Re: [PATCH 2/9] xfs: add a on-disk log header cycle array accessor

[Christoph Hellwig <hch@lst.de>]

On Tue, Oct 14, 2025 at 02:27:17PM -0700, Darrick J. Wong wrote:
> > @@ -1543,14 +1538,6 @@ xlog_pack_data(
> >  	if (xfs_has_logv2(log->l_mp)) {
> >  		xlog_in_core_2_t *xhdr = iclog->ic_data;
> >  
> > -		for ( ; i < BTOBB(size); i++) {
> > -			j = i / XLOG_CYCLE_DATA_SIZE;
> > -			k = i % XLOG_CYCLE_DATA_SIZE;
> > -			xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
> > -			*(__be32 *)dp = cycle_lsn;
> > -			dp += BBSIZE;
> > -		}
> 
> Just to orient myself -- this is the code that stamps (swazzled) LSN
> cycle numbers into the log headers, right?

Yes.

> > +static inline __be32 *xlog_cycle_data(struct xlog_rec_header *rhead, unsigned i)
> > +{
> > +	if (i >= XLOG_CYCLE_DATA_SIZE) {
> 
> Does this need a xfs_has_logv2() check?  The old callsites all seem to
> have them.
> 
> What should happen if i >= XLOG_CYCLE_DATA_SIZE && !logv2 ?  Should
> this helper return NULL so that callers can return EFSCORRUPTED or
> something like that?

It can't happen.  It is bound by l_iclog_hsize or h_len, which
can't go into the subsequent blocks for v1 logs.  For the recovery
case we also check for h_len being smaller than h_size, and h_size
is fixed to XLOG_BIG_RECORD_BSIZE for v1 logs.