Re: blktap2 implementation in NetBSD

[Roger Pau Monné <roger.pau%entel.upc.edu@localhost>]

> Problem with netbsd-iscsi-initiator is that you use vnd(4) to access the
> storage. I've never got the block/character device options to work (though
> haven't tried hard) as really you need all the devices for the partitions
> (e.g. bd0a, rbd0, bd0b, rbd0d, etc.) as the tools generally expect them to
> be there like they are in /dev.

I don't think we need all the partitions, since Xen only uses the 'd'
partition (whole disk), the *only* requirement is to create a block
device that contains the whole disk, there's no need to know what
partitions are inside, or if the image is actually not a disk image,
but only a single partition. Xen and your config file take care of
knowing if the attached device is a partition or a whole disk (by
adding xvda or xvda1 to your config file). Also, there's no need to
use vnd if the partition is attached to a guest, unless you want to
open it locally in your dom0, that's a whole different story. Also the
Linux blktap driver uses char devices, but I still haven't figured out
how or why.

>
> When using vnd(4) if the userland process restarts (such as reconnecting to
> an iSCSI target) it kills the whole filesystem layer requiring a forced
> reboot.
>
> I think you should really be looking at libvdisk in -current.

I've googled libvdisk and I'm not able to find anything, the path
where I think libvdisk should reside is empty:

http://cvsweb.netbsd.org/bsdweb.cgi/src/lib/libvdisk/

Can you provide some information about libvdisk please?

>
> --
> Stephen
>

I'm not an expert, so maybe some of my statements are wrong, it would
be of great help to get some opinions about the best way to do this.

Regards, Roger.