Re: [PATCH v7 09/11] scsi: scsi_debug: Refuse a short WRITE ATOMIC (16) before writing it

[Niklas Cassel <cassel@kernel.org>]

Hello John,

On Fri, Sep 25, 2026 at 10:30:11AM +0100, John Garry wrote:
> > ---
> > Tested with:
> >
> >    modprobe scsi_debug sector_size=512 physblk_exp=3 dev_size_mb=128 \
> >        atomic_wr=1 lbpu=1
> >
> > issuing a WRITE ATOMIC (16) of eight blocks through SG_IO with a buffer
> > of four blocks of 0xaa. It fails with DID_ERROR before and after this
> > patch, but before it the first four blocks read back as 0xaa afterwards,
> > and after it all eight read back as they were. A WRITE ATOMIC (16) with
> > a full buffer writes all eight blocks, as before.
> >
>
> What is the command which you use?

SG_IO, with a CDB for eight blocks and a buffer of four:
sg_raw -s 2048 -i aa4 /dev/sdX 9c 00 00 00 00 00 00 00 00 40 00 00 00 08 00 00


>
> I am just wondering if this check should go into sd_setup_atomic_cmnd() or
> somewhere else higher up. I mean, this check is not really specific to
> scsi_debug, right?

sd can't produce a short atomic write: sd_setup_read_write_cmnd() derives
nr_blocks from blk_rq_sectors(), and the data buffer is the same request,
so the transfer length and the buffer always match.

A mismatch is only possible through SG_IO, where the caller supplies both.

SBC-6 4.28.2 puts the requirement on the device server: a failed atomic
write must leave the LBAs unaltered. scsi_debug is the device server here,
so I think this is the right place for it.


Kind regards,
Niklas