AIX Mail stored in AFS space

Paul Blackburn mpb@hursley.ibm.com
Tue, 13 Feb 2001 18:39:09 +0000


Hello Rick,

My experience with this is that it works well.
We run sendmail daemon and a sibling reauth daemon
sharing the same PAG.

Mail is appended to ${HOME}/.mail with the
appropriate ACLs to allow the authenticated sendmail
daemon to write.

    mpb@rest $ ls -l .mail
l    rwxr-xr-x   1 mpb      system        13 Jan 15 1999  .mail ->
afsmail/.mail
    mpb@rest $ fs la afsmail
    Access list for afsmail is
    Normal rights:
      system:administrators rlidwka
      mpb rlidwka
      sendmail rliwk

Users are responsible for their space quota
instead of the system admins worrying about
shared /var/spool/mail/.

We deliberately disable sendmail from processing .forward files
because it could be easily abused. Also, whatever is done in
.forward could equally be done in a managed global /etc/aliases file.

One really neat thing is that if you have multiple mailservers
with same MX preference records for your DNS domain
with each mailserver delivering mail into /afs/@cell/
then you have a scalable mailserver cluster. Whoohoo! ;-)

One benefit of this is that you can take out
one of the mailserver cluster machines with
no impact on mail delivery.

My bundle of scripts and notes for this are available here:
    http://www.angelfire.com/hi/plutonic/images/authsendmail.tar.Z

I hope this helps.
--
cheers
paul                             http://acm.org/~mpb

R McGee wrote:

> We currently are in production at AIX 4.3.3 and AFS 3.5.  We have our mail
> set up through a central mail  server and NFS mount /var/spool/mail/ from
> that server to every client machine.  Of course, anytime we have a problem
> with our mail server, EVERYONE feels it and grumbles loudly.  We would like
> to convert to a mail scheme where the mail is delivered and stored in the
> users AFS home directory and the need for a NFS mount is eliminated.  I
> would be interested in hearing from anyone who has done this, with any
> associated problems, nuances, etc. and some direction on how to get it
> implemented.
>
> All input is appreciated.
>
> Rick McGee
> IBM Raleigh
> mcgee@us.ibm.com
> (919) 543-9299