Root dataset (or any other dataset) on new zpool smaller than it should be

[indigophox]

The following is so far inexplicable. The numbers add up just fine in the first case, but even with (power of 2)+2 disks in the second case, a bunch of space has gone AWOL:

root@medusa:# zpool create -o ashift=12 medusa0 raidz2 $(ls /dev/disk/by-id/scsi-SATA_ST3000DM001-9YN_???????? | head -n 6)
root@medusa:# zpool list
NAME SIZE ALLOC FREE CAP DEDUP HEALTH ALTROOT
medusa0 16.2T 1.58M 16.2T 0% 1.00x ONLINE -
root@medusa:~# zfs list
NAME USED AVAIL REFER MOUNTPOINT
medusa0 919K 10.7T 272K /medusa0

(4/6) * 16.2T = 10.8T

root@medusa:# zpool create -o ashift=12 medusa0 raidz2 $(ls /dev/disk/by-id/scsi-SATA_ST3000DM001-9YN_???????? | head -n 10)
root@medusa:# zpool list
NAME SIZE ALLOC FREE CAP DEDUP HEALTH ALTROOT
medusa0 27.2T 1.35M 27.2T 0% 1.00x ONLINE -
root@medusa:~# zfs list
NAME USED AVAIL REFER MOUNTPOINT
medusa0 1.03M 20.4T 311K /medusa0

(8/10) * 27.2T = 21.76T

root@medusa:# zpool create -o ashift=12 medusa0 raidz2 $(ls /dev/disk/by-id/scsi-SATA_ST3000DM001-9YN_???????? | head -n 12)
root@medusa:# zpool list
NAME SIZE ALLOC FREE CAP DEDUP HEALTH ALTROOT
medusa0 32.5T 1.58M 32.5T 0% 1.00x ONLINE -
root@medusa:~# zfs list
NAME USED AVAIL REFER MOUNTPOINT
medusa0 1.03M 24.4T 311K /medusa0

(10/12) * 32.5T = 27T

[indigophox]

So, here are the ratios of actual available space on ZFS vs "ideal" (pool "size" * (non-parity disks / total disks)), multiplied by 64/63 to account for metadata block reservation, each for 3..12 disks:

raidz1:
1.0002346 0.9667880 0.9990663 0.9631981 0.9793372 0.9586220 0.9982507 0.9876543 0.9759153 0.9684182

raidz2:
0.9983580 0.9674093 0.9884843 1.0064668 0.9356725 0.9444202 0.9809246 0.9523810 0.9312169 0.9152234