Hi,
I'm trying to set up a PXE boot environment.
Now the diskless machine reads in vmlinuz and initrd.lz from the /tftpboot directory of the dhcp server machine (host name: antilop, runing slackware 11.0).
There is a NFS export named /usr/data/ubuntu on antilop (which can be mounted from the client when runing mandriva mcnlive Toronto)

When I'm trying to boot with PXE, the kernel and initrd starts, eth0 gets a valid ip address.
Then a lot of error messages appear saying:
NFS over TCP not available from 192.168.1.48 <- this is antilop, my DHCP server and NFS server too.

Here is the content of the /tftpboot directory:

-rw-r--r-- 1 root root 34 2010-02-11 20:49 boot.txt
-rwxr--r-- 1 tjp users 75564 2009-10-05 16:08 gpxelinux.0*
-r--r--r-- 1 root root 5755032 2009-10-28 00:00 initrd.lz
drwxr-xr-x 2 root root 2592 2010-02-12 12:59 pxelinux.cfg/
-r--r--r-- 1 root root 3890400 2009-10-16 00:00 vmlinuz

Here is the contents of my /pxelinux.cfg/default:

default aaaa

label aaaa
kernel vmlinuz
append boot=casper netboot=nfs nfsroot=192.168.1.48:/usr/data/ubuntu initrd=initrd.lz

There is no process listening on tcp port 2049 (on antilop)


I asked my friend google but it didn't give a satisfying answer so far.
Some said that no nfs-over-tcp exists on slackware, some said nfs clients trying over tcp must fallback to udp.
I am quite lost. Can anyone help me?

Might have to switch over to gpxe and use http or some other tcp.

Not sure about your topology so I can't say why it seems to be falling back to udp. Might be your version. Not sure if the newest if hardcoded to use tcp first.

I guess it could be some firewall issue.

Thank for your post.
Since english is not my native language perhaps I was not clear enough...
Consider the following:
server side:
-tftp server works, no problem
-dhcp server works, no problem
-nfs server works, listens only on 2049 udp port

client side:
-gpxelinux.0 loads
-vmlinuz loads
-initrd.lz loads
-/scripts/casper-premount (initial script) starts
-ifconfig brings up eth0 with dhcp - ok.
-tries to mount myserver:/usr/data/ubuntu (where the content of the cd image reside)

the error message refers to that it attempts nfs mount over tcp. At this point I thought that it ought to fall back to udp, since tcp failed. (according to some google result)
Obviously I wanted to assure that my nfs server works correctly. I booted mandriva with this client machine and I was able to mount myserver:/usr/data/ubuntu. I think no firewall problem is present.

One can ask why I insist on slackware with its nfs capability.
The problem is the constrains on the server side. The server hardware was probed with several distros and slackware 11.0 was the only solution - I don't know why.

You English is good.

Sorry I am of no help.

I would have thought a nfs server would have been configured for tcp.