[Asterisk-Dev] Multiple IP's For sip.

[Tilghman Lesher]

On Wednesday 14 July 2004 12:57, Wolfgang S. Rupprecht wrote:
> tilghman at mail.jeffandtilghman.com (Tilghman Lesher) writes:
> > On Tuesday 13 July 2004 14:07, Alberto Fernandez wrote:
> > > Correct In sip.conf i bind the ip to 0.0.0.0 and it still doesnt let
> > > the phones register. Im not folowing what you are trying to tell me.
> > > Please elaborate.
> >
> > I suspect I have already been too elaborate, and you actually want a more
> > simplistic explanation:  the IP that Asterisk uses for SIP is based upon
> > your route table.  If there is no explicit route for a destination, then
> > Asterisk will use the device associated with the default route to specify
> > that IP.
>
> I think he might be seeing the same bug I see here, but haven't
> bothered to track down further.  When Asterisk does a wildcard bind,
> something screws up.  I solved it by the expediency of simply choosing
> my outside interface and limiting asterisk to binding to that one.
>
> If asterisk really does choose (or let the kernel choose) the source
> IP based only on routing then I think I see the problem.  Asterisk
> must take the src/dest addresses from the incoming packet and flip
> them around (along with flipping the port numbers).  Doing otherwise
> prevents the other side from associating the reply packet with its
> query packet.

It does.  I wrote the code.  See acl.c.  And the only problem with your
approach is that Asterisk may or may not have corresponding incoming
packet headers when it chooses to send a packet to a remote host.  What
should Asterisk do in that case?

And this masks the real problem, here:  the SIP protocol has a braindead
way of operating:  namely, by encoding data from a lower level protocol
into a higher level protocol, which, as you correctly pointed out, breaks
stuff.  There's an interesting document that came out from ICANN this week
which said much the same about Verisign's SiteFinder, but I digress.

-- 
Tilghman