Currently, O_DIRECT writes in a hole or outside the 'i_size' was not
actually honored, while using a buffered io instead.

It hurts/cheats the 'O_DIRECT' semantics somehow, users may be confusing
about the data lost after a crash when performing the odirect append writes.

A straightforward way is to flush the pagecache like O_SYNC did to guarantee
the real touch to disk before writes get returned.

Signed-off-by: Tristan Ye <[email protected]>
---
 fs/ocfs2/file.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 2b10b36..97b41f8 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -2321,7 +2321,7 @@ out_dio:
        BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
 
        if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
-           ((file->f_flags & O_DIRECT) && has_refcount)) {
+           ((file->f_flags & O_DIRECT) && !direct_io)) {
                ret = filemap_fdatawrite_range(file->f_mapping, pos,
                                               pos + count - 1);
                if (ret < 0)
-- 
1.5.5


_______________________________________________
Ocfs2-devel mailing list
[email protected]
http://oss.oracle.com/mailman/listinfo/ocfs2-devel

Reply via email to