grub2 disk error with kvm

grub2 disk error with kvm

[Daniel Dehennin]

[-- Attachment #1: Type: text/plain, Size: 1724 bytes --]

Hello,

I post a bug report here as asked by nyu on #grub.

=== IRC LOG ===
[13:32] <nyu> nebuchadnezzar: could you report the grub one first?  the problem
    is grub_biosdisk_get_diskinfo_standard() fails, but this call wasn't used
    when probing for drives (and I think it should be)
=== IRC LOG ===

I have a kvm machine featuring debian lenny (now upgraded to sid).
It has only one disk configured, with one partition as LVM so I
install grub2 1.96+20080724-7.

When booting I have the following error:

===
error: unknown drive hd15
Entering rescue mode...
===

This does not come from the configuration, I renamed grub.cfg and the
same happens.

Nyu ask me to remove the grub_error call line 53:
grub_error (GRUB_ERR_BAD_DEVICE, "unknown device %s", name);

Then it gives me "error: no such disk".

In rescue mode:
- I can insmod normal and enter in normal mode to boot my system.
- ls gives me (lenny-var) (lenny-usr) (lenny-root) (lenny-boot) (hd0) (hd0,1) (hd1) (hd2)...(hd15)

The problem happens only when using the kvm -drive option:
sudo /usr/bin/kvm -M pc -m 256 -smp 1 -name lenny -monitor stdio \
    -boot c -drive file=/dev/hati/lenny,if=virtio,index=0,boot=on \
    -net nic,model=virtio,vlan=0 -net tap,vlan=0,ifname=vnet0 \
    -serial pty -parallel none -k fr

Replacing this by -hda makes it work:

sudo /usr/bin/kvm -M pc -m 256 -smp 1 -name lenny -monitor stdio \
    -boot c -hda dev/hati/lenny \
    -net nic,model=virtio,vlan=0 -net tap,vlan=0,ifname=vnet0 \
    -serial pty -parallel none -k fr

I provide the grub.cfg as attachement, I can provide lspci, dmesg or
what ever you need.

Regards.
-- 
Daniel Dehennin
Récupérer ma clef GPG:
gpg --keyserver pgp.mit.edu --recv-keys 0x6A2540D1


[-- Attachment #2: grub.cfg --]
[-- Type: text/plain, Size: 1385 bytes --]

#
# DO NOT EDIT THIS FILE
#
# It is automatically generated by /usr/sbin/update-grub using templates
# from /etc/grub.d and settings from /etc/default/grub
#

### BEGIN /etc/grub.d/00_header ###
set default=0
set timeout=5
insmod lvm
set root=(lenny-usr)
search --fs-uuid --set b67ea215-b6ec-40cc-bfe8-56e1284a42c8
if font /share/grub/ascii.pff ; then
  set gfxmode=640x480
  insmod gfxterm
  insmod vbe
  terminal gfxterm
fi
### END /etc/grub.d/00_header ###

### BEGIN /etc/grub.d/05_debian_theme ###
set menu_color_normal=cyan/blue
set menu_color_highlight=white/blue
### END /etc/grub.d/05_debian_theme ###

### BEGIN /etc/grub.d/10_hurd ###
### END /etc/grub.d/10_hurd ###

### BEGIN /etc/grub.d/10_linux ###
insmod lvm
set root=(lenny-boot)
search --fs-uuid --set 5b03d02f-e69e-418d-8f03-b0eb4e3185de
menuentry "Debian GNU/Linux, linux 2.6.25-2-amd64" {
	linux	/vmlinuz-2.6.25-2-amd64 root=/dev/mapper/lenny-root ro  vga=0x318
	initrd	/initrd.img-2.6.25-2-amd64
}
menuentry "Debian GNU/Linux, linux 2.6.25-2-amd64 (single-user mode)" {
	linux	/vmlinuz-2.6.25-2-amd64 root=/dev/mapper/lenny-root ro single 
	initrd	/initrd.img-2.6.25-2-amd64
}
### END /etc/grub.d/10_linux ###

### BEGIN /etc/grub.d/30_os-prober ###
### END /etc/grub.d/30_os-prober ###

### BEGIN /etc/grub.d/40_custom ###
# This file is an example on how to add custom entries
### END /etc/grub.d/40_custom ###

Re: grub2 disk error with kvm

[Robert Millan]

On Mon, Aug 11, 2008 at 02:08:58PM +0200, Daniel Dehennin wrote:
> Hello,
> 
> I post a bug report here as asked by nyu on #grub.
> 
> === IRC LOG ===
> [13:32] <nyu> nebuchadnezzar: could you report the grub one first?  the problem
>     is grub_biosdisk_get_diskinfo_standard() fails, but this call wasn't used
>     when probing for drives (and I think it should be)
> === IRC LOG ===
> 
> I have a kvm machine featuring debian lenny (now upgraded to sid).
> It has only one disk configured, with one partition as LVM so I
> install grub2 1.96+20080724-7.
> 
> When booting I have the following error:
> 
> ===
> error: unknown drive hd15
> Entering rescue mode...
> ===

Hi,

Summary of the problem, based on what Daniel told me:

  - grub_biosdisk_iterate() probes for hard drives by trying to read their first
    sector with grub_biosdisk_rw_standard().  In Daniel's setup, it finds 16
    hard disks (the maximum), although only one exists.

  - raid.mod (or anything else) sees the hard disks and tries to use them.

  - grub_biosdisk_open() fails with "cannot get C/H/S values" because
    grub_biosdisk_get_diskinfo_standard() is not usable.

A possible solution would be to allow grub_biosdisk_get_diskinfo_standard()
to fail as long as we have total_sectors, and let CHS values be set to 0
(AFAIK, as long as LBA works they're not essential).  This wouldn't make the
fake drives disappear (it's a KVM bug after all), but would prevent GRUB from
aborting when this happens [1].

Another one would be to integrate grub_biosdisk_get_diskinfo_standard() as
part of the probing routine in grub_biosdisk_iterate(), thus preventing the
disks from appearing at all.

Or another one would be to do nothing, since it will be fixed in KVM sooner
or later [2].  However, I think it's good if we can make GRUB behave more
robustly, just in case we find similar bugs in propietary BIOSes.

[1] Then again, I think the last RAID patch from Felix would archieve the
    same effect.  No reason we can't have both things though.

[2] Daniel, feel free to report this.  You can point to this mail for their
    reference.

-- 
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."