From owner-freebsd-hackers  Mon Jan  8 01:21:37 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id BAA24491
          for hackers-outgoing; Mon, 8 Jan 1996 01:21:37 -0800 (PST)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id BAA24446
          for <hackers@FreeBSD.ORG>; Mon, 8 Jan 1996 01:21:08 -0800 (PST)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id UAA19835; Mon, 8 Jan 1996 20:18:35 +1100
Date: Mon, 8 Jan 1996 20:18:35 +1100
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199601080918.UAA19835@godzilla.zeta.org.au>
To: msmith@atrad.adelaide.edu.au, phk@critter.tfs.com
Subject: Re: Add new slice to running system, comments?
Cc: hackers@FreeBSD.ORG
Sender: owner-hackers@FreeBSD.ORG
Precedence: bulk

>Poul-Henning Kamp stands accused of saying:
>> > This still means that swap operations have to go through the FAT filesystem
>> > code, which is slow and buggy.  I'm looking for a performance solution
>> > here, not a crumb to throw to people with space problems.
>> Well, the right solution is to fix the msdosfs to have a decent performance
>> in the cases needed and to bug davidg & dyson to implement swapping on
>> any random vnode...

Fix ufs to have decent performance for swapping on a vn device first :-).
(Perhaps it is already decent.)

>That still doesn't achieve what I want, which is for 'native' FreeBSD
>disk slices to be able to exist _inside_ a FAT filesystem.

>> More as an architectural principle...  It's really bde's code, so you'd
>> better ask him.

Slices are supposed to be disjoint.  There are enough complications for
overlapping partitions inside slices.  Since the windows swap file is in
a DOS slice, it seems best to create a label for this slice and describe
the swap file as the b partition in the label.  You will need a label
anyway since swapping only works on the b partition.

The DIOCSDINFO ioctl can be used to set an in-core label without writing
it to the disk.  You can experiment with this by removing the disk
write(s) from `disklabel -r'.

Bruce