On Fri, Oct 1, 2010 at 2:18 PM, Sake Blok
<sake@xxxxxxxxxx> wrote:
On 1 okt 2010, at 19:53, Fulko Hew wrote:
> Imagine my surprise when Wireshark failed to decode the
> AgentX protocol inside some captured packets. It all
> depends on where the packets originated from (which OS).
>
> Attached are two capture sessions of AgentX traffic.
>
> One decodes... Between a Linux box and a Linux box.
> One doesn't... Between a Windows box and a Linux box.
>
> I'm not sure what triggers the failure, but in one case
> Wireshark successfully decodes the AgentX traffic inside
> the TCP PDU and in the other case it doesn't. The top
> protocol window (when it doesn't decode) also tags the
> packets as '[TCP segment of a reassembled PDU]'
The difference is that in the non-working example, there is a flag that indicates that multibyte values are in BigEndian representation and the agentX dissector does not seem to honor this. When it then sees "00 00 00 20" as length, it does not interpret this as 32 bytes, but as 536870912. So then it tries to read that many bytes to reassemble the PDU. Of course it fails at that.
Could you please open a bug report at http://bugs.wireshark.org and attach the two tracefiles so that we don't lose track of it?