The admin page offered "User Plugin Uploads" and "Require Admin
Approval", but nothing in Settings used them: there was no upload, no
uninstall and no way to ask for approval, and the admin had no list of
requests. The approval code that existed pinned a hash of the code only
and showed the admin a manifest the browser sent, never the code.
Now:
- Settings > Plugins has an install button (when uploads are allowed),
a remove button for the user's own plugins, and lists in plain
language what each plugin may do, high-risk items first
- an install sends the ZIP to the server, which reads the manifest and
code itself and pins the request to a hash of both
- the admin Plugins tab lists requests with the code and approves or
denies them; a user may have 10 open requests
- a user plugin is checked again on every start, so a revoked approval
or turning user plugins off takes effect
- the consent dialog and the plugin settings are translated, use the
theme colours, and show the sites a plugin may send data to
The Settings upload, the admin upload, the marketplace install and the
marketplace preview each parsed the ZIP and manifest with their own copy
of the code, and the copies had drifted. A `permissions` string skipped
validation and was then matched by substring, a small ZIP could inflate
to hundreds of megabytes on the server, the Settings upload dropped
`locales` and took origin lists unchecked, and settings schemas were
never checked at all.
All of them now go through `lib/extension-package.ts`:
- every file is inflated through a byte cap, with a limit on entries
- one manifest check: types, permissions, entrypoint, version,
minAppVersion, settingsSchema, configSchema, locales and origin lists
- only checked fields are stored
Users with several accounts can save the list of accounts in their browser
(address, server, label, colour) on the webmail server and bring it back on
another device. The list is encrypted in the browser with a password the user
chooses (PBKDF2-SHA-256, 600,000 iterations, AES-256-GCM, owner bound as
additional data); the server stores only the envelope, one file per owner
under SETTINGS_DATA_DIR/account-archives. No mailbox passwords are included.
Off by default behind the admin feature gate accountArchiveEnabled, pinned
off in Lite. Every method of /api/account-archive needs a session cookie of
the owner, matched without regard to case, whose credential is checked
against the mail server on each call; listing included.
Restored accounts this browser lacks are listed as awaiting sign-in: they
survive a reload, and opening one leads to the sign-in form with the address
filled in. Accounts already present get their archived label and colour.
The agenda sidecar queried only the session's primary calendar account,
so events in a shared or group account (a family or team calendar) never
reached the Calendar Agenda plugin, although the calendar app shows them.
Read every account the calendar app reads: the primary account plus
accounts that advertise the calendar capability or are not personal,
mirroring the client's getCalendarCapableAccountIds(). Event and
calendar ids repeat across accounts, so the DTO id is scoped as
<accountId>:<id> and carries accountId; calendar colours are looked up
per account. A shared account whose calendars cannot be read is skipped
instead of failing the whole agenda; a failure of the primary account is
still reported.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Any failed method in the preview request answered 502, so when the
server named the delivered message and it was read fine, a failing
Inbox query still turned the notification into a generic "New mail".
The named message is now a complete preview on its own.
A response without the optional `total` was also treated as a failure;
it now counts the returned ids.
A JMAP method error comes back with HTTP 200. The preview route read a
failed Mailbox/query or Email/query as an empty Inbox and answered
"nothing unread", so the service worker dropped the push in silence.
It now answers 502, and the worker shows its generic notification.
When the server names the message that arrived and it cannot be read -
deleted since, or filed where this login cannot see it - the route no
longer substitutes the newest unread message in the Inbox, which was
some other mail announced as if it had just arrived. It names no
message; the unread total still comes back for the generic toast.
Some servers build a message's preview from its HTML body without skipping
the <style> element, so a marketing mail's preview - in the list and in its
push notification - can read "@media screen and (min-width:600px){.hide{
display:none!important;…" and never reach the text.
Drop whole leading rules from the preview before showing it; a preview that
ends inside the style sheet was nothing but CSS and becomes empty, so the
usual "no preview" text shows instead. A rule only counts as CSS when its
block holds declarations or further rules, so prose with a brace in it
("the meeting is at {time} tomorrow") is left alone.
Three files carried a literal NUL byte inside a string: the admin
check's cache key in `app/api/admin/auth/route.ts`, the global search
hit key in `lib/global-search/types.ts` and two TNEF fixtures in
`lib/__tests__/tnef.test.ts`. Git treats any file containing a NUL as
binary, so it showed no diffs for them, never normalised their line
endings, and reported every pull request that touched the admin route
as a conflict even when a text merge was clean.
The escape produces the same string at runtime. The admin route was
stored with CRLF while git saw it as binary; it is now stored with LF
like the rest of the tree, so its diff covers the whole file (the only
content change is on line 70).
`/api/auth/sso/complete` only accepted cookie slots 0-4 and quietly
used slot 0 for anything higher. Since the account limit rose to
`MAX_ACCOUNT_SLOTS` (50) on HTTP/2, adding a sixth account through
server-side SSO wrote its refresh token over the first account's cookie,
while the new account was registered on slot 5 with no cookie behind it.
The route and the OAuth callback's pending-slot check now use the same
bound as every other auth route.
Retry interrupted or invalid runtime configuration requests with bounded backoff and per-attempt timeouts. Validate the complete response before caching it, and prevent stale HTTP-cached configuration.
Cover request and body timeouts, shared hook consumers, retry bounds, the server response contract, and both Lite configuration paths.
The app refuses a JMAP server over plain http (loopback aside), yet the
desktop issued a code, reported "Signed in on your phone" once it was
redeemed, and the phone failed. Linking now stops before the step-up
with an explanation.
The token proxy trusted the endpoint sealed into a phone's refresh token
for good: after a server was removed or its provider moved, it kept
posting there with the plain fetch and whichever secret the server id
now maps to. Trust is now re-derived on every renewal; an endpoint the
webmail is no longer configured for gets neither the secret nor the
unguarded fetch.
The step-up for SSO accounts asks the provider for a new login with
`prompt=login`, but a provider that ignores it answers from its existing
session, so anyone at an unlocked desktop could link a phone. When the
ID token's `auth_time` is older than the step-up, the re-auth is now
refused.
The step-up's failure count was keyed on the cookie slot, which the
client chooses, so a session could spread guesses over 50 slots; and
even one key allowed 480 failures a day, well above Stalwart's ban of
100 a day per address, which would ban the webmail's own IP for every
user. Failures now count per server and login name, 5 per 15 minutes
and 20 a day.
The mobile handoff returned the provider's token endpoint and the
webmail's client id as they were. With a confidential client the phone
could not renew (it has no secret), and a provider on another host was
refused by the app outright. The handoff now returns the same bundle a
paired phone gets: such grants renew through `/api/auth/pair/token` with
a sealed refresh token.
"Link Mobile App" only appeared for OAuth sessions, its step-up could
only go through the identity provider (which failed without an OAuth
client and bounced password+TOTP users to the provider's login page),
and it handed the phone the desktop's own refresh token. Revoking that
token on sign-out, or a rotating provider, signed the phone out too; a
confidential client stopped the phone at its first token expiry; and an
identity provider on another host was refused by the app.
The step-up now mints the phone's own grant:
- Password and password+TOTP accounts confirm the current password (and
code) in the dialog. The server uses it once to get an OAuth grant
from Stalwart's structured login for the app's own client, retries
with the webmail's client when the server insists on a registered one,
and falls back to an app password ("Bulwark mobile (paired <date>)")
or, on servers that offer neither, the verified password. Failed
attempts are limited per account so the webmail's address cannot be
banned by guessing through it.
- SSO accounts keep the provider round trip, now with `offline_access`
where offered; its fresh tokens become the phone's grant instead of
being thrown away, and must belong to the account being paired.
- The account and server always come from the slot's verified session
context, never from the request.
Grants live 5 minutes server-side, codes 2 minutes, both single use; the
store is pinned on `globalThis`. Redeem answers `invalid_code`,
`expired_code` or `used_code`, and the desktop polls a separate status
id to replace the QR with "Signed in on your phone". A phone whose
client is confidential, or whose provider is on another host, renews
through `/api/auth/pair/token` with a sealed refresh token, so current
app builds work unchanged. Signing out discards a pending grant but no
longer affects a paired phone.
The dialog shows for every account (not in Lite), with a copyable
sign-in link, a TOTP field when needed, and the app's own "Scan a
sign-in code" wording.
The step-up for linking a device sends the user to the identity
provider with `prompt=login`. The login route accepted that pending
state too, so a code from the re-auth could be spent there and sign this
browser in as whichever account answered the prompt. It now refuses a
re-auth state; only the pairing route completes one.
The callback page also handles its code once. React's development
double effect ran it twice: the second pass no longer saw the re-auth
flag and raced the pairing route for the single-use code.
The password+TOTP login drove Stalwart's structured login (`/api/auth`,
then `/auth/token` with PKCE) inline. Linking the mobile app needs the
same exchange to mint a phone's own grant, so it now lives in
`lib/auth/stalwart-password-login.ts` and the route only maps its
result to the responses it always gave.
The mock JMAP server behind `/api/dev-jmap` accepts any password and
shipped in every image, held back only by `DEV_MOCK_JMAP=true` at
runtime. It now also needs the flag at build time (inlined through
`next.config.ts`), so a build made without it cannot switch the mock on,
and the Docker image drops the route before building.
Several endpoints answered anyone with what fingerprints a deployment:
`/api/health?detailed=true` (versions, memory), the update status
(exact version and the text of a pending security advisory, even with
`loginShowVersion` off), `/api/plugins` (installed plugins and the hosts
they may reach), the admin policy (push relays, sidebar app URLs,
plugin lists) and `/api/config` (every domain each server serves).
Visitors without a session now get:
- the plain liveness answer from `/api/health`;
- the update status only where the login page shows the version, and
never the advisory text;
- a 401 from `/api/plugins`, which the client only syncs after signing
in;
- the policy without plugin lists, push relays and sidebar apps, marked
`X-Bulwark-Policy-Scope: public`; the client fetches the full policy
once it signs in;
- server domain lists only when servers are picked by domain.
A mailbox session, a remembered session cookie or the admin session
counts as signed in. Deployments without a session secret cannot tell
sessions apart and keep the old answers.
A WOPI token carries the user's credentials to the editor server and
stayed valid for its whole six-hour lifetime, so an editor opened
before signing out could still read and save the document afterwards.
Each token now records the browser (a random id in an HttpOnly cookie,
hashed) and account slot that opened it. Signing out records a
revocation time for that browser and slot, or for every slot on a full
sign-out, and WOPI calls with a token minted before it are refused.
Signing out on another device leaves this browser's editors alone.
Revocations live in the state directory, so they survive a restart and
reach replicas sharing the state volume, and are dropped once the
tokens they cover have expired.
`/api/auth/verify` checked any username and password against the JMAP
server without a limit. Every probe reaches the server from the
webmail's own address, so the route was a password oracle and enough
wrong passwords could trip Stalwart's authentication ban against the
webmail itself, locking out every user.
Wrong passwords now use up a budget of five per client and thirty
across all clients per fifteen minutes; right ones are refunded. Over
budget the route answers `inconclusive` and the browser probes the JMAP
server itself, as it did before the pre-check existed.
Used impersonation token ids were only remembered in process memory, so
a captured handoff link could be redeemed a second time within its five
minute lifetime once the process restarted, or on another replica. Each
used id now also leaves a marker file in the state directory, created
exclusively, which every replica sharing the state volume sees; markers
are pruned once the token could no longer verify.
The absolute session limit comes from the eight-hour app password that
Stalwart itself expires.
`/api/favicon` is unauthenticated and answers on the webmail origin,
and it passed the upstream Content-Type through unchecked. Should the
favicon service ever return `image/svg+xml` or `text/html`, opening
that URL on its own would run script as the webmail. Only raster image
types are served now, with `nosniff` and a sandboxing CSP; anything
else falls back to the transparent placeholder.
Clicking a new-mail notification opened `/mail/message/<id>` in
whichever account was active. With several accounts signed in, that
looked the id up in the wrong account: nothing opened, or a different
message did and was marked read.
The preview API now returns the cookie slot of the login that owns the
notified account, the service worker adds it to the permalink (or links
that account's Inbox for a group notification), and the app switches to
the login on that slot before opening the message.
The code exchange, token refresh and revocation POSTs carry the
authorization code, refresh token and client secret, and used a plain
`fetch` that follows redirects. The endpoints come from a validated
discovery document, but a redirect from them (or from an endpoint that
changed after discovery) re-sent the body to a host nobody validated.
These requests now fail on any redirect. A failed refresh still keeps
the refresh token, as for any other outage.
A failed TOTP login returned the first 500 bytes of the upstream error
body. With custom JMAP endpoints allowed and no server configured, the
upstream is any address the caller names, so the route read other
servers' responses on the caller's behalf. The detail is now only
passed on from an admin-configured server.
A Bearer login may claim a bare local part, which is accepted when the
server's `Session.username` has the same local part (TOTP users who
type `john` log in this way). Settings were keyed on that claim, so on a
multi-domain server `john@b.example` could claim `john` and read and
overwrite the synced settings of the `john@a.example` who signs in as
`john` with a password.
`resolveJmapIdentity` now returns the canonical account name for a bare
Bearer claim, the context cookie carries it as `accountName`, and the
settings route keys on it. Basic logins and fully-qualified claims keep
their existing key.
The theme sanitizer was a short denylist. It missed protocol-relative
`url(//host)`, schemes and keywords spelled with CSS escapes
(`url(\68ttps:...)`, `@\69mport`), `image-set("https://...")` and
`@font-face` sources, so a theme could load remote resources from the
app document. `validateThemeSelectors` was never called, so theme CSS
could also target `input[type=password]` or restyle real UI. The admin
routes and the zip validator only sanitized when the safety check
failed, and CSS returned by the `onThemeBeforeApply` plugin hook was
injected as-is.
- every `url()` except a `#fragment` and every `image-set()` is removed
with its arguments; a banned construct spelled with escapes drops the
whole stylesheet
- theme CSS keeps only `:root` and `.dark` rules (no nested rules),
plus `@font-face`, `@keyframes` and `@media`/`@supports` wrappers
- sanitizing is unconditional everywhere, including plugin-transformed
CSS; compiled token values may not contain escapes or `url()`
The Stalwart admin check only bailed while `allowCustomJmapEndpoint` was
on. A context cookie minted while it was on keeps naming the
user-chosen server, and the envelope has no expiry, so after the admin
turned the switch off that server could still answer the admin probe
and mint a Bulwark admin session. The check now also requires the
cookie's server to be an admin-configured one.
Linus Rath
lucamzanon
sudandClaude Fable 5.1
Lucius7nya