The bot HTTP API carries the bot key as a URL path segment
(/rooms/:room_id/:bot_key/...). config.filter_parameters redacts query
and form parameters but never path segments, so the key was written
verbatim to the request log (the "Started POST ..." line) and to any
log line echoing the pagination Link header.
Add a log formatter that redacts the bot-key path segment wherever it
appears in a formatted line, and wire it into the production logger.
Action Cable authorizes a Connection once at the WebSocket handshake and
never re-checks it. Destroying the session record refuses future handshakes
and HTTP requests bearing the cookie, but a socket opened before sign out
keeps its handshake-time current_user and keeps authorizing new
subscriptions and delivering frames as the signed-out user.
Reset the user's remote connections when the session is terminated. Clients
tear down and reconnect: the signed-out device carries a destroyed session
and cleared cookie and is rejected at the fresh handshake, while the user's
other devices with still-valid sessions reconnect and stay live. This reuses
the existing reset_remote_connections primitive already used on membership
removal, for the same reason.
Run the disconnect last and best-effort, after the session record and cookie
are already gone, so sign out completes even when the realtime service is
unreachable.
Active Storage mounts its direct-upload write endpoints -- POST
/rails/active_storage/direct_uploads and the disk-service PUT at
/rails/active_storage/disk/:token -- on framework controllers that inherit
from ActiveStorage::BaseController, so they never pass through
ApplicationController's require_authentication. Anyone who can read the
public login page can lift a CSRF token and Rails session cookie, POST to
the metadata endpoint, and receive a signed disk PUT URL without holding a
Campfire session_token.
That is enough to allocate ActiveStorage::Blob rows and persist bytes to
disk anonymously. The blobs stay unattached (no message can be created
without an account) and nothing purges them, so an unauthenticated caller
can grow storage without bound. Because the recommended self-host layout
co-locates uploaded files and the SQLite database on one /rails/storage
volume, that growth eventually makes database writes fail -- blocking login
and messaging until an administrator frees space and purges the blobs.
Campfire uploads attachments through MessagesController as a normal
multipart POST and does not use direct uploads at all, so these endpoints
have no legitimate anonymous caller. Require a valid Campfire session
before the metadata endpoint allocates a blob or the disk endpoint accepts
an upload; both return 401 to anonymous callers. Serving (disk#show,
representations, blob redirects) is unchanged.
* Guard push-subscription endpoints against SSRF
Web push delivery POSTed to the endpoint URL a user supplied when
registering a subscription, with no scheme, host, or private-network
check -- unlike the OpenGraph unfurl path, which already routes through
the shared SSRF address policy (surfguard). Any authenticated user could
register a subscription whose endpoint pointed at an internal address and
have the server fetch it on every chat message: blind SSRF for internal
recon and reachability probing, plus a thread-pool DoS on the delivery
pool.
Validate the endpoint when the subscription is saved: it must be HTTPS,
its host must belong to a known browser push service (allowlist), and it
must resolve to a public IP. On every delivery, re-resolve the host and
pin the connection to that public IP so a later DNS rebind can't redirect
the request to an internal address. If no public IP resolves at delivery
time -- a rebind, or a subscription that predates this validation --
delivery is skipped rather than falling back to re-resolving the raw
host.
Adds model, controller, and delivery-pinning tests, plus a DNS stub
helper for deterministic resolution in tests.
* Route push endpoint guarding through RestrictedHTTP::PrivateNetworkGuard
The endpoint SSRF check already delegated its private-network classification
to surfguard, but called Surfguard.resolve_public_ips directly rather than
through RestrictedHTTP::PrivateNetworkGuard -- the hostname-in, address-out
shim #241 established as the app's single guarded-outbound entry point and
that Opengraph::Fetch resolves through. Route push resolution through the same
guard so once-campfire keeps one place that resolves and classifies outbound
addresses. Behavior is unchanged: the guard raises Violation when a permitted
host resolves only to blocked addresses (previously an empty list -> nil) and
propagates Surfguard::Unresolvable for a host that resolves to nothing; both
map to a nil endpoint IP, which fails validation and skips delivery. The
allowlist, HTTPS/443 constraints, and per-delivery IP pinning are unchanged.
* Address push-SSRF review: defer DNS off enqueue path, disable proxy on pinned path, revalidate on re-registration
- Resolve the guarded endpoint IP lazily inside WebPush::Notification#deliver
(on the bounded delivery worker) instead of eagerly when the notification is
built on the serial enqueue path, so a slow resolver can't stall the push job
before any delivery starts. resolved_endpoint_ip only reads the already-loaded
endpoint attribute, so it is safe off the AR connection.
- Pin the delivery socket with an explicit nil proxy address so http_proxy/
https_proxy can't route the request through a proxy that re-resolves the host
and defeats the ipaddr pin.
- Revalidate an existing subscription on re-registration so a row predating
endpoint validation gets the same 422 as a fresh create instead of being kept
alive by touch.
- Regression tests: resolution deferred to delivery, pin survives proxy env,
legacy invalid row rejected with 422.
* Bound the web-push delivery queue (max_queue, not the ignored queue_size)
Concurrent::ThreadPoolExecutor takes :max_queue; :queue_size was silently
ignored, leaving the delivery backlog unbounded (max_queue: 0). A flood of
valid push subscriptions could accumulate queued deliveries without limit --
more acute now that each delivery task also resolves DNS. Using max_queue: 10000
activates the intended cap; overflow raises RejectedExecutionError under the
default :abort policy, which deliver_later already rescues (push is best-effort,
retried on the next message).
* Trim push-SSRF guard comments to match house style
Apply the review suggestions on the push-subscription SSRF guard: replace
the verbose rationale comments with terse one-liners (or drop them where
the code speaks for itself). No behavior change -- the Surfguard-shim
routing, per-delivery public-IP pin, and bounded delivery queue are
untouched.
Three pins, two values: Dockerfile said 3.4.5, Dockerfile-export said
3.4.7, .ruby-version said 3.4.5 — the export image had already been
bumped on its own and nobody noticed the other two lagging. Both
Dockerfiles carry a comment telling you to keep them in step.
3.4.10 is the current 3.4.x. CI resolves its Ruby from .ruby-version,
so this is the version the tests now run under.
v1.295.0 predates Ruby 3.4.10 — its baked-in version index stops at
3.4.9, so asking for anything newer fails with "Unknown version 3.4.10
for ruby on ubuntu-24.04". The next commit needs 3.4.10, and pinning
our Ruby to whatever an old action happens to know is backwards.
SHA verified against the v1.321.0 tag.
The base image tag is a build arg — `ARG RUBY_VERSION` plus
`FROM ruby:$RUBY_VERSION-slim` — and Dependabot's docker updater matches
literal tags, so this entry has never had anything to propose. It ran
green every week and reported nothing, which reads as coverage and
isn't.
No other repo in the fleet configures a docker ecosystem, and none
could: they all either interpolate a variable or pull from the internal
registry. Inlining the tag here to buy coverage would make this
Dockerfile the outlier instead, against Rails-generated boilerplate.
Ruby bumps stay a manual, human-decided step.
The docker block copied bundler's cooldown wholesale, but Dependabot only
accepts semver-major/minor/patch-days for ecosystems whose versions it
classifies as semver, and container tags aren't. One invalid property
invalidates the entire file rather than the block it sits in, so since
this config landed in #248 the version updater has not run for any
ecosystem at all:
Your .github/dependabot.yml contained invalid details
The property '#/updates/2/cooldown/semver-major-days' is not supported
for the package ecosystem 'docker'. (and -minor-, -patch-)
That is why #249's cooldown exclude for brakeman never took effect, and
why #250 had to bump the workflow linter pins by hand while every other
repo got a Dependabot PR. It also left `bin/brakeman --ensure-latest 15`
armed with nothing to disarm it: the lock pins brakeman 8.0.6, and CI
would have gone red roughly 15 days after 8.0.7 shipped.
Security updates were never affected — those don't read this file, which
is why the only four Dependabot runs here are single-gem security bumps.
docker keeps default-days, which is supported for every ecosystem.
Dependabot version updates are disabled for this repo, so the weekly
github-actions group never proposed these. The pins had been frozen since
March: zizmor-action v0.5.2 ships zizmor 1.23.1, which flags
secrets-outside-env at the default persona -- that moved to the auditor
persona in zizmor 1.24.0. v0.6.2 ships zizmor 1.29.0.
Both linters pass locally at the new versions.
* Take the SSRF address policy from surfguard instead of keeping our own copy
Four other apps carried this same classification and the five had drifted into
four different ideas of what "internal" means. It now comes from the surfguard
gem, which is their union. resolve and Violation keep their shapes, so the
opengraph callers are unchanged.
Two verdicts change.
SIIT (::ffff:0:0:0/96) is now recognised. It is the third way an IPv4 address
rides inside an IPv6 one and the only one Ruby has no predicate for --
ipv4_mapped?, ipv4_compat?, private?, loopback? and link_local? are all false
for ::ffff:0:a9fe:a9fe, so it fell through to the IPv6 branch unrecognised and
reached the metadata endpoint. Note the extra group: ::ffff:0:0:0/96 is not the
IPv4-mapped ::ffff:0:0/96 the guard already refused, and the two do not overlap.
The RFC 8215 local-use NAT64 block is now refused whole rather than decoded.
Reading its low 32 bits as an embedded IPv4 is only correct for a /96 Pref64;
the block can host any length from /32 to /96 and the position is not
recoverable from the address alone (RFC 6052 2.2), so the decode reads the
wrong octets. It is never globally routed, so refusing it costs nothing. The
well-known /96 is still decoded and re-checked, so DNS64 for public sites on
IPv6-only hosts keeps working.
Resolution moves from Resolv.getaddress to Resolv.getaddresses, so the guard
sees every address a host answers with rather than only the first.
* Distinguish a DNS lookup failure from a private-IP block in the guard
Advance the surfguard pin so resolve_public_ips raises Unresolvable when a
host resolves to nothing and returns an empty list only when it resolves to a
blocked address. The shim lets Unresolvable propagate as a lookup failure --
matching the old Resolv.getaddress behavior -- and reserves Violation for a
resolved-but-blocked address, so a transient DNS miss is no longer reported as
an SSRF attempt.
* Bump brakeman to 8.0.6
Brakeman 8.0.6 shipped 2026-08-12: corrected Rails 8.0 EOL date, added
Rails 8.1 and Ruby 4.0 EOL dates, and fixed command-injection false
positives.
brakeman's only runtime dependency is racc and its required ruby is
>= 3.2.0, both unchanged since 7.1.2, so this is a version bump with no
other movement in the graph.
* Give brakeman's --ensure-latest a 15-day grace period
Bare --ensure-latest exits 5 the moment a newer brakeman exists, so a
release turns this build red before anyone has a chance to react. That
is what happened on 2026-08-12 when 8.0.6 shipped.
The flag takes an optional minimum age in days and only complains once
the latest release is at least that old. 15 is the maximum it accepts;
brakeman rejects anything outside 1-15.
* Exempt brakeman from the dependabot cooldown
The grace period on --ensure-latest is only headroom if the bump lands
inside it. A weekly schedule plus a 7-day cooldown can take 14 days to
so much as open the PR, leaving a single day to merge it.
Excluding brakeman from the cooldown caps the delay at the weekly
schedule, comfortably inside the 15 days.
The advisory (email address spoofing via malformed RFC 2047
encoded-words, CVE-2026-63435) landed in ruby-advisory-db on
2026-08-18, turning the gem audit red since the audit pulls a live
advisory DB. 2.9.1 is the patch release for exactly this CVE.
Bots can only create. A lifecycle notification — an alert that fires and then
resolves, a deploy that starts and finishes, a backup that runs — therefore has
to post a second message, and the room becomes an append-only log of states
rather than a view of the current one.
Adds PATCH and DELETE inside the existing bot_key scope, routed to
Messages::ByBotsController. The body is read the way create reads it, so
updating a message is the same request shape as posting one.
No new authorization: both actions already run through ensure_can_administer,
and can_administer? grants access only to a record the user created, so a bot
key reaches that bot's own messages and no others. set_room narrows it again by
looking the room up through the bot's own memberships. A leaked bot key gains
what it could already do by posting: write to rooms that bot belongs to.
update answers head :ok rather than the redirect, which meant extracting the
update and its broadcast into update_message — calling super and then head
would double render, since the parent redirects inside the action. destroy
needs no split, because the parent renders implicitly like create does.
- Use jbuilder instead of hashes
- Use resource instead of direct HTTP verbs
Verbs only make sense if you have one or two routes, if there are
multiple that emulate what resource does then it's better to use resource.
- Paginate using link headers
- Cache responses
* Bump thruster 0.1.15 → 0.1.23
* Scrub disallowed attributes on allowed tags in message rendering
SanitizeTags removes disallowed tags from message presentation, but
attributes on the tags it allows passed through untouched. Extend the
content-filter chain with a SanitizeAttributes filter that runs Rails'
safe-list sanitizer over the remaining markup, stripping event-handler
attributes and unsafe URI schemes as defense-in-depth alongside the
existing Content-Security-Policy.
Running it after SanitizeTags with the same allowed-tags list makes the
sanitizer's tag pass a no-op, preserving SanitizeTags' remove-not-unwrap
semantics while scrubbing attributes. The attribute allowlist is the
standard ActionText set (which keeps attachments intact) plus class,
which presentation styling relies on.
UnreadRoomsChannel streamed from a hardcoded global name, and every message
published its room id to it. Any authenticated user, including one with no
memberships at all, could subscribe and watch which rooms were active and exactly
when, across every closed room and direct conversation on the account. The browser
filters ids it doesn't recognise, but that happens after delivery.
Its sibling ReadRoomsChannel is already scoped per user; this mirrors it, and
message creation fans the notice out to the room's members instead of broadcasting
it to everyone. No message content was exposed either way, only the timing.
Message content is delivered over turbo streams, which ran on the stock
Turbo::StreamsChannel. That channel verifies the signature on the stream name and
nothing else. The name carries no expiry and no binding to a user, so one read off
the page while a member kept working after the membership was revoked.
Revocation made this worse rather than better. Membership#after_destroy_commit
disconnects the user with reconnect: true, and the client replays its subscriptions
on the new socket: RoomChannel re-checks membership and rejects, while the turbo
subscription re-verified only the signature and was accepted.
RoomMessagesChannel re-checks membership on every subscribe, deriving the room from
the verified stream name so there is no parameter to point elsewhere. Since the
subscriber names the channel it wants, the stock channel would otherwise be a way
around that check, so it now turns these stream names away and this is the only door.
Rooms::DirectsController relaxes ensure_can_administer to true, because every
participant in a direct room may administer it. set_room was inherited unscoped,
though, so that relaxation applied to any room the caller was merely a member of:
DELETE /rooms/directs/<id> destroyed open and closed rooms and all their messages.
The same unscoped lookup let a direct room be loaded by the opens and closeds
controllers, where force_room_type promoted it. Promoting a DM to open grants every
user on the account membership and republishes the whole conversation, including the
other participant's messages; converting it to closed lets the initiator revise who
is in it and lock the other participant out.
Each controller now narrows room_scope to the types it may act on. Opens and closeds
keep reach into each other, since converting between them is a feature. Neither can
reach a direct room, and directs can only reach directs.
Room also refuses to change type away from Rooms::Direct, so the invariant holds for
any future caller of becomes! rather than only these two controllers.
By default, ONCE pauses the application container when taking a backup.
This guarantees a consistent snapshot of the data, but for a large
installation and/or a slow destination path (like a network mount), that
pause could be a bit disruptive to live traffic.
But ONCE also support a pause-free path: if a `pre-backup` hook is
present in the container, ONCE will run that and skip the pause. So by
supplying necessary hooks, we can avoid any disruption from backups.
- `pre-backup` runs the existing script/admin/prepare-backup, which uses
SQLite's online backup API to get a consistent snapshop.
- `post-restore` moves our snaphotted file back into place, and removes
the orphaned sidecar files
Nothing in the repository records that unrestricted bot webhook delivery is
deliberate, so the missing private-network guard reads as an oversight next to
the guarded unfurl path. Researchers report it as server-side request forgery,
repeatedly.
and add test coverage for (un)supported file types.
The avatar and logo variants move into the models and return nil for content
types that are no longer variable, so the controllers fall back to the initials
avatar and stock logo icon instead of raising `ActiveStorage::InvariableError`.
The guard uses IPAddr but relied on something else loading it first;
require it explicitly. And the rebinding tests lost their port assertion
when the matchers were loosened for newer Net::HTTP keyword args, so
check the port alongside the IP again.
The RFC1918, loopback, and link-local ranges in DISALLOWED_IPV4
duplicated the private?/loopback?/link_local? checks that run right
before the list scan, so they could never be the deciding factor. Keep
only the ranges the predicates don't catch.
Matches the fizzy guard: the local-use NAT64 prefix (64:ff9b:1::/48,
RFC 8215) embeds an IPv4 target in its low 32 bits just like the
well-known prefix, so run it through the same embedded-IPv4 recheck.
Local-use NAT64 to a public address now resolves (keeping unfurls
working for self-hosters on such networks) while local-use NAT64 to an
internal address stays blocked.
The local-use NAT64 prefix (64:ff9b:1::/48, RFC 8215) embeds an IPv4
target like the well-known prefix does, but at a deployment-chosen
position we can't extract, so block the whole range outright.
Also block the IPv6 benchmarking range (2001:2::/48, RFC 5180) to match
the IPv4 benchmarking block on 198.18.0.0/15.
Net::HTTP now passes an open_timeout: option to TCPSocket.open, so the mock
that matched exact positional arguments no longer matches. Match on the
host instead.
The guard blocked the usual private, loopback, and link-local ranges (and
the IPv4-mapped/-compatible IPv6 forms), but let through NAT64, 6to4, and
Teredo addresses, which can point at an internal IPv4, and CGNAT.
Now it pulls the IPv4 out of a NAT64 address and checks that (so NAT64 to a
public site still works), blocks 6to4 and Teredo outright, and adds the
missing IPv4 and IPv6 ranges.
Jeremy Daer
Jeremy Daer
dependabot[bot]
Mike Dalessio
Stanko Krtalić
Ronald Lokers
Kevin McConnell
Donal McBreen
Donal McBreen