Skip to contentSkip to conversation

Btrfs: save us a mutex_lock usage when doing quota rescan

PATCH NAK Wang Shilong 2 participants 3 messages May 7, 2013 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Btrfs: save us a mutex_lock usage when doing quota rescan

If qgroup_rescan worker is in progress, we should ignore the extent that has not been dealt with qgroup_rescan worker,just let them dealt later otherwise we may get wrong qgroup accounting.

However, we have checked this before find_all_roots() without spin_lock. When doing qgroup accounting, we don't have to check it again, because during this period,qgroup_rescan worker can deal with more extents and qgroup_rescan_extent->objectid can only go larger, so here the check is unnecessary.

Just remove this check, so that we don't need hold qgroup_rescan_lock when doing qgroup accounting.

---
 fs/btrfs/qgroup.c |    9 ---------
 1 files changed, 0 insertions(+), 9 deletions(-)
fs/btrfs/qgroup.c +0 -9
index d059d86..2710784 100644
@@ -1445,15 +1445,7 @@ int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 	if (ret < 0)
 		return ret;
 
-	mutex_lock(&fs_info->qgroup_rescan_lock);
 	spin_lock(&fs_info->qgroup_lock);
-	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
-		if (fs_info->qgroup_rescan_progress.objectid <= node->bytenr) {
-			ret = 0;
-			goto unlock;
-		}
-	}
-
 	quota_root = fs_info->quota_root;
 	if (!quota_root)
 		goto unlock;
@@ -1492,7 +1484,6 @@ int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 
 unlock:
 	spin_unlock(&fs_info->qgroup_lock);
-	mutex_unlock(&fs_info->qgroup_rescan_lock);
 	ulist_free(roots);
 
 	return ret;
Signature · 1 line
1.7.7.6
#1 lore raw
Quoted 4 lines · Wang Shilong

If qgroup_rescan worker is in progress, we should ignore the extent that has not been dealt with qgroup_rescan worker,just let them dealt later otherwise we may get wrong qgroup accounting.

Quoted 8 lines

However, we have checked this before find_all_roots() without spin_lock. When doing qgroup accounting, we don't have to check it again, because during this period,qgroup_rescan worker can deal with more extents and qgroup_rescan_extent->objectid can only go larger, so here the check is unnecessary.

Just remove this check, so that we don't need hold qgroup_rescan_lock when doing qgroup accounting.

NAK.

After a discussion on that lock the last thing in this thread I see is ...

Quoted 2 lines · Jan Schmidt

Now I see what you mean. The second check is only required when we start a rescan operation after the initial check in btrfs_qgroup_account_ref.

Please continue on that argument, your commit message doesn't explain at all why we should be safe to remove this check.

-Jan

Quoted 34 lines
---
 fs/btrfs/qgroup.c |    9 ---------
 1 files changed, 0 insertions(+), 9 deletions(-)
fs/btrfs/qgroup.c +0 -9
index d059d86..2710784 100644
@@ -1445,15 +1445,7 @@ int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 	if (ret < 0)
 		return ret;
 
-	mutex_lock(&fs_info->qgroup_rescan_lock);
 	spin_lock(&fs_info->qgroup_lock);
-	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
-		if (fs_info->qgroup_rescan_progress.objectid <= node->bytenr) {
-			ret = 0;
-			goto unlock;
-		}
-	}
-
 	quota_root = fs_info->quota_root;
 	if (!quota_root)
 		goto unlock;
@@ -1492,7 +1484,6 @@ int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 
 unlock:
 	spin_unlock(&fs_info->qgroup_lock);
-	mutex_unlock(&fs_info->qgroup_rescan_lock);
 	ulist_free(roots);
 
 	return ret;
#2 lore raw

Hello Jan,

Quoted 16 lines · Wang Shilong
Quoted 4 lines

If qgroup_rescan worker is in progress, we should ignore the extent that has not been dealt with qgroup_rescan worker,just let them dealt later otherwise we may get wrong qgroup accounting.

Quoted 8 lines

However, we have checked this before find_all_roots() without spin_lock. When doing qgroup accounting, we don't have to check it again, because during this period,qgroup_rescan worker can deal with more extents and qgroup_rescan_extent->objectid can only go larger, so here the check is unnecessary.

Just remove this check, so that we don't need hold qgroup_rescan_lock when doing qgroup accounting.

NAK.

Quoted 8 lines

After a discussion on that lock the last thing in this thread I see is ...

Quoted 2 lines · Jan Schmidt

Now I see what you mean. The second check is only required when we start a rescan operation after the initial check in btrfs_qgroup_account_ref.

Please continue on that argument, your commit message doesn't explain at all why we should be safe to remove this check.

You are right!

Thanks,
Wang

Quoted 39 lines

-Jan

Quoted 34 lines
---
 fs/btrfs/qgroup.c |    9 ---------
 1 files changed, 0 insertions(+), 9 deletions(-)
fs/btrfs/qgroup.c +0 -9
index d059d86..2710784 100644
@@ -1445,15 +1445,7 @@ int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 	if (ret < 0)
 		return ret;
 
-	mutex_lock(&fs_info->qgroup_rescan_lock);
 	spin_lock(&fs_info->qgroup_lock);
-	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
-		if (fs_info->qgroup_rescan_progress.objectid <= node->bytenr) {
-			ret = 0;
-			goto unlock;
-		}
-	}
-
 	quota_root = fs_info->quota_root;
 	if (!quota_root)
 		goto unlock;
@@ -1492,7 +1484,6 @@ int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
 
 unlock:
 	spin_unlock(&fs_info->qgroup_lock);
-	mutex_unlock(&fs_info->qgroup_rescan_lock);
 	ulist_free(roots);
 
 	return ret;