Installing grub manually for a VM image

PeterSteele · 07-11-2014, 05:39 PM

I am preparing a bootable image manually (via a script) for use with virt-install and the --import option. The bootable image has three main partitions, the first which will be mounted as /boot on the VM, the second as swap, and the third as / (root).

I've got the image populated with the files I need and all I have left is to install grub so that the image will be bootable. I've done some searches and I'm not entirely sure what steps are needed. Let's say my bootable image is test.img. I'm running kpartx to give me access to the partitions in my image:

# kpartx -av test.img
add map loop0p1 (253:0): 0 249967 linear /dev/loop0 34
add map loop0p2 (253:1): 0 8388608 linear /dev/loop0 251904
add map loop0p3 (253:2): 0 91357184 linear /dev/loop0 8640512

I need to get grub installed correctly into the /boot partition. I can access it after doing the kpartx command via the /dev/mapper/loop0p1 device (the real block device is actually /dev/dm-0; loop0p1 is a symbolic link). I'm just not sure which grub commands I need to make this a proper bootable image. Can anyone enlighten me as to the procedure I need to follow?

PeterSteele · 07-11-2014, 08:08 PM

I should add that I'm using CentOS 6.4/6.5, and the end result needs to be a process that can be automated.

ReaperX7 · 07-12-2014, 08:59 PM

Are you using Grub-1.x or Grub-2.x?

PeterSteele · 07-12-2014, 10:46 PM

Neither. The reports itself as being 0.97:

# grub --version
grub (GNU GRUB 0.97)

These are the basic steps I'm following:

- Step 1: Create image file that will host the VM
# dd if=/dev/zero of=test.img count=0 bs=1 seek=50GB

- Step 2: Partition the image as needed (boot, swap, root)
# parted -s test.img mklabel gpt
# parted -s test.img mkpart ext4 0 128M
# parted -s test.img mkpart swap 129M 4400MB
# parted -s test.img mkpart ext4 4400MB 100%

- Step 3: Format partitions
# kpartx -av test.img
# mkfs.ext4 /dev/dm-0
# mkfs.ext4 /dev/dm-2

- Step 4: Populate image with custom CentOS template
# mkdir -p /mnt/sysimage
# mount -o loop /dev/dm-2 /mnt/sysimage
# mkdir -o /mnt/sysimage/boot
# mount -o loop /dev/dm-0 /mnt/sysimage/boot
# tar -zxpf custom-img.tar.gz -C /mnt/sysimage

At this point I have everything completed except for the final step of creating grub.conf and preparing the image so that it will boot when launched as a VM. Assuming I get grub setup properly, I'll need to run virt-install as the final step:

# virt-install --blah --blah --import --disk path=test.img

I just can't quite figure out the grub magic to put a bow on this procedure. I've read lots of examples but nothing that works for my specific case.

Note: I have a separate procedure that I use to create my custom CentOS VM.

yancek · 07-12-2014, 11:07 PM

Quote:

grub --version
grub (GNU GRUB 0.97)

That means it still uses Grub Legacy. According to the Archlinux site below, neither Grub Legacy nor Lilo support GPT which you are using.

https://wiki.archlinux.org/index.php...artition_Table

Quote:

GRUB Legacy and LILO do not support GPT.

PeterSteele · 07-12-2014, 11:32 PM

The CentOS grub support gpt just fine. If I use the CentOS kickstart facility to duplicate the steps I'm attempting to do manually, the end result is a bootable image that is configured with gpt partitions:

Code:

# parted -s test.img print
Model: Unknown (unknown)
Disk test.img: 51.2GB
Sector size (logical/physical): 512B/4096B
Partition Table: gpt

Number  Start   End     Size    File system     Name     Flags
 1      17.4kB  128MB   128MB   ext4            primary  boot
 2      129MB   4424MB  4295MB  linux-swap(v1)
 3      4424MB  51.2GB  46.8GB  ext4

So I know I can accomplish what I need to do since CentOS itself does it during an OS install. I can duplicate all of the steps except for what CentOS does to make the image bootable and the specific grub commands it runs. Pages like this describe grub well but I cannot workout what's needed in my more specialized case.

PeterSteele · 07-13-2014, 12:55 AM

For what its worth, the kickstart script I use to accomplish the layout I need uses a %pre section with these commands:

Code:

parted -s /dev/vda mklabel gpt
parted -s /dev/vda mkpart primary ext4 0 128
mkfs.ext4 /dev/vda1

and the main kickstart header section specifies these options:

Code:

bootloader --location=mbr --append="console=tty0 console=ttyS0,115200 selinux=0"
part /boot --onpart=vda1 
part swap  --ondrive=vda --asprimary --fstype swap --size=4096
part /     --ondrive=vda --asprimary --fstype=ext4 --size=500 --grow

The bootloader option ends up performing the exact grub install I am trying to reverse engineer.

yancek · 07-13-2014, 07:47 AM

I just found a link to the CentOS forums confirming what you already knew, that CentOS 6 and later supports GPT. Just noticed that my ArchLinux site had no date on it so don't know if it was written last week or ten years ago. I'm afraid I can't help you because I've not used Red Hat or derivatives much a little with Fedora and never had much success even trying to do a remaster.

PeterSteele · 07-13-2014, 10:11 AM

That's quite alright. This specialized stuff and will need advice from the CentOS experts...

PeterSteele · 07-13-2014, 10:27 AM

In looking at the logs that are creating by anaconda during a kickstart controlled VM install, the commands it appears to perform are

- Switch to perspective of VM image just created
# chroot /mnt/sysimage

- Do first step of grub install
# grub-install --just-copy

- Then complete grub install using
# grub --batch --no-floppy --device-map=/boot/grub/device.map
grub> root (hd0,0)
grub> install --stage2=/boot/grub/stage2 /grub/stage1 d (hd0) /grub/stage2 p (hd0,0)/grub/grub.conf

When I try the same sequence of commands after chrooting to my own prepared image, it fails on the "root (hd0,0)" command saying that the selected disk does not exist. There's obviously another step that's not obvious from the logs that makes hd0 visible to Anaconda's chroot environment that I'm missing.

PeterSteele · 07-13-2014, 12:45 PM

I've got the solution. Before launching my chroot session, I need to setup hard links representing the devices in my image:

ln /dev/loop0 /dev/vda
ln /dev/dm-0 /dev/vda1
ln /dev/dm-2 /dev/vda3

and create device.map to map (hd0) to /dev/vda. After doing this the grub commands all worked and I was able to create and boot a VM using the manually prepared image.