I'm not responsible if you ruin your system, this guide functions as documentation for future me. Remember to back up your data.
Why do this instead of luks
I wanted to remove a layer from the File to Disk layering, before it was ZFS -> LUKS -> disk, now it's ZFS -> disk.
I just got a new laptop and wanted to just migrate the data, luckily the old laptop was using ZFS as well, so the data could be sent/received though native ZFS means.
The actual setup
Set up your root pool with the encryption key, it will be inherited by all child datasets, no child datasets will be allowed to be unencrypted.
In my case the pool name was
slaanesh-zp00, so I ran the following to create the fresh pool.
zpool create -O encryption=on -O keyformat=passphrase zfstest /dev/zvol/slaanesh-zp00/zfstest
After that just go on and create your datasets as normal, transfer old data as needed (it'll be encrypted as it's written). See /wiki.gentoo.org/wiki/ZFS for a good general guide on setting up your datasets.
decrypting at boot
If you are using dracut it should just work. No changes to what you pass on the kernel command line are needed. The code is upstream in /github.com/zfsonlinux/zfs/blob/master/contrib/dracut/90zfs/zfs-load-key.sh.in
Make sure you install from git master, there was a disk format change for encrypted datasets that just went in a week or so ago.