[Devel] Parse Errors, error_route and fixing common NAT ALG bugs

[Aron Rosenberg]

This is a discussion question mostly, but part bug report. Our
application targets general consumers and as such we see a huge amount
of "broken" SIP message traffic come to our servers. These are packets
where "friendly" NATs/routers try to fixup the SIP traffic to undo the
NAT behavior. While most of these routers do the right thing, we have
noticed several that have bugs in their implementations. These bugs
manifest themselves as a mangling of the SIP headers in one or another.

 

The major issue that we see in consumer routers is that the Via: header
has been reconstructed incorrectly. The OpenSER bug that we see here is
that an incorrect Via header does NOT result in error_route being
invoked, nor does it result in a 400 error packet being sent to the
originator either.

 

The discussion point is should OpenSER try to "fix" these broken routers
so that the traffic can be used correctly. Two of the most common errors
that our servers see are below

 

Issue 1. The via header has a port that is totally out of range. This
error comes from the default DSL modem that Embarq / Sprint.net ships to
their customers.

 

Via: SIP/2.0/UDP
1.1.1.1:13088001;branch=z9hG4bK-d87543-0865e7125c508513-1--d87543-;rport

 

 

Issue 2. The via header originally had a rport param from our client,
but the router stripped it off incorrectly, notice how it ends in just a
; We believe this error is caused by Linksys WRT54G routers that don't
have recent firmware. 

 

Via: SIP/2.0/UDP
1.1.1.1:56261;branch=z9hG4bK-d87543-9b572d31e0515e44-1--d87543-;

 

In both of these cases the NAT / router is changing the existing VIA
header rather than adding a new VIA line

 

 

The question is should OpenSER allows these errors to proceed to the
script layer? Right now it just swallows the packet.

 

Issue 1, if allowed to reach the script layer would be fixed by our call
to nat_uac_test("23") and fix_nated_contact(). 

 

Issue 2, I am not so sure what the behavior would be if it reached the
script layer.

 

Thoughts?

 

-Aron Rosenberg

SightSpeed Inc.

http://www.sightspeed.com <http://www.sightspeed.com/> 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://openser.org/pipermail/devel/attachments/20070311/66627efa/attachment.htm