Re: Linus GIT - INFO: possible circular locking dependency detected

From: Andrew Morton
Date: Tue Nov 08 2011 - 18:57:18 EST


On Sat, 5 Nov 2011 13:31:46 +0400
Vasiliy Kulikov <segoon@xxxxxxxxxxxx> wrote:

> (cc'ed Andrew and Alexey)
>
> On Thu, Nov 03, 2011 at 20:49 +0000, Luis Henriques wrote:
> > On Thu, Nov 03, 2011 at 11:57:20AM -0400, Miles Lane wrote:
> > ...
> >
> > I'm hitting the exact same problem, using a minimal .config file (can send
> > it if required), by just running "find /".
> >
> > I have bisected the problem and found that commit
> > aa6afca5bcaba8101f3ea09d5c3e4100b2b9f0e5 seems to be the cause of it.

Well, let's tell the -stable maintainer(s?) that aa6afca5bca ("proc:
fix races against execve() of /proc/PID/fd**") is known to cause a
regression.

> procfs holds sig->cred_guard_mutex to ensure the target's credentials are
> not changed. It is held for a little timeslice. From the stack trace I
> don't understand how sys_execve() can happen with ->cred_guard_mutex
> held:
>
> static struct dentry *proc_lookupfd_common(struct inode *dir,
> struct dentry *dentry,
> instantiate_t instantiate)
> {
> ...
> if (lock_trace(task))
> goto out;
>
> result = instantiate(dir, dentry, task, &fd);
> unlock_trace(task);
> ...
> }
>
>
> static int lock_trace(struct task_struct *task)
> {
> int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
> if (err)
> return err;
> if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
> mutex_unlock(&task->signal->cred_guard_mutex);
> return -EPERM;
> }
> return 0;
> }
>
> proc_lookupfd_common() always exits without ->cred_guard_mutex held.
>

Yes, it's a strange trace. udev is a strange thing and can be
triggered by the kernel at odd times. I wonder if it's possible that
some other process is now synchronously running udevd while holding
cred_guard_mutex. That wouldn't show up in the backtrace. But I doubt
if lockdep would notice it either...

Either way, it would be prudent to revert aa6afca5bca from mainline if
we can't get this fixed up soon.
--
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/