Every page parsed the importmap and joined each of its URLs against the
request's base URL, about a millisecond of CPU a page, though the answer
changes only with the importmap and the host. Keep the last answer, keyed
by both. One entry, so a request with a made-up Host header can't grow it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
If the password changes again between this browser's change and its new session, that session is refused, and the browser is sent to sign in instead of an error.
A password change ends every session, but a sign-in or sign-in link checked
against the old password just before the change could still save its session
just after, and that session outlived the change.
The session now checks, as it's saved, that the user's password is still the
one the request loaded. The save's transaction holds SQLite's write lock from
the start, so a change commits either before the check, which fails it, or
after the insert, and ends this session with the others. Sessions start
through the user's association so the check sees the user that was
authenticated, and a refused sign-in is answered like a wrong password.
The profile asked for the current password when the request's password
field was filled in. The check now follows the save itself: the model
checks the challenge whenever the password digest is about to change,
and the profile always hands it the challenge. Editing a name, avatar
or email without a new password still needs none, and scripts that
reset a password, which set no challenge, are unaffected.
Only a string counts as a challenge, and the rate limit counts every
non-empty one, so what's limited is exactly what gets checked.
The lookup under the write lock repeated the one made before
validation, so the query cache could answer it with that earlier miss
even after another connection had registered the endpoint. It runs
uncached now.
Enabling notifications unsubscribed the browser on any refusal, so a
session signed out between authenticating and saving lost the browser's
subscription too, and the next session had nothing to post. Only a
refused endpoint unsubscribes it now.
Only room pages posted the subscription again, so a browser that
changed its password and closed the profile page went without
notifications until it next opened a room. A small controller in the
layout does it now, still once per session, and the bell is back to how
it was.
Two tabs posting at once after a password change could both miss the
row and insert two, so the endpoint is looked up again under the same
write lock as the session check.
A push subscription belongs to the user, not a session, so a device
signed out by a password change kept getting notifications. Changing
the password now removes every subscription along with the sessions,
in the same transaction.
A browser still signed in keeps its own subscription in its service
worker, so the notifications controller posts it once per session: the
bell names the session by a digest of its id, never its token, and the
browser remembers the last one it posted for. That post never
unsubscribes the browser when the server refuses it.
A request that authenticated just before the change could put a
subscription back after it. The create action now checks the session
still exists and saves in one immediate transaction, after validating
the endpoint outside it.
Changing the password now ends every other session, so whoever holds
one could shut the owner out of their own account with a new password.
The profile asks for the current password whenever the password
changes; without it, or with a wrong one, nothing is saved and the page
says why. Other profile edits don't need it, and the admin
reset-password script still works without it.
Attempts are rate limited like signing in, 10 in 3 minutes per address,
so the current password can't be guessed here instead.
Keeping the session the password was changed from also kept any copy of
its cookie working, and a copied cookie is one of the ways someone else
gets into an account. Every session ends now, and the profile page
signs the browser that changed the password back in with a new one.
The sessions end wherever the password changes, but only the profile
page closed the live connections, so a change from a console left a
signed-out device receiving messages until it dropped. Close them after
commit from User, logging rather than failing if the realtime service
can't be reached, since the password is saved by then.
Changing the password is how someone shuts out whoever else is signed
in to their account, but every session opened with the old password
stayed valid. Now the others end and their live connections close, as
on sign-out. The session the password was changed from stays signed in.
Changing the password outside a request, as from a console, ends every
session.
An asset host's path can contain the asset directory too, as in
https://cdn.example.com/assets, and cutting each URL at the first one would
allow everything the host serves under it. Cut at the last one instead.
Behind a proxy that terminates TLS without X-Forwarded-Proto, Rails sees
http while the page is https, and an http source on a non-default port
doesn't match. A source without a scheme takes the page's own.
Uploads are served from this origin too, so script-src 'self' would run one
that a cache still holds under a script type from before the upload guard.
Application pages now allow each importmap URL up to its asset directory
(relative root and asset host included), the inline importmap by hash, and
the service worker by worker-src.
With DISABLE_CONTENT_SECURITY_POLICY=true there's no application policy,
and the hash callback started an empty one, so pages sent an empty
Content-Security-Policy header. Skip the callback then.
script-src 'self', object-src 'none' and base-uri 'self', enforced. The
two inline scripts the importmap helper emits are allowed by their
SHA-256, and the origins its URLs point to (a configured asset host, in
any form Rails accepts) are allowed by name. There is no per-request
nonce, so cached pages stay valid.
DISABLE_CONTENT_SECURITY_POLICY=true leaves the policy off, for an
install whose proxy or customization injects scripts.
X-Total-Count on GET /rooms/:id/:bot_key/messages was COUNT(*) of the
room on every page. Serve it from rooms.messages_count updated by SQLite
triggers so Rails, bulk SQL, and foreign writers stay in step — without
ActiveRecord counter_cache callbacks those paths skip.
Fixes#309.
Co-authored-by: Thomas Klemm <github@tklemm.eu>
Transfer the C completed-response cache lesson into Rails, keeping authentication, room checks and cookies per request. A persistent read-only SQLite observer detects local and foreign commits and rejects racing admission. Whole-page misses render fresh to avoid stale nested fragments; message ETags reflect token-neutral presentation.
* Derive message DOM ids from the server id, not client_message_id
A message's DOM id was derived from the browser-chosen client_message_id
via a Message#to_key override, so dom_id(message) was
"message_<client_message_id>". Turbo's append de-dups by DOM id, so a room
member who posted a message reusing a victim's client_message_id displaced
the victim's message element in every connected member's live view; editing
the attacker's own message then broadcast onto the victim's presentation id.
Drop the to_key override so every message DOM id and broadcast target derives
from the record's primary key. Two distinct records can no longer share a DOM
id regardless of stored client_message_id, so the collision is impossible with
no data migration and no uniqueness constraint. to_param and the fragment
cache key already used the primary key, so message URLs and per-message cache
entries are unchanged.
The composer's optimistic pending message still uses client_message_id as its
placeholder DOM id, which no longer matches the server broadcast's PK-based id.
Reconcile instead by rendering data-client-message-id on the real message and
having the messages controller drop the matching pending placeholder on
connect. Only client-side placeholders (data-pending-message) are removed, so a
message another member posts reusing the same client_message_id can never
displace a real one through the reconciliation path either.
Also point the boost broadcast target at the PK-based dom_id(message, :boosts)
to match the rebuilt container id.
GHSA-3v99-4vxh-xg84
* Bust cached message fragments rendered with client_message_id DOM ids
The message fragment cache keys on the record and the template digest, and
removing the to_key override changes neither. Fragments cached by an earlier
release would keep their message_<client_message_id> ids, so edit, delete and
boost broadcasts, which now target primary-key ids, would miss those messages
in other members' live views until the cache entry expired.
* Locate messages by record id in the client_message_id collision tests
Assert on data-message-id rather than the new primary-key DOM ids, so the tests
describe the behavior instead of the fix and fail on the vulnerable code for the
real reason. Edit the attacker's message to new text and wait for it to arrive,
so the edit path is exercised rather than passing vacuously. Add a request test
that two messages sharing a client_message_id render as distinct elements.
---------
Co-authored-by: Jeremy Daer <jeremy@37signals.com>
QrCodeController#show is open without signing in and decodes its id as
URL-safe base64. An id that isn't base64 raised ArgumentError, and a
URL longer than a QR code holds raised QRCodeRunTimeError, so both
answered 500 and reached error reporting as server errors. Answer them
with 400 instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwoX7uRy5vFZSNKJhSqMSG
The deadline in Opengraph::Fetch started after the lookup of the pasted
host, and an unfurl looks up more hosts outside any fetch: the canonical
URL's, and the image's, once to check its content type and again to
validate it. Each of those waits as long as the resolver takes to give
up. Move the deadline up to UnfurlLinksController, around everything the
link leads to; Opengraph::Fetch keeps its per-operation timeouts and
makes no retries.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Bj8KnxpTf9sj2Ysa8aLAVa
Room#destroy destroyed every message inside the room's own
transaction, which holds SQLite's write lock until the last one: on a
room with many messages, every other write in the app waited and
failed. The request now takes the room away from its members and
leaves the rest to Room::DestroyJob, which destroys the messages one
at a time, each in its own short transaction, and then the room.
An open room is closed in the request, so that someone who joins the
account before the job ends isn't given it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Bj8KnxpTf9sj2Ysa8aLAVa
Administrators see banned users in the account settings, but only the
first page counted them: the next pages listed active users alone. A
banned member before the page boundary shifted the offset, so one active
member was never listed. Both pages now share User.visible_to.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0142qgjggdJ2KDdGk7RF9Xm9
Since administrators were grouped apart from members (b52c318), the account
settings page loads every user to split them and renders all of them. The
lazy next page still starts at the 501st user, so on an account with more
than 500 people, scrolling down lists those users a second time.
Load administrators on their own and page only members, both on the settings
page and on the pages that follow it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0142qgjggdJ2KDdGk7RF9Xm9
Directs merge Room.directs and order by recency. Shared rooms reuse
with_ordered_room and without_direct_rooms, with the STI filter coming
from Room.without_directs so the join alias stays rooms.
Co-authored-by: Thomas Klemm <github@tklemm.eu>
Chaining without_direct_rooms with with_ordered_room joined rooms as
`room` while still ordering on `rooms.name`. Load shared rooms in one
scope instead.
Co-authored-by: Thomas Klemm <github@tklemm.eu>
Load visible direct memberships ordered by room recency and other
rooms ordered by name, instead of hydrating every membership and
splitting them in Ruby.
Fixes#307.
Co-authored-by: Thomas Klemm <github@tklemm.eu>
Search showed the last 100 matches by created_at, so SQLite collected every
message containing the words and sorted them before keeping a page. A common
word in a large account meant sorting most of its history on every search.
Ordering by the full-text index's rowid, which is the message id, lets SQLite
walk the index from the newest match and stop once the page is full.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JVFo3Lt9T8M5NR7KxVvsZ2
* Preload only uncached messages and reduce rendering overhead
* Keep benchmark summaries without raw JSON results
* Use Ruby benchmark drivers and keep generated results out of the repo