This implements new group consistency algorithm described in
<https://github.com/deltachat/deltachat-core-rust/issues/6401>
New `Chat-Group-Member-Timestamps` header is added
to send timestamps of member additions and removals.
Member is part of the chat if its addition timestamp
is greater or equal to the removal timestamp.
Received messages shouldn't mingle with just sent ones and appear somewhere in the middle of the
chat, so we go after the newest non fresh message.
But if a received outgoing message is older than some `InSeen` message, better sort the received
message purely by timestamp (this is an heuristic in order not to break the Gmail-like case
simulated by `verified_chats::test_old_message_4()`). We could place the received message just
before that `InSeen` message, but anyway the user may not notice it.
At least this fixes outgoing messages sorting for shared accounts where messages from other devices
should be sorted the same way as incoming ones.
The "Cannot establish guaranteed end-to-end encryption with ..." info
message can have lots of causes, and it happened twice to us now that it
took us some time to figure out which one it is.
So, include some more detail in the info message by simply adding the
non-translated error message in parantheses.
If we want to put in some more effort for nicer error messages, we
could:
- Introduce one new translated string "Cannot establish guaranteed
end-to-end encryption with …. Cause: %2$s" or similar (and remove the
old stock string)
- And/Or: Introduce new translated strings for all the possible errors
- And/Or: Maybe reword it in order to account better for the case that
the chat already is marked as g-e2ee, or use a different wording
(because if the chat is marked as g-e2ee then it might be nice to notify
the user that something may have gone wrong, but it's still working,
just that maybe the other side doesn't have us verified now)
This protects Bob (the joiner) of sending unexpected-unencrypted messages during an otherwise nicely
running SecureJoin.
If things get stuck, however, we do not want to block communication -- the chat is just
opportunistic as usual, but that needs to be communicated:
1. If Bob's chat with Alice is `Unprotected` and a SecureJoin is started, then add info-message
"Establishing guaranteed end-to-end encryption, please wait..." and let `Chat::can_send()` return
`false`.
2. Once the info-message "Messages are guaranteed to be e2ee from now on" is added, let
`Chat::can_send()` return `true`.
3. If after SECUREJOIN_WAIT_TIMEOUT seconds `2.` did not happen, add another info-message "Could not
yet establish guaranteed end-to-end encryption but you may already send a message" and also let
`Chat::can_send()` return `true`.
Both `2.` and `3.` require the event `ChatModified` being sent out so that UI pick up the change wrt
`Chat::can_send()` (this is the same way how groups become updated wrt `can_send()` changes).
SECUREJOIN_WAIT_TIMEOUT should be 10-20 seconds so that we are reasonably sure that the app remains
active and receiving also on mobile devices. If the app is killed during this time then we may need
to do step 3 for any pending Bob-join chats (right now, Bob can only join one chat at a time).
Secure-Join-Group is only expected by old core in vg-request-with-auth.
There is no reason to leak group ID in unencrypted vg-request.
Besides that, Secure-Join-Group is deprecated
as Alice knows Group ID corresponding to the auth code,
so the header can be removed completely eventually.
Mark 1:1 chat as verified as soon as Alice is forward-verified
so Bob can already start sending Chat-Verified headers.
This way Alice and Bob can scan each other's QR codes
and even if all Secure-Join headers are dropped from the network,
still get forward verifications via QR-code scans
and backward verifications via Chat-Verified messages in 1:1 chat.
As per the comment in `receive_imf.rs`, `chat.protected` must be maintained regardless of the
`Config::VerifiedOneOnOneChats`. The only thing that mustn't be done if `VerifiedOneOnOneChats` is
unset (i.e. for non-supporting UIs) is marking chats as "protection broken" because this needs
showing the corresponding dialog to a user.
Before in some places it was correctly calculated by passing the "sent" timestamp to
`calc_sort_timestamp()`, but in other places just the system time was used. In some complex
scenarios like #5088 (restoration of a backup made before a contact verification) it led to wrong
sort timestamps of protection messages and also messages following by them.
But to reduce number of args passed to functions needing to calculate the sort timestamp, add
message timestamps to `struct MimeMessage` which is anyway passed everywhere.
- Remove "Detected Autocrypt-mime message" logs printed for every incoming Autocrypt message.
- Print only a single line at the beginning of receive_imf with both the Message-ID and seen flag.
- Print Securejoin step only once, inside handle_securejoin_handshake or observe_securejoin_on_other_device.
- Do not log "Not creating ad-hoc group" every time ad-hoc group is not created, log when it is created instead.
- Log ID of the chat where Autocrypt-Gossip for all members is received.
- Do not print "Secure-join requested." for {vg,vc}-request, we already log the step.
- Remove ">>>>>>>>>>>>>>>>>>>>>>>>>" noise from securejoin logs.
Otherwise it looks like the message creating a protected group is not verified. For this, use
`sent_timestamp` of the received message as an upper limit of the sort timestamp (`msgs.timestamp`)
of the protection message. As the protection message is added to the chat earlier, this way its
timestamp is always less or eq than the received message's timestamp.
Message.set_text() and Message.get_text() are modified accordingly
to accept String and return String.
Messages which previously contained None text
are now represented as messages with empty text.
Use Message.set_text("".to_string())
instead of Message.set_text(None).
Moved custom ToSql trait including Send + Sync from lib.rs to sql.rs.
Replaced most params! and paramsv! macro usage with tuples.
Replaced paramsv! and params_iterv! with params_slice!,
because there is no need to construct a vector.
- Return Result from set_verified() so that it can't be missed.
- Pass Fingerprint to set_verified() by value to avoid cloning it inside. This optimises out an
extra clone() if we already have a value that can be moved at the caller side. However, this may
add an extra clone() if set_verified() fails, but let's not optimise the fail scenario.
This esp. speeds up receive_imf a bit when we recreate the member list (recreate_member_list == true).
It's a preparation for https://github.com/deltachat/deltachat-core-rust/issues/3768, which will be a one-line-fix, but recreate the member list more often, so that we want to optimize this case a bit.
It also adds a benchmark for this case. It's not that easy to make the benchmark non-flaky, but by closing all other programs and locking the CPU to 1.5GHz it worked. It is consistently a few percent faster than ./without-optim:
```
Receive messages/Receive 100 Chat-Group-Member-{Added|Removed} messages
time: [52.257 ms 52.569 ms 52.941 ms]
change: [-3.5301% -2.6181% -1.6697%] (p = 0.00 < 0.05)
Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
4 (4.00%) high mild
3 (3.00%) high severe
```
This makes the contact ID its own newtype instead of being a plain
u32. The change purposefully does not yet try and reap any benefits
from this yet, instead aiming for a boring change that's easy to
review. Only exception is the ToSql/FromSql as not doing that yet
would also have created churn in the database code and it is easier to
go straight for the right solution here.
The state bob needs to maintain during a secure-join process when
exchanging messages used to be stored on the context. This means if
the process was killed this state was lost and the securejoin process
would fail. Moving this state into the database should help this.
This still only allows a single securejoin process at a time, this may
be relaxed in the future. For now any previous securejoin process
that was running is killed if a new one is started (this was already
the case).
This can remove some of the complexity around BobState handling: since
the state is in the database we can already make state interactions
transactional and correct. We no longer need the mutex around the
state handling. This means the BobStateHandle construct that was
handling the interactions between always having a valid state and
handling the mutex is no longer needed, resulting in some nice
simplifications.
Part of #2777.
