Re: Thread-unsafe access to lock->owner in PR_Lock

On Thu, Dec 19, 2013 at 3:47 PM, Wan-Teh Chang <[email protected]> wrote:
>
> Hi Alexander,
>
> PRLock is a non-recursive lock. So the PR_ASSERT(lock->owner != me) on
> line 200 is trying to detect an attempt to lock a PRLock recursively.
> It is intended as a debugging aid for NSPR users.
>
> This debugging aid is useful, so ideally we should find a way to do it
> the right way.
>
> If we swap lines 200 and 201, it will only work for a _PR_MD_LOCK
> implementation that allows a lock to be acquired recursively. As Dave
> Hylands pointed out, in a _PR_MD_LOCK implementation that deadlocks,
> we won't be able to see the assertion failure. However, since we only
> use the PR_Lock code in prulock.c on Windows, where _PR_MD_LOCK
> (EnterCriticalSection) is recursive, we can make the change you
> suggested.

I filed an NSPR bug report and wrote a patch:
https://bugzilla.mozilla.org/show_bug.cgi?id=952621

Wan-Teh Chang