I had a weird issue show up the other day. I am using Knoppix 6.7.0 to do PXE booting. I cannot use 6.7.1 at this time, so I do not know if this is fixed there.

Client -> Knoppix 6.7
Server -> Fedora 14. PXE files from Knoppix 6.7 CD.

problem:
Have an old machine with onboard 1 Gig nic, and pci 10/100 nic. Ethernet cord is plugged into onboard nic, boots, gets IP, TFTP Starts, select the kernel and go. Kernel loads, gets to the udhcp part of getting an IP. Times out.

Playing with it, I re-started the process, and when udhcp was trying to get an ip, I moved the ethernet cord to the 10/100 pci nic. Knoppix found an ip and completed booting.

Investigating further, it comes to show that the pci 10/100 board was grabbing eth0, and not the onboard Nic!

Further investigation shows that the kernels attempt to load drivers in parallel, and that the ethX numbering scheme is based in part, on driver load order. Therefore, I can't really force that, and I'm also in an environment of lots of machines, so it's not like I can do a simple fix for 1 and be done.

I then unpacked the minirt.gz and examined the init script. The issue is in the DHCP loop, it calls ifconfig up ethX. If ifconfig returns a non zero number, it is supposed to continue. For some reason, (driver? ifconfig bug?) calling ifconfig on eth0 (with no cord) results in a "ok" response.

I have seen ifconfig up seem to "work" before on nics without a cord. So i suspect that it's not safe to depend on the return from it. I decided to change the init code to instead check for the presence of the cable using the /sys/class/net/ethX/operstate file.

Works like a charm. I wanted to share the issue here, along with the code I did to change it, and see about putting that into future versions. I don't know how easy it will be to reproduce this, i saw it happen 18 times out of several hundred machines.

note: my logic checks for the word "down" in the operstate file. I do not check for up, as there are a couple of other possible values, which may indicate there is a cord there. but it's always down when its down.

I also check for the existence of the file, as ifconfig does not even raise an error doing "up" on a non-existent interface, like eth2. maybe ifconfig just needs to be upgraded, it seems pretty dumb.

I just did ifconfig --version, and for Knoppix 6.7, it shows net-tools: 1.60 and ifconfig 1.42 (2001-04-13)

Anyway, the fix:
open init, search for DHCP and replace that part with this:


Code:
DHCP=""
 # How many network cards should we probe for?
 for i in 0 1 2 3; do
  ifconfig "eth$i" up >/dev/null 2>&1 || continue
  sleep 2
  if [ -f /sys/class/net/eth$i/operstate ]; then
          if grep -Fq "down" /sys/class/net/eth$i/operstate
          then
            continue
          fi
  else
    continue
  fi
 
  message "${CRE}${BLUE}DHCP Broadcasting for IP address (${MAGENTA}eth$i${BLUE})... ${NORMAL}"
  if [ -n "$HOSTNAME" ]; then
   udhcpc -i eth$i -H $HOSTNAME && DHCP="eth$i"
  else
   udhcpc -i eth$i && DHCP="eth$i"
  fi
  [ -n "$DHCP" ] && break
  message "${RED}No answer from network.${NORMAL}"
 done