Re: [PATCH] mtd: fix: avoid race condition when accessing mtd->usecount

[Brian Norris]

On Thu, May 07, 2015 at 05:17:45PM -0700, Brian Norris wrote:
> On Thu, May 07, 2015 at 05:10:12PM -0700, Brian Norris wrote:
> > On Tue, Apr 21, 2015 at 12:20:22PM +0200, Giuseppe Cantavenera wrote:
> > > @@ -484,7 +486,7 @@ int del_mtd_blktrans_dev(struct mtd_blktrans_dev *old)
> > >  	if (old->open) {
> > >  		if (old->tr->release)
> > >  			old->tr->release(old);
> > > -		__put_mtd_device(old->mtd);
> > > +		put_mtd_device(old->mtd);
> > 
> > This looks wrong. See:
> [...]
> > deregister_mtd_blktrans()
> > |_ mutex_lock(&mtd_table_mutex)
> > |_ tr->remove_dev() -> inftl_remove_dev()
> >    |_ del_mtd_blktrans_dev()
> >       |_ put_mtd_device()
> >          |_ mutex_lock(&mtd_table_mutex) <--- AA deadlock
> 
> What's more, this code in del_mtd_blktrans_dev() makes it obvious that
> this hunk is wrong:
> 
> int del_mtd_blktrans_dev(struct mtd_blktrans_dev *old)
> {
>         unsigned long flags;
> 
>         if (mutex_trylock(&mtd_table_mutex)) {
>                 mutex_unlock(&mtd_table_mutex);
>                 BUG();
>         }
> 	...
> 
> So rather than a comment, the code is showing that it's a BUG() to not
> be holding mtd_table_mutex already.

As an alternative to your patch, how about the following?

BTW, this does still leave a usecount race in
drivers/mtd/maps/vmu-flash.c. But that driver should really be using
mtd->_get_device(), if it actually wants its own refcount.

Signed-off-by: Brian Norris <computersforpeace@xxxxxxxxx>
---

diff --git a/drivers/mtd/mtd_blkdevs.c b/drivers/mtd/mtd_blkdevs.c
index 2b0c52870999..df7c6c70757a 100644
--- a/drivers/mtd/mtd_blkdevs.c
+++ b/drivers/mtd/mtd_blkdevs.c
@@ -197,6 +197,7 @@ static int blktrans_open(struct block_device *bdev, fmode_t mode)
 		return -ERESTARTSYS; /* FIXME: busy loop! -arnd*/
 
 	mutex_lock(&dev->lock);
+	mutex_lock(&mtd_table_mutex);
 
 	if (dev->open)
 		goto unlock;
@@ -220,6 +221,7 @@ static int blktrans_open(struct block_device *bdev, fmode_t mode)
 
 unlock:
 	dev->open++;
+	mutex_unlock(&mtd_table_mutex);
 	mutex_unlock(&dev->lock);
 	blktrans_dev_put(dev);
 	return ret;
@@ -230,6 +232,7 @@ error_release:
 error_put:
 	module_put(dev->tr->owner);
 	kref_put(&dev->ref, blktrans_dev_release);
+	mutex_unlock(&mtd_table_mutex);
 	mutex_unlock(&dev->lock);
 	blktrans_dev_put(dev);
 	return ret;
@@ -243,6 +246,7 @@ static void blktrans_release(struct gendisk *disk, fmode_t mode)
 		return;
 
 	mutex_lock(&dev->lock);
+	mutex_lock(&mtd_table_mutex);
 
 	if (--dev->open)
 		goto unlock;
@@ -256,6 +260,7 @@ static void blktrans_release(struct gendisk *disk, fmode_t mode)
 		__put_mtd_device(dev->mtd);
 	}
 unlock:
+	mutex_unlock(&mtd_table_mutex);
 	mutex_unlock(&dev->lock);
 	blktrans_dev_put(dev);
 }
--
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/