Wireshark-dev: Re: [Wireshark-dev] [Wireshark-commits] rev 20908: /trunk/ /trunk/epan/dissector

From: Michael Tuexen <Michael.Tuexen@xxxxxxxxxxxxxxxxx>
Date: Mon, 26 Feb 2007 19:08:18 +0100
Comments in-line.

Best regards
Michael

On Feb 26, 2007, at 5:32 PM, Luis Ontanon wrote:

I got it clear now.

But I have thought of another way to relate both sides: TSN.

Other than the table indexed by [spt,dpt,vtag] from where we look for
known "half associations" we can have a table indexed by
(spt,dpt,highest_tsn_seen_so_far).

If we find an entry in that table that is equal to the
cummulative_tsn_ack of a sack's half association, its very unlikelly
that we get anything but the right other half.
You are assuming one thing:
You do not see multiple association using the ame TSN range.
This is true when
1. you have short captures
2. the initial TSN is selected by random

The second point is true for a lot of implementations but I know
of implementations which always start with TSN=0. This is fully
compliant with the SCTP spec.

I have traces where there's no way to correlate the IP of two different paths.

IP_A1->IP_B1 PT_A PT_B VTAG_AB
IP_A2->IP_B2 PT_B PT_A VTAG_BA

Are you sure? Are you talking about two end-points A and B
having port number PT_A and PT_B. So the situation should look like

IP_A1->IP_B1 PT_A PT_B VTAG_AB
IP_B2->IP_A2 PT_B PT_A VTAG_BA

RFC 2960 states that if a packet with DATA chunks is sent from IP_A1
to IP_B1 the corresponding SACK should be sent to IP_A1. But
this might not apply if both send continuously data and A uses
IP_B1 as its primary and B uses A_2 as its primary path. This is
valid. However, this would mean the A should send HEARTBEATS to
IP_B2. and you should be able to tie both halves together by these
messages.

So I think the following would be appropriate:
- consider triple od src port, dst port and v-tag identifying half associations.
- use the addresses to typ two half associations together by looking at
  at INIT-ACK chunks, DATA/SACK pairs, HEARTBEAT/HEARTBEAT-ACK pairs,
INIT-ACK/COOKIE pairs, COOKIE/COOKIE-ACK pairs. The pairing is based on
  IP-addresses.
This should be a pretty good heuristic.
If you are not able to tie two half associations together, you could try
to use DATA/SACK paris based on TSN, as you are suggesting.
But I would only do this if the above does not work out. This heuristic makes assumptions which are not covered by the RFC, but if you only us it when the rest is not working, it is better to do something which is possibly wrong,
than to do nothing.
on the other hand for all traces I have TSN values are different (by
far) for every direction on every association.

Is there any reason why this should not work?

n 2/24/07, Michael Tuexen <Michael.Tuexen@xxxxxxxxxxxxxxxxx> wrote:
Hi Luis,

see my comments in-line.

Best regards
Michael
On Feb 23, 2007, at 11:14 PM, Luis Ontanon wrote:

It's heuristic, not having the setup of the association.

I mantain two tables.
pl_table conatinig a list of assocs indexed by "port_labels" a 32bit
label out of the ports being used (low_pt << 16 | high_pt)
THis will break in scenarios where the same port is used on
both sides and on multiple associations. This is pretty common
on SIGTRAN szenarios where all sides use the registered port.

and plvt_table indexed by port_label and verification_tag of one
direction which I assume to be unique.
That is OK. Experience has shown that you can use the port number pair
and the vtag as an identifier for one direction of an association.

if match in plvt_table then we got it.

if match on pl_table then
   for each assoc in list
     if assoc is missing the other direction then
        we got this and add it to the plvt_table.

if no assoc was found so far
     this is a new assoc add it to both tables


I'm not sure it will always work, but so far (with the traces I have
available) it appears to do so... at least the perl prototype against
which I played text files derived from captures did.

I think what you need to do is the following:
- Identify one direction of an association by the pair of port numbers
   and the v-tag.
- Add information about the addresses to it while you are going through
   the trace file.
- Connect both directions based on IP-addresses. For example if you
   find DATA chunk from A -> B and a SACK from B->A, the port numbers
   are OK, then tie the two association directions together.

This is done (and more complex stuff) in the sctp related code
in the gtk directory.

AFAIU, there's very little chances to have two different associations
match... if I actually see it happening I'll start to play the
lottery!
 From what I understand this is pretty likely: You assume that there
in randomness in the port numbers. This is recommended in general but
not used in the SIGTRAN scenarios. It is pretty likely that
multiple association use all the same port number.

I have still problems matching the CTSN ack to the right TSN frames
without falling in an infinite loop but that's another story. And
serial arithmetic makes that a hard thing to deal with.

BTW, if you have captures where the counter cycles I would love to
have them. Or else I'll have to hope that an association on the lab
I'm working stays up long enough and does not catch me unprepared when it happens.Or I'll have to generate fake packets but my experience as
a telecom troubleshooter tells me that the fact that something works
with generated traffic does not imply it will work in the real world.
I think I can provide you with a trace. The BSD stack (which runs on
Mac OS X) has a socket option to set the Initial TSN for debugging....

As per Association Restart I do not think I'll ever implement it, I'll
treat the restarted Association as a new one (I need traces for this
too, but this given slack time in the lab I can force it to happen).
We consider it also a new association...

Luis.

On 2/23/07, Michael Tuexen <Michael.Tuexen@xxxxxxxxxxxxxxxxx> wrote:
Hi Lego,

I'm wondering how you tie together both directions of an SCTP
association?

Best regards
Michael

On Feb 23, 2007, at 8:57 PM, lego@xxxxxxxxxxxxx wrote:

http://anonsvn.wireshark.org/viewvc/viewvc.cgi?
view=rev&revision=20908

User: lego
Date: 2007/02/23 08:57 PM

Log:
 fix some bugs introduced in the latest releases and add
value_strings for param, evt, sig and stat ids s well as "sub-
parameters".

Directory: /trunk/epan/dissectors/
  Changes    Path                     Action
  +39 -33    packet-h248.c            Modified
  +20 -14    packet-h248.h            Modified
  +103 -39   packet-h248_3gpp.c       Modified
  +4 -4      packet-h248_annex_c.c    Modified
  +83 -30    packet-h248_annex_e.c    Modified
  +23 -11    packet-h248_q1950.c      Modified
  +486 -52   packet-sctp.c            Modified

Directory: /trunk/asn1/h248/
  Changes    Path                      Action
  +36 -30    packet-h248-template.c    Modified
  +20 -14    packet-h248-template.h    Modified

_______________________________________________
Wireshark-commits mailing list
Wireshark-commits@xxxxxxxxxxxxx
http://www.wireshark.org/mailman/listinfo/wireshark-commits




--
This information is top security. When you have read it, destroy
yourself.
-- Marshall McLuhan


--
This information is top security. When you have read it, destroy
yourself.
-- Marshall McLuhan
_______________________________________________
Wireshark-dev mailing list
Wireshark-dev@xxxxxxxxxxxxx
http://www.wireshark.org/mailman/listinfo/wireshark-dev

_______________________________________________
Wireshark-dev mailing list
Wireshark-dev@xxxxxxxxxxxxx
http://www.wireshark.org/mailman/listinfo/wireshark-dev



--
This information is top security. When you have read it, destroy yourself.
-- Marshall McLuhan
_______________________________________________
Wireshark-dev mailing list
Wireshark-dev@xxxxxxxxxxxxx
http://www.wireshark.org/mailman/listinfo/wireshark-dev