This fixes a DCHECK during teardown in the case when the primary
DataChannelTranspot (eg. DatagramTransport) is successfully negotiated.
DatagramTransport expects the DataSink to be unset before it's deleted.
This was not caught by existing tests because the fallback transport
(SctpDataChannelTransport) does not have the same DCHECK.
Also adds a regression test for the issue, in which SCTP is available
as a fallback but DataChannelTransport is negotiated successfully.
Bug: webrtc:9719
Change-Id: I414d964d3c85d3d01cdb5e34d6b248659a613c39
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/154365
Commit-Queue: Bjorn Mellem <mellem@webrtc.org>
Reviewed-by: Seth Hampson <shampson@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#29292}
This is a reland of 487f9a17e426fd14bb06b13e861071b3f15d119b
Original change's description:
> Reland "Refactor SCTP data channels to use DataChannelTransportInterface."
>
> Also clears SctpTransport before deleting JsepTransport.
>
> SctpTransport is ref-counted, but the underlying transport is deleted when
> JsepTransport clears the rtp_dtls_transport. This results in crashes when
> usrsctp attempts to send outgoing packets through a dangling pointer to the
> underlying transport.
>
> Clearing SctpTransport before DtlsTransport removes the pointer to the
> underlying transport before it becomes invalid.
>
> This fixes a crash in chromium's web platform tests (see
> https://chromium-review.googlesource.com/c/chromium/src/+/1776711).
>
> Original change's description:
> > Refactor SCTP data channels to use DataChannelTransportInterface.
> >
> > This change moves SctpTransport to be owned by JsepTransport, which now
> > holds a DataChannelTransport implementation for SCTP when it is used for
> > data channels.
> >
> > This simplifies negotiation and fallback to SCTP. Negotiation can now
> > use a composite DataChannelTransport, just as negotiation for RTP uses a
> > composite RTP transport.
> >
> > PeerConnection also has one fewer way it needs to manage data channels.
> > It now handles SCTP and datagram- or media-transport-based data channels
> > the same way.
> >
> > There are a few leaky abstractions left. For example, PeerConnection
> > calls Start() on the SctpTransport at a particular point in negotiation,
> > but does not need to call this for other transports. Similarly, PC
> > exposes an interface to the SCTP transport directly to the user; there
> > is no equivalent for other transports.
>
> Bug: webrtc:9719
> Change-Id: I64e94b88afb119fdbf5f22750f88c8a084d53937
> Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/151981
> Reviewed-by: Benjamin Wright <benwright@webrtc.org>
> Reviewed-by: Steve Anton <steveanton@webrtc.org>
> Commit-Queue: Benjamin Wright <benwright@webrtc.org>
> Commit-Queue: Bjorn Mellem <mellem@webrtc.org>
> Cr-Commit-Position: refs/heads/master@{#29120}
Bug: webrtc:9719
Change-Id: I28481a3de64a3506bc57748106383eeba4ef205c
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/152740
Reviewed-by: Artem Titov <titovartem@webrtc.org>
Reviewed-by: Benjamin Wright <benwright@webrtc.org>
Reviewed-by: Seth Hampson <shampson@webrtc.org>
Commit-Queue: Bjorn Mellem <mellem@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#29290}
This reverts commit 487f9a17e426fd14bb06b13e861071b3f15d119b.
Reason for revert: speculative revert
Original change's description:
> Reland "Refactor SCTP data channels to use DataChannelTransportInterface."
>
> Also clears SctpTransport before deleting JsepTransport.
>
> SctpTransport is ref-counted, but the underlying transport is deleted when
> JsepTransport clears the rtp_dtls_transport. This results in crashes when
> usrsctp attempts to send outgoing packets through a dangling pointer to the
> underlying transport.
>
> Clearing SctpTransport before DtlsTransport removes the pointer to the
> underlying transport before it becomes invalid.
>
> This fixes a crash in chromium's web platform tests (see
> https://chromium-review.googlesource.com/c/chromium/src/+/1776711).
>
> Original change's description:
> > Refactor SCTP data channels to use DataChannelTransportInterface.
> >
> > This change moves SctpTransport to be owned by JsepTransport, which now
> > holds a DataChannelTransport implementation for SCTP when it is used for
> > data channels.
> >
> > This simplifies negotiation and fallback to SCTP. Negotiation can now
> > use a composite DataChannelTransport, just as negotiation for RTP uses a
> > composite RTP transport.
> >
> > PeerConnection also has one fewer way it needs to manage data channels.
> > It now handles SCTP and datagram- or media-transport-based data channels
> > the same way.
> >
> > There are a few leaky abstractions left. For example, PeerConnection
> > calls Start() on the SctpTransport at a particular point in negotiation,
> > but does not need to call this for other transports. Similarly, PC
> > exposes an interface to the SCTP transport directly to the user; there
> > is no equivalent for other transports.
>
> Bug: webrtc:9719
> Change-Id: I64e94b88afb119fdbf5f22750f88c8a084d53937
> Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/151981
> Reviewed-by: Benjamin Wright <benwright@webrtc.org>
> Reviewed-by: Steve Anton <steveanton@webrtc.org>
> Commit-Queue: Benjamin Wright <benwright@webrtc.org>
> Commit-Queue: Bjorn Mellem <mellem@webrtc.org>
> Cr-Commit-Position: refs/heads/master@{#29120}
TBR=steveanton@webrtc.org,mellem@webrtc.org,benwright@webrtc.org
Change-Id: Ibd1a7f30931c114212c90824fec414d276d3f915
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: webrtc:9719
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/152421
Reviewed-by: Qingsi Wang <qingsi@webrtc.org>
Commit-Queue: Qingsi Wang <qingsi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#29141}
Also clears SctpTransport before deleting JsepTransport.
SctpTransport is ref-counted, but the underlying transport is deleted when
JsepTransport clears the rtp_dtls_transport. This results in crashes when
usrsctp attempts to send outgoing packets through a dangling pointer to the
underlying transport.
Clearing SctpTransport before DtlsTransport removes the pointer to the
underlying transport before it becomes invalid.
This fixes a crash in chromium's web platform tests (see
https://chromium-review.googlesource.com/c/chromium/src/+/1776711).
Original change's description:
> Refactor SCTP data channels to use DataChannelTransportInterface.
>
> This change moves SctpTransport to be owned by JsepTransport, which now
> holds a DataChannelTransport implementation for SCTP when it is used for
> data channels.
>
> This simplifies negotiation and fallback to SCTP. Negotiation can now
> use a composite DataChannelTransport, just as negotiation for RTP uses a
> composite RTP transport.
>
> PeerConnection also has one fewer way it needs to manage data channels.
> It now handles SCTP and datagram- or media-transport-based data channels
> the same way.
>
> There are a few leaky abstractions left. For example, PeerConnection
> calls Start() on the SctpTransport at a particular point in negotiation,
> but does not need to call this for other transports. Similarly, PC
> exposes an interface to the SCTP transport directly to the user; there
> is no equivalent for other transports.
Bug: webrtc:9719
Change-Id: I64e94b88afb119fdbf5f22750f88c8a084d53937
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/151981
Reviewed-by: Benjamin Wright <benwright@webrtc.org>
Reviewed-by: Steve Anton <steveanton@webrtc.org>
Commit-Queue: Benjamin Wright <benwright@webrtc.org>
Commit-Queue: Bjorn Mellem <mellem@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#29120}
This reverts commit 4c85828ab272d9bd58789bad7b135b6287395f97.
Reason for revert:
Speculatively reverting this because it makes several web platform tests relating to RTCDataChannel flaky, see first failing roll:
https://chromium-review.googlesource.com/c/chromium/src/+/1776711
Original change's description:
> Refactor SCTP data channels to use DataChannelTransportInterface.
>
> This change moves SctpTransport to be owned by JsepTransport, which now
> holds a DataChannelTransport implementation for SCTP when it is used for
> data channels.
>
> This simplifies negotiation and fallback to SCTP. Negotiation can now
> use a composite DataChannelTransport, just as negotiation for RTP uses a
> composite RTP transport.
>
> PeerConnection also has one fewer way it needs to manage data channels.
> It now handles SCTP and datagram- or media-transport-based data channels
> the same way.
>
> There are a few leaky abstractions left. For example, PeerConnection
> calls Start() on the SctpTransport at a particular point in negotiation,
> but does not need to call this for other transports. Similarly, PC
> exposes an interface to the SCTP transport directly to the user; there
> is no equivalent for other transports.
>
> Bug: webrtc:9719
> Change-Id: I0d3151c48c1a511368277981fc4cf818a9f8ebb4
> Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/150341
> Reviewed-by: Steve Anton <steveanton@webrtc.org>
> Reviewed-by: Benjamin Wright <benwright@webrtc.org>
> Commit-Queue: Bjorn Mellem <mellem@webrtc.org>
> Cr-Commit-Position: refs/heads/master@{#29012}
TBR=steveanton@webrtc.org,mellem@webrtc.org,benwright@webrtc.org
Change-Id: I074b9e68f298d20d0cabb4239084b4843e76e910
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: webrtc:9719
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/150944
Reviewed-by: Henrik Boström <hbos@webrtc.org>
Commit-Queue: Henrik Boström <hbos@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#29025}
This change moves SctpTransport to be owned by JsepTransport, which now
holds a DataChannelTransport implementation for SCTP when it is used for
data channels.
This simplifies negotiation and fallback to SCTP. Negotiation can now
use a composite DataChannelTransport, just as negotiation for RTP uses a
composite RTP transport.
PeerConnection also has one fewer way it needs to manage data channels.
It now handles SCTP and datagram- or media-transport-based data channels
the same way.
There are a few leaky abstractions left. For example, PeerConnection
calls Start() on the SctpTransport at a particular point in negotiation,
but does not need to call this for other transports. Similarly, PC
exposes an interface to the SCTP transport directly to the user; there
is no equivalent for other transports.
Bug: webrtc:9719
Change-Id: I0d3151c48c1a511368277981fc4cf818a9f8ebb4
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/150341
Reviewed-by: Steve Anton <steveanton@webrtc.org>
Reviewed-by: Benjamin Wright <benwright@webrtc.org>
Commit-Queue: Bjorn Mellem <mellem@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#29012}
