1021854 - nsFileStreamBase::DoOpen() leaking fds

Status: RESOLVED DUPLICATE of bug 1028456

(NSPR :: NSPR, defect)

Description

[Andrew McCreight [:mccr8]]

Attachment: stacks for leaking allocations

Running content/base/test/ locally, m5 and bc1 on TBPL.

Comment 1

[Andrew McCreight [:mccr8]]

Unfortunately on TBPL the stack is only showing through nsLocalFile::OpenNSPRFileDesc() (maybe due to that file or nsFileStreamBase somehow not being compiled with frame pointers?) so my suppression will have to be pretty generic.

Comment 2

[Nicholas Nethercote [inactive]]

I ran this command:

> mach mochitest-plain --keep-open content/base/test/

When it completed, I enabled memory.system_memory_reporter in about:config. The fds listed for the process were these:

├─────73 (05.16%) -- process(/home/njn/moz/trunk/co64/dist/bin/firefox, pid=23161)
│     ├──36 (02.54%) -- files
│     │  ├──32 (02.26%) -- tmp/tmpGk3PRH
│     │  │  ├───5 (00.35%) ++ places.sqlite
│     │  │  ├───5 (00.35%) ++ places.sqlite-wal
│     │  │  ├───4 (00.28%) ++ Cache
│     │  │  ├───2 (00.14%) ++ webappsstore.sqlite
│     │  │  ├───2 (00.14%) ++ webappsstore.sqlite-wal
│     │  │  ├───1 (00.07%) ── .parentlock/11
│     │  │  ├───1 (00.07%) ── OfflineCache/index.sqlite/56
│     │  │  ├───1 (00.07%) ── _CACHE_CLEAN_/47
│     │  │  ├───1 (00.07%) ── cert8.db/33
│     │  │  ├───1 (00.07%) ── content-prefs.sqlite/39
│     │  │  ├───1 (00.07%) ── cookies.sqlite/28
│     │  │  ├───1 (00.07%) ── formhistory.sqlite/64
│     │  │  ├───1 (00.07%) ── healthreport.sqlite-shm/63
│     │  │  ├───1 (00.07%) ── healthreport.sqlite-wal/57
│     │  │  ├───1 (00.07%) ── healthreport.sqlite/55
│     │  │  ├───1 (00.07%) ── key3.db/34
│     │  │  ├───1 (00.07%) ── permissions.sqlite/35
│     │  │  ├───1 (00.07%) ── places.sqlite-shm/38
│     │  │  └───1 (00.07%) ── webappsstore.sqlite-shm/44
│     │  └───4 (00.28%) -- (2 tiny)
│     │      ├──2 (00.14%) -- dev
│     │      │  ├──1 (00.07%) ── pts/13/0
│     │      │  └──1 (00.07%) ── urandom/17
│     │      └──2 (00.14%) -- proc
│     │         ├──1 (00.07%) ── 23161/fd/72
│     │         └──1 (00.07%) ── 71


Compared to the same measurements taken just after starting the browser, the only difference of note is the |Cache| line.

It's not conclusive whether this is a true fd leak -- mccr8's report only has two missing blocks. But it's not obviously a problem.

Comment 3

[Andrew McCreight [:mccr8]]

Here's another test this happens with, which at least partially explains the Moth failure:
  ./mach mochitest-chrome dom/apps/tests/

Comment 4

[Andrew McCreight [:mccr8]]

Oops, sorry, that one does not involve nsLocalFile::OpenNSPRFileDesc, so that's a different issue, bug 1021302.

Comment 5

[Andrew McCreight [:mccr8]]

This may be fixed now, so I'll do a try run with the suppression removed.

Comment 6

[Andrew McCreight [:mccr8]]

This is probably also actually an NSPR issue, if it still exists.

Comment 7

[Andrew McCreight [:mccr8]]

This is probably just a specific instance of whatever fd caching issue there is in NSPR. I'll remove the specific suppression for this if I can.