"Nathan J. Williams" <nathanw@wasabisystems.com> writes:
> Frank van der Linden <fvdl@NetBSD.org> writes:
>> On Tue, Jan 18, 2005 at 01:30:52PM -0500, Perry E. Metzger wrote:
>> > Yes, but as I recall, it gets the files information for the CD off of
>> > the directory it slurps in, rather than from a metadata file. (This
>> > may be unimportant in practice, I'm not sure.)
>> 
>> Hmm.. I see that. But then again, are permissions on a CD image important?
>
> Not usually. The most common mkisofs invocation for this purpose is
> -r, which forces uid and gid to 0, sets all read bits, sets all
> execute bits if any execute bits are set, and clears all
> sticky/sgid/suid bits.
>
> A "live" CD would have different needs, but that doesn't seem to be
> what we're talking about here.

Two more points, just to muddy the water more.

First, I'm not entirely sure that the old SSTO install mechanism we've
used on x86 for some years is as useful any more. It was really
designed for the world where you were going to boot off of floppy and
would need several to hold the image. Doing root on an embedded MFS
made a lot of sense in that context -- it may make less sense
now, especially on a platform like x86-64 that doesn't have legacy
machines with microscopic amounts of memory and floppy drives. Perhaps
root on isofs would make the most sense -- or perhaps it is a really
bad move. It *would* give you a lot more flexibility during install
though.

Second, these days my usual way of installing boxes isn't CDs but
disk-on-key USB frobs. We might want to distribute a disk-on-keyable
image as well. (I've discovered, btw, that the key to making USB flash
drives boot is putting a proper MBR on them...)

Perry