[OpenAFS-devel] Read-only file system

Chris Huebsch chris@huebsch-gemacht.de
Sun, 11 Mar 2007 21:10:48 +0100 (CET)


Hello Erik,

On Sun, 11 Mar 2007, Erik Osterman wrote:

[everything RO]

> What would cause this unexpected behavior?

Is it possible, that there is a chain of RO-Volumes from /afs/ down to
the user-volumes?

AFS picks RO-Volumes first.

Let me make an example.

consider following volumes:
root.afs, root.cell, homes, user.chu each of them with one RO-Clone
(root.afs.readonly, root.cell.readonly, homes.readonly, user.chu.readonly)

If you now create this mount-chain:

/afs -> root.afs
/afs/cell -> root.cell
/afs/cell/homes -> homes
/afs/cell/homes/c/ -> (just a plain directory)
/afs/cell/homes/c/chu -> user.chu

then AFS can select for each mountpoint an RO-Clone. And it will do so.

You can break this behaviour by explicitely mounting a volume as a
RW-Mountpoint (specifying -rw to the fs mkmount).

I suggest the volume most down the path (eg. user.chu in my example).

It's best practice to mount root.cell at /afs/.cell (mention the
extra-point) as RW to generate alway rw-paths starting witch
/afs/.cell/....

Oh. And if you do not have any rw-volume available for any mount, remove
all ro's for any volume of your choice and mount root.afs into that
volume temporarily.

Or use dynroot, afaik it generates the /afs/.cell-mountpoint too.

HTH


Chris
-- 
web: http://chu.in-chemnitz.de | Chemnitzer Linux-Tage 2008, 1.-2. Maerz
gpg: 7F2B4DBA                  |     http://chemnitzer.linux-tage.de