Skip to content

concurrency.md

Latest commit

 

History

History
63 lines (42 loc) · 2.33 KB

concurrency.md

File metadata and controls

63 lines (42 loc) · 2.33 KB
Raw

Concurrency Model

Current Model

The scanner uses Boost.Asio coroutines to run many in-flight connection attempts while limiting parallelism with a simple queue and an inflight counter.

Port Scans

  • run_scans queues (address, port) pairs and spawns coroutines until max_inflight is reached.
  • Each coroutine performs one scan (TCP connect, TCP banner, or UDP) and calls the result callback.
  • When a coroutine finishes, it decrements inflight and launches more work.
  • A steady_timer acts as a latch to wait until all work completes.

Ping Mode

  • ping_loop performs a full scan cycle, tracks state changes, then sleeps for the configured interval before repeating.
  • Only changes are reported after the first pass.

ICMP Ping Mode

  • icmp_scan_hosts resolves/expands targets, then runs ICMP pings sequentially per address.
  • icmp_ping_loop repeats the ICMP scan cycle on an interval and reports changes.

Known Characteristics

  • Fairness: Ports for a host are processed in queue order; no priority between hosts.
  • Backpressure: max_inflight caps concurrent scans but does not adapt to network conditions.
  • Single io_context: All work is scheduled onto one executor.

Improvement Ideas

  1. Per-host rate limits

    • Limit concurrent scans per host to avoid hot-spotting a single target.

  2. Adaptive concurrency

    • Scale max_inflight based on recent timeouts/errors to avoid overwhelming slow networks.

  3. Address rotation

    • Round-robin across resolved addresses instead of scanning all ports per address in one batch.

  4. Work stealing queues

    • Use multiple queues (per host or per protocol) to reduce head-of-line blocking.

  5. Batch DNS resolution

    • Cache resolved addresses across ping cycles to reduce resolver load when --ping is used.

  6. Timeout tuning

    • Separate connect timeout from read timeout for banner mode to reduce false negatives.

  7. Parallel ping loops

    • Run ICMP and TCP/UDP scan cycles concurrently for mixed modes.

  8. Multi-threaded io_context

    • Run multiple threads with shared io_context for CPU-bound parsing or logging overhead.

  9. Structured metrics

    • Track per-host latency and error rates; feed into adaptive scheduling.

  10. Cancellation / stop signals

    • Add signal handling to stop in-flight scans gracefully.