Transition POSIX shmem to use lockf(), not semaphores.

Transition POSIX shmem to use lockf(), not semaphores.  Eliminates
resource leak possibility; allows complete sharing of shmem by fd
without name.
Add actual multiprocess shmem/locking unit test.
Fix flaky shmem unit test.

agl: please review.
Amanda: please look over as you've been here before.


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=9235

[agl, 2009-02-05 00:48:08]

LGTM apart from the comment below

http://codereview.chromium.org/21063/diff/1/3
File base/shared_memory_posix.cc (right):

http://codereview.chromium.org/21063/diff/1/3#newcode252
Line 252: while (lockf(mapped_file_, F_LOCK, 0) < 0) {
From man lockf, it looks like we want to handle the errors differently here.
Specifically, EAGAIN suggests a programming error because we aren't using TLOCK.

while (lockf(mapped_file_, F_LOCK, 0) < 0) {
  if (errno == EINTR)
    continue;
  if (errno == ENOLCK) {
    // temporary kernel resource exaustion
    PlatformThread::Sleep(500);
    continue;
  }
  NOTREACHED() << "Failed to lock shared memory region"
               << " fd:" << mapped_file_
               << " errno:" << errno;
               << " msg:" << strerror(errno);
}

http://codereview.chromium.org/21063/diff/1/3#newcode260
Line 260: DCHECK(errno == EAGAIN || errno == EINTR);
Ditto with the above

[Avi (use Gerrit), 2009-02-05 01:00:49]

http://codereview.chromium.org/21063/diff/1/4
File base/shared_memory.h (right):

http://codereview.chromium.org/21063/diff/1/4#newcode127
Line 127: // lock at the same time.  There are serveral solutions for this
spelling

http://codereview.chromium.org/21063/diff/1/2
File base/shared_memory_unittest.cc (right):

http://codereview.chromium.org/21063/diff/1/2#newcode170
Line 170: // (make it not flaky) but starting with 1 thread, then
...by starting...

http://codereview.chromium.org/21063/diff/1/2#newcode171
Line 171: // intentionally don't clean up it's shmem before running with
its

[Amanda Walker, 2009-02-05 01:47:51]

LGTM with agl's & avi's comments applied.

[John Grabowski, 2009-02-05 02:34:54]

On 2009/02/05 01:47:51, Amanda Walker wrote:
> LGTM with agl's & avi's comments applied.

Thanks for the feedback guys and girls.
Diffs uploaded.

To fully disclose, I snuck in one more change to fix a purify problem (in
AnonymousPrivate I moved 2 "new blah[]"s into scoped_array<>s).  I will wait
until morning to be sure you have a chance to object should you wish.

[Amanda Walker, 2009-02-05 03:34:46]

Still LGTM