Commit Graph
5 Commits
Author SHA1 Message Date
Juliaanddakkar 583df3ec63 Merge commit from fork
none of our endpoints will ever contain `..` (they might, maybe, at
some point, contain `.`, as in `something/get.html`?), so every
`Mk:api()` call to an endpoint that contains `..` can't work: let's
reject it outright

Co-authored-by: dakkar <dakkar@thenautilus.net>
2025-04-29 18:06:39 +09:00
Juliaanddakkar d10fdfe973 Merge commit from fork
* SP-2025-03.1 always wrap icon&thumbnail URLs

if they're not HTTP URLs, the frontend won't be able to display them
anyway (`<img src="mailto:…">` or '<div stile="background-image:
url(nntp:…)">` aren't going to work!), so let's always run them through the
media proxy, which will fail harder (fetching a `javascript:` URL
won't do anything in the backend, might do something in the frontend)
and will always protect the client's address in cases like `gemini:`
where the browser could try to fetch

* SP-2025-03.2 use object binding for more styles

interpolating a random (remote-controlled!) string into a `style`
attribute is a bad idea; using VueJS object binding, we should get
proper quoting and therefore safe parse failures instead of CSS
injections / XSS

* SP-2025-03.3 slightly more robust "self" URL handling

parse URLs instead of treating them as strings; this is still not
perfect, but the `URL` class only handles full URLs, not relative
ones, so there's so way to ask it "give me a URL object that
represents this resource relative to this base URL"

notice that passing very weird URLs to `MkUrl` and `MkUrlPreview` will
break the frontend (in dev mode) because there's an untrapped `new
URL(…)` that may explode; production builds seem to safely ignore the
error, though

---------

Co-authored-by: dakkar <dakkar@thenautilus.net>
2025-04-29 08:15:54 +09:00
Juliaandsyuilo b9cb949eb1 Merge commit from fork
* Fix poll update spoofing

* fix: Disallow negative poll counts

---------

Co-authored-by: syuilo <4439005+syuilo@users.noreply.github.com>
2024-11-21 08:24:50 +09:00
5f675201f2 Merge commit from fork
* enhance: Add a few validation fixes from Sharkey

See the original MR on the GitLab instance:
https://activitypub.software/TransFem-org/Sharkey/-/merge_requests/484

Co-Authored-By: Dakkar <dakkar@thenautilus.net>

* fix: primitive 2: acceptance of cross-origin alternate

Co-Authored-By: Laura Hausmann <laura@hausmann.dev>

* fix: primitive 3: validation of non-final url

* fix: primitive 4: missing same-origin identifier validation of collection-wrapped activities

* fix: primitives 5 & 8: reject activities with non
string identifiers

Co-Authored-By: Laura Hausmann <laura@hausmann.dev>

* fix: primitive 6: reject anonymous objects that were fetched by their id

* fix: primitives 9, 10 & 11: http signature validation
doesn't enforce required headers or specify auth header name

Co-Authored-By: Laura Hausmann <laura@hausmann.dev>

* fix: primitive 14: improper validation of outbox, followers, following & shared inbox collections

* fix: code style for primitive 14

* fix: primitive 15: improper same-origin validation for
note uri and url

Co-Authored-By: Laura Hausmann <laura@hausmann.dev>

* fix: primitive 16: improper same-origin validation for user uri and url

* fix: primitive 17: note same-origin identifier validation can be bypassed by wrapping the id in an array

* fix: code style for primitive 17

* fix: check attribution against actor in notes

While this isn't strictly required to fix the exploits at hand, this
mirrors the fix in `ApQuestionService` for GHSA-5h8r-gq97-xv69, as a
preemptive countermeasure.

* fix: primitive 18: `ap/get` bypasses access checks

One might argue that we could make this one actually preform access
checks against the returned activity object, but I feel like that's a
lot more work than just restricting it to administrators, since, to me
at least, it seems more like a debugging tool than anything else.

* fix: primitive 19 & 20: respect blocks and hide more

Ideally, the user property should also be hidden (as leaving it in leaks
information slightly), but given the schema of the note endpoint, I
don't think that would be possible without introducing some kind of
"ghost" user, who is attributed for posts by users who have you blocked.

* fix: primitives 21, 22, and 23: reuse resolver

This also increases the default `recursionLimit` for `Resolver`, as it
theoretically will go higher that it previously would and could possibly
fail on non-malicious collection activities.

* fix: primitives 25-33: proper local instance checks

* revert: fix: primitive 19 & 20

This reverts commit 465a9fe6591de90f78bd3d084e3c01e65dc3cf3c.

---------

Co-authored-by: Dakkar <dakkar@thenautilus.net>
Co-authored-by: Laura Hausmann <laura@hausmann.dev>
Co-authored-by: syuilo <4439005+syuilo@users.noreply.github.com>
2024-11-21 08:20:09 +09:00
Julia 6fd4de246c Make post form attachments accessible (#14666)
* fix(frontend): Make post form attachments accessible

Adds a role="button", tabindex, and @keydown to MkPostFormAttaches in
order to make it accessible to keyboard users.

* Fix for linter

* Add spacing in type signature
2024-10-03 09:09:08 +09:00