with the user ID of the knocking user in the JSON body. Rejecting a knock
over federation has a slight catch, however.
When the knocking user is on another homeserver, the homeserver of the
rejecting user needs to send the `leave` event over federation to the
knocking homeserver. However, this is a bit tricky as it is currently very
difficult to have events from a room propagate over federation when the
receiving homeserver is not in the room. This is due to the remote homeserver
being unable to verify that the event being sent is actually from a
homeserver in the room - and that the homeserver in the room had the required
power level to send it. This is a problem that currently affects other,
similar operations, such as disinviting or unbanning a federated user. In
both cases, they won't be notified as their homeserver is not in the room.
While we could send easily send the leave event as part of a generic
transaction to the remote homeserver, that homeserver would have no way to
validate the `prev_events` and `auth_events` that the event references. We
could send those events over as well, but those will also reference other
events that require validation and so on.
A simple thing we could easily do here is to trust the leave event implicitly
if it is sent by the homeserver we originally knocked through. We know this
homeserver is (or at least was) in the room, so they're probably telling the
truth. This is almost an edge case though, as while you'll knock through one
homeserver in the room, there's no guarantee that the admin that denies your
knock will be on the same homeserver you knocked through. Perhaps the
homeserver you knocked through could listen for this and then send the event
back to you - but what if it goes offline in the meantime?
As such, informing remote homeservers about the rejection of knocks over
federation is de-scoped for now, and left to a future MSC which can solve
this class of problem in a suitable way. Rejections should still work for the
homeservers that are in the room, as they can validate the leave event for
they have access to the events it references.
### Membership change to `ban`
The knock has been rejected by someone in the room and the user has been
banned, and is unable to send further knocks.
This one is fairly straightforward. Someone with the appropriate power levels
in the room bans the user. This will have the same effect as rejecting the
knock, and in addition prevent any further knocks by this user from being
allowed into the room.
If the user is unbanned, they will be able to send a new knock which could be
accepted.
To ban the user, the client should call [`POST
/_matrix/client/r0/rooms/{roomId}/ban`](https://matrix.org/docs/spec/client_server/r0.6.1#post-matrix-client-r0-rooms-roomid-ban) with the user ID of the knocking user in the JSON body.
Informing the knocking user about the update is the same as rejecting the
knock.
## Client-Server API
A new endpoint is introduced in the Client-Server API: `POST
/_matrix/client/r0/knock/{roomIdOrAlias}`. This allows the client to state
@ -385,9 +227,10 @@ the underlying push rules architecture can support it.
## Server-Server API
Similarly to join and leave over federation, a ping-pong game with two new
endpoints is introduced: `make_knock` and `send_knock`. Both endpoints must
be protected via server ACLs.
Similarly to [join](https://matrix.org/docs/spec/server_server/r0.1.4#joining-rooms)
and [leave](https://matrix.org/docs/spec/server_server/r0.1.4#leaving-rooms-rejecting-invites)
over federation, a ping-pong game with two new endpoints is introduced: `make_knock`
and `send_knock`. Both endpoints must be protected via server ACLs.
with the user ID of the knocking user in the JSON body. Rejecting a knock
over federation has a slight catch, however.
When the knocking user is on another homeserver, the homeserver of the
rejecting user needs to send the `leave` event over federation to the
knocking homeserver. However, this is a bit tricky as it is currently very
difficult to have events from a room propagate over federation when the
receiving homeserver is not in the room. This is due to the remote homeserver
being unable to verify that the event being sent is actually from a
homeserver in the room - and that the homeserver in the room had the required
power level to send it. This is a problem that currently affects other,
similar operations, such as disinviting or unbanning a federated user. In
both cases, they won't be notified as their homeserver is not in the room.
While we could send easily send the leave event as part of a generic
transaction to the remote homeserver, that homeserver would have no way to
validate the `prev_events` and `auth_events` that the event references. We
could send those events over as well, but those will also reference other
events that require validation and so on.
A simple thing we could easily do here is to trust the leave event implicitly
if it is sent by the homeserver we originally knocked through. We know this
homeserver is (or at least was) in the room, so they're probably telling the
truth. This is almost an edge case though, as while you'll knock through one
homeserver in the room, there's no guarantee that the admin that denies your
knock will be on the same homeserver you knocked through. Perhaps the
homeserver you knocked through could listen for this and then send the event
back to you - but what if it goes offline in the meantime?
As such, informing remote homeservers about the rejection of knocks over
federation is de-scoped for now, and left to a future MSC which can solve
this class of problem in a suitable way. Rejections should still work for the
homeservers that are in the room, as they can validate the leave event for
they have access to the events it references.
### Membership change to `ban`
The knock has been rejected by someone in the room and the user has been
banned, and is unable to send further knocks.
This one is fairly straightforward. Someone with the appropriate power levels
in the room bans the user. This will have the same effect as rejecting the
knock, and in addition prevent any further knocks by this user from being
allowed into the room.
If the user is unbanned, they will be able to send a new knock which could be
accepted.
To ban the user, the client should call [`POST
/_matrix/client/r0/rooms/{roomId}/ban`](https://matrix.org/docs/spec/client_server/r0.6.1#post-matrix-client-r0-rooms-roomid-ban) with the user ID of the knocking user in the JSON body.
Informing the knocking user about the update is the same as rejecting the
knock.
# Potential issues
This new feature would allow users to send events into rooms that they don't
partake in. That is why this proposal enables the a `knock` join rule, in