md/raid5: fix a potential deadlock of raid5 reshape

There is a potential deadlock if mount/umount happens when raid5_finish_reshape() tries to grow the size of emulated disk.

How the deadlock happens?
1) The raid5 resync thread finished reshape (expanding array).
2) The mount or umount thread holds VFS sb->s_umount lock and tries to write through critical data into raid5 emulated block device. So it waits for raid5 kernel thread handling stripes in order to finish it I/Os.
3) In the routine of raid5 kernel thread, md_check_recovery() will be called first in order to reap the raid5 resync thread. That is, raid5_finish_reshape() will be called. In this function, it will try to update conf and call VFS revalidate_disk() to grow the raid5 emulated block device. It will try to acquire VFS sb->s_umount lock. The raid5 kernel thread cannot continue, so no one can handle mount/ umount I/Os (stripes). Once the write-through I/Os cannot be finished, mount/umount will not release sb->s_umount lock. The deadlock happens.

The raid5 kernel thread is an emulated block device. It is responible to handle I/Os (stripes) from upper layers. The emulated block device should not request any I/Os on itself. That is, it should not call VFS layer functions. (If it did, it will try to acquire VFS locks to guarantee the I/Os sequence.) So we have the resync thread to send resync I/O requests and to wait for the results.

For solving this potential deadlock, we can put the size growth of the emulated block device as the final step of reshape thread.

Reported-by Alex Wu <[email protected]>

Reviewed-by Alex Wu <[email protected]>

Reviewed-by Chung-Chiang Cheng <[email protected]>

Signed-off-by BingJing Chang <[email protected]>

---
 drivers/md/md.c    | 13 +++++++++++++
 drivers/md/raid5.c |  8 +-------

 2 files changed, 14 insertions(+), 7 deletions(-)

drivers/md/md.c +13 -0

index 4e4dee0..77d6d85 100644
@@ -8527,6 +8527,19 @@ void md_do_sync(struct md_thread *thread)
       * raid */
      set_mask_bits(&mddev->sb_flags, 0,
                BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
+
+    if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
+            !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
+            mddev->delta_disks > 0 &&
+            mddev->pers->finish_reshape &&
+            mddev->pers->size &&
+            mddev->queue) {
+        mddev_lock_nointr(mddev);
+        md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
+        mddev_unlock(mddev);
+        set_capacity(mddev->gendisk, mddev->array_sectors);
+        revalidate_disk(mddev->gendisk);
+    }
 
      spin_lock(&mddev->lock);
      if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {

drivers/md/raid5.c +1 -7

index 98ce427..0f68faa 100644
@@ -8001,13 +8001,7 @@ static void raid5_finish_reshape(struct mddev 
 *mddev)
 
      if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
 
-        if (mddev->delta_disks > 0) {
-            md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
-            if (mddev->queue) {
-                set_capacity(mddev->gendisk, mddev->array_sectors);
-                revalidate_disk(mddev->gendisk);
-            }
-        } else {
+        if (mddev->delta_disks <= 0) {
              int d;
              spin_lock_irq(&conf->device_lock);

             mddev->degraded = raid5_calc_degraded(conf);

Quoted 26 lines

There is a potential deadlock if mount/umount happens when raid5_finish_reshape() tries to grow the size of emulated disk.

How the deadlock happens?
1) The raid5 resync thread finished reshape (expanding array).
2) The mount or umount thread holds VFS sb->s_umount lock and tries to write through critical data into raid5 emulated block device. So it waits for raid5 kernel thread handling stripes in order to finish it I/Os.
3) In the routine of raid5 kernel thread, md_check_recovery() will be called first in order to reap the raid5 resync thread. That is, raid5_finish_reshape() will be called. In this function, it will try to update conf and call VFS revalidate_disk() to grow the raid5 emulated block device. It will try to acquire VFS sb->s_umount lock. The raid5 kernel thread cannot continue, so no one can handle mount/ umount I/Os (stripes). Once the write-through I/Os cannot be finished, mount/umount will not release sb->s_umount lock. The deadlock happens.

The raid5 kernel thread is an emulated block device. It is responible to handle I/Os (stripes) from upper layers. The emulated block device should not request any I/Os on itself. That is, it should not call VFS layer functions. (If it did, it will try to acquire VFS locks to guarantee the I/Os sequence.) So we have the resync thread to send resync I/O requests and to wait for the results.

Quoted 2 lines

For solving this potential deadlock, we can put the size growth of the emulated block device as the final step of reshape thread.

Maybe another way to avoid the deadlock is to ensure the reshaping array can't be mount at all. One reason is, if the reshaping is finished, but fs still don't know about  the new array size since metadata is not refreshed. Just my $0.02.

Quoted 9 lines

Reported-by Alex Wu <[email protected]>

Reviewed-by Alex Wu <[email protected]>

Reviewed-by Chung-Chiang Cheng <[email protected]>

Signed-off-by BingJing Chang <[email protected]>

---
 drivers/md/md.c    | 13 +++++++++++++
 drivers/md/raid5.c |  8 +-------

 2 files changed, 14 insertions(+), 7 deletions(-)

I think raid10 has the similar issue, so at least you need to change raid10 either.

Thanks,
Guoqing

Quoted 45 lines

drivers/md/md.c +13 -0

index 4e4dee0..77d6d85 100644
@@ -8527,6 +8527,19 @@ void md_do_sync(struct md_thread *thread)
       * raid */
      set_mask_bits(&mddev->sb_flags, 0,
                BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
+
+    if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
+            !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
+            mddev->delta_disks > 0 &&
+            mddev->pers->finish_reshape &&
+            mddev->pers->size &&
+            mddev->queue) {
+        mddev_lock_nointr(mddev);
+        md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
+        mddev_unlock(mddev);
+        set_capacity(mddev->gendisk, mddev->array_sectors);
+        revalidate_disk(mddev->gendisk);
+    }
 
      spin_lock(&mddev->lock);
      if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {

drivers/md/raid5.c +1 -7

index 98ce427..0f68faa 100644
@@ -8001,13 +8001,7 @@ static void raid5_finish_reshape(struct mddev 
 *mddev)
 
      if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
 
-        if (mddev->delta_disks > 0) {
-            md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
-            if (mddev->queue) {
-                set_capacity(mddev->gendisk, mddev->array_sectors);
-                revalidate_disk(mddev->gendisk);
-            }
-        } else {
+        if (mddev->delta_disks <= 0) {
              int d;
              spin_lock_irq(&conf->device_lock);

             mddev->degraded = raid5_calc_degraded(conf);

Guoqing Jiang 於 2017-12-28 15:59 寫到:

Quoted 27 lines · bingjingc

Quoted 26 lines

There is a potential deadlock if mount/umount happens when raid5_finish_reshape() tries to grow the size of emulated disk.

How the deadlock happens?
1) The raid5 resync thread finished reshape (expanding array).
2) The mount or umount thread holds VFS sb->s_umount lock and tries to write through critical data into raid5 emulated block device. So it waits for raid5 kernel thread handling stripes in order to finish it I/Os.
3) In the routine of raid5 kernel thread, md_check_recovery() will be called first in order to reap the raid5 resync thread. That is, raid5_finish_reshape() will be called. In this function, it will try to update conf and call VFS revalidate_disk() to grow the raid5 emulated block device. It will try to acquire VFS sb->s_umount lock. The raid5 kernel thread cannot continue, so no one can handle mount/ umount I/Os (stripes). Once the write-through I/Os cannot be finished, mount/umount will not release sb->s_umount lock. The deadlock happens.

The raid5 kernel thread is an emulated block device. It is responible to handle I/Os (stripes) from upper layers. The emulated block device should not request any I/Os on itself. That is, it should not call VFS layer functions. (If it did, it will try to acquire VFS locks to guarantee the I/Os sequence.) So we have the resync thread to send resync I/O requests and to wait for the results.

Quoted 8 lines

Quoted 2 lines

For solving this potential deadlock, we can put the size growth of the emulated block device as the final step of reshape thread.

Maybe another way to avoid the deadlock is to ensure the reshaping array can't be mount at all. One reason is, if the reshaping is finished, but fs still don't know about  the new array size since metadata is not refreshed. Just my $0.02.

It may take hours or couples of days to expand an array. So I bet that it's a feature that an expanding array can still serve fs requests. Once it finished, the new array size for metadata can be passed to fs layer via revalidate_disk(). From then, fs is able to be expanded, too. There are no needs to restart that array. Based on this point, I don't want to break this feature.

Quoted 38 lines

Guoqing Jiang 於 2017-12-28 15:59 寫到:

Quoted 27 lines · bingjingc

Quoted 26 lines

There is a potential deadlock if mount/umount happens when raid5_finish_reshape() tries to grow the size of emulated disk.

How the deadlock happens?
1) The raid5 resync thread finished reshape (expanding array).
2) The mount or umount thread holds VFS sb->s_umount lock and tries to write through critical data into raid5 emulated block device. So it waits for raid5 kernel thread handling stripes in order to finish it I/Os.
3) In the routine of raid5 kernel thread, md_check_recovery() will be called first in order to reap the raid5 resync thread. That is, raid5_finish_reshape() will be called. In this function, it will try to update conf and call VFS revalidate_disk() to grow the raid5 emulated block device. It will try to acquire VFS sb->s_umount lock. The raid5 kernel thread cannot continue, so no one can handle mount/ umount I/Os (stripes). Once the write-through I/Os cannot be finished, mount/umount will not release sb->s_umount lock. The deadlock happens.

The raid5 kernel thread is an emulated block device. It is responible to handle I/Os (stripes) from upper layers. The emulated block device should not request any I/Os on itself. That is, it should not call VFS layer functions. (If it did, it will try to acquire VFS locks to guarantee the I/Os sequence.) So we have the resync thread to send resync I/O requests and to wait for the results.

Quoted 8 lines

Quoted 2 lines

For solving this potential deadlock, we can put the size growth of the emulated block device as the final step of reshape thread.

Maybe another way to avoid the deadlock is to ensure the reshaping array can't be mount at all. One reason is, if the reshaping is finished, but fs still don't know about  the new array size since metadata is not refreshed. Just my $0.02.

Quoted 5 lines

It may take hours or couples of days to expand an array. So I bet that it's a feature that an expanding array can still serve fs requests. Once it finished, the new array size for metadata can be passed to fs layer via revalidate_disk(). From then, fs is able to be expanded, too. There are no needs to restart that array. Based on this point, I don't want to break this feature.

Sorry, the metadata which I mentioned before is the metadata for file system. How fs know about the new size without update fs's metadata? Pls see the wiki:
https://raid.wiki.kernel.org/index.php/Growing#Extending_the_filesystem

Thanks,
Guoqing

Subject: [PATCH] md: fix a potential deadlock of raid5/raid10 reshape

There is a potential deadlock if mount/umount happens when raid5_finish_reshape() tries to grow the size of emulated disk.

How the deadlock happens?
1) The raid5 resync thread finished reshape (expanding array).
2) The mount or umount thread holds VFS sb->s_umount lock and tries to write through critical data into raid5 emulated block device. So it waits for raid5 kernel thread handling stripes in order to finish it I/Os.
3) In the routine of raid5 kernel thread, md_check_recovery() will be called first in order to reap the raid5 resync thread. That is, raid5_finish_reshape() will be called. In this function, it will try to update conf and call VFS revalidate_disk() to grow the raid5 emulated block device. It will try to acquire VFS sb->s_umount lock.
The raid5 kernel thread cannot continue, so no one can handle mount/ umount I/Os (stripes). Once the write-through I/Os cannot be finished, mount/umount will not release sb->s_umount lock. The deadlock happens.

The raid5 kernel thread is an emulated block device. It is responible to handle I/Os (stripes) from upper layers. The emulated block device should not request any I/Os on itself. That is, it should not call VFS layer functions. (If it did, it will try to acquire VFS locks to guarantee the I/Os sequence.) So we have the resync thread to send resync I/O requests and to wait for the results.

For solving this potential deadlock, we can put the size growth of the emulated block device as the final step of reshape thread.

2017/12/29:
Thanks to Guoqing Jiang <[email protected]>,
we confirmed that there is the same deadlock issue in raid10. It's reproducible and can be fixed by this patch. For raid10.c, we can remove the similar code to prevent deadlock as well since they has been called before.

Reported-by Alex Wu <[email protected]>

Reviewed-by Alex Wu <[email protected]>

Reviewed-by Chung-Chiang Cheng <[email protected]>

Signed-off-by BingJing Chang <[email protected]>

---
 drivers/md/md.c     | 13 +++++++++++++
 drivers/md/raid10.c |  8 +-------
 drivers/md/raid5.c  |  8 +-------
 3 files changed, 15 insertions(+), 14 deletions(-)

drivers/md/md.c +13 -0

index 4e4dee0..52092de 100644
@@ -8528,6 +8528,19 @@ void md_do_sync(struct md_thread *thread)
 	set_mask_bits(&mddev->sb_flags, 0,
 		      BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
 
+	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
+			!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
+			mddev->delta_disks > 0 &&
+			mddev->pers->finish_reshape &&
+			mddev->pers->size &&
+			mddev->queue) {
+		mddev_lock_nointr(mddev);
+		md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
+		mddev_unlock(mddev);
+		set_capacity(mddev->gendisk, mddev->array_sectors);
+		revalidate_disk(mddev->gendisk);
+	}
+
 	spin_lock(&mddev->lock);
 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
 		/* We completed so min/max setting can be forgotten if used. */

drivers/md/raid10.c +1 -7

index c131835..8a46473 100644
@@ -4818,17 +4818,11 @@ static void raid10_finish_reshape(struct mddev
 *mddev)
 		return;
 
 	if (mddev->delta_disks > 0) {
-		sector_t size = raid10_size(mddev, 0, 0);
-		md_set_array_sectors(mddev, size);
 		if (mddev->recovery_cp > mddev->resync_max_sectors) {
 			mddev->recovery_cp = mddev->resync_max_sectors;
 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 		}
-		mddev->resync_max_sectors = size;
-		if (mddev->queue) {
-			set_capacity(mddev->gendisk, mddev->array_sectors);
-			revalidate_disk(mddev->gendisk);
-		}
+		mddev->resync_max_sectors = mddev->array_sectors;
 	} else {
 		int d;

drivers/md/raid10.c +0 -0

 		rcu_read_lock();

drivers/md/raid5.c +1 -7

index 98ce427..0f68faa 100644
@@ -8001,13 +8001,7 @@ static void raid5_finish_reshape(struct mddev *mddev)
 
 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
 
-		if (mddev->delta_disks > 0) {
-			md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
-			if (mddev->queue) {
-				set_capacity(mddev->gendisk, mddev->array_sectors);
-				revalidate_disk(mddev->gendisk);
-			}
-		} else {
+		if (mddev->delta_disks <= 0) {
 			int d;
 			spin_lock_irq(&conf->device_lock);
 			mddev->degraded = raid5_calc_degraded(conf);

Signature · 5 lines

2.7.4



Guoqing Jiang 於 2017-12-29 14:42 寫到:

Quoted 63 lines · bingjingc

Quoted 52 lines

Guoqing Jiang 於 2017-12-28 15:59 寫到:

Quoted 41 lines · bingjingc

Quoted 36 lines

There is a potential deadlock if mount/umount happens when raid5_finish_reshape() tries to grow the size of emulated disk.

How the deadlock happens?
1) The raid5 resync thread finished reshape (expanding array).
2) The mount or umount thread holds VFS sb->s_umount lock and tries to
   write through critical data into raid5 emulated block device. So it
   waits for raid5 kernel thread handling stripes in order to finish it
   I/Os.
3) In the routine of raid5 kernel thread, md_check_recovery() will be
   called first in order to reap the raid5 resync thread. That is, raid5_finish_reshape() will be called. In this function, it will try
   to update conf and call VFS revalidate_disk() to grow the raid5 emulated block device. It will try to acquire VFS sb->s_umount lock.
The raid5 kernel thread cannot continue, so no one can handle mount/ umount I/Os (stripes). Once the write-through I/Os cannot be finished,
mount/umount will not release sb->s_umount lock. The deadlock happens.

The raid5 kernel thread is an emulated block device. It is responible to
handle I/Os (stripes) from upper layers. The emulated block device should not request any I/Os on itself. That is, it should not call VFS
layer functions. (If it did, it will try to acquire VFS locks to guarantee the I/Os sequence.) So we have the resync thread to send resync I/O requests and to wait for the results.

Quoted 3 lines

For solving this potential deadlock, we can put the size growth of the
emulated block device as the final step of reshape thread.

Quoted 8 lines

Maybe another way to avoid the deadlock is to ensure the reshaping array
can't be mount at all. One reason is, if the reshaping is finished, but fs still
don't know about  the new array size since metadata is not refreshed. Just
my $0.02.

Quoted 9 lines

It may take hours or couples of days to expand an array. So I bet that it's a
feature that an expanding array can still serve fs requests. Once it finished,
the new array size for metadata can be passed to fs layer via revalidate_disk().
From then, fs is able to be expanded, too. There are no needs to restart that
array. Based on this point, I don't want to break this feature.

Quoted 5 lines

Sorry, the metadata which I mentioned before is the metadata for file system.
How fs know about the new size without update fs's metadata? Pls see the wiki:
https://raid.wiki.kernel.org/index.php/Growing#Extending_the_filesystem

Sorry, you're right. It requires few more commands to grow the filesystem's size. I used wrong wordings. In fact, what I want to mention is the metadata of the emulated disk. In fs/block_dev.c, revalidate_disk() will call check_disk_size_change(disk, bdev) to adjust the monitored block device size. I guess that these structures belong to VFS. So for MD, it keeps the same behavior to issue the upper layer about the size change.

Thanks,
BingJing

Quoted 3 lines

Thanks,
Guoqing