security review 2: five rows for the confirmed serious findings (R-1406..R-1410), R-984 noted and re-ranked P4 -> P2 (254 -> 259 rows; docs only)
gates / gates (push) Successful in 2m9s

This commit is contained in:
admin committed 2026-11-19 09:04:34 +01:00
1 parent 064941f3b1
commit 3e7c05020b
1 file changed
+8 -3
+8 -3
View File
@@ -245,13 +245,18 @@ stopping line that lies.
| **R-1373** | Storage & devices | P3 | **Three endings have a second door that does less than the first.** Found 2026-11-10 (the endings list, `documentation/audits/endings-list-2026-11-10/TABLE.md` endings EC03, EC13, EA08; read in source, not measured on a box). (1) **Orphan delete** of an app (`internal/stacks/delete.go:305-460`, `internal/api/router.go:1150-1178`) removes the containers, the volumes and the stack record; it does not remove the undo copy, the update record or the backup folders that the normal remove takes away. (2) **A move of ONE app to another drive** (`internal/stacks/migrate.go:616-654`) removes the app's data and its backup unit from the old drive; the app's restore copy, a set-aside backup and kept data stay there. Only the move of all apps clears the old drive. (3) **A drive retired by a signed operator job** (agent `internal/signedjobs/decommission.go:86-103`) sets the record and does not take the drive out of the running guest, which the household's own door does (`internal/localapi/disks.go:559-567`). **Members (answers marked N):** EC03-4, EC13-4, EA08-4. | OPEN — filed 2026-11-10 (endings list) | — | Controller: make orphan delete call the same clean-up steps as the normal remove; make a one-app move take the app's other folders with it, or list them. Agent: say in `03-host-agent.md` whether the signed retire is meant to leave the drive in the guest. | CC |
| **R-1378** | Storage & devices | P3 | **Three endings can stop half-way with no finish, or have no way back that nobody chose.** Found 2026-11-10 (the endings list, `documentation/audits/endings-list-2026-11-10/TABLE.md` endings EC15, EC03, EC27; read in source, not measured). (1) **Wipe of a drive** (`internal/web/storage_handlers.go:911-1000`): the handler removes the drive from the registered places BEFORE the format. If the format fails, or the controller stops between the two steps, the drive is not registered and still holds the old data; there is no record of a wipe in progress, and a failed eject is only a warning. The handler does not look for apps on the drive. (2) **Orphan delete** writes no 'remove pending' record, so a controller that stops in the middle does not finish it at the next start (the normal remove does). (3) **Import over an installed app** (`internal/web/handler_export.go:312-356`, `internal/appexport/restore.go:288`) replaces the app's volumes and makes no undo copy first. **Members (answers marked N):** EC15-4, EC15-7, EC03-7, EC27-5. | OPEN — filed 2026-11-10 (endings list) | — | Controller: format first and unregister after, or write a wipe-in-progress record that the next start reads; give orphan delete the pending record; decide if an import makes an undo copy (a restore does). | CC |
## Security & access — 28 rows (P1 2, P2 5, P3 14, P4 7)
## Security & access — 33 rows (P1 2, P2 10, P3 14, P4 7)
| ID | Category | Sev | What | State | Blocked on | Next action | Owner |
|---|---|---|---|---|---|---|---|
| **R-1241** | Security & access | P1 | **2026-10-24: the Felhom half is DONE** — hub 0.168.0 deployed (the tokens sealed; every copy leaving the volume is clean, the live file's old bytes accepted by the operator, R-975 closed), the hub volume out of the Longhorn backup (D40, R-1243 VERIFY), the two old plain copies deleted by the operator. **What stays here is the homelab half:** the open NFS export and Longhorn's login (the homelab session). **A hacked app pod on DooPlex can read the hub's nightly database copies — with every customer's unsealed Cloudflare, tunnel and git token (R-975) — through the Longhorn NFS export, and by a published file-handle escape the operator signing keys on the same disk: whoever breaks one homelab app can take over the boxes' tunnels and sign agent updates.** MEASURED 2026-10-23 (Plex2Shell impact audit, `audits/plex2shell-felhom-impact-2026-10-23/`): export `/mnt/5_hdd/backup/longhorn-pvc` rw, `no_root_squash`, `no_subtree_check`, a sub-directory of sda1, which also holds `/mnt/5_hdd/felhom.eu/felhom-op-*` (no passphrase, R-924); `secure` is void in pods (containerd sets `ip_unprivileged_port_start=0`; a uid-1000 probe with no capabilities bound port 700); the four `felhom-system` volumes have `encrypted=false`; Longhorn's API on :9500 answers any pod with no login (backups, targets, volumes). In the Plex2Shell window neither door was used — verdict NOT REACHED (nfsd accepted 0 new connections 10-10 → 10-23, positive control; no off-schedule Longhorn backup or volume). 2026-10-23 ~15:18Z, operator yes: the export was narrowed from `*` to `10.42.0.0/16` — the home network and the tailnet are closed, pods are not (Longhorn connects from its pod address). Not exercised: the escape itself (Tier-2 machine). | OPEN | the homelab session (DooPlex host + cluster); the operator (R-924) | Homelab: a NetworkPolicy so only `longhorn-system` reaches :2049/:111 and :9500, or the backup store on its own filesystem. Felhom: R-975 sealed in hub 0.168.0 (deploy pending); the hub's plain Longhorn copies → R-1243; a passphrase on the signing keys (R-924) | operator |
| **R-1139** | Security & access | P2 | **Tester 2's break-glass root password was not rotated with the others (R-954, operator decision D1 = A, 2026-10-17): the box has been offline since 2026-10-04.** demo-hp, Tester 1 and demo-felhom were rotated and proven on 2026-10-17 (`audits/agent-allowlist-rotation-2026-10-17/P3/`). | OPEN — filed 2026-10-17 **2026-10-22: closes when Tester 2 is reinstalled (operator ruling 2026-10-22 14:11 — a fresh install sets and vaults a new console password).** | Tester 2 reporting again | When it reports: the installer's step 4b alone (`P3/rotate-step4b.sh` shape), then the vaulted value logs in (PVE API 200) and the old one is refused | CC |
| **R-1305** | Security & access | P2 | **The file manager on every box (FileBrowser Quantum 1.5.6-stable) is inside 36 published notices; 2 are fixed in our own channel (1.5.7-stable)** — GHSA-qx86-4v5r-26g5 (high: with the victim's password, an anonymous request re-enrolls TOTP and skips the second factor; our households have no second factor there, so low for us) and GHSA-3846-gh75-gp3m (medium: a public upload share replaces files). Most others name fixes only on the 2.x line. `files.<domain>` is public through the tunnel (behind the controller's forwardAuth, password sign-in on). Found by the first security-watch run (R-1303). `audits/security-watch-2026-10-26/ADVISORIES-2026-10-26.md` | **NARROWED 2026-10-27 — the two notices fixed in our channel are RELEASED: controller v0.346.1 pins `gtstef/filebrowser:1.5.8-stable` (1.5.7-stable skipped: it broke existing TOTP sign-ins), global floor 15:50:15Z; live on Tester 1, demo-hp, demo-felhom and 9202 (sign-in walk: listing 200, 401 without a cookie; R-1281 probe: an app container cannot reach it) — `audits/security-app-updates-2026-10-27/part5/`. LEFT: triage the 34 other notices against the 1.5.x line.** **2026-10-28 (R-1332): HTTP2Shell does not reach it** — 1.5.8-stable carries Alpine `brotli-libs` 1.2.0-r1 (2.0.2-stable: 1.2.1-r0, a new main version), but the Go program does not link libbrotli (`ldd`: musl only; no process maps it, all three boxes) — the library is there for libcurl. NOT a step today. Earlier: OPEN — filed 2026-10-26 **2026-11-02:** a new notice, GHSA-892f-9h5p-mfc8 (low: `/api/settings` shows secrets to an admin's token), fixed on 2.1.0-stable only. | a controller release | raise `FileBrowserImage` to 1.5.7-stable in the next controller release; triage the 34 others against 1.5.x (which apply to the stable line) | CC |
| **R-1406** | Security & access | P2 | **A taken-over controller (root in the household's guest) can most likely become root on the Proxmox host: two local-API disk routes of the agent act as root on paths the caller names, with no check that the path is an enrolled drive, and no refusal of a link.** The agent mounts a filesystem by an id the caller sends (shape checked only), then changes an owner and binds a folder under that mount by path. The row „guest ↔ Proxmox host: holds” of `13` §5 does not hold for this caller. Harm: the host, the agent's keys and, on a BYO host, the owner's other guests. Not usable by a stranger alone: it needs the controller first. Static, with the three host facts it rests on measured on demo-hp (2026-11-19); the path itself was not run. Second read: holds. Fix: the assign route accepts only a disk the unclaimed-disk filter offers; the attach route requires a registered user-data drive; one root helper does the steps with no-follow opens; `nosuid,nodev` on drive mounts. Test: an attach where the namespace folder is a link changes no owner and binds nothing. Security review 2, 2026-11-19 (`audits/security-review-2-2026-11-19/FINDINGS.md`, S2-01; detail held off-git for the operator, `/mnt/5_hdd/felhom.eu/evidence/security-review-2-2026-11-19-PRIVATE/`). | OPEN — filed 2026-11-19, technical severity Critical | — | Fix task (agent release, operator vouch); project Claude checks the finding first | CC |
| **R-1407** | Security & access | P2 | **Importing an app bundle (`.fab`) skips the box's privilege allow-list and its path rules: the bundle's own app definition is written and started unchecked, its drive data is unpacked where the bundle says, and its name may be one of our own infrastructure stacks.** Whoever made the file gets what the allow-list (R-971) exists to refuse — in the end root in the guest — after one import click by the household; a bundle in a network drive's `exports` folder can be placed or changed by anyone who can write that share. Static; second read: holds (both parts). Fix: ask the allow-list before anything is stopped or written; accept only the known file names; refuse infrastructure names; require the data path to be this app's folder on a registered drive. Test: a bundle that asks for a refused setting, or for a path outside the app's folder, ends „refused” with the running app untouched. Security review 2, 2026-11-19 (`audits/security-review-2-2026-11-19/FINDINGS.md`, S2-03 and S2-04; detail off-git, `/mnt/5_hdd/felhom.eu/evidence/security-review-2-2026-11-19-PRIVATE/`). | OPEN — filed 2026-11-19, technical severity High | — | Fix task (controller release) | CC |
| **R-1408** | Security & access | P2 | **Dashboard pages put a job's error text into the page without escaping, and that text can hold a file name chosen by an app, or a field of an imported bundle.** A hacked app (or a prepared bundle) can make script run in the dashboard with the household's session when the household exports, moves or imports that app — every dashboard action, R-1407's import included. No page rule (a script policy) stops it. Static; second read: holds (export and import traced end to end; the move route's trigger is not proven). Fix: escape at every such place in the page scripts, add a template gate for it, then a script policy header. Test: an error text holding markup shows as text. Security review 2, 2026-11-19 (`audits/security-review-2-2026-11-19/FINDINGS.md`, S2-05; detail off-git, `/mnt/5_hdd/felhom.eu/evidence/security-review-2-2026-11-19-PRIVATE/`). | OPEN — filed 2026-11-19, technical severity High | — | Fix task (controller release) | CC |
| **R-1409** | Security & access | P2 | **The privilege allow-list judges one file, but the app's settings reach the compose tool as its environment with no rule on their NAMES, and the tool is not told which file to use — so a setting named like one of the tool's own control variables changes which definition runs.** A bad catalog commit can therefore still run a refused definition on every box that has the app (the case R-971 was closed for); for an installed app no household act is needed. The same gap is open through an imported bundle's settings. Static, and the tool's behaviour measured on scratch guest 9202 (2026-11-19: a second file named by the environment was applied in a render). Second read: holds. Fix: name the file and the project on every compose call; refuse the tool's and the loader's variable names where settings are read. Test: such a name is dropped at load, and the call carries the file argument. Security review 2, 2026-11-19 (`audits/security-review-2-2026-11-19/FINDINGS.md`, S2-06; detail off-git, `/mnt/5_hdd/felhom.eu/evidence/security-review-2-2026-11-19-PRIVATE/`). | OPEN — filed 2026-11-19, technical severity High | — | Fix task (controller release) | CC |
| **R-1410** | Security & access | P2 | **The hub's operator sign-in has three gaps that add up: the scrambled form (hash) of the start password has been in the public repository's hub manifest since 2026-02-16; nothing limits wrong tries on the sign-in form or on Basic sign-in; and an operator session cannot be ended (no sign-out; a password change keeps old sessions for up to 7 days).** If the February password is still the live one, a stranger can guess it offline with no trace on the hub, and that one password opens everything, each box's break-glass root password included — then this row is P1. If a newer password was set on the Configuration page, the public hash is the fallback only (it returns if the stored one is cleared or unreadable). Which is live was NOT measured (the read of the running hub was refused to CC). Static; second read: holds (all three). Fix: the start hash out of git into a Secret, a new password, a try limit by address and in total before the slow compare on both paths, sign-out, and a password change that ends sessions. Test: the manifest gate refuses a non-empty hash; N+1 wrong tries are refused on both paths. Security review 2, 2026-11-19 (`audits/security-review-2-2026-11-19/FINDINGS.md`, S2-10, S2-12 and S2-13; detail off-git, `/mnt/5_hdd/felhom.eu/evidence/security-review-2-2026-11-19-PRIVATE/`). | OPEN — filed 2026-11-19, technical severity High (conditional). **Operator: say whether the hub password was changed on the Configuration page after 2026-02-16** | — | Operator answers; then a fix task (hub release) | CC |
| **R-1042** | Security & access | P3 | **The boxes' Cloudflare keys cannot change zone settings, so Cloudflare's own „Always Use HTTPS” cannot be switched on.** Measured 2026-10-14 on all three zones (`audits/https-only-2026-10-14/``1A-1-zones-and-keys.txt`): each `cf_api_token` is valid and sees only its own zone, but GET/PATCH `settings/always_use_https`, `ssl`, `security_header` answer 403; the redirect and transform rule phases too. The https-only fix therefore lives on the box (decision 246). A second layer at Cloudflare needs „Zone Settings: Edit” (and, for headers, „Transform Rules”) added to each customer's key — the hub's zone-reach check stays. | **WAITING-ON-OPERATOR** — filed 2026-10-14 | a key permission only the operator can grant | Optional: widen each customer key, then switch „Always Use HTTPS” on per zone | operator |
| **R-1242** | Security & access | P3 | **The demo NAS export `/mnt/5_hdd/felhom-data` → 192.168.0.162 (demo-felhom) is a sub-directory of DooPlex's sda1 with `no_subtree_check`, so a compromised demo-felhom host could walk the whole disk as uid 101000 and read every world-readable file there (the GPG secrets exports are 0644, every repo, the public keys).** Measured 2026-10-23 (`audits/plex2shell-felhom-impact-2026-10-23/` Part 1; `exportfs -v`: `all_squash,anonuid=101000`). 0600 files (the signing keys) stay out of reach. Not exercised. | OPEN | a DooPlex host change (fence) | Give the demo share its own filesystem, or `subtree_check`; or remove other-read from `/mnt/5_hdd/backup/secrets/exports` | operator |
| **R-1104** | Security & access | P3 | **Check https-only on Tester 2 when it reports again.** R-1033 (plain http on every public name) is closed on the three boxes that report (35 names → 301, 2026-10-16). Tester 2 (`sajatfelhom.hu`) has been off since 2026-10-04 on controller 0.292.0; its addresses answer Cloudflare 530 (tunnel down), so nothing is served in plain http while it is off. When it comes back it takes the floor (≥ 0.312.0 carries the redirect). | OPEN — filed 2026-10-16 | Tester 2 reporting again | Run `scripts/https_only_check.py` with Tester 2 in it; close | CC |
@@ -277,10 +282,11 @@ stopping line that lies.
| **R-1248** | Security & access | P3 | **An app's own numbers can lie: an attacker with code inside an app's container can answer the app's API with whatever matches its flood, and the upload watch (R-1244) would then call it explained.** Measured remedy, not built: a per-container upload ceiling with `tc` from a helper container in the app's network namespace — on scratch 9202 (unprivileged) a flood was held at 19.9–20.1 Mbit/s and the app kept answering (`audits/SPIKE-upload-watch-2026-10-23.md` §5). Costs: a helper image with `tc` (a new external image, or `tc` in the controller image), `NET_ADMIN` for that helper, re-applied after each app restart. | OPEN | an operator decision (the new image / capability) | Decide; if yes, CC builds the ceiling for the download and media apps first | operator |
| **R-1309** | Security & access | P4 | **What no automated source shows: a fix announced only by mail, forum or a blog we do not read; a security fix nobody calls one (cloudflared 2026.10.2's „bump distroless image"); closed-source apps (Plex, Emby).** The watch (R-1303) shows these only as BEHIND (weekly). The vendor mails the operator receives stay a source. `audits/security-watch-2026-10-26/SPIKE.md` | OPEN — filed 2026-10-26 (a stated limit) | — | none until a vendor offers a feed; add a vendor's feed to `security_watch.FEEDS` when one exists | CC |
## Box system & updates — 20 rows (P3 14, P4 6)
## Box system & updates — 20 rows (P2 1, P3 14, P4 5)
| ID | Category | Sev | What | State | Blocked on | Next action | Owner |
|---|---|---|---|---|---|---|---|
| **R-984** | Box system & updates | P2 | **Host and agent hardening, low risk each: the backup-target route has no role check; the unsigned guest spec has no size bounds; signed ops have no maximum validity; the guarded mkfs does not refuse LVs, zvols or RAID devices; the local API is reachable from the LAN by a static route; a sudoers inspect pattern is wider than needed; install passes secrets on the command line; guest settings files are world-readable; the host offers rpcbind and DNS on the LAN.** Fix each in its own small change. Security review 2026-10-12 (`audits/security-review-2026-10-12/`); detail held off-git for the operator (`/mnt/5_hdd/felhom.eu/evidence/security-review-2026-10-12-PRIVATE/`). | OPEN **2026-11-19 (security review 2, S2-02): the backup-target clause is larger than „low risk”: the route accepts any mount point that is not the system disk, with a caller-chosen storage id, so a taken-over controller can move the whole-guest backup to a place it can write, erase or forge — with no operator signature (`03` §4). Second read: holds; technical severity High. The mkfs clause is confirmed again (S2-31). Re-ranked P4 → P2 on 2026-11-19 for the backup-target clause; split it out when the fix is planned. Detail off-git, `/mnt/5_hdd/felhom.eu/evidence/security-review-2-2026-11-19-PRIVATE/`.** | — | Batch into hardening work | CC |
| **R-1260** | Box system & updates | P3 | **Spring 2027 (2027-03-28): two clock-change effects in the controller's night (review round 2, HUNT-4 H2/H3).** The update leg's deadline is W + 5 h of REAL time while the full-system backup's gate counts wall-clock minutes, so a step can start after the gate stopped waiting (`felhom-controller/controller/internal/stacks/unattended.go:225`; `internal/quiesce/quiesce.go:922-933`); with W in 02:00–02:59 the DB dump and the second-drive copy start at the same instant (`internal/scheduler/scheduler.go:509`; VERIFIED: Go maps 2027-03-28 02:30 Europe/Budapest to 03:30 CEST). **2026-11-11:** the Clock list leans on this row (`documentation/audits/clock-list-2026-11-11/TABLE.md` CC02, CC06-6, CC07-6, CC22-6). | OPEN | — | before 2027-03-28: one wall-clock basis for both, and a test at the spring date | CC |
| **R-1367** | Box system & updates | P3 | **The host journal fills a small system disk: 61% of its lines are `pam_unix(sudo:session)` open/close pairs from the agent's sudo calls.** MEASURED 2026-11-09, one hour, same on demo-hp and the N100: ~14,000 journal lines, ~8,500 of them sudo session lines; the top commands are `smartctl -a` (1,074 per hour — one every ~3 s), `lxc-info -n` (662), `mkdir -p` (540), `lvs` (537), `pct exec` (~520). The journal grows by journald's default cap (10% of the filesystem, max 4 GB): the N100 sits at 4 GB, demo-hp was 3.8 GB on 2026-11-07 and reached 0.7 GB again two days after a vacuum to 500 MB. On a 39 GB root that space is what the whole-guest local copy needs (R-548). | OPEN — filed 2026-11-09 **2026-11-10 (space list, `documentation/audits/space-list-2026-11-10/TABLE.md`):** place H7: the N100 (4.28 GB, 74 files) and Tester 1 (4.29 GB, 81 files) sit at the journal's own cap; Tester 1 writes about 150 MB a day. Lines written by `sudo` are 92 / 92 / 90 % of all lines in one hour — this counts three lines for each call (the command and the session pair), while this row's 61 % counts the session pair only: the same fact, counted two ways. No rank change is proposed: the journal stops at its own cap (about 4.3 GB), so it costs space and log history, not safety. | — | Agent, through the signed config bundle: `Defaults:felhom-agent !pam_session` (and `!syslog` only if the command log has no reader — check the audit trail's needs first) in the sudoers file; a journald cap drop-in (e.g. `SystemMaxUse=500M`) on the host; and read why `smartctl -a` runs every ~3 s (the SMART poll's interval or a loop). Red-proof each in the agent's tests; measure the line rate after on demo-hp. Evidence: `/mnt/5_hdd/felhom.eu/evidence/demo-hp-images-2026-11-09/journal-noise.txt` (off-git) | CC |
| **R-1368** | Box system & updates | P3 | **Five places on the host's system disk only grow: nothing deletes them and no rule says how many to keep.** MEASURED 2026-11-10 on demo-hp, demo-felhom and Tester 1 (the space list, `documentation/audits/space-list-2026-11-10/TABLE.md` places H5, H6, H9, H12, H13). (1) **Old kernels** (`/lib/modules`, `/boot`): 8 / 8 / 3 kernels = 8.15 / 8.15 / 3.05 GB. The ring-0 boxes got 7 kernels in 37 days and no kernel was ever removed: the update wrapper refuses a plan that removes a package (`11-os-updates.md` rule R4), and nothing runs `apt autoremove` (searched: architecture folder, both registers, runbooks, three source trees — 0 hits). The N100's system disk grew 4.6 GiB in 26 days, about a kernel a week. (2) **The agent's self-update folder** `/var/lib/felhom-agent/selfupdate`: 49 / 48 / 33 binaries = 740 / 724 / 509 MB, one for each agent version the box ever got; no document names the folder. (3) **Guest template archives** `vzdump-lxc-9100-*` in `/var/lib/vz/dump`: 1 / 4 / 1 = 0.65 / 2.56 / 0.65 GB. (4) **Old agent binary copies** `/usr/local/bin/felhom-agent.bak-*`: 1 / 11 / 0 files, up to 155 MB. (5) **Logs of failed whole-guest backup runs** in the dump folder (small; 9 on demo-hp). **Members (answers marked N):** H5-3, H5-4, H6-3, H9-3, H9-4, H9-8, H12-3, H12-4, H13-3, H13-4. | NARROWED — **2026-11-15 06:35 — the FIRST real night removal: the N100, by itself, at 05:58 CET.** 8 kernels before, 2 after (it kept 7.0.14-26 and 7.0.14-28, the running one), 6 packages purged in 35 s, 6.81 GB gained (the wrapper's own figure; 65.1 GB free after), 0 left-over paths; the default kernel and the pinned file are unchanged (the file's time is still 2026-11-08); the hub's „Kernel step” cell reads „cleaned”; 0 failed units, 9 of 9 containers healthy, clock +0.06 s to DooPlex. demo-hp had a QUIET night (2 installed, nothing to remove) — it does not count; its 6 old kernels went on 2026-11-10 at 13:51, by day. **D82 count: the N100 1 of 2, demo-hp 0 of 2.** demo-hp can count only after two more kernels arrive (it holds 2 and keeps 2). Evidence: `audits/night-build-2026-11-14/E/`. Earlier: NARROWED — **2026-11-14 16:06 — RULED (D82, `09` §3 decision 307): a household's box may remove its old kernels, after two REAL night removals on each ring-0 box (the N100 and demo-hp); quiet nights do not count. Count so far: the N100 0, demo-hp 0 (the Tester 1 removal of 2026-11-14 was started by hand and is not a night). It needs a later agent release to switch it on.** Earlier: NARROWED — **2026-11-14 (release 2a): the old-kernel part is BUILT — agent 0.174.0, vouched 15:21 CET (`11` §5.11.1 `[FACT — built]`).** A ring-0 box removes its old kernels at night by the wrapper's own rule. On boxes: a real removal on Tester 1 (3 → 2 kernels, 1.12 GB, 8.5 s; the default, the pinned file, the generators and the state file unchanged; a second run removed nothing; one clean restart) — started by hand with a ring-0 mark set for the test on the operator's yes, because a night cannot be started by day (R-1394); dry runs on demo-hp (nothing to remove; refusals P1, P2, P3) and the N100 (6 of 8 will go). **LEFT for this part: the step by the agent at night — read the N100's and demo-hp's night of 2026-11-15; then a ruling for ring-1 boxes (D82).** Evidence `audits/release-2a-2026-11-14/`. The other four places of this row are unchanged. Earlier: OPEN — filed 2026-11-10 (space list) **2026-11-10 12:22 — RULED (D67 = yes, as a separate step; `09` §3 decision 286):** the box may remove old kernels by itself; rule R4 stays as it is for update plans; the clean-up is its own step with its own narrow rule (old kernel packages only, only after a good boot on the newest kernel). **2026-11-10:** the endings list leans on this row (`documentation/audits/endings-list-2026-11-10/TABLE.md` EA01-4, EA23-4, EA24-4, EA26-4). **2026-11-10 — the kernel part's fix shape is PROVEN by hand on demo-hp** (`documentation/audits/kernel-cleanup-spike-2026-11-10/`; design `11-os-updates.md` §5.11.1, not built): the rule removed 6 of 8 kernels and nothing else, 6.4 GB gained, boot settings unchanged, one orderly reboot back on the same kernel, a second run removed nothing. Not proven: a box whose running kernel is not its newest; a boot of the kept old kernel. **2026-11-10 — part (2) FIXED IN SOURCE, NOT RELEASED** (felhom-agent `776025c8`): after an update that ended well the agent removes every staged binary but the running version's; nothing is removed while an update is pending. `TestPrune_*` (6 tests, one runs the wrapper's real commit and rollback verbs), red seen three times. A box cleans its folder at the first good update after the release that carries it. Part (4), the `.bak-*` copies, is not reached (root-owned, made by a person). Parts (1), (3), (5) stay open. **2026-11-10 14:46 UTC — the kept old kernel boots:** on demo-hp the one-shot flag booted 7.0.14-26-pve to a healthy Line truncated
@@ -288,7 +294,6 @@ stopping line that lies.
| **R-1370** | Box system & updates | P3 | **The host's system disk has no space budget: no document says how much the host's own files may take, and the installer's sizing rule covers the guest only.** Found 2026-11-10 (the space list, `documentation/audits/space-list-2026-11-10/TABLE.md` place H1, answer H1-3; search S5 — 0 hits). Today: demo-hp 58 % of 41 GB, N100 38 % of 101 GB, Tester 1 34 % of 63 GB — and the parts that grow are the journal (R-1367), old kernels and agent binaries (R-1368), old backup copies (R-1087) and the local whole-guest copy (R-548). Second half, known but with no row until now: on a small disk the guest's own guards read the guest's 70 GiB while the thin pool behind it holds 54 GiB (demo-hp: 102 GiB promised on a 53.9 GiB pool, 70.67 % used), so only the hub's thin-pool alarm stands before a full pool. **Member:** H1-3. | NARROWED — **2026-11-15 (build of release 4): BUILT, not released** — installer v1.38.0 on `main`, NOT tagged (`16` §14.5 `[FACT — built, not released]`): `check_system_disk` first in the preflight; an appliance below 200 × 10⁹ bytes stops; `--accept-small-disk` goes on, prints the cost, records `small_disk`; a BYO host is warned only; an unreadable size stops and names the flag. 8 tests, 15 mutations seen red. **Not run on a machine.** What it does not do is R-1398. For the release: the proof VM gets a 200G disk; demo-hp's reinstall needs the flag. Earlier: OPEN — **2026-11-15 12:49 — RULED (D83 = refuse, with one named flag; `09` §3 decision 308).** Fix shape: the installer reads the size of the disk that holds the `pve` volume group before any change; below 200 × 10⁹ bytes an appliance install stops with the designed sentence; `--accept-small-disk` goes on, prints the cost, records `small_disk: true`; a BYO host is warned only; a size that cannot be read refuses on an appliance and names the flag (`16` §14.5). Earlier: OPEN — **2026-11-14 (night build):** the open point of D75 has a written design with a pick, not built: `16-storage-physical-identity.md` §14.5 — refuse below 200 GB on an appliance install, with one named flag for the operator; warn only on a BYO host; 8 edge cases. It waits for the operator's yes. OPEN — filed 2026-11-10 (space list) **2026-11-10 12:22 — RULED (D69 = yes; `09` §3 decision 288):** the section goes inside `16-storage-physical-identity.md`, no new document. **2026-11-10 (afternoon) — the section is written:** `16-storage-physical-identity.md` §14 (45 places, the list „no limit, by ruling”, the budget, the open design points); `PROMPT-TEMPLATE.md` §N.5 keeps it true. The budget from measured numbers: the root needs about 14 GB when the whole-box copies go to a backup drive, and 28–54 GB at keep 3 when they do not. **D75 (for the operator):** the smallest system disk Felhom accepts. CC's pick: 128 GB only with a backup drive, 200 GB without one. The other option, 128 GB always, lets a box with many apps and no backup drive stop its local copies. With no ruling the installer accepts any disk. Open after the ruling: the installer's refusal. **2026-11-10 15:19 — RULED (D75 = one number, 200 GB, with or without a backup drive; `09` §3 decision 295).** This replaces CC's two-step pick. **Open, not ruled:** what the installer does below 200 GB — refuse, or warn. demo-hp (128 GB) stays as a test box. `16` §14.3 carries the ruling. | — | Design first: what the installer does below 200 GB (refuse, or warn) — a later session. Then the installer. | CC |
| **R-1398** | Box system & updates | P3 | **A machine that the installer refuses for a small system disk tells nobody but its own screen, and the hub does not read the „small disk” mark.** Found 2026-11-15 while building D83 (installer v1.38.0, not tagged). (1) On the public image the first-boot unit (`scripts/iso/felhom-bootstrap.sh`, „retry until success”) runs the installer again every 30 seconds: a refused 128 GB machine prints the refusal on its screen for ever, and the hub shows a bound appliance that never installs, with no reason. (2) An install that went on with `--accept-small-disk` sends `"small_disk":true` in its enrolment request; the hub ignores the field, so no page shows which boxes were accepted small (`16-storage-physical-identity.md` §14.5, the last row of the built table). | OPEN — **2026-11-15 evening (second build): part (2) is BUILT, not released** (hub `main` after 0.182.0): the hub keeps `small_disk` (and a size, when an installer sends one) from the enrolment and the box's page shows „System disk: small — accepted below 200 GB”. **Part (1) is open, and one more thing was read:** the hub's bind form has NO field for extra installer arguments (the handler takes `extra_args`, the page does not offer it; the household's own bind sends none), so `--accept-small-disk` cannot reach an install from the public image today (`audits/build-release-4b-2026-11-15/E/small-disk-flag-on-the-public-image.md`). Earlier: OPEN — filed 2026-11-15 | the next installer ISO for (1) | (1) a final exit code for a refusal that no retry can cure; the first-boot unit stops retrying on it, shows the sentence once, and tells the hub the reason; (2) the hub stores `small_disk` at the enrolment and shows it on the box's page | CC |
| **R-1400** | Monitoring & notifications | P4 | **Every household mail but one prints the event's raw details as a line „Megjegyzés”.** Found 2026-11-15 while building the receipt mail of D84: the hub's household frame adds the event's details JSON under „Megjegyzés” (a note); the new receipt mail leaves it out, the other household mails keep it. A household reads a line of field names and values. Not measured: which mails, and how it reads in an inbox (`hub/internal/notify/testdata/mail_goldens/`). | OPEN — filed 2026-11-15 | — | read the household mail frame: which event types carry details worth a sentence, and drop the raw line for the rest | CC |
| **R-984** | Box system & updates | P4 | **Host and agent hardening, low risk each: the backup-target route has no role check; the unsigned guest spec has no size bounds; signed ops have no maximum validity; the guarded mkfs does not refuse LVs, zvols or RAID devices; the local API is reachable from the LAN by a static route; a sudoers inspect pattern is wider than needed; install passes secrets on the command line; guest settings files are world-readable; the host offers rpcbind and DNS on the LAN.** Fix each in its own small change. Security review 2026-10-12 (`audits/security-review-2026-10-12/`); detail held off-git for the operator (`/mnt/5_hdd/felhom.eu/evidence/security-review-2026-10-12-PRIVATE/`). | OPEN | — | Batch into hardening work | CC |
| **R-979** | Box system & updates | P3 | **Controller updates are trusted by registry tag alone (no digest, no signature), and the Update button offers any newer tag, even an unvouched one; the built-in traefik, cloudflared and FileBrowser images are also pinned by tag only.** Whoever can push to the registry decides what runs. Fix: pin by digest from the hub's vouched list; offer only vouched versions. Security review 2026-10-12 (`audits/security-review-2026-10-12/`); detail held off-git for the operator (`/mnt/5_hdd/felhom.eu/evidence/security-review-2026-10-12-PRIVATE/`). | OPEN — design-proposal | — | Design note, then build | CC |
| **R-974** | Box system & updates | P3 | **A compromised host agent can likely become root on the host through the root wrapper that applies PBS settings: the wrapper accepts a folder path inside the agent's own writable tree and then acts on it as root.** Static (the symlink behaviour was reasoned, not run). Same class as R-861. Fix: a fixed root-owned folder, and refuse symlinks before any root action. Security review 2026-10-12 (`audits/security-review-2026-10-12/`); detail held off-git for the operator (`/mnt/5_hdd/felhom.eu/evidence/security-review-2026-10-12-PRIVATE/`). | OPEN | — | Fix the wrapper with a test on a throwaway box | CC |
| **R-1214** | Box system & updates | P3 | **On the 8 GB box with the swap fix, the night's second-drive backup still pushes memory „full" pressure above 20 % for about 3 minutes; the box keeps answering.** Night soak 2026-10-21→22 (`audits/night-2026-10-21/P1-soak/03-rec.jsonl`, 10-s samples; fresh box from the public ISO, agent with swap = RAM, 8 apps, the rehearsal uploads for 2 h): at 01:30–01:34 UTC (the 03:30 tier2 job) host 39 % / guest 45 % over 10 s, over 60 s at most 20 % / 25 %; IO „full" 66 %; guest swap 1.35 → 2.1 GB; `/api/health` answered every probe (slowest 4.1 s, largest interval 16 s); 0 OOM kills. Once more in the upload window (Immich loading its models, 31 % / 35 % for one sample). The brief's soak pass line was „full" < 20 %, so the soak missed that one line; the freeze itself (96 % for an hour, R-1192) did not come back. Not diagnosed: which part of the tier2 job (the copy's page cache, the volume dumps) causes it. | WATCHING — **2026-10-24 02:4xZ dated check read (weekend-night session, read only):** Tester 1 (8 GB host, guest 4 GB, controller 0.341.0) ran its 03:30 copy at 01:30:00–01:30:13Z (3 apps, 157 MB + 86 MB + 16 KB; no „capped copy container did not start" WARN, no leftover `felhom-tier2-copy-*`); the guest cgroup's memory „full" stall counter on the host (`/sys/fs/cgroup/lxc/9201/memory.pressure`) read **total 444 µs** over the whole 11.1 h since the host started (15:32Z), so „full" stayed far under 20 % (20 % of 10 s = 2 s); 0 OOM kills. **Not shown:** that the copy ran in the capped container (the INFO log does not name it; Docker keeps only ~200 events, from 02:36Z). **Weak load:** 0.24 GB, not the soak's large copy — so the row stays WATCHING; close after a night whose copy moves ≥ 1 GB, read from the same counter. WATCHING — **RELEASED 2026-10-23 in controller 0.340.0** (floor 06:41 UTC; on all three test boxes, start-up „no leftover”, 0 copy containers; Tester 1's manual second-drive run: 3 apps, 0 containers left). **Open only for the consequence:** dated check 2026-10-24 — Tester 1's first night copy ran in the capped container and memory „full” pressure stayed under 20 % (the box's own night log). READY — rides controller 0.340.0. READY — waits for the controller 0.340.0 release. **2026-10-22 night — cause measured, fix built and proven on 9202, waits for the controller release.** Cause: the copy's page cache is charged to the guest and evicts the apps' working set. On 9202 (4 GB RAM, 1.8 GB held by a hog, a 6 GB copy): plain rsync full stall 5 702 / 2 464 ms, peak „full” 16.8 / 6.9 %; the same copy in a 512 MB memory-capped container 374 / 320 ms, 0.9 / 1.0 %. Controller main `03c0a69` + `ae39467` (unreleased, rides 0.340.0): the copy runs capped only for clean `/mnt/` paths with no symlink (security review); a docker error (exit 125) falls back to plain rsync. A real `POST /api/backup/tier2` on rc.4: 6 capped containers, each exit 0; Immich's copy equals its source (dry-run 0 differences, 44 = 44 files). Evidence `audits/night-2026-10-22/P4-R1214/`. **2026-10-23 night (stray-copy fix, night brief Part 1).** The capped copy's container outlived the copy (a timeout or a controller restart killed only the `docker` CLI; `--delete` went on outside the R-1180 guard). Controller main `66a69d4` + `912e592` (unreleased, rides 0.340.0): a fixed name + label per copy, removed by name on any error, every leftover removed at start before the scheduler and the web server, a restore refused while a copy runs or may run; `--mount` (a path the daemon does not see is refused, not created empty) and the image ID with `--pull never` (review C4, C7). Proven on 9202 with rc.5 (`audits/night-2026-10-23/P1-straycopy/`): a restart in the middle of a 16 GB copy → the new controller removed `felhom-tier2-copy-immich-948f2280` at start, the restore probe was refused during the copy, the redone copy equals the source; a 15 s test build → the timed-out copy's container removed, the partial copy stopped growing. A second review of that code (N2–N5) and the commit's own security review (two races) → `9d9e404`, `1babfe9`: a stray clears when a later removal works, the start-up check never removes this process's own copy and holds the destination's lock, one copy per destination at a time, `RunTier2` refuses during a restore for every caller and records itself in one step with that check, every leg asks again. Re-proven on 9202 with rc.8 (`audits/night-2026-10-23/P1-straycopy/1*`). 9202 back on 0.339.0. | — | Release 0.340.0; then re-measure the 03:30 job on the 8 GB soak box and close if „full” stays low | CC |