Returning objects in a collection vs. IDs

[mariusor@metalhead.club]

@silverpill do you mean that the "malicious" attachment is not a facsimile of an actual note produced by that actor, but a forgery?

In these cases, I'll agree with
@grishka that some validation based on the ID should be necessary.

For embedded object attachments on the other hand (like mastodon produces), probably the validation needs to check that attributedTo corresponds to the one of the parent object or missing.

Interesting corner case.

@technical-discussion

[silverpill@mitra.social]

@mariusor Yes, a forged note. I've come up with a more realistic example:

{
  "type": "Create",
  "id": "https://social.example/activity/345",
  "actor": "https://social.example/alice"
  "object": {
    "type": "Note",
    "id": "https://social.example/note/123",
    "attributedTo": "https://social.example/alice",
    "content": "This is just a note, nothing to see here",
    "replies": {
      "type": "Collection",
      "id": "https://social.example/note/123/replies",
      "items": [{
        "type": Note",
        "id": "https://social.example/note/987",
        "attributedTo": "https://social.example/bob",
        "inReplyTo": "https://social.example/note/123",
        "content": "Ha ha ha... Yes!"
      }]
    }
  }
}

If the originating server doesn't check the embedded replies collection, a recipient that processes replies and trusts same-origin embeddings unconditionally may end up trusting the forged note.

What we can do?

- Sender: find all embedded objects with local id and reject activity if they are not known.
- Recipient: trust embedded object only if the wrapping object has the same owner.

I think the second solution is much easier to implement. It reduces the utility of embedding in the use case described by @julian, but to be honest I doubt that embedding significantly reduces the number of required HTTP requests in that case.

@grishka

[mariusor@metalhead.club]

> - Recipient: trust embedded object only if the wrapping object has the same owner.

@silverpill no, dereference object and use that instead. The canonical version of an object is the one retrieved from the originating service.

Mastodon has popularised this behaviour where embedding collections (like your replies) is done by servers in the name of "optimizing" for request counts. But this introduces issues and personally I think it's a "code smell" for ActivityPub. Embedding should be restricted to anonymous objects. When an ID exists it should be used most of the time.

@technical-discussion @julian @grishka

[julian@activitypub.space]

mariusor@metalhead.club silverpill@mitra.social C2S brings with it a whole other rat's nest of security concerns.

In an S2S context same origin content ought to be trusted as having been verified. I'd argue a server blindly reflecting received AP content is a vulnerability.

[mariusor@metalhead.club]

@julian I'm not sure what "blindly reflecting" means, but it's at most as vulnerable as using iframes and way less than trusting CDN scripts.

The way GoActivityPub uses C2S is through clients that validate and sanitize content that they serve back to users, or store in a persistence layer.

Personally I don't understand why it would make it different than S2S?

Are you thinking about C2S from a JavaScript client perspective only?

@silverpill

[silverpill@mitra.social]

@mariusor This is basically what my FEP currently recommends: you can trust embedded anonymous objects, fragments and object of Create. Everything else should be authenticated using a different method (e.g. fetched from origin).

@julian @grishka

[mariusor@metalhead.club]

@silverpill oh, I see. I must have missed the context for the discussion, sorry.

@technical-discussion @julian @grishka