Re: [RFC][PATCH v2 02/34] fix automount/automount race properly

[Al Viro]

On Sat, Feb 22, 2020 at 06:07:39PM -0800, Linus Torvalds wrote:
> On Sat, Feb 22, 2020 at 5:16 PM Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
> >
> > +
> > +discard2:
> > +       namespace_unlock();
> > +discard1:
> > +       inode_unlock(dentry->d_inode);
> > +discard:
> >         /* remove m from any expiration list it may be on */
> 
> Would you mind re-naming those labels?
> 
> I realize that the numbering may help show that the error handling is
> done in the reverse order, but I bet that a nice name could so that
> too.

Umm...  A bit of reordering in the beginning eliminates discard1, suggesting
s/discard2/discard_locked/...  Incremental would be

diff --git a/fs/namespace.c b/fs/namespace.c
index 6228fd1ef94f..777c3116e62e 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2844,22 +2844,22 @@ int finish_automount(struct vfsmount *m, struct path *path)
 	 * got", not "try to mount it on top".
 	 */
 	inode_lock(dentry->d_inode);
+	namespace_lock();
 	if (unlikely(cant_mount(dentry))) {
 		err = -ENOENT;
-		goto discard1;
+		goto discard_locked;
 	}
-	namespace_lock();
 	rcu_read_lock();
 	if (unlikely(__lookup_mnt(path->mnt, dentry))) {
 		rcu_read_unlock();
 		err = 0;
-		goto discard2;
+		goto discard_locked;
 	}
 	rcu_read_unlock();
 	mp = get_mountpoint(dentry);
 	if (IS_ERR(mp)) {
 		err = PTR_ERR(mp);
-		goto discard2;
+		goto discard_locked;
 	}
 
 	err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
@@ -2869,9 +2869,8 @@ int finish_automount(struct vfsmount *m, struct path *path)
 	mntput(m);
 	return 0;
 
-discard2:
+discard_locked:
 	namespace_unlock();
-discard1:
 	inode_unlock(dentry->d_inode);
 discard:
 	/* remove m from any expiration list it may be on */