[OpenAFS-devel] device nodes...

[Derek Atkins]

Love <lha@stacken.kth.se> writes:

> Jeffrey Hutzelman <jhutz@cmu.edu> writes:
> 
> > In order to come close to being safe, I would propose something like the
> > following, if device nodes are to be supported at all...
> > 
> > Device nodes are represented in AFS as magic symlinks, in the same way
> > as mount points.  
> > [...]
> 
> When do you need device-nodes in AFS space ? I would guess that you'll need

I need them when, for example, I'm building an initrd for the MIT/SIPB
RedHat-Athena installation diskette.  We download the initrd file tree
(which includes /dev and a bunch of device inodes) into AFS, make some
changes to it, and then build new disk images.  I don't want to use a
local disk, because I'd like to be able to build Boot Disks from any
number of machines that we have (MIT has a distributed computing
environment -- there is no guarantee that I'll get the same machine
twice, and even then machines are considered 'dataless' and can be
reinstalled at any time.  Any long-term storage must be done in AFS).

Basically, I'd like to keep the whole initrd (expanded) in AFS space
and then cpio it into a loopback'd file in order to build a disk
image.  Unfortunately I cannot do this because I cannot store the
device nodes in AFS, so every time I upgrade and download a new source
tree from RedHat I have to search for all the device inodes and then
write them down so that after I cpio the files into the disk image I
can 'mknod' each device.  This is a major PitA.  I'd much rather be
able to 'store' the device inodes in AFS in the first place, with the
rest of the Red Hat initrd.

> Use a memfs/ramfs/tmpfs/nullfs/devfs instead, and created the nodes
> if needed.

There is no long-term storage in any of these systems.  How does that
help me?

> Love

-derek

-- 
       Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
       Member, MIT Student Information Processing Board  (SIPB)
       URL: http://web.mit.edu/warlord/    PP-ASEL-IA     N1NWH
       warlord@MIT.EDU                        PGP key available