linux-block.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] block: loop: fix deadlock between open and remove
@ 2021-06-05 14:09 Christoph Hellwig
  2021-06-05 15:00 ` Colin Ian King
                   ` (2 more replies)
  0 siblings, 3 replies; 7+ messages in thread
From: Christoph Hellwig @ 2021-06-05 14:09 UTC (permalink / raw)
  To: axboe; +Cc: ming.lei, pasha.tatashin, linux-block, Colin Ian King

Commit c76f48eb5c08 ("block: take bd_mutex around delete_partitions in
del_gendisk") adds disk->part0->bd_mutex in del_gendisk(), this way
causes the following AB/BA deadlock between removing loop and opening
loop:

 1) loop_control_ioctl(LOOP_CTL_REMOVE)
     -> mutex_lock(&loop_ctl_mutex)
     -> del_gendisk
         -> mutex_lock(&disk->part0->bd_mutex)

 2) blkdev_get_by_dev
     -> mutex_lock(&disk->part0->bd_mutex)
     -> lo_open
         -> mutex_lock(&loop_ctl_mutex)

Add a new Lo_deleting state to remove the need for clearing
->private_data and thus holding loop_ctl_mutex in the ioctl
LOOP_CTL_REMOVE path.

Based on an analysis and earlier patch from
Ming Lei <ming.lei@redhat.com>.

Reported-by: Colin Ian King <colin.king@canonical.com>
Fixes: c76f48eb5c08 ("block: take bd_mutex around delete_partitions in del_gendisk")
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/block/loop.c | 25 +++++++------------------
 drivers/block/loop.h |  1 +
 2 files changed, 8 insertions(+), 18 deletions(-)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index d58d68f3c7cd..76e12f3482a9 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -1879,29 +1879,18 @@ static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
 
 static int lo_open(struct block_device *bdev, fmode_t mode)
 {
-	struct loop_device *lo;
+	struct loop_device *lo = bdev->bd_disk->private_data;
 	int err;
 
-	/*
-	 * take loop_ctl_mutex to protect lo pointer from race with
-	 * loop_control_ioctl(LOOP_CTL_REMOVE), however, to reduce contention
-	 * release it prior to updating lo->lo_refcnt.
-	 */
-	err = mutex_lock_killable(&loop_ctl_mutex);
-	if (err)
-		return err;
-	lo = bdev->bd_disk->private_data;
-	if (!lo) {
-		mutex_unlock(&loop_ctl_mutex);
-		return -ENXIO;
-	}
 	err = mutex_lock_killable(&lo->lo_mutex);
-	mutex_unlock(&loop_ctl_mutex);
 	if (err)
 		return err;
-	atomic_inc(&lo->lo_refcnt);
+	if (lo->lo_state == Lo_deleting)
+		err = -ENXIO;
+	else
+		atomic_inc(&lo->lo_refcnt);
 	mutex_unlock(&lo->lo_mutex);
-	return 0;
+	return err;
 }
 
 static void lo_release(struct gendisk *disk, fmode_t mode)
@@ -2285,7 +2274,7 @@ static long loop_control_ioctl(struct file *file, unsigned int cmd,
 			mutex_unlock(&lo->lo_mutex);
 			break;
 		}
-		lo->lo_disk->private_data = NULL;
+		lo->lo_state = Lo_deleting;
 		mutex_unlock(&lo->lo_mutex);
 		idr_remove(&loop_index_idr, lo->lo_number);
 		loop_remove(lo);
diff --git a/drivers/block/loop.h b/drivers/block/loop.h
index a3c04f310672..5beb959b94d3 100644
--- a/drivers/block/loop.h
+++ b/drivers/block/loop.h
@@ -22,6 +22,7 @@ enum {
 	Lo_unbound,
 	Lo_bound,
 	Lo_rundown,
+	Lo_deleting,
 };
 
 struct loop_func_table;
-- 
2.30.2


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH] block: loop: fix deadlock between open and remove
  2021-06-05 14:09 [PATCH] block: loop: fix deadlock between open and remove Christoph Hellwig
@ 2021-06-05 15:00 ` Colin Ian King
  2021-06-07  0:11 ` Ming Lei
  2021-06-11 17:51 ` Jens Axboe
  2 siblings, 0 replies; 7+ messages in thread
From: Colin Ian King @ 2021-06-05 15:00 UTC (permalink / raw)
  To: Christoph Hellwig, axboe; +Cc: ming.lei, pasha.tatashin, linux-block

On 05/06/2021 15:09, Christoph Hellwig wrote:
> Commit c76f48eb5c08 ("block: take bd_mutex around delete_partitions in
> del_gendisk") adds disk->part0->bd_mutex in del_gendisk(), this way
> causes the following AB/BA deadlock between removing loop and opening
> loop:
> 
>  1) loop_control_ioctl(LOOP_CTL_REMOVE)
>      -> mutex_lock(&loop_ctl_mutex)
>      -> del_gendisk
>          -> mutex_lock(&disk->part0->bd_mutex)
> 
>  2) blkdev_get_by_dev
>      -> mutex_lock(&disk->part0->bd_mutex)
>      -> lo_open
>          -> mutex_lock(&loop_ctl_mutex)
> 
> Add a new Lo_deleting state to remove the need for clearing
> ->private_data and thus holding loop_ctl_mutex in the ioctl
> LOOP_CTL_REMOVE path.
> 
> Based on an analysis and earlier patch from
> Ming Lei <ming.lei@redhat.com>.
> 
> Reported-by: Colin Ian King <colin.king@canonical.com>
> Fixes: c76f48eb5c08 ("block: take bd_mutex around delete_partitions in del_gendisk")
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  drivers/block/loop.c | 25 +++++++------------------
>  drivers/block/loop.h |  1 +
>  2 files changed, 8 insertions(+), 18 deletions(-)
> 
> diff --git a/drivers/block/loop.c b/drivers/block/loop.c
> index d58d68f3c7cd..76e12f3482a9 100644
> --- a/drivers/block/loop.c
> +++ b/drivers/block/loop.c
> @@ -1879,29 +1879,18 @@ static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
>  
>  static int lo_open(struct block_device *bdev, fmode_t mode)
>  {
> -	struct loop_device *lo;
> +	struct loop_device *lo = bdev->bd_disk->private_data;
>  	int err;
>  
> -	/*
> -	 * take loop_ctl_mutex to protect lo pointer from race with
> -	 * loop_control_ioctl(LOOP_CTL_REMOVE), however, to reduce contention
> -	 * release it prior to updating lo->lo_refcnt.
> -	 */
> -	err = mutex_lock_killable(&loop_ctl_mutex);
> -	if (err)
> -		return err;
> -	lo = bdev->bd_disk->private_data;
> -	if (!lo) {
> -		mutex_unlock(&loop_ctl_mutex);
> -		return -ENXIO;
> -	}
>  	err = mutex_lock_killable(&lo->lo_mutex);
> -	mutex_unlock(&loop_ctl_mutex);
>  	if (err)
>  		return err;
> -	atomic_inc(&lo->lo_refcnt);
> +	if (lo->lo_state == Lo_deleting)
> +		err = -ENXIO;
> +	else
> +		atomic_inc(&lo->lo_refcnt);
>  	mutex_unlock(&lo->lo_mutex);
> -	return 0;
> +	return err;
>  }
>  
>  static void lo_release(struct gendisk *disk, fmode_t mode)
> @@ -2285,7 +2274,7 @@ static long loop_control_ioctl(struct file *file, unsigned int cmd,
>  			mutex_unlock(&lo->lo_mutex);
>  			break;
>  		}
> -		lo->lo_disk->private_data = NULL;
> +		lo->lo_state = Lo_deleting;
>  		mutex_unlock(&lo->lo_mutex);
>  		idr_remove(&loop_index_idr, lo->lo_number);
>  		loop_remove(lo);
> diff --git a/drivers/block/loop.h b/drivers/block/loop.h
> index a3c04f310672..5beb959b94d3 100644
> --- a/drivers/block/loop.h
> +++ b/drivers/block/loop.h
> @@ -22,6 +22,7 @@ enum {
>  	Lo_unbound,
>  	Lo_bound,
>  	Lo_rundown,
> +	Lo_deleting,
>  };
>  
>  struct loop_func_table;
> 

Tested and no longer hangs with the stress-ng loop test. Thank you for
the speedy turnaround on the fix.

Tested-by: Colin Ian King <colin.king@canonical.com>

Colin

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH] block: loop: fix deadlock between open and remove
  2021-06-05 14:09 [PATCH] block: loop: fix deadlock between open and remove Christoph Hellwig
  2021-06-05 15:00 ` Colin Ian King
@ 2021-06-07  0:11 ` Ming Lei
  2021-06-11 17:51 ` Jens Axboe
  2 siblings, 0 replies; 7+ messages in thread
From: Ming Lei @ 2021-06-07  0:11 UTC (permalink / raw)
  To: Christoph Hellwig; +Cc: axboe, pasha.tatashin, linux-block, Colin Ian King

On Sat, Jun 05, 2021 at 05:09:50PM +0300, Christoph Hellwig wrote:
> Commit c76f48eb5c08 ("block: take bd_mutex around delete_partitions in
> del_gendisk") adds disk->part0->bd_mutex in del_gendisk(), this way
> causes the following AB/BA deadlock between removing loop and opening
> loop:
> 
>  1) loop_control_ioctl(LOOP_CTL_REMOVE)
>      -> mutex_lock(&loop_ctl_mutex)
>      -> del_gendisk
>          -> mutex_lock(&disk->part0->bd_mutex)
> 
>  2) blkdev_get_by_dev
>      -> mutex_lock(&disk->part0->bd_mutex)
>      -> lo_open
>          -> mutex_lock(&loop_ctl_mutex)
> 
> Add a new Lo_deleting state to remove the need for clearing
> ->private_data and thus holding loop_ctl_mutex in the ioctl
> LOOP_CTL_REMOVE path.
> 
> Based on an analysis and earlier patch from
> Ming Lei <ming.lei@redhat.com>.
> 
> Reported-by: Colin Ian King <colin.king@canonical.com>
> Fixes: c76f48eb5c08 ("block: take bd_mutex around delete_partitions in del_gendisk")
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Reviewed-by: Ming Lei <ming.lei@redhat.com>

-- 
Ming


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH] block: loop: fix deadlock between open and remove
  2021-06-05 14:09 [PATCH] block: loop: fix deadlock between open and remove Christoph Hellwig
  2021-06-05 15:00 ` Colin Ian King
  2021-06-07  0:11 ` Ming Lei
@ 2021-06-11 17:51 ` Jens Axboe
  2 siblings, 0 replies; 7+ messages in thread
From: Jens Axboe @ 2021-06-11 17:51 UTC (permalink / raw)
  To: Christoph Hellwig; +Cc: ming.lei, pasha.tatashin, linux-block, Colin Ian King

On 6/5/21 8:09 AM, Christoph Hellwig wrote:
> Commit c76f48eb5c08 ("block: take bd_mutex around delete_partitions in
> del_gendisk") adds disk->part0->bd_mutex in del_gendisk(), this way
> causes the following AB/BA deadlock between removing loop and opening
> loop:
> 
>  1) loop_control_ioctl(LOOP_CTL_REMOVE)
>      -> mutex_lock(&loop_ctl_mutex)
>      -> del_gendisk
>          -> mutex_lock(&disk->part0->bd_mutex)
> 
>  2) blkdev_get_by_dev
>      -> mutex_lock(&disk->part0->bd_mutex)
>      -> lo_open
>          -> mutex_lock(&loop_ctl_mutex)
> 
> Add a new Lo_deleting state to remove the need for clearing
> ->private_data and thus holding loop_ctl_mutex in the ioctl
> LOOP_CTL_REMOVE path.
> 
> Based on an analysis and earlier patch from
> Ming Lei <ming.lei@redhat.com>.

Applied, thanks.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH] block: loop: fix deadlock between open and remove
  2021-06-04  6:57 ` Christoph Hellwig
@ 2021-06-04  8:56   ` Ming Lei
  0 siblings, 0 replies; 7+ messages in thread
From: Ming Lei @ 2021-06-04  8:56 UTC (permalink / raw)
  To: Christoph Hellwig; +Cc: Jens Axboe, linux-block, Colin Ian King, Pavel Tatashin

On Fri, Jun 04, 2021 at 08:57:33AM +0200, Christoph Hellwig wrote:
> This looks good, but I think we can do simple by just adding add new
> Lo_deleting state.  Completely untested as I'm at the tail end of a vacation
> with a broken laptop:

Indeed, adding one deleting state is simpler to kill the lock of loop_ctl_mutex
in lo_open():

Reviewed-by: Ming Lei <ming.lei@redhat.com>

-- 
Ming


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH] block: loop: fix deadlock between open and remove
  2021-06-04  0:04 Ming Lei
@ 2021-06-04  6:57 ` Christoph Hellwig
  2021-06-04  8:56   ` Ming Lei
  0 siblings, 1 reply; 7+ messages in thread
From: Christoph Hellwig @ 2021-06-04  6:57 UTC (permalink / raw)
  To: Ming Lei
  Cc: Jens Axboe, Christoph Hellwig, linux-block, Colin Ian King,
	Pavel Tatashin

This looks good, but I think we can do simple by just adding add new
Lo_deleting state.  Completely untested as I'm at the tail end of a vacation
with a broken laptop:

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index d58d68f3c7cd..76e12f3482a9 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -1879,29 +1879,18 @@ static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
 
 static int lo_open(struct block_device *bdev, fmode_t mode)
 {
-	struct loop_device *lo;
+	struct loop_device *lo = bdev->bd_disk->private_data;
 	int err;
 
-	/*
-	 * take loop_ctl_mutex to protect lo pointer from race with
-	 * loop_control_ioctl(LOOP_CTL_REMOVE), however, to reduce contention
-	 * release it prior to updating lo->lo_refcnt.
-	 */
-	err = mutex_lock_killable(&loop_ctl_mutex);
-	if (err)
-		return err;
-	lo = bdev->bd_disk->private_data;
-	if (!lo) {
-		mutex_unlock(&loop_ctl_mutex);
-		return -ENXIO;
-	}
 	err = mutex_lock_killable(&lo->lo_mutex);
-	mutex_unlock(&loop_ctl_mutex);
 	if (err)
 		return err;
-	atomic_inc(&lo->lo_refcnt);
+	if (lo->lo_state == Lo_deleting)
+		err = -ENXIO;
+	else
+		atomic_inc(&lo->lo_refcnt);
 	mutex_unlock(&lo->lo_mutex);
-	return 0;
+	return err;
 }
 
 static void lo_release(struct gendisk *disk, fmode_t mode)
@@ -2285,7 +2274,7 @@ static long loop_control_ioctl(struct file *file, unsigned int cmd,
 			mutex_unlock(&lo->lo_mutex);
 			break;
 		}
-		lo->lo_disk->private_data = NULL;
+		lo->lo_state = Lo_deleting;
 		mutex_unlock(&lo->lo_mutex);
 		idr_remove(&loop_index_idr, lo->lo_number);
 		loop_remove(lo);
diff --git a/drivers/block/loop.h b/drivers/block/loop.h
index a3c04f310672..5beb959b94d3 100644
--- a/drivers/block/loop.h
+++ b/drivers/block/loop.h
@@ -22,6 +22,7 @@ enum {
 	Lo_unbound,
 	Lo_bound,
 	Lo_rundown,
+	Lo_deleting,
 };
 
 struct loop_func_table;

^ permalink raw reply	[flat|nested] 7+ messages in thread

* [PATCH] block: loop: fix deadlock between open and remove
@ 2021-06-04  0:04 Ming Lei
  2021-06-04  6:57 ` Christoph Hellwig
  0 siblings, 1 reply; 7+ messages in thread
From: Ming Lei @ 2021-06-04  0:04 UTC (permalink / raw)
  To: Jens Axboe, Christoph Hellwig
  Cc: linux-block, Ming Lei, Colin Ian King, Pavel Tatashin

Commit c76f48eb5c08 ("block: take bd_mutex around delete_partitions in
del_gendisk") adds disk->part0->bd_mutex in del_gendisk(), this way
causes the following AB/BA deadlock between removing loop and opening
loop:

1) loop_control_ioctl(LOOP_CTL_REMOVE)
- mutex_lock(&loop_ctl_mutex)
- mutex_lock(&disk->part0->bd_mutex)	//del_gendisk

2) open look device
- mutex_lock(&disk->part0->bd_mutex)	//blkdev_get_by_dev
- mutex_lock(&loop_ctl_mutex)		//lo_open() <- __blkdev_get

Fixes the issue by not holding loop_ctl_mutex in lo_open(), and cover
the protection on bdev->bd_disk->private_data via disk->part0->bd_mutex.

Reported-by: Colin Ian King <colin.king@canonical.com>
Fixes: c76f48eb5c08 ("block: take bd_mutex around delete_partitions in del_gendisk")
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 drivers/block/loop.c | 31 +++++++++++++++----------------
 1 file changed, 15 insertions(+), 16 deletions(-)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index d58d68f3c7cd..b03d8f4c1cdf 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -1883,20 +1883,14 @@ static int lo_open(struct block_device *bdev, fmode_t mode)
 	int err;
 
 	/*
-	 * take loop_ctl_mutex to protect lo pointer from race with
-	 * loop_control_ioctl(LOOP_CTL_REMOVE), however, to reduce contention
-	 * release it prior to updating lo->lo_refcnt.
+	 * both ->private_data and ->lo_refcnt are covered by disk's
+	 * open_mutex, so race between open and remove can be avoided
 	 */
-	err = mutex_lock_killable(&loop_ctl_mutex);
-	if (err)
-		return err;
 	lo = bdev->bd_disk->private_data;
-	if (!lo) {
-		mutex_unlock(&loop_ctl_mutex);
+	if (!lo)
 		return -ENXIO;
-	}
+
 	err = mutex_lock_killable(&lo->lo_mutex);
-	mutex_unlock(&loop_ctl_mutex);
 	if (err)
 		return err;
 	atomic_inc(&lo->lo_refcnt);
@@ -2272,21 +2266,26 @@ static long loop_control_ioctl(struct file *file, unsigned int cmd,
 		ret = loop_lookup(&lo, parm);
 		if (ret < 0)
 			break;
-		ret = mutex_lock_killable(&lo->lo_mutex);
+		/* cover removing vs. opening loop device */
+		ret = mutex_lock_killable(&lo->lo_disk->part0->bd_mutex);
 		if (ret)
 			break;
-		if (lo->lo_state != Lo_unbound) {
-			ret = -EBUSY;
-			mutex_unlock(&lo->lo_mutex);
+		ret = mutex_lock_killable(&lo->lo_mutex);
+		if (ret) {
+			mutex_unlock(&lo->lo_disk->part0->bd_mutex);
 			break;
 		}
-		if (atomic_read(&lo->lo_refcnt) > 0) {
+		if (lo->lo_state != Lo_unbound ||
+				atomic_read(&lo->lo_refcnt) > 0) {
 			ret = -EBUSY;
 			mutex_unlock(&lo->lo_mutex);
+			mutex_unlock(&lo->lo_disk->part0->bd_mutex);
 			break;
 		}
-		lo->lo_disk->private_data = NULL;
 		mutex_unlock(&lo->lo_mutex);
+		lo->lo_disk->private_data = NULL;
+		mutex_unlock(&lo->lo_disk->part0->bd_mutex);
+
 		idr_remove(&loop_index_idr, lo->lo_number);
 		loop_remove(lo);
 		break;
-- 
2.29.2


^ permalink raw reply	[flat|nested] 7+ messages in thread

end of thread, other threads:[~2021-06-11 17:51 UTC | newest]

Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-06-05 14:09 [PATCH] block: loop: fix deadlock between open and remove Christoph Hellwig
2021-06-05 15:00 ` Colin Ian King
2021-06-07  0:11 ` Ming Lei
2021-06-11 17:51 ` Jens Axboe
  -- strict thread matches above, loose matches on Subject: below --
2021-06-04  0:04 Ming Lei
2021-06-04  6:57 ` Christoph Hellwig
2021-06-04  8:56   ` Ming Lei

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).