Re: Regression parsing GPT (EFI) partition tables

[Davidlohr Bueso]

On Thu, 2013-10-10 at 15:29 -0700, Doug Anderson wrote:
> Hi,
[...]
> > Do you happen to know what tool was used to create the GPT partition(s)?
> > I ask because it is up to the partitioning program to set sz correctly,
> > and most if not all set it to either the disk size - 1 or 0xFFFFFFFF.
> 
> See Bill's email.  It was created using "cgpt" to create a binary
> file, which was then "dd"ed onto removable media.
> 
> 
> FWIW, I can "fix" my problems also by tweaking my image:
> 
> setword.py /.../chromiumos_test_image.bin 0x1ca 0xffffffff
> 0x000001ca: 0x004b9fbf => 0xffffffff
> 
> Now things boot up nicely.  :)  Similarly I can get things to boot by running:

Right, now sz == 0xffffffff so you're "using" the whole disk, but you
probably want 0xecdfff since the disk is smaller than that.

> 
> cgpt boot -p /dev/mmcblk1
> 
> ...which appears to put the right size into this field.
> 
> 
> >> ...so basically it looks like we're now considering something an error
> >> that used to be considered a warning.
> >
> > It simply wasn't checked before.
> >
> > It's worthwhile finding out if this scenario also occurs without the
> > recent GPT changes. If so, then we could go ahead and just use a warning
> > and not consider it an error for the sake of booting. Otherwise there's
> > something wrong with c2ebdc2 and replacing 'struct partition' with
> > 'struct _gpt_mbr_record' caused some sort of structure offset problem
> > and some fields are holding the wrong data.
> >
> > To check this, you can revert all patches, starting from c2ebdc2, then
> > add the following (untested) to pmbr_part_valid():
> >
> > diff --git a/block/partitions/efi.c b/block/partitions/efi.c
> > index c85fc89..7144d8a 100644
> > --- a/block/partitions/efi.c
> > +++ b/block/partitions/efi.c
> > @@ -152,6 +152,8 @@ static u64 last_lba(struct block_device *bdev)
> >  static inline int
> >  pmbr_part_valid(struct partition *part)
> >  {
> > +       printk("part->nr_sects = %d\n", le32_to_cpu(part->nr_sects));
> > +
> >          if (part->sys_ind == EFI_PMBR_OSTYPE_EFI_GPT &&
> >              le32_to_cpu(part->start_sect) == 1UL)
> >                  return 1;
> >
> > Thanks,
> > Davidlohr
> 
> I can run this test if you need, but it sounds like we're tracked it
> down pretty well.
> 
> 
> My dumb summary of this without digging and understanding everything is:
> 
> * It used to be "OK" if the sz was wrong.

Yes, because it was never checked.

> 
> * If you're "dd"ing an image from a smaller device to a bigger device,
> the "sz" will likely be wrong.  It would be nice if this were a
> warning not an error since this can be a useful thing to do.
> 
> * We could fix our tool to not specify "sz" (aka use "-1") when
> creating our images and it would work.

You should use force_gpt and forget about MBR all together.

> 
> * This is not exactly the same as the GPT/alternate GPT error, since
> it's a different header field.
> 
> 
> Does that sound reasonable?
> 

Since it used to work before, I blame my change and it's a regression,
I'll happily ack the "warn instead of not recognizing approach".

Thanks,
Davidlohr


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/