Re: [PATCH] null_blk: allow byte aligned memory offsets

[Keith Busch <kbusch@kernel.org>]

On Tue, Nov 04, 2025 at 10:48:25AM +0900, Damien Le Moal wrote:
> >  		if (valid_len) {
> > -			err = copy_from_nullb(nullb, page, off,
> > -				sector, valid_len);
> > -			off += valid_len;
> > +			err = copy_from_nullb(nullb, p, pos, valid_len);
> 
> Not your fault, but if this fails, we will still do the nullb_fill_pattern()
> below which I do not think is correct... ? May be we should have:
> 
> 			if (err)
> 				return err;
> 
> here ? But not sure if we should still call flush_dcache_page() even on error
> though.

It does look odd. copy_from_nullb() only returns success though, so
maybe we just drop the return value entirely.

> > @@ -1276,25 +1275,26 @@ static blk_status_t null_handle_data_transfer(struct nullb_cmd *cmd,
> >  	struct nullb *nullb = cmd->nq->dev->nullb;
> >  	int err = 0;
> >  	unsigned int len;
> > -	sector_t sector = blk_rq_pos(rq);
> > -	unsigned int max_bytes = nr_sectors << SECTOR_SHIFT;
> > -	unsigned int transferred_bytes = 0;
> > +	loff_t pos = blk_rq_pos(rq) << SECTOR_SHIFT;
> > +	unsigned int nr_bytes = nr_sectors << SECTOR_SHIFT;
> 
> Overflow potential here ?

Should be okay: nr_sectors comes from blk_rq_sectors().

The same calculation already exist just above, I just changed the name.
Actually, I don't know why I changed it, so I'll leave it alone in the
next version.