Possible IP fragmentation issue

[kenmcmil]

I'm using the Cisco 3725 model in GNS3. I observed an odd behavior with IP fragmentation when experimenting with randomly generated IP datagrams. The input to the router was a fragmented datagram with these fields:

more = 0
don't fragment = 1
offset = 1

The forwarded datagram had these fields:

more = 0
don't fragment = 1
offset = 0

In other words, the fragment was converted to a whole datagram with only part of the data.

Arguably the input datagram should never have been sent. My question, though, is whether this behavior accurately reflects the behavior of the actual Cisco hardware, or whether it is an artifact of the model.

[grossmj]

I would say it is very likely this is because of Dynamips and it is not the behavior of the actual Cisco hardware. It would be interesting to try with the Cisco IOSv appliance to compare results.

How have you generated the IP datagrams?

[kenmcmil]

That's a good thought. I could try setting that up.

I am using a custom IP test generator running in a docker container. Would it be helpful to post pcap files for the test?