[Serusers] cancel and 487 relay

[Jiri Kuthan]

UAC must receive a 487 on CANCEL, that's correct. What is not ideal is that
we generate 487 localy. That helps some issues with some broken UASs but
introduces an additional race condition. the dev branch does not do that.
See a recent thread with Maxim on that.

-jiri

At 10:29 AM 10/18/2004, Richard wrote:
>Hi Jiri,
>
>I probably didn't make myself clear. In this case, it is not really a race
>condition.
>
>It happens when one user picks up a branch. Ser sends out CANCEL to all
>other branches. The issue is that ser propagates the 487 to the upstream
>caller. It should absorb the 487 because it is a ser generated branch. And
>the 487 to the caller can disrupt the call depending on the UA
>implementation.
>
>Thanks,
>Richard
>
>
>> -----Original Message-----
>> From: Jiri Kuthan [mailto:jiri at iptel.org]
>> Sent: Sunday, October 17, 2004 5:55 PM
>> To: Richard; 'ser users'
>> Subject: Re: [Serusers] cancel and 487 relay
>> 
>> 
>> At 11:33 PM 10/17/2004, Richard wrote:
>> >Hi,
>> >
>> >I ran into some issues. Hope that someone can enlighten me. When ser does
>> a
>> >parallel forking, it uses its uac to initiate a new call. The scenario is
>> >Phone A calls phone B via ser. In ser, it forks into two branches, one
>> for
>> >phone B and one for phone C. To call phone C, ser's uac initiates the
>> call.
>> >When Phone B picks up the call, ser sends a CANCEL to phone C and cancesl
>> >the call. The following diagram illustrates the sip call flow. For
>> >simplicity, I didn't include any message involving Phone B.
>> >
>> >Packet #    Phone A ------- (ser UAC) -------- ser --------- Phone C
>> >0               |     INVITE ---------------->  |               |
>> >1               |               |   INVITE ->   |               |
>> >2               |     <----------- 100 trying   |               |
>> >3               |               |               |  INVITE ->    |
>> >4               |               |               | <- 100 trying |
>> >5               |               |               |  <- 183       |
>> >6               |     <-------------- 183       |               |
>> >7               |               |   CANCEL ->   |               |
>> >8               |               |<- 200canceling|               |
>> >9               |               |               |   CANCEL ->   |
>> >10              |               |               |    <- 487     |
>> >11              |               |               |  <- 200 OK    |
>> >12              |               |               |    ACK ->     |
>> >13              |               |    <- 487     |               |
>> >14              |               |    ACK ->     |               |
>> >15              |               |               |    ACK ->     |
>> >16              |    <---------------  487      |               |
>> >17              |     ACK --------------->      |               |
>> >18              |               |               |    ACK ->     |
>> >
>> >Per rfc, 487 and its ACK are hop-by-hop, i.e. ACK is generated at each
>> hop.
>> >Ser correctly does it at Packet 12. The problem happens after packet 14.
>> >Somehow ser doesn't think this ACK of Packet 14 is an acknowledgement of
>> >Packet 13.
>> 
>> that's strange. I don't know why it is this way without seeing the
>> packets. (I have to admit that even if I received them, my current
>> worload would not allow me to study them.)
>> 
>> > So it is forwarded to Phone C. Not sure if I can use ser.cfg to
>> >absorb it without forwarding. Anyway, not a big deal for an extra ACK.
>> The
>> >big problem is that somehow ser relays the 487 back to Phone A. It
>> doesn't
>> >seem right. In Packet 10, 487 is to cancel the original INVITE at Packet
>> 3
>> >which has a record-route of Phone A, ser. So 487 has the same "Via" field
>> >back to Phone A. When this 487 is relayed back to Phone A, it already
>> gets
>> >the 200 OK from Phone B and has an active conversation.
>> 
>> well that's a race condition -- if caller cancelled, then the 487 should
>> propagate to UAC. If at the same time one of the called parties answered,
>> the UAC will see the 200 too. The case that caller hangs up at the same
>> point of time when called party happens simply happens.
>> 
>> > Depending on the
>> >implementation, Phone A may choose to close the call because of the 487.
>> >
>> >Any help is highly appreciated.
>> >
>> >Thanks,
>> >Richard
>> >
>> >
>> >
>> >
>> >
>> >
>> >_______________________________________________
>> >Serusers mailing list
>> >serusers at lists.iptel.org
>> >http://lists.iptel.org/mailman/listinfo/serusers
>> 
>> --
>> Jiri Kuthan            http://iptel.org/~jiri/

--
Jiri Kuthan            http://iptel.org/~jiri/