ZFS Problem with DirtyFrag Fix Kernel

[utkonos]

I am unable to boot from the new kernel 6.17.13-7. Any help finding a workaround is greatly appreciated.

I have performed the kernel update to fix the DirtyFrag vulnerability, and there is a ZFS incompatibility with the new kernel. This instance was running 6.17.13-6 and the only updates that were performed were:

proxmox-kernel-6.17: 6.17.13-6 -> 6.17.13-7


After the update, boot was unable to finish due to a ZFS feature not supported:
com.truenas:block_cloning_endian (Fixes BRT ZAP endianness on new pools.)



I am able to boot fine by choosing 6.17.13-6 at GRUB. Looking at the settings, I see the following:

# zpool get com.truenas:block_cloning_endian rpool
NAME   PROPERTY                          VALUE                             SOURCE
rpool  com.truenas:block_cloning_endian  -                                 default

# zpool get all rpool | grep block_cloning_endian
rpool  feature@block_cloning_endian   active                         local

I am not able to set it to disabled, however:

# zpool set feature@block_cloning=disabled rpool
cannot set property for 'rpool': property 'feature@block_cloning' can only be set to 'disabled' at creation time

 

[Kingneutron]

Look into zfsbootmenu, might allow you to boot from usb stick / SD card media

https://docs.zfsbootmenu.org/en/latest/general/portable.html

 

[utkonos]

Look into zfsbootmenu, might allow you to boot from usb stick / SD card media

https://docs.zfsbootmenu.org/en/latest/general/portable.html

I am able to boot from the previous kernel as I noted above. Is there a workaround for booting from the new kernel with the security fix?

 

[mystica555]

I am able to boot from the previous kernel as I noted above. Is there a workaround for booting from the new kernel with the security fix?

If you aren't averse to kernel 7, 7.0.2-2-pve works fine with zfs for me.

 

[utkonos]

If you aren't averse to kernel 7, 7.0.2-2-pve works fine with zfs for me.

Thanks, but is there a bug in the 6.17.13-7 kernel? The zfs feature was supported in 6.17.13-6.

 

[uzumo]

Apart from commit e057618, it doesn't appear that there have been any changes from the previous version...

https://github.com/proxmox/pve-kernel/commits/trixie-6.17/

 

[utkonos]

Apart from commit e057618, it doesn't appear that there have been any changes from the previous version...

https://github.com/proxmox/pve-kernel/commits/trixie-6.17/

Thanks! I think what happened is the ABI bump triggered a rebuild of the ZFS DKMS module and that is a potential root cause of losing block_cloning_endian support. This is a testable hypothesis. I will need to run both the before and after through binary ninja and then perform a bindiff. That may show the precise change that happened.