We should now have the logic in place to handle this properly
without regressing on the write performance, so re-enable
the sync writes.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
        if (test_clear_buffer_dirty(bh)) {
                get_bh(bh);
                bh->b_end_io = end_buffer_write_sync;
-               ret = submit_bh(WRITE, bh);
+               ret = submit_bh(WRITE_SYNC, bh);
                wait_on_buffer(bh);
                if (buffer_eopnotsupp(bh)) {
                        clear_buffer_eopnotsupp(bh);