updater:
- /opt/furtka/current is swapped via rename() instead of unlink+symlink,
so a power cut can't leave the box without a current version
- OSError after the flip (disk full writing the Caddyfile) now rolls back
too; rollback restores the previous Caddyfile and reloads caddy +
daemon-reload
- the health check keeps polling on read timeouts / RemoteDisconnected
instead of escaping and skipping the rollback
- run_update writes an 'error' stage for any failure (download, sha,
disk full); the settings page shows it instead of spinning on
'Downloading…'
auth:
- users.json that exists but is empty/corrupt/unreadable raises instead of
reading as 'no admin yet' — before, a truncated file made /login an open
create-admin form for the whole LAN. /login answers 503 with SSH recovery
steps instead
- save_users fsyncs before the rename
- first-run admin creation is serialised (threaded server)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Wq1hzUp4ZmCKHDuLKCnu73
Ten wrong passwords from the same (username, client-IP) tuple within
15 minutes now return 429 with Retry-After for the next 15 minutes;
authenticate() isn't even called while locked, so the 429 response is
identical whether the password would have been correct — no oracle.
Tuple keying prevents an attacker from one IP from locking the real
admin out of their own box: a different IP (or an ISP reconnect) keeps
them in. The client IP comes from the rightmost X-Forwarded-For entry,
which is what Caddy appends and thus trustworthy (no upstream proxy in
front of Caddy). First-run setup bypasses the lockout — otherwise a
clumsy operator could lock themselves out before an admin exists.
State is in-memory (parallel to SessionStore), so `systemctl restart
furtka` clears a stuck lockout.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
One-admin, one-password model — all of /apps, /api/*, /, and
/settings/ now require a signed-in session. Passwords are werkzeug
PBKDF2-hashed in /var/lib/furtka/users.json (mode 0600, atomic write
via the same .tmp+chmod+rename dance installer.write_env uses).
Sessions are secrets.token_urlsafe(32) tokens held in a module-level
SessionStore dict (thread-safe lock included for when we swap to
ThreadingHTTPServer). Cookies are HttpOnly, SameSite=Strict, and
Path=/, with Secure set when X-Forwarded-Proto from Caddy says HTTPS.
Two bootstrap paths:
* Fresh install — webinstaller step-1 collects Linux user + password,
the chroot post-install step hashes the password and writes
users.json on the target partition. First browser visit lands on
/login with the account already present.
* Upgrade from 26.10-alpha — no users.json yet, so /login detects
setup_needed() and renders a first-run setup form. POST creates
the admin and immediately logs in.
POST /logout revokes the server session and clears the cookie.
Unauthenticated HTML requests 302 to /login; unauthenticated API
requests 401 JSON so fetch() callers see a clean error. A sleep(0.5)
on failed logins is the brute-force speed bump on top of werkzeug's
~600k-iter PBKDF2.
Caddyfile gains /login* and /logout* handle blocks in the shared
furtka_routes snippet so both :80 and the HTTPS hostname block
forward the auth endpoints to localhost:7000. Without this Caddy
would 404 from the static file server.
Test surface:
* tests/test_auth.py (new, 19 cases): hash roundtrip, users.json
I/O, session create/lookup/expire/revoke.
* tests/test_api.py: new admin_session fixture; existing HTTP
tests updated to send the cookie; new tests cover login setup,
login success, wrong-password 401, logout revocation, and the
guard's 302/401 split.
* tests/test_webinstaller_assets.py: new case that unpacks the
users.json _write_file_cmd body and verifies the werkzeug hash
round-trips against the step-1 password.
Bumped version to 26.11-alpha and rolled CHANGELOG. Also folded in
the ruff-format fix that was pending from 26.10-alpha's lint red.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
