Re: Possible memory leak via alloc_pid()

[Eric W. Biederman]

Catalin Marinas <catalin.marinas@xxxxxxx> writes:

> On Sun, 2009-08-02 at 18:44 -0700, Eric W. Biederman wrote:
>> Hmm.  I'm starting to wonder if kmemleak is right.  I don't know how
>> it works but something about the way pids are used might be confusing it.
>
> It could as well be a false positive but I can't find its source.
>
> Basically, the pid structure for the dead Xorg is still allocated
> minutes after Xorg died with a pid->count of 2. Kmemleak scans the data
> and bss sections, task stacks and most of the allocated objects (which
> are not reported as leaks) but cannot find a pointer to this pid
> structure (or anywhere inside it like pid->number.pid_chain).
>
> The supposedly leaked pid structure also have pid_chain.pprev ==
> LIST_POISON2 which means that it was already removed from the pid_hash
> (this block of memory is scanned by kmemleak anyway).
>
> The free_pid() function was also called on this object according to the
> pid->rcu values but put_pid() couldn't free it because of pid->count.
>
> If this structure in not on pid_hash, is there any other place where its
> pointer may be stored for a long time? Otherwise it looks like a real
> leak (though not a big one).

It depends on how it is used. Pids not on the pid_hash are perfectly
fine.  We use these kinds of long standing pid references to prevent
any chance that pid rollover would be a problem.   There are corner
cases in the SIGIO path and a few other places where we have
these kind of long standing pid references.

Perhaps it comes from the tty switching code?

> I'll do more tests in the next few days as suggested by Oleg.

Thanks, and thanks for spotting into and looking into this.

Eric
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/