[patch] add dentry revalidate to follow mount.

[patch] add dentry revalidate to follow mount.

[pravin shelar]

[-- Attachment #1: Type: text/plain, Size: 333 bytes --]

Hi
          i have seen error in lustre while setting permission on fs root
from a client. problem is permission are not getting propagated to
other clients.

this is because of do_lookup() call path which does not revalidate
fs root dentry in follow_mount()

attached patch adds revalidation call in follow_mount.

Thanks,
Pravin


[-- Attachment #2: follow-mount-2630.patch --]
[-- Type: text/x-diff, Size: 1891 bytes --]

Signed-off-by: pravin shelar <pravin.shelar@sun.com>

Index: linux-2.6-latest/fs/namei.c
===================================================================
--- linux-2.6-latest.orig/fs/namei.c
+++ linux-2.6-latest/fs/namei.c
@@ -691,9 +691,11 @@ int follow_up(struct vfsmount **mnt, str
 /* no need for dcache_lock, as serialization is taken care in
  * namespace.c
  */
-static int __follow_mount(struct path *path)
+static int __follow_mount(struct nameidata *nd, struct path *path)
 {
 	int res = 0;
+	struct dentry *dentry;
+
 	while (d_mountpoint(path->dentry)) {
 		struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
 		if (!mounted)
@@ -703,6 +705,18 @@ static int __follow_mount(struct path *p
 			mntput(path->mnt);
 		path->mnt = mounted;
 		path->dentry = dget(mounted->mnt_root);
+		dentry = path->dentry;
+
+		if (dentry->d_op && dentry->d_op->d_revalidate){
+			dentry = do_revalidate(dentry, nd);
+			if (!dentry)
+				return -EINVAL;
+			if (IS_ERR(dentry))
+				return PTR_ERR(dentry);
+		}
+
+		path->dentry = dentry;
+
 		res = 1;
 	}
 	return res;
@@ -788,6 +802,7 @@ static int do_lookup(struct nameidata *n
 {
 	struct vfsmount *mnt = nd->path.mnt;
 	struct dentry *dentry = __d_lookup(nd->path.dentry, name);
+	int rc;
 
 	if (!dentry)
 		goto need_lookup;
@@ -796,7 +811,9 @@ static int do_lookup(struct nameidata *n
 done:
 	path->mnt = mnt;
 	path->dentry = dentry;
-	__follow_mount(path);
+	rc = __follow_mount(nd, path);
+	if (rc < 0)
+		return rc;
 	return 0;
 
 need_lookup:
@@ -1747,11 +1764,15 @@ do_last:
 	if (flag & O_EXCL)
 		goto exit_dput;
 
-	if (__follow_mount(&path)) {
-		error = -ELOOP;
-		if (flag & O_NOFOLLOW)
+	error = __follow_mount(&nd, &path);
+	if (error == 1) {
+		if (flag & O_NOFOLLOW) {
+			error = -ELOOP;
 			goto exit_dput;
+		}
 	}
+	if (error < 0)
+		goto exit_dput;
 
 	error = -ENOENT;
 	if (!path.dentry->d_inode)

Re: [patch] add dentry revalidate to follow mount.

[Al Viro]

On Mon, Jun 08, 2009 at 07:14:15PM +0530, pravin shelar wrote:
> Hi
>          i have seen error in lustre while setting permission on fs root
> from a client. problem is permission are not getting propagated to
> other clients.
>
> this is because of do_lookup() call path which does not revalidate
> fs root dentry in follow_mount()
>
> attached patch adds revalidation call in follow_mount.

... and makes umount() of such thing impossible.  NAK.  That's not a solution.

Re: [patch] add dentry revalidate to follow mount.

[Al Viro]

On Tue, Jun 09, 2009 at 01:53:56AM +0530, pravin shelar wrote:
> Al Viro wrote:
>> On Mon, Jun 08, 2009 at 07:14:15PM +0530, pravin shelar wrote:
>>> Hi
>>>          i have seen error in lustre while setting permission on fs root
>>> from a client. problem is permission are not getting propagated to
>>> other clients.
>>>
>>> this is because of do_lookup() call path which does not revalidate
>>> fs root dentry in follow_mount()
>>>
>>> attached patch adds revalidation call in follow_mount.
>>
>> ... and makes umount() of such thing impossible.  NAK.  That's not a solution.
>
> i am not sure how this patch could cause problem for umount. i have 
> tested patch with nfs and it worked fine.
> can u elaborate what is problem?

Think what happens if revalidate fails and keeps failing.  Would do
wonders to pathname resolution in umount(2)...

Re: [patch] add dentry revalidate to follow mount.

[pravin shelar]

Al Viro wrote:
> On Mon, Jun 08, 2009 at 07:14:15PM +0530, pravin shelar wrote:
>> Hi
>>          i have seen error in lustre while setting permission on fs root
>> from a client. problem is permission are not getting propagated to
>> other clients.
>>
>> this is because of do_lookup() call path which does not revalidate
>> fs root dentry in follow_mount()
>>
>> attached patch adds revalidation call in follow_mount.
> 
> ... and makes umount() of such thing impossible.  NAK.  That's not a solution.

i am not sure how this patch could cause problem for umount. i have tested patch 
with nfs and it worked fine.
can u elaborate what is problem?

thanks,
Pravin.

Re: [patch] add dentry revalidate to follow mount.

[Andreas Dilger]

On Jun 08, 2009  21:17 +0100, Al Viro wrote:
> On Tue, Jun 09, 2009 at 01:53:56AM +0530, pravin shelar wrote:
> > Al Viro wrote:
> >> On Mon, Jun 08, 2009 at 07:14:15PM +0530, pravin shelar wrote:
> >>> Hi
> >>>          i have seen error in lustre while setting permission on fs root
> >>> from a client. problem is permission are not getting propagated to
> >>> other clients.
> >>>
> >>> this is because of do_lookup() call path which does not revalidate
> >>> fs root dentry in follow_mount()
> >>>
> >>> attached patch adds revalidation call in follow_mount.
> >>
> >> ... and makes umount() of such thing impossible.  NAK.  That's not a solution.
> >
> > i am not sure how this patch could cause problem for umount. i have 
> > tested patch with nfs and it worked fine.
> > can u elaborate what is problem?
> 
> Think what happens if revalidate fails and keeps failing.  Would do
> wonders to pathname resolution in umount(2)...

Would having a limit on the number of revalidations be a suitable
compromise?

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.