[Dnsmasq-discuss] RaspberryPi PCAP file

[Geert Stappers]

On Sat, Dec 04, 2021 at 12:42:52PM +0100, Community Member via Dnsmasq-discuss wrote:
> On Fri, Dec 3, 2021 at 12:50 PM Geert Stappers wrote:
> > On Thu, Dec 02, 2021 at 10:02:27PM +0000, Simon Kelley wrote:
> > > I've not played with Pi, but assuming it's some variety of the PXE
> > > protocol, that usually tries  pathnames constructed from the MAC
> > > address, then the IP address with decreasing specificity (Ie
> > > 192.168.1.100, then 192.168.1, then 192.168, then 192) and finally it
> > > tries a fixed pathname. If you don't want to have machine-specific boot
> > > files, just don't provide those files and the boot process will drop
> > > though top the generic ones.
> >
> >
> > What would help is providing a fairly complete  network packet capture
> > of a Raspi attempting to netboot.
> >
> wget http://now.zapto.org/temp/RPI3bp13.pcap
> client Raspberry Pi 3 Model B Plus Rev 1.3

My observation on the PCAP file
* there are several stages
* each stage starts with a (partial) DHCP session


The first stage:
* DHCP Discover with option 97, Client Identifier,
  f6 27 6d 7a f6 27 6d 7a f6 27 6d 7a f6 27 6d 7a
  from client
* DHCP Offer with filename 'ipxe/undionly.kpxe'
  from server
* No DHCP Request, No DHCP Acknowledge
* Client TFTP Requests 'bootcode.bin'
* Server TFTP sents that 'bootcode.bin'
* Client TFTP Requests 'bootsig.bin'
* Server TFTP sents 'bootcode.bin'


The second stage:
* DHCP Discover with option 97, Client Identifier,
  f6 27 6d 7a f6 27 6d 7a f6 27 6d 7a f6 27 6d 7a
  from client  (yes, that is the same identifier)
* DHCP Offer with filename 'ipxe/undionly.kpxe'
  from server (yes, that is the same filename)
* No DHCP Request, No DHCP Acknowledge
* Client TFTP Requests '7a6d27f6/start.elf'
* Server response not clear to me
* Client TFTP Requests '7a6d27f6/start.elf'
* Server response not clear to me
* Client TFTP Requests '7a6d27f6/autoboot.txt

At that point there was enough information for me.  I'm only interested
in a better dnsmasq. For a better netbooting raspi am I NO stakeholder.[1]

I do see a match in the begin of the Client Identifier ( f6 27 6d 7a )
and the "serial number" (7a6d27f6). More network captures should prove
that match.[2]


 
> server log from one boot:
> 
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-dhcp[1584]: DHCPDISCOVER(eth-local) b8:27:eb:6d:27:f6
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-dhcp[1584]: DHCPOFFER(eth-local) 10.21.0.162 b8:27:eb:6d:27:f6
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-tftp[1584]: file /srv/tftp/bootsig.bin not found
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-tftp[1584]: sent /srv/tftp/bootcode.bin to 10.21.0.162
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-dhcp[1584]: DHCPDISCOVER(eth-local) b8:27:eb:6d:27:f6
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-dhcp[1584]: DHCPOFFER(eth-local) 10.21.0.162 b8:27:eb:6d:27:f6
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-tftp[1584]: error 0 Early terminate received from 10.21.0.162
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-tftp[1584]: failed sending /srv/tftp/7a6d27f6/start.elf to 10.21.0.162
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-tftp[1584]: file /srv/tftp/7a6d27f6/autoboot.txt not found
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-tftp[1584]: error 0 Early terminate received from 10.21.0.162
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-tftp[1584]: failed sending /srv/tftp/7a6d27f6/start.elf to 10.21.0.162
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-tftp[1584]: sent /srv/tftp/7a6d27f6/config.txt to 10.21.0.162
> Dec  3 16:51:43 rpi-cb-1f-f7 dnsmasq-tftp[1584]: file /srv/tftp/7a6d27f6/recovery.elf not found
> Dec  3 16:51:47 rpi-cb-1f-f7 dnsmasq-tftp[1584]: sent /srv/tftp/7a6d27f6/start.elf to 10.21.0.162
> Dec  3 16:51:47 rpi-cb-1f-f7 dnsmasq-tftp[1584]: sent /srv/tftp/7a6d27f6/fixup.dat to 10.21.0.162
> Dec  3 16:51:48 rpi-cb-1f-f7 dnsmasq-tftp[1584]: file /srv/tftp/7a6d27f6/recovery.elf not found

In https://lists.thekelleys.org.uk/pipermail/dnsmasq-discuss/2021q4/015955.html
is a reference to --dhcp-script as in the dnsmasq manual page.  The dhcp-script has
access to the environment variable DNSMASQ_CLIENT_ID

I think that can solve the original problem (
 https://lists.thekelleys.org.uk/pipermail/dnsmasq-discuss/2021q4/015954.html
).


Groeten
Geert Stappers

Foot notes:
[1] Make me stakeholder by providing enough RaspberryPi hardware
    Even better provide me with hardware that should get better
    netbooting.
[2] I'm willing to check more .pcap files
-- 
Silence is hard to parse