Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[WIPish] MSC1777: peeking over federation (via server pseudousers) #1777

Closed
wants to merge 5 commits into from
Closed
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
84 changes: 84 additions & 0 deletions proposals/1777-peeking-over-federation.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,84 @@
# Proposal for implementing peeking over federation (server pseudousers)

## Problem

Currently you can't peek over federation, as it was never designed or
implemented due to time constraints when peeking was originally added to Matrix
in 2016.

As well as stopping users from previewing rooms before joining, the fact that
servers can't participate in remote rooms without joining them first is
inconvenient in many other ways:

* You can't reliably participate in E2E encryption in rooms you're invited to
unless the server is actually participating in the room
(https://github.com/vector-im/riot-web/issues/2713)
* Rooms wink out of existence when all participants leave (but if servers were
able to participate even when their users have left, this would go away)
(https://github.com/matrix-org/matrix-doc/issues/534)
* You can't use rooms as generic pubsub mechanisms for synchronising data like
profiles, groups, device-lists etc if you can't peek into them remotely.
* Search engines can't work if they can't peek remote rooms.

## Solution

If a client asks to peek into a room which its server is not currently joined
to, the server should attempt to join the room using a pseudouser account which
represents the server itself.

This allows the server to participate in the room and peek the data being
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should the special user be allowed to post anything in the room? Because they'd be joined and hidden from the memberlist, it opens up a vector for abuse which is difficult to mitigate. Enforcing that the special case user is read only (with the exception of join and leave explicitly) would prevent most forms of abuse.

requested without disclosing the identity of the peeking user, and without
bloating the membership table of the room by joining on behalf of every local
peeking user.

This proposal suggests `@:server` is standardised as the special form of the
server pseudo-user (changing the rules for user IDs to allow zero-length
localparts).

Clients must not show this pseudouser in their membership lists, and the
pseudouser membership event must not be used when calculating room names.
However, clients may choose to show the existence of the member in advanced
details about a given room.

`m.room.join_rules` is extended with a new type: `peekable`, which describes
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How does this work with other join_rules? What if you want knock but not peak? But the other way round? What if you want to be able to peak some things in a private room? etc.

It feels like it might be time to make the join_rules contain a set of features instead?

public-joinable rooms which may also be joined by `@:server` pseudousers.
Otherwise, server pseudousers must not be allowed to join the room, unless a
user from that server has already joined or been invited.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What happens to pseudousers if the join_rules become more restrictive after they have joined it. Does it work like current room semantics and keeps them joined, keeping the room peekable for servers who have peeked it once already?

(i.e I can see the situation where lots of servers peek your room, you downgrade from peekable and are surprised to learn it doesn't help you.)


This replaces the `world_readable` setting on `m.room.history_visibility`.

XXX: Presumably this requires a room version upgrade.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Most definitely. It might be worth the extra effort to split out this particular change into a dedicated proposal so it can make it in to v2, leaving the specifics of how peeking work to this proposal. Only real reason would be we're cutting rooms v2 at the end of the month, and it'd be a bit annoying to have a v3 very shortly after (particularly if the change is just peeking - that's not a lot of motivation to update rooms, particularly when compared to v2's changelog).


This also solves the 'rooms wink out of existence' bug
(https://github.com/matrix-org/matrix-doc/issues/534)
if servers which have aliases pointing to a room also join their pseudouser to
the room in order to keep the room 'alive' (and thus the alias working) even
if everyone leaves.

Pseudousers could potentially also act on behalf of ASes within a room without
the AS having to unpleasantly join/part a bot to interact with it
(https://github.com/matrix-org/matrix-doc/issues/544)

## Security considerations

This has potential to allow users to unilaterally invite servers into their rooms,
which could be a DoS vector. If a user creates a peekable room, and invites a
remote user in, it's now possible for that server to join via their pseudouser
in order to (say) participate in E2E... even if the user themselves hasn't
acted on the invitation. Care must be taken in being lured into peekable rooms.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we're bumping the room version, might as well enforce the user being effectively read-only at the same time.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I assume this would break Pseudousers could potentially also act on behalf of ASes since that would requite the user to write into the room?


## Tradeoffs

We could instead create a new m.room.server_membership event type. But whilst
slightly semantically clearer, it complicates the implementation even more,
whereas here we can leverage most of the existing behaviour of m.room.membership
events.

## Dependencies

This unblocks MSC1769 (profiles as rooms) and MSC1772 (groups as rooms)
and is required for MSC1776 (peeking via /sync) to be of any use.

## History

This would close https://github.com/matrix-org/matrix-doc/issues/913