What was the hardest part, Range header handling or the slowloris mitigations? Both seem like they'd be a nightmare without higher-level state machines.
slowloris mititgations actually weren't too bad, just a couple syscalls to setitimer(), sigaction(), and setsockopt(). Range header parsing was awful, so was content length. I'd say all in all, probably Range headers. Just string parsing in general is pretty awful in assembly.