Wednesday, September 10, 2008

Running TAHI's v6eval suite on linux systems

For many years I've been verifying our products stack at work with the TAHI organization's v6eval engine and conformance test suite. One of the good and bad things about it is that it ran on FreeBSD -- pretty much only on FreeBSD.

This is because in the beginning this organization also were also sponsoring the KAME stack that you could drop right into the FreeBSD kernel sources and come out with a highly compliant ipv6 stack. Well, like anything that was FreeBSD only, that made it difficult to run in a company that did Linux development.

It wasn't long before people tasked with ipv6 verification and strapped by Linux systems began attempting to port it. IBM first posted a patch to get the v6eval engine running, and they did a really good job. That was posted on some long lost IBM site, but I had a copy and saved it off to help people out with their Linux systems here. This patch worked for a long time, as long as you didn't upgrade your old 2.2 or 2.4 based Linux system.

That patch I have archived here: v6eval-2.2b6_linux.patch

Recently the v6eval code was updated so that it would, a. run on modern FreeBSD systems, and b. compile on modern C++ compilers. This change was good, because it actually made the code more portable in a way. A while back on the TAHI users email list a patch was posted for Fedora 8. This patch was based on the Debian port and makes changes for Fedora 8.

That patch I have archived here: v6eval_fedora8.diff

Here's the original message from the list (you can pull it from the TAHI user archives if you choose -- I removed the emails addresses/names).

Date: Thu, 3 Apr 2008 13:31:45 -0700
Subject: [users:00660] Re: Running Tahi on Linux

Here is the patch file for v6eval-3.0.12. Naturally the Tahi folk released a new version while I was building and verifying this patch file.

To use this, install the source from Tahi onto a Fedora 8 system and use the 'patch' command to apply this patch. You should then be able to use 'make' as before to
build the system and 'make install' to install it in /usr/local/v6eval. Note that there are probably some dependancies that I'm glossing over. If you have problems
building, let me know and I'll try to help you figure out what packages you need to install. The compiler does issue a lot of warnings about virtual functions that I
didn't chase down, but the system seems to still work for me.

While this patch is meant to be applied to the base source as available from Tahi, I won't claim to be the one who did all the necessary modifications. I started
with the tahi-linux package that according to the change log came from Kazuo Hiekata at IBM, then made changes to get it to compile on Fedora 8. My modification
mostly involved the Makefiles. Guess I should add my name to the change log as well. Maybe next time.

Let me know if there are any issues.

The Fedora/Debian patch has some good bits but makes WAY too many unnecessary changes (reformatting code, rewriting Makefiles -- no big deal as it was someones rough personal copy posted to a list). When I ported to the automated test system here I did the following and was able to port it in just under a day:

  • extract the latest tarball
  • grab 'pmake' package for my distribution and install (*BSD style make)
  • grab 'byacc' package for my distribution and install (*BSD style yacc)
  • apply minimalist 2.2b6 patch (read the instructions at the top of the patch file text)
  • attempt to compile, and fix issues.
  • pull very specific Linux OS related issues from Fedora/Debian patch over to extracted area.

That's it.

Monday, September 1, 2008

Patchiness of source control

I maintain several systems that I use day to day at my Job. One of these is a box I use for mostly code compilation that run Gentoo. The machine runs on a Tyan S2822-D, aka Thunder K8SD Pro.

When I decided to wipe this machine, which was running CentOS 3.x at the time, and drop Gentoo on it, I had been out of the Linux Kernel compilation game for quite a while. To make a long story short, I could not find the correct Kernel compile options (which have become inordinately numerous and bloated) in the 2.6 kernel to make this beast boot. The internet was very little use as to which ones I needed as well. After a couple days of free time spent trying to get the correct options, I gave up and used what did boot -- Gentoo's genkernel package.

Since I ply mostly with FreeBSD systems, having a 'GENERIC' kernel was very familiar to me, and I wrote it down as a 'good move overall' for the future. However, after many moons of regular 'emerge --sync' and 'emerge -vuD world' success stories, I came across a update that broke my genkernel compilation. I couldn't find any hints from internet groups or google searches. Last week I broke down and decided to fix the damn thing myself.

The error was fleeting to say the least and NOT logged in /var/log/genkernel.log. The compilation always died after the kernel compilation and during the busybox compilation. busybox in genkernel is 1.7.4 and gentoo's genkernel ships several patches to it. The compilation would die during this patch process. The error would appear only on the terminal and would not be logged:

patch: **** File shell/ash.c seems to be locked by somebody else under Perforce

There was a whole bunch of other failures that would make this scroll by, so unless you happened to troll the backbuffer you wouldn't catch this as the failure. To it's credit the gentoo genkernel process does mention the failure to patch as the key failure.

So I simulated the patch process, I ran 'tar -xjpf /usr/portage/distfiles/busybox-1.7.4.tar.bz2' in /var/tmp and ran 'patch -p1 < /usr/share/genkernel/patches/busybox/1.7.4/1.7.4-ash-timeout.diff' from the extracted directory. Sure enough, same error.

It turns out that since permissions are extracted by default with the tar command under the build process, patch has '-g' ON BY DEFAULT (from the manpage):

-g num or --get=num
This option controls patch's actions when a file is under RCS or SCCS control, and does not exist or is read-only and matches the default ver-
sion, or when a file is under ClearCase or Perforce control and does not exist. If num is positive, patch gets (or checks out) the file from
the revision control system; if zero, patch ignores RCS, ClearCase, Perforce, and SCCS and does not get the file; and if negative, patch asks
the user whether to get the file. The default value of this option is given by the value of the PATCH_GET environment variable if it is set;
if not, the default value is zero if patch is conforming to POSIX, negative otherwise.

I was not amused. Anyway, I moved patch from /usr/bin/ to /usr/bin/ and wrote this wrapper in /usr/bin/patch:

/usr/bin/ -g0 $@
exit $?

genkernel now compiles.