IP fragmentation corrupted in NAT network

[Oleg]

I am currently working on DTLS implementation, and I am using Ubuntu 11.04 guest OS under Windows 7 host OS. My VirtualBox version is 4.0.12r72916.

I stumbled upon a very annoying networking problem with IP fragmentation. I am using the NAT network option, and the incoming large IP packets (larger than external network MTU) from the outside world are not getting delivered to the guest OS correctly. Only the very first packet in the IP fragmentation sequence has the correct remote address; the rest of the packets have bogus remote address 10.0.2.2 (the default "internal" gateway IP). I checked the VirtualBox support history and I found that this exactly bug was discovered and fixed a while ago (version 4.0.4):

​http://www.virtualbox.org/ticket/8701

But I am seeing it now, version 4.0.12, Linux guest OS.

There are also some problems with egress IP fragmentation, but I am still discovering it.

Does anybody know about this problem ? Is there a fix or workaround ? So some reasons, I can use only NAT network option, and this problem is very annoying for me. I cannot exchange any large UDP packets with my outside server larger that MTU (1400+ bytes).

Thanks ! Oleg

Primary OS: MS Windows 7 VBox Version: OSE other Guest OSses: Ubuntu

[Oleg]

I installed 4.1.0, and I found that this problem is fixed in the new release. Thanks !

[Oleg]

I ran more testing on IP fragmentation functionality in 4.1.0, and I actually found another problem NAT networking. The UDP packets larger than 8K are not getting delivered correctly - the "tail" of the packet, beyond 8K size, is corrupted. I send a packet from Linux guest OS to another system, connected by the Ethernet cable, and the packet is delivered without errors, and seemingly correctly, but the content of the post-8K bytes is incorrect. When I try the same test with bridged networking, or without virtualization, then the packet is fine, so this is a NAT-only problem.

[Valery Ushakov]

As you confirm in comment:1, fragmentation issue was fixed (#7711 and #8701).

The 8KB limit should be fixed in the the forthcoming 4.3.16. Details are in bug #12136. Closing this one as a duplicate.