Rainer Gerhards

Long time no post ;) It has been busy days, with finaly a healthy discussion on the IETF syslog-sec mailing list. Still, there are (too) few participants, but it looks like the recent events got the group some momentum. The WG is now in danger of being shut down and that seems to drive activity. A new charter is being discussed. It looks like the rejection of previous work will lead to a really good alternative. It is still too early to be sure all will have a good outcome, but in my opinion it looks more promising than any time the past month – especially if you think about a spec becoming implemented.

Sam Hartman (IETF Security Area Director) has rejected the syslog-protocol draft due to missing support in the last IETF meeting. I do not yet know which new non-concensus turned up. I fear it is an re-iteration of arguments already exchanged. I am very curios to have a look at the minutes. Anyhow, if it is yet another re-iteration, I seriously begin to doubt if that activity makes any sense at all… Maybe it is a much better idea just to create some simple TCP-based syslog format, talk to the other implementors… and do it ;)

I’ve talked with a lot of people about rfc3195 to lots the past days. I’ve a mixed feeling. Since spring, rfc 3195 is getting momentum. On the other side, the IETF syslog-sec WG is considering removing some parts from RFC 3195 (namely the COOKED profile). The adoption rate in practice is also very low…

Anyhow, the discussions indicated that a lot of folks seem to work on rfc 3195 (well, “a lot” in my terms…), but most of them somewhat isolated. I will now try to solve this issue with a new mailing list. Maybe we can even get some IHE folks onboard.

The list charter is as follows:

###
The rfc3195 list is targeted towards people interested in RFC 3195-based solutions. It is primarily aimed at implementors, protocol-designers and operators who would like to have insight into the protocol and the various implementations. It carries deeply technical content about protocol interpretations, interoperability of different RFC 3195-based solutions, and discussion about the future of RFC 3195. It also covers news and annoucements about RFC 3195-related projects and products. These items should not be marketingish but rather help inform the community of new arrivals and other important events.
####

Subscription information is available at

http://lists.adiscon.net/mailman/listinfo/rfc3195

I hope this is a useful tool for the community. Let’s see how it evolves.

Today, I have released rsyslog 1.11.1. It now supports BSD config file extension for hostname and program blocks. Initialy, it caught me by surprise that BSD has such considerable extensions. Actually, it looks like I picked the wrong starting point with sysklogd. BSD’s syslogd is much more capable (e.g. it has IPv6!) and the code looks somewhat cleaner. Anyhow, now we are too far in the game to reconsider anything. Plus, there is not much code left from the orginal sysklogd. In lines of code, I think rsyslogd has roughly 2.5 times as much code then sysklogd (hopefully not only consisting of bugs ;)).

Today, I’ve also brought over some patches from sysklogd to those remaining code pieces in rsyslogd. The sysklogd team is *very* conservative with updating the package. In its CVS, there are a number of non-intrusive yet slightly important patches which have not yet made it into the source. The “current” version is now rougly 2 years old… Anyhow, I think I have finally build a considerably more capable and also reliable syslogd than what I started with. It’s still a way to go, but I begin to like what I have done. :)

I am a bit late in posting this. After “last call”, syslog-protocol went to AD review. On September 19th, 2005, Sam Hartman has replied with a number of very good questions and suggestions. Obviously, that has pushed us a bit back, but it also has fingerpointed to some very important issues. While I am not happy that syslog-protocol will probably take much longer to complete, I am happy to get it as good as possible. Sam has also mentioned that he will have it reviewed by a number of operators, which is a big plus given the sparse feedback we had so far from that community.

In the mean time, I have looked at most issues on Sam’s list and done some text changes. I am right now finalizing on Unicode security, the last thing with outstanding suggestions and feedback by the WG. Hopefully, I’ll be able to find consensus some time next week.

Chris has called for a meeting at the next IETF (in November), and I hope to meet the October 24th draft submission deadline. Then, the meeting can hopefully reach decision. It’s a shame I can’t attend, but it is hardly justifiable for Adiscon to travel to Vancouver for an one hour slot…

An interesting fact is that Chris also has put re-consideration of RFC 3195 on the agenda. It is only very slowly getting implemented. My personal view is that it is too early to dump it, implementations are now beginning to be seen. I am most interested how this continues…

Long time, no post. But now I am really excited. Finally, I have managed to get RFC 3195 functionality into rsyslog. The 1.11.0 release contains just the listener (aka “server ;)), and the code definitely can be improved. But, after all, this is a big step for rsyslog. Esepcially, if you consider that the original project goals called for immediate implementation of RFC 3195. Even the name – rsyslog – stems back to RFC 3195 (reliable syslog!). Well, it turned out that other needs were in much more in demand, and so RFC 3195 was postponed and postponed…

Still, we do not have the initiator (sender) and not the greatest code. But at least we can now see if there is growing demand. I expect so, but only slowly. I will see that I can integrate the initiator shortly, but I will first have a look into multithread-enabling rsyslogd. That would facilitate some of the RFC 3195 (synchronous) requirements and it probably is also needed to implement openssl with sufficiently performance (and thus low to no message loss).

But for now, I am happy with where I am arrived ;)

Finally, I have been able to do some major feature editons. Now the development branch somewhat gets off. See the change log for rsyslog 1.10.0 to see what I mean ;)

I have to admit that creating this release was more effort than I initially estimated. While I was right that the base design allowed neat integration of the new filtering, a lot of infrastructure (counted strings, parser, …) had to be added. Of course, these additions will help in the future, too. But as I said… it took much longer than expected. I ended up adding 1,000 lines of code and also modifying others. So roughly about 1/6th to 1/7th of the project needed to be changed to make this feature a reality…

It is finally done: rsyslog 1.0.0 is released. This is the first (officially ;)) stable branch. I’ve taken some weeks of close-to-no modifications to make sure as many bugs as posisble have been found and removed. I am very positive this release is really stable. I also hope that the availability of a stable release will help grow acceptance of rsyslog. All in all, we are now scoring within the top 5,000 projects on freshmeat.net (by popularity), which I think is not bad for such a new project in the logging area. I also already got some helping hands (more soon), so this is also an excellent sign.

Now that 1.0.0 is out, I will begin new feature enhancements – which will bring even more usefulness tp rsyslog.

I have finally begun preparations for the rsyslog 1.0 stable release. I’ve checked all bug trackers, done some minor doc updates and other such things. Made some very minor changes. As I heared no bad feedback after the 0.9.8 release (now two days ago), I do not accept any more troubles. So I think I am basically set to release by the end of the week or monday.

I have now branched off a stable branch in CVS and think I won’t need to modify it. Now I can turn back to new features. I will start by implementing a core counted-string library, which will hopefully enhance the security ones it is used all over rsyslog. To be honest, that will probably take quite some while. And there is also another reasoning for the library: to support syslog-protocol, I must also support strings with embedded characters, which makes it impossible to use with the standard library routines. So it is nice to even get some more security out of a thing that I need to do anyhow…

After some silence, I have released rsyslog 0.9.8 today. I was on vacation the last two weeks and used this time to see how 0.9.7 worked in the field. No bad news, but some user activity. So the problems with the new make procedure seem to be solved. I still had some minor changes, which I didn’t like to apply before going on vacation (I had email with me, but… ;)). So I released these few changes today as 0.9.8. I think they won’t break anything. If that assumption is right, we will have a final 1.0 within the next days :).