lwext4's mkfs sets the csum_seed incompat bit (0x2000) but doesn't actually enable metadata_csum, and it seeds checksums from the UUID regardless of s_checksum_seed. We handle that fine on our side now — 0.1.1 stopped wrongly degrading these volumes to read-only. But I haven't checked what Linux makes of a volume newfs_fskit produced.
Someone should newfs_fskit -t ext4 a volume, then e2fsck -f it from a Linux box and confirm it comes back clean. If e2fsck complains about the seed / checksums, the right fix is probably to not set that incompat bit at mkfs time at all (an lwext4 patch), or to write a correct seed.
Filing this so the "plays nice with Linux" promise is actually verified for the format path, not just for mounting drives Linux already made.
lwext4's mkfs sets the
csum_seedincompat bit (0x2000) but doesn't actually enablemetadata_csum, and it seeds checksums from the UUID regardless ofs_checksum_seed. We handle that fine on our side now — 0.1.1 stopped wrongly degrading these volumes to read-only. But I haven't checked what Linux makes of a volumenewfs_fskitproduced.Someone should
newfs_fskit -t ext4a volume, thene2fsck -fit from a Linux box and confirm it comes back clean. If e2fsck complains about the seed / checksums, the right fix is probably to not set that incompat bit at mkfs time at all (an lwext4 patch), or to write a correct seed.Filing this so the "plays nice with Linux" promise is actually verified for the format path, not just for mounting drives Linux already made.