Another fix for that is to switch the ZFS license to GPL before building:

Just put this line:

sed -i "s/CDDL/GPL/g" META

in the prepare() section of the PKGBUILD.

This will use the in-kernel FPU functions for cryptography and other things like raidz calculations that got EXPORT_SYMBOL_GPL'ed later if the original plumbing is still in the codebase. It's probably not offically recommend and I'm not sure how much these codepaths are tested anymore but I assume that quite a few people that don't need to redistribute anything but need the performance use a hack like this or something similiar. It works fine for me.

Personally I lost a ton of respect for the Linux kernel developers after introducing that kernel fpu EXPORT_SYMBOL_GPL thing and backporting it to all kernels and Linus badmouthing ZFS and somehow assuming Oracle is still involved in OpenZFS which is mostly FreeBSD and some sponsors spread out over the world but not related to Oracle.

This is a great read. It explains in detail the process of enabling debug symbols and using perf to root cause a performance issue.

I've never done this type of work, but probably will in the future now that I have an example to follow.

It is refreshing to see the author explain how and why all the pieces were gathered. Most guides/instructions/bug-reports only lists the end result, omitting information that helps the reader to get a better understanding and be better equipped to solve related issues.

On Arch, zfs-linux package has an exact version of linux as a dependency. linux package gets updated fairly often, and by the time zfs-linux package catches up, linux is already on a newer version and you are unable to upgrade.

Sometimes it takes quite a while before those two are in sync and you can upgrade your system.

I heard a rumor that ZFS native encryption is somewhat abandoned by its maintainer following it having landed in the release, that it was one person and they moved on/lost interest. I don’t know if this is true or not.

I’m still using ZFS on LUKS2 for now, given the issues like this. (You also can’t specify multiple keys in ZFS native encryption.)

Channelling Erik Dubious: “it’s ok, you see the error message, and you navigate to etsy.

This config we find is the answer. I lean into neovim this time, today, because it’s all a lessons. And then we type update, and it’s all okay!”

The thing he humbly leaves out is “why aren’t you an arcolinux fanboy yet!?”

Sadly there have seemed to be a number of rough edges that have stuck around with ZFS native encryption, which is unfortunate given that the cross platform and remote replication advantages are fundamentally very compelling. Honestly it's kind of curious that polishing that experience hasn't been more of a priority, for commercial interests like iXsystems if no one else. Speaking of which, for people running their ZFS off a NAS the idea of just being able to offload all the encryption and FS computation to that is a compelling part of the value equation beyond sheer storage in principle too. Though that in turn (if one does it with iSCSI) runs into another seemingly neglected area of ZFS (zvols). Nothing is perfect.

More details on this discussion: https://github.com/openzfs/zfs/issues/6824#issuecomment-1817... . Basically he is too busy to continue developing the encryption features but is able to review the related works.

There is also discussion about using ZFS on LUKS in the same thread: https://github.com/openzfs/zfs/issues/6824#issuecomment-1819...

ZFS on top of LUKS seems to have it's own issues though. :(

https://github.com/openzfs/zfs/issues/15533

See this discussion about the state of native encryption and the linked list of open bugs: https://discourse.practicalzfs.com/t/is-native-encryption-re...