vfs pile 2 (automount work)

This pile consists mostly automount patchset by dhowells plus
fixes for related mainline bugs. There's more, but I'd rather keep pull requests separate; I'll send another one tonight and there will be more on Monday. Please, pull from

git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6.git/ for-linus

Shortlog:
Al Viro (3):
do_lookup() fix
fix old umount_tree() breakage
sanitize vfsmount refcounting changes

David Howells (11):
Add a dentry op to handle automounting rather than abusing follow_link() Add a dentry op to allow processes to be held during pathwalk transit Add an AT_NO_AUTOMOUNT flag to suppress terminal automount
AFS: Use d_automount() rather than abusing follow_link()
NFS: Use d_automount() rather than abusing follow_link()
CIFS: Use d_automount() rather than abusing follow_link() Remove the automount through follow_link() kludge code from pathwalk Remove a further kludge from __do_follow_link()
Allow d_manage() to be used in RCU-walk mode
Unexport do_add_mount() and add in follow_automount(), not ->d_automount()
autofs4: Merge the remaining dentry ops tables

Ian Kent (9):
autofs4: Add d_automount() dentry operation
autofs4: Add d_manage() dentry operation
autofs4: Remove unused code
autofs4: Clean up inode operations
autofs4: Clean up dentry operations
autofs4: Clean up autofs4_free_ino()
autofs4: Fix wait validation
autofs4: Add v4 pseudo direct mount support
autofs4: Bump version

Diffstat:

 Documentation/filesystems/Locking |    4 +
 Documentation/filesystems/vfs.txt |   43 +++
 drivers/mtd/mtdchar.c             |    2 +-
 drivers/staging/autofs/dirhash.c  |    5 +-
 fs/afs/dir.c                      |    1 +
 fs/afs/inode.c                    |    3 +-
 fs/afs/internal.h                 |    1 +
 fs/afs/mntpt.c                    |   56 +---
 fs/anon_inodes.c                  |    2 +-
 fs/autofs4/autofs_i.h             |   99 ++++--
 fs/autofs4/dev-ioctl.c            |    2 +-
 fs/autofs4/expire.c               |   51 ++--
 fs/autofs4/inode.c                |   26 +--
 fs/autofs4/root.c                 |  673 ++++++++++++++++---------------------
 fs/autofs4/waitq.c                |   17 +-
 fs/cifs/cifs_dfs_ref.c            |  120 +++----
 fs/cifs/cifsfs.h                  |    6 +
 fs/cifs/dir.c                     |    2 +
 fs/cifs/inode.c                   |    8 +-
 fs/dcache.c                       |    5 +-
 fs/fs_struct.c                    |   35 ++-
 fs/internal.h                     |    5 +
 fs/namei.c                        |  379 +++++++++++++++------
 fs/namespace.c                    |  179 +++++-----
 fs/nfs/dir.c                      |    4 +-
 fs/nfs/inode.c                    |    4 +-
 fs/nfs/internal.h                 |    1 +
 fs/nfs/namespace.c                |   79 ++---
 fs/nfsd/vfs.c                     |    5 +-
 fs/pipe.c                         |    2 +-
 fs/stat.c                         |    4 +-
 fs/super.c                        |    2 +-
 include/linux/auto_fs4.h          |    2 +-
 include/linux/dcache.h            |   16 +-
 include/linux/fcntl.h             |    1 +
 include/linux/fs.h                |    2 +
 include/linux/mount.h             |   11 +-
 include/linux/namei.h             |    5 +-
 include/linux/nfs_fs.h            |    1 -
 include/linux/path.h              |    2 -
 40 files changed, 1010 insertions(+), 855 deletions(-)

Hi,

f03c65993b98eeb909a4012ce7833c5857d74755 - sanitize vfsmount refcounting changes

Breaks my ARM !CONFIG_SMP compile

CC fs/namespace.o
fs/namespace.c: In function 'commit_tree':
fs/namespace.c:629: error: 'struct vfsmount' has no member named 'mnt_longterm' fs/namespace.c: In function 'umount_tree':
fs/namespace.c:1192: error: 'struct vfsmount' has no member named 'mnt_longterm' fs/namespace.c: In function 'mnt_make_longterm':
fs/namespace.c:2246: error: 'struct vfsmount' has no member named 'mnt_longterm' fs/namespace.c: In function 'mnt_make_shortterm':
fs/namespace.c:2251: error: 'struct vfsmount' has no member named 'mnt_longterm' fs/namespace.c:2254: error: 'struct vfsmount' has no member named 'mnt_longterm' fs/namespace.c: In function 'dup_mnt_ns':
fs/namespace.c:2295: error: 'struct vfsmount' has no member named 'mnt_longterm' fs/namespace.c:2299: error: 'struct vfsmount' has no member named 'mnt_longterm' fs/namespace.c:2305: error: 'struct vfsmount' has no member named 'mnt_longterm' fs/namespace.c: In function 'create_mnt_ns':
fs/namespace.c:2351: error: 'struct vfsmount' has no member named 'mnt_longterm' make[1]: *** [fs/namespace.o] Error 1
make: *** [fs] Error 2

My configuration is arch/arm/configs/at91rm9200_defconfig.

regards
Joachim Eastwood

Aaaaargh... Brown paperbag time - all testing was on SMP boxen ;-/ The fix follows. Linus, I've dropped that into for-linus in usual place
(git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6.git/), so if you prefer to pull it, it's there.

commit 91ed228ab50daa7aca54faa8bf5208b97e3448b8

Author: Al Viro <[email protected]>
Date:   Sun Jan 16 16:32:11 2011 -0500

mnt_longterm is there only on SMP

index 48809e2..9f544f3 100644
@@ -611,6 +611,21 @@ static void attach_mnt(struct vfsmount *mnt, struct path *path)
 	list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
 }
 
+static inline void __mnt_make_longterm(struct vfsmount *mnt)
+{
+#ifdef CONFIG_SMP
+	atomic_inc(&mnt->mnt_longterm);
+#endif
+}
+
+/* needs vfsmount lock for write */
+static inline void __mnt_make_shortterm(struct vfsmount *mnt)
+{
+#ifdef CONFIG_SMP
+	atomic_dec(&mnt->mnt_longterm);
+#endif
+}
+
 /*
  * vfsmount lock must be held for write
  */

@@ -626,7 +641,7 @@ static void commit_tree(struct vfsmount *mnt)
 	list_add_tail(&head, &mnt->mnt_list);
 	list_for_each_entry(m, &head, mnt_list) {
 		m->mnt_ns = n;
-		atomic_inc(&m->mnt_longterm);
+		__mnt_make_longterm(m);
 	}
 
 	list_splice(&head, n->list.prev);

@@ -1189,7 +1204,7 @@ void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
 		list_del_init(&p->mnt_list);
 		__touch_mnt_namespace(p->mnt_ns);
 		p->mnt_ns = NULL;
-		atomic_dec(&p->mnt_longterm);
+		__mnt_make_shortterm(p);
 		list_del_init(&p->mnt_child);
 		if (p->mnt_parent != p) {
 			p->mnt_parent->mnt_ghosts++;

@@ -2243,16 +2258,18 @@ static struct mnt_namespace *alloc_mnt_ns(void)
 
 void mnt_make_longterm(struct vfsmount *mnt)
 {
-	atomic_inc(&mnt->mnt_longterm);
+	__mnt_make_longterm(mnt);
 }
 
 void mnt_make_shortterm(struct vfsmount *mnt)
 {
+#ifdef CONFIG_SMP
 	if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
 		return;
 	br_write_lock(vfsmount_lock);
 	atomic_dec(&mnt->mnt_longterm);
 	br_write_unlock(vfsmount_lock);
+#endif
 }
 
 /*

@@ -2292,17 +2309,17 @@ static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
 	q = new_ns->root;
 	while (p) {
 		q->mnt_ns = new_ns;
-		atomic_inc(&q->mnt_longterm);
+		__mnt_make_longterm(q);
 		if (fs) {
 			if (p == fs->root.mnt) {
 				fs->root.mnt = mntget(q);
-				atomic_inc(&q->mnt_longterm);
+				__mnt_make_longterm(q);
 				mnt_make_shortterm(p);
 				rootmnt = p;
 			}
 			if (p == fs->pwd.mnt) {
 				fs->pwd.mnt = mntget(q);
-				atomic_inc(&q->mnt_longterm);
+				__mnt_make_longterm(q);
 				mnt_make_shortterm(p);
 				pwdmnt = p;
 			}

@@ -2348,7 +2365,7 @@ struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
 	new_ns = alloc_mnt_ns();
 	if (!IS_ERR(new_ns)) {
 		mnt->mnt_ns = new_ns;
-		atomic_inc(&mnt->mnt_longterm);
+		__mnt_make_longterm(mnt);
 		new_ns->root = mnt;
 		list_add(&new_ns->list, &new_ns->root->mnt_list);
 	}

Thanks,
The patch fixed it.

regards
Joachim Eastwood

In fact, any non-SMP compile, it's not ARM-specific.

The simple fix for the build breakage should be to just move the mnt_longterm thing out of the #ifdef CONFIG_SMP in include/linux/mount.h. But I do worry that it would cause some count imbalance, because there are some accesses that are still inside that CONFIG_SMP case in mntput_no_expire().

Al, please take a look,

Linus

Already fixed. Actually, taking it out of ifdef would work (the only place that actually cares about the value of that sucker is SMP side of mntput()), but we are obviously better off just not touching it on UP at all - why do pointless work and waste space?

See the patch upthread. ->mnt_longterm is SMP-only optimization of mntput(); it's there only to free the common case of mntput() from cacheline bouncing and on UP it's needed at all.

You push "mnt_longterm is there only on SMP" patch also for "for-next" GIT branch?

- Sedat -
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

PS: the patch does survive UP beating. Could you pull
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6.git/ for-linus ? There's only one patch at the moment:

Al Viro (1):
mnt_longterm is there only on SMP

 fs/namespace.c |   31 ++++++++++++++++++++++++-------
 1 files changed, 24 insertions(+), 7 deletions(-)