matrix-spec-proposals/proposals/2828-restrict-user-ids.md

Proposal to restrict allowed user IDs over federation

Currently the spec has an appendix specifying allowed user ID grammar. The spec makes it seem like user IDs can only consist of printable ASCII characters, excluding space and :. However, the user ID grammar is not actually a part of the event authorization rules, meaning most server implementations will accept user IDs with arbitrary unicode characters.

Synapse and Dendrite will accept arbitrary unicode, while Construct and Conduit believed the spec appendix.

Proposal

The proposed solution is applying the historical user ID grammar in step 5 of the room v1 event authorization rules. A new step is added between 5.a. and 5.b.:

If the localpart of the user ID in the state_key contains characters not allowed by the extended_user_id_char grammar, reject.

extended_user_id_char = %x21-39 / %x3B-7E  ; all ascii printing chars except : and space

The changes are applied on top of room v6.

Note about the user ID grammar spec

The user ID grammar spec appendix needs to be updated to mention that room versions prior to this new version can contain arbitrary unicode user IDs, so clients and servers must be able to handle such users.

Alternatives

Don't restrict user IDs

We could simply add the note about arbitrary unicode being allowed, and not create a new room version that restricts user IDs in event auth. Something like MSC1228 will eventually overhaul the entire user ID system anyway.

Potential issues

Existing users with user IDs outside the historical user ID grammar would be unable to join rooms after this new room version. However, there shouldn't be any actual users with such user IDs, as normal registrations have always required usernames to follow the historical grammar.

Unstable prefix

Implementations can use net.maunium.msc2828 as the room version until this proposal is added to an official room version.