Suricata

It looks like the current git master has 2 regressions after the http pipelining work was merged.

The first is related to the removal of the de_state_m lock and would be addressed by https://redmine.openinfosecfoundation.org/issues/1418

The 2nd is visible mostly with DNS traffic with many transactions. Here the new per-tx de_state logic is doing too many list walks. I'm thinking about keeping a "de_state" counter in the state to keep this under control.