NFS - fix recent breakage to NFS error handling.
NeilBrown [Wed, 16 Nov 2011 00:46:31 +0000 (11:46 +1100)]
commit 2edb6bc3852c681c0d948245bd55108dc6407604 upstream.

From c6d615d2b97fe305cbf123a8751ced859dca1d5e Mon Sep 17 00:00:00 2001
From: NeilBrown <neilb@suse.de>
Date: Wed, 16 Nov 2011 09:39:05 +1100
Subject: NFS - fix recent breakage to NFS error handling.

commit 02c24a82187d5a628c68edfe71ae60dc135cd178 made a small and
presumably unintended change to write error handling in NFS.

Previously an error from filemap_write_and_wait_range would only be of
interest if nfs_file_fsync did not return an error.  After this commit,
an error from filemap_write_and_wait_range would mean that (the rest of)
nfs_file_fsync would not even be called.

This means that:
 1/ you are more likely to see EIO than e.g. EDQUOT or ENOSPC.
 2/ NFS_CONTEXT_ERROR_WRITE remains set for longer so more writes are
    synchronous.

This patch restores previous behaviour.

Cc: Josef Bacik <josef@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

fs/nfs/file.c

index babaf3a..b76be2f 100644 (file)
@@ -321,13 +321,13 @@ nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
                        datasync);
 
        ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
-       if (ret)
-               return ret;
        mutex_lock(&inode->i_mutex);
 
        nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
        have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
        status = nfs_commit_inode(inode, FLUSH_SYNC);
+       if (status >= 0 && ret < 0)
+               status = ret;
        have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
        if (have_error)
                ret = xchg(&ctx->error, 0);