Re: 3.14-rc2 XFS backtrace because irqs_disabled.

[Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>]

On Tue, Feb 11, 2014 at 5:09 PM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> Slap the check in vfs_create(), see if interrupts had been disabled by it or
> by something in ->create().  Since it's reproducible...

path_openat() starts off with a get_empty_filp(), which allocates a
file pointer with GFP_KERNEL. So that should have triggered the
might_sleep warning if irq's were already disabled at that point.

So it's not before that - in particular, it's not in the signal
handling or do_coredump() paths.

Also, at least xfs_buf_lock() - which is much deeper in that chain -
does a down(&bp->b_sema). I'm disguested that that doesn't have a
might_sleep() in it.

Dave, mind adding a "might_sleep()" to the top of
"down[_interruptible]()". It's silly to not have coverage of semaphore
use in bad contexts.

             Linus