Re: [PATCH] elevator: Fix a race in elevator switching and md device initialization

[Vivek Goyal <vgoyal@redhat.com>]

On Thu, Aug 29, 2013 at 07:29:07PM +0000, Tomoki Sekiyama wrote:
> On 8/29/13 14:43 , "Vivek Goyal" <vgoyal@redhat.com> wrote:
> >On Thu, Aug 29, 2013 at 02:33:10PM -0400, Vivek Goyal wrote:
> >> On Mon, Aug 26, 2013 at 09:45:15AM -0400, Tomoki Sekiyama wrote:
> >> > The soft lockup below happes at the boot time of the system using dm
> >> > multipath and automated elevator switching udev rules.
> >> > 
> >> > [  356.127001] BUG: soft lockup - CPU#3 stuck for 22s! [sh:483]
> >> > [  356.127001] RIP: 0010:[<ffffffff81072a7d>]  [<ffffffff81072a7d>]
> >>lock_timer_base.isra.35+0x1d/0x50
> >> > ...
> >> > [  356.127001] Call Trace:
> >> > [  356.127001]  [<ffffffff81073810>] try_to_del_timer_sync+0x20/0x70
> >> > [  356.127001]  [<ffffffff8118b08a>] ?
> >>kmem_cache_alloc_node_trace+0x20a/0x230
> >> > [  356.127001]  [<ffffffff810738b2>] del_timer_sync+0x52/0x60
> >> > [  356.127001]  [<ffffffff812ece22>] cfq_exit_queue+0x32/0xf0
> >> > [  356.127001]  [<ffffffff812c98df>] elevator_exit+0x2f/0x50
> >> > [  356.127001]  [<ffffffff812c9f21>] elevator_change+0xf1/0x1c0
> >> > [  356.127001]  [<ffffffff812caa50>] elv_iosched_store+0x20/0x50
> >> > [  356.127001]  [<ffffffff812d1d09>] queue_attr_store+0x59/0xb0
> >> > [  356.127001]  [<ffffffff812143f6>] sysfs_write_file+0xc6/0x140
> >> > [  356.127001]  [<ffffffff811a326d>] vfs_write+0xbd/0x1e0
> >> > [  356.127001]  [<ffffffff811a3ca9>] SyS_write+0x49/0xa0
> >> > [  356.127001]  [<ffffffff8164e899>] system_call_fastpath+0x16/0x1b
> >> > 
> >> 
> >> Tokomi, 
> >> 
> >> As you noticed, there is a fedora bug open with similar signature. May
> >> be this patch will fix that issue also.
> >> 
> >> https://bugzilla.redhat.com/show_bug.cgi?id=902012
> >> 
> >> 
> >> > This is caused by a race between md device initialization and sysfs
> >>knob
> >> > to switch the scheduler.
> >> > 
> >> > * multipathd:
> >> >  SyS_ioctl -> do_vfs_ioctl -> dm_ctl_ioctl -> ctl_ioctl ->  table_load
> >> >   -> dm_setup_md_queue -> blk_init_allocated_queue -> elevator_init:
> >> > 
> >> >     q->elevator = elevator_alloc(q, e); // not yet initialized
> >> > 
> >> > * sh -c 'echo deadline > /sys/$DEVPATH/queue/scheduler'
> >> >  SyS_write -> vfs_write -> sysfs_write_file -> queue_attr_store
> >> >      ( mutex_lock(&q->sysfs_lock) here. )
> >> >   -> elv_iosched_store -> elevator_change:
> >> > 
> >> >   elevator_exit(old); // try to de-init uninitialized elevator and
> >>hang up
> >> > 
> >
> >If problem in this case is that we are trying to exit() the elevator
> >which has not been properly initialized, then we should not attach
> >the elevator to the queue yet.
> >
> >In cfq_init_queue(), can we move following code towards the end of
> >function.
> >
> >        spin_lock_irq(q->queue_lock);
> >        q->elevator = eq;
> >        spin_unlock_irq(q->queue_lock);
> >
> >So till elevator is initialized, we will not attach it to queue and
> >elevator_switch() will return as it will not find a valid elevator
> >on the queue.
> >
> >
> >elevator_change() {
> >	        if (!q->elevator)
> >                return -ENXIO;
> >}
> >
> >Thanks
> >Vivek
> 
> I think it also works, though I prefer introducing explicit locking,
> as you said, so that code won't break again in some future.

I agree. Providing explicit locking and making sure only one elevator
can be initializing at a time on a queue and others wait till that 
operation is complete, will make up the code more readable and less
bug suspecible.

Thanks
Vivek