The most significant changes are:
- Use UMA zone instead of own chunk of memory.
- Lock each hash entry separately.
- Expire items "actively" - interrupt method can expire flows
from hash slot, when it searches through it.
- Remove global tailqueue. Make callout thread search through
every hash slot.
- Export datagram is detached from private data and filled. If
it is incomplete, it is attached back. Another thread will
continue working with it.
Lesser, but also important speedups:
- Flows in hash slot are stored in tailqueue. Whenever a flow is
hit, it is moved to the begging, so it can be located quicker.
- When callout thread works with hash slot it bails out if
slot mutex is contested.
