Prevent races between encrypt and decrypt calls, and other read/write
operations on the session store by serializing session io ops per
device.
Possible fix for #535
// FREEBIE
Because remote clients will delete all sessions in response to an end
session message, regardless of which device it came from, when our
linked device sends an end session message, we must also end all
sessions with the destination.
This change moves the end session flag processing to processDecrypted,
which is shared between handlers of sent messages, data messages, and
messages which are re-tried after resolving identity conflicts.
// FREEBIE
Help debug bad session errors by logging some envelope info about the
message we are about to decrypt. With this, if there is a decryption
error (e.g., bad mac or no session) it is clear from the logs what
number and device message sent the bad message.
Also log when we send and receive end session messages and when we close
sessions for certain devices.
// FREEBIE
Previously we would delete our own key from the trust store when
re-linking, in case it changed. We can also immediately take one step
further and store the new key. Typically this happens in the course of
requesting sync info from the master device, except in the case of
standalone clients.
Closes#596
// FREEBIE
Sometimes an error is thrown while processing groups from a group sync
message. We still want to fire the groupsync event when we're done
handling all the data, even if some of it was bad.
Similar in function to an xhr request, a textsecure.SyncRequest object
is initialized from a message sender and receiver pair and initiates a
request for sync from the master device. It later fires a success event
when both contacts and groups are done syncing, or a timeout event after
one minute.
// FREEBIE
We should not ack envelope protobufs that fail to decode correctly. If
the server happens to send us such a thing it probably indicates a
protocol mismatch between it and the client, in which case the client
needs to update and re-receive the failed message.
// FREEBIE
This line was never being hit due to the duplicate one above (#L166).
Peeking at the server code shows that 403 is only given in response to
registration attempts, where it does in fact indicate an invalid code.
// FREEBIE
Fix bug where resending due to failed network connection would fail
silently.
Broken in 7b6820 refactor which changed arguments to transmitMessage
// FREEBIE
Previously, libtextsecure would send a sync message automatically
when appropriate. This fails if any recipient has a key conflict
or if our network connection fails mid-send.
Instead, when appropriate, return a the DataMessage encoded as an array
buffer for later syncing. This lets the application choose when to send
it, which we now do after any successful send to a recipient, rather
than after all recipients are successfully sent to.
Eventually we should move the DataMessage protobuf construction and
group sending logic to the application layer entirely, in which case
we wouldn't need libtextsecure to construct the sync message either.
Fixes#408
Fix bug where re-installing would cause sync messages to fail and
produce unresolvable error indicators on messages that were actually
sent.
// FREEBIE
encryptMessageFor now requires an array buffer instead of a protobuf.
By converting the message to an array buffer outside the Promise.all, we
avoid allocating a new copy for each device.
// FREEBIE
Follow up to b0da4910. When inferring membership of the sender in an
unknown group, remember to save and return the group members.
Generally, this should only effect standalone clients unless someone
managed to clear their groups db table, since linked clients get group
info synced at registration.
// FREEBIE
If you're the last member of the group, it doesn't make sense to send
messages to it. Previously, we would wait forever for a callback,
causing a stuck pending state indicator.
// FREEBIE
When we re-register, our deviceId might change, which makes our sessions
are no longer valid since the recipient will see us as a new device.
Fixes#388
Fix a bad loop scope bug in getKeysForNumber by using forEach.
Refactor the initial process of establishing key material for devices
that do not have open sessions.
// FREEBIE
These functions accept an array buffer and extract an AES and MAC key
from it without verifying it has the appropriate length. Ciphertext
messages are similarly dissected. The slice function does not raise an
error on out of bounds accesses but instead returns an empty or
partially-filled array. Empty or short arrays will be passed through to
the window.crypto.subtle API, where they will raise an error. We should
not rely on the Web Crypto API to validate key lengths or for MAC checks
to fail. Instead, validate the lengths of given parameters before
extracting their components.
// FREEBIE
This may increase processing latency a bit, particularly with large
attachments, but will ensure that messages are dispatched in the order
they are received.
It would be nice to enforce ordering on only the dispatch step, so that
we could, for example, decrypt the next websocket message while waiting
for an attachment to download, but that will require a more complicated
refactor. Will stick with the quick fix for now and revisit later.
Fixes#342
// FREEBIE
It's rare that we get in a state where we have a device record without a
session, but we should handle errors gracefully in that case. Catch them
and register them, except for identity key errors which are registered
in handleResult.
// FREEBIE
fixup error handling // FREEBIE
This function dynamically declares a bunch of functions which bind to
its input arguments. Instead, use a new prototypal class to define
these functions within the context of a particular message.
// FREEBIE
Add a pendingMessages object to MessageSender. This object holds
one promise per recipient number. We init this promise with
Promise.resolve(), and chain on promises for message sending, replacing
the previous promise with the newly chained promise each time. If the
current promise resolves and finds that it is still the last promise
in the chain, it removes itself.
Websocket resources should have their keepalive timers reset whenever a
message comes in. This is a nicety that slightly reduces the amount of
traffic we send when actively messaging.
Previously this was handled by MessageReceiver, but it's a bit cleaner
to just have the WebsocketResource add an extra 'message' event handler.
// FREEBIE
Split into separate encrypt and transmit functions. Let the encryption
function also handle all wire formatting (ie, jsonification and base64
encoding), which simplifes TextSecureServer.sendMessages, removes a
TODO, and lets us save fewer params to make network errors replayable.
// FREEBIE
