This commit fixes a few issues with communities encryption:
Key distribution was disconnected from the community description, this created a case where the key would arrive after the community description and that would result in the client thinking that it was kicked.
To overcome this, we added a message that signals the user that is kicked. Also, we distribute the key with the community description so that there's no more issues with timing.
This is a bit expensive for large communities, and it will require some further optimizations.
Key distribution is now also connected to the request to join response, so there are no timing issues.
Fixes an issue with key distribution (race condition) where the community would be modified before being compared, resulting in a comparison of two identical communities, which would result in no key being distributed. This commit only partially address the issue.
This commit adds basic syncing capabilities with peers if they are both
online.
It updates the work done on MVDS, but I decided to create the code in
status-go instead, since it's very tight to the application (similarly
the code that was the inspiration for mvds, bramble, is all tight
together at the database level).
I reused parts of the protobufs.
The flow is:
1) An OFFER message is sent periodically with a bunch of message-ids and
group-ids.
2) Anyone can REQUEST some of those messages if not present in their
database.
3) The peer will then send over those messages.
It's disabled by default, but I am planning to add a way to set up the
flags.
This commit adds a test for out of order messages, which were only
implicitly tested.
It also deletes them after being processed, otherwise they would be
reproceessed each time a message was sent
This commit fixes 3 issues:
1) In some cases, the hash ratchet was not correctly found
2) Out of order messages were not processed correctly as the wrong error
was returned
3) Batched non datasync messages were not processed correctly
Fixes: #4170
This commit changes the format of the encryption id to be based off 3
things:
1) The group id
2) The timestamp
3) The actual key
Previously this was solely based on the timestamp and the group id, but
this might lead to conflicts. Moreover the format of the key was an
uint32 and so it would wrap periodically.
The migration is a bit tricky, so first we cleared the cache of keys,
that's easier than migrating, and second we set the new field hash_id to
the concatenation of group_id / key_id.
This might lead on some duplication in case keys are re-received, but it
should not have an impact on the correctness of the code.
I have added 2 tests covering compatibility between old/new clients, as
this should not be a breaking change.
It also adds a new message to rekey in a single go, instead of having to
send multiple messages
- use protected topics for communities
- associate chats to pubsub topics and populate these depending if the chat belongs to a community or not
- mailserver functions should be aware of pubsub topics
- generate private key for pubsub topic protection when creating a community
- add shard cluster and index to communities
- setup shards for existing communities
- distribute pubsubtopic password
- fix: do not send the requests to join and cancel in the protected topic
- fix: undefined shard values for backward compatibility
- refactor: use shard message in protobuffers
refactor: associate chats to pubsub topics and populate these depending if the chat belongs to a community or not
refactor: add pubsub topic to mailserver batches
chore: ensure default relay messages continue working as they should
refactor: mailserver functions should be aware of pubsub topics
fix: use []byte for communityIDs
- distribute ratchet keys at both community and channel levels
- use explicit `HashRatchetGroupID` in ecryption layer, instead of
inheriting `groupID` from `CommunityID`
- populate `HashRatchetGroupID` with `CommunityID+ChannelID` for
channels, and `CommunityID` for whole community
- hydrate channels with members; channel members are now subset of
community members
- include channel permissions in periodic permissions check
closes: status-im/status-desktop#10998
Sometimes the message scheduled & message sent notifications are
received out of order.
Now we use a single channel for both so order is maintained.
This was causing the tests to be flaky and it might have happened in
production as well.
