freebsd stable - Oct 2017 - System fails to boot with zfs root on nvme mirror when not using full disks

Hello list,

I have a particularly odd problem, where I can't figure out what's going
on, or whether I'm possibly doing something
stupid...

Short summary:
- Supermicro X10DRI-T motherboard, using UEFI boot, 128G RAM, 2 CPUs
- 2 Intel DC P3700 NVME cards, to be used as mirrored zfs root and mirrored log
devices for a data pool
- FreeBSD-11.1 release installs fine off memstick, and the built system boots
correctly, but this system uses the entire
  disks (so I need to shrink the system to make room for log partitions)
- so I then rebooted into usb live system, and did a gpart backup of the nvme
drives (see later). zfs snapshot -r, zfs
  send -R > backup, gpart delete last index and recreate with shorter size on
both drives, recreate zroot pool with
  correct ashift, restore with zfs receive from backup, set bootfs, reboot
- the rebooting system bootloader finds the zroot pool correctly, and proceeds
to load the kernel. However, when it's
  supposed to mount the root filesystem, I get:
    Trying to mount root from zfs:zroot/ROOT/default []...
    Mounting from zfs:zroot/ROOT/default failed with error 6.
- when I list the available boot devices, all partitions of the nvme disks are
listed
- I can import this pool without any issues with the usb live system, there are
no errors on import.
- I then redid the whole exercise, but restored the previously backed up
partition tables (full size zfs
  partition), did exactly the same steps as described above to restore the
previous zfs filesystem, rebooted, and the
  system started up normally.

So, my impression is: the kernel is doing something odd when trying to import
the zroot pool, if that pool isn't
using the whole physical disk space. Is there some way to enable debugging for
the "mount a root zfs pool" process of
the kernel, so see where this is failing?

Here's a more detailled history of my commands to reduce the pool size:

{I'm on a data pool directory here, where I can store stuff. I've booted
from usb stick into live system}
zpool import -Nf zroot
zfs snapshot -r zroot at backup
zfs send -R zroot at backup > zroot.zfs
zpool destroy zroot
gpart delete -i 3 nvd0
gpart delete -i 3 nvd1
gpart add -a 4k -s 50G -t freebsd-zfs -l zfs0 nvd0
gpart add -a 4k -s 50G -t freebsd-zfs -l zfs1 nvd1
sysctl vfs.zfs.min_auto_ashift=12
zpool create -f -o altroot=/mnt -o cachefile=/var/tmp/zpool.cache zroot mirror
/dev/gpt/zfs0 /dev/gpt/zfs1
zfs receive -Fud zroot < zroot.zfs
zpool set bootfs=zroot/ROOT/default zroot
reboot

Also, here's a diff between full and partial gpart backup:

--- nvd0.gpart  2017-10-20 14:04:26.583846000 +0200
+++ nvd0.gpart.new      2017-10-20 15:39:20.184445000 +0200
@@ -1,4 +1,4 @@
 GPT 152
 1            efi        40    409600 efiboot0 
 2   freebsd-swap    411648  33554432 swap0 
-3    freebsd-zfs  33966080 747456512 zfs0 
+3    freebsd-zfs  33966080 104857600 zfs0 


I'm really at a loss to see the reason why this is failing. I can provide
detailled dmesg output of the booting system,
and screenshots of the kvm-efi-console of the failing boot. 

Thanks for any help:)

Cheers,
Markus

On Fri, Oct 20, 2017 at 05:13:45PM +0200, Markus Wild
wrote:
> Hello list,
> 
> I have a particularly odd problem, where I can't figure out what's
going on, or whether I'm possibly doing something
> stupid...
> 
> Short summary:
> - Supermicro X10DRI-T motherboard, using UEFI boot, 128G RAM, 2 CPUs
> - 2 Intel DC P3700 NVME cards, to be used as mirrored zfs root and mirrored
log devices for a data pool
> - FreeBSD-11.1 release installs fine off memstick, and the built system
boots correctly, but this system uses the entire
>   disks (so I need to shrink the system to make room for log partitions)
> - so I then rebooted into usb live system, and did a gpart backup of the
nvme drives (see later). zfs snapshot -r, zfs
>   send -R > backup, gpart delete last index and recreate with shorter
size on both drives, recreate zroot pool with
>   correct ashift, restore with zfs receive from backup, set bootfs, reboot
> - the rebooting system bootloader finds the zroot pool correctly, and
proceeds to load the kernel. However, when it's
>   supposed to mount the root filesystem, I get:
>     Trying to mount root from zfs:zroot/ROOT/default []...
>     Mounting from zfs:zroot/ROOT/default failed with error 6.
> - when I list the available boot devices, all partitions of the nvme disks
are listed
My suspicion is that you didn't make a backup of the new
/boot/zfs/zpool.cache
>From the USB live booted system, when you recreated the zpool you
likely changed parameters that live in the cache.  You need to copy that
to the new boot pool or ZFS can get angry

Just a suspicion

Regards,

Gary