Repository navigation
Changelog
1.4.0
Faster scans of large ranges (/16 and up) and scan timing in the output.
Added
- Scan start time, end time and durations (scan, saving results, total) plus rate in units/s
(a unit is onehost:port) in the output. --concurrency(default1000, lowered to fitulimit -nwhen it is low): real limit of
simultaneous connections. A warning is printed if the chosen value exceeds the open files limit.--partitions(default: CPU cores): the work (hosts x ports) is calculated up front and split
into balanced parts (sizes differ by at most 1) that are scanned in parallel.--timeout(default500ms): connect/read timeout perhost:port.- Defaults (
--concurrency 1000,--timeout 500) favour speed. Measured on a/18, port 22:
5000/2000 found 20 banners in 8s (lost SYNs from the burst), 1000/500 found 31 in 9s,
1000/3000 found 48 in 51s, 300/2000 found 54 in 1m54s. Use--timeout 3000to miss fewer hosts. connect/timeoutin the error breakdown shows a yellow hint:try increasing the timeout with --timeout <ms> - now: 500ms, default: 500ms(no colors ifNO_COLORis set).- Scan plan printed before starting (units, partitions, concurrency, timeout).
- Summary with banners, errors, resource errors (file descriptors/ports exhausted on this machine)
and units not processed. Resource errors are no longer hidden as "host did not answer". - Error breakdown in the summary by stage and reason (
connect/timeout,connect/econnrefused,
recv/timeout= connected but no banner,recv/closed, ...), so dropped hosts can be explained. --errors-output <file>: saves every failedhost:port stage reasontooutput/<file>for
debugging (opt-in: it keeps one row per failure in memory).- The docker launcher shows local time instead of UTC on Linux hosts.
- The docker launcher raises the container open files limit (
--ulimit nofile=65536:65536); the
docker default of 1024 made most connections of a/16fail withemfile.
Changed
- Results are stored one row per result and progress uses counters. Registering a result no longer
copies the whole result list (it was quadratic:/16with more ports got much slower). - Progress refreshes 10 times per second reading only counters (it was a busy loop copying all results).
- The total of hosts is calculated, not enumerated.
Binoculo.Maestrois no longer a GenServer;Maestro.run/2runs the plan. A crashed
partition does not stop the others and is reported in the summary.Worker.get_banner/4accepts awrite_payloadoption (read once instead of oneConfigcall
per connection), closes the socket and returns the errorreason.
Fixed
- Files created in
output/were owned by root and could not be deleted from the host. The launcher
now runs the container as your user and createsoutput/itself. Ifoutput/already belongs to
root (older runs), it stops with thesudo chowncommand to fix it once. - A response header without a space after the colon (
X-XSS-Protection:1; mode=block) crashed the
run while saving to Meilisearch, before the summary. The parser now accepts it and a failing result
can no longer abort the run. - The old
max_concurrency: 200did not limit the connections (it only limited acast), so a
big range opened all of them at once. The limit is real now. - The scan could hang forever when every worker finished before the first progress check.
- A descending range (
192.168.0.10..192.168.0.1) crashed instead of returning an invalid range error. - Concluding a port no longer removes the other ports of the same host from the in-progress state.