docs: CHANGELOG v0.113.0 + REPORT + CONTEXT + README NAS section + REUSE seams

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PSK5g6qYLknKj8u3QAFEr6
This commit is contained in:
2026-07-11 10:31:07 +02:00
parent a65dcff85a
commit 5d06ecf374
5 changed files with 186 additions and 58 deletions
+112 -57
View File
@@ -1,68 +1,123 @@
# REPORT — v0.112.0: self-update without credentials (anonymous registry mode)
# REPORT — v0.113.0: NAS verify-before-commit + page redesign + protocol-honest guidance
**Date:** 2026-07-10 · **Version:** v0.112.0 (from v0.111.0) · **Pairs with:** hub v0.43.1
**Spec:** TASK — controller self-update without credentials (anonymous registry mode)
**Date:** 2026-07-11 · **Version:** controller v0.113.0 (from v0.112.0) · **Pairs with:** agent
v0.81.0 + felhom.eu host-install v1.13.0
**Spec:** TASK — NAS verify-before-commit + page redesign + protocol-honest guidance
**Evidence base:** `felhom.eu/documentation/audits/SPIKE-nas-verify-2026-07-11.md` (b57f6c1)
## Root cause (live on Peti's box)
## Confirmed baselines → shipped
The updater piggybacked on the Git Sync credentials and REFUSED when absent (`queryRegistry` +
`pullImage` guards) — but the registry serves the public package anonymously (Docker v2 token dance,
verified empirically 2026-07-10). A fresh customer without a private catalog silently lost version
discovery + self-update for no reason. Credentials are now what they were meant to be: **optional,
private-catalog only**.
| Repo | Baseline (`main`) | Shipped commits | Version |
|---|---|---|---|
| felhom-agent | `300f06722b` (v0.80.0) | `added9d` | **v0.81.0** — deployed on felhom-pve |
| felhom-controller | `3db9126121` (v0.112.0) | `bb8737a` (orchestration) + `a65dcff` (UI) + this docs commit | **v0.113.0** — live on 9201 |
| felhom.eu | `e80e14d6` | `27e2fb0` | host-install **v1.13.0** + feature doc |
## What shipped
## What shipped (files)
- **Part 1 — `queryRegistry` anonymous mode** (`internal/selfupdate/updater.go`): both creds empty →
plain GET; on 401 parse the `WWW-Authenticate` Bearer challenge (realm + service **from the header**
— never hardcoded; quoted/bare values, any order, commas-in-quotes handled); GET the realm with
`service` + `repository:<image>:pull` scope and NO credentials; retry tags/list with the Bearer.
Creds present → the BasicAuth path unchanged. Half-configured pair → loud
"hiányos registry hitelesítő adatok". A genuinely-denying registry →
**"registry denied anonymous access — a private registry requires Git Sync credentials"** (never the
old "credentials missing"). Registry base URL now derives from the image ref (was a hardcoded host).
- **Part 2 — `pullImage` without login:** no creds → the `docker login` step skipped entirely
(docker's native anonymous flow covers public packages); creds → login/pull/logout unchanged
(token still stdin-only). The old refuse-guard removed; a denied anonymous pull surfaces docker's
own error.
- **Part 3 — settings truthfulness:** "Verzió és frissítés" gains
"Registry: nyilvános (hitelesítés nélkül)" / "Registry: hitelesített"; credential-less is no longer
an error state; the Hiba row appears only on real failure. `DryRun.PullCapable` counts anonymous as
capable (false only on a half-configured pair).
- **Part 4 (hub v0.43.1):** Git Sync form hint — "Opcionális — csak privát alkalmazás-katalógushoz.
A verziófrissítés enélkül is működik." (one template string).
**Agent (`added9d`)**`internal/storage/netmount.go` (NFS `retry=0` + exported
`NetworkMountedAt`/`NetworkEndpointReachable`), NEW `internal/storage/netverify.go`
(`ClassifyNetVerifyFailure`, Q4-verbatim table), NEW `internal/localapi/netverifyjob.go` (in-memory
single-slot detached verify + auto-rollback + `GET /netstorage/verify-status`),
`internal/localapi/netstorage.go` (sync fast-fail: full validation + 2 s TCP pre-probe before ANY
install; verify-job start), `server.go` (seams + route). **No new sudoers grants** — journal read is
unprivileged via the `systemd-journal` group.
## Tests + red-proof (executed; failures verbatim)
**Controller (`bb8737a` + `a65dcff`)**`internal/agentapi/client.go` (verify fields, typed
`NetAddRefusedError`, `NetVerifyStatus`), NEW `internal/web/netprobe.go` + `netprobe_linux.go` +
`netprobe_other.go` (uid-1000 re-exec probe; `--netprobe` hidden mode in `cmd/controller/main.go`),
NEW `internal/web/netstorage_job.go` (detached single-flight orchestration + §3.2 Hungarian map),
`netstorage_handlers.go` (job-start add + status endpoint + orphan rows + `netListFn` seam),
`storage_handlers.go` (route), `templates/storage_network.html` (full redesign).
- **Red-proof** — the old creds-required guard restored in `queryRegistry` → all three anonymous tests
FAILED with the old message visible:
`anonymous queryRegistry failed: registry hitelesítő adatok hiányoznak`. Restored → green.
- Non-hollow coverage (`registry_anon_test.go`, httptest fake registry + fake CLI runner): full token
dance with ZERO creds (token request carries NO Authorization, correct
`service`+`scope=repository:admin/felhom-controller:pull`, highest semver returned, non-semver tags
skipped); creds path unchanged (BasicAuth on tags/list, token endpoint never hit); BOTH denial paths
(token 401, tags-401-with-Bearer) → the new clear error and NOT the old one; `parseWWWAuthenticate`
table (quoted/bare/order/comma-inside-quotes/missing-realm/non-Bearer/empty); pull with no creds →
**no login/logout invocation recorded**, exactly one `docker pull`; creds → login→pull→logout order +
token via stdin; partial creds refuse everywhere with zero docker commands run.
- Green gate both repos (`go build && go vet && go test ./...`) + controller template gates
(`template_id_gate.py`, `emoji_gate.py`) pass.
**felhom.eu (`27e2fb0`)**`scripts/felhom-host-install.sh` v1.13.0 (`usermod -aG systemd-journal
felhom-agent`, idempotent; header/const drift v1.11.0-vs-1.12.0 fixed), NEW
`documentation/controller/network-storage-nas.md` (authoritative feature doc).
## Deploy + live validation (demo, credential-less for real)
## Tests + companion red-proofs (all: mutation run → FAIL observed → reverted → suite green)
- Controller `0.112.0` → 9201 (bootstrap mechanism), `Up (healthy)`. Hub `0.43.1` → GitOps sync,
Synced/Healthy, rollout complete.
- The demo box's `controller.yaml` has `git.username: ""` / `git.token: ""` (verified quoted-empty →
parses to empty strings) — a genuinely credential-less box.
- **Live anonymous check against the real registry:** `POST /api/selfupdate/check` (in-guest, via the
container IP + Host header, hub-API-key bearer) →
`{"ok":true, "current_version":"0.112.0", "latest_version":"0.112.0", "update_available":false}`
NO error. Pre-0.112.0 this exact call failed with "registry hitelesítő adatok hiányoznak".
- **Settings page live:** renders `Registry — nyilvános (hitelesítés nélkül)` and
`Legújabb verzió — 0.112.0`; no Hiba row.
Test funcs added: agent storage +3 (netverify_test.go incl. the pure `networkMountedIn` table),
agent localapi +5 (netverifyjob_test.go), controller web +9 (netstorage_job_test.go +
storage_network_template_test.go).
## Pending operator step (per spec — Peti's box untouched by this task)
| # | Test | Red-proof mutant | Observed failure (verbatim core) |
|---|---|---|---|
| A1 | `TestMountOptions_NFSRetry0_SMBWithout` (+ NFS rendering test) | reverted `retry=0` | `NFS options missing retry=0 (Q4-vi): "vers=4.1,...,_netdev"` |
| A2 | `TestClassifyNetVerifyFailure` (Q4-verbatim table incl. merged `nfs_export` + empty-journal degradation) | exit-code classifier (`return mount_failed` — everything is rc=32) | every non-generic row: `= "mount_failed", want "unreachable"/...` — an exit-code mutant cannot split smb_auth/smb_share |
| A3 | `TestNetVerify_MountFailed_RollsBackAndClassifies` (+ journal-unavailable sibling) | rollback call dropped | `RemoveNetworkMount not called for the failed install: removed=[]` + `creds file must be removed` |
| A4 | `TestNetVerify_TruthTable` (§8: ReadDir-ok+unmounted=FAIL; EACCES+mounted=OK) | readability-based verdict (`terr == nil`) | BOTH rows failed: `a readable-but-unmounted path must FAIL verify, got done` / `unreadable-but-mounted must PASS, got failed` |
| A5 | `TestNetVerify_UnreachablePreProbe_InstallsNothing` | pre-probe skipped | `unreachable add: got 200 want 502` (the install would have run) |
| A6 | `TestNetVerify_SingleFlight_AndNoJobShape` | running-check dropped | `second add while verifying: got 200 want 409` |
| C1 | `TestNetAdd_HappyPath_RegisterOnlyAfterProbe` | register moved BEFORE the probe | `phase = failed (category=register_failed detail=... already registered)` + C2's `must NOT be registered (got 1 paths)` |
| C2 | `TestNetAdd_ProbeFail_RollsBackNotRegistered` | rollback dropped | `probe-fail must roll the agent install back: removes=[]` |
| C3 | `TestNetAdd_AgentVerifyFailed_MappedMessage` | (kill-surface shared with C4's mutant) | asserts the EXACT §3.2 merged nfs_export string + probe-not-run + no controller double-remove |
| C4 | `TestNetAdd_VerifyLost_RollsBack` | `none` treated as success | `probe must not run when the verify was lost` |
| C5 | `TestNetProbeChild` (ok / unwritable→2→not_writable / nonce-tamper→3→probe_io / cleanup-fail→4=OK+warn) | — (pure child body + verdict table; Credential wiring live-validated in Scenario C) | — |
| C6 | `TestNetAdd_SingleFlight` | acquire check dropped | `second add: got 200 want 409` |
| C7 | `TestNetStorage_OrphanRow` | orphan sweep disabled (registry-only filter) | `items = 1, want 2 (registered + orphan)` |
| C8 | `TestStorageNetworkTemplate_CanonicalClasses` (renders through the PRODUCTION template tree) | `form-input` reintroduced | `rendered page still contains the banned pattern "form-input"` |
After Viktor's floor bump rolls Peti to 0.112.0: delete the temporary Git Sync creds from Peti's hub
config → next config pull → Peti's settings page should show a clean "Registry: nyilvános
(hitelesítés nélkül)" with zero credentials on the box.
Green gates: agent `go build && go vet && go test ./...` PASS (one hit of the KNOWN
`TestGenerateRecoveryCode` wordlist flake — clean on re-run, pre-documented); controller full suite
PASS; `template_id_gate.py` + `emoji_gate.py` green.
## Deployed + verified
- **Agent v0.81.0** on felhom-pve: `.bak-0.80.0` kept; `usermod -aG systemd-journal felhom-agent`
applied BEFORE restart; `felhom-agent --version` → 0.81.0; `id felhom-agent`
`groups=990(felhom-agent),999(systemd-journal)`; journal shows a clean start (enrolled drive
bound under shared parent, local-api listening, hub desired-state gen 10, no capability
degradation).
- **Controller v0.113.0** on 9201 (golden/bootstrap mechanism): `docker ps`
`gitea.dooplex.hu/admin/felhom-controller:0.113.0 Up (healthy)`.
## Live validation (anti-F9: the exact endpoint the UI invokes)
Method: `curl` from inside guest 9201 to the controller container
(`http://172.17.0.2:8080` + `Host: felhom.demo-felhom.eu` — the dashboard is host-routed; the demo
has no password so auth/CSRF do not apply) — POST `/api/storage/netstorage/add` + status-poll, the
byte-identical server pipeline behind the UI; the residual is client-side rendering only.
Sim NAS = isolated `/srv/nas-spike3/*` on DooPlex (`.bak-nasspike3` backups; pre-counts exports=2,
smb_sections=6; NO iptables — the unreachable case used the ping+neigh-verified-unused
192.168.0.199). Throwaway users `spike3b` (uid 1060) + `spike3smb`; passwords never committed.
| Scenario | Result | Key evidence |
|---|---|---|
| **A** bogus export on a reachable server (THE bug) | **PASS** — failed in **4 s**, category `nfs_export`, the exact merged Hungarian message, the REAL journal in detail (`reason given by server: No such file or directory` — the unprivileged journal read works), list EMPTY, **zero spike3 units left on felhom-pve** — the "Készenlét forever" behavior is dead |
| **B** SMB wrong password / wrong share | **PASS**`smb_auth` („Hibás SMB felhasználónév vagy jelszó.") vs `smb_share` (distinct message), both 4 s, **creds file gone after each**, no units |
| **C** squash trap (`anonuid=1000`, NO all_squash — mounts fine) | **PASS** — agent verify passed → **uid-1000 probe refused**`not_writable` with the Route-A message incl. the computed **101000**; full rollback (no units/mounts/registration); **no probe file** left on the export |
| **D** happy paths | **PASS** — d1 (NFS anonuid=101000), d2 (NFS **Route A**, alien uid 1060), d3 (SMB plain user, no force user): all `done` in **4 s**, registered Schedulable, health `ok`. d2 production proof: the installed unit carries `retry=0`; a uid-1000 write shows guest-view `65534:65534` and lands **server-side `1060:1060`** |
| **E** unreachable IP | **PASS** — failed in **2.0 s** (the agent's sync TCP pre-probe), category `unreachable`, message names the IP, **nothing installed** |
| **G** single-flight | unit-proven (A6 + C6); not re-proven live — the 4 s happy windows make a live race impractical |
**Teardown verified:** d1/d2/d3 removed via the real remove endpoint (list empty); 180 restored from
`.bak-nasspike3` (post-counts exports=2, smb_sections=6, 0 spike3 exports, both users deleted,
scratch gone); felhom-pve: 0 spike3 mounts/units/creds, mountpoint dirs removed,
`systemctl reset-failed` cleared the residual failed-unit listings; guest 9201: 0 spike3 mounts,
controller healthy.
## NOT yet live-validated
- Guest-restart trigger survival (Q1c follow-up — needs a restart window; no guest restart allowed).
- Synology/QNAP appliance pass (virtual-dsm) — the sim-vs-real caveat stands.
- Peti rollout: floor bump + his `usermod -aG systemd-journal felhom-agent` step — when the NAS
feature reaches him. STOP honored: no publish train (agent 0.81.0 NOT published to Gitea, Day-0
manifest untouched), Peti's box untouched, guest 9201 never restarted.
- Scenario F live (agent restart mid-verify) — covered by unit tests A6/C4 + the orphan surface
(C7); a live kill inside a 4 s verify window is impractical.
## Observations (documented, NOT acted on)
- The demo host had NO pre-existing `nas-media` unit (the spike-era shares were fully torn down) —
nothing predating `retry=0` exists; the "don't rewrite installed units" rule was moot in practice.
- `RemoveNetworkMount` leaves the EMPTY mountpoint dir under `/mnt/felhom-drives/` and does not
`reset-failed` a failed mount unit's residual state (the unit FILES are removed; a `not-found
failed` listing lingers until reset-failed/reboot). Cosmetic; cleaned by hand this run.
- Pre-existing (since 2026-07-08, unrelated to this task): `lanresolver: cannot list provisioned
guests: permission denied` — `/var/lib/felhom-agent/guests` is root-0700 while the agent runs
non-root.
- host-install header/SCRIPT_VERSION drift (v1.11.0 vs 1.12.0): the v1.12.0 bump had shipped with
no changelog entry and no header sync; both fixed at v1.13.0.
- The controller container publishes NO ports (bridge-only, traefik-fronted): in-guest API testing
needs the container IP + `Host:` header — the "POST to in-guest 127.0.0.1:8080" note in older
session memory is stale.