Skip to contentSkip to conversation

btrfs: destroy qgroup extent records on transaction abort

PATCH In review Qu Wenruo 3 participants 3 messages Feb 11, 2020 – Feb 14, 2020 Btrfs file system

lore
Share & tools
lore

Qu Wenruo

author
#0 lore raw

PATCH v2 1/1 btrfs: destroy qgroup extent records on transaction abort

From: Jeff Mahoney <[email protected]>

We clean up the delayed references when we abort a transaction but we leave the pending qgroup extent records behind, leaking memory.

This patch destroyes the extent records when we destroy the delayed refs and checks to ensure they're gone before releasing the transaction.

Fixes 3368d001ba5df (btrfs: qgroup: Record possible quota-related extent for qgroup.)

[Rebased to latest upstream, remove to_qgroup() helper, use
rbtree_postorder_for_each_entry_safe() wrapper]

---
Changelog:
v2:
- Update the commit message to remove to_qgroup()

---
 fs/btrfs/disk-io.c     |  1 +
 fs/btrfs/qgroup.c      | 13 +++++++++++++
 fs/btrfs/qgroup.h      |  1 +
 fs/btrfs/transaction.c |  2 ++
 4 files changed, 17 insertions(+)
fs/btrfs/disk-io.c +1 -0
index 7fa9bb79ad08..e8154e2747da 100644
@@ -4275,6 +4275,7 @@ static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
 		cond_resched();
 		spin_lock(&delayed_refs->lock);
 	}
+	btrfs_qgroup_destroy_extent_records(trans);
 
 	spin_unlock(&delayed_refs->lock);
 
fs/btrfs/qgroup.c +13 -0
index 98d9a50352d6..184e3f66ee94 100644
@@ -4002,3 +4002,16 @@ int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
 	}
 	return ret;
 }
+
+void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
+{
+	struct btrfs_delayed_ref_root *delayed_refs = &trans->delayed_refs;
+	struct btrfs_qgroup_extent_record *entry;
+	struct btrfs_qgroup_extent_record *next;
+	struct rb_root *root = &delayed_refs->dirty_extent_root;
+
+	rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
+		ulist_free(entry->old_roots);
+		kfree(entry);
+	}
+}
fs/btrfs/qgroup.h +1 -0
index 236f12224d52..1bc654459469 100644
@@ -414,5 +414,6 @@ int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
 		u64 last_snapshot);
 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
 		struct btrfs_root *root, struct extent_buffer *eb);
+void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans);
 
 #endif
fs/btrfs/transaction.c +2 -0
index 33dcc88b428a..beb6c69cd1e5 100644
@@ -121,6 +121,8 @@ void btrfs_put_transaction(struct btrfs_transaction *transaction)
 		BUG_ON(!list_empty(&transaction->list));
 		WARN_ON(!RB_EMPTY_ROOT(
 				&transaction->delayed_refs.href_root.rb_root));
+		WARN_ON(!RB_EMPTY_ROOT(
+				&transaction->delayed_refs.dirty_extent_root));
 		if (transaction->delayed_refs.pending_csums)
 			btrfs_err(transaction->fs_info,
 				  "pending csums is %llu",
Signature · 1 line
2.25.0
Reviewed-by on v2 1/1
#1 lore raw
Quoted 5 lines · Qu Wenruo

From: Jeff Mahoney <[email protected]>

We clean up the delayed references when we abort a transaction but we leave the pending qgroup extent records behind, leaking memory.

Quoted 8 lines

This patch destroyes the extent records when we destroy the delayed refs and checks to ensure they're gone before releasing the transaction.

Fixes 3368d001ba5df (btrfs: qgroup: Record possible quota-related extent for qgroup.)

[Rebased to latest upstream, remove to_qgroup() helper, use
rbtree_postorder_for_each_entry_safe() wrapper]

Thanks,

Josef

#2 lore raw
In reply to Qu Wenruo
Quoted 10 lines · Qu Wenruo

From: Jeff Mahoney <[email protected]>

We clean up the delayed references when we abort a transaction but we leave the pending qgroup extent records behind, leaking memory.

This patch destroyes the extent records when we destroy the delayed refs and checks to ensure they're gone before releasing the transaction.

Fixes 3368d001ba5df (btrfs: qgroup: Record possible quota-related extent for qgroup.)
Quoted 7 lines

[Rebased to latest upstream, remove to_qgroup() helper, use
rbtree_postorder_for_each_entry_safe() wrapper]

---
Changelog:
v2:
- Update the commit message to remove to_qgroup()

Added to misc-next, thanks.

fs/btrfs/qgroup.c int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans, · line 4011

Quoted 6 lines
fs/btrfs/qgroup.c +6 -0
+void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
+{
+	struct btrfs_delayed_ref_root *delayed_refs = &trans->delayed_refs;
+	struct btrfs_qgroup_extent_record *entry;
+	struct btrfs_qgroup_extent_record *next;
+	struct rb_root *root = &delayed_refs->dirty_extent_root;

I've removed the delayed_refs varaible indirection:

	root = &trans->delayed_refs.dirty_extent_root;
Quoted 5 lines
fs/btrfs/qgroup.c +5 -0
+	rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
+		ulist_free(entry->old_roots);
+		kfree(entry);
+	}
+}