For all messages where the last fragment was _not_ put on the wire, the
send queue is responsible for generating lifecycle events, but once all
fragments have been put on the wire, it's the retransmission queue that
is responsible. It does that by marking the final fragment of a message
with the lifecycle identifier, and once that message has been fully
acked by the cumulative ack TSN, it's clear that the peer has fully
received all fragments of the message, as the last fragment was acked.
For abandoned messages - where FORWARD-TSNs are sent, those will be
replied to by a SACK as well, and then we report abandoned messages
separately, to later trigger `OnLifecycleMessageExpired`.
This CL adds support in OutstandingData, which doesn't generate the
callbacks itself, but just reports them in the AckInfo.
Bug: webrtc:5696
Change-Id: I64092f13bcfda685443b7df9967b04d54aedd36a
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/264124
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#37448}
Minor refactoring of the API, to put optional arguments last. Also
changed internal structures to reflect that order, for consistency.
Also reduced size of Item from 88 to 72 bytes, by packing fields better.
Bug: webrtc:5696
Change-Id: I1b9d50831a8e9a358224682d06a782a3269b8416
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/264123
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#37403}
The current code assumed that chunks that were scheduled for fast
retransmission would never be abandoned, as chunks marked for fast
retransmission would be immediately sent after the SACK has been
processed, giving no time for them to be abandoned.
But fuzzers keep on fuzzing, and can craft a sequence of chunks that
result in a SACK that both marks the chunks for fast retransmission and
later (while processing the same SACK) abandons them.
Bug: chromium:1331087
Change-Id: Id218607e18a6f3a9d6d51044dccb920e1e77372a
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/264960
Commit-Queue: Florent Castelli <orphis@webrtc.org>
Auto-Submit: Victor Boivie <boivie@webrtc.org>
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#37108}
This CL makes OutstandingData keep track of chunks that are eligible for
fast retransmission. When the socket goes into fast recovery, the
reported missing chunks can be retransmitted quickly (ignoring the
congestion window) according to
https://datatracker.ietf.org/doc/html/rfc4960#section-7.2.4.
The CL also adds the new API to OutstandingData to retrieve only the
chunks that are eligible for fast retransmission, and moves the
remaining chunks to the ordinary list of chunks to be retransmitted
later.
This solves an issue where the retransmission timer wouldn't start if
there wouldn't be any chunks to fast-retransmit.
It doesn't, however, make sure that chunks that should be fast
retransmitted can send even when the congestion window is full. That
will be solved in the follow-up CL.
Bug: webrtc:13969
Change-Id: If4012d1cb284ef4a2d815683ed60cbbbff5b3c3b
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/259865
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36721}
This is mainly an issue when sending items with partial reliability,
with high bandwidth on a link with packet loss.
In SCTP, when a fragment isn't included in the SACK a number of times,
it's scheduled to be retransmitted or abandoned, if it has been
retransmitted too many times already (depending configuration). Before
this CL, if a fragment was NACKed and scheduled for retransmission, but
couldn't be retransmitted immediately (due to congestion window not
allowing it), future received SACKs - that would still indicate that the
fragment hasn't been received yet - would still increment the
retransmission counter. Which wasn't fair, because this fragment hasn't
had a chance to be retransmitted yet.
With this CL, the fragment will only have its retransmission counter
increased when it is not already scheduled to be retransmitted, but
actually sent on the wire and considered in-flight again.
Bug: webrtc:12943
Change-Id: I2af08255650221c044cc14591a5835c885e94c58
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/259825
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36683}
It's put in the public folder since the intention is to expose it in
SendOptions.
Additionally, use TimeMs::InfiniteFuture() to represent sending a
message with no limited lifetime (i.e. to send it reliably).
One benefit for these two is avoiding using absl::optional more than
necessary, as it results in larger struct sizes for the outstanding
data chunks.
Bug: webrtc:12943
Change-Id: I87a340f0e0905342878fe9d2a74869bfcd6b0076
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/235984
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35323}
RetransmissionQueue was growing too long (almost 1000 lines), and as
there is reason to believe that more changes are necessary in it for
performance reasons, the data structure that handles managing the
in-flight outstanding data has been extracted as a separate class with
its own test cases. What remains in RetransmissionQueue is that it holds
OutstandingData, fetch data from the SendQueue and manage all congestion
control variables and algorithms.
Bug: webrtc:12943
Change-Id: I46062a774e0e76b44e36c66f836b7d992508bf5f
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/235980
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35279}
