[c2k8]: Hackathon Summary Part 6

c2k8 General Hackathon (Part 6) - June 7-15, 2008, Edmonton, Alberta, Canada When you have a developer like this on the team, even if you don't know anything about OpenBSD, you know that there must be something special about it to attract the likes of him.

Read on to learn why this developer is so unique:

Can Erkin Acar (canacar@) has a really beautiful name. His cvs account name is just as nice. It's normal for everyone in the developer community to call or reference each other by their cvs account name. I've heard some refer to one developer as "unpronounceable"; referring to his really difficult to pronounce cvs account name. Anyhow, I was curious how to pronounce Can's real name. It turns out that "Can" is pronounced somewhat like "John" as "C" in Turkish is more like "J" in English. I'm sure that developers will still call him "canacar" with a hard "C", but at least now, they know how to pronounce his first name properly :-).

Can has been around for a long time and besides his unique name, so far he and Okan Demirmen (okan@) are the only two Turkish developers in OpenBSD. I often wonder why we don't have more developers from other countries. I'm not referring to smaller countries but the larger ones such as China and India with large populations. Obviously, it helps to be able to communicate in English. I'm not sure about China, but I know that most educated people in India learn English. They even have a thriving IT sector. Granted, the OpenBSD developer and user community may not be the most maternal of communities but there must be other reasons.

The OpenBSD developer community have the following mantra: "shut up and hack(code)". Can articulated this nicely here when he said, "...OpenBSD does not run on requests, or polls or democracy." It is a meritocracy of sorts. People who use or follow OpenBSD know this quite well. However, it may be somewhat difficult for people new to the Project, and especially so if they are not very comfortable communicating in English. It gets even more complex when one considers cultural differences. The n2k8 hackathon was the first mini-hackathon in Asia and a great example of what an Asian cultural exchange can be like. I look forward to more of them; cultural exchanges and more code :-).

Of the most notable things that Can brought to OpenBSD was pftop and a lot of work on kernel pppoe(4), acpi(4) and tcpdump(8). As you will read below, Can is all over the place improving OpenBSD. Here is what Can had to say about his time at the c2k8 hackathon:

When I am not slacking, I work on bits and pieces all around the kernel and userland. This hackathon was no different, and I had great time meeting with old and new friends, and doing some fun and enjoyable work. Below are the main things I worked on during the hackathon.

== ACPI ==

I started working on the ACPI code three years ago. At that time my main motivation was to get the wireless button on my MSI laptop working. It was one of the earlier versions of the ACPI code and Jordan Hargrave (jordan@) and Marco Peereboom (marco@) rewrote the parser two more times since then.

The final version of the parser was committed just before the hackathon and it worked much better then the previous versions. The first thing I did at the hackathon was to hunt for some memory leaks. I managed to track down and fix some of it, and when Jordan arrived, a couple of days later, he already had a fix for the remaining leaks.

With the memory leaks out of the way, Jordan worked on and committed some interpreter optimizations that prevent nested conditionals to use a lot of scopes. Gordon Willem Klok (gwk@) worked on improving CPU frequency settings through ACPI. I started tracking down some brightness control issues, Marco fixed some other remaining bugs, and on the last day it was decided that the parser was stable enough. Work has now started on suspend/resume support. As Bob loves to say, "suspend is easy, resume is the problem".

Since I worked on kernel BPF packet capture interface and the tcpdump program, I usually end up being the point of contact for tcpdump related issues. Todd Fries (todd@) reported a problem where tcpdump was reporting "truncated-ip6 - 48 bytes missing" when listening to IPv6 traffic on the enc0 interface (which is used for IPSec).

I could not find any problems on the bpf->libpcap->tcpdump path. Once Todd set up an ipv4-in-ipv6 IPSec tunnel for testing, it was relatively easy to trace the issue to the following fragment of code in the kernel IPSec code in netinet/ipsec_input.c

@@ -427,8 +427,7 @@
  		}

   		ip6 = mtod(m, struct ip6_hdr *);
-		ip6->ip6_plen = htons(m->m_pkthdr.len -
-		    sizeof(struct ip6_hdr));
+		ip6->ip6_plen = htons(m->m_pkthdr.len - skip);

  		/* Save protocol */
  		m_copydata(m, protoff, 1, (unsigned char *) &prot);

However, this did not fix the problem. No change at all. After many more attempts, I managed to spot this fragment a few lines earlier in the code:

@@ -412,7 +412,7 @@

  #ifdef INET6
  	/* Fix IPv6 header */
-	if (af == INET6)
+	if (af == AF_INET6)
  	{
  		if (m->m_len < sizeof(struct ip6_hdr) &&
  		    (m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {

We basically looked at this fragment hundreds of times without realizing that the check was not using the correct #define. We traced this error back to the first version of the file, committed about 8.5 years ago:

   1.1  (angelos  09-Dec-99):  if (af == INET6)

The incorrect condition skipped some extra checks on the IPv6 headers and was only really visible in tcpdump. The inner packet made it into the stack just fine, minus a few sanity checks. A quick search on the whole kernel for similar INET and INET6 misuse did not find any other issues.

== Systat ==

A long time ago, I have written a utility called 'pftop' for realtime visualizing the 'pf' packet filter in OpenBSD. I continue to maintain and add features to it. It has a quite flexible text-based display code that can be used to present information in tables interactively.

More recently, I did some work to modify the systat(1) utility to use the display engine of pftop. The result was a systat implementation that can also display 'pf' states, and contains more 'modern' features such as the ability to scroll lists.

At this hackathon, I got an OK from Theo to commit my systat changes. It is now a part of the base system. It took about 2 minutes before I got my first bug report. It is still under development, and not really 100% compatible with the previous systat commands. I will be working on it to add the missing functionality, and new features.

== APS sensor ==

IBM/Lenovo Thinkpad computers which are quite popular among OpenBSD developers feature an Active Protection System (APS) that shuts down the hard drive when the notebook is shaken or dropped. This is done by accelerometers inside the laptop. The interface to the accelerometers was reverse engineered some years back by some engineers at IBM (I have no idea why they had to reverse engineer their own product).

From these specifications, a driver for the accelerometer was written for Linux, and later for BSDs.

The OpenBSD driver did not work on my T61, and I was meaning to look at it for some time. I found the opportunity on the last day, and using some documentation referenced from the Linux driver, I updated our driver to correctly use the underlying embedded controller interface, and also fixed the detection of the newer Thinkpad models.

This work is not yet committed, mainly because I need to test its interaction with the ACPI which also accesses the same embedded controller to do its work.

Can

I would like to thank Can for bringing us pftop. He continues to fix bugs besides adding features in the kernel and in userland. Maybe one day, we may be able to enjoy a mini-hackathon in Turkey ;-).

(c2k8 hackathon summary to be continued)