[pjsip] Bug in sdp_neg/match_offer with prefer_remote_codec=0

Régis Montoya r3gis.3r at gmail.com
Fri Oct 7 07:17:54 EDT 2011


Hi,

I found a bug in sdp_neg.c "match_offer" method when you disable 
prefer_remote_codec_order (set PJMEDIA_SDP_NEG_PREFER_REMOTE_CODEC_ORDER 
to 0).
(Tested on pjsip-1.x but regarding code should also affect pjsip-2.x)

In this case master and slave pjmedia_sdp_media are inverted.
As consequence, when tests are done on rtpmap the or_ parameter here 
http://trac.pjsip.org/repos/browser/pjproject/trunk/pjmedia/src/pjmedia/sdp_neg.c#L1255
can be the one generated by pjsip and lr the one of the offer.

In this case the test made at this line is not valid cause obviously 
or_.param will be empty in case of a single channel codec (pjsip does 
not add channel counts if 1 which is fine, but other side could add...).
So test will fail if the other side add the channel count 1, while it 
should have succeeded because we have something like remote AMR/8000/1 
and local AMR/8000 .

So, in this case if remote party add channel count in rtpmap and 
prefer_remote_codec_order is disabled, it will result in unexpected 
behavior.

On my side I did a quick fix :
-------------------------------
             pjmedia_sdp_rtpmap remote_r = or_;
             if(!prefer_remote_codec_order){
                 remote_r = lr;
             }
             if (!pj_stricmp(&or_.enc_name, &lr.enc_name) &&
                 or_.clock_rate == lr.clock_rate &&
                 (pj_stricmp(&or_.param, &lr.param)==0 ||
                  (remote_r.param.slen==1 && *remote_r.param.ptr=='1')))
             {
....
-------------------------------
But I don't know if it is the best way to do that (maybe a pointer to 
param would be a better idea?).

Regards,
Régis



More information about the pjsip mailing list