This reverts commit 3180a5ad0663900a39adf4b9974052c356c835fe.
Reason for revert: Speculative revert due to failures in downstream tests.
Original change's description:
> dcsctp: Reset send queue when recreating TCB
>
> This is an issue found in fuzzer, and doesn't really happen in WebRTC
> as it never closes the connection and reconnects.
>
> The issue is that the send queue outlives any connection since you're
> allowed to send messages (well, enqueue them) before the association is
> fully connected. So the send queue is always present but the TCB
> (information about the connection) is torn down when the connection is
> closed for example. And the TCB holds the Stream Reset handler, which is
> responsible for e.g. keeping track of stream reset sequence numbers and
> such - which is tied to the connection.
>
> So to ensure that the Stream Reset Handler is in charge of deciding
> if a stream reset is taking place, make sure that the send queue is in
> a known good state when the Stream Reset handler is created.
>
> Bug: webrtc:13994, chromium:1320194
> Change-Id: I940e4690ac9237ac99dd69a9ffc060cdac61711d
> Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/261260
> Reviewed-by: Harald Alvestrand <hta@webrtc.org>
> Commit-Queue: Victor Boivie <boivie@webrtc.org>
> Cr-Commit-Position: refs/heads/main@{#36779}
Bug: webrtc:13994, chromium:1320194
Change-Id: I89bb9cae60adc53902c1304e79047d18e72594a5
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/261302
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Auto-Submit: Björn Terelius <terelius@webrtc.org>
Bot-Commit: rubber-stamper@appspot.gserviceaccount.com <rubber-stamper@appspot.gserviceaccount.com>
Owners-Override: Björn Terelius <terelius@webrtc.org>
Reviewed-by: Victor Boivie <boivie@google.com>
Commit-Queue: Harald Alvestrand <hta@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36783}
This is an issue found in fuzzer, and doesn't really happen in WebRTC
as it never closes the connection and reconnects.
The issue is that the send queue outlives any connection since you're
allowed to send messages (well, enqueue them) before the association is
fully connected. So the send queue is always present but the TCB
(information about the connection) is torn down when the connection is
closed for example. And the TCB holds the Stream Reset handler, which is
responsible for e.g. keeping track of stream reset sequence numbers and
such - which is tied to the connection.
So to ensure that the Stream Reset Handler is in charge of deciding
if a stream reset is taking place, make sure that the send queue is in
a known good state when the Stream Reset handler is created.
Bug: webrtc:13994, chromium:1320194
Change-Id: I940e4690ac9237ac99dd69a9ffc060cdac61711d
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/261260
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36779}
When streams were to be reset, but there was already an ongoing
stream reset command in-flight, those streams wouldn't be properly
reset. When multiple streams were reset close to each other (within
an RTT), some streams would not have their SSNs reset, which resulted
in the stream resuming the SSN sequence. This could result in ordered
streams not delivering all messages as the receiver wouldn't deliver any
messages with SSN different from the expected SSN=0.
In WebRTC data channels, this would be triggered if multiple channels
were closed at roughly the same time, then re-opened, and continued
to be used in ordered mode. Unordered messages would still be delivered,
but the stream state could be wrong as the DATA_CHANNEL_ACK message is
sent ordered, and possibly not delivered.
There were unit tests for this, but not on the socket level using
real components, but just on the stream reset handler using mocks,
where this issue wasn't found. Also, those mocks didn't validate that
the correct parameters were provided, so that's fixed now.
The root cause was the PrepareResetStreams was only called if there
wasn't an ongoing stream reset operation in progress. One may try to
solve it by always calling PrepareResetStreams also when there is an
ongoing request, or to call it when the request has finished. One would
then realize that when the response of the outgoing stream request is
received, and CommitResetStreams is called, it would reset all paused
and (prepared) to-be-reset streams - not just the ones in the outgoing
stream request.
One cause of this was the lack of a single source of truth of the stream
states. The SendQueue kept track of which streams that were paused, but
the stream reset handler kept track of which streams that were
resetting. As that's error prone, this CL moves the source of truth
completely to the SendQueue and defining explicit stream pause states. A
stream can be in one of these possible states:
* Not paused. This is the default for an active stream.
* Pending to be paused. This is when it's about to be reset, but
there is a message that has been partly sent, with fragments
remaining to be sent before it can be paused.
* Paused, with no partly sent message. In this state, it's ready to
be reset.
* Resetting. A stream transitions into this state when it has been
paused and has been included in an outgoing stream reset request.
When this request has been responded to, the stream can really be
reset (SSN=0, MID=0).
This CL also improves logging, and adds socket tests to catch this
issue.
Bug: webrtc:13994, chromium:1320194
Change-Id: I883570d1f277bc01e52b1afad62d6be2aca930a2
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/261180
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36771}
Before this CL, fast retransmission didn't follow the SHOULDs:
https://datatracker.ietf.org/doc/html/rfc4960#section-7.2.4
* "the sender SHOULD ignore the value of cwnd (...)"
* "(...) and SHOULD NOT delay retransmission for this single
packet."
With this CL, chunks that are eligible for fast retransmission (limited
to what can fit in a single packet) will be sent just after having
received the SACK that reported them missing and transitioned the socket
into fast recovery, and they will be sent even if the congestion window
is full.
Bug: webrtc:13969
Change-Id: I12c7e191a8ffd67973db7f083bad8a6061549fa2
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/259866
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36724}
This proved to be not very efficient unfortunately, so revert it and
keep bundling FORWARD-TSN with other packets to be more efficient.
https://github.com/sctplab/usrsctp/issues/597 is still unresolved.
Note that this is not a clean revert; The logic to rate limit the
sending of FORWARD-TSN is kept, as it still makes sense.
This partly reverts commit 0ca62e3752149ad37f73bf074db0a5f8fcaf6585.
Bug: webrtc:12961
Change-Id: I42728434290e7ece19e9c23f24ef6f3d3b171315
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/259520
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36584}
When resetting several streams in sequence, only the first stream will
be included in the first RE_CONFIG chunk as it's created eagerly
whenever someone calls ResetStreams. The remaining ones are queued as
pending. When the first request finishes, the remaining ones should
continue to be processed, but this wasn't done prior to this commit.
This would only happen if two streams would be reset with shorter time
between them than a RTT, so that there would be an outstanding request
forcing the second reset to be enqueued.
Bug: chromium:1312009
Change-Id: Id74b375d1d1720406a3bca4ec60df5780ca7edba
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/257306
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36404}
This is a solution to some problems that have been found locally when
running the fuzzer for a long time. The fuzzer keeps on fuzzing, and has
found a way to trigger a consistency check to fail when a client
intentionally sends different messages - unordered and ordered - using
the same TSNs. As the reassembly queue has different handling of ordered
and unordered chunks due to how they are reassembled, it will not notice
if it receives two different chunks with the same TSN. They will both go
to their respective reassembly streams, as those are separate by design.
The data tracker - which keeps track of all received DATA chunks as it
needs to generate SACKs, has a global understanding of all received
chunks. By having it indicate if this is a duplicate received chunk, the
socket can avoid forwarding both chunks to the reassembly queue; only
one chunk will get there.
Bug: None
Change-Id: I602a8552a9a4c853684fcf105309ec3d8073f2c2
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/256110
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36316}
When a FORWARD-TSN is received as the first chunk on an ordered stream,
it will fail to set the new "next expected SSN" that is present in the
FORWARD-TSN as that stream hasn't been allocated yet. It's allocated
when the first DATA is received on that stream.
This is a non-issue for ordinary data channels as the first message on
any stream will be the "Data Channel Establishment Protocol" messages,
which are always sent reliably. But if prenegotiated channels are used,
and the very first packet received on an ordered data channel is lost
_and_ signaled to the receiver as lost _before_ the receiver has
received any other fragments on that data channel, future messages will
not be delivered on that channel.
Bug: webrtc:13799
Change-Id: Ide5c656243b3a51a2ed9d76615cfc3631cfe900c
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/253902
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36155}
Following https://abseil.io/tips/122 to make tests easier to understand
and adds a bit of flexibility to create sockets with custom parameters.
This also simplifies handover tests.
Additionally, AdvanceTime will now also run timers, as that was easily
forgotten previously.
Bug: None
Change-Id: Ieb5eece7aca51c98a7634ed1c61646383ad1712d
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/253782
Reviewed-by: Sergey Sukhanov <sergeysu@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36141}
To disable dcSCTP and fallback to usrsctp, you can use the field trial
WebRTC-DataChannel-Dcsctp/Disabled/
Also remove a hidden no-break space in dcSCTP logging causing issues in
some log parsing.
Bug: chromium:1243702
Change-Id: I46136a8913a6d803a3c63c710f3ed29523e4d773
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/251867
Auto-Submit: Florent Castelli <orphis@webrtc.org>
Reviewed-by: Victor Boivie <boivie@google.com>
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Harald Alvestrand <hta@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36027}
In SCTP, sending ACKs at the right time is presumably important in order
not to cause unnecessary retransmissions.
This CL sets the ACK timers to use high timer precision. This unblocks
experimentally lowering the precision of the default, "low", timers in
WebRTC.
// All bots are green, but mac_chromium_compile is randomly timing out
// independently of this CL and has been doing so for several days...
NOTRY=True
Bug: webrtc:13604
Change-Id: I1c81e3d0eeb477c3e00277d35649114c5edd249c
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/249090
Reviewed-by: Victor Boivie <boivie@webrtc.org>
Commit-Queue: Henrik Boström <hbos@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35813}
Context: The timer precision of PostDelayedTask() is about to be lowered
to include up to 17 ms leeway. In order not to break use cases that
require high precision timers, PostDelayedHighPrecisionTask() will
continue to have the same precision that PostDelayedTask() has today.
webrtc::TaskQueueBase has an enum (kLow, kHigh) to decide which
precision to use when calling PostDelayedTaskWithPrecision().
See go/postdelayedtask-precision-in-webrtc for motivation and a table of
delayed task use cases in WebRTC that are "high" or "low" precision.
Most timers in DCSCTP are believed to only be needing low precision (see
table), but the delayed_ack_timer_ of DataTracker[1] is an example of a
use case that is likely to break if the timer precision is lowered (if
ACK is sent too late, retransmissions may occur). So this is considered
a high precision use case.
This CL makes it possible to specify the precision of dcsctp::Timer.
In a follow-up CL we will update delayed_ack_timer_ to kHigh precision.
[1] https://source.chromium.org/chromium/chromium/src/+/main:third_party/webrtc/net/dcsctp/rx/data_tracker.cc;l=340
Bug: webrtc:13604
Change-Id: I8eec5ce37044096978b5dd1985fbb00bc0d8fb7e
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/249081
Reviewed-by: Victor Boivie <boivie@webrtc.org>
Reviewed-by: Tomas Gunnarsson <tommi@webrtc.org>
Commit-Queue: Henrik Boström <hbos@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35809}
As WebRTC now supports C++17, simplify the code of dcSCTP by binding
return values from std::pair or std::tuple to separate names.
Bug: webrtc:13220
Change-Id: Ie49154ff4c823e1528deaef7e372cbc550923bc2
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/246442
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35773}
the new spelling is more standard and more compact, in particular doesn't need extra include and thus dependency
Bug: None
Change-Id: Iaea69d2154e4d9eff2468514f5734cb3fe016ff8
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/245080
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Danil Chapovalov <danilchap@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35709}
This avoids copying the payload at all. Future CL will change the
transport.
In performance tests, memcpy was visible in the performance profiles
prior to this change.
Bug: webrtc:12943
Change-Id: I507a1a316165db748e73cf0d58c1be62cc76a2d2
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/236346
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35428}
Networks tests were previously disabled if building in debug mode as
debug mode adds DCHECKs, and when DCHECKs are enabled, a lot of the
components in dcSCTP will add consistency checks, and they can be really
expensive to run in these network tests.
However, if running in with TSAN or MSAN sanitizers and with DCHECKs
enabled, they also take a long time.
Current run-time on my relatively fast CPU (with is_debug=false):
(no sanitizer) always_dcheck=false: 2.5s
(no sanitizer) always_dcheck=true: 31s
is_tsan=true, always_dcheck=false: 53s
is_tsan=true, always_dcheck=true: 5m50s <-- too slow
is_asan=true, always_dcheck=false: 13s
is_asan=true, always_dcheck=true: 47s
is_msan=true, always_dcheck=false: 35s
is_msan=true, always_dcheck=true: 1m53s <-- too slow
Note that buildbots may be much slower than my computer.
Bug: webrtc:12943
Change-Id: If044ee9936372d54c9899b4864156c9f680af0b6
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/236581
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35276}
This is a follow-up to change 232904 that also validates that the
timestamp from the heartbeat ack isn't negative (which the fuzzer
managed to set it to).
Bug: chromium:1252515
Change-Id: Idaac570589dbdaaee67b7785f6232b60226e88e1
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/234582
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35168}
The DcSctpSocket is not thread safe and must be called from a single
thread or from a task queue that serializes access to it. This is now
validated at run-time in debug builds.
Bug: None
Change-Id: I3ed816924c20f6ed7e84a3273bee5a3f8f74112b
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/233420
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35127}
This initial version contains a few tests, testing both lossy, non-lossy
and bandwidth limited networks. It uses simulated time, and runs much
faster than wall time on release builds, but slower on debug when there
is a lot of outstanding data (high throughput) as there are consistency
checks on outstanding data. Because of that, some tests had to be
disabled in debug build.
Bug: webrtc:12943
Change-Id: I9323f2dc99bca4e40558d05a283e99ce7dded7f1
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/225201
Reviewed-by: Artem Titov <titovartem@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35124}
The previous manual way of triggering the deferred callbacks was very
error-prone, and this was also forgotten at a few places.
We can do better.
Using the RAII programming idiom, the callbacks are now ensured to be
called before returning from public methods.
Also added additional debug checks to ensure that there is a
ScopedDeferrer active whenever callbacks are deferred.
Bug: webrtc:13217
Change-Id: I16a8343b52c00fb30acb018d3846acd0a64318e0
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/233242
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35117}
It's to be used for clients to record metrics and to e.g. attribute
metrics to which SCTP implementation the peer was using.
This is not explicitly signaled, so heuristics are used. These are not
guaranteed to come to the correct conclusion, and the data is not always
available.
Note: The behavior of dcSCTP will not change depending on the assumed
implementation - only by explicitly signaled capabilities.
Bug: webrtc:13216
Change-Id: I2f58054d17d53d947ed5845df7a08f974d42f918
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/233100
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35103}
When a HEARTBEAT is sent out, the current timestamp is stored in the
parameter that will be returned by the HEARTBEAT-ACK. If the timestamp
from the HEARTBEAT-ACK would be from the future (either by jumping
clocks, bit errors, exploiting peer or a fuzzer), the measured RTT would
become really large, and when it was calculated, it would result in an
integer overflow; a wrapping subtraction.
This isn't a problem, as RetransmissionTimeout::ObserveRTT method would
discard measurements that were negative or too large, but it would
trigger an UBSAN violation.
Adding an additional check so that the subtraction doesn't ever wrap.
Bug: chromium:1252515
Change-Id: I1f97b1e773a4639b8193a528716ebd6a27fcb740
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/232904
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35095}
This allows build targets that need only HandoverReadinessStatus
to depend only on public:socket, and not on socket:dcsctp_socket.
Bug: webrtc:13154
Change-Id: I29f41910cdb5baed96b57fd7284b96fc50a56ba4
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/232331
Reviewed-by: Victor Boivie <boivie@webrtc.org>
Commit-Queue: Sergey Sukhanov <sergeysu@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35037}
dcSCTP library users can set their custom
g_handover_state_transformer_for_test that can serialize and
deserialize the state. All dcSCTP handover tests call
g_handover_state_transformer_for_test. If some part of the state is
serialized incorrectly or is forgotten, high chance that it will
fail the tests.
Bug: webrtc:13154
Change-Id: I251a099be04dda7611e9df868d36e3a76dc7d1e1
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/232325
Commit-Queue: Sergey Sukhanov <sergeysu@webrtc.org>
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Reviewed-by: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35035}
The buffer of reassembled messages in ReassemblyQueue is only to be
used while processing a DATA/I-DATA or FORWARD-TSN as processing these
chunks may result in assembling messages.
When the socket is idle - between API calls - it's supposed to be empty.
Instead of having it as a member in ReassemblyQueue, it could be
provided as an argument to ReassemblyQueue::Add and
ReassemblyQueue::Handle(ForwardTSN), but that would be a quite big
refactoring. That will be investigated separately.
Bug: None
Change-Id: I41238de28f32f2a622c1d045debe3ea11e7c23f6
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/232000
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35014}
This will be useful in future socket handover tests.
Bug: webrtc:13154
Change-Id: Ia789ae971edd9d2832be088f2f8f7dd50c9ce52d
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/231222
Reviewed-by: Victor Boivie <boivie@webrtc.org>
Commit-Queue: Sergey Sukhanov <sergeysu@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#34931}
By allowing the max timer backoff duration to be limited, a socket can
fast recover in case of intermittent network issues. Before this CL, the
exponential backoff algorithm could result in very long retry durations
(in the order of minutes), when connection has been lost or been flaky
for a long while.
Note that limiting the maximum backoff duration might require
compensating the maximum retransmission limit to avoid closing the
socket prematurely due to reaching the maximum retransmission limit much
faster than previously.
Bug: webrtc:13129
Change-Id: Ib94030d666433e3fa1a2c8ef69750a1afab8ef94
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/230702
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#34913}
The restart limit for timers can already be limitless, but the
RetransmissionErrorCounter didn't support this. With this change, the
max_retransmissions and max_init_retransmits can be absl::nullopt to
indicate that there should be infinite retries.
Bug: webrtc:13129
Change-Id: Ia6e91cccbc2e1bb77b3fdd7f37436290adc2f483
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/230701
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#34882}
dcSCTP seems to be able to provoke usrsctp to send ABORT in some
situations, as described in
https://github.com/sctplab/usrsctp/issues/597. Using a packetdrill
script, it seems as a contributing factor to this behavior is when a
FORWARD-TSN chunk is bundled with a DATA chunk. This is a valid and
recommended pattern in RFC3758:
"F2) The data sender SHOULD always attempt to bundle an outgoing
FORWARD TSN with outbound DATA chunks for efficiency."
However, that seems to be a rare event in usrsctp, which generally sends
each FORWARD-TSN in a separate packet.
By doing the same, the assumption is that this scenario will generally
be avoided.
With many browsers and other clients using usrsctp, and which will not
be upgraded for a long time, there is an advantage of avoiding the issue
even if it is according to specification.
Before this change, a FORWARD-TSN was bundled with outgoing DATA and due
to this, it wasn't rate limited as the overhead was very little. With
this change, a rate limiting behavior has been added to avoid sending
too many FORWARD-TSN in small packets. It will be sent every RTT, or
200 ms, whichever is smallest. This is also described in the RFC.
Bug: webrtc:12961
Change-Id: I3d8036a34f999f405958982534bfa0e99e330da3
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/229101
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#34801}
The congestion window is unlikely to be even divisible by the size
of a packet, so when the congestion window is almost full, there is
often just a few bytes remaining in it. Before this change, a small
packet was created to fill the remaining bytes in the congestion window,
to make it really full.
Small packets don't add much. The cost of sending a small packet is
often the same as sending a large one, and you usually get lower
throughput sending many small packets compared to few larger ones.'
This mode will only be enabled when the congestion window is large, so
if the congestion window is small - e.g. due to poor network conditions,
it will allow packets to become fragmented into small parts, in order to
fully utilize the congestion window.
Bug: webrtc:12943
Change-Id: I8522459174bc72df569edd57f5cc4a494a4b93a8
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/228526
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#34778}
This is mainly a refactoring commit, to break out packet sending to a
dedicated component.
Bug: webrtc:12943
Change-Id: I78f18933776518caf49737d3952bda97f19ef335
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/228565
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#34772}
It is now allowed in WebRTC, so let's use it.
Bug: webrtc:12943
Change-Id: I74a0f2fd9c1b9e7b5613ae1c592cf26842b8dddd
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/228564
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Reviewed-by: Florent Castelli <orphis@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#34768}
Before this change, there was no way for a client to indicate to the
dcSCTP library if a packet that was supposed to be sent, was actually
sent. It was assumed that it always was.
To handle temporary failures better, such as retrying to send packets
that failed to be sent when the send buffer was full, this information
is propagated to the library.
Note that this change only covers the API and adaptations to clients.
The actual implementation to make use of this information is done as a
follow-up change.
Bug: webrtc:12943
Change-Id: I8f9c62e17f1de1566fa6b0f13a57a3db9f4e7684
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/228563
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Reviewed-by: Niels Moller <nisse@webrtc.org>
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#34767}
Some deployments, e.g. Chromium, has a limited send buffer. It's
reasonable that it's quite small, as it avoids queuing too much, which
typically results in increased latency for real-time communication. To
avoid SCTP to fill up the entire buffer at once - especially when doing
fast retransmissions - limit the amount of packets that are sent in one
go.
In a typical scenario, SCTP will not send more than three packets for
each incoming packet, which is is the case when a SACK is received which
has acknowledged two large packets, and which also adds the MTU to the
congestion window (due to in slow-start mode), which then may result in
sending three packets. So setting this value to four makes any
retransmission not use that much more of the send buffer.
This is analogous to usrsctp_sysctl_set_sctp_fr_max_burst_default in
usrsctp, which also has the default value of four (4).
Bug: webrtc:12943
Change-Id: Iff76a1668beadc8776fab10312ef9ee26f24e442
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/228480
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Victor Boivie <boivie@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#34744}
