Subject: Re: more on CFS
To: None <>
From: Steven M. Bellovin <>
List: tech-crypto
Date: 05/01/2001 11:14:59
In message <>, Adam Laurie writes:
>Steve Bellovin wrote:
>> I should have mentioned one more thing:  on 1.5.1, at least, there are
>> some broken system #include files.  See PR 12761 for details and the
>> fix.  I have no idea what other versions of NetBSD, if any, are
>> affected.
>current hpcmips is clean, but the story isn't over - i'm now at the
>stage where i can get the mount request through rpc, but it then fails:
>  gnugget-root# /sbin/mount_nfs localhost:/null /crypt
>  mount_nfs: rpcbind on server: RPC: Program not registered
>so i had to add the "-4" flag to mount_nfs to switch off ipv6 stuff
>(which i'll submit patches for once this all works):
>  gnugget-root# /sbin/mount_nfs -4 localhost:/null /crypt
>  mount_nfs: localhost:/null on /crypt: Protocol not supported
>but this message would seem to come from kernel level - anyone got any
>idea where to go from here? did you have to do any further tweaking on

I'm firing up CFS this way:

mountd && \
        /usr/local/sbin/cfsd && \
        mount -o intr,-2 /crypt

-2 gets the right version of NFS.  I have no idea why I have to say instead of localhost; your -4 option may be working around 
the same problem.  (Btw, I'd suggest a different name than -4, since 
-2 and -3 control NFS versions.)

Also ensure that you compile with -DCFS_PORT=2049.  (And I'm not sure 
if my note on the lseek changes got through.)
>btw, if you're in touch with matt blaze is there any chance you can get
>the export filter switched off on the homepage now that the u.s. has
>dropped the restrictions?
I'll talk to Matt about that.  The issue may be the AT&T bureaucracy as 

		--Steve Bellovin,