Hi,

To start off with, I would like to say that I am quite new to Linux, so I have little knowledge of advanced, or even some basic, system commands. Thank you for your attention.

Now, to the problem...
I am experimenting with booting diskless client. To minimize network complexity, the current setup consists of:
1 - Server running Slackware 9.1 / Bootp / Tftp Server
2 - A client running nothing of course (diskless) booted with a Netboot floppy
3 - A single switch connecting the computers, without internet access

What happens currently, is that the diskless boots, sets up the NE2000 compatible Accton 1660 ISA NIC, receives it IP (10.101.160.16) from the server (10.101.160.15), negotiates transfer options with the tftp server and attempts to transfer the netbootable image (bootImage.nbi).
Now, watching the show from Ethereal on the server, it goes pretty much like this:

CL: Broadcasts Boot Request with correct MAC
SV: Boot Reply
CL: Arp request goes out
SV: Answers arp request
CL: TFTP Request (bootImage.nbi)
SV: Option acknowledgment
CL: Acknowledgment block 0
SV: Sends block 1
CL: Acks block 1
SV: Sends block 2
CL: Acks block 1
CL: Acks block 1
CL: Acks block 1
...

At some point is stalls with the client never not acknowledging the last block, and the server not resending it. Which afaik, it should do when it receives a new request for the previous block. Longest I ever got was to block 31 acknowledged, but it rarely gets past block 3.
Sometimes a single UDP packet goes from the server to the client. Server port 33-thousand-and-something to client port 1079.

And to the system log it on each attempted boot of the client and startup of the server is printed that:
Jan 9 16:56:17 redbird bootpd[1002]: can not determine size of file "/bootImage.nbi"
Which to me makes no sense, since output on the client screen indicates that it does:
"Options: Blocksize 1432, Filesize 1054208, Timeout 15"

Finally, I will include some records from /etc/bootptab and /etc/inetd.conf:

Bootptab extract:
.mediatek:\
:hn:dn=server.svgs:\
:td=/tftpboot:\
:to=auto:
:bf=bootImage.nbi:\
:bs=auto:\
:ht=ethernet:\
:ha=0000E84B2BA2:\
:ip=10.101.160.16:\
:sm=255.255.255.0:\
:sa=10.101.160.15:\
:gw=10.101.160.1:

Inetd.conf extract:
tftp dgram udp wait root /usr/sbin/tcpd in.tftpd -s /tftpboot
bootps dgram udp wait root /usr/sbin/bootpd bootpd

This might be to little information, maybe not the right info... But I hope someone can make something of it. If you more, please tell me how to get it.
Furthermore, I will probably not have net access in the weekend, so I will not be replying to any answers until late Sunday or on Monday.

Thanks in advance to any helpful replies...

Hello.

First, I would like to say that i'm also new to Linux but not new to computers and networking. So, I'll give you some general tips:

1. Connect the workstation directly to the server using a cross-over patch cable,
this will eliminate the possibility that the router/switch is the problem.

2. Disable all firewalls on the server, it is possible that a rule is preventing the boot image from loading.

3. When making changes to any ".conf" file make sure you restart the service that uses that configuration. For example, if you edit "dhcpd.conf" restart dhcpd by using these commands: "service dhcpd stop" and then "service dhcpd start".

4. Make sure that all services required to boot the diskless workstation are enabled, and running on the server. Here is a list of some commands and possible outputs you might get:

Command: ps -e | grep portmap
Output: 1156 ? 00:00:00 portmap

Command: ps -e | grep nfs
Output: 1886 ? 00:00:00 nfsd

Command: ps -e | grep mountd
Output: 1907 ? 00:00:00 rpc.mountd

Command: ps -e | grep dhcpd
Output: 1882 ? 00:00:00 dhcpd

ps = report process status, the "-e" options is to select all processes.
grep = searches the named input.
portmap, nfs, mountd, and dhcpd = Services that you might/will need to boot the diskless workstation.

Note: The "Output" may not match the above, but if you get anything this means it's working.

P.S. I have set up my diskless workstations using Mandrake 9.2 server. and LTSP. So I won't be able to give you anything very specific for your setup.

well it is very difficult to tell what is going on.

can u tell me what the screen is doing on the client.

Also what is the max image size for your netboot pxe.
There is a hardware limit of 500k on x86 hardware.
try compiling a bzImage and test that.

Also is your kernel sane, might it be getting the image but it is corrupt.


So what to do.
1 find out the max kernel image size for your bootloader.
2 if it is * try and boot a normal kernel from slackware
3 look up doc's about your boot loader are there errors, or lack of errors
being reported.


also do u have a machine with onboard network card, as they can network boot with a build in PXE. try that and see what happens. remember it will have a 500k image limit.

look up pxelinux and read about it maybe booting that first will fix work problems. ?????
network booting an entire linux distro is not an easy thing, and booting the kernel is the easy bit. all i can say is read read read, i am only a student and have been using linux for about 2 years, but boy do i read shit loads of man pages, howto docs, javadocs everything i can get my eyes on. once I have read enough info i then fire up my mac(server) and start playing with my client untill the dam thing works.
will have to admin netbooting is fast, love to see it boot over gigbit with a raided server.

Thanks a lot for the replies, I solved it now, and not a moment too early, I was about to take a bite out of my keyboard when it finally worked.
It was the 500k image limit... My boot image was approximately 1 mb. I just filled a text file with some garbage HTML until it was about 500k and put that as the boot image, worked great. Except it didn't boot of course.

But right now, in between the waves of joy, I feel like there is a manual somewhere that I should have read. Leckie, anyone else, can you point me to some info on that image limit? I never read anything about that in the howtos or different documents I checked...

Again, thanks a lot for the help, I am off to read some kernel howtos, to shrink my system a tad bit.

Yep have fun i have spend weeks trying to get linux +network +nfs to be smaller then 500k. and i have come to the conclusion it is impossible. So best thing todo is use the bootloader to load pxelinux(this is a bootloader which can load really big kernels). pxelinux is only about 100k. Once pxelinux loads it will try and configure it's ipaddy then it downloads a config file from tftp. this file tells it which kernel to download. good idea, very configurable.

I have once loaded a 5mb linux+initrd.img with it.

It is really good as i just get the default kernel with mandrake add nft root and kernel autoconfig and u have instant network boot kernel.
then i just open the initrd.img and put some modules in there change the linuxrc script to pivot on nfs and u have a workable system.

Docs. well i first started at www.tldp.org there are a few there, but they are out of date and do not work(still good read). then u need to read some theory in the linux source. There is a documentation folder in there read about initrd and nfs.
also man pages on pivotroot. ummm yep that is about it.

ow yeah the kernel limit is bound to 19bytes of data. this is an old rule which has not been changed due to it not being used much. this is not the case on other architectures such as commercial unix which use network booting to install the operating system. It has been changed with 64bit processors i think.

That solution is something I never would have thought of leckie... Just tried it, worked great. A few days of tweaking, and I should be ready to set up some workstations.

Again, thanks a lot, you saved me a lot of time setting this up.