[Dnsmasq-discuss] dnsmasq v2.40: pxe network boot is failing

[Simon Kelley]

Wayne Sherman wrote:
> I cannot get network booting to work.
> 
> Client Computer:
>   Dell Dimension 4600 with built-in 10/100 NIC
> 
> Server running Gentoo:
>   Linux 2.6.23-gentoo-r3 SMP Thu Dec 6 04:36:06 PST 2007 i686
>   AMD Athlon 64 X2 Dual Core Processor 3800+ GNU/Linux
>   Asus M2A-VM AMD-690 motherboard with integrated gigabit NIC
> 
> If I don't set "tftp-no-blocksize", it doesn't get very far,
> and after setting it still fails with errors.
> 
> See here for full dnsmasq.conf, more complete log excerpt and screen shots:
> 
>   http://www.systemdesignworks.com/dnsmasq
> 
> 
> Here is an excerpt from the log:
> 
> <START log excerpt>
> Dec 30 23:36:30 wopr dnsmasq[10164]: TFTP failed sending
>   /home/netboot/pxelinux.cfg/default to 192.168.1.178
> <END log excerpt>
> 
> Here is part of my dnsmasq.conf file
> 
> #dhcp-boot=pxegrub
> #dhcp-option=150,grub.conf
> 
> dhcp-boot=pxelinux.0
> enable-tftp
> tftp-no-blocksize
> tftp-root=/home/netboot/
> 
> Any help would be appreciated.
> 

I'm not exactly what's going on, but the following observations might help.

1) I don't think there's any configuration problem: DHCP is working 
fine, and the client seems to be trying to do the right thing.

2) TFTP seems to be failing. TFTP is a pretty simple-minded protocol, so 
it's fairly easy to trace what's going on in a packet-trace captured 
using wireshark.

3) The logging is a bit misleading: when it says "sent file <name>...." 
that means that the file has been found and the transfer started, not 
that it's complete. The "failed" logs indicate that the transfer times out.

4) Attemped tranfers which fail with "file not found" are not logged, so 
all the address-related filenames which pxelinux tries are not logged.

5) You say that your server is gigabit and the client 10/100. What's 
doing the rate-adaption between the two? I have a consumer-grade 10/100 
switch which works great when all the clients are at the same speed, but 
  is useless with mixed speed. There's no significant buffering and it 
just drops packets going from fast links to slow. TFTP is a stupid 
protocol and doesn't tolerate bad networks. Maybe your Gb switch is similar?


Cheers,

Simon.