Re: [PATCH] Install to LVM PVs

[Dryden Personalis]

Andrei Borzenkov schreef op 08-05-2016 10:47:
> 08.05.2016 09:05, Andrei Borzenkov пишет:
> > > +#ifdef GRUB_UTIL
> > > +int
> > > +grub_util_is_lvm(grub_disk_t disk)
> > > +{
> > > +  struct grub_diskfilter_pv_id id;
> > > +  struct grub_diskfilter_vg *vg;
> > > +  grub_disk_addr_t start_sector;
> > > +  vg = grub_lvm_detect(disk, &id, &start_sector);
> > > +  if (! vg)
> > > +    return 0;
> > > +  /* don't free the vg, it's held by grub_diskfilter_vg_register */
> > > +  grub_free(id.uuid);
> > > +  return 1;
> > > +}
> > > +
> >
> > This has side effect of adding duplicate VG definitions; this may 
> > later
> > confuse grub. What about just checking array->driver for LVM? Go 
> > through
> > registered arrays, find disk match and check array driver. See
> > scan_disk_partition_iter () for example.
>
> Which is basically call grub_diskfilter_get_pv_from_disk() and check
> vg_out->driver.

This method also has the downside that is_lvm will fail without any 
indication as to why, when no volume group has been created. lvm_detect 
will return null because it can only return a VG. And, you cannot error 
out on that if a non-existing LVM should be okay.

So this call to lvm_detect is pretty annoying.

If it had more info, I could at least give an error in setup.c.

I am not sure whether installation is *possible* without a VG but it 
would be odd if it couldn't?

Anyway I would like a better error message on this because I just wasted 
2 hours trying to get GRUB_UTIL to work and failing. That means as per 
your method I can probably distinguish between PV present and VG present 
(or not present).

Then the only way I know how to report back is to give more values to 
is_lvm, but do we need a VG? Why error out on no VG? I will test.