This is a backport of 84e13e9.
We caught it in logs, race condition led to this error:
ActiveRecord::RecordNotUnique
(PG::UniqueViolation: ERROR: duplicate key value violates unique constraint "user_statuses_pkey"
DETAIL: Key (user_id)=(15) already exists.)
The reason the problem happened was that we were checking if a user has status and if not inserting status:
if user_status
...
else
self.user_status = UserStatus.create!(status)
end
The problem is that it's possible that another request will insert status just after we check if status exists and just before our request call `UserStatus.create!(status)`. Using `upsert` fixes the problem because under the hood `upsert` generates the only SQL request that uses "INSERT ... ON CONFLICT DO UPDATE". So we do everything in one SQL query, and that query takes care of resolving possible conflicts.
Currently `Topic#pm_topic_count` is a count of all personal messages tagged for a given tag. As a result, any user with access to PM tags can poll a sensitive tag to determine if a new personal message has been created using that tag even if the user does not have access to the personal message. We classify this as a minor leak in sensitive information.
With this commit, `Topic#pm_topic_count` is hidden from users by default unless the `display_personal_messages_tag_counts` site setting is enabled.
The presence service would retry `/presence/update` requests every second (or immediately in tests) in case where server returns 429 (rate limit) errors. That could lead to infinite spamming (until user refreshed tab/tabs)
Co-authored-by: David Taylor <david@taylorhq.com>
Our fork was needed for OpenSSL 3 and Ruby 2.X compatibility.
The OpenSSL 3 part was merged into the gem for version 3.
Discourse dropped support for Ruby 2.X.
That means we don't need our fork anymore.
Co-authored-by: Rafael dos Santos Silva <xfalcox@gmail.com>
The `git` version in our discourse_test docker image was recently updated to include a permissions check before running any git commands. For this to pass, the owner of the discourse directory needs to match the user running any git commands.
Under GitHub actions, by default the working directory is created with uid=1000 as the owner. We run all our tests as `root`, so this mismatch causes git to raise the permissions error. We can't switch to run the entire workflow as the `discourse (uid=1000)` user because our discourse_test image is not configured to allow `discourse` access to postgres/redis directories. For now, this commit updates the working directory's owner to match the user running the workflow.
Under some situations, we would inadvertently return a public (unauthenticated) result to an authenticated API request. This commit adds the `Api-Key` header to our anonymous cache bypass logic.
The check used to be necessary because we validated the referrer too and
this bypass was a workaround a bug that is present in some browsers that
do not send the correct referrer.
When creating a group membership request, there is no character
limit on the 'reason' field. This can be potentially be used by
an attacker to create enormous amount of data in the database.
Only allow maximum of `50_000` characters for chat drafts. A hidden `max_chat_draft_length` setting can control this limit. A migration is also provided to delete any abusive draft in the database.
The number of drafts loaded on current user has also been limited and ordered by most recent update.
Note that spec files moved are not directly related to the fix.
When a user checks "Open all external links in a new tab" preference
he expects not to be overruled by unrelated text selections.
Yet if text is selected during a link click the link is followed on
the same tab. This change corrects that.
There was an issue where if hashtag-cooked HTML was sent
to the ExcerptParser without the keep_svg option, we would
end up with empty </use> and </svg> tags on the parts of the
excerpt where the hashtag was, in this case when a post
push notification was sent.
Fixed this, and also added a way to only display a plaintext
version of the hashtag for cases like this via PrettyText#excerpt.
When checking whether an existing upload should be secure
based on upload references, do not count deleted posts, since
there is still a reference attached to them. This can lead to
issues where e.g. an upload is used for a post then later on
a custom emoji.
Learn more about skidding here: https://popper.js.org/docs/v2/modifiers/offset/#skidding-1
This change has two goals:
- Fixes an issue when the user had zoomed the viewport and the popper would position on the opposite side
- Makes msg actions arguably more pleasant to the eye by preventing it to be right aligned with the message container
Prior to this fix trashed channels would still prevent a channel with the same slug to be created. This commit generates a new slug on trash and frees the slug for future usage.
The format used for the slug is: `YYYYMMDD-HHMM-OLD_SLUG-deleted` truncated to the max length of a channel name.
Meta topic: https://meta.discourse.org/t/android-keyboard-overlaps-text-when-flagging-with-something-else/249687?u=osama
On Android, it's currently not possible to scroll modals that take input from the user (such as the flagging modal) when the keyboard is open which means that the keyboard can cover up part of the modal with no way for the user to see the covered part without closing the keyboard. This commit adds some CSS to make these modals scrollable when the keyboard is open.
In the group member bulk edit menu we are displaying staff-only options
to non-staff. The requests are blocked by the back-end, so there is no
harm other than to the user experience.
Notably the individual user edit menu is correctly filtering out
unavailable options. This change brings the bulk edit menu in line with
that.
When EmbeddableHost is configured for a specific category and that category is deleted, then EmbeddableHost should be deleted as well.
In addition, migration was added to fix existing data.
This fixes a longstanding issue for sites with the
secure_uploads setting enabled. What would happen is a scenario
like this, since we did not check all places an upload could be
linked to whenever we used UploadSecurity to check whether an
upload should be secure:
* Upload is created and used for site setting, set to secure: false
since site setting uploads should not be secure. Let's say favicon
* Favicon for the site is used inside a post in a private category,
e.g. via a Onebox
* We changed the secure status for the upload to true, since it's been
used in a private category and we don't check if it's originator
was a public place
* The site favicon breaks :'(
This was a source of constant consternation. Now, when an upload is _not_
being created, and we are checking if an existing upload should be
secure, we now check to see what the first record in the UploadReference
table is for that upload. If it's something public like a site setting,
then we will never change the upload to `secure`.
Specifying wildcard characters which also happen to be regex
meta characters for `auto_approve_email_domains`, `allowed_email_domains`
and `blocked_email_domains` site settings currently breaks email
validation.
This change prevents these characters from being specified for these
site settings. It does this by switching the site setting type
from `list` to `host_list`. The `host_list` validator checks for these
characters.
In addition, this change also improves the site setting descriptions and
introduces a migration to fix existing records.
This commit fixes the following issue:
* User creates a post
* Akismet or some other thing like requiring posts to be approved puts
the post in the review queue, deleting it
* Admin approves the post
* Email is never sent to mailing list mode subscribers
We intentionally do not enqueue this for every single post when
recovering a topic (i.e. recovering the first post) since the topics
could have a lot of posts with emails already sent, and we don't want
to clog sidekiq with thousands of notify jobs.
We refer to the channel name rather than title elsewhere
(including the new channel modal), so we should be consistent.
Title is an internal abstraction, since DM channels cannot have
names (currently).
Also change the name field on channel edit to a input type="text"
rather than a textarea, since we don't want a huge input here.
Added in c2013865d7,
this migration was supposed to only turn off the hashtag
setting for existing sites (since that was the old default)
but its doing it for new ones too because we run all migrations
on new sites.
Instead, we should only run this if the first migration was
only just created, meaning its a new site.
If the enable_experimental_hashtag_autocomplete setting is
enabled, then we should autolink hashtag references to the
archived channels (e.g. #blah::channel) for a nicer UX, and
just show the channel name if not (since doing #channelName
can lead to weird inconsistent results).
In Discourse, there are many migration files where we CREATE INDEX CONCURRENTLY which requires us to set disable_ddl_transaction!. Setting disable_ddl_transaction! in a migration file runs the SQL statements outside of a transaction. The implication of this is that there is no ROLLBACK should any of the SQL statements fail.
We have seen lock timeouts occuring when running CREATE INDEX CONCURRENTLY. When that happens, the index would still have been created but marked as invalid by Postgres.
Per the postgres documentation:
> If a problem arises while scanning the table, such as a deadlock or a uniqueness violation in a unique index, the CREATE INDEX command will fail but leave behind an “invalid” index. This index will be ignored for querying purposes because it might be incomplete; however it will still consume update overhead.
> The recommended recovery method in such cases is to drop the index and try again to perform CREATE INDEX CONCURRENTLY . (Another possibility is to rebuild the index with REINDEX INDEX CONCURRENTLY ).
When such scenarios happen, we are supposed to either drop and create the index again or run a REINDEX operation. However, I noticed today that we have not been doing so in Discourse. Instead, we’ve been incorrectly working around the problem by checking for the index existence before creating the index in order to make the migration idempotent. What this potentially mean is that we might have invalid indexes which are lying around in the database which PG will ignore for querying purposes.
This commits adds a migration which queries for all the
invalid indexes in the `public` namespace and reindexes them.
Since the poll post handler runs very early in the post creation
process, it's possible to run the handler on an obiviously invalid post.
This change ensures the post's `raw` value is present before
proceeding.
