It's possible to run NetBSD/xen in the DomUs of many hosting providers
which disavow any NetBSD support. I've added mount_ext2fs and
newfs_ext2fs to my installation ramdisk to create and mount the
disklabel slice where pv-grub looks to find its configuration file and
kernels. Yes, pv-grub does have some ffs support, but it doesn't seem
to able to cope with filesystems created with -O2 (I was going to
write UFS2, but the whole ffs2 vs ufs2 nomenclature and taxonomy
confuses me).
If they are added, it seems i386 and amd64 should both get them.
How much more space is needed?
It sounds like {mount,newfs}_ext2fs are needed to set up a small
partition to store the pvgrub config/kernel, and then you are using ffs
for NetBSD to use as a real partition.
Did you consider just using ext2fs for the entire netbsd disk? As a
long-time BSD user, that feels odd, but I wonder if people have
opinions/data on how well it works. I've had the impression ext2fs had
consistency issues with unclean shutdown, but also that those were fixed
in Linux, and of course our code is different.
As for ffs/ufs, it is confusing, but basically "-O 2" leads to
UFS2/FFSv2. FFSv1 can have various sublevels; see -c option to
fsck_ffs. It is common to use -c 5 which leads to ufs2-style
superblocks in a ffsv1 layout.
dumpfs is helpful; it will show you the fs-level magic number and the
superblock format, and a fs level.
i think 'newfs -O 1' leads to the same as fsck-ffs -c 4, but I'm not
sure.
You said pvgrub can read ffsv1. Do you know if that's with ufs2
superblocks, or the older superblocks, and what the version of pvgrub
is?
I don't know of any compelling reasons to use ufs2 vs ffsv1 for smallish
filesystems (where smallish is certainly up to tens of GB).
I tend to use ffsv1 (with v2 superblocks) for root filesystems, and ufs2
for the bigger data ones.
Attachment:
pgplknTeENy9C.pgp
Description: PGP signature