On Wed, Jan 14, 2009 at 12:27 PM, Jan Kara [off-list ref] wrote:

On Tue 13-01-09 23:24:02, Theodore Tso wrote:

quoted

On Tue, Jan 13, 2009 at 04:14:11PM -0600, Eric Sandeen wrote:

quoted

This looks sane to me, and it does fix the below testcase.

Care to formally propose it?

Can we confirm what is being proposed?  From following this thread, I
think what folks are suggesting is:

1)  Revert the current "ext3/4: wait on all pending ocmmits in ext3/4_sync_fs"

 Yes.

quoted

2)  Apply Jan's patch "jbd[2]: Fix return value of journal_start_commit()"

 Yes.

quoted

3)  Also apply Jan's patch "jbd2: Skip commit of a transaction without
any buffers" since it appears to be a good optimization (although it's
not clear it would happen once we revert (1), above.

 Yes, it's an optimization but I'm still a bit afraid about something
relying on jbd2_journal_force_commit() implying a barrier which would not
always be a case after this patch... So we should probably audit all users of
ext4_force_commit() and check that this change is fine with them.

Ted/Jan/Eric,

I just wanted to followup on this to see what the plan is.  Items 1
and 2 haven't occurred in any of the ext4.git branches that I can see.
 I could be missing something but it seems this may have slipped
through the ext[34] cracks?

Mike