Not proxying these requests through camo is a security concern.
Furthermore, on the desktop client, any embed image which is hosted on
a server with an expired or otherwise invalid certificate will trigger
a blocking modal window with no clear source and a confusing error
message; see zulip/zulip-desktop#1119.
Rewrite all `message_embed_image` URLs through camo, if it is enabled.
Supporting URL percent-encoded bytes is possible using `%%20`, but this
is not necessarily very understandable to end-users, even those that
understand percent encoding.
Allow `%20` in linkifier URL format strings, and transform them into
`%%20` in the pattern just before they are applied in markdown
translation. Care must be taken here, such that already-escaped `%`s
are not escaped an extra time.
We do this before rendering, and not before storage, as
a simplification; the JS-side linkifier at present only understands
`%(foo)s` and thus needs no changes, and to avoid an un-escaping pass
before showing in the admin UI.
og:image is supposed to be an absolute URL, but some sites incorrectly
provide a relative URL. In this case, it makes more sense to
interpret it relative to the full page URL after redirects, rather
than relative to just the domain part of the page URL before
redirects.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Zulip attempts to validate that the regular expressions that admins
enter for linkifiers are well-formatted, and only contain a specific
subset of regex grammar. The process of checking these
properties (via a regex!) can cause denial-of-service via
backtracking.
Furthermore, this validation itself does not prevent the creation of
linkifiers which themselves cause denial-of-service when they are
executed. As the validator accepts literally anything inside of a
`(?P<word>...)` block, any quadratic backtracking expression can be
hidden therein.
Switch user-provided linkifier patterns to be matched in the Markdown
processor by the `re2` library, which is guaranteed constant-time.
This somewhat limits the possible features of the regular
expression (notably, look-head and -behind, and back-references);
however, these features had never been advertised as working in the
context of linkifiers.
A migration removes any existing linkifiers which would not function
under re2, after printing them for posterity during the upgrade; they
are unlikely to be common, and are impossible to fix automatically.
The denial-of-service in the linkifier validator was discovered by
@erik-krogh and @yoff, as GHSL-2021-118.
This check was copied from upstream python-markdown's "safe mode"
before they removed that feature. The upstream history is that they
introduced this check in
2db5d1c8e4,
which was not a complete security check, and then added the
immediately following check (with an allowlist of schemes) in
0b4ffbb60e.
Their first, incomplete check provides no security benefit and makes
the code hard to reason about, so we remove it.
This adds the X-Smokescreen-Role header to proxy connections, to track
usage from various codepaths, and enforces a timeout. Timeouts were
kept consistent with their previous values, or set to 5s if they had
none previously.
This commits removes some unnecessary checks for `self.md.zulip_message`,
which were put there historically, as earlier we used to add the additional
properties like mentions_user_ids, alert_words, etc. to Message dict
only. These were later moved to MessageRenderingResult class in commit
75cea329b but the checks weren't removed.
This is important because while rendering the messages imported from
other chat tools (like Rocket.Chat), the Message dict is not passed to
the markdown, due to which the checks for `self.md.zerver_message` fails
and hence, things like user mentions, stream/topic mentions are not
rendered in the imported messages properly.
These changes are all independent of each other; I just didn’t feel
like making dozens of commits for them.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
This way we can stop reading as soon as we get to the body. Also,
send an Accept header, check that the request was actually successful,
use lxml.etree.iterparse instead of a broken hand-rolled state
machine, and support XHTML, all for negative 28 lines of code.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
This reverts commit 1965584eec.
This syntax has a bad interaction with table syntax and needs to be
rethought.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
This adds a new class called MessageRenderingResult to contain the
additional properties we added to the Message object (like alert_words)
as well as the rendered content to ensure typesafe reference. No
behavioral change is made except changes in typing.
This is a preparatory change for adding django-stubs to the backend.
Related: #18777
This should help with #17425, where messages with lots of LaTeX are
lost, due to the large expansion factor.
This isn't a total fix for this - large messages with lots of LaTeX
can still end up larger than 1MB, and rendering could timeout, but
this fix should help significantly.
1MB is still small enough that I don't expect we'll run into any DOS
problems - my testing didn't show any problems rendering messages that
contain ~1MB of LaTeX.
This will offer users who are self-hosting to adjust
this value. Moreover, this will help to reduce the
overall time taken to test `test_markdown.py` (since
this can be now overridden with `override_settings`
Django decorator).
This is done as a prep commit for #18641.
?dl=1 causes Dropbox to send Content-Type: application/binary, which
can’t be interpreted by Camo. Use ?raw=1 instead.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Prior to this, we only supported direct mention to
the user groups. This commit extends that support
to silent mention for the user groups.
A related test case is also added.
Fixes: #11711.
Earlier, USER_GROUP_MENTIONS_RE was:
r"(?<![^\s\'\"\(,:<])@(\*[^\*]+\*)"
For the syntax: *foo*, this was unnecessarily capturing it as
*foo* and the extraction of `foo` was done using another helper
function: `extract_user_group`.
This is now changed as:
r"(?<![^\s\'\"\(,:<])@(\*(?P<match>[^\*]+)\*)"
and extraction of `foo` can be done just by using the named capture
group `match`.
This change also helps to simplify its related code path.
Earlier, MENTIONS_RE was:
r"(?<![^\s\'\"\(,:<])@(?P<silent>_?)(?P<match>\*\*[^\*]+\*\*)"
For the syntax: **foo**, this was unnecessarily capturing it as
**foo** and adding extra operation for the extraction of `foo`.
This is now changed as:
r"(?<![^\s\'\"\(,:<])@(?P<silent>_?)(\*\*(?P<match>[^\*]+)\*\*)"
and extraction of `foo` can be done just by using the named capture
group `match`.
This change also helps to simplify its related code path.
Following the convention, we use uppercase for
regex. Also, `user_group_mentions` is given a
conventional name ending with `*_RE`: `USER_GROUP_MENTIONS_RE`.
A message containing wildcard mention when quoted (which
is turned into a silent mention) or message with silent
wildcard mention notifies the users by sending desktop,
sound, and missed message email notifications. This
is clearly a bug which is fixed by this commit.
Fixes: #18354.
Requesting external images is a privacy risk, so route all external
images through Camo.
Tweaked by tabbott for better test coverage, more comments, and to fix
bugs.
This logic likely never ran due to a combination of bugs.
* Running `maybe_update_markdown_engines` unconditionally meant that
`if md_engine_key in md_engines` was likely always true.
* Introduced in 65838bb: DEFAULT_MARKDOWN_KEY could never be in
md_engines, so should we have ever reached that code path, we'd have
tried to rebuild all markdown engines every time.
And it also wasn't clearly helpful -- because we fetch all linkifiers
for a realm on every request anyway, we don't really save database
queries by doing a bulk fetch on startup, and doing so would likely
result in a material regression to Zulip's overall startup time that
we were creating markdown engines for large numbers of realms in bulk
during process startup.
