nfs client should it use tcp or udp to nfs serve rportmapper
 

If I specify the udp protocol on a nfs mount the intiial communications to the nfs server portmapper (port 111 ) is done useing tcp not udp. All the the rest is done via udp (mountd nfsd etc).

question should the initial portmapper exchange also occur useing udp to port 111 of the nfs server?

mount -t nfs -o udp server:/export /mountpoint

I also tried to force nfsvers2 and it also did the inital portmapper exchange via tcp.

I see this on a redhat as 2.1 and a Mandrake 10.0

Placeing a 32768 in the /proc/sys/sunrpc/rpc_debug did not provide any clues.
I used tcpdump to capture the traffic between the nfs client and server.


Thanks

afaik the client will always use tcp for the handshake(identification) with the rpc (portmapper). the option to use udp or tcp only applies to the later nfs-traffic.

sl mritch.

thanks
 

Yep looks like linux nfs clients do an initail rpc dump port request to get all the rpc ports on the nfs server. I can see that approach has an advantage.

Other unixes nfs clients only do a getport mountd for the one port for mountd, but will need other get port request for other services.

I didnt even know this was happening till a linux nfs client banged it head against a firewall.

Two approaches to the same question (what port to use)

Now I have to convince the firewall folk to make adjustments.

Thanks