[Dnsmasq-discuss] dhcp-script and netboot pi

[Simon Kelley]

On 25/02/2024 23:24, Carl Karsten wrote:
> Either dhcp-script isn't doing what it is expected, or I'd like it to do more.
> 
> I am netbooting raspberry pi.  so some dhcp client in the pi firmware
> get's an IP and netboot params, then tftp client gets files.
> 
> the dhcp traffic happens and is shown in the logs, but not dhcp-script:
> 
> sudo journalctl --follow -u dnsmasq.service
> Feb 25 16:49:44 base dnsmasq-dhcp[47924]: DHCPDISCOVER(eth-local)
> b8:27:eb:2f:5d:08
> Feb 25 16:49:44 base dnsmasq-dhcp[47924]: DHCPOFFER(eth-local)
> 10.21.0.102 b8:27:eb:2f:5d:08
> Feb 25 16:49:50 base dnsmasq-tftp[47924]: file /srv/tftp/bootsig.bin
> not found for 10.21.0.102
> Feb 25 16:49:50 base dnsmasq-tftp[47924]: sent /srv/tftp/bootcode.bin
> to 10.21.0.102
> Feb 25 16:49:50 base dnsmasq-dhcp[47924]: DHCPDISCOVER(eth-local)
> b8:27:eb:2f:5d:08
> Feb 25 16:49:50 base dnsmasq-dhcp[47924]: DHCPOFFER(eth-local)
> 10.21.0.102 b8:27:eb:2f:5d:08
> Feb 25 16:49:50 base dnsmasq-tftp[47924]: error 0 Early terminate
> received from 10.21.0.102
> Feb 25 16:49:50 base dnsmasq-tftp[47924]: failed sending
> /srv/tftp/042f5d08/start.elf to 10.21.0.102
> Feb 25 16:49:50 base dnsmasq-tftp[47924]: file
> /srv/tftp/042f5d08/autoboot.txt not found for 10.21.0.102
> Feb 25 16:49:50 base dnsmasq-tftp[47924]: error 0 Early terminate
> received from 10.21.0.102
> Feb 25 16:49:50 base dnsmasq-tftp[47924]: failed sending
> /srv/tftp/042f5d08/start.elf to 10.21.0.102
> Feb 25 16:49:50 base dnsmasq-tftp[47924]: sent
> /srv/tftp/042f5d08/config.txt to 10.21.0.102
> 
> log from
> https://github.com/CarlFK/pici/blob/main/ansible/roles/site/files/pib/pistat/scripts/send_stat.py#L83
> 
> Namespace(action='tftp', mac='52476', ip='10.21.0.102',
> hostname='/srv/tftp/bootcode.bin') dsh='(none)'
> Namespace(action='tftp', mac='2545', ip='10.21.0.102',
> hostname='/srv/tftp/042f5d08/config.txt') dsh='(none)'
> Namespace(action='tftp', mac='2979264', ip='10.21.0.102',
> hostname='/srv/tftp/042f5d08/start.elf') dsh='(none)'
> 
> I suspect the problem is with the client not DHCPREQUEST and so the
> server doesn't ACK and thus I guess an IP hasn't actually be
> allocated, thus no "add" event has happened which calls the script.
> if this is the case, can we add an "offer" action?
> 

You're right that the script actions are triggered by changes in the 
DHCP lease database, and DISCOVER/OFFER doesn't change the database so 
there's no action. Because of the link with the lease database and the 
lack of a lease at OFFER stage, adding a new action there is severely 
non trivial.

I'd suggest that the the Pi netboot code is pretty buggy: a client 
shouldn't use IP address until it gets a lease, and it doesn't get the 
lease until it gets a DHCPACK message. It's perfectly possible to have a 
two-message interaction: just set the rapid-commit option in the 
DISCOVER packet and the server will go straight to DHCPACK and 
committing a lease. Buggy netboot code is far from uncommon, in my 
experience.

Simon.