982111 - Leaks in nestegg_read_packet called by WebMReader::NextPacket

Status: RESOLVED WORKSFORME

(Core :: Audio/Video, defect)

Description

[Andrew McCreight [:mccr8]]

Attachment: webm.txt

There are a total of about 11KB of leaks with stacks that look like this:

Indirect leak of 9141 byte(s) in 33 object(s) allocated from:
    #0 0x446434 in calloc /builds/slave/moz-toolchain/src/llvm/projects/compiler-rt/lib/asan/asan_malloc_linux.cc:90
    #1 0x7f4cf4883b78 in ne_alloc /somepath/media/libnestegg/src/nestegg.c:507
    #2 0x7f4cf4883b78 in ne_read_block /somepath/media/libnestegg/src/nestegg.c:1356
    #3 0x7f4cf4883b78 in nestegg_read_packet /somepath/media/libnestegg/src/nestegg.c:2190
    #4 0x7f4cf29fec95 in mozilla::WebMReader::NextPacket(mozilla::WebMReader::TrackType) /somepath/content/media/webm/WebMReader.cpp:788
    #5 0x7f4cf29ff4fd in mozilla::WebMReader::DecodeVideoFrame(bool&, long) /somepath/content/media/webm/WebMReader.cpp:869
    #6 0x7f4cf28c4411 in mozilla::MediaDecoderStateMachine::DecodeLoop() /somepath/content/media/MediaDecoderStateMachine.cpp:700

Comment 1

[Andrew McCreight [:mccr8]]

Attachment: WebM leak stacks

I got more complete stacks by running LSAN locally.

Comment 2

[Nathan Froyd [:froydnj]]

Some of the things in the vicinity of WebMReader could stand to be leak-checked.  I think the memory management for nestegg_packets is correct and we are leaking the containing object(s) somewhere.

Comment 3

[Andrew McCreight [:mccr8]]

Ah, yeah, the "indirect leak" in comment 0 means that LSAN found another block that holds onto this one, and it thinks the other one doesn't have any references.  I'll try to dig around more when I get the chance.

Comment 4

[Andrew McCreight [:mccr8]]

Hmm, okay, in the full set of leak stacks, there is one that's a "direct leak", so LSAN at least thinks some of it is being leaked directly.

Comment 5

[Andrew McCreight [:mccr8]]

Attachment: OGG reader leaks

I suspect these indirect leaks of the allocations in the Ogg reader are related.

Comment 6

[Matthew Gregan [:kinetik]]

(In reply to Andrew McCreight [:mccr8] from comment #5)
> Created attachment 8395852 [details]
> OGG reader leaks
> 
> I suspect these indirect leaks of the allocations in the Ogg reader are
> related.

Looks like MediaMetadataManager::DispatchMetadataIfNeeded is leaking these.  Filed bug 990892.

Comment 7

[Matthew Gregan [:kinetik]]

Also filed bug 990817 to clean up WebMReader's memory management in a couple of places.  But I have no idea where/why it's leaking.

Comment 8

[Andrew McCreight [:mccr8]]

I don't see this leak any more, so I'm going to remove the suppression in bug 1201096. Maybe bug 1198107 fixed this.