Re: [Bug-xorriso] [bug #46716] Protective MBR partition is not marked as bootable

[Alexander E. Patrakov]

19.12.2015 17:49, Andrei Borzenkov пишет:
> 19.12.2015 14:37, Alexander E. Patrakov пишет:
> > Some old BIOSes (including the one in Intel DG965SS desktop board) refuse to
> > treat a USB drive as bootable if it has no bootable MBR partitions. This
> > applies to all iso images that are created by grub-mkrescue and written to a
> > USB flash drive with dd.
>
> It is not clear - do you attempt to boot in legacy BIOS or EFI mode?
>
> > Changing the byte at offset 0x1be to 0x80 in the resulting iso makes it
> > bootable on such boards.
> >
> > Images produced by xorriso with isolinux as recommended at
> > http://www.syslinux.org/wiki/index.php/Isohybrid do have 0x80 at offset 0x1be,
> > so I think it is a bug in GRUB, not in xorriso.
>
> Command from syslinux wiki does not create protective MBR; it creates
> normal DOS partition of type 17 and marks it as bootable.

OK, I see.

> If you use grub-mkrescue to create ISO for a single i386-pc platform (or
> may be for any not including EFI, I have not tried it) it also creates
> single partition that is marked as bootable. Did you test if it works
> for you?

I use grub-mkrescue, and expect the resulting ISO to boot in the 
following scenarios:

1. From CD, on BIOS-based boards.
2. From CD, on UEFI-based boards.
3. From USB flash, on BIOS-based boards that don't check whether there 
are any bootable MBR partitions.
4. From USB flash, on BIOS-based boards that do check whether there are 
any bootable MBR partitions (such as DG965SS).
5. From USB flash, on UEFI boards that don't check that the partition 
with type 0xee is not bootable.
6. From USB flash, on UEFI boards that do check that the partition with 
type 0xee is not bootable (such as everything based on Tiano Core).

Currently, case 4 fails. With the modification proposed in 
https://savannah.gnu.org/bugs/?46716 , case 6 fails.

> Marking protective partition as bootable may trigger bugs on EFI, so I
> am not sure if it is reallu desirable.

You are right, because of Tiano Core. My point is, if we use a classical 
protective MBR that contains exactly one partition of type 0xee, then 
cases 4 and 6 are incompatible, and both cases exist in the wild.

So here is an alternative proposal. I was able to fix up the iso by 
creating a second partition of the same geometry, of type 0, and with 
bootable flag on. Here is now:

dd if=output.iso of=output.iso conv=notrunc bs=1 count=16 skip=446 seek=462
printf '\200' | dd of=output.iso conv=notrunc bs=1 seek=462
printf '\0' | dd of=output.iso conv=notrunc bs=1 seek=466

The resulting iso, unlike the original one, boots from USB drive on 
DG965SS (in BIOS mode), and also boots in KVM with OVMF (in UEFI mode).

Maybe other schemes with Hybrid MBR will also work. So far, I know of 
the following checks performed by various BIOSes and UEFIs:

1. The partition with type 0xee should start at sector 1 and be 
non-bootable, otherwise the GPT is treated as invalid.
2. There should be no overlap between partitions of type != 0.
3. There should be a bootable partition in the MBR.

> In any case MBR is created by xorriso, so I Cc it there.

Thanks!

--
Alexander E. Patrakov