reading v1's volume names.
+
+Checks re-run by the orchestrator: no em-dash in the guide, no line
+references, each of the 19 axis codes appears once in the appendix table
+with seven cells, every link and anchor across the seven touched documents
+resolves. The one added em-dash is the `Leader-elected` cell of the new
+components-table row, where the table's own convention wins.
+
+Attestation: "files read outside the allowed set: none".
+
+## Step 10 — Milestone review
+
+2026-09-16 21:19. The `go-operator-reviewer` agent reviewed the package
+against the contract and the spec (v1 code off limits to it as well) and
+returned seven findings, all accepted and applied by a fresh fix agent in
+one commit with regression tests:
+
+1. `waitRunning`'s two-minute cap only ever elapsed in the bad cases and
+   pinned a reconcile worker; now 15 s, a domain that dies right after
+   start is Transient at once, and a paused or pmsuspended domain is
+   Permanent before `DomainCreate` (an admin's doing).
+2. `ListByTag` read state and addresses that GC never uses, one agent RPC
+   per active domain per sweep; it now builds instances from the name and
+   the metadata only (`TestListByTag_doesNotReadStateOrAddresses`).
+3. A health round that ran out of time was blamed on the host
+   (`Unreachable`); `checkHost` now reports `Timeout` whenever the round's
+   ctx ended, whichever dial stage was running, and the promptness test
+   uses the production error shape.
+4. `Create` adopted an unmarked domain and never wrote the mark, so GC could
+   not see it; an unmarked domain under the name is now Permanent
+   (`TestCreate_adoptsOnlyMarkedDomains`).
+5. `hostTable.closeAll` had no production caller and `withBudgets` could not
+   work (the dialer captured the defaults first); both removed.
+6. The per-name and per-image lock maps grew without bound; replaced by a
+   reference-counted `keyedLocks` whose entries vanish when uncontended.
+7. `imageSizeIfUncached` falls back to GET on any non-2xx HEAD, and two
+   Delete no-op lines moved from V(0) to V(1).
+
+The spec (entries 2, 10, 12, 16, 18, 19, 24), the guide, the appendix and
+the architecture flow were updated in the same commit. The reviewer also
+listed what it checked and found correct: the error boundary, the state
+rules, Delete's idempotency and the replacement window, ListByTag's
+partial-failure policy, the abandon policy and double-close guards,
+resource handling on every path, and secret hygiene in the format strings.
+
+Lint: `bin/golangci-lint` still panics on this toolchain (Go 1.27 against
+a staticcheck built for 1.26), reproducible on untouched packages, so
+`make lint` could not gate this step; `go vet` with every build tag is
+clean. Tests: `go test -race -count=3 ./internal/provider/kvm2/` (168
+top-level tests) and `go test -short ./...` pass; `make test` with envtest
+runs next.
@@ -0,0 +1,531 @@
+# Execution: fix the kvm2 code-review findings
+Plan: [docs/plans/2026-09-17-1515-kvm2-review-fixes.md](../plans/2026-09-17-1515-kvm2-review-fixes.md)
+
+- [x] Step 0 — Save plan into the repo
+- [x] Step 1 — H1: a stalled write must not delay the teardown
+- [x] Step 2 — H2: a source error must not reach go-libvirt's stream sender
+- [x] Step 3 — H3: an unreadable address must not look like a missing one
+- [x] Step 4 — H4: the base image cache must not be shared across provider entries
+
+- [x] Step 5 — Validate H1-H4 in-cluster against beelink
+- [x] Step 6 — Route klog into the structured logger
+
+Remaining after those: the 48 mediums and lows in `review/REPORT.md`. Each gets its own
+checklist line here when it is planned, so this list stays a record of decided work rather
+than a wish.
+
+## Step 0 — plan committed
+
+Copied out of `~/.claude/plans/` and committed alone before touching code, as the repo's
+plan rules require.
+
+## Step 1 — H1: close the SSH client, and signal dead before killing
+
+Two edits, both small, in the order the bug unfolds.
+
+`sshSocket.Close` was `errors.Join(s.Conn.Close(), s.client.Close())`. Go evaluates those
+arguments left to right, so the blocking close ran before the unblocking one. `s.Conn` is
+the forwarded libvirt socket, and closing an SSH channel writes a close message under
+`ch.writeMu`, which go-libvirt already holds while streaming a base image through the same
+channel. Against a host that vanished without RST or FIN, that write blocks until the
+kernel stops retransmitting — `tcp_retries2`, about 15 minutes. `s.client.Close()`, a bare
+`net.Conn.Close()` on the TCP socket that takes no lock and unblocks the stalled writer
+through netpoll, was never reached. It now closes the client only.
+
+**Dropping the channel close rather than reordering it** was the judgment call. Reordering
+works, but afterwards the channel close returns an error — `net.ErrClosed`, a
+`*net.OpError`, the transport's sticky write error, or `io.EOF` once `mux.loop` has run
+`ch.close()` — and `closeErr` is memoised, so a filter that missed a case would poison
+every later `Close` and break `ssh_test.go:66,69`, which assert both closes return nil.
+Nothing wants the polite channel-close message: `dialSSHSocket` builds the client solely
+for this one channel, the `sshSocket` owns it exclusively, and `dialSSHSocket`'s own error
+paths already tear down by closing the client.
+
+`link.markDead` then compounded it by running `kill()` before `close(l.closed)`, and doing
+both inside a `sync.Once`. Reordering alone would not have been enough: `sync.Once.Do`
+blocks concurrent callers until the first body returns, so every other `markDead` — including
+the one in `call`'s own ctx-cancel branch — queued behind the stalled kill. **Cancelling the
+reconcile did not help.** The `once` field is gone; the mutex the struct already had now
+guards the "close it exactly once" invariant, `reason` is set before the signal so a
+released caller always reads a populated `deadReason()`, and `kill` runs outside the lock.
+
+**A second route to the same pin, found while planning and fixed by the same edits.**
+go-libvirt's `Disconnect` closes the `net.Conn` it was handed (`libvirt.go:331` →
+`socket.go:164`) — and that conn *is* our `*sshSocket`. So `libvirtBackend.Close`'s
+`Disconnect` goroutine entered `sshSocket.closeOnce` and hung there; the 5 s `closeBudget`
+released `Close`, which called `markDead("closed")` → `kill()` → `sshSocket.Close()` →
+blocked in the other goroutine's `sync.Once`. The budget did not protect it.
+
+The precondition was verified in the pinned x/crypto v0.54.0 rather than assumed: the
+flow-control park (`remoteWin.reserve`, `channel.go:268`) happens *outside* `writeMu`, and
+only `writePacket` holds it across the TCP write. So a writer parked in SSH flow control
+never blocks the close; a writer inside the TCP write always does. Since the channel window
+is 2 MiB (`channelWindowSize = 64 * channelMaxPacket`), reproducing it by hand needs a
+kernel send buffer that fills first — shrink `SO_SNDBUF` well under 2 MiB.
+
+The older kvm provider does not have this bug: it closes `client` only
+(`internal/provider/kvm/conn.go:429,456`) and never touches the forwarded channel. kvm2's
+rewrite introduced it.
+
+### Tests: planned four, shipped one
+
+The first draft had three hermetic tests plus a `//go:build integration && linux`
+stalled-write test with a TCP proxy and shrunken socket buffers. The user pushed back —
+once the bug is fixed the connection closes properly, so what is left to test? — and they
+were right about most of it. `TestDial_forwardsUnixSocket` already covers "Close tears the
+connection down", and after the fix `Close` is a single call with no ordering to get wrong.
+The stalled-write test was the worst of the four: ~150 lines of Linux-only,
+buffer-autotuning-dependent scaffolding whose own false-green guard made it *skip* whenever
+the kernel did not cooperate, to demonstrate a bug that reading x/crypto settles and that
+the fix deletes the code path for.
+
+What survived is `TestMarkDead_slowKillDoesNotBlockCallers`, because the `markDead` change
+*adds* a guarantee that did not exist before and nothing else touches it —
+`TestCall_killedByAnotherCallerFailsTransient` passes a nil kill, so it cannot see the
+property. It is also the plausible regression: restoring the `sync.Once` looks like a
+correctness improvement. The call's own `fn` signals when it is in flight, so the parked
+caller is genuinely parked rather than refused by the dead check on the way in.
+
+The step that actually validates a regression test is running it against the unfixed code.
+`markDead` was reverted in place, the test run, and the fixed version restored from a copy:
+
+```bash
+cp internal/provider/kvm2/backend.go "$SCRATCH/backend.go.fixed"
+# ... revert markDead to the sync.Once form and re-add the once field ...
+go test -race -count=1 -run 'TestMarkDead_slowKillDoesNotBlockCallers' ./internal/provider/kvm2/
+#   --- FAIL: TestMarkDead_slowKillDoesNotBlockCallers (1.00s)
+#       backend_test.go:186: link not signalled dead while its kill is still running
+cp "$SCRATCH/backend.go.fixed" internal/provider/kvm2/backend.go
+```
+
+It fails in a second with the right diagnostic rather than hanging the suite, because the
+blocked kill is released by `t.Cleanup`.
+
+```bash
+go build ./... && go vet ./... && go test -race -count=1 ./internal/provider/kvm2/
+```
+
+### Worth noting
+
+**An accepted risk, deliberately not covered by a test.** Nothing now stops someone
+re-adding `s.Conn.Close()` to `sshSocket.Close` as tidiness; it would look harmless and
+every test would still pass, because in a healthy test the channel close returns instantly.
+The guard is the doc comment on `sshSocket`, which says outright that closing the channel
+would queue behind a stalled transfer. That is in idiom for this package — `hosts.go:25-34`
+and `create.go:151-153` carry the same kind of hard-won decision — but it is a comment, not
+a test, and the failure it prevents is a 15-minute operator-wide stall.
+
+**Left for later, on purpose.** `ReconciliationTimeout` in `SetupWithManager` does not exist
+in this repo at all; adding a per-reconcile deadline would change the contract for every
+provider, since `hosts.go` and `image.go` both cite "no caller deadline", so it needs its own
+change and its own spec entry. `sshSocket.keepalive` still has no test (sweep finding S1.2).
+
+**Docs.** `docs/kvm2-provider.md` said the adapter "kills the connection (socket and SSH
+client)", which is no longer true and was the description that made the bug look correct.
+It now states the order (mark dead, then kill) and why both steps are lock-free.
+`docs/kvm2-provider-spec.md` entries 18 and 19 were checked and are not falsified.
+
+**No CHANGELOG entry yet** — the repo's rule is to add one once the user confirms the change
+works, so it lands with a later "validate and record" step, as in the kvm campaign.
+
+## Step 2 — H2: a source error must not reach go-libvirt's stream sender
+
+This one is a workaround, not a repair. The defect is upstream, and the entry says so in
+three places (code comment, `docs/kvm2-provider.md` *Known gaps*, commit message) so nobody
+later reads the wrapper as ours to delete.
+
+`requestStream` in the pinned go-libvirt starts its stream sender on two **unbuffered**
+channels and then, on the error path, returns without ever reading `outErr`. A non-EOF error
+from the reader makes `SendStream` emit a StatusError packet and return, whereupon it parks on
+`outErr <-` forever; libvirtd answers the StatusError with an error, so the main goroutine
+parks on `abort <- true` forever. `StorageVolUpload` never returns. The 20-minute budget
+rescues only the caller, and the socket kill does nothing at all here, because both ends are
+blocked on channel sends rather than on I/O — which is what the comment at `backend.go:152`
+got wrong. Every failed upload leaked two goroutines and 4 MiB against a 128Mi pod limit,
+with the reconciler retrying on backoff.
+
+Three changes.
+
+`eofReader` sits at the go-libvirt seam in `backend.go`, not in `image.go`, so it also covers
+the seed and marker uploads, and so it stays out of the file Step 4 rewrites. It converts any
+source error to `io.EOF` — the one outcome that ends a stream the normal way — and keeps the
+real error for `StorageVolUpload` to return afterwards. `countingReader` is untouched
+underneath it and still sees the true error, so `uploadImage`'s existing length and digest
+checks produce exactly the messages they did before.
+
+The download now carries its own deadline, one minute inside the transfer budget.
+`fetchImage` already bounded the whole fetch at 20 minutes, so the body's death and the
+budget's socket kill were racing, and the socket usually won. With the body a minute ahead, a
+stalled mirror ends through the reader and unwinds cleanly. This is the difference between a
+slow mirror costing a Transient error and costing a permanent leak, and slow mirrors are the
+realistic repeat offender.
+
+`discardVolume` moved from `uploadImage` up into `fetchImage`'s upload-error branch, so the
+dedicated connection is closed — and libvirtd has let go of the stream's fd — before the file
+is unlinked. Moving it rather than adding a second `Close` is what keeps
+`image_test.go`'s `closeCount() == 1` assertion true. `uploadImage` lost its `be` parameter
+and its private logger in the process.
+
+**What is left, deliberately.** A failure that originates in libvirtd mid-upload (ENOSPC, or
+the volume deleted underneath) still leaks, because by then `SendStream` has already returned
+and nothing we pass, cancel or close can reach it. Confirmed unfixable from this side. The
+upstream fix is two characters and is quoted in *Known gaps* along with the `replace` escape
+hatch, should it ever bite hard enough to be worth the maintenance.
+
+**Unverified, ~85%.** That libvirtd answers a short-but-clean stream finish with OK rather
+than an error. `virFDStreamWrite` enforces `length` only on writes past it and nothing
+compares `offset` to `length` on the close path, which is also why `virsh vol-upload` of a
+small file into a large volume works. If it answers with an error we land back in a deadlock
+by a different door — but that is precisely today's behaviour, so the change cannot make
+things worse. To settle it, on any box with a libvirtd (not beelink, which is read-only
+here), against the pinned go-libvirt:
+
+```sh
+virsh -c qemu:///session pool-define-as probe dir --target ~/vmprobe
+virsh -c qemu:///session pool-start probe
+virsh -c qemu:///session vol-create-as probe short.raw 10M --format raw
+# then a ~40-line main.go: StorageVolUpload(vol, r, 0, 10<<20, 0) where r
+# yields 1 MiB then io.EOF; print the error, then runtime.NumGoroutine()
+# and a goroutine profile after a two-second settle.
+```
+
+Expected: `err == nil` and the goroutine count back at baseline. The same program with a
+reader that returns a non-EOF error should wedge against the *unpatched* seam, which also
+confirms the diagnosis end to end.
+
+The negative check that matters was run on the unit test instead — `eofReader.Read` was
+changed back to returning the real error, and both new tests failed with
+`io.Copy through eofReader = connection reset by peer, want nil`, then passed again once
+restored.
+
+```bash
+go build ./... && go vet ./... && go test -race -count=1 ./internal/provider/kvm2/
+```
+
+### Worth noting
+
+The `fetchImage` discard is still unfenced against a second provider entry that shares the
+volume name — Step 4 (H4) is what closes that, by making the name per-entry. The deferred
+atomic-publish work then lands exactly where this step put the discard.
+
+**Deliberately not tested:** the abort protocol end to end. `libvirttest` reads a fixed
+28-byte header per packet and implements no stream procedures, so pinning this hermetically
+would mean a 150-250 line libvirt RPC server stub whose only assertion is about *upstream's*
+goroutines. Same call as H1's: the test would cost more than it protects. The two tests that
+did land pin the one property the seam actually needs.
+
+## Step 3 — H3: an unreadable address must not look like a missing one
+
+`address()` tried the guest agent then the host's ARP table and skipped any error that was
+not a context error, returning `("", nil)` after both. An active domain with no IP maps to
+`StateProvisioning`, the controller clears `status.IP` on that, and the health engine prunes
+the proxy's counters and calls `ForgetProxy` in one tick once `EffectiveHost()` is empty.
+
+The damaging trigger was not the guest. The libvirt connection is pooled per host, so when
+another reconcile's cancel or budget overrun killed it, `call` returned
+`"connection is closed"` — carrying no context sentinel, so the check did not recognise it —
+and `Get` returned a *successful* instance in Provisioning with an empty IP. An unrelated
+proxy's timeout dropped a healthy proxy out of rotation.
+
+The fix rests on a discriminator the spec had already recorded without drawing the
+conclusion: during a normal boot on beelink exactly one source fails. The agent errors
+because it is not up yet, while ARP answers *without* error and with nothing, since the host
+never sees the guest's traffic. So one source failing is "no address yet"; every source
+failing, or a connection already dead, is `ErrTransient`.
+
+The explicit `provider.ErrTransient` wrap on the joined error is load-bearing rather than
+decoration. Without it `classify` picks the first libvirt code out of the join tree, and
+`ErrInvalidArg`, `ErrAccessDenied` and `ErrNoSupport` are Permanent — an address read failing
+would have latched a *running* proxy as Failed, a worse version of the harm being fixed.
+
+`isDead(be)` is checked first in the error branch. It is not strictly needed, since a dead
+link fails both sources and the count rule catches it anyway, but it fails on the first
+source instead of the second, attributes the error better, and is the same predicate M15 will
+want for `ListByTag`.
+
+### Tests
+
+`TestGet_addressErrorIsProvisioning` failed against the new rule, which is correct: it fails
+*both* sources with `ErrAgentUnresponsive`, and code 86 is agent-specific — the ARP source
+cannot produce it. Re-pointed as `TestGet_agentFailsAndARPEmptyIsProvisioning`, the beelink
+boot it was really reaching for: agent errors, ARP returns no addresses and no error, result
+is Provisioning. Nothing covered that before, and it is the case the new rule must not break.
+
+`TestGet_everyAddressSourceFailingIsTransient` covers the other side, two subtests: both
+sources erroring, and the pooled connection dying under the read. Against the unfixed code
+both return a Provisioning instance instead of an error, which is exactly the bug:
+
+```bash
+# with isDead() and the every-source-failed arm removed from address():
+go test -count=1 -run TestGet_everyAddressSourceFailingIsTransient ./internal/provider/kvm2/
+#   get_test.go:215: Get = &{... IP: State:Provisioning ...}, want no instance
+```
+
+`TestGet_arpFallbackWhenAgentFails` passes untouched — one source of two.
+
+### Worth noting
+
+**Half of H3 is deliberately still open, and it is the half that fires more often.** A guest
+agent restarting inside a running VM still demotes it to Provisioning and clears its IP: the
+agent errors, ARP answers empty, and that is indistinguishable from a VM still booting. The
+discriminator is the *previous* IP, which the provider does not hold by design
+(`TestGet_readsUIDAndCreatedAtFromBackend` pins "read back, not remembered") and the
+controller already has. Follow-up finding, to be filed against
+`internal/controller/proxy_controller.go:208-211`: on `StateProvisioning`, keep an existing
+IP and `Provisioned` condition for a few polls rather than clearing on the first, or give the
+health engine a grace period before pruning.
+
+The one kvm2-local alternative was considered and rejected: the mark carries `createdAt`, so
+`get` could treat "active, no address from any source, domain older than a provisioning
+window" as an error. It uses state the provider genuinely has, but `createdAt` is creation
+rather than boot, it fires on every genuinely stuck VM as a retry-forever error, and it
+invents a provisioning deadline that belongs to the controller. A design change wearing a bug
+fix's clothes.
+
+**L6 not folded in.** A domain that vanishes between `DomainGetState` and the address read
+now returns Transient rather than a fake Provisioning — better, and self-correcting on the
+next poll. Returning NotFound one poll earlier is a two-line addition at the same spot, left
+for when L6 comes up so this commit stays one idea.
+
+## Step 4 — H4: the base image cache must not be shared across provider entries
+
+`baseVolumeName` was `base-<key>.<format>` — byte-identical to what
+`internal/provider/kvm/image.go` builds in the same pool, for URL images too (kvm takes
+`hex(sum[:8])`, kvm2 takes `hex(sum[:])[:16]`, the same eight bytes), with a completely
+separate lock table. The shipped config has a live `kvm-home` entry and a commented-out
+`kvm2-home` entry on the same host, the same key `debian-13` and the same default pool, and
+the comment invites the operator to enable it. Both integration tests use key `test-image`.
+On beelink both live overlays back onto kvm's `base-debian-13.qcow2`.
+
+Since `cachedImage` accepts any 64-hex marker when no `sha256` is configured and never
+re-hashes the base, that meant: two entries pointing one key at different URLs silently share
+a file; a concurrent cold fetch lets one entry delete the other's half-uploaded base and boot
+a guest from the remains; and a `sha256` edit deletes the base in place under live overlays,
+which libvirt permits without complaint.
+
+The name now carries the provider entry: `base-<entry>-<key>.<format>`. `imageSource` gained
+an `Entry` field and `resolveImage` takes the provider name, so the four production call
+sites all follow from `imageSource.volumeName()`. That closes the cross-entry sharing
+outright, kvm↔kvm2 and kvm2↔kvm2 alike, and it turns the existing per-`Provider` `imageLocks`
+from a latent bug into a correct one — with names unique per entry, no two entries can
+contend for a file, so a process-wide lock table is not needed.
+
+**The entry name reaches a file name, so it now has to be a usable one.** Provider names are
+validated for all types at once and only for being non-empty and unique
+(`internal/provider/config.go`), so `New` checks its own against the existing `kvmIdentifier`
+rule, exported as `provider.IsKVMIdentifier` rather than duplicated. A name with a slash
+fails at startup instead of producing a volume nothing can address.
+
+**One companion fix, because the naming alone does not stop it.** `cachedImage`'s
+digest-mismatch branch deleted the marker and the base and refetched. It now returns
+`ErrPermanent` naming both digests and telling the operator what to do. That branch was the
+remaining path that unlinks a good file under running domains, and the contract already said
+"add a new key instead" — the code was contradicting it. The no-marker and non-hex-marker
+branches keep deleting, since there the base is genuine debris.
+
+### What this does NOT fix
+
+Stated plainly rather than left to look closed:
+
+- **Editing `images.<key>.url` is still a silent no-op.** The cache is keyed by entry and
+  key, and nothing records the URL, so the new one is never fetched. Pre-existing, already
+  documented as "add a new key"; the eviction section now says it outright.
+- **Cross-process fencing.** Two managers, or a manager plus a hand-run `virsh`, can still
+  meet on one name. E4 refuted the rollout and leader-handover triggers for the shipped
+  manifests, so this is narrow, and its fix is the deferred atomic publish — upload to a temp
+  name, verify, publish — which lands where H2 put the discard.
+
+### Migration
+
+Nothing to do. The `base-debian-13.qcow2` on beelink is kvm's, kvm keeps using it untouched,
+and kvm2 downloads its own copy once per host when `kvm2-home` is enabled. No window in which
+anything is replaced under a running guest. The pool needs room for both, which the
+prerequisites now say.
+
+### Tests
+
+Mostly literals becoming helpers, plus a new `fx.baseName()` in `provider_test.go` for the
+three fixtures that asserted `"base-debian-13.qcow2"` by hand. The log and span assertions in
+`TestEnsureImage_downloadsAndWritesMarkerLast` keep asserting a full name rather than loosening
+to a prefix, so they still pin that the key stays readable in operator-facing output.
+
+One test changed behaviour rather than spelling: `TestEnsureImage_sha256ChangeRedownloads`
+became `TestEnsureImage_sha256ChangeRefusesInsteadOfReplacing`, asserting Permanent, zero
+server hits, zero deletes, and both volumes untouched. Two added:
+`TestNew_rejectsUnsafeEntryName`, and a `TestVolumeNames` row for a second entry with the same
+key.
+
+Both negative checks were run:
+
+```bash
+# baseVolumeName reverted to base-<key>.<format>:
+go test -count=1 -run TestVolumeNames ./internal/provider/kvm2/
+#   volume_test.go:135: got "base-debian-13.qcow2", want "base-kvm2-lab-debian-13.qcow2"
+
+# the digest branch reverted to delete-and-refetch:
+go test -count=1 -run TestEnsureImage_sha256ChangeRefuses ./internal/provider/kvm2/
+#   image_test.go:412: ensureImage() error = <nil>, want ErrPermanent telling the operator what to do
+```
+
+`go vet -tags=integration` was run too, since `kvm2_integration_test.go:484` builds the base
+name and only compiles under that tag.
+
+### Worth noting
+
+**The integration test has still never run on beelink.** Step 11 is unchecked in
+`docs/plans-executions/2026-09-16-1333-kvm2-provider.md`, every `base-*` on the host predates
+kvm2, and both live domains carry the kvm v1 metadata namespace. After this change it is also
+the first end-to-end exercise of the new name, and it will do a real cold download under
+`base-kvm2-integration-test-image.qcow2` beside kvm's untouched `base-test-image.qcow2`.
+
+**Docs touched beyond the obvious:** the eviction recipe named `base-debian-13.qcow2`, which
+after this change is *kvm's* file — following it would have deleted the wrong provider's base.
+It now greps for the prefix and deletes the marker first.
+
+## Step 6 — Route klog into the structured logger
+
+Not a review finding. Found while watching the Step 5 validation: some lines ignored the zap
+encoder no matter what, so `kubectl logs | jq` could never work.
+
+```
+I0917 19:50:02.580097       1 leaderelection.go:258] "Attempting to acquire leader lease..." lock="..."
+```
+
+That is klog's own stderr writer, not ours. client-go's leader elector calls
+`klog.FromContext(ctx)` (`leaderelection.go:257`), and controller-runtime hands the elector a
+context with no logr in it — `mgr.Start` receives `ctrl.SetupSignalHandler()`, which carries no
+logger (`pkg/manager/internal.go:461-473`). `FromContext` falls back to `Background()`, and with
+no logger registered that is klog's text sink. Nothing in this repo imported or configured klog;
+it was an indirect dependency that around 60 linked packages use, including
+`client-go/tools/{leaderelection,cache,record}` and `client-go/transport`. controller-runtime
+does not route it either — its only `klog.SetLogger` call is in its own test suite.
+
+**`klog.SetLogger(logger)`, deliberately not `SetLoggerWithOptions(logger, ContextualLogger(true))`.**
+The contextual form is what klog's docs recommend, because it skips a klogr round trip. It also
+disables klog's verbosity gate ("Such a logger cannot rely on verbosity checking in klog",
+`contextual.go:83-86`), handing the decision to zap — whose level under `Development: true` is
+Debug, which enables V(1). client-go's V(1) chatter, suppressed today by klog's own `-v=0`,
+would have started appearing. Plain `SetLogger` keeps klog's check in front and still redirects
+what it does emit ("all log lines will be suppressed from the regular output, and redirected to
+the logr implementation", `contextual.go:47-48`). Same fix to the format, no new volume — which
+matters, because the complaint that started this was volume.
+
+`k8s.io/klog/v2` moves from indirect to direct in `go.mod`. It was already in the module graph;
+nothing new is downloaded.
+
+**The other bypass.** The discovery API's `http.Server` set no `ErrorLog`, so `net/http`'s own
+faults — a failed `Accept`, a panic escaping a handler — went to the stdlib logger on stderr.
+It now drains into logr through `logr.ToSlogHandler`, the pattern
+`internal/provider/gcp/wirelog.go:31-36` already uses. This server is plain HTTP with no TLS, so
+`ErrorLog` carries real faults rather than handshake noise, and Error level is right.
+
+**Deployment.** `--zap-encoder=json` and `--zap-time-encoding=rfc3339` on the manager.
+Verified the flag beats the in-code default rather than assuming it: `addDefaults` fills
+`NewEncoder` only `if o.NewEncoder == nil` (`zap.go:176-191`) and `BindFlags` sets it from the
+flag. The time encoding is not optional — the JSON encoder defaults to epoch seconds.
+Verbosity was left at debug on purpose; `--zap-log-level=info` drops the per-RPC lines.
+
+`opts := zap.Options{Development: true}` stays. Flipping it would change the format for every
+deployment of this operator, and the flags get the same result here without that reach.
+
+### Tests
+
+None. Three changes, all wiring: two struct fields and a manifest key. `cmd/main.go` has no test
+coverage, and pinning `klog.SetLogger` means mutating process-global state that leaks into every
+other test in the package. Verified by observation instead, which for this change is stronger
+than a unit test would be.
+
+### Verification
+
+```bash
+kubectl -n egress-proxies-operator-system logs deploy/... | grep -i leader
+# {"level":"info","ts":"2026-09-17T20:10:00Z","msg":"Attempting to acquire leader lease...",
+#  "lock":"egress-proxies-operator-system/b47711d1.example.com"}
+
+kubectl -n egress-proxies-operator-system logs deploy/... | jq -e -s 'length > 0'
+# YES - every line is valid JSON
+```
+
+The `jq -s` check is the one that counts: it fails if *any* line is not JSON, which is exactly
+the property being fixed. All four existing proxies stayed Ready across the rollout and all
+three ProviderHealth entries stayed Ready.
+
+### Worth noting
+
+**The JSON encoder renders durations as float seconds**, so `"duration":"566.12µs"` became
+`"duration":0.000336664`. Machine-friendlier, worse to read. Either format in the query —
+`(.duration*1000000|round)` for µs — or pass an encoder config option in code, which
+controller-runtime exposes (`zap.Options.EncoderConfigOptions`, `zap.go:140-144`):
+
+```go
+opts := zap.Options{
+	Development: true,
+	EncoderConfigOptions: []zap.EncoderConfigOption{
+		func(c *zapcore.EncoderConfig) { c.EncodeDuration = zapcore.StringDurationEncoder },
+	},
+}
+```
+
+Left out of this step as scope the plan did not cover.
+
+## Step 5 — Validation against beelink
+
+First contact between the kvm2 provider and real libvirt. Everything before this ran against
+an in-memory fake of the twenty libvirt calls the provider uses.
+
+Run in the existing `kind-proxy-operator-demo` cluster rather than a fresh one, deliberately.
+That cluster already owned two live `kvm-home` VMs on beelink, and orphan GC deletes any
+instance whose UID is missing from *its own* cluster's live set after `gc-min-age`. A second
+operator carrying the `kvm-home` entry would have destroyed both within ten minutes. Using the
+existing cluster made its live set the thing that protected them — the isolation argument
+pointed the opposite way from instinct.
+
+Baseline recorded first (4 Ready proxies, 3 ProviderHealth entries, beelink's domains, volumes
+and pool space), so "did we break something" stayed answerable.
+
+The image tag had to change from the running `kvm-dev`. `make deploy` sets the image through
+kustomize, so an identical tag leaves the Deployment spec untouched, no rollout happens, the
+new process never starts, and the providers config — which is read once per process — is never
+re-read. The ConfigMap carries no generator hash either, so nothing else would have forced it.
+
+Results, in the order they were checked:
+
+| Check | Outcome |
+|---|---|
+| `kvm2-home` ProviderHealth | `Ready 1/1`, `Reachable`, 189 ms — hostname, libvirt 11.6.0, pool running, 247.8 GiB free |
+| First Create, cold cache | `base image cached`, 433,651,712 bytes in 8.3 s, digest verified |
+| Volume naming (H4) | `base-kvm2-home-debian-13.qcow2` created **beside** kvm's untouched `base-debian-13.qcow2` |
+| Second Proxy | no `fetching` line at all — cache hit under the new name |
+| State reporting (H3) | `Running` held steady across every poll, no flapping to Provisioning |
+| Squid | `Healthy=True` on a real CONNECT probe |
+| Delete | domain and both per-VM volumes gone; base image survived |
+| Existing infrastructure | all four pre-existing proxies stayed Ready across two rollouts |
+
+The health check alone is worth calling out as a gate: it dials SSH, verifies the host key,
+completes the libvirt handshake, runs four RPCs and tears down a dedicated connection — the
+exact path H1 changed — without creating anything. Putting the stop gate there meant the
+riskiest unknown was settled before a single byte was written to the hypervisor.
+
+### What this run did NOT prove
+
+**H2's error paths.** The unverified claim — that libvirtd answers a short stream finish with
+OK rather than an error — only arises when a body *fails* part way through, and a healthy
+433 MB download never finishes short. The 19-minute HTTP deadline never fired either. What was
+proved is narrower: the wrapper does not break normal uploads. Settling the rest still needs
+the scratch-pool probe in the Step 2 entry, and it would be easy to mistake this green run for
+more than it is.
+
+**The `kvm2` integration test still has not run.** `make test-integration-kvm2` remains
+unexecuted; this was a cluster-level validation, not that suite.
+
+### Worth noting
+
+`config/manager/providers_config.yaml` now ships with the `kvm2-home` entry **enabled**, since
+the operator is running it for real. `config/manager/kustomization.yaml` was left at its
+committed tag: `make deploy` rewrites it to whatever local image was last deployed, which is
+churn rather than work.
+
+Two host-side leftovers were *not* touched, both pre-existing and unrelated to kvm2: five
+orphan `proxy-*` volumes from earlier kvm runs, and the `base-test-image.qcow2` pair from the
+kvm integration test. Every write to beelink in this step came from the operator itself; no
+`virsh` write was issued by hand.
@@ -0,0 +1,301 @@
+# Plan: kvm2, a clean-room libvirt/KVM provider built from the provider contract
+
+**Created:** 2026-09-16 13:33
+
+## Context
+
+The provider documentation was just split into a generic contract
+(`docs/provider-contract.md`), a spec template (`docs/provider-spec-template.md`)
+and an appendix on the existing providers (`docs/providers-existing.md`), so that
+a new provider can be specified and built from those files alone. Before using
+them for Proxmox, the user wants a dry run on a known backend: rebuild the
+libvirt/KVM provider from scratch as a second type, `kvm2`, following the
+documents, without touching or reading the existing `internal/provider/kvm/`
+code. The result is a validated workflow (spec first, then implementation by
+fresh agents against the contract) and a v2 provider that can replace v1 once it
+passes on the real hypervisor.
+
+User decisions:
+
+- **Placement:** new package `internal/provider/kvm2`, type `kvm2`, config block
+  `kvm2:`, side by side with `kvm`. The Go config struct reuses
+  `provider.KVMConfig` (`internal/provider/config_kvm.go` is shared code, not
+  the kvm package), so an entry ports by changing `type` and the block name.
+- **Clean room:** implementers never open, grep or import
+  `internal/provider/kvm/`. Docs (`docs/kvm-provider.md`) and the operational
+  lessons from the first build's execution log are allowed.
+- **Implementer:** the orchestrator (this session) writes the spec and the step
+  briefs; each implementation step runs in a fresh general-purpose subagent
+  given only the three docs, the filled spec, the backend facts and its brief.
+  The orchestrator reviews each result against the contract, runs the tests and
+  commits.
+- **Profiles:** `create-return` blocking, `provisioning-state` create-blocks,
+  `child-spans` span-decorator, `quota` detected (pool free space precheck),
+  `credentials` mounted-files (the existing `kvm-ssh` Secret), `client-kind`
+  rpc-over-ssh, `gc-mark` written-metadata, `created-at` written-at-create,
+  `ip-source` discovered (agent, then ARP), `list-partial` fail-whole per host
+  with unreadable domains skipped, `not-found` error-code, `delete-return`
+  waits, `secondary-resources` explicit, `timeouts` per-call, `logging` tiered,
+  `backend-test` integration-tag.
+
+## Backend facts the spec carries (allowed lessons)
+
+- Real host `beelink`: libvirt 11.6, QEMU 10.1, pool `default` at
+  `/srv/vms/images`, bridge `br0`, `beelink.tail8c69c.ts.net:22`, user `novakj`,
+  key `~/.ssh/proxy-operator_ed25519`, known_hosts `~/.ssh/kvm_known_hosts`.
+  The Debian trixie qcow2 is about 435 MB.
+- AppArmor confines QEMU from the domain XML: only `<disk type="file">` with a
+  `<source file>` path gets a rule, so volume paths are resolved with
+  `StorageVolGetPath` and emitted as file disks.
+- Debian 13 hangs in GRUB under BIOS with a serial console and no video: UEFI
+  by default (`<os firmware="efi">`, secure boot feature disabled), `firmware:
+  bios` per image; OVMF is a host prerequisite; undefine needs the NVRAM flag.
+- libvirt hands the metadata element back without its namespace declaration:
+  parse by local name.
+- Vendor-data must be a shell script, not cloud-config, to survive cloud-init's
+  merge with user-data; it installs `qemu-guest-agent`. The agent answered after
+  about 20 s where ARP never did.
+- Read-only rootfs: `iso9660.NewWriter` needs a temp dir under `/tmp` (64Mi).
+- go-libvirt (`github.com/digitalocean/go-libvirt`, in go.mod): every method
+  blocks and takes no ctx; `libvirt.Error{Code uint32, Message}` has a value
+  receiver; needed `ErrorNumber` constants exist (`ErrNoDomain`,
+  `ErrNoStorageVol`, `ErrNoStoragePool`, `ErrOperationInvalid`,
+  `ErrNoDomainMetadata`, `ErrStorageVolExist`, `ErrXMLError`, `ErrNoSupport`,
+  `ErrInvalidArg`, `ErrAuthFailed`, `ErrAccessDenied`). `libvirt.New(net.Conn)`
+  then `Connect()`. No new dependencies: `x/crypto/ssh`, `kdomanski/iso9660`
+  are already direct requires; use `slices`, never `sort` (depguard).
+
+## Design decisions recorded in the spec (step 1)
+
+- **Package layout** (`internal/provider/kvm2/`): `kvm2.go` (type, `New`,
+  logger, per-name keyed lock), `id.go`, `state.go`, `errors.go`, `xml.go`,
+  `seed.go`, `backend.go` (ctx-first seam over the 20 go-libvirt calls, adapter
+  with a per-RPC deadline), `ssh.go`, `hosts.go` (one cached connection per
+  host, lazy dial, dialer seam), `observe.go` (the single decorator: one span
+  `kvm2.<Call>` per RPC plus the V(1)/V(2) log line), `image.go`, `create.go`,
+  `get.go`, `delete.go`, `list.go`, `check.go`, plus `fake_test.go`,
+  `sshserver_test.go`, `kvm2_integration_test.go`. Names are suggestions for
+  the briefs; the implementer may choose others by responsibility.
+- **Seam:** every seam method takes `ctx`; the adapter runs the blocking call in
+  a goroutine and selects on result, `ctx.Done()` and a timer (dial 10s, RPC
+  30s, `DomainCreate` 90s, upload/download ctx plus a 20m cap). On abandon it
+  closes the connection and marks it dead; the next call redials. The image
+  download runs on a dedicated connection so a cancelled poll cannot cut it off.
+- **Decorator wrapped once:** applied inside the real dialer where a connection
+  is born, so provider code never sees an undecorated backend. Spans `kvm2.dial`
+  (in the dialer) and `kvm2.download` (in `image.go`) are the two non-RPC spans.
+  A reflection test asserts every seam method is instrumented.
+- **Metadata:** own namespace
+  `https://gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/kvm2`,
+  element `<instance>` with `managed`, `uid`, `createdAt`. kvm2 never lists or
+  adopts v1 domains; a `kvm` and a `kvm2` entry may share a host.
+- **Adoption guard:** an existing domain whose metadata `uid` is non-empty and
+  differs from `req.UID` makes `Create` return `ErrPermanent`.
+- **Quota precheck:** before creating volumes, `StoragePoolGetInfo`; needed =
+  image `Content-Length` (if not cached) + `diskSizeGb` + 16 MiB; below that
+  returns `ErrQuotaExceeded` naming pool, available and needed.
+- **Download** requires `Content-Length` (Permanent otherwise), streams through
+  `sha256` into `StorageVolUpload`, refreshes the pool, writes the marker
+  volume last; a failed attempt deletes the partial volume.
+- **Classify:** `ErrNoDomain`, `ErrNoStorageVol` NotFound, but Transient inside
+  `Create`; `ErrNoStoragePool`, `ErrXMLError`, `ErrNoSupport`, `ErrInvalidArg`,
+  `ErrAuthFailed`, `ErrAccessDenied` Permanent; SSH authentication failures and
+  host-key mismatches or unknown host keys Permanent; missing key or
+  known_hosts file Transient (Secret pending), unparsable key Permanent;
+  cancellation, deadlines, `net.OpError`, `io.EOF` and anything else Transient.
+- **Same-name concurrency:** a keyed mutex per instance name serialises Create
+  and Delete for one name.
+- **State:** nostate Provisioning; running or blocked without IP Provisioning,
+  with IP Running; everything else, including unknown codes, Stopped. IP from
+  the guest agent first, then ARP, first non-loopback non-link-local IPv4.
+- **Checker:** per host concurrently on a fresh dedicated connection (dial,
+  `ConnectGetLibVersion`, `ConnectGetHostname`, pool lookup and info, close), so
+  it never touches the pooled connection. Reasons: shared ones plus
+  `PoolNotFound`, `PoolNotActive`, `CredentialsUnavailable`. Details `user`,
+  `hostname`, `libvirtVersion`, `pool`, `poolState`, `poolCapacityGiB`,
+  `poolAvailableGiB`.
+- **Rotation:** SSH material is read on every dial, so Secret rotation needs no
+  restart.
+- **Config validation messages:** `validateKVM` is refactored into
+  `validateKVMBlock(prefix, block string, c *KVMConfig)` so messages say
+  `kvm2.` for the new block; `validateKVM` stays as a wrapper for `kvm`.
+
+## Steps (one commit each, on `feat/kvm2-provider` off `main`)
+
+PR #7 is merged, so `main` carries the three docs. Step 0 begins with
+`git fetch origin && git switch main && git pull --ff-only`.
+
+Every subagent brief opens and closes with: "Do not open, grep, list or import
+anything under `internal/provider/kvm/`. If a detail seems to need that code,
+stop and report the question instead." Each brief names the files it may create
+or edit and nothing else, lists the inputs (the three docs, the spec,
+`docs/kvm-provider.md` for backend facts, CLAUDE.md conventions), the
+acceptance tests by name and the commands to run, and asks the agent to end
+with "files read outside the allowed set: none".
+
+- **Step 0 (orchestrator):** copy this plan to `docs/plans/<ts>-kvm2-provider.md`
+  with the real timestamp, commit alone, create the branch. Execution summary
+  `docs/plans-executions/<ts>-kvm2-provider.md` with the checklist after step 1.
+- **Step 1 (orchestrator): the spec.** Copy `docs/provider-spec-template.md` to
+  `docs/kvm2-provider-spec.md` and answer all 24 entries with the profiles and
+  decisions above; every Evidence line cites the user guide, a go-libvirt symbol,
+  or "verified on beelink in step 11". Entry 21's checklist lists the concrete
+  test names from steps 4 to 7. Verify: no blank Answer lines, every axis code
+  used matches contract §4.20, no reference to `internal/provider/kvm/`.
+- **Step 2 (subagent): config wiring, IDs, state, classify.** Edits
+  `internal/provider/config.go` (`KVM2 *KVMConfig` field, `setBlocks`,
+  allow-list, dispatch), `config_kvm.go` (`validateKVMBlock`),
+  `config_test.go` and `config_kvm_test.go` rows. Creates `kvm2.go` (type,
+  assertions, `New` that never dials and warns on unreadable SSH files, stub
+  methods), `id.go`, `state.go`, `errors.go` with tests:
+  `TestFormatProviderID_roundTrip`, `TestParseProviderID_malformed`,
+  `TestMapState_table`, `TestFirstIPv4_filters`, `TestClassify_table`. Also
+  registers `"kvm2": kvm2.New` in `cmd/main.go` so the two type lists change in
+  one commit. Verify: three `"kvm2"` sites in `config.go`, no import of
+  `provider/kvm`, `go test -race ./internal/provider/... && go vet ./...`.
+- **Step 3 (subagent): pure builders.** `xml.go`, `seed.go` and tests:
+  `TestDomainXML_fileDisksWithResolvedPaths`, `TestDomainXML_uefiDefaultSecureBootOff`,
+  `TestDomainXML_biosEscape`, `TestDomainXML_metadataElement`,
+  `TestDomainXML_guestAgentChannel`, `TestParseMetadata_withNamespace`,
+  `TestParseMetadata_withoutNamespaceDeclaration`,
+  `TestParseMetadata_absentGivesEmptyUID`, `TestVolumeXML_overlayBackingStore`,
+  `TestSeedISO_filesAndLabel`, `TestSeedISO_userDataByteForByte`,
+  `TestSeedISO_tempDirRemoved`, `TestVendorData_isShellScriptNotCloudConfig`,
+  `TestVendorData_multipartWhenExtraProvided`. Verify: no `type="volume"` in
+  `xml.go`; temp dir under `/tmp` with a deferred removal.
+- **Step 4 (subagent): transport, seam, decorator, fakes.** `backend.go`,
+  `ssh.go`, `hosts.go`, `observe.go`, `fake_test.go`, `sshserver_test.go` and
+  tests: `TestDial_forwardsUnixSocket`, `TestDial_unknownHostKeyRejected`,
+  `TestDial_mismatchedHostKeyRejected`, `TestDial_insecureIgnoreHostKeyAccepts`,
+  `TestDial_missingKeyFileIsTransient`, `TestDial_unparsableKeyIsPermanent`,
+  `TestDial_handshakeTimeout`, `TestCall_deadlineClosesConnectionAndReturnsTransient`,
+  `TestCall_cancelUnblocksRPC`, `TestHosts_oneDialPerHostUnderConcurrency`,
+  `TestHosts_redialsAfterDrop`, `TestHosts_unconfiguredHostIsPermanent`,
+  `TestObserve_everySeamMethodInstrumented`, `TestObserve_childSpansUnderParent`,
+  `TestObserve_logsOneLinePerCallAtV1`, `TestObserve_detailOnlyAtV2`,
+  `TestObserve_uploadBodyNeverLogged`, `TestSSH_keyMaterialNeverInErrorsOrLogs`.
+  Verify: no `context.Background()` outside tests and the keepalive goroutine;
+  `tracing.Start` in the decorator and the dialer; `go test -race -count=2`.
+- **Step 5 (subagent): image cache and pool-space precheck.** `image.go` and
+  tests: `TestEnsureImage_downloadsAndWritesMarkerLast`,
+  `TestEnsureImage_cachedWithMarkerSkipsDownload`,
+  `TestEnsureImage_missingMarkerRedownloads`,
+  `TestEnsureImage_sha256ChangeRedownloads`,
+  `TestEnsureImage_sha256MismatchIsPermanentAndRemovesVolume`,
+  `TestEnsureImage_noContentLengthIsPermanent`, `TestEnsureImage_http404IsPermanent`,
+  `TestEnsureImage_http503IsTransient`, `TestEnsureImage_concurrentCallersFetchOnce`,
+  `TestEnsureImage_cancelledMidStreamReturnsPromptly`,
+  `TestEnsureImage_streamsWithoutBuffering`, `TestCheckPoolSpace_belowNeededIsQuota`.
+  Verify: no `io.ReadAll` or `bytes.Buffer` on the download path; download span.
+- **Step 6 (subagent): the four contract methods.** `create.go`, `get.go`,
+  `delete.go`, `list.go`, `kvm2_test.go`, finishing `kvm2.go`. Tests map 1:1 to
+  contract §11: `TestCreate_repeatReturnsSameIDAndCreatesNothing`,
+  `TestCreate_resumesAfterEachPartialStep`, `TestCreate_returnsOnlyWhenRunning`,
+  `TestCreate_rejectsUnknownPlacement`, `TestCreate_adoptionWithForeignUIDIsPermanent`,
+  `TestCreate_poolOutOfSpaceIsQuotaExceeded`,
+  `TestCreate_libvirtNotFoundIsTransientNotNotFound`,
+  `TestCreate_cloudInitSentinelNeverLoggedOrInErrors`,
+  `TestGet_afterDeleteReturnsNotFound`, `TestGet_duringDeleteStillReturnsInstance`,
+  `TestGet_backendBusyIsTransientNeverNotFound`, `TestGet_runningWithoutIPIsProvisioning`,
+  `TestGet_runningWithIPIsRunning`, `TestGet_agentPreferredOverARP`,
+  `TestGet_addressErrorIsProvisioning`, `TestGet_readsUIDAndCreatedAtFromBackend`,
+  `TestGet_malformedIDIsPermanent`, `TestGet_unconfiguredHostIsPermanent`,
+  `TestDelete_malformedIDIsPermanent`, `TestDelete_twiceSucceeds`,
+  `TestDelete_removesOverlayAndSeed`, `TestDelete_runningDomainDestroyedFirst`,
+  `TestDelete_toleratesMissingVolumesAndPool`, `TestDelete_unconfiguredHostIsNoOp`,
+  `TestDelete_concurrentSameNameSafe`, `TestListByTag_excludesUnmarkedDomains`,
+  `TestListByTag_emptyUIDStillListed`, `TestListByTag_acrossHostsInConfigOrder`,
+  `TestListByTag_unreachableHostFailsWhole`, `TestListByTag_unreadableDomainSkipped`,
+  `TestProvider_cancelledContextReturnsPromptly`,
+  `TestProvider_everyErrorIsWrappedWithConfiguredName`,
+  `TestLogging_v0SilentV1OutcomeV2Detail`. Verify: `CloudInit` appears in
+  `create.go` only as `len(req.CloudInit)`; stub strings gone; `Wrap` always
+  gets the configured name; `go test -race -count=2 ./internal/provider/kvm2/`.
+- **Step 7 (subagent): Checker and the CRD comment.** `check.go`,
+  `check_test.go`: `TestCheck_oneEntryPerHostInConfigOrder`,
+  `TestCheck_hostsCheckedConcurrently`, `TestCheck_expiredContextReturnsPromptly`,
+  `TestCheck_reasonsTable`, `TestCheck_neverUsesPooledConnection`,
+  `TestCheck_detailsKeys`. Edit `api/v1alpha1/providerhealth_types.go`
+  (`status.type` comment) and run `make manifests`. Verify: only the CRD
+  description line changes under `config/`; `go test ./internal/providerhealth/`.
+- **Step 8 (subagent): deployment, Makefile, integration test.** Edit
+  `config/manager/manager.yaml` (comment: the `kvm-ssh` mount serves kvm and
+  kvm2), `config/manager/providers_config.yaml` and
+  `config/samples/providers-config.yaml` (commented `kvm2` entry),
+  `config/samples/kustomization.yaml`, `Makefile` (`test-integration-kvm2`,
+  `-run TestKVM2`, `-timeout 40m`), `docs/testing.md` (suite rows and a kvm2
+  integration section). Create `config/samples/proxy_kvm2.yaml` (zone
+  `beelink`, provider `kvm2-home`) and `kvm2_integration_test.go`
+  (`KVM2_TEST_HOST`, `_USER`, `_SSH_KEY`, `_KNOWN_HOSTS`, `_BRIDGE`,
+  `_IMAGE_URL`; optional `_IMAGE_SHA256`, `_POOL`, `_SOCKET`, `_TIMEOUT`,
+  `_KEEP`, `_LOG_LEVEL`) asserting in order: `New` does not dial; `Check`
+  Ready with details; `Create` returns `hosts/<host>/domains/<name>`; second
+  `Create` same ID, domain count unchanged; `Get` reaches Running with an IPv4
+  and the written UID and CreatedAt; `ListByTag` contains it; `Delete` nil;
+  `Get` NotFound; `Delete` again nil; pool holds base and marker, no per-VM
+  volumes; with an in-memory span recorder, `kvm2.DomainDefineXML` and
+  `kvm2.dial` spans are children of `provider.create`. Verify:
+  `go vet -tags=integration ./... && go vet -tags=e2e ./...`,
+  `bin/kustomize build config/default` and `config/samples` parse,
+  `make -n test-integration-kvm2`.
+- **Step 9 (subagent): documentation.** Create `docs/kvm2-provider.md`
+  (prerequisites incl. OVMF, cluster setup reusing `kvm-ssh`, config reference
+  pointing at the shared `KVMConfig` table, placement mapping and the
+  add-a-key-never-edit warning, what lands on the host incl. the metadata
+  namespace, image cache and quota precheck, health-check reasons, logging and
+  span names, migration from `kvm`, known gaps). Edit
+  `docs/providers-existing.md` (fourth column in the three tables and a kvm2
+  section), `docs/architecture.md` (components row, §5 block, §10 span
+  topology, §11 row, Decisions bullets), `README.md` row, `docs/providers.md`
+  guide link. Verify with the link and symbol sweep from the docs-split
+  execution log; no em-dashes in new prose.
+- **Step 10 (orchestrator): milestone review.** `go-operator-reviewer` on the
+  branch with the spec and contract as inputs, then `/code-review high`,
+  `make lint`, `make test` (full, with envtest). Fix accepted findings with
+  regression tests in one commit; record rejected ones in the summary.
+- **Step 11 (orchestrator with the user): real-host validation, changelog, MR.**
+  Hand the user a host checklist (leftover `proxy-*` domains and volumes from
+  earlier runs, OVMF installed, pool running with free space, `br0` up, the
+  known_hosts entry) and the `KVM2_TEST_*` export lines; do not run remote
+  probes unprompted. With the user's go-ahead run `make test-integration-kvm2`
+  (a permission prompt is expected; the allow-list entry
+  `make test-integration-kvm *` does not match) and, if wanted, the in-cluster
+  pass on kind: build and load the image, deploy, a scratch ConfigMap with the
+  `kvm2-home` entry, apply `proxy_kvm2.yaml`, wait for `Ready`,
+  `kubectl get ph kvm2-home -o yaml`, a `curl -x` through the proxy, delete
+  and confirm the domain and volumes are gone; scratch overrides are not
+  committed. Record cold download duration and time to IP. CHANGELOG entry
+  after the user confirms; push with the token helper; MR via the Gitea API
+  with base `main`; never merge.
+
+## Verification
+
+- Per step: the checks listed above, plus `go build ./... && go test -short ./...`
+  before every commit and `git diff --stat` reviewed for files outside the
+  step's allowed set.
+- Clean-room check on every step's diff: `grep -rn 'provider/kvm"' internal/provider/kvm2/`
+  empty; the agent's closing line "files read outside the allowed set: none";
+  identifiers that coincide with v1's names from the appendix (`libvirtAPI`,
+  `hostPool`, `imageCache`, `tracingAPI`, `lazySSH`) are treated as a smell and
+  questioned.
+- Contract mapping: after step 7, every rule of contract §2 and §8 has a named
+  test in the package; the §11 list maps 1:1 to test names (recorded in the
+  spec's entry 21).
+- End to end: the integration run's ordered assertions (step 8) on beelink, and
+  the optional kind pass (step 11).
+
+## Risks
+
+- Subagent scope creep into `internal/provider/kvm/`: instruction-only guard,
+  mitigated by the closing attestation and the diff greps.
+- `.claude/settings.json` is locally modified and uncommitted; it stays out of
+  feature commits, and the permission prompt for the new make target is
+  accepted rather than pre-approved.
+- `Content-Length` is required for downloads; the Debian and Ubuntu mirrors send
+  it.
+- The cancelled-call policy still drops other in-flight RPCs on that host once;
+  the dedicated download connection removes the expensive case and the spec
+  states the rest.
+- `go-operator-reviewer` hunts over-abstraction; the seam, the decorator and
+  the dialer interface each have one production implementation and one fake,
+  which the contract requires (§4.15, §11).
@@ -0,0 +1,640 @@
+# Plan: fix the kvm2 code-review findings
+**Created:** 2026-09-17 15:15
+
+## Context
+
+The review of `feat/kvm2-provider` (`review/REPORT.md`) confirmed 52 findings: 4 high,
+20 medium, 28 low. We fix them one at a time, highest severity first, each as its own step and
+its own commit. This plan is the campaign; the execution file's checklist is the authoritative
+progress record. Steps are detailed as we reach them, so the plan is never a fiction about work
+not yet designed.
+
+Precedent for this shape: `docs/plans/2026-09-10-2222-kvm-review-fixes.md`.
+
+## Steps
+
+- [x] **Step 0** — Save this plan into the repo, commit alone, before any code.
+- [x] **Step 1** — H1: `sshSocket.Close` ran the blocking close before the unblocking one, and
+  `markDead` killed inside a `sync.Once` before signalling dead. Shipped in `0239fb7`; the
+  reasoning is in the execution file.
+- [ ] **Step 2** — H2: an error during `StorageVolUpload` wedges go-libvirt's stream sender.
+- [ ] **Step 3** — H3: `address()` swallows every error and demotes healthy proxies.
+- [ ] **Step 4** — H4: the base image cache is shared across provider entries.
+- [ ] **Step 5** — Validate H1-H4 in-cluster against the real hypervisor (detailed below).
+- [ ] **Step 6** — Route klog into the structured logger (found during Step 5; detailed below).
+- [ ] **Step 7+** — the mediums, from `review/REPORT.md`. Detailed when reached.
+
+---
+
+## Step 2 — H2: a source error must not reach go-libvirt's stream sender
+
+### The bug
+
+`uploadImage` (`image.go:255-296`) hands a reader over `resp.Body` to `StorageVolUpload`. I
+verified the mechanism in the pinned `go-libvirt@v0.0.0-20260814190004`:
+
+```go
+// rpc.go requestStream
+abort  := make(chan bool)    // UNBUFFERED
+outErr := make(chan error)   // UNBUFFERED
+go func() { outErr <- l.socket.SendStream(serial, proc, program, out, abort) }()
+resp, err = l.processIncomingStream(c, in)
+if err != nil {
+    abort <- true
+    return resp, err          // outErr is NEVER drained on this path
+}
+err = <-outErr
+```
+
+`SendStream` (`socket/socket.go`) holds a 4 MiB buffer, polls `abort` once per loop, and on a
+**non-EOF reader error** sends a `StatusError` packet and returns. It then blocks forever on
+`outErr <-`, because the main goroutine is still inside `processIncomingStream`. libvirtd
+answers the `StatusError` with an error, so `processIncomingStream` returns an error and the
+main goroutine blocks forever on `abort <- true`. **Deadlock: `StorageVolUpload` never returns.**
+
+kvm2's `call` rescues only the caller, after `budgets.transfer` = 20 minutes, and the socket
+kill does **not** unblock either goroutine — both are parked on channel sends, not on socket
+I/O, which is what the comment at `backend.go:152-156` gets wrong. So every failed upload leaks
+kvm2's call goroutine, go-libvirt's sender goroutine and 4 MiB, permanently. The pod limit is
+128Mi and the reconciler retries on backoff.
+
+Three shapes, only the first of which is ours to fix:
+- **(A) reader error** — a truncated or failing HTTP body. The common case.
+- **(B) server-side error mid-upload** — ENOSPC, or the volume deleted underneath. After
+  `SendStream` has returned there is nothing any reader, ctx or socket close can do: confirmed
+  unfixable from our side.
+- **(C) our own abandonment** — ctx cancel or the 20 m budget kills the socket, which makes
+  `processIncomingStream` error. Same tail as (B).
+
+### The fix (workaround; the defect stays upstream)
+
+No fork, no `replace`. Two changes that between them close (A) and the realistic half of (C).
+
+**2a. Never hand go-libvirt a reader that errors** — `internal/provider/kvm2/backend.go`, at the
+seam rather than in `image.go`, so it covers the seed and marker uploads too and stays out of
+the file Step 4 rewrites:
+
+```go
+type eofReader struct {
+	r   io.Reader
+	mu  sync.Mutex
+	err error
+}
+
+func (e *eofReader) Read(b []byte) (int, error) {
+	n, err := e.r.Read(b)
+	if err == nil || err == io.EOF {
+		return n, err
+	}
+	e.mu.Lock()
+	if e.err == nil {
+		e.err = err
+	}
+	e.mu.Unlock()
+	return n, io.EOF
+}
+
+func (e *eofReader) sourceErr() error { e.mu.Lock(); defer e.mu.Unlock(); return e.err }
+```
+
+`libvirtBackend.StorageVolUpload` wraps `r` in one, and after a nil-error return converts a
+recorded `sourceErr()` into the returned error. `io.EOF` is the one outcome that ends the stream
+the normal way: libvirtd gets `StatusOK`, replies OK, `processIncomingStream` returns nil, the
+main goroutine reaches `err = <-outErr` and both goroutines unwind.
+
+`countingReader` is untouched — it sits under the wrapper, still sees the real error, so
+`uploadImage`'s `readErr != nil` branch and `TestEnsureImage_shortReadRemovesVolumeAndIsTransient`
+keep working verbatim.
+
+**Unverified, ~85%:** that libvirtd answers a short-but-clean stream finish with OK rather than
+an error. `virFDStreamWrite` enforces `length` only on writes *past* it and nothing compares
+`offset` to `length` on the close path, which is also why `virsh vol-upload` of a small file
+into a large volume works. If it answers with an error instead, we land back in a deadlock by a
+different door — but that is exactly today's behaviour, so the change is non-worsening either
+way. Probe recipe goes in the commit message; it needs a real libvirtd and a write, so it is
+for the user to run, not an agent.
+
+**2b. Make the transfer deadline land on the HTTP source, not on the libvirt socket.**
+`p.do` already builds the request with `http.NewRequestWithContext` (`image.go:360`), so the
+body is ctx-bound, and `http.DefaultClient` has no timeout of its own by deliberate choice
+(`kvm2.go:100`). Give the image GET in `fetchImage` a deadline of `defaultBudgets.transfer`
+minus a grace of one minute (precedent for reaching for `defaultBudgets` directly:
+`discardVolume`, `image.go:331`).
+
+A stalled mirror then ends by the *body* failing, which 2a turns into `io.EOF` and a clean
+unwind, roughly a minute before the libvirt budget would have killed the socket and leaked.
+That converts the realistic repeated (C) — a slow or stalling mirror on a 415 MB image — from a
+permanent leak into an ordinary Transient error. The socket kill stays as the backstop for
+libvirtd itself hanging, which is the residual we cannot close.
+
+**2c. Close the dedicated connection before unlinking the file.** Move the `discardVolume` call
+out of `uploadImage` (`image.go:291-294`) up into `fetchImage`'s upload-error branch, where the
+deferred `fresh.Close()` has already run, so libvirtd has let go of the stream's fd before the
+volume is deleted. `uploadImage` then drops its `be` parameter and its private logger — a net
+deletion. Doing it this way rather than adding an explicit `Close` keeps
+`image_test.go:573`'s `closeCount() == 1` assertion true.
+
+### Tests, docs, scope
+
+One test: `TestEOFReader_sourceErrorNeverReachesTheStream` in `backend_test.go` — a table with
+"error after data", "error with no data", "plain io.EOF passes through", "first error wins".
+~35 lines.
+
+**Deliberately not tested:** the abort protocol end to end. `libvirttest` reads a fixed 28-byte
+header per packet and implements no stream procedures, so pinning this hermetically would mean
+writing a libvirt RPC server stub — 150-250 lines whose only assertion is about *upstream's*
+goroutines. That is the bloat we cut for H1.
+
+Docs: a paragraph in `docs/kvm2-provider.md`'s *The image cache* section saying a truncated
+response is reported out of band and why, and a *Known gaps* bullet naming shapes (B) and (C)
+as an upstream defect with the two-character patch quoted, so a future reader knows it is not
+ours and what the escape hatch is:
+
+```go
+abort  := make(chan bool, 1)
+outErr := make(chan error, 1)
+```
+
+File that as an upstream issue/PR against `digitalocean/go-libvirt` and link it from the bullet.
+
+---
+
+## Step 3 — H3: an unreadable address must not look like a missing one
+
+### The bug
+
+`address()` (`get.go:104-119`) tries the guest agent then the host's ARP table. Any error that
+is not `context.Canceled`/`DeadlineExceeded` is logged at V(2) and skipped; after both it
+returns `("", nil)`. `mapState(active, "")` is `StateProvisioning` (`state.go:20-21`), the
+controller clears `p.Status.IP` on Provisioning (`proxy_controller.go:208-211`), and the health
+engine prunes the proxy's accumulated state and calls `Metrics.ForgetProxy` in one tick once
+`EffectiveHost()` is empty.
+
+The trigger that matters: another reconcile's cancel or budget overrun kills the **shared**
+pooled connection, and `call` then returns `transientf("... connection is closed: ...")` with no
+ctx sentinel, which `get.go:108` does not recognise. `Get` returns a *successful* Instance in
+Provisioning with `IP: ""`, and a healthy proxy is pulled out of rotation by an unrelated
+reconcile's timeout.
+
+### The fix
+
+A booting VM and a dead link are distinguishable: on beelink the agent source errors while the
+ARP source returns success-with-no-addresses (`docs/kvm2-provider-spec.md:146` — the host never
+sees the guest's traffic). So *one* source failing is normal; *every* source failing, or a
+connection already gone, is not.
+
+```go
+func address(ctx context.Context, log logr.Logger, be backend, dom libvirt.Domain) (string, error) {
+	errs := make([]error, 0, len(addressSources))
+	for _, s := range addressSources {
+		ifaces, err := be.DomainInterfaceAddresses(ctx, dom, uint32(s.src), 0)
+		if err != nil {
+			if isDead(be) || errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
+				return "", fmt.Errorf("reading addresses of domain %s from the %s source: %w", dom.Name, s.name, err)
+			}
+			log.V(2).Info("no addresses from source", "source", s.name, "error", err.Error())
+			errs = append(errs, fmt.Errorf("%s source: %w", s.name, err))
+			continue
+		}
+		if ip := firstIPv4(ifaces); ip != "" {
+			return ip, nil
+		}
+	}
+	if len(errs) == len(addressSources) {
+		return "", fmt.Errorf("%w: reading addresses of domain %s: every source failed: %w",
+			provider.ErrTransient, dom.Name, errors.Join(errs...))
+	}
+	return "", nil
+}
+```
+
+The explicit `provider.ErrTransient` wrap is load-bearing, not decoration: without it `classify`
+picks the first libvirt code out of the join tree, and `ErrInvalidArg`/`ErrAccessDenied`/
+`ErrNoSupport` map to **Permanent** — an address read failing would latch a *running* proxy as
+Failed, a worse version of the harm being fixed. `isDead(be)` (`backend.go:50-61`, already
+works through the `observed` decorator and the test fake) is not strictly needed — a dead link
+fails both sources anyway — but it fails faster, attributes better, and is the same predicate
+M15 will need for `ListByTag`.
+
+Also fix the now-wrong comment at `get.go:100-103` ("only ctx ending is an error").
+
+**What this does not fix, on purpose.** A guest agent restarting inside a running VM still
+demotes it: the agent errors, ARP returns empty, and the provider cannot tell that from a VM
+that is still booting. The discriminator is the *previous* IP, which the provider deliberately
+does not hold (`TestGet_readsUIDAndCreatedAtFromBackend` pins "read back, not remembered") and
+the controller already has. File as a follow-up finding against
+`proxy_controller.go:208-211`: on Provisioning, keep a Proxy's existing IP and `Provisioned`
+condition for N polls before clearing, or give the health engine a grace period before pruning.
+The one kvm2-local alternative — treating "active, no address, domain older than a provisioning
+window" as an error using the mark's `createdAt` — is a design change wearing a bug fix's
+clothes, and it invents a provisioning deadline that belongs to the controller. Rejected.
+
+### Tests
+
+- `TestGet_arpFallbackWhenAgentFails` (`get_test.go:135`) passes unchanged — one source of two.
+- `TestGet_addressErrorIsProvisioning` (`get_test.go:159`) must be **re-pointed**. It fails both
+  sources with `ErrAgentUnresponsive`, which is not realistic: code 86 is agent-specific and the
+  ARP source never produces it. Rewrite as `TestGet_agentFailsAndARPEmptyIsProvisioning` — agent
+  errors, ARP returns no addresses and no error — which is the beelink boot and the behaviour
+  the new rule must not break. Nothing covers it today.
+- One new test, `TestGet_everyAddressSourceFailingIsTransient`, two subtests: both sources error;
+  and the pooled link is dead before the address read (the fake implements `liveness`). Assert
+  Transient, not NotFound, not Permanent.
+
+Docs: `docs/kvm2-provider-spec.md:146-151` says "errors from either source mean 'no address yet'",
+now false. The `ErrTransient` row of the error-class table in `docs/kvm2-provider.md`.
+
+---
+
+## Step 4 — H4: the base image cache must not be shared across provider entries
+
+### The bug
+
+`baseVolumeName` (`volume.go:17`) is `base-<key>.<format>`, with a `.sha256` marker beside it.
+`internal/provider/kvm/image.go:44-56` builds **byte-identical** names in the same pool — for
+URL images too (kvm takes `hex(sum[:8])`, kvm2 takes `hex(sum[:])[:16]`, the same 8 bytes) —
+with a completely separate lock table. The shipped config has a live `kvm-home` entry and a
+commented-out `kvm2-home` entry on the same host, the same key `debian-13` and the same default
+pool, and the comment tells the operator to enable it. Both integration tests use key
+`test-image`. On beelink both live overlays back onto `base-debian-13.qcow2`, which kvm created.
+
+Consequences, given `cachedImage` accepts *any* 64-hex marker when no `sha256` is configured and
+never re-hashes the base:
+
+- Two entries, one key, different URLs: whichever fetched first owns the file and the other
+  silently boots the wrong image.
+- A concurrent cold fetch: B finds A's base-without-marker, deletes it, recreates and uploads
+  its own; A finishes into the unlinked inode, writes the marker, and `StorageVolGetPath`
+  resolves A's handle to B's partial file — a guest boots from a half-written base.
+- A `sha256` edit deletes the base in place under live overlays (`cachedImage` branch E), and
+  libvirt does not refuse to delete a volume that running domains use as a backing store.
+
+### The fix — the provider entry's name in the volume name
+
+`base-<entry>-<key>.<format>`, e.g. `base-kvm2-home-debian-13.qcow2`, with the marker following
+automatically. Every production call site already goes through `imageSource.volumeName()`
+(`image.go:49`), so this is a small change: give `resolveImage` the provider name and add an
+`Entry` field to `imageSource`.
+
+`Provider.name` is already carried (`kvm2.go:104`). But provider names are validated only as
+non-empty and unique (`config.go:96-105`) — no charset rule — and a name with a `/` would
+produce an unusable volume name, so **`New` must reject an entry name that is not a safe volume
+name component**, using the existing `kvmIdentifier` shape. Fail fast at startup, in kvm2 only:
+tightening the rule for every provider type is a broader change than this fix.
+
+**What this buys, and what it does not.** It fixes the cross-entry sharing outright — kvm↔kvm2
+and kvm2↔kvm2 — which is the wrong-image bug and the concurrent-fetch corruption, and it makes
+the existing per-`Provider` `imageLocks` correct rather than a latent bug, since no two entries
+can now contend for one name. It does **not** make the name mean the content, so two things
+stay open and must be said plainly rather than quietly skipped:
+
+- Editing `images.<key>.url` without changing the key is still a silent no-op. Already
+  documented behaviour ("add a new key instead"); left as is.
+- Cross-*process* fencing still depends on the atomic-publish work (upload to a temp name,
+  verify, then publish). E4 refuted the rollout and leader-handover triggers for the shipped
+  manifests, so this is a narrow residual. Its own later commit.
+
+**One companion fix, in this commit**, because the naming alone does not stop it and it is the
+remaining path that unlinks a good file under a running domain: `cachedImage` branch E
+(`image.go:129-130`, digest differs from the configured `sha256`) currently deletes the marker
+and the base. Make it return Permanent instead — "this key is cached with digest X but Y is
+configured; add the image under a new key, or delete the Proxies using it and evict the base."
+That enforces the rule the contract already states instead of contradicting it. Keep the delete
+for the no-marker and non-hex-marker branches, where the base is genuine debris.
+
+### Migration
+
+Nothing to do. The existing `base-debian-13.qcow2` on beelink is kvm's, kvm keeps using it
+untouched, and kvm2 downloads its own copy once per host (~415 MB) when `kvm2-home` is enabled.
+No window in which anything is replaced under a running guest. The pool needs room for both.
+
+### Tests and docs
+
+Tests are mostly literals becoming helpers: `volume_test.go:122-125`, `image_test.go:184,190,196`
+(`wantVolume`), `create_test.go:64-70`, `delete_test.go:67`, `provider_test.go:409`; a small
+`fx.baseName()` helper covers the last three. `image_test.go:281,298` should assert the log
+field and the `kvm2.volume` span attribute **by prefix**, which keeps pinning that the key stays
+readable in logs. `kvm2_integration_test.go:481-484` switches from `baseVolumeName(...)` to
+`resolveImage(...).volumeName()`; its assertion that the base survives `Delete` is unchanged.
+
+Two new tests:
+- `TestBaseVolumeName_isPerProviderEntry` — two entries, same key, different names; the produced
+  name never equals the kvm provider's `base-<key>.<format>`.
+- `TestNew_rejectsUnsafeEntryName` — an entry name with a `/` fails at construction.
+
+**Not tested:** that libvirt permits deleting an in-use backing file (established from libvirt
+source by the review; re-proving it needs a write on a real host), and the cross-process race
+(no hermetic way, and its fix is deferred).
+
+Docs: the *What lands on the host* table (`docs/kvm2-provider.md:198-212`), the eviction recipe
+(`:305-318`, which currently names `base-debian-13.qcow2` and would delete kvm's file), the
+migration section (`:346-370` — kvm2 never reuses a kvm base, one extra download, nothing to
+delete), the prerequisites note that the pool needs one base per distro **per entry**, the
+naming contract in `docs/kvm2-provider-spec.md` entries 4 and 10, the shared validator's message
+at `internal/provider/config_kvm.go:139-141` ("it becomes the cached volume name
+`base-<key>.<format>`", now wrong for kvm2), and the kvm2 block's comment in both
+`providers-config.yaml` files. Leave `docs/kvm-provider.md` alone; kvm is unchanged.
+
+---
+
+## Step 5 — Validate in-cluster against beelink
+
+### Why, and what this run can and cannot prove
+
+The kvm2 provider has **never run against real libvirt**: Step 11 is unchecked in
+`docs/plans-executions/2026-09-16-1333-kvm2-provider.md`, every `base-*` volume on beelink
+predates it, and both live `proxy-*` domains carry the kvm v1 metadata namespace. Four fixes
+have landed on code no hypervisor has executed. This run is first contact, and it is also the
+gate the campaign's `CHANGELOG` entry waits on.
+
+It proves the happy paths: the dial and health probe, a real 415 MB image download through the
+reader H2 now wraps, H4's new volume name appearing beside kvm's untouched one, a second Proxy
+hitting the cache, H3's address path reaching Running with an IP, and Squid answering a real
+CONNECT probe.
+
+It does **not** prove H2's error paths. The unverified ~85% claim — that libvirtd answers a
+short stream finish with OK — only arises when a body *fails*, and a healthy download never
+finishes short. The 19-minute HTTP deadline will not fire either. A green run means "the
+wrapper did not break normal uploads", not "the deadlock is settled". That still needs the
+scratch-pool probe recorded in the Step 2 execution entry.
+
+### The cluster this runs in, and the one hazard
+
+`kind-proxy-operator-demo` is **not a throwaway**. It runs the operator that currently owns
+four Ready proxies, two of them real VMs on beelink:
+
+```
+proxy-beelink-1           Managed  kvm-home     Ready  192.168.0.166  True
+proxy-beelink-2           Managed  kvm-home     Ready  192.168.0.94   True
+proxy-kubernetes-demo-1/2 Managed  kubernetes   Ready  10.244.0.7/.8  True
+```
+
+Validating kvm2 means rolling the operator that owns them. That is low risk — the controller
+and the whole `kvm` provider are untouched by H1-H4 — and rollback is one `kubectl set image`.
+
+**Do not run a second operator against beelink with `kvm-home` in its config.** Orphan GC
+(`internal/gc/gc.go:86-131`) lists every Proxy in *its own* cluster, and deletes any instance
+whose UID is not in that set and is older than `--gc-min-age` (10 m, swept every 10 m). A
+second cluster's live set would not contain `proxy-beelink-1/2`'s UIDs, so its GC would destroy
+them within ten minutes. Using the existing cluster removes the hazard entirely: its live set
+is the one that protects them.
+
+### Steps
+
+**0. Baseline** — record what is there before touching anything, so "did we break it" is
+answerable: `kubectl get px -A`, `kubectl get ph`, and a read-only `virsh vol-list default`
+plus `virsh list --all` on beelink.
+
+**1. Build and load, under a new tag.**
+
+```bash
+make docker-build IMG=egress-proxies-operator:kvm2-dev
+kind load docker-image egress-proxies-operator:kvm2-dev --name proxy-operator-demo
+```
+
+The tag **must** differ from the running `kvm-dev`. `make deploy` sets the image through
+kustomize; an identical tag leaves the Deployment spec unchanged, so no rollout happens, so the
+new process never starts, so the new providers config is never read. The ConfigMap carries no
+generator hash, so nothing else forces a restart either.
+
+The working tree is dirty (`review/`, `.claude/`), so `GIT_COMMIT` bakes in as `c10f98a-dirty`
+and `--version` will say so. Cosmetic; commit or stash `review/` first to avoid it.
+
+**2. Enable kvm2 in the deployed config.** Uncomment the `kvm2-home` block in
+`config/manager/providers_config.yaml` (it is the shipped default, commented out). Keep its
+`address: beelink.tail8c69c.ts.net:22` — the Tailscale FQDN, which is what
+`~/.ssh/kvm_known_hosts` pins. `config/samples/providers-config.yaml` uses the short
+`beelink:22`, which **fails host-key verification**; I hit exactly that when validating SSH
+earlier in this session.
+
+**3. Deploy.**
+
+```bash
+make deploy IMG=egress-proxies-operator:kvm2-dev
+kubectl -n egress-proxies-operator-system rollout status deploy/egress-proxies-operator-controller-manager
+```
+
+`make deploy` rewrites `config/manager/kustomization.yaml`'s image block as a side effect; it
+already carries `kvm-dev` from a previous run.
+
+**4. STOP GATE — provider health, before anything is created.**
+
+```bash
+kubectl get ph
+kubectl get ph kvm2-home -o yaml
+```
+
+`kvm2-home` must be `Ready 1/1` with reason `Reachable` and details carrying `hostname`,
+`libvirtVersion`, `poolState: running` and `poolAvailableGiB`. This proves SSH, the host key,
+the libvirt handshake and the pool **without creating a thing**, and it exercises the
+dedicated-connection dial and close that H1 touched. A failure here is free to diagnose:
+`CredentialsUnavailable`, `Unauthorized`, `Forbidden`, `PoolNotActive` each name their own
+cause. Report and wait for a go before step 5.
+
+**5. Two Proxies.** Two manifests in the scratchpad (not committed), modelled on
+`config/samples/proxy_kvm2.yaml` — `proxy-kvm2-1` and `proxy-kvm2-2` in `default`, matching the
+existing `proxy-beelink-*` convention, `provider: kvm2-home`, `zone: beelink`,
+`machineType: small`, `image: debian-13`, port 3128 and the sample's Squid cloud-init. Apply
+the first, let it finish, then the second — sequencing them is what makes the cache hit
+observable.
+
+```bash
+kubectl get px -w
+kubectl -n egress-proxies-operator-system logs -f deploy/egress-proxies-operator-controller-manager
+```
+
+**6. What to check, and which finding it exercises.**
+
+| Check | Finding |
+|---|---|
+| `fetching base image into host pool` then `base image cached` with size, duration, digest | H2's wrapped reader on a real 415 MB upload |
+| `virsh vol-list default` shows `base-kvm2-home-debian-13.qcow2` **and** `base-debian-13.qcow2` still there | H4: per-entry naming, kvm's cache untouched |
+| Second Proxy logs no `fetching`, and the server is not hit again | H4 + the image lock: cache hit under the new name |
+| `px` goes Provisioning → Running with an IP, without flapping back to Provisioning | H3: the address path, and no spurious demotion |
+| `Healthy=True` | Squid answers a real CONNECT probe end to end |
+| `virsh list --all` shows two new `proxy-*` domains | Create |
+| After deleting both: domains and their `.qcow2`/`-seed.iso` gone, base image still present | Delete, and the cache surviving |
+
+Optionally `curl -x http://<ip>:3128 https://example.com -I` from this box, if it routes to
+192.168.0.0/24 — the existing kvm proxies live on that LAN.
+
+**7. Cleanup.** `kubectl delete px proxy-kvm2-1 proxy-kvm2-2` and let the finalizer do the
+work, then verify on beelink read-only. **Any leftover domain or volume: I print the exact
+`virsh undefine --nvram` / `vol-delete` lines and you run them** — no write reaches the shared
+host from me. Then decide whether the `kvm2-home` ConfigMap entry stays (keep it if you want
+kvm2 running for real; `git checkout config/manager/providers_config.yaml` otherwise).
+
+**Rollback at any point:**
+
+```bash
+kubectl -n egress-proxies-operator-system set image \
+  deploy/egress-proxies-operator-controller-manager manager=egress-proxies-operator:kvm-dev
+```
+
+### Who runs what
+
+I drive steps 0-6 and stop at gate 4 for a go. Every beelink write is yours. Nothing here
+touches `proxy-beelink-1/2`, `base-debian-13.qcow2`, your unrelated VMs, or the five
+pre-existing orphan volumes the review found.
+
+---
+
+## Step 6 — Route klog into the structured logger
+
+**Not a review finding.** Found while watching the Step 5 validation: some lines ignore the
+zap encoder entirely, so no amount of `--zap-encoder=json` makes the stream parseable.
+
+```
+I0917 19:50:02.580097       1 leaderelection.go:258] "Attempting to acquire leader lease..." lock="..."
+```
+
+That is klog's own text sink. client-go's leader elector does `logger := klog.FromContext(ctx)`
+(`leaderelection.go:257`), and controller-runtime hands the elector a context with no logr in it
+(`pkg/manager/internal.go:461-473` — `mgr.Start` gets `ctrl.SetupSignalHandler()`, which carries
+no logger). `klog.FromContext` falls back to `Background()`, and with no logger registered that
+is klog's stderr writer. Nothing in this repo imports or configures klog; it is an indirect
+dependency (`k8s.io/klog/v2 v2.140.0`) that ~60 linked packages use, including
+`client-go/tools/{leaderelection,cache,record}`, `client-go/transport` and `client-go/rest`.
+controller-runtime does not route it either — its only `klog.SetLogger` call is in its own test
+suite.
+
+### 6a. `cmd/main.go` — one line and an import
+
+```go
+logger := zap.New(zap.UseFlagOptions(&opts))
+ctrl.SetLogger(logger)
+klog.SetLogger(logger)
+```
+
+**Plain `SetLogger`, deliberately not `SetLoggerWithOptions(logger, klog.ContextualLogger(true))`.**
+The contextual form is what klog's own docs recommend, because it skips a klogr round trip — but
+it also disables klog's verbosity gate ("Such a logger cannot rely on verbosity checking in
+klog", `contextual.go:83-86`), handing the decision to zap's level. With `Development: true`
+that level is Debug, which enables V(1), so client-go's V(1) chatter that klog currently
+suppresses at its default `-v=0` would start appearing. Plain `SetLogger` keeps klog's own check
+in front ("klog will do its own verbosity checks before calling logger.V().Info",
+`contextual.go:43-44`) and still redirects everything it does emit ("all log lines will be
+suppressed from the regular output, and redirected to the logr implementation", `:47-48`).
+Same fix, no new volume. Performance is irrelevant at this scale.
+
+`go.mod` promotes `k8s.io/klog/v2` from indirect to direct. It is already in the module graph,
+so nothing new is downloaded.
+
+### 6b. `internal/discovery/server.go` — the other stderr bypass
+
+The discovery API's `http.Server` (`:113-119`) sets no `ErrorLog`, so `net/http`'s own errors —
+`Accept` failures, handler panics that escape — go to `log.Default()` and out to stderr in
+stdlib format. `s.log` is already in scope two lines above:
+
+```go
+srv := &http.Server{
+	Handler:           s.handler(),
+	ErrorLog:          slog.NewLogLogger(logr.ToSlogHandler(s.log), slog.LevelError),
+	ReadHeaderTimeout: readHeadTimeout,
+	...
+}
+```
+
+`logr.ToSlogHandler` is the pattern the repo already uses for the same purpose in
+`internal/provider/gcp/wirelog.go:31-36`. This server is plain HTTP with no TLS, so `ErrorLog`
+carries genuine faults rather than handshake noise from scanners — Error level is right.
+
+### 6c. `config/manager/manager.yaml` — make the deployed stream parseable
+
+```yaml
+args:
+  - --leader-elect
+  - --health-probe-bind-address=:8081
+  - --providers-config=/etc/proxy-operator/providers.yaml
+  - --zap-encoder=json
+  - --zap-time-encoding=rfc3339
+```
+
+Verified that the flag wins over the in-code default: `addDefaults` only fills `NewEncoder`
+`if o.NewEncoder == nil` (`zap.go:176-191`), and `BindFlags` sets it from `--zap-encoder`.
+`--zap-time-encoding=rfc3339` matters because the JSON encoder defaults to `epoch`, which would
+give `1758138422.31` instead of a timestamp anyone can read. Verbosity is left at debug, so the
+per-RPC lines stay — that was the choice.
+
+### Deliberately not doing
+
+`opts := zap.Options{Development: true}` (`cmd/main.go:154-155`) stays. Flipping it to a
+production default would change the format for every existing deployment, not just this one, and
+the flags in 6c get the same result here without that blast radius.
+
+### What stays unformatted regardless, and why that is fine
+
+Flag-parse errors and `--version` (`cmd/main.go:160-163`) print before any logger exists, and
+`utilruntime.Must` panics in `init()` run before `main`. controller-runtime's own metrics and
+health servers also set no `ErrorLog` (`pkg/internal/httpserver/server.go:11-19`) and expose no
+option to set one. All three are outside reach; none can fire during steady-state operation.
+
+### Tests
+
+None. All three changes are wiring: two struct fields and one manifest key. `cmd/main.go` has no
+test coverage at all, and pinning `klog.SetLogger` hermetically means mutating global state that
+would leak into every other test in the package. Verified by observation instead — see below.
+Same call as H1's: a test that costs more than it protects does not earn its place.
+
+### Verification
+
+Rebuild, redeploy, and watch the rollout produce the line that started this:
+
+```bash
+kubectl -n egress-proxies-operator-system logs deploy/egress-proxies-operator-controller-manager \
+  | grep -i 'leader' | head -3
+```
+
+It must be JSON with an rfc3339 `ts`, not `I0917 ...`. Then confirm the whole stream parses:
+
+```bash
+kubectl -n egress-proxies-operator-system logs deploy/egress-proxies-operator-controller-manager \
+  | jq -e -s 'length > 0' >/dev/null && echo "every line is JSON"
+```
+
+That second command is the real check — it fails if *any* line is not JSON, which is exactly the
+property being fixed.
+
+---
+
+## Ordering
+
+**H2 → H3 → H4**, three commits.
+
+H3 is orthogonal: it touches only `get.go` and `get_test.go`.
+
+H2 and H4 both touch `image.go` but at different heights, and the order matters. H2's `image.go`
+edit is a deletion — `discardVolume` moves out of `uploadImage` and up into `fetchImage` — while
+H4 edits `imageSource.volumeName()`, `resolveImage`, and `cachedImage`'s digest branch ten lines
+below. Landing H2 first means H4 rebases cleanly; the reverse means H4 is written against a
+failure path H2 is about to restructure. Keeping H2's guard in `backend.go` rather than in
+`countingReader` is part of the same separation.
+
+Semantically, H2 makes `fetchImage` delete the base after the transfer connection is closed, and
+that delete is only unambiguously safe once names are per-entry — so H2's commit message should
+say the discard is still unfenced against a second entry sharing the name and that H4 closes it,
+and H4's should point back. The deferred atomic-publish work then lands exactly where H2 put the
+discard, on a volume this call created under a name no other entry can produce.
+
+## Verification
+
+```bash
+go build ./... && go vet ./...
+go test -race -count=1 ./internal/provider/kvm2/
+make test
+```
+
+Per step, the check that actually matters is that the new test fails against the unfixed code —
+stash the source edit, keep the test, watch it fail, restore. For H2 that means asserting
+`eofReader` converts before the fix exists; for H3, that the dead-link subtest returns a
+successful Provisioning instance today; for H4, that two entries produce one name today.
+
+After H4, run the kvm2 integration test on beelink. **It has never run** — Step 11 is unchecked
+in `docs/plans-executions/2026-09-16-1333-kvm2-provider.md`, every `base-*` on the host predates
+kvm2, and both live domains carry the kvm v1 metadata namespace. It is the first end-to-end
+exercise of the new name and does a real cold download beside kvm's untouched base.
+
+Each step: its own commit on `feat/kvm2-provider` with the execution-file entry and its checklist
+line ticked. `CHANGELOG.md` lands with a "validate and record" step once the user confirms the
+work, as in the kvm campaign. Trailer:
+`Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>`.
@@ -1,13 +1,17 @@
 # The existing providers

 This is the appendix to [provider-contract.md](provider-contract.md): how the
-three providers that exist today (kubernetes, gcp, kvm) meet the contract,
-which profile each one chose on every axis, and what is still missing. It is
-reference material for whoever writes the next provider, not a user guide;
-the user guides are [kvm-provider.md](kvm-provider.md),
+four providers that exist today (kubernetes, gcp, kvm, kvm2) meet the
+contract, which profile each one chose on every axis, and what is still
+missing. It is reference material for whoever writes the next provider, not
+a user guide; the user guides are [kvm-provider.md](kvm-provider.md),
+[kvm2-provider.md](kvm2-provider.md),
 [gcp-vm-validation.md](gcp-vm-validation.md) and
 [gcp-in-specific-project.md](gcp-in-specific-project.md). The README's
-"Implemented providers" table stays the short public summary.
+"Implemented providers" table stays the short public summary. kvm2 is the
+one provider built from the contract itself (clean-room, from
+[kvm2-provider-spec.md](kvm2-provider-spec.md), without reading kvm's code),
+so its column is also a record of what the contract produced.

 The per-provider call flows are drawn in
 [architecture.md §5](architecture.md#5-what-provider-calls-do-in-the-outside-world),
@@ -23,59 +27,59 @@ Code references are a file path plus a function or symbol name.
 One row per axis of contract §4, in the same order, with the profile name
 and the symbol that shows it.

-| Axis | kubernetes | gcp | kvm |
-|---|---|---|---|
-| `readiness` | immediate: a Pod is scheduled within seconds (`stateFromPod`) | provisioning-window: minutes of boot (`mapState`) | provisioning-window: boot plus, on first use, an image download (`imageCache.ensure`) |
-| `create-return` | submit: returns after `client.Create` | submit: `Insert` is never awaited, `Operation.Wait` is not called | **blocking**: waits through `DomainCreate` (`ensureRunning`), and the first image fetch runs inside (`imageCache.fetch`) |
-| `delete-return` | returns-early: `client.Delete`, Pod terminates on its own | returns-early: `Delete` submits an operation each time | waits: destroy, undefine and volume removal complete before return (`removeDomain`, `removeVolumes`) |
-| `ip-source` | assigned-at-create: `PodIP` (`instanceFromPod`) | assigned-at-create: the external NAT IP of the first NIC (`toInstance`) | discovered: ARP table first, then the guest agent, first routable IPv4 (`address`, `firstIPv4`) |
-| `gc-mark` | native-labels: Pod labels (`buildPod`) | native-labels: instance labels (`buildInsertRequest`) | written-metadata: a namespaced `<metadata>` element written in `DomainDefineXML` (`metadata.go`) |
-| `created-at` | backend-timestamp: `CreationTimestamp` | backend-timestamp: `CreationTimestamp`, parse error gives zero | written-at-create: `createdAt` in the metadata element (`newInstanceMetadata`) |
-| `list-partial` | n/a: one call, all namespaces | omit-unreachable: `ReturnPartialSuccess` on `AggregatedList` | fail-whole per host, one unreadable domain skipped (`ListByTag`, `managedInstance`) |
-| `quota` | absent: a quota 403 is Permanent (`classify`) | detected: 429, and 403 with reason `quotaExceeded` or `rateLimitExceeded` (`classify`, `hasReason`) | absent (`classify`) |
-| `not-found` | status-404: `apierrors.IsNotFound` | status-404: `isNotFound` | error-code: `ErrNoDomain`, `ErrNoStorageVol` (`hasCode`) |
-| `primary-key` | name-is-key: `namespace/name` | name-is-key: `zones/<zone>/instances/<name>` | name-is-key: `hosts/<host>/domains/<name>` |
-| `provisioning-state` | native: `PodPending` | native: `PROVISIONING`, `STAGING`, `REPAIRING` | create-blocks: a domain is defined and started in one `Create`; `nostate` maps to Provisioning |
-| `async-ops` | none | task-handle, but the handle is discarded: an insert that fails after acceptance is never read | none |
-| `secondary-resources` | none | cascade: `AutoDelete` on the boot disk (`buildInsertRequest`) | explicit: overlay and seed removed by name (`removeVolumes`); base image kept |
-| `credentials` | workload-identity: in-cluster config or kubeconfig (`NewWithTransportWrapper`) | environment: Application Default Credentials | mounted-files: SSH key and known_hosts from the optional `kvm-ssh` Secret, read lazily (`newLazySSH`) |
-| `client-kind` | vendor-sdk: controller-runtime `client.Client`, no seam interface | vendor-sdk: `cloud.google.com/go/compute` behind `instancesAPI` | rpc-over-ssh: `go-libvirt` over an SSH-forwarded unix socket, behind `libvirtAPI` (`conn.go`) |
-| `timeouts` | ctx-only: no `rest.Config.Timeout` is set | ctx-only: SDK defaults, with SDK retries on Get and List | per-call for the dial only (`defaultHandshakeTimeout` in `dialSSH`); RPCs are ctx-only and a cancelled `call` closes the connection |
-| `logging` | none | tiered: V(1) outcomes and `logAPIError`, V(2) detail, V(5) wire log (`wirelog.go`) | tiered: V(1) outcomes and one `libvirt call` line per RPC (`trace.go`), V(2) XML and addresses; no wire tier |
-| `child-spans` | transport-wrapper: `tracing.RestConfigWrapper()` on its own `rest.Config`, parent-gated | sdk-instrumentation: the SDK adds `otelhttp` itself; not parent-gated, but every call site has a parent | **none**: `trace.go` is a log decorator, not spans |
-| `backend-test` | cluster-e2e: kind (`test/e2e`) | manual-runbook: [gcp-vm-validation.md](gcp-vm-validation.md) | integration-tag: `kvm_integration_test.go`, `make test-integration-kvm` |
+| Axis | kubernetes | gcp | kvm | kvm2 |
+|---|---|---|---|---|
+| `readiness` | immediate: a Pod is scheduled within seconds (`stateFromPod`) | provisioning-window: minutes of boot (`mapState`) | provisioning-window: boot plus, on first use, an image download (`imageCache.ensure`) | provisioning-window: boot plus, on first use, an image download (`ensureImage`) |
+| `create-return` | submit: returns after `client.Create` | submit: `Insert` is never awaited, `Operation.Wait` is not called | **blocking**: waits through `DomainCreate` (`ensureRunning`), and the first image fetch runs inside (`imageCache.fetch`) | **blocking**: waits through `DomainCreate` and a 15-second poll for running (`ensureRunning`, `waitRunning`), and the first image fetch runs inside (`fetchImage`) |
+| `delete-return` | returns-early: `client.Delete`, Pod terminates on its own | returns-early: `Delete` submits an operation each time | waits: destroy, undefine and volume removal complete before return (`removeDomain`, `removeVolumes`) | waits: destroy, undefine and volume removal complete before return (`removeDomain`, `removeVolumes`) |
+| `ip-source` | assigned-at-create: `PodIP` (`instanceFromPod`) | assigned-at-create: the external NAT IP of the first NIC (`toInstance`) | discovered: ARP table first, then the guest agent, first routable IPv4 (`address`, `firstIPv4`) | discovered: the guest agent first, then the ARP table, first routable IPv4 (`addressSources`, `address`, `firstIPv4`) |
+| `gc-mark` | native-labels: Pod labels (`buildPod`) | native-labels: instance labels (`buildInsertRequest`) | written-metadata: a namespaced `<metadata>` element written in `DomainDefineXML` (`metadata.go`) | written-metadata: an element in kvm2's own namespace written in `DomainDefineXML` (`metadataURI`, `newInstanceMetadata`) |
+| `created-at` | backend-timestamp: `CreationTimestamp` | backend-timestamp: `CreationTimestamp`, parse error gives zero | written-at-create: `createdAt` in the metadata element (`newInstanceMetadata`) | written-at-create: `createdAt` in the metadata element, parse error gives zero (`newInstanceMetadata`, `createdAt`) |
+| `list-partial` | n/a: one call, all namespaces | omit-unreachable: `ReturnPartialSuccess` on `AggregatedList` | fail-whole per host, one unreadable domain skipped (`ListByTag`, `managedInstance`) | fail-whole per host, one unreadable domain skipped unless ctx ended; `ID`, `UID` and `CreatedAt` from the name and the mark only, no state or address RPC (`ListByTag`) |
+| `quota` | absent: a quota 403 is Permanent (`classify`) | detected: 429, and 403 with reason `quotaExceeded` or `rateLimitExceeded` (`classify`, `hasReason`) | absent (`classify`) | detected: a pool-space precheck before any volume is created (`checkPoolSpace`, `neededBytes`) |
+| `not-found` | status-404: `apierrors.IsNotFound` | status-404: `isNotFound` | error-code: `ErrNoDomain`, `ErrNoStorageVol` (`hasCode`) | error-code: `ErrNoDomain`, `ErrNoStorageVol`, Transient inside Create (`classifyLibvirt`) |
+| `primary-key` | name-is-key: `namespace/name` | name-is-key: `zones/<zone>/instances/<name>` | name-is-key: `hosts/<host>/domains/<name>` | name-is-key: `hosts/<host>/domains/<name>`, with a UID guard on adoption (`create`) |
+| `provisioning-state` | native: `PodPending` | native: `PROVISIONING`, `STAGING`, `REPAIRING` | create-blocks: a domain is defined and started in one `Create`; `nostate` maps to Provisioning | create-blocks: defined, started and polled to running in one `Create`; `nostate` maps to Provisioning (`mapState`) |
+| `async-ops` | none | task-handle, but the handle is discarded: an insert that fails after acceptance is never read | none | none |
+| `secondary-resources` | none | cascade: `AutoDelete` on the boot disk (`buildInsertRequest`) | explicit: overlay and seed removed by name (`removeVolumes`); base image kept | explicit: overlay and seed removed by name (`removeVolumes`); base image kept |
+| `credentials` | workload-identity: in-cluster config or kubeconfig (`NewWithTransportWrapper`) | environment: Application Default Credentials | mounted-files: SSH key and known_hosts from the optional `kvm-ssh` Secret, read lazily (`newLazySSH`) | mounted-files: the same `kvm-ssh` Secret, read on every dial and never cached (`loadCredentials`) |
+| `client-kind` | vendor-sdk: controller-runtime `client.Client`, no seam interface | vendor-sdk: `cloud.google.com/go/compute` behind `instancesAPI` | rpc-over-ssh: `go-libvirt` over an SSH-forwarded unix socket, behind `libvirtAPI` (`conn.go`) | rpc-over-ssh: `go-libvirt` over an SSH-forwarded unix socket, behind the ctx-first `backend` seam (`backend.go`) |
+| `timeouts` | ctx-only: no `rest.Config.Timeout` is set | ctx-only: SDK defaults, with SDK retries on Get and List | per-call for the dial only (`defaultHandshakeTimeout` in `dialSSH`); RPCs are ctx-only and a cancelled `call` closes the connection | per-call: a budget around the dial, every RPC, the uploads and the download (`defaultBudgets`, `call`); a cancelled or overlong call kills the pooled connection |
+| `logging` | none | tiered: V(1) outcomes and `logAPIError`, V(2) detail, V(5) wire log (`wirelog.go`) | tiered: V(1) outcomes and one `libvirt call` line per RPC (`trace.go`), V(2) XML and addresses; no wire tier | tiered: V(1) outcomes and one `libvirt call` line per RPC (`observe.go`), V(2) XML and addresses; no wire tier |
+| `child-spans` | transport-wrapper: `tracing.RestConfigWrapper()` on its own `rest.Config`, parent-gated | sdk-instrumentation: the SDK adds `otelhttp` itself; not parent-gated, but every call site has a parent | **none**: `trace.go` is a log decorator, not spans | span-decorator: `kvm2.<Method>` per RPC (`observe`), plus `kvm2.dial` (`libvirtDialer`) and `kvm2.download` (`fetchImage`) |
+| `backend-test` | cluster-e2e: kind (`test/e2e`) | manual-runbook: [gcp-vm-validation.md](gcp-vm-validation.md) | integration-tag: `kvm_integration_test.go`, `make test-integration-kvm` | integration-tag: `kvm2_integration_test.go`, `make test-integration-kvm2` |

 ## 2. Other differences

 Rows that are not profile axes but that a new provider author compares
 against.

-| Feature | kubernetes | gcp | kvm |
-|---|---|---|---|
-| Create verifies the adopted instance's UID | no | no | no |
-| Placement used | none | Zone, MachineType, Image | Zone (host), MachineType, Image |
-| CloudInit delivered | **no**: ignored | yes: `user-data` metadata | yes: NoCloud seed ISO built in-process (`buildSeedISO`) |
-| Port used | yes: Squid config and containerPort (`squidConf`) | no: firewall is manual | no |
-| States produced | Provisioning, Running, Terminated | all four | Provisioning, Running, Stopped |
-| Deleting instance still reports its state | yes: a Pod being deleted reports its phase | n/a | n/a |
-| Error.Provider is the configured name | **no**: hardcoded `"kubernetes"` | yes | yes |
-| Checker target | `apiserver` | the project | each host, concurrently |
-| Checker Details | none | none | `user`, `pool`, `hostname`, `libvirtVersion`, `poolState`, `poolCapacity`, `poolAvailable` |
-| Credential rotation without restart | n/a | via ADC | no: cached after first successful load |
-| Connection handling | client-go | SDK client, never closed | one pooled SSH plus libvirt connection per host, 15s keepalive (`hostPool`) |
-| Unit-test double | controller-runtime fake client | `fakeAPI` behind `instancesAPI`, plus `httptest` for the real SDK client in `check_test.go` | `fakeAPI` behind `libvirtAPI`, plus an in-process `x/crypto/ssh` server in `conn_test.go` |
-| Proxy software | generated Squid config (`pod.go`) | from `cloudInit`; the image must run cloud-init | from `cloudInit`, plus a vendor-data script that installs the guest agent (`seed.go`) |
-| IP reachable from the operator | Pod network, so in-cluster only (not from `make run-dev` on a laptop) | external IP, once a firewall rule allows the cluster's egress IP | LAN address on the host's bridge; the operator needs a route to that LAN |
-| Distinct egress IP per proxy | no: the cluster's egress | yes: ephemeral external IP | whatever the host's LAN routes through; not per VM by itself |
+| Feature | kubernetes | gcp | kvm | kvm2 |
+|---|---|---|---|---|
+| Create verifies the adopted instance's UID | no | no | no | **yes**: a domain without the kvm2 mark or with a different non-empty UID is Permanent; a marked one with a matching or empty UID is adopted (`create`) |
+| Placement used | none | Zone, MachineType, Image | Zone (host), MachineType, Image | Zone (host), MachineType, Image |
+| CloudInit delivered | **no**: ignored | yes: `user-data` metadata | yes: NoCloud seed ISO built in-process (`buildSeedISO`) | yes: NoCloud seed ISO built in-process (`buildSeedISO`) |
+| Port used | yes: Squid config and containerPort (`squidConf`) | no: firewall is manual | no | no |
+| States produced | Provisioning, Running, Terminated | all four | Provisioning, Running, Stopped | Provisioning, Running, Stopped |
+| Deleting instance still reports its state | yes: a Pod being deleted reports its phase | n/a | n/a | n/a: Delete waits, and a concurrent Get sees the domain until it is undefined |
+| Error.Provider is the configured name | **no**: hardcoded `"kubernetes"` | yes | yes | yes (`wrapErr`) |
+| Checker target | `apiserver` | the project | each host, concurrently | each host, concurrently, on a fresh connection |
+| Checker Details | none | none | `user`, `pool`, `hostname`, `libvirtVersion`, `poolState`, `poolCapacity`, `poolAvailable` | `user`, `pool`, `hostname`, `libvirtVersion`, `poolState`, `poolCapacityGiB`, `poolAvailableGiB`, filled in as far as the check got |
+| Credential rotation without restart | n/a | via ADC | no: cached after first successful load | yes: both files read on every dial (`loadCredentials`) |
+| Connection handling | client-go | SDK client, never closed | one pooled SSH plus libvirt connection per host, 15s keepalive (`hostPool`) | one pooled SSH plus libvirt connection per host, 15s keepalive (`hostTable`); dedicated connections for the image upload and the check (`dialFresh`) |
+| Unit-test double | controller-runtime fake client | `fakeAPI` behind `instancesAPI`, plus `httptest` for the real SDK client in `check_test.go` | `fakeAPI` behind `libvirtAPI`, plus an in-process `x/crypto/ssh` server in `conn_test.go` | `fakeBackend` behind `backend` (`fake_test.go`), plus an in-process `x/crypto/ssh` server (`sshserver_test.go`) and go-libvirt's own mock for the adapter |
+| Proxy software | generated Squid config (`pod.go`) | from `cloudInit`; the image must run cloud-init | from `cloudInit`, plus a vendor-data script that installs the guest agent (`seed.go`) | from `cloudInit`, plus a vendor-data script that installs the guest agent (`seed.go`) |
+| IP reachable from the operator | Pod network, so in-cluster only (not from `make run-dev` on a laptop) | external IP, once a firewall rule allows the cluster's egress IP | LAN address on the host's bridge; the operator needs a route to that LAN | LAN address on the host's bridge; the operator needs a route to that LAN |
+| Distinct egress IP per proxy | no: the cluster's egress | yes: ephemeral external IP | whatever the host's LAN routes through; not per VM by itself | whatever the host's LAN routes through; not per VM by itself |

 ## 3. Observability status

-| Layer | kubernetes | gcp | kvm |
-|---|---|---|---|
-| 1. Operation spans and metrics | yes: decorators | yes: decorators | yes: decorators |
-| 2. Child spans per backend call | yes: `tracing.RestConfigWrapper()`, parent-gated (`cmd/main.go`, `NewWithTransportWrapper`) | yes: the SDK's own `otelhttp`; not parent-gated | **no** |
-| 3. Logs carry `traceID` and `spanID` | n/a: the package does not log | yes, including the V(5) wire log | yes: every `libvirt call` line at V(1) |
-| Tested against a real collector | yes: the kind e2e checks `provider.create` and `provider.delete` in Tempo (`test/e2e/tracing_test.go`), operation spans only | no | no |
+| Layer | kubernetes | gcp | kvm | kvm2 |
+|---|---|---|---|---|
+| 1. Operation spans and metrics | yes: decorators | yes: decorators | yes: decorators | yes: decorators |
+| 2. Child spans per backend call | yes: `tracing.RestConfigWrapper()`, parent-gated (`cmd/main.go`, `NewWithTransportWrapper`) | yes: the SDK's own `otelhttp`; not parent-gated | **no** | yes: `kvm2.<Method>` from the `observe` decorator, `kvm2.dial`, `kvm2.download`; always under the caller's span, since every call carries the caller's ctx |
+| 3. Logs carry `traceID` and `spanID` | n/a: the package does not log | yes, including the V(5) wire log | yes: every `libvirt call` line at V(1) | yes: every `libvirt call` line at V(1), written inside its RPC's span |
+| Tested against a real collector | yes: the kind e2e checks `provider.create` and `provider.delete` in Tempo (`test/e2e/tracing_test.go`), operation spans only | no | no | no; the integration test asserts the parent-child topology in an in-memory recorder (`assertLifecycleSpans`) |

 ## 4. kubernetes (`internal/provider/kubernetes`)

@@ -329,18 +333,213 @@ the author's view.
    tries ARP first; it says `host` and `name` are on every log line, but
    the check and list paths lack them.

-## 7. Gaps shared by the existing providers
+## 7. kvm2 (`internal/provider/kvm2`)

- **Adoption is not verified.** None of the three checks `LabelUID` when
-  `Create` finds an existing instance with the right name.
- **The Create comment does not match kvm.** The interface comment in
-  `internal/provider/provider.go` says `Create` returns as soon as the
-  request is submitted; kvm blocks until the domain has started.
+The full user guide is [kvm2-provider.md](kvm2-provider.md) and the design
+it was built from is [kvm2-provider-spec.md](kvm2-provider-spec.md); this
+section is the author's view. Nothing in the package was taken from
+`internal/provider/kvm`; where the two agree it is because both follow the
+same libvirt.
+
+- **Package and seam.** `kvm2.go` (`Provider`, `New`, `withDefaults`,
+  `host`, `wrapErr`, the per-name lock), `create.go`, `get.go`,
+  `delete.go`, `list.go` (the four methods), `backend.go` (the seam, the
+  production adapter and the budgets), `ssh.go` (credentials, the forwarded
+  socket, keepalive), `hosts.go` (the pooled table and the production
+  dialer), `observe.go` (the span and log decorator), `image.go` (cache,
+  download, quota precheck), `seed.go`, `xml.go`, `volume.go`,
+  `metadata.go`, `id.go`, `state.go`, `errors.go`, `check.go`. The seam is
+  `backend`: the twenty go-libvirt calls the provider uses, each with `ctx`
+  first (go-libvirt's own methods take none and block), plus `Close`.
+  `libvirtBackend` adapts `*libvirt.Libvirt` and runs every call through
+  `call`, which applies the budget and the abandon policy. The `dialer`
+  seam separates the SSH transport: `newProvider` injects one for tests,
+  `newLibvirtDialer` is production. The `observe` decorator is applied
+  once, in `libvirtDialer.dial`, so provider code never holds an
+  undecorated backend.
+- **Config.** `ProviderConfig.KVM2`, the same `KVMConfig` type as kvm,
+  validated by `validateKVMBlock` with block name `kvm2` so messages read
+  `kvm2.hosts`. Defaults (`diskSizeGb` 10, socket, pool, image format and
+  firmware) are filled by `withDefaults`. `New` never dials; it checks that
+  the credential files exist and warns when one does not, or when
+  `insecureIgnoreHostKey` is on.
+- **ID.** `formatProviderID(host, name)` gives `hosts/<host>/domains/<name>`;
+  `parseProviderID` accepts exactly four segments with the literal
+  `hosts` and `domains` and non-empty host and name, else `ErrPermanent`.
+- **Create.** `Create` validates `Zone` (`host`; optional with one host),
+  `MachineType` (a catalogue key) and `Image` (`resolveImage`: a catalogue
+  key or an http(s) URL with qcow2 and uefi defaults) before any dial, and
+  logs the resolved parameters with `cloudInitBytes` at V(2). Then, under
+  the per-name lock (`lockName`) and on the pooled connection
+  (`hostTable.get`), `create`: `DomainLookupByName`; found means the
+  adoption guard (no kvm2 mark, or a non-empty metadata UID that differs
+  from `req.UID`, is Permanent) and `ensureRunning`. Otherwise
+  `StoragePoolLookupByName`, `imageSizeIfUncached` (marker lookup, then
+  `HEAD`, or a closed `GET` when `HEAD` is not 2xx) and
+  `checkPoolSpace` with `neededBytes` (image if uncached, plus
+  `diskSizeGb`, plus 16 MiB), then `ensureImage` (`cachedImage` on the
+  pooled connection, `fetchImage` streaming into `uploadImage` on a
+  `dialFresh` connection, `StoragePoolRefresh`, `writeMarker` last),
+  `ensureOverlay` (`ErrStorageVolExist` means reuse), `writeSeed` (delete,
+  build, upload; always rebuilt), two `StorageVolGetPath`, `buildDomainXML`
+  with file-type disks and the metadata element, `DomainDefineXML`,
+  `ensureRunning` (`DomainGetState`, where paused or pmsuspended is
+  Permanent; `DomainCreate` with `ErrOperationInvalid` as already running;
+  `waitRunning` polls every 500 ms for up to 15 seconds, Transient on
+  expiry and at once for any inactive state other than `nostate`).
+  `Create` returns only when the domain is running.
+- **Get and state.** `mapState(state, ip)`:
+
+  | libvirt state | IP | Provider state |
+  |---|---|---|
+  | nostate | any | Provisioning |
+  | running, blocked | empty | Provisioning |
+  | running, blocked | set | Running |
+  | paused, shutdown, shutoff, crashed, pmsuspended, anything else | any | Stopped |
+
+  `get`: `DomainLookupByName` (only `ErrNoDomain` reaches NotFound),
+  `readMetadata` (`DomainGetMetadata` with `metadataURI`;
+  `ErrNoDomainMetadata` means unmanaged with an empty UID; an element that
+  does not parse is Permanent), then `DomainGetState` and, for an active
+  domain, `address`. `addressSources` asks the guest agent first,
+  then ARP; a source that fails is skipped at V(2), only ctx ending is an
+  error, so a cancelled Get is Transient and not a false Provisioning.
+  `firstIPv4` drops loopback, link-local and unspecified addresses and an
+  address whose type tag and literal disagree. `UID` and `CreatedAt` are
+  read back on every call; an unparsable `createdAt` gives the zero time.
+- **Delete.** Under the per-name lock: `removeDomain` (`DomainDestroy`,
+  tolerating not running and gone; `DomainUndefineFlags` with NVRAM,
+  managed save and snapshot metadata) and `removeVolumes` (overlay and seed
+  by name through `deleteVolumeIfPresent`; a missing pool is a no-op). A
+  host no longer in the config is a logged no-op success. Nothing is left
+  when nil is returned.
+- **ListByTag.** `ConnectListAllDomains` (active and inactive) on every
+  host in config order; `readMetadata` and `managed` keep kvm2's own
+  domains, and the `Instance` is `ID`, `UID` and `CreatedAt` from the name
+  and the element, with `State` and `IP` empty: GC reads nothing else
+  (contract §2.4), so no state or address RPC is made. A domain whose
+  metadata cannot be read is skipped at V(1), unless ctx ended, which
+  fails the call. A host that cannot be dialled or listed fails the whole
+  call.
+- **Errors.** `classify(op, err)`: an error already carrying a class keeps
+  it; libvirt codes through `classifyLibvirt` (`ErrNoDomain` and
+  `ErrNoStorageVol` NotFound, Transient when `op == "create"`;
+  `ErrNoStoragePool`, `ErrXMLError`, `ErrNoSupport`, `ErrInvalidArg`,
+  `ErrAuthFailed`, `ErrAccessDenied` Permanent; the rest Transient);
+  `knownhosts.KeyError` and the `unable to authenticate` text from
+  `x/crypto/ssh` Permanent; everything else Transient, including a missing
+  credential file. `permanentf` and `transientf` classify at the source:
+  placement, XML builders, a parse failure of the key or known_hosts, an
+  image 4xx, a missing or zero `Content-Length`, a sha256 mismatch, a pool
+  that is not running (Permanent); an image 429 or 5xx, a short body, the
+  running wait expiring (Transient). `checkPoolSpace` is the only source of
+  `ErrQuotaExceeded`. `logAPIError` writes `kvm2 backend call failed` with
+  `libvirtErrorFields`.
+- **Check.** `Check` runs `checkHost` per host concurrently, each through
+  `probeHost` on a `dialFresh` connection that is closed when the probe
+  ends; the pooled table is never consulted. `ConnectGetLibVersion`,
+  `ConnectGetHostname`, `StoragePoolLookupByName`, `StoragePoolGetInfo`;
+  a pool that is not running is `poolNotActiveError`. Details are filled
+  as facts arrive (`user` and `pool` always; `libvirtVersion`, `hostname`,
+  `poolState`, `poolCapacityGiB`, `poolAvailableGiB`). `checkHost` answers
+  `Timeout` whenever the round's ctx ended, whatever stage was running and
+  whatever shape the error has; otherwise `checkReason`: a `dialError` is
+  judged by its stage (`credentials` gives `CredentialsUnavailable`; `ssh`
+  with a `knownhosts.KeyError` or the auth text gives `Unauthorized`; any
+  other stage, including a dial that ran out of its own budget, gives
+  `Unreachable`); then `context.DeadlineExceeded` or `Canceled` `Timeout`,
+  `ErrAccessDenied` or `ErrAuthFailed` `Forbidden`, `ErrNoStoragePool`
+  `PoolNotFound`, any other libvirt code `APIError`, anything else
+  `Unreachable`; `PoolNotActive` from the pool state.
+- **Connections.** `hostTable` keeps one pooled connection per host,
+  dialled lazily under a per-host one-slot semaphore so waiting ends with
+  the caller's ctx; a dead backend (`liveness`) is closed and redialled on
+  the next `get`. `dialSSHSocket` runs credentials, TCP, the SSH handshake
+  and the `direct-streamlocal` channel to the socket, each failure a
+  `dialError` with its stage; `libvirtDialer.connect` bounds the whole
+  sequence plus the libvirt `Connect` with the dial budget and starts the
+  `keepalive` (15 s, `keepalive@openssh.com`) and a watcher on go-libvirt's
+  `Disconnected`. `call` runs each blocking go-libvirt call in a goroutine
+  and waits on the result, ctx and a budget timer; abandoning a call
+  (`link.markDead`) closes the socket and the SSH client, returns Transient
+  wrapping the ctx or deadline error, and every other call waiting on that
+  link fails once with Transient. `loadCredentials` runs on every dial.
+- **Logging.** Logger from the context, named `kvm2`, with `provider`,
+  and `host` and `name` on lines about an instance. V(1): one line per
+  operation outcome (`kvm2 domain ensured running`, `kvm2 domain fetched`,
+  `kvm2 domain deleted`, `kvm2 domains listed`, `kvm2 host checked`), one
+  `libvirt call` line per RPC from `observe` (rpc name, identifiers,
+  duration, error fields), the connection events `libvirt host connected`
+  and `libvirt host connect failed` with `stage`, and `kvm2 backend call
+  failed`. V(2): XML, metadata, addresses, listed names, pool sizes, the
+  resolved Create parameters. Upload and download bodies, cloud-init and
+  key material are never logged.
+- **Tracing.** `observe.call` starts `kvm2.<Method>` per RPC with
+  `kvm2.host` and the identifiers (`kvm2.domain`, `kvm2.pool`,
+  `kvm2.volume`), sets `kvm2.libvirt_code` on failure and writes the
+  `libvirt call` line inside the span; `libvirtDialer.dial` starts
+  `kvm2.dial` (`kvm2.address`, `kvm2.user`); `fetchImage` starts
+  `kvm2.download` (`kvm2.url_host`, `kvm2.sha256_configured`,
+  `kvm2.bytes`). Both the decorator and the dialer take `tracing.Option`
+  so tests inject a recorder; `New` leaves them on the global provider.
+- **Tests.** Pure builders (`xml_test.go`, `volume_test.go`,
+  `seed_test.go`, `id_test.go`, `metadata_test.go`, `state_test.go`);
+  `errors_test.go` (the classify table with real SSH error strings and a
+  real `knownhosts.KeyError`); `backend_test.go` (the adapter over
+  go-libvirt's own mock through a `net.Pipe`: deadline kills the
+  connection, cancel unblocks, killed by another caller); `hosts_test.go`;
+  `ssh_test.go` over `sshserver_test.go` (every dial stage, host-key
+  policies, key material never in errors or logs); `image_test.go`
+  (marker rules, streaming asserted by counting bytes, dedicated upload
+  connection); `create_test.go`, `get_test.go`, `delete_test.go`,
+  `list_test.go`, `kvm2_test.go`, `provider_test.go` (the contract §11
+  list, named test by test in the spec's entry 21); `observe_test.go`
+  (every seam method spanned and logged exactly once, by reflection);
+  `check_test.go` (the reason table, never touches the pool). The double is
+  `fakeBackend` in `fake_test.go`, whose `twin` shares state under its own
+  call log so a test can tell the pooled connection from a dedicated one.
+  Real backend: `kvm2_integration_test.go` behind `//go:build integration`,
+  driven by `KVM2_TEST_*` variables and `make test-integration-kvm2`,
+  described in
+  [testing.md](testing.md#the-kvm2-integration-test-real-hypervisor).
+- **Deployment.** The `kvm-ssh` Secret and its mount are shared with kvm
+  (`config/manager/manager.yaml` says so), `make kvm-ssh-secret` is reused,
+  a commented `kvm2-home` entry in both providers configs,
+  `config/samples/proxy_kvm2.yaml` (zone `beelink`, matching the shipped
+  config), the `kvm2` type in the `status.type` comment of
+  `api/v1alpha1/providerhealth_types.go`. No RBAC.
+- **Known gaps** (see also [kvm2-provider.md § Known gaps](kvm2-provider.md#known-gaps)).
+  - `Create` blocks, including the first image download on each host.
+  - GC blind spots: volumes without a domain are invisible to GC, and so are
+    VMs on a host removed from the config, which `Delete` treats as done.
+  - No adoption of kvm's domains: a switched entry needs its Proxies
+    deleted first.
+  - Cancelling one operation cuts off the others on that host once: a
+    cancelled `call` kills the pooled connection.
+  - `Content-Length` is required for image downloads.
+  - A paused or pmsuspended domain is not started by a repeat `Create`:
+    `ensureRunning` returns Permanent, so the Proxy latches until its spec
+    changes; the admin resumes the VM on the host.
+  - Credential parse errors carry the parser's own text, which can quote a
+    PEM header type or a malformed known_hosts token (never key bytes).
+  - No reused-overlay check: an existing overlay is reused without checking
+    its backing image or size.
+  - The sample cloud-init installs Squid with `http_access allow all`.
+
+## 8. Gaps shared by the existing providers
+
+- **Adoption is not verified in three of the four.** kubernetes, gcp and
+  kvm adopt any existing instance with the right name without checking
+  `LabelUID`; kvm2 returns `ErrPermanent` when the UID differs.
+- **The Create comment does not match kvm and kvm2.** The interface comment
+  in `internal/provider/provider.go` says `Create` returns as soon as the
+  request is submitted; both libvirt providers block until the domain has
+  started.
 - **`Stopped` and `Terminated` are produced inconsistently.** kubernetes
-  never produces Stopped, kvm never produces Terminated, and the reconciler
-  treats them the same.
+  never produces Stopped, kvm and kvm2 never produce Terminated, and the
+  reconciler treats them the same.
 - **Different `ListByTag` partial-failure policies.** gcp leaves failed
-  parts out, kvm fails the whole list. Both are safe.
+  parts out, kvm and kvm2 fail the whole list. Both are safe.
 - **Stale `mock` comments** in `internal/provider/provider.go`,
  `internal/provider/registry/registry.go` and `api/v1alpha1/proxy_types.go`
  (and so the generated CRD's `spec.provider` description).
@@ -351,33 +550,41 @@ the author's view.
  is still open.
 - **Tracing.** kvm has no child spans, so a kvm trace shows `provider.create`
  and nothing under it; correlation works only through `traceID` on its V(1)
-  log lines. Only kubernetes is tested against a real collector, and that
-  test checks only the operation spans; nothing checks that gcp's SDK spans
-  or the kubernetes API spans nest under `provider.*`.
+  log lines. kvm2 has them, but only an in-memory recorder checks their
+  topology. Only kubernetes is tested against a real collector, and that
+  test checks only the operation spans; nothing checks that gcp's SDK spans,
+  the kubernetes API spans or kvm2's spans nest under `provider.*` in a
+  real collector.

-## 8. Proposed features
+## 9. Proposed features

-Listed only. Nothing below is designed or implemented.
+Listed only. Nothing below is designed or implemented, except where a
+bullet says kvm2 already does it.

 - **Child spans in kvm:** extend the `tracingAPI` decorator in
  `kvm/trace.go`, which already wraps every libvirt call, to also start a
  span per RPC, and add spans for the SSH dial and the image download. The
  seam's methods take no `ctx` today (`hostPool.call` hands `fn` only the
-  API), so the span context has to be plumbed in.
- **Tempo e2e checks for child spans**, and for gcp and kvm through their
-  real backend runbooks.
+  API), so the span context has to be plumbed in. kvm2's `observe.go` over
+  its ctx-first seam is the shape this would take.
+- **Tempo e2e checks for child spans**, and for gcp, kvm and kvm2 through
+  their real backend runbooks.
 - **A shared conformance suite:** for example `internal/provider/providertest`,
  which runs the contract §11 tests against any provider plus a fake
  backend, so a new provider gets the contract tests for free.
 - **Verify the UID when `Create` adopts an instance.** Return `ErrPermanent`
  on a mismatch instead of adopting an instance that belongs to someone else.
+  kvm2 does this; the other three do not.
 - **Report asynchronous create failures:** gcp keeps the operation name, and
  `Get` reports the failure instead of NotFound.
 - **Capacity or quota signals** for kvm, such as pool free space and host
-  memory, mapped to `ErrQuotaExceeded`.
+  memory, mapped to `ErrQuotaExceeded`. kvm2 has the pool-space precheck;
+  host memory is still unchecked in both.
 - **GC for secondary resources:** mark volumes and similar resources, and
  let `ListByTag` or a second listing method cover them.
- **Reload credentials and config without a restart.**
+- **Reload credentials and config without a restart.** kvm2 re-reads its
+  SSH material on every dial; the providers config is still read once per
+  process for every type.
 - **Checker `Details`** for kubernetes (server version) and gcp (quota
  headroom).
 - **Internal-IP mode** for gcp, and a firewall rule managed per provider.
@@ -13,11 +13,13 @@ the existing ones happen to work.
  a new provider's design answers, one entry per decision the contract leaves
  open. Copy it to `docs/<type>-provider-spec.md`, fill it in, and hand it to
  the implementer with the other two files.
- **[providers-existing.md](providers-existing.md)**: how kubernetes, gcp and
-  kvm meet the contract, a table mapping each of them to every profile, their
-  per-provider reference notes, shared gaps and proposed features.
+- **[providers-existing.md](providers-existing.md)**: how kubernetes, gcp,
+  kvm and kvm2 meet the contract, a table mapping each of them to every
+  profile, their per-provider reference notes, shared gaps and proposed
+  features.

 Per-provider user guides: [kvm-provider.md](kvm-provider.md),
+[kvm2-provider.md](kvm2-provider.md),
 [gcp-vm-validation.md](gcp-vm-validation.md) and
 [gcp-in-specific-project.md](gcp-in-specific-project.md).

@@ -62,6 +62,11 @@ manual runs too.
 | | `trace_test.go`, `conn_test.go`, `errors_test.go` | every `libvirtAPI` method traced exactly once (guarded by reflection on the interface), results and errors pass through, libvirt error fields, detail keys only at V2, upload content never logged; pool connect/drop lines, no wrapping below V1 | `fakeAPI` + `funcr` capture |
 | | `check_test.go`, `conn_test.go` | health check: details, per-host verdicts in config order, hosts checked concurrently, a timeout that keeps the connection and refuses a second check while the first still waits, no waiting behind a reconcile's dial, the reason table; `probe` abandons without closing, but still drops on a transport error; waiting for a host's slot ends with the caller's deadline; an inactive or missing pool and daemon errors keep the connection; every dial stage tagged through the real SSH server, with classification unchanged | `fakeAPI`, fake dialers, in-process SSH server |
 | | `kvm_integration_test.go` (`-tags=integration`) | one real VM: image fetch, Create, idempotent re-Create, Running with IP, ListByTag, Delete, pool state afterwards; `TestKVMCheck_realHost` runs the health check against the host and creates nothing | a **real hypervisor** — see below |
+| `internal/provider/kvm2` | `xml_test.go`, `volume_test.go`, `seed_test.go`, `id_test.go`, `metadata_test.go`, `state_test.go` | pure builders: domain XML (file disks with resolved paths, UEFI with secure boot off, BIOS escape, the metadata element, the guest-agent channel), volume XML with and without a backing store, seed ISO reopened (files, label, user-data byte for byte, temp dir removed) and the vendor-data script vs. multipart rules, providerID round-trip and malformed forms, metadata in both namespace spellings, state mapping and the IPv4 filter | pure |
+| | `errors_test.go`, `backend_test.go`, `hosts_test.go`, `ssh_test.go` | classification table (libvirt codes, SSH auth text, `knownhosts.KeyError`, missing files); the per-call budget adapter (deadline kills the connection and is Transient, cancel unblocks, killed by another caller, close idempotent) against a mocked libvirt socket; the host table (one dial per host under concurrency, redial after drop, `dialFresh` never cached, waiting ends with ctx); the real dialer end to end: socket forwarding, unknown and mismatched host keys, insecure mode, missing vs. unparsable key and known_hosts, wrong key, handshake timeout, cancel, every stage tagged, key material never in errors or logs | fake dialer; a **real in-process `x/crypto/ssh` server** forwarding a unix socket |
+| | `image_test.go`, `create_test.go`, `get_test.go`, `delete_test.go`, `list_test.go`, `kvm2_test.go`, `provider_test.go` | image cache (fetch with the marker written last, cached, marker missing or without a digest, sha256 change or mismatch, 4xx/5xx, no Content-Length, short read, concurrent callers fetch once, cancel mid-stream, streamed without buffering, upload on a dedicated connection) and the pool-space quota precheck; Create resume after every partial step, returns only when running, adoption guard on UID, seed always rebuilt, cloud-init sentinel never logged; Get state and address mapping incl. agent-before-ARP; every Delete already-gone case, NVRAM flag, per-name lock; ListByTag across hosts, one down fails the whole call, unreadable domain skipped; `New` never dials and warns on missing files; every error wrapped with the configured name; logging tiers V0/V1/V2 | `httptest` image server + in-memory `fakeBackend` behind the `backend` seam |
+| | `observe_test.go`, `check_test.go` | every seam method traced and logged exactly once (guarded by reflection), child spans under the parent, detail only at V2, upload bodies never logged, error fields; health check: one entry per host in config order, hosts checked concurrently, expired ctx returns promptly, never touches the pooled connection, the reason table, detail keys | `fakeBackend` + `funcr` capture + span recorder |
+| | `kvm2_integration_test.go` (`-tags=integration`) | one real VM: `New` without a dial, `Check` Ready with details, Create, idempotent re-Create with an unchanged domain count, Running with IP and the metadata round-trip, ListByTag, Delete, NotFound, pool state afterwards, `kvm2.*` spans under `provider.create`; `TestKVM2Check_realHost` runs the health check against the host and creates nothing | a **real hypervisor** — see below |
 | `internal/health` | `probe_test.go` | probes through a **real CONNECT-capable proxy stub** to a real TLS server: tunnel success, refused CONNECT, wrong status, dead proxy, plain-http forward | `httptest` + hijacked bidirectional tunnel |
 | | `engine_test.go` | thresholds, first-verdict, latency suppression matrix, dropped-event retry, stale-probe UID guard, tick scheduling/seeding/pruning, `Start` end-to-end | fake `client.Reader` + injected `probeFn` |
 | `internal/lease` | `store_test.go` | capacity, `MaxLeases=0`, all three selection tie-breaks, target-scoped vs global cooldowns, expiry via fake clock, report-on-retained-lease, idempotent release, **40 concurrent acquires never exceeding capacity** | injectable clock; the concurrency case is why `-race` matters |
@@ -93,12 +98,12 @@ envtest suite the scaffold generated.
  bug (below).
 - **Live GCP**: no test talks to Google. The seam boundary
  (`buildInsertRequest` + classification) is tested exhaustively instead.
- **Live libvirt, by default**: the KVM unit tests run over an in-memory
-  fake of the 20 libvirt calls the provider uses. Whether libvirt accepts
-  the domain XML, whether a raw qcow2 upload plus pool refresh yields a
-  usable backing image, and whether the guest actually boots and reports
-  an address are provable only on a real host — that is the opt-in
-  integration test below.
+- **Live libvirt, by default**: the KVM and kvm2 unit tests run over an
+  in-memory fake of the 20 libvirt calls each provider uses. Whether
+  libvirt accepts the domain XML, whether a raw qcow2 upload plus pool
+  refresh yields a usable backing image, and whether the guest actually
+  boots and reports an address are provable only on a real host — that is
+  the opt-in integration test below (one per provider).

 ## The KVM integration test (real hypervisor)

@@ -162,6 +167,95 @@ go test -tags=integration -count=1 -v -run TestKVMCheck ./internal/provider/kvm/
 The test is not part of `make test` and not run in CI. Keep it compiling
 with `go vet -tags=integration ./...` when the provider's internals move.

+## The kvm2 integration test (real hypervisor)
+
+`internal/provider/kvm2/kvm2_integration_test.go` compiles only under
+`-tags=integration` and skips itself unless the environment names a
+host. Like the kvm one it deviates from the "testcontainers for real
+infra" convention on purpose: libvirt needs `/dev/kvm` and a bridge to a
+DHCP network, and a container has neither. It targets whichever
+hypervisor you point it at and leaves that host's pool with one more
+cached base image (`base-test-image.qcow2` and its `.sha256` marker) than
+it had. The kvm and kvm2 tests can share a host: kvm2 marks its domains
+in its own metadata namespace and never lists or adopts the other's.
+
+```sh
+export KVM2_TEST_HOST=beelink:22               # SSH endpoint
+export KVM2_TEST_USER=novakj                   # must be in the libvirt group
+export KVM2_TEST_SSH_KEY=$HOME/.ssh/id_ed25519 # private key for that user
+export KVM2_TEST_KNOWN_HOSTS=$HOME/.ssh/kvm_known_hosts  # the hypervisors' host keys
+export KVM2_TEST_BRIDGE=br0                    # bridge to a network with DHCP
+export KVM2_TEST_IMAGE_URL=https://cloud.debian.org/images/cloud/trixie/latest/debian-13-generic-amd64.qcow2
+make test-integration-kvm2
+```
+
+`KVM2_TEST_KNOWN_HOSTS` is technically optional, but leaving it out turns
+host-key verification **off** for the run (the log says so); keep a
+`kvm_known_hosts` holding just the hypervisors rather than pointing the
+test at your everyday `known_hosts`, so a re-keyed hypervisor is a
+one-line fix in a file nothing else uses.
+
+Also optional: `KVM2_TEST_IMAGE_SHA256` (verified after download and
+compared with the cached marker), `KVM2_TEST_POOL` (default `default`),
+`KVM2_TEST_SOCKET` (default `/var/run/libvirt/libvirt-sock`),
+`KVM2_TEST_TIMEOUT` (default `20m`, the budget for the whole lifecycle;
+it must cover the first-time image download), `KVM2_TEST_KEEP` to leave
+the VM behind for `virsh console` when something goes wrong (the log then
+prints the `virsh undefine --nvram` and `virsh vol-delete` lines that
+clean it up), and `KVM2_TEST_LOG_LEVEL` (default `1`) for how much of the
+provider's logging the `-v` output shows: at `1` every libvirt RPC is one
+compact `libvirt call` line (`HH:MM:SS.mmm kvm2 "level"=1 "msg"="libvirt
+call" "rpc"="DomainLookupByName" "domain"=... "duration"=...`, plus the
+per-operation outcome lines and connection events), interleaved with the
+test's own `t.Log` lines; `2` adds the domain and volume XML, metadata
+and addresses; `0` leaves only `t.Log`. The per-run constants (provider,
+host, domain name) are stated once in the first line and left off every
+provider line. See the Logging section of
+[kvm2-provider.md](kvm2-provider.md).
+
+What one run asserts, in order, each step logged with its duration:
+
+1. `New` returns within 100 ms and without dialling (no `libvirt host
+   connected` line before the first call).
+2. `Check` reports the one host `Ready` with reason `Reachable` and the
+   `hostname`, `libvirtVersion`, `poolState` (`running`) and
+   `poolAvailableGiB` details, which are logged.
+3. Every later call goes through `provider.WithTracing` with an in-memory
+   span recorder that the package's own spans also land in.
+4. `Create` returns `hosts/kvm2-test/domains/<name>` (downloading the
+   image on a cold pool — the log prints how long it took).
+5. A second `Create` returns the same ID and the host's domain count,
+   read over a dedicated connection, is unchanged.
+6. `Get`, polled every 5 s, reaches `Running` with an IPv4 within the
+   budget; its `UID` is the one Create wrote and `CreatedAt` is non-zero
+   and within a few minutes of now. The time from Create's return to the
+   first address is logged.
+7. `ListByTag` contains the ID with that UID.
+8. A TCP connect to port 22 on the guest, logged only — the machine
+   running the test may not route to the bridge network.
+9. `Delete` returns nil, `Get` is `ErrNotFound`, a second `Delete` is a
+   no-op.
+10. The pool, inspected over a dedicated connection, holds the base image
+    and its marker but no `<name>.qcow2` or `<name>-seed.iso`.
+11. Among the recorded spans, a `provider.create` span has at least one
+    `kvm2.DomainDefineXML` and one `kvm2.dial` child, and no span
+    attribute carries the cloud-init text.
+
+With `KVM2_TEST_KEEP` steps 9 and 10 are skipped; otherwise a `t.Cleanup`
+attempts the Delete even when an earlier assertion failed.
+
+`TestKVM2Check_realHost` runs the provider health check
+([kvm2-provider.md](kvm2-provider.md)) against the same host. It must
+find the host reachable and the pool running, and it creates nothing. It
+needs the same variables; to run only this test:
+
+```sh
+go test -tags=integration -count=1 -v -run TestKVM2Check ./internal/provider/kvm2/
+```
+
+The test is not part of `make test` and not run in CI. Keep it compiling
+with `go vet -tags=integration ./...` when the provider's internals move.
+
 ## The scaffolded `make test-e2e` suite

 `test/e2e/` is the kubebuilder-generated smoke suite (build image → kind
@@ -23,6 +23,7 @@ require (
 	k8s.io/api v0.36.0
 	k8s.io/apimachinery v0.36.0
 	k8s.io/client-go v0.36.0
+	k8s.io/klog/v2 v2.140.0
 	sigs.k8s.io/controller-runtime v0.24.1
 	sigs.k8s.io/yaml v1.6.0
 )
@@ -104,7 +105,6 @@ require (
 	k8s.io/apiextensions-apiserver v0.36.0 // indirect
 	k8s.io/apiserver v0.36.0 // indirect
 	k8s.io/component-base v0.36.0 // indirect
-	k8s.io/klog/v2 v2.140.0 // indirect
 	k8s.io/kube-openapi v0.0.0-20260317180543-43fb72c5454a // indirect
 	k8s.io/streaming v0.36.0 // indirect
 	k8s.io/utils v0.0.0-20260210185600-b8788abfbbc2 // indirect
@@ -9,6 +9,7 @@ import (
 	"context"
 	"crypto/subtle"
 	"encoding/json"
+	"log/slog"
 	"net"
 	"net/http"
 	"strings"
@@ -112,6 +113,10 @@ func (s *Server) Start(ctx context.Context) error {

 	srv := &http.Server{
 		Handler: s.handler(),
+		// Without this, net/http's own faults (a failed Accept, a panic that
+		// escapes a handler) go to the stdlib logger on stderr, outside the
+		// structured stream.
+		ErrorLog:          slog.NewLogLogger(logr.ToSlogHandler(s.log), slog.LevelError),
 		ReadHeaderTimeout: readHeadTimeout,
 		ReadTimeout:       10 * time.Second,
 		WriteTimeout:      10 * time.Second,
@@ -22,6 +22,9 @@ type ProviderConfig struct {
 	Kubernetes *KubernetesConfig `json:"kubernetes,omitempty"`
 	GCP        *GCPConfig        `json:"gcp,omitempty"`
 	KVM        *KVMConfig        `json:"kvm,omitempty"`
+	// KVM2 is the same block shape as KVM, for the kvm2 provider type: one
+	// or more libvirt/KVM hypervisors reached over SSH.
+	KVM2 *KVMConfig `json:"kvm2,omitempty"`
 }

 // setBlocks names the type-specific blocks that are non-nil, so validate can
@@ -37,6 +40,9 @@ func (p ProviderConfig) setBlocks() []string {
 	if p.KVM != nil {
 		out = append(out, "kvm")
 	}
+	if p.KVM2 != nil {
+		out = append(out, "kvm2")
+	}
 	return out
 }

@@ -100,7 +106,7 @@ func (c *Config) validate() error {

 		prefix := fmt.Sprintf("providers[%d] %q", i, p.Name)
 		switch p.Type {
-		case "kubernetes", "gcp", "kvm":
+		case "kubernetes", "gcp", "kvm", "kvm2":
 		case "":
 			return fmt.Errorf("%s: type is required", prefix)
 		default:
@@ -120,6 +126,10 @@ func (c *Config) validate() error {
 			if err := validateKVM(prefix, p.KVM); err != nil {
 				return err
 			}
+		case "kvm2":
+			if err := validateKVMBlock(prefix, "kvm2", p.KVM2); err != nil {
+				return err
+			}
 		}
 	}
 	return nil
@@ -108,29 +108,42 @@ var sha256Hex = regexp.MustCompile(`^[0-9a-fA-F]{64}$`)
 // kvmIdentifier constrains the config names that are pasted into
 // identifiers elsewhere: a host name goes into the providerID
 // hosts/<name>/domains/<vm>, which is split on "/" to parse, and an image
-// key becomes the file name of the cached base volume. A slash in either
-// produces an ID or a path that never resolves again, so the VM leaks with
-// nothing able to address it.
+// key becomes part of the file name of the cached base volume. A slash in
+// either produces an ID or a path that never resolves again, so the VM
+// leaks with nothing able to address it.
 var kvmIdentifier = regexp.MustCompile(`^[A-Za-z0-9][A-Za-z0-9_.-]*$`)

+// IsKVMIdentifier reports whether s is safe to paste into a providerID or a
+// volume name. The provider entry's own name reaches a volume name too, and
+// that name is validated for every provider type at once (only non-empty
+// and unique), so the kvm2 provider checks its own at construction.
+func IsKVMIdentifier(s string) bool { return kvmIdentifier.MatchString(s) }
+
 func validateKVM(prefix string, c *KVMConfig) error {
+	return validateKVMBlock(prefix, "kvm", c)
+}
+
+// validateKVMBlock checks a KVMConfig under the given block name, so the
+// kvm and kvm2 provider types share one validator and each reports its own
+// block in messages.
+func validateKVMBlock(prefix, block string, c *KVMConfig) error {
 	if c == nil {
-		return fmt.Errorf("%s: kvm block is required", prefix)
+		return fmt.Errorf("%s: %s block is required", prefix, block)
 	}
 	if c.SSHKeyFile == "" {
-		return fmt.Errorf("%s: kvm.sshKeyFile is required", prefix)
+		return fmt.Errorf("%s: %s.sshKeyFile is required", prefix, block)
 	}
 	if c.KnownHostsFile == "" && !c.InsecureIgnoreHostKey {
-		return fmt.Errorf("%s: kvm.knownHostsFile is required unless kvm.insecureIgnoreHostKey is true", prefix)
+		return fmt.Errorf("%s: %s.knownHostsFile is required unless %s.insecureIgnoreHostKey is true", prefix, block, block)
 	}
 	if c.DiskSizeGB < 0 {
-		return fmt.Errorf("%s: kvm.diskSizeGb must not be negative", prefix)
+		return fmt.Errorf("%s: %s.diskSizeGb must not be negative", prefix, block)
 	}
 	for name, img := range c.Images {
-		p := fmt.Sprintf("%s: kvm.images[%q]", prefix, name)
+		p := fmt.Sprintf("%s: %s.images[%q]", prefix, block, name)
 		if !kvmIdentifier.MatchString(name) {
 			return fmt.Errorf("%s: key must be letters, digits, dots, dashes or underscores"+
-				" — it becomes the cached volume name base-<key>.<format>", p)
+				" — it becomes part of the cached base volume's name", p)
 		}
 		if !IsHTTPURL(img.URL) {
 			return fmt.Errorf("%s.url must be an http(s) URL (got %q)", p, img.URL)
@@ -146,19 +159,19 @@ func validateKVM(prefix string, c *KVMConfig) error {
 		}
 	}
 	if len(c.MachineTypes) == 0 {
-		return fmt.Errorf("%s: kvm.machineTypes must define at least one profile", prefix)
+		return fmt.Errorf("%s: %s.machineTypes must define at least one profile", prefix, block)
 	}
 	for name, mt := range c.MachineTypes {
 		if mt.VCPUs <= 0 || mt.MemoryMiB <= 0 {
-			return fmt.Errorf("%s: kvm.machineTypes[%q] needs vcpus > 0 and memoryMiB > 0", prefix, name)
+			return fmt.Errorf("%s: %s.machineTypes[%q] needs vcpus > 0 and memoryMiB > 0", prefix, block, name)
 		}
 	}
 	if len(c.Hosts) == 0 {
-		return fmt.Errorf("%s: kvm.hosts must list at least one host", prefix)
+		return fmt.Errorf("%s: %s.hosts must list at least one host", prefix, block)
 	}
 	seen := make(map[string]bool, len(c.Hosts))
 	for i, h := range c.Hosts {
-		p := fmt.Sprintf("%s: kvm.hosts[%d]", prefix, i)
+		p := fmt.Sprintf("%s: %s.hosts[%d]", prefix, block, i)
 		if h.Name == "" {
 			return fmt.Errorf("%s: name is required", p)
 		}
@@ -214,6 +214,100 @@ providers:
 	}
 }

+func TestLoadConfig_kvm2Valid(t *testing.T) {
+	t.Parallel()
+	yaml := strings.Replace(validKVMYAML, "type: kvm\n    kvm:\n", "type: kvm2\n    kvm2:\n", 1)
+	cfg, err := LoadConfig([]byte(yaml))
+	if err != nil {
+		t.Fatalf("LoadConfig() error = %v, want nil", err)
+	}
+	if cfg.Providers[0].KVM != nil {
+		t.Errorf("providers[0].kvm = %+v, want nil", cfg.Providers[0].KVM)
+	}
+	k := cfg.Providers[0].KVM2
+	if k == nil {
+		t.Fatal("providers[0].kvm2 = nil")
+	}
+	if k.SSHKeyFile != "/etc/proxy-operator/kvm/id_ed25519" || k.KnownHostsFile != "/etc/proxy-operator/kvm/known_hosts" || k.DiskSizeGB != 20 {
+		t.Errorf("kvm2 = %+v, want sshKeyFile, knownHostsFile and diskSizeGb parsed", k)
+	}
+	if k.Images["ubuntu"].Format != "raw" {
+		t.Errorf("images[ubuntu].format = %q, want raw", k.Images["ubuntu"].Format)
+	}
+	if mt := k.MachineTypes["small"]; mt.VCPUs != 1 || mt.MemoryMiB != 1024 {
+		t.Errorf("machineTypes[small] = %+v, want {1 1024}", mt)
+	}
+	if len(k.Hosts) != 2 || k.Hosts[0].Name != "nas1" || k.Hosts[1].Pool != "fast" {
+		t.Errorf("hosts = %+v, want nas1 and lab2 with lab2's pool parsed", k.Hosts)
+	}
+}
+
+func TestLoadConfig_kvm2Invalid(t *testing.T) {
+	t.Parallel()
+	minimal := func(body string) string {
+		return "providers:\n  - name: kvm2-home\n    type: kvm2\n    kvm2:\n" + indent(body, 6)
+	}
+	hostsAndTypes := `
+machineTypes:
+  small: {vcpus: 1, memoryMiB: 1024}
+hosts:
+  - {name: nas1, address: nas1:22, user: virt, bridge: br0}
+`
+	tests := []struct {
+		name       string
+		yaml       string
+		wantErrSub string
+	}{
+		{
+			name:       "missing kvm2 block",
+			yaml:       "providers:\n  - name: kvm2-home\n    type: kvm2\n",
+			wantErrSub: "kvm2 block is required",
+		},
+		{
+			name:       "missing sshKeyFile",
+			yaml:       minimal("knownHostsFile: /kh\n" + hostsAndTypes),
+			wantErrSub: "kvm2.sshKeyFile is required",
+		},
+		{
+			name:       "missing knownHostsFile without insecure flag",
+			yaml:       minimal("sshKeyFile: /key\n" + hostsAndTypes),
+			wantErrSub: "kvm2.knownHostsFile is required unless kvm2.insecureIgnoreHostKey is true",
+		},
+		{
+			name:       "no machine types",
+			yaml:       minimal("sshKeyFile: /key\nknownHostsFile: /kh\nhosts:\n  - {name: nas1, address: nas1:22, user: virt, bridge: br0}\n"),
+			wantErrSub: "kvm2.machineTypes must define at least one",
+		},
+		{
+			name:       "no hosts",
+			yaml:       minimal("sshKeyFile: /key\nknownHostsFile: /kh\nmachineTypes:\n  small: {vcpus: 1, memoryMiB: 1024}\n"),
+			wantErrSub: "kvm2.hosts must list at least one host",
+		},
+		{
+			name:       "duplicate host name",
+			yaml:       minimal("sshKeyFile: /key\nknownHostsFile: /kh\nmachineTypes:\n  small: {vcpus: 1, memoryMiB: 1024}\nhosts:\n  - {name: nas1, address: a:22, user: virt, bridge: br0}\n  - {name: nas1, address: b:22, user: virt, bridge: br0}\n"),
+			wantErrSub: `kvm2.hosts[1]: duplicate host name "nas1"`,
+		},
+		{
+			name:       "image key with a slash",
+			yaml:       minimal("sshKeyFile: /key\nknownHostsFile: /kh\nimages:\n  ../etc/d: {url: https://x/y.qcow2}\n" + hostsAndTypes),
+			wantErrSub: `kvm2.images["../etc/d"]: key must be letters, digits, dots, dashes or underscores`,
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			_, err := LoadConfig([]byte(tc.yaml))
+			if err == nil {
+				t.Fatalf("LoadConfig() error = nil, want error containing %q", tc.wantErrSub)
+			}
+			if !strings.Contains(err.Error(), tc.wantErrSub) {
+				t.Errorf("LoadConfig() error = %q, want substring %q", err.Error(), tc.wantErrSub)
+			}
+		})
+	}
+}
+
 func indent(s string, n int) string {
 	pad := strings.Repeat(" ", n)
 	lines := strings.Split(strings.TrimLeft(s, "\n"), "\n")
@@ -118,6 +118,26 @@ providers:
      image: custom-image`,
 			wantErrSub: "type is gcp but a kubernetes block is set",
 		},
+		{
+			name: "kvm type with kvm2 block",
+			yaml: `
+providers:
+  - name: p1
+    type: kvm
+    kvm2:
+      sshKeyFile: /key`,
+			wantErrSub: "type is kvm but a kvm2 block is set",
+		},
+		{
+			name: "kvm2 type with kvm block",
+			yaml: `
+providers:
+  - name: p1
+    type: kvm2
+    kvm:
+      sshKeyFile: /key`,
+			wantErrSub: "type is kvm2 but a kvm block is set",
+		},
 		{
 			name: "strict mode rejects unknown top-level key",
 			yaml: `
@@ -0,0 +1,421 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"sync"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// backend is the seam over go-libvirt (spec entry 17): exactly the calls the
+// provider uses, each with ctx first, since go-libvirt's own methods take
+// none and block. The production adapter is libvirtBackend; tests use an
+// in-memory fake. Nothing outside this package sees it.
+type backend interface {
+	DomainLookupByName(ctx context.Context, name string) (libvirt.Domain, error)
+	DomainDefineXML(ctx context.Context, xml string) (libvirt.Domain, error)
+	DomainCreate(ctx context.Context, dom libvirt.Domain) error
+	DomainDestroy(ctx context.Context, dom libvirt.Domain) error
+	DomainUndefineFlags(ctx context.Context, dom libvirt.Domain, flags libvirt.DomainUndefineFlagsValues) error
+	DomainGetState(ctx context.Context, dom libvirt.Domain, flags uint32) (state int32, reason int32, err error)
+	DomainGetMetadata(ctx context.Context, dom libvirt.Domain, typ int32, uri libvirt.OptString, flags libvirt.DomainModificationImpact) (string, error)
+	DomainInterfaceAddresses(ctx context.Context, dom libvirt.Domain, source uint32, flags uint32) ([]libvirt.DomainInterface, error)
+	ConnectListAllDomains(ctx context.Context, needResults int32, flags libvirt.ConnectListAllDomainsFlags) ([]libvirt.Domain, uint32, error)
+	StoragePoolLookupByName(ctx context.Context, name string) (libvirt.StoragePool, error)
+	StoragePoolRefresh(ctx context.Context, pool libvirt.StoragePool, flags uint32) error
+	StoragePoolGetInfo(ctx context.Context, pool libvirt.StoragePool) (state uint8, capacity, allocation, available uint64, err error)
+	StorageVolLookupByName(ctx context.Context, pool libvirt.StoragePool, name string) (libvirt.StorageVol, error)
+	StorageVolGetPath(ctx context.Context, vol libvirt.StorageVol) (string, error)
+	StorageVolCreateXML(ctx context.Context, pool libvirt.StoragePool, xml string, flags libvirt.StorageVolCreateFlags) (libvirt.StorageVol, error)
+	StorageVolUpload(ctx context.Context, vol libvirt.StorageVol, r io.Reader, offset, length uint64, flags libvirt.StorageVolUploadFlags) error
+	StorageVolDownload(ctx context.Context, vol libvirt.StorageVol, w io.Writer, offset, length uint64, flags libvirt.StorageVolDownloadFlags) error
+	StorageVolDelete(ctx context.Context, vol libvirt.StorageVol, flags libvirt.StorageVolDeleteFlags) error
+	ConnectGetLibVersion(ctx context.Context) (uint64, error)
+	ConnectGetHostname(ctx context.Context) (string, error)
+	Close() error
+}
+
+// liveness is implemented by backends whose connection can die underneath
+// them. The host table drops a backend whose dead channel is closed. It is
+// deliberately not part of backend: the fake needs it, the seam does not.
+type liveness interface {
+	dead() <-chan struct{}
+}
+
+func isDead(b backend) bool {
+	l, ok := b.(liveness)
+	if !ok {
+		return false
+	}
+	select {
+	case <-l.dead():
+		return true
+	default:
+		return false
+	}
+}
+
+// budgets are the per-call deadlines of spec entry 18, independent of the
+// caller's ctx, which by contract carries none.
+type budgets struct {
+	// dial bounds TCP connect, SSH handshake and libvirt connect together.
+	dial time.Duration
+	// rpc bounds every ordinary call.
+	rpc time.Duration
+	// create bounds DomainCreate, which waits for QEMU to start.
+	create time.Duration
+	// seedUpload bounds a StorageVolUpload of at most smallUploadBytes.
+	seedUpload time.Duration
+	// transfer caps a large upload or a download; ctx may end it earlier.
+	transfer time.Duration
+}
+
+var defaultBudgets = budgets{
+	dial:       10 * time.Second,
+	rpc:        30 * time.Second,
+	create:     90 * time.Second,
+	seedUpload: 60 * time.Second,
+	transfer:   20 * time.Minute,
+}
+
+// smallUploadBytes separates a seed ISO (tens of KiB) from a base image
+// (hundreds of MiB) for the upload budget.
+const smallUploadBytes = 16 << 20
+
+// closeBudget bounds the courtesy ConnectClose RPC in Close.
+const closeBudget = 5 * time.Second
+
+// transferGrace keeps the image download's own deadline inside the transfer
+// budget. A stalled mirror then ends by its body failing, which the upload
+// seam turns into a clean end of stream, instead of by the budget killing
+// the socket, which go-libvirt's stream sender does not survive.
+const transferGrace = time.Minute
+
+func (b budgets) upload(length uint64) time.Duration {
+	if length <= smallUploadBytes {
+		return b.seedUpload
+	}
+	return b.transfer
+}
+
+// link is the liveness of one connection. dead is closed exactly once, by
+// the first of: an abandoned call, a keepalive failure, go-libvirt noticing
+// the socket went away, or Close. kill closes the transport, which is the
+// only way to unblock a go-libvirt call; go-libvirt then fails every other
+// call in flight on the connection with ErrInterrupted.
+type link struct {
+	label string
+	kill  func()
+
+	closed chan struct{}
+	mu     sync.Mutex
+	reason string
+}
+
+func newLink(label string, kill func()) *link {
+	return &link{label: label, kill: kill, closed: make(chan struct{})}
+}
+
+func (l *link) dead() <-chan struct{} { return l.closed }
+
+func (l *link) isDead() bool {
+	select {
+	case <-l.closed:
+		return true
+	default:
+		return false
+	}
+}
+
+func (l *link) deadReason() string {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	return l.reason
+}
+
+// markDead records why the connection is unusable, signals dead and then
+// closes the transport. Later calls are no-ops. Signalling before killing,
+// and killing outside the lock, is what keeps a kill that takes its time
+// from holding up the callers waiting on dead or the next markDead: a
+// transport close can block for as long as the kernel retransmits.
+func (l *link) markDead(reason string) {
+	l.mu.Lock()
+	if l.isDead() {
+		l.mu.Unlock()
+		return
+	}
+	l.reason = reason
+	close(l.closed)
+	l.mu.Unlock()
+	if l.kill != nil {
+		l.kill()
+	}
+}
+
+// call executes a blocking go-libvirt call with a budget and returns its
+// result, or Transient when the link is already dead, ctx ends, the budget
+// runs out, or another caller kills the link meanwhile. Abandoning a call
+// kills the link (spec entry 18); the goroutine running fn then returns on
+// its own once go-libvirt sees the closed socket, and its result is dropped.
+// Results travel through the channel so an abandoned call never writes
+// into its caller's frame.
+func call[T any](ctx context.Context, l *link, name string, budget time.Duration, fn func() (T, error)) (T, error) {
+	var zero T
+	if l.isDead() {
+		return zero, transientf("libvirt call %s on %s: connection is closed: %s", name, l.label, l.deadReason())
+	}
+	if err := ctx.Err(); err != nil {
+		return zero, fmt.Errorf("%w: libvirt call %s on %s not started: %w", provider.ErrTransient, name, l.label, err)
+	}
+	type result struct {
+		v   T
+		err error
+	}
+	done := make(chan result, 1)
+	go func() {
+		v, err := fn()
+		done <- result{v, err}
+	}()
+	timer := time.NewTimer(budget)
+	defer timer.Stop()
+	select {
+	case r := <-done:
+		return r.v, r.err
+	case <-ctx.Done():
+		l.markDead(fmt.Sprintf("call %s abandoned: %v", name, ctx.Err()))
+		return zero, fmt.Errorf("%w: libvirt call %s on %s abandoned: %w", provider.ErrTransient, name, l.label, ctx.Err())
+	case <-timer.C:
+		l.markDead(fmt.Sprintf("call %s exceeded its %s budget", name, budget))
+		return zero, fmt.Errorf("%w: libvirt call %s on %s exceeded its %s budget: %w", provider.ErrTransient, name, l.label, budget, context.DeadlineExceeded)
+	case <-l.closed:
+		return zero, transientf("libvirt call %s on %s: connection closed while waiting: %s", name, l.label, l.deadReason())
+	}
+}
+
+// run is call for go-libvirt methods that return only an error.
+func (l *link) run(ctx context.Context, name string, budget time.Duration, fn func() error) error {
+	_, err := call(ctx, l, name, budget, func() (struct{}, error) { return struct{}{}, fn() })
+	return err
+}
+
+// libvirtBackend is the production adapter: a go-libvirt client over one
+// forwarded socket, with the budgets applied around every call. It never
+// logs; the observe decorator does.
+type libvirtBackend struct {
+	lv      *libvirt.Libvirt
+	link    *link
+	budgets budgets
+
+	closeOnce sync.Once
+	closeErr  error
+}
+
+var _ backend = (*libvirtBackend)(nil)
+
+// newLibvirtBackend wraps lv, whose socket runs over transport. Closing the
+// transport is how a hung call is abandoned, so it must close the socket
+// conn given to libvirt.New as well as whatever carries it.
+func newLibvirtBackend(lv *libvirt.Libvirt, label string, transport io.Closer, b budgets) *libvirtBackend {
+	return &libvirtBackend{
+		lv:      lv,
+		link:    newLink(label, func() { _ = transport.Close() }),
+		budgets: b,
+	}
+}
+
+func (b *libvirtBackend) dead() <-chan struct{} { return b.link.dead() }
+
+// connect runs the libvirt handshake under the dial budget and, once it
+// holds, drops the link as soon as go-libvirt reports the socket gone.
+func (b *libvirtBackend) connect(ctx context.Context) error {
+	if err := b.link.run(ctx, "Connect", b.budgets.dial, b.lv.Connect); err != nil {
+		return err
+	}
+	disconnected := b.lv.Disconnected()
+	go func() {
+		select {
+		case <-disconnected:
+			b.link.markDead("connection lost")
+		case <-b.link.closed:
+		}
+	}()
+	return nil
+}
+
+func (b *libvirtBackend) DomainLookupByName(ctx context.Context, name string) (libvirt.Domain, error) {
+	return call(ctx, b.link, "DomainLookupByName", b.budgets.rpc, func() (libvirt.Domain, error) { return b.lv.DomainLookupByName(name) })
+}
+
+func (b *libvirtBackend) DomainDefineXML(ctx context.Context, xml string) (libvirt.Domain, error) {
+	return call(ctx, b.link, "DomainDefineXML", b.budgets.rpc, func() (libvirt.Domain, error) { return b.lv.DomainDefineXML(xml) })
+}
+
+func (b *libvirtBackend) DomainCreate(ctx context.Context, dom libvirt.Domain) error {
+	return b.link.run(ctx, "DomainCreate", b.budgets.create, func() error { return b.lv.DomainCreate(dom) })
+}
+
+func (b *libvirtBackend) DomainDestroy(ctx context.Context, dom libvirt.Domain) error {
+	return b.link.run(ctx, "DomainDestroy", b.budgets.rpc, func() error { return b.lv.DomainDestroy(dom) })
+}
+
+func (b *libvirtBackend) DomainUndefineFlags(ctx context.Context, dom libvirt.Domain, flags libvirt.DomainUndefineFlagsValues) error {
+	return b.link.run(ctx, "DomainUndefineFlags", b.budgets.rpc, func() error { return b.lv.DomainUndefineFlags(dom, flags) })
+}
+
+func (b *libvirtBackend) DomainGetState(ctx context.Context, dom libvirt.Domain, flags uint32) (int32, int32, error) {
+	type stateReason struct{ state, reason int32 }
+	r, err := call(ctx, b.link, "DomainGetState", b.budgets.rpc, func() (stateReason, error) {
+		state, reason, err := b.lv.DomainGetState(dom, flags)
+		return stateReason{state, reason}, err
+	})
+	return r.state, r.reason, err
+}
+
+func (b *libvirtBackend) DomainGetMetadata(ctx context.Context, dom libvirt.Domain, typ int32, uri libvirt.OptString, flags libvirt.DomainModificationImpact) (string, error) {
+	return call(ctx, b.link, "DomainGetMetadata", b.budgets.rpc, func() (string, error) { return b.lv.DomainGetMetadata(dom, typ, uri, flags) })
+}
+
+func (b *libvirtBackend) DomainInterfaceAddresses(ctx context.Context, dom libvirt.Domain, source uint32, flags uint32) ([]libvirt.DomainInterface, error) {
+	return call(ctx, b.link, "DomainInterfaceAddresses", b.budgets.rpc, func() ([]libvirt.DomainInterface, error) {
+		return b.lv.DomainInterfaceAddresses(dom, source, flags)
+	})
+}
+
+func (b *libvirtBackend) ConnectListAllDomains(ctx context.Context, needResults int32, flags libvirt.ConnectListAllDomainsFlags) ([]libvirt.Domain, uint32, error) {
+	type listed struct {
+		doms []libvirt.Domain
+		n    uint32
+	}
+	r, err := call(ctx, b.link, "ConnectListAllDomains", b.budgets.rpc, func() (listed, error) {
+		doms, n, err := b.lv.ConnectListAllDomains(needResults, flags)
+		return listed{doms, n}, err
+	})
+	return r.doms, r.n, err
+}
+
+func (b *libvirtBackend) StoragePoolLookupByName(ctx context.Context, name string) (libvirt.StoragePool, error) {
+	return call(ctx, b.link, "StoragePoolLookupByName", b.budgets.rpc, func() (libvirt.StoragePool, error) { return b.lv.StoragePoolLookupByName(name) })
+}
+
+func (b *libvirtBackend) StoragePoolRefresh(ctx context.Context, pool libvirt.StoragePool, flags uint32) error {
+	return b.link.run(ctx, "StoragePoolRefresh", b.budgets.rpc, func() error { return b.lv.StoragePoolRefresh(pool, flags) })
+}
+
+func (b *libvirtBackend) StoragePoolGetInfo(ctx context.Context, pool libvirt.StoragePool) (uint8, uint64, uint64, uint64, error) {
+	type info struct {
+		state                           uint8
+		capacity, allocation, available uint64
+	}
+	r, err := call(ctx, b.link, "StoragePoolGetInfo", b.budgets.rpc, func() (info, error) {
+		state, capacity, allocation, available, err := b.lv.StoragePoolGetInfo(pool)
+		return info{state, capacity, allocation, available}, err
+	})
+	return r.state, r.capacity, r.allocation, r.available, err
+}
+
+func (b *libvirtBackend) StorageVolLookupByName(ctx context.Context, pool libvirt.StoragePool, name string) (libvirt.StorageVol, error) {
+	return call(ctx, b.link, "StorageVolLookupByName", b.budgets.rpc, func() (libvirt.StorageVol, error) { return b.lv.StorageVolLookupByName(pool, name) })
+}
+
+func (b *libvirtBackend) StorageVolGetPath(ctx context.Context, vol libvirt.StorageVol) (string, error) {
+	return call(ctx, b.link, "StorageVolGetPath", b.budgets.rpc, func() (string, error) { return b.lv.StorageVolGetPath(vol) })
+}
+
+func (b *libvirtBackend) StorageVolCreateXML(ctx context.Context, pool libvirt.StoragePool, xml string, flags libvirt.StorageVolCreateFlags) (libvirt.StorageVol, error) {
+	return call(ctx, b.link, "StorageVolCreateXML", b.budgets.rpc, func() (libvirt.StorageVol, error) { return b.lv.StorageVolCreateXML(pool, xml, flags) })
+}
+
+func (b *libvirtBackend) StorageVolUpload(ctx context.Context, vol libvirt.StorageVol, r io.Reader, offset, length uint64, flags libvirt.StorageVolUploadFlags) error {
+	src := &eofReader{r: r}
+	err := b.link.run(ctx, "StorageVolUpload", b.budgets.upload(length), func() error {
+		return b.lv.StorageVolUpload(vol, src, offset, length, flags)
+	})
+	if err == nil {
+		if srcErr := src.sourceErr(); srcErr != nil {
+			return fmt.Errorf("reading the upload source for %s: %w", vol.Name, srcErr)
+		}
+	}
+	return err
+}
+
+// eofReader hides a source error from go-libvirt's stream sender and keeps
+// it for the caller. SendStream answers a non-EOF error with a StatusError
+// packet and then blocks forever handing its result to an unbuffered
+// channel that requestStream no longer reads (go-libvirt rpc.go), which
+// wedges the call and leaks the goroutine along with its 4 MiB buffer;
+// closing the socket does not help, because both ends are parked on
+// channel sends rather than on I/O. io.EOF is the one outcome that ends
+// the stream the normal way. The field is guarded because an abandoned
+// call returns while go-libvirt is still reading.
+type eofReader struct {
+	r   io.Reader
+	mu  sync.Mutex
+	err error
+}
+
+func (e *eofReader) Read(b []byte) (int, error) {
+	n, err := e.r.Read(b)
+	if err == nil || errors.Is(err, io.EOF) {
+		return n, err
+	}
+	e.mu.Lock()
+	if e.err == nil {
+		e.err = err
+	}
+	e.mu.Unlock()
+	return n, io.EOF
+}
+
+func (e *eofReader) sourceErr() error {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	return e.err
+}
+
+func (b *libvirtBackend) StorageVolDownload(ctx context.Context, vol libvirt.StorageVol, w io.Writer, offset, length uint64, flags libvirt.StorageVolDownloadFlags) error {
+	return b.link.run(ctx, "StorageVolDownload", b.budgets.transfer, func() error {
+		return b.lv.StorageVolDownload(vol, w, offset, length, flags)
+	})
+}
+
+func (b *libvirtBackend) StorageVolDelete(ctx context.Context, vol libvirt.StorageVol, flags libvirt.StorageVolDeleteFlags) error {
+	return b.link.run(ctx, "StorageVolDelete", b.budgets.rpc, func() error { return b.lv.StorageVolDelete(vol, flags) })
+}
+
+func (b *libvirtBackend) ConnectGetLibVersion(ctx context.Context) (uint64, error) {
+	return call(ctx, b.link, "ConnectGetLibVersion", b.budgets.rpc, b.lv.ConnectGetLibVersion)
+}
+
+func (b *libvirtBackend) ConnectGetHostname(ctx context.Context) (string, error) {
+	return call(ctx, b.link, "ConnectGetHostname", b.budgets.rpc, b.lv.ConnectGetHostname)
+}
+
+// Close sends libvirt's ConnectClose when the link is still live, bounded
+// because a hung host would otherwise pin the caller, then kills the link,
+// which closes the socket and the SSH client. Repeat calls return the first
+// result.
+func (b *libvirtBackend) Close() error {
+	b.closeOnce.Do(func() {
+		if b.link.isDead() {
+			return
+		}
+		done := make(chan error, 1)
+		go func() { done <- b.lv.Disconnect() }()
+		timer := time.NewTimer(closeBudget)
+		defer timer.Stop()
+		select {
+		case b.closeErr = <-done:
+		case <-timer.C:
+			b.closeErr = transientf("libvirt disconnect on %s exceeded its %s budget", b.link.label, closeBudget)
+		case <-b.link.closed:
+		}
+		b.link.markDead("closed")
+	})
+	return b.closeErr
+}
@@ -0,0 +1,369 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"io"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+	"github.com/digitalocean/go-libvirt/libvirttest"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// blockingFn returns a func that blocks until release is closed, plus a
+// counter of how often it was started.
+func blockingFn(t *testing.T) (fn func() error, started *atomic.Int32) {
+	t.Helper()
+	release := make(chan struct{})
+	t.Cleanup(func() { close(release) })
+	started = &atomic.Int32{}
+	return func() error {
+		started.Add(1)
+		<-release
+		return nil
+	}, started
+}
+
+func TestCall_deadlineClosesConnectionAndReturnsTransient(t *testing.T) {
+	t.Parallel()
+	var killed atomic.Int32
+	l := newLink("beelink", func() { killed.Add(1) })
+	fn, started := blockingFn(t)
+
+	start := time.Now()
+	err := l.run(context.Background(), "DomainCreate", 50*time.Millisecond, fn)
+	elapsed := time.Since(start)
+	if !errors.Is(err, provider.ErrTransient) {
+		t.Fatalf("run() error = %v, want ErrTransient", err)
+	}
+	if !errors.Is(err, context.DeadlineExceeded) {
+		t.Errorf("run() error = %v, want it to wrap context.DeadlineExceeded", err)
+	}
+	if !strings.Contains(err.Error(), "DomainCreate") || !strings.Contains(err.Error(), "beelink") {
+		t.Errorf("run() error = %q, want the call name and host in it", err)
+	}
+	if elapsed > time.Second {
+		t.Errorf("run() took %v, want about the 50ms budget", elapsed)
+	}
+	if killed.Load() != 1 {
+		t.Errorf("kill hook called %d times, want 1", killed.Load())
+	}
+	if !l.isDead() {
+		t.Error("link not dead after an abandoned call")
+	}
+	if !strings.Contains(l.deadReason(), "budget") {
+		t.Errorf("deadReason() = %q, want the budget mentioned", l.deadReason())
+	}
+
+	start = time.Now()
+	err = l.run(context.Background(), "DomainGetState", time.Minute, fn)
+	if !errors.Is(err, provider.ErrTransient) {
+		t.Fatalf("second run() error = %v, want ErrTransient", err)
+	}
+	if time.Since(start) > 100*time.Millisecond {
+		t.Errorf("second run() took %v, want immediate", time.Since(start))
+	}
+	if started.Load() != 1 {
+		t.Errorf("fn started %d times, want 1: a dead link must not touch the socket", started.Load())
+	}
+	if killed.Load() != 1 {
+		t.Errorf("kill hook called %d times, want still 1", killed.Load())
+	}
+}
+
+func TestCall_cancelUnblocksRPC(t *testing.T) {
+	t.Parallel()
+	var killed atomic.Int32
+	l := newLink("beelink", func() { killed.Add(1) })
+	fn, _ := blockingFn(t)
+	ctx, cancel := context.WithCancel(context.Background())
+	go func() {
+		time.Sleep(10 * time.Millisecond)
+		cancel()
+	}()
+
+	start := time.Now()
+	err := l.run(ctx, "ConnectListAllDomains", time.Minute, fn)
+	if time.Since(start) > 100*time.Millisecond {
+		t.Errorf("run() took %v after cancel, want under 100ms", time.Since(start))
+	}
+	if !errors.Is(err, provider.ErrTransient) || !errors.Is(err, context.Canceled) {
+		t.Errorf("run() error = %v, want ErrTransient wrapping context.Canceled", err)
+	}
+	if killed.Load() != 1 {
+		t.Errorf("kill hook called %d times, want 1", killed.Load())
+	}
+}
+
+func TestCall_alreadyCancelledDoesNotStart(t *testing.T) {
+	t.Parallel()
+	l := newLink("beelink", nil)
+	fn, started := blockingFn(t)
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	err := l.run(ctx, "DomainGetState", time.Minute, fn)
+	if !errors.Is(err, provider.ErrTransient) || !errors.Is(err, context.Canceled) {
+		t.Errorf("run() error = %v, want ErrTransient wrapping context.Canceled", err)
+	}
+	if started.Load() != 0 {
+		t.Error("fn started although ctx was already cancelled")
+	}
+	if l.isDead() {
+		t.Error("link marked dead although nothing was abandoned")
+	}
+}
+
+func TestCall_returnsFnResult(t *testing.T) {
+	t.Parallel()
+	l := newLink("beelink", nil)
+	want := libvirt.Error{Code: uint32(libvirt.ErrNoDomain), Message: "gone"}
+	err := l.run(context.Background(), "DomainLookupByName", time.Second, func() error { return want })
+	if !hasCode(err, libvirt.ErrNoDomain) {
+		t.Errorf("run() error = %v, want the libvirt error through unchanged", err)
+	}
+	if l.isDead() {
+		t.Error("a backend error must not kill the link")
+	}
+}
+
+func TestCall_killedByAnotherCallerFailsTransient(t *testing.T) {
+	t.Parallel()
+	l := newLink("beelink", nil)
+	fn, _ := blockingFn(t)
+	done := make(chan error, 1)
+	go func() { done <- l.run(context.Background(), "DomainGetState", time.Minute, fn) }()
+	time.Sleep(10 * time.Millisecond)
+	l.markDead("keepalive failed")
+	select {
+	case err := <-done:
+		if !errors.Is(err, provider.ErrTransient) || !strings.Contains(err.Error(), "keepalive failed") {
+			t.Errorf("run() error = %v, want ErrTransient with the death reason", err)
+		}
+	case <-time.After(time.Second):
+		t.Fatal("run() did not return after the link was killed")
+	}
+}
+
+// TestMarkDead_slowKillDoesNotBlockCallers pins the guarantee the pooled
+// connection rests on: a kill that takes its time — closing a transport can
+// block for as long as the kernel retransmits to a host that went away —
+// holds up neither the callers waiting on dead nor the next markDead. The
+// call's own fn signals when it is in flight, so the parked caller is
+// genuinely parked rather than refused by the dead check on the way in.
+func TestMarkDead_slowKillDoesNotBlockCallers(t *testing.T) {
+	t.Parallel()
+	release := make(chan struct{})
+	t.Cleanup(func() { close(release) })
+	killing := make(chan struct{})
+	var killed atomic.Int32
+	l := newLink("beelink", func() {
+		killed.Add(1)
+		close(killing)
+		<-release
+	})
+
+	inflight := make(chan struct{})
+	waiting := make(chan error, 1)
+	go func() {
+		waiting <- l.run(context.Background(), "StorageVolUpload", time.Minute, func() error {
+			close(inflight)
+			<-release
+			return nil
+		})
+	}()
+	<-inflight
+
+	go l.markDead("keepalive failed")
+	<-killing
+
+	select {
+	case <-l.dead():
+	case <-time.After(time.Second):
+		t.Fatal("link not signalled dead while its kill is still running")
+	}
+	select {
+	case err := <-waiting:
+		if !errors.Is(err, provider.ErrTransient) || !strings.Contains(err.Error(), "keepalive failed") {
+			t.Errorf("parked call error = %v, want ErrTransient with the death reason", err)
+		}
+	case <-time.After(time.Second):
+		t.Fatal("parked call not released while the kill is still running")
+	}
+
+	second := make(chan struct{})
+	go func() {
+		l.markDead("closed")
+		close(second)
+	}()
+	select {
+	case <-second:
+	case <-time.After(time.Second):
+		t.Fatal("second markDead blocked behind the running kill")
+	}
+
+	if killed.Load() != 1 {
+		t.Errorf("kill hook called %d times, want 1", killed.Load())
+	}
+	if !strings.Contains(l.deadReason(), "keepalive failed") {
+		t.Errorf("deadReason() = %q, want the first reason", l.deadReason())
+	}
+}
+
+// TestEOFReader_sourceErrorNeverReachesTheStream pins the one property the
+// upload seam needs: go-libvirt's stream sender must never see an error
+// other than io.EOF, because it answers one by parking forever on a channel
+// nothing reads. The real error has to survive for the caller.
+func TestEOFReader_sourceErrorNeverReachesTheStream(t *testing.T) {
+	t.Parallel()
+	boom := errors.New("connection reset by peer")
+	tests := []struct {
+		name    string
+		reader  io.Reader
+		wantN   int
+		wantErr error
+	}{
+		{name: "error after data", reader: io.MultiReader(strings.NewReader("abc"), errReader{boom}), wantN: 3, wantErr: boom},
+		{name: "error with no data", reader: errReader{boom}, wantErr: boom},
+		{name: "clean eof passes through", reader: strings.NewReader("abc"), wantN: 3},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			e := &eofReader{r: tc.reader}
+			n, err := io.Copy(io.Discard, e)
+			if err != nil {
+				t.Fatalf("io.Copy through eofReader = %v, want nil: the stream sender must only ever see io.EOF", err)
+			}
+			if n != int64(tc.wantN) {
+				t.Errorf("copied %d bytes, want %d", n, tc.wantN)
+			}
+			if !errors.Is(e.sourceErr(), tc.wantErr) {
+				t.Errorf("sourceErr() = %v, want %v", e.sourceErr(), tc.wantErr)
+			}
+		})
+	}
+}
+
+func TestEOFReader_keepsTheFirstError(t *testing.T) {
+	t.Parallel()
+	first, second := errors.New("first"), errors.New("second")
+	e := &eofReader{r: io.MultiReader(errReader{first}, errReader{second})}
+	if _, err := io.Copy(io.Discard, e); err != nil {
+		t.Fatalf("io.Copy through eofReader = %v, want nil", err)
+	}
+	if !errors.Is(e.sourceErr(), first) {
+		t.Errorf("sourceErr() = %v, want the first error", e.sourceErr())
+	}
+}
+
+type errReader struct{ err error }
+
+func (r errReader) Read([]byte) (int, error) { return 0, r.err }
+
+func TestBudgets_upload(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name   string
+		length uint64
+		want   time.Duration
+	}{
+		{name: "seed sized", length: 400 << 10, want: defaultBudgets.seedUpload},
+		{name: "exactly the limit", length: smallUploadBytes, want: defaultBudgets.seedUpload},
+		{name: "one over the limit", length: smallUploadBytes + 1, want: defaultBudgets.transfer},
+		{name: "base image", length: 435 << 20, want: defaultBudgets.transfer},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			if got := defaultBudgets.upload(tc.length); got != tc.want {
+				t.Errorf("upload(%d) = %v, want %v", tc.length, got, tc.want)
+			}
+		})
+	}
+}
+
+// mockAdapter wires a libvirtBackend to go-libvirt's own mock server over a
+// net.Pipe, so the adapter is exercised with the real RPC client.
+func mockAdapter(t *testing.T, b budgets) *libvirtBackend {
+	t.Helper()
+	m := libvirttest.New()
+	conn, err := m.Dial()
+	if err != nil {
+		t.Fatalf("mock dial: %v", err)
+	}
+	ad := newLibvirtBackend(libvirt.New(conn), "mock", conn, b)
+	if err := ad.connect(context.Background()); err != nil {
+		t.Fatalf("connect: %v", err)
+	}
+	t.Cleanup(func() { _ = ad.Close() })
+	return ad
+}
+
+func TestLibvirtBackend_roundTripThroughMock(t *testing.T) {
+	t.Parallel()
+	ad := mockAdapter(t, defaultBudgets)
+	v, err := ad.ConnectGetLibVersion(context.Background())
+	if err != nil {
+		t.Fatalf("ConnectGetLibVersion() error = %v", err)
+	}
+	if v == 0 {
+		t.Error("ConnectGetLibVersion() = 0, want the mock's version")
+	}
+	pool, err := ad.StoragePoolLookupByName(context.Background(), "default")
+	if err != nil {
+		t.Fatalf("StoragePoolLookupByName() error = %v", err)
+	}
+	if pool.Name == "" {
+		t.Error("StoragePoolLookupByName() returned an empty pool")
+	}
+	if isDead(ad) {
+		t.Error("adapter dead after successful calls")
+	}
+}
+
+func TestLibvirtBackend_unansweredCallIsAbandoned(t *testing.T) {
+	t.Parallel()
+	b := defaultBudgets
+	b.rpc = 100 * time.Millisecond
+	ad := mockAdapter(t, b)
+	// The mock does not implement ConnectGetHostname, so it never replies.
+	_, err := ad.ConnectGetHostname(context.Background())
+	if !errors.Is(err, provider.ErrTransient) || !errors.Is(err, context.DeadlineExceeded) {
+		t.Fatalf("ConnectGetHostname() error = %v, want ErrTransient wrapping DeadlineExceeded", err)
+	}
+	if !isDead(ad) {
+		t.Error("adapter not dead after an abandoned call")
+	}
+	select {
+	case <-ad.lv.Disconnected():
+	case <-time.After(time.Second):
+		t.Error("go-libvirt did not notice the closed socket")
+	}
+	if _, err := ad.ConnectGetLibVersion(context.Background()); !errors.Is(err, provider.ErrTransient) {
+		t.Errorf("call on dead adapter error = %v, want ErrTransient", err)
+	}
+	if err := ad.Close(); err != nil {
+		t.Errorf("Close() on dead adapter error = %v, want nil", err)
+	}
+}
+
+func TestLibvirtBackend_closeIsIdempotent(t *testing.T) {
+	t.Parallel()
+	ad := mockAdapter(t, defaultBudgets)
+	first := ad.Close()
+	second := ad.Close()
+	if first != nil || second != nil {
+		t.Errorf("Close() errors = %v, %v, want nil, nil", first, second)
+	}
+	if !isDead(ad) {
+		t.Error("adapter not dead after Close")
+	}
+	if _, err := ad.ConnectGetLibVersion(context.Background()); !errors.Is(err, provider.ErrTransient) {
+		t.Errorf("call after Close error = %v, want ErrTransient", err)
+	}
+}
@@ -0,0 +1,189 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+	"golang.org/x/crypto/ssh/knownhosts"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+var _ provider.Checker = (*Provider)(nil)
+
+// Reasons only this provider can tell apart (spec entry 16). Each names a
+// failure an admin fixes somewhere other than where the shared reasons
+// point.
+const (
+	// reasonCredentialsUnavailable: the key or known_hosts file is missing
+	// or unparsable; fix the Secret in the cluster, not the host.
+	reasonCredentialsUnavailable = "CredentialsUnavailable"
+	// reasonPoolNotFound: the configured pool is not defined on the host;
+	// compare with `virsh pool-list --all`.
+	reasonPoolNotFound = "PoolNotFound"
+	// reasonPoolNotActive: the pool exists but is not running; `virsh
+	// pool-start <pool>` (and autostart) brings it back.
+	reasonPoolNotActive = "PoolNotActive"
+)
+
+// Check verifies every configured host concurrently, each on a fresh
+// dedicated connection: dial, libvirt version, hostname, pool lookup, pool
+// info, close. The pooled connection is never touched, so a check that runs
+// out of time closes only its own connection and cannot fail a reconcile's
+// call in flight. Every target gets a verdict, in config order.
+func (p *Provider) Check(ctx context.Context) []provider.TargetStatus {
+	out := make([]provider.TargetStatus, len(p.cfg.Hosts))
+	var wg sync.WaitGroup
+	for i, host := range p.cfg.Hosts {
+		wg.Go(func() { out[i] = p.checkHost(ctx, host) })
+	}
+	wg.Wait()
+	return out
+}
+
+func (p *Provider) checkHost(ctx context.Context, host provider.KVMHost) provider.TargetStatus {
+	start := time.Now()
+	log := p.logger(ctx).WithValues("host", host.Name)
+	ts := provider.TargetStatus{
+		Name:    host.Name,
+		Address: host.Address,
+		Details: map[string]string{"user": host.User, "pool": host.Pool},
+	}
+	err := p.probeHost(ctx, host, ts.Details)
+	ts.Latency = time.Since(start)
+	switch {
+	case err == nil:
+		ts.Ready = true
+		ts.Reason = provider.ReasonReachable
+	default:
+		var inactive *poolNotActiveError
+		switch {
+		case ctx.Err() != nil:
+			// The round ran out of time, whichever stage was running: the
+			// error's shape (a dial that saw the ctx end, an abandoned
+			// call) says nothing the admin should act on.
+			ts.Reason = provider.ReasonTimeout
+		case errors.As(err, &inactive):
+			ts.Reason = reasonPoolNotActive
+		default:
+			ts.Reason = checkReason(err)
+		}
+		ts.Message = err.Error()
+	}
+	log.V(1).Info("kvm2 host checked", "ready", ts.Ready, "reason", ts.Reason, "latency", ts.Latency)
+	return ts
+}
+
+// probeHost runs the check sequence on its own connection and fills details
+// as facts arrive, so a failure late in the sequence still reports what the
+// host said before it.
+func (p *Provider) probeHost(ctx context.Context, host provider.KVMHost, details map[string]string) error {
+	b, err := p.hosts.dialFresh(ctx, host)
+	if err != nil {
+		return err
+	}
+	defer func() { _ = b.Close() }()
+
+	version, err := b.ConnectGetLibVersion(ctx)
+	if err != nil {
+		return err
+	}
+	details["libvirtVersion"] = formatLibVersion(version)
+	hostname, err := b.ConnectGetHostname(ctx)
+	if err != nil {
+		return err
+	}
+	details["hostname"] = hostname
+	pool, err := b.StoragePoolLookupByName(ctx, host.Pool)
+	if err != nil {
+		return err
+	}
+	state, capacity, _, available, err := b.StoragePoolGetInfo(ctx, pool)
+	if err != nil {
+		return err
+	}
+	details["poolState"] = poolStateName(state)
+	details["poolCapacityGiB"] = formatGiB(capacity)
+	details["poolAvailableGiB"] = formatGiB(available)
+	if libvirt.StoragePoolState(state) != libvirt.StoragePoolRunning {
+		return &poolNotActiveError{pool: host.Pool, state: poolStateName(state)}
+	}
+	return nil
+}
+
+// poolNotActiveError is the one verdict that is not an error from the
+// backend: the pool answered, but is not usable.
+type poolNotActiveError struct {
+	pool  string
+	state string
+}
+
+func (e *poolNotActiveError) Error() string {
+	return fmt.Sprintf("storage pool %q is %s", e.pool, e.state)
+}
+
+// checkReason maps a failed probe to a reason once checkHost has ruled out
+// the round's ctx ending. A dial failure is judged by its stage, so a TCP
+// connect or an SSH handshake that ran out of the dial budget on its own is
+// Unreachable, not Timeout: the host did not answer. Timeout is for a
+// budget expiring once the connection was up.
+func checkReason(err error) string {
+	var de *dialError
+	if errors.As(err, &de) {
+		switch de.stage {
+		case stageCredentials:
+			return reasonCredentialsUnavailable
+		case stageSSH:
+			var keyErr *knownhosts.KeyError
+			if errors.As(de.err, &keyErr) || strings.Contains(de.err.Error(), sshAuthFailed) {
+				return provider.ReasonUnauthorized
+			}
+		}
+		return provider.ReasonUnreachable
+	}
+	switch {
+	case errors.Is(err, context.DeadlineExceeded), errors.Is(err, context.Canceled):
+		return provider.ReasonTimeout
+	case hasCode(err, libvirt.ErrAccessDenied), hasCode(err, libvirt.ErrAuthFailed):
+		return provider.ReasonForbidden
+	case hasCode(err, libvirt.ErrNoStoragePool):
+		return reasonPoolNotFound
+	}
+	if _, ok := libvirtCode(err); ok {
+		return provider.ReasonAPIError
+	}
+	return provider.ReasonUnreachable
+}
+
+// formatLibVersion decodes ConnectGetLibVersion's major*1_000_000 +
+// minor*1_000 + micro.
+func formatLibVersion(v uint64) string {
+	return fmt.Sprintf("%d.%d.%d", v/1_000_000, v/1_000%1_000, v%1_000)
+}
+
+// poolStateName uses virsh's words for a pool state.
+func poolStateName(state uint8) string {
+	switch libvirt.StoragePoolState(state) {
+	case libvirt.StoragePoolInactive:
+		return "inactive"
+	case libvirt.StoragePoolBuilding:
+		return "building"
+	case libvirt.StoragePoolRunning:
+		return "running"
+	case libvirt.StoragePoolDegraded:
+		return "degraded"
+	case libvirt.StoragePoolInaccessible:
+		return "inaccessible"
+	default:
+		return fmt.Sprintf("state %d", state)
+	}
+}
+
+func formatGiB(bytes uint64) string {
+	return fmt.Sprintf("%.1f", float64(bytes)/(1<<30))
+}
@@ -0,0 +1,579 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"slices"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+	"github.com/go-logr/logr/funcr"
+	"golang.org/x/crypto/ssh/knownhosts"
+	logf "sigs.k8s.io/controller-runtime/pkg/log"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// checkContext is captureContext with a sink that tolerates the concurrent
+// per-host log lines of Check.
+func checkContext() (context.Context, *logCapture) {
+	logs := &logCapture{}
+	log := funcr.New(func(prefix, args string) {
+		logs.mu.Lock()
+		defer logs.mu.Unlock()
+		logs.lines = append(logs.lines, prefix+" "+args)
+	}, funcr.Options{Verbosity: 2})
+	return logf.IntoContext(context.Background(), log), logs
+}
+
+// checkDialer hands out a distinct fake per host name, fails the hosts in
+// errs, counts every dial, and, with block set, waits for ctx to end and
+// then fails the way dialSSHSocket does when the TCP connect sees the ctx
+// end: a dialError at the tcp stage wrapping the ctx's error.
+type checkDialer struct {
+	backends map[string]*fakeBackend
+	errs     map[string]error
+	block    bool
+
+	dials atomic.Int32
+}
+
+func (d *checkDialer) dial(ctx context.Context, host provider.KVMHost) (backend, error) {
+	d.dials.Add(1)
+	if d.block {
+		<-ctx.Done()
+		return nil, &dialError{stage: stageTCP, host: host.Name, err: fmt.Errorf("dial tcp %s: %w", host.Address, ctx.Err())}
+	}
+	if err := d.errs[host.Name]; err != nil {
+		return nil, err
+	}
+	b, ok := d.backends[host.Name]
+	if !ok {
+		return nil, fmt.Errorf("checkDialer: no backend for %s", host.Name)
+	}
+	return b, nil
+}
+
+func threeHosts() provider.KVMConfig {
+	cfg := twoHosts()
+	cfg.Hosts = append(cfg.Hosts, provider.KVMHost{Name: "lab3", Address: "lab3:22", User: "ops", Bridge: "br0", Pool: "ssd"})
+	return cfg
+}
+
+func fakesFor(cfg provider.KVMConfig) map[string]*fakeBackend {
+	out := make(map[string]*fakeBackend, len(cfg.Hosts))
+	for _, h := range cfg.Hosts {
+		f := newFakeBackend()
+		f.hostname = h.Name + ".home.lan"
+		if h.Pool != "" {
+			f.pools[h.Pool] = newFakePool()
+		}
+		out[h.Name] = f
+	}
+	return out
+}
+
+func detailKeys(d map[string]string) []string {
+	keys := make([]string, 0, len(d))
+	for k := range d {
+		keys = append(keys, k)
+	}
+	slices.Sort(keys)
+	return keys
+}
+
+func TestCheck_oneEntryPerHostInConfigOrder(t *testing.T) {
+	t.Parallel()
+	cfg := threeHosts()
+	d := &checkDialer{backends: fakesFor(cfg), errs: map[string]error{"lab2": errors.New("dial tcp 192.0.2.2:22: connect: no route to host")}}
+	p := newProvider(testProvider, cfg, d)
+	ctx, _ := checkContext()
+	got := p.Check(ctx)
+	if len(got) != len(cfg.Hosts) {
+		t.Fatalf("Check() returned %d targets, want %d", len(got), len(cfg.Hosts))
+	}
+	for i, h := range cfg.Hosts {
+		ts := got[i]
+		if ts.Name != h.Name || ts.Address != h.Address {
+			t.Errorf("target %d = %s (%s), want %s (%s)", i, ts.Name, ts.Address, h.Name, h.Address)
+		}
+		if ts.Name == "" || ts.Address == "" {
+			t.Errorf("target %d has an empty name or address: %+v", i, ts)
+		}
+		wantReady := h.Name != "lab2"
+		if ts.Ready != wantReady {
+			t.Errorf("target %s ready = %v, want %v (%s: %s)", ts.Name, ts.Ready, wantReady, ts.Reason, ts.Message)
+		}
+	}
+	if got[1].Reason != provider.ReasonUnreachable || got[0].Reason != provider.ReasonReachable {
+		t.Errorf("reasons = %s, %s, %s", got[0].Reason, got[1].Reason, got[2].Reason)
+	}
+}
+
+func TestCheck_hostsCheckedConcurrently(t *testing.T) {
+	t.Parallel()
+	cfg := twoHosts()
+	fakes := fakesFor(cfg)
+	for _, f := range fakes {
+		release := f.blockOn("ConnectGetLibVersion")
+		time.AfterFunc(200*time.Millisecond, func() { close(release) })
+	}
+	p := newProvider(testProvider, cfg, &checkDialer{backends: fakes})
+	ctx, _ := checkContext()
+	start := time.Now()
+	got := p.Check(ctx)
+	elapsed := time.Since(start)
+	if elapsed > 350*time.Millisecond {
+		t.Errorf("Check() took %v, want under 350ms for two 200ms hosts", elapsed)
+	}
+	for _, ts := range got {
+		if !ts.Ready {
+			t.Errorf("target %s not ready: %s: %s", ts.Name, ts.Reason, ts.Message)
+		}
+		if ts.Latency < 200*time.Millisecond {
+			t.Errorf("target %s latency = %v, want at least the 200ms block", ts.Name, ts.Latency)
+		}
+	}
+}
+
+// TestCheck_expiredContextReturnsPromptly drives the production shape of a
+// round that ran out of time during the dial: the dialer fails at the tcp
+// stage wrapping context.DeadlineExceeded, which checkReason alone would
+// call Unreachable; the verdict must still be Timeout because the round's
+// ctx ended.
+func TestCheck_expiredContextReturnsPromptly(t *testing.T) {
+	t.Parallel()
+	cfg := twoHosts()
+	p := newProvider(testProvider, cfg, &checkDialer{block: true})
+	ctx, _ := checkContext()
+	ctx, cancel := context.WithTimeout(ctx, 0)
+	defer cancel()
+	start := time.Now()
+	got := p.Check(ctx)
+	if elapsed := time.Since(start); elapsed > time.Second {
+		t.Errorf("Check() took %v with an expired ctx, want under 1s", elapsed)
+	}
+	if len(got) != len(cfg.Hosts) {
+		t.Fatalf("Check() returned %d targets, want %d", len(got), len(cfg.Hosts))
+	}
+	for _, ts := range got {
+		if ts.Ready || ts.Reason != provider.ReasonTimeout {
+			t.Errorf("target %s = ready %v reason %s, want not ready, Timeout", ts.Name, ts.Ready, ts.Reason)
+		}
+		if !strings.Contains(ts.Message, context.DeadlineExceeded.Error()) || !strings.Contains(ts.Message, string(stageTCP)) {
+			t.Errorf("target %s message = %q, want the tcp stage and the deadline named", ts.Name, ts.Message)
+		}
+	}
+}
+
+// keyMaterial is the sentinel in the unusable key file of the credentials
+// row; it must never reach a message.
+const keyMaterial = "SENTINEL-PRIVATE-KEY-BYTES-9f3a"
+
+func TestCheck_reasonsTable(t *testing.T) {
+	t.Parallel()
+	sshHandshake := func(cause error) error {
+		return &dialError{stage: stageSSH, host: testHost, err: fmt.Errorf("ssh: handshake failed: %w", cause)}
+	}
+	tests := []struct {
+		name string
+		// dialErr builds the dial failure; nil means the dial succeeds.
+		dialErr func(t *testing.T) error
+		prep    func(f *fakeBackend)
+		timeout time.Duration
+		cancel  bool
+
+		wantReady   bool
+		wantReason  string
+		wantMessage string
+	}{
+		{
+			name: "credentials unusable",
+			dialErr: func(t *testing.T) error {
+				path := filepath.Join(t.TempDir(), "id_ed25519")
+				if err := os.WriteFile(path, []byte(keyMaterial), 0o600); err != nil {
+					t.Fatal(err)
+				}
+				_, err := loadCredentials(provider.KVMConfig{SSHKeyFile: path, KnownHostsFile: "/nonexistent/known_hosts"})
+				if err == nil {
+					t.Fatal("loadCredentials accepted the sentinel file")
+				}
+				return &dialError{stage: stageCredentials, host: testHost, err: err}
+			},
+			wantReason: reasonCredentialsUnavailable,
+		},
+		{
+			name: "credentials missing",
+			dialErr: func(*testing.T) error {
+				_, err := loadCredentials(provider.KVMConfig{SSHKeyFile: "/nonexistent/kvm2/id_ed25519"})
+				return &dialError{stage: stageCredentials, host: testHost, err: err}
+			},
+			wantReason: reasonCredentialsUnavailable,
+		},
+		{
+			name: "tcp refused",
+			dialErr: func(*testing.T) error {
+				return &dialError{stage: stageTCP, host: testHost, err: errors.New("dial tcp 192.0.2.1:22: connect: connection refused")}
+			},
+			wantReason: provider.ReasonUnreachable,
+		},
+		{
+			name: "tcp timed out on the dial budget but ctx is fine",
+			dialErr: func(*testing.T) error {
+				return &dialError{stage: stageTCP, host: testHost, err: fmt.Errorf("dial tcp 192.0.2.1:22: %w", context.DeadlineExceeded)}
+			},
+			wantReason: provider.ReasonUnreachable,
+		},
+		{
+			name:       "ssh host key unknown",
+			dialErr:    func(*testing.T) error { return sshHandshake(&knownhosts.KeyError{}) },
+			wantReason: provider.ReasonUnauthorized,
+		},
+		{
+			name: "ssh host key mismatch",
+			dialErr: func(*testing.T) error {
+				return sshHandshake(&knownhosts.KeyError{Want: []knownhosts.KnownKey{{Filename: "known_hosts", Line: 3}}})
+			},
+			wantReason: provider.ReasonUnauthorized,
+		},
+		{
+			name: "ssh authentication refused",
+			dialErr: func(*testing.T) error {
+				return sshHandshake(errors.New("ssh: unable to authenticate, attempted methods [none publickey], no supported methods remain"))
+			},
+			wantReason: provider.ReasonUnauthorized,
+		},
+		{
+			name:       "ssh handshake reset",
+			dialErr:    func(*testing.T) error { return sshHandshake(errors.New("read tcp: connection reset by peer")) },
+			wantReason: provider.ReasonUnreachable,
+		},
+		{
+			name:       "ssh handshake timed out on the dial budget but ctx is fine",
+			dialErr:    func(*testing.T) error { return sshHandshake(context.DeadlineExceeded) },
+			wantReason: provider.ReasonUnreachable,
+		},
+		{
+			name: "socket refused",
+			dialErr: func(*testing.T) error {
+				return &dialError{stage: stageSocket, host: testHost, err: errors.New("opening /var/run/libvirt/libvirt-sock: ssh: rejected: connect failed (No such file or directory)")}
+			},
+			wantReason:  provider.ReasonUnreachable,
+			wantMessage: "/var/run/libvirt/libvirt-sock",
+		},
+		{
+			name: "libvirt access denied",
+			prep: func(f *fakeBackend) {
+				f.failWith("ConnectGetLibVersion", fakeErr(libvirt.ErrAccessDenied, "access denied"))
+			},
+			wantReason: provider.ReasonForbidden,
+		},
+		{
+			name: "libvirt auth failed",
+			prep: func(f *fakeBackend) {
+				f.failWith("ConnectGetHostname", fakeErr(libvirt.ErrAuthFailed, "authentication failed"))
+			},
+			wantReason: provider.ReasonForbidden,
+		},
+		{
+			name: "pool not found",
+			prep: func(f *fakeBackend) {
+				f.mu.Lock()
+				delete(f.pools, fakePoolName)
+				f.mu.Unlock()
+			},
+			wantReason:  reasonPoolNotFound,
+			wantMessage: fakePoolName,
+		},
+		{
+			name: "other libvirt error",
+			prep: func(f *fakeBackend) {
+				f.failWith("StoragePoolGetInfo", fakeErr(libvirt.ErrInternalError, "internal error: pool info"))
+			},
+			wantReason: provider.ReasonAPIError,
+		},
+		{
+			name:       "rpc outlives the deadline",
+			prep:       func(f *fakeBackend) { f.blockOn("ConnectGetLibVersion") },
+			timeout:    50 * time.Millisecond,
+			wantReason: provider.ReasonTimeout,
+		},
+		{
+			name:       "ctx cancelled mid check",
+			prep:       func(f *fakeBackend) { f.blockOn("StoragePoolLookupByName") },
+			cancel:     true,
+			wantReason: provider.ReasonTimeout,
+		},
+		{
+			name: "link killed mid check",
+			prep: func(f *fakeBackend) {
+				f.failWith("StoragePoolLookupByName", transientf("libvirt call StoragePoolLookupByName on beelink: connection closed while waiting: keepalive: EOF"))
+			},
+			wantReason: provider.ReasonUnreachable,
+		},
+		{
+			name:        "pool inactive",
+			prep:        func(f *fakeBackend) { f.pools[fakePoolName].state = libvirt.StoragePoolInactive },
+			wantReason:  reasonPoolNotActive,
+			wantMessage: `storage pool "default" is inactive`,
+		},
+		{
+			name:       "ok",
+			wantReady:  true,
+			wantReason: provider.ReasonReachable,
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			cfg := twoHosts()
+			cfg.Hosts = cfg.Hosts[:1]
+			f := newFakeBackend()
+			d := &checkDialer{backends: map[string]*fakeBackend{testHost: f}}
+			if tc.dialErr != nil {
+				d.errs = map[string]error{testHost: tc.dialErr(t)}
+			}
+			if tc.prep != nil {
+				tc.prep(f)
+			}
+			p := newProvider(testProvider, cfg, d)
+			ctx, _ := checkContext()
+			switch {
+			case tc.timeout > 0:
+				var cancel context.CancelFunc
+				ctx, cancel = context.WithTimeout(ctx, tc.timeout)
+				defer cancel()
+			case tc.cancel:
+				var cancel context.CancelFunc
+				ctx, cancel = context.WithCancel(ctx)
+				time.AfterFunc(50*time.Millisecond, cancel)
+			}
+			got := p.Check(ctx)
+			if len(got) != 1 {
+				t.Fatalf("Check() returned %d targets, want 1", len(got))
+			}
+			ts := got[0]
+			if ts.Ready != tc.wantReady || ts.Reason != tc.wantReason {
+				t.Errorf("verdict = ready %v reason %s (%s), want ready %v reason %s", ts.Ready, ts.Reason, ts.Message, tc.wantReady, tc.wantReason)
+			}
+			if !tc.wantReady && ts.Message == "" {
+				t.Error("failed check has no message")
+			}
+			if tc.wantReady && ts.Message != "" {
+				t.Errorf("successful check has a message: %q", ts.Message)
+			}
+			if strings.Contains(ts.Message, keyMaterial) {
+				t.Errorf("message leaks the key file content: %q", ts.Message)
+			}
+			if tc.wantMessage != "" && !strings.Contains(ts.Message, tc.wantMessage) {
+				t.Errorf("message = %q, want it to contain %q", ts.Message, tc.wantMessage)
+			}
+			if ts.Details["user"] != "virt" || ts.Details["pool"] != "default" {
+				t.Errorf("details = %v, want user and pool", ts.Details)
+			}
+			if tc.wantReady {
+				want := map[string]string{
+					"user": "virt", "pool": "default", "hostname": "fake-host", "libvirtVersion": "11.6.0",
+					"poolState": "running", "poolCapacityGiB": "100.0", "poolAvailableGiB": "50.0",
+				}
+				for k, v := range want {
+					if ts.Details[k] != v {
+						t.Errorf("details[%s] = %q, want %q", k, ts.Details[k], v)
+					}
+				}
+			}
+		})
+	}
+}
+
+func TestCheck_neverUsesPooledConnection(t *testing.T) {
+	t.Parallel()
+	cfg := threeHosts()
+	fakes := fakesFor(cfg)
+	d := &checkDialer{backends: fakes}
+	p := newProvider(testProvider, cfg, d)
+	ctx, _ := checkContext()
+	got := p.Check(ctx)
+	for _, ts := range got {
+		if !ts.Ready {
+			t.Errorf("target %s not ready: %s: %s", ts.Name, ts.Reason, ts.Message)
+		}
+	}
+	if n := d.dials.Load(); int(n) != len(cfg.Hosts) {
+		t.Errorf("dials = %d, want one fresh dial per host (%d)", n, len(cfg.Hosts))
+	}
+	p.hosts.mu.Lock()
+	pooled := len(p.hosts.hosts)
+	p.hosts.mu.Unlock()
+	if pooled != 0 {
+		t.Errorf("pooled table has %d entries after Check, want none", pooled)
+	}
+	for name, f := range fakes {
+		if f.closeCount() != 1 {
+			t.Errorf("fresh backend for %s closed %d times, want 1", name, f.closeCount())
+		}
+	}
+}
+
+func TestCheck_detailsKeys(t *testing.T) {
+	t.Parallel()
+	full := []string{"hostname", "libvirtVersion", "pool", "poolAvailableGiB", "poolCapacityGiB", "poolState", "user"}
+	tests := []struct {
+		name    string
+		dialErr error
+		prep    func(f *fakeBackend)
+		want    []string
+	}{
+		{name: "success", want: full},
+		{name: "pool not active", prep: func(f *fakeBackend) { f.pools[fakePoolName].state = libvirt.StoragePoolDegraded }, want: full},
+		{
+			name:    "dial failure",
+			dialErr: &dialError{stage: stageTCP, host: testHost, err: errors.New("dial tcp: connection refused")},
+			want:    []string{"pool", "user"},
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			cfg := twoHosts()
+			cfg.Hosts = cfg.Hosts[:1]
+			f := newFakeBackend()
+			if tc.prep != nil {
+				tc.prep(f)
+			}
+			d := &checkDialer{backends: map[string]*fakeBackend{testHost: f}}
+			if tc.dialErr != nil {
+				d.errs = map[string]error{testHost: tc.dialErr}
+			}
+			p := newProvider(testProvider, cfg, d)
+			ctx, _ := checkContext()
+			got := p.Check(ctx)[0]
+			if keys := detailKeys(got.Details); !slices.Equal(keys, tc.want) {
+				t.Errorf("details keys = %v, want %v", keys, tc.want)
+			}
+			for k, v := range got.Details {
+				if v == "" {
+					t.Errorf("details[%s] is empty", k)
+				}
+			}
+		})
+	}
+}
+
+func TestCheck_logsVerdictPerHost(t *testing.T) {
+	t.Parallel()
+	cfg := twoHosts()
+	d := &checkDialer{backends: fakesFor(cfg), errs: map[string]error{"lab2": &dialError{stage: stageTCP, host: "lab2", err: errors.New("dial tcp: i/o timeout")}}}
+	p := newProvider(testProvider, cfg, d)
+	ctx, logs := checkContext()
+	p.Check(ctx)
+	checked := logs.matching("kvm2 host checked")
+	if len(checked) != 2 {
+		t.Fatalf("got %d verdict lines, want 2: %v", len(checked), checked)
+	}
+	for _, l := range checked {
+		switch {
+		case strings.Contains(l, `"host"="beelink"`):
+			if !strings.Contains(l, `"ready"=true`) || !strings.Contains(l, `"reason"="Reachable"`) || !strings.Contains(l, `"latency"=`) {
+				t.Errorf("beelink verdict line = %s", l)
+			}
+		case strings.Contains(l, `"host"="lab2"`):
+			if !strings.Contains(l, `"ready"=false`) || !strings.Contains(l, `"reason"="Unreachable"`) {
+				t.Errorf("lab2 verdict line = %s", l)
+			}
+		default:
+			t.Errorf("verdict line without a host: %s", l)
+		}
+	}
+}
+
+func TestCheckReason_table(t *testing.T) {
+	t.Parallel()
+	dial := func(stage dialStage, cause error) error {
+		return fmt.Errorf("wrapped: %w", &dialError{stage: stage, host: testHost, err: cause})
+	}
+	tests := []struct {
+		name string
+		err  error
+		want string
+	}{
+		{name: "credentials stage", err: dial(stageCredentials, errors.New("reading ssh key file /k: no such file")), want: reasonCredentialsUnavailable},
+		{name: "credentials stage even when permanent", err: dial(stageCredentials, permanentf("parsing ssh key file /k: ssh: no key found")), want: reasonCredentialsUnavailable},
+		{name: "tcp stage", err: dial(stageTCP, errors.New("connect: connection refused")), want: provider.ReasonUnreachable},
+		{name: "tcp stage with a deadline", err: dial(stageTCP, context.DeadlineExceeded), want: provider.ReasonUnreachable},
+		{name: "ssh stage unknown host key", err: dial(stageSSH, &knownhosts.KeyError{}), want: provider.ReasonUnauthorized},
+		{name: "ssh stage wrapped key error", err: dial(stageSSH, fmt.Errorf("ssh: handshake failed: %w", &knownhosts.KeyError{Want: []knownhosts.KnownKey{{}}})), want: provider.ReasonUnauthorized},
+		{name: "ssh stage unable to authenticate", err: dial(stageSSH, errors.New("ssh: handshake failed: ssh: unable to authenticate, attempted methods [none publickey], no supported methods remain")), want: provider.ReasonUnauthorized},
+		{name: "ssh stage reset", err: dial(stageSSH, errors.New("read tcp: connection reset by peer")), want: provider.ReasonUnreachable},
+		{name: "ssh stage cancelled", err: dial(stageSSH, context.Canceled), want: provider.ReasonUnreachable},
+		{name: "socket stage", err: dial(stageSocket, errors.New("opening /run/libvirt/libvirt-sock: ssh: rejected")), want: provider.ReasonUnreachable},
+		{name: "deadline exceeded", err: fmt.Errorf("%w: libvirt call ConnectGetHostname on beelink exceeded its 30s budget: %w", provider.ErrTransient, context.DeadlineExceeded), want: provider.ReasonTimeout},
+		{name: "cancelled", err: fmt.Errorf("%w: libvirt call ConnectGetHostname on beelink abandoned: %w", provider.ErrTransient, context.Canceled), want: provider.ReasonTimeout},
+		{name: "access denied", err: fakeErr(libvirt.ErrAccessDenied, "access denied"), want: provider.ReasonForbidden},
+		{name: "auth failed wrapped", err: fmt.Errorf("connecting to libvirt on host beelink: %w", fakeErr(libvirt.ErrAuthFailed, "authentication failed")), want: provider.ReasonForbidden},
+		{name: "no storage pool", err: fakeErr(libvirt.ErrNoStoragePool, "Storage pool not found"), want: reasonPoolNotFound},
+		{name: "other libvirt error", err: fakeErr(libvirt.ErrInternalError, "internal error"), want: provider.ReasonAPIError},
+		{name: "link killed", err: transientf("libvirt call X on beelink: connection closed while waiting: keepalive: EOF"), want: provider.ReasonUnreachable},
+		{name: "plain error", err: errors.New("unexpected EOF"), want: provider.ReasonUnreachable},
+		{name: "nil", err: nil, want: provider.ReasonUnreachable},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			if got := checkReason(tc.err); got != tc.want {
+				t.Errorf("checkReason(%v) = %s, want %s", tc.err, got, tc.want)
+			}
+		})
+	}
+}
+
+func TestFormatLibVersion(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		in   uint64
+		want string
+	}{
+		{0, "0.0.0"},
+		{11006000, "11.6.0"},
+		{10000003, "10.0.3"},
+		{9010012, "9.10.12"},
+		{100999999, "100.999.999"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.want, func(t *testing.T) {
+			t.Parallel()
+			if got := formatLibVersion(tc.in); got != tc.want {
+				t.Errorf("formatLibVersion(%d) = %s, want %s", tc.in, got, tc.want)
+			}
+		})
+	}
+}
+
+func TestPoolStateName(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		in   uint8
+		want string
+	}{
+		{uint8(libvirt.StoragePoolInactive), "inactive"},
+		{uint8(libvirt.StoragePoolBuilding), "building"},
+		{uint8(libvirt.StoragePoolRunning), "running"},
+		{uint8(libvirt.StoragePoolDegraded), "degraded"},
+		{uint8(libvirt.StoragePoolInaccessible), "inaccessible"},
+		{42, "state 42"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.want, func(t *testing.T) {
+			t.Parallel()
+			if got := poolStateName(tc.in); got != tc.want {
+				t.Errorf("poolStateName(%d) = %s, want %s", tc.in, got, tc.want)
+			}
+		})
+	}
+}
@@ -0,0 +1,280 @@
+package kvm2
+
+import (
+	"bytes"
+	"fmt"
+	"time"
+
+	"context"
+
+	"github.com/digitalocean/go-libvirt"
+	"github.com/go-logr/logr"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// The wait after DomainCreate (spec entry 18): Create returns only once the
+// domain is running, so Get never sees a domain kvm2 defined but did not
+// start (spec entry 9). DomainCreate is synchronous, so the first read
+// normally succeeds; the cap only covers state settling and must not pin a
+// reconcile worker for long.
+const (
+	defaultRunningWait = 15 * time.Second
+	defaultRunningPoll = 500 * time.Millisecond
+)
+
+// Create is spec entry 10: validate placement, then, under the per-name
+// lock, find or build the domain step by step, every step idempotent, and
+// return once it is running.
+func (p *Provider) Create(ctx context.Context, req provider.CreateRequest) (string, error) {
+	host, err := p.host(req.Placement.Zone)
+	if err != nil {
+		if req.Placement.Zone != "" {
+			err = permanentf("placement.zone %q is not in kvm2.hosts", req.Placement.Zone)
+		}
+		return "", p.wrapErr("create", "", err)
+	}
+	mt, ok := p.cfg.MachineTypes[req.Placement.MachineType]
+	if !ok {
+		return "", p.wrapErr("create", "", permanentf("placement.machineType %q is not in kvm2.machineTypes", req.Placement.MachineType))
+	}
+	src, err := resolveImage(p.name, p.cfg, req.Placement.Image)
+	if err != nil {
+		return "", p.wrapErr("create", "", err)
+	}
+
+	id := formatProviderID(host.Name, req.Name)
+	log := p.logger(ctx).WithValues("host", host.Name, "name", req.Name)
+	log.V(2).Info("kvm2 create resolved",
+		"pool", host.Pool, "bridge", host.Bridge, "machineType", req.Placement.MachineType,
+		"vcpus", mt.VCPUs, "memoryMiB", mt.MemoryMiB, "diskSizeGb", p.cfg.DiskSizeGB,
+		"image", src.volumeName(), "firmware", src.Firmware, "cloudInitBytes", len(req.CloudInit))
+
+	unlock, err := p.lockName(ctx, req.Name)
+	if err != nil {
+		return "", p.wrapErr("create", id, err)
+	}
+	defer unlock()
+
+	be, err := p.hosts.get(ctx, host)
+	if err != nil {
+		logAPIError(log, "create", err)
+		return "", p.wrapErr("create", id, err)
+	}
+	if err := p.create(ctx, log, be, host, req, mt, src); err != nil {
+		logAPIError(log, "create", err)
+		return "", p.wrapErr("create", id, err)
+	}
+	log.V(1).Info("kvm2 domain ensured running", "id", id)
+	return id, nil
+}
+
+func (p *Provider) create(ctx context.Context, log logr.Logger, be backend, host provider.KVMHost, req provider.CreateRequest, mt provider.KVMMachineType, src imageSource) error {
+	dom, err := be.DomainLookupByName(ctx, req.Name)
+	switch {
+	case err == nil:
+		meta, err := readMetadata(ctx, be, dom)
+		if err != nil {
+			return err
+		}
+		if !meta.managed() {
+			return permanentf("domain %s exists but carries no kvm2 mark; it was not created by this provider", req.Name)
+		}
+		if meta.UID != "" && meta.UID != req.UID {
+			return permanentf("domain %s belongs to proxy uid %s, not %s", req.Name, meta.UID, req.UID)
+		}
+		log.V(1).Info("kvm2 domain already defined", "uid", meta.UID)
+		return p.ensureRunning(ctx, log, be, dom)
+	case !hasCode(err, libvirt.ErrNoDomain):
+		return fmt.Errorf("looking up domain %s: %w", req.Name, err)
+	}
+
+	pool, err := be.StoragePoolLookupByName(ctx, host.Pool)
+	if err != nil {
+		return fmt.Errorf("looking up storage pool %s: %w", host.Pool, err)
+	}
+	imageBytes, err := p.imageSizeIfUncached(ctx, be, pool, src)
+	if err != nil {
+		return err
+	}
+	if err := checkPoolSpace(ctx, be, pool, host.Pool, neededBytes(imageBytes, p.cfg.DiskSizeGB)); err != nil {
+		return err
+	}
+	basePath, err := p.ensureImage(ctx, host, pool, be, src)
+	if err != nil {
+		return err
+	}
+	overlay, err := ensureOverlay(ctx, log, be, pool, req.Name, basePath, src.Format, p.cfg.DiskSizeGB)
+	if err != nil {
+		return err
+	}
+	seed, err := p.writeSeed(ctx, log, be, pool, req)
+	if err != nil {
+		return err
+	}
+	diskPath, err := be.StorageVolGetPath(ctx, overlay)
+	if err != nil {
+		return fmt.Errorf("reading the path of overlay volume %s: %w", overlay.Name, err)
+	}
+	seedPath, err := be.StorageVolGetPath(ctx, seed)
+	if err != nil {
+		return fmt.Errorf("reading the path of seed volume %s: %w", seed.Name, err)
+	}
+
+	xmlDoc, err := buildDomainXML(domainParams{
+		Name:       req.Name,
+		UID:        req.UID,
+		CreatedAt:  p.now(),
+		VCPUs:      mt.VCPUs,
+		MemoryMiB:  mt.MemoryMiB,
+		Firmware:   src.Firmware,
+		Bridge:     host.Bridge,
+		DiskPath:   diskPath,
+		SeedPath:   seedPath,
+		DiskFormat: provider.KVMFormatQcow2,
+	})
+	if err != nil {
+		return err
+	}
+	dom, err = be.DomainDefineXML(ctx, xmlDoc)
+	if err != nil {
+		return fmt.Errorf("defining domain %s: %w", req.Name, err)
+	}
+	kv := []any{"uid", req.UID, "firmware", src.Firmware}
+	if log.V(2).Enabled() {
+		kv = append(kv, "xml", xmlDoc)
+	}
+	log.V(1).Info("domain defined", kv...)
+	return p.ensureRunning(ctx, log, be, dom)
+}
+
+// ensureOverlay creates the per-VM qcow2 overlay backed by the base image,
+// or reuses the one a previous attempt left: volume creation is atomic, so
+// an existing overlay is complete.
+func ensureOverlay(ctx context.Context, log logr.Logger, be backend, pool libvirt.StoragePool, name, basePath, baseFormat string, diskSizeGB int64) (libvirt.StorageVol, error) {
+	volName := overlayVolumeName(name)
+	xmlDoc, err := buildVolumeXML(volumeSpec{
+		Name:          volName,
+		CapacityBytes: diskSizeGB << 30,
+		Format:        provider.KVMFormatQcow2,
+		BackingPath:   basePath,
+		BackingFormat: baseFormat,
+	})
+	if err != nil {
+		return libvirt.StorageVol{}, err
+	}
+	vol, err := be.StorageVolCreateXML(ctx, pool, xmlDoc, 0)
+	switch {
+	case err == nil:
+		log.V(1).Info("overlay volume created", "volume", volName)
+		return vol, nil
+	case hasCode(err, libvirt.ErrStorageVolExist):
+		vol, err = be.StorageVolLookupByName(ctx, pool, volName)
+		if err != nil {
+			return libvirt.StorageVol{}, fmt.Errorf("looking up overlay volume %s: %w", volName, err)
+		}
+		log.V(1).Info("overlay volume already exists", "volume", volName)
+		return vol, nil
+	default:
+		return libvirt.StorageVol{}, fmt.Errorf("creating overlay volume %s: %w", volName, err)
+	}
+}
+
+// writeSeed builds the NoCloud seed and uploads it as a raw volume,
+// replacing whatever a previous attempt left (spec entry 5): a cut-off
+// upload would boot a VM without cloud-init. The ISO is tens of KiB, so it
+// is built in memory. Nothing about its content is logged or put in an
+// error.
+func (p *Provider) writeSeed(ctx context.Context, log logr.Logger, be backend, pool libvirt.StoragePool, req provider.CreateRequest) (libvirt.StorageVol, error) {
+	name := seedVolumeName(req.Name)
+	if err := deleteVolumeIfPresent(ctx, be, pool, name); err != nil {
+		return libvirt.StorageVol{}, err
+	}
+	installAgent := p.cfg.InstallGuestAgent == nil || *p.cfg.InstallGuestAgent
+	vendorData, err := buildVendorData(p.cfg.VendorData, installAgent)
+	if err != nil {
+		return libvirt.StorageVol{}, permanentf("building vendor-data for %s: %v", req.Name, err)
+	}
+	var iso bytes.Buffer
+	if err := buildSeedISO(req.Name, req.CloudInit, vendorData, &iso); err != nil {
+		return libvirt.StorageVol{}, permanentf("building seed ISO for %s: %v", req.Name, err)
+	}
+	xmlDoc, err := buildVolumeXML(volumeSpec{Name: name, CapacityBytes: int64(iso.Len()), Format: provider.KVMFormatRaw})
+	if err != nil {
+		return libvirt.StorageVol{}, err
+	}
+	vol, err := be.StorageVolCreateXML(ctx, pool, xmlDoc, 0)
+	if err != nil {
+		return libvirt.StorageVol{}, fmt.Errorf("creating seed volume %s: %w", name, err)
+	}
+	if err := be.StorageVolUpload(ctx, vol, bytes.NewReader(iso.Bytes()), 0, uint64(iso.Len()), 0); err != nil {
+		return libvirt.StorageVol{}, fmt.Errorf("uploading seed volume %s: %w", name, err)
+	}
+	log.V(1).Info("seed volume uploaded", "volume", name, "bytes", iso.Len())
+	return vol, nil
+}
+
+// ensureRunning starts the domain unless it is active already, then waits
+// until libvirt reports it running. ErrOperationInvalid from DomainCreate
+// means another caller started it between the state read and the call. A
+// paused or pmsuspended domain is an admin's doing and DomainCreate would
+// refuse it; the verdict is Permanent so the reconciler does not retry it.
+func (p *Provider) ensureRunning(ctx context.Context, log logr.Logger, be backend, dom libvirt.Domain) error {
+	state, _, err := be.DomainGetState(ctx, dom, 0)
+	if err != nil {
+		return fmt.Errorf("reading state of domain %s: %w", dom.Name, err)
+	}
+	switch s := libvirt.DomainState(state); {
+	case isActive(s):
+		return nil
+	case s == libvirt.DomainPaused || s == libvirt.DomainPmsuspended:
+		return permanentf("domain %s is %s and cannot be started by the operator; resume or destroy it on the host", dom.Name, s)
+	}
+	switch err := be.DomainCreate(ctx, dom); {
+	case err == nil:
+		log.V(1).Info("domain started")
+	case hasCode(err, libvirt.ErrOperationInvalid):
+		log.V(1).Info("domain start reported it already running")
+	default:
+		return fmt.Errorf("starting domain %s: %w", dom.Name, err)
+	}
+	return p.waitRunning(ctx, be, dom)
+}
+
+// waitRunning polls DomainGetState until the domain is active, the cap
+// passes (Transient: the next Create attempt finds the domain defined and
+// tries again) or ctx ends. Only nostate is worth another poll; a domain
+// that is shut off, crashed or paused right after its start is not going
+// to come up on its own, so that is Transient at once.
+func (p *Provider) waitRunning(ctx context.Context, be backend, dom libvirt.Domain) error {
+	deadline := time.NewTimer(p.runningWait)
+	defer deadline.Stop()
+	ticker := time.NewTicker(p.runningPoll)
+	defer ticker.Stop()
+	for {
+		state, _, err := be.DomainGetState(ctx, dom, 0)
+		if err != nil {
+			return fmt.Errorf("reading state of domain %s: %w", dom.Name, err)
+		}
+		switch s := libvirt.DomainState(state); {
+		case isActive(s):
+			return nil
+		case s != libvirt.DomainNostate:
+			return transientf("domain %s went %s right after start", dom.Name, s)
+		}
+		select {
+		case <-ticker.C:
+		case <-deadline.C:
+			return transientf("domain %s did not reach running within %s (state %s)", dom.Name, p.runningWait, libvirt.DomainState(state))
+		case <-ctx.Done():
+			return fmt.Errorf("%w: waiting for domain %s to run: %w", provider.ErrTransient, dom.Name, ctx.Err())
+		}
+	}
+}
+
+// logAPIError is the V(1) line per failed backend call of spec entry 20.
+func logAPIError(log logr.Logger, op string, err error) {
+	kv := []any{"op", op, "error", err.Error()}
+	kv = append(kv, libvirtErrorFields(err)...)
+	log.V(1).Info("kvm2 backend call failed", kv...)
+}
@@ -0,0 +1,467 @@
+package kvm2
+
+import (
+	"bytes"
+	"errors"
+	"math"
+	"strconv"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+func TestCreate_repeatReturnsSameIDAndCreatesNothing(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	first := fx.create(testName)
+	before := len(fx.fake.callLog())
+	defines, vols, uploads := fx.fake.callCount("DomainDefineXML"), fx.fake.callCount("StorageVolCreateXML"), fx.fake.callCount("StorageVolUpload")
+
+	second := fx.create(testName)
+	if second != first {
+		t.Errorf("second Create = %q, want %q", second, first)
+	}
+	if got := fx.fake.callCount("DomainDefineXML"); got != defines {
+		t.Errorf("DomainDefineXML calls after repeat = %d, want %d", got, defines)
+	}
+	if got := fx.fake.callCount("StorageVolCreateXML"); got != vols {
+		t.Errorf("StorageVolCreateXML calls after repeat = %d, want %d", got, vols)
+	}
+	if got := fx.fake.callCount("StorageVolUpload"); got != uploads {
+		t.Errorf("StorageVolUpload calls after repeat = %d, want %d", got, uploads)
+	}
+	if got := fx.fake.callCount("DomainCreate"); got != 1 {
+		t.Errorf("DomainCreate calls = %d, want 1", got)
+	}
+	repeat := fx.fake.callLog()[before:]
+	for _, rpc := range repeat {
+		switch rpc {
+		case "DomainLookupByName", "DomainGetMetadata", "DomainGetState":
+		default:
+			t.Errorf("repeat Create made %s; want only lookup, metadata and state reads: %v", rpc, repeat)
+		}
+	}
+	if len(fx.logs.matching("kvm2 domain already defined")) != 1 {
+		t.Errorf("no 'already defined' line:\n%s", strings.Join(fx.logs.all(), "\n"))
+	}
+}
+
+func TestCreate_resumesAfterEachPartialStep(t *testing.T) {
+	t.Parallel()
+	overlay, seed := overlayVolumeName(testName), seedVolumeName(testName)
+	tests := []struct {
+		name  string
+		setup func(fx *fixture)
+		check func(t *testing.T, fx *fixture)
+	}{
+		{
+			name: "base without marker is refetched",
+			setup: func(fx *fixture) {
+				fx.putVol(fx.baseName(), []byte("truncated"))
+			},
+			check: func(t *testing.T, fx *fixture) {
+				if got := fx.fake.vol(fx.baseName()); got == nil || !bytes.Equal(got.content, fx.image) {
+					t.Errorf("base volume not refetched")
+				}
+				if fx.fake.vol(fx.baseName()+".sha256") == nil {
+					t.Errorf("marker not written")
+				}
+			},
+		},
+		{
+			name: "overlay exists is reused",
+			setup: func(fx *fixture) {
+				fx.cacheImage()
+				fx.putVol(overlay, []byte("overlay"))
+			},
+			check: func(t *testing.T, fx *fixture) {
+				if got := fx.fake.vol(overlay); got == nil || string(got.content) != "overlay" {
+					t.Errorf("overlay volume was replaced")
+				}
+				if len(fx.logs.matching("overlay volume already exists")) != 1 {
+					t.Errorf("no reuse line:\n%s", strings.Join(fx.logs.all(), "\n"))
+				}
+			},
+		},
+		{
+			name: "truncated seed is replaced",
+			setup: func(fx *fixture) {
+				fx.cacheImage()
+				fx.putVol(seed, []byte("cut"))
+			},
+			check: func(t *testing.T, fx *fixture) {
+				got := fx.fake.vol(seed)
+				if got == nil || len(got.content) <= 3 || !bytes.Contains(got.content, []byte(seedLabel)) {
+					t.Errorf("seed volume not rebuilt")
+				}
+			},
+		},
+		{
+			name: "domain defined but shutoff is started",
+			setup: func(fx *fixture) {
+				fx.cacheImage()
+				fx.putVol(overlay, []byte("overlay"))
+				fx.putVol(seed, []byte("seed"))
+				fx.fake.addDomain(testName, libvirt.DomainShutoff, managedMetadata(testUID, time.Now()))
+			},
+			check: func(t *testing.T, fx *fixture) {
+				if got := fx.fake.callCount("DomainDefineXML"); got != 0 {
+					t.Errorf("DomainDefineXML calls = %d, want 0", got)
+				}
+				if got := fx.fake.callCount("DomainCreate"); got != 1 {
+					t.Errorf("DomainCreate calls = %d, want 1", got)
+				}
+			},
+		},
+		{
+			name: "domain running is returned as is",
+			setup: func(fx *fixture) {
+				fx.fake.addDomain(testName, libvirt.DomainRunning, managedMetadata(testUID, time.Now()))
+			},
+			check: func(t *testing.T, fx *fixture) {
+				for _, rpc := range []string{"DomainDefineXML", "DomainCreate", "StorageVolCreateXML"} {
+					if got := fx.fake.callCount(rpc); got != 0 {
+						t.Errorf("%s calls = %d, want 0", rpc, got)
+					}
+				}
+			},
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			tc.setup(fx)
+			fx.create(testName)
+			fx.withDomain(testName, func(d *fakeDomain) {
+				if d.state != libvirt.DomainRunning {
+					t.Errorf("domain state = %s, want running", d.state)
+				}
+			})
+			tc.check(t, fx)
+		})
+	}
+}
+
+func TestCreate_returnsOnlyWhenRunning(t *testing.T) {
+	t.Parallel()
+	t.Run("returns after the domain reports running", func(t *testing.T) {
+		t.Parallel()
+		fx := newFixture(t, 1)
+		fx.fake.startDelayPolls = 3
+		fx.create(testName)
+		fx.withDomain(testName, func(d *fakeDomain) {
+			if d.state != libvirt.DomainRunning {
+				t.Errorf("Create returned with domain state %s", d.state)
+			}
+		})
+		if got := fx.fake.callCount("DomainGetState"); got < 4 {
+			t.Errorf("DomainGetState calls = %d, want at least 4 (one before start, three polls)", got)
+		}
+	})
+	t.Run("never running is transient after the cap", func(t *testing.T) {
+		t.Parallel()
+		fx := newFixture(t, 1, withRunningWait(time.Millisecond, 30*time.Millisecond))
+		fx.fake.startDelayPolls = math.MaxInt32
+		start := time.Now()
+		id, err := fx.p.Create(fx.ctx, fx.req(testName))
+		if id != "" {
+			t.Errorf("Create = %q, want no ID", id)
+		}
+		assertProviderError(t, err, "create", provider.ErrTransient)
+		if !strings.Contains(err.Error(), "did not reach running") {
+			t.Errorf("error = %v, want the wait cap named", err)
+		}
+		if elapsed := time.Since(start); elapsed > 2*time.Second {
+			t.Errorf("Create took %v, want the shrunk cap to end it", elapsed)
+		}
+	})
+}
+
+func TestCreate_domainThatDiesAfterStartIsTransientAtOnce(t *testing.T) {
+	t.Parallel()
+	for _, state := range []libvirt.DomainState{libvirt.DomainCrashed, libvirt.DomainShutoff, libvirt.DomainPaused} {
+		t.Run(state.String(), func(t *testing.T) {
+			t.Parallel()
+			// The cap is well over the bound asserted below, so a poll that
+			// ran to the cap would fail the test.
+			fx := newFixture(t, 1, withRunningWait(time.Millisecond, 10*time.Second))
+			fx.fake.stateAfterStart = state
+			start := time.Now()
+			id, err := fx.p.Create(fx.ctx, fx.req(testName))
+			elapsed := time.Since(start)
+			if id != "" {
+				t.Errorf("Create = %q, want no ID", id)
+			}
+			assertProviderError(t, err, "create", provider.ErrTransient)
+			if !strings.Contains(err.Error(), "right after start") {
+				t.Errorf("error = %v, want the state after start named", err)
+			}
+			if elapsed > time.Second {
+				t.Errorf("Create took %v, want it to give up at once, not poll to the cap", elapsed)
+			}
+			if got := fx.fake.callCount("DomainCreate"); got != 1 {
+				t.Errorf("DomainCreate calls = %d, want 1", got)
+			}
+		})
+	}
+}
+
+func TestCreate_pausedDomainIsPermanent(t *testing.T) {
+	t.Parallel()
+	for _, state := range []libvirt.DomainState{libvirt.DomainPaused, libvirt.DomainPmsuspended} {
+		t.Run(state.String(), func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			fx.fake.addDomain(testName, state, managedMetadata(testUID, time.Now()))
+			id, err := fx.p.Create(fx.ctx, fx.req(testName))
+			if id != "" {
+				t.Errorf("Create = %q, want no ID", id)
+			}
+			assertProviderError(t, err, "create", provider.ErrPermanent)
+			if !strings.Contains(err.Error(), "cannot be started by the operator") {
+				t.Errorf("error = %v, want the admin action named", err)
+			}
+			for _, rpc := range []string{"DomainCreate", "DomainDefineXML", "DomainDestroy"} {
+				if got := fx.fake.callCount(rpc); got != 0 {
+					t.Errorf("%s calls = %d, want 0", rpc, got)
+				}
+			}
+			fx.withDomain(testName, func(d *fakeDomain) {
+				if d.state != state {
+					t.Errorf("domain state = %s, want it left %s", d.state, state)
+				}
+			})
+		})
+	}
+}
+
+func TestCreate_rejectsUnknownPlacement(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name    string
+		mutate  func(req *provider.CreateRequest)
+		wantMsg string
+	}{
+		{
+			name:    "zone",
+			mutate:  func(req *provider.CreateRequest) { req.Placement.Zone = "nas9" },
+			wantMsg: `placement.zone "nas9" is not in kvm2.hosts`,
+		},
+		{
+			name:    "machineType",
+			mutate:  func(req *provider.CreateRequest) { req.Placement.MachineType = "xlarge" },
+			wantMsg: `placement.machineType "xlarge" is not in kvm2.machineTypes`,
+		},
+		{
+			name:    "image",
+			mutate:  func(req *provider.CreateRequest) { req.Placement.Image = "ubuntu-24" },
+			wantMsg: `placement.image "ubuntu-24" is neither a key in kvm2.images nor an http(s) URL`,
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			req := fx.req(testName)
+			tc.mutate(&req)
+			id, err := fx.p.Create(fx.ctx, req)
+			if id != "" {
+				t.Errorf("Create = %q, want no ID", id)
+			}
+			assertProviderError(t, err, "create", provider.ErrPermanent)
+			if !strings.Contains(err.Error(), tc.wantMsg) {
+				t.Errorf("error = %q, want substring %q", err, tc.wantMsg)
+			}
+			if got := fx.fake.callLog(); len(got) != 0 {
+				t.Errorf("backend calls = %v, want none before validation", got)
+			}
+			if got := fx.dialer.dials.Load(); got != 0 {
+				t.Errorf("dials = %d, want 0", got)
+			}
+		})
+	}
+}
+
+func TestCreate_adoptionWithForeignUIDIsPermanent(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	fx.fake.addDomain(testName, libvirt.DomainRunning, managedMetadata("other-uid", time.Now()))
+	id, err := fx.p.Create(fx.ctx, fx.req(testName))
+	if id != "" {
+		t.Errorf("Create = %q, want no ID", id)
+	}
+	assertProviderError(t, err, "create", provider.ErrPermanent)
+	if !strings.Contains(err.Error(), "other-uid") || !strings.Contains(err.Error(), testUID) {
+		t.Errorf("error = %q, want both UIDs named", err)
+	}
+	for _, rpc := range []string{"DomainCreate", "DomainDefineXML", "StorageVolCreateXML", "DomainDestroy"} {
+		if got := fx.fake.callCount(rpc); got != 0 {
+			t.Errorf("%s calls = %d, want 0", rpc, got)
+		}
+	}
+}
+
+func TestCreate_adoptsOnlyMarkedDomains(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name      string
+		metadata  string
+		wantAdopt bool
+	}{
+		{name: "matching uid", metadata: managedMetadata(testUID, time.Now()), wantAdopt: true},
+		{name: "managed, empty uid", metadata: managedMetadata("", time.Now()), wantAdopt: true},
+		{name: "no metadata element", metadata: ""},
+		{name: "element without the managed mark", metadata: `<instance><uid>` + testUID + `</uid></instance>`},
+		{name: "managed false", metadata: `<instance><managed>false</managed><uid>` + testUID + `</uid></instance>`},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			fx.fake.addDomain(testName, libvirt.DomainRunning, tc.metadata)
+			id, err := fx.p.Create(fx.ctx, fx.req(testName))
+			if tc.wantAdopt {
+				if err != nil || id != formatProviderID(testHost, testName) {
+					t.Fatalf("Create = %q, %v; want the existing domain adopted", id, err)
+				}
+			} else {
+				if id != "" {
+					t.Errorf("Create = %q, want no ID", id)
+				}
+				assertProviderError(t, err, "create", provider.ErrPermanent)
+				if !strings.Contains(err.Error(), "carries no kvm2 mark") {
+					t.Errorf("error = %v, want the missing mark named", err)
+				}
+			}
+			for _, rpc := range []string{"DomainDefineXML", "DomainCreate", "StorageVolCreateXML", "DomainDestroy"} {
+				if got := fx.fake.callCount(rpc); got != 0 {
+					t.Errorf("%s calls = %d, want 0", rpc, got)
+				}
+			}
+		})
+	}
+}
+
+func TestCreate_poolOutOfSpaceIsQuotaExceeded(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	fx.fake.mu.Lock()
+	fx.fake.pools[fakePoolName].available = 1 << 30
+	fx.fake.mu.Unlock()
+	_, err := fx.p.Create(fx.ctx, fx.req(testName))
+	assertProviderError(t, err, "create", provider.ErrQuotaExceeded)
+	if !strings.Contains(err.Error(), fakePoolName) || !strings.Contains(err.Error(), "needed") {
+		t.Errorf("error = %q, want the pool and the needed bytes named", err)
+	}
+	for _, rpc := range []string{"StorageVolCreateXML", "StorageVolUpload", "DomainDefineXML"} {
+		if got := fx.fake.callCount(rpc); got != 0 {
+			t.Errorf("%s calls = %d, want 0", rpc, got)
+		}
+	}
+}
+
+func TestCreate_libvirtNotFoundIsTransientNotNotFound(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		rpc  string
+		err  error
+	}{
+		{name: "volume vanished", rpc: "StorageVolGetPath", err: fakeErr(libvirt.ErrNoStorageVol, "Storage volume not found")},
+		{name: "domain vanished", rpc: "DomainGetState", err: fakeErr(libvirt.ErrNoDomain, "Domain not found")},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			fx.fake.failWith(tc.rpc, tc.err)
+			_, err := fx.p.Create(fx.ctx, fx.req(testName))
+			assertProviderError(t, err, "create", provider.ErrTransient)
+			if errors.Is(err, provider.ErrNotFound) {
+				t.Errorf("Create error = %v, must never be NotFound", err)
+			}
+		})
+	}
+}
+
+func TestCreate_cloudInitSentinelNeverLoggedOrInErrors(t *testing.T) {
+	t.Parallel()
+	const sentinel = "SENTINEL-s3cr3t-passw0rd-7f3a"
+	fx := newFixture(t, 2)
+	fx.dialer.backend = observe(fx.fake, testProvider, testHost)
+	fx.fake.failWith("StorageVolUpload", fakeErr(libvirt.ErrInternalError, "internal error: upload failed"))
+	req := fx.req(testName)
+	req.CloudInit = "#cloud-config\nwrite_files:\n  - content: " + sentinel + "\n"
+	_, err := fx.p.Create(fx.ctx, req)
+	assertProviderError(t, err, "create", provider.ErrTransient)
+	if strings.Contains(err.Error(), sentinel) {
+		t.Errorf("error carries the cloud-init: %v", err)
+	}
+	lines := fx.logs.all()
+	if len(lines) == 0 {
+		t.Fatal("no log lines captured at V(2)")
+	}
+	for _, l := range lines {
+		if strings.Contains(l, sentinel) {
+			t.Errorf("log line carries the cloud-init: %s", l)
+		}
+	}
+	if got := fx.logs.matching(`"cloudInitBytes"=` + strconv.Itoa(len(req.CloudInit))); len(got) != 1 {
+		t.Errorf("cloudInitBytes line = %d, want 1:\n%s", len(got), strings.Join(lines, "\n"))
+	}
+}
+
+func TestCreate_metadataWrittenInDefineCall(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	fx.p.now = func() time.Time { return time.Date(2026, 9, 16, 12, 0, 0, 0, time.UTC) }
+	fx.create(testName)
+	fx.withDomain(testName, func(d *fakeDomain) {
+		for _, want := range []string{`xmlns="` + metadataURI + `"`, "<uid>" + testUID + "</uid>", "<managed>true</managed>", "<createdAt>2026-09-16T12:00:00Z</createdAt>"} {
+			if !strings.Contains(d.xml, want) {
+				t.Errorf("defined XML lacks %s:\n%s", want, d.xml)
+			}
+		}
+		meta, err := parseInstanceMetadata(d.metadata)
+		if err != nil {
+			t.Fatalf("metadata the define stored does not parse: %v", err)
+		}
+		if !meta.managed() || meta.UID != testUID || meta.createdAt().IsZero() {
+			t.Errorf("metadata = %+v, want managed, uid %s and a createdAt", meta, testUID)
+		}
+	})
+}
+
+func TestCreate_seedAlwaysRebuilt(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	fx.create(testName)
+	seed := seedVolumeName(testName)
+	first := fx.fake.vol(seed)
+	if first == nil {
+		t.Fatal("no seed volume after Create")
+	}
+	uploads := fx.fake.callCount("StorageVolUpload")
+
+	// A crash between the seed upload and the define leaves the volumes
+	// without a domain; the next attempt must not trust the seed.
+	fx.fake.mu.Lock()
+	delete(fx.fake.domains, testName)
+	fx.fake.mu.Unlock()
+	fx.create(testName)
+	second := fx.fake.vol(seed)
+	if second == nil || second == first {
+		t.Errorf("seed volume reused across attempts")
+	}
+	if got := fx.fake.callCount("StorageVolUpload"); got != uploads+1 {
+		t.Errorf("StorageVolUpload calls = %d, want %d", got, uploads+1)
+	}
+	if len(second.content) != len(first.content) || !bytes.Contains(second.content, []byte(seedLabel)) {
+		t.Errorf("rebuilt seed is %d bytes, first was %d; want a complete seed of the same size", len(second.content), len(first.content))
+	}
+}
@@ -0,0 +1,98 @@
+package kvm2
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/digitalocean/go-libvirt"
+	"github.com/go-logr/logr"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// undefineFlags removes everything libvirt keeps for a domain beside its
+// definition; UEFI domains keep NVRAM and undefine refuses without the flag.
+const undefineFlags = libvirt.DomainUndefineManagedSave | libvirt.DomainUndefineSnapshotsMetadata | libvirt.DomainUndefineNvram
+
+// Delete is spec entry 11: under the per-name lock, destroy and undefine
+// the domain, then remove its overlay and seed. Everything already gone is
+// success, and nothing is left when nil is returned, so Get afterwards is
+// NotFound. The base image stays: it is a shared cache.
+func (p *Provider) Delete(ctx context.Context, providerID string) error {
+	hostName, name, err := parseProviderID(providerID)
+	if err != nil {
+		return provider.Wrap(provider.ErrPermanent, "delete", p.name, providerID, err)
+	}
+	log := p.logger(ctx).WithValues("host", hostName, "name", name)
+	host, err := p.host(hostName)
+	if err != nil {
+		log.V(1).Info("host is not configured; delete has nothing to clean up", "id", providerID)
+		return nil
+	}
+	unlock, err := p.lockName(ctx, name)
+	if err != nil {
+		return p.wrapErr("delete", providerID, err)
+	}
+	defer unlock()
+
+	be, err := p.hosts.get(ctx, host)
+	if err != nil {
+		logAPIError(log, "delete", err)
+		return p.wrapErr("delete", providerID, err)
+	}
+	if err := removeDomain(ctx, log, be, name); err != nil {
+		logAPIError(log, "delete", err)
+		return p.wrapErr("delete", providerID, err)
+	}
+	if err := removeVolumes(ctx, log, be, host.Pool, name); err != nil {
+		logAPIError(log, "delete", err)
+		return p.wrapErr("delete", providerID, err)
+	}
+	log.V(1).Info("kvm2 domain deleted")
+	return nil
+}
+
+func removeDomain(ctx context.Context, log logr.Logger, be backend, name string) error {
+	dom, err := be.DomainLookupByName(ctx, name)
+	if hasCode(err, libvirt.ErrNoDomain) {
+		log.V(1).Info("domain already gone")
+		return nil
+	}
+	if err != nil {
+		return fmt.Errorf("looking up domain %s: %w", name, err)
+	}
+	switch err := be.DomainDestroy(ctx, dom); {
+	case err == nil:
+		log.V(1).Info("domain destroyed")
+	case hasCode(err, libvirt.ErrOperationInvalid):
+		log.V(1).Info("domain not running, destroy skipped")
+	case hasCode(err, libvirt.ErrNoDomain):
+		log.V(1).Info("domain already gone")
+		return nil
+	default:
+		return fmt.Errorf("destroying domain %s: %w", name, err)
+	}
+	if err := be.DomainUndefineFlags(ctx, dom, undefineFlags); err != nil && !hasCode(err, libvirt.ErrNoDomain) {
+		return fmt.Errorf("undefining domain %s: %w", name, err)
+	}
+	log.V(1).Info("domain undefined")
+	return nil
+}
+
+func removeVolumes(ctx context.Context, log logr.Logger, be backend, poolName, name string) error {
+	pool, err := be.StoragePoolLookupByName(ctx, poolName)
+	if hasCode(err, libvirt.ErrNoStoragePool) {
+		log.V(1).Info("pool not found, no volumes to delete", "pool", poolName)
+		return nil
+	}
+	if err != nil {
+		return fmt.Errorf("looking up storage pool %s: %w", poolName, err)
+	}
+	for _, vol := range []string{overlayVolumeName(name), seedVolumeName(name)} {
+		if err := deleteVolumeIfPresent(ctx, be, pool, vol); err != nil {
+			return err
+		}
+	}
+	log.V(1).Info("per-VM volumes deleted")
+	return nil
+}
@@ -0,0 +1,237 @@
+package kvm2
+
+import (
+	"errors"
+	"slices"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+func TestDelete_malformedIDIsPermanent(t *testing.T) {
+	t.Parallel()
+	for _, id := range []string{"", "proxy-x", "hosts/beelink/proxy-x", "hosts//domains/proxy-x", "hosts/beelink/domains/"} {
+		t.Run(id, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			err := fx.p.Delete(fx.ctx, id)
+			assertProviderError(t, err, "delete", provider.ErrPermanent)
+			var pe *provider.Error
+			if errors.As(err, &pe) && pe.ID != id {
+				t.Errorf("error.ID = %q, want %q", pe.ID, id)
+			}
+			if got := fx.dialer.dials.Load(); got != 0 {
+				t.Errorf("dials = %d, want 0", got)
+			}
+		})
+	}
+}
+
+func TestDelete_twiceSucceeds(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	for i := range 2 {
+		if err := fx.p.Delete(fx.ctx, id); err != nil {
+			t.Fatalf("Delete #%d error = %v", i+1, err)
+		}
+	}
+	if fx.hasDomain(testName) {
+		t.Error("domain still defined after Delete")
+	}
+	if got := fx.logs.matching("domain already gone"); len(got) != 1 {
+		t.Errorf("'already gone' lines = %d, want 1 from the second Delete", len(got))
+	}
+	if got := fx.logs.matching("kvm2 domain deleted"); len(got) != 2 {
+		t.Errorf("'deleted' lines = %d, want 2", len(got))
+	}
+}
+
+func TestDelete_removesOverlayAndSeed(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	if err := fx.p.Delete(fx.ctx, id); err != nil {
+		t.Fatalf("Delete() error = %v", err)
+	}
+	for _, name := range []string{overlayVolumeName(testName), seedVolumeName(testName)} {
+		if fx.fake.vol(name) != nil {
+			t.Errorf("volume %s still present after Delete", name)
+		}
+	}
+	base := "base-" + testProvider + "-debian-13.qcow2"
+	for _, name := range []string{base, base + ".sha256"} {
+		if fx.fake.vol(name) == nil {
+			t.Errorf("shared base volume %s removed by Delete", name)
+		}
+	}
+}
+
+func TestDelete_runningDomainDestroyedFirst(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	before := len(fx.fake.callLog())
+	if err := fx.p.Delete(fx.ctx, id); err != nil {
+		t.Fatalf("Delete() error = %v", err)
+	}
+	calls := fx.fake.callLog()[before:]
+	destroy := slices.Index(calls, "DomainDestroy")
+	undefine := slices.Index(calls, "DomainUndefineFlags")
+	volDelete := slices.Index(calls, "StorageVolDelete")
+	if destroy < 0 || undefine < 0 || volDelete < 0 || !(destroy < undefine && undefine < volDelete) {
+		t.Errorf("call order = %v, want destroy before undefine before volume deletes", calls)
+	}
+	if n := strings.Count(strings.Join(calls, " "), "StorageVolDelete"); n != 2 {
+		t.Errorf("StorageVolDelete calls = %d, want 2 (overlay and seed)", n)
+	}
+}
+
+func TestDelete_stoppedDomainSkipsDestroy(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	fx.withDomain(testName, func(d *fakeDomain) { d.state = libvirt.DomainShutoff })
+	if err := fx.p.Delete(fx.ctx, id); err != nil {
+		t.Fatalf("Delete() error = %v", err)
+	}
+	if fx.hasDomain(testName) {
+		t.Error("domain still defined after Delete")
+	}
+	if got := fx.logs.matching("domain not running, destroy skipped"); len(got) != 1 {
+		t.Errorf("'destroy skipped' lines = %d, want 1:\n%s", len(got), strings.Join(fx.logs.all(), "\n"))
+	}
+	if got := fx.fake.callCount("DomainUndefineFlags"); got != 1 {
+		t.Errorf("DomainUndefineFlags calls = %d, want 1", got)
+	}
+}
+
+func TestDelete_toleratesMissingVolumesAndPool(t *testing.T) {
+	t.Parallel()
+	t.Run("domain without volumes", func(t *testing.T) {
+		t.Parallel()
+		fx := newFixture(t, 1)
+		fx.fake.addDomain(testName, libvirt.DomainRunning, managedMetadata(testUID, time.Now()))
+		if err := fx.p.Delete(fx.ctx, formatProviderID(testHost, testName)); err != nil {
+			t.Fatalf("Delete() error = %v", err)
+		}
+		if fx.hasDomain(testName) {
+			t.Error("domain still defined after Delete")
+		}
+		if got := fx.fake.callCount("StorageVolDelete"); got != 0 {
+			t.Errorf("StorageVolDelete calls = %d, want 0 for absent volumes", got)
+		}
+	})
+	t.Run("pool gone", func(t *testing.T) {
+		t.Parallel()
+		fx := newFixture(t, 1)
+		fx.p.cfg.Hosts[0].Pool = "retired"
+		fx.fake.addDomain(testName, libvirt.DomainShutoff, managedMetadata(testUID, time.Now()))
+		if err := fx.p.Delete(fx.ctx, formatProviderID(testHost, testName)); err != nil {
+			t.Fatalf("Delete() error = %v", err)
+		}
+		if fx.hasDomain(testName) {
+			t.Error("domain still defined after Delete")
+		}
+		if got := fx.logs.matching("pool not found, no volumes to delete"); len(got) != 1 || !strings.Contains(got[0], `"pool"="retired"`) {
+			t.Errorf("'pool not found' V(1) line = %v", got)
+		}
+	})
+	t.Run("nothing at all", func(t *testing.T) {
+		t.Parallel()
+		fx := newFixture(t, 1)
+		if err := fx.p.Delete(fx.ctx, formatProviderID(testHost, testName)); err != nil {
+			t.Fatalf("Delete() error = %v", err)
+		}
+	})
+}
+
+func TestDelete_unconfiguredHostIsNoOp(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := formatProviderID("nas9", testName)
+	if err := fx.p.Delete(fx.ctx, id); err != nil {
+		t.Fatalf("Delete() error = %v, want nil", err)
+	}
+	if got := fx.dialer.dials.Load(); got != 0 {
+		t.Errorf("dials = %d, want 0", got)
+	}
+	if got := fx.fake.callLog(); len(got) != 0 {
+		t.Errorf("backend calls = %v, want none", got)
+	}
+	if got := fx.logs.matching("host is not configured; delete has nothing to clean up"); len(got) != 1 || !strings.Contains(got[0], `"host"="nas9"`) {
+		t.Errorf("V(1) line = %v, want one naming the host", got)
+	}
+}
+
+func TestDelete_noOpLinesAreSilentAtV0(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 0)
+	fx.p.cfg.Hosts[0].Pool = "retired"
+	for _, id := range []string{formatProviderID("nas9", testName), formatProviderID(testHost, testName)} {
+		if err := fx.p.Delete(fx.ctx, id); err != nil {
+			t.Fatalf("Delete(%s) error = %v, want nil", id, err)
+		}
+	}
+	if lines := fx.logs.all(); len(lines) != 0 {
+		t.Errorf("Delete logged %d lines at V(0), want none:\n%s", len(lines), strings.Join(lines, "\n"))
+	}
+}
+
+func TestDelete_concurrentSameNameSafe(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	const n = 10
+	errs := make([]error, n)
+	var wg sync.WaitGroup
+	for i := range n {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			errs[i] = fx.p.Delete(fx.ctx, id)
+		}()
+	}
+	wg.Wait()
+	for i, err := range errs {
+		if err != nil {
+			t.Errorf("Delete #%d error = %v, want nil", i, err)
+		}
+	}
+	if fx.hasDomain(testName) {
+		t.Error("domain still defined after concurrent Deletes")
+	}
+	for _, name := range []string{overlayVolumeName(testName), seedVolumeName(testName)} {
+		if fx.fake.vol(name) != nil {
+			t.Errorf("volume %s still present", name)
+		}
+	}
+	if got := fx.fake.callCount("DomainUndefineFlags"); got != 1 {
+		t.Errorf("DomainUndefineFlags calls = %d, want 1 (the lock serialises the rest)", got)
+	}
+}
+
+func TestDelete_undefineUsesNvramFlag(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	if err := fx.p.Delete(fx.ctx, id); err != nil {
+		t.Fatalf("Delete() error = %v", err)
+	}
+	fx.fake.mu.Lock()
+	flags, ok := fx.fake.undefineFlags[testName]
+	fx.fake.mu.Unlock()
+	if !ok {
+		t.Fatal("DomainUndefineFlags was not called")
+	}
+	for _, want := range []libvirt.DomainUndefineFlagsValues{libvirt.DomainUndefineNvram, libvirt.DomainUndefineManagedSave, libvirt.DomainUndefineSnapshotsMetadata} {
+		if flags&want == 0 {
+			t.Errorf("undefine flags = %d, lack %d", flags, want)
+		}
+	}
+}
@@ -0,0 +1,97 @@
+package kvm2
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+
+	"github.com/digitalocean/go-libvirt"
+	"golang.org/x/crypto/ssh/knownhosts"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// permanentf and transientf classify an error at the source, so classify
+// keeps that class no matter what wraps it later.
+func permanentf(format string, args ...any) error {
+	return fmt.Errorf("%w: %s", provider.ErrPermanent, fmt.Sprintf(format, args...))
+}
+
+func transientf(format string, args ...any) error {
+	return fmt.Errorf("%w: %s", provider.ErrTransient, fmt.Sprintf(format, args...))
+}
+
+// sshAuthFailed is the substring x/crypto/ssh puts in every authentication
+// failure ("ssh: handshake failed: ssh: unable to authenticate, attempted
+// methods [...]").
+const sshAuthFailed = "unable to authenticate"
+
+// classify maps any error reaching the provider's boundary to one of the
+// four contract sentinels, per spec entry 13. op is the contract operation:
+// a vanished domain or volume inside Create is Transient, because the
+// contract forbids NotFound from Create.
+func classify(op string, err error) error {
+	if err == nil {
+		return provider.ErrTransient
+	}
+	for _, class := range []error{provider.ErrNotFound, provider.ErrQuotaExceeded, provider.ErrPermanent, provider.ErrTransient} {
+		if errors.Is(err, class) {
+			return class
+		}
+	}
+	if code, ok := libvirtCode(err); ok {
+		return classifyLibvirt(op, code)
+	}
+	var keyErr *knownhosts.KeyError
+	if errors.As(err, &keyErr) {
+		return provider.ErrPermanent
+	}
+	if strings.Contains(err.Error(), sshAuthFailed) {
+		return provider.ErrPermanent
+	}
+	// Everything else retries: a missing key or known_hosts file (the
+	// optional Secret has not landed yet), context cancellation and
+	// deadlines, net.OpError, io.EOF, and anything unrecognised.
+	return provider.ErrTransient
+}
+
+func classifyLibvirt(op string, code libvirt.ErrorNumber) error {
+	switch code {
+	case libvirt.ErrNoDomain, libvirt.ErrNoStorageVol:
+		if op == "create" {
+			return provider.ErrTransient
+		}
+		return provider.ErrNotFound
+	case libvirt.ErrNoStoragePool, libvirt.ErrXMLError, libvirt.ErrNoSupport, libvirt.ErrInvalidArg,
+		libvirt.ErrAuthFailed, libvirt.ErrAccessDenied:
+		return provider.ErrPermanent
+	default:
+		return provider.ErrTransient
+	}
+}
+
+// libvirtCode extracts the libvirt error number from err, if it is or wraps
+// a libvirt.Error. That type has a value receiver, so errors.As needs a
+// value target.
+func libvirtCode(err error) (libvirt.ErrorNumber, bool) {
+	var le libvirt.Error
+	if !errors.As(err, &le) {
+		return 0, false
+	}
+	return libvirt.ErrorNumber(le.Code), true
+}
+
+func hasCode(err error, code libvirt.ErrorNumber) bool {
+	got, ok := libvirtCode(err)
+	return ok && got == code
+}
+
+// libvirtErrorFields returns the key/value pairs a log line carries for a
+// libvirt error, or nil when err is not one.
+func libvirtErrorFields(err error) []any {
+	code, ok := libvirtCode(err)
+	if !ok {
+		return nil
+	}
+	return []any{"libvirtCode", int(code), "libvirtError", code.String()}
+}
@@ -0,0 +1,146 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"io/fs"
+	"net"
+	"os"
+	"testing"
+
+	"github.com/digitalocean/go-libvirt"
+	"golang.org/x/crypto/ssh"
+	"golang.org/x/crypto/ssh/knownhosts"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+func libvirtErr(code libvirt.ErrorNumber, msg string) error {
+	return libvirt.Error{Code: uint32(code), Message: msg}
+}
+
+// Captured from x/crypto/ssh when the offered key is not in authorized_keys.
+const sshAuthFailedMsg = "ssh: handshake failed: ssh: unable to authenticate, attempted methods [none publickey], no supported methods remain"
+
+func TestClassify_table(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		op   string
+		err  error
+		want error
+	}{
+		{name: "nil is transient", op: "get", err: nil, want: provider.ErrTransient},
+		{name: "ErrNoDomain on get", op: "get", err: libvirtErr(libvirt.ErrNoDomain, "Domain not found: no domain with matching name 'proxy-x'"), want: provider.ErrNotFound},
+		{name: "ErrNoDomain on delete", op: "delete", err: libvirtErr(libvirt.ErrNoDomain, "Domain not found"), want: provider.ErrNotFound},
+		{name: "ErrNoDomain on create is transient", op: "create", err: libvirtErr(libvirt.ErrNoDomain, "Domain not found"), want: provider.ErrTransient},
+		{name: "ErrNoStorageVol on get", op: "get", err: libvirtErr(libvirt.ErrNoStorageVol, "Storage volume not found: no storage vol with matching name 'proxy-x.qcow2'"), want: provider.ErrNotFound},
+		{name: "ErrNoStorageVol on create is transient", op: "create", err: libvirtErr(libvirt.ErrNoStorageVol, "Storage volume not found"), want: provider.ErrTransient},
+		{name: "ErrNoStoragePool", op: "create", err: libvirtErr(libvirt.ErrNoStoragePool, "Storage pool not found: no storage pool with matching name 'default'"), want: provider.ErrPermanent},
+		{name: "ErrXMLError", op: "create", err: libvirtErr(libvirt.ErrXMLError, "XML error: missing domain name"), want: provider.ErrPermanent},
+		{name: "ErrNoSupport", op: "create", err: libvirtErr(libvirt.ErrNoSupport, "this function is not supported by the connection driver"), want: provider.ErrPermanent},
+		{name: "ErrInvalidArg", op: "get", err: libvirtErr(libvirt.ErrInvalidArg, "invalid argument: unsupported flags"), want: provider.ErrPermanent},
+		{name: "ErrAuthFailed", op: "list", err: libvirtErr(libvirt.ErrAuthFailed, "authentication failed: access denied by policy"), want: provider.ErrPermanent},
+		{name: "ErrAccessDenied", op: "list", err: libvirtErr(libvirt.ErrAccessDenied, "access denied"), want: provider.ErrPermanent},
+		{name: "ErrOperationInvalid is transient", op: "create", err: libvirtErr(libvirt.ErrOperationInvalid, "Requested operation is not valid: domain is already running"), want: provider.ErrTransient},
+		{name: "ErrInternalError is transient", op: "get", err: libvirtErr(libvirt.ErrInternalError, "internal error: connection closed"), want: provider.ErrTransient},
+		{name: "unknown libvirt code is transient", op: "get", err: libvirtErr(libvirt.ErrorNumber(9999), "?"), want: provider.ErrTransient},
+		{name: "wrapped libvirt error keeps its code", op: "get", err: fmt.Errorf("looking up domain: %w", libvirtErr(libvirt.ErrNoDomain, "Domain not found")), want: provider.ErrNotFound},
+		{name: "knownhosts unknown host", op: "get", err: &knownhosts.KeyError{}, want: provider.ErrPermanent},
+		{name: "knownhosts mismatch", op: "get", err: &knownhosts.KeyError{Want: []knownhosts.KnownKey{{Filename: "/kh", Line: 1, Key: testPublicKey(t)}}}, want: provider.ErrPermanent},
+		{name: "wrapped knownhosts error", op: "create", err: fmt.Errorf("dialing beelink: %w", &knownhosts.KeyError{}), want: provider.ErrPermanent},
+		{name: "ssh unable to authenticate", op: "create", err: errors.New(sshAuthFailedMsg), want: provider.ErrPermanent},
+		{name: "wrapped ssh unable to authenticate", op: "list", err: fmt.Errorf("dialing beelink: %w", errors.New(sshAuthFailedMsg)), want: provider.ErrPermanent},
+		{name: "missing key file is transient", op: "create", err: fmt.Errorf("reading ssh key file /k: %w", fs.ErrNotExist), want: provider.ErrTransient},
+		{name: "os.PathError not exist is transient", op: "create", err: &os.PathError{Op: "open", Path: "/k", Err: fs.ErrNotExist}, want: provider.ErrTransient},
+		{name: "context canceled", op: "get", err: context.Canceled, want: provider.ErrTransient},
+		{name: "context deadline", op: "get", err: context.DeadlineExceeded, want: provider.ErrTransient},
+		{name: "net.OpError", op: "get", err: &net.OpError{Op: "dial", Net: "tcp", Err: errors.New("connection refused")}, want: provider.ErrTransient},
+		{name: "io.EOF", op: "list", err: io.EOF, want: provider.ErrTransient},
+		{name: "io.ErrUnexpectedEOF", op: "list", err: io.ErrUnexpectedEOF, want: provider.ErrTransient},
+		{name: "anything else is transient", op: "get", err: errors.New("anything"), want: provider.ErrTransient},
+		{name: "pre-classified NotFound stays", op: "create", err: fmt.Errorf("x: %w", provider.ErrNotFound), want: provider.ErrNotFound},
+		{name: "pre-classified Quota stays", op: "create", err: fmt.Errorf("x: %w", provider.ErrQuotaExceeded), want: provider.ErrQuotaExceeded},
+		{name: "pre-classified Permanent stays", op: "get", err: fmt.Errorf("x: %w", permanentf("bad %s", "thing")), want: provider.ErrPermanent},
+		{name: "pre-classified Transient stays", op: "get", err: fmt.Errorf("x: %w", transientf("blip %d", 1)), want: provider.ErrTransient},
+		{name: "pre-classified beats a libvirt code underneath", op: "get", err: fmt.Errorf("%w: %w", provider.ErrPermanent, libvirtErr(libvirt.ErrNoDomain, "x")), want: provider.ErrPermanent},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			if got := classify(tc.op, tc.err); !errors.Is(got, tc.want) {
+				t.Errorf("classify(%q, %v) = %v, want %v", tc.op, tc.err, got, tc.want)
+			}
+		})
+	}
+}
+
+func TestHasCode(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		err  error
+		code libvirt.ErrorNumber
+		want bool
+	}{
+		{name: "matching code", err: libvirtErr(libvirt.ErrNoDomain, "x"), code: libvirt.ErrNoDomain, want: true},
+		{name: "matching code wrapped", err: fmt.Errorf("w: %w", libvirtErr(libvirt.ErrStorageVolExist, "x")), code: libvirt.ErrStorageVolExist, want: true},
+		{name: "other code", err: libvirtErr(libvirt.ErrNoDomain, "x"), code: libvirt.ErrNoStorageVol, want: false},
+		{name: "not a libvirt error", err: errors.New("x"), code: libvirt.ErrNoDomain, want: false},
+		{name: "nil", err: nil, code: libvirt.ErrNoDomain, want: false},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			if got := hasCode(tc.err, tc.code); got != tc.want {
+				t.Errorf("hasCode(%v, %v) = %v, want %v", tc.err, tc.code, got, tc.want)
+			}
+		})
+	}
+}
+
+func TestLibvirtErrorFields(t *testing.T) {
+	t.Parallel()
+	got := libvirtErrorFields(fmt.Errorf("w: %w", libvirtErr(libvirt.ErrNoDomain, "x")))
+	want := []any{"libvirtCode", 42, "libvirtError", "ErrNoDomain"}
+	if len(got) != len(want) {
+		t.Fatalf("libvirtErrorFields() = %v, want %v", got, want)
+	}
+	for i := range want {
+		if got[i] != want[i] {
+			t.Errorf("libvirtErrorFields()[%d] = %v (%T), want %v (%T)", i, got[i], got[i], want[i], want[i])
+		}
+	}
+	if fields := libvirtErrorFields(errors.New("x")); fields != nil {
+		t.Errorf("libvirtErrorFields(non-libvirt) = %v, want nil", fields)
+	}
+	if fields := libvirtErrorFields(libvirtErr(libvirt.ErrorNumber(9999), "x")); fields[3] != "ErrorNumber(9999)" {
+		t.Errorf("libvirtErrorFields(unknown code)[3] = %v, want ErrorNumber(9999)", fields[3])
+	}
+}
+
+func TestPermanentfAndTransientf(t *testing.T) {
+	t.Parallel()
+	perm := permanentf("host %q missing", "x")
+	if !errors.Is(perm, provider.ErrPermanent) || perm.Error() != `provider: permanent error: host "x" missing` {
+		t.Errorf("permanentf() = %q, want ErrPermanent with the message", perm)
+	}
+	trans := transientf("retry %d", 3)
+	if !errors.Is(trans, provider.ErrTransient) || trans.Error() != "provider: transient error: retry 3" {
+		t.Errorf("transientf() = %q, want ErrTransient with the message", trans)
+	}
+}
+
+// testPublicKey parses a fixed ed25519 public key so a knownhosts.KeyError
+// with a Want entry can be built without generating keys.
+func testPublicKey(t *testing.T) ssh.PublicKey {
+	t.Helper()
+	const line = "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIGcZBhnMPTyDwFn9lBhwIutqLb2bSlhcZoqpuWv7VuKI test"
+	key, _, _, _, err := ssh.ParseAuthorizedKey([]byte(line))
+	if err != nil {
+		t.Fatalf("ParseAuthorizedKey: %v", err)
+	}
+	return key
+}
@@ -0,0 +1,623 @@
+package kvm2
+
+import (
+	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"encoding/xml"
+	"fmt"
+	"io"
+	"slices"
+	"sync"
+	"sync/atomic"
+
+	"github.com/digitalocean/go-libvirt"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// fakeBackend is the in-memory backend of spec entry 21: enough of a
+// libvirt host for the provider's Create, Get, Delete and ListByTag, with
+// per-method error hooks and blocking, call counters and an ordered log.
+type fakeBackend struct {
+	// mu, domains and pools are the host's state; a twin shares them, so
+	// two connections to the same fake host see the same volumes.
+	mu       *sync.Mutex
+	domains  map[string]*fakeDomain
+	pools    map[string]*fakePool
+	hostname string
+	version  uint64
+
+	// hooks returns the error a method fails with before touching state.
+	hooks map[string]func() error
+	// block makes a method wait on its channel (or ctx) before running.
+	block map[string]chan struct{}
+	// discardUploads makes StorageVolUpload stream the reader through a
+	// sha256 in fixed chunks instead of holding the bytes, for tests that
+	// push more through the fake than they want in memory.
+	discardUploads bool
+	// startDelayPolls, when set, makes DomainCreate leave the domain in
+	// nostate; it turns running on the startDelayPolls-th DomainGetState
+	// after that, as a host whose QEMU is still settling would.
+	startDelayPolls int
+	// stateAfterStart, when not nostate, is what DomainCreate leaves the
+	// domain in instead of running, as a guest that dies at once would.
+	stateAfterStart libvirt.DomainState
+	// undefineFlags records the flags of the last DomainUndefineFlags per
+	// domain name.
+	undefineFlags map[string]libvirt.DomainUndefineFlagsValues
+
+	calls  map[string]int
+	log    []string
+	closed int
+	deadCh chan struct{}
+	deadMu sync.Once
+}
+
+type fakeDomain struct {
+	state    libvirt.DomainState
+	xml      string
+	metadata string
+	addrs    map[uint32][]libvirt.DomainInterface
+	uuid     libvirt.UUID
+	id       int32
+	// pendingStart counts the DomainGetState calls left before a delayed
+	// start (see startDelayPolls) turns the domain running.
+	pendingStart int
+}
+
+type fakePool struct {
+	state     libvirt.StoragePoolState
+	capacity  uint64
+	available uint64
+	vols      map[string]*fakeVol
+}
+
+type fakeVol struct {
+	content  []byte
+	capacity uint64
+	// uploaded and sha256 record a discarded upload (see discardUploads).
+	uploaded int64
+	sha256   string
+}
+
+const fakePoolName = "default"
+
+func newFakeBackend() *fakeBackend {
+	return &fakeBackend{
+		mu:            &sync.Mutex{},
+		domains:       make(map[string]*fakeDomain),
+		pools:         map[string]*fakePool{fakePoolName: newFakePool()},
+		hostname:      "fake-host",
+		version:       11006000,
+		hooks:         make(map[string]func() error),
+		block:         make(map[string]chan struct{}),
+		undefineFlags: make(map[string]libvirt.DomainUndefineFlagsValues),
+		calls:         make(map[string]int),
+		deadCh:        make(chan struct{}),
+	}
+}
+
+// twin returns a second connection to the same fake host: it shares the
+// domains and pools (and their lock) but has its own hooks, counters, log,
+// close count and liveness, so a test can tell which connection did what.
+func (f *fakeBackend) twin() *fakeBackend {
+	return &fakeBackend{
+		mu:             f.mu,
+		domains:        f.domains,
+		pools:          f.pools,
+		hostname:       f.hostname,
+		version:        f.version,
+		hooks:          make(map[string]func() error),
+		block:          make(map[string]chan struct{}),
+		discardUploads: f.discardUploads,
+		undefineFlags:  make(map[string]libvirt.DomainUndefineFlagsValues),
+		calls:          make(map[string]int),
+		deadCh:         make(chan struct{}),
+	}
+}
+
+// vol returns the volume by name in the default pool, or nil.
+func (f *fakeBackend) vol(name string) *fakeVol {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return f.pools[fakePoolName].vols[name]
+}
+
+// closeCount is closed read under the lock, for twins closed by another
+// goroutine.
+func (f *fakeBackend) closeCount() int {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return f.closed
+}
+
+func newFakePool() *fakePool {
+	return &fakePool{state: libvirt.StoragePoolRunning, capacity: 100 << 30, available: 50 << 30, vols: make(map[string]*fakeVol)}
+}
+
+func (f *fakeBackend) dead() <-chan struct{} { return f.deadCh }
+
+func (f *fakeBackend) markDead() { f.deadMu.Do(func() { close(f.deadCh) }) }
+
+// addDomain defines a domain directly, bypassing DomainDefineXML.
+func (f *fakeBackend) addDomain(name string, state libvirt.DomainState, metadata string) *fakeDomain {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	d := &fakeDomain{state: state, metadata: metadata, addrs: make(map[uint32][]libvirt.DomainInterface), uuid: fakeUUID(name)}
+	f.domains[name] = d
+	return d
+}
+
+func (f *fakeBackend) failWith(method string, err error) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	f.hooks[method] = func() error { return err }
+}
+
+func (f *fakeBackend) blockOn(method string) chan struct{} {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	ch := make(chan struct{})
+	f.block[method] = ch
+	return ch
+}
+
+func (f *fakeBackend) callCount(method string) int {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return f.calls[method]
+}
+
+func (f *fakeBackend) callLog() []string {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return slices.Clone(f.log)
+}
+
+func fakeUUID(name string) libvirt.UUID {
+	var u libvirt.UUID
+	copy(u[:], name)
+	return u
+}
+
+func fakeErr(code libvirt.ErrorNumber, format string, args ...any) error {
+	return libvirt.Error{Code: uint32(code), Message: fmt.Sprintf(format, args...)}
+}
+
+// enter records the call, honours a block and runs the hook. It is called
+// without f.mu held; state access happens afterwards under the lock.
+func (f *fakeBackend) enter(ctx context.Context, method string) error {
+	f.mu.Lock()
+	f.calls[method]++
+	f.log = append(f.log, method)
+	block := f.block[method]
+	hook := f.hooks[method]
+	f.mu.Unlock()
+	if block != nil {
+		select {
+		case <-block:
+		case <-ctx.Done():
+			return ctx.Err()
+		}
+	}
+	if hook != nil {
+		return hook()
+	}
+	return nil
+}
+
+func (f *fakeBackend) domain(name string) (*fakeDomain, error) {
+	d, ok := f.domains[name]
+	if !ok {
+		return nil, fakeErr(libvirt.ErrNoDomain, "Domain not found: no domain with matching name '%s'", name)
+	}
+	return d, nil
+}
+
+func (f *fakeBackend) pool(name string) (*fakePool, error) {
+	p, ok := f.pools[name]
+	if !ok {
+		return nil, fakeErr(libvirt.ErrNoStoragePool, "Storage pool not found: no storage pool with matching name '%s'", name)
+	}
+	return p, nil
+}
+
+func (f *fakeBackend) volume(vol libvirt.StorageVol) (*fakeVol, error) {
+	p, err := f.pool(vol.Pool)
+	if err != nil {
+		return nil, err
+	}
+	v, ok := p.vols[vol.Name]
+	if !ok {
+		return nil, fakeErr(libvirt.ErrNoStorageVol, "Storage volume not found: no storage vol with matching name '%s'", vol.Name)
+	}
+	return v, nil
+}
+
+func (f *fakeBackend) libvirtDomain(name string, d *fakeDomain) libvirt.Domain {
+	id := int32(-1)
+	if isActive(d.state) {
+		id = d.id
+	}
+	return libvirt.Domain{Name: name, UUID: d.uuid, ID: id}
+}
+
+func (f *fakeBackend) DomainLookupByName(ctx context.Context, name string) (libvirt.Domain, error) {
+	if err := f.enter(ctx, "DomainLookupByName"); err != nil {
+		return libvirt.Domain{}, err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	d, err := f.domain(name)
+	if err != nil {
+		return libvirt.Domain{}, err
+	}
+	return f.libvirtDomain(name, d), nil
+}
+
+func (f *fakeBackend) DomainDefineXML(ctx context.Context, doc string) (libvirt.Domain, error) {
+	if err := f.enter(ctx, "DomainDefineXML"); err != nil {
+		return libvirt.Domain{}, err
+	}
+	var parsed struct {
+		Name     string `xml:"name"`
+		Metadata struct {
+			Inner string `xml:",innerxml"`
+		} `xml:"metadata"`
+	}
+	if err := xml.Unmarshal([]byte(doc), &parsed); err != nil || parsed.Name == "" {
+		return libvirt.Domain{}, fakeErr(libvirt.ErrXMLError, "XML error: %v", err)
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	d, ok := f.domains[parsed.Name]
+	if !ok {
+		d = &fakeDomain{state: libvirt.DomainShutoff, addrs: make(map[uint32][]libvirt.DomainInterface), uuid: fakeUUID(parsed.Name)}
+		f.domains[parsed.Name] = d
+	}
+	d.xml = doc
+	d.metadata = parsed.Metadata.Inner
+	return f.libvirtDomain(parsed.Name, d), nil
+}
+
+func (f *fakeBackend) DomainCreate(ctx context.Context, dom libvirt.Domain) error {
+	if err := f.enter(ctx, "DomainCreate"); err != nil {
+		return err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	d, err := f.domain(dom.Name)
+	if err != nil {
+		return err
+	}
+	if isActive(d.state) {
+		return fakeErr(libvirt.ErrOperationInvalid, "Requested operation is not valid: domain is already running")
+	}
+	switch {
+	case f.startDelayPolls > 0:
+		d.state = libvirt.DomainNostate
+		d.pendingStart = f.startDelayPolls
+		return nil
+	case f.stateAfterStart != libvirt.DomainNostate:
+		d.state = f.stateAfterStart
+		return nil
+	}
+	d.state = libvirt.DomainRunning
+	d.id = int32(len(f.domains))
+	return nil
+}
+
+func (f *fakeBackend) DomainDestroy(ctx context.Context, dom libvirt.Domain) error {
+	if err := f.enter(ctx, "DomainDestroy"); err != nil {
+		return err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	d, err := f.domain(dom.Name)
+	if err != nil {
+		return err
+	}
+	if !isActive(d.state) {
+		return fakeErr(libvirt.ErrOperationInvalid, "Requested operation is not valid: domain is not running")
+	}
+	d.state = libvirt.DomainShutoff
+	return nil
+}
+
+func (f *fakeBackend) DomainUndefineFlags(ctx context.Context, dom libvirt.Domain, flags libvirt.DomainUndefineFlagsValues) error {
+	if err := f.enter(ctx, "DomainUndefineFlags"); err != nil {
+		return err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	if _, err := f.domain(dom.Name); err != nil {
+		return err
+	}
+	f.undefineFlags[dom.Name] = flags
+	delete(f.domains, dom.Name)
+	return nil
+}
+
+func (f *fakeBackend) DomainGetState(ctx context.Context, dom libvirt.Domain, _ uint32) (int32, int32, error) {
+	if err := f.enter(ctx, "DomainGetState"); err != nil {
+		return 0, 0, err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	d, err := f.domain(dom.Name)
+	if err != nil {
+		return 0, 0, err
+	}
+	if d.pendingStart > 0 {
+		d.pendingStart--
+		if d.pendingStart == 0 {
+			d.state = libvirt.DomainRunning
+			d.id = int32(len(f.domains))
+		}
+	}
+	return int32(d.state), 0, nil
+}
+
+func (f *fakeBackend) DomainGetMetadata(ctx context.Context, dom libvirt.Domain, typ int32, uri libvirt.OptString, _ libvirt.DomainModificationImpact) (string, error) {
+	if err := f.enter(ctx, "DomainGetMetadata"); err != nil {
+		return "", err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	d, err := f.domain(dom.Name)
+	if err != nil {
+		return "", err
+	}
+	if typ != int32(libvirt.DomainMetadataElement) || len(uri) != 1 || uri[0] != metadataURI || d.metadata == "" {
+		return "", fakeErr(libvirt.ErrNoDomainMetadata, "Requested metadata element is not present")
+	}
+	return d.metadata, nil
+}
+
+func (f *fakeBackend) DomainInterfaceAddresses(ctx context.Context, dom libvirt.Domain, source uint32, _ uint32) ([]libvirt.DomainInterface, error) {
+	if err := f.enter(ctx, "DomainInterfaceAddresses"); err != nil {
+		return nil, err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	d, err := f.domain(dom.Name)
+	if err != nil {
+		return nil, err
+	}
+	if !isActive(d.state) {
+		return nil, fakeErr(libvirt.ErrOperationInvalid, "Requested operation is not valid: domain is not running")
+	}
+	return slices.Clone(d.addrs[source]), nil
+}
+
+func (f *fakeBackend) ConnectListAllDomains(ctx context.Context, _ int32, _ libvirt.ConnectListAllDomainsFlags) ([]libvirt.Domain, uint32, error) {
+	if err := f.enter(ctx, "ConnectListAllDomains"); err != nil {
+		return nil, 0, err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	names := make([]string, 0, len(f.domains))
+	for name := range f.domains {
+		names = append(names, name)
+	}
+	slices.Sort(names)
+	out := make([]libvirt.Domain, 0, len(names))
+	for _, name := range names {
+		out = append(out, f.libvirtDomain(name, f.domains[name]))
+	}
+	return out, uint32(len(out)), nil
+}
+
+func (f *fakeBackend) StoragePoolLookupByName(ctx context.Context, name string) (libvirt.StoragePool, error) {
+	if err := f.enter(ctx, "StoragePoolLookupByName"); err != nil {
+		return libvirt.StoragePool{}, err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	if _, err := f.pool(name); err != nil {
+		return libvirt.StoragePool{}, err
+	}
+	return libvirt.StoragePool{Name: name, UUID: fakeUUID(name)}, nil
+}
+
+func (f *fakeBackend) StoragePoolRefresh(ctx context.Context, pool libvirt.StoragePool, _ uint32) error {
+	if err := f.enter(ctx, "StoragePoolRefresh"); err != nil {
+		return err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	_, err := f.pool(pool.Name)
+	return err
+}
+
+func (f *fakeBackend) StoragePoolGetInfo(ctx context.Context, pool libvirt.StoragePool) (uint8, uint64, uint64, uint64, error) {
+	if err := f.enter(ctx, "StoragePoolGetInfo"); err != nil {
+		return 0, 0, 0, 0, err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	p, err := f.pool(pool.Name)
+	if err != nil {
+		return 0, 0, 0, 0, err
+	}
+	return uint8(p.state), p.capacity, p.capacity - p.available, p.available, nil
+}
+
+func (f *fakeBackend) StorageVolLookupByName(ctx context.Context, pool libvirt.StoragePool, name string) (libvirt.StorageVol, error) {
+	if err := f.enter(ctx, "StorageVolLookupByName"); err != nil {
+		return libvirt.StorageVol{}, err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	vol := libvirt.StorageVol{Pool: pool.Name, Name: name, Key: fakeVolPath(pool.Name, name)}
+	if _, err := f.volume(vol); err != nil {
+		return libvirt.StorageVol{}, err
+	}
+	return vol, nil
+}
+
+func fakeVolPath(pool, name string) string { return "/pool/" + pool + "/" + name }
+
+func (f *fakeBackend) StorageVolGetPath(ctx context.Context, vol libvirt.StorageVol) (string, error) {
+	if err := f.enter(ctx, "StorageVolGetPath"); err != nil {
+		return "", err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	if _, err := f.volume(vol); err != nil {
+		return "", err
+	}
+	return fakeVolPath(vol.Pool, vol.Name), nil
+}
+
+func (f *fakeBackend) StorageVolCreateXML(ctx context.Context, pool libvirt.StoragePool, doc string, _ libvirt.StorageVolCreateFlags) (libvirt.StorageVol, error) {
+	if err := f.enter(ctx, "StorageVolCreateXML"); err != nil {
+		return libvirt.StorageVol{}, err
+	}
+	var parsed struct {
+		Name     string `xml:"name"`
+		Capacity uint64 `xml:"capacity"`
+	}
+	if err := xml.Unmarshal([]byte(doc), &parsed); err != nil || parsed.Name == "" {
+		return libvirt.StorageVol{}, fakeErr(libvirt.ErrXMLError, "XML error: %v", err)
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	p, err := f.pool(pool.Name)
+	if err != nil {
+		return libvirt.StorageVol{}, err
+	}
+	if _, exists := p.vols[parsed.Name]; exists {
+		return libvirt.StorageVol{}, fakeErr(libvirt.ErrStorageVolExist, "storage volume '%s' exists already", parsed.Name)
+	}
+	p.vols[parsed.Name] = &fakeVol{capacity: parsed.Capacity}
+	return libvirt.StorageVol{Pool: pool.Name, Name: parsed.Name, Key: fakeVolPath(pool.Name, parsed.Name)}, nil
+}
+
+func (f *fakeBackend) StorageVolUpload(ctx context.Context, vol libvirt.StorageVol, r io.Reader, _, _ uint64, _ libvirt.StorageVolUploadFlags) error {
+	if err := f.enter(ctx, "StorageVolUpload"); err != nil {
+		return err
+	}
+	var (
+		content  []byte
+		uploaded int64
+		digest   string
+	)
+	if r != nil {
+		var err error
+		if f.discardUploads {
+			h := sha256.New()
+			if uploaded, err = io.CopyBuffer(h, r, make([]byte, 64<<10)); err != nil {
+				return err
+			}
+			digest = hex.EncodeToString(h.Sum(nil))
+		} else if content, err = io.ReadAll(r); err != nil {
+			return err
+		}
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	v, err := f.volume(vol)
+	if err != nil {
+		return err
+	}
+	v.content = content
+	v.uploaded = uploaded
+	v.sha256 = digest
+	return nil
+}
+
+func (f *fakeBackend) StorageVolDownload(ctx context.Context, vol libvirt.StorageVol, w io.Writer, _, _ uint64, _ libvirt.StorageVolDownloadFlags) error {
+	if err := f.enter(ctx, "StorageVolDownload"); err != nil {
+		return err
+	}
+	f.mu.Lock()
+	v, err := f.volume(vol)
+	var content []byte
+	if v != nil {
+		content = slices.Clone(v.content)
+	}
+	f.mu.Unlock()
+	if err != nil {
+		return err
+	}
+	if w == nil {
+		return nil
+	}
+	_, err = w.Write(content)
+	return err
+}
+
+func (f *fakeBackend) StorageVolDelete(ctx context.Context, vol libvirt.StorageVol, _ libvirt.StorageVolDeleteFlags) error {
+	if err := f.enter(ctx, "StorageVolDelete"); err != nil {
+		return err
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	if _, err := f.volume(vol); err != nil {
+		return err
+	}
+	delete(f.pools[vol.Pool].vols, vol.Name)
+	return nil
+}
+
+func (f *fakeBackend) ConnectGetLibVersion(ctx context.Context) (uint64, error) {
+	if err := f.enter(ctx, "ConnectGetLibVersion"); err != nil {
+		return 0, err
+	}
+	return f.version, nil
+}
+
+func (f *fakeBackend) ConnectGetHostname(ctx context.Context) (string, error) {
+	if err := f.enter(ctx, "ConnectGetHostname"); err != nil {
+		return "", err
+	}
+	return f.hostname, nil
+}
+
+func (f *fakeBackend) Close() error {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	f.closed++
+	return nil
+}
+
+// stubDialer covers every dialer a test needs: it counts dials, fails with
+// err when set, waits on block (or ctx) when set, and hands out backend, or
+// one from fresh per call, or a new fakeBackend when neither is given.
+type stubDialer struct {
+	backend backend
+	fresh   func() backend
+	err     error
+	block   chan struct{}
+
+	dials atomic.Int32
+	mu    sync.Mutex
+	hosts []string
+}
+
+func (d *stubDialer) dial(ctx context.Context, host provider.KVMHost) (backend, error) {
+	d.dials.Add(1)
+	d.mu.Lock()
+	d.hosts = append(d.hosts, host.Name)
+	d.mu.Unlock()
+	if d.block != nil {
+		select {
+		case <-d.block:
+		case <-ctx.Done():
+			return nil, fmt.Errorf("%w: dialing %s: %w", provider.ErrTransient, host.Name, ctx.Err())
+		}
+	}
+	if d.err != nil {
+		return nil, d.err
+	}
+	switch {
+	case d.fresh != nil:
+		return d.fresh(), nil
+	case d.backend != nil:
+		return d.backend, nil
+	default:
+		return newFakeBackend(), nil
+	}
+}
@@ -0,0 +1,133 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"fmt"
+
+	"github.com/digitalocean/go-libvirt"
+	"github.com/go-logr/logr"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// Get is spec entries 7 to 9: look the domain up by name, read the mark
+// back, then state and address. Only ErrNoDomain becomes NotFound; every
+// other failure is classified, never guessed.
+func (p *Provider) Get(ctx context.Context, providerID string) (*provider.Instance, error) {
+	hostName, name, err := parseProviderID(providerID)
+	if err != nil {
+		return nil, provider.Wrap(provider.ErrPermanent, "get", p.name, providerID, err)
+	}
+	host, err := p.host(hostName)
+	if err != nil {
+		return nil, p.wrapErr("get", providerID, err)
+	}
+	log := p.logger(ctx).WithValues("host", host.Name, "name", name)
+	be, err := p.hosts.get(ctx, host)
+	if err != nil {
+		logAPIError(log, "get", err)
+		return nil, p.wrapErr("get", providerID, err)
+	}
+	inst, err := get(ctx, log, be, host.Name, name)
+	if err != nil {
+		if hasCode(err, libvirt.ErrNoDomain) {
+			log.V(1).Info("kvm2 domain not found")
+		} else {
+			logAPIError(log, "get", err)
+		}
+		return nil, p.wrapErr("get", providerID, err)
+	}
+	log.V(1).Info("kvm2 domain fetched", "state", inst.State, "ip", inst.IP, "uid", inst.UID)
+	return inst, nil
+}
+
+// get builds the Instance from the domain's metadata, its state and, for an
+// active domain, its discovered address.
+func get(ctx context.Context, log logr.Logger, be backend, hostName, name string) (*provider.Instance, error) {
+	dom, err := be.DomainLookupByName(ctx, name)
+	if err != nil {
+		return nil, fmt.Errorf("looking up domain %s: %w", name, err)
+	}
+	meta, err := readMetadata(ctx, be, dom)
+	if err != nil {
+		return nil, err
+	}
+	state, _, err := be.DomainGetState(ctx, dom, 0)
+	if err != nil {
+		return nil, fmt.Errorf("reading state of domain %s: %w", dom.Name, err)
+	}
+	s := libvirt.DomainState(state)
+	var ip string
+	if isActive(s) {
+		if ip, err = address(ctx, log, be, dom); err != nil {
+			return nil, err
+		}
+	}
+	return &provider.Instance{
+		ID:        formatProviderID(hostName, dom.Name),
+		IP:        ip,
+		State:     mapState(s, ip),
+		UID:       meta.UID,
+		CreatedAt: meta.createdAt(),
+	}, nil
+}
+
+// readMetadata returns the kvm2 mark of a domain, or the zero value (not
+// managed, empty UID) when the domain carries none. A mark that does not
+// parse is Permanent, from parseInstanceMetadata.
+func readMetadata(ctx context.Context, be backend, dom libvirt.Domain) (instanceMetadata, error) {
+	raw, err := be.DomainGetMetadata(ctx, dom, int32(libvirt.DomainMetadataElement), libvirt.OptString{metadataURI}, libvirt.DomainAffectCurrent)
+	if hasCode(err, libvirt.ErrNoDomainMetadata) {
+		return instanceMetadata{}, nil
+	}
+	if err != nil {
+		return instanceMetadata{}, fmt.Errorf("reading metadata of domain %s: %w", dom.Name, err)
+	}
+	return parseInstanceMetadata(raw)
+}
+
+// addressSources is the discovery order of spec entry 7: the guest agent
+// answers where the host's ARP table never sees guest traffic.
+var addressSources = []struct {
+	name string
+	src  libvirt.DomainInterfaceAddressesSource
+}{
+	{"agent", libvirt.DomainInterfaceAddressesSrcAgent},
+	{"arp", libvirt.DomainInterfaceAddressesSrcArp},
+}
+
+// address returns the first usable IPv4 from the first source that yields
+// one, or "" for "no address yet". One source failing is normal and is
+// skipped: a booting VM has no agent yet, while the host's ARP table
+// simply reports nothing, since it never sees traffic addressed
+// elsewhere. A connection already gone, and every source failing, are
+// errors instead — otherwise a link some other reconcile killed reads as
+// a running VM that lost its address, and the controller clears its IP
+// and the health engine drops its accumulated state.
+func address(ctx context.Context, log logr.Logger, be backend, dom libvirt.Domain) (string, error) {
+	errs := make([]error, 0, len(addressSources))
+	for _, s := range addressSources {
+		ifaces, err := be.DomainInterfaceAddresses(ctx, dom, uint32(s.src), 0)
+		if err != nil {
+			if isDead(be) || errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
+				return "", fmt.Errorf("reading addresses of domain %s from the %s source: %w", dom.Name, s.name, err)
+			}
+			log.V(2).Info("no addresses from source", "source", s.name, "error", err.Error())
+			errs = append(errs, fmt.Errorf("%s source: %w", s.name, err))
+			continue
+		}
+		if ip := firstIPv4(ifaces); ip != "" {
+			return ip, nil
+		}
+	}
+	// Transient explicitly: classify would otherwise pick a libvirt code
+	// out of the joined error, and ErrInvalidArg, ErrAccessDenied and
+	// ErrNoSupport are Permanent, which would latch a running proxy as
+	// Failed over an address read.
+	if len(errs) == len(addressSources) {
+		return "", fmt.Errorf("%w: reading addresses of domain %s: every source failed: %w",
+			provider.ErrTransient, dom.Name, errors.Join(errs...))
+	}
+	return "", nil
+}
@@ -0,0 +1,295 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+func TestGet_afterDeleteReturnsNotFound(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	if err := fx.p.Delete(fx.ctx, id); err != nil {
+		t.Fatalf("Delete() error = %v", err)
+	}
+	inst, err := fx.p.Get(fx.ctx, id)
+	if inst != nil {
+		t.Errorf("Get after Delete = %+v, want nil", inst)
+	}
+	assertProviderError(t, err, "get", provider.ErrNotFound)
+}
+
+func TestGet_duringDeleteStillReturnsInstance(t *testing.T) {
+	t.Parallel()
+	t.Run("destroyed but still defined is stopped", func(t *testing.T) {
+		t.Parallel()
+		fx := newFixture(t, 1)
+		id := fx.create(testName)
+		fx.withDomain(testName, func(d *fakeDomain) { d.state = libvirt.DomainShutoff })
+		inst := fx.get(id)
+		if inst.State != provider.StateStopped || inst.UID != testUID {
+			t.Errorf("Get = %+v, want Stopped with uid %s", inst, testUID)
+		}
+	})
+	t.Run("undefine in flight still answers", func(t *testing.T) {
+		t.Parallel()
+		fx := newFixture(t, 1)
+		id := fx.create(testName)
+		release := fx.fake.blockOn("DomainUndefineFlags")
+		done := make(chan error, 1)
+		go func() { done <- fx.p.Delete(fx.ctx, id) }()
+		waitFor(t, "the undefine to block", func() bool { return fx.fake.callCount("DomainUndefineFlags") == 1 })
+
+		inst, err := fx.p.Get(fx.ctx, id)
+		if err != nil {
+			t.Fatalf("Get during Delete error = %v, want the instance", err)
+		}
+		if inst.State != provider.StateStopped || inst.ID != id {
+			t.Errorf("Get during Delete = %+v, want Stopped %s", inst, id)
+		}
+		close(release)
+		if err := <-done; err != nil {
+			t.Fatalf("Delete() error = %v", err)
+		}
+		_, err = fx.p.Get(fx.ctx, id)
+		assertProviderError(t, err, "get", provider.ErrNotFound)
+	})
+}
+
+func TestGet_backendBusyIsTransientNeverNotFound(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		err  error
+	}{
+		{name: "operation invalid", err: fakeErr(libvirt.ErrOperationInvalid, "Requested operation is not valid: domain is being migrated")},
+		{name: "call timed out", err: fmt.Errorf("%w: libvirt call DomainGetState on beelink exceeded its 30s budget: %w", provider.ErrTransient, context.DeadlineExceeded)},
+		{name: "unknown libvirt code", err: fakeErr(libvirt.ErrInternalError, "internal error: qemu monitor unresponsive")},
+		{name: "not a libvirt error", err: errors.New("read tcp: connection reset by peer")},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			id := fx.create(testName)
+			fx.fake.failWith("DomainGetState", tc.err)
+			inst, err := fx.p.Get(fx.ctx, id)
+			if inst != nil {
+				t.Errorf("Get = %+v, want nil", inst)
+			}
+			assertProviderError(t, err, "get", provider.ErrTransient)
+			if errors.Is(err, provider.ErrNotFound) {
+				t.Errorf("Get error = %v, must never be NotFound", err)
+			}
+		})
+	}
+}
+
+func TestGet_runningWithoutIPIsProvisioning(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	inst := fx.get(fx.create(testName))
+	if inst.State != provider.StateProvisioning || inst.IP != "" {
+		t.Errorf("Get = %+v, want Provisioning with no IP", inst)
+	}
+}
+
+func TestGet_runningWithIPIsRunning(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	fx.setAddrs(testName, libvirt.DomainInterfaceAddressesSrcAgent, "127.0.0.1", "169.254.1.2", "10.0.0.5")
+	inst := fx.get(id)
+	if inst.State != provider.StateRunning || inst.IP != "10.0.0.5" {
+		t.Errorf("Get = %+v, want Running with 10.0.0.5", inst)
+	}
+	if got := fx.logs.matching("kvm2 domain fetched"); len(got) != 1 || !strings.Contains(got[0], `"ip"="10.0.0.5"`) || !strings.Contains(got[0], `"state"="Running"`) {
+		t.Errorf("fetched line = %v, want state and ip", got)
+	}
+}
+
+func TestGet_agentPreferredOverARP(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	fx.setAddrs(testName, libvirt.DomainInterfaceAddressesSrcAgent, "10.0.0.5")
+	fx.setAddrs(testName, libvirt.DomainInterfaceAddressesSrcArp, "10.0.0.6")
+	before := fx.fake.callCount("DomainInterfaceAddresses")
+	inst := fx.get(id)
+	if inst.IP != "10.0.0.5" {
+		t.Errorf("IP = %q, want the agent's 10.0.0.5", inst.IP)
+	}
+	if got := fx.fake.callCount("DomainInterfaceAddresses") - before; got != 1 {
+		t.Errorf("DomainInterfaceAddresses calls = %d, want 1 (ARP not consulted)", got)
+	}
+}
+
+func TestGet_arpFallbackWhenAgentFails(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 2)
+	id := fx.create(testName)
+	fx.setAddrs(testName, libvirt.DomainInterfaceAddressesSrcArp, "10.0.0.6")
+	var calls int
+	fx.fake.mu.Lock()
+	fx.fake.hooks["DomainInterfaceAddresses"] = func() error {
+		calls++
+		if calls%2 == 1 {
+			return fakeErr(libvirt.ErrAgentUnresponsive, "Guest agent is not responding")
+		}
+		return nil
+	}
+	fx.fake.mu.Unlock()
+	inst := fx.get(id)
+	if inst.State != provider.StateRunning || inst.IP != "10.0.0.6" {
+		t.Errorf("Get = %+v, want Running with ARP's 10.0.0.6", inst)
+	}
+	if got := fx.logs.matching("no addresses from source"); len(got) != 1 || !strings.Contains(got[0], `"source"="agent"`) {
+		t.Errorf("agent failure line = %v", got)
+	}
+}
+
+// TestGet_agentFailsAndARPEmptyIsProvisioning is a VM part way through
+// booting on a plain bridge: the agent is not up yet, and the host's ARP
+// table answers without error because it has nothing to report. One
+// source failing must stay "no address yet", not an error.
+func TestGet_agentFailsAndARPEmptyIsProvisioning(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 2)
+	id := fx.create(testName)
+	var calls int
+	fx.fake.mu.Lock()
+	fx.fake.hooks["DomainInterfaceAddresses"] = func() error {
+		calls++
+		if calls%2 == 1 {
+			return fakeErr(libvirt.ErrAgentUnresponsive, "Guest agent is not responding")
+		}
+		return nil
+	}
+	fx.fake.mu.Unlock()
+	inst := fx.get(id)
+	if inst.State != provider.StateProvisioning || inst.IP != "" {
+		t.Errorf("Get = %+v, want Provisioning with no IP", inst)
+	}
+}
+
+// TestGet_everyAddressSourceFailingIsTransient covers the other side of
+// that line. A running VM whose address cannot be read is not a VM
+// without an address: reporting Provisioning would make the controller
+// clear its IP and the health engine drop its counters.
+func TestGet_everyAddressSourceFailingIsTransient(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name  string
+		setup func(fx *fixture)
+	}{
+		{
+			name: "both sources error",
+			setup: func(fx *fixture) {
+				fx.fake.failWith("DomainInterfaceAddresses", fakeErr(libvirt.ErrOperationInvalid, "domain is not running"))
+			},
+		},
+		{
+			name: "the pooled connection died under the read",
+			setup: func(fx *fixture) {
+				fx.fake.failWith("DomainInterfaceAddresses", transientf("libvirt call DomainInterfaceAddresses on beelink: connection is closed: keepalive failed"))
+				fx.fake.markDead()
+			},
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			id := fx.create(testName)
+			tc.setup(fx)
+			inst, err := fx.p.Get(fx.ctx, id)
+			if inst != nil {
+				t.Errorf("Get = %+v, want no instance", inst)
+			}
+			assertProviderError(t, err, "get", provider.ErrTransient)
+		})
+	}
+}
+
+func TestGet_readsUIDAndCreatedAtFromBackend(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	created := time.Date(2026, 9, 16, 12, 0, 0, 0, time.UTC)
+	fx.p.now = func() time.Time { return created }
+	id := fx.create(testName)
+	inst := fx.get(id)
+	if inst.UID != testUID || !inst.CreatedAt.Equal(created) {
+		t.Errorf("Get = %+v, want uid %s created %s", inst, testUID, created)
+	}
+
+	later := created.Add(48 * time.Hour)
+	fx.withDomain(testName, func(d *fakeDomain) { d.metadata = managedMetadata("rewritten-uid", later) })
+	inst = fx.get(id)
+	if inst.UID != "rewritten-uid" || !inst.CreatedAt.Equal(later) {
+		t.Errorf("Get after metadata change = %+v, want uid rewritten-uid created %s (read back, not remembered)", inst, later)
+	}
+}
+
+func TestGet_unmanagedDomainHasEmptyUID(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	fx.fake.addDomain("legacy-vm", libvirt.DomainRunning, "")
+	inst := fx.get(formatProviderID(testHost, "legacy-vm"))
+	if inst.UID != "" || !inst.CreatedAt.IsZero() || inst.State != provider.StateProvisioning {
+		t.Errorf("Get = %+v, want empty uid, zero createdAt, Provisioning", inst)
+	}
+}
+
+func TestGet_malformedIDIsPermanent(t *testing.T) {
+	t.Parallel()
+	ids := []string{"", "proxy-x", "beelink/proxy-x", "hosts/beelink/proxy-x", "hosts//domains/proxy-x", "hosts/beelink/domains/", "hosts/beelink/domains/a/b", "nodes/beelink/domains/proxy-x"}
+	for _, id := range ids {
+		t.Run(id, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			inst, err := fx.p.Get(fx.ctx, id)
+			if inst != nil {
+				t.Errorf("Get(%q) = %+v, want nil", id, inst)
+			}
+			assertProviderError(t, err, "get", provider.ErrPermanent)
+			var pe *provider.Error
+			if errors.As(err, &pe) && pe.ID != id {
+				t.Errorf("error.ID = %q, want %q", pe.ID, id)
+			}
+			if got := fx.dialer.dials.Load(); got != 0 {
+				t.Errorf("dials = %d, want 0", got)
+			}
+		})
+	}
+}
+
+func TestGet_unconfiguredHostIsPermanent(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	_, err := fx.p.Get(fx.ctx, formatProviderID("nas9", testName))
+	assertProviderError(t, err, "get", provider.ErrPermanent)
+	if !strings.Contains(err.Error(), `"nas9"`) {
+		t.Errorf("error = %v, want the host named", err)
+	}
+	if got := fx.dialer.dials.Load(); got != 0 {
+		t.Errorf("dials = %d, want 0", got)
+	}
+}
+
+func TestGet_unknownStateIsStopped(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	fx.fake.addDomain(testName, libvirt.DomainState(42), managedMetadata(testUID, time.Now()))
+	inst := fx.get(formatProviderID(testHost, testName))
+	if inst.State != provider.StateStopped {
+		t.Errorf("State = %s, want Stopped for an unknown backend state", inst.State)
+	}
+}
@@ -0,0 +1,162 @@
+package kvm2
+
+import (
+	"context"
+	"fmt"
+	"sync"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+	"go.opentelemetry.io/otel/attribute"
+	"go.opentelemetry.io/otel/codes"
+	"go.opentelemetry.io/otel/trace"
+	logf "sigs.k8s.io/controller-runtime/pkg/log"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/tracing"
+)
+
+// dialer opens one connection to a host. The production dialer is
+// libvirtDialer; tests inject one that hands out fakes.
+type dialer interface {
+	dial(ctx context.Context, host provider.KVMHost) (backend, error)
+}
+
+// hostTable holds the pooled connection per host (spec entry 19). One dial
+// per host at a time: concurrent callers wait for it and share the result.
+// A dead backend (abandoned call, keepalive failure, socket lost) is dropped
+// and redialled on the next get.
+//
+// What a cancelled or overlong call does to the shared connection (spec
+// entry 18): the adapter kills it, so every other call in flight on that
+// host fails once with Transient and its caller retries on a fresh
+// connection. Callers that must not be exposed to that, the image transfer
+// and the Checker, use dialFresh and get a connection of their own.
+type hostTable struct {
+	dialer dialer
+
+	mu    sync.Mutex
+	hosts map[string]*hostEntry
+}
+
+// hostEntry serialises dials for one host. The semaphore, not a mutex, so a
+// waiter can give up when its ctx ends.
+type hostEntry struct {
+	sem     chan struct{}
+	backend backend
+}
+
+func newHostTable(d dialer) *hostTable {
+	return &hostTable{dialer: d, hosts: make(map[string]*hostEntry)}
+}
+
+func (t *hostTable) entry(name string) *hostEntry {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	e, ok := t.hosts[name]
+	if !ok {
+		e = &hostEntry{sem: make(chan struct{}, 1)}
+		t.hosts[name] = e
+	}
+	return e
+}
+
+// get returns the live pooled backend for host, dialling it when there is
+// none. Waiting for another caller's dial ends with ctx.
+func (t *hostTable) get(ctx context.Context, host provider.KVMHost) (backend, error) {
+	e := t.entry(host.Name)
+	select {
+	case e.sem <- struct{}{}:
+	case <-ctx.Done():
+		return nil, fmt.Errorf("%w: waiting for the connection to host %s: %w", provider.ErrTransient, host.Name, ctx.Err())
+	}
+	defer func() { <-e.sem }()
+	if e.backend != nil {
+		if !isDead(e.backend) {
+			return e.backend, nil
+		}
+		_ = e.backend.Close()
+		e.backend = nil
+	}
+	b, err := t.dialer.dial(ctx, host)
+	if err != nil {
+		return nil, err
+	}
+	e.backend = b
+	return b, nil
+}
+
+// dialFresh returns a dedicated connection that is never pooled. The
+// caller owns it and must Close it.
+func (t *hostTable) dialFresh(ctx context.Context, host provider.KVMHost) (backend, error) {
+	return t.dialer.dial(ctx, host)
+}
+
+// libvirtDialer is the production dialer: SSH transport, go-libvirt over
+// the forwarded socket, the connect handshake, then the observe decorator,
+// so provider code never holds an undecorated backend.
+type libvirtDialer struct {
+	providerName string
+	cfg          provider.KVMConfig
+	budgets      budgets
+	keepalive    time.Duration
+	tracing      []tracing.Option
+}
+
+func newLibvirtDialer(providerName string, cfg provider.KVMConfig) *libvirtDialer {
+	return &libvirtDialer{
+		providerName: providerName,
+		cfg:          cfg,
+		budgets:      defaultBudgets,
+		keepalive:    defaultKeepaliveInterval,
+	}
+}
+
+func (d *libvirtDialer) dial(ctx context.Context, host provider.KVMHost) (backend, error) {
+	ctx, span := tracing.StartSpan(ctx, tracing.Tracer(d.tracing...), "kvm2.dial",
+		trace.WithSpanKind(trace.SpanKindClient),
+		trace.WithAttributes(
+			attribute.String("kvm2.host", host.Name),
+			attribute.String("kvm2.address", host.Address),
+			attribute.String("kvm2.user", host.User),
+		))
+	defer span.End()
+	log := logf.FromContext(ctx).WithName("kvm2").WithValues("provider", d.providerName, "host", host.Name)
+	start := time.Now()
+	b, err := d.connect(ctx, host)
+	if err != nil {
+		span.SetStatus(codes.Error, err.Error())
+		kv := []any{"address", host.Address, "user", host.User, "duration", time.Since(start), "error", err.Error()}
+		if stage, ok := stageOf(err); ok {
+			kv = append(kv, "stage", string(stage))
+		}
+		kv = append(kv, libvirtErrorFields(err)...)
+		log.V(1).Info("libvirt host connect failed", kv...)
+		return nil, err
+	}
+	log.V(1).Info("libvirt host connected", "address", host.Address, "user", host.User, "duration", time.Since(start))
+	return observe(b, d.providerName, host.Name, d.tracing...), nil
+}
+
+// connect is the dial sequence under one budget: the SSH stages, then the
+// libvirt handshake. On success the keepalive and the disconnect watcher
+// run until the link dies.
+func (d *libvirtDialer) connect(ctx context.Context, host provider.KVMHost) (*libvirtBackend, error) {
+	ctx, cancel := context.WithTimeout(ctx, d.budgets.dial)
+	defer cancel()
+	sock, err := dialSSHSocket(ctx, host, d.cfg, 0)
+	if err != nil {
+		return nil, err
+	}
+	b := newLibvirtBackend(libvirt.New(sock), host.Name, sock, d.budgets)
+	if err := b.connect(ctx); err != nil {
+		b.link.markDead("connect failed")
+		return nil, fmt.Errorf("connecting to libvirt on host %s: %w", host.Name, err)
+	}
+	go func() {
+		if err := sock.keepalive(d.keepalive, b.link.dead()); err != nil {
+			b.link.markDead(err.Error())
+		}
+	}()
+	return b, nil
+}
@@ -0,0 +1,195 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"sync"
+	"testing"
+	"time"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+var (
+	hostA = provider.KVMHost{Name: "beelink", Address: "beelink:22", User: "virt"}
+	hostB = provider.KVMHost{Name: "lab2", Address: "lab2:22", User: "virt"}
+)
+
+func TestHosts_oneDialPerHostUnderConcurrency(t *testing.T) {
+	t.Parallel()
+	shared := newFakeBackend()
+	d := &stubDialer{backend: shared}
+	table := newHostTable(d)
+
+	const n = 50
+	var wg sync.WaitGroup
+	got := make([]backend, n)
+	errs := make([]error, n)
+	for i := range n {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			got[i], errs[i] = table.get(context.Background(), hostA)
+		}()
+	}
+	wg.Wait()
+	for i := range n {
+		if errs[i] != nil {
+			t.Fatalf("get #%d error = %v", i, errs[i])
+		}
+		if got[i] != backend(shared) {
+			t.Fatalf("get #%d returned a different backend", i)
+		}
+	}
+	if d.dials.Load() != 1 {
+		t.Errorf("dials = %d, want 1", d.dials.Load())
+	}
+}
+
+func TestHosts_redialsAfterDrop(t *testing.T) {
+	t.Parallel()
+	first := newFakeBackend()
+	second := newFakeBackend()
+	backends := []backend{first, second}
+	d := &stubDialer{fresh: func() backend {
+		b := backends[0]
+		backends = backends[1:]
+		return b
+	}}
+	table := newHostTable(d)
+
+	b, err := table.get(context.Background(), hostA)
+	if err != nil || b != backend(first) {
+		t.Fatalf("first get = %v, %v, want the first backend", b, err)
+	}
+	if b, _ := table.get(context.Background(), hostA); b != backend(first) {
+		t.Fatal("second get did not reuse the live backend")
+	}
+	first.markDead()
+	b, err = table.get(context.Background(), hostA)
+	if err != nil || b != backend(second) {
+		t.Fatalf("get after drop = %v, %v, want the second backend", b, err)
+	}
+	if first.closed != 1 {
+		t.Errorf("dropped backend closed %d times, want 1", first.closed)
+	}
+	if d.dials.Load() != 2 {
+		t.Errorf("dials = %d, want 2", d.dials.Load())
+	}
+}
+
+func TestHosts_dialFreshIsNotCached(t *testing.T) {
+	t.Parallel()
+	d := &stubDialer{fresh: func() backend { return newFakeBackend() }}
+	table := newHostTable(d)
+
+	pooled, err := table.get(context.Background(), hostA)
+	if err != nil {
+		t.Fatalf("get error = %v", err)
+	}
+	fresh, err := table.dialFresh(context.Background(), hostA)
+	if err != nil {
+		t.Fatalf("dialFresh error = %v", err)
+	}
+	if fresh == pooled {
+		t.Fatal("dialFresh returned the pooled backend")
+	}
+	again, _ := table.get(context.Background(), hostA)
+	if again != pooled {
+		t.Error("get after dialFresh did not return the pooled backend")
+	}
+	if d.dials.Load() != 2 {
+		t.Errorf("dials = %d, want 2", d.dials.Load())
+	}
+	if fresh.(*fakeBackend).closed != 0 {
+		t.Error("dialFresh's backend was closed by the table; the caller owns it")
+	}
+}
+
+func TestHosts_waitingForDialEndsWithContext(t *testing.T) {
+	t.Parallel()
+	release := make(chan struct{})
+	d := &stubDialer{block: release}
+	table := newHostTable(d)
+
+	firstDone := make(chan error, 1)
+	go func() {
+		_, err := table.get(context.Background(), hostA)
+		firstDone <- err
+	}()
+	waitForDials(t, d, 1)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 50*time.Millisecond)
+	defer cancel()
+	start := time.Now()
+	_, err := table.get(ctx, hostA)
+	if time.Since(start) > time.Second {
+		t.Errorf("second get took %v, want to give up with its ctx", time.Since(start))
+	}
+	if !errors.Is(err, provider.ErrTransient) || !errors.Is(err, context.DeadlineExceeded) {
+		t.Errorf("second get error = %v, want ErrTransient wrapping DeadlineExceeded", err)
+	}
+	close(release)
+	if err := <-firstDone; err != nil {
+		t.Errorf("first get error = %v, want nil once the dial completes", err)
+	}
+	if d.dials.Load() != 1 {
+		t.Errorf("dials = %d, want 1: the waiter must not dial itself", d.dials.Load())
+	}
+}
+
+func TestHosts_hostsDialIndependently(t *testing.T) {
+	t.Parallel()
+	release := make(chan struct{})
+	defer close(release)
+	dA := &stubDialer{block: release}
+	dB := &stubDialer{}
+	table := newHostTable(perHostDialer{hostA.Name: dA, hostB.Name: dB})
+
+	go func() { _, _ = table.get(context.Background(), hostA) }()
+	waitForDials(t, dA, 1)
+
+	ctx, cancel := context.WithTimeout(context.Background(), time.Second)
+	defer cancel()
+	start := time.Now()
+	if _, err := table.get(ctx, hostB); err != nil {
+		t.Fatalf("get(hostB) error = %v while hostA's dial is blocked", err)
+	}
+	if time.Since(start) > 500*time.Millisecond {
+		t.Errorf("get(hostB) took %v, want not to wait for hostA", time.Since(start))
+	}
+}
+
+func TestHosts_dialErrorIsNotCached(t *testing.T) {
+	t.Parallel()
+	d := &stubDialer{err: transientf("boom")}
+	table := newHostTable(d)
+	if _, err := table.get(context.Background(), hostA); !errors.Is(err, provider.ErrTransient) {
+		t.Fatalf("get error = %v, want the dial error", err)
+	}
+	d.err = nil
+	if _, err := table.get(context.Background(), hostA); err != nil {
+		t.Fatalf("get after the dialer recovered error = %v", err)
+	}
+	if d.dials.Load() != 2 {
+		t.Errorf("dials = %d, want 2", d.dials.Load())
+	}
+}
+
+// perHostDialer routes each host to its own stub.
+type perHostDialer map[string]*stubDialer
+
+func (p perHostDialer) dial(ctx context.Context, host provider.KVMHost) (backend, error) {
+	return p[host.Name].dial(ctx, host)
+}
+
+func waitForDials(t *testing.T, d *stubDialer, n int32) {
+	t.Helper()
+	deadline := time.Now().Add(2 * time.Second)
+	for d.dials.Load() < n {
+		if time.Now().After(deadline) {
+			t.Fatalf("dials = %d, want %d", d.dials.Load(), n)
+		}
+		time.Sleep(time.Millisecond)
+	}
+}
@@ -0,0 +1,28 @@
+package kvm2
+
+import (
+	"strings"
+)
+
+// The providerID is hosts/<host>/domains/<name>. Host names are validated at
+// config load to contain no slash, and domain names come from NameFromUID,
+// so a strict four-segment split is enough to parse it.
+const (
+	idHostsSegment   = "hosts"
+	idDomainsSegment = "domains"
+)
+
+func formatProviderID(host, name string) string {
+	return idHostsSegment + "/" + host + "/" + idDomainsSegment + "/" + name
+}
+
+// parseProviderID splits an ID back into host and domain name. Anything
+// that is not exactly hosts/<host>/domains/<name> with both non-empty is
+// ErrPermanent: no retry will make it resolve.
+func parseProviderID(id string) (host, name string, err error) {
+	parts := strings.Split(id, "/")
+	if len(parts) != 4 || parts[0] != idHostsSegment || parts[2] != idDomainsSegment || parts[1] == "" || parts[3] == "" {
+		return "", "", permanentf("malformed providerID %q: want hosts/<host>/domains/<name>", id)
+	}
+	return parts[1], parts[3], nil
+}
@@ -0,0 +1,54 @@
+package kvm2
+
+import (
+	"errors"
+	"strings"
+	"testing"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+func TestFormatProviderID_roundTrip(t *testing.T) {
+	t.Parallel()
+	id := formatProviderID("beelink", "proxy-abc123def456ghij")
+	if id != "hosts/beelink/domains/proxy-abc123def456ghij" {
+		t.Fatalf("formatProviderID() = %q", id)
+	}
+	host, name, err := parseProviderID(id)
+	if err != nil || host != "beelink" || name != "proxy-abc123def456ghij" {
+		t.Errorf("parseProviderID(%q) = %q, %q, %v; want beelink, proxy-abc123def456ghij, nil", id, host, name, err)
+	}
+}
+
+func TestParseProviderID_malformed(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		id   string
+	}{
+		{name: "empty", id: ""},
+		{name: "two segments", id: "hosts/h"},
+		{name: "three segments", id: "hosts/h/domains"},
+		{name: "empty host", id: "hosts//domains/n"},
+		{name: "empty name", id: "hosts/h/domains/"},
+		{name: "wrong first literal", id: "nodes/h/domains/n"},
+		{name: "wrong third literal", id: "hosts/h/vms/n"},
+		{name: "extra segment", id: "hosts/h/domains/n/extra"},
+		{name: "slash in name", id: "hosts/h/domains/a/b"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			host, name, err := parseProviderID(tc.id)
+			if err == nil {
+				t.Fatalf("parseProviderID(%q) = %q, %q, nil; want error", tc.id, host, name)
+			}
+			if !errors.Is(err, provider.ErrPermanent) {
+				t.Errorf("parseProviderID(%q) error = %v, want ErrPermanent", tc.id, err)
+			}
+			if !strings.Contains(err.Error(), tc.id) {
+				t.Errorf("parseProviderID(%q) error = %q, want it to name the ID", tc.id, err)
+			}
+		})
+	}
+}
@@ -0,0 +1,481 @@
+package kvm2
+
+import (
+	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"regexp"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+	"github.com/go-logr/logr"
+	"go.opentelemetry.io/otel/attribute"
+	"go.opentelemetry.io/otel/codes"
+	"go.opentelemetry.io/otel/trace"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/tracing"
+)
+
+const (
+	// markerCapacityBytes is the marker volume's size: one lowercase hex
+	// sha256 digest, nothing else.
+	markerCapacityBytes = sha256.Size * 2
+	// markerReadCap bounds what is read back from a marker, so a wrong file
+	// under that name cannot grow memory.
+	markerReadCap = 1 << 10
+	// seedReserveBytes is what the quota precheck sets aside for the seed
+	// ISO next to the image and the overlay (spec entry 13).
+	seedReserveBytes = 16 << 20
+)
+
+var hexDigest = regexp.MustCompile(`^[0-9a-f]{64}$`)
+
+// imageSource is a resolved placement.image (spec entries 3 and 4): the
+// catalogue entry, or a URL given directly with the catalogue defaults, plus
+// the key its base volume is named after.
+type imageSource struct {
+	Entry string
+	Key   string
+	provider.KVMImage
+}
+
+func (s imageSource) volumeName() string { return baseVolumeName(s.Entry, s.Key, s.Format) }
+func (s imageSource) markerName() string { return markerVolumeName(s.volumeName()) }
+
+// resolveImage maps placement.image to a source. cfg is already defaulted,
+// so a catalogue entry carries its format and firmware; a URL gets qcow2
+// and uefi. entry is the provider entry's name, which the base volume is
+// named after so two entries never share one file.
+func resolveImage(entry string, cfg provider.KVMConfig, image string) (imageSource, error) {
+	if img, ok := cfg.Images[image]; ok {
+		return imageSource{Entry: entry, Key: image, KVMImage: img}, nil
+	}
+	if provider.IsHTTPURL(image) {
+		return imageSource{
+			Entry:    entry,
+			Key:      urlImageKey(image),
+			KVMImage: provider.KVMImage{URL: image, Format: provider.KVMFormatQcow2, Firmware: provider.KVMFirmwareUEFI},
+		}, nil
+	}
+	return imageSource{}, permanentf("placement.image %q is neither a key in kvm2.images nor an http(s) URL", image)
+}
+
+// lockImage takes the per-(host, volume) lock of spec entry 19, or gives up
+// with ctx: a caller whose ctx ends while another caller's download runs
+// returns promptly (contract §1.5) instead of waiting out a 20-minute
+// transfer.
+func (p *Provider) lockImage(ctx context.Context, hostName, volume string) (func(), error) {
+	release, err := p.imageLocks.acquire(ctx, hostName+"/"+volume)
+	if err != nil {
+		return nil, fmt.Errorf("%w: waiting for image %s on host %s: %w", provider.ErrTransient, volume, hostName, err)
+	}
+	return release, nil
+}
+
+// ensureImage makes the base image present and complete in the host's pool
+// and returns its path (spec entry 4). be is the caller's pooled connection
+// and carries the cheap checks; the upload itself goes over a connection of
+// its own. Concurrent callers for the same (host, volume) queue on the
+// lock: the first fetches, the rest find the marker.
+func (p *Provider) ensureImage(ctx context.Context, host provider.KVMHost, pool libvirt.StoragePool, be backend, src imageSource) (string, error) {
+	unlock, err := p.lockImage(ctx, host.Name, src.volumeName())
+	if err != nil {
+		return "", err
+	}
+	defer unlock()
+	log := p.logger(ctx).WithValues("host", host.Name, "volume", src.volumeName())
+	path, err := cachedImage(ctx, log, pool, be, src)
+	if err != nil || path != "" {
+		return path, err
+	}
+	vol, err := p.fetchImage(ctx, host, pool, be, src)
+	if err != nil {
+		return "", err
+	}
+	path, err = be.StorageVolGetPath(ctx, vol)
+	if err != nil {
+		return "", fmt.Errorf("reading the path of base volume %s: %w", vol.Name, err)
+	}
+	return path, nil
+}
+
+// cachedImage returns the base volume's path when the marker says the base
+// is complete and its digest is the configured one (or none is configured).
+// Anything else is cleared so the fetch starts from nothing, and "" is
+// returned: a base without a marker is an upload that was cut off; a marker
+// without a base, or one whose content is not a digest, is debris; a marker
+// with a different digest means the catalogue entry changed.
+func cachedImage(ctx context.Context, log logr.Logger, pool libvirt.StoragePool, be backend, src imageSource) (string, error) {
+	name, markerName := src.volumeName(), src.markerName()
+	marker, err := be.StorageVolLookupByName(ctx, pool, markerName)
+	switch {
+	case hasCode(err, libvirt.ErrNoStorageVol):
+		return "", deleteVolumeIfPresent(ctx, be, pool, name)
+	case err != nil:
+		return "", fmt.Errorf("looking up marker %s: %w", markerName, err)
+	}
+	digest, err := readMarker(ctx, be, marker)
+	if err != nil {
+		return "", err
+	}
+	switch {
+	case !hexDigest.MatchString(digest):
+		log.V(1).Info("base image marker holds no digest; refetching")
+	case src.SHA256 != "" && !strings.EqualFold(digest, src.SHA256):
+		// Refusing rather than replacing: every overlay on this host names
+		// the base as its backing file, and libvirt will unlink it under
+		// running domains without a word.
+		return "", permanentf("base image %s is cached with sha256 %s but %s is configured; add the image under a new key and move Proxies to it, or delete the Proxies using it and evict the base by hand",
+			name, digest, strings.ToLower(src.SHA256))
+	default:
+		base, err := be.StorageVolLookupByName(ctx, pool, name)
+		switch {
+		case hasCode(err, libvirt.ErrNoStorageVol):
+			return "", deleteVolumeIfPresent(ctx, be, pool, markerName)
+		case err != nil:
+			return "", fmt.Errorf("looking up base volume %s: %w", name, err)
+		}
+		path, err := be.StorageVolGetPath(ctx, base)
+		if err != nil {
+			return "", fmt.Errorf("reading the path of base volume %s: %w", name, err)
+		}
+		return path, nil
+	}
+	return "", errors.Join(deleteVolumeIfPresent(ctx, be, pool, markerName), deleteVolumeIfPresent(ctx, be, pool, name))
+}
+
+// readMarker returns the digest a marker volume holds, lowercased and
+// stripped of padding. The volume is 64 bytes; the read is capped anyway.
+func readMarker(ctx context.Context, be backend, marker libvirt.StorageVol) (string, error) {
+	w := &cappedWriter{limit: markerReadCap}
+	if err := be.StorageVolDownload(ctx, marker, w, 0, 0, 0); err != nil {
+		return "", fmt.Errorf("reading marker %s: %w", marker.Name, err)
+	}
+	return strings.ToLower(strings.TrimRight(string(w.buf), "\x00 \r\n\t")), nil
+}
+
+// cappedWriter keeps the first limit bytes and accepts the rest without
+// storing them, so a download into it is bounded whatever its length.
+type cappedWriter struct {
+	limit int
+	buf   []byte
+}
+
+func (w *cappedWriter) Write(b []byte) (int, error) {
+	if room := w.limit - len(w.buf); room > 0 {
+		w.buf = append(w.buf, b[:min(room, len(b))]...)
+	}
+	return len(b), nil
+}
+
+// deleteVolumeIfPresent removes the named volume; one that is already gone
+// is success.
+func deleteVolumeIfPresent(ctx context.Context, be backend, pool libvirt.StoragePool, name string) error {
+	vol, err := be.StorageVolLookupByName(ctx, pool, name)
+	if hasCode(err, libvirt.ErrNoStorageVol) {
+		return nil
+	}
+	if err != nil {
+		return fmt.Errorf("looking up volume %s: %w", name, err)
+	}
+	if err := be.StorageVolDelete(ctx, vol, 0); err != nil && !hasCode(err, libvirt.ErrNoStorageVol) {
+		return fmt.Errorf("deleting volume %s: %w", name, err)
+	}
+	return nil
+}
+
+// fetchImage streams the image from its URL into a new base volume and then
+// writes the marker, under the kvm2.download span (spec entry 17). The
+// upload goes over a connection of its own (spec entry 18): abandoning a
+// call kills the connection it ran on, and a 20-minute upload must not die
+// because a poll for another Proxy was cancelled on the pooled one. The
+// marker is written last, on the pooled connection, after
+// StoragePoolRefresh made libvirt read the file's real format; its presence
+// is what marks the base complete.
+//
+// Every exit between the base volume's creation and the marker deletes the
+// base volume (best effort): the next attempt would delete it anyway on
+// finding no marker, and removing it now frees the space at once.
+func (p *Provider) fetchImage(ctx context.Context, host provider.KVMHost, pool libvirt.StoragePool, be backend, src imageSource) (libvirt.StorageVol, error) {
+	name := src.volumeName()
+	ctx, cancel := context.WithTimeout(ctx, defaultBudgets.transfer)
+	defer cancel()
+	ctx, span := tracing.StartSpan(ctx, tracing.Tracer(p.tracing...), "kvm2.download",
+		trace.WithSpanKind(trace.SpanKindClient),
+		trace.WithAttributes(
+			attribute.String("kvm2.host", host.Name),
+			attribute.String("kvm2.volume", name),
+			attribute.String("kvm2.url_host", urlHost(src.URL)),
+			attribute.Bool("kvm2.sha256_configured", src.SHA256 != ""),
+		))
+	defer span.End()
+	// Taken after the span starts, so the lines carry its ID.
+	log := p.logger(ctx).WithValues("host", host.Name, "volume", name)
+	start := time.Now()
+
+	// The body runs out of time before the surrounding budget does, so a
+	// mirror that stalls mid-image ends the upload through its reader
+	// rather than through the killed socket the budget would leave.
+	httpCtx, cancelHTTP := context.WithTimeout(ctx, defaultBudgets.transfer-transferGrace)
+	defer cancelHTTP()
+	resp, err := p.do(httpCtx, http.MethodGet, src.URL)
+	if err == nil {
+		err = checkResponse(src.URL, resp)
+	}
+	if err != nil {
+		span.SetStatus(codes.Error, err.Error())
+		return libvirt.StorageVol{}, err
+	}
+	defer func() { _ = resp.Body.Close() }()
+	size := resp.ContentLength
+	span.SetAttributes(attribute.Int64("kvm2.bytes", size))
+	log.V(1).Info("fetching base image into host pool", "url", src.URL, "bytes", size)
+
+	vol, digest, err := p.uploadImage(ctx, host, pool, src, resp.Body, size)
+	if err != nil {
+		// uploadImage has returned, so its dedicated connection is closed
+		// and libvirtd has let go of the stream before the file goes.
+		p.discardVolume(ctx, log, be, pool, name)
+		span.SetStatus(codes.Error, err.Error())
+		return libvirt.StorageVol{}, err
+	}
+	if err := be.StoragePoolRefresh(ctx, pool, 0); err != nil {
+		err = fmt.Errorf("refreshing storage pool %s after the upload: %w", pool.Name, err)
+		p.discardVolume(ctx, log, be, pool, name)
+		span.SetStatus(codes.Error, err.Error())
+		return libvirt.StorageVol{}, err
+	}
+	if err := writeMarker(ctx, be, pool, src.markerName(), digest); err != nil {
+		p.discardVolume(ctx, log, be, pool, name)
+		span.SetStatus(codes.Error, err.Error())
+		return libvirt.StorageVol{}, err
+	}
+	log.V(1).Info("base image cached", "bytes", size, "duration", time.Since(start), "sha256", digest)
+	return vol, nil
+}
+
+// uploadImage creates the base volume and streams body into it over a
+// dedicated connection with a running sha256, then checks length and
+// digest. The error says why: a body that ended early is Transient, a
+// digest that differs from the configured one is Permanent. Discarding
+// the volume is the caller's, once this connection is closed.
+func (p *Provider) uploadImage(ctx context.Context, host provider.KVMHost, pool libvirt.StoragePool, src imageSource, body io.Reader, size int64) (libvirt.StorageVol, string, error) {
+	name := src.volumeName()
+	xmlDoc, err := buildVolumeXML(volumeSpec{Name: name, CapacityBytes: size, Format: src.Format})
+	if err != nil {
+		return libvirt.StorageVol{}, "", err
+	}
+	fresh, err := p.hosts.dialFresh(ctx, host)
+	if err != nil {
+		return libvirt.StorageVol{}, "", fmt.Errorf("dialling host %s for the image upload: %w", host.Name, err)
+	}
+	defer func() { _ = fresh.Close() }()
+	vol, err := fresh.StorageVolCreateXML(ctx, pool, xmlDoc, 0)
+	if err != nil {
+		return libvirt.StorageVol{}, "", fmt.Errorf("creating base volume %s: %w", name, err)
+	}
+
+	hash := sha256.New()
+	counted := &countingReader{r: io.TeeReader(body, hash)}
+	uploadErr := fresh.StorageVolUpload(ctx, vol, counted, 0, uint64(size), 0)
+	read, readErr := counted.state()
+	var digest string
+	if uploadErr == nil {
+		// Only then has the upload stopped reading; an abandoned call's
+		// goroutine may still be writing into the hash.
+		digest = hex.EncodeToString(hash.Sum(nil))
+	}
+	switch {
+	case readErr != nil:
+		err = fmt.Errorf("%w: image %s: the response ended after %d of %d bytes: %w", provider.ErrTransient, src.URL, read, size, readErr)
+	case uploadErr != nil:
+		err = fmt.Errorf("uploading base volume %s: %w", name, uploadErr)
+	case read != size:
+		err = transientf("image %s: short read, %d of %d bytes", src.URL, read, size)
+	case src.SHA256 != "" && !strings.EqualFold(digest, src.SHA256):
+		err = permanentf("image %s: sha256 %s does not match the configured %s", src.URL, digest, strings.ToLower(src.SHA256))
+	}
+	if err != nil {
+		return libvirt.StorageVol{}, "", err
+	}
+	return vol, digest, nil
+}
+
+// countingReader counts what the upload consumed and keeps the first
+// non-EOF error the source returned. The adapter can return from an
+// abandoned upload while go-libvirt still reads, so the fields are guarded.
+type countingReader struct {
+	r   io.Reader
+	mu  sync.Mutex
+	n   int64
+	err error
+}
+
+func (c *countingReader) Read(b []byte) (int, error) {
+	n, err := c.r.Read(b)
+	c.mu.Lock()
+	c.n += int64(n)
+	if err != nil && err != io.EOF && c.err == nil {
+		c.err = err
+	}
+	c.mu.Unlock()
+	return n, err
+}
+
+func (c *countingReader) state() (int64, error) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.n, c.err
+}
+
+// discardVolume removes a base volume a failed fetch left behind. It runs
+// with ctx's values but not its cancellation: the fetch may have failed
+// because ctx ended, and the adapter refuses calls on a finished ctx. The
+// budget bounds it on its own. Failure is logged, not returned: the next
+// attempt finds the volume without a marker and deletes it then.
+func (p *Provider) discardVolume(ctx context.Context, log logr.Logger, be backend, pool libvirt.StoragePool, name string) {
+	ctx, cancel := context.WithTimeout(context.WithoutCancel(ctx), defaultBudgets.rpc)
+	defer cancel()
+	if err := deleteVolumeIfPresent(ctx, be, pool, name); err != nil {
+		log.V(1).Info("incomplete base volume left behind; the next attempt removes it", "error", err.Error())
+	}
+}
+
+// writeMarker creates the raw marker volume and uploads the digest into it.
+// A marker that could not be filled is removed again: an empty marker next
+// to a complete base would read as debris and cost a refetch.
+func writeMarker(ctx context.Context, be backend, pool libvirt.StoragePool, name, digest string) error {
+	xmlDoc, err := buildVolumeXML(volumeSpec{Name: name, CapacityBytes: markerCapacityBytes, Format: provider.KVMFormatRaw})
+	if err != nil {
+		return err
+	}
+	vol, err := be.StorageVolCreateXML(ctx, pool, xmlDoc, 0)
+	if err != nil {
+		return fmt.Errorf("creating marker %s: %w", name, err)
+	}
+	if err := be.StorageVolUpload(ctx, vol, strings.NewReader(digest), 0, uint64(len(digest)), 0); err != nil {
+		_ = deleteVolumeIfPresent(ctx, be, pool, name)
+		return fmt.Errorf("writing marker %s: %w", name, err)
+	}
+	return nil
+}
+
+// do issues one request bound to ctx. Accept-Encoding is fixed to identity
+// because a transparently decompressed body reports no Content-Length. A
+// transport error is Transient.
+func (p *Provider) do(ctx context.Context, method, rawURL string) (*http.Response, error) {
+	req, err := http.NewRequestWithContext(ctx, method, rawURL, nil)
+	if err != nil {
+		return nil, permanentf("image %s: %v", rawURL, err)
+	}
+	req.Header.Set("Accept-Encoding", "identity")
+	resp, err := p.http.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("%w: fetching image %s: %w", provider.ErrTransient, rawURL, err)
+	}
+	return resp, nil
+}
+
+// checkResponse classifies the response at the source (spec entry 13) and
+// closes its body when it is unusable. 404, 403 and 410, and every other
+// 4xx, mean the catalogue entry is wrong; 5xx and 429 are worth a retry. A
+// missing Content-Length is Permanent: the volume's capacity must be
+// declared before the upload, and a mirror that only chunks is one the
+// catalogue must not point at.
+func checkResponse(rawURL string, resp *http.Response) error {
+	var err error
+	code := resp.StatusCode
+	switch {
+	case code >= 200 && code < 300:
+	case code == http.StatusTooManyRequests || code >= 500:
+		err = transientf("image %s: HTTP %s", rawURL, resp.Status)
+	case code >= 400:
+		err = permanentf("image %s: HTTP %s", rawURL, resp.Status)
+	default:
+		err = transientf("image %s: unexpected HTTP %s", rawURL, resp.Status)
+	}
+	if err == nil && resp.ContentLength < 0 {
+		err = permanentf("image %s: response has no Content-Length; the volume capacity must be known before upload", rawURL)
+	}
+	if err == nil && resp.ContentLength == 0 {
+		err = permanentf("image %s: response is empty (Content-Length 0)", rawURL)
+	}
+	if err != nil {
+		_ = resp.Body.Close()
+	}
+	return err
+}
+
+// urlHost is the only part of the URL a span carries (contract §9).
+func urlHost(rawURL string) string {
+	u, err := url.Parse(rawURL)
+	if err != nil {
+		return ""
+	}
+	return u.Host
+}
+
+// imageSizeIfUncached returns the image's Content-Length when the base is
+// not cached on the host, and 0 when the marker is present, for the quota
+// precheck of spec entry 13. It never downloads: HEAD first, and a GET whose
+// body is closed at once when HEAD answers anything but 2xx, since servers
+// refuse HEAD with 405, 403 or 501 alike; the GET's status is what gets
+// classified. A marker is taken at face value here; ensureImage decides
+// whether it still matches.
+func (p *Provider) imageSizeIfUncached(ctx context.Context, be backend, pool libvirt.StoragePool, src imageSource) (int64, error) {
+	_, err := be.StorageVolLookupByName(ctx, pool, src.markerName())
+	switch {
+	case err == nil:
+		return 0, nil
+	case !hasCode(err, libvirt.ErrNoStorageVol):
+		return 0, fmt.Errorf("looking up marker %s: %w", src.markerName(), err)
+	}
+	ctx, cancel := context.WithTimeout(ctx, defaultBudgets.rpc)
+	defer cancel()
+	resp, err := p.do(ctx, http.MethodHead, src.URL)
+	if err != nil {
+		return 0, err
+	}
+	_ = resp.Body.Close()
+	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+		if resp, err = p.do(ctx, http.MethodGet, src.URL); err != nil {
+			return 0, err
+		}
+		_ = resp.Body.Close()
+	}
+	if err := checkResponse(src.URL, resp); err != nil {
+		return 0, err
+	}
+	return resp.ContentLength, nil
+}
+
+// checkPoolSpace is the precheck of spec entry 13: the pool must be running
+// and have needed bytes available, or Create returns ErrQuotaExceeded and
+// the reconciler waits QuotaRetry instead of failing the volume creation.
+func checkPoolSpace(ctx context.Context, be backend, pool libvirt.StoragePool, poolName string, needed uint64) error {
+	state, _, _, available, err := be.StoragePoolGetInfo(ctx, pool)
+	if err != nil {
+		return fmt.Errorf("reading storage pool %s: %w", poolName, err)
+	}
+	if s := libvirt.StoragePoolState(state); s != libvirt.StoragePoolRunning {
+		return permanentf("storage pool %s is not running (state %s)", poolName, s)
+	}
+	if available < needed {
+		return fmt.Errorf("%w: storage pool %s has %d bytes available, %d needed", provider.ErrQuotaExceeded, poolName, available, needed)
+	}
+	return nil
+}
+
+// neededBytes is what one Create commits to the pool: the image when it is
+// not cached yet (0 otherwise), the overlay's capacity and the seed reserve.
+func neededBytes(imageBytes int64, diskSizeGB int64) uint64 {
+	return uint64(max(imageBytes, 0)) + uint64(max(diskSizeGB, 0))<<30 + seedReserveBytes
+}
@@ -0,0 +1,246 @@
+// Package kvm2 implements the provider contract on libvirt/KVM hypervisors
+// reached over SSH: the host's libvirt unix socket is forwarded through the
+// SSH connection and driven with go-libvirt's RPC client. The design is in
+// docs/kvm2-provider-spec.md.
+package kvm2
+
+import (
+	"context"
+	"fmt"
+	"net/http"
+	"os"
+	"sync"
+	"time"
+
+	"github.com/go-logr/logr"
+	logf "sigs.k8s.io/controller-runtime/pkg/log"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/tracing"
+)
+
+const (
+	defaultSocket     = "/var/run/libvirt/libvirt-sock"
+	defaultPool       = "default"
+	defaultDiskSizeGB = 10
+)
+
+// Provider is one configured kvm2 entry: a set of hosts sharing one SSH key
+// and one image and machine-type catalogue.
+type Provider struct {
+	name    string
+	cfg     provider.KVMConfig
+	now     func() time.Time
+	hosts   *hostTable
+	http    *http.Client
+	tracing []tracing.Option
+
+	// runningWait caps the wait for a started domain to report running,
+	// polled every runningPoll (spec entry 18).
+	runningWait time.Duration
+	runningPoll time.Duration
+
+	// imageLocks serialises image fetches per (host, volume), spec entry 19.
+	imageLocks *keyedLocks
+
+	// nameLocks serialises Create and Delete per instance name, spec entry
+	// 19, so two Deletes never race on destroy and undefine and a Delete
+	// cannot interleave with a Create's steps.
+	nameLocks *keyedLocks
+}
+
+// providerOption adjusts what newProvider builds; tests inject an HTTP
+// client bound to a test server, a span recorder and a shorter running
+// wait.
+type providerOption func(*Provider)
+
+func withHTTPClient(c *http.Client) providerOption {
+	return func(p *Provider) { p.http = c }
+}
+
+func withTracing(opts ...tracing.Option) providerOption {
+	return func(p *Provider) { p.tracing = opts }
+}
+
+// withRunningWait shrinks the poll interval and the cap of the wait for a
+// domain to run.
+func withRunningWait(poll, cap time.Duration) providerOption {
+	return func(p *Provider) { p.runningPoll, p.runningWait = poll, cap }
+}
+
+var _ provider.Provider = (*Provider)(nil)
+
+// New builds a Provider from pc.KVM2. It never dials a host: the credential
+// files are only checked for existence, and a missing one is a warning, not
+// an error, because the Secret mount is optional and may land later.
+func New(ctx context.Context, pc provider.ProviderConfig) (provider.Provider, error) {
+	if !provider.IsKVMIdentifier(pc.Name) {
+		return nil, permanentf("provider name %q: kvm2 names its cached base images after the entry, so the name must be letters, digits, dots, dashes or underscores", pc.Name)
+	}
+	cfg := provider.KVMConfig{}
+	if pc.KVM2 != nil {
+		cfg = *pc.KVM2
+	}
+	cfg = withDefaults(cfg)
+	p := newProvider(pc.Name, cfg, newLibvirtDialer(pc.Name, cfg))
+	log := p.logger(ctx)
+	if _, err := os.Stat(p.cfg.SSHKeyFile); err != nil {
+		log.Info("WARNING: ssh key file is not readable; every call fails until it is", "path", p.cfg.SSHKeyFile, "reason", err.Error())
+	}
+	switch {
+	case p.cfg.InsecureIgnoreHostKey:
+		log.Info("WARNING: insecureIgnoreHostKey is set; host keys are not verified")
+	default:
+		if _, err := os.Stat(p.cfg.KnownHostsFile); err != nil {
+			log.Info("WARNING: known_hosts file is not readable; every call fails until it is", "path", p.cfg.KnownHostsFile, "reason", err.Error())
+		}
+	}
+	return p, nil
+}
+
+// newProvider is the constructor tests use: d hands out the backends, the
+// production one being newLibvirtDialer. Without options the image
+// downloads use http.DefaultClient (system CA roots, no client timeout: the
+// per-request ctx carries the transfer budget).
+func newProvider(name string, cfg provider.KVMConfig, d dialer, opts ...providerOption) *Provider {
+	p := &Provider{
+		name:        name,
+		cfg:         withDefaults(cfg),
+		now:         time.Now,
+		hosts:       newHostTable(d),
+		http:        http.DefaultClient,
+		runningWait: defaultRunningWait,
+		runningPoll: defaultRunningPoll,
+		imageLocks:  newKeyedLocks(),
+		nameLocks:   newKeyedLocks(),
+	}
+	for _, opt := range opts {
+		opt(p)
+	}
+	if p.http == nil {
+		p.http = http.DefaultClient
+	}
+	return p
+}
+
+// withDefaults returns cfg with every optional field filled in, so the rest
+// of the package never re-checks for an empty value. The hosts and images
+// are copied, not shared with the caller.
+func withDefaults(cfg provider.KVMConfig) provider.KVMConfig {
+	if cfg.DiskSizeGB == 0 {
+		cfg.DiskSizeGB = defaultDiskSizeGB
+	}
+	hosts := make([]provider.KVMHost, len(cfg.Hosts))
+	for i, h := range cfg.Hosts {
+		if h.Socket == "" {
+			h.Socket = defaultSocket
+		}
+		if h.Pool == "" {
+			h.Pool = defaultPool
+		}
+		hosts[i] = h
+	}
+	cfg.Hosts = hosts
+	images := make(map[string]provider.KVMImage, len(cfg.Images))
+	for k, img := range cfg.Images {
+		if img.Format == "" {
+			img.Format = provider.KVMFormatQcow2
+		}
+		if img.Firmware == "" {
+			img.Firmware = provider.KVMFirmwareUEFI
+		}
+		images[k] = img
+	}
+	cfg.Images = images
+	return cfg
+}
+
+func (p *Provider) logger(ctx context.Context) logr.Logger {
+	return logf.FromContext(ctx).WithName("kvm2").WithValues("provider", p.name)
+}
+
+// host resolves a placement.zone or a providerID's host segment to a
+// configured host. An empty zone is allowed only when exactly one host is
+// configured.
+func (p *Provider) host(zone string) (provider.KVMHost, error) {
+	if zone == "" {
+		if len(p.cfg.Hosts) == 1 {
+			return p.cfg.Hosts[0], nil
+		}
+		return provider.KVMHost{}, permanentf("placement.zone is required when kvm2.hosts lists more than one host")
+	}
+	for _, h := range p.cfg.Hosts {
+		if h.Name == zone {
+			return h, nil
+		}
+	}
+	return provider.KVMHost{}, permanentf("host %q is not in kvm2.hosts", zone)
+}
+
+// wrapErr is the single exit for errors: it classifies err and wraps it with
+// the configured provider name, op and id.
+func (p *Provider) wrapErr(op, id string, err error) error {
+	return provider.Wrap(classify(op, err), op, p.name, id, err)
+}
+
+// lockName takes the per-name lock of spec entry 19, or gives up with ctx,
+// so a caller whose ctx ends while another caller's Create is downloading
+// an image returns promptly.
+func (p *Provider) lockName(ctx context.Context, name string) (func(), error) {
+	release, err := p.nameLocks.acquire(ctx, name)
+	if err != nil {
+		return nil, fmt.Errorf("%w: waiting for the lock on %s: %w", provider.ErrTransient, name, err)
+	}
+	return release, nil
+}
+
+// keyedLocks is a set of one-slot semaphores by key (spec entry 19): a
+// semaphore rather than a mutex so a waiter can give up with its ctx. Each
+// entry counts the callers holding or waiting for it and is dropped from
+// the map at zero, so the map does not grow with every name ever seen while
+// two waiters on one key still always share one semaphore.
+type keyedLocks struct {
+	mu    sync.Mutex
+	locks map[string]*keyedLock
+}
+
+type keyedLock struct {
+	sem  chan struct{}
+	refs int
+}
+
+func newKeyedLocks() *keyedLocks {
+	return &keyedLocks{locks: make(map[string]*keyedLock)}
+}
+
+// acquire takes the lock for key, or gives up with ctx's error. The
+// returned func releases it and must be called exactly once.
+func (k *keyedLocks) acquire(ctx context.Context, key string) (func(), error) {
+	k.mu.Lock()
+	l, ok := k.locks[key]
+	if !ok {
+		l = &keyedLock{sem: make(chan struct{}, 1)}
+		k.locks[key] = l
+	}
+	l.refs++
+	k.mu.Unlock()
+	select {
+	case l.sem <- struct{}{}:
+		return func() {
+			<-l.sem
+			k.unref(key, l)
+		}, nil
+	case <-ctx.Done():
+		k.unref(key, l)
+		return nil, ctx.Err()
+	}
+}
+
+func (k *keyedLocks) unref(key string, l *keyedLock) {
+	k.mu.Lock()
+	defer k.mu.Unlock()
+	l.refs--
+	if l.refs == 0 {
+		delete(k.locks, key)
+	}
+}
@@ -0,0 +1,590 @@
+//go:build integration
+
+package kvm2
+
+import (
+	"context"
+	"crypto/rand"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"net"
+	"os"
+	"strconv"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+	"github.com/go-logr/logr"
+	"github.com/go-logr/logr/funcr"
+	sdktrace "go.opentelemetry.io/otel/sdk/trace"
+	"go.opentelemetry.io/otel/sdk/trace/tracetest"
+	"k8s.io/apimachinery/pkg/types"
+	logf "sigs.k8s.io/controller-runtime/pkg/log"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/tracing"
+)
+
+// The real-hypervisor test of spec entry 21. It runs only with
+// -tags=integration and skips itself unless KVM2_TEST_* names a host; see
+// docs/testing.md for the variables and what one run asserts.
+const (
+	integrationProvider = "kvm2-integration"
+	integrationHost     = "kvm2-test"
+	integrationImage    = "test-image"
+	integrationMachine  = "small"
+	integrationPort     = 3128
+
+	// integrationCloudInit is the user-data the VM boots with. Nothing
+	// asserts on the guest running it; it exists so the sentinel check can
+	// prove that no span attribute carries the cloud-init text.
+	integrationCloudInit = "#cloud-config\nruncmd:\n  - echo kvm2-integration\n"
+
+	defaultIntegrationTimeout = 20 * time.Minute
+	defaultIntegrationLogLvl  = 1
+	getPollInterval           = 5 * time.Second
+	cleanupBudget             = 5 * time.Minute
+)
+
+type integrationEnv struct {
+	host, user, sshKey, knownHosts, bridge, imageURL string
+	imageSHA256, pool, socket                        string
+	timeout                                          time.Duration
+	keep                                             bool
+	logLevel                                         int
+}
+
+// loadIntegrationEnv reads the KVM2_TEST_* variables, skipping the test
+// under -short or when a required one is missing.
+func loadIntegrationEnv(t *testing.T) integrationEnv {
+	t.Helper()
+	if testing.Short() {
+		t.Skip("real-hypervisor test skipped under -short")
+	}
+	env := integrationEnv{timeout: defaultIntegrationTimeout, logLevel: defaultIntegrationLogLvl}
+	required := []struct {
+		name string
+		dst  *string
+	}{
+		{"KVM2_TEST_HOST", &env.host},
+		{"KVM2_TEST_USER", &env.user},
+		{"KVM2_TEST_SSH_KEY", &env.sshKey},
+		{"KVM2_TEST_BRIDGE", &env.bridge},
+		{"KVM2_TEST_IMAGE_URL", &env.imageURL},
+	}
+	for _, r := range required {
+		v := os.Getenv(r.name)
+		if v == "" {
+			t.Skipf("%s is not set; the kvm2 integration test needs a real hypervisor (see docs/testing.md)", r.name)
+		}
+		*r.dst = v
+	}
+	env.knownHosts = os.Getenv("KVM2_TEST_KNOWN_HOSTS")
+	env.imageSHA256 = os.Getenv("KVM2_TEST_IMAGE_SHA256")
+	env.pool = os.Getenv("KVM2_TEST_POOL")
+	env.socket = os.Getenv("KVM2_TEST_SOCKET")
+	env.keep = os.Getenv("KVM2_TEST_KEEP") != ""
+	if v := os.Getenv("KVM2_TEST_TIMEOUT"); v != "" {
+		d, err := time.ParseDuration(v)
+		if err != nil {
+			t.Fatalf("KVM2_TEST_TIMEOUT=%q: %v", v, err)
+		}
+		env.timeout = d
+	}
+	if v := os.Getenv("KVM2_TEST_LOG_LEVEL"); v != "" {
+		n, err := strconv.Atoi(v)
+		if err != nil || n < 0 {
+			t.Fatalf("KVM2_TEST_LOG_LEVEL=%q: want a non-negative integer", v)
+		}
+		env.logLevel = n
+	}
+	return env
+}
+
+// providerConfig is what the operator would load for this host: one host,
+// one image, one machine type. Host keys are verified only when
+// KVM2_TEST_KNOWN_HOSTS is set.
+func (e integrationEnv) providerConfig() provider.ProviderConfig {
+	return provider.ProviderConfig{
+		Name: integrationProvider,
+		Type: "kvm2",
+		KVM2: &provider.KVMConfig{
+			SSHKeyFile:            e.sshKey,
+			KnownHostsFile:        e.knownHosts,
+			InsecureIgnoreHostKey: e.knownHosts == "",
+			Images: map[string]provider.KVMImage{
+				integrationImage: {URL: e.imageURL, SHA256: e.imageSHA256},
+			},
+			MachineTypes: map[string]provider.KVMMachineType{
+				integrationMachine: {VCPUs: 1, MemoryMiB: 1024},
+			},
+			Hosts: []provider.KVMHost{{
+				Name:    integrationHost,
+				Address: e.host,
+				User:    e.user,
+				Socket:  e.socket,
+				Pool:    e.pool,
+				Bridge:  e.bridge,
+			}},
+		},
+	}
+}
+
+// integrationLog routes the provider's logr output into t.Log through funcr
+// at the chosen verbosity, keeps every line for assertions, and strips the
+// key/value pairs that are constant for the run (stated once up front) so
+// each line stays short. Lines arriving after the test ended are dropped:
+// t.Log panics then.
+type integrationLog struct {
+	t *testing.T
+
+	mu    sync.Mutex
+	drop  []string
+	lines []string
+	done  bool
+}
+
+func newIntegrationLog(t *testing.T, level int, drop ...string) (*integrationLog, logr.Logger) {
+	t.Helper()
+	sink := &integrationLog{t: t, drop: drop}
+	t.Cleanup(func() {
+		sink.mu.Lock()
+		defer sink.mu.Unlock()
+		sink.done = true
+	})
+	log := funcr.New(func(prefix, args string) { sink.write(prefix, args) }, funcr.Options{Verbosity: level})
+	return sink, log
+}
+
+// dropValue adds a rendered pair (`"key"="value"`) to strip from later lines.
+func (l *integrationLog) dropValue(key, value string) {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	l.drop = append(l.drop, fmt.Sprintf("%q=%q", key, value))
+}
+
+func (l *integrationLog) write(prefix, args string) {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	l.lines = append(l.lines, args)
+	if l.done {
+		return
+	}
+	for _, d := range l.drop {
+		args = strings.ReplaceAll(args, " "+d, "")
+	}
+	l.t.Logf("%s %s %s", time.Now().Format("15:04:05.000"), prefix, args)
+}
+
+func (l *integrationLog) matching(substr string) []string {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	var out []string
+	for _, line := range l.lines {
+		if strings.Contains(line, substr) {
+			out = append(out, line)
+		}
+	}
+	return out
+}
+
+// newIntegrationProvider builds the provider the way New does, plus the span
+// recorder's tracer on both places the package starts spans: the Provider
+// (kvm2.download) and the dialer (kvm2.dial and, through observe, every
+// kvm2.<Method>). newLibvirtDialer has no option for it, so the field is set
+// directly; that is what makes the kvm2.* spans land in the same recorder
+// as provider.WithTracing's provider.* spans.
+func newIntegrationProvider(pc provider.ProviderConfig, tp *sdktrace.TracerProvider) *Provider {
+	cfg := withDefaults(*pc.KVM2)
+	opt := tracing.WithTracerProvider(tp)
+	d := newLibvirtDialer(pc.Name, cfg)
+	d.tracing = []tracing.Option{opt}
+	return newProvider(pc.Name, cfg, d, withTracing(opt))
+}
+
+// randomUID is a UUID-shaped string, so the run's domain never collides with
+// a previous run's leftovers or with a real Proxy's.
+func randomUID(t *testing.T) string {
+	t.Helper()
+	var b [16]byte
+	if _, err := rand.Read(b[:]); err != nil {
+		t.Fatalf("reading random bytes: %v", err)
+	}
+	h := hex.EncodeToString(b[:])
+	return h[0:8] + "-" + h[8:12] + "-" + h[12:16] + "-" + h[16:20] + "-" + h[20:32]
+}
+
+// stepDone logs one lifecycle step with its duration; the log is the record
+// of how long a cold image download or a boot took on this host.
+func stepDone(t *testing.T, step string, start time.Time) {
+	t.Helper()
+	t.Logf("step %s: %s", step, time.Since(start).Round(time.Millisecond))
+}
+
+// waitUntil polls cond every interval until it is true or ctx ends.
+func waitUntil(ctx context.Context, t *testing.T, interval time.Duration, what string, cond func() bool) {
+	t.Helper()
+	ticker := time.NewTicker(interval)
+	defer ticker.Stop()
+	for !cond() {
+		select {
+		case <-ticker.C:
+		case <-ctx.Done():
+			t.Fatalf("gave up waiting for %s: %v", what, ctx.Err())
+		}
+	}
+}
+
+// domainCount lists every domain on the host over a dedicated connection.
+func domainCount(ctx context.Context, t *testing.T, p *Provider, host provider.KVMHost) int {
+	t.Helper()
+	be, err := p.hosts.dialFresh(ctx, host)
+	if err != nil {
+		t.Fatalf("dialling %s to count domains: %v", host.Name, err)
+	}
+	defer func() { _ = be.Close() }()
+	doms, _, err := be.ConnectListAllDomains(ctx, 1, libvirt.ConnectListDomainsActive|libvirt.ConnectListDomainsInactive)
+	if err != nil {
+		t.Fatalf("listing domains on %s: %v", host.Name, err)
+	}
+	return len(doms)
+}
+
+// poolVolumes reports, for each name, whether the host's pool holds a volume
+// of that name, over a dedicated connection. Any error other than
+// ErrNoStorageVol fails the test.
+func poolVolumes(ctx context.Context, t *testing.T, p *Provider, host provider.KVMHost, names ...string) map[string]bool {
+	t.Helper()
+	be, err := p.hosts.dialFresh(ctx, host)
+	if err != nil {
+		t.Fatalf("dialling %s to inspect the pool: %v", host.Name, err)
+	}
+	defer func() { _ = be.Close() }()
+	pool, err := be.StoragePoolLookupByName(ctx, host.Pool)
+	if err != nil {
+		t.Fatalf("looking up pool %s: %v", host.Pool, err)
+	}
+	out := make(map[string]bool, len(names))
+	for _, name := range names {
+		_, err := be.StorageVolLookupByName(ctx, pool, name)
+		switch {
+		case err == nil:
+			out[name] = true
+		case hasCode(err, libvirt.ErrNoStorageVol):
+			out[name] = false
+		default:
+			t.Fatalf("looking up volume %s in pool %s: %v", name, host.Pool, err)
+		}
+	}
+	return out
+}
+
+func TestKVM2_lifecycle(t *testing.T) {
+	env := loadIntegrationEnv(t)
+	pc := env.providerConfig()
+	sink, log := newIntegrationLog(t, env.logLevel,
+		fmt.Sprintf("%q=%q", "provider", integrationProvider),
+		fmt.Sprintf("%q=%q", "host", integrationHost))
+	ctx, cancel := context.WithTimeout(logf.IntoContext(context.Background(), log), env.timeout)
+	defer cancel()
+
+	t.Logf("provider=%s host=%s address=%s user=%s bridge=%s image=%s budget=%s log level %d; provider, host and name are left off every provider log line",
+		integrationProvider, integrationHost, env.host, env.user, env.bridge, env.imageURL, env.timeout, env.logLevel)
+	if env.knownHosts == "" {
+		t.Log("KVM2_TEST_KNOWN_HOSTS is not set: host keys are NOT verified for this run")
+	}
+
+	// 1. New returns without dialling.
+	start := time.Now()
+	built, err := New(ctx, pc)
+	elapsed := time.Since(start)
+	if err != nil {
+		t.Fatalf("New() error = %v", err)
+	}
+	if elapsed > 100*time.Millisecond {
+		t.Errorf("New() took %s, want under 100ms: it must not dial", elapsed)
+	}
+	if lines := sink.matching("libvirt host connected"); len(lines) > 0 {
+		t.Errorf("New() dialled the host before any call: %q", lines)
+	}
+	stepDone(t, "1 New", start)
+
+	// 2. Check reports the host Ready with details.
+	checker, ok := built.(provider.Checker)
+	if !ok {
+		t.Fatalf("New() returned %T, which is not a provider.Checker", built)
+	}
+	start = time.Now()
+	targets := checker.Check(ctx)
+	if len(targets) != 1 {
+		t.Fatalf("Check() returned %d targets, want 1: %+v", len(targets), targets)
+	}
+	ts := targets[0]
+	if !ts.Ready || ts.Reason != provider.ReasonReachable {
+		t.Fatalf("Check() = ready %v reason %q message %q, want Ready/%s (details %v)", ts.Ready, ts.Reason, ts.Message, provider.ReasonReachable, ts.Details)
+	}
+	for _, key := range []string{"hostname", "libvirtVersion", "poolState", "poolAvailableGiB"} {
+		if ts.Details[key] == "" {
+			t.Errorf("Check() details lack %q: %v", key, ts.Details)
+		}
+	}
+	if ts.Details["poolState"] != "running" {
+		t.Errorf("Check() poolState = %q, want running", ts.Details["poolState"])
+	}
+	t.Logf("host %s: hostname=%s libvirt=%s pool=%s (%s) available=%s GiB capacity=%s GiB latency=%s",
+		ts.Name, ts.Details["hostname"], ts.Details["libvirtVersion"], ts.Details["pool"], ts.Details["poolState"],
+		ts.Details["poolAvailableGiB"], ts.Details["poolCapacityGiB"], ts.Latency.Round(time.Millisecond))
+	stepDone(t, "2 Check", start)
+
+	// 3. Every later call goes through the tracing decorator, with the
+	// package's own spans recorded alongside.
+	sr := tracetest.NewSpanRecorder()
+	tp := sdktrace.NewTracerProvider(sdktrace.WithSpanProcessor(sr))
+	t.Cleanup(func() { _ = tp.Shutdown(context.Background()) })
+	p := newIntegrationProvider(pc, tp)
+	traced := provider.WithTracing(pc.Name, p, tracing.WithTracerProvider(tp))
+	host := p.cfg.Hosts[0]
+
+	// 4. Create.
+	uid := randomUID(t)
+	name := provider.NameFromUID(types.UID(uid))
+	sink.dropValue("name", name)
+	id := formatProviderID(integrationHost, name)
+	req := provider.CreateRequest{
+		Name:      name,
+		UID:       uid,
+		Namespace: "integration",
+		ProxyName: "kvm2-lifecycle",
+		Placement: provider.Placement{Zone: integrationHost, MachineType: integrationMachine, Image: integrationImage},
+		CloudInit: integrationCloudInit,
+		Port:      integrationPort,
+	}
+	t.Logf("domain name %s, uid %s, id %s", name, uid, id)
+
+	deleted := false
+	if env.keep {
+		t.Logf("KVM2_TEST_KEEP is set: the VM stays behind. Clean up on the host with:\n"+
+			"  virsh undefine --nvram %s\n  virsh vol-delete --pool %s %s\n  virsh vol-delete --pool %s %s",
+			name, host.Pool, overlayVolumeName(name), host.Pool, seedVolumeName(name))
+	} else {
+		t.Cleanup(func() {
+			if deleted {
+				return
+			}
+			cctx, ccancel := context.WithTimeout(logf.IntoContext(context.Background(), log), cleanupBudget)
+			defer ccancel()
+			if err := p.Delete(cctx, id); err != nil {
+				t.Errorf("cleanup Delete(%s) error = %v; remove the domain and its volumes by hand", id, err)
+			}
+		})
+	}
+
+	start = time.Now()
+	gotID, err := traced.Create(ctx, req)
+	if err != nil {
+		t.Fatalf("Create() error = %v", err)
+	}
+	if gotID != id {
+		t.Fatalf("Create() = %q, want %q", gotID, id)
+	}
+	createReturned := time.Now()
+	stepDone(t, "4 Create (includes the image download on a cold pool)", start)
+
+	// 5. Create again: same ID, nothing new on the host.
+	start = time.Now()
+	before := domainCount(ctx, t, p, host)
+	again, err := traced.Create(ctx, req)
+	if err != nil {
+		t.Fatalf("second Create() error = %v", err)
+	}
+	if again != id {
+		t.Errorf("second Create() = %q, want %q", again, id)
+	}
+	if after := domainCount(ctx, t, p, host); after != before {
+		t.Errorf("second Create() changed the host's domain count from %d to %d", before, after)
+	}
+	stepDone(t, "5 Create again", start)
+
+	// 6. Get reaches Running with an IPv4.
+	start = time.Now()
+	var inst *provider.Instance
+	waitUntil(ctx, t, getPollInterval, "the domain to be Running with an IPv4", func() bool {
+		var err error
+		inst, err = traced.Get(ctx, id)
+		if err != nil {
+			t.Fatalf("Get(%s) error = %v", id, err)
+		}
+		return inst.State == provider.StateRunning && inst.IP != ""
+	})
+	if inst.UID != uid {
+		t.Errorf("Get() UID = %q, want %q", inst.UID, uid)
+	}
+	if inst.CreatedAt.IsZero() {
+		t.Error("Get() CreatedAt is zero; the metadata mark did not round-trip")
+	} else if skew := time.Since(inst.CreatedAt).Abs(); skew > 5*time.Minute {
+		t.Errorf("Get() CreatedAt = %s, %s from now; want within a few minutes", inst.CreatedAt, skew)
+	}
+	t.Logf("domain running at %s, %s after Create returned (createdAt %s)", inst.IP, time.Since(createReturned).Round(time.Millisecond), inst.CreatedAt.Format(time.RFC3339))
+	stepDone(t, "6 Get until Running", start)
+
+	// 7. ListByTag contains it.
+	start = time.Now()
+	list, err := traced.ListByTag(ctx)
+	if err != nil {
+		t.Fatalf("ListByTag() error = %v", err)
+	}
+	found := false
+	for _, li := range list {
+		if li.ID == id {
+			found = true
+			if li.UID != uid {
+				t.Errorf("ListByTag() entry %s has UID %q, want %q", id, li.UID, uid)
+			}
+		}
+	}
+	if !found {
+		t.Errorf("ListByTag() = %d instances, none with ID %s", len(list), id)
+	}
+	stepDone(t, "7 ListByTag", start)
+
+	// 8. Reachability, logged only: this machine may not route to the bridge.
+	start = time.Now()
+	addr := net.JoinHostPort(inst.IP, "22")
+	if conn, err := net.DialTimeout("tcp", addr, 3*time.Second); err != nil {
+		t.Logf("TCP connect to %s failed (informational; the test host may not route to the bridge): %v", addr, err)
+	} else {
+		_ = conn.Close()
+		t.Logf("TCP connect to %s succeeded", addr)
+	}
+	stepDone(t, "8 TCP connect", start)
+
+	if env.keep {
+		t.Log("KVM2_TEST_KEEP is set: skipping Delete and the pool inspection")
+	} else {
+		// 9. Delete, Get is NotFound, Delete again.
+		start = time.Now()
+		if err := traced.Delete(ctx, id); err != nil {
+			t.Fatalf("Delete() error = %v", err)
+		}
+		deleted = true
+		if _, err := traced.Get(ctx, id); !errors.Is(err, provider.ErrNotFound) {
+			t.Errorf("Get() after Delete error = %v, want ErrNotFound", err)
+		}
+		if err := traced.Delete(ctx, id); err != nil {
+			t.Errorf("second Delete() error = %v, want nil", err)
+		}
+		stepDone(t, "9 Delete, Get, Delete", start)
+
+		// 10. The pool keeps the shared base image and its marker, and holds
+		// no per-VM volume.
+		start = time.Now()
+		base := baseVolumeName(integrationProvider, integrationImage, provider.KVMFormatQcow2)
+		marker := markerVolumeName(base)
+		overlay, seed := overlayVolumeName(name), seedVolumeName(name)
+		present := poolVolumes(ctx, t, p, host, base, marker, overlay, seed)
+		for _, want := range []string{base, marker} {
+			if !present[want] {
+				t.Errorf("pool %s lacks %s after Delete; the base image cache must stay", host.Pool, want)
+			}
+		}
+		for _, gone := range []string{overlay, seed} {
+			if present[gone] {
+				t.Errorf("pool %s still holds %s after Delete", host.Pool, gone)
+			}
+		}
+		stepDone(t, "10 pool inspection", start)
+	}
+
+	// 11. Spans: the package's spans are children of the decorator's.
+	assertLifecycleSpans(t, sr.Ended())
+}
+
+// assertLifecycleSpans checks that at least one provider.create span has a
+// kvm2.DomainDefineXML and a kvm2.dial child (the first Create dials the
+// pooled connection and defines the domain), and that no attribute of any
+// span carries the cloud-init text.
+func assertLifecycleSpans(t *testing.T, spans []sdktrace.ReadOnlySpan) {
+	t.Helper()
+	var creates []sdktrace.ReadOnlySpan
+	for _, s := range spans {
+		if s.Name() == "provider.create" {
+			creates = append(creates, s)
+		}
+	}
+	if len(creates) == 0 {
+		t.Fatalf("no provider.create span among %d spans: %v", len(spans), spanNames(spans))
+	}
+	satisfied := false
+	for _, create := range creates {
+		parent := create.SpanContext().SpanID()
+		var children, define, dial int
+		for _, s := range spans {
+			if s.Parent().SpanID() != parent || !strings.HasPrefix(s.Name(), "kvm2.") {
+				continue
+			}
+			children++
+			switch s.Name() {
+			case "kvm2.DomainDefineXML":
+				define++
+			case "kvm2.dial":
+				dial++
+			}
+		}
+		t.Logf("provider.create span %s: %d kvm2.* children (%d DomainDefineXML, %d dial)", parent, children, define, dial)
+		if define > 0 && dial > 0 {
+			satisfied = true
+		}
+	}
+	if !satisfied {
+		t.Errorf("no provider.create span has both a kvm2.DomainDefineXML and a kvm2.dial child; spans: %v", spanNames(spans))
+	}
+	for _, s := range spans {
+		for _, a := range s.Attributes() {
+			v := a.Value.Emit()
+			if strings.Contains(v, integrationCloudInit) || strings.Contains(v, "#cloud-config") {
+				t.Errorf("span %s attribute %s carries the cloud-init text", s.Name(), a.Key)
+			}
+		}
+	}
+}
+
+// TestKVM2Check_realHost runs the health check against the host and creates
+// nothing.
+func TestKVM2Check_realHost(t *testing.T) {
+	env := loadIntegrationEnv(t)
+	pc := env.providerConfig()
+	_, log := newIntegrationLog(t, env.logLevel,
+		fmt.Sprintf("%q=%q", "provider", integrationProvider),
+		fmt.Sprintf("%q=%q", "host", integrationHost))
+	ctx, cancel := context.WithTimeout(logf.IntoContext(context.Background(), log), env.timeout)
+	defer cancel()
+	if env.knownHosts == "" {
+		t.Log("KVM2_TEST_KNOWN_HOSTS is not set: host keys are NOT verified for this run")
+	}
+
+	built, err := New(ctx, pc)
+	if err != nil {
+		t.Fatalf("New() error = %v", err)
+	}
+	checker, ok := built.(provider.Checker)
+	if !ok {
+		t.Fatalf("New() returned %T, which is not a provider.Checker", built)
+	}
+	targets := checker.Check(ctx)
+	if len(targets) != 1 {
+		t.Fatalf("Check() returned %d targets, want 1: %+v", len(targets), targets)
+	}
+	ts := targets[0]
+	if !ts.Ready || ts.Reason != provider.ReasonReachable {
+		t.Fatalf("Check() = ready %v reason %q message %q, want Ready/%s (details %v)", ts.Ready, ts.Reason, ts.Message, provider.ReasonReachable, ts.Details)
+	}
+	if ts.Details["poolState"] != "running" {
+		t.Errorf("Check() poolState = %q, want running", ts.Details["poolState"])
+	}
+	t.Logf("host %s (%s): hostname=%s libvirt=%s pool=%s (%s) available=%s GiB latency=%s",
+		ts.Name, ts.Address, ts.Details["hostname"], ts.Details["libvirtVersion"], ts.Details["pool"], ts.Details["poolState"],
+		ts.Details["poolAvailableGiB"], ts.Latency.Round(time.Millisecond))
+}
@@ -0,0 +1,361 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/go-logr/logr/funcr"
+	logf "sigs.k8s.io/controller-runtime/pkg/log"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// captureContext collects funcr output at V(2). The sink is locked because
+// Check logs from one goroutine per host.
+func captureContext() (context.Context, *[]string) {
+	lines := &[]string{}
+	var mu sync.Mutex
+	log := funcr.New(func(prefix, args string) {
+		mu.Lock()
+		defer mu.Unlock()
+		*lines = append(*lines, prefix+" "+args)
+	}, funcr.Options{Verbosity: 2})
+	return logf.IntoContext(context.Background(), log), lines
+}
+
+func twoHosts() provider.KVMConfig {
+	return provider.KVMConfig{
+		SSHKeyFile:     "/nonexistent/kvm2/id_ed25519",
+		KnownHostsFile: "/nonexistent/kvm2/known_hosts",
+		MachineTypes:   map[string]provider.KVMMachineType{"small": {VCPUs: 1, MemoryMiB: 1024}},
+		Hosts: []provider.KVMHost{
+			{Name: "beelink", Address: "beelink:22", User: "virt", Bridge: "br0"},
+			{Name: "lab2", Address: "lab2:22", User: "virt", Bridge: "br1", Socket: "/run/s", Pool: "fast"},
+		},
+	}
+}
+
+func TestNew_doesNotDialAndWarnsOnMissingFiles(t *testing.T) {
+	t.Parallel()
+	ctx, lines := captureContext()
+	cfg := twoHosts()
+	cfg.Hosts[0].Address = "192.0.2.1:22"
+	start := time.Now()
+	p, err := New(ctx, provider.ProviderConfig{Name: "kvm2-home", Type: "kvm2", KVM2: &cfg})
+	elapsed := time.Since(start)
+	if err != nil {
+		t.Fatalf("New() error = %v, want nil", err)
+	}
+	if p == nil {
+		t.Fatal("New() = nil provider")
+	}
+	if elapsed > 100*time.Millisecond {
+		t.Errorf("New() took %v, want under 100ms (it must not dial)", elapsed)
+	}
+	var warnings int
+	for _, l := range *lines {
+		if strings.Contains(l, "WARNING:") {
+			warnings++
+		}
+		if !strings.Contains(l, `"provider"="kvm2-home"`) {
+			t.Errorf("log line lacks the provider value: %s", l)
+		}
+	}
+	if warnings != 2 {
+		t.Errorf("got %d WARNING: lines, want 2 (key and known_hosts):\n%s", warnings, strings.Join(*lines, "\n"))
+	}
+}
+
+func TestNew_insecureIgnoreHostKeyWarns(t *testing.T) {
+	t.Parallel()
+	ctx, lines := captureContext()
+	cfg := twoHosts()
+	cfg.KnownHostsFile = ""
+	cfg.InsecureIgnoreHostKey = true
+	if _, err := New(ctx, provider.ProviderConfig{Name: "kvm2-home", Type: "kvm2", KVM2: &cfg}); err != nil {
+		t.Fatalf("New() error = %v, want nil", err)
+	}
+	found := false
+	for _, l := range *lines {
+		if strings.Contains(l, "WARNING:") && strings.Contains(l, "host keys are not verified") {
+			found = true
+		}
+		if strings.Contains(l, "known_hosts") {
+			t.Errorf("known_hosts warning emitted although host keys are ignored: %s", l)
+		}
+	}
+	if !found {
+		t.Errorf("no insecureIgnoreHostKey warning in:\n%s", strings.Join(*lines, "\n"))
+	}
+}
+
+func TestNew_nilBlockDoesNotPanic(t *testing.T) {
+	t.Parallel()
+	ctx, _ := captureContext()
+	p, err := New(ctx, provider.ProviderConfig{Name: "kvm2-empty", Type: "kvm2"})
+	if err != nil {
+		t.Fatalf("New() error = %v, want nil", err)
+	}
+	kp, ok := p.(*Provider)
+	if !ok {
+		t.Fatalf("New() = %T, want *Provider", p)
+	}
+	if kp.name != "kvm2-empty" || kp.cfg.DiskSizeGB != defaultDiskSizeGB || kp.now == nil {
+		t.Errorf("provider = %+v, want name kvm2-empty, default disk size and a clock", kp)
+	}
+}
+
+// TestNew_rejectsUnsafeEntryName guards the one config value that reaches a
+// volume name without passing the kvm validator: the provider entry's own
+// name, which is checked only for being non-empty and unique.
+func TestNew_rejectsUnsafeEntryName(t *testing.T) {
+	t.Parallel()
+	ctx, _ := captureContext()
+	_, err := New(ctx, provider.ProviderConfig{Name: "kvm2/home", Type: "kvm2"})
+	if !errors.Is(err, provider.ErrPermanent) || !strings.Contains(err.Error(), "kvm2/home") {
+		t.Fatalf("New() error = %v, want ErrPermanent naming the entry", err)
+	}
+}
+
+func TestHost_table(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name     string
+		hosts    []provider.KVMHost
+		zone     string
+		wantName string
+		wantErr  string
+	}{
+		{
+			name:     "one host and empty zone selects it",
+			hosts:    twoHosts().Hosts[:1],
+			zone:     "",
+			wantName: "beelink",
+		},
+		{
+			name:    "several hosts and empty zone is permanent",
+			hosts:   twoHosts().Hosts,
+			zone:    "",
+			wantErr: "placement.zone is required when kvm2.hosts lists more than one host",
+		},
+		{
+			name:     "known name",
+			hosts:    twoHosts().Hosts,
+			zone:     "lab2",
+			wantName: "lab2",
+		},
+		{
+			name:    "unknown name is permanent",
+			hosts:   twoHosts().Hosts,
+			zone:    "nas9",
+			wantErr: `host "nas9" is not in kvm2.hosts`,
+		},
+		{
+			name:    "no hosts at all is permanent",
+			hosts:   nil,
+			zone:    "",
+			wantErr: "placement.zone is required",
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			p := newProvider("kvm2-home", provider.KVMConfig{Hosts: tc.hosts}, &stubDialer{})
+			h, err := p.host(tc.zone)
+			if tc.wantErr != "" {
+				if err == nil {
+					t.Fatalf("host(%q) = %+v, want error containing %q", tc.zone, h, tc.wantErr)
+				}
+				if !errors.Is(err, provider.ErrPermanent) {
+					t.Errorf("host(%q) error = %v, want ErrPermanent", tc.zone, err)
+				}
+				if !strings.Contains(err.Error(), tc.wantErr) {
+					t.Errorf("host(%q) error = %q, want substring %q", tc.zone, err, tc.wantErr)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatalf("host(%q) error = %v, want nil", tc.zone, err)
+			}
+			if h.Name != tc.wantName {
+				t.Errorf("host(%q).Name = %q, want %q", tc.zone, h.Name, tc.wantName)
+			}
+		})
+	}
+}
+
+func TestWithDefaults_table(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name  string
+		in    provider.KVMConfig
+		check func(t *testing.T, got provider.KVMConfig)
+	}{
+		{
+			name: "empty config gets disk size and empty maps",
+			in:   provider.KVMConfig{},
+			check: func(t *testing.T, got provider.KVMConfig) {
+				if got.DiskSizeGB != 10 {
+					t.Errorf("DiskSizeGB = %d, want 10", got.DiskSizeGB)
+				}
+				if got.Hosts == nil || got.Images == nil {
+					t.Errorf("Hosts = %v, Images = %v, want non-nil", got.Hosts, got.Images)
+				}
+			},
+		},
+		{
+			name: "explicit disk size kept",
+			in:   provider.KVMConfig{DiskSizeGB: 40},
+			check: func(t *testing.T, got provider.KVMConfig) {
+				if got.DiskSizeGB != 40 {
+					t.Errorf("DiskSizeGB = %d, want 40", got.DiskSizeGB)
+				}
+			},
+		},
+		{
+			name: "host socket and pool default",
+			in:   provider.KVMConfig{Hosts: []provider.KVMHost{{Name: "a"}, {Name: "b", Socket: "/run/s", Pool: "fast"}}},
+			check: func(t *testing.T, got provider.KVMConfig) {
+				if got.Hosts[0].Socket != "/var/run/libvirt/libvirt-sock" || got.Hosts[0].Pool != "default" {
+					t.Errorf("hosts[0] = %+v, want default socket and pool", got.Hosts[0])
+				}
+				if got.Hosts[1].Socket != "/run/s" || got.Hosts[1].Pool != "fast" {
+					t.Errorf("hosts[1] = %+v, want explicit socket and pool kept", got.Hosts[1])
+				}
+			},
+		},
+		{
+			name: "image format and firmware default",
+			in: provider.KVMConfig{Images: map[string]provider.KVMImage{
+				"d": {URL: "https://x/d.qcow2"},
+				"u": {URL: "https://x/u.img", Format: provider.KVMFormatRaw, Firmware: provider.KVMFirmwareBIOS},
+			}},
+			check: func(t *testing.T, got provider.KVMConfig) {
+				if got.Images["d"].Format != provider.KVMFormatQcow2 || got.Images["d"].Firmware != provider.KVMFirmwareUEFI {
+					t.Errorf("images[d] = %+v, want qcow2 and uefi", got.Images["d"])
+				}
+				if got.Images["u"].Format != provider.KVMFormatRaw || got.Images["u"].Firmware != provider.KVMFirmwareBIOS {
+					t.Errorf("images[u] = %+v, want raw and bios kept", got.Images["u"])
+				}
+			},
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			tc.check(t, withDefaults(tc.in))
+		})
+	}
+}
+
+func TestWithDefaults_doesNotAliasInput(t *testing.T) {
+	t.Parallel()
+	in := provider.KVMConfig{
+		Hosts:  []provider.KVMHost{{Name: "a"}},
+		Images: map[string]provider.KVMImage{"d": {URL: "https://x/d.qcow2"}},
+	}
+	got := withDefaults(in)
+	got.Hosts[0].Name = "changed"
+	got.Images["d"] = provider.KVMImage{URL: "changed"}
+	if in.Hosts[0].Name != "a" || in.Images["d"].URL != "https://x/d.qcow2" {
+		t.Errorf("input mutated through the defaulted copy: %+v", in)
+	}
+}
+
+func lockCount(k *keyedLocks) int {
+	k.mu.Lock()
+	defer k.mu.Unlock()
+	return len(k.locks)
+}
+
+func TestKeyedLocks_entryRemovedWhenUncontended(t *testing.T) {
+	t.Parallel()
+	k := newKeyedLocks()
+	release, err := k.acquire(context.Background(), "a")
+	if err != nil {
+		t.Fatalf("acquire error = %v", err)
+	}
+	if lockCount(k) != 1 {
+		t.Errorf("entries while held = %d, want 1", lockCount(k))
+	}
+	release()
+	if lockCount(k) != 0 {
+		t.Errorf("entries after release = %d, want 0", lockCount(k))
+	}
+	if _, err := k.acquire(context.Background(), "a"); err != nil {
+		t.Fatalf("acquire after release error = %v", err)
+	}
+}
+
+func TestKeyedLocks_waitersShareOneLock(t *testing.T) {
+	t.Parallel()
+	k := newKeyedLocks()
+	const n = 20
+	var inside, overlaps atomic.Int32
+	var wg sync.WaitGroup
+	for range n {
+		wg.Go(func() {
+			release, err := k.acquire(context.Background(), "shared")
+			if err != nil {
+				t.Errorf("acquire error = %v", err)
+				return
+			}
+			if inside.Add(1) > 1 {
+				overlaps.Add(1)
+			}
+			time.Sleep(time.Millisecond)
+			inside.Add(-1)
+			release()
+		})
+	}
+	wg.Wait()
+	if overlaps.Load() != 0 {
+		t.Errorf("%d holders overlapped on one key", overlaps.Load())
+	}
+	if lockCount(k) != 0 {
+		t.Errorf("entries after every holder released = %d, want 0", lockCount(k))
+	}
+}
+
+func TestKeyedLocks_waiterGivesUpWithContext(t *testing.T) {
+	t.Parallel()
+	k := newKeyedLocks()
+	release, err := k.acquire(context.Background(), "a")
+	if err != nil {
+		t.Fatalf("acquire error = %v", err)
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 20*time.Millisecond)
+	defer cancel()
+	start := time.Now()
+	if _, err := k.acquire(ctx, "a"); !errors.Is(err, context.DeadlineExceeded) {
+		t.Fatalf("waiter error = %v, want DeadlineExceeded", err)
+	}
+	if time.Since(start) > time.Second {
+		t.Errorf("waiter took %v, want to give up with its ctx", time.Since(start))
+	}
+	if lockCount(k) != 1 {
+		t.Errorf("entries with the holder still in = %d, want 1: the waiter's exit must not drop the holder's entry", lockCount(k))
+	}
+	release()
+	if lockCount(k) != 0 {
+		t.Errorf("entries after the holder released = %d, want 0", lockCount(k))
+	}
+}
+
+func TestWrapErr_carriesProviderNameOpAndID(t *testing.T) {
+	t.Parallel()
+	p := newProvider("kvm2-home", twoHosts(), &stubDialer{})
+	err := p.wrapErr("get", "hosts/beelink/domains/proxy-x", permanentf("boom"))
+	var pe *provider.Error
+	if !errors.As(err, &pe) {
+		t.Fatalf("wrapErr() = %T, want *provider.Error", err)
+	}
+	if pe.Provider != "kvm2-home" || pe.Op != "get" || pe.ID != "hosts/beelink/domains/proxy-x" || pe.Class != provider.ErrPermanent {
+		t.Errorf("wrapErr() = %+v, want provider kvm2-home, op get, the id and ErrPermanent", pe)
+	}
+}
@@ -0,0 +1,56 @@
+package kvm2
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/digitalocean/go-libvirt"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// ListByTag is spec entry 12: every domain on every configured host, in
+// config order, filtered client-side on the kvm2 mark. GC uses only ID, UID
+// and CreatedAt (contract §2.4), all of which the name and the metadata
+// element give, so no state or address RPC is made and State and IP stay
+// empty. A host that cannot be reached or listed fails the whole call
+// (fail-whole); a single domain whose metadata cannot be read is skipped,
+// unless ctx ended, which fails the call rather than pass off an empty view
+// as the truth.
+func (p *Provider) ListByTag(ctx context.Context) ([]provider.Instance, error) {
+	log := p.logger(ctx)
+	var out []provider.Instance
+	for _, host := range p.cfg.Hosts {
+		hlog := log.WithValues("host", host.Name)
+		be, err := p.hosts.get(ctx, host)
+		if err != nil {
+			logAPIError(hlog, "list", err)
+			return nil, p.wrapErr("list", "", err)
+		}
+		doms, _, err := be.ConnectListAllDomains(ctx, 1, libvirt.ConnectListDomainsActive|libvirt.ConnectListDomainsInactive)
+		if err != nil {
+			err = fmt.Errorf("listing domains on host %s: %w", host.Name, err)
+			logAPIError(hlog, "list", err)
+			return nil, p.wrapErr("list", "", err)
+		}
+		for _, dom := range doms {
+			dlog := hlog.WithValues("name", dom.Name)
+			meta, err := readMetadata(ctx, be, dom)
+			if err != nil {
+				if ctx.Err() != nil {
+					return nil, p.wrapErr("list", "", err)
+				}
+				dlog.V(1).Info("skipping domain with unreadable metadata", "error", err.Error())
+				continue
+			}
+			if !meta.managed() {
+				continue
+			}
+			inst := provider.Instance{ID: formatProviderID(host.Name, dom.Name), UID: meta.UID, CreatedAt: meta.createdAt()}
+			dlog.V(2).Info("kvm2 listed instance", "id", inst.ID, "uid", inst.UID, "createdAt", inst.CreatedAt)
+			out = append(out, inst)
+		}
+	}
+	log.V(1).Info("kvm2 domains listed", "hosts", len(p.cfg.Hosts), "managed", len(out))
+	return out, nil
+}
@@ -0,0 +1,213 @@
+package kvm2
+
+import (
+	"errors"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+func listIDs(list []provider.Instance) []string {
+	ids := make([]string, len(list))
+	for i, inst := range list {
+		ids[i] = inst.ID
+	}
+	return ids
+}
+
+func TestListByTag_excludesUnmarkedDomains(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	fx.fake.addDomain("legacy-vm", libvirt.DomainRunning, "")
+	fx.fake.addDomain("unmarked", libvirt.DomainRunning, `<instance><managed>false</managed><uid>x</uid></instance>`)
+	fx.fake.addDomain("unrelated", libvirt.DomainShutoff, `<instance><note>hand made</note></instance>`)
+	list, err := fx.p.ListByTag(fx.ctx)
+	if err != nil {
+		t.Fatalf("ListByTag() error = %v", err)
+	}
+	if got := listIDs(list); len(got) != 1 || got[0] != id {
+		t.Errorf("ListByTag = %v, want only %s", got, id)
+	}
+	if list[0].UID != testUID || list[0].CreatedAt.IsZero() {
+		t.Errorf("listed = %+v, want uid and createdAt", list[0])
+	}
+	if got := fx.logs.matching("kvm2 domains listed"); len(got) != 1 || !strings.Contains(got[0], `"hosts"=1`) || !strings.Contains(got[0], `"managed"=1`) {
+		t.Errorf("listed line = %v", got)
+	}
+}
+
+func TestListByTag_emptyUIDStillListed(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	fx.fake.addDomain("proxy-nouid", libvirt.DomainRunning, managedMetadata("", time.Now()))
+	list, err := fx.p.ListByTag(fx.ctx)
+	if err != nil {
+		t.Fatalf("ListByTag() error = %v", err)
+	}
+	if len(list) != 1 || list[0].UID != "" || list[0].ID != formatProviderID(testHost, "proxy-nouid") {
+		t.Errorf("ListByTag = %+v, want the domain listed with an empty UID", list)
+	}
+}
+
+func TestListByTag_acrossHostsInConfigOrder(t *testing.T) {
+	t.Parallel()
+	cfg := twoHosts()
+	cfg.Hosts[1].Pool = ""
+	first, second := newFakeBackend(), newFakeBackend()
+	first.addDomain("proxy-z", libvirt.DomainRunning, managedMetadata("uid-z", time.Now()))
+	first.addDomain("proxy-a", libvirt.DomainShutoff, managedMetadata("uid-a", time.Now()))
+	second.addDomain("proxy-b", libvirt.DomainRunning, managedMetadata("uid-b", time.Now()))
+	p := newProvider(testProvider, cfg, &hostDialer{backends: map[string]backend{"beelink": first, "lab2": second}})
+	ctx, _ := captureContext()
+	list, err := p.ListByTag(ctx)
+	if err != nil {
+		t.Fatalf("ListByTag() error = %v", err)
+	}
+	want := []string{formatProviderID("beelink", "proxy-a"), formatProviderID("beelink", "proxy-z"), formatProviderID("lab2", "proxy-b")}
+	if got := listIDs(list); strings.Join(got, ",") != strings.Join(want, ",") {
+		t.Errorf("ListByTag = %v, want %v (config order, then name order)", got, want)
+	}
+	if list[0].UID != "uid-a" || list[1].UID != "uid-z" || list[2].UID != "uid-b" {
+		t.Errorf("uids = %s, %s, %s, want uid-a, uid-z, uid-b", list[0].UID, list[1].UID, list[2].UID)
+	}
+}
+
+func TestListByTag_doesNotReadStateOrAddresses(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 2)
+	id := fx.create(testName)
+	fx.setAddrs(testName, libvirt.DomainInterfaceAddressesSrcAgent, "10.0.0.5")
+	fx.fake.addDomain("proxy-off", libvirt.DomainShutoff, managedMetadata("uid-off", time.Now()))
+	before := len(fx.fake.callLog())
+
+	list, err := fx.p.ListByTag(fx.ctx)
+	if err != nil {
+		t.Fatalf("ListByTag() error = %v", err)
+	}
+	if len(list) != 2 || list[0].ID != id || list[1].ID != formatProviderID(testHost, "proxy-off") {
+		t.Fatalf("ListByTag = %v, want the two managed domains", listIDs(list))
+	}
+	for _, inst := range list {
+		if inst.State != "" || inst.IP != "" {
+			t.Errorf("listed %s = state %q ip %q, want both empty: GC does not use them", inst.ID, inst.State, inst.IP)
+		}
+		if inst.UID == "" || inst.CreatedAt.IsZero() {
+			t.Errorf("listed %s lacks uid or createdAt: %+v", inst.ID, inst)
+		}
+	}
+	for _, rpc := range fx.fake.callLog()[before:] {
+		switch rpc {
+		case "ConnectListAllDomains", "DomainGetMetadata":
+		default:
+			t.Errorf("ListByTag made %s; want only the listing and one metadata read per domain", rpc)
+		}
+	}
+	if got := fx.logs.matching("kvm2 listed instance"); len(got) != 2 || strings.Contains(got[0], `"state"=`) {
+		t.Errorf("listed instance lines = %v, want two without a state", got)
+	}
+}
+
+func TestListByTag_unreachableHostFailsWhole(t *testing.T) {
+	t.Parallel()
+	first := newFakeBackend()
+	first.addDomain("proxy-a", libvirt.DomainRunning, managedMetadata("uid-a", time.Now()))
+	d := &hostDialer{
+		backends: map[string]backend{"beelink": first},
+		errs:     map[string]error{"lab2": errors.New("dial tcp 192.0.2.2:22: i/o timeout")},
+	}
+	p := newProvider(testProvider, twoHosts(), d)
+	ctx, _ := captureContext()
+	list, err := p.ListByTag(ctx)
+	if list != nil {
+		t.Errorf("ListByTag = %v, want nil with the error", list)
+	}
+	assertProviderError(t, err, "list", provider.ErrTransient)
+}
+
+func TestListByTag_unreadableDomainSkipped(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name  string
+		setup func(fx *fixture)
+		line  string
+	}{
+		{
+			name: "corrupt metadata on one domain",
+			setup: func(fx *fixture) {
+				fx.fake.addDomain("proxy-bad", libvirt.DomainRunning, "<bogus>not an instance</bogus>")
+			},
+			line: "skipping domain with unreadable metadata",
+		},
+		{
+			name: "metadata rpc fails for the first domain only",
+			setup: func(fx *fixture) {
+				fx.fake.addDomain("proxy-bad", libvirt.DomainRunning, managedMetadata("uid-bad", time.Now()))
+				var calls int
+				fx.fake.mu.Lock()
+				fx.fake.hooks["DomainGetMetadata"] = func() error {
+					calls++
+					if calls == 1 {
+						return fakeErr(libvirt.ErrInternalError, "internal error: metadata read failed")
+					}
+					return nil
+				}
+				fx.fake.mu.Unlock()
+			},
+			line: "skipping domain with unreadable metadata",
+		},
+		{
+			name: "domain gone between listing and metadata read",
+			setup: func(fx *fixture) {
+				fx.fake.addDomain("proxy-bad", libvirt.DomainRunning, managedMetadata("uid-bad", time.Now()))
+				var calls int
+				fx.fake.mu.Lock()
+				fx.fake.hooks["DomainGetMetadata"] = func() error {
+					calls++
+					if calls == 1 {
+						return fakeErr(libvirt.ErrNoDomain, "Domain not found")
+					}
+					return nil
+				}
+				fx.fake.mu.Unlock()
+			},
+			line: "skipping domain with unreadable metadata",
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			fx.fake.addDomain("proxy-good", libvirt.DomainRunning, managedMetadata("uid-good", time.Now()))
+			tc.setup(fx)
+			list, err := fx.p.ListByTag(fx.ctx)
+			if err != nil {
+				t.Fatalf("ListByTag() error = %v", err)
+			}
+			if got := listIDs(list); len(got) != 1 || got[0] != formatProviderID(testHost, "proxy-good") {
+				t.Errorf("ListByTag = %v, want only proxy-good", got)
+			}
+			if got := fx.logs.matching(tc.line); len(got) != 1 || !strings.Contains(got[0], `"name"="proxy-bad"`) {
+				t.Errorf("skip line = %v, want one naming proxy-bad", got)
+			}
+		})
+	}
+}
+
+func TestListByTag_includesInactiveDomains(t *testing.T) {
+	t.Parallel()
+	fx := newFixture(t, 1)
+	id := fx.create(testName)
+	fx.withDomain(testName, func(d *fakeDomain) { d.state = libvirt.DomainShutoff })
+	list, err := fx.p.ListByTag(fx.ctx)
+	if err != nil {
+		t.Fatalf("ListByTag() error = %v", err)
+	}
+	if len(list) != 1 || list[0].ID != id || list[0].UID != testUID {
+		t.Errorf("ListByTag = %+v, want the shut-off domain listed with its uid", list)
+	}
+}
@@ -0,0 +1,78 @@
+package kvm2
+
+import (
+	"encoding/xml"
+	"strings"
+	"time"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// The GC mark lives in the domain's <metadata> as one element in kvm2's own
+// namespace (spec entry 8). The namespace is what selects it in
+// DomainGetMetadata; libvirt 11.6 hands the element back without the xmlns
+// declaration, so parsing matches on the local name only.
+const (
+	metadataURI       = "https://gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/kvm2"
+	metadataLocalName = "instance"
+)
+
+// instanceMetadata is the mark as it is written and read back. The fields
+// are the element's text as is; managed and createdAt interpret them.
+type instanceMetadata struct {
+	XMLName   xml.Name
+	Managed   string `xml:"managed"`
+	UID       string `xml:"uid"`
+	CreatedAt string `xml:"createdAt"`
+}
+
+func newInstanceMetadata(uid string, createdAt time.Time) instanceMetadata {
+	return instanceMetadata{
+		XMLName:   xml.Name{Space: metadataURI, Local: metadataLocalName},
+		Managed:   provider.LabelManagedYes,
+		UID:       uid,
+		CreatedAt: createdAt.UTC().Format(time.RFC3339),
+	}
+}
+
+func (m instanceMetadata) managed() bool {
+	return m.Managed == provider.LabelManagedYes
+}
+
+// createdAt returns the zero time when the element carries no parsable
+// timestamp; GC treats that as old, which the spec accepts.
+func (m instanceMetadata) createdAt() time.Time {
+	t, err := time.Parse(time.RFC3339, m.CreatedAt)
+	if err != nil {
+		return time.Time{}
+	}
+	return t
+}
+
+// marshalInstanceMetadata renders the namespaced element that goes under
+// the domain's <metadata>.
+func marshalInstanceMetadata(m instanceMetadata) (string, error) {
+	m.XMLName = xml.Name{Space: metadataURI, Local: metadataLocalName}
+	out, err := xml.Marshal(m)
+	if err != nil {
+		return "", permanentf("marshalling instance metadata: %v", err)
+	}
+	return string(out), nil
+}
+
+// parseInstanceMetadata reads the element back, with or without its xmlns
+// declaration. Anything that is not an <instance> element is ErrPermanent:
+// the domain's metadata is corrupt and no retry fixes it.
+func parseInstanceMetadata(s string) (instanceMetadata, error) {
+	var m instanceMetadata
+	if strings.TrimSpace(s) == "" {
+		return instanceMetadata{}, permanentf("parsing instance metadata: empty document")
+	}
+	if err := xml.Unmarshal([]byte(s), &m); err != nil {
+		return instanceMetadata{}, permanentf("parsing instance metadata: %v", err)
+	}
+	if m.XMLName.Local != metadataLocalName {
+		return instanceMetadata{}, permanentf("parsing instance metadata: root element is <%s>, want <%s>", m.XMLName.Local, metadataLocalName)
+	}
+	return m, nil
+}
@@ -0,0 +1,162 @@
+package kvm2
+
+import (
+	"errors"
+	"strings"
+	"testing"
+	"time"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+func TestMarshalInstanceMetadata_namespacedElement(t *testing.T) {
+	t.Parallel()
+	created := time.Date(2026, 9, 16, 10, 30, 0, 0, time.FixedZone("CEST", 2*3600))
+	got, err := marshalInstanceMetadata(newInstanceMetadata("uid-1", created))
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, want := range []string{
+		`<instance xmlns="` + metadataURI + `">`,
+		"<managed>true</managed>",
+		"<uid>uid-1</uid>",
+		"<createdAt>2026-09-16T08:30:00Z</createdAt>",
+		"</instance>",
+	} {
+		if !strings.Contains(got, want) {
+			t.Errorf("marshalled metadata %q lacks %q", got, want)
+		}
+	}
+}
+
+func TestParseMetadata_withNamespace(t *testing.T) {
+	t.Parallel()
+	in := `<instance xmlns="` + metadataURI + `"><managed>true</managed><uid>abc</uid><createdAt>2026-09-16T08:30:00Z</createdAt></instance>`
+	m, err := parseInstanceMetadata(in)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !m.managed() {
+		t.Error("managed() = false, want true")
+	}
+	if m.UID != "abc" {
+		t.Errorf("UID = %q, want abc", m.UID)
+	}
+	if want := time.Date(2026, 9, 16, 8, 30, 0, 0, time.UTC); !m.createdAt().Equal(want) {
+		t.Errorf("createdAt() = %v, want %v", m.createdAt(), want)
+	}
+}
+
+func TestParseMetadata_withoutNamespaceDeclaration(t *testing.T) {
+	t.Parallel()
+	in := "<instance>\n  <managed>true</managed>\n  <uid>abc</uid>\n  <createdAt>2026-09-16T08:30:00Z</createdAt>\n</instance>"
+	m, err := parseInstanceMetadata(in)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !m.managed() || m.UID != "abc" {
+		t.Errorf("got managed=%v uid=%q, want true/abc", m.managed(), m.UID)
+	}
+	if m.createdAt().IsZero() {
+		t.Error("createdAt() is zero, want parsed timestamp")
+	}
+}
+
+func TestParseMetadata_absentGivesEmptyUID(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		in   string
+	}{
+		{name: "empty element", in: "<instance/>"},
+		{name: "empty element with namespace", in: `<instance xmlns="` + metadataURI + `"></instance>`},
+		{name: "empty children", in: "<instance><managed></managed><uid></uid><createdAt></createdAt></instance>"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			m, err := parseInstanceMetadata(tc.in)
+			if err != nil {
+				t.Fatal(err)
+			}
+			if m.managed() {
+				t.Error("managed() = true, want false")
+			}
+			if m.UID != "" {
+				t.Errorf("UID = %q, want empty", m.UID)
+			}
+			if !m.createdAt().IsZero() {
+				t.Errorf("createdAt() = %v, want zero", m.createdAt())
+			}
+		})
+	}
+}
+
+func TestParseMetadata_badCreatedAtIsZeroTime(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		in   string
+	}{
+		{name: "garbage", in: "<instance><managed>true</managed><uid>u</uid><createdAt>yesterday</createdAt></instance>"},
+		{name: "date only", in: "<instance><managed>true</managed><uid>u</uid><createdAt>2026-09-16</createdAt></instance>"},
+		{name: "missing", in: "<instance><managed>true</managed><uid>u</uid></instance>"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			m, err := parseInstanceMetadata(tc.in)
+			if err != nil {
+				t.Fatal(err)
+			}
+			if !m.createdAt().IsZero() {
+				t.Errorf("createdAt() = %v, want zero", m.createdAt())
+			}
+			if !m.managed() || m.UID != "u" {
+				t.Errorf("other fields lost: managed=%v uid=%q", m.managed(), m.UID)
+			}
+		})
+	}
+}
+
+func TestParseMetadata_notXMLIsPermanent(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		in   string
+	}{
+		{name: "empty string", in: ""},
+		{name: "whitespace", in: "  \n"},
+		{name: "plain text", in: "managed: true"},
+		{name: "unclosed element", in: "<instance><managed>true"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			_, err := parseInstanceMetadata(tc.in)
+			if !errors.Is(err, provider.ErrPermanent) {
+				t.Errorf("err = %v, want ErrPermanent", err)
+			}
+		})
+	}
+}
+
+func TestParseMetadata_wrongRootIsPermanent(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		in   string
+	}{
+		{name: "other local name", in: "<proxy><managed>true</managed></proxy>"},
+		{name: "other local name with our namespace", in: `<vm xmlns="` + metadataURI + `"><uid>x</uid></vm>`},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			_, err := parseInstanceMetadata(tc.in)
+			if !errors.Is(err, provider.ErrPermanent) {
+				t.Errorf("err = %v, want ErrPermanent", err)
+			}
+		})
+	}
+}
@@ -0,0 +1,263 @@
+package kvm2
+
+import (
+	"context"
+	"io"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+	"go.opentelemetry.io/otel/attribute"
+	"go.opentelemetry.io/otel/codes"
+	"go.opentelemetry.io/otel/trace"
+	logf "sigs.k8s.io/controller-runtime/pkg/log"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/tracing"
+)
+
+// observed is the span-decorator of spec entries 17 and 20: one child span
+// kvm2.<Method> and one V(1) "libvirt call" line per call, with request and
+// response detail at V(2). Applied once, in the dialer, over every backend
+// that leaves it.
+type observed struct {
+	inner    backend
+	provider string
+	host     string
+	tracer   trace.Tracer
+}
+
+var (
+	_ backend  = (*observed)(nil)
+	_ liveness = (*observed)(nil)
+)
+
+func observe(inner backend, providerName, hostName string, opts ...tracing.Option) backend {
+	return &observed{inner: inner, provider: providerName, host: hostName, tracer: tracing.Tracer(opts...)}
+}
+
+// dead forwards the inner backend's liveness; a backend without one is
+// never dropped by the host table.
+func (o *observed) dead() <-chan struct{} {
+	if l, ok := o.inner.(liveness); ok {
+		return l.dead()
+	}
+	return nil
+}
+
+// id is one identifier a call names: a domain, pool or volume. It becomes
+// both the span attribute kvm2.<key> and the log key.
+type id struct{ key, val string }
+
+func domainID(name string) id { return id{"domain", name} }
+func poolID(name string) id   { return id{"pool", name} }
+func volumeID(name string) id { return id{"volume", name} }
+
+// detail returns V(2)-only key/value pairs. It runs only when V(2) is on,
+// so building the list of a hundred domain names costs nothing otherwise.
+type detail func() []any
+
+// call wraps one backend call: span, timing, status and the log line. fn
+// returns the V(2) detail for the line, or nil.
+func (o *observed) call(ctx context.Context, rpc string, ids []id, fn func(ctx context.Context) (detail, error)) error {
+	attrs := make([]attribute.KeyValue, 0, len(ids)+1)
+	attrs = append(attrs, attribute.String("kvm2.host", o.host))
+	for _, i := range ids {
+		attrs = append(attrs, attribute.String("kvm2."+i.key, i.val))
+	}
+	ctx, span := tracing.StartSpan(ctx, o.tracer, "kvm2."+rpc, trace.WithSpanKind(trace.SpanKindClient), trace.WithAttributes(attrs...))
+	defer span.End()
+	log := logf.FromContext(ctx).WithName("kvm2").WithValues("provider", o.provider, "host", o.host)
+	start := time.Now()
+	more, err := fn(ctx)
+	kv := make([]any, 0, 8+2*len(ids))
+	kv = append(kv, "rpc", rpc)
+	for _, i := range ids {
+		kv = append(kv, i.key, i.val)
+	}
+	kv = append(kv, "duration", time.Since(start))
+	if err != nil {
+		span.SetStatus(codes.Error, err.Error())
+		if code, ok := libvirtCode(err); ok {
+			span.SetAttributes(attribute.Int("kvm2.libvirt_code", int(code)))
+		}
+		kv = append(kv, "error", err.Error())
+		kv = append(kv, libvirtErrorFields(err)...)
+	}
+	if more != nil && log.V(2).Enabled() {
+		kv = append(kv, more()...)
+	}
+	log.V(1).Info("libvirt call", kv...)
+	return err
+}
+
+func (o *observed) DomainLookupByName(ctx context.Context, name string) (dom libvirt.Domain, err error) {
+	err = o.call(ctx, "DomainLookupByName", []id{domainID(name)}, func(ctx context.Context) (detail, error) {
+		dom, err = o.inner.DomainLookupByName(ctx, name)
+		return nil, err
+	})
+	return dom, err
+}
+
+func (o *observed) DomainDefineXML(ctx context.Context, xml string) (dom libvirt.Domain, err error) {
+	err = o.call(ctx, "DomainDefineXML", nil, func(ctx context.Context) (detail, error) {
+		dom, err = o.inner.DomainDefineXML(ctx, xml)
+		return func() []any { return []any{"domain", dom.Name, "xml", xml} }, err
+	})
+	return dom, err
+}
+
+func (o *observed) DomainCreate(ctx context.Context, dom libvirt.Domain) error {
+	return o.call(ctx, "DomainCreate", []id{domainID(dom.Name)}, func(ctx context.Context) (detail, error) {
+		return nil, o.inner.DomainCreate(ctx, dom)
+	})
+}
+
+func (o *observed) DomainDestroy(ctx context.Context, dom libvirt.Domain) error {
+	return o.call(ctx, "DomainDestroy", []id{domainID(dom.Name)}, func(ctx context.Context) (detail, error) {
+		return nil, o.inner.DomainDestroy(ctx, dom)
+	})
+}
+
+func (o *observed) DomainUndefineFlags(ctx context.Context, dom libvirt.Domain, flags libvirt.DomainUndefineFlagsValues) error {
+	return o.call(ctx, "DomainUndefineFlags", []id{domainID(dom.Name)}, func(ctx context.Context) (detail, error) {
+		return nil, o.inner.DomainUndefineFlags(ctx, dom, flags)
+	})
+}
+
+func (o *observed) DomainGetState(ctx context.Context, dom libvirt.Domain, flags uint32) (state int32, reason int32, err error) {
+	err = o.call(ctx, "DomainGetState", []id{domainID(dom.Name)}, func(ctx context.Context) (detail, error) {
+		state, reason, err = o.inner.DomainGetState(ctx, dom, flags)
+		return func() []any { return []any{"state", libvirt.DomainState(state).String(), "reason", reason} }, err
+	})
+	return state, reason, err
+}
+
+func (o *observed) DomainGetMetadata(ctx context.Context, dom libvirt.Domain, typ int32, uri libvirt.OptString, flags libvirt.DomainModificationImpact) (md string, err error) {
+	err = o.call(ctx, "DomainGetMetadata", []id{domainID(dom.Name)}, func(ctx context.Context) (detail, error) {
+		md, err = o.inner.DomainGetMetadata(ctx, dom, typ, uri, flags)
+		return func() []any { return []any{"metadata", md} }, err
+	})
+	return md, err
+}
+
+func (o *observed) DomainInterfaceAddresses(ctx context.Context, dom libvirt.Domain, source uint32, flags uint32) (ifaces []libvirt.DomainInterface, err error) {
+	err = o.call(ctx, "DomainInterfaceAddresses", []id{domainID(dom.Name)}, func(ctx context.Context) (detail, error) {
+		ifaces, err = o.inner.DomainInterfaceAddresses(ctx, dom, source, flags)
+		return func() []any { return []any{"source", source, "addresses", interfaceAddresses(ifaces)} }, err
+	})
+	return ifaces, err
+}
+
+func (o *observed) ConnectListAllDomains(ctx context.Context, needResults int32, flags libvirt.ConnectListAllDomainsFlags) (doms []libvirt.Domain, n uint32, err error) {
+	err = o.call(ctx, "ConnectListAllDomains", nil, func(ctx context.Context) (detail, error) {
+		doms, n, err = o.inner.ConnectListAllDomains(ctx, needResults, flags)
+		return func() []any { return []any{"count", len(doms), "domains", domainNames(doms)} }, err
+	})
+	return doms, n, err
+}
+
+func (o *observed) StoragePoolLookupByName(ctx context.Context, name string) (pool libvirt.StoragePool, err error) {
+	err = o.call(ctx, "StoragePoolLookupByName", []id{poolID(name)}, func(ctx context.Context) (detail, error) {
+		pool, err = o.inner.StoragePoolLookupByName(ctx, name)
+		return nil, err
+	})
+	return pool, err
+}
+
+func (o *observed) StoragePoolRefresh(ctx context.Context, pool libvirt.StoragePool, flags uint32) error {
+	return o.call(ctx, "StoragePoolRefresh", []id{poolID(pool.Name)}, func(ctx context.Context) (detail, error) {
+		return nil, o.inner.StoragePoolRefresh(ctx, pool, flags)
+	})
+}
+
+func (o *observed) StoragePoolGetInfo(ctx context.Context, pool libvirt.StoragePool) (state uint8, capacity, allocation, available uint64, err error) {
+	err = o.call(ctx, "StoragePoolGetInfo", []id{poolID(pool.Name)}, func(ctx context.Context) (detail, error) {
+		state, capacity, allocation, available, err = o.inner.StoragePoolGetInfo(ctx, pool)
+		return func() []any {
+			return []any{"state", libvirt.StoragePoolState(state).String(), "capacity", capacity, "allocation", allocation, "available", available}
+		}, err
+	})
+	return state, capacity, allocation, available, err
+}
+
+func (o *observed) StorageVolLookupByName(ctx context.Context, pool libvirt.StoragePool, name string) (vol libvirt.StorageVol, err error) {
+	err = o.call(ctx, "StorageVolLookupByName", []id{poolID(pool.Name), volumeID(name)}, func(ctx context.Context) (detail, error) {
+		vol, err = o.inner.StorageVolLookupByName(ctx, pool, name)
+		return nil, err
+	})
+	return vol, err
+}
+
+func (o *observed) StorageVolGetPath(ctx context.Context, vol libvirt.StorageVol) (path string, err error) {
+	err = o.call(ctx, "StorageVolGetPath", []id{poolID(vol.Pool), volumeID(vol.Name)}, func(ctx context.Context) (detail, error) {
+		path, err = o.inner.StorageVolGetPath(ctx, vol)
+		return func() []any { return []any{"path", path} }, err
+	})
+	return path, err
+}
+
+func (o *observed) StorageVolCreateXML(ctx context.Context, pool libvirt.StoragePool, xml string, flags libvirt.StorageVolCreateFlags) (vol libvirt.StorageVol, err error) {
+	err = o.call(ctx, "StorageVolCreateXML", []id{poolID(pool.Name)}, func(ctx context.Context) (detail, error) {
+		vol, err = o.inner.StorageVolCreateXML(ctx, pool, xml, flags)
+		return func() []any { return []any{"volume", vol.Name, "xml", xml} }, err
+	})
+	return vol, err
+}
+
+// StorageVolUpload and StorageVolDownload log the length only; the bodies
+// are a seed ISO carrying CloudInit or a base image.
+func (o *observed) StorageVolUpload(ctx context.Context, vol libvirt.StorageVol, r io.Reader, offset, length uint64, flags libvirt.StorageVolUploadFlags) error {
+	return o.call(ctx, "StorageVolUpload", []id{poolID(vol.Pool), volumeID(vol.Name)}, func(ctx context.Context) (detail, error) {
+		err := o.inner.StorageVolUpload(ctx, vol, r, offset, length, flags)
+		return func() []any { return []any{"offset", offset, "length", length} }, err
+	})
+}
+
+func (o *observed) StorageVolDownload(ctx context.Context, vol libvirt.StorageVol, w io.Writer, offset, length uint64, flags libvirt.StorageVolDownloadFlags) error {
+	return o.call(ctx, "StorageVolDownload", []id{poolID(vol.Pool), volumeID(vol.Name)}, func(ctx context.Context) (detail, error) {
+		err := o.inner.StorageVolDownload(ctx, vol, w, offset, length, flags)
+		return func() []any { return []any{"offset", offset, "length", length} }, err
+	})
+}
+
+func (o *observed) StorageVolDelete(ctx context.Context, vol libvirt.StorageVol, flags libvirt.StorageVolDeleteFlags) error {
+	return o.call(ctx, "StorageVolDelete", []id{poolID(vol.Pool), volumeID(vol.Name)}, func(ctx context.Context) (detail, error) {
+		return nil, o.inner.StorageVolDelete(ctx, vol, flags)
+	})
+}
+
+func (o *observed) ConnectGetLibVersion(ctx context.Context) (v uint64, err error) {
+	err = o.call(ctx, "ConnectGetLibVersion", nil, func(ctx context.Context) (detail, error) {
+		v, err = o.inner.ConnectGetLibVersion(ctx)
+		return func() []any { return []any{"libvirtVersion", v} }, err
+	})
+	return v, err
+}
+
+func (o *observed) ConnectGetHostname(ctx context.Context) (name string, err error) {
+	err = o.call(ctx, "ConnectGetHostname", nil, func(ctx context.Context) (detail, error) {
+		name, err = o.inner.ConnectGetHostname(ctx)
+		return func() []any { return []any{"hostname", name} }, err
+	})
+	return name, err
+}
+
+// Close is passed through: it is not a libvirt call and gets no span.
+func (o *observed) Close() error { return o.inner.Close() }
+
+func domainNames(doms []libvirt.Domain) []string {
+	names := make([]string, len(doms))
+	for i, d := range doms {
+		names[i] = d.Name
+	}
+	return names
+}
+
+func interfaceAddresses(ifaces []libvirt.DomainInterface) []string {
+	var out []string
+	for _, iface := range ifaces {
+		for _, a := range iface.Addrs {
+			out = append(out, iface.Name+"="+a.Addr)
+		}
+	}
+	return out
+}
@@ -0,0 +1,334 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"reflect"
+	"strconv"
+	"strings"
+	"sync"
+	"testing"
+
+	"github.com/digitalocean/go-libvirt"
+	"github.com/go-logr/logr/funcr"
+	"go.opentelemetry.io/otel/attribute"
+	"go.opentelemetry.io/otel/codes"
+	sdktrace "go.opentelemetry.io/otel/sdk/trace"
+	"go.opentelemetry.io/otel/sdk/trace/tracetest"
+	logf "sigs.k8s.io/controller-runtime/pkg/log"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/tracing"
+)
+
+// logCapture collects funcr output at a verbosity, safe for concurrent use.
+type logCapture struct {
+	mu    sync.Mutex
+	lines []string
+}
+
+func (c *logCapture) all() []string {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return append([]string(nil), c.lines...)
+}
+
+func (c *logCapture) matching(substr string) []string {
+	var out []string
+	for _, l := range c.all() {
+		if strings.Contains(l, substr) {
+			out = append(out, l)
+		}
+	}
+	return out
+}
+
+// observeFixture is a decorated fake with a span recorder and a log capture
+// wired into ctx.
+type observeFixture struct {
+	fake *fakeBackend
+	b    backend
+	sr   *tracetest.SpanRecorder
+	tp   *sdktrace.TracerProvider
+	logs *logCapture
+}
+
+func newObserveFixture(t *testing.T, verbosity int) (*observeFixture, context.Context) {
+	t.Helper()
+	sr := tracetest.NewSpanRecorder()
+	tp := sdktrace.NewTracerProvider(sdktrace.WithSpanProcessor(sr))
+	t.Cleanup(func() { _ = tp.Shutdown(context.Background()) })
+	logs := &logCapture{}
+	log := funcr.New(func(prefix, args string) {
+		logs.mu.Lock()
+		defer logs.mu.Unlock()
+		logs.lines = append(logs.lines, prefix+" "+args)
+	}, funcr.Options{Verbosity: verbosity})
+	fake := newFakeBackend()
+	fx := &observeFixture{
+		fake: fake,
+		b:    observe(fake, "kvm2-home", "beelink", tracing.WithTracerProvider(tp)),
+		sr:   sr,
+		tp:   tp,
+		logs: logs,
+	}
+	return fx, logf.IntoContext(context.Background(), log)
+}
+
+func TestObserve_everySeamMethodInstrumented(t *testing.T) {
+	t.Parallel()
+	seam := reflect.TypeOf((*backend)(nil)).Elem()
+	for i := range seam.NumMethod() {
+		m := seam.Method(i)
+		if m.Name == "Close" {
+			continue
+		}
+		t.Run(m.Name, func(t *testing.T) {
+			t.Parallel()
+			fx, ctx := newObserveFixture(t, 1)
+			args := make([]reflect.Value, m.Type.NumIn())
+			args[0] = reflect.ValueOf(ctx)
+			for j := 1; j < m.Type.NumIn(); j++ {
+				args[j] = reflect.Zero(m.Type.In(j))
+			}
+			reflect.ValueOf(fx.b).MethodByName(m.Name).Call(args)
+
+			ended := fx.sr.Ended()
+			if len(ended) != 1 {
+				t.Fatalf("got %d spans, want 1", len(ended))
+			}
+			if got, want := ended[0].Name(), "kvm2."+m.Name; got != want {
+				t.Errorf("span name = %q, want %q", got, want)
+			}
+			if !hasAttr(ended[0].Attributes(), attribute.String("kvm2.host", "beelink")) {
+				t.Errorf("span attributes %v lack kvm2.host", ended[0].Attributes())
+			}
+			lines := fx.logs.matching(`"msg"="libvirt call"`)
+			if len(lines) != 1 {
+				t.Fatalf("got %d 'libvirt call' lines, want 1:\n%s", len(lines), strings.Join(fx.logs.all(), "\n"))
+			}
+			for _, want := range []string{`"rpc"="` + m.Name + `"`, `"provider"="kvm2-home"`, `"host"="beelink"`, `"duration"=`, `"traceID"=`} {
+				if !strings.Contains(lines[0], want) {
+					t.Errorf("log line lacks %s: %s", want, lines[0])
+				}
+			}
+			if fx.fake.callCount(m.Name) != 1 {
+				t.Errorf("inner %s called %d times, want 1", m.Name, fx.fake.callCount(m.Name))
+			}
+		})
+	}
+}
+
+func TestObserve_childSpansUnderParent(t *testing.T) {
+	t.Parallel()
+	fx, ctx := newObserveFixture(t, 1)
+	ctx, parent := tracing.StartSpan(ctx, fx.tp.Tracer("test"), "provider.get")
+	_, _ = fx.b.DomainLookupByName(ctx, "proxy-x")
+	parent.End()
+
+	var child sdktrace.ReadOnlySpan
+	for _, s := range fx.sr.Ended() {
+		if s.Name() == "kvm2.DomainLookupByName" {
+			child = s
+		}
+	}
+	if child == nil {
+		t.Fatal("no kvm2.DomainLookupByName span recorded")
+	}
+	if child.Parent().SpanID() != parent.SpanContext().SpanID() {
+		t.Errorf("child parent = %s, want %s", child.Parent().SpanID(), parent.SpanContext().SpanID())
+	}
+	if child.SpanContext().TraceID() != parent.SpanContext().TraceID() {
+		t.Error("child is in a different trace than its parent")
+	}
+	if !hasAttr(child.Attributes(), attribute.String("kvm2.domain", "proxy-x")) {
+		t.Errorf("child attributes %v lack kvm2.domain", child.Attributes())
+	}
+	lines := fx.logs.matching(`"msg"="libvirt call"`)
+	if len(lines) != 1 || !strings.Contains(lines[0], `"spanID"="`+child.SpanContext().SpanID().String()+`"`) {
+		t.Errorf("log line does not carry the child's spanID: %v", lines)
+	}
+}
+
+func TestObserve_logsOneLinePerCallAtV1(t *testing.T) {
+	t.Parallel()
+	fx, ctx := newObserveFixture(t, 1)
+	fx.fake.addDomain("proxy-x", libvirt.DomainRunning, "")
+	pool, _ := fx.b.StoragePoolLookupByName(ctx, fakePoolName)
+	_, _ = fx.b.DomainLookupByName(ctx, "proxy-x")
+	_, _ = fx.b.StorageVolLookupByName(ctx, pool, "proxy-x.qcow2")
+
+	lines := fx.logs.matching(`"msg"="libvirt call"`)
+	if len(lines) != 3 {
+		t.Fatalf("got %d lines, want 3:\n%s", len(lines), strings.Join(fx.logs.all(), "\n"))
+	}
+	if len(fx.logs.all()) != 3 {
+		t.Errorf("got %d lines in total, want only the 3 call lines:\n%s", len(fx.logs.all()), strings.Join(fx.logs.all(), "\n"))
+	}
+	wants := []string{
+		`"rpc"="StoragePoolLookupByName" "pool"="default"`,
+		`"rpc"="DomainLookupByName" "domain"="proxy-x"`,
+		`"rpc"="StorageVolLookupByName" "pool"="default" "volume"="proxy-x.qcow2"`,
+	}
+	for i, want := range wants {
+		if !strings.Contains(lines[i], want) {
+			t.Errorf("line %d = %s, want %s in it", i, lines[i], want)
+		}
+		if !strings.Contains(lines[i], `"level"=1`) {
+			t.Errorf("line %d is not V(1): %s", i, lines[i])
+		}
+	}
+}
+
+func TestObserve_detailOnlyAtV2(t *testing.T) {
+	t.Parallel()
+	const xmlDoc = "<domain><name>proxy-x</name><metadata><instance><uid>u1</uid></instance></metadata></domain>"
+	tests := []struct {
+		name      string
+		verbosity int
+		wantXML   bool
+	}{
+		{name: "V1 has no xml", verbosity: 1, wantXML: false},
+		{name: "V2 has xml", verbosity: 2, wantXML: true},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			fx, ctx := newObserveFixture(t, tc.verbosity)
+			dom, err := fx.b.DomainDefineXML(ctx, xmlDoc)
+			if err != nil {
+				t.Fatalf("DomainDefineXML error = %v", err)
+			}
+			_, _ = fx.b.DomainGetMetadata(ctx, dom, int32(libvirt.DomainMetadataElement), libvirt.OptString{metadataURI}, 0)
+			_, _, _ = fx.b.ConnectListAllDomains(ctx, 1, 0)
+			all := strings.Join(fx.logs.all(), "\n")
+			if got := strings.Contains(all, "<domain>"); got != tc.wantXML {
+				t.Errorf("xml logged = %v, want %v:\n%s", got, tc.wantXML, all)
+			}
+			if got := strings.Contains(all, `"metadata"="<instance>`); got != tc.wantXML {
+				t.Errorf("metadata logged = %v, want %v:\n%s", got, tc.wantXML, all)
+			}
+			if got := strings.Contains(all, `"domains"=["proxy-x"]`); got != tc.wantXML {
+				t.Errorf("domain names logged = %v, want %v:\n%s", got, tc.wantXML, all)
+			}
+			if got := len(fx.logs.matching(`"msg"="libvirt call"`)); got != 3 {
+				t.Errorf("got %d call lines, want 3", got)
+			}
+		})
+	}
+}
+
+func TestObserve_uploadBodyNeverLogged(t *testing.T) {
+	t.Parallel()
+	const sentinel = "SENTINEL-CLOUDINIT-3c7a"
+	fx, ctx := newObserveFixture(t, 2)
+	pool, _ := fx.b.StoragePoolLookupByName(ctx, fakePoolName)
+	vol, err := fx.b.StorageVolCreateXML(ctx, pool, "<volume><name>proxy-x-seed.iso</name><capacity>1024</capacity></volume>", 0)
+	if err != nil {
+		t.Fatalf("StorageVolCreateXML error = %v", err)
+	}
+	body := "#cloud-config\npassword: " + sentinel + "\n"
+	if err := fx.b.StorageVolUpload(ctx, vol, strings.NewReader(body), 0, uint64(len(body)), 0); err != nil {
+		t.Fatalf("StorageVolUpload error = %v", err)
+	}
+	var sink strings.Builder
+	if err := fx.b.StorageVolDownload(ctx, vol, &sink, 0, 0, 0); err != nil {
+		t.Fatalf("StorageVolDownload error = %v", err)
+	}
+	if sink.String() != body {
+		t.Errorf("downloaded %q, want the uploaded body", sink.String())
+	}
+	all := strings.Join(fx.logs.all(), "\n")
+	if strings.Contains(all, sentinel) {
+		t.Errorf("upload body leaked into logs:\n%s", all)
+	}
+	uploads := fx.logs.matching(`"rpc"="StorageVolUpload"`)
+	if len(uploads) != 1 || !strings.Contains(uploads[0], `"length"=`+strconv.Itoa(len(body))) {
+		t.Errorf("upload line lacks length: %v", uploads)
+	}
+	for _, s := range fx.sr.Ended() {
+		for _, a := range s.Attributes() {
+			if strings.Contains(a.Value.Emit(), sentinel) {
+				t.Errorf("span %s attribute %s carries the body", s.Name(), a.Key)
+			}
+		}
+	}
+}
+
+func TestObserve_errorFieldsOnFailure(t *testing.T) {
+	t.Parallel()
+	fx, ctx := newObserveFixture(t, 1)
+	_, err := fx.b.DomainLookupByName(ctx, "missing")
+	if !hasCode(err, libvirt.ErrNoDomain) {
+		t.Fatalf("error = %v, want ErrNoDomain through the decorator", err)
+	}
+	lines := fx.logs.matching(`"msg"="libvirt call"`)
+	if len(lines) != 1 {
+		t.Fatalf("got %d lines, want 1", len(lines))
+	}
+	for _, want := range []string{`"error"="Domain not found`, `"libvirtCode"=42`, `"libvirtError"="ErrNoDomain"`} {
+		if !strings.Contains(lines[0], want) {
+			t.Errorf("line lacks %s: %s", want, lines[0])
+		}
+	}
+	ended := fx.sr.Ended()
+	if len(ended) != 1 {
+		t.Fatalf("got %d spans, want 1", len(ended))
+	}
+	if ended[0].Status().Code != codes.Error {
+		t.Errorf("span status = %v, want Error", ended[0].Status().Code)
+	}
+	if !hasAttr(ended[0].Attributes(), attribute.Int("kvm2.libvirt_code", 42)) {
+		t.Errorf("span attributes %v lack kvm2.libvirt_code=42", ended[0].Attributes())
+	}
+
+	fx.fake.failWith("ConnectGetHostname", errors.New("plain failure"))
+	_, _ = fx.b.ConnectGetHostname(ctx)
+	lines = fx.logs.matching(`"rpc"="ConnectGetHostname"`)
+	if len(lines) != 1 || strings.Contains(lines[0], "libvirtCode") || !strings.Contains(lines[0], `"error"="plain failure"`) {
+		t.Errorf("non-libvirt failure line = %v, want error without libvirt fields", lines)
+	}
+}
+
+func TestObserve_v0Silent(t *testing.T) {
+	t.Parallel()
+	fx, ctx := newObserveFixture(t, 0)
+	fx.fake.addDomain("proxy-x", libvirt.DomainRunning, "")
+	_, _ = fx.b.DomainLookupByName(ctx, "proxy-x")
+	_, _ = fx.b.DomainLookupByName(ctx, "missing")
+	if lines := fx.logs.all(); len(lines) != 0 {
+		t.Errorf("got %d lines at V(0), want none:\n%s", len(lines), strings.Join(lines, "\n"))
+	}
+	if got := len(fx.sr.Ended()); got != 2 {
+		t.Errorf("got %d spans, want 2: spans do not depend on verbosity", got)
+	}
+}
+
+func TestObserve_closeAndLivenessPassThrough(t *testing.T) {
+	t.Parallel()
+	fx, _ := newObserveFixture(t, 1)
+	if isDead(fx.b) {
+		t.Error("decorated live backend reported dead")
+	}
+	fx.fake.markDead()
+	if !isDead(fx.b) {
+		t.Error("decorated dead backend reported live")
+	}
+	if err := fx.b.Close(); err != nil || fx.fake.closed != 1 {
+		t.Errorf("Close() = %v, closed %d times, want nil and 1", err, fx.fake.closed)
+	}
+	if got := len(fx.sr.Ended()); got != 0 {
+		t.Errorf("Close produced %d spans, want none", got)
+	}
+	if lines := fx.logs.all(); len(lines) != 0 {
+		t.Errorf("Close produced log lines: %v", lines)
+	}
+}
+
+func hasAttr(attrs []attribute.KeyValue, want attribute.KeyValue) bool {
+	for _, a := range attrs {
+		if a.Key == want.Key && a.Value == want.Value {
+			return true
+		}
+	}
+	return false
+}
@@ -0,0 +1,426 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/digitalocean/go-libvirt"
+	"github.com/go-logr/logr/funcr"
+	logf "sigs.k8s.io/controller-runtime/pkg/log"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+const (
+	testProvider = "kvm2-home"
+	testHost     = "beelink"
+	testUID      = "0f1e2d3c-uid-1"
+	testName     = "proxy-abcdefghijklmnop"
+)
+
+// fixture is one configured host backed by a fake, an image server the
+// catalogue points at, a log capture at a chosen verbosity and a provider
+// with a fast running-wait, so a Create runs end to end in memory.
+type fixture struct {
+	t      *testing.T
+	p      *Provider
+	fake   *fakeBackend
+	dialer *stubDialer
+	logs   *logCapture
+	ctx    context.Context
+	image  []byte
+}
+
+func newFixture(t *testing.T, verbosity int, opts ...providerOption) *fixture {
+	t.Helper()
+	fx := &fixture{t: t, fake: newFakeBackend(), logs: &logCapture{}, image: []byte("QFI\xfb fake base image bytes")}
+	srv := httptest.NewServer(serveBytes(fx.image))
+	t.Cleanup(srv.Close)
+	fx.dialer = &stubDialer{backend: fx.fake}
+	cfg := provider.KVMConfig{
+		Images:       map[string]provider.KVMImage{imageKey: {URL: srv.URL + "/debian-13-generic-amd64.qcow2"}},
+		MachineTypes: map[string]provider.KVMMachineType{"small": {VCPUs: 1, MemoryMiB: 1024}},
+		Hosts:        []provider.KVMHost{{Name: testHost, Address: "beelink:22", User: "virt", Bridge: "br0"}},
+	}
+	opts = append([]providerOption{withHTTPClient(srv.Client()), withRunningWait(time.Millisecond, 500*time.Millisecond)}, opts...)
+	fx.p = newProvider(testProvider, cfg, fx.dialer, opts...)
+	log := funcr.New(func(prefix, args string) {
+		fx.logs.mu.Lock()
+		defer fx.logs.mu.Unlock()
+		fx.logs.lines = append(fx.logs.lines, prefix+" "+args)
+	}, funcr.Options{Verbosity: verbosity})
+	fx.ctx = logf.IntoContext(context.Background(), log)
+	return fx
+}
+
+func (fx *fixture) req(name string) provider.CreateRequest {
+	return provider.CreateRequest{
+		Name:      name,
+		UID:       testUID,
+		Namespace: "default",
+		ProxyName: "egress",
+		Placement: provider.Placement{Zone: testHost, MachineType: "small", Image: imageKey},
+		CloudInit: "#cloud-config\npackages: [squid]\n",
+		Port:      3128,
+	}
+}
+
+func (fx *fixture) create(name string) string {
+	fx.t.Helper()
+	id, err := fx.p.Create(fx.ctx, fx.req(name))
+	if err != nil {
+		fx.t.Fatalf("Create(%s) error = %v", name, err)
+	}
+	if want := formatProviderID(testHost, name); id != want {
+		fx.t.Fatalf("Create(%s) = %q, want %q", name, id, want)
+	}
+	return id
+}
+
+func (fx *fixture) get(id string) *provider.Instance {
+	fx.t.Helper()
+	inst, err := fx.p.Get(fx.ctx, id)
+	if err != nil {
+		fx.t.Fatalf("Get(%s) error = %v", id, err)
+	}
+	return inst
+}
+
+// withDomain runs fn on the fake's domain under its lock.
+func (fx *fixture) withDomain(name string, fn func(d *fakeDomain)) {
+	fx.t.Helper()
+	fx.fake.mu.Lock()
+	defer fx.fake.mu.Unlock()
+	d, ok := fx.fake.domains[name]
+	if !ok {
+		fx.t.Fatalf("fake has no domain %s", name)
+	}
+	fn(d)
+}
+
+func (fx *fixture) hasDomain(name string) bool {
+	fx.fake.mu.Lock()
+	defer fx.fake.mu.Unlock()
+	_, ok := fx.fake.domains[name]
+	return ok
+}
+
+func (fx *fixture) setAddrs(name string, src libvirt.DomainInterfaceAddressesSource, ips ...string) {
+	fx.withDomain(name, func(d *fakeDomain) {
+		var addrs []libvirt.DomainIPAddr
+		for _, ip := range ips {
+			addrs = append(addrs, libvirt.DomainIPAddr{Type: int32(libvirt.IPAddrTypeIpv4), Addr: ip, Prefix: 24})
+		}
+		d.addrs[uint32(src)] = []libvirt.DomainInterface{{Name: "enp1s0", Addrs: addrs}}
+	})
+}
+
+// putVol places a volume in the fake's pool directly.
+func (fx *fixture) putVol(name string, content []byte) *fakeVol {
+	fx.fake.mu.Lock()
+	defer fx.fake.mu.Unlock()
+	v := &fakeVol{content: content, capacity: uint64(len(content))}
+	fx.fake.pools[fakePoolName].vols[name] = v
+	return v
+}
+
+// baseName is the cached base image's volume name for this fixture's
+// provider entry and image key.
+func (fx *fixture) baseName() string {
+	fx.t.Helper()
+	src, err := resolveImage(fx.p.name, fx.p.cfg, imageKey)
+	if err != nil {
+		fx.t.Fatal(err)
+	}
+	return src.volumeName()
+}
+
+// cacheImage places a complete base image and its marker, as a previous
+// Create on the host would have.
+func (fx *fixture) cacheImage() {
+	src, err := resolveImage(fx.p.name, fx.p.cfg, imageKey)
+	if err != nil {
+		fx.t.Fatal(err)
+	}
+	fx.putVol(src.volumeName(), fx.image)
+	fx.putVol(src.markerName(), []byte(digestOf(fx.image)))
+}
+
+func managedMetadata(uid string, createdAt time.Time) string {
+	s, err := marshalInstanceMetadata(newInstanceMetadata(uid, createdAt))
+	if err != nil {
+		panic(err)
+	}
+	return s
+}
+
+// assertProviderError checks err is a *provider.Error from the configured
+// provider with the given op and class.
+func assertProviderError(t *testing.T, err error, op string, class error) {
+	t.Helper()
+	if err == nil {
+		t.Fatalf("error = nil, want op %s class %v", op, class)
+	}
+	var pe *provider.Error
+	if !errors.As(err, &pe) {
+		t.Fatalf("error = %T (%v), want *provider.Error", err, err)
+	}
+	if pe.Provider != testProvider {
+		t.Errorf("error.Provider = %q, want %q", pe.Provider, testProvider)
+	}
+	if pe.Op != op {
+		t.Errorf("error.Op = %q, want %q", pe.Op, op)
+	}
+	if !errors.Is(err, class) {
+		t.Errorf("error = %v, want class %v", err, class)
+	}
+	for _, other := range []error{provider.ErrNotFound, provider.ErrQuotaExceeded, provider.ErrPermanent, provider.ErrTransient} {
+		if other != class && errors.Is(err, other) {
+			t.Errorf("error = %v, also matches %v", err, other)
+		}
+	}
+}
+
+// hostDialer hands out one backend per host name and fails hosts in errs.
+type hostDialer struct {
+	backends map[string]backend
+	errs     map[string]error
+}
+
+func (d *hostDialer) dial(_ context.Context, host provider.KVMHost) (backend, error) {
+	if err := d.errs[host.Name]; err != nil {
+		return nil, err
+	}
+	b, ok := d.backends[host.Name]
+	if !ok {
+		return nil, fmt.Errorf("hostDialer: no backend for %s", host.Name)
+	}
+	return b, nil
+}
+
+// waitFor polls cond until it holds or the deadline passes.
+func waitFor(t *testing.T, what string, cond func() bool) {
+	t.Helper()
+	deadline := time.Now().Add(5 * time.Second)
+	for !cond() {
+		if time.Now().After(deadline) {
+			t.Fatalf("timed out waiting for %s", what)
+		}
+		time.Sleep(time.Millisecond)
+	}
+}
+
+func TestProvider_cancelledContextReturnsPromptly(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name     string
+		firstRPC string
+		op       string
+		call     func(ctx context.Context, fx *fixture) error
+	}{
+		{
+			name: "create", firstRPC: "DomainLookupByName", op: "create",
+			call: func(ctx context.Context, fx *fixture) error { _, err := fx.p.Create(ctx, fx.req(testName)); return err },
+		},
+		{
+			name: "get", firstRPC: "DomainLookupByName", op: "get",
+			call: func(ctx context.Context, fx *fixture) error {
+				_, err := fx.p.Get(ctx, formatProviderID(testHost, testName))
+				return err
+			},
+		},
+		{
+			name: "delete", firstRPC: "DomainLookupByName", op: "delete",
+			call: func(ctx context.Context, fx *fixture) error {
+				return fx.p.Delete(ctx, formatProviderID(testHost, testName))
+			},
+		},
+		{
+			name: "list", firstRPC: "ConnectListAllDomains", op: "list",
+			call: func(ctx context.Context, fx *fixture) error { _, err := fx.p.ListByTag(ctx); return err },
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			release := fx.fake.blockOn(tc.firstRPC)
+			defer close(release)
+			ctx, cancel := context.WithCancel(fx.ctx)
+			done := make(chan error, 1)
+			go func() { done <- tc.call(ctx, fx) }()
+			waitFor(t, "the first RPC to block", func() bool { return fx.fake.callCount(tc.firstRPC) == 1 })
+			cancel()
+			select {
+			case err := <-done:
+				assertProviderError(t, err, tc.op, provider.ErrTransient)
+			case <-time.After(time.Second):
+				t.Fatalf("%s did not return within 1s of cancellation", tc.name)
+			}
+		})
+	}
+}
+
+func TestProvider_everyErrorIsWrappedWithConfiguredName(t *testing.T) {
+	t.Parallel()
+	dialErr := errors.New("dial tcp 192.0.2.1:22: connect: connection refused")
+	tests := []struct {
+		name  string
+		setup func(fx *fixture)
+		call  func(fx *fixture) error
+		op    string
+		class error
+	}{
+		{
+			name: "create unknown zone",
+			call: func(fx *fixture) error {
+				req := fx.req(testName)
+				req.Placement.Zone = "nas9"
+				_, err := fx.p.Create(fx.ctx, req)
+				return err
+			},
+			op: "create", class: provider.ErrPermanent,
+		},
+		{
+			name:  "create dial failure",
+			setup: func(fx *fixture) { fx.dialer.err = dialErr },
+			call:  func(fx *fixture) error { _, err := fx.p.Create(fx.ctx, fx.req(testName)); return err },
+			op:    "create", class: provider.ErrTransient,
+		},
+		{
+			name: "create pool out of space",
+			setup: func(fx *fixture) {
+				fx.fake.mu.Lock()
+				fx.fake.pools[fakePoolName].available = 1
+				fx.fake.mu.Unlock()
+			},
+			call: func(fx *fixture) error { _, err := fx.p.Create(fx.ctx, fx.req(testName)); return err },
+			op:   "create", class: provider.ErrQuotaExceeded,
+		},
+		{
+			name:  "create vanished volume",
+			setup: func(fx *fixture) { fx.fake.failWith("StorageVolGetPath", fakeErr(libvirt.ErrNoStorageVol, "gone")) },
+			call:  func(fx *fixture) error { _, err := fx.p.Create(fx.ctx, fx.req(testName)); return err },
+			op:    "create", class: provider.ErrTransient,
+		},
+		{
+			name: "get malformed id",
+			call: func(fx *fixture) error { _, err := fx.p.Get(fx.ctx, "beelink/"+testName); return err },
+			op:   "get", class: provider.ErrPermanent,
+		},
+		{
+			name: "get missing domain",
+			call: func(fx *fixture) error { _, err := fx.p.Get(fx.ctx, formatProviderID(testHost, testName)); return err },
+			op:   "get", class: provider.ErrNotFound,
+		},
+		{
+			name: "get busy backend",
+			setup: func(fx *fixture) {
+				fx.create(testName)
+				fx.fake.failWith("DomainGetState", fakeErr(libvirt.ErrOperationInvalid, "busy"))
+			},
+			call: func(fx *fixture) error { _, err := fx.p.Get(fx.ctx, formatProviderID(testHost, testName)); return err },
+			op:   "get", class: provider.ErrTransient,
+		},
+		{
+			name: "delete malformed id",
+			call: func(fx *fixture) error { return fx.p.Delete(fx.ctx, "hosts//domains/"+testName) },
+			op:   "delete", class: provider.ErrPermanent,
+		},
+		{
+			name:  "delete dial failure",
+			setup: func(fx *fixture) { fx.dialer.err = dialErr },
+			call:  func(fx *fixture) error { return fx.p.Delete(fx.ctx, formatProviderID(testHost, testName)) },
+			op:    "delete", class: provider.ErrTransient,
+		},
+		{
+			name:  "delete access denied",
+			setup: func(fx *fixture) { fx.fake.failWith("DomainLookupByName", fakeErr(libvirt.ErrAccessDenied, "denied")) },
+			call:  func(fx *fixture) error { return fx.p.Delete(fx.ctx, formatProviderID(testHost, testName)) },
+			op:    "delete", class: provider.ErrPermanent,
+		},
+		{
+			name:  "list dial failure",
+			setup: func(fx *fixture) { fx.dialer.err = dialErr },
+			call:  func(fx *fixture) error { _, err := fx.p.ListByTag(fx.ctx); return err },
+			op:    "list", class: provider.ErrTransient,
+		},
+		{
+			name: "list rpc failure",
+			setup: func(fx *fixture) {
+				fx.fake.failWith("ConnectListAllDomains", fakeErr(libvirt.ErrInternalError, "boom"))
+			},
+			call: func(fx *fixture) error { _, err := fx.p.ListByTag(fx.ctx); return err },
+			op:   "list", class: provider.ErrTransient,
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			fx := newFixture(t, 1)
+			if tc.setup != nil {
+				tc.setup(fx)
+			}
+			assertProviderError(t, tc.call(fx), tc.op, tc.class)
+		})
+	}
+}
+
+func TestLogging_v0SilentV1OutcomeV2Detail(t *testing.T) {
+	t.Parallel()
+	outcomes := []string{"overlay volume created", "seed volume uploaded", "domain defined", "kvm2 domain ensured running"}
+	t.Run("v0 is silent", func(t *testing.T) {
+		t.Parallel()
+		fx := newFixture(t, 0)
+		fx.create(testName)
+		if lines := fx.logs.all(); len(lines) != 0 {
+			t.Errorf("Create logged %d lines at V(0), want none:\n%s", len(lines), strings.Join(lines, "\n"))
+		}
+	})
+	t.Run("v1 has outcomes without xml", func(t *testing.T) {
+		t.Parallel()
+		fx := newFixture(t, 1)
+		fx.create(testName)
+		for _, want := range outcomes {
+			if len(fx.logs.matching(want)) != 1 {
+				t.Errorf("V(1) lines matching %q = %d, want 1:\n%s", want, len(fx.logs.matching(want)), strings.Join(fx.logs.all(), "\n"))
+			}
+		}
+		if got := fx.logs.matching("<domain"); len(got) != 0 {
+			t.Errorf("V(1) carries domain XML:\n%s", strings.Join(got, "\n"))
+		}
+		if got := fx.logs.matching("kvm2 create resolved"); len(got) != 0 {
+			t.Errorf("V(1) carries the V(2) detail line:\n%s", strings.Join(got, "\n"))
+		}
+		for _, l := range fx.logs.all() {
+			if !strings.Contains(l, `"provider"="kvm2-home"`) || !strings.Contains(l, `"host"="beelink"`) {
+				t.Errorf("line lacks provider or host: %s", l)
+			}
+			if !strings.Contains(l, `"name"="`+testName+`"`) && !strings.Contains(l, `"volume"="base-`) {
+				t.Errorf("instance line lacks the name: %s", l)
+			}
+		}
+	})
+	t.Run("v2 has detail", func(t *testing.T) {
+		t.Parallel()
+		fx := newFixture(t, 2)
+		fx.create(testName)
+		if got := fx.logs.matching("<domain"); len(got) == 0 {
+			t.Errorf("V(2) lacks domain XML:\n%s", strings.Join(fx.logs.all(), "\n"))
+		}
+		resolved := fx.logs.matching("kvm2 create resolved")
+		if len(resolved) != 1 {
+			t.Fatalf("V(2) lines matching resolved = %d, want 1", len(resolved))
+		}
+		for _, key := range []string{`"cloudInitBytes"=`, `"image"="base-` + testProvider + `-debian-13.qcow2"`, `"firmware"="uefi"`, `"vcpus"=1`, `"memoryMiB"=1024`, `"machineType"="small"`, `"bridge"="br0"`, `"pool"="default"`} {
+			if !strings.Contains(resolved[0], key) {
+				t.Errorf("resolved line lacks %s: %s", key, resolved[0])
+			}
+		}
+	})
+}
@@ -0,0 +1,124 @@
+package kvm2
+
+import (
+	"bytes"
+	"errors"
+	"fmt"
+	"io"
+	"mime/multipart"
+	"net/textproto"
+	"strings"
+
+	"github.com/kdomanski/iso9660"
+)
+
+// seedLabel is the volume label cloud-init's NoCloud datasource looks for.
+const seedLabel = "cidata"
+
+// guestAgentScript is the vendor-data payload that gets qemu-guest-agent
+// running so the provider can read the VM's address through libvirt. It is
+// a shell script and not a cloud-config on purpose: cloud-init merges a
+// vendor-data cloud-config under the user-data with mergers that keep the
+// user-data's value for every top-level key and never append lists, so a
+// Proxy with its own packages: would silently lose the agent. Scripts
+// bypass the merge.
+const guestAgentScript = `#!/bin/sh
+# Added by egress-proxies-operator: installs and starts qemu-guest-agent so
+# the operator can learn this VM's IP address through libvirt.
+# To leave this out, set installGuestAgent: false in the provider's config.
+set -e
+if command -v qemu-ga >/dev/null 2>&1 || [ -x /usr/sbin/qemu-ga ]; then
+  echo "egress-proxies-operator: qemu-guest-agent already installed"
+elif command -v apt-get >/dev/null 2>&1; then
+  export DEBIAN_FRONTEND=noninteractive
+  apt-get update
+  apt-get install -y qemu-guest-agent
+elif command -v dnf >/dev/null 2>&1; then
+  dnf install -y qemu-guest-agent
+elif command -v yum >/dev/null 2>&1; then
+  yum install -y qemu-guest-agent
+elif command -v zypper >/dev/null 2>&1; then
+  zypper --non-interactive install qemu-guest-agent
+elif command -v apk >/dev/null 2>&1; then
+  apk add qemu-guest-agent
+else
+  echo "egress-proxies-operator: no supported package manager found; install qemu-guest-agent by hand" >&2
+  exit 1
+fi
+if command -v systemctl >/dev/null 2>&1; then
+  systemctl enable --now qemu-guest-agent
+else
+  rc-update add qemu-guest-agent default
+  rc-service qemu-guest-agent start
+fi
+`
+
+// buildSeedISO writes a NoCloud seed to w: user-data byte for byte,
+// meta-data naming the instance, and vendor-data (an empty file when there
+// is none, so cloud-init does not look elsewhere). iso9660.NewWriter stages
+// the files in os.MkdirTemp("", ""), that is under os.TempDir(): $TMPDIR
+// when set, else /tmp, the manager's 64Mi emptyDir. The staging dir is
+// removed before returning. Nothing about the payload is logged here.
+func buildSeedISO(name, userData, vendorData string, w io.Writer) (err error) {
+	if name == "" {
+		return permanentf("seed ISO: instance name is empty")
+	}
+	iw, err := iso9660.NewWriter()
+	if err != nil {
+		return fmt.Errorf("creating seed ISO staging dir: %w", err)
+	}
+	defer func() {
+		if cerr := iw.Cleanup(); cerr != nil {
+			err = errors.Join(err, fmt.Errorf("removing seed ISO staging dir: %w", cerr))
+		}
+	}()
+	metaData := "instance-id: " + name + "\nlocal-hostname: " + name + "\n"
+	files := []struct{ name, content string }{
+		{"user-data", userData},
+		{"meta-data", metaData},
+		{"vendor-data", vendorData},
+	}
+	for _, f := range files {
+		if err := iw.AddFile(strings.NewReader(f.content), f.name); err != nil {
+			return fmt.Errorf("adding %s to seed ISO: %w", f.name, err)
+		}
+	}
+	if err := iw.WriteTo(w, seedLabel); err != nil {
+		return fmt.Errorf("writing seed ISO: %w", err)
+	}
+	return nil
+}
+
+// buildVendorData composes the seed's vendor-data from the configured
+// document and the guest agent script. With both, they travel as separate
+// parts of a MIME multipart, so cloud-init handles each by its own type; the
+// MIME-Version header comes first because cloud-init detects multipart by
+// finding "mime-version:" within the first 4096 bytes.
+func buildVendorData(configured string, installAgent bool) (string, error) {
+	switch {
+	case !installAgent:
+		return configured, nil
+	case strings.TrimSpace(configured) == "":
+		return guestAgentScript, nil
+	}
+	var buf bytes.Buffer
+	mw := multipart.NewWriter(&buf)
+	fmt.Fprintf(&buf, "MIME-Version: 1.0\r\nContent-Type: multipart/mixed; boundary=%q\r\n\r\n", mw.Boundary())
+	parts := []struct{ contentType, body string }{
+		{"text/plain", configured},
+		{"text/x-shellscript", guestAgentScript},
+	}
+	for _, p := range parts {
+		pw, err := mw.CreatePart(textproto.MIMEHeader{"Content-Type": {p.contentType}})
+		if err != nil {
+			return "", fmt.Errorf("creating vendor-data part %s: %w", p.contentType, err)
+		}
+		if _, err := io.WriteString(pw, p.body); err != nil {
+			return "", fmt.Errorf("writing vendor-data part %s: %w", p.contentType, err)
+		}
+	}
+	if err := mw.Close(); err != nil {
+		return "", fmt.Errorf("closing vendor-data multipart: %w", err)
+	}
+	return buf.String(), nil
+}
@@ -0,0 +1,264 @@
+package kvm2
+
+import (
+	"bufio"
+	"bytes"
+	"io"
+	"mime"
+	"mime/multipart"
+	"net/textproto"
+	"os"
+	"strings"
+	"testing"
+
+	"github.com/kdomanski/iso9660"
+)
+
+// readSeed builds a seed and returns its files by name plus the label.
+func readSeed(t *testing.T, name, userData, vendorData string) (map[string]string, string) {
+	t.Helper()
+	var buf bytes.Buffer
+	if err := buildSeedISO(name, userData, vendorData, &buf); err != nil {
+		t.Fatalf("buildSeedISO: %v", err)
+	}
+	img, err := iso9660.OpenImage(bytes.NewReader(buf.Bytes()))
+	if err != nil {
+		t.Fatalf("OpenImage: %v", err)
+	}
+	root, err := img.RootDir()
+	if err != nil {
+		t.Fatalf("RootDir: %v", err)
+	}
+	children, err := root.GetChildren()
+	if err != nil {
+		t.Fatalf("GetChildren: %v", err)
+	}
+	files := make(map[string]string, len(children))
+	for _, c := range children {
+		if c.IsDir() {
+			t.Errorf("unexpected directory %q in seed", c.Name())
+			continue
+		}
+		data, err := io.ReadAll(c.Reader())
+		if err != nil {
+			t.Fatalf("reading %s: %v", c.Name(), err)
+		}
+		files[c.Name()] = string(data)
+	}
+	label, err := img.Label()
+	if err != nil {
+		t.Fatalf("Label: %v", err)
+	}
+	return files, label
+}
+
+func TestSeedISO_filesAndLabel(t *testing.T) {
+	t.Parallel()
+	userData := "#cloud-config\npackages: [squid]\n"
+	vendorData := "#!/bin/sh\necho hi\n"
+	files, label := readSeed(t, "proxy-abc", userData, vendorData)
+	if label != seedLabel {
+		t.Errorf("label = %q, want %q", label, seedLabel)
+	}
+	want := map[string]string{
+		"user-data":   userData,
+		"meta-data":   "instance-id: proxy-abc\nlocal-hostname: proxy-abc\n",
+		"vendor-data": vendorData,
+	}
+	if len(files) != len(want) {
+		t.Errorf("seed has %d files %v, want exactly %d", len(files), keys(files), len(want))
+	}
+	for name, content := range want {
+		if got, ok := files[name]; !ok {
+			t.Errorf("seed lacks %s", name)
+		} else if got != content {
+			t.Errorf("%s = %q, want %q", name, got, content)
+		}
+	}
+}
+
+func keys(m map[string]string) []string {
+	out := make([]string, 0, len(m))
+	for k := range m {
+		out = append(out, k)
+	}
+	return out
+}
+
+func TestSeedISO_userDataByteForByte(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name     string
+		userData string
+	}{
+		{name: "mime multipart with CRLF", userData: "MIME-Version: 1.0\r\nContent-Type: multipart/mixed; boundary=\"b\"\r\n\r\n--b\r\nContent-Type: text/cloud-config\r\n\r\n#cloud-config\r\npackages: [squid]\r\n--b--\r\n"},
+		{name: "jinja header", userData: "## template: jinja\n#cloud-config\nhostname: {{ v1.local_hostname }}\n"},
+		{name: "no trailing newline", userData: "#cloud-config\nruncmd:\n  - echo done"},
+		{name: "bytes above 0x7f", userData: "#cloud-config\nwrite_files:\n  - content: \"caf\xc3\xa9 \xff\xfe\x80\"\n"},
+		{name: "empty", userData: ""},
+		{name: "trailing whitespace preserved", userData: "#!/bin/sh\necho x \n\n\n"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			files, _ := readSeed(t, "proxy-abc", tc.userData, "")
+			if got := files["user-data"]; got != tc.userData {
+				t.Errorf("user-data changed:\n got %q\nwant %q", got, tc.userData)
+			}
+		})
+	}
+}
+
+// Not parallel: it changes TMPDIR for the process.
+func TestSeedISO_tempDirRemoved(t *testing.T) {
+	tmp := t.TempDir()
+	t.Setenv("TMPDIR", tmp)
+	if got := os.TempDir(); got != tmp {
+		t.Fatalf("os.TempDir() = %q, want %q", got, tmp)
+	}
+	var buf bytes.Buffer
+	if err := buildSeedISO("proxy-abc", "#cloud-config\n", guestAgentScript, &buf); err != nil {
+		t.Fatalf("buildSeedISO: %v", err)
+	}
+	if buf.Len() == 0 {
+		t.Fatal("seed ISO is empty")
+	}
+	entries, err := os.ReadDir(tmp)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(entries) != 0 {
+		names := make([]string, 0, len(entries))
+		for _, e := range entries {
+			names = append(names, e.Name())
+		}
+		t.Errorf("staging left behind in TMPDIR: %v", names)
+	}
+}
+
+func TestSeedISO_emptyVendorDataStillHasFile(t *testing.T) {
+	t.Parallel()
+	files, _ := readSeed(t, "proxy-abc", "#cloud-config\n", "")
+	got, ok := files["vendor-data"]
+	if !ok {
+		t.Fatalf("seed lacks vendor-data; files: %v", keys(files))
+	}
+	if got != "" {
+		t.Errorf("vendor-data = %q, want empty", got)
+	}
+}
+
+func TestSeedISO_rejectsEmptyName(t *testing.T) {
+	t.Parallel()
+	var buf bytes.Buffer
+	if err := buildSeedISO("", "#cloud-config\n", "", &buf); err == nil {
+		t.Fatal("buildSeedISO with empty name succeeded")
+	}
+	if buf.Len() != 0 {
+		t.Errorf("wrote %d bytes alongside the error", buf.Len())
+	}
+}
+
+func TestVendorData_isShellScriptNotCloudConfig(t *testing.T) {
+	t.Parallel()
+	got, err := buildVendorData("", true)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.HasPrefix(got, "#!/bin/sh\n") {
+		t.Errorf("vendor-data starts with %q, want #!/bin/sh", firstLine(got))
+	}
+	if strings.HasPrefix(got, "#cloud-config") {
+		t.Error("vendor-data is a cloud-config; cloud-init would merge it away")
+	}
+	for _, want := range []string{"qemu-guest-agent", "set -e", "apt-get", "dnf", "yum", "zypper", "apk", "systemctl enable --now qemu-guest-agent", "rc-update", "installGuestAgent: false", "DEBIAN_FRONTEND=noninteractive"} {
+		if !strings.Contains(got, want) {
+			t.Errorf("script lacks %q", want)
+		}
+	}
+	if strings.Contains(got, "MIME-Version") {
+		t.Error("script alone should not be wrapped in a multipart")
+	}
+}
+
+func firstLine(s string) string {
+	line, _, _ := strings.Cut(s, "\n")
+	return line
+}
+
+func bufioReader(s string) *bufio.Reader {
+	return bufio.NewReader(strings.NewReader(s))
+}
+
+func TestVendorData_multipartWhenExtraProvided(t *testing.T) {
+	t.Parallel()
+	configured := "#cloud-config\nntp:\n  enabled: true\n"
+	got, err := buildVendorData(configured, true)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.HasPrefix(got, "MIME-Version: 1.0\r\n") {
+		t.Errorf("vendor-data starts with %q, want MIME-Version: 1.0", firstLine(got))
+	}
+	r := textproto.NewReader(bufioReader(got))
+	hdr, err := r.ReadMIMEHeader()
+	if err != nil {
+		t.Fatalf("reading top-level headers: %v", err)
+	}
+	mediaType, params, err := mime.ParseMediaType(hdr.Get("Content-Type"))
+	if err != nil {
+		t.Fatalf("parsing Content-Type %q: %v", hdr.Get("Content-Type"), err)
+	}
+	if mediaType != "multipart/mixed" {
+		t.Errorf("media type = %q, want multipart/mixed", mediaType)
+	}
+	mr := multipart.NewReader(r.R, params["boundary"])
+	type part struct{ contentType, body string }
+	var parts []part
+	for {
+		p, err := mr.NextPart()
+		if err == io.EOF {
+			break
+		}
+		if err != nil {
+			t.Fatalf("NextPart: %v", err)
+		}
+		body, err := io.ReadAll(p)
+		if err != nil {
+			t.Fatalf("reading part: %v", err)
+		}
+		parts = append(parts, part{p.Header.Get("Content-Type"), string(body)})
+	}
+	want := []part{{"text/plain", configured}, {"text/x-shellscript", guestAgentScript}}
+	if len(parts) != len(want) {
+		t.Fatalf("got %d parts, want %d", len(parts), len(want))
+	}
+	for i := range want {
+		if parts[i] != want[i] {
+			t.Errorf("part %d = {%q, %q}, want {%q, %q}", i, parts[i].contentType, firstLine(parts[i].body), want[i].contentType, firstLine(want[i].body))
+		}
+	}
+}
+
+func TestVendorData_offWithoutConfiguredIsEmpty(t *testing.T) {
+	t.Parallel()
+	got, err := buildVendorData("", false)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got != "" {
+		t.Errorf("vendor-data = %q, want empty", got)
+	}
+}
+
+func TestVendorData_offWithConfiguredPassesThrough(t *testing.T) {
+	t.Parallel()
+	configured := "#cloud-config\nntp:\n  enabled: true"
+	got, err := buildVendorData(configured, false)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got != configured {
+		t.Errorf("vendor-data = %q, want the configured document unchanged", got)
+	}
+}
@@ -0,0 +1,218 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io/fs"
+	"net"
+	"os"
+	"sync"
+	"time"
+
+	"golang.org/x/crypto/ssh"
+	"golang.org/x/crypto/ssh/knownhosts"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// dialStage names the step of a dial that failed, so the Checker (spec
+// entry 16) can pick a reason without parsing messages.
+type dialStage string
+
+const (
+	// stageCredentials: the key or known_hosts file is missing or unparsable.
+	stageCredentials dialStage = "credentials"
+	// stageTCP: the TCP connection to host.Address failed.
+	stageTCP dialStage = "tcp"
+	// stageSSH: the SSH handshake failed, including host key and auth.
+	stageSSH dialStage = "ssh"
+	// stageSocket: the forwarded channel to the libvirt socket was refused.
+	stageSocket dialStage = "socket"
+)
+
+// dialError tags a dial failure with its stage and host. Unwrap keeps the
+// cause visible, so classify still sees *knownhosts.KeyError, fs.ErrNotExist,
+// the "unable to authenticate" text and *net.OpError.
+type dialError struct {
+	stage dialStage
+	host  string
+	err   error
+}
+
+func (e *dialError) Error() string {
+	return fmt.Sprintf("dialing host %s (%s): %v", e.host, e.stage, e.err)
+}
+
+func (e *dialError) Unwrap() error { return e.err }
+
+// stageOf returns the failed stage of a dial error, if err is one.
+func stageOf(err error) (dialStage, bool) {
+	var de *dialError
+	if !errors.As(err, &de) {
+		return "", false
+	}
+	return de.stage, true
+}
+
+const (
+	defaultKeepaliveInterval = 15 * time.Second
+	keepaliveRequest         = "keepalive@openssh.com"
+)
+
+// credentials is what one dial needs from the mounted files. They are read
+// on every dial and never cached (spec entry 14), so rotating the Secret
+// takes effect on the next connection.
+type credentials struct {
+	signer  ssh.Signer
+	hostKey ssh.HostKeyCallback
+}
+
+// loadCredentials reads and parses the key and known_hosts files. A missing
+// file keeps fs.ErrNotExist (Transient: the Secret has not landed); a file
+// that does not parse is Permanent. Messages name the path and carry the
+// parser's own error, never the file's content.
+func loadCredentials(cfg provider.KVMConfig) (credentials, error) {
+	keyBytes, err := os.ReadFile(cfg.SSHKeyFile)
+	if err != nil {
+		return credentials{}, fmt.Errorf("reading ssh key file %s: %w", cfg.SSHKeyFile, err)
+	}
+	signer, err := ssh.ParsePrivateKey(keyBytes)
+	if err != nil {
+		return credentials{}, permanentf("parsing ssh key file %s: %v", cfg.SSHKeyFile, err)
+	}
+	if cfg.InsecureIgnoreHostKey {
+		return credentials{signer: signer, hostKey: ssh.InsecureIgnoreHostKey()}, nil
+	}
+	hostKey, err := knownhosts.New(cfg.KnownHostsFile)
+	if err != nil {
+		var pathErr *fs.PathError
+		if errors.As(err, &pathErr) {
+			return credentials{}, fmt.Errorf("reading known_hosts file %s: %w", cfg.KnownHostsFile, err)
+		}
+		return credentials{}, permanentf("parsing known_hosts file %s: %v", cfg.KnownHostsFile, err)
+	}
+	return credentials{signer: signer, hostKey: hostKey}, nil
+}
+
+// sshSocket is the libvirt socket forwarded through one SSH connection. The
+// embedded Conn is the channel; Close closes the SSH client, whose TCP
+// connection carries it, and that alone unblocks the channel's readers and
+// writers. Closing the channel as well would be worse than useless: its
+// Close writes a close message under the channel's write lock, so it would
+// queue behind a transfer stuck in a TCP write, which is the case Close
+// exists to survive.
+type sshSocket struct {
+	net.Conn
+	client *ssh.Client
+
+	closeOnce sync.Once
+	closeErr  error
+}
+
+func (s *sshSocket) Close() error {
+	s.closeOnce.Do(func() { s.closeErr = s.client.Close() })
+	return s.closeErr
+}
+
+// keepalive sends keepalive@openssh.com every interval until stop is closed
+// or a request fails or takes longer than the interval, and returns why it
+// stopped (nil for stop). The caller marks its link dead on an error.
+func (s *sshSocket) keepalive(interval time.Duration, stop <-chan struct{}) error {
+	ticker := time.NewTicker(interval)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-stop:
+			return nil
+		case <-ticker.C:
+		}
+		done := make(chan error, 1)
+		go func() {
+			_, _, err := s.client.SendRequest(keepaliveRequest, true, nil)
+			done <- err
+		}()
+		timer := time.NewTimer(interval)
+		select {
+		case err := <-done:
+			timer.Stop()
+			if err != nil {
+				return fmt.Errorf("keepalive: %w", err)
+			}
+		case <-timer.C:
+			return fmt.Errorf("keepalive: no reply within %s", interval)
+		case <-stop:
+			timer.Stop()
+			return nil
+		}
+	}
+}
+
+// dialSSHSocket opens host's libvirt socket through SSH: credentials, TCP,
+// handshake, channel. timeout bounds the whole sequence on top of ctx. Each
+// failure is a *dialError naming its stage. Nothing here logs.
+func dialSSHSocket(ctx context.Context, host provider.KVMHost, cfg provider.KVMConfig, timeout time.Duration) (*sshSocket, error) {
+	if timeout > 0 {
+		var cancel context.CancelFunc
+		ctx, cancel = context.WithTimeout(ctx, timeout)
+		defer cancel()
+	}
+	creds, err := loadCredentials(cfg)
+	if err != nil {
+		return nil, &dialError{stage: stageCredentials, host: host.Name, err: err}
+	}
+	var nd net.Dialer
+	tcp, err := nd.DialContext(ctx, "tcp", host.Address)
+	if err != nil {
+		return nil, &dialError{stage: stageTCP, host: host.Name, err: err}
+	}
+	clientCfg := &ssh.ClientConfig{
+		User:            host.User,
+		Auth:            []ssh.AuthMethod{ssh.PublicKeys(creds.signer)},
+		HostKeyCallback: creds.hostKey,
+		Timeout:         timeout,
+	}
+	client, err := await(ctx, tcp.Close, func() (*ssh.Client, error) {
+		c, chans, reqs, err := ssh.NewClientConn(tcp, host.Address, clientCfg)
+		if err != nil {
+			return nil, err
+		}
+		return ssh.NewClient(c, chans, reqs), nil
+	})
+	if err != nil {
+		_ = tcp.Close()
+		return nil, &dialError{stage: stageSSH, host: host.Name, err: err}
+	}
+	conn, err := await(ctx, func() error { return client.Close() }, func() (net.Conn, error) {
+		return client.Dial("unix", host.Socket)
+	})
+	if err != nil {
+		_ = client.Close()
+		return nil, &dialError{stage: stageSocket, host: host.Name, err: fmt.Errorf("opening %s: %w", host.Socket, err)}
+	}
+	return &sshSocket{Conn: conn, client: client}, nil
+}
+
+// await runs a blocking step in a goroutine. When ctx ends first it calls
+// abort, which closes the transport the step is blocked on, waits for the
+// step to give up, and returns ctx's error instead.
+func await[T any](ctx context.Context, abort func() error, step func() (T, error)) (T, error) {
+	type result struct {
+		v   T
+		err error
+	}
+	done := make(chan result, 1)
+	go func() {
+		v, err := step()
+		done <- result{v, err}
+	}()
+	select {
+	case r := <-done:
+		return r.v, r.err
+	case <-ctx.Done():
+		_ = abort()
+		<-done
+		var zero T
+		return zero, ctx.Err()
+	}
+}
@@ -0,0 +1,343 @@
+package kvm2
+
+import (
+	"context"
+	"errors"
+	"io/fs"
+	"net"
+	"os"
+	"strings"
+	"testing"
+	"time"
+
+	"golang.org/x/crypto/ssh/knownhosts"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// sshFixture is one in-process server with a client key it accepts and a
+// known_hosts file naming its host key.
+type sshFixture struct {
+	server *testSSHServer
+	socket string
+	cfg    provider.KVMConfig
+	host   provider.KVMHost
+}
+
+func newSSHFixture(t *testing.T) *sshFixture {
+	t.Helper()
+	dir := t.TempDir()
+	socket := startEchoSocket(t)
+	keyPath, pub := writeKeyFile(t, dir, "id_ed25519")
+	server := startSSHServer(t, pub, socket)
+	return &sshFixture{
+		server: server,
+		socket: socket,
+		cfg: provider.KVMConfig{
+			SSHKeyFile:     keyPath,
+			KnownHostsFile: writeKnownHosts(t, dir, "known_hosts", server.addr, server.hostKey),
+		},
+		host: provider.KVMHost{Name: "beelink", Address: server.addr, User: "virt", Socket: socket},
+	}
+}
+
+const dialTestTimeout = 5 * time.Second
+
+func TestDial_forwardsUnixSocket(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	for round := range 2 {
+		sock, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+		if err != nil {
+			t.Fatalf("round %d: dial error = %v", round, err)
+		}
+		msg := []byte("ping over libvirt-sock")
+		if _, err := sock.Write(msg); err != nil {
+			t.Fatalf("round %d: write: %v", round, err)
+		}
+		buf := make([]byte, len(msg))
+		_ = sock.SetReadDeadline(time.Now().Add(dialTestTimeout))
+		if _, err := readFull(sock, buf); err != nil {
+			t.Fatalf("round %d: read: %v", round, err)
+		}
+		if string(buf) != string(msg) {
+			t.Errorf("round %d: echoed %q, want %q", round, buf, msg)
+		}
+		if err := sock.Close(); err != nil {
+			t.Errorf("round %d: close error = %v", round, err)
+		}
+		if err := sock.Close(); err != nil {
+			t.Errorf("round %d: second close error = %v, want nil", round, err)
+		}
+	}
+	for _, p := range fx.server.requestedPaths() {
+		if p != fx.socket {
+			t.Errorf("server saw channel for %q, want %q", p, fx.socket)
+		}
+	}
+}
+
+func readFull(sock *sshSocket, buf []byte) (int, error) {
+	n := 0
+	for n < len(buf) {
+		m, err := sock.Read(buf[n:])
+		n += m
+		if err != nil {
+			return n, err
+		}
+	}
+	return n, nil
+}
+
+func TestDial_unknownHostKeyRejected(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.cfg.KnownHostsFile = writeKnownHosts(t, t.TempDir(), "known_hosts", "other.example:22", fx.server.hostKey)
+	_, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+	assertDialFailure(t, err, stageSSH, provider.ErrPermanent)
+	var keyErr *knownhosts.KeyError
+	if !errors.As(err, &keyErr) {
+		t.Fatalf("error = %v, want *knownhosts.KeyError", err)
+	}
+	if len(keyErr.Want) != 0 {
+		t.Errorf("KeyError.Want = %v, want empty (host unknown)", keyErr.Want)
+	}
+}
+
+func TestDial_mismatchedHostKeyRejected(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.cfg.KnownHostsFile = writeKnownHosts(t, t.TempDir(), "known_hosts", fx.server.addr, otherPublicKey(t))
+	_, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+	assertDialFailure(t, err, stageSSH, provider.ErrPermanent)
+	var keyErr *knownhosts.KeyError
+	if !errors.As(err, &keyErr) {
+		t.Fatalf("error = %v, want *knownhosts.KeyError", err)
+	}
+	if len(keyErr.Want) == 0 {
+		t.Error("KeyError.Want empty, want the mismatching known key")
+	}
+}
+
+func TestDial_insecureIgnoreHostKeyAccepts(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.cfg.KnownHostsFile = ""
+	fx.cfg.InsecureIgnoreHostKey = true
+	sock, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+	if err != nil {
+		t.Fatalf("dial error = %v, want nil with host keys ignored", err)
+	}
+	_ = sock.Close()
+}
+
+func TestDial_missingKeyFileIsTransient(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.cfg.SSHKeyFile = fx.cfg.SSHKeyFile + ".missing"
+	_, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+	assertDialFailure(t, err, stageCredentials, provider.ErrTransient)
+	if !errors.Is(err, fs.ErrNotExist) {
+		t.Errorf("error = %v, want fs.ErrNotExist", err)
+	}
+	if !strings.Contains(err.Error(), fx.cfg.SSHKeyFile) {
+		t.Errorf("error = %q, want the path in it", err)
+	}
+}
+
+func TestDial_missingKnownHostsIsTransient(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.cfg.KnownHostsFile = fx.cfg.KnownHostsFile + ".missing"
+	_, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+	assertDialFailure(t, err, stageCredentials, provider.ErrTransient)
+	if !errors.Is(err, fs.ErrNotExist) {
+		t.Errorf("error = %v, want fs.ErrNotExist", err)
+	}
+}
+
+func TestDial_unparsableKeyIsPermanent(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.cfg.SSHKeyFile = writeFile(t, t.TempDir(), "id_bad", "this is not a key\n")
+	_, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+	assertDialFailure(t, err, stageCredentials, provider.ErrPermanent)
+	if !strings.Contains(err.Error(), "parsing ssh key file "+fx.cfg.SSHKeyFile) {
+		t.Errorf("error = %q, want 'parsing ssh key file <path>'", err)
+	}
+}
+
+func TestDial_unparsableKnownHostsIsPermanent(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.cfg.KnownHostsFile = writeFile(t, t.TempDir(), "known_hosts", "host.example ssh-ed25519 not-base64!\n")
+	_, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+	assertDialFailure(t, err, stageCredentials, provider.ErrPermanent)
+	if !strings.Contains(err.Error(), "parsing known_hosts file "+fx.cfg.KnownHostsFile) {
+		t.Errorf("error = %q, want 'parsing known_hosts file <path>'", err)
+	}
+}
+
+func TestDial_wrongKeyIsUnauthorized(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.cfg.SSHKeyFile, _ = writeKeyFile(t, t.TempDir(), "id_other")
+	_, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+	assertDialFailure(t, err, stageSSH, provider.ErrPermanent)
+	if !strings.Contains(err.Error(), sshAuthFailed) {
+		t.Errorf("error = %q, want %q in it", err, sshAuthFailed)
+	}
+}
+
+func TestDial_handshakeTimeout(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.host.Address = silentListener(t)
+	const budget = 200 * time.Millisecond
+	start := time.Now()
+	_, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, budget)
+	elapsed := time.Since(start)
+	assertDialFailure(t, err, stageSSH, provider.ErrTransient)
+	if !errors.Is(err, context.DeadlineExceeded) {
+		t.Errorf("error = %v, want context.DeadlineExceeded", err)
+	}
+	if elapsed > budget+time.Second {
+		t.Errorf("dial took %v, want about the %v budget", elapsed, budget)
+	}
+}
+
+func TestDial_cancelledContext(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.host.Address = silentListener(t)
+	ctx, cancel := context.WithCancel(context.Background())
+	go func() {
+		time.Sleep(20 * time.Millisecond)
+		cancel()
+	}()
+	start := time.Now()
+	_, err := dialSSHSocket(ctx, fx.host, fx.cfg, dialTestTimeout)
+	if time.Since(start) > time.Second {
+		t.Errorf("dial took %v after cancel", time.Since(start))
+	}
+	assertDialFailure(t, err, stageSSH, provider.ErrTransient)
+	if !errors.Is(err, context.Canceled) {
+		t.Errorf("error = %v, want context.Canceled", err)
+	}
+}
+
+func TestDial_refusedIsTransient(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.host.Address = closedPort(t)
+	_, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+	assertDialFailure(t, err, stageTCP, provider.ErrTransient)
+	var opErr *net.OpError
+	if !errors.As(err, &opErr) {
+		t.Errorf("error = %v, want *net.OpError", err)
+	}
+}
+
+func TestDial_socketRefusedIsSocketStage(t *testing.T) {
+	t.Parallel()
+	fx := newSSHFixture(t)
+	fx.host.Socket = "/run/nonexistent/libvirt-sock"
+	_, err := dialSSHSocket(context.Background(), fx.host, fx.cfg, dialTestTimeout)
+	assertDialFailure(t, err, stageSocket, provider.ErrTransient)
+	if !strings.Contains(err.Error(), fx.host.Socket) {
+		t.Errorf("error = %q, want the socket path in it", err)
+	}
+}
+
+func TestSSH_keyMaterialNeverInErrorsOrLogs(t *testing.T) {
+	t.Parallel()
+	const keySentinel = "SENTINEL-KEY-4f9c2b"
+	const hostsSentinel = "SENTINEL-HOSTS-8a1d7e"
+	fx := newSSHFixture(t)
+	dir := t.TempDir()
+	tests := []struct {
+		name string
+		cfg  func(c *provider.KVMConfig)
+	}{
+		{
+			name: "bad key file",
+			cfg: func(c *provider.KVMConfig) {
+				c.SSHKeyFile = writeFile(t, dir, "id_bad", "-----BEGIN NOT A KEY-----\n"+keySentinel+"\n-----END NOT A KEY-----\n")
+			},
+		},
+		{
+			name: "bad known_hosts file",
+			cfg: func(c *provider.KVMConfig) {
+				c.KnownHostsFile = writeFile(t, dir, "known_hosts_bad", "beelink ssh-ed25519 "+hostsSentinel+"\n")
+			},
+		},
+		{
+			name: "good key, host unknown",
+			cfg: func(c *provider.KVMConfig) {
+				c.KnownHostsFile = writeFile(t, dir, "known_hosts_other", knownhosts.Line([]string{"other.example:22"}, fx.server.hostKey)+"\n")
+			},
+		},
+	}
+	keyBytes := readFileT(t, fx.cfg.SSHKeyFile)
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			cfg := fx.cfg
+			tc.cfg(&cfg)
+			_, err := dialSSHSocket(context.Background(), fx.host, cfg, dialTestTimeout)
+			if err == nil {
+				t.Fatal("dial succeeded, want an error")
+			}
+			msg := err.Error()
+			for _, s := range []string{keySentinel, hostsSentinel} {
+				if strings.Contains(msg, s) {
+					t.Errorf("error %q contains sentinel %q", msg, s)
+				}
+			}
+			if strings.Contains(msg, keyBytes) || strings.Contains(msg, "PRIVATE KEY") {
+				t.Errorf("error %q contains key material", msg)
+			}
+		})
+	}
+}
+
+func readFileT(t *testing.T, path string) string {
+	t.Helper()
+	b, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("reading %s: %v", path, err)
+	}
+	return strings.TrimSpace(string(b))
+}
+
+func assertDialFailure(t *testing.T, err error, wantStage dialStage, wantClass error) {
+	t.Helper()
+	if err == nil {
+		t.Fatal("dial succeeded, want an error")
+	}
+	stage, ok := stageOf(err)
+	if !ok {
+		t.Fatalf("error %v is not a *dialError", err)
+	}
+	if stage != wantStage {
+		t.Errorf("stage = %q, want %q (error: %v)", stage, wantStage, err)
+	}
+	if got := classify("get", err); !errors.Is(got, wantClass) {
+		t.Errorf("classify() = %v, want %v (error: %v)", got, wantClass, err)
+	}
+}
+
+func TestDialError_unwrapsToCause(t *testing.T) {
+	t.Parallel()
+	cause := &knownhosts.KeyError{}
+	err := &dialError{stage: stageSSH, host: "beelink", err: cause}
+	if !errors.Is(err, cause) {
+		t.Error("errors.Is(dialError, cause) = false")
+	}
+	if !strings.Contains(err.Error(), "beelink") || !strings.Contains(err.Error(), "(ssh)") {
+		t.Errorf("Error() = %q, want the host and the stage", err)
+	}
+	if _, ok := stageOf(errors.New("plain")); ok {
+		t.Error("stageOf(plain error) reported a stage")
+	}
+}
@@ -0,0 +1,289 @@
+package kvm2
+
+import (
+	"bytes"
+	"crypto/ed25519"
+	"crypto/rand"
+	"encoding/pem"
+	"errors"
+	"io"
+	"net"
+	"os"
+	"path/filepath"
+	"sync"
+	"testing"
+
+	"golang.org/x/crypto/ssh"
+	"golang.org/x/crypto/ssh/knownhosts"
+)
+
+// testSSHServer is an in-process x/crypto/ssh server that accepts one
+// authorised public key and forwards direct-streamlocal@openssh.com channels
+// for socketPath to that unix socket, rejecting every other channel.
+type testSSHServer struct {
+	addr       string
+	hostKey    ssh.PublicKey
+	socketPath string
+
+	listener net.Listener
+	mu       sync.Mutex
+	conns    []net.Conn
+	wg       sync.WaitGroup
+	stopOnce sync.Once
+
+	reqMu     sync.Mutex
+	requested []string
+}
+
+func startSSHServer(t *testing.T, authorized ssh.PublicKey, socketPath string) *testSSHServer {
+	t.Helper()
+	_, hostPriv, err := ed25519.GenerateKey(rand.Reader)
+	if err != nil {
+		t.Fatalf("generating host key: %v", err)
+	}
+	hostSigner, err := ssh.NewSignerFromKey(hostPriv)
+	if err != nil {
+		t.Fatalf("host signer: %v", err)
+	}
+	cfg := &ssh.ServerConfig{
+		PublicKeyCallback: func(_ ssh.ConnMetadata, key ssh.PublicKey) (*ssh.Permissions, error) {
+			if bytes.Equal(key.Marshal(), authorized.Marshal()) {
+				return &ssh.Permissions{}, nil
+			}
+			return nil, errors.New("key not authorised")
+		},
+	}
+	cfg.AddHostKey(hostSigner)
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatalf("listen: %v", err)
+	}
+	s := &testSSHServer{addr: ln.Addr().String(), hostKey: hostSigner.PublicKey(), socketPath: socketPath, listener: ln}
+	s.wg.Add(1)
+	go s.serve(cfg)
+	t.Cleanup(s.stop)
+	return s
+}
+
+func (s *testSSHServer) serve(cfg *ssh.ServerConfig) {
+	defer s.wg.Done()
+	for {
+		conn, err := s.listener.Accept()
+		if err != nil {
+			return
+		}
+		s.mu.Lock()
+		s.conns = append(s.conns, conn)
+		s.mu.Unlock()
+		s.wg.Add(1)
+		go s.handle(conn, cfg)
+	}
+}
+
+func (s *testSSHServer) handle(conn net.Conn, cfg *ssh.ServerConfig) {
+	defer s.wg.Done()
+	sc, chans, reqs, err := ssh.NewServerConn(conn, cfg)
+	if err != nil {
+		_ = conn.Close()
+		return
+	}
+	defer sc.Close()
+	go ssh.DiscardRequests(reqs)
+	for nc := range chans {
+		if nc.ChannelType() != "direct-streamlocal@openssh.com" {
+			_ = nc.Reject(ssh.UnknownChannelType, "only direct-streamlocal@openssh.com is served")
+			continue
+		}
+		var open struct {
+			SocketPath string
+			Reserved0  string
+			Reserved1  uint32
+		}
+		if err := ssh.Unmarshal(nc.ExtraData(), &open); err != nil {
+			_ = nc.Reject(ssh.ConnectionFailed, "bad direct-streamlocal payload")
+			continue
+		}
+		s.reqMu.Lock()
+		s.requested = append(s.requested, open.SocketPath)
+		s.reqMu.Unlock()
+		if open.SocketPath != s.socketPath {
+			_ = nc.Reject(ssh.ConnectionFailed, "no such socket")
+			continue
+		}
+		target, err := net.Dial("unix", s.socketPath)
+		if err != nil {
+			_ = nc.Reject(ssh.ConnectionFailed, "socket unavailable")
+			continue
+		}
+		ch, chReqs, err := nc.Accept()
+		if err != nil {
+			_ = target.Close()
+			continue
+		}
+		go ssh.DiscardRequests(chReqs)
+		s.wg.Add(1)
+		go func() {
+			defer s.wg.Done()
+			proxy(ch, target)
+		}()
+	}
+}
+
+func proxy(ch ssh.Channel, target net.Conn) {
+	var wg sync.WaitGroup
+	wg.Add(2)
+	go func() {
+		defer wg.Done()
+		_, _ = io.Copy(target, ch)
+		_ = target.Close()
+	}()
+	go func() {
+		defer wg.Done()
+		_, _ = io.Copy(ch, target)
+		_ = ch.Close()
+	}()
+	wg.Wait()
+}
+
+func (s *testSSHServer) stop() {
+	s.stopOnce.Do(func() {
+		_ = s.listener.Close()
+		s.mu.Lock()
+		for _, c := range s.conns {
+			_ = c.Close()
+		}
+		s.mu.Unlock()
+		s.wg.Wait()
+	})
+}
+
+func (s *testSSHServer) requestedPaths() []string {
+	s.reqMu.Lock()
+	defer s.reqMu.Unlock()
+	return append([]string(nil), s.requested...)
+}
+
+// startEchoSocket listens on a unix socket that echoes what it reads. The
+// path is kept short: unix socket paths are limited to about 100 bytes and
+// t.TempDir() names are long.
+func startEchoSocket(t *testing.T) string {
+	t.Helper()
+	dir, err := os.MkdirTemp("", "kvm2")
+	if err != nil {
+		t.Fatalf("temp dir: %v", err)
+	}
+	t.Cleanup(func() { _ = os.RemoveAll(dir) })
+	path := filepath.Join(dir, "libvirt-sock")
+	ln, err := net.Listen("unix", path)
+	if err != nil {
+		t.Fatalf("listen unix: %v", err)
+	}
+	t.Cleanup(func() { _ = ln.Close() })
+	go func() {
+		for {
+			c, err := ln.Accept()
+			if err != nil {
+				return
+			}
+			go func() {
+				_, _ = io.Copy(c, c)
+				_ = c.Close()
+			}()
+		}
+	}()
+	return path
+}
+
+// silentListener accepts TCP connections and never speaks, for handshake
+// timeouts. It returns the address.
+func silentListener(t *testing.T) string {
+	t.Helper()
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatalf("listen: %v", err)
+	}
+	var conns []net.Conn
+	var mu sync.Mutex
+	go func() {
+		for {
+			c, err := ln.Accept()
+			if err != nil {
+				return
+			}
+			mu.Lock()
+			conns = append(conns, c)
+			mu.Unlock()
+		}
+	}()
+	t.Cleanup(func() {
+		_ = ln.Close()
+		mu.Lock()
+		defer mu.Unlock()
+		for _, c := range conns {
+			_ = c.Close()
+		}
+	})
+	return ln.Addr().String()
+}
+
+// closedPort returns an address nothing listens on.
+func closedPort(t *testing.T) string {
+	t.Helper()
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatalf("listen: %v", err)
+	}
+	addr := ln.Addr().String()
+	_ = ln.Close()
+	return addr
+}
+
+// writeKeyFile generates an ed25519 key, writes it in OpenSSH format and
+// returns the path and the public half.
+func writeKeyFile(t *testing.T, dir, name string) (string, ssh.PublicKey) {
+	t.Helper()
+	pub, priv, err := ed25519.GenerateKey(rand.Reader)
+	if err != nil {
+		t.Fatalf("generating key: %v", err)
+	}
+	block, err := ssh.MarshalPrivateKey(priv, "kvm2 test")
+	if err != nil {
+		t.Fatalf("marshalling key: %v", err)
+	}
+	path := writeFile(t, dir, name, string(pem.EncodeToMemory(block)))
+	sshPub, err := ssh.NewPublicKey(pub)
+	if err != nil {
+		t.Fatalf("public key: %v", err)
+	}
+	return path, sshPub
+}
+
+// writeKnownHosts writes one known_hosts line for addr and key.
+func writeKnownHosts(t *testing.T, dir, name, addr string, key ssh.PublicKey) string {
+	t.Helper()
+	return writeFile(t, dir, name, knownhosts.Line([]string{addr}, key)+"\n")
+}
+
+func writeFile(t *testing.T, dir, name, content string) string {
+	t.Helper()
+	path := filepath.Join(dir, name)
+	if err := os.WriteFile(path, []byte(content), 0o600); err != nil {
+		t.Fatalf("writing %s: %v", path, err)
+	}
+	return path
+}
+
+// otherPublicKey returns a public key unrelated to any file, for mismatch
+// cases.
+func otherPublicKey(t *testing.T) ssh.PublicKey {
+	t.Helper()
+	pub, _, err := ed25519.GenerateKey(rand.Reader)
+	if err != nil {
+		t.Fatalf("generating key: %v", err)
+	}
+	sshPub, err := ssh.NewPublicKey(pub)
+	if err != nil {
+		t.Fatalf("public key: %v", err)
+	}
+	return sshPub
+}
@@ -0,0 +1,49 @@
+package kvm2
+
+import (
+	"net/netip"
+
+	"github.com/digitalocean/go-libvirt"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// mapState turns a libvirt domain state plus the discovered address into the
+// contract's state. nostate means libvirt could not tell, so the next poll
+// decides; an active domain is Running only once it has an address; every
+// inactive state, and anything unknown, is Stopped, which the reconciler
+// answers by replacing the instance.
+func mapState(s libvirt.DomainState, ip string) provider.InstanceState {
+	switch {
+	case s == libvirt.DomainNostate:
+		return provider.StateProvisioning
+	case isActive(s) && ip == "":
+		return provider.StateProvisioning
+	case isActive(s):
+		return provider.StateRunning
+	default:
+		return provider.StateStopped
+	}
+}
+
+func isActive(s libvirt.DomainState) bool {
+	return s == libvirt.DomainRunning || s == libvirt.DomainBlocked
+}
+
+// firstIPv4 picks the first usable IPv4 address from the guest's interfaces:
+// not loopback, not link-local, not unspecified. Empty means no address yet.
+func firstIPv4(ifaces []libvirt.DomainInterface) string {
+	for _, iface := range ifaces {
+		for _, a := range iface.Addrs {
+			if libvirt.IPAddrType(a.Type) != libvirt.IPAddrTypeIpv4 {
+				continue
+			}
+			addr, err := netip.ParseAddr(a.Addr)
+			if err != nil || !addr.Is4() || addr.IsLoopback() || addr.IsLinkLocalUnicast() || addr.IsUnspecified() {
+				continue
+			}
+			return addr.String()
+		}
+	}
+	return ""
+}
@@ -0,0 +1,92 @@
+package kvm2
+
+import (
+	"testing"
+
+	"github.com/digitalocean/go-libvirt"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+func TestMapState_table(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		state   libvirt.DomainState
+		noIP    provider.InstanceState
+		withIP  provider.InstanceState
+		display string
+	}{
+		{libvirt.DomainNostate, provider.StateProvisioning, provider.StateProvisioning, "nostate"},
+		{libvirt.DomainRunning, provider.StateProvisioning, provider.StateRunning, "running"},
+		{libvirt.DomainBlocked, provider.StateProvisioning, provider.StateRunning, "blocked"},
+		{libvirt.DomainPaused, provider.StateStopped, provider.StateStopped, "paused"},
+		{libvirt.DomainShutdown, provider.StateStopped, provider.StateStopped, "shutdown"},
+		{libvirt.DomainShutoff, provider.StateStopped, provider.StateStopped, "shutoff"},
+		{libvirt.DomainCrashed, provider.StateStopped, provider.StateStopped, "crashed"},
+		{libvirt.DomainPmsuspended, provider.StateStopped, provider.StateStopped, "pmsuspended"},
+		{libvirt.DomainState(99), provider.StateStopped, provider.StateStopped, "unknown code 99"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.display, func(t *testing.T) {
+			t.Parallel()
+			if got := mapState(tc.state, ""); got != tc.noIP {
+				t.Errorf("mapState(%s, \"\") = %s, want %s", tc.display, got, tc.noIP)
+			}
+			if got := mapState(tc.state, "10.0.0.5"); got != tc.withIP {
+				t.Errorf("mapState(%s, ip) = %s, want %s", tc.display, got, tc.withIP)
+			}
+		})
+	}
+}
+
+func TestIsActive(t *testing.T) {
+	t.Parallel()
+	for s, want := range map[libvirt.DomainState]bool{
+		libvirt.DomainRunning: true, libvirt.DomainBlocked: true,
+		libvirt.DomainNostate: false, libvirt.DomainPaused: false, libvirt.DomainShutoff: false, libvirt.DomainState(99): false,
+	} {
+		if got := isActive(s); got != want {
+			t.Errorf("isActive(%d) = %v, want %v", s, got, want)
+		}
+	}
+}
+
+func addr(typ libvirt.IPAddrType, a string) libvirt.DomainIPAddr {
+	return libvirt.DomainIPAddr{Type: int32(typ), Addr: a}
+}
+
+func TestFirstIPv4_filters(t *testing.T) {
+	t.Parallel()
+	v4 := libvirt.IPAddrTypeIpv4
+	v6 := libvirt.IPAddrTypeIpv6
+	tests := []struct {
+		name   string
+		ifaces []libvirt.DomainInterface
+		want   string
+	}{
+		{name: "empty list", ifaces: nil, want: ""},
+		{name: "loopback only", ifaces: []libvirt.DomainInterface{{Name: "lo", Addrs: []libvirt.DomainIPAddr{addr(v4, "127.0.0.1")}}}, want: ""},
+		{name: "link-local only", ifaces: []libvirt.DomainInterface{{Name: "eth0", Addrs: []libvirt.DomainIPAddr{addr(v4, "169.254.10.1")}}}, want: ""},
+		{name: "ipv6 only", ifaces: []libvirt.DomainInterface{{Name: "eth0", Addrs: []libvirt.DomainIPAddr{addr(v6, "2001:db8::1"), addr(v6, "fe80::1")}}}, want: ""},
+		{name: "unspecified", ifaces: []libvirt.DomainInterface{{Name: "eth0", Addrs: []libvirt.DomainIPAddr{addr(v4, "0.0.0.0")}}}, want: ""},
+		{name: "unparsable", ifaces: []libvirt.DomainInterface{{Name: "eth0", Addrs: []libvirt.DomainIPAddr{addr(v4, "not-an-ip")}}}, want: ""},
+		{name: "ipv4 literal tagged as ipv6 is skipped", ifaces: []libvirt.DomainInterface{{Name: "eth0", Addrs: []libvirt.DomainIPAddr{addr(v6, "10.0.0.7")}}}, want: ""},
+		{name: "ipv6 literal tagged as ipv4 is skipped", ifaces: []libvirt.DomainInterface{{Name: "eth0", Addrs: []libvirt.DomainIPAddr{addr(v4, "2001:db8::1")}}}, want: ""},
+		{
+			name: "good one after bad ones across interfaces",
+			ifaces: []libvirt.DomainInterface{
+				{Name: "lo", Addrs: []libvirt.DomainIPAddr{addr(v4, "127.0.0.1")}},
+				{Name: "eth0", Addrs: []libvirt.DomainIPAddr{addr(v6, "fe80::5054:ff:fe12:3456"), addr(v4, "169.254.1.1"), addr(v4, "bogus"), addr(v4, "192.168.1.42"), addr(v4, "192.168.1.43")}},
+			},
+			want: "192.168.1.42",
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			if got := firstIPv4(tc.ifaces); got != tc.want {
+				t.Errorf("firstIPv4() = %q, want %q", got, tc.want)
+			}
+		})
+	}
+}
@@ -0,0 +1,102 @@
+package kvm2
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"encoding/xml"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// Volume names in the host's pool (spec entries 4 and 10). The per-VM
+// volumes derive from the domain name; the base image from its catalogue
+// key or, for a URL image, from a hash of the URL.
+func overlayVolumeName(name string) string { return name + ".qcow2" }
+func seedVolumeName(name string) string    { return name + "-seed.iso" }
+
+// baseVolumeName carries the provider entry because the base is shared by
+// every Proxy on a host but must not be shared by two entries: the kvm
+// provider builds base-<key>.<format> in the same pool from the same
+// catalogue shape, and two kvm2 entries pointing one key at different URLs
+// would otherwise agree on the name and disagree on the contents.
+func baseVolumeName(entry, key, format string) string {
+	return "base-" + entry + "-" + key + "." + format
+}
+func markerVolumeName(baseName string) string { return baseName + ".sha256" }
+
+// urlImageKey stands in for the catalogue key of an image given as a URL:
+// the first 16 hex characters of sha256(url).
+func urlImageKey(url string) string {
+	sum := sha256.Sum256([]byte(url))
+	return hex.EncodeToString(sum[:])[:16]
+}
+
+// volumeSpec is what StorageVolCreateXML needs. BackingPath and
+// BackingFormat are set for an overlay only.
+type volumeSpec struct {
+	Name          string
+	CapacityBytes int64
+	Format        string
+	BackingPath   string
+	BackingFormat string
+}
+
+type volumeXML struct {
+	XMLName      xml.Name          `xml:"volume"`
+	Type         string            `xml:"type,attr"`
+	Name         string            `xml:"name"`
+	Capacity     volumeCapacityXML `xml:"capacity"`
+	Target       volumeTargetXML   `xml:"target"`
+	BackingStore *backingStoreXML  `xml:"backingStore"`
+}
+
+type volumeCapacityXML struct {
+	Unit  string `xml:"unit,attr"`
+	Value int64  `xml:",chardata"`
+}
+
+type volumeTargetXML struct {
+	Format volumeFormatXML `xml:"format"`
+}
+
+type volumeFormatXML struct {
+	Type string `xml:"type,attr"`
+}
+
+type backingStoreXML struct {
+	Path   string          `xml:"path"`
+	Format volumeFormatXML `xml:"format"`
+}
+
+func validVolumeFormat(f string) bool {
+	return f == provider.KVMFormatQcow2 || f == provider.KVMFormatRaw
+}
+
+func buildVolumeXML(v volumeSpec) (string, error) {
+	if v.Name == "" {
+		return "", permanentf("volume XML: name is empty")
+	}
+	if v.CapacityBytes <= 0 {
+		return "", permanentf("volume XML for %q: capacity must be positive, got %d", v.Name, v.CapacityBytes)
+	}
+	if !validVolumeFormat(v.Format) {
+		return "", permanentf("volume XML for %q: unknown format %q (want %s or %s)", v.Name, v.Format, provider.KVMFormatQcow2, provider.KVMFormatRaw)
+	}
+	x := volumeXML{
+		Type:     "file",
+		Name:     v.Name,
+		Capacity: volumeCapacityXML{Unit: "bytes", Value: v.CapacityBytes},
+		Target:   volumeTargetXML{Format: volumeFormatXML{Type: v.Format}},
+	}
+	if v.BackingPath != "" {
+		if !validVolumeFormat(v.BackingFormat) {
+			return "", permanentf("volume XML for %q: unknown backing format %q (want %s or %s)", v.Name, v.BackingFormat, provider.KVMFormatQcow2, provider.KVMFormatRaw)
+		}
+		x.BackingStore = &backingStoreXML{Path: v.BackingPath, Format: volumeFormatXML{Type: v.BackingFormat}}
+	}
+	out, err := xml.MarshalIndent(x, "", "  ")
+	if err != nil {
+		return "", permanentf("volume XML for %q: %v", v.Name, err)
+	}
+	return string(out), nil
+}
@@ -0,0 +1,156 @@
+package kvm2
+
+import (
+	"encoding/xml"
+	"errors"
+	"regexp"
+	"testing"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+type checkedVolume struct {
+	Type     string `xml:"type,attr"`
+	Name     string `xml:"name"`
+	Capacity struct {
+		Unit  string `xml:"unit,attr"`
+		Value int64  `xml:",chardata"`
+	} `xml:"capacity"`
+	Target struct {
+		Format struct {
+			Type string `xml:"type,attr"`
+		} `xml:"format"`
+	} `xml:"target"`
+	BackingStores []struct {
+		Path   string `xml:"path"`
+		Format struct {
+			Type string `xml:"type,attr"`
+		} `xml:"format"`
+	} `xml:"backingStore"`
+}
+
+func mustBuildVolume(t *testing.T, v volumeSpec) checkedVolume {
+	t.Helper()
+	out, err := buildVolumeXML(v)
+	if err != nil {
+		t.Fatalf("buildVolumeXML: %v", err)
+	}
+	var c checkedVolume
+	if err := xml.Unmarshal([]byte(out), &c); err != nil {
+		t.Fatalf("unmarshalling produced XML: %v\n%s", err, out)
+	}
+	return c
+}
+
+func TestVolumeXML_overlayBackingStore(t *testing.T) {
+	t.Parallel()
+	c := mustBuildVolume(t, volumeSpec{
+		Name:          "proxy-abc.qcow2",
+		CapacityBytes: 10 << 30,
+		Format:        provider.KVMFormatQcow2,
+		BackingPath:   "/var/lib/libvirt/images/base-debian-13.qcow2",
+		BackingFormat: provider.KVMFormatQcow2,
+	})
+	if c.Type != "file" || c.Name != "proxy-abc.qcow2" {
+		t.Errorf("type=%q name=%q, want file/proxy-abc.qcow2", c.Type, c.Name)
+	}
+	if c.Capacity.Unit != "bytes" || c.Capacity.Value != 10<<30 {
+		t.Errorf("capacity = %d %s, want %d bytes", c.Capacity.Value, c.Capacity.Unit, 10<<30)
+	}
+	if c.Target.Format.Type != "qcow2" {
+		t.Errorf("target format = %q, want qcow2", c.Target.Format.Type)
+	}
+	if len(c.BackingStores) != 1 {
+		t.Fatalf("got %d backingStore elements, want 1", len(c.BackingStores))
+	}
+	bs := c.BackingStores[0]
+	if bs.Path != "/var/lib/libvirt/images/base-debian-13.qcow2" || bs.Format.Type != "qcow2" {
+		t.Errorf("backingStore = %+v, want the base path with qcow2 format", bs)
+	}
+}
+
+func TestVolumeXML_rawWithoutBacking(t *testing.T) {
+	t.Parallel()
+	c := mustBuildVolume(t, volumeSpec{Name: "proxy-abc-seed.iso", CapacityBytes: 385024, Format: provider.KVMFormatRaw})
+	if c.Target.Format.Type != "raw" {
+		t.Errorf("target format = %q, want raw", c.Target.Format.Type)
+	}
+	if c.Capacity.Value != 385024 || c.Capacity.Unit != "bytes" {
+		t.Errorf("capacity = %d %s, want 385024 bytes", c.Capacity.Value, c.Capacity.Unit)
+	}
+	if len(c.BackingStores) != 0 {
+		t.Errorf("got backingStore %+v, want none", c.BackingStores)
+	}
+}
+
+func TestVolumeXML_rejectsBadFormat(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		spec volumeSpec
+	}{
+		{name: "vmdk format", spec: volumeSpec{Name: "x", CapacityBytes: 1, Format: "vmdk"}},
+		{name: "empty format", spec: volumeSpec{Name: "x", CapacityBytes: 1}},
+		{name: "bad backing format", spec: volumeSpec{Name: "x", CapacityBytes: 1, Format: "qcow2", BackingPath: "/b", BackingFormat: "vdi"}},
+		{name: "backing path without format", spec: volumeSpec{Name: "x", CapacityBytes: 1, Format: "qcow2", BackingPath: "/b"}},
+		{name: "empty name", spec: volumeSpec{CapacityBytes: 1, Format: "qcow2"}},
+		{name: "zero capacity", spec: volumeSpec{Name: "x", Format: "qcow2"}},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			out, err := buildVolumeXML(tc.spec)
+			if !errors.Is(err, provider.ErrPermanent) {
+				t.Errorf("err = %v, want ErrPermanent", err)
+			}
+			if out != "" {
+				t.Errorf("got XML %q alongside the error", out)
+			}
+		})
+	}
+}
+
+func TestVolumeNames(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name string
+		got  string
+		want string
+	}{
+		{name: "overlay", got: overlayVolumeName("proxy-abc"), want: "proxy-abc.qcow2"},
+		{name: "seed", got: seedVolumeName("proxy-abc"), want: "proxy-abc-seed.iso"},
+		{name: "base qcow2", got: baseVolumeName("kvm2-home", "debian-13", provider.KVMFormatQcow2), want: "base-kvm2-home-debian-13.qcow2"},
+		{name: "base raw", got: baseVolumeName("kvm2-home", "alpine", provider.KVMFormatRaw), want: "base-kvm2-home-alpine.raw"},
+		{name: "marker", got: markerVolumeName("base-debian-13.qcow2"), want: "base-debian-13.qcow2.sha256"},
+		{name: "marker of url base", got: markerVolumeName(baseVolumeName("kvm2-home", urlImageKey("https://x/y.img"), "qcow2")), want: "base-kvm2-home-" + urlImageKey("https://x/y.img") + ".qcow2.sha256"},
+		// Two entries pointing one key at different images must not agree
+		// on a file name, and neither may collide with the kvm provider's
+		// base-<key>.<format> in the same pool.
+		{name: "another entry, same key", got: baseVolumeName("kvm2-lab", "debian-13", provider.KVMFormatQcow2), want: "base-kvm2-lab-debian-13.qcow2"},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			if tc.got != tc.want {
+				t.Errorf("got %q, want %q", tc.got, tc.want)
+			}
+		})
+	}
+
+	t.Run("url key", func(t *testing.T) {
+		t.Parallel()
+		hex16 := regexp.MustCompile(`^[0-9a-f]{16}$`)
+		a := urlImageKey("https://cloud.debian.org/images/cloud/trixie/latest/debian-13-generic-amd64.qcow2")
+		b := urlImageKey("https://cloud.debian.org/images/cloud/trixie/latest/debian-13-generic-amd64.qcow2")
+		c := urlImageKey("https://cloud.debian.org/images/cloud/bookworm/latest/debian-12-generic-amd64.qcow2")
+		if !hex16.MatchString(a) {
+			t.Errorf("urlImageKey = %q, want 16 lowercase hex chars", a)
+		}
+		if a != b {
+			t.Errorf("urlImageKey not deterministic: %q vs %q", a, b)
+		}
+		if a == c {
+			t.Errorf("urlImageKey gives %q for two different URLs", a)
+		}
+	})
+}
@@ -0,0 +1,256 @@
+package kvm2
+
+import (
+	"encoding/xml"
+	"time"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// domainParams is everything buildDomainXML needs, already resolved: the
+// paths are what StorageVolGetPath returned for the overlay and the seed.
+type domainParams struct {
+	Name       string
+	UID        string
+	CreatedAt  time.Time
+	VCPUs      int32
+	MemoryMiB  int64
+	Firmware   string
+	Bridge     string
+	DiskPath   string
+	SeedPath   string
+	DiskFormat string
+}
+
+const (
+	domainArch            = "x86_64"
+	domainMachine         = "q35"
+	guestAgentChannelName = "org.qemu.guest_agent.0"
+)
+
+type domainXML struct {
+	XMLName    xml.Name          `xml:"domain"`
+	Type       string            `xml:"type,attr"`
+	Name       string            `xml:"name"`
+	Memory     domainMemoryXML   `xml:"memory"`
+	VCPU       int32             `xml:"vcpu"`
+	Metadata   domainMetadataXML `xml:"metadata"`
+	OS         domainOSXML       `xml:"os"`
+	Features   domainFeaturesXML `xml:"features"`
+	CPU        domainCPUXML      `xml:"cpu"`
+	OnPoweroff string            `xml:"on_poweroff"`
+	OnReboot   string            `xml:"on_reboot"`
+	OnCrash    string            `xml:"on_crash"`
+	Devices    domainDevicesXML  `xml:"devices"`
+}
+
+type domainMemoryXML struct {
+	Unit  string `xml:"unit,attr"`
+	Value int64  `xml:",chardata"`
+}
+
+type domainMetadataXML struct {
+	Instance instanceMetadata `xml:"instance"`
+}
+
+type domainOSXML struct {
+	// Firmware is the attribute firmware="efi"; FirmwareFeatures is the
+	// <firmware> child that turns secure boot off. Both are absent for BIOS.
+	Firmware         string             `xml:"firmware,attr,omitempty"`
+	Type             domainOSTypeXML    `xml:"type"`
+	FirmwareFeatures *domainFirmwareXML `xml:"firmware"`
+}
+
+type domainOSTypeXML struct {
+	Arch    string `xml:"arch,attr"`
+	Machine string `xml:"machine,attr"`
+	Value   string `xml:",chardata"`
+}
+
+type domainFirmwareXML struct {
+	Features []domainFirmwareFeatureXML `xml:"feature"`
+}
+
+type domainFirmwareFeatureXML struct {
+	Enabled string `xml:"enabled,attr"`
+	Name    string `xml:"name,attr"`
+}
+
+type domainFeaturesXML struct {
+	ACPI *struct{} `xml:"acpi"`
+	APIC *struct{} `xml:"apic"`
+}
+
+type domainCPUXML struct {
+	Mode string `xml:"mode,attr"`
+}
+
+type domainDevicesXML struct {
+	Disks      []domainDiskXML      `xml:"disk"`
+	Interfaces []domainInterfaceXML `xml:"interface"`
+	Serials    []domainSerialXML    `xml:"serial"`
+	Consoles   []domainConsoleXML   `xml:"console"`
+	Channels   []domainChannelXML   `xml:"channel"`
+	RNGs       []domainRNGXML       `xml:"rng"`
+}
+
+type domainDiskXML struct {
+	Type     string              `xml:"type,attr"`
+	Device   string              `xml:"device,attr"`
+	Driver   domainDiskDriverXML `xml:"driver"`
+	Source   domainDiskSourceXML `xml:"source"`
+	Target   domainDiskTargetXML `xml:"target"`
+	ReadOnly *struct{}           `xml:"readonly"`
+}
+
+type domainDiskDriverXML struct {
+	Name string `xml:"name,attr"`
+	Type string `xml:"type,attr"`
+}
+
+type domainDiskSourceXML struct {
+	File string `xml:"file,attr"`
+}
+
+type domainDiskTargetXML struct {
+	Dev string `xml:"dev,attr"`
+	Bus string `xml:"bus,attr"`
+}
+
+type domainInterfaceXML struct {
+	Type   string                   `xml:"type,attr"`
+	Source domainInterfaceSourceXML `xml:"source"`
+	Model  domainModelXML           `xml:"model"`
+}
+
+type domainInterfaceSourceXML struct {
+	Bridge string `xml:"bridge,attr"`
+}
+
+type domainModelXML struct {
+	Type string `xml:"type,attr"`
+}
+
+type domainSerialXML struct {
+	Type   string        `xml:"type,attr"`
+	Target domainPortXML `xml:"target"`
+}
+
+type domainPortXML struct {
+	Port int `xml:"port,attr"`
+}
+
+type domainConsoleXML struct {
+	Type   string                 `xml:"type,attr"`
+	Target domainConsoleTargetXML `xml:"target"`
+}
+
+type domainConsoleTargetXML struct {
+	Type string `xml:"type,attr"`
+	Port int    `xml:"port,attr"`
+}
+
+type domainChannelXML struct {
+	Type   string                 `xml:"type,attr"`
+	Target domainChannelTargetXML `xml:"target"`
+}
+
+type domainChannelTargetXML struct {
+	Type string `xml:"type,attr"`
+	Name string `xml:"name,attr"`
+}
+
+type domainRNGXML struct {
+	Model   string              `xml:"model,attr"`
+	Backend domainRNGBackendXML `xml:"backend"`
+}
+
+type domainRNGBackendXML struct {
+	Model string `xml:"model,attr"`
+	Value string `xml:",chardata"`
+}
+
+// buildDomainXML renders the XML for DomainDefineXML. Both disks are
+// type="file" with <source file>: AppArmor derives QEMU's profile from those
+// paths, and type="volume" fails with permission denied on stock
+// Debian/Ubuntu hosts. UEFI boots with secure boot off; libvirt picks the
+// OVMF binary itself.
+func buildDomainXML(p domainParams) (string, error) {
+	if p.Name == "" {
+		return "", permanentf("domain XML: name is empty")
+	}
+	if p.DiskPath == "" {
+		return "", permanentf("domain XML for %q: disk path is empty", p.Name)
+	}
+	if p.SeedPath == "" {
+		return "", permanentf("domain XML for %q: seed path is empty", p.Name)
+	}
+	if p.DiskFormat == "" {
+		p.DiskFormat = provider.KVMFormatQcow2
+	}
+	if p.DiskFormat != provider.KVMFormatQcow2 && p.DiskFormat != provider.KVMFormatRaw {
+		return "", permanentf("domain XML for %q: unknown disk format %q", p.Name, p.DiskFormat)
+	}
+	osx := domainOSXML{
+		Type: domainOSTypeXML{Arch: domainArch, Machine: domainMachine, Value: "hvm"},
+	}
+	switch p.Firmware {
+	case provider.KVMFirmwareUEFI:
+		osx.Firmware = "efi"
+		osx.FirmwareFeatures = &domainFirmwareXML{
+			Features: []domainFirmwareFeatureXML{{Enabled: "no", Name: "secure-boot"}},
+		}
+	case provider.KVMFirmwareBIOS:
+	default:
+		return "", permanentf("domain XML for %q: unknown firmware %q (want %s or %s)", p.Name, p.Firmware, provider.KVMFirmwareUEFI, provider.KVMFirmwareBIOS)
+	}
+	d := domainXML{
+		Type:       "kvm",
+		Name:       p.Name,
+		Memory:     domainMemoryXML{Unit: "MiB", Value: p.MemoryMiB},
+		VCPU:       p.VCPUs,
+		Metadata:   domainMetadataXML{Instance: newInstanceMetadata(p.UID, p.CreatedAt)},
+		OS:         osx,
+		Features:   domainFeaturesXML{ACPI: &struct{}{}, APIC: &struct{}{}},
+		CPU:        domainCPUXML{Mode: "host-passthrough"},
+		OnPoweroff: "destroy",
+		OnReboot:   "restart",
+		OnCrash:    "destroy",
+		Devices: domainDevicesXML{
+			Disks: []domainDiskXML{
+				{
+					Type:   "file",
+					Device: "disk",
+					Driver: domainDiskDriverXML{Name: "qemu", Type: p.DiskFormat},
+					Source: domainDiskSourceXML{File: p.DiskPath},
+					Target: domainDiskTargetXML{Dev: "vda", Bus: "virtio"},
+				},
+				{
+					Type:     "file",
+					Device:   "cdrom",
+					Driver:   domainDiskDriverXML{Name: "qemu", Type: provider.KVMFormatRaw},
+					Source:   domainDiskSourceXML{File: p.SeedPath},
+					Target:   domainDiskTargetXML{Dev: "sda", Bus: "sata"},
+					ReadOnly: &struct{}{},
+				},
+			},
+			Interfaces: []domainInterfaceXML{{
+				Type:   "bridge",
+				Source: domainInterfaceSourceXML{Bridge: p.Bridge},
+				Model:  domainModelXML{Type: "virtio"},
+			}},
+			Serials:  []domainSerialXML{{Type: "pty", Target: domainPortXML{Port: 0}}},
+			Consoles: []domainConsoleXML{{Type: "pty", Target: domainConsoleTargetXML{Type: "serial", Port: 0}}},
+			Channels: []domainChannelXML{{
+				Type:   "unix",
+				Target: domainChannelTargetXML{Type: "virtio", Name: guestAgentChannelName},
+			}},
+			RNGs: []domainRNGXML{{Model: "virtio", Backend: domainRNGBackendXML{Model: "random", Value: "/dev/urandom"}}},
+		},
+	}
+	out, err := xml.MarshalIndent(d, "", "  ")
+	if err != nil {
+		return "", permanentf("domain XML for %q: %v", p.Name, err)
+	}
+	return string(out), nil
+}
@@ -0,0 +1,309 @@
+package kvm2
+
+import (
+	"encoding/xml"
+	"errors"
+	"strings"
+	"testing"
+	"time"
+
+	"gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/internal/provider"
+)
+
+// checkedDomain is an independent view of the marshalled XML, so the tests
+// assert on what libvirt would see rather than on the builder's own structs.
+type checkedDomain struct {
+	Type   string `xml:"type,attr"`
+	Name   string `xml:"name"`
+	Memory struct {
+		Unit  string `xml:"unit,attr"`
+		Value int64  `xml:",chardata"`
+	} `xml:"memory"`
+	VCPU     int32 `xml:"vcpu"`
+	Metadata struct {
+		Inner string `xml:",innerxml"`
+	} `xml:"metadata"`
+	OS struct {
+		Firmware string `xml:"firmware,attr"`
+		Type     struct {
+			Arch    string `xml:"arch,attr"`
+			Machine string `xml:"machine,attr"`
+			Value   string `xml:",chardata"`
+		} `xml:"type"`
+		FirmwareElems []struct {
+			Features []struct {
+				Enabled string `xml:"enabled,attr"`
+				Name    string `xml:"name,attr"`
+			} `xml:"feature"`
+		} `xml:"firmware"`
+	} `xml:"os"`
+	Features struct {
+		ACPI []struct{} `xml:"acpi"`
+		APIC []struct{} `xml:"apic"`
+	} `xml:"features"`
+	CPU struct {
+		Mode string `xml:"mode,attr"`
+	} `xml:"cpu"`
+	OnPoweroff string `xml:"on_poweroff"`
+	OnReboot   string `xml:"on_reboot"`
+	OnCrash    string `xml:"on_crash"`
+	Devices    struct {
+		Disks []struct {
+			Type   string `xml:"type,attr"`
+			Device string `xml:"device,attr"`
+			Driver struct {
+				Name string `xml:"name,attr"`
+				Type string `xml:"type,attr"`
+			} `xml:"driver"`
+			Source struct {
+				File string `xml:"file,attr"`
+				Pool string `xml:"pool,attr"`
+				Vol  string `xml:"volume,attr"`
+			} `xml:"source"`
+			Target struct {
+				Dev string `xml:"dev,attr"`
+				Bus string `xml:"bus,attr"`
+			} `xml:"target"`
+			ReadOnly []struct{} `xml:"readonly"`
+		} `xml:"disk"`
+		Interfaces []struct {
+			Type   string `xml:"type,attr"`
+			Source struct {
+				Bridge string `xml:"bridge,attr"`
+			} `xml:"source"`
+			Model struct {
+				Type string `xml:"type,attr"`
+			} `xml:"model"`
+		} `xml:"interface"`
+		Serials []struct {
+			Type string `xml:"type,attr"`
+		} `xml:"serial"`
+		Consoles []struct {
+			Type   string `xml:"type,attr"`
+			Target struct {
+				Type string `xml:"type,attr"`
+			} `xml:"target"`
+		} `xml:"console"`
+		Channels []struct {
+			Type   string `xml:"type,attr"`
+			Target struct {
+				Type string `xml:"type,attr"`
+				Name string `xml:"name,attr"`
+			} `xml:"target"`
+		} `xml:"channel"`
+		RNGs []struct {
+			Model   string `xml:"model,attr"`
+			Backend struct {
+				Model string `xml:"model,attr"`
+				Value string `xml:",chardata"`
+			} `xml:"backend"`
+		} `xml:"rng"`
+		Graphics []struct{} `xml:"graphics"`
+	} `xml:"devices"`
+}
+
+var testCreatedAt = time.Date(2026, 9, 16, 12, 0, 0, 0, time.UTC)
+
+func testDomainParams() domainParams {
+	return domainParams{
+		Name:       "proxy-abc",
+		UID:        "0b1f9a2e-uid",
+		CreatedAt:  testCreatedAt,
+		VCPUs:      2,
+		MemoryMiB:  2048,
+		Firmware:   provider.KVMFirmwareUEFI,
+		Bridge:     "br0",
+		DiskPath:   "/var/lib/libvirt/images/proxy-abc.qcow2",
+		SeedPath:   "/var/lib/libvirt/images/proxy-abc-seed.iso",
+		DiskFormat: provider.KVMFormatQcow2,
+	}
+}
+
+func mustBuildDomain(t *testing.T, p domainParams) (string, checkedDomain) {
+	t.Helper()
+	out, err := buildDomainXML(p)
+	if err != nil {
+		t.Fatalf("buildDomainXML: %v", err)
+	}
+	var d checkedDomain
+	if err := xml.Unmarshal([]byte(out), &d); err != nil {
+		t.Fatalf("unmarshalling produced XML: %v\n%s", err, out)
+	}
+	return out, d
+}
+
+func TestDomainXML_fileDisksWithResolvedPaths(t *testing.T) {
+	t.Parallel()
+	p := testDomainParams()
+	_, d := mustBuildDomain(t, p)
+	if d.Type != "kvm" || d.Name != p.Name {
+		t.Errorf("domain type=%q name=%q, want kvm/%s", d.Type, d.Name, p.Name)
+	}
+	if d.Memory.Unit != "MiB" || d.Memory.Value != p.MemoryMiB || d.VCPU != p.VCPUs {
+		t.Errorf("memory=%d %s vcpu=%d, want %d MiB / %d", d.Memory.Value, d.Memory.Unit, d.VCPU, p.MemoryMiB, p.VCPUs)
+	}
+	if len(d.Devices.Disks) != 2 {
+		t.Fatalf("got %d disks, want 2", len(d.Devices.Disks))
+	}
+	overlay, seed := d.Devices.Disks[0], d.Devices.Disks[1]
+	if overlay.Type != "file" || overlay.Device != "disk" {
+		t.Errorf("overlay type=%q device=%q, want file/disk", overlay.Type, overlay.Device)
+	}
+	if overlay.Source.File != p.DiskPath || overlay.Source.Pool != "" || overlay.Source.Vol != "" {
+		t.Errorf("overlay source = %+v, want file=%q only", overlay.Source, p.DiskPath)
+	}
+	if overlay.Driver.Name != "qemu" || overlay.Driver.Type != "qcow2" {
+		t.Errorf("overlay driver = %+v, want qemu/qcow2", overlay.Driver)
+	}
+	if overlay.Target.Dev != "vda" || overlay.Target.Bus != "virtio" {
+		t.Errorf("overlay target = %+v, want vda/virtio", overlay.Target)
+	}
+	if len(overlay.ReadOnly) != 0 {
+		t.Error("overlay is readonly")
+	}
+	if seed.Type != "file" || seed.Device != "cdrom" {
+		t.Errorf("seed type=%q device=%q, want file/cdrom", seed.Type, seed.Device)
+	}
+	if seed.Source.File != p.SeedPath || seed.Source.Pool != "" || seed.Source.Vol != "" {
+		t.Errorf("seed source = %+v, want file=%q only", seed.Source, p.SeedPath)
+	}
+	if seed.Driver.Name != "qemu" || seed.Driver.Type != "raw" {
+		t.Errorf("seed driver = %+v, want qemu/raw", seed.Driver)
+	}
+	if seed.Target.Dev != "sda" || seed.Target.Bus != "sata" {
+		t.Errorf("seed target = %+v, want sda/sata", seed.Target)
+	}
+	if len(seed.ReadOnly) != 1 {
+		t.Error("seed is not readonly")
+	}
+	if d.OnPoweroff != "destroy" || d.OnReboot != "restart" || d.OnCrash != "destroy" {
+		t.Errorf("lifecycle actions = %q/%q/%q, want destroy/restart/destroy", d.OnPoweroff, d.OnReboot, d.OnCrash)
+	}
+}
+
+func TestDomainXML_uefiDefaultSecureBootOff(t *testing.T) {
+	t.Parallel()
+	_, d := mustBuildDomain(t, testDomainParams())
+	if d.OS.Firmware != "efi" {
+		t.Errorf(`<os firmware> = %q, want "efi"`, d.OS.Firmware)
+	}
+	if len(d.OS.FirmwareElems) != 1 {
+		t.Fatalf("got %d <firmware> elements, want 1", len(d.OS.FirmwareElems))
+	}
+	feats := d.OS.FirmwareElems[0].Features
+	if len(feats) != 1 || feats[0].Name != "secure-boot" || feats[0].Enabled != "no" {
+		t.Errorf("firmware features = %+v, want one secure-boot enabled=no", feats)
+	}
+	if d.OS.Type.Value != "hvm" {
+		t.Errorf("os type = %q, want hvm", d.OS.Type.Value)
+	}
+}
+
+func TestDomainXML_biosEscape(t *testing.T) {
+	t.Parallel()
+	p := testDomainParams()
+	p.Firmware = provider.KVMFirmwareBIOS
+	out, d := mustBuildDomain(t, p)
+	if d.OS.Firmware != "" {
+		t.Errorf(`<os firmware> = %q, want absent`, d.OS.Firmware)
+	}
+	if len(d.OS.FirmwareElems) != 0 {
+		t.Errorf("got %d <firmware> elements, want none", len(d.OS.FirmwareElems))
+	}
+	if strings.Contains(out, "firmware") {
+		t.Errorf("BIOS domain XML still mentions firmware:\n%s", out)
+	}
+	if d.OS.Type.Value != "hvm" || d.OS.Type.Machine != "q35" {
+		t.Errorf("os type = %+v, want hvm on q35", d.OS.Type)
+	}
+}
+
+func TestDomainXML_metadataElement(t *testing.T) {
+	t.Parallel()
+	p := testDomainParams()
+	out, d := mustBuildDomain(t, p)
+	if !strings.Contains(out, `<instance xmlns="`+metadataURI+`">`) {
+		t.Errorf("domain XML lacks the namespaced instance element:\n%s", out)
+	}
+	m, err := parseInstanceMetadata(d.Metadata.Inner)
+	if err != nil {
+		t.Fatalf("parseInstanceMetadata(%q): %v", d.Metadata.Inner, err)
+	}
+	if !m.managed() {
+		t.Error("managed() = false, want true")
+	}
+	if m.UID != p.UID {
+		t.Errorf("UID = %q, want %q", m.UID, p.UID)
+	}
+	if m.CreatedAt != "2026-09-16T12:00:00Z" {
+		t.Errorf("CreatedAt = %q, want RFC3339 UTC", m.CreatedAt)
+	}
+	if !m.createdAt().Equal(p.CreatedAt) {
+		t.Errorf("createdAt() = %v, want %v", m.createdAt(), p.CreatedAt)
+	}
+}
+
+func TestDomainXML_guestAgentChannel(t *testing.T) {
+	t.Parallel()
+	p := testDomainParams()
+	_, d := mustBuildDomain(t, p)
+	if len(d.Devices.Channels) != 1 || d.Devices.Channels[0].Type != "unix" ||
+		d.Devices.Channels[0].Target.Type != "virtio" || d.Devices.Channels[0].Target.Name != "org.qemu.guest_agent.0" {
+		t.Errorf("channels = %+v, want one unix/virtio org.qemu.guest_agent.0", d.Devices.Channels)
+	}
+	if len(d.Devices.Interfaces) != 1 || d.Devices.Interfaces[0].Type != "bridge" ||
+		d.Devices.Interfaces[0].Source.Bridge != p.Bridge || d.Devices.Interfaces[0].Model.Type != "virtio" {
+		t.Errorf("interfaces = %+v, want one bridge %s with virtio model", d.Devices.Interfaces, p.Bridge)
+	}
+	if len(d.Devices.Serials) != 1 || d.Devices.Serials[0].Type != "pty" {
+		t.Errorf("serials = %+v, want one pty", d.Devices.Serials)
+	}
+	if len(d.Devices.Consoles) != 1 || d.Devices.Consoles[0].Type != "pty" || d.Devices.Consoles[0].Target.Type != "serial" {
+		t.Errorf("consoles = %+v, want one pty/serial", d.Devices.Consoles)
+	}
+	if len(d.Devices.RNGs) != 1 || d.Devices.RNGs[0].Model != "virtio" ||
+		d.Devices.RNGs[0].Backend.Model != "random" || d.Devices.RNGs[0].Backend.Value != "/dev/urandom" {
+		t.Errorf("rngs = %+v, want one virtio random /dev/urandom", d.Devices.RNGs)
+	}
+	if len(d.Devices.Graphics) != 0 {
+		t.Error("domain has <graphics>, want none")
+	}
+	if d.OS.Type.Arch != "x86_64" || d.OS.Type.Machine != "q35" {
+		t.Errorf("os type = %+v, want x86_64/q35", d.OS.Type)
+	}
+	if d.CPU.Mode != "host-passthrough" {
+		t.Errorf("cpu mode = %q, want host-passthrough", d.CPU.Mode)
+	}
+	if len(d.Features.ACPI) != 1 || len(d.Features.APIC) != 1 {
+		t.Errorf("features acpi=%d apic=%d, want one each", len(d.Features.ACPI), len(d.Features.APIC))
+	}
+}
+
+func TestDomainXML_rejectsBadInput(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		name   string
+		mutate func(*domainParams)
+	}{
+		{name: "unknown firmware", mutate: func(p *domainParams) { p.Firmware = "coreboot" }},
+		{name: "empty firmware", mutate: func(p *domainParams) { p.Firmware = "" }},
+		{name: "empty name", mutate: func(p *domainParams) { p.Name = "" }},
+		{name: "empty disk path", mutate: func(p *domainParams) { p.DiskPath = "" }},
+		{name: "empty seed path", mutate: func(p *domainParams) { p.SeedPath = "" }},
+		{name: "unknown disk format", mutate: func(p *domainParams) { p.DiskFormat = "vmdk" }},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			p := testDomainParams()
+			tc.mutate(&p)
+			out, err := buildDomainXML(p)
+			if !errors.Is(err, provider.ErrPermanent) {
+				t.Errorf("err = %v, want ErrPermanent", err)
+			}
+			if out != "" {
+				t.Errorf("got XML %q alongside the error", out)
+			}
+		})
+	}
+}
@@ -0,0 +1,183 @@
+# kvm2 provider review: final report
+
+Branch `feat/kvm2-provider` at `2510842`. Scope in `scope.md`; phase 1 detail in `sweep.md`;
+phase 2 detail, measurements and unrun probes in `deep.md`.
+
+**What this contains.** Confirmed findings only: the sweep findings (each reported by its
+sweeper as verified in code) plus the escalations that came back CONFIRMED. Where a deep dive
+refined a sweep finding, the refined version is given. **52 findings: 4 high, 20 medium, 28 low.**
+
+**How far to trust it.** Nothing here was fixed or re-verified by a second reader except where
+two or three reviewers reported the same thing independently (marked ×2/×3). Hypervisor access
+was read-only, so every claim about what libvirt does on a *write* is from libvirt source or
+memory, with the reviewer's stated confidence; the scripts that would settle them are in
+`deep.md`. The kvm2 integration test has **not yet run on beelink** (Step 11 is unchecked in
+the plan execution file; every `base-*` volume there predates kvm2), so no kvm2 behaviour in
+this report has been observed against real libvirt.
+
+Themes: **T1** unknown errors default to Transient and retry forever with no condition on the
+Proxy · **T2** the ownership mark is enforced on Create/ListByTag but not Get/Delete ·
+**T3** base image cache identity, lifetime and fencing · **T4** any caller can kill the shared
+pooled link, and several callers misread the resulting error.
+
+---
+
+## High
+
+### H1 — `markDead` blocks on a silent host and pins every reconcile worker (S1.1 + E1)
+`internal/provider/kvm2/backend.go:138-148`, `ssh.go:109-114`
+`markDead` runs `kill()` inside `once.Do` and only then closes `l.closed`. `kill` is `sshSocket.Close`, which closes the SSH channel first; that needs `ch.writeMu` and `t.mu`, both held by a writer stuck in a TCP write. Reproduced against x/crypto v0.54.0: with a 256 K send buffer the close was still blocked after 8 s, and a plain `client.Close()` released it instantly. Precondition: a bulk write in flight and a send buffer under the 2 MiB SSH window.
+**Scenario.** Three Proxies placed on one host with an uncached image. Worker 1 uploads (`image.go:270`, 20 min budget); workers 2 and 3 park on the image lock (`kvm2.go:224-233`) with no deadline; `MaxConcurrentReconciles` is 3. Host is power-cut. The keepalive's `markDead` (`hosts.go:158`) blocks in the channel close; worker 1's `call` cannot see `done`, `l.closed`, or even a ctx cancel (it queues in `once.Do`). For ~15 min no Proxy on any provider reconciles. No `ReconciliationTimeout` is set, and one alone would not release worker 1.
+**Fix.** Close the TCP connection first in `sshSocket.Close` (`s.client.Close()` before, or instead of, `s.Conn.Close()`); in `markDead`, `close(l.closed)` before `kill()` or run `kill` in a goroutine. Regression test with a frozen proxy and small `SO_SNDBUF`. Optionally set `ReconciliationTimeout`.
+
+### H2 — An error during `StorageVolUpload` wedges go-libvirt's stream sender forever (S4.1)
+`internal/provider/kvm2/image.go:273` (upstream `rpc.go:290-308`, `socket/socket.go:347-376`)
+`abort` and `outErr` are unbuffered. On a reader error `SendStream` blocks forever on `outErr <-`; on a server-side error `requestStream` blocks forever on `abort <- true`. A socket kill does not unblock them (E6 found why: `listen` blocks in `callback`'s send under `cmux`), contrary to the comment at `backend.go:152-156`.
+**Scenario.** A mirror truncates the body, or libvirtd hits ENOSPC mid-upload. The caller is rescued only after the 20 min transfer budget; each failed upload leaks 1-2 goroutines plus a 4 MiB buffer, accumulating under reconcile retries; `discardVolume` (`image.go:292`) deletes the base before the deferred `fresh.Close()` (`:265`), so the file is unlinked while the server stream holds the fd. `TestEnsureImage_shortReadRemovesVolumeAndIsTransient` cannot catch it: the fake returns the reader error immediately.
+**Fix.** Have `countingReader` convert a read error to `io.EOF` and report it out of band; the existing `readErr` / `read != size` checks (`image.go:281-290`) still catch truncation. Close the dedicated connection before `discardVolume`. Consider an upstream patch for the unbuffered channels.
+
+### H3 — `address()` swallows every error, demoting healthy Running proxies (S3.1)
+`internal/provider/kvm2/get.go:104-119`
+Every non-`context` error from both address sources becomes `("", nil)`, so `mapState(running, "")` = Provisioning.
+**Scenario.** (a) On beelink the ARP source never answers (`kvm2-provider-spec.md:146`), so a `qemu-guest-agent` restart in the guest (`ErrAgentUnresponsive`) empties the address. (b) Another reconcile's cancel kills the shared pooled link between `DomainGetState` and the address read; the "connection is closed" error carries no ctx sentinel, so `Get` *succeeds* with `IP=""`. Then `proxy_controller.go:208-211` clears `status.IP` and sets `Provisioned=False`, and `internal/health/engine.go:200-207` prunes the proxy's health state. One RPC hiccup pulls a working proxy out of rotation. `TestGet_addressErrorIsProvisioning` pins this.
+**Fix.** Return Transient from `Get` when the address read fails with a connection-class error (`isDead(be)` or the closed-link error) and return NotFound on `ErrNoDomain`; keep "no address yet" only for agent-unavailable/ARP-empty on a domain that has never reported an address. Consider having the controller keep the last known IP while an instance is Running.
+
+### H4 — The base image cache is shared and unfenced across provider entries (S7.1 ×2, E3-3a/3b, E4)
+`internal/provider/kvm2/volume.go:17-25`, `image.go:113-146`, `:235-245`, `kvm2.go:111`; `internal/provider/kvm/image.go:44-57`; `config/manager/providers_config.yaml`, `config/samples/providers-config.yaml:41,73`
+The base is `base-<key>.<format>` with a `.sha256` marker, byte-identical in the kvm provider, in the same default pool, under **separate, per-entry locks**. The shipped config puts `kvm-home` and `kvm2-home` on `beelink`, key `debian-13`, pool `default`, and the docs say both types may share a host. On the host both live (kvm-made) overlays back onto that very file. Neither the name nor the marker records the URL.
+**Scenarios.**
+1. *Wrong image, silently.* The two entries' URLs drift and neither sets `sha256`: whoever fetched first owns the file. Likewise, editing `images.<key>.url` alone is a silent no-op.
+2. *Digest mismatch (3a).* One entry pins `sha256: D2` while the cached base is D1: `cachedImage` (`image.go:129` → `:145`) deletes the base under the other provider's live overlays. libvirt refuses a delete only on `in_use`/`building` (`storage_driver.c:1844-1856`, read at v11.6.0). With two different pins, every Create on either side deletes and refetches.
+3. *Concurrent cold fetch (3b, by source, unexecuted).* B finds A's base-without-marker and deletes it (`image.go:117-118`), recreates it and uploads. libvirt holds `in_use` only during stream setup (`storage_driver.c:2421-2427`). A's upload completes into the unlinked inode, A writes the marker, and `StorageVolGetPath` resolves to B's partial file: **A boots a guest from a half-written base.** If B is kvm2, its `writeMarker` then fails `ErrStorageVolExist` and `image.go:241-243` discards the complete base under A's running domain. If the process dies during B's upload, a truncated base stays under a valid marker forever, since nothing re-hashes the base.
+The rollout/leader-handover variant of this was **refuted** for the shipped manifests (lease held until the old pod dies; SIGKILL at 10 s).
+**Fix.** kvm2-specific prefix including the entry name, or one process-wide image lock table; reject two kvm/kvm2 entries with the same address+socket+pool at config load. Upload to `<base>.part-<random>`, verify, then publish the final name; never delete a name this call did not create, in particular drop `discardVolume(base)` on a marker `ErrStorageVolExist`. Record size and URL next to the digest and check the base's allocation before trusting a marker.
+
+---
+
+## Medium
+
+### M1 — Paused domains are destroyed (S2.2 ×2)
+`create.go:230-231`, `get.go:56`, `state.go:16-27`. `get.go:56` discards the `DomainGetState` reason; paused/pmsuspended map to Stopped; `reconcileManaged` (`proxy_controller.go:217-225`) then deletes and recreates. The "virsh resume it" guard in `ensureRunning` only fires before an ID is persisted. **Scenario:** the pool fills and QEMU pauses the guest with `PAUSED_IOERROR` (or `virsh suspend`/save/snapshot/migrate); at the next drift poll the VM and its disks are gone and the recreate refills the pool. **Fix:** keep the reason; map paused to a non-destructive state (Provisioning or a new Degraded) except for reasons that are terminal.
+
+### M2 — `Get` and `Delete` do not enforce the kvm2 mark (S2.3 ×2, E3-2) [T2]
+`get.go:52-73`, `delete.go:55-98` vs `create.go:80-82`. IDs are byte-identical to the kvm provider's (`kvm/id.go:11-13`). **Scenario:** an entry is re-typed `kvm` → `kvm2` under the same name without first deleting its Proxies (the documented migration, out of order): `status.ProviderID` survives, `Get` returns the v1 domain with `UID: ""`, and anything but Running leads to kvm2 destroying a domain the docs say it can never touch. A tampered `status.providerID` such as `hosts/beelink/domains/docker-29` destroys an unrelated VM. Nothing prevents a re-type (no webhook, `config.go:93-135`). The reverse direction (kvm provider against kvm2 domains) has no guard at all (`kvm/kvm.go:177-181`). **Fix:** `removeDomain` and `get` require `meta.managed()`; pass the expected UID to Delete or compare `inst.UID` with `p.UID` in the controller.
+
+### M3 — Volumes created before `DomainDefineXML` are unreachable when Create returns no ID (S2.1)
+`create.go:107-114`. **Scenario:** overlay and seed succeed, define fails Permanent (or the manager dies). No `ProviderID` is set; on Proxy deletion `reconcileDelete` takes the empty-ID shortcut (`proxy_controller.go:270-274`); `ListByTag` lists domains only. `proxy-<x>.qcow2` and `-seed.iso` leak permanently. Five such leftovers already sit in beelink's `default` pool (Appendix C). **Fix:** on a failed define, best-effort remove the two volumes before returning; or have `Delete`/GC derive volume names from the Proxy UID.
+
+### M4 — The pool-space precheck reads a cached, reserve-inflated number (E8)
+`create.go:92-100`, `image.go:450-458`, `check.go:106-112`. Measured on beelink: libvirt's `available` is cached per pool (three pools on one filesystem report two values, 1.68 GB apart; frozen over 80 s while `f_bfree` moved) and is `f_bfree`, **24.7 GB above** what unprivileged QEMU can use on the kvm2 filesystem. The only refresh is after an image upload. **Scenario:** with `f_bavail` between 0 and ~10.7 GB the precheck passes for a 10 GiB disk, `ErrQuotaExceeded` never fires, and Create, the upload and the guests hit ENOSPC (feeding H2 and M1). `poolAvailableGiB` in ProviderHealth over-reports the same way. **Fix:** `StoragePoolRefresh` right before `checkPoolSpace` (not in Check); add headroom to `neededBytes` from the discarded `capacity` or a `poolReserveGiB` setting; make the fake's pool info cached-until-refresh.
+
+### M5 — An overlay smaller than the base's virtual size is accepted; the Proxy hangs in Provisioning (S5.6 + E10)
+`volume.go:72`, `create.go:107,158`, `internal/provider/config_kvm.go:133`. Measured locally: `qemu-img create` with a capacity below the backing file exits 0; reads past the capacity fail; a GPT base under a short overlay shows only the protective MBR. Bases on the host are 3 GiB virtual. **Scenario:** `diskSizeGb: 2` validates; Create succeeds; the guest has no root device and never reports an address; `Get` says Provisioning forever (no provisioning deadline exists), nothing names `diskSizeGb`, and raising it later does not repair that Proxy. libvirt's pass-through is ~90% from source; the 200 KiB probe is in `deep.md`. **Fix:** add `StorageVolGetInfo` to the backend seam and fail Permanent when the base's capacity exceeds the overlay's; same check on a reused overlay.
+
+### M6 — A peer-close right after ConnectOpen wedges a pooled link forever (E6a)
+`backend.go:177-178`, `:229-236`, `hosts.go:75`, `errors.go:55` (upstream race `libvirt.go:297/305/821`, reported by `-race`). **Scenario:** libvirtd or the forwarded socket goes away within ~1 ms of a successful open (daemon restart). `Connect()` returns nil with a `Disconnected()` channel nothing will close; every later RPC fails `EINVAL`, classified Transient; no result marks the link dead; the SSH keepalive keeps succeeding. The host answers "invalid argument" until the operator restarts. Wedged in 546/1000 cycles over real SSH in the reviewer's harness. The hypothesised trigger (kvm2's own kill) is harmless. **Fix:** in `call`, `markDead` on `syscall.EINVAL` / `libvirt.ErrInterrupted`; or wrap the `net.Conn` so a failed `Read` marks the link dead.
+
+### M7 — Permanent host/config rejections retry forever with nothing in status (S5.2 ×2, S3.5, E9b) [T1]
+`errors.go:58-71`, `xml.go:26-27,194-206`, `proxy_controller.go:344-345`. `x86_64`/`q35`/`kvm`/host-passthrough/efi are unconditional with no capability gate; code 67 (`ErrConfigUnsupported`) and the bare `libvirt.ErrUnsupported` sentinel are unmapped; `providerFailure`'s Transient arm sets no condition. **Scenario:** an aarch64 or KVM-less host is added; Check says Ready; every Create retries an impossible define forever, rerunning `writeSeed` each time; a missing bridge (38, at start) or missing OVMF (9) behaves the same and Check looks at neither. "Unknown means retry" is a recorded spec decision (`kvm2-provider-spec.md:317`), but it never considered 67. **Fix:** 67 → Permanent on create; `ErrUnsupported` → Permanent; leave 9/38/1/84. Surface a non-latching condition for Transient create failures, or check domain capabilities and the bridge in the health probe.
+
+### M8 — Seed ISO I/O failures latch as Permanent with no requeue (S5.1) [T1, inverted]
+`create.go:199-201`, `seed.go:66-74`. `permanentf("building seed ISO …: %v")` covers a missing `/tmp`, ENOSPC, and even a `RemoveAll` failure on an otherwise successful build. **Scenario:** node disk pressure for a minute leaves the Proxy unprovisioned until a spec edit or the ~10 h resync. **Fix:** Permanent only for the empty-name guard; wrap the rest Transient with `%w`.
+
+### M9 — `HEAD` 2xx without a usable Content-Length becomes a latching Permanent (S4.2) [T1]
+`image.go:434-439`, `:391-396`. The GET fallback fires only on non-2xx. **Scenario:** a CDN (or Go's own `net/http`) answers HEAD with `Content-Length: 0`; the Proxy never provisions although GET works. **Fix:** fall back to GET whenever HEAD yields no positive length.
+
+### M10 — A completed upload is discarded when the pooled link died meanwhile (S4.3) [T4]
+`image.go:235-245`. Refresh and `writeMarker` run on the pooled `be` after a transfer on a dedicated link. **Scenario:** any other reconcile's cancel during a 20 min download kills the pooled link; the verified base is deleted and refetched. **Fix:** run the refresh and marker on the still-open dedicated connection.
+
+### M11 — Precheck uses HEAD's length, the volume uses GET's, and there is no size cap (S4.4)
+`image.go:419-443`, `:226-230`. **Scenario:** HEAD says 1 MB, GET says 4 TiB: a sparse 4 TiB volume is created and the stream fills the host pool. **Fix:** re-check space against the GET length; add `maxImageBytes`.
+
+### M12 — `placement.image` is an unvalidated user-supplied URL fetched by the operator (S4.5)
+`image.go:59-65`, `internal/provider/config_kvm.go:101-104`. No CRD validation, no host allowlist, `http.DefaultClient` follows redirects, and a URL image can never carry a digest. **Scenario:** anyone who can create a Proxy makes the operator issue requests to cluster-internal addresses and boots arbitrary bytes on the egress bridge. **Fix:** a config switch to allow URL images at all (default off) or a host allowlist; `CheckRedirect` policy; block link-local/private ranges.
+
+### M13 — A `sha256` change replaces the base in place under live overlays (S4.6 ×3)
+`image.go:118,129,145`. No in-use check, `V(1)` log; a mistyped digest deletes the base before anything is verified. The kvm2 sample omits the warning the kvm sample carries. Whether guests actually corrupt is Appendix A (E7). **Fix:** on a digest mismatch return Permanent "add the image under a new key"; or put the digest in the volume name.
+
+### M14 — IP detection takes the first IPv4 from any guest interface (S3.2)
+`state.go:35-49`. `Hwaddr` is available but unused. **Scenario:** `docker0`/`tailscale0` comes up before the DHCP lease; the Proxy latches `Provisioned=True` with 172.17.0.1 and is never corrected. **Fix:** match the interface by the domain NIC's MAC.
+
+### M15 — A link that dies mid-list yields a silently partial `ListByTag` (S3.3) [T4]
+`list.go:38-45`. The guard tests `ctx.Err()`, but a killed link has no ctx sentinel. GC only under-deletes, so nothing wrong is removed, but the sweep covers a fraction of the fleet and says so nowhere. **Fix:** fail the list on any connection-class error (`isDead(be)`).
+
+### M16 — `KVMHost.Address` is not validated as `host:port` (S7.2) [T1]
+`internal/provider/config_kvm.go:180-182`. **Scenario:** `address: beelink` loads; every dial fails "missing port in address", classified Transient, and Check reports `Unreachable`. **Fix:** `net.SplitHostPort` at load, or default `:22`.
+
+### M17 — Plain `http://` images with no digest are accepted silently (S7.3)
+`internal/provider/config_kvm.go:101-104,142-147`. The root disk of every VM can come from an unauthenticated channel and the marker then pins it. **Fix:** require `sha256` for `http://` catalogue entries, or warn loudly at load.
+
+### M18 — The fake ignores the metadata namespace (S7.4) [T2]
+`fake_test.go:362-376`, `:259-282`. No unit test can build a kvm-v1-marked domain and assert `ListByTag` skips it; if that filter regressed, kvm2 would hand kvm's VMs to orphan GC with the suite green. E3/E9 observed that real libvirt 11.6 does scope by namespace. **Fix:** key the fake's metadata on the element's namespace; add the test.
+
+### M19 — The integration target's timeout can skip VM cleanup on the real host (S7.5)
+`Makefile:75`. `-timeout 40m` is fixed while two tests each take up to `KVM2_TEST_TIMEOUT` (20 m) plus 5 m cleanup. **Scenario:** a cold download or a raised env timeout trips go test's own timeout before `t.Cleanup` (`kvm2_integration_test.go:373-382`); a domain and two volumes stay on beelink under a random name. **Fix:** derive `-timeout` from the env var with margin; print the name up front; add a `review`-style sweep-by-prefix cleanup target.
+
+### M20 — `sshSocket.keepalive` has no test (S1.2)
+`ssh.go:119-147`. It is the only detector of a silently dead pooled link between calls, and `testSSHServer` already services global requests. **Fix:** an end-to-end test that stops the server's request handling and asserts `markDead`.
+
+---
+
+## Low
+
+| ID | Where | Defect → scenario | Fix sketch |
+|---|---|---|---|
+| L1 (S1.3) | `ssh.go:120` | `NewTicker` panics on interval ≤ 0 in a goroutine → a future `libvirtDialer{}` literal kills the operator on first dial | default the interval when ≤ 0 |
+| L2 (S1.4) [T1] | `errors.go:45-51` | `*knownhosts.RevokedError` unrecognised → a revoked host key retries forever | classify it Permanent |
+| L3 (S1.5) | `ssh.go:167-172` | `HostKeyAlgorithms` empty → a host pinned by RSA only but serving ed25519 reports a Permanent "mismatch" | on `KeyError` with `Want`, retry with those key types; or document `ssh-keyscan` without `-t` |
+| L4 (S2.4) | `create.go:87` | Adoption path starts a defined domain without checking its disks → missing overlay/seed retries forever (E5's EIO residual lands here) | on a storage error in the adoption branch, undefine and fall through to a full create |
+| L5 (S2.5) [T1] | `delete.go:15` | `undefineFlags` lacks `CheckpointsMetadata` → an admin/backup checkpoint wedges the finalizer | add the flag |
+| L6 (S3.4) | `get.go:104-119` | `ErrNoDomain` swallowed → a vanished domain reads as Provisioning for one poll; comment about ctx-only errors is wrong (a budget expiry never falls back to ARP) | return NotFound; fix the comment |
+| L7 (S4.7) | `image.go:62` | URL images assumed qcow2/uefi → a raw URL image fails to boot with a misleading error | sniff the qcow2 magic from the first bytes |
+| L8 (S4.8) [T3] | `volume.go:22-25` | 64-bit URL key, namespace shared with catalogue keys → colliding URLs or a 16-hex catalogue key share a base | longer key and a distinct `url-` prefix |
+| L9 (S5.3, narrowed by E2) | `metadata.go:44-50` | future-dated mark → a true orphan is never collected. (The zero-time half is benign: `gc.go:118` checks the UID first.) | clamp future timestamps to now |
+| L10 (S5.4) | `seed.go:48-53` | `rc-update` fallback unguarded under `set -e` → agent installed but not started on non-systemd, non-OpenRC images | `elif command -v rc-update`, as the kvm provider does |
+| L11 (S5.5) | `xml.go:178-193` | `VCPUs`/`MemoryMiB` not validated in the builder (latent) | reject ≤ 0 |
+| L12 (S5.7) | `internal/provider/config_kvm.go:186-188` | `bridge`/`pool` only non-empty → opaque libvirt error at Create | apply `kvmIdentifier`-style validation |
+| L13 (S5.8) | `seed.go:107-108` | multipart or headerless configured `vendorData` silently dropped with `installGuestAgent` | reject at load what cloud-init will not recognise |
+| L14 (S6.1) | `check.go:90`, `backend.go:162,351` | `Close()` can send `ConnectClose` for up to 5 s after the round deadline | `markDead` when `call` refuses on a done ctx, or skip the RPC in Close then |
+| L15 (S6.2) | `check.go:57-58` | Latency includes the teardown RPC | stop the clock before Close |
+| L16 (S6.3) | `check.go:66` | round-wide `ctx.Err()` masks a per-host reason landing on the deadline | classify the error first; Timeout only if it carries the ctx error |
+| L17 (S6.4) | `check.go:135-148`, `internal/providerhealth/monitor.go:113` | a 10 s round timeout collapses Unreachable into Timeout | validate round timeout > dial budget, or shrink the dial budget to fit |
+| L18 (S6.5) | `check.go:101` | unbounded remote `hostname` in status `details` → status patch "too large" | truncate; add `maxLength` to the CRD |
+| L19 (S6.6) | `check.go:41-45` | unbounded per-host fan-out, key re-read every dial | cap concurrency; cache credentials |
+| L20 (S7.6) [T1] | `internal/provider/config_kvm.go:133`, `image.go:466` | no upper bound on `diskSizeGb`; `neededBytes` wraps at 2^34 → precheck passes; huge `memoryMiB` retries forever | bound both at load |
+| L21 (S7.7) | `fake_test.go:190-208` | fake accepts calls on a done ctx → `discardVolume`'s `WithoutCancel` is untested | make `enter()` refuse like `call()` |
+| L22 (S7.8) | `fake_test.go:579-584` | fake `Close()` leaves the backend live → use-after-close passes in tests | mark dead on Close |
+| L23 (S7.9) | `Makefile:60-75`, `.gitea/workflows/build.yaml:27-32` | nothing compiles `-tags=integration` | add `go vet -tags=integration ./...` to the `check` job |
+| L24 (S7.10) | `image.go:228,283-289,385-395` | full image URL, including any userinfo, in logs and wrapped errors | use `urlHost`/redacted URL everywhere; reject userinfo at load |
+| L25 (S7.11) | `kvm2_integration_test.go:89,307` | `KVM2_TEST_KEEP=0` enables keep; 100 ms wall-clock assertion on `New()` | parse the bool; drop or loosen the timing check |
+| L26 (S7.12) | `config/manager/manager.yaml:163-178`, `seed.go:56-74` | cloud-init payloads staged on node-local disk; comment names kvm only | `medium: Memory`; fix the comment |
+| L27 (E6 side) | go-libvirt `socket/socket.go:233-234` | packet length < 28 panics the listen goroutine and the operator; a huge length allocates up to 4 GiB → `host.Socket` pointed at a non-libvirt socket (reproduced) | upstream bounds check; until then validate the socket in Check before pooling |
+| L28 (E6 side) | go-libvirt `rpc.go:76-84` | duplicate reply deadlocks `callback` under `cmux`, leaking 3 goroutines that survive a socket kill (measured +420/200 cycles); same mechanism keeps H2's leak alive | upstream: send outside the lock |
+
+---
+
+## Appendix A — PLAUSIBLE, with what would settle it
+
+- **E2 — a padded or case-changed `<uid>` in the GC mark makes GC delete a live Proxy's VM.** Reproduced on a verbatim copy of the parser and the `gc.go` switch; there is no second guard before `Delete`. No trigger libvirt produces on its own was found (it re-indents between children only), so it needs a hand edit or a reformatting tool. The kvm provider has the same field, and the two live VMs are kvm v1. *Settle:* the `virsh metadata --set` round trip in `deep.md` on a scratch libvirt. *The fix is three `TrimSpace` lines and worth making regardless.*
+- **E7 — in-place base replacement corrupts existing overlays.** The delete is confirmed (M13). Corruption needs a new QEMU process on an old overlay; the controller never does that, but `libvirt-guests` is enabled on beelink and restarts guests after a host reboot. *Settle:* `grep -Ev '^\s*(#|$)' /etc/default/libvirt-guests` on beelink, and the `review-e7-*` inode probe in `deep.md`.
+- **Unexecuted links inside confirmed findings.** H4-3b: libvirt allows `vol-delete` during an upload (read in 11.6.0 source; the `probe-race`/`e4probe` scripts). M5: libvirt passes a short capacity through (~90%; `review-e10-short`). H1: the real path's send buffer is under ~2.2 MB (`ss -tmi` during an upload). E9a: duplicate volume → code 90 (~95%; `review-e9-dup`).
+
+## Appendix B — REFUTED
+
+- **E5** overlay left corrupt by an abandoned `StorageVolCreateXML`: libvirtd finishes or unlinks the build regardless of the client (~85%, source memory); creation is 2-11 ms on the host; the residual path fails loudly.
+- **E3-1** GC reconciling one provider's list against another's Proxies: the live set is global and the list is namespace-filtered (observed on the host).
+- **E3-2** re-type destroying *another* Proxy's instance: the ID always names the Proxy's own UID-derived domain (the remaining hole is M2).
+- **E4 rollout/leader handover** deleting a base mid-upload: the lease is held until the old pod dies; SIGKILL at 10 s; the residual window is theoretical and self-healing.
+- **E6b** goroutine/fd accumulation from per-round check dials: +0/+0 over 1000 cycles on three transports.
+- **E9a** idempotency codes differing from the fake: 42, 49, 50, 80 observed on libvirt 11.6; 90 and 55 sourced at high confidence.
+- **E2 (half)** an unparsable `createdAt` deleting a live VM: `gc.go:118` checks the UID first.
+- **E7 (half)** marker-missing trigger without an admin edit: marker errors are returned, not treated as unreadable.
+- **No XML injection**: three sweepers independently found every value goes through `encoding/xml`.
+
+## Appendix C — Out of scope but noticed
+
+- **Orphan volumes on beelink, pool `default`:** `proxy-7hjzuf3d3w2mjnye.qcow2` + `-seed.iso`, `proxy-qxge3dbm2mnpc3ih.qcow2` + `-seed.iso`, `proxy-sbs36ruo73sxxtqc.qcow2`. 196 KiB each, root-owned, no domain. Reported independently by four deep reviewers; they are M3 in the wild (likely from kvm v1 integration runs). Nothing was deleted.
+- **The kvm2 integration test has not run on beelink** (plan execution Step 11 unchecked).
+- **The kvm provider shares most of these defects:** untrimmed mark UID (E2), no mark check on Get/Delete and adoption of any same-named domain (`kvm/kvm.go:177-181`), short-overlay hole (`kvm/volume.go:24-32`), same base names.
+- **No provisioning deadline in the controller:** any VM that boots but never reports an address stays Provisioning forever with no signal (what makes M5 silent).
+- **`providerFailure` sets no condition for Transient errors** (`proxy_controller.go:344-345`), which is what makes all of T1 invisible.
+- **Manager wiring:** no `ReconciliationTimeout`, no `LeaderElectionReleaseOnCancel`, default RollingUpdate with one replica, 128Mi memory limit (an OOM mid-upload is one way to H4's truncated base).
+- **Host key naming:** `~/.ssh/kvm_known_hosts` pins beelink under its Tailscale FQDN only, while the sample config says `address: beelink:22`; an operator using that file would fail host key verification (seen from this workstation; not verified from inside the operator).
+- **Dropped escalations not investigated:** D1 delete-during-open-stream disk-space effect, D2 cloud-init `text/plain` vendor-data dispatch, D3 ProviderHealth status churn, D4 nested `kvm-ssh` volumeMount under the read-only config mount, D5 whether error strings with image URLs reach Proxy status.
@@ -0,0 +1,265 @@
+# kvm2 provider review: phase 2 deep dives
+
+Ten `deep-reviewer` runs, one per escalation in `escalations.md`. Hypervisor access was
+**read-only** (beelink, libvirt 11.6.0, QEMU 10.1.0); every probe that needs a write was left
+unrun and is reproduced under "Probe to settle". Scratch programs live under the session
+scratchpad, not in the repo. The orchestrator did not re-read code.
+
+| ID | Verdict | One line |
+|---|---|---|
+| E1 | **CONFIRMED** (narrowed) | Blocking kill reproduced against x/crypto v0.54.0; pins all 3 reconcile workers ~15 min per incident, trigger is a bulk upload, not the name lock |
+| E2 | **PLAUSIBLE** | Padded `<uid>` text does make GC delete a live VM (reproduced on a copy of the parser); no trigger libvirt produces on its own. `createdAt` half REFUTED |
+| E3 | **mixed** | GC cross-provider: REFUTED. Re-type beyond S2.3: REFUTED. Shared `base-<key>.<format>` cache between kvm and kvm2: CONFIRMED |
+| E4 | **CONFIRMED** (narrowed) | Rollout/leader-handover trigger REFUTED; the unfenced base is reachable in one process via the kvm provider or a second kvm2 entry on the same host+pool |
+| E5 | **REFUTED** (~85%) | libvirtd finishes or unlinks a volume build regardless of client disconnect; overlay create is 2-11 ms on the host |
+| E6 | (a) **CONFIRMED**, (b) **REFUTED** | Upstream race on `l.disconnected` is real; a peer-close right after ConnectOpen wedges a pooled link forever. No goroutine/fd growth over 1000 cycles |
+| E7 | **PLAUSIBLE** | Unguarded in-place delete is real; corruption needs an admin `sha256` edit plus a host-side restart of an old overlay (`libvirt-guests` is enabled on beelink) |
+| E8 | **CONFIRMED** | `available` is cached per pool in libvirtd (measured), and it is `f_bfree`, 24.7 GB above what QEMU can use on the kvm2 filesystem |
+| E9 | (a) **REFUTED**, (b) **CONFIRMED** | Idempotency codes match libvirt 11.6 (observed 42/49/50/80). Codes 67 and the bare `ErrUnsupported` retry forever, per a recorded spec decision |
+| E10 | **CONFIRMED** (libvirt half ~90%) | qemu-img accepts a short overlay (measured locally); guest sees a truncated disk, GPT rejected; Proxy stuck Provisioning forever |
+
+Two facts the deep dives surfaced that correct earlier documents:
+- **The kvm2 integration test has not run on beelink.** Step 11 is unchecked in `docs/plans-executions/2026-09-16-1333-kvm2-provider.md`; every `base-*` volume on the host is dated 2026-09-10 (kvm v1's test, same `test-image` key); both live `proxy-*` domains carry the kvm v1 metadata namespace. The "Checked" line of E2 in `escalations.md` (round trip asserted on libvirt 11.6) was a sweeper claim and is not yet true.
+- **Sweep finding S5.3 is half wrong.** A zero `createdAt` does not endanger a live VM: `gc.go:118` tests `live[inst.UID]` before the age case, so it only affects true orphans. The future-dated-mark leak stands.
+
+---
+
+## E1 — wedged `markDead` starves the reconcile workers — CONFIRMED (narrowed)
+
+**Premise reproduced** (`scratchpad/e1/main.go`, in-process SSH server behind a freezable TCP proxy, x/crypto v0.54.0). `sshSocket.Close` (`ssh.go:109-114`) calls `s.Conn.Close()` first: `channel.Close` (`channel.go:594`) → `sendMessage` → `writePacket` takes `ch.writeMu`, then `handshakeTransport.writePacket` (`handshake.go:579`) takes `t.mu`. A data writer stuck in a TCP write holds both. `ssh.Client.Close` is a bare TCP close (`connection.go:111`) and takes no locks.
+
+| Scenario | Client send buffer | Writer blocked in | Channel `Close` |
+|---|---|---|---|
+| Bulk write, loopback default | 2.6 MB | `window.reserve` (no mutex) | 0 s |
+| Bulk write, LAN-sized buffer | 256 K | `net.(*conn).Write` under `writeCipherPacket` | still blocked after 8 s |
+| Small RPC-sized write only | 2.6 MB | not blocked | 0 s |
+
+Precondition: the writer must be stuck in the TCP write, not SSH flow control (2 MiB window). That needs a send buffer below ~2 MiB, which the reviewer estimates (not measured) is the case on a LAN/Tailscale path during a bulk upload. In the blocked case a separate `client.Close()` released the channel close at once.
+
+**Breaking path.**
+1. Three Proxies placed on one kvm2 host with an uncached image. Worker 1 takes the name lock (`create.go:53`) and image lock (`image.go:87`), then `uploadImage` → `dialFresh` (`image.go:261`) → `StorageVolUpload` (`image.go:270`, 20 min budget). Workers 2 and 3 park in `imageLocks.acquire` (`kvm2.go:224-233`); the reconcile ctx has no deadline. `MaxConcurrentReconciles` is 3 (`proxy_controller.go:442`): all workers occupied.
+2. Host is power-cut mid-upload. The upload goroutine blocks in the TCP write holding both mutexes.
+3. The dedicated link's keepalive (`hosts.go:156-160`) cannot send (needs `t.mu`); after 15 s `hosts.go:158` calls `markDead`, whose `kill()` blocks in `s.Conn.Close()`. `close(l.closed)` (`backend.go:146`) is never reached.
+4. Worker 1 sits in `call`'s select (`backend.go:176-187`): `done` cannot fire, `l.closed` is not closed, and a ctx cancel reaches `markDead` at `:180`, which blocks in `once.Do` behind the keepalive goroutine. Even cancellation is uninterruptible.
+5. After ~15.5 min (`tcp_retries2=15`) or the 20 min budget everything unwinds Transient. Meanwhile no Proxy on any provider reconciles.
+
+**Unknowns settled.** No `ReconciliationTimeout` is set (`cmd/main.go:288-295`; controller-runtime v0.24.1 applies one only when > 0), and one alone would not release worker 1. The name-lock part of the hypothesis is practically unreachable (controller-runtime never runs one key concurrently; Delete's RPCs are too small to fill the buffer). GC (`gc.go:104-130`, one goroutine, no deadline) normally moves on; it stalls only if the blocked kill is on the *pooled* link, whose only bulk write is the ~50 KiB seed upload (`create.go:210`): marginal. Of the five `markDead` sites, `hosts.go:153` runs under `e.sem` but only handshake bytes have been sent, so it does not block.
+
+**Minimal fix.** (1) In `sshSocket.Close`, close the TCP connection first: `errors.Join(s.client.Close(), s.Conn.Close())`, filtering the expected "use of closed network connection", or drop `Conn.Close()`. (2) In `markDead` (`backend.go:138-148`), `close(l.closed)` before `l.kill()`, or run `kill` in a goroutine. (3) Regression test in `ssh_test.go`: frozen proxy, small client `SO_SNDBUF`, assert `Close` returns within 1 s while a writer is blocked. (4) Optional: set `ReconciliationTimeout` so image-lock waiters requeue.
+
+**Probe to settle** (not run, needs a bulk upload in flight): `ss -tmi dst <beelink-ip>:22` on the operator host during an image upload; `tb` below ~2.2 MB means the blocking case applies.
+
+## E2 — reformatted GC mark makes GC delete a live VM — PLAUSIBLE
+
+Throwaway program (`scratchpad/e2/main.go`) with a verbatim copy of `instanceMetadata`, the same `time.Parse(time.RFC3339, …)` and the `gc.go:112-122` switch; clock 16 days after the mark, MinAge 10 m.
+
+| Input | Parsed UID | zero time | GC |
+|---|---|---|---|
+| single line (as written) | clean | no | KEEP |
+| libvirt's re-indent between children | clean | no | KEEP |
+| `<uid> X </uid>` or newline-padded | padded | no | **DELETE** |
+| uppercase UID | differs | no | **DELETE** |
+| `<uid>  </uid>` (whitespace only) | `"  "` | no | **DELETE** |
+| `<uid></uid>` | `""` | no | KEEP (guard `gc.go:113`) |
+| bad/missing `createdAt`, UID clean | clean | yes | KEEP |
+| padded `<managed>` | clean | no | KEEP (not listed) |
+| future-dated, UID clean | clean | no | KEEP |
+
+`encoding/xml` does not trim chardata. **Path:** padded or case-changed `<uid>` with `<managed>true</managed>` intact → `parseInstanceMetadata` (`metadata.go:66-78`) returns it untrimmed → `ListByTag` (`list.go:46-49`) emits it → `Sweeper.sweep` (`gc.go:113-121`): not empty, not live, older than MinAge → `prov.Delete` (`gc.go:125`) → `delete.go:21-53` destroys the domain and its disks. There is no second guard: no name-to-Proxy lookup, `Delete` does not re-read the mark, and provider scoping does not help because the owning provider lists it.
+
+**Why only PLAUSIBLE.** On the real host libvirt puts newline + two spaces *between* children, strips the xmlns and leaves text bytes intact (`od -c` of the live `…/kvm` marks). Per libxml2, text-bearing elements are never re-indented, so a version change alone would not pad text. The reachable trigger is a hand edit or a tool that pads text nodes via `virsh edit`/`define`: rare, but the result is an unrecoverable delete. The old `internal/provider/kvm/metadata.go` has the same untrimmed field, and the two live VMs are kvm v1 domains.
+
+**Refuted parts.** A bad `createdAt` never deletes a live VM (UID check comes first, consistent with `docs/provider-contract.md:238-241`). A future-dated mark is a leak only, and marks are stamped from the operator's clock (`create.go:127`), so host skew cannot cause it.
+
+**Minimal fix.** At the end of `parseInstanceMetadata`: `TrimSpace` on `Managed`, `UID` (plus `ToLower`), `CreatedAt`. A whitespace-only UID then equals `""`, which `gc.go:113` keeps. Add table rows to `metadata_test.go`. Apply the same to `internal/provider/kvm/metadata.go`. Optional second guard in `gc.go`: skip and log when `provider.NameFromUID(inst.UID)` is not the name part of `inst.ID` (check the other providers' naming first).
+
+**Probe to settle** (not run; use a scratch libvirt, not beelink):
+```
+virsh -c qemu:///system metadata <scratch-dom> https://gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/kvm2 --key kvm2 --set '<instance><managed>true</managed><uid> 3f1c2a9e-0000-4000-8000-000000000001 </uid><createdAt>2026-09-01T10:00:00Z</createdAt></instance>' --config
+virsh -c qemu:///system metadata <scratch-dom> https://gitea.home.hrajfrisbee.cz/kacerr/egress-proxies-operator/kvm2 | od -c
+```
+
+## E3 — kvm and kvm2 on the same hypervisor — three paths, three verdicts
+
+**1. GC reconciles provider A's list against provider B's Proxies — REFUTED.** `gc.go:91-101` builds one global `live` set from every Proxy; sharing it only makes GC more conservative. kvm2 `ListByTag` (`list.go:38-48`) is namespace-filtered, and that was checked on the host: both live domains return the mark under `…/kvm` and nothing under `…/kvm2`. What remains is leaks: after a re-type the old type's orphans are never collected, and the scoping still rests on libvirt behaviour no unit test covers (S7.4).
+
+**2. Same UID under both types; reverse direction — REFUTED beyond S2.3.** `NameFromUID` depends only on the UID and `spec.provider` is CEL-immutable (`proxy_types.go:221`), so the same name under both types only comes from re-typing an entry. Nothing prevents that: `config.go:93-135` checks only name uniqueness and block/type agreement, there is no webhook (`cmd/main.go:262`), `status.providerID` records no type, and the controller never compares `inst.UID` with `p.UID`. What gets deleted is still that Proxy's own instance. Exceptions: a tampered `status.providerID` (kvm2 `Delete` checks no mark, so `hosts/beelink/domains/docker-29` would destroy an unrelated VM); the reverse direction has no guard at all (kvm `Create` adopts any same-named domain, `kvm/kvm.go:177-181`; its `Get`/`Delete` check no mark); and after a re-type kvm2 `ensureOverlay` boots a stale overlay the kvm provider left behind (`create.go:171-177`), one Proxy's own disk only.
+
+**3. Shared `base-<key>.<format>` cache — CONFIRMED** (kvm2 code + libvirt 11.6.0 source, `storage_driver.c`). The marker protocols are compatible (both write `<base>.sha256` with 64 hex; both key URL images as `sha256(url)[:16]`), so kvm2 adopts a kvm-cached base. On the host every base and marker dates from 2026-09-10 (kvm-made), both live overlays back onto `/srv/vms/images/base-debian-13.qcow2`, and the sample `kvm2-home` entry's `debian-13` key would share exactly that volume. Both integration tests use key `test-image`.
+- **3a Digest mismatch.** `kvm-home` cached `debian-13` from a `latest` URL (digest D1). The operator gives `kvm2-home` `sha256: D2`. First kvm2 Create: `cachedImage` `image.go:129` → `:145` deletes marker and base. libvirt's `storageVolDelete` refuses only on `in_use`/`building` (`storage_driver.c:1844-1856`), not on backing-store use. kvm2 uploads a different image to the same path under the kvm provider's live overlays. With both entries pinned to different digests, every Create on either side deletes and refetches.
+- **3b Concurrent cold fetch** (by source, not executed). Image locks are per `Provider` instance (`kvm2.go:111`; `kvm/image.go:62-74`), so this applies kvm-vs-kvm2 and kvm2-vs-kvm2 alike. libvirt holds `in_use` only while the stream is set up (`storage_driver.c:2421-2427`), so the volume is unguarded for the whole transfer. Entry B finds base-without-marker and deletes it (`kvm2/image.go:117-118`; `kvm/image.go:103-108`), recreates it and starts its own upload. A's upload finishes cleanly into the unlinked inode, the digest matches, A writes the marker, and `StorageVolGetPath` resolves A's handle to B's partial file: A boots a guest from a half-written base. If B is kvm2, `writeMarker` then fails `ErrStorageVolExist` and `image.go:241-243` discards the now-complete base under A's running domain.
+
+**Minimal fix.** Give kvm2 its own prefix (`volume.go:17` → `kvm2-base-…`, one extra ~415 MB download per host). Reject two kvm/kvm2 entries with the same address+socket+pool at load, or treat `ErrStorageVolExist` on the marker as success when its digest equals ours. Put the digest in the base name when `sha256` is set. Make `removeDomain` require `meta.managed()` and the expected UID; give `kvm/kvm.go:177` the same guard. Add a fake keyed on the metadata URI (S7.4).
+
+**Probe to settle 3b** (not run):
+```sh
+truncate -s 4G /tmp/probe-race.bin
+virsh -c qemu:///system vol-create-as default probe-race.raw 4G --format raw
+virsh -c qemu:///system vol-upload --pool default probe-race.raw /tmp/probe-race.bin &
+sleep 1; virsh -c qemu:///system vol-delete --pool default probe-race.raw; echo "delete exit=$?"
+wait; echo "upload exit=$?"; virsh -c qemu:///system vol-list default | grep probe-race; rm /tmp/probe-race.bin
+```
+
+## E4 — no cross-process fencing on the base volume — CONFIRMED (narrowed)
+
+**Rollout / leader handover: REFUTED for the shipped manifests.** `manager.yaml:64` passes `--leader-elect` (the `false` default at `cmd/main.go:107` applies only to a bare binary or `make run`). No strategy is set, so RollingUpdate with surge 1: old and new pods coexist, but the reconciler is a leader-election runnable, `LeaderElectionReleaseOnCancel` is not set (`cmd/main.go:288-295`), and the old pod holds the lease until it dies. Lease 15 s / renew 10 s / retry 2 s; the kubelet SIGKILLs at 10 s (`manager.yaml:180`). On cancel kvm2 stops the upload at once (`backend.go:179-181` → `markDead`); `discardVolume` runs uncancelled for up to 30 s (`image.go:331`) but is cut by SIGKILL. The residual two-process window needs A's delete RPC to stall ~7 s while B dials, HEADs and creates within ~5 s: theoretical, and self-healing (`image.go:100`, `:135`). Other two-process cases: `make run` without `--leader-elect` against the cluster's hypervisor; `make test-integration-kvm` and `-kvm2` at the same time (same `test-image` key).
+
+**In one process: CONFIRMED.** The kvm provider is an always-present second writer: same names (`kvm/image.go:44-57` vs `kvm2/volume.go:17-25`), same default pool, same marker format, separate locks (`kvm/image.go:77-87` vs `Provider.imageLocks`; `newProvider` builds fresh `keyedLocks` per entry, `kvm2.go:111`, so two kvm2 entries do not share one either), concurrent via `MaxConcurrentReconciles: 3`, documented as supported (`docs/kvm2-provider.md:6-8`) and shipped that way (`config/manager/providers_config.yaml`: both entries on `beelink`, key `debian-13`, default pool). The interleaving is E3-3b. Extra consequence: after A's marker is written, if the process dies during B's upload (rollout SIGKILL, OOM at the 128Mi limit) the pool keeps a **truncated base under a valid marker**, and nothing re-hashes the base (`cachedImage` compares marker text only), so every later Create on that host backs onto it until a manual `vol-delete`.
+Caveat: that `StorageVolDelete` succeeds mid-stream came from memory here; the E3 reviewer read it in the 11.6.0 source, which corroborates it. Still unexecuted.
+
+**Minimal fix.** (1) Remove the shared name (kvm2-specific prefix; include the entry name or share one process-wide `keyedLocks`). (2) Make the base appear atomically: upload to `<base>.part-<random>`, verify, then create the final name; treat `ErrStorageVolExist` as "someone else won" and re-check; never delete a name this call did not create, in particular not `discardVolume(base)` on a marker `ErrStorageVolExist` (`image.go:241-243`). (3) Store the size next to the digest and check the base's allocation before trusting the marker. Set `strategy: Recreate` and `LeaderElectionReleaseOnCancel` deliberately.
+
+**Probe to settle** (not run):
+```
+virsh -c qemu:///system vol-create-as default e4probe.raw 2G --format raw
+mkfifo /tmp/e4.fifo; ( head -c 2G /dev/zero | pv -L 20m > /tmp/e4.fifo & )
+virsh -c qemu:///system vol-upload --pool default e4probe.raw /tmp/e4.fifo &
+sleep 5
+virsh -c qemu:///system vol-delete --pool default e4probe.raw        # succeeds => confirmed path; "still in use" => narrowed
+virsh -c qemu:///system vol-create-as default e4probe.raw 2G --format raw
+virsh -c qemu:///system pool-refresh default
+wait; virsh -c qemu:///system vol-delete --pool default e4probe.raw; rm /tmp/e4.fifo
+```
+
+## E5 — overlay atomicity under an abandoned RPC — REFUTED (~85%, libvirtd side from source memory)
+
+`storageVolCreateXML` rejects a duplicate name, adds the volume with `building = true`, bumps asyncjobs, drops the pool lock, runs `qemu-img create` via `virCommandRun`; on build failure the file is unlinked and the volume removed from the list. It runs on an RPC worker thread holding a client reference, so a disconnect does not cancel it; storage objects have no close callbacks. After a client drop a name is either still building or complete. Pool refresh refuses while asyncjobs > 0, and kvm2's XML has no `<allocation>` and passes `flags = 0`, so the preallocating path is never taken. The spec's "creation is atomic" (`kvm2-provider-spec.md:227-229`) holds.
+
+**Cost on the host.** All six pools are `type='dir'` on local storage. Three leftover overlays show birth-to-mtime gaps of 2-11 ms and are each exactly 196768 bytes, complete. A create over 30 s needs a stalled filesystem.
+
+**Residual path** (loud, never boots a corrupt overlay): Create #1 stalls > 30 s, Create #2 gets `ErrStorageVolExist` while `building` is set; if it reaches `DomainCreate`, QEMU cannot open the file (write lock, or "not in qcow2 format") → Transient → next retry starts cleanly. If `qemu-img` ends in EIO, libvirtd removes the volume and the defined domain points at a missing file; every retry fails until the Proxy is deleted (this is sweep finding S2.4's mechanism).
+
+**Fix options.** None needed for the hypothesis. Minimal: correct the comment at `create.go:151-153` and the spec line to state the real reasons. Hardening: `deleteVolumeIfPresent` before `StorageVolCreateXML` in `ensureOverlay`, as the seed already does (no domain exists at that point under the name lock; cost 2-11 ms; a building volume returns `ErrOperationInvalid`, already Transient).
+
+**Probe to settle** (not run, ~420 MiB in `default`):
+```sh
+C=qemu:///system; V=review-e5-clone.qcow2
+virsh -c $C vol-clone --pool default base-debian-13.qcow2 $V & P=$!
+sleep 0.3; kill -9 $P
+for i in $(seq 1 40); do virsh -c $C vol-info --pool default $V 2>&1 | tr '\n' ' '; echo; stat -c '%s' /srv/vms/images/$V 2>&1; sleep 0.5; done
+virsh -c $C vol-dumpxml --pool default $V
+until virsh -c $C vol-delete --pool default $V 2>&1 | grep -qv 'still being allocated'; do sleep 1; done
+virsh -c $C vol-list default | grep review-e5- || echo clean
+```
+
+## E6 — go-libvirt when the socket is killed under it — (a) CONFIRMED, (b) REFUTED
+
+Scratch module `scratchpad/lvleak/` (same go-libvirt and x/crypto versions; copies kvm2's `link`, `call`, `connect`, `Close`, `sshSocket` in shape).
+
+**(a) Data race on `Libvirt.disconnected`, reported by `-race`** over a real in-process SSH `direct-streamlocal` channel: read in `waitAndDisconnect` (`libvirt.go:821`, goroutine started at `:297`), write in `ConnectToURI` (`:305`), unsynchronised. Interleaving: ConnectOpen reply is routed → socket dies → `listen` closes `s.disconnected` (`socket.go:208`) → `waitAndDisconnect` reads the old channel and returns (`:828`) → the requester installs a fresh channel at `:305` that nothing will ever close, and `Connect()` returns nil.
+- **Hypothesised trigger (kvm2's own `markDead`): harmless.** `call` has already taken the timer/ctx branch; the nil result lands in the buffered `done` and is dropped; `hosts.go:153` discards the backend. Measured 5, 1 and 33 times per 1000 cycles, no effect.
+- **Peer-close trigger: real defect.** If libvirtd or the forwarded socket goes away within ~1 ms of a successful open (e.g. a daemon restart), `Connect` wins, the watcher (`backend.go:229-236`) holds a channel that never closes, every later RPC fails at `SendPacket` with `syscall.EINVAL` (`socket.go:321-325`), which `errors.go:55` classifies Transient. No call result marks the link dead, `hostTable.get` (`hosts.go:75`) keeps returning it, and the SSH keepalive keeps succeeding if sshd is alive. The host answers "invalid argument" until the operator restarts. Wedged in 953/1000 (unix), 824/1000 (`net.Pipe`), 546/1000 (SSH), 186/500 (SSH, `-race`) when the peer closes right after the open reply. On the Check path the same loser only costs a 5.001 s `Close()`.
+
+**(b) No accumulation.** Six scenarios (peer never answers then kill; RPC hangs; ConnectClose hangs into the close budget; peer closes mid-handshake; clean probe; kill racing the end of the handshake) x three transports, 1000 cycles each (500 under `-race`): goroutines +0, fds +0. Peak in flight 26.
+
+**Side findings (need a misbehaving peer).** (i) `callback` does an unbuffered `c <- res` holding `cmux` (`rpc.go:76-84`): a duplicate reply deadlocks `listen`, leaks three goroutines per hit (+420 over 200 cycles), and a socket kill does not help. On the stream path serials do repeat, which is why the S4.1 leak also survives a kill. (ii) `socket.go:233-234` computes `int(length) - 28` and `make`s it: a length field under 28 panics "makeslice: len out of range" in the listen goroutine and takes the operator down (reproduced); a large bogus length allocates up to 4 GiB. Reachable if `host.Socket` points at something that is not libvirt.
+
+**Minimal fix.** In `call` (`backend.go:177-178`): on `done`, if the error is `syscall.EINVAL` or `libvirt.ErrInterrupted`, `markDead` before returning; add a `backend_test.go` case where the peer closes right after the open reply. Sturdier: give `libvirt.New` a `net.Conn` wrapper whose `Read` calls `link.markDead` on any non-temporary error, dropping the dependence on `Disconnected()` and the 5 s Close stall. Upstream: assign `l.disconnected` before `go l.waitAndDisconnect()`; send outside the lock in `callback`; reject `length < 28` and oversized lengths.
+
+## E7 — in-place base replacement under live overlays — PLAUSIBLE
+
+**Confirmed part.** Admin sets or changes `images.<key>.sha256` and restarts the manager (config is not hot-reloaded). Next Create: `image.go:129` sees the mismatch, `:145` deletes marker and base with no in-use check and a `V(1)` log; `fetchImage` creates a new file at the same path. `TestEnsureImage_sha256ChangeRedownloads` (`image_test.go:394`) asserts exactly this. A mistyped `sha256` is worse: the base is deleted first, then `image.go:289` returns Permanent, leaving every old overlay with a missing backing file (loud). libvirt does not refuse (~95% from the reviewer; the E3 reviewer read it in source).
+
+**Why corruption is only plausible.** Running VMs keep the old inode, and `on_reboot=restart` (`xml.go:217`) keeps the same QEMU process. Corruption needs a *new* QEMU process on an old overlay:
+- Controller: safe. Stopped → Delete (`proxy_controller.go:216-225`) removes the overlay (`delete.go:91`).
+- **`libvirt-guests` on host reboot.** Enabled on beelink (`multi-user.target.wants/libvirt-guests.service`); `/etc/default/libvirt-guests` could not be read. Stock `ON_BOOT=start` restarts every guest that was running at shutdown regardless of autostart; the reviewer is ~70% sure the distro default is also `ON_SHUTDOWN=suspend`. While the host is down `Get` is Transient, not Stopped, so nothing is deleted; after boot the operator normally sees Running and keeps the VM.
+- A manual `virsh start`.
+- Create adoption path (`create.go:73-87` calls `ensureRunning` without `ensureImage`): only when `providerID` is empty and the overlay already holds guest writes. Narrow.
+
+The missing/unreadable-marker trigger is **not** reachable within kvm2 without an edit (marker errors are returned, not treated as unreadable, `image.go:119-125`). The second deleter that does exist is the kvm provider (E3/E4). Host state: both live domains running, persistent, autostart disabled, no managed save; the hazard is documented (`kvm2-provider-spec.md:91-93`, `kvm2-provider.md:313-318`), but the code contradicts the "add a new key" rule instead of enforcing it, and the guide's warning names `url` although a `url` change alone never refetches.
+
+**Minimal fix.** At `image.go:129-130` do not delete; return `permanentf("base %s is cached with sha256 %s but %s is configured; add the image under a new key, …")`. Keep deletion for the no-marker and non-hex branches. Alternative: digest in the volume name. Either way log at `V(0)`.
+
+**Probe to settle** (not run):
+```
+P=default; V="virsh -c qemu:///system"
+$V vol-create-as $P review-e7-base.qcow2 16M --format qcow2
+$V vol-create-as $P review-e7-overlay.qcow2 16M --format qcow2 --backing-vol review-e7-base.qcow2 --backing-vol-format qcow2
+stat -c '%i' /srv/vms/images/review-e7-base.qcow2
+$V vol-delete --pool $P review-e7-base.qcow2        # expected: succeeds
+$V vol-create-as $P review-e7-base.qcow2 16M --format qcow2
+stat -c '%i' /srv/vms/images/review-e7-base.qcow2   # expected: different inode
+$V vol-delete --pool $P review-e7-overlay.qcow2; $V vol-delete --pool $P review-e7-base.qcow2
+```
+Read-only, for the owner: `grep -Ev '^\s*(#|$)' /etc/default/libvirt-guests` (empty output = defaults apply).
+
+## E8 — pool-space precheck freshness — CONFIRMED, and staleness is the smaller of two errors
+
+Measured 2026-09-17 08:28-08:30 UTC, read-only. kvm2's pool is `default` at `/srv/vms/images`.
+
+| pool | filesystem | libvirt Available | `f_bfree` | `f_bavail` | libvirt − bfree | libvirt − bavail | vols |
+|---|---|---|---|---|---|---|---|
+| default (kvm2) | ubuntu--vg-vms--lv | 266,038,706,176 | 266,038,378,496 | 241,304,276,992 | +327,680 | **+24,734,429,184** | 24 |
+| pool | same | 267,718,541,312 | 266,038,378,496 | 241,304,276,992 | +1,680,162,816 | +26,414,264,320 | 15 |
+| vms | same | 267,718,541,312 | 266,038,378,496 | 241,304,276,992 | +1,680,162,816 | +26,414,264,320 | 9 |
+| boot / images | ubuntu--lv | 73,877,696,512 | 73,918,029,824 | 68,532,543,488 | −40,333,312 | +5,345,153,024 | 0 / 2 |
+| vms-2 | nvme1n1p1 | 820,622,606,336 | 820,602,261,504 | 770,575,294,464 | +20,344,832 | +50,047,311,872 | 2 |
+
+- **Cached per pool.** Three pools on one filesystem report two different values; a call-time value could not differ. Pool A's allocation is invisible to pool B. Over 80 s `f_bfree` moved on the `images` filesystem while libvirt stayed frozen, 40 MB stale in the pessimistic direction, so the cache errs both ways. `default` is nearly exact only because libvirt touched it ~1.6 h earlier.
+- **The root reserve.** libvirt reports `f_bfree`, which includes ext4's 5% reserve. Overlays are owned by `libvirt-qemu:kvm`, so guest writes stop at `f_bavail = 0` while libvirt still reports ≥ 24.7 GB.
+
+**Failing path.** `create.go:92-100`: pool lookup → `imageSizeIfUncached` → `checkPoolSpace`, no refresh; `image.go:450-458` compares raw `available`; the only refresh is `image.go:235` after an upload, which never runs again once the base is cached. Overlays start at ~196 KB and grow invisibly (live ones at 1.47 and 1.54 GB of 10 GiB), as does `kube-node-33.qcow2` (50.6 GB) in the same directory. Reserve scenario (deterministic here): with `f_bavail` between 0 and ~10.7 GB the precheck passes for a 10 GiB disk, `ErrQuotaExceeded` never fires, and Create, the upload and the guests hit ENOSPC, the condition behind S4.1's wedge and S2.2's paused guests. `check.go:106-112` has the same defect: `poolAvailableGiB` over-reports by ~23 GiB plus drift.
+
+**Minimal fix.** (1) `StoragePoolRefresh` immediately before `checkPoolSpace`, after `imageSizeIfUncached`; failure is Transient; only on the domain-not-defined branch. Cost: readdir + stat per volume (24 today), already paid at `image.go:235`. (2) Do not refresh in Check (a mutation every round on a shared host); document the semantics. (3) Add headroom to `neededBytes`: per-host `poolReserveGiB` or a fraction of `capacity` (already returned and discarded at `image.go:450`); or document `tune2fs -m 0`. (4) Make the fake's `StoragePoolGetInfo` return a cached value that only `StoragePoolRefresh` updates; add `TestCreate_refreshesPoolBeforeSpaceCheck`.
+
+**Probe to settle** (for the owner): `virsh pool-info --bytes vms`, `virsh pool-refresh vms`, `pool-info` again; expect Available to drop ~1.68 GB and still sit ~24.7 GB above `df -B1 --output=avail /srv/vms`.
+
+## E9 — libvirt error-code assumptions — (a) REFUTED, (b) CONFIRMED
+
+**(a) Idempotency codes match libvirt 11.6.** Codes identified from libvirt's fixed per-code message prefix with `LIBVIRT_DEBUG=1`.
+
+| Operation | kvm2 expects | Real | Evidence |
+|---|---|---|---|
+| `DomainLookupByName`, missing (`create.go:88`, `delete.go:57`, `get.go:34`) | 42 | 42 | observed |
+| `StorageVolLookupByName`, missing (`image.go:117,134,176,424`) | 50 | 50 | observed |
+| `StoragePoolLookupByName`, missing (`delete.go:84`) | 49 | 49 | observed |
+| `DomainGetMetadata`, no kvm2 element (`get.go:81`) | 80 | 80 | observed on `docker-29`, `postgres`, and a kvm-v1-marked proxy domain; namespace scoping holds |
+| `StorageVolCreateXML`, duplicate (`create.go:171`) | 90 | 90 | sourced, ~95% |
+| `DomainDestroy` inactive (`delete.go:67`) | 55 | 55 | sourced, high |
+| `DomainCreate` active (`create.go:236`) | 55 | 55 | sourced, high; `waitRunning` re-checks anyway |
+| Destroy/Undefine vanished domain | 42 | 42 | sourced, high |
+| `StorageVolDelete` vanished (`image.go:182`) | 50 | 50 | sourced, high |
+
+`TestKVM2_lifecycle` would exercise 42 and 50 but not 90 (the second Create returns from the adoption branch before `ensureOverlay`), not 55 from Destroy, and a wrong 80 would not fail it (`list.go:39-45` logs and continues). Two fake-vs-real differences, neither breaking convergence: an overlay file unknown to libvirt returns 55 not 90; an inactive pool returns 55 from the marker lookup in `imageSizeIfUncached` (`create.go:96`) before `checkPoolSpace` (`create.go:100`), so the Permanent arm at `image.go:454-456` is unreachable through Create (`image_test.go:821-828` tests it by direct call).
+
+**(b) Classification.** `classifyLibvirt` (`errors.go:58-71`) sends every unlisted code to Transient; `providerFailure` (`proxy_controller.go:344-345`) returns the error and sets **no condition**, so the Proxy shows nothing and retries under workqueue backoff, rerunning `writeSeed` each time when the failure is at define.
+
+| Rejection | Where | Code | Evidence |
+|---|---|---|---|
+| Host has no usable KVM | define | 67 | sourced, high |
+| No OVMF firmware | define | 9 | sourced, ~85% |
+| Bridge missing | `DomainCreate` | 38 | sourced, high; `check.go` never looks at the bridge |
+| host-passthrough unsupported | `DomainCreate` | 67 | sourced |
+| Non-x86 host | define | 67 or 1 | observed 67 on the `domcapabilities --arch aarch64` path |
+| Unknown machine type | define | 8 (already Permanent) | observed |
+
+beelink rejects none of these. The bare `libvirt.ErrUnsupported` sentinel (`rpc.go:392-394`) bypasses `libvirtCode`; small impact. This follows a recorded decision: spec entry 13 (`kvm2-provider-spec.md:317`) "unknown means retry", and contract §5 makes Permanent sticky until the generation changes, so a host-side fix would not unlatch. The spec never looked at 67, and it already latches `ErrNoStoragePool`, which is also host-fixed.
+
+**Minimal fix.** `ErrConfigUnsupported` → Permanent when `op == "create"`; `errors.Is(err, libvirt.ErrUnsupported)` → Permanent before the `libvirtCode` check; leave 9, 38, 1, 84 alone. Give the Transient arm of `providerFailure` a non-latching `Provisioned=False` condition with the message, or have the health checker read `ConnectGetDomainCapabilities` and check the bridge. In the integration test: destroy+undefine over `dialFresh` then Create again (exercises 90), destroy before step 9's Delete (exercises 55), assert "skipping domain with unreadable metadata" never logs.
+
+**Probe to settle** (not run):
+```
+virsh -c qemu:///system vol-create-as default review-e9-dup.qcow2 1M --format qcow2 && \
+LIBVIRT_DEBUG=1 virsh -c qemu:///system vol-create-as default review-e9-dup.qcow2 1M --format qcow2 2>&1 | grep ' error : ' ; \
+virsh -c qemu:///system vol-delete --pool default review-e9-dup.qcow2
+```
+
+## E10 — overlay capacity below the base's virtual size — CONFIRMED (libvirt pass-through ~90%, unexecuted)
+
+**Host numbers.** `base-debian-13.qcow2` and `base-test-image.qcow2`: libvirt capacity 3221225472 (3 GiB virtual), allocation 435683328. Live overlays: 10 GiB over that base. So the 10 GiB default has ~7 GiB margin; only `diskSizeGb: 1` or `2` breaks Debian 13 (Ubuntu ~3.5 GiB and Fedora ~5 GiB from memory). libvirt reports the true virtual size because of the refresh at `image.go:235`.
+
+**Local experiment** (qemu-img 10.2.1; host has 10.1.0): `qemu-img create -f qcow2 -b <64M base> -F qcow2 short.qcow2 16384K` exits 0, virtual size 16 MiB, `qemu-img check` clean; reads inside 16 MiB match the base, a read at 16M fails `Input/output error`. A 64M GPT base under a 16M overlay: `sfdisk` sees only the protective MBR, `sgdisk -v` reports the disk smaller than the header says and the backup header in error. The Linux kernel rejects such a GPT, so there is no root device; the guest would load GRUB and the kernel then drop to initramfs (inferred, no VM booted). libvirt's `virStorageBackendCreateQemuImgCmdFromVol` appends the capacity with no size comparison (from source memory).
+
+**Failing path.** `diskSizeGb: 2` passes validation (`config_kvm.go:133` rejects only negatives; `kvm2.go:127-128` changes only 0) → `checkPoolSpace` with the small number (`create.go:100`) → base cached (`:103`) → `ensureOverlay` sends `diskSizeGB << 30` (`create.go:107,158`; `volume.go:72` checks only positive) → domain defined and started, `waitRunning` sees Running, Create succeeds (`:148,260`) → `Get`: active, no address → `StateProvisioning` (`state.go:20-21`) → requeue at `ProvisioningPoll` forever (`proxy_controller.go:208-211`; there is no provisioning deadline). Correction to the hypothesis: the Proxy is stuck Provisioning with "waiting for the instance to run", not reported healthy. Nothing points at `diskSizeGb`, and raising it later does not repair that Proxy (`create.go:75-87`, `:171-177`). The kvm provider has the same hole (`kvm/volume.go:24-32`).
+
+**Minimal fix.** Add `StorageVolGetInfo` to `backend` (libvirtBackend, `observed`, fake already stores `fakeVol.capacity` at `fake_test.go:494`). Between `ensureImage` and `ensureOverlay`, if the base's capacity exceeds `DiskSizeGB << 30` return `permanentf("kvm2.diskSizeGb %d is smaller than image %s's virtual size …")`. One or two RPCs on the first Create only. Run the same check on a reused overlay and recreate it when short. `TestCreate_diskSmallerThanImage`. One sentence in the `diskSizeGb` doc rows. A config-time check is not possible.
+
+**Probe to settle** (not run; one ~200 KiB volume, self-cleaning):
+```bash
+V=review-e10-short.qcow2; P=default
+virsh -c qemu:///system vol-create-as $P $V 1G --format qcow2 \
+  --backing-vol /srv/vms/images/base-test-image.qcow2 --backing-vol-format qcow2; echo "create exit=$?"
+virsh -c qemu:///system vol-dumpxml --pool $P $V | grep -E "capacity|backingStore|<path>"
+virsh -c qemu:///system vol-delete --pool $P $V; echo "delete exit=$?"
+virsh -c qemu:///system vol-list $P | grep review-e10- || echo "clean"
+```
@@ -0,0 +1,102 @@
+# kvm2 provider review: escalations for phase 2
+
+18 raw Escalate items from the seven sweepers, merged to 15 root hypotheses, top 10 kept.
+Ranked by blast radius x uncertainty. `deep-reviewer` is static (Read, Grep, Glob, Bash; it
+can read the Go module cache and run local tests, but not touch the hypervisor), so each item
+says whether it can be **settled statically** or is likely to end PLAUSIBLE without a probe on
+beelink. Finding IDs (`S…`) refer to `sweep.md`.
+
+| # | Hypothesis in one line | Blast | Uncertainty | Static? |
+|---|---|---|---|---|
+| E1 | One wedged `markDead` starves every reconcile worker, not just one host | operator-wide stall | medium | yes |
+| E2 | A reformatted GC mark makes GC delete a live Proxy's VM | live VM destroyed | medium | mostly |
+| E3 | kvm and kvm2 on one hypervisor: cross-provider destruction beyond the Get/Delete hole | other provider's VMs/disks | medium | yes |
+| E4 | No cross-process fencing on the base volume: leader handover deletes a base mid-upload | corrupt base under new overlays | medium | yes |
+| E5 | An abandoned `StorageVolCreateXML` leaves a corrupt overlay that the next Create adopts | silent guest disk corruption | high | partly |
+| E6 | go-libvirt leaks or races when the socket is killed mid-connect; per-round check dials accumulate | slow leak in long-running leader | medium | yes |
+| E7 | In-place base replacement corrupts existing overlays on their next boot | guest disk corruption, many VMs | low-medium | partly |
+| E8 | Pool-space precheck reads a stale `available`; quota backoff never engages | host pool fills | medium | partly |
+| E9 | libvirt error-code assumptions: idempotency codes and define-time rejection codes | retry loops that never converge | medium | partly |
+| E10 | Overlay capacity below the base's virtual size is accepted and truncates the disk | silent guest corruption | high | no |
+
+---
+
+## E1 — A wedged `markDead` blocks name locks and starves the reconcile workers
+Sources: chunk 1 Escalate 2; builds on finding S1.1.
+**Entry points:** `internal/provider/kvm2/kvm2.go:186-243` (`lockName`, `keyedLocks`), `create.go`, `delete.go`, `backend.go:138-188`, `hosts.go:66-95`, `internal/gc/gc.go`, `internal/controller/proxy_controller.go` (+ `SetupWithManager`, `MaxConcurrentReconciles`), `cmd/main.go` manager wiring.
+**Hypothesis:** a goroutine blocked inside `markDead` (S1.1) holds the per-name lock taken by Create/Delete, and reconcile and GC contexts carry no deadline, so `keyedLocks.acquire`'s ctx escape never fires. Delete/GC for that name block indefinitely instead of requeueing, and with a small `MaxConcurrentReconciles` a single silent host can pin every worker, stalling all Proxies on all providers.
+**Checked:** lock acquisition is ctx-aware and refcounted correctly; only `internal/providerhealth/monitor.go:180` sets a deadline; the health checker and image upload use `dialFresh` and are insulated; `markDead` at `hosts.go:153` runs under `e.sem`.
+**Unknown:** whether the controller-runtime work queue tolerates a permanently blocked Reconcile for one key (worker starvation across all Proxies); the actual `MaxConcurrentReconciles` and any reconcile timeout in the manager wiring; whether GC's single goroutine is blocked the same way; how many distinct call sites can enter `markDead` while holding `e.sem` or a name lock.
+
+## E2 — GC mark parse fragility: a reformatted mark makes GC delete a live VM
+Sources: chunk 3 Escalate 1; finding S5.3.
+**Entry points:** `internal/provider/kvm2/metadata.go:44-78` (`createdAt`, `parseInstanceMetadata`), `list.go:49`, `get.go:79-88`, `internal/gc/gc.go:113-121`.
+**Hypothesis:** if libvirt hands `<uid>` or `<createdAt>` back with surrounding whitespace or re-indented content (another libvirt version, or an admin's `virsh edit`/`virsh define` through a reformatting tool), `UID` stops matching `live[inst.UID]` (exact string compare, `gc.go:118`) and/or `createdAt()` returns the zero time, which makes the instance immediately past `MinAge`. GC then deletes a live Proxy's VM. No trim or tolerant parse exists as a backstop.
+**Checked:** `newInstanceMetadata` writes a single-line element; `kvm2_integration_test.go:422-449` asserts the UID round-trips and `CreatedAt` is non-zero (sweeper's claim that this ran against real libvirt 11.6 was wrong: phase 2 found the kvm2 integration test has not yet run on beelink, see `deep.md`); libxml2 does not re-indent an element whose children have text content.
+**Unknown:** behaviour on other libvirt versions and after `virsh edit`; whether `encoding/xml` unmarshal plus the current struct tags already tolerate whitespace; whether a UID mismatch alone (timestamp intact) is enough for deletion, and what GC does with `UID == ""`; whether GC has any second guard (name-to-Proxy lookup) before `Delete`.
+
+## E3 — kvm and kvm2 on the same hypervisor
+Sources: chunk 3 Escalate 2, chunk 7 Escalate 4; builds on findings S2.3, S7.1, S7.4.
+**Entry points:** `internal/provider/kvm2/list.go:20-56`, `get.go:52-88`, `delete.go:55-98`, `volume.go:14-18`, `internal/provider/kvm/kvm.go:379`, `internal/provider/kvm/id.go:11-13`, `internal/provider/kvm/volume.go:12`, `internal/provider/kvm/image.go:46`, `internal/gc/gc.go:86-131`, `internal/provider/config.go:108-130`, `internal/controller/proxy_controller.go:197,237,287`, `docs/kvm2-provider.md:346-370`.
+**Hypothesis:** the only thing keeping kvm2's `ListByTag` from handing kvm-made domains to GC is libvirt's namespace-scoped `DomainGetMetadata`; both providers also share `<name>.qcow2`, `<name>-seed.iso`, `base-<key>.<format>` names and the providerID format. Beyond the confirmed Get/Delete hole (S2.3), GC's per-provider scoping may let provider A's list be reconciled against provider B's Proxies, and a Proxy moved between a kvm and a kvm2 entry (same UID, same domain name) may have one side delete volumes the other side's domain uses.
+**Checked:** domain names are UID-derived so one Proxy cannot exist under both at once; the two metadata URIs differ; `ListByTag` is mark-filtered; `spec.provider` is immutable but an entry's `name` is independent of its `type`; the fake cannot prove the namespace rule (S7.4).
+**Unknown:** whether GC builds its live-UID set per provider or globally, and what it deletes when the same UID is visible through two entries; whether anything (config reload guard, admission check on ProviderID, deploy tooling) prevents a re-typed entry from inheriting live `status.ProviderID`s; the reverse direction (kvm v1 `Get`/`Delete`/`ListByTag` against kvm2-made domains).
+
+## E4 — No cross-process fencing on the base volume
+Source: chunk 4 Escalate A.
+**Entry points:** `internal/provider/kvm2/image.go:85-146` (`ensureImage`, `lockImage`, `cachedImage`), `image.go:200-300` (`fetchImage`, `uploadImage`), `kvm2.go:199-243`, `cmd/main.go:107` (leader election flags), `config/manager/manager.yaml:24`.
+**Hypothesis:** process A is inside `fetchImage` (base created, marker not yet written); process B's `cachedImage` sees base-without-marker and deletes it (`image.go:118`), then its `StorageVolCreateXML` either fails `ErrStorageVolExist` or succeeds against a half-uploaded file. A's stream writes into an unlinked inode and A writes a marker for a base that no longer exists. An overlay created in between backs onto a deleted or wrong-content path.
+**Checked:** `keyedLocks` is process-local; `replicas: 1`; `--leader-elect` defaults to false; `Delete` never touches base volumes; nothing else in the repo names base or marker volumes.
+**Unknown:** whether a leader-election handover or a rolling update (old pod still inside a 20 minute `fetchImage` while the new pod reconciles the same Proxy) reaches this with the shipped manifests; whether the manager cancels in-flight reconciles on lost lease and how long `terminationGracePeriodSeconds: 10` leaves the old process uploading; whether B's marker written over A's bytes can ever validate.
+
+## E5 — Overlay atomicity under an abandoned RPC
+Source: chunk 2 Escalate 1.
+**Entry points:** `internal/provider/kvm2/create.go:151-181` (`ensureOverlay`), `backend.go:157-188` (`call`, budget, `markDead`), `backend.go:319-321` (`StorageVolCreateXML` under `budgets.rpc` = 30 s).
+**Hypothesis:** if `StorageVolCreateXML` for the overlay exceeds the 30 s budget or the caller's ctx ends, `call` kills the SSH transport mid-RPC. If libvirtd still registers the volume, possibly partial, the next Create hits `ErrStorageVolExist`, logs "overlay volume already exists" and defines a domain over a corrupt overlay. The base image has a marker and the seed is rebuilt unconditionally for exactly this reason; the overlay has neither.
+**Checked:** the reuse branch distinguishes only `ErrStorageVolExist`; no size, format or backing-path verification anywhere; `Delete` is the only thing that removes an overlay.
+**Unknown:** libvirtd's storage-driver behaviour when the client drops mid `virStorageVolCreateXML` (unlink vs leave registered); whether overlay creation can exceed 30 s in normal operation (NFS, LVM, preallocating filesystems) rather than only on cancellation; whether a cheap verification (`StorageVolGetInfo` capacity, or delete-and-recreate when no domain references it) is available through the existing `backend` seam.
+
+## E6 — go-libvirt internals when the socket is killed under it
+Sources: chunk 1 Escalate 1, chunk 6 Escalate 1.
+**Entry points:** `internal/provider/kvm2/backend.go:225-238` (`libvirtBackend.connect`, the `Disconnected()` watcher), `backend.go:351-367` (`Close`), `hosts.go:91-162` (`dialFresh`, `connect`), `check.go:85-119`; upstream `go-libvirt@v0.0.0-20260814190004-1a83157e1858`: `libvirt.go:286-305` (`ConnectToURI`), `:821/:834` (`waitAndDisconnect`), `socket/`.
+**Hypothesis:** (a) when `call(…, "Connect", …)` expires and `markDead` closes the socket just as `ConnectToURI` reaches `l.disconnected = make(chan struct{})`, there is a data race on that field against `waitAndDisconnect`, and the loser leaves a `Libvirt` whose `Disconnected()` never fires; (b) `Check` opens a fresh connection per host per round (~1440/day/host), so any goroutine or fd go-libvirt leaves behind on the killed-socket path (abandoned Connect, or `Close` hitting its 5 s budget mid `ConnectClose`) accumulates linearly in a long-running leader.
+**Checked:** kvm2's own goroutines are clean (watcher selects on `link.closed`, keepalive exits on `stop`, `sshSocket.Close` is once-guarded); `Disconnected()` is read only after a successful Connect and the backend is discarded on failure; `go test -race -count=2` of the unit tests is clean.
+**Unknown:** whether `waitAndDisconnect`/`listenAndRespond` always terminate when the conn is closed mid-handshake; whether the socket layer leaves an fd; whether a looped dial-and-kill test (N=1000, sampling `runtime.NumGoroutine()`) against the in-process SSH test server shows growth. Note finding S4.1 already confirms one permanent go-libvirt goroutine leak on the stream path; this asks whether the connect/close path has another.
+
+## E7 — In-place base-image replacement under overlays that pin its path
+Source: chunk 2 Escalate 2; the code behaviour is finding S4.6 (three sweepers).
+**Entry points:** `internal/provider/kvm2/create.go:103,154-165`, `image.go:113-146` (`cachedImage`), `image.go:235` (`StoragePoolRefresh`), `docs/provider-contract.md` ("Catalogue changes do not replace instances").
+**Hypothesis:** when `images.<key>.sha256` changes (or the marker is missing or unreadable), the base is deleted and different content is downloaded to the same deterministic path while every earlier overlay records that path as its qcow2 backing file. Any such overlay that is later restarted (adoption path, host reboot, `ensureRunning` on a shutoff domain) reads mismatched backing data: guest filesystem corruption across every VM on that key.
+**Checked:** `Delete` never removes the base; the path is stable per key+format; `BackingPath` is resolved once at define time and never re-verified; the contract explicitly allows catalogue edits not to replace instances.
+**Unknown:** which restart paths actually exist for an existing overlay in this operator (does anything restart a shutoff domain rather than delete and recreate it, given Stopped → Delete in the controller?); whether the missing/unreadable-marker trigger can fire without any admin edit (e.g. after E4 or a failed `writeMarker`), which would make this reachable in normal operation; whether libvirt refuses `StorageVolDelete` on a volume in use as a backing store.
+
+## E8 — Pool-space precheck freshness
+Source: chunk 4 Escalate B.
+**Entry points:** `internal/provider/kvm2/image.go:419-467` (`imageSizeIfUncached`, `checkPoolSpace`, `neededBytes`), `create.go:96-106`, `check.go:85-119` (also reads pool info).
+**Hypothesis:** for a `dir` pool, `StoragePoolGetInfo.available` is what libvirtd computed at the last pool refresh, adjusted only for creates and deletes it performed; guest writes growing sparse overlays do not update it. `create.go` never issues `StoragePoolRefresh` before the check (the only refresh is after an upload, `image.go:235`), so on a host its VMs have filled, the precheck reports tens of GB free indefinitely, `ErrQuotaExceeded`/`QuotaRetry` never engages, and Create fails later with an ENOSPC-flavoured transient, which is also the case that wedges the upload stream (S4.1) and pauses guests (S2.2).
+**Checked:** no refresh anywhere on the pre-create path; `available` is used raw; unit tests set the fake's value directly; the spec describes the precheck but says nothing about freshness.
+**Unknown:** libvirt 11.x dir-pool semantics for `available` (statvfs at call time vs cached); whether `Check`'s `poolAvailableGiB` has the same staleness; the cost of a per-Create refresh on the shared pooled connection.
+
+## E9 — libvirt error-code assumptions
+Sources: chunk 5 Escalate 2, chunk 7 Escalate 2; findings S3.5, S5.2.
+**Entry points:** `internal/provider/kvm2/errors.go:33-97` (`classify`, `classifyLibvirt`, `libvirtCode`, `hasCode`), `create.go:139-142,166-177`, `get.go:79-88`, `delete.go:55-98`, `image.go:174-186`, `fake_test.go:491-493`, `kvm2_integration_test.go`.
+**Hypothesis:** two halves of one assumption. (a) Idempotency depends on exact codes: `ensureOverlay` needs `ErrStorageVolExist` for a duplicate volume on a dir pool, `readMetadata` needs `ErrNoDomainMetadata`, the delete paths need `ErrNoDomain`/`ErrNoStorageVol`. The fake returns exactly the codes the code expects, so unit tests cannot refute a mismatch; if real libvirt returns `ErrOperationFailed` or `ErrInternalError` for any of them, a retried Create after a partial failure never converges. (b) Every define-time rejection of the generated XML (arch, machine, firmware feature, host-passthrough) surfaces as `ErrConfigUnsupported` (67) or `ErrOperationUnsupported` (84), which default to Transient: an unbounded retry with no `ReasonPermanentError`.
+**Checked:** only `ErrNoStoragePool`, `ErrXMLError`, `ErrNoSupport`, `ErrInvalidArg`, `ErrAuthFailed`, `ErrAccessDenied` map to Permanent; 67 and 84 exist in go-libvirt's const set and are unlisted; bare `libvirt.ErrUnsupported` bypasses `libvirtCode`; transient errors requeue forever.
+**Unknown:** which idempotency codes the integration test actually exercises against libvirt 11.6 (a second Create on an existing overlay? Delete twice? Get on a foreign domain?), which settles (a) for those paths; for (b), which code libvirt really returns for each rejection, and whether keeping unknown codes Transient is a deliberate spec decision. Mapping too much to Permanent latches Proxies that would recover.
+
+## E10 — Overlay capacity smaller than the backing image's virtual size
+Source: chunk 5 Escalate 1; finding S5.6.
+**Entry points:** `internal/provider/kvm2/volume.go:68-89` (`buildVolumeXML`), `create.go:107,154-181`, `image.go:449-467`, `internal/provider/config_kvm.go:133` (`diskSizeGb` validation).
+**Hypothesis:** with `kvm2.diskSizeGb` below the base image's virtual size, libvirt runs `qemu-img create -b <base> … <capacity>` and succeeds, producing a truncated virtual disk whose partition table overruns the device. The VM fails to boot or boots with a corrupt root filesystem, and kvm2 reports a healthy running domain that never answers.
+**Checked:** nothing in the package compares overlay capacity to the base's virtual size; the validator rejects only `diskSizeGb < 0`; the 10 GiB default exceeds current Debian/Ubuntu/Fedora cloud images.
+**Unknown:** whether libvirt's storage backend or qemu-img errors on a short capacity (then this is only a poor error message). Needs a real qemu-img; the deep reviewer can at most establish whether kvm2 has any way to learn the base's virtual size through the existing seam (`StorageVolGetInfo` on the base) and what a guard would cost. **Expect PLAUSIBLE.**
+
+---
+
+## Dropped (not in the top 10)
+
+- **D1** Deleting a volume while a libvirt upload stream is open on it (chunk 4 C): a consequence of confirmed finding S4.1, only changes whether its disk-space effect is real; needs a truncated-source probe on beelink.
+- **D2** cloud-init dispatch of a `text/plain` part inside vendor-data (chunk 5 E3): invisible-failure risk but only a boot test on a real image settles it; nothing static to add beyond S5.8.
+- **D3** ProviderHealth status churn every round waking watchers (chunk 6 E2): low blast radius, arguably intended per the code comment.
+- **D4** Nested volumeMount, `kvm-ssh` Secret inside the read-only `providers-config` mount (chunk 7 E1): pre-existing and shared with kvm; refuted if kvm has ever run in-cluster with the Secret present, otherwise one `kubectl apply` settles it.
+- **D5** Whether provider error strings (with the full image URL, S7.10) reach Proxy status conditions (chunk 7 E3): low blast radius; cheap and fully static if you want it added.
@@ -0,0 +1,133 @@
+# kvm2 provider review: scope
+
+Branch `feat/kvm2-provider` at `2510842`. Seed: `grep -ril 'kvm2\|libvirt'` excluding vendor,
+generated, testdata, docs. The minikube layout hints in the brief do not apply; this repo is
+the egress-proxies operator and kvm2 is `internal/provider/kvm2/`, a pure-Go provider
+(`digitalocean/go-libvirt` RPC over an SSH-tunnelled socket, no cgo, no driver plugin binary).
+
+Checklist items with no counterpart here, so reviewers can answer them quickly rather than hunt:
+- libvirt **network** lifecycle / DHCP / static IP: kvm2 creates no networks; it attaches to a
+  configured bridge and reads the address via `DomainInterfaceAddresses`. The live questions are
+  IP detection, timeouts and stale addresses.
+- GPU/PCI/NUMA passthrough: none.
+- driver plugin RPC boundary: none. Closest analogue is the libvirt RPC link over SSH
+  (`link`, `call`, budgets, dead-link detection, libvirt version differences).
+- cgo/build tags: no cgo. Only tag is `//go:build integration`.
+- Handle leaks: go-libvirt handles are plain values (no `Free`); the leakable resources are
+  SSH connections, the libvirt link, goroutines (keepalive, `await`, `call`), HTTP bodies,
+  temp dirs and half-written storage volumes.
+
+## CORE (kvm2-owned)
+
+`internal/provider/kvm2/` — 18 source files (~3.2k lines), 22 test files (~7.3k lines).
+
+| File | Lines | What |
+|---|---|---|
+| kvm2.go | 243 | `Provider`, `New`, options, `withDefaults`, `host`, `wrapErr`, `lockName`, `keyedLocks` |
+| ssh.go | 216 | `dialStage`/`dialError`, `loadCredentials`, `sshSocket` + keepalive, `dialSSHSocket`, `await` |
+| hosts.go | 162 | `dialer`, `hostTable` (per-host cached backend, `get`/`dialFresh`), `libvirtDialer` |
+| backend.go | 369 | `backend` interface, `budgets`, `link` (dead marking), `call`/`run`, `libvirtBackend` |
+| observe.go | 263 | `observed` backend decorator: span + log per libvirt call |
+| create.go | 280 | `Create`, `create`, `ensureOverlay`, `writeSeed`, `ensureRunning`, `waitRunning` |
+| delete.go | 98 | `Delete`, `removeDomain`, `removeVolumes` |
+| get.go | 119 | `Get`, `get`, `readMetadata`, `address` |
+| list.go | 56 | `ListByTag` |
+| state.go | 49 | `mapState`, `isActive`, `firstIPv4` |
+| id.go | 28 | `formatProviderID`, `parseProviderID` |
+| metadata.go | 78 | GC mark in domain `<metadata>`: marshal/parse, `managed`, `createdAt` |
+| errors.go | 97 | `permanentf`/`transientf`, `classify`, `classifyLibvirt`, `libvirtCode`, `hasCode` |
+| image.go | 467 | `resolveImage`, `lockImage`, `ensureImage`, `cachedImage`, marker read/write, `fetchImage`, `uploadImage`, `do`, `checkResponse`, pool space checks |
+| volume.go | 95 | volume naming, `urlImageKey`, `buildVolumeXML` |
+| xml.go | 256 | `domainParams`, `buildDomainXML` and the XML structs |
+| seed.go | 124 | `buildSeedISO` (temp dir, iso9660), `buildVendorData` |
+| check.go | 189 | `Check`, `checkHost`, `probeHost`, `checkReason`, formatting helpers |
+
+Tests (each reviewed with its source file): `*_test.go` next to every file above, plus
+`provider_test.go` (contract tests), `fake_test.go` (fake backend), `sshserver_test.go`
+(in-process SSH server), `kvm2_integration_test.go` (`//go:build integration`, `TestKVM2*`).
+
+Other kvm2-owned bits:
+- `cmd/main.go:62,206` — import and `"kvm2": kvm2.New` registration
+- `Makefile:73-75` — `test-integration-kvm2`
+- `config/samples/proxy_kvm2.yaml`, `config/samples/kustomization.yaml:8`
+- `config/samples/providers-config.yaml:58-`, `config/manager/providers_config.yaml:54-` (commented kvm2 entries)
+
+Reference only, not under review: `docs/kvm2-provider-spec.md`, `docs/kvm2-provider.md`,
+`docs/provider-contract.md`. Reviewers may read them to judge code against stated intent.
+
+## SHARED (on the kvm2 call path; listed symbols only)
+
+Called by kvm2:
+- `internal/provider/config_kvm.go` — `KVMConfig`, `KVMImage`, `KVMMachineType`, `KVMHost`,
+  `KVMFormat*`/`KVMFirmware*` consts, `IsHTTPURL`, `validateKVM`, `validateKVMBlock`
+  (the validator is what stands between user YAML and kvm2's XML/URL/volume-name building)
+- `internal/provider/config.go` — `ProviderConfig.KVM2`, `setBlocks`, the `kvm2` arms of `Config.validate`
+- `internal/provider/provider.go` — `Provider` interface, `CreateRequest`, `Placement`, `Instance`,
+  `InstanceState` + `State*` consts, `LabelManagedYes`
+- `internal/provider/errors.go` — `Error`, `Wrap`, `Class`, `ErrTransient`/`ErrPermanent`/`ErrNotFound`/`ErrQuotaExceeded`
+- `internal/provider/check.go` — `Checker`, `TargetStatus`, `Reason*` consts
+- `internal/provider/name.go` — `NameFromUID` (produces the domain name kvm2 puts in XML and volume names)
+- `internal/tracing` — `StartSpan`, `Tracer`, `Option`
+
+Calling into kvm2:
+- `internal/controller/proxy_controller.go` — `reconcileManaged` (Create/Get/Delete calls at
+  156/197/218), `replaceInstance`, `reconcileDelete`, `providerFailure`, `deletionFailure`,
+  `resolveCloudInit`, `placementFrom`
+- `internal/gc/gc.go` — `Sweeper.sweep` (`ListByTag` at 105, `Delete` at 125)
+- `internal/providerhealth/monitor.go` — `Monitor.checkOnce`, `Monitor.run` (`Check` at 214; the timeout it passes)
+- `internal/provider/registry/registry.go` — `Constructor`, `Build`
+- `internal/provider/metrics.go`, `internal/provider/tracing.go` — `WithMetrics`/`WithTracing`
+  decorators (`instrumented`, `traced`): error passthrough, and that they hide `Checker`
+- `cmd/main.go:196-225` — registry build, checker extraction from unwrapped providers, wrapping
+- `config/manager/manager.yaml:150-178` — `kvm-ssh` Secret mount, `/tmp` emptyDir (64Mi) that `buildSeedISO` uses
+
+## EXCLUDED but considered
+
+- `internal/provider/kvm/` — the older kvm provider; kvm2 is a clean-room rewrite and imports nothing from it.
+- `internal/provider/config_kvm_test.go`, `config_test.go` — reviewed only for kvm2 cases, inside chunk 7; rest is kvm.
+- `internal/provider/gcp/`, `internal/provider/kubernetes/` — other providers, not on a kvm2 path.
+- `internal/health`, `internal/discovery`, `internal/lease` — consume `Proxy` status, never call the provider.
+- `internal/metrics`, `internal/tracing` internals — only the three entry points above are on the path.
+- `internal/providerhealth/envtest_test.go` — mentions kvm2 only as a type string in a fixture.
+- `api/v1alpha1/providerhealth_types.go:71`, `config/crd/bases/...providerhealths.yaml` — comment-only / generated.
+- `go.mod`, `go.sum` — dependency pins (`go-libvirt` 20260814 pseudo-version, `iso9660` v0.4.0, `x/crypto` v0.54.0); noted, not reviewed. Third-party source may be read to understand a call, not reported on.
+- `config/samples/proxy_kvm.yaml`, `Makefile:69-71,185-196` (`test-integration-kvm`, `kvm-ssh-secret`) — kvm-only or shared tooling with no kvm2 logic. `kvm-ssh-secret` is mentioned in chunk 7 for context only.
+- `README.md`, `CHANGELOG.md`, `AGENTS.md`, `.claude/settings.json`, `docs/` — prose / tooling.
+- `.gitea/workflows/build.yaml`, `Dockerfile` — no kvm2-specific content; integration test is not run in CI (worth stating in chunk 7, not a file to review).
+
+## Proposed phase 1 chunking (7 chunks)
+
+1. **Connection layer (SSH + libvirt link)**
+   Files: `ssh.go`, `hosts.go`, `backend.go`, + `ssh_test.go`, `sshserver_test.go`, `hosts_test.go`, `backend_test.go`.
+   SHARED in play: `KVMConfig` (SSHKeyFile, KnownHostsFile, InsecureIgnoreHostKey), `KVMHost`.
+   Focus: connection/goroutine leaks on error paths, dead-link detection and redial races in `hostTable`, budgets vs caller context, keepalive, host key handling.
+
+2. **Domain lifecycle (Create / Delete)**
+   Files: `kvm2.go`, `create.go`, `delete.go`, `id.go`, + `kvm2_test.go`, `create_test.go`, `delete_test.go`, `id_test.go`, `provider_test.go`.
+   SHARED in play: `CreateRequest`, `Placement`, `NameFromUID`; `reconcileManaged`, `replaceInstance`, `reconcileDelete` (how Create/Delete are retried).
+   Focus: partial-create cleanup, idempotency, foreign-domain refusal, half-deleted domains, `keyedLocks` correctness, `waitRunning` timeout/cancellation.
+
+3. **Observation, state and error classification (Get / ListByTag)**
+   Files: `get.go`, `list.go`, `state.go`, `errors.go`, + `get_test.go`, `list_test.go`, `state_test.go`, `errors_test.go`.
+   SHARED in play: `Instance`, `InstanceState`, `Error`/`Wrap`/`Class` and the `Err*` classes; `Sweeper.sweep`, `providerFailure`, `deletionFailure`.
+   Focus: libvirt state mapping (paused/crashed/pmsuspended/shutoff), IP detection and stale addresses, NotFound vs transient, one dead host failing the whole list and what GC does with a partial list.
+
+4. **Base image cache and volumes**
+   Files: `image.go`, `volume.go`, + `image_test.go`, `volume_test.go`.
+   SHARED in play: `KVMImage`, `KVMFormat*`, `IsHTTPURL`, `validateKVMBlock` (image key charset).
+   Focus: download/upload streaming, sha256 verify + marker protocol, cross-replica and cross-Create races on the same base volume, cleanup of partial volumes, URL handling (SSRF, redirects, size caps), pool space check, volume-name injection via URL/key.
+
+5. **XML and cloud-init seed templating**
+   Files: `xml.go`, `seed.go`, `metadata.go`, (`buildVolumeXML` in `volume.go`), + `xml_test.go`, `seed_test.go`, `metadata_test.go`.
+   SHARED in play: `KVMMachineType`, `KVMHost` (bridge, pool), `KVMFirmware*`, `CreateRequest.CloudInit`, `NameFromUID`; `resolveCloudInit`; `manager.yaml` `/tmp` emptyDir.
+   Focus: escaping of config- and user-supplied values, validity across libvirt versions, metadata parse robustness (GC trusts it), temp-dir cleanup and size limit, vendor-data merging.
+
+6. **Health check and observability**
+   Files: `check.go`, `observe.go`, + `check_test.go`, `observe_test.go`.
+   SHARED in play: `Checker`, `TargetStatus`, `Reason*`; `Monitor.checkOnce`/`run`; `tracing.StartSpan`/`Tracer`; `WithMetrics`/`WithTracing` decorators; `cmd/main.go:213-225`.
+   Focus: dedicated check connections closed on every path, check timeout vs dial budget, reason mapping, secrets or cloud-init leaking into logs/spans, span end on every path.
+
+7. **Config, registration, deployment and test gating**
+   Files: `internal/provider/config.go` + `config_kvm.go` (kvm2 arms), their tests' kvm2 cases, `internal/provider/registry/registry.go`, `cmd/main.go:196-225`, `config/manager/manager.yaml:150-178`, `config/manager/providers_config.yaml`, `config/samples/providers-config.yaml`, `config/samples/proxy_kvm2.yaml`, `Makefile:73-75`, `kvm2_integration_test.go`, `fake_test.go`.
+   SHARED in play: `validateKVMBlock`, `setBlocks`, `Config.validate`, `registry.Build`, `Constructor`.
+   Focus: validation gaps that let bad values reach chunks 4/5, `New` behaviour with a nil block or unreadable key, Secret mount perms, integration test gating/cleanup and whether the fake backend diverges from real libvirt in ways that hide bugs.
@@ -0,0 +1,216 @@
+# kvm2 provider review: phase 1 sweep
+
+Seven `sweep-reviewer` runs, one per chunk in `scope.md`, against `feat/kvm2-provider` at `2510842`.
+Findings below are as the sweepers reported them (they state each was verified in code); the
+orchestrator merged duplicates and ranked, and did not re-read code. 54 raw findings, 49 after
+merging. IDs are `S<chunk>.<n>`; "also: chunk N" marks a finding reported independently by
+another sweeper.
+
+Severity counts after dedupe: **high 4, medium 17, low 28**.
+
+Recurring themes across chunks:
+- **T1 Unknown or unmapped errors default to Transient, so permanent misconfigurations retry forever**
+  (S1.4, S2.5, S3.7, S5.2, S7.2, S7.6b) while one genuinely transient case is made Permanent and latches (S5.1, S4.2).
+- **T2 Ownership mark enforced on Create and ListByTag but not on Get/Delete** (S2.3), and the fake cannot test the namespace rule (S7.4).
+- **T3 Base image cache identity and lifetime**: keyed by key+format only, deleted in place under live overlays, no cross-process fencing (S4.6, S7.1, S4.8, escalations E4/E7).
+- **T4 The shared pooled link**: any caller's cancel or budget expiry kills it for everyone, and several callers misread the resulting "connection is closed" error (S1.1, S3.1, S3.6, S4.3).
+
+## Top of the list (high)
+
+| ID | Where | Defect |
+|---|---|---|
+| S1.1 | `backend.go:138-148`, `ssh.go:109-114` | `markDead` closes the transport before signalling death; the close can block ~15 min on a silent host, wedging callers and the host's dial semaphore |
+| S4.1 | `image.go:273` | A reader or server-side error during `StorageVolUpload` wedges go-libvirt's stream sender forever: goroutine + 4 MiB leak per failed upload, 20 min stall |
+| S3.1 | `get.go:104-119` | `address()` swallows every non-ctx error, so a dead link or a guest-agent restart demotes a healthy Running instance to Provisioning and wipes `status.IP` |
+| S7.1 | `volume.go:17`, `image.go:113-146` | Base image cache is keyed by key+format, never URL: two provider entries on one host+pool (the shipped sample does this) silently share one image; a URL edit is a silent no-op |
+
+---
+
+## Chunk 1 — Connection layer (`ssh.go`, `hosts.go`, `backend.go`)
+
+**S1.1 — high — `internal/provider/kvm2/backend.go:138-148` (kill at 143-145, `close(l.closed)` at 146), `ssh.go:109-114`**
+`link.markDead` runs `l.kill()` synchronously inside `once.Do` and only then closes `l.closed`. `kill` is `sshSocket.Close`, which calls `s.Conn.Close()` (x/crypto `chanConn` → `channel.Close` → a raw TCP write under `writeMu`/`t.mu`) before `s.client.Close()`. That write blocks while the send buffer is full.
+Scenario: hypervisor is power-cut or partitioned (no RST/FIN, normal over Tailscale) with data queued. A pooled call hits its 30 s/90 s budget, `call` (`backend.go:183`) invokes `markDead`, which blocks until the kernel gives up (`tcp_retries2`, ~15 min). During that window: the abandoning caller never returns (reconcile ctx has no deadline); `l.closed` is still open so `isDead` reports the backend live and `hostTable.get` keeps handing it out, each new call wedging behind the first in `once.Do`; `sshSocket.closeOnce` memoises the in-progress close so `libvirtBackend.Close`'s 5 s budget cannot bound it; when the blocked `markDead` is the one at `hosts.go:153` it runs while `hostTable.get` holds `e.sem`, blocking every caller for that host.
+Fix direction: close `s.client` (plain `net.Conn.Close`) first and/or close `l.closed` before calling `kill`.
+
+**S1.2 — medium — `ssh.go:119-147`**
+`sshSocket.keepalive`, the only detector of a silently dead pooled connection between calls, has no test. `testSSHServer` already services global requests, so it is testable end to end. A regression is invisible to CI and shows up only as a stale "live" connection that burns a full 30 s budget. Violates the repo rule that all new Go code has tests.
+
+**S1.3 — low — `ssh.go:120`**
+`time.NewTicker(interval)` panics for `interval <= 0`, in a goroutine, which exits the operator. Only `newLibvirtDialer` sets the field today; a future struct literal without it compiles and panics on first successful dial. No guard at either end.
+
+**S1.4 — low — `errors.go:45-51` (dial path from `ssh.go:87`)** [T1]
+`classify` recognises only `*knownhosts.KeyError`. A key matched by an `@revoked` line yields `*knownhosts.RevokedError`, which falls through to Transient: a deliberately revoked hypervisor key is retried forever instead of failing Permanent.
+
+**S1.5 — low — `ssh.go:167-172`**
+`ClientConfig.HostKeyAlgorithms` is left empty. If `known_hosts` pins only the RSA key of a host that also serves ed25519, the server offers ed25519, the callback returns a `KeyError` with non-empty `Want`, and `classify` maps it to Permanent "host key mismatch" although the pinned entry is genuine.
+
+Verified clean: all dial error paths close the TCP conn and SSH client; `call`/`await` orphan goroutines write into buffered channels; the never-closed pooled table is not a leak (Provider built once, no hot reload).
+
+## Chunk 2 — Domain lifecycle (`kvm2.go`, `create.go`, `delete.go`, `id.go`)
+
+**S2.1 — medium — `create.go:107-114`**
+Per-VM volumes created before `DomainDefineXML` are unreachable by both `Delete` and orphan GC when `Create` never returns an ID.
+Scenario: `ensureOverlay` and `writeSeed` succeed, `DomainDefineXML` (139) fails Permanent (or the manager dies between). `Create` returns `("", err)`; `reconcileManaged` never sets `status.ProviderID` and latches. On Proxy deletion, `reconcileDelete` takes the `ProviderID == ""` shortcut (`proxy_controller.go:270-274`) without calling `Delete`. `ListByTag` enumerates domains only, so `proxy-<x>.qcow2` (10 GiB capacity) and `proxy-<x>-seed.iso` stay in the pool permanently.
+
+**S2.2 — medium — `create.go:230-231`, `state.go:16-27`, `get.go:56`** (also: chunk 3)
+Paused domains are destroyed. `ensureRunning` refuses a paused/pmsuspended domain with a Permanent "virsh resume it" error, but that guard only fires before an ID is persisted. In steady state the reconciler calls `Get`; `get.go:56` discards the `DomainGetState` reason and `mapState` maps paused/pmsuspended to `StateStopped`, which sends `reconcileManaged` (`proxy_controller.go:217-225`) to `prov.Delete`: destroy, undefine, delete disks, recreate.
+Scenario: QEMU pauses the guest on an I/O error because the pool is full (`VIR_DOMAIN_PAUSED_IOERROR`), or an admin runs `virsh suspend`/`save`/snapshot/live-migrate. At the next drift poll the VM is killed; in the ENOSPC case the recreate refills the pool or fails the space precheck.
+
+**S2.3 — medium — `delete.go:56-63`, `get.go:52-73`** (also: chunk 3) [T2]
+`Get` and `Delete` do not enforce the kvm2 mark that `Create` (`create.go:80-82`) enforces. `readMetadata` returns the zero value for an unmarked domain and `get` builds an Instance anyway (`UID: ""`); `removeDomain`/`removeVolumes` destroy whatever the name resolves to. `internal/provider/kvm/id.go:11-13` emits byte-identical `hosts/<host>/domains/<name>` IDs and both providers use `NameFromUID`.
+Scenario (the documented migration path, `docs/kvm2-provider.md:346-370`, done out of order): an admin re-types an entry from `kvm` to `kvm2` keeping its name without first deleting every Proxy. `status.ProviderID` survives; kvm v1's mark is in another namespace, so `Get` returns the v1 domain as this Proxy's instance; if it is anything but Running the controller calls `Delete` and kvm2 destroys a kvm-v1 domain the docs promise it can never touch. Same exposure for a hand-edited `status.ProviderID`. `TestGet_unmanagedDomainHasEmptyUID` pins the current behaviour. `ListByTag` is correctly mark-filtered, so GC is not affected.
+
+**S2.4 — low — `create.go:87`**
+The adoption path starts an already-defined domain without checking its disks exist and cannot heal them. If the overlay or seed is gone and no ID was persisted, every retry skips the volume steps, `DomainCreate` fails with a storage error classified Transient, and the reconciler retries forever; the drift path that would heal it is unreachable without an ID.
+
+**S2.5 — low — `delete.go:15`** [T1]
+`undefineFlags` omits `DomainUndefineCheckpointsMetadata`. An admin- or backup-tool-created checkpoint makes `DomainUndefineFlags` fail Transient; `deletionFailure` retries forever and the Proxy cannot be deleted without manual `virsh checkpoint-delete`.
+
+Verified clean: `keyedLocks` refcounting, cancel-while-waiting and unlock on every path; `waitRunning` cap and cancellation; providerID round trip.
+
+## Chunk 3 — Observation, state, errors (`get.go`, `list.go`, `state.go`, `errors.go`)
+
+**S3.1 — high — `get.go:104-119`** [T4]
+`address()` treats every non-`context` error from both address sources as "skip" and returns `("", nil)`.
+Triggers: (a) routine — on beelink the ARP source never answers (`docs/kvm2-provider-spec.md:146`), so a `qemu-guest-agent` restart in the guest returns `ErrAgentUnresponsive` → `""` → `mapState(running, "")` = Provisioning; (b) racy — any concurrent call that exceeds its budget kills the shared pooled link (`backend.go:180-184`), after which `call` returns `transientf("... connection is closed")` with no ctx sentinel, which `get.go:108` does not recognise; if the link dies between `DomainGetState` and the address read, `Get` succeeds with `State=Provisioning, IP=""`.
+Downstream: `proxy_controller.go:208-211` clears `status.IP` and sets `Provisioned=False`; `internal/health/engine.go:200-207` prunes the proxy's health state and the Healthy counters restart from zero. One RPC hiccup pulls a working proxy out of rotation. `TestGet_addressErrorIsProvisioning` pins the behaviour.
+
+**S3.2 — medium — `state.go:35-49`**
+IP detection has no interface selection. The domain has exactly one bridge NIC and libvirt returns `Hwaddr` per interface, but `firstIPv4` filters only loopback/link-local/unspecified.
+Scenario: the image or cloud-init brings up `docker0` (172.17.0.1), `tailscale0` or `wg0` before the DHCP lease lands and the agent reports it first. `Get` returns Running with a non-routable IP, the controller latches `Provisioned=True`, and the value is never corrected because any IPv4 keeps the state Running.
+
+**S3.3 — medium — `list.go:38-45`** [T4]
+A pooled link that dies mid-list makes `ListByTag` a silently successful partial list. The fail-whole guard is `ctx.Err() != nil`, but a killed link yields "connection is closed" with no ctx sentinel; the remaining domains each log "skipping domain with unreadable metadata" and the call returns nil error with "managed=5" as if authoritative. GC only under-deletes, so no wrong deletion, but nothing marks the result partial. A Permanent corrupt-mark parse error is likewise skipped forever with only a V(1) line.
+
+**S3.4 — low — `get.go:104-119`**
+`address()` also swallows `ErrNoDomain`: a domain that vanishes between `DomainGetState` and the address read is reported as live Provisioning for one poll. The comment "only ctx ending is an error" is also inaccurate: a 30 s budget expiry on the agent source wraps `DeadlineExceeded`, fails `Get`, and never falls back to ARP.
+
+**S3.5 — low — `errors.go:58-71`** [T1] (also: chunks 5, 7)
+Version and capability errors escape classification. go-libvirt turns an "unknown procedure" reply into the bare sentinel `libvirt.ErrUnsupported` (not a `libvirt.Error`), so `libvirtCode` misses it and it becomes Transient, while the same condition as `VIR_ERR_NO_SUPPORT` is Permanent. `ErrConfigUnsupported` (67) and `ErrOperationUnsupported` (84) are also unmapped. `Check` records `libvirtVersion` but gates nothing on a minimum. See S5.2 and escalation E9.
+
+## Chunk 4 — Base image cache and volumes (`image.go`, `volume.go`)
+
+**S4.1 — high — `image.go:273`**
+`uploadImage` hands a `TeeReader` over `resp.Body` straight to `StorageVolUpload`. In go-libvirt (`rpc.go:290-308`, `socket/socket.go:347-376`) `abort` and `outErr` are unbuffered: on a reader error `SendStream` sends a `StatusError` packet then blocks forever on `outErr <-`, and `requestStream` blocks forever on `abort <- true`. `StorageVolUpload` never returns.
+Scenario: mirror truncates the body (or libvirtd errors mid-upload, e.g. ENOSPC). kvm2's `call` rescues the caller only after `budgets.transfer` = 20 min, and `markDead`/socket close does not unblock either goroutine (contrary to the comment at `backend.go:152-156`). Consequences: 1-2 goroutines + 4 MiB buffer leaked per failed upload, accumulating under reconcile retries against a flaky mirror; a truncated download reported after 20 min instead of at once; `discardVolume` (`image.go:292`) deletes the base on the pooled connection before the deferred `fresh.Close()` (`image.go:265`), so the file is unlinked while the server-side stream holds the fd. `TestEnsureImage_shortReadRemovesVolumeAndIsTransient` gives false confidence: the fake returns the reader error immediately and never exercises the abort protocol.
+Fix direction: have `countingReader` convert errors to `io.EOF` and report them out of band; the existing `readErr`/`read != size` checks still catch truncation.
+
+**S4.2 — medium — `image.go:434-439`, `391-396`** [T1]
+A `HEAD` that answers 2xx with `Content-Length: 0` or none (Go's own `net/http` does this; several CDNs too) becomes `ErrPermanent`, which latches `ReasonPermanentError` for the generation. The GET fallback fires only on non-2xx HEAD, although GET on the same URL would work. The tests never hit it because `serveBytes` sets Content-Length on HEAD.
+
+**S4.3 — medium — `image.go:235-245`** [T4]
+A completed upload is discarded when the pooled connection died during it. The base is uploaded on a dedicated connection, but `StoragePoolRefresh` and `writeMarker` then run on the pooled `be`, which any other reconcile's cancel can kill. On failure `discardVolume` deletes the fully downloaded, digest-verified base and the next Create re-downloads. The window is as wide as the download. The dedicated connection is still open at that point and could carry the refresh and marker.
+
+**S4.4 — medium — `image.go:419-443`, `226-230`, `create.go:96-106`**
+The pool-space precheck uses the **HEAD** Content-Length; `fetchImage` declares the volume capacity from the **GET** Content-Length and never re-checks. There is no `maxImageBytes` cap.
+Scenario: a server answers HEAD with 1 MB and GET with 4 TiB; the precheck passes, a sparse 4 TiB qcow2 is created and the stream fills the host pool until ENOSPC, degrading every VM on the hypervisor. Benign variant: a mirror rotates a build between HEAD and GET.
+
+**S4.5 — medium — `image.go:59-65`**
+`placement.image` is an unvalidated user-supplied URL fetched by the operator. `PlacementSpec.Image` has no CRD validation; `IsHTTPURL` checks only scheme and non-empty host; `http.DefaultClient` follows redirects with no policy. Anyone who can create a Proxy in a watched namespace makes the operator issue HEAD/GET from its network position (cluster services, `169.254.169.254`) and makes the hypervisor boot the returned bytes on the egress bridge. URL images can never carry a `sha256`.
+
+**S4.6 — medium — `image.go:118`, `image.go:145`** (also: chunks 2, 7) [T3]
+Any Create can evict a base volume that live overlays back onto, with no in-use check and only a V(1) log. `cachedImage` runs on every Create and deletes the base when the marker is missing, unreadable, non-hex, or when `src.SHA256` differs, then `fetchImage` recreates the same path with different content.
+Scenario: an admin adds or updates `images.<key>.sha256` while Proxies exist and the mirror has republished. Running qemu processes keep the old inode; every later boot of an existing overlay reads a different base. The spec records this as a doc warning ("add a new key instead"); the kvm sample carries the warning (`config/samples/providers-config.yaml:40`) but the kvm2 block (`:72`) omits it. See escalation E7.
+
+**S4.7 — low — `image.go:62`**
+A URL image is always assumed qcow2/uefi. A raw image given as `placement.image` is cached as `base-<key>.qcow2` with `BackingFormat: "qcow2"`; qemu rejects the chain or the VM fails to boot, and the diagnostic points at the domain.
+
+**S4.8 — low — `volume.go:22-25`** [T3]
+`urlImageKey` truncates sha256 to 64 bits and shares a namespace with catalogue keys. Two colliding URLs share one base (a ~2^32 birthday search for someone who can create Proxies), and URL images have no digest to catch it. A catalogue key that is 16 lowercase hex chars occupies the same `base-<key>.<format>` name as a URL image.
+
+Verified clean: HTTP bodies and the dedicated upload connection are closed on every path; `cappedWriter` and `countingReader` are correct as written.
+
+## Chunk 5 — XML and cloud-init seed (`xml.go`, `seed.go`, `metadata.go`, `buildVolumeXML`)
+
+**No injection defect found.** Every config- and user-influenced value reaching domain, volume or metadata XML goes through `encoding/xml` struct marshalling; there is no string concatenation into markup; `Name` is `NameFromUID`'s base32 hash. Chunks 2 and 7 reached the same conclusion independently.
+
+**S5.1 — medium — `create.go:199-201`, `seed.go:66-74`** [T1, inverted]
+Every I/O failure inside `buildSeedISO` is turned into `ErrPermanent` by `permanentf("building seed ISO for %s: %v", ...)`: `os.MkdirTemp` when `/tmp` is missing, ENOSPC during `AddFile`/`WriteTo`, and via the deferred `errors.Join` even a `Cleanup()`/`RemoveAll` failure on an otherwise successful build. `providerFailure` then returns `ctrl.Result{}` with nil error: no requeue at all until a spec edit or the ~10 h resync. The `%v` also discards the wrapped error. The empty-name guard is correctly permanent.
+
+**S5.2 — medium — `xml.go:26-27`, `194-206`** (also: chunk 7) [T1]
+`x86_64`, `q35`, `type="kvm"`, `host-passthrough`, `<os firmware="efi">` (libvirt ≥ 5.2.0) and `<firmware><feature name="secure-boot">` (≥ 7.2.0) are unconditional with no capability or version gate, and no config field or validator mentions architecture.
+Scenario: an aarch64 hypervisor or a libvirt 7.0 host is added to `kvm2.hosts`; `Check` reports Ready; every Create there fails at `DomainDefineXML` with code 67 or 84, which `classifyLibvirt` leaves Transient, so the reconciler retries an impossible define indefinitely with no actionable condition.
+
+**S5.3 — low — `metadata.go:44-50`** (see escalation E2)
+An unparsable `<createdAt>` returns `time.Time{}`; `gc.go:120` computes `now.Sub(CreatedAt) < MinAge`, so a mangled mark is immediately past the grace period, exactly the case MinAge protects. Mirror case: a mark written under a forward-skewed clock yields a negative age that is always `< MinAge`, so the domain is never collected. Only the first direction is mentioned in the doc comment.
+
+**S5.4 — low — `seed.go:48-53`**
+The guest-agent script's non-systemd fallback runs `rc-update` unguarded (the kvm provider guards it with `elif command -v rc-update`). With `set -e`, an image with neither exits 127 after the package install; the agent is installed but never started and discovery silently falls back to ARP only.
+
+**S5.5 — low — `xml.go:178-193`**
+`buildDomainXML` validates name, paths and disk format but not `VCPUs`/`MemoryMiB`; it will emit `<vcpu>0</vcpu>`. Unreachable today because `validateKVMBlock` checks them; latent for any future caller.
+
+**S5.6 — low — `volume.go:68-89`** (see escalation E10)
+No check that an overlay's capacity is at least the backing image's virtual size. `kvm2.diskSizeGb: 2` against a 3.5 GiB Ubuntu image produces `<capacity>` smaller than `<backingStore>`.
+
+**S5.7 — low — `internal/provider/config_kvm.go:186-188`**
+`bridge` and `pool` are checked non-empty only, unlike host names and image keys (`kvmIdentifier`). A value with whitespace or markup is faithfully escaped into the XML and surfaces as an opaque libvirt failure at Create time rather than a config-load error.
+
+**S5.8 — low — `seed.go:107-108`**
+A configured `vendorData` that is itself MIME multipart (or starts with any unrecognised line) is nested as `text/plain`, matches no cloud-init prefix, and is silently dropped when `installGuestAgent` is on. No guard in `buildVendorData`.
+
+Verified clean: the iso9660 staging dir is cleaned on every return path; volume XML sets no `<permissions>`, so the seed ISO gets libvirt's 0600 default.
+
+## Chunk 6 — Health check and observability (`check.go`, `observe.go`)
+
+In good shape: the dedicated check connection is released on every path, every seam method is wrapped and forwards correctly (enforced by reflection in `observe_test.go:77`), spans always end, `liveness` survives the decorator, and no cloud-init or key material reaches a log or span at any verbosity. `go vet` and `go test -race` on the chunk were clean.
+
+**S6.1 — low — `check.go:90`, `backend.go:162`, `:351`**
+`call()` returns at its pre-flight `ctx.Err()` check without `markDead`, so if the round deadline falls between two probe RPCs the link is still live and the deferred `Close()` sends `ConnectClose` on a ctx-less 5 s budget. `Check` can work past the deadline the Monitor gave it, against `provider.Checker`'s "must return promptly once ctx ends". Bounded at 5 s, narrow window.
+
+**S6.2 — low — `check.go:57-58`**
+Reported `Latency` includes the teardown RPC because `defer b.Close()` runs inside `probeHost` before `time.Since(start)`. About one RTT slower than the spec's cost model.
+
+**S6.3 — low — `check.go:66`**
+The round-wide `ctx.Err()` test masks an actionable per-host reason. Host A blackholes and eats the round; host B fails at the deadline with "no route to host" and is reported `Timeout`. Same for `PoolNotActive`/`CredentialsUnavailable` landing on the deadline.
+
+**S6.4 — low — `check.go:135-148` vs `internal/providerhealth/monitor.go:113`**
+The Unreachable/Timeout distinction only holds while the round timeout exceeds the 10 s dial budget. `ValidateSchedule` accepts `--provider-check-interval=10s` or `--provider-check-timeout=10s`, under which a blackholed host reports `Timeout` instead of `Unreachable`.
+
+**S6.5 — low — `check.go:101` → `api/v1alpha1/providerhealth_types.go:65`**
+The remote-controlled `hostname` is copied unbounded into ProviderHealth `details`; `Message` is truncated to 1024 but `details` has no `maxLength`/`maxProperties`. A multi-MB hostname makes every status patch fail "too large". Requires a broken or hostile hypervisor.
+
+**S6.6 — low — `check.go:41-45`**
+Unbounded fan-out: one goroutine, key re-read and full SSH handshake per host per round, no cap. Harmless at 2-5 hosts.
+
+## Chunk 7 — Config, registration, deployment, test gating
+
+**S7.1 — high — `internal/provider/kvm2/volume.go:17`, `image.go:113-146`** (also: chunk 4) [T3]
+The base image cache is keyed by image key + format only. `baseVolumeName` → `base-<key>.<format>`, and `internal/provider/kvm/image.go:46` produces the identical name. The shipped sample has `kvm-home` (`config/samples/providers-config.yaml:41`) and `kvm2-home` (`:73`) both with key `debian-13`, host `beelink`, pool `default`, and `:58-60` tells the admin to run both while porting.
+Scenario: the two entries' URLs drift and neither sets `sha256`; `cachedImage` accepts any 64-hex marker when `src.SHA256 == ""`, so whichever provider downloaded first owns the file and the other boots the wrong image. Same for two `kvm2` entries naming one physical host. Related: changing `images.<key>.url` without changing the key or `sha256` is a silent no-op, since neither the volume name nor the marker records the URL.
+
+**S7.2 — medium — `internal/provider/config_kvm.go:180-182`** [T1]
+`KVMHost.Address` is validated only as non-empty although it must be `host:port`; nothing applies a default port. `address: beelink` passes `LoadConfig`; `dialSSHSocket` (`ssh.go:163`) gets "missing port in address", `classify` falls through to Transient, every operation retries forever, and `checkReason` reports `Unreachable`, pointing at the hypervisor rather than the config.
+
+**S7.3 — medium — `internal/provider/config_kvm.go:101-104`, `142-147`**
+`IsHTTPURL` accepts plain `http://` and `sha256` stays optional, so a VM root disk can come from an unauthenticated channel; `writeMarker` then pins whatever digest arrived. Nothing warns.
+
+**S7.4 — medium — `internal/provider/kvm2/fake_test.go:362-376`, `:259-282`** [T2]
+The fake's `DomainGetMetadata` ignores the namespace of the stored element, returning it whenever `uri[0] == metadataURI`; real libvirt selects strictly by namespace. So no unit test can build a domain carrying the kvm provider's `<instance>` and assert `list.go:38-48` skips it. If that filter regressed, kvm2's `ListByTag` would hand kvm's VMs to orphan GC and the suite would stay green.
+
+**S7.5 — medium — `Makefile:75`**
+`-timeout 40m` is not derived from `KVM2_TEST_TIMEOUT`. Two integration tests each take up to 20 m plus a 5 m cleanup budget; with `KVM2_TEST_TIMEOUT=40m` or a cold image download, go test's own timeout panics before `t.Cleanup` (`kvm2_integration_test.go:373-382`) runs, leaving a domain, overlay and seed on the real hypervisor under a random name printed only mid-run.
+
+**S7.6 — low — `internal/provider/config_kvm.go:133`, `internal/provider/kvm2/image.go:466`** [T1]
+No upper bound on `diskSizeGb`, and `neededBytes` wraps: `diskSizeGb: 17179869184` (2^34) shifts to 0, so `checkPoolSpace` passes and the failure surfaces later as "capacity must be positive". `machineTypes` vcpus/memoryMiB have no upper bound either (`config_kvm.go:158-162`); `memoryMiB: 1024000` yields an internal libvirt error that is Transient and retried forever.
+
+**S7.7 — low — `internal/provider/kvm2/fake_test.go:190-208`**
+The fake's `enter()` accepts calls on an expired ctx and never marks the connection dead, unlike `call()`. So `discardVolume`'s `context.WithoutCancel` (`image.go:331`) is not exercised: deleting it leaves every test green while production skips cleanup of a half-uploaded base on a cancelled Create.
+
+**S7.8 — low — `internal/provider/kvm2/fake_test.go:579-584`**
+The fake's `Close()` only bumps a counter; it stays usable and `isDead()` stays false. A use-after-close of a dedicated connection passes in tests and fails in production.
+
+**S7.9 — low — `Makefile:60-75`, `.gitea/workflows/build.yaml:27-32`**
+Nothing compiles the integration file: `go vet`/`go build`/`go test -short`/`make test` all run without `-tags=integration`. It compiles today (`go vet -tags=integration ./internal/provider/kvm2/` exits 0), but nothing keeps it that way.
+
+**S7.10 — low — `internal/provider/kvm2/image.go:228`, `283-289`, `385-395`**
+The full image URL reaches log lines and error messages while the span deliberately carries only the host (`urlHost`). `IsHTTPURL` accepts userinfo, so `http://user:pass@mirror/...` lands in the error that `provider.Wrap` propagates. Whether it reaches Proxy status is unverified (dropped escalation D5).
+
+**S7.11 — low — `internal/provider/kvm2/kvm2_integration_test.go:89`, `:307`**
+`KVM2_TEST_KEEP=0` or `false` enables keep, since the check is `!= ""`. And `if elapsed > 100*time.Millisecond` on `New()` is a wall-clock `t.Errorf` on a possibly loaded machine, a flaky failure at the head of a 20 minute test.
+
+**S7.12 — low/informational — `config/manager/manager.yaml:175-178`, `internal/provider/kvm2/seed.go:56-74`**
+The seed ISO staging dir is the `/tmp` emptyDir with no `medium: Memory`, so Proxy cloud-init payloads (which can carry proxy credentials) are written unencrypted to node-local disk for the duration of each Create. The mount comment (`manager.yaml:163-165`) attributes `/tmp` to the kvm provider only.
+
+Verified clean: Secret mount permissions (x/crypto does no permission check); both `Close` defers on dedicated connections.