Skip to contentSkip to conversation

ocfs2: fix dead lock caused by ocfs2_defrag_extent

PATCH Superseded Larry Chen 4 participants 5 messages Sep 2, 2018 – Nov 1, 2018 OCFS2

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Larry Chen

author
#0 lore raw

PATCH v2 1/1 ocfs2: fix dead lock caused by ocfs2_defrag_extent

ocfs2_defrag_extent may fall into dead lock.

ocfs2_ioctl_move_extents
ocfs2_ioctl_move_extents
ocfs2_move_extents
ocfs2_defrag_extent
ocfs2_lock_allocators_move_extents

ocfs2_reserve_clusters
inode_lock GLOBAL_BITMAP_SYSTEM_INODE

__ocfs2_flush_truncate_log
inode_lock GLOBAL_BITMAP_SYSTEM_INODE

As back trace shows above, ocfs2_reserve_clusters will call inode_lock against the global bitmap if local allocator has not sufficient cluters. Once global bitmap could meet the demand, ocfs2_reserve_cluster will return success with global bitmap locked.

After ocfs2_reserve_cluster, if truncate log is full, __ocfs2_flush_truncate_log will definitely fall into dead lock because it needs to inode_lock global bitmap, which has already been locked.

To fix this bug, we could remove from ocfs2_lock_allocators_move_extents the code which intends to lock global allocator, and put the removed code after __ocfs2_flush_truncate_log

The ocfs2_lock_allocators_move_extents has been refered by 2 places, one is here, the other does not need the data allocator context, which means this patch does not affect the caller so far.

Change log:
1. rename ocfs2_lock_allocators_move_extents to
ocfs2_lock_meta_allocator_move_extents

2. add some comments

Signed-off-by Larry Chen <[email protected]>
---
 fs/ocfs2/move_extents.c | 40 +++++++++++++++++++++++++---------------
 1 file changed, 25 insertions(+), 15 deletions(-)
fs/ocfs2/move_extents.c +25 -15
index 7eb3b0a6347e..d85dc8a02bd6 100644
@@ -162,7 +162,7 @@ static int __ocfs2_move_extent(handle_t *handle,
  * in some cases, we don't need to reserve clusters, just let data_ac
  * be NULL.
  */
-static int ocfs2_lock_allocators_move_extents(struct inode *inode,
+static int ocfs2_lock_meta_allocator_move_extents(struct inode *inode,
 					struct ocfs2_extent_tree *et,
 					u32 clusters_to_move,
 					u32 extents_to_split,

@@ -192,13 +192,6 @@ static int ocfs2_lock_allocators_move_extents(struct inode *inode,
 		goto out;
 	}
 
-	if (data_ac) {
-		ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac);
-		if (ret) {
-			mlog_errno(ret);
-			goto out;
-		}
-	}
 
 	*credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el);
 

@@ -257,10 +250,11 @@ static int ocfs2_defrag_extent(struct ocfs2_move_extents_context *context,
 		}
 	}
 
-	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, *len, 1,
-						 &context->meta_ac,
-						 &context->data_ac,
-						 extra_blocks, &credits);
+	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
+						*len, 1,
+						&context->meta_ac,
+						&context->data_ac,
+						extra_blocks, &credits);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;

@@ -283,6 +277,21 @@ static int ocfs2_defrag_extent(struct ocfs2_move_extents_context *context,
 		}
 	}
 
+	/*
+	 * Make sure ocfs2_reserve_cluster is called after
+	 * __ocfs2_flush_truncate_log, otherwise, dead lock may happen.
+	 *
+	 * If ocfs2_reserve_cluster is called
+	 * before __ocfs2_flush_truncate_log, dead lock on global bitmap
+	 * may happen.
+	 *
+	 */
+	ret = ocfs2_reserve_clusters(osb, *len, &context->data_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_unlock_mutex;
+	}
+
 	handle = ocfs2_start_trans(osb, credits);
 	if (IS_ERR(handle)) {
 		ret = PTR_ERR(handle);

@@ -600,9 +609,10 @@ static int ocfs2_move_extent(struct ocfs2_move_extents_context *context,
 		}
 	}
 
-	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1,
-						 &context->meta_ac,
-						 NULL, extra_blocks, &credits);
+	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
+						len, 1,
+						&context->meta_ac,
+						NULL, extra_blocks, &credits);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;
Signature · 1 line
2.13.7
#1 lore raw

Folks, could we please review this patch for upstream inclusion?

Thanks.

From: Larry Chen <[email protected]>
Subject: ocfs2: fix deadlock caused by ocfs2_defrag_extent

ocfs2_defrag_extent may fall into deadlock.

ocfs2_ioctl_move_extents
ocfs2_ioctl_move_extents
ocfs2_move_extents
ocfs2_defrag_extent
ocfs2_lock_allocators_move_extents

ocfs2_reserve_clusters
inode_lock GLOBAL_BITMAP_SYSTEM_INODE

__ocfs2_flush_truncate_log
inode_lock GLOBAL_BITMAP_SYSTEM_INODE

As backtrace shows above, ocfs2_reserve_clusters() will call inode_lock against the global bitmap if local allocator has not sufficient cluters. Once global bitmap could meet the demand, ocfs2_reserve_cluster will return success with global bitmap locked.

After ocfs2_reserve_cluster(), if truncate log is full, __ocfs2_flush_truncate_log() will definitely fall into deadlock because it needs to inode_lock global bitmap, which has already been locked.

To fix this bug, we could remove from ocfs2_lock_allocators_move_extents() the code which intends to lock global allocator, and put the removed code after __ocfs2_flush_truncate_log().

ocfs2_lock_allocators_move_extents() is refered by 2 places, one is here, the other does not need the data allocator context, which means this patch does not affect the caller so far.

[lchen at suse.com: rename ocfs2_lock_allocators_move_extents() to ocfs2_lock_meta_allocator_move_extents(), add some comments]
Link https://urldefense.proofpoint.com/v2/url?u=http-3A__lkml.kernel.org_r_20180902091455.23862-2D1-2Dlchen-40suse.com&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=C7gAd4uDxlAvTdc0vmU6X8CMk6L2iDY8-HD0qT6Fo7Y&m=K8aRgOiMescamJW-IuHq-cW4mWedGQv2JTT-2KPy1RI&s=3fQHFtwgixirPeQ5Px4fRBnd_UcSx3rvSsLKLidwESo&e=
Link https://urldefense.proofpoint.com/v2/url?u=http-3A__lkml.kernel.org_r_20180827080121.31145-2D1-2Dlchen-40suse.com&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=C7gAd4uDxlAvTdc0vmU6X8CMk6L2iDY8-HD0qT6Fo7Y&m=K8aRgOiMescamJW-IuHq-cW4mWedGQv2JTT-2KPy1RI&s=PQOJREyeElsg_1IlINdSGefNqA6w_cUrI1ezdVxyi7o&e=
Signed-off-by Larry Chen <[email protected]>
Cc Mark Fasheh <[email protected]>
Cc Joel Becker <[email protected]>
Cc Junxiao Bi <[email protected]>
Cc Joseph Qi <[email protected]>
Cc Changwei Ge <[email protected]>
Signed-off-by Andrew Morton <[email protected]>

---

 fs/ocfs2/move_extents.c |   40 +++++++++++++++++++++++---------------
 1 file changed, 25 insertions(+), 15 deletions(-)
fs/ocfs2/move_extents.c renamed from fs/ocfs2/move_extents.c~fix-dead-lock-caused-by-ocfs2_defrag_extent +25 -15
@@ -162,7 +162,7 @@ out:
  * in some cases, we don't need to reserve clusters, just let data_ac
  * be NULL.
  */
-static int ocfs2_lock_allocators_move_extents(struct inode *inode,
+static int ocfs2_lock_meta_allocator_move_extents(struct inode *inode,
 					struct ocfs2_extent_tree *et,
 					u32 clusters_to_move,
 					u32 extents_to_split,

@@ -192,13 +192,6 @@ static int ocfs2_lock_allocators_move_ex
 		goto out;
 	}
 
-	if (data_ac) {
-		ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac);
-		if (ret) {
-			mlog_errno(ret);
-			goto out;
-		}
-	}
 
 	*credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el);
 

@@ -257,10 +250,11 @@ static int ocfs2_defrag_extent(struct oc
 		}
 	}
 
-	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, *len, 1,
-						 &context->meta_ac,
-						 &context->data_ac,
-						 extra_blocks, &credits);
+	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
+						*len, 1,
+						&context->meta_ac,
+						&context->data_ac,
+						extra_blocks, &credits);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;

@@ -283,6 +277,21 @@ static int ocfs2_defrag_extent(struct oc
 		}
 	}
 
+	/*
+	 * Make sure ocfs2_reserve_cluster is called after
+	 * __ocfs2_flush_truncate_log, otherwise, dead lock may happen.
+	 *
+	 * If ocfs2_reserve_cluster is called
+	 * before __ocfs2_flush_truncate_log, dead lock on global bitmap
+	 * may happen.
+	 *
+	 */
+	ret = ocfs2_reserve_clusters(osb, *len, &context->data_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_unlock_mutex;
+	}
+
 	handle = ocfs2_start_trans(osb, credits);
 	if (IS_ERR(handle)) {
 		ret = PTR_ERR(handle);

@@ -600,9 +609,10 @@ static int ocfs2_move_extent(struct ocfs
 		}
 	}
 
-	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1,
-						 &context->meta_ac,
-						 NULL, extra_blocks, &credits);
+	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
+						len, 1,
+						&context->meta_ac,
+						NULL, extra_blocks, &credits);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;

_

#2 lore raw

Hi Larry,

I still have a few tiny comments for your patch, please refer to them.

Quoted 58 lines · Andrew Morton

Folks, could we please review this patch for upstream inclusion?

Thanks.

From: Larry Chen <[email protected]>
Subject: ocfs2: fix deadlock caused by ocfs2_defrag_extent

ocfs2_defrag_extent may fall into deadlock.

ocfs2_ioctl_move_extents
ocfs2_ioctl_move_extents
ocfs2_move_extents
ocfs2_defrag_extent
ocfs2_lock_allocators_move_extents

ocfs2_reserve_clusters
inode_lock GLOBAL_BITMAP_SYSTEM_INODE

__ocfs2_flush_truncate_log
inode_lock GLOBAL_BITMAP_SYSTEM_INODE

As backtrace shows above, ocfs2_reserve_clusters() will call inode_lock against the global bitmap if local allocator has not sufficient cluters. Once global bitmap could meet the demand, ocfs2_reserve_cluster will return success with global bitmap locked.

After ocfs2_reserve_cluster(), if truncate log is full, __ocfs2_flush_truncate_log() will definitely fall into deadlock because it needs to inode_lock global bitmap, which has already been locked.

To fix this bug, we could remove from ocfs2_lock_allocators_move_extents() the code which intends to lock global allocator, and put the removed code after __ocfs2_flush_truncate_log().

ocfs2_lock_allocators_move_extents() is refered by 2 places, one is here, the other does not need the data allocator context, which means this patch does not affect the caller so far.

[lchen at suse.com: rename ocfs2_lock_allocators_move_extents() to ocfs2_lock_meta_allocator_move_extents(), add some comments]
Link https://urldefense.proofpoint.com/v2/url?u=http-3A__lkml.kernel.org_r_20180902091455.23862-2D1-2Dlchen-40suse.com&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=C7gAd4uDxlAvTdc0vmU6X8CMk6L2iDY8-HD0qT6Fo7Y&m=K8aRgOiMescamJW-IuHq-cW4mWedGQv2JTT-2KPy1RI&s=3fQHFtwgixirPeQ5Px4fRBnd_UcSx3rvSsLKLidwESo&e=
Link https://urldefense.proofpoint.com/v2/url?u=http-3A__lkml.kernel.org_r_20180827080121.31145-2D1-2Dlchen-40suse.com&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=C7gAd4uDxlAvTdc0vmU6X8CMk6L2iDY8-HD0qT6Fo7Y&m=K8aRgOiMescamJW-IuHq-cW4mWedGQv2JTT-2KPy1RI&s=PQOJREyeElsg_1IlINdSGefNqA6w_cUrI1ezdVxyi7o&e=
Signed-off-by Larry Chen <[email protected]>
Cc Mark Fasheh <[email protected]>
Cc Joel Becker <[email protected]>
Cc Junxiao Bi <[email protected]>
Cc Joseph Qi <[email protected]>
Cc Changwei Ge <[email protected]>
Signed-off-by Andrew Morton <[email protected]>

---

  fs/ocfs2/move_extents.c |   40 +++++++++++++++++++++++---------------
  1 file changed, 25 insertions(+), 15 deletions(-)

fs/ocfs2/move_extents.c out: · line 168

Quoted 8 lines
fs/ocfs2/move_extents.c renamed from fs/ocfs2/move_extents.c~fix-dead-lock-caused-by-ocfs2_defrag_extent +1 -1
@@ -162,7 +162,7 @@ out:
   * in some cases, we don't need to reserve clusters, just let data_ac
   * be NULL.
   */
-static int ocfs2_lock_allocators_move_extents(struct inode *inode,
+static int ocfs2_lock_meta_allocator_move_extents(struct inode *inode,
  					struct ocfs2_extent_tree *et,
  					u32 clusters_to_move,
  					u32 extents_to_split,

After removing 'reserving clusters logic' out of this function, I suggest to also change the function header comments. And I think there is no need for arguments _data_ac_ now.
Otherwise this patch looks sane to me.

Thanks,
Changwei

Quoted 74 lines
fs/ocfs2/move_extents.c renamed from fs/ocfs2/move_extents.c~fix-dead-lock-caused-by-ocfs2_defrag_extent +24 -14
@@ -192,13 +192,6 @@ static int ocfs2_lock_allocators_move_ex
  		goto out;
  	}
 
-	if (data_ac) {
-		ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac);
-		if (ret) {
-			mlog_errno(ret);
-			goto out;
-		}
-	}
 
  	*credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el);
 

@@ -257,10 +250,11 @@ static int ocfs2_defrag_extent(struct oc
  		}
  	}
 
-	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, *len, 1,
-						 &context->meta_ac,
-						 &context->data_ac,
-						 extra_blocks, &credits);
+	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
+						*len, 1,
+						&context->meta_ac,
+						&context->data_ac,
+						extra_blocks, &credits);
  	if (ret) {
  		mlog_errno(ret);
  		goto out;

@@ -283,6 +277,21 @@ static int ocfs2_defrag_extent(struct oc
  		}
  	}
 
+	/*
+	 * Make sure ocfs2_reserve_cluster is called after
+	 * __ocfs2_flush_truncate_log, otherwise, dead lock may happen.
+	 *
+	 * If ocfs2_reserve_cluster is called
+	 * before __ocfs2_flush_truncate_log, dead lock on global bitmap
+	 * may happen.
+	 *
+	 */
+	ret = ocfs2_reserve_clusters(osb, *len, &context->data_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_unlock_mutex;
+	}
+
  	handle = ocfs2_start_trans(osb, credits);
  	if (IS_ERR(handle)) {
  		ret = PTR_ERR(handle);

@@ -600,9 +609,10 @@ static int ocfs2_move_extent(struct ocfs
  		}
  	}
 
-	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1,
-						 &context->meta_ac,
-						 NULL, extra_blocks, &credits);
+	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
+						len, 1,
+						&context->meta_ac,
+						NULL, extra_blocks, &credits);
  	if (ret) {
  		mlog_errno(ret);
  		goto out;

_

_______________________________________________
Ocfs2-devel mailing list
Ocfs2-devel at oss.oracle.com
https://oss.oracle.com/mailman/listinfo/ocfs2-devel

author
#3 lore raw

Hi Changwei,

Thanks for your review, I'll improve my patch and resend it later.

Thanks
Larry

Quoted 154 lines · Changwei Ge

Hi Larry,

I still have a few tiny comments for your patch, please refer to them.

Quoted 58 lines · Andrew Morton

Folks, could we please review this patch for upstream inclusion?

Thanks.

From: Larry Chen <[email protected]>
Subject: ocfs2: fix deadlock caused by ocfs2_defrag_extent

ocfs2_defrag_extent may fall into deadlock.

ocfs2_ioctl_move_extents
ocfs2_ioctl_move_extents
ocfs2_move_extents
ocfs2_defrag_extent
ocfs2_lock_allocators_move_extents

ocfs2_reserve_clusters
inode_lock GLOBAL_BITMAP_SYSTEM_INODE

__ocfs2_flush_truncate_log
inode_lock GLOBAL_BITMAP_SYSTEM_INODE

As backtrace shows above, ocfs2_reserve_clusters() will call inode_lock against the global bitmap if local allocator has not sufficient cluters. Once global bitmap could meet the demand, ocfs2_reserve_cluster will return success with global bitmap locked.

After ocfs2_reserve_cluster(), if truncate log is full, __ocfs2_flush_truncate_log() will definitely fall into deadlock because it needs to inode_lock global bitmap, which has already been locked.

To fix this bug, we could remove from ocfs2_lock_allocators_move_extents() the code which intends to lock global allocator, and put the removed code after __ocfs2_flush_truncate_log().

ocfs2_lock_allocators_move_extents() is refered by 2 places, one is here, the other does not need the data allocator context, which means this patch does not affect the caller so far.

[lchen at suse.com: rename ocfs2_lock_allocators_move_extents() to ocfs2_lock_meta_allocator_move_extents(), add some comments]

    Link: https://urldefense.proofpoint.com/v2/url?u=http-3A__lkml.kernel.org_r_20180902091455.23862-2D1-2Dlchen-40suse.com&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=C7gAd4uDxlAvTdc0vmU6X8CMk6L2iDY8-HD0qT6Fo7Y&m=K8aRgOiMescamJW-IuHq-cW4mWedGQv2JTT-2KPy1RI&s=3fQHFtwgixirPeQ5Px4fRBnd_UcSx3rvSsLKLidwESo&e=
Link https://urldefense.proofpoint.com/v2/url?u=http-3A__lkml.kernel.org_r_20180827080121.31145-2D1-2Dlchen-40suse.com&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=C7gAd4uDxlAvTdc0vmU6X8CMk6L2iDY8-HD0qT6Fo7Y&m=K8aRgOiMescamJW-IuHq-cW4mWedGQv2JTT-2KPy1RI&s=PQOJREyeElsg_1IlINdSGefNqA6w_cUrI1ezdVxyi7o&e=
Signed-off-by Larry Chen <[email protected]>
Cc Mark Fasheh <[email protected]>
Cc Joel Becker <[email protected]>
Cc Junxiao Bi <[email protected]>
Cc Joseph Qi <[email protected]>
Cc Changwei Ge <[email protected]>
Signed-off-by Andrew Morton <[email protected]>

---

   fs/ocfs2/move_extents.c |   40 +++++++++++++++++++++++---------------
   1 file changed, 25 insertions(+), 15 deletions(-)

fs/ocfs2/move_extents.c out: · line 168

Quoted 8 lines
fs/ocfs2/move_extents.c renamed from fs/ocfs2/move_extents.c~fix-dead-lock-caused-by-ocfs2_defrag_extent +1 -1
@@ -162,7 +162,7 @@ out:
    * in some cases, we don't need to reserve clusters, just let data_ac
    * be NULL.
    */
-static int ocfs2_lock_allocators_move_extents(struct inode *inode,
+static int ocfs2_lock_meta_allocator_move_extents(struct inode *inode,
   					struct ocfs2_extent_tree *et,
   					u32 clusters_to_move,
   					u32 extents_to_split,

After removing 'reserving clusters logic' out of this function, I suggest to also change the function header comments. And I think there is no need for arguments _data_ac_ now.
Otherwise this patch looks sane to me.

Thanks,
Changwei

Quoted 74 lines
fs/ocfs2/move_extents.c renamed from fs/ocfs2/move_extents.c~fix-dead-lock-caused-by-ocfs2_defrag_extent +24 -14
@@ -192,13 +192,6 @@ static int ocfs2_lock_allocators_move_ex
   		goto out;
   	}
 
-	if (data_ac) {
-		ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac);
-		if (ret) {
-			mlog_errno(ret);
-			goto out;
-		}
-	}
 
   	*credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el);
 

@@ -257,10 +250,11 @@ static int ocfs2_defrag_extent(struct oc
   		}
   	}
 
-	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, *len, 1,
-						 &context->meta_ac,
-						 &context->data_ac,
-						 extra_blocks, &credits);
+	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
+						*len, 1,
+						&context->meta_ac,
+						&context->data_ac,
+						extra_blocks, &credits);
   	if (ret) {
   		mlog_errno(ret);
   		goto out;

@@ -283,6 +277,21 @@ static int ocfs2_defrag_extent(struct oc
   		}
   	}
 
+	/*
+	 * Make sure ocfs2_reserve_cluster is called after
+	 * __ocfs2_flush_truncate_log, otherwise, dead lock may happen.
+	 *
+	 * If ocfs2_reserve_cluster is called
+	 * before __ocfs2_flush_truncate_log, dead lock on global bitmap
+	 * may happen.
+	 *
+	 */
+	ret = ocfs2_reserve_clusters(osb, *len, &context->data_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_unlock_mutex;
+	}
+
   	handle = ocfs2_start_trans(osb, credits);
   	if (IS_ERR(handle)) {
   		ret = PTR_ERR(handle);

@@ -600,9 +609,10 @@ static int ocfs2_move_extent(struct ocfs
   		}
   	}
 
-	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1,
-						 &context->meta_ac,
-						 NULL, extra_blocks, &credits);
+	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
+						len, 1,
+						&context->meta_ac,
+						NULL, extra_blocks, &credits);
   	if (ret) {
   		mlog_errno(ret);
   		goto out;

_

_______________________________________________
Ocfs2-devel mailing list
Ocfs2-devel at oss.oracle.com
https://oss.oracle.com/mailman/listinfo/ocfs2-devel

#4 lore raw
In reply to Changwei Ge
Quoted 60 lines · Changwei Ge

Hi Larry,

I still have a few tiny comments for your patch, please refer to them.

Quoted 55 lines · Andrew Morton

Folks, could we please review this patch for upstream inclusion?

Thanks.

From: Larry Chen <[email protected]>
Subject: ocfs2: fix deadlock caused by ocfs2_defrag_extent

ocfs2_defrag_extent may fall into deadlock.

ocfs2_ioctl_move_extents
ocfs2_ioctl_move_extents
ocfs2_move_extents
ocfs2_defrag_extent
ocfs2_lock_allocators_move_extents

ocfs2_reserve_clusters
inode_lock GLOBAL_BITMAP_SYSTEM_INODE

__ocfs2_flush_truncate_log
inode_lock GLOBAL_BITMAP_SYSTEM_INODE

As backtrace shows above, ocfs2_reserve_clusters() will call inode_lock against the global bitmap if local allocator has not sufficient cluters. Once global bitmap could meet the demand, ocfs2_reserve_cluster will return success with global bitmap locked.

After ocfs2_reserve_cluster(), if truncate log is full, __ocfs2_flush_truncate_log() will definitely fall into deadlock because it needs to inode_lock global bitmap, which has already been locked.

To fix this bug, we could remove from ocfs2_lock_allocators_move_extents() the code which intends to lock global allocator, and put the removed code after __ocfs2_flush_truncate_log().

ocfs2_lock_allocators_move_extents() is refered by 2 places, one is here, the other does not need the data allocator context, which means this patch does not affect the caller so far.

[lchen at suse.com: rename ocfs2_lock_allocators_move_extents() to ocfs2_lock_meta_allocator_move_extents(), add some comments]
Link https://urldefense.proofpoint.com/v2/url?u=http-3A__lkml.kernel.org_r_20180902091455.23862-2D1-2Dlchen-40suse.com&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=C7gAd4uDxlAvTdc0vmU6X8CMk6L2iDY8-HD0qT6Fo7Y&m=K8aRgOiMescamJW-IuHq-cW4mWedGQv2JTT-2KPy1RI&s=3fQHFtwgixirPeQ5Px4fRBnd_UcSx3rvSsLKLidwESo&e=
Link https://urldefense.proofpoint.com/v2/url?u=http-3A__lkml.kernel.org_r_20180827080121.31145-2D1-2Dlchen-40suse.com&d=DwICAg&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=C7gAd4uDxlAvTdc0vmU6X8CMk6L2iDY8-HD0qT6Fo7Y&m=K8aRgOiMescamJW-IuHq-cW4mWedGQv2JTT-2KPy1RI&s=PQOJREyeElsg_1IlINdSGefNqA6w_cUrI1ezdVxyi7o&e=
Signed-off-by Larry Chen <[email protected]>
Cc Mark Fasheh <[email protected]>
Cc Joel Becker <[email protected]>
Cc Junxiao Bi <[email protected]>
Cc Joseph Qi <[email protected]>
Cc Changwei Ge <[email protected]>
Signed-off-by Andrew Morton <[email protected]>

---

  fs/ocfs2/move_extents.c |   40 +++++++++++++++++++++++---------------
  1 file changed, 25 insertions(+), 15 deletions(-)
Quoted 16 lines

fs/ocfs2/move_extents.c out: · line 168

Quoted 11 lines
fs/ocfs2/move_extents.c renamed from fs/ocfs2/move_extents.c~fix-dead-lock-caused-by-ocfs2_defrag_extent +1 -1
@@ -162,7 +162,7 @@ out:
   * in some cases, we don't need to reserve clusters, just let data_ac
   * be NULL.
   */
-static int ocfs2_lock_allocators_move_extents(struct inode *inode,
+static int ocfs2_lock_meta_allocator_move_extents(struct inode *inode,
  					struct ocfs2_extent_tree *et,
  					u32 clusters_to_move,
  					u32 extents_to_split,

After removing 'reserving clusters logic' out of this function, I suggest to also change the function header comments. And I think there is no need for arguments _data_ac_ now.
Otherwise this patch looks sane to me.

Agree, I think I have commented this out before.

Thanks,
Joseph

Quoted 83 lines

Thanks,
Changwei

Quoted 69 lines
fs/ocfs2/move_extents.c renamed from fs/ocfs2/move_extents.c~fix-dead-lock-caused-by-ocfs2_defrag_extent +24 -14
@@ -192,13 +192,6 @@ static int ocfs2_lock_allocators_move_ex
  		goto out;
  	}
 
-	if (data_ac) {
-		ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac);
-		if (ret) {
-			mlog_errno(ret);
-			goto out;
-		}
-	}
 
  	*credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el);
 

@@ -257,10 +250,11 @@ static int ocfs2_defrag_extent(struct oc
  		}
  	}
 
-	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, *len, 1,
-						 &context->meta_ac,
-						 &context->data_ac,
-						 extra_blocks, &credits);
+	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
+						*len, 1,
+						&context->meta_ac,
+						&context->data_ac,
+						extra_blocks, &credits);
  	if (ret) {
  		mlog_errno(ret);
  		goto out;

@@ -283,6 +277,21 @@ static int ocfs2_defrag_extent(struct oc
  		}
  	}
 
+	/*
+	 * Make sure ocfs2_reserve_cluster is called after
+	 * __ocfs2_flush_truncate_log, otherwise, dead lock may happen.
+	 *
+	 * If ocfs2_reserve_cluster is called
+	 * before __ocfs2_flush_truncate_log, dead lock on global bitmap
+	 * may happen.
+	 *
+	 */
+	ret = ocfs2_reserve_clusters(osb, *len, &context->data_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_unlock_mutex;
+	}
+
  	handle = ocfs2_start_trans(osb, credits);
  	if (IS_ERR(handle)) {
  		ret = PTR_ERR(handle);

@@ -600,9 +609,10 @@ static int ocfs2_move_extent(struct ocfs
  		}
  	}
 
-	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1,
-						 &context->meta_ac,
-						 NULL, extra_blocks, &credits);
+	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
+						len, 1,
+						&context->meta_ac,
+						NULL, extra_blocks, &credits);
  	if (ret) {
  		mlog_errno(ret);
  		goto out;

_

Quoted 5 lines

_______________________________________________
Ocfs2-devel mailing list
Ocfs2-devel at oss.oracle.com
https://oss.oracle.com/mailman/listinfo/ocfs2-devel

_______________________________________________
Ocfs2-devel mailing list
Ocfs2-devel at oss.oracle.com
https://oss.oracle.com/mailman/listinfo/ocfs2-devel