[PATCH 1/6] Add a dentry op to handle automounting rather than abusing follow_link() [ver #2]

[David Howells]

Ian Kent <raven at themaw.net> wrote:

> > Does it make autofs easier if d_op->d_automount() is allowed to return
> > -EXDEV to request this?  Then you can return it in Oz mode to allow the
> > daemon to see/use the underlying mountpoint without recursing back into
> > d_automount().
> 
> Yes, it's really useful.

I think what's required, then, is if d_automount() returns -EXDEV then:

 (1) If the dentry is terminal in the lookup path, then we just return -EXDEV
     to indicate to __follow_mount() that we really do want to stop there.

 (2) If the dentry is not terminal, then we convert the error to -EREMOTE to
     indicate that we can't complete the pathwalk as one of the earlier
     components is inaccessible.

See the attached patch.

David
---
diff --git a/fs/namei.c b/fs/namei.c
index c154112..6c385d4 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -653,8 +653,20 @@ static int follow_automount(struct path *path, unsigned flags, int res)
 		return -ELOOP;
 
 	mnt = path->dentry->d_op->d_automount(path);
-	if (IS_ERR(mnt))
+	if (IS_ERR(mnt)) {
+		/*
+		 * The filesystem is allowed to return -EXDEV here to indicate
+		 * they don't want to automount.  For instance, autofs would do
+		 * this so that its userspace daemon can mount on this dentry.
+		 *
+		 * However, we can only permit this if it's a terminal point in
+		 * the path being looked up; if it wasn't then the remainder of
+		 * the path is inaccessible and we should say so.
+		 */
+		if (PTR_ERR(mnt) == -EXDEV && (flags & LOOKUP_CONTINUE))
+			return -EREMOTE;
 		return PTR_ERR(mnt);
+	}
 	if (!mnt) /* mount collision */
 		return 0;