bug: mountd

[Jeff Nieusma]

We're having some really interesting problems here with mountd under 4.1.
I rebuilt the libc.so so it uses the nameserver just because I like to be
up on all the modern technology ;-)

It seems our nameserver likes to return hostnames as:
hostname.Colorado.EDU rather than hostname.colorado.edu

this actually causes mountd die with a segmentation fault when I have all
lowercase character in my /etc/exports file.  Is this a feature?  Has
anyone else seen this happening?

Ok, rather than fully qualify all my hostname in /etc/exports and
capitalize the correct letters, I kludged my way around it...  For ease of
use and quickness of operation, I did the following just to make the
machines work:

mkdir /usr/lib/kludge
cp /usr/lib/libc.s[oa].1.5 /usr/lib/kludge	 # the original sun lib's

and now I'm starting mountd like this from rc.local:

        ( LD_LIBRARY_PATH=/usr/lib/kludge; export LD_LIBRARY_PATH;\
          cd /usr/lib/kludge;  rpc.mountd -n)

This seems to work fine as long as everyone sending a mount request is in
my /etc/hosts file.  However, if I go to another workstation and change
the IP number and try to mount a filesystem, my server mountd process dies
with a segmentation fault.  Gee, does this sound like someone is
dereferencing a null pointer?

So, to combat this problem temporarily until our source code arrives in
the mail, I have added the following line to crontab:

5,20,35,50 * * * * if ps ax | grep 'rpc.mountd$' >/dev/null; then : ; else ( LD_LIBRARY_PATH=/usr/lib/kludge; export LD_LIBRARY_PATH; /usr/etc/rpc.mountd ); rm -f /core; fi

It very infrequently dumps core, but I put the rm /core in there just in
case.

 Jeff Nieusma                         Logical:  nieusma at boulder.colorado.edu |
 System Administrator/Programmer      Audible:  (303) 492-0677               |
 Computer Science Department         Physical:  Campus Box 430               |
 University of Colorado                         Boulder, CO  80309-0430      |