ext4: fix error handling on inode bitmap corruption

When insert_inode_locked() fails in ext4_new_inode() it most likely means inode
bitmap got corrupted and we allocated again inode which is already in use. Also
doing unlock_new_inode() during error recovery is wrong since the inode does
not have I_NEW set. Fix the problem by jumping to fail: (instead of fail_drop:)
which declares filesystem error and does not call unlock_new_inode().

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>

fs/ext4/ialloc.c

@@ -885,8 +885,12 @@ struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
 	if (IS_DIRSYNC(inode))
 		ext4_handle_sync(handle);
 	if (insert_inode_locked(inode) < 0) {
-		err = -EINVAL;
-		goto fail_drop;
+		/*
+		 * Likely a bitmap corruption causing inode to be allocated
+		 * twice.
+		 */
+		err = -EIO;
+		goto fail;
 	}
 	spin_lock(&sbi->s_next_gen_lock);
 	inode->i_generation = sbi->s_next_generation++;