Re: [Qemu-devel] [PATCH V2] hd-geometry.c: Integrate HDIO_GETGEO in guessing

[Markus Armbruster <armbru@redhat.com>]

Einar Lueck <elelueck@linux.vnet.ibm.com> writes:

> This patch extends the function hd_geometry_guess. If no geo could
> be guessed via guess_disk_lchs, a new function called guess_disk_pchs is
> called. The latter utilizes HDIO_GET_GEO ioctl to ask the underlying disk
> for geometry.
> If this is not successful (e.g. image files) geometry is derived
> from the size of the disk (as before).
> The new HDIO_GETGEO logic is required for two use cases:
> a) Support for geometries of Direct Attached Storage Disks (DASD)
> on s390x configured as backing of virtio block devices.
> b) Support for FCP attached SCSI disks that do not yet have a
> partition table. Without this patch, fdisk -l on the host would
> return different results then fdisk -l in the guest.

I'm afraid this could mess up existing, working disks.

Consider a disk where guess_disk_lchs() fails and HDIO_GETGEO succeeds.

The old code arbitrarily picks a "standard" physical geometry then.

Your code picks the one returned by HDIO_GETGEO.

Unless the two geometries happen to lead to a compatible address
translation, any guest that actually uses the translation now sees
different disk contents, with predictably bad results.

Can this happen?  If no, why not?