[Opalvoip-devel] Bug in rtp.cxx

[Southanon R. <sra...@la...>]

There seems to be a bug in rtp.cxx that causes the jitterLevel to jump 
up in some cases. The bug is seen when the difference in time goes from 
large to small to large again. This seems to be present in the opal 
3.10.9 release.

This is the line of the bug in rtp.cxx, function is Internal_OnReceiveData.

         // As per RFC3550 Appendix 8
         diff *= GetJitterTimeUnits(); // Convert to timestamp units
         long variance = diff - lastTransitTime;

Both "diff" and "lastTransitTime" are unsigned. If lastTransitTime is 
bigger than diff, this produces a very large unsigned number. Not the 
small negative number we are expecting.

Here is sample output demonstrating what is happening:

   0:11.293        RTP Jitter:0xe3494910 rtp.cxx(1346)  RTP     Session 
1, ssrc=3735373445, jitterLevel = 21
   0:11.318        RTP Jitter:0xe3494910          rtp.cxx(1305) RTP     
Session 1, ssrc=3735373445, tick = 170:25:21.136, lastReceivedPacketTime 
= 170:25:21.110, diff = 26
   0:11.318        RTP Jitter:0xe3494910          rtp.cxx(1324) RTP     
Session 1, ssrc=3735373445, diff = 208, lastTransitTime = 152, variance 
= 56, variance2 = 56

   0:11.318        RTP Jitter:0xe3494910          rtp.cxx(1346) RTP     
Session 1, ssrc=3735373445, jitterLevel = 76
   0:11.333        RTP Jitter:0xe3494910          rtp.cxx(1305) RTP     
Session 1, ssrc=3735373445, tick = 170:25:21.151, lastReceivedPacketTime 
= 170:25:21.136, diff = 15
   0:11.333        RTP Jitter:0xe3494910          rtp.cxx(1324) RTP     
Session 1, ssrc=3735373445, diff = 120, lastTransitTime = 208, variance 
= 4294967208, variance2 = -88

   0:11.333        RTP Jitter:0xe3494910          rtp.cxx(1346) RTP     
Session 1, ssrc=3735373445, jitterLevel = 4294967279
   0:11.353        RTP Jitter:0xe3494910          rtp.cxx(1305) RTP     
Session 1, ssrc=3735373445, tick = 170:25:21.170, lastReceivedPacketTime 
= 170:25:21.151, diff = 19
   0:11.353        RTP Jitter:0xe3494910          rtp.cxx(1324) RTP     
Session 1, ssrc=3735373445, diff = 152, lastTransitTime = 120, variance 
= 32, variance2 = 32

Notice the jump in the variance value when diff is smaller than 
lastTransitTime. variance2 is the same calculation with lastTransitTime 
type casted to a long. Since this variance is part of the jitterLevel 
calculation, this subsequently causes the jitterLevel to jump in value 
dramatically.

Best Regards,
Southanon