Any time a new session is established, call syncRosterToDisk() to ensure
that on subsequent restoreFromDatabase() calls, the roster is actually
available. This is important so that initAccountServices() can properly
initialize the SessionMap. This prevents a race condition where after
adding a new account and initiating sessions with it, if the app is
killed (e.g. by reinstall) before triggering a syncRosterToDisk(),
subsequent restores will not have the roster available, leading to
missing XmppAxolotlSessions in the SessionMap cache. As a result of
this, a new session was initiated when sending a new message, and
received messages could not be tagged with the originating session's
fingerprint.
As an added sanity check, go to the database to confirm no records are
present before creating fresh XmppAxolotlSession objects (both in the
sending and receiving case).
Wipe session cache to prevent stale sessions being used. Wipe fetch
status cache to enable recreation of sessions. Regenerate deviceId, so
that foreign devices will talk to us again.
Messages are now tagged with the IdentityKey fingerprint of the
originating session. IdentityKeys have one of three trust states:
undecided (default), trusted, and untrusted/not yet trusted.
EditAccountActivity now show own fingerprint, and gives an option to
regenerate local keying material (and wipe all sessions associated with
the old keys in the process).
It also now displays a list of other own devices, and gives an option to
remove all but the current device.
No longer store own device ID (so that we don't encrypt messages for
ourselves), verify that own device ID is present in update list
(otherwise republish), reflect update in UI.
When receiving a message, only remember the XmppAxolotlSession wrapper
if the prospective session was actually established. This prevents us
from erroneously adding empty sessions that are never established using
received PreKeyWhisperMessages, which would lead to errors if we try to
use them for sending.
The trust-on-first-use policy leads to problems when receiving messages
from two different devices of a contact before sending a message to them
(as their IdentityKeys will not have been added yet). Since session
trust will be managed externally anyway, this change is not a security
problem, and will allow us to decrypt messages from yet-untrusted
sessions.
We now track preKeys used to establish incoming sessions with us. On
each new established session, we remove the used prekey from PEP. We
have to do this because libaxolotl-java internally clears the used
preKey from its storage, so we will not be able to establish any future
sessions using that key.
XmppConnectionService.sendMessage() now dispatches messages to the
AxolotlService, where they only are prepared for sending and cached.
AxolotlService now triggers a XmppConnectionService.resendMessage(),
which then handles sending the cached message packet.
This transparently fixes, e.g., handling of messages sent while we are
offline.
Previously, the sender was assumed to be the conversation counterpart.
This broke carboned own-device messages. We now track the sender
properly, and also set the status (sent by one of the own devices vs
received from the counterpart) accordingly.
Now checks which part(s) are out of sync w/ local storage, and updates
only those, rather than assuming the entire node corrupt and
overwriting it all (especially relevant for preKey list)
We need a session object in order to build a session from a
PreKeyWhisperMessage, so add an empty one when none exists on receiving
a message.
Warning: this will break right now if the session can not be constructed
from the received message.There will be an invalid session which will
break if we try to send using it.
Bundles are now fetched on demand when a session needs to be
established. This should lessen the chance of changes to the bundles
occuring before they're used, as well as lessen the load of fetching
bundles.
Also, the message encryption is now done in a background thread, as this
can be somewhat costly if many sessions are present. This is probably
not going to be an issue in real use, but it's good practice anyway.
Andreas Straub