statfs() gives ESTALE error
authorMenyhart Zoltan <Zoltan.Menyhart@bull.net>
Sun, 12 Sep 2010 23:55:26 +0000 (19:55 -0400)
committerTrond Myklebust <Trond.Myklebust@netapp.com>
Sun, 12 Sep 2010 23:55:26 +0000 (19:55 -0400)
Hi,

An NFS client executes a statfs("file", &buff) call.
"file" exists / existed, the client has read / written it,
but it has already closed it.

user_path(pathname, &path) looks up "file" successfully in the
directory-cache  and restarts the aging timer of the directory-entry.
Even if "file" has already been removed from the server, because the
lookupcache=positive option I use, keeps the entries valid for a while.

nfs_statfs() returns ESTALE if "file" has already been removed from the
server.

If the user application repeats the statfs("file", &buff) call, we
are stuck: "file" remains young forever in the directory-cache.

Signed-off-by: Zoltan Menyhart <Zoltan.Menyhart@bull.net>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: stable@kernel.org
fs/nfs/super.c

index ec3966e4706b2f70d8709199f86456392a6e6584..f4cbf0c306c64d6583cbfffaa14cfbc7a467dfaa 100644 (file)
@@ -431,7 +431,15 @@ static int nfs_statfs(struct dentry *dentry, struct kstatfs *buf)
                goto out_err;
 
        error = server->nfs_client->rpc_ops->statfs(server, fh, &res);
+       if (unlikely(error == -ESTALE)) {
+               struct dentry *pd_dentry;
 
+               pd_dentry = dget_parent(dentry);
+               if (pd_dentry != NULL) {
+                       nfs_zap_caches(pd_dentry->d_inode);
+                       dput(pd_dentry);
+               }
+       }
        nfs_free_fattr(res.fattr);
        if (error < 0)
                goto out_err;