Merging upstream version 5.10.216.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

1 files changed, 62 insertions, 29 deletions

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index c900a3966..591caac2b 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -4007,7 +4007,8 @@ int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
 	struct btrfs_block_rsv block_rsv;
 	u64 root_flags;
 	int ret;
-	int err;
+
+	down_write(&fs_info->subvol_sem);
 
 	/*
 	 * Don't allow to delete a subvolume with send in progress. This is
@@ -4020,25 +4021,25 @@ int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
 		btrfs_warn(fs_info,
 			   "attempt to delete subvolume %llu during send",
 			   dest->root_key.objectid);
-		return -EPERM;
+		ret = -EPERM;
+		goto out_up_write;
 	}
 	if (atomic_read(&dest->nr_swapfiles)) {
 		spin_unlock(&dest->root_item_lock);
 		btrfs_warn(fs_info,
 			   "attempt to delete subvolume %llu with active swapfile",
 			   root->root_key.objectid);
-		return -EPERM;
+		ret = -EPERM;
+		goto out_up_write;
 	}
 	root_flags = btrfs_root_flags(&dest->root_item);
 	btrfs_set_root_flags(&dest->root_item,
 			     root_flags | BTRFS_ROOT_SUBVOL_DEAD);
 	spin_unlock(&dest->root_item_lock);
 
-	down_write(&fs_info->subvol_sem);
-
-	err = may_destroy_subvol(dest);
-	if (err)
-		goto out_up_write;
+	ret = may_destroy_subvol(dest);
+	if (ret)
+		goto out_undead;
 
 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 	/*
@@ -4046,13 +4047,13 @@ int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
 	 * two for dir entries,
 	 * two for root ref/backref.
 	 */
-	err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
-	if (err)
-		goto out_up_write;
+	ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
+	if (ret)
+		goto out_undead;
 
 	trans = btrfs_start_transaction(root, 0);
 	if (IS_ERR(trans)) {
-		err = PTR_ERR(trans);
+		ret = PTR_ERR(trans);
 		goto out_release;
 	}
 	trans->block_rsv = &block_rsv;
@@ -4062,7 +4063,6 @@ int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
 
 	ret = btrfs_unlink_subvol(trans, dir, dentry);
 	if (ret) {
-		err = ret;
 		btrfs_abort_transaction(trans, ret);
 		goto out_end_trans;
 	}
@@ -4080,7 +4080,6 @@ int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
 					dest->root_key.objectid);
 		if (ret) {
 			btrfs_abort_transaction(trans, ret);
-			err = ret;
 			goto out_end_trans;
 		}
 	}
@@ -4090,7 +4089,6 @@ int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
 				  dest->root_key.objectid);
 	if (ret && ret != -ENOENT) {
 		btrfs_abort_transaction(trans, ret);
-		err = ret;
 		goto out_end_trans;
 	}
 	if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
@@ -4100,7 +4098,6 @@ int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
 					  dest->root_key.objectid);
 		if (ret && ret != -ENOENT) {
 			btrfs_abort_transaction(trans, ret);
-			err = ret;
 			goto out_end_trans;
 		}
 	}
@@ -4111,20 +4108,20 @@ out_end_trans:
 	trans->block_rsv = NULL;
 	trans->bytes_reserved = 0;
 	ret = btrfs_end_transaction(trans);
-	if (ret && !err)
-		err = ret;
 	inode->i_flags |= S_DEAD;
 out_release:
 	btrfs_subvolume_release_metadata(root, &block_rsv);
-out_up_write:
-	up_write(&fs_info->subvol_sem);
-	if (err) {
+out_undead:
+	if (ret) {
 		spin_lock(&dest->root_item_lock);
 		root_flags = btrfs_root_flags(&dest->root_item);
 		btrfs_set_root_flags(&dest->root_item,
 				root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
 		spin_unlock(&dest->root_item_lock);
-	} else {
+	}
+out_up_write:
+	up_write(&fs_info->subvol_sem);
+	if (!ret) {
 		d_invalidate(dentry);
 		btrfs_prune_dentries(dest);
 		ASSERT(dest->send_in_progress == 0);
@@ -4136,7 +4133,7 @@ out_up_write:
 		}
 	}
 
-	return err;
+	return ret;
 }
 
 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
@@ -8971,8 +8968,6 @@ static int btrfs_rename_exchange(struct inode *old_dir,
 		/* force full log commit if subvolume involved. */
 		btrfs_set_log_full_commit(trans);
 	} else {
-		btrfs_pin_log_trans(root);
-		root_log_pinned = true;
 		ret = btrfs_insert_inode_ref(trans, dest,
 					     new_dentry->d_name.name,
 					     new_dentry->d_name.len,
@@ -8989,8 +8984,6 @@ static int btrfs_rename_exchange(struct inode *old_dir,
 		/* force full log commit if subvolume involved. */
 		btrfs_set_log_full_commit(trans);
 	} else {
-		btrfs_pin_log_trans(dest);
-		dest_log_pinned = true;
 		ret = btrfs_insert_inode_ref(trans, root,
 					     old_dentry->d_name.name,
 					     old_dentry->d_name.len,
@@ -9021,6 +9014,29 @@ static int btrfs_rename_exchange(struct inode *old_dir,
 				BTRFS_I(new_inode), 1);
 	}
 
+	/*
+	 * Now pin the logs of the roots. We do it to ensure that no other task
+	 * can sync the logs while we are in progress with the rename, because
+	 * that could result in an inconsistency in case any of the inodes that
+	 * are part of this rename operation were logged before.
+	 *
+	 * We pin the logs even if at this precise moment none of the inodes was
+	 * logged before. This is because right after we checked for that, some
+	 * other task fsyncing some other inode not involved with this rename
+	 * operation could log that one of our inodes exists.
+	 *
+	 * We don't need to pin the logs before the above calls to
+	 * btrfs_insert_inode_ref(), since those don't ever need to change a log.
+	 */
+	if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
+		btrfs_pin_log_trans(root);
+		root_log_pinned = true;
+	}
+	if (new_ino != BTRFS_FIRST_FREE_OBJECTID) {
+		btrfs_pin_log_trans(dest);
+		dest_log_pinned = true;
+	}
+
 	/* src is a subvolume */
 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
 		ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
@@ -9270,8 +9286,6 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 		/* force full log commit if subvolume involved. */
 		btrfs_set_log_full_commit(trans);
 	} else {
-		btrfs_pin_log_trans(root);
-		log_pinned = true;
 		ret = btrfs_insert_inode_ref(trans, dest,
 					     new_dentry->d_name.name,
 					     new_dentry->d_name.len,
@@ -9295,6 +9309,25 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
 		ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
 	} else {
+		/*
+		 * Now pin the log. We do it to ensure that no other task can
+		 * sync the log while we are in progress with the rename, as
+		 * that could result in an inconsistency in case any of the
+		 * inodes that are part of this rename operation were logged
+		 * before.
+		 *
+		 * We pin the log even if at this precise moment none of the
+		 * inodes was logged before. This is because right after we
+		 * checked for that, some other task fsyncing some other inode
+		 * not involved with this rename operation could log that one of
+		 * our inodes exists.
+		 *
+		 * We don't need to pin the logs before the above call to
+		 * btrfs_insert_inode_ref(), since that does not need to change
+		 * a log.
+		 */
+		btrfs_pin_log_trans(root);
+		log_pinned = true;
 		ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
 					BTRFS_I(d_inode(old_dentry)),
 					old_dentry->d_name.name,