PXE + TFTP boot problem

[srd]

I'm having an issue PXE booting from another host. I can TFTP as localhost or from other machine from the command line, but the diskless appears to not be able to download the kernel upon boot, even though it seems to be getting the IP just fine. Any ideas?

This is the error that appears on the diskless nodes console.

[machspeed]

hey srd

I literally had this issue 3 days ago. Check that tftpd is actually running.

I got the same error message, "PXE-E32: TFTP open time-out" and I had foolishly uninstalled my tftpd at some stage. Reinstalled, made sure it was running (simple "ps ajfx") and then my PXE booted systems worked.

[srd]

Thanks, but my ftp server is running. I was able to test it locally and from another machine and download the files just fine. But when booting from the diskless node, I receive the following error.

[machspeed]

Edit: On re-reading everything. Maybe you'll only need to shift the "next-server" line into the subnet section. It may be a case of the PXE loader on the local machine looking to your router at 10.0.0.1 instead of your PXE server at 10.0.0.100

Ah ok, maybe this is an issue of your node locating the file on the server.

I'd assume you have pxelinux.0 in the /diskless directory along with your pxelinux.cfg directory. Pays to double check.

Something else I've noticed is in your dhcpd config, you have a group specified with the filename option there. Can you try adding the filename option directly into the host section? Also, your next-server probably should go into the subnet part.

I'll add part of mine for reference. It's nothing glamorous but it works.

[krinn]

https://forums.gentoo.org/viewtopic-p-7592928.html#7592928

[srd]

I've still been unable to get these diskless nodes to boot.

The following is from running "tcpdump port 69" on the master.

[szatox]

My working dhcpd config, comments and non-pxe zone striped, the key positions set bold

[srd]

Here is /etc/conf.d/in.tftpd for tftp-hpa.

[szatox]

I had been using wireshark to sniff on the network traffix during dhcp negotiation. This let's you see at what point things go wrong.
Mine setup is very different

[srd]

Ugh ... so I replaced the bad broadcast address and made other changes like what you had, but nothing has changed, still seeing the TFTP server timeout.

I also put wireshark on and it looks as if the DHCP server is working fine, 1) because the host is showing the correct DHCP IP address on it's boot screen, and 2) wireshark is showing the Discover, Offer, Request, and Ack between the server and the client node, and 3) /var/log/messages on the server side is showing a RRQ from the proper node ip (as given by DHCP) for the filename /pxelinux.0.

From what I understand, this line in /var/log/messages is telling me that the client is doing a read request from the tftp server, and it looks correct. The file is there, has read perms ...

[apecksec]

I have struggled with the same problem for a LONG time. This was very difficult to figure out...

For me it was an MTU issue. Network MTu was 1476 but TFTP default MTU is 1500 and some TFTP clients didn't seem to work like that, while others were fine.

tftp will lkely be trying to send 1500 byte packets, but your network will be fragmenting those packets if the mtu is <1500 (like when going through a tunnel). So you have to tell the tftp server to use an MTU which is less than the MTU of your traffic flow. In my case I just used 1300 knowing it's well below my mtu of 1450 or so.

So, try adding '-B 1300' to your tftpd config and restarting the tftpd daemon. Mine looks like this:

/usr/sbin/in.tftpd -B 1300 -l -v -v -v -m /etc/tftpd.conf -s /var/public/tftproot

hope this helps someone out there in the future