ksmbd: fix a mount write count leak in ksmbd_vfs_kern_path_locked()

If the call of ksmbd_vfs_lock_parent() fails, we drop the parent_path
references and return an error.  We need to drop the write access we
just got on parent_path->mnt before we drop the mount reference - callers
assume that ksmbd_vfs_kern_path_locked() returns with mount write
access grabbed if and only if it has returned 0.

Fixes: 864fb5d ("ksmbd: fix possible deadlock in smb2_open")
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>

Author: Al Viro

fs/smb/server/vfs.c

@@ -1282,6 +1282,7 @@ int ksmbd_vfs_kern_path_locked(struct ksmbd_work *work, char *name,
 
 		err = ksmbd_vfs_lock_parent(parent_path->dentry, path->dentry);
 		if (err) {
+			mnt_drop_write(parent_path->mnt);
 			path_put(path);
 			path_put(parent_path);
 		}