Bug 244838

Comment 1 Kyle Evans 2020-03-16 03:35:49 UTC

(In reply to Victor Sudakov from comment #0)

Indeed, bhyve support is a little more complicated. Currently, only the BIOS loaders can do the bits needed for activate -t -- notably, UEFI cannot and userboot cannot. I do not have a solid idea of whether it would be painful to add the necessary support to userboot.

UEFI support for -t is even more important IMHO. More and more installations (including hypervisors) abandon BIOS for UEFI. If UEFI could do it - I could run FreeBSD guests in bhyve in UEFI mode.

Yes, under UEFI "activate -t" does not work either.

Comment 4 Kyle Evans 2020-07-25 16:03:18 UTC

CC tsoome@, because I can't recall if his work covers userboot, too.

I have a similar issue but without bhyve. The system is running in a virtual machine (KVM under Arch GNU/Linux). I have the following BEs:

1. default ... releng/12.2 (upgraded from 12.0 -> releng/12.1 -> releng/12.2) 
2. fbsd13t ... HEAD @ r364739 (before the OpenZFS import/switch)
3. head ... HEAD @ main-c255645-g74bd20769706

In BE 1 and 2 'bectl activate -t' (and zfsbootcfg) works. In BE 3 i have to run bectl without "-t" to activate the BE. 

Can anyone confirm this? I can test with a clean install (latest HEAD snapshot) later.

Has this been addressed yet and does this bug occur with OpenZFS again?

1. Fresh AUTO-ZFS install from latest CURRENT ISO (FreeBSD-13.0-CURRENT-amd64-20210107-f2b794e1e90-255641-disc1.iso): 'bectl activate -t' works.

2. Fresh AUTO-ZFS install from latest STABLE ISO (FreeBSD-12.2-STABLE-amd64-20201231-r368787-disc1.iso): 'bectl activate -t' works. 

3. After upgrading STABLE VM (from 2.) to HEAD (main-c255827-gdcdad299479e) 'bectl activate -t' no longer works:

# bectl list
BE      Active Mountpoint Space Created
default NR     /          775M  2021-01-10 12:47
# bectl create test
# bectl activate -t test
Successfully activated boot environment test
for next boot
# bectl list
BE      Active Mountpoint Space Created
default NR     /          775M  2021-01-10 12:47
test    T      -          8K    2021-01-10 14:10
# shutdown -r now
# bectl list
BE      Active Mountpoint Space Created
default NR     /          775M  2021-01-10 12:47
test    -      -          308K  2021-01-10 14:10

'bectl activate' works:

# bectl activate test
Successfully activated boot environment test
# bectl list
BE      Active Mountpoint Space Created
default N      /          308K  2021-01-10 12:47
test    R      -          775M  2021-01-10 14:10
# shutdown -r now
# bectl list
BE      Active Mountpoint Space Created
default -      -          588K  2021-01-10 12:47
test    NR     /          775M  2021-01-10 14:10

Sorry, I forgot to add:

'zpool upgrade -a' and 'gpart bootcode -b /boot/pmbr -p /boot/gptzfsboot -i 1 vtbd0' would resolve the issue. But then of course the old 12.2-STABLE VM is no longer bootable:

Mounting from zfs:zroot/ROOT/default failed with error 45: retrying for 3 more seconds

Comment 8 Toomas Soome 2021-01-10 14:24:00 UTC

(In reply to Herbert J. Skuhra from comment #7)

Do not upgrade pool, just install gptzfsboot.

(In reply to Toomas Soome from comment #8)
Thanks. After only installing gptzfsboot from HEAD:

- Both BEs (12.2-STABLE and HEAD) are bootable 
- But now 'bectl activate -t' only works in the new BE (HEAD).

Comment 10 Toomas Soome 2021-01-10 18:06:08 UTC

(In reply to Herbert J. Skuhra from comment #9)

Yes, because we are using more flexible data structure (nvlist) in pool label with updated implementation. Unfortunately this does make jumping back and forth of old and new implementation a bit rocky, but can not really help about it.

Comment 11 Mark Linimon 2021-05-13 13:10:03 UTC

^Triage: assign to mailing list.