The review found the store key readable from a copy of the disk. The
desktop app only added its OS layer with the lock on, the record from
before a lock stayed in LevelDB and opened the mail with no PIN, any
frame could unwrap the PIN record and guess at it, a lock left the key
in the renderer's memory, and a store waiting after a refused login went
to whoever signed in next under the same username@host.
- The desktop app wraps the store key with safeStorage, lock or no lock.
A record without the layer gets it at its next open. A refusal is said
out loud, and only a computer with no key store goes without, which
settings say. So do they for a browser, which has no such layer.
- A record that gained layers, and every record after a changed PIN, asks
for a new key: `sqlite3_backup` into a new file under a new key while
the store keeps working, one write to the record, then the old file
goes. A changed PIN makes a new lock key, since the old PIN's record
stays on disk too. The passkey and OS unlock go with the old key.
- `wrap` and `unwrap` come through `claim()`, and the key under the PIN
takes a secret from main, one counted trip per guess.
- A lock on the desktop ends the renderer rather than reload in it.
- A sign-in compares the waiting store's JMAP account and server first,
and wipes on a mismatch. The account id keeps its form: every store,
setting and server record of an account is keyed by it.
- The 30 days also run in an app that stays open.
Changes an existing assertion: turning the lock off now keeps the
safeStorage layer, by decision, so that test expects it to stay.
Review of the local store export. A cancel, sign-out or wipe during the
final step still downloaded the file; now the sink takes the job's signal
and drops the Blob, and a job that outlasts the wait is forgotten when it
ends. Finished exports are in a list a later wipe or sign-out can empty,
also after the dialog closed.
- `localExportAllowed` uses `scopeDecided`, so a domain or group scope
can't be missed before the subject arrives. It's checked before the
save picker, in the same tick as the click.
- WebKit holds the final Blob in memory: exports stop at 1 GB, and a
larger choice or one over `storage.estimate()` is refused up front.
- An empty folder that refuses during a full listing counts as empty,
only when the store's own count agrees. Any other refusal stops the run.
- A picked file goes on an abort or an all-failed run; if it can't, the
dialog names it. Failure texts come from our strings, not the browser's.
- A selection keeps to the chosen folders on a retry.
The test that the organisation shows as the reason now flips the policy
through the policy store, because the old one replaced the guard and so
never reached the watch that a real change goes through. The spec moves
to an account of its own, dso-export: other specs running at the same
time work in ent-dave, and its folder went missing there mid-run.
A refused login used to delete the account's data, and with it everyone's
S/MIME and OpenPGP keys, also with offline off. The usual cause is a password
changed on another device, which Stalwart 0.16 and 1.0 both answer with
`invalid_grant`, so that lost unsent mail for little protection.
- A refused login signs out at once and hides the data. A new password brings
it back, unsent mail included. Otherwise it goes after 30 days. The sign-in
page gives a count and a date. The 24-hour timer is gone.
- `organisation.devices.wipeOnRefusedLogin`, off by default, asks for deletion
at once, of that account's own data only.
- The session limit no longer counts as a refusal on any route. It hides the
data and says nothing.
- A wipe runs under a Web Lock, tells other windows when it starts and ends,
and queues its report before deleting anything. It reports only once the
deletion worked and no plugin database is left.
- More goes with a wipe: Pro tabs, search history, Files and WebDAV state,
templates, pop-out drafts, shown notifications, push state, and Web Push
with the last account.
- The unsent count is per account, and unknown rather than 0 when the store
can't say. A new row is checked against the account that asked, and a
sign-out drops its row.
Existing tests changed because the rule did: `auth-store-keep-local.test.ts`
expected a wipe on `invalid_grant` and at the session limit, and
`device-wipe.test.ts` expected shared keys to go on a refused login.
The lock is off by default, and an OS policy can require it and set the
timeout. It locks after a time without use counted across all windows and
tabs, on OS screen lock in the desktop app, and from Lock now.
Locking saves open composers, stops the outbox at its next message, closes
the store and stops its worker with every key, then reloads the page, so
nothing of the old page stays in reach. The new page holds the app back at a
Suspense boundary until someone unlocks. Pop-outs get no session while
locked, push notifications name nobody, and subjects leave plain storage.
One random lock key wraps each account's key as a layer. The PIN wraps the
lock key with PBKDF2 at 600,000 iterations, with a wait after wrong tries
that a reload doesn't reset. A passkey with PRF and the desktop's OS storage
wrap it too. Details and limits are in the decisions file.
A persisted account id with no subject yet read as decided: Everyone's
noOffline wiped mail a group scope allowed, and a wider scope went out at
every reload. A public, partial policy counted as the organisation's word,
and turned on the desktop default in a forbidding organisation.
- an organisation's cap on one file now binds a pin
- Clear and the off switch read what would be lost when asked, messages,
changes and drafts, and Clear waits for a connection
- turning one account off survives the last one going off, and off then on
no longer warns of an eviction
- a tab that takes over the store sends no scope until it has heard the policy
- a dropped account is judged by what applies to anyone
- a choice reaches the store once it settles; progress announces phases only
- restored Pro tabs ask for their titles
A forced value could be got around from inside the app. Only a forced
value may now decide what's allowed: the keys that do are forced-only,
so a recommended one no longer acts as forced. With a plugin list a
package can name any id, so uploads are refused and only managed
plugins can match. A server taken off AllowedJmapServers stops its
accounts without signing out: client, push and sync stop, the switcher
drops them, restore and refresh answer 403 with the cookie kept, and
listing it again resumes with no sign-in. OAuth and SSO starts check
the list too.
- `policy.onWipeOfflineData` is gone from the bridge: main runs the wipe
- `KeystoreAllowedOrigins` added, and `POLICY_CONSUMERS` names who acts
on each key or who still owes it
- a Bulwark server's name must be plain after parsing, a wildcard in the
auth lists needs a domain of two labels
One attachment scope: the kept-attachments code, which takes chunks with a
file and spares pinned mail. A draft only on this device stays through a
tightened window. Mail's budget still gives way to what Files keeps.
The flush keeps one `outbox.dropOrphans` caller, the mail branch's daily
one with its 7 days, and drafts' own query says what's needed, so
`addHeldFileSource` and its test are gone. `unsentCount` counts local
drafts, and a draft a waiting send names counts once. The scope check in
`recordBlob` and the permanent-status rule moved to
`lib/store/mail/attachments.ts`, with tests.
Eviction now goes by the base's signal: `storeLost` and the `store-reset`
notice, with `bulwark-store-opened` as the marker and `prune-files` for files
with no key. The storage page and the lost count hang off that. What a user
picks is stored under `bulwark-offline-scope`, the key `offlineScopeFor`
reads, and `email-snapshot` follows the base's rule. `OfflineEnabled` is read
from the desktop bridge when it's there.
The offline section of Settings > Backup & reset now shows, per account, the
space each part takes, the last sync, a first sync's progress with the time
left, the window and attachment choices, and clearing the account. All
conflicts, calendar and contacts too, are listed there and counted in the
banner.
- the organisation's caps beat the user's choice and the desktop default, and
every path that takes a scope goes through `effectiveScope`
- the desktop app keeps mail without asking, says so once, and stays off for
an account where the user turned it off
- `navigator.storage.persist()` is asked on turning it on, and a refusal
warns with a way to ask again; Safari gets the 7-day hint when not installed
- a store the browser deleted is noticed, says what was lost and syncs again
- a full disk says so, with what to do
- those who had offline mode keep it, with the 30 days they had
- Keep offline in the Files app's menu
A draft saved offline now stays on the device under a stand-in id and
goes to the server by itself when it's back. Until now every save was
refused and the composer showed "not saved".
- `lib/store/drafts`: one pending create per draft, however often it's
saved. A draft from the server edited offline is replaced by create,
then destroy, so a failure leaves a copy. Each create has a fixed
Message-ID, looked up in Drafts after a lost response. A refused
draft stays on the device with its error and can be opened. Changed
on another device: both are kept and the user is told.
- `JMAPClient`: stand-in ids in Email/get and Email/set resolve like
calendar's, also when answering from the device. A file attached
offline stays readable, and is uploaded before a save that names it.
- Held files and kept mail files are stored in 1 MB chunks. Orphans are
dropped at the end of a flush, counting what drafts need.
- `lib/store/mail/attachments.ts`: attachments, inline images and raw
messages per scope, newest first, after everything else in a pass,
with eviction that spares pinned mail and never sees held files.
- `mail.rawBlob` and `mail.rawStatus` for an offline export.
Changed test assertions: `outbox.heldFile` and `mail.getBlob` return
`blob` instead of `data`, so a file isn't held whole in memory. The
attachment tests ignore raw downloads, and mail-sync runs with
attachments off, because raw messages add a commit per message.
Any site could POST a form to the share target on load: with no clicks
the composer opened with its subject and file, and autosave stored a
real draft on the server 2 seconds later. The worker can't see where a
POST came from, and Chromium's link capturing accepts form submissions,
so an installed-app window proves nothing either.
- every share landing now asks first, in every window and browser; a
text landing address gets the same question. Nothing goes up before
the answer.
- handed-in text and files are the composer's starting state: no
autosave, page-close save or sign-out flush before the first real
edit.
- the worker ignores frames, keeps at most 5 entries and 100 MB for 15
minutes, sweeps on activate, on each share and from a page timer,
and seals what it keeps with a key passed in the landing fragment.
Offline wipes clear it too.
- the server fallback uses the fragment and caps by encoded length:
3000 Japanese characters made a 27 KB `Location` and a 431.
- a second compose handoff waits until the composer closes, also one
set aside by reading a message, where it used to replace the first.
- a launch leaving the mail page saves what was just typed first.
- an expired handoff or a missing stored file now says so; shares get
the control-character stripping and limits of `parseMailto`.
Existing tests changed because the landing moved from the query to the
fragment, stored entries gained a sealed part, and a share now asks
first.
The IndexedDB cache, its crypto, the responder and the page-side sync
go. lib/offline keeps its structure, policy, wipes, sign-out gates and
notices; answers, the outbox and sync now live in lib/store, in a worker
behind lib/store/client/api.ts.
What users get besides that:
- every folder is kept, not just the Inbox and opened ones
- delete for good and empty trash work offline
- a change the server refuses, or that meets a message changed on
another device, is shown instead of dropped
- a send whose response got lost is looked up by a Message-ID fixed when
it was queued, then finished, submitted or sent again
- with the store on, a start paints from the local copy, then connects
An account the app drops by itself no longer wipes its data: only a
sign-out or an explicit removal does. The outbox of the old database
moves over once, in lib/offline/migrate-legacy.ts, and the old database
goes only after the store committed. `isNetworkFailure` now counts
timeouts and proxy 5xx. Sends and outbox requests are never answered
from the device.
Tests moved with the code: cache.test.ts and responder.test.ts are
replaced by lib/store/__tests__/, and review-fixes.test.ts, outbox.test.ts
and client-fallback.test.ts now run the real engine on node:sqlite.
Changed assertions: an uncertain send is looked up rather than left alone
forever, destroy and empty trash are queued rather than refused, and #9
(opened folders) became "a folder the server deleted goes", since all
folders are kept.
The review of offline mode found ways to send a message twice, to lose
unsent mail without a word, and to send mail without its signing,
encryption or check plugin. This fixes them and the smaller findings:
- each outbox message is taken for sending in one database transaction
under a Web Lock per account; only a failure before the submission
request (SendNotStartedError from the client) puts it back, any other
one marks it "may have gone out"; read-backs after the submission no
longer fail a send; a send of a closed tab turns uncertain after a minute
- every sign-out the user didn't confirm names the deleted messages on
the sign-in page, or in a toast when another account stays signed in
- nothing is queued while an enabled plugin isn't running (after a start
without the network none can); the composer says why
- the outbox waits for a fresh policy and holds messages the rules have
since made stricter (a required label, a required send check)
- another tab switching offline mode off sticks: only switching it on
may create the database
- only attachments of kept messages are kept, filed with their message,
deleted with it and under the cap between syncs; file names and types
are sealed; folders that went away leave the scope, and a folder that
couldn't be asked keeps its mail without holding the others back
- a group scope that forbids offline mode waits until the account's
groups are known; one server down no longer holds another's mail
- an offline start only when the server is really unreachable, and the
real sign-in follows by itself once it answers
- the service worker files pages and chunks under the build that served
them, and the shell list of another build isn't kept under this one
- a raw send counts as the composer's only from the plugin running its
takeover, for that identity
- errors and refusals are in all languages; the banner tells "offline"
from "server can't be reached"; outbox buttons name their message;
changes and messages go in the order they were made
Linus Rath