1 files changed, 48 insertions, 12 deletions

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 7755a0362a3a..c89e85a7da7d 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -6830,7 +6830,7 @@ static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
 
-	err = btrfs_find_free_ino(root, &objectid);
+	err = btrfs_find_free_objectid(root, &objectid);
 	if (err)
 		goto out_unlock;
 
@@ -6894,7 +6894,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
 
-	err = btrfs_find_free_ino(root, &objectid);
+	err = btrfs_find_free_objectid(root, &objectid);
 	if (err)
 		goto out_unlock;
 
@@ -7039,7 +7039,7 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
 
-	err = btrfs_find_free_ino(root, &objectid);
+	err = btrfs_find_free_objectid(root, &objectid);
 	if (err)
 		goto out_fail;
 
@@ -9751,8 +9751,6 @@ 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,
@@ -9768,8 +9766,6 @@ 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,
@@ -9797,6 +9793,29 @@ 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);
@@ -9911,7 +9930,7 @@ static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
 	u64 objectid;
 	u64 index;
 
-	ret = btrfs_find_free_ino(root, &objectid);
+	ret = btrfs_find_free_objectid(root, &objectid);
 	if (ret)
 		return ret;
 
@@ -10046,8 +10065,6 @@ 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,
@@ -10071,6 +10088,25 @@ 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,
@@ -10380,7 +10416,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
 
-	err = btrfs_find_free_ino(root, &objectid);
+	err = btrfs_find_free_objectid(root, &objectid);
 	if (err)
 		goto out_unlock;
 
@@ -10663,7 +10699,7 @@ static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
 
-	ret = btrfs_find_free_ino(root, &objectid);
+	ret = btrfs_find_free_objectid(root, &objectid);
 	if (ret)
 		goto out;