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

Fully specify room versions #3432

Merged
merged 14 commits into from
Nov 1, 2021
4 changes: 3 additions & 1 deletion content/client-server-api/_index.md
Original file line number Diff line number Diff line change
Expand Up @@ -1598,7 +1598,9 @@ content from the databases. Servers should include a copy of the
when serving the redacted event to clients.

The exact algorithm to apply against an event is defined in the [room
version specification](/rooms).
version specification](/rooms), as are the criteria homeservers should
use when deciding whether to accept a redaction event from a remote
homeserver.

When a client receives an `m.room.redaction` event, it should change
the affected event in the same way a server does.
Expand Down
144 changes: 144 additions & 0 deletions content/rooms/fragments/v1-auth-rules.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,144 @@
The types of state events that affect authorization are:

- `m.room.create`
- `m.room.member`
- `m.room.join_rules`
- `m.room.power_levels`
- `m.room.third_party_invite`

{{% boxes/note %}}
Power levels are inferred from defaults when not explicitly supplied.
For example, mentions of the `sender`'s power level can also refer to
the default power level for users in the room.
{{% /boxes/note %}}

The rules are as follows:

1. If type is `m.room.create`:
1. If it has any previous events, reject.
2. If the domain of the `room_id` does not match the domain of the
`sender`, reject.
3. If `content.room_version` is present and is not a recognised
version, reject.
4. If `content` has no `creator` field, reject.
5. Otherwise, allow.
2. Reject if event has `auth_events` that:
1. have duplicate entries for a given `type` and `state_key` pair
2. have entries whose `type` and `state_key` don't match those
specified by the [auth events
selection](/server-server-api#auth-events-selection)
algorithm described in the server specification.
3. If event does not have a `m.room.create` in its `auth_events`,
reject.
4. If type is `m.room.aliases`:
1. If event has no `state_key`, reject.
2. If sender's domain doesn't matches `state_key`, reject.
3. Otherwise, allow.
5. If type is `m.room.member`:
1. If no `state_key` key or `membership` key in `content`, reject.
2. If `membership` is `join`:
1. If the only previous event is an `m.room.create` and the
`state_key` is the creator, allow.
2. If the `sender` does not match `state_key`, reject.
3. If the `sender` is banned, reject.
4. If the `join_rule` is `invite` then allow if membership
state is `invite` or `join`.
5. If the `join_rule` is `public`, allow.
6. Otherwise, reject.
3. If `membership` is `invite`:
1. If `content` has `third_party_invite` key:
1. If *target user* is banned, reject.
2. If `content.third_party_invite` does not have a `signed`
key, reject.
3. If `signed` does not have `mxid` and `token` keys,
reject.
4. If `mxid` does not match `state_key`, reject.
5. If there is no `m.room.third_party_invite` event in the
current room state with `state_key` matching `token`,
reject.
6. If `sender` does not match `sender` of the
`m.room.third_party_invite`, reject.
7. If any signature in `signed` matches any public key in
the `m.room.third_party_invite` event, allow. The public
keys are in `content` of `m.room.third_party_invite` as:
1. A single public key in the `public_key` field.
2. A list of public keys in the `public_keys` field.
8. Otherwise, reject.
2. If the `sender`'s current membership state is not `join`,
reject.
3. If *target user*'s current membership state is `join` or
`ban`, reject.
4. If the `sender`'s power level is greater than or equal to
the *invite level*, allow.
5. Otherwise, reject.
4. If `membership` is `leave`:
1. If the `sender` matches `state_key`, allow if and only if
that user's current membership state is `invite` or `join`.
2. If the `sender`'s current membership state is not `join`,
reject.
3. If the *target user*'s current membership state is `ban`,
and the `sender`'s power level is less than the *ban level*,
reject.
4. If the `sender`'s power level is greater than or equal to
the *kick level*, and the *target user*'s power level is
less than the `sender`'s power level, allow.
5. Otherwise, reject.
5. If `membership` is `ban`:
1. If the `sender`'s current membership state is not `join`,
reject.
2. If the `sender`'s power level is greater than or equal to
the *ban level*, and the *target user*'s power level is less
than the `sender`'s power level, allow.
3. Otherwise, reject.
6. Otherwise, the membership is unknown. Reject.
6. If the `sender`'s current membership state is not `join`, reject.
7. If type is `m.room.third_party_invite`:
1. Allow if and only if `sender`'s current power level is greater
than or equal to the *invite level*.
8. If the event type's *required power level* is greater than the
`sender`'s power level, reject.
9. If the event has a `state_key` that starts with an `@` and does not
match the `sender`, reject.
10. If type is `m.room.power_levels`:
1. If `users` key in `content` is not a dictionary with keys that
are valid user IDs with values that are integers (or a string
that is an integer), reject.
2. If there is no previous `m.room.power_levels` event in the room,
allow.
3. For the keys `users_default`, `events_default`, `state_default`,
`ban`, `redact`, `kick`, `invite` check if they were added,
changed or removed. For each found alteration:
1. If the current value is higher than the `sender`'s current
power level, reject.
2. If the new value is higher than the `sender`'s current power
level, reject.
4. For each entry being added, changed or removed in both the
`events` and `users` keys:
1. If the current value is higher than the `sender`'s current
power level, reject.
2. If the new value is higher than the `sender`'s current power
level, reject.
5. For each entry being changed under the `users` key, other than
the `sender`'s own entry:
1. If the current value is equal to the `sender`'s current
power level, reject.
6. Otherwise, allow.
11. If type is `m.room.redaction`:
1. If the `sender`'s power level is greater than or equal to the
*redact level*, allow.
2. If the domain of the `event_id` of the event being redacted is
the same as the domain of the `event_id` of the
`m.room.redaction`, allow.
3. Otherwise, reject.
12. Otherwise, allow.

{{% boxes/note %}}
Some consequences of these rules:

- Unless you are a member of the room, the only permitted operations
(apart from the initial create/join) are: joining a public room;
accepting or rejecting an invitation to a room.
- To unban somebody, you must have power level greater than or equal
to both the kick *and* ban levels, *and* greater than the target
user's power level.
{{% /boxes/note %}}
3 changes: 3 additions & 0 deletions content/rooms/fragments/v1-canonical-json.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
Servers MUST NOT strictly enforce the JSON format specified in the
[appendices](/appendices#canonical-json) for the reasons
described there.
29 changes: 29 additions & 0 deletions content/rooms/fragments/v1-redactions.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
Upon receipt of a redaction event, the server must strip off any keys
not in the following list:

- `event_id`
- `type`
- `room_id`
- `sender`
- `state_key`
- `content`
- `hashes`
- `signatures`
- `depth`
- `prev_events`
- `prev_state`
- `auth_events`
- `origin`
- `origin_server_ts`
- `membership`

The content object must also be stripped of all keys, unless it is one
of one of the following event types:

- `m.room.member` allows key `membership`.
- `m.room.create` allows key `creator`.
- `m.room.join_rules` allows key `join_rule`.
- `m.room.power_levels` allows keys `ban`, `events`, `events_default`,
`kick`, `redact`, `state_default`, `users`, `users_default`.
- `m.room.aliases` allows key `aliases`.
- `m.room.history_visibility` allows key `history_visibility`.
165 changes: 165 additions & 0 deletions content/rooms/fragments/v2-state-res.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,165 @@
The room state *S*′(*E*) after an event *E* is defined in terms of the
room state *S*(*E*) before *E*, and depends on whether *E* is a state
event or a message event:

- If *E* is a message event, then *S*′(*E*) = *S*(*E*).
- If *E* is a state event, then *S*′(*E*) is *S*(*E*), except that its
entry corresponding to *E*'s `event_type` and `state_key` is
replaced by *E*'s `event_id`.

The room state *S*(*E*) before *E* is the *resolution* of the set of
states {*S*′(*E*<sub>1</sub>), *S*′(*E*<sub>2</sub>), …} consisting of
the states after each of *E*'s `prev_event`s
{*E*<sub>1</sub>, *E*<sub>2</sub>, …}, where the resolution of a set of
states is given in the algorithm below.

#### Definitions

The state resolution algorithm for version 2 rooms uses the following
definitions, given the set of room states
{*S*<sub>1</sub>, *S*<sub>2</sub>, …}:

Power events
A *power event* is a state event with type `m.room.power_levels` or
`m.room.join_rules`, or a state event with type `m.room.member` where
the `membership` is `leave` or `ban` and the `sender` does not match the
`state_key`. The idea behind this is that power events are events that
might remove someone's ability to do something in the room.

Unconflicted state map and conflicted state set
The *unconflicted state map* is the state where the value of each key
exists and is the same in each state *S*<sub>*i*</sub>. The *conflicted
state set* is the set of all other state events. Note that the
unconflicted state map only has one event per `(event_type, state_key)`,
whereas the conflicted state set may have multiple events.

Auth difference
The *auth difference* is calculated by first calculating the full auth
chain for each state *S*<sub>*i*</sub>, that is the union of the auth
chains for each event in *S*<sub>*i*</sub>, and then taking every event
that doesn't appear in every auth chain. If *C*<sub>*i*</sub> is the
full auth chain of *S*<sub>*i*</sub>, then the auth difference is
 ∪ *C*<sub>*i*</sub> −  ∩ *C*<sub>*i*</sub>.

Full conflicted set
The *full conflicted set* is the union of the conflicted state set and
the auth difference.

Reverse topological power ordering
The *reverse topological power ordering* of a set of events is the
lexicographically smallest topological ordering based on the DAG formed
by auth events. The reverse topological power ordering is ordered from
earliest event to latest. For comparing two topological orderings to
determine which is the lexicographically smallest, the following
comparison relation on events is used: for events *x* and *y*,
*x* &lt; *y* if

1. *x*'s sender has *greater* power level than *y*'s sender, when
looking at their respective `auth_event`s; or
2. the senders have the same power level, but *x*'s `origin_server_ts`
is *less* than *y*'s `origin_server_ts`; or
3. the senders have the same power level and the events have the same
`origin_server_ts`, but *x*'s `event_id` is *less* than *y*'s
`event_id`.

The reverse topological power ordering can be found by sorting the
events using Kahn's algorithm for topological sorting, and at each step
selecting, among all the candidate vertices, the smallest vertex using
the above comparison relation.

Mainline ordering
Given an `m.room.power_levels` event *P*, the *mainline of* *P* is the
list of events generated by starting with *P* and recursively taking the
`m.room.power_levels` events from the `auth_events`, ordered such that
*P* is last. Given another event *e*, the *closest mainline event to*
*e* is the first event encountered in the mainline when iteratively
descending through the `m.room.power_levels` events in the `auth_events`
starting at *e*. If no mainline event is encountered when iteratively
descending through the `m.room.power_levels` events, then the closest
mainline event to *e* can be considered to be a dummy event that is
before any other event in the mainline of *P* for the purposes of
condition 1 below.

The *mainline ordering based on* *P* of a set of events is the ordering,
from smallest to largest, using the following comparison relation on
events: for events *x* and *y*, *x* &lt; *y* if

1. the closest mainline event to *x* appears *before* the closest
mainline event to *y*; or
2. the closest mainline events are the same, but *x*'s
`origin_server_ts` is *less* than *y*'s `origin_server_ts`; or
3. the closest mainline events are the same and the events have the
same `origin_server_ts`, but *x*'s `event_id` is *less* than *y*'s
`event_id`.

Iterative auth checks
The *iterative auth checks algorithm* takes as input an initial room
state and a sorted list of state events, and constructs a new room state
by iterating through the event list and applying the state event to the
room state if the state event is allowed by the [authorization
rules](/server-server-api#authorization-rules).
If the state event is not allowed by the authorization rules, then the
event is ignored. If a `(event_type, state_key)` key that is required
for checking the authorization rules is not present in the state, then
the appropriate state event from the event's `auth_events` is used if
the auth event is not rejected.

#### Algorithm

The *resolution* of a set of states is obtained as follows:

1. Take all *power events* and any events in their auth chains,
recursively, that appear in the *full conflicted set* and order them
by the *reverse topological power ordering*.
2. Apply the *iterative auth checks algorithm*, starting from the
*unconflicted state map*, to the list of events from the previous
step to get a partially resolved state.
3. Take all remaining events that weren't picked in step 1 and order
them by the mainline ordering based on the power level in the
partially resolved state obtained in step 2.
4. Apply the *iterative auth checks algorithm* on the partial resolved
state and the list of events from the previous step.
5. Update the result by replacing any event with the event with the
same key from the *unconflicted state map*, if such an event exists,
to get the final resolved state.

#### Rejected events

Events that have been rejected due to failing auth based on the state at
the event (rather than based on their auth chain) are handled as usual
by the algorithm, unless otherwise specified.

Note that no events rejected due to failure to auth against their auth
chain should appear in the process, as they should not appear in state
(the algorithm only uses events that appear in either the state sets or
in the auth chain of the events in the state sets).

{{% boxes/rationale %}}
This helps ensure that different servers' view of state is more likely
to converge, since rejection state of an event may be different. This
can happen if a third server gives an incorrect version of the state
when a server joins a room via it (either due to being faulty or
malicious). Convergence of state is a desirable property as it ensures
that all users in the room have a (mostly) consistent view of the state
of the room. If the view of the state on different servers diverges it
can lead to bifurcation of the room due to e.g. servers disagreeing on
who is in the room.

Intuitively, using rejected events feels dangerous, however:

1. Servers cannot arbitrarily make up state, since they still need to
pass the auth checks based on the event's auth chain (e.g. they
can't grant themselves power levels if they didn't have them
before).
2. For a previously rejected event to pass auth there must be a set of
state that allows said event. A malicious server could therefore
produce a fork where it claims the state is that particular set of
state, duplicate the rejected event to point to that fork, and send
the event. The duplicated event would then pass the auth checks.
Ignoring rejected events would therefore not eliminate any potential
attack vectors.
{{% /boxes/rationale %}}

Rejected auth events are deliberately excluded from use in the iterative
auth checks, as auth events aren't re-authed (although non-auth events
are) during the iterative auth checks.
Loading