At 08:39 PM 8/28/2003, Gavin Bensom wrote:

Here are the files mentioned. Also, Linux kernel 2.4.20-6smp ser 0.8.11pre29 - main.c, v 1.161

Thanks for the report.

It looks like a fairly stupid Windows messenger's spec violation, potentialy two of them. Bill Gates is actually the proper contact person in this case.

See html-ized dumps at http://www.iptel.org/~jiri/etc/ms/invalid/messenger50.html. What happens is gnoah(.49, RTC 1.2) sends an INVITE to bkern (.128, RTC 1.2) through ser (.49). That's a Microsoft's proprietary spec violation -- there is no standard for messaging sessions using INVITE. It can only work between two messengers.

Now the surprising part of it, that it does not work even between two messengers. The recepient instantly accepts the session (200) and terminates it (BYE). When the sender's MESSAGE is sent, it already reaches a terminated session and is replied with 481.

We will try to see if we can align SER in a way which will make Messenger happy, but I can't promise it -- we can only try guessing how to do it. SER's proyxing mode is spec-compliant and very conservative -- the proxied messages are modified only very slightly. My first guess is that Messenger uses some invalid scheme for validation of Record-routes -- try turning record-routing off (comment "record_route") and see if it works. If not, downgrade to pre-5 version of Messenger -- this is really broken.

-Jiri

ps -- in your script, you apply an aliases lookup. You should do so before the location lookup. The place you have it in here now would cause uri translation even for requests outside your domain, which is not correct.