We have a memory leak in the core module.
When there is fragmentation involved, sk_buffs are queued. When translation is successful, Jool asks the kernel to drop the packet. Because the kernel doesn't know the sk_buffs are queued, it only frees the first one.
This is that one bug that only applies to 3.2.0. It does not apply to the 3.1 series.
By testing the checksums again, I realized that there's no reason to not update inner packet checksums anymore, and wireshark would judge Jool's packets better if we do.
Therefore, I'm going to refactor Translate a bit again. Checkpoint.
Merging version 3.2.1 into master, hereby makingthe changes official.
Version 3.2.1 is 3.2.0 with issues #57, #106, #108 and #109 fixed.
Issue #107 has been marked as duplicate and postponed to 3.3.0.
rting with '#' will be ignored, and an empty message aborts