bdev: fix NULL pointer dereference in sync()/close() race

Commit Message

Vegard Nossum Aug. 29, 2016, 9:33 p.m. UTC

On 08/29/2016 09:55 PM, Tejun Heo wrote:
> On Sat, Aug 27, 2016 at 11:30:22AM +0200, Vegard Nossum wrote:
>> If people who are more savvy in block/fs code could ack the locking bits
>> I think we should apply the patch ASAP because it's an easy local DOS if
>> you have (open/read) access to any block device.
>
> I think the right thing to do there is doing blkdev_get() /
> blkdev_put() around func() invocation in iterate_bdevs() rather than
> holding bd_mutex across the callback.  Can you please verify whether
> that works?

Didn't work for me, I kept getting use-after-free in __blkdev_get() on
bdev->bd_invalidated after it calls bdev->bd_disk->fops->open(). I tried
a few related things without much luck.

The only thing that worked for me without holding the mutex across the
call was this:

  		if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
@@ -1906,7 +1907,19 @@ void iterate_bdevs(void (*func)(struct 
block_device *, void *), void *arg)
  		iput(old_inode);
  		old_inode = inode;

-		func(I_BDEV(inode), arg);
+		bdev = I_BDEV(inode);
+
+		mutex_lock(&bdev->bd_mutex);
+		bdev->bd_openers++;
+		bdev->bd_holders++;
+		mutex_unlock(&bdev->bd_mutex);
+
+		func(bdev, arg);
+
+		mutex_lock(&bdev->bd_mutex);
+		bdev->bd_openers--;
+		bdev->bd_holders--;
+		mutex_unlock(&bdev->bd_mutex);

  		spin_lock(&blockdev_superblock->s_inode_list_lock);
  	}

I'm guessing that's too simple to work in general (especially when you
bring in partitions and stuff; I'm just opening /dev/sr0 in my reproducer).

It's been a long day, I'll have a look tomorrow and see if I didn't just
do something stupid.


Vegard
--
To unsubscribe from this list: send the line "unsubscribe linux-block" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Tejun Heo Aug. 29, 2016, 11:49 p.m. UTC | #1

Hello,

On Mon, Aug 29, 2016 at 11:33:41PM +0200, Vegard Nossum wrote:
> On 08/29/2016 09:55 PM, Tejun Heo wrote:
> > I think the right thing to do there is doing blkdev_get() /
> > blkdev_put() around func() invocation in iterate_bdevs() rather than
> > holding bd_mutex across the callback.  Can you please verify whether
> > that works?
> 
> Didn't work for me, I kept getting use-after-free in __blkdev_get() on
> bdev->bd_invalidated after it calls bdev->bd_disk->fops->open(). I tried
> a few related things without much luck.

I see.  It could be that it's doing blkdev_get() on a dying device.

> The only thing that worked for me without holding the mutex across the
> call was this:
...
> +		mutex_lock(&bdev->bd_mutex);
> +		bdev->bd_openers++;
> +		bdev->bd_holders++;
> +		mutex_unlock(&bdev->bd_mutex);
> +
> +		func(bdev, arg);
> +
> +		mutex_lock(&bdev->bd_mutex);
> +		bdev->bd_openers--;
> +		bdev->bd_holders--;
> +		mutex_unlock(&bdev->bd_mutex);

And this might not be too far fetched.  I think what we want is

* Bump bd_openers if there are other users already; otherwise, skip.

* blkdev_put() after the callback is finished.

Thanks.

Patch

diff --git a/fs/block_dev.c b/fs/block_dev.c
index 08ae993..586d745 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -1885,6 +1885,7 @@  void iterate_bdevs(void (*func)(struct 
block_device *, void *), void *arg)
  	spin_lock(&blockdev_superblock->s_inode_list_lock);
  	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
  		struct address_space *mapping = inode->i_mapping;
+		struct block_device *bdev;

  		spin_lock(&inode->i_lock);