[SOLVED] Installing on ZFS root: "ZFS: cannot find bootfs" on boot.

Question

I have been experimenting with ZFS filesystems on external HDDs for some time now to get more comfortable with using ZFS in the hopes of one day reinstalling my system on a ZFS root.
Today, I tried installing a system on an USB external HDD, as my first attempt to install on ZFS (I wanted to try in a safe, disposable environment before I try this on my main system).
My partition configuration (from gdisk):
Command (? for help): p
Disk /dev/sdb: 3907024896 sectors, 1.8 TiB
Logical sector size: 512 bytes
Disk identifier (GUID): 2FAE5B61-CCEF-4E1E-A81F-97C8406A07BB
Partition table holds up to 128 entries
First usable sector is 34, last usable sector is 3907024862
Partitions will be aligned on 8-sector boundaries
Total free space is 0 sectors (0 bytes)
Number Start (sector) End (sector) Size Code Name
1 34 2047 1007.0 KiB EF02 BIOS boot partition
2 2048 264191 128.0 MiB 8300 Linux filesystem
3 264192 3902828543 1.8 TiB BF00 Solaris root
4 3902828544 3907024862 2.0 GiB 8300 Linux filesystem
Partition #1 is for grub, obviously. Partition #2 is an ext2 partition that I mount on /boot in the new system. Partition #3 is where I make my ZFS pool.
Partition #4 is an ext4 filesystem containing another minimal Arch system for recovery and setup purposes. GRUB is installed on the other system on partition #4, not in the new ZFS system.
I let grub-mkconfig generate a config file from the system on partition #4 to boot that. Then, I manually edited the generated grub.cfg file to add this menu entry for my ZFS system:
menuentry 'ZFS BOOT' --class arch --class gnu-linux --class gnu --class os {
load_video
set gfxpayload=keep
insmod gzio
insmod part_gpt
insmod ext2
set root='hd0,gpt2'
echo 'Loading Linux core repo kernel ...'
linux /vmlinuz-linux zfs=bootfs zfs_force=1 rw quiet
echo 'Loading initial ramdisk ...'
initrd /initramfs-linux.img
My ZFS configuration:
# zpool list
NAME SIZE ALLOC FREE CAP DEDUP HEALTH ALTROOT
External2TB 1.81T 6.06G 1.81T 0% 1.00x ONLINE -
# zpool status :(
pool: External2TB
state: ONLINE
scan: none requested
config:
NAME STATE READ WRITE CKSUM
External2TB ONLINE 0 0 0
usb-WD_Elements_1048_575836314135334C32383131-0:0-part3 ONLINE 0 0 0
errors: No known data errors
# zpool get bootfs
NAME PROPERTY VALUE SOURCE
External2TB bootfs External2TB/ArchSystemMain local
# zfs list
NAME USED AVAIL REFER MOUNTPOINT
External2TB 14.6G 1.77T 30K none
External2TB/ArchSystemMain 293M 1.77T 293M /
External2TB/PacmanCache 5.77G 1.77T 5.77G /var/cache/pacman/pkg
External2TB/Swap 8.50G 1.78T 20K -
The reason for the above configuration is that after I get this system to work, I want to install a second system in the same zpool on a different dataset, and have them share a pacman cache.
GRUB "boots" successfully, in that it loads the kernel and the initramfs as expected from the 2nd GPT partition. The problem is that the kernel does not load the ZFS:
ERROR: device '' not found. Skipping fsck.
ZFS: Cannot find bootfs.
ERROR: Failed to mount the real root device.
Bailing out, you are on your own. Good luck.
and I am left in busybox in the initramfs.
What am I doing wrong?
Also, here is my /etc/fstab in the new system:
# External2TB/ArchSystemMain
#External2TB/ArchSystemMain / zfs rw,relatime,xattr 0 0
# External2TB/PacmanCache
#External2TB/PacmanCache /var/cache/pacman/pkg zfs rw,relatime,xattr 0 0
UUID=8b7639e2-c858-4ff6-b1d4-7db9a393578f /boot ext4 rw,relatime 0 2
UUID=7a37363e-9adf-4b4c-adfc-621402456c55 none swap defaults 0 0
I also tried to boot using "zfs=External2TB/ArchSystemMain" in the kernel options, since that was the more logical way to approach my intention of having multiple systems on different datasets. It would allow me to simply create separate grub menu entries for each, with different boot datasets in the kernel parameters. I also tried setting the mount points to "legacy" and uncommenting the zfs entries in my fstab above. That didn't work either and produced the same results, and that was why I decided to try to use "bootfs" (and maybe have a script for switching between the systems by changing the ZFS bootfs and mountpoints before reboot, reusing the same grub menuentry).
Thanks in advance for any help.
Last edited by tajjada (2013-12-30 20:03:09)

Answer 1

Sounds like a zpool.cache issue. I'm guessing your zpool.cache inside your arch-chroot is not up to date. So on boot the ZFS hook cannot find the bootfs. At least, that's what I assume the issue is, because of this line:
ERROR: device '' not found. Skipping fsck.
If your zpool.cache was populated, it would spit out something other than an empty string.
Some assumptions:
- You're using the ZFS packages provided by demizer (repository or AUR).
- You're using the Arch Live ISO or some version of it.
On cursory glance your configuration looks good. But verify anyway. Here are the steps you should follow to make sure your zpool.cache is correct and up to date:
Outside arch-chroot:
- Import pools (not using '-R') and verify the mountpoints.
- Make a copy of the /etc/zfs/zpool.cache before you export any pools. Again, make a copy of the /etc/zfs/zpool.cache before you export any pools. The reason for this is once you export a pool the /etc/zfs/zpool.cache gets updated and removes any reference to the exported pool. This is likely the cause of your issue, as you would have an empty zpool.cache.
- Import the pool containing your root filesystem using the '-R' flag, and mount /boot within.
- Make sure to copy your updated zpool.cache to your arch-chroot environment.
Inside arch-chroot:
- Make sure your bootloader is configured properly (i.e. read 'mkinitcpio -H zfs').
- Use the 'udev' hook and not the 'systemd' one in your mkinitcpio.conf. The zfs-utils package does not have a ported hook (as of 0.6.2_3.12.6-1).
- Update your initramfs.
Outside arch-chroot:
- Unmount filesystems.
- Export pools.
- Reboot.
Inside new system:
- Make sure to update the hostid then rebuild your initramfs. Then you can drop the 'zfs_force=1'.
Good luck. I enjoy root on ZFS myself. However, I wouldn't recommend swap on ZFS. Despite what the ZoL tracker says, I still ran into deadlocks on occasion (as of a month ago). However, I cannot say definitely the cause of the issue; but it resolved when I moved swap off ZFS to a dedicated partition.
Last edited by NVS (2013-12-29 14:56:44)