> I have an NFS server with fddi and ethernet interfaces.
> At present [will be changed tonight], the hop count to this subnet is
> the same via le0 and nf0.  So occasionally the server updates his
> routing table to use the other interface....

> SunOS systems are not fussed at all, so I assume that when they send
> NFS requests to the ethernet address and get a reply from the fddi
> address they happily take it.

That matches my experience, too.

> The NetBSD (sparc-current 95-08-18) on the other hand does:

> 12:49:38.32 144.136.188.252.11c > 144.136.48.60.801: 128 getattr [|nfs]
> 12:49:38.32 144.136.188.165.801 > 144.136.188.252.11c: reply ok 96
> 12:49:38.32 144.136.188.252 > 144.136.188.165: icmp: 144.136.188.252 udp port 1034 unreachable

> and of course it hangs...

> Anyone got a patch for this?

The "patch" is to do the mount with the -c option (-c to mount_nfs, or
-o -c to mount) to have it not connect the socket.  If you're booting
diskless and this is your root filesystem, build your kernel with

	options	"NFS_BOOT_OPTIONS=NFSMNT_NOCONN"

(assuming a relatively recent sys/nfs/nfs_vfsops.c - for older source,
you'll have to hack the source, either to add NFSMNT_NOCONN or add
handling for the NFS_BOOT_OPTIONS config option).

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu