Re: [PATCH 1/8] xfs: don't try to write a start record into every iclog

["Darrick J. Wong" <darrick.wong@oracle.com>]

On Thu, Mar 26, 2020 at 07:08:28AM -0400, Brian Foster wrote:
> On Thu, Mar 26, 2020 at 10:25:00AM +1100, Dave Chinner wrote:
> > On Wed, Mar 25, 2020 at 08:33:14AM -0400, Brian Foster wrote:
> > > On Tue, Mar 24, 2020 at 06:44:52PM +0100, Christoph Hellwig wrote:
> > > > From: Dave Chinner <dchinner@redhat.com>
> > > > 
> > > > The xlog_write() function iterates over iclogs until it completes
> > > > writing all the log vectors passed in. The ticket tracks whether
> > > > a start record has been written or not, so only the first iclog gets
> > > > a start record. We only ever pass single use tickets to
> > > > xlog_write() so we only ever need to write a start record once per
> > > > xlog_write() call.
> > > > 
> > > > Hence we don't need to store whether we should write a start record
> > > > in the ticket as the callers provide all the information we need to
> > > > determine if a start record should be written. For the moment, we
> > > > have to ensure that we clear the XLOG_TIC_INITED appropriately so
> > > > the code in xfs_log_done() still works correctly for committing
> > > > transactions.
> > > > 
> > > > Signed-off-by: Dave Chinner <dchinner@redhat.com>
> > > > [hch: use an need_start_rec bool]
> > > > Signed-off-by: Christoph Hellwig <hch@lst.de>
> > > > ---
> > > >  fs/xfs/xfs_log.c | 63 ++++++++++++++++++++++++------------------------
> > > >  1 file changed, 32 insertions(+), 31 deletions(-)
> > > > 
> > > > diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c
> > > > index 2a90a483c2d6..bf071552094a 100644
> > > > --- a/fs/xfs/xfs_log.c
> > > > +++ b/fs/xfs/xfs_log.c
> > > ...
> > > > @@ -2372,25 +2359,29 @@ xlog_write(
> > > >  	int			record_cnt = 0;
> > > >  	int			data_cnt = 0;
> > > >  	int			error = 0;
> > > > +	bool			need_start_rec = true;
> > > >  
> > > >  	*start_lsn = 0;
> > > >  
> > > > -	len = xlog_write_calc_vec_length(ticket, log_vector);
> > > >  
> > > >  	/*
> > > >  	 * Region headers and bytes are already accounted for.
> > > >  	 * We only need to take into account start records and
> > > >  	 * split regions in this function.
> > > >  	 */
> > > > -	if (ticket->t_flags & XLOG_TIC_INITED)
> > > > -		ticket->t_curr_res -= sizeof(xlog_op_header_t);
> > > > +	if (ticket->t_flags & XLOG_TIC_INITED) {
> > > > +		ticket->t_curr_res -= sizeof(struct xlog_op_header);
> > > > +		ticket->t_flags &= ~XLOG_TIC_INITED;
> > > > +	}
> > > >  
> > > >  	/*
> > > >  	 * Commit record headers need to be accounted for. These
> > > >  	 * come in as separate writes so are easy to detect.
> > > >  	 */
> > > > -	if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
> > > > -		ticket->t_curr_res -= sizeof(xlog_op_header_t);
> > > > +	if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS)) {
> > > > +		ticket->t_curr_res -= sizeof(struct xlog_op_header);
> > > > +		need_start_rec = false;
> > > > +	}
> > > 
> > > Hmm.. I was asking for a comment update in v1 for this logic change.
> > > Looking through it again, what happens here if
> > > xfs_log_write_unmount_record() clears the UNMOUNT_TRANS flag for that
> > > summary counter check? That looks like a potential behavior change wrt
> > > to the start record..
> > 
> > xfs_log_write_unmount_record() clears the XLOG_TIC_INITED
> > flag before calling xlog_write(), so the current code never writes
> > out a start record for the unmount record. i.e. the unmount
> > record is a single region with the unmount log item in it, and
> > AFAICT this code does not change the behaviour of the unmount record
> > write at all.
> > 
> 
> I'm referring to the UNMOUNT_TRANS flag, not t_flags. With this patch,
> we actually would write a start record in some cases because
> need_start_rec is toggled based on the flags parameter and the summary
> counter check zeroes it.
> 
> > FWIW, that error injection code looks dodgy - it turns the unmount
> > record into an XFS_LOG transaction type with an invalid log item
> > type (0). That probably should be flagged as corruption, not be
> > silently ignored by recovery. IOWs, I think the error injection code
> > was wrong to begin with - if we want to ensure the log is dirty at
> > unmount, we should just skip writing the unmount record altogether.
> > 
> 
> That may be true... TBH I wasn't totally clear on what that logic was
> for (it isn't purely error injection). From the commit (f467cad95f5e3)
> log, the intent appears to be to "skip writing the unmount record," but
> that doesn't quite describe the behavior. Darrick might want to
> comment..? If we do revisit this, I'm mainly curious on whether there's
> a change in recovery behavior between having this specially crafted
> record vs. just writing nothing. For example, does recovery still set
> the head/tail based on this record even though we don't mark the log
> clean? If so, do we care..?

I'll have a look later today, but I think Dave is correct that the
summary recalc force code should bail out of xlog_unmount_write without
writing anything.

It's curious that recovery doesn't complain about this, but at this
point we've potentially been writing out logs with oh_flags == 0...

...though now that I look at xfs_log_recover.c, we don't ever actually
look for XLOG_UNMOUNT_TYPE, do we... heh.

--D

> 
> Brian
> 
> > Cheers,
> > 
> > Dave.
> > -- 
> > Dave Chinner
> > david@fromorbit.com
> > 
>