paperlined.org
apps > slax > PXE_booting
document updated 14 years ago, on Mar 1, 2010
When I'm finished collecting all the details for this, I'll post this bug report in the forums with the title "Bugfix for PXE booting".
In Slax v6.1.2 there's a bug with the "Slax as PXE server" boot option.

Explanation

To boot successfully, a PXE client has to fetch these files, in order:

The bug occurs when it fails to load the 'default' file. PXELINUX has the notion of a "path prefix". PXELINUX's documentation says that in this case, it should decide that the path prefix is "/pxelinux.cfg/". This means that it will try to load tftp://10.*.*.1/pxelinux.cfg/pxelinux.cfg/default, which won't work.

The Wireshark capture at http://www.paperlined.org/apps/slax/PXE_booting/gpxe_with_v6.1.2.pcap shows this problem happening.

If this is a PXELINUX standard feature, then why wasn't this problem found in previous testing? Answer: Some PXE clients seem to pass the "/pxelinux.cfg/pxelinux.0" path to PXELINUX, while others seem to prevent PXELINUX from knowing what its path is, thereby obscuring this bug. Clients that always fail with v6.1.2 include: gPXE. Clients that always succeed with v6.1.2 include: VirtualBox's BIOS.

TODO: test this theory out on a larger number of computers in the lab, especially physical computers

Workaround

When using v6.1.2, a quick fix is to create /etc/dnsmasq.conf with these two lines: 

dhcp-option-force=208,f1:00:74:7e    # pxelinux.magic   (without this, the next setting will be ignored)
dhcp-option-force=210,/              # pxelinux.pathprefix
And then run /boot/pxelinux.cfg/start to restart dnsmasq.

Long-term fix

For the official release, it's probably better to make those changes to dnsmasq's command line, inside the /boot/pxelinux.cfg/start script.