MSC3083: Restricting room membership based on membership in other rooms

proposals/3083-restricted-rooms.md

@@ -0,0 +1,220 @@
+# Restricting room membership based on space membership
+
+A desirable feature is to give room admins the power to restrict membership of
+their room based on the membership of one or more spaces from
+[MSC1772](https://github.com/matrix-org/matrix-doc/pull/1772), for example:
+
+> members of the #doglovers space can join this room without an invitation<sup id="a1">[1](#f1)</sup>
+
+## Proposal
+
+In a future room version a new `join_rule` (`restricted`) will be used to reflect
+a cross between `invite` and `public` join rules. The content of the join rules
+would include the rooms to trust for membership. For example:
+
+```json
+{
+    "type": "m.room.join_rules",
+    "state_key": "",
+    "content": {
+        "join_rule": "restricted",
+        "allow": [
+            {
+                "type": "m.room_membership",
+                "room_id": "!mods:example.org",
+                "via": ["example.org"]
+            },
+            {
+                "type": "m.room_membership",
+                "room_id": "!users:example.org",
+                "via": ["example.org"]
+            }
+        ]
+    }
+}
+```
+
+This means that a user must be a member of the `!mods:example.org` room or
+`!users:example.org` room in order to join without an invite<sup id="a2">[2](#f2)</sup>.
+Membership in a single room is enough.
+
+If the `allow` key is an empty list (or not a list at all), then no users are
+allowed to join without an invite. Each entry is expected to be an object with the
+following keys:
+
+* `type`: `"m.room_membership"` to describe that we are allowing access via room
+  membership. Future MSCs may define other types.
+* `room_id`: The room ID to check the membership of.
+* `via`: A list of servers which may be used to peek for membership of the room.
+
+Any entries in the list which do not match the expected format are ignored. Thus,
+if all entries are invalid, the list behaves as if empty and all users without
+an invite are rejected.
+
+From the perspective of the [auth rules](https://spec.matrix.org/unstable/rooms/v1/#authorization-rules),
+the `restricted` join rule has the same behavior as `public`. Additional checks
+against the `allow` rules are performed during event generation, as described below.
+
+When a homeserver receives a `/join` request from a client or a `/make_join` / `/send_join`
+request from a server, the request should only be permitted if the user has a valid
+invite or is in one of the listed rooms.
+
+If the user is not a member of at least one of the rooms, the homeserver should
+return an error  response with HTTP status code of 403 and an `errcode` of `M_FORBIDDEN`.
+
+It is possible for a homeserver receiving a `/make_join` / `/send_join` request
+to not know if the user is in a particular room (due to not participating in any
+of the necessary rooms). In this case the homeserver should reject the join,
+the requesting server may wish to attempt to join via other homeservers.
+
+Unlike the `invite` join rule, confirmation that the `allow` rules were properly
+checked cannot be enforced over federation by event authorisation, so servers in
+the room are trusted not to allow invalid users to join.<sup id="a3">[3](#f3)</sup>
+
+## Summary of the behaviour of join rules
+
+See the [join rules](https://matrix.org/docs/spec/client_server/r0.6.1#m-room-join-rules)
+specification for full details, but the summary below should highlight the differences
+between `public`, `invite`, and `restricted`.
+
+* `public`: anyone can join, subject to `ban` and `server_acls`, as today.
+* `invite`: only people with membership `invite` can join, subject to `ban` and
+  `server_acls`, as today.
+* `knock`: the same as `invite`, except anyone can knock, subject to `ban` and
+  `server_acls`. See [MSC2403](https://github.com/matrix-org/matrix-doc/pull/2403).
+* `private`: This is reserved, but unspecified.
+* `restricted`: the same as `public` from the perspective of the [auth rules](https://spec.matrix.org/unstable/rooms/v1/#authorization-rules),
+  but with the additional caveat that servers must check the `allow` rules before
+  generating a `join` event (whether for a local or a remote user).
+
+## Security considerations
+
+The `allow` feature for `join_rules` places increased trust in the servers in the
+room. Any server which is joined to the room will be able to issue join events
+for the room, there are situations which no individual server in the room can
+verify that the membership event was issued in good faith.
+
+We consider this acceptable: if you don't want evil servers randomly joining
+spurious users into your rooms, then:
+
+1. Don't let evil servers in your room in the first place
+2. Don't use `allow` lists, given the expansion increases the attack surface anyway
+   by letting members in other rooms dictate who's allowed into your room.
+
+## Unstable prefix
+
+The `restricted` join rule will be included in a future room version to allow
+servers and clients to opt-into the new functionality.
+
+During development, an unstable room version of `org.matrix.msc3083` will be used.
+Since the room version namespaces the behaviour, the `allow` key and value, as well
+as the `restricted` join rule value do not need unstable prefixes.
+
+## Alternatives
+
+It may seem that just having the `allow` key with `public` join rules is enough
+(as originally suggested in [MSC2962](https://github.com/matrix-org/matrix-doc/pull/2962)),
+but there are concerns that having a `public` join rule that is restricted may
+cause issues if an implementation has not been updated to understand the semantics
+of the `allow` keyword. This could be solved by introducing a new room version,
+but in that case it seems clearer to introduce the `restricted` join rule, as
+described above.
+
+Using an `allow` key with `invite` join rules to broaden who can join was rejected
+as an option since it requires weakening the [auth rules](https://spec.matrix.org/unstable/rooms/v1/#authorization-rules).
+From the perspective of the auth rules, the `restricted` join rule is identical
+to `public` (since the checking of whether a member is in the room is done during
+the call to `/join` or `/make_join` / `/send_join` regardless).
+
+It was also considered to limit servers which can issue join membership events
+to those in the `via` field (or some other list of trusted servers). This is
+undesirable since it would increase centralization (e.g. a server already in the
+room couldn't issue membership events for another user on that server). It is
+unclear that this would significantly increase the security of the room.
+
+## Future extensions
+
+### Checking room membership over federation
+
+If a server is not in a room (and thus doesn't know the membership of a room) it
+cannot enforce membership of a room during a call to `/make_join`, or `/send_join`.
+Peeking over federation,  as described in [MSC2444](https://github.com/matrix-org/matrix-doc/pull/2444),
+could be used to establish if the user is in any of the proper rooms.
+
+Note that there are additional security considerations with this, namely that
+the peek server has significant power. For example, a poorly chosen peek
+server could lie about the room membership and add an `@evil_user:example.org`
+to a room to gain membership to a room.
+
+As iterated above, this MSC recommends rejecting the join, potentially allowing
+the requesting homeserver to retry via another homeserver.
+
+### Kicking users out when they leave the allowed space
+
+In the above example, suppose `@bob:server.example` leaves `!users:example.org`:
+should they be removed from the room? Likely not, by analogy with what happens
+when you switch the join rules from public to invite. Join rules currently govern
+joins, not existing room membership.
+
+It is left to a future MSC to consider this, but some potential thoughts are
+given below.
+
+If you assume that a user *should* be removed in this case, one option is to
+leave the departure up to Bob's server `server.example`, but this places a
+relatively high level of trust in that server. Additionally, if `server.example`
+were offline, other users in the room would still see Bob in the room (and their
+servers would attempt to send message traffic to it).
+
+Another consideration is that users may have joined via a direct invite, not via
+access through a room.
+
+Fixing this is thorny. Some sort of annotation on the membership events might
+help. but it's unclear what the desired semantics are:
+
+* Assuming that users in a given space are *not* kicked when that space is
+  removed from `allow`, are those users then given a pass to remain
+  in the room indefinitely? What happens if the space is added back to
+  `allow` and *then* the user leaves it?
+* Suppose a user joins a room via a space (SpaceA). Later, SpaceB is added to
+  the `allow` list and SpaceA is removed. What should happen when the
+  user leaves SpaceB? Are they exempt from the kick?
+
+It is possible that completely different state should be kept, or a different
+`m.room.member` state could be used in a more reasonable way to track this.
+
+### Inheriting join rules
+
+If you make a parent space invite-only, should that (optionally?) cascade into
+child rooms? This would have some of the same problems as inheriting power levels,
+as discussed in [MSC2962](https://github.com/matrix-org/matrix-doc/pull/2962).
+
+### Additional allow types
+
+Future MSCs may wish to define additional values for the `type` argument, potentially
+restricting access via:
+
+* MXIDs or servers.
+* A shared secret (room password).
+
+These are just examples are not fully thought through for this MSC, but it should
+be possible to add these behaviors in the future.
+
+## Footnotes
+
+<a id="f1"/>[1]: The converse restriction, "anybody can join, provided they are not members
+of the '#catlovers' space" is less useful since:
+
+1. Users in the banned room could simply leave it at any time
+2. This functionality is already partially provided by
+   [Moderation policy lists](https://matrix.org/docs/spec/client_server/r0.6.1#moderation-policy-lists). [↩](#a1)
+
+<a id="f2"/>[2]: Note that there is nothing stopping users sending and
+receiving invites in `public` rooms today, and they work as you might expect.
+The only difference is that you are not *required* to hold an invite when
+joining the room. [↩](#a2)
+
+<a id="f3"/>[3]: This is a marginal decrease in security from the current
+situation. Currently, a misbehaving server can allow unauthorised users to join
+any room by first issuing an invite to that user. In theory that can be
+prevented by raising the PL required to send an invite, but in practice that is
+rarely done. [↩](#a3)