Re: [PATCH] swap: fix set_blocksize race during swapon/swapoff

[Krzysztof Kozlowski]

On Fri, 2013-10-11 at 12:02 -0700, Andrew Morton wrote:
> (cc Hugh)
> 
> On Fri, 11 Oct 2013 11:54:22 +0200 Krzysztof Kozlowski <k.kozlowski@xxxxxxxxxxx> wrote:
> 
> > Swapoff used old_block_size from swap_info which could be overwritten by
> > concurrent swapon.
> > 
> > Reported-by: Weijie Yang <weijie.yang.kh@xxxxxxxxx>
> > Signed-off-by: Krzysztof Kozlowski <k.kozlowski@xxxxxxxxxxx>
> > ---
> >  mm/swapfile.c |    4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> > 
> > diff --git a/mm/swapfile.c b/mm/swapfile.c
> > index 3963fc2..de7c904 100644
> > --- a/mm/swapfile.c
> > +++ b/mm/swapfile.c
> > @@ -1824,6 +1824,7 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
> >  	struct filename *pathname;
> >  	int i, type, prev;
> >  	int err;
> > +	unsigned int old_block_size;
> >  
> >  	if (!capable(CAP_SYS_ADMIN))
> >  		return -EPERM;
> > @@ -1914,6 +1915,7 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
> >  	}
> >  
> >  	swap_file = p->swap_file;
> > +	old_block_size = p->old_block_size;
> >  	p->swap_file = NULL;
> >  	p->max = 0;
> >  	swap_map = p->swap_map;
> > @@ -1938,7 +1940,7 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
> >  	inode = mapping->host;
> >  	if (S_ISBLK(inode->i_mode)) {
> >  		struct block_device *bdev = I_BDEV(inode);
> > -		set_blocksize(bdev, p->old_block_size);
> > +		set_blocksize(bdev, old_block_size);
> >  		blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
> >  	} else {
> >  		mutex_lock(&inode->i_mutex);
> 
> I find it worrying that a swapon can run concurrently with any of this
> swapoff code.  It just seem to be asking for trouble and the code
> really isn't set up for this and races here will be poorly tested for
> 
> I'm wondering if we should just extend swapon_mutex a lot and eliminate
> the concurrency?

It seems there are even more races here between swapoff & swapon (and
swapon with swapon). Simple script:
	for i in `seq 1000`
	do
		swapoff -a &
		swapon -a &
	done
causes frequent switches of block size of devices (jumping from 512 to 4096).


Best regards,
Krzysztof



--
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/