[OpenAFS] AFS always readonly?

Brandon S. Allbery KF8NH allbery@ece.cmu.edu
Thu, 18 Oct 2007 17:20:08 -0400


On Oct 18, 2007, at 15:26 , Tim OBrien wrote:

> machine. However, from the other machine, I can't create a  
> mountpoint in afs or
> do much of anything, since it states that /afs is mounted read  
> only. I used the

Normally if a read only replica exists it will be mounted by  
preference.  The usefulness of r/o replicas is in fact dependent on / 
afs mounting read-only by default.

Usually one creates an explicit read-write mountpoint for the cell (/ 
afs/.cell vs. /afs/cell) while the initial root.cell and root.afs are  
still not replicated.  Once you have the latter, you can always  
create a read-write mount for root.afs or other volumes.  (-dynroot  
should do this automatically)

If you have inadvertently created a cell without a forced r/w mount  
in your root.afs, there are two ways to fix it:

- create such r/w mount in some other cell that has an r/w mount and  
knows how to reach yours (requires you to auth to *your* cell, so  
crossrealm/cross-cell auth is not necessary, but you may need to  
insure that using your cell's Kerberos tickets won't cripple your  
access to the machine in the foreign realm/cell)

- bring up a client with -dynroot, which should get you an automatic  
r/w mountpoint, and fix the static mounts from there.

(Several years back when a combination of a corrupted r/w site for  
root.afs and a pair of inexperienced admins caused our /afs to  
suddenly become quite empty, afs didn't have dynroot.  Luckily I had  
a test cell running which knew how to reach the main one, so I was  
able to repopulate root.afs remotely.  These days -dynroot is almost  
certainly the best way to go for this.)

As for the difference between the two machines:  you probably haven't  
restarted the client on the first machine yet, so its initial read- 
write mount of /afs is still intact.  You should probably use that to  
make sure the /afs/.cell forced read-write mount of root.cell is  
present, and rerelease.

-- 
brandon s. allbery [solaris,freebsd,perl,pugs,haskell] allbery@kf8nh.com
system administrator [openafs,heimdal,too many hats] allbery@ece.cmu.edu
electrical and computer engineering, carnegie mellon university    KF8NH