[PATCH 2/2] loop : Don't hold the lo_ctl_mutex while fput in loop_clr_fd

[PATCH 2/2] loop : Don't hold the lo_ctl_mutex while fput in loop_clr_fd

[Dave Young]

[Fixing bug 10504]

If the bakingfile is a block device file, losetuo -d will trigger lockdep
warning of "circular locking dependency".

open/release lock order : bdev->bd_mutex ---> lo->lo_ctl_mutex
loop_clr_fd lock order  : lo->lo_ctl_mutex ---> bdev->bd_mutex (fput)

Don't hold the lo_ctl_mutex while fput in loop_clr_fd to fix it. It's safe
because all loop device state will be consistent here.

Signed-off-by: Dave Young <hidave.darkstar@gmail.com>

---
drivers/block/loop.c |    5 +++++
1 file changed, 5 insertions(+)

diff -upr linux/drivers/block/loop.c linux.new/drivers/block/loop.c
--- linux/drivers/block/loop.c	2008-04-25 10:10:25.000000000 +0800
+++ linux.new/drivers/block/loop.c	2008-04-25 10:10:35.000000000 +0800
@@ -923,7 +923,12 @@ static int loop_clr_fd(struct loop_devic
 	bd_set_size(bdev, 0);
 	mapping_set_gfp_mask(filp->f_mapping, gfp);
 	lo->lo_state = Lo_unbound;
+
+	/* unlock the lo_ctl_mutex to silence lockdep in case
+	 the backingfile is a block device */
+	mutex_unlock(&lo->lo_ctl_mutex);
 	fput(filp);
+	mutex_lock(&lo->lo_ctl_mutex);
 	/* This is safe: open() is still holding a reference. */
 	module_put(THIS_MODULE);
 	if (max_part > 0)

Re: [PATCH 2/2] loop : Don't hold the lo_ctl_mutex while fput in loop_clr_fd

[Al Viro]

On Mon, Apr 28, 2008 at 10:10:22AM +0800, Dave Young wrote:
> [Fixing bug 10504]
> 
> If the bakingfile is a block device file, losetuo -d will trigger lockdep
> warning of "circular locking dependency".
> 
> open/release lock order : bdev->bd_mutex ---> lo->lo_ctl_mutex
> loop_clr_fd lock order  : lo->lo_ctl_mutex ---> bdev->bd_mutex (fput)
> 
> Don't hold the lo_ctl_mutex while fput in loop_clr_fd to fix it. It's safe
> because all loop device state will be consistent here.

Explain.

Re: [PATCH 2/2] loop : Don't hold the lo_ctl_mutex while fput in loop_clr_fd

[Al Viro]

On Mon, Apr 28, 2008 at 03:13:09AM +0100, Al Viro wrote:
> On Mon, Apr 28, 2008 at 10:10:22AM +0800, Dave Young wrote:
> > [Fixing bug 10504]
> > 
> > If the bakingfile is a block device file, losetuo -d will trigger lockdep
> > warning of "circular locking dependency".
> > 
> > open/release lock order : bdev->bd_mutex ---> lo->lo_ctl_mutex
> > loop_clr_fd lock order  : lo->lo_ctl_mutex ---> bdev->bd_mutex (fput)
> > 
> > Don't hold the lo_ctl_mutex while fput in loop_clr_fd to fix it. It's safe
> > because all loop device state will be consistent here.
> 
> Explain.

BTW, explain also why open() at that spot will not do interesting things
wrt BLKRRPART done a bit below, please.

Re: [PATCH 2/2] loop : Don't hold the lo_ctl_mutex while fput in loop_clr_fd

[Dave Young]

On Mon, Apr 28, 2008 at 10:15 AM, Al Viro <viro@zeniv.linux.org.uk> wrote:
>
> On Mon, Apr 28, 2008 at 03:13:09AM +0100, Al Viro wrote:
>  > On Mon, Apr 28, 2008 at 10:10:22AM +0800, Dave Young wrote:
>  > > [Fixing bug 10504]
>  > >
>  > > If the bakingfile is a block device file, losetuo -d will trigger lockdep
>  > > warning of "circular locking dependency".
>  > >
>  > > open/release lock order : bdev->bd_mutex ---> lo->lo_ctl_mutex
>  > > loop_clr_fd lock order  : lo->lo_ctl_mutex ---> bdev->bd_mutex (fput)
>  > >
>  > > Don't hold the lo_ctl_mutex while fput in loop_clr_fd to fix it. It's safe
>  > > because all loop device state will be consistent here.
>  >
>  > Explain.

Hi, Maybe I have some mis-understood, thanks for pointing out in advance.

IMO the lo_ctl_mutex is for the lo->* state locking,  they are all set
to proper values before fput(filp), so I think it's safe to do that.

>
>  BTW, explain also why open() at that spot will not do interesting things
>  wrt BLKRRPART done a bit below, please.
>

open with a loop device unbound? Sorry, I don't understand.  Could you
explain a bit? Thanks

Regards
dave

Re: [PATCH 2/2] loop : Don't hold the lo_ctl_mutex while fput in loop_clr_fd

[Jiri Slaby]

On 28.4.2008 04:33, Dave Young wrote:
> On Mon, Apr 28, 2008 at 10:15 AM, Al Viro<viro@zeniv.linux.org.uk>  wrote:
>> On Mon, Apr 28, 2008 at 03:13:09AM +0100, Al Viro wrote:
>>   >  On Mon, Apr 28, 2008 at 10:10:22AM +0800, Dave Young wrote:
>>   >  >  [Fixing bug 10504]
>>   >  >
>>   >  >  If the bakingfile is a block device file, losetuo -d will trigger lockdep
>>   >  >  warning of "circular locking dependency".
>>   >  >
>>   >  >  open/release lock order : bdev->bd_mutex --->  lo->lo_ctl_mutex
>>   >  >  loop_clr_fd lock order  : lo->lo_ctl_mutex --->  bdev->bd_mutex (fput)
>>   >  >
>>   >  >  Don't hold the lo_ctl_mutex while fput in loop_clr_fd to fix it. It's safe
>>   >  >  because all loop device state will be consistent here.
>>   >
>>   >  Explain.
>
> Hi, Maybe I have some mis-understood, thanks for pointing out in advance.
>
> IMO the lo_ctl_mutex is for the lo->* state locking,  they are all set
> to proper values before fput(filp), so I think it's safe to do that.
>
>>   BTW, explain also why open() at that spot will not do interesting things
>>   wrt BLKRRPART done a bit below, please.
>>
>
> open with a loop device unbound? Sorry, I don't understand.  Could you
> explain a bit? Thanks

Any ideas here? Al? The bug survived almost a year...