nfs: stat(2) fails during cthon04 basic test5 on NFSv4.0
authorChuck Lever <chuck.lever@oracle.com>
Wed, 6 May 2015 22:26:58 +0000 (18:26 -0400)
committerTrond Myklebust <trond.myklebust@primarydata.com>
Wed, 13 May 2015 18:56:03 +0000 (14:56 -0400)
When running the Connectathon basic tests against a Solaris NFS
server over NFSv4.0, test5 reports that stat(2) returns a file size
of zero instead of 1MB.

On success, nfs_commit_inode() can return a positive result; see
other call sites such as nfs_file_fsync_commit() and
nfs_commit_unstable_pages().

The call site recently added in nfs_wb_all() does not prevent that
positive return value from leaking to its callers. If it leaks
through nfs_sync_inode() back to nfs_getattr(), that causes stat(2)
to return a positive return value to user space while also not
filling in the passed-in struct stat.

Additional clean up: the new logic in nfs_wb_all() is rewritten in
bfields-normal form.

Fixes: 5bb89b4702e2 ("NFSv4.1/pnfs: Separate out metadata . . .")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
fs/nfs/write.c

index d12a4be613a5ced58599f8095822f4659b32f9a4..dfc19f1575a19d00bee1b0aeef6575e4416a9ef9 100644 (file)
@@ -1845,12 +1845,15 @@ int nfs_wb_all(struct inode *inode)
        trace_nfs_writeback_inode_enter(inode);
 
        ret = filemap_write_and_wait(inode->i_mapping);
-       if (!ret) {
-               ret = nfs_commit_inode(inode, FLUSH_SYNC);
-               if (!ret)
-                       pnfs_sync_inode(inode, true);
-       }
+       if (ret)
+               goto out;
+       ret = nfs_commit_inode(inode, FLUSH_SYNC);
+       if (ret < 0)
+               goto out;
+       pnfs_sync_inode(inode, true);
+       ret = 0;
 
+out:
        trace_nfs_writeback_inode_exit(inode, ret);
        return ret;
 }