On Dec 21, 2007, at 4:11 PM, Abhik Sarkar wrote:
I am facing an issue with dissection of SMPP PDUs with invalid length
(though the matter might effect any other dissector which use TCP
reassembly too).
...as well as affecting the receiving SMPP implementation. :-)
In my test, I send two PDUs in two different TCP frames each of length
40.
The term that's used is "TCP segment".
TCP is a byte-stream protocol, so it doesn't expose a notion of
"frames" to protocols running atop it - the receiving system should
operate the same regardless of whether you sent:
one TCP segment with length 80;
two TCP segments with length 40;
80 TCP segments with length 1;
etc..
However, the PDUs have their PDU length field incorrectly set to
41. With reassembly on in this case, the TCP dissector "steals" one
byte from the second frame for the dissection of the first SMPP PDU.
However, because the second PDU has now lost the first byte, it is no
longer recognized as a valid SMPP PDU and hence is not dissected.
With TCP reassembly turned off, the behaviour is that each frame is
dissected as an SMPP PDU, but as malformed PDUs.
Is this something to be expected,
Yes. The packets are being interpreted the way the SMPP spec says
they should be, i.e. there's a 41-byte message split between two TCP
segments, with the rest of the second TCP segment being the first 39
bytes of another 41-byte message.
