`PRAGMA busy_timeout = 5000` tells SQLite to keep trying for up to 5000 ms
when there is lock contention, rather than aborting immediately. This should
hopefully prevent the vast majority of "database is locked" panics observed
since the async transition.
`PRAGMA synchronous = NORMAL` trades better performance for a small potential
loss in durability (the default is `FULL`). The SQLite docs recommend `NORMAL`
as "a good choice for most applications running in WAL mode".
The added endpoints work the same as in their upstream implementations.
Upstream also implements `/api/ip`. This seems to include the server's public
IP address (the one that should be hidden behind Cloudflare), which doesn't
seem like a great idea.
The current limit of 19 is an artifact of the implementation, which can be
easily rewritten in terms of a more general string generation function.
The new limit is 255 (max value of a `u8`); using a larger type would
probably be overkill.
The default code is 307 (temporary) to make it easier to test different icon
services, but once a service has been decided on, users should ideally switch
to using permanent redirects for cacheability.
If an external icon service is configured, icon requests return an HTTP
redirect to the corresponding icon at the external service.
An external service may be useful for various reasons, such as if:
* The Vaultwarden instance has no external network connectivity.
* The Vaultwarden instance has trouble handling large bursts of icon requests.
* There are concerns that an attacker may probe the instance to try to detect
whether icons for certain sites have been cached, which would suggest that
the instance contains entries for those sites.
* The external icon service does a better job of providing icons than the
built-in fetcher.
Rejecting an emergency access request should transition the grantor/grantee
relationship back into the `Confirmed` state, and the grantor's encrypted key
should remain in escrow rather than being cleared, or else future emergency
access requsts from that grantee will fail.
For one of these flags to be in effect for a cipher, upstream requires all of
(rather than any of) the collections the cipher is in to have that flag set.
Also, some of the logic for loading access restrictions was wrong. I think
that only malicious clients that also had knowledge of the UUIDs of ciphers
they didn't have access to would have been able to take advantage of that.
An incomplete 2FA login is one where the correct master password was provided,
but the 2FA token or action required to complete the login was not provided
within the configured time limit. This potentially indicates that the user's
master password has been compromised, but the login was blocked by 2FA.
Be aware that the 2FA step can usually still be completed after the email
notification has already been sent out, which could be confusing. Therefore,
the incomplete 2FA time limit should be long enough that this situation would
be unlikely. This feature can also be disabled entirely if desired.
When `show_password_hint` is enabled but mail is not configured, the previous
implementation returned a differentiable response for non-existent email
addresses.
Even if mail is enabled, there is a timing side channel since mail is sent
synchronously. Add a randomized sleep to mitigate this somewhat.
When creating a new cipher, the mobile clients seem to set this field to an
invalid value, which causes a warning to be logged:
Error parsing LastKnownRevisionDate '0001-01-01T00:00:00': premature end of input
Avoid this by dropping the `LastKnownRevisionDate` field on cipher creation.
Note: The original Vaultwarden implementation of Bitwarden Send would always
hide the email address, while the upstream implementation would always show it.
Upstream PR: https://github.com/bitwarden/server/pull/1234
Upstream will soon auto-delete trashed items after 30 days, but some people
use the trash as an archive folder, so to avoid unexpected data loss, this
implementation requires the user to explicitly enable auto-deletion.
Adjust checks for max access count, expiration date, and deletion date.
The date checks aren't that important, but the access count check
currently allows one more access than it should.
* The Safari extension apparently now uses the origin `file://` and expects
that to be returned (see bitwarden/browser#1311, bitwarden/server#800).
* The `Access-Control-Allow-Origin` header was reflecting the value of the
`Origin` header without checking whether the origin was actually allowed.
This effectively allows any origin to interact with the server, which
defeats the purpose of CORS.
The org name in the invitation email was made customizable in 8867626, but
the org name is still hardcoded as "bitwarden_rs" in the confirmation email.
Currently, when starting up for the first time (running standalone, outside
of Docker), bitwarden_rs panics when the `openssl` tool isn't able to create
`data/rsa_key.pem` due to the `data` dir not existing. Instead, print a more
helpful error message telling the user to create the directory.
The implementation of the `Manager` user type (#1242) introduced a regression
whereby owner/admin users are incorrectly denied access to certain collection
APIs if their access control for collections isn't set to "access all".
Owner/admin users should always have full access to collection APIs, per
https://bitwarden.com/help/article/user-types-access-control/#access-control:
> Assigning Admins and Owners to Collections via Access Control will only
> impact which Collections appear readily in the Filters section of their
> Vault. Admins and Owners will always be able to access "un-assigned"
> Collections via the Organization view.
The bitwarden_rs code is still cross-compiled exactly as before, but Docker
Buildx is used to rewrite the resulting Docker images with correct platform
metadata (reflecting the target platform instead of the build platform).
Buildx also now handles building and pushing the multi-arch manifest lists.
* For clarity, add `UTC` suffix for datetimes in the `Diagnostics` admin tab.
* Format datetimes in the local timezone in the `Users` admin tab.
* Refactor some datetime code and add doc comments.
Prevent clients from updating a cipher if the local copy is stale.
Validation is only performed when the client provides its last known
revision date; this date isn't provided when using older clients,
or when the operation doesn't involve updating an existing cipher.
Upstream PR: https://github.com/bitwarden/server/pull/994
If org owners/admins set their org access to only include selected
collections, then ciphers from non-selected collections shouldn't
appear in "My Vault". This matches the upstream behavior.
Currently, favorites are tracked at the cipher level. For org-owned ciphers,
this means that if one user sets it as a favorite, it automatically becomes a
favorite for all other users that the cipher has been shared with.
Note that Diesel implements its own parser for MySQL connection URIs, so it
probably doesn't accept the full range of syntax that would be accepted by
MySQL's client libraries, whereas for PostgreSQL, Diesel simply passes the
connection string/URI to PostgreSQL's libpq for processing.
The script works by reading the relevant files from the upstream Bitwarden
source repo and generating a matching JSON file. It could potentially be
integrated into the build/release process, but for now it can be run manually
as needed.
Jeremy LinandDaniel García
