Theodore Tso wrote:

On Fri, May 16, 2008 at 02:09:56PM -0500, Eric Sandeen wrote:

quoted

To ensure that bits are truly on-disk after an fsync,
we should call blkdev_issue_flush if barriers are supported.

This patch isn't necessary, and in fact will cause a double flush.
When you call fsync(), it calls ext4_force_commit(), and we do a the
equivalent of a blkdev_issue_flush() today (which is what happenes
when you do a submit_bh(WRITE_BARRIER, bh), which is what setting
set_ordered_mode(bh) ends up causing.

ISTR fsync() on ext3 did not always force a commit, if in-place data
writes did not change any metadata.  Has this been fixed in ext4 but
not ext3 then?

Does WRITE_BARRIER always cause a flush?  It does not have to
according to Documentation/block/barrier.txt.  There are caveats about
tagged queuing "not yet implemented" in the text, but can we rely on
that?  The documentation is older than the current implementation;
those caveats might no longer apply.

-- Jamie