nfs: Fix cache_validity check in nfs_write_pageuptodate()
authorScott Mayhew <smayhew@redhat.com>
Fri, 20 Jun 2014 12:44:42 +0000 (08:44 -0400)
committerTrond Myklebust <trond.myklebust@primarydata.com>
Tue, 24 Jun 2014 22:46:56 +0000 (18:46 -0400)
NFS_INO_INVALID_DATA cannot be ignored, even if we have a delegation.

We're still having some problems with data corruption when multiple
clients are appending to a file and those clients are being granted
write delegations on open.

To reproduce:

Client A:
vi /mnt/`hostname -s`
while :; do echo "XXXXXXXXXXXXXXX" >>/mnt/file; sleep $(( $RANDOM % 5 )); done

Client B:
vi /mnt/`hostname -s`
while :; do echo "YYYYYYYYYYYYYYY" >>/mnt/file; sleep $(( $RANDOM % 5 )); done

What's happening is that in nfs_update_inode() we're recognizing that
the file size has changed and we're setting NFS_INO_INVALID_DATA
accordingly, but then we ignore the cache_validity flags in
nfs_write_pageuptodate() because we have a delegation.  As a result,
in nfs_updatepage() we're extending the write to cover the full page
even though we've not read in the data to begin with.

Signed-off-by: Scott Mayhew <smayhew@redhat.com>
Cc: <stable@vger.kernel.org> # v3.11+
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
fs/nfs/write.c

index 3ee5af4e738efb0f014d5b9dbd9963a179de5dc5..98ff061ccaf3650c6069b251e879a5537d6a8d7f 100644 (file)
@@ -934,12 +934,14 @@ static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
 
        if (nfs_have_delegated_attributes(inode))
                goto out;
-       if (nfsi->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE))
+       if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
                return false;
        smp_rmb();
        if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
                return false;
 out:
+       if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
+               return false;
        return PageUptodate(page) != 0;
 }