Re: [PATCH v13 3/8] block: add block layer APIs resembling Linux ZonedBlockDevice ioctls

[Stefan Hajnoczi]

On Wed, Nov 30, 2022 at 10:24:10AM +0800, Sam Li wrote:
> Stefan Hajnoczi <stefanha@gmail.com> 于2022年11月30日周三 10:01写道：
> > On Thu, 27 Oct 2022 at 11:46, Sam Li <faithilikerun@gmail.com> wrote:
> > > @@ -1374,9 +1428,11 @@ static int hdev_probe_blocksizes(BlockDriverState 
> > > *bs, BlockSizes *bsz)
> > >      int ret;
> > >
> > >      /* If DASD, get blocksizes */
> > > +#ifndef CONFIG_BLKZONED
> > >      if (check_for_dasd(s->fd) < 0) {
> > >          return -ENOTSUP;
> > >      }
> > > +#endif
> >
> > What is the purpose of this #ifndef? .bdrv_probe_blocksizes() should
> > only return block sizes for s390 DASD devices. I don't think zoned
> > storage needs block size probing here.
> 
> Zoned storage needs to be virtualized with the correct physical block
> size and logical block size. And the probing here can guarantee that.
> Or virtio-blk may send wrong block size to the guest. If manually set
> block size in the command line as before, it is somewhat inaccurate.

I see. I/O won't work if the guest block size differs from the physical
zoned device's block size.

However, we must not do this for regular host_device BlockDriverStates.
The block size is manually controlled from those devices and defaults to
512B. That way the blocksize doesn't change across live migration and
break the guest.

Please use a run-time check instead of an #ifdef. Only probe blocksizes
for dasd and zoned devices.

Stefan

signature.asc
Description: PGP signature