Got my first SuSE machine at work (well, not mine, but I'm setting it up), and I'm running into a weird problem with ypbind. If I call ypbind on its own -- no arguments -- it'll work. man page sez it's parsing /etc/yp.conf, which has the line "domain foo broadcast", and sure enough it broadcasts on a nice privileged port and binds to the server for domain foo. If I call ypbind with the -d argument, it stays in foreground, prints debuggin messages and fails like so:
do_broadcast() for domain 'foo' is called
broadcast: RPC: Can't encode arguments.
leave do_broadcast() for domain 'foo'
Signal (2) for quitting program arrived.
Well, crap. That's weird. After some searching, found [Debian bug
(tentatively, but) on libc. And but so there's these other bugs from, you know, Novell/SuSE, which also sound similar. And holy crap, where the hell have I been that I haven't heard of:
echo 65535 > /proc/sys/sunrpc/rpc_debug
echo 65535 > /proc/sys/sunrpc/nfs_debug
And other interesting behaviour from that second bug:
The problem in that bug was that immediately following a reboot, the NFS client will end up opening the same TCP port it used before, so it tries to establish a TCP connection from client:1234 -> server:2049. The server still has a TCP control block for this, and replies with a single ACK containing what it thinks are the right sequence numbers. That ACK is eaten by the conntrack module because the connection isn't yet in state ESTABLISHED.
Okay but back to our original bug: which appears to have been fixed
now by adding one line to /etc/yp.conf:
broadcast
domain foo broadcast
Why the fuck that should work is beyond me...