Mailing List Archive

James S. Martin wrote:
> I haven't encountered this problem before. This is from a Compaq
> DL320, with Intel Boot Agent 4.0.17 and tftp-hpa-0.28 running on
> Solaris. The PXE stack gets its IP address just fine, but it times out
> on the TFTP. While snooping on the Solaris box and grepping for TFTP, I
> found that the box gets one block of the TFTP transfer just fine.
>
> TFTP Read "pxelinux.0" (octet)
> TFTP Error: not defined
> TFTP Read "pxelinux.0" (octet)
> TFTP Ack block 0
>
> and continues with the Acks until it times out.. I have no problems
> pxebooting my othe Compaq boxes with the same IBA version. The "not
> defined" message is bugging me. Any ideas?
>

That's very odd indeed. The not defined is almost certainly an option
negotiation failure, most PXE stacks seem to force one on purpose before
starting the real transfer.

I hate to suggest this, but could you by any chance have either an
address conflict on your network or a hardware/cabling failure? The
fact that it only affects one box our of a set that is known to work
well is sticky...

-hpa

H. Peter Anvin wrote:
>>
>
> That's very odd indeed. The not defined is almost certainly an option
> negotiation failure, most PXE stacks seem to force one on purpose before
> starting the real transfer.
>
> I hate to suggest this, but could you by any chance have either an
> address conflict on your network or a hardware/cabling failure? The
> fact that it only affects one box our of a set that is known to work
> well is sticky...
>
> -hpa
>
>

Thanks for the help. We have 3 network interfaces on the boot server.
hme0, hme1, and hme2. hme2 is doing the DHCP, and is on the "81"
network, hme0 is on the "83" network and serving TFTP. The machine we
are building is on the "81" network. In the dhcpd.conf we had the
next-server line explicitly pointing to the IP address for hme0 (83
network). After I removed that line and restarted the dhcp server, it
worked fine. So it seems that for some reason the PXE stack doesn't
like having a next-server line if the machine being built is on the same
network as the DHCP server. That really doesn't make any sense to me.
But it works now :).

James

James S. Martin wrote:
>
> Thanks for the help. We have 3 network interfaces on the boot server.
> hme0, hme1, and hme2. hme2 is doing the DHCP, and is on the "81"
> network, hme0 is on the "83" network and serving TFTP. The machine we
> are building is on the "81" network. In the dhcpd.conf we had the
> next-server line explicitly pointing to the IP address for hme0 (83
> network). After I removed that line and restarted the dhcp server, it
> worked fine. So it seems that for some reason the PXE stack doesn't
> like having a next-server line if the machine being built is on the same
> network as the DHCP server. That really doesn't make any sense to me.
> But it works now :).
>

It could be a simple PXE stack bug, but since you're referring to these
interfaces as "hme", I'm guessing it's a Sun box. It could mean it's an
issue with the handling of ARP inside the PXE stack, since Suns have the
same MAC on all interfaces.

-hpa