Avoid FFmpeg destroying active resources during demuxer stop.

Avoid FFmpeg destroying active resources during demuxer stop.

The root cause is a combination of BufferedDataSource::Stop()
requiring a lock and BlockingURLProtocol::Read() returning
control to FFmpeg which may destory resources in use by an
ongoing BufferedDataSource::ReadCallback() call.

The lock in BDS::Stop() shared by BDS::ReadCallback() prevents
the stop signal from taking affect until after the read finishes,
however calling BUP::Abort() will immediately returns control to
FFmpeg which may destroy the resources in use by BDS::ReadCallback().

Sadly, given the racy nature and multi-class dependencies I've been
unable to craft a unit test for this, but I've manually verified this
fix works by arbitrarily delaying BDS::Stop() under an ASAN build.

BUG=411318
TEST=none

Committed: https://crrev.com/1973f0b6f7497b1887397de9bb36dbff4faf5e39
Cr-Commit-Position: refs/heads/master@{#293995}

[DaleCurtis, 2014-09-05 21:08:10]

dalecurtis@chromium.org changed reviewers:
+ scherkus@chromium.org

[DaleCurtis, 2014-09-05 21:09:02]

Patchset #1 (id:1) has been deleted

[DaleCurtis, 2014-09-05 21:12:13]

Since BufferedDataSource lives in content/ it's not used by any media_unittests.
 Faking delayed reads with FileDataSource doesn't work either without
implementing a FileDataSource::Stop() with the same restrictions as
BufferedDataSource.  So no test unless you've got a clever idea :(

[scherkus (not reviewing), 2014-09-09 17:29:28]

would writing a test be easier now that acolwell@ moved our code to
media/blink/?

[DaleCurtis, 2014-09-09 17:30:16]

Ah, I didn't realize he was moving BufferedDataSource too.  Let me think about
it.

[DaleCurtis, 2014-09-09 17:50:22]

Not really, the media/blink code is part of a new media_blink target, so we
can't use it directly in the FFmpegDemuxer unittest. We'd need to create a new
unittest file which lives in media_blink for FFmpegDemuxer and extract common
code from the BufferedDataSource unittest which mocks several things out.

I think there's some value in having some real BufferedDataSource+FFmpegDemuxer
tests, but that's a larger task than this fix (which ideally will be mergeable).

[scherkus (not reviewing), 2014-09-09 18:03:41]

lgtm

[DaleCurtis, 2014-09-09 18:05:55]

The CQ bit was checked by dalecurtis@chromium.org

[commit-bot: I haz the power, 2014-09-09 18:11:16]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/dalecurtis@chromium.org/544843005/20001

[commit-bot: I haz the power, 2014-09-09 20:04:45]

Committed patchset #1 (id:20001) as a8cffdefe2f7888d0dc01e8c7cf5ef312adf5611

[commit-bot: I haz the power, 2014-09-10 03:55:10]

Patchset 1 (id:??) landed as
https://crrev.com/1973f0b6f7497b1887397de9bb36dbff4faf5e39
Cr-Commit-Position: refs/heads/master@{#293995}