From: Steve Dickson <SteveD@redhat.com>
Subject: Re: [RHEL5.1][PATCH] NFS4: closes and umounts are racing.
Date: Mon, 20 Aug 2007 13:10:33 -0400
Bugzilla: 245062
Message-Id: <46C9CB09.2090103@RedHat.com>
Changelog: [nfs] NFS4: closes and umounts are racing




Steve Dickson wrote:
>The attached patch avoids an oops during a
>race between that occurs between the rpciod daemon
>trying to do a nfs4_close and the filesystem being unmounted.
>The window is small and only happens when all of the NFS
>debugging is turned (although i was never able to reproduce
>the oops) but it does exist.
>
>Note, the key word is 'avoids'. This patch does not fix the
>problem, just avoids the oops...
>
>The true fix is currently being worked on in upstream
>which basically means there a patch but has very
>little (i.e. none) testing. So I've decided to submit
>this patch (since today is the deadline) with the idea
>of updating the patch when the final solution is a bit
>more baked.
>
>The bz for this is:
>https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=225515
Here is the upstream version that fixes this close/umount race.
I'm still not able to reproduce this problem (with or without this
patch) and the customer has not come back with any hard results
either.

But the patch does look reasonable and has been "baking" in upstream
since the beginning of the summer...

This bz is a blocker bz, which are due today, so if could
please take a look sooner verses later, that would good...

tia,

steved.


commit a0356862bcbeb20acf64bc1a82d28a4c5bb957a7
Author: Trond Myklebust <Trond.Myklebust@netapp.com>

    NFS: Fix nfs_reval_fsid()
    
    We don't need to revalidate the fsid on the root directory. It suffices to
    revalidate it on the current directory.
    
    Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>


--- linux-2.6.18.i686/fs/nfs/dir.c.orig	2007-08-20 10:51:29.094361000 -0400
+++ linux-2.6.18.i686/fs/nfs/dir.c	2007-08-20 12:30:09.080228000 -0400
@@ -871,14 +871,13 @@ int nfs_is_exclusive_create(struct inode
 	return (nd->intent.open.flags & O_EXCL) != 0;
 }
 
-static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir,
-				 struct nfs_fh *fh, struct nfs_fattr *fattr)
+static inline int nfs_reval_fsid(struct inode *dir, const struct nfs_fattr *fattr)
 {
 	struct nfs_server *server = NFS_SERVER(dir);
 
 	if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
-		/* Revalidate fsid on root dir */
-		return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode);
+		/* Revalidate fsid using the parent directory */
+		return __nfs_revalidate_inode(server, dir);
 	return 0;
 }
 
@@ -920,7 +919,7 @@ static struct dentry *nfs_lookup(struct 
 		res = ERR_PTR(error);
 		goto out_unlock;
 	}
-	error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr);
+	error = nfs_reval_fsid(dir, &fattr);
 	if (error < 0) {
 		res = ERR_PTR(error);
 		goto out_unlock;
--- linux-2.6.18.i686/fs/nfs/inode.c.orig	2007-08-20 10:51:42.320416000 -0400
+++ linux-2.6.18.i686/fs/nfs/inode.c	2007-08-20 12:34:52.103026000 -0400
@@ -903,8 +903,8 @@ static int nfs_update_inode(struct inode
 		goto out_changed;
 
 	server = NFS_SERVER(inode);
-	/* Update the fsid if and only if this is the root directory */
-	if ( inode->i_sb->s_root && inode == inode->i_sb->s_root->d_inode
+	/* Update the fsid? */
+	if (S_ISDIR(inode->i_mode)
 			&& !nfs_fsid_equal(&server->fsid, &fattr->fsid))
 		server->fsid = fattr->fsid;