btrfs: ensure releasing squota reserve on head refs

A reservation goes through a 3 step lifetime:

- generated during delalloc
- released/counted by ordered_extent allocation
- freed by running delayed ref

That third step depends on must_insert_reserved on the head ref, so the
head ref with that field set owns the reservation. Once you prepare to
run the head ref, must_insert_reserved is unset, which means that
running the ref must free the reservation, whether or not it succeeds,
or else the reservation is leaked. That results in either a risk of
spurious ENOSPC if the fs stays writeable or a warning on unmount if it
is readonly.

The existing squota code was aware of these invariants, but missed a few
cases. Improve it by adding a helper function to use in the cleanup
paths and call it from the existing early returns in running delayed
refs. This also simplifies btrfs_record_squota_delta and struct
btrfs_quota_delta.

This fixes (or at least improves the reliability of) generic/475 with
"mkfs -O squota". On my machine, that test failed ~4/10 times without
this patch and passed 100/100 times with it.

Signed-off-by: Boris Burkov <[e-mail]>
Signed-off-by: David Sterba <[e-mail]>

fs/btrfs/extent-tree.c

···
 	return ret;
 }
 
+static void free_head_ref_squota_rsv(struct btrfs_fs_info *fs_info,
+				     struct btrfs_delayed_ref_head *href)
+{
+	u64 root = href->owning_root;
+
+	/*
+	 * Don't check must_insert_reserved, as this is called from contexts
+	 * where it has already been unset.
+	 */
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE ||
+	    !href->is_data || !is_fstree(root))
+		return;
+
+	btrfs_qgroup_free_refroot(fs_info, root, href->reserved_bytes,
+				  BTRFS_QGROUP_RSV_DATA);
+}
+
 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 				struct btrfs_delayed_ref_head *href,
 				struct btrfs_delayed_ref_node *node,
···
 		struct btrfs_squota_delta delta = {
 			.root = href->owning_root,
 			.num_bytes = node->num_bytes,
-			.rsv_bytes = href->reserved_bytes,
 			.is_data = true,
 			.is_inc	= true,
 			.generation = trans->transid,
···
 						 flags, ref->objectid,
 						 ref->offset, &key,
 						 node->ref_mod, href->owning_root);
+		free_head_ref_squota_rsv(trans->fs_info, href);
 		if (!ret)
 			ret = btrfs_record_squota_delta(trans->fs_info, &delta);
-		else
-			btrfs_qgroup_free_refroot(trans->fs_info, delta.root,
-						  delta.rsv_bytes, BTRFS_QGROUP_RSV_DATA);
 	} else if (node->action == BTRFS_ADD_DELAYED_REF) {
 		ret = __btrfs_inc_extent_ref(trans, node, parent, ref->root,
 					     ref->objectid, ref->offset,
···
 		struct btrfs_squota_delta delta = {
 			.root = href->owning_root,
 			.num_bytes = fs_info->nodesize,
-			.rsv_bytes = 0,
 			.is_data = false,
 			.is_inc = true,
 			.generation = trans->transid,
···
 	int ret = 0;
 
 	if (TRANS_ABORTED(trans)) {
-		if (insert_reserved)
+		if (insert_reserved) {
 			btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
+			free_head_ref_squota_rsv(trans->fs_info, href);
+		}
 		return 0;
 	}
 
···
 				  struct btrfs_delayed_ref_root *delayed_refs,
 				  struct btrfs_delayed_ref_head *head)
 {
+	u64 ret = 0;
+
 	/*
 	 * We had csum deletions accounted for in our delayed refs rsv, we need
 	 * to drop the csum leaves for this update from our delayed_refs_rsv.
···
 
 		btrfs_delayed_refs_rsv_release(fs_info, 0, nr_csums);
 
-		return btrfs_calc_delayed_ref_csum_bytes(fs_info, nr_csums);
+		ret = btrfs_calc_delayed_ref_csum_bytes(fs_info, nr_csums);
 	}
-	if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE &&
-	    head->must_insert_reserved && head->is_data)
-		btrfs_qgroup_free_refroot(fs_info, head->owning_root,
-					  head->reserved_bytes, BTRFS_QGROUP_RSV_DATA);
+	/* must_insert_reserved can be set only if we didn't run the head ref. */
+	if (head->must_insert_reserved)
+		free_head_ref_squota_rsv(fs_info, head);
 
-	return 0;
+	return ret;
 }
 
 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
···
 		 * spin lock.
 		 */
 		must_insert_reserved = locked_ref->must_insert_reserved;
+		/*
+		 * Unsetting this on the head ref relinquishes ownership of
+		 * the rsv_bytes, so it is critical that every possible code
+		 * path from here forward frees all reserves including qgroup
+		 * reserve.
+		 */
 		locked_ref->must_insert_reserved = false;
 
 		extent_op = locked_ref->extent_op;
···
 		struct btrfs_squota_delta delta = {
 			.root = delayed_ref_root,
 			.num_bytes = num_bytes,
-			.rsv_bytes = 0,
 			.is_data = is_data,
 			.is_inc = false,
 			.generation = btrfs_extent_generation(leaf, ei),
···
 		.root = root_objectid,
 		.num_bytes = ins->offset,
 		.generation = trans->transid,
-		.rsv_bytes = 0,
 		.is_data = true,
 		.is_inc = true,
 	};

fs/btrfs/qgroup.c

···
 	*root = RB_ROOT;
 }
 
+void btrfs_free_squota_rsv(struct btrfs_fs_info *fs_info, u64 root, u64 rsv_bytes)
+{
+	if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE)
+		return;
+
+	if (!is_fstree(root))
+		return;
+
+	btrfs_qgroup_free_refroot(fs_info, root, rsv_bytes, BTRFS_QGROUP_RSV_DATA);
+}
+
 int btrfs_record_squota_delta(struct btrfs_fs_info *fs_info,
 			      struct btrfs_squota_delta *delta)
 {
···
 
 out:
 	spin_unlock(&fs_info->qgroup_lock);
-	if (!ret && delta->rsv_bytes)
-		btrfs_qgroup_free_refroot(fs_info, root, delta->rsv_bytes,
-					  BTRFS_QGROUP_RSV_DATA);
 	return ret;
 }

fs/btrfs/qgroup.h

···
 	u64 root;
 	/* The number of bytes in the extent being counted. */
 	u64 num_bytes;
-	/* The number of bytes reserved for this extent. */
-	u64 rsv_bytes;
 	/* The generation the extent was created in. */
 	u64 generation;
 	/* Whether we are using or freeing the extent. */
···
 		struct btrfs_root *root, struct extent_buffer *eb);
 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans);
 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info);
+void btrfs_free_squota_rsv(struct btrfs_fs_info *fs_info, u64 root, u64 rsv_bytes);
 int btrfs_record_squota_delta(struct btrfs_fs_info *fs_info,
 			      struct btrfs_squota_delta *delta);