Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Introduction

Network Scope is a portable network discovery and inventory tool written in Rust. One code base provides three front ends:

  • an egui desktop app (network-scope.exe on Windows),
  • a scriptable CLI (network-scope-cli.exe on Windows, a console-subsystem build of the same program),
  • a ratatui live terminal view (scan --tui).

It finds the hosts on a network, identifies their services, TLS and SMB configuration, operating system family, manufacturer and device type, derives risk findings and version-based CVE hints, and stores every run in a local SQLite inventory. Snapshots can be compared, exported (HTML, Markdown, JSON, JSON Lines, CSV, nmap XML) and monitored over time with alerts.

Network Scope has a strong focus on industrial (OT) networks: Modbus/TCP, S7, EtherNet/IP, BACnet/IP and OPC UA devices are identified with read-only requests only, and a passive mode reads pcap/pcapng files or a live interface to build an OT asset inventory without transmitting anything. Safe Mode is enabled by default.

Normal TCP connect scans work without administrator rights or Npcap.

What it is not

  • It is not a vulnerability scanner. CVE entries are hints derived from version strings; distributions backport fixes without changing versions.
  • It is not a drop-in replacement for other scanners. It reads and writes nmap XML for interchange only.
  • It does not exploit, authenticate, write to devices or change device configuration.

Status

Version 0.5. The changelog lists what changed and, importantly, what has not been tested. In short: everything that needs Npcap/libpcap on a real interface (SYN, ARP sweep, TCP traceroute, live capture, DCP) has only been tested against a fake wire, the OT decoders only against synthetic frames, and no live production network was scanned during development.

Where to go next

Installation

Windows

Download from the latest GitHub release:

FileWhat it is
NetworkScope-<version>-setup.exeInno Setup installer
network-scope-<version>-windows-x64.zipPortable binaries
SHA256SUMSSHA-256 checksums of all release files

Verify a download in PowerShell:

Get-FileHash .\network-scope-0.5.1-windows-x64.zip -Algorithm SHA256

and compare the hash with the line in SHA256SUMS.

Installer

The installer installs per user by default (no administrator rights needed; the setup dialog lets you choose an all-users install). Optional tasks: a desktop shortcut and adding the install folder to your user PATH so network-scope-cli works from any shell. It installs network-scope.exe, network-scope-cli.exe, the README, third-party notices, the example scripts and the lab files.

The installer and binaries are not code-signed; Windows SmartScreen will warn. The installer script has not been compiled with iscc yet; the first release build is its first compile.

Portable

Unzip anywhere and run network-scope.exe (GUI) or network-scope-cli.exe (CLI).

Two executables

  • network-scope-cli.exe is a console-subsystem build: cmd.exe and PowerShell wait for it, and %ERRORLEVEL% / $LASTEXITCODE reflect the run (1 on error). Use it for scripting.
  • network-scope.exe is a GUI-subsystem program for double-click use. Started from a shell with arguments it attaches to the console and runs the command, but the shell does not wait for it and its exit code is not seen (workaround: Start-Process -Wait -NoNewWindow or | Out-Host). Redirected stdout/stderr stay redirected.

Without arguments both binaries open the desktop. On Linux and macOS both binaries are built and behave identically.

Preferences and the default SQLite database live in %LOCALAPPDATA%\NetworkScope.

Privileges

Most features need no administrator rights and no driver: TCP connect scanning, service, TLS and SMB detection, ICMP and ARP lookups through Windows APIs, UDP probes, inventory, reports, monitoring and pcap file analysis.

FeatureNeeds
SYN scan (--fast --syn)Npcap/libpcap, capture rights, --interface, --source-ip, --source-mac
ARP sweep (--fast --arp-sweep)Npcap/libpcap, capture rights, interface and source addresses
TCP traceroute (--trace-method tcp)Npcap/libpcap, capture rights, --fast
UDP traceroute (--trace-method udp)Administrator / root (raw ICMP socket), --fast
ICMPv6 traceroute on Linux/macOSroot (raw ICMPv6 socket)
Passive capture (listen, ot-listen --interface)Npcap/libpcap, capture rights
Profinet DCP Identify (dcp)Npcap/libpcap, capture rights, --source-mac

Npcap (Windows)

Install Npcap separately; it is under its own license and not bundled. Network Scope never installs drivers or requests elevation. If Npcap was installed with “Restrict Npcap driver’s access to Administrators only”, run Network Scope as administrator for capture features. Npcap’s dependent DLLs are loaded from the driver’s directory and Windows default trusted locations.

network-scope-cli interfaces lists the capture interfaces and their IDs. Source IPv4 and MAC values must match that interface.

If the driver or configuration is missing, a SYN scan falls back to TCP connect with the host note “SYN unavailable; TCP connect fallback: …”; the ARP sweep falls back to per-host lookups.

Linux and macOS

libpcap is loaded at run time from libpcap.A.dylib/libpcap.dylib (macOS) or libpcap.so.1/libpcap.so/libpcap.so.0.8 (Linux). ICMP uses an unprivileged datagram socket when the OS allows it and falls back to a raw socket (root). Run capture features as root; granting cap_net_raw,cap_net_admin with setcap should also work on Linux but is untested.

Linux and macOS binaries

Release archives network-scope-<version>-linux-x86_64.tar.gz and network-scope-<version>-macos-arm64.tar.gz contain network-scope and network-scope-cli. The Linux and macOS platform glue (routes, neighbor states, raw sockets) is not compiled on the development machine; CI is its first compile, and it has not been exercised on a real network.

Container image

Each release tag publishes the CLI as ghcr.io/sparkz400/network-scope:<version> and :latest (linux/amd64 only):

docker run --rm ghcr.io/sparkz400/network-scope:latest scan 192.168.1.0/24 --tcp-only
  • The entrypoint is network-scope-cli; without arguments the image prints the help.
  • The image runs as an unprivileged user (scanner, UID 10001). TCP connect scanning, service and TLS detection and UDP probes need nothing more.
  • Raw sockets, ICMP fallbacks, ARP and capture features (SYN, ARP sweep, listen, dcp, TCP/UDP traceroute) need --user root --cap-add NET_RAW (capture may additionally need --cap-add NET_ADMIN). A capability granted with --cap-add reaches a process running as root in the container; the unprivileged default user does not receive it. This combination has not been tested yet.
  • Use --network host (Linux) to see the local LAN directly; on the default bridge network ARP, neighbor discovery and capture only see the container’s own network.
  • Safe Mode is on by default inside the container as well.
  • The inventory database defaults to /data/NetworkScope/inventory.sqlite; mount a volume to keep it: -v network-scope:/data.
  • The runtime image contains libpcap (libpcap0.8t64) for capture features and CA certificates for HTTPS webhooks; the GUI libraries are not included and the desktop app does not run in the container.

Build the image yourself with docker build -t network-scope . from the repository root.

Verifying downloads and the image

Every release has three independent proofs of origin (details in SECURITY.md):

# GitHub artifact attestation (Sigstore-signed SLSA build provenance) for a release file
gh attestation verify network-scope-0.5.1-linux-x86_64.tar.gz --repo Sparkz400/network-scope

# SLSA level 3 provenance from slsa-github-generator (network-scope-<version>.intoto.jsonl)
slsa-verifier verify-artifact network-scope-0.5.1-linux-x86_64.tar.gz \
  --provenance-path network-scope-0.5.1.intoto.jsonl \
  --source-uri github.com/Sparkz400/network-scope --source-tag v0.5.1

# Container image
gh attestation verify oci://ghcr.io/sparkz400/network-scope:0.5.1 --owner Sparkz400

Build from source

Requires a current stable Rust toolchain (rustup).

git clone https://github.com/Sparkz400/network-scope
cd network-scope
cargo build --release --locked

The binaries are target/release/network-scope and target/release/network-scope-cli (.exe on Windows). cargo run --release starts the GUI; cargo run --release --bin network-scope-cli -- scan … runs the CLI.

Linux GUI build dependencies (Debian/Ubuntu):

sudo apt-get install libxkbcommon-dev libwayland-dev libx11-dev libgl1-mesa-dev \
  libxcb-render0-dev libxcb-shape0-dev libxcb-xfixes0-dev

Windows installer from source

With Inno Setup 6:

cargo build --release --locked
iscc /DAppVersion=0.5.1 installer\network-scope.iss

writes dist\NetworkScope-0.5.1-setup.exe. Pass /DExePath=… and /DCliExePath=… for other build directories.

Shell completions

network-scope-cli completions powershell | Out-String | Invoke-Expression
source <(network-scope completions bash)

The scripts cover both binary names.

Quick start

Read Responsible use first: only scan networks you are authorized to scan.

Desktop

  1. Start network-scope.exe (or network-scope / cargo run --release).
  2. Enter an IP, an IPv4 range (192.168.1.10-50), a CIDR block (192.168.1.0/24) or several comma/space-separated addresses.
  3. Press Start. Nothing scans automatically.
  4. Click a host for its identity, probe errors and notes. Open Findings, Devices and Inventory from the menu.

See the GUI guide.

Command line

On Windows use network-scope-cli for the command line. Everything below works the same with network-scope on Linux and macOS.

# Discover a /24 and identify services, OS and UDP services, with the live terminal view
network-scope-cli scan 192.168.1.0/24 --deep --tui

# Industrial identification on standard OT ports, saved as JSON as well
network-scope-cli scan 192.168.1.10 --arp --ot --output inventory.json

# What was saved, and what changed between scan 1 and scan 2
network-scope-cli history
network-scope-cli diff 1 2 --format text

# Findings and an HTML report against a baseline
network-scope-cli findings 2 --min-severity medium
network-scope-cli report 2 --baseline 1 --output report.html

# Passive OT asset discovery from a capture file (sends nothing)
network-scope-cli ot-listen --pcap plant.pcapng --progress
network-scope-cli ot-assets 3 --format csv --output assets.csv

Every scan is stored in the inventory database (%LOCALAPPDATA%\NetworkScope\inventory.sqlite on Windows; --db FILE selects another). The numbers in diff 1 2 are the snapshot IDs printed by history.

Faster scans

Safe Mode (the default) keeps traffic gentle: at most eight hosts at a time and at least 100 ms between application probes. On networks without fragile equipment, --fast lifts those limits:

network-scope-cli scan 10.0.0.0/22 --fast --tcp-only --workers 1024 --ports 22,80,443,445

With Npcap/libpcap installed, SYN scanning is available too; see Installation and the Scanning engine.

Next steps

Responsible use

Network Scope sends network traffic to the targets you give it. Use it responsibly.

  • Authorized networks only. Scan only networks and devices you own or are explicitly authorized to test. Unauthorized scanning may be illegal and may violate the terms of your network provider or employer.
  • OT and other fragile equipment. Industrial controllers, medical devices and other embedded systems can fault under active probing, even with read-only requests. Safe Mode is the default; keep it on, prefer passive discovery, and coordinate with the people who operate the equipment. See Safe Mode and OT safety.
  • No warranty. Network Scope is provided “as is”, without warranty of any kind, under the MIT license. Findings and CVE entries are hints, not verified vulnerabilities.
  • Your responsibility. You are responsible for complying with the laws, regulations and policies that apply to your use, including data protection rules for the inventory data you collect (host names, MAC addresses, captured metadata).
  • Sharing results. Scan databases, reports and captures describe a network in detail. Store them accordingly, and redact IP addresses, MAC addresses and host names before posting them in issues.

GUI guide

The desktop app (network-scope.exe, or either binary started without arguments) runs the same engine as the CLI.

Scanning

Enter an IP, an IPv4 range, a CIDR block, or comma/space-separated addresses, then Start. Nothing scans automatically. The desktop accepts up to 65,536 addresses per scan.

  • Settings enables ARP discovery, industrial identification, services/UDP/OS detection, traceroute and the opt-in transmitting features (ARP sweep, adaptive SYN, trace method and port; these are greyed out in Safe Mode).
  • CVE rules file in Settings (type a path or Browse…; Clear returns to the default rules) merges your own CVE/CPE rules over the built-in ones, like --cves FILE; see Findings and CVE hints. It is saved with the other preferences and loaded when you leave the field, press Enter or pick a file. A file that fails to parse is reported under the field and the previous rules stay in use; findings on screen are recomputed with new rules.
  • Use industrial ports selects the relevant TCP ports for OT identification. To find ICMP-silent TCP services, also enable scanning of nonresponders.
  • Safe Mode is on by default; see Safe Mode and OT safety. The status bar at the bottom shows the current profile: a green Safe Mode badge (at most 8 hosts at a time, at least 100 ms between probes; no SYN, ARP sweep or TCP/UDP traceroute) or an amber Fast badge whose tooltip lists the enabled opt-in features (SYN, ARP sweep, rate cap). It follows the setting immediately, and a click opens Settings.
  • Fast selects async TCP discovery with up to 1,024 hosts at once; Settings allows up to 4,096. These are concurrency limits, not benchmark results.
  • Pause lets in-flight hosts finish; Stop aborts asynchronous work and keeps hosts that had already finished (an already issued native request can still finish later).
  • The Standard preset and Restore defaults keep interface, source IPv4/MAC, script path and probe file.

Click a host for its identity fields, probe errors and notes. Notes are stored with the host (empty notes are not stored). Preferences are saved about a second after a change.

Scan → Clear results and captures clears active results and passive observations from the window; the saved inventory is untouched. Starting a scan does not clear passive observations and starting a capture does not clear scan results, so the Devices window merges both.

Inventory

Inventory reopens SQLite snapshots, selects baseline A and scan B for a comparison, and exports JSON, CSV, HTML or a Markdown summary. Completed and stopped scans are stored separately. Comparisons require completed scans with matching scope and settings; a timeout never proves that a device is offline.

Exports are written in the background; the status line shows “Saved FILE” or the error. Exports of a reopened snapshot carry its stored scan time; exports of live results carry the current time.

Findings and Devices

Findings and Devices sort by column header (click again to reverse) and filter by severity (Devices: highest finding, Risk column) and text. Clicking a row selects and opens that host in the main table. See Findings and CVE hints.

Passive / DCP

Select an Ethernet interface and a capture duration (needs Npcap/libpcap). Passive capture sends no discovery packets and decodes mDNS, SSDP/UPnP, LLDP, NetBIOS, DHCP and ARP. The separate DCP Identify button sends one read-only Profinet multicast request; it never assigns IPs or changes a device. MAC-keyed observations retain devices that advertise no IP.

Capture visibility depends on the network interface, multicast delivery and switch mirroring; it does not expose all traffic on a switched LAN. Observations are bounded to 10,000; a full capture queue or the cap marks the capture partial and collection continues with what still fits.

OT passive asset discovery

The same window has an OT passive section:

  • Open pcap… reads a capture file on a worker thread with a progress bar (percent, frames/s) and Cancel.
  • Listen (OT) listens receive-only on the selected interface for the chosen number of seconds (disabled while another capture or a scan runs); Stop ends it early.
  • Cancelled or stopped runs are marked partial.
  • The asset table lists address, vendor, product, firmware, serial, protocols, roles, first/last seen and control traffic; rows with observed control traffic are red.
  • Export assets CSV/JSON writes the ot-assets formats.
  • Each finished OT capture is saved to the inventory like ot-listen; closing the window saves a running one as partial, and reopening a saved ot-passive: capture refills the asset table.

See OT / industrial.

Scripts

Host details → Test script… dry-runs the configured or a chosen .rhai file on a copy of the host; Apply adds the result to the host in the window only. See Scripting.

Topology

The Topology window draws the merged path map from traceroute data (enable traceroute in Settings).

Screenshot mode

network-scope --screenshot FILE.png renders the window to a PNG without scanning or touching preferences or the inventory. NETWORK_SCOPE_SCREENSHOT_STATE=main|findings|devices|ot|script renders a larger window with sample data in that view (append -small for the default 800x450 size, then -fast for Safe Mode off, e.g. main-small-fast). This is used for the documentation images.

CLI reference

On Windows use network-scope-cli (console subsystem: shells wait and see the exit code). network-scope behaves the same on Linux and macOS. network-scope-cli --help prints the built-in summary.

General rules:

  • Exit status: see Exit codes. scan, listen and dcp save what was collected when an error occurs or the run is stopped early (marked partial) and exit with status 2; errors with nothing collected fail with status 1 without saving. A forced process kill cannot save in-memory work.
  • --quiet (before or after the command, any command) suppresses progress lines on stderr, including monitor’s per-cycle lines; errors still print and the exit status is unchanged.
  • --db FILE selects the inventory database (default %LOCALAPPDATA%\NetworkScope\inventory.sqlite on Windows).
  • --cves FILE merges extra CVE/CPE rules for any command; see Findings and CVE hints.
  • --feeds FILE selects the KEV/EPSS cache (default feeds.json next to the user cves.json); without a cache findings are not enriched. See Known exploited and EPSS.
  • --format and --output FILE select an export where a command produces one. Without --output the result is printed to stdout.

Exit codes

CodeMeaning
0Success
1Error: bad arguments, I/O, database (also: nothing was saved because the scan or capture failed before collecting anything)
2Partial result: a scan or capture stopped early and its results were saved (marked partial in the inventory)
3findings / diff found something at or above --min-severity, only when --fail-on-findings is given

Without --fail-on-findings, findings and diff exit 0 however much they report. The flag is an error on any other command (status 1).

What counts as partial (2): scan stopped with Q/Esc in the --tui view, or ended after errors (for example a terminal failure or a worker that disconnected) with at least some results; listen and dcp that hit a capture error after observing something. In each case the snapshot is saved with complete = false (shown as partial), the result is still written to stdout or --output, and an error line is printed on stderr. listen --seconds N running to the end is a normal, complete run (0). ot-listen and monitor are not covered by code 2: monitor reports incomplete passive captures as capture-incomplete alerts and keeps running, and Ctrl+C or a process kill ends the process without saving (the shell’s own status for the signal).

findings ID --fail-on-findings exits 3 when at least one finding at or above --min-severity (default info) is listed. diff OLD NEW --fail-on-findings exits 3 when the comparison contains at least one new finding at or above --min-severity (default info). Other changes (hosts or ports appearing, changed identity), resolved findings and findings the new scan did not re-check do not count; use --format json and its summary for other gates. On diff, --min-severity only sets this threshold and never filters the output. The output is written in full before the process exits with 3, so stdout still carries the structured result.

network-scope-cli findings 3 --min-severity high --fail-on-findings --quiet > findings.json
if ($LASTEXITCODE -eq 3) { "High findings present" }
network-scope-cli diff 2 3 --format json --fail-on-findings --min-severity medium

network-scope with arguments (the GUI binary) returns the same codes; on Windows its shell does not wait for it, so use network-scope-cli in scripts.

Commands

CommandPurpose
(no arguments)Open the desktop GUI
scan TARGETS [options]Active scan, saved to SQLite
listen --interface ID [--seconds 60]Passive capture only (sends nothing)
dcp --interface ID --source-mac MAC [--seconds 60]Read-only Profinet DCP Identify
interfacesList capture interfaces
history [--db FILE]List saved scans as JSON
diff OLD_ID NEW_ID [--format lines|json|text] [--ignore KEY,KEY] [--fail-on-findings [--min-severity SEV]]Compare two snapshots
export ID --format json|csv|html|nmap|md|jsonl [--baseline ID] [--output FILE]Export a snapshot
report ID [--baseline ID] [--output FILE]HTML report with charts and map
import FILE.xmlImport an nmap XML scan into the inventory
findings ID [--min-severity low] [--sort exploited|severity] [--fail-on-findings]Weak TLS/SSH, risky services, CVE hints
feeds import [--kev FILE] [--epss FILE] [--feeds FILE]Import CISA KEV (JSON) and FIRST EPSS (plain CSV) files you downloaded; no network access
feeds status [--feeds FILE]Show what the feeds cache holds (JSON)
monitor TARGETS [scan options] --interval 300 [...]Repeat scans and alert on changes
alerts [--db FILE]Alerts raised by monitor runs
scripts PATHCheck Rhai scripts (use --script PATH with scan)
ot-listen --interface ID --seconds N or ot-listen --pcap FILE [--progress] [--no-reassembly]Passive OT asset discovery
ot-assets ID [--format json|csv]OT asset inventory from a passive capture
devices ID [--format json|csv]One record per device (active and passive merged)
devices --history MAC|IP [--format text|json]Timeline of one device across saved scans
topology ID [--format ascii|dot]Network path map from traceroute data
trace TARGET [--max-hops 20] [--timeout 1000] [--fast --trace-method udp|tcp --trace-port N]Traceroute to one IP
probes [--probes FILE]List and validate the TCP and UDP probe database
completions powershell|bashPrint a shell completion script
help, --help, -hPrint the help text

Targets

TARGETS is an IP address, an IPv4 dash range, a CIDR block, or several of them separated by commas or spaces. Ranges are expanded lazily. The CLI accepts up to 16,777,216 addresses (a /8); above 65,536 only responding hosts are kept in memory and saved. See Scanning engine for how network and broadcast addresses are skipped.

Every CLI scan and monitor run also checks the ports of addresses that did not answer ping.

Scan options

These apply to scan and monitor.

OptionDefaultMeaning
--ports LIST80,443,8080; with --ot and no --ports: 80,443,102,502,4840,44818TCP ports; top:N adds the N most common ports
--top-ports NAdd the N most common TCP ports (1-100)
--workers N1024Hosts scanned concurrently (1-4096; Safe Mode caps at 8)
--timeout MS2000Port and ping timeout (1-10000 ms)
--retries N1Extra attempts for unanswered ports (0-5); the timeout doubles each time
--rate PPS0Probe rate cap outside Safe Mode (0 = unlimited)
--parallel N64Concurrent port probes per host outside Safe Mode (1-1024; Safe Mode: 4)
--tcp-onlyoffSkip ICMP ping
--arpoffARP discovery on the local IPv4 network
--dnsoffResolve host names (system DNS)
--otoffRead-only industrial identification (Modbus, S7, EtherNet/IP, BACnet, OPC UA)
--unit N1Modbus unit ID
--rack N / --slot N0 / 2S7 rack and slot
--servicesoffIdentify services: banners, HTTP, TLS, SMB, RDP, databases…
--udpoffProbe UDP: DNS, DHCP, NTP, NetBIOS, SNMP (community public), SSDP, mDNS
--udp-ports LIST53,67,123,137,161,1900,5353Replace the UDP port list (implies --udp; OT and printer ports are skipped)
--osoffOS family, manufacturer (IEEE OUI) and device type guess
--deepoffSame as --services --udp --os
--traceoffRecord the router path to each responding host
--max-hops N20Traceroute hop limit (1-64)
--trace-method icmp|tcp|udpicmpTraceroute method (implies --trace; tcp/udp only with --fast)
--trace-port Nfirst open portTCP traceroute port (never OT or printer ports)
--probes FILEExtra service probes (TCP and UDP), merged over data/probes.json by name
--script PATHRhai script, or folder of .rhai files, run against every host
--fastoffLeave Safe Mode (read-only OT still applies)
--synoffIPv4 SYN scan via Npcap/libpcap (routed targets via the gateway MAC); else TCP connect
--adaptive-synoffRTT-based SYN reply windows and congestion window (with --syn --fast)
--adaptive-tcpoffPer-host TCP connect timeout from the host’s measured connect RTT; only shortens waits (also in Safe Mode)
--arp-sweepoffFast ARP sweep through the capture interface (with --fast; needs Npcap/libpcap)
--interface IDCapture interface (see interfaces); required for SYN, ARP sweep, TCP trace, capture
--source-ip IPV4 / --source-mac MACSource addresses of that interface
--tuioffLive terminal view
--format FMT / --output FILEjsonAlso write an export of the finished scan
--db FILEInventory database

TUI keys: Space pauses/resumes; Q, q or Escape stops and saves a partial result; Up/Down/PgUp/PgDn/Home/End select a row; Enter shows the host’s ports, services, findings and identity fields (Enter or Backspace returns); S cycles the sort order (newest, address, open ports).

Export formats

--format json|csv|html|nmap|md|jsonl for scan and export:

FormatContent
jsonFull snapshot
csvOne row per host; cells starting with =, +, -, @, TAB or CR get a leading ' (purely numeric cells such as -1 are unchanged)
htmlSame as report: self-contained HTML report
mdMarkdown summary (scope, totals, findings, findings by host, devices, changes since --baseline)
jsonlOne JSON object per responding host
nmapnmap-compatible XML

Exports from scan, listen, dcp and export carry the stored scan time. Details in Inventory, diffs and reports.

diff options

OptionMeaning
--format lines (default)JSON array of change lines
--format jsonStructured report for CI and monitors
--format textPlain lines
--ignore K,KExtra ignored keys: exact keys, globs (http.*.title) or suffixes (.server)
--fail-on-findingsExit 3 when a new finding at or above --min-severity appeared; see Exit codes
--min-severity SEVThreshold for --fail-on-findings (default info); does not filter the output

findings options

--min-severity info|low|medium|high|critical (default info), --cves FILE, --feeds FILE, --fail-on-findings (exit 3 when anything at or above the threshold is listed; see Exit codes), --sort exploited|severity (default exploited: known exploited first, then severity, then EPSS, when feeds are imported; the same as severity without feeds).

feeds options

feeds import needs --kev FILE (known_exploited_vulnerabilities.json) and/or --epss FILE (plain CSV; run gunzip on the .csv.gz first). The part not given keeps its previous content. --feeds FILE writes to another cache file.

monitor options

All scan options, plus:

OptionDefaultMeaning
--interval S300Seconds between cycles (1-86400)
--cycles N0 (endless)Number of cycles
--learnoffRecord devices without alerting
--min-severity SEVMinimum alert severity
--report-missingoffAlert when a known device is not seen
--alerts FILEAppend alerts as JSON lines
--dhcp-allow IP,IPLegitimate DHCP servers
--dhcp-discoveroffBroadcast one DHCPDISCOVER per cycle and collect offers
--interface ID / --listen-seconds N30Passive DHCP listening per cycle
--webhook http(s)://URL,...POST alerts as JSON; HTTPS is verified against the OS certificate store
--syslog [udp://|tcp://]HOST[:PORT],...RFC 5424 syslog
--alert-timeout S5Per-operation alert delivery timeout (1-60)
--no-dedupoffAlert every cycle
--renotify-hours H24Re-alert a persisting condition after H hours (0 = only after it resolved)

See Monitoring and alerts.

Passive and OT options

OptionDefaultMeaning
--seconds Nlisten/dcp: 60 (max 86400); ot-listen: 300 (max 604800)Capture duration
--pcap FILERead a pcap/pcapng file instead of an interface (ot-listen)
--progressoffPrint percent, frames and frames/s on stderr while reading a file
--no-reassemblyoffDisable OT TCP reassembly

trace options

trace TARGET takes --max-hops (default 20), --timeout (default 1000 ms), --trace-method icmp|tcp|udp (tcp/udp need --fast), --trace-port N, --rate PPS, and for TCP the SYN interface options --interface, --source-ip, --source-mac.

devices options

devices ID [--format json|csv], or devices --history MAC|IP [--format text|json]. The MAC can be written AA:BB:.., aa-bb-.., aabb.cc.. or aabbcc….

Examples

network-scope-cli scan 192.168.1.0/24 --tcp-only --fast --workers 1024 --ports 80,443
network-scope-cli scan 192.168.1.10 --arp --ot --tui --output inventory.json
network-scope-cli scan 192.168.1.0/24 --deep --trace --script scripts\example.rhai --format md --output scan.md
network-scope-cli scan 10.0.0.0/22 --fast --syn --adaptive-syn --arp-sweep --interface "INTERFACE_ID" --source-ip 10.0.0.9 --source-mac "AA:BB:CC:DD:EE:FF"
network-scope-cli history
network-scope-cli diff 1 2 --format json --ignore http.*.title,.server
network-scope-cli export 2 --format html --output report.html
network-scope-cli export 3 --format jsonl --output hosts.jsonl
network-scope-cli findings 3 --min-severity medium --cves my-cves.json
network-scope-cli feeds import --kev known_exploited_vulnerabilities.json --epss epss_scores-current.csv
network-scope-cli findings 3 --sort exploited
network-scope-cli report 3 --baseline 2 --output report.html
network-scope-cli export 3 --format nmap --output scan.xml
network-scope-cli import nmap-result.xml
network-scope-cli devices --history AA:BB:CC:DD:EE:FF
network-scope-cli monitor 192.168.1.0/24 --deep --interval 300 --dhcp-allow 192.168.1.1 --alerts alerts.jsonl --syslog udp://10.0.0.5
network-scope-cli ot-listen --pcap plant.pcapng --progress
network-scope-cli ot-assets 4 --format csv --output assets.csv
network-scope-cli trace 10.20.0.1 --fast --trace-method udp
network-scope-cli interfaces
network-scope-cli listen --interface "INTERFACE_ID" --seconds 60
network-scope-cli dcp --interface "INTERFACE_ID" --source-mac "AA:BB:CC:DD:EE:FF"
network-scope-cli completions powershell | Out-String | Invoke-Expression

Inventory, diffs and reports

Inventory

Every scan and passive capture is stored in a SQLite database, by default %LOCALAPPDATA%\NetworkScope\inventory.sqlite on Windows; --db FILE selects another. SQLite is bundled into the executable. Scan settings are recorded per snapshot; notes are retained in JSON/CSV exports. Completed, stopped and partial runs are stored and marked as such.

network-scope history lists the saved snapshots as JSON.

Diff

network-scope-cli diff OLD NEW [--format lines|json|text] [--ignore K,K]
  • lines (default) is a JSON array of change lines.
  • json is a structured report for CI and monitors: a summary with highest_new_severity, changes tagged by kind, new/resolved/unconfirmed findings, ignored keys.
  • text is plain lines.

A finding counts as resolved only if the new scan looked at its evidence again (host answered; TCP ports checked; UDP port probed in both); otherwise it is unconfirmed.

--ignore adds exact keys, globs (http.*.title) or suffixes (.server) to the always-ignored volatile keys *.days_left, *.clock_skew and *.boot_time.

Comparisons are meaningful only between completed scans with matching scope and settings. A timeout never proves that a device is offline.

Device history

devices --history MAC|IP [--format text|json] gives the timeline of one device across all saved scans and captures: first/last seen, addresses, MACs, names, the open ports of the latest completed scan, and events (first-seen, ip-changed, mac-changed, hostname-changed, port-opened, port-closed, udp-answering, udp-silent).

  • Only sightings create events: a scan that does not mention the device produces no “not seen” event.
  • Port events come only from completed active scans that checked the device’s ports with the same TCP port set; UDP is compared only for ports probed in both; SMB-only names are not compared.

Reports

network-scope-cli report ID [--baseline ID] --output report.html

One self-contained HTML file with summary cards, SVG charts, findings with NVD links, a findings-by-host table (highest severity, count per severity), a CPE column when any host has a CPE, changes since a baseline (“New findings”, “Resolved findings” and “Findings not re-checked”), a network map from traceroute data, a device table, per-host details and passive observations.

The report contains no scripts, sets a restrictive Content-Security-Policy, and HTML-escapes every scanned string. Captured metadata is treated as untrusted input.

export ID --format html produces the same report.

Exports

FormatContent
jsonFull snapshot
csvOne row per host. CSV cells that start with =, +, -, @, TAB or CR get a leading ' (CSV injection); purely numeric cells such as -1 are unchanged
mdMarkdown summary: scope, totals, findings, findings by host, devices, changes since --baseline
jsonlOne JSON object per responding host (passive-only devices: use devices ID --format json)
htmlThe HTML report
nmapnmap-compatible XML

Exports carry the real scan time: the stored time for saved or reopened scans, the current time for live GUI results.

nmap XML interchange

  • export ID --format nmap writes nmap XML (scanner="network-scope"), with product names in nmap spelling (Microsoft-IIS → Microsoft IIS httpd, Apache → Apache httpd, …; OpenSSH for Windows 8.1 → product OpenSSH, version for_Windows_8.1) and svc.<port>.cpe as <cpe> elements.
  • import FILE.xml reads nmap XML (up to 256 MB) into the inventory as a normal scan and maps product names back, so the CVE rules fire on imports and Windows OpenSSH does not look like a vulnerable portable OpenSSH. <cpe> children become svc.<port>.cpe.
  • The svc.<port> summary keeps the imported wording except for files written by Network Scope, so export → import is lossless.
  • Imports stored by 0.4 have the original spellings in svc.<port>.product, so a diff against a new import shows those keys once.
  • Down hosts and non-open ports are not imported.

Network Scope only reads and writes the documented XML format; it is not a drop-in replacement for nmap and contains no nmap code or data.

Monitoring and alerts

network-scope-cli monitor TARGETS [scan options] --interval 300 [--cycles N]

monitor repeats the scan, stores every run, and compares it with the previous completed scan of the same targets and with all devices ever seen. Alerts go to stdout as JSON lines, to --alerts FILE, and to the alerts command. See the CLI reference for all options.

Alert kinds

AlertSeverity
New deviceHigh for OT devices (including a device whose only hint is 2222/tcp), else Medium
TCP/UDP port newly openedHigh for risky ports
Port no longer openInfo (a timeout does not prove a port closed)
New findingSeverity of the finding
Address change for a known MAC (address-changed)Medium
Device not seen (device-missing, only with --report-missing)Low
Passive capture lost data (capture-incomplete)Low
Rogue DHCP serverCritical with --dhcp-allow, else High when more than one server answers
  • Only ports probed by both scans are compared, and partial scans never raise change alerts.
  • The first run (or --learn) records devices without alerting.
  • --min-severity filters; with --report-missing a hint names the alert kinds it hides.

De-duplication

A condition that persists is alerted once, again after a complete cycle without it, or after --renotify-hours (default 24; 0 = only after it resolved). --no-dedup alerts every cycle. A restarted monitor restores this state from the alerts it stored: a condition it raised before the restart is not repeated while it persists, and the --renotify-hours timer continues from the original alert (with 0, any stored alert counts, however old). Only alerts of scans of the same targets count, so several monitors with different targets can share one database without suppressing each other’s alerts (two monitors with the identical targets text share their history). One limit remains: a condition that resolved and came back during the downtime, or in the first cycle after the restart, cannot be told from one that persisted, and is not raised until it has resolved once more or the renotify time has passed.

Rogue DHCP

--dhcp-allow IP,IP lists legitimate servers (anything else is critical; without it, more than one server is high). Servers come from:

  • DHCPINFORM replies (UDP scan),
  • passive OFFER/ACK traffic (--interface ID --listen-seconds N; server = option 54 or the IP source),
  • --dhcp-discover, which broadcasts one DHCPDISCOVER and collects offers. It never sends a REQUEST.

DHCP observations keep the client in ip/mac and record servers per dhcp_server.<ip>. A capture error or the 10,000-observation cap never ends monitoring.

Alert delivery

Opt-in; nothing is sent unless a sink is configured.

  • Webhook: --webhook http(s)://host[:port]/path[,…] POSTs each alert as JSON (2xx = delivered). The default port is 80 for http:// and 443 for https://; user:pass@ URLs and other schemes are rejected.
    • HTTPS verifies the server certificate against the operating system’s certificate store (Windows, macOS, or the system CA bundle on Linux), so public CAs and a corporate CA your administrator installed both work. The host name in the URL must match the certificate (it is also sent as SNI). An IP-literal host works only if the certificate lists that IP address as a subject alternative name. There is no option to skip verification: an untrusted, expired or mismatching certificate fails the delivery with a warning that says why, and no alert is sent. For a receiver with a private CA, install that CA in the system store (on Linux, in the system CA bundle, or point SSL_CERT_FILE at it) or use http:// on a trusted network.
    • Plain http:// sends the alerts unencrypted; use it only for localhost or a trusted network.
  • Syslog: --syslog udp://host[:514], tcp://host[:601] (RFC 6587 octet counting) or host[:port] sends RFC 5424 messages, facility local0, severity critical→crit … info→informational. Control characters become spaces; messages are cut at 2048 bytes.
  • Each operation times out after --alert-timeout seconds (default 5, 1-60). After the first failure a destination’s remaining alerts of that cycle are skipped, at most 30 s per destination per cycle. Failures are printed and never stop the monitor.

Webhook and syslog delivery are tested against loopback test listeners only, not against real receivers. For HTTPS this means: the rejection of untrusted and mismatching certificates and the delivery to a server whose CA is added as an extra root are tested; the system-store path with a publicly trusted certificate (for example a hosted chat or incident service) has not been tried.

Upgrade notes

Routed hosts merged by SMB GUID + NetBIOS name (new in 0.5) keep the device id of the lowest address; an existing monitor database may raise one new-device or device-missing alert for such hosts after the upgrade.

Scripting

Network Scope runs Rhai scripts against every scanned host to add your own identity fields, findings and notes, for example site policies (“Telnet is not allowed”).

network-scope-cli scripts scripts/example.rhai                    # check a script or folder
network-scope-cli scan 192.168.1.0/24 --deep --script scripts/example.rhai

--script takes a file or a folder of .rhai files. In the desktop, set the script path in Settings; Host details → Test script… dry-runs it on a copy of one host, and Apply adds the result to that host in the window only.

Host API

A script defines fn on_host(h), which runs after each host is scanned.

MemberMeaning
h.ip, h.hostname, h.mac, h.ttl, h.pingAddress, name, MAC, TTL, ping time
h.open, h.udpOpen TCP ports, answering UDP ports
h.os, h.vendor, h.deviceOS guess, manufacturer, device type
h.has_port(n)Is TCP port n open
h.has_service("name")Was a service with that name detected
h.service(port)Service summary of a port
h.identity("key")Any identity field, e.g. svc.22.product
h.set("key", "value")Add the identity field script.key
h.finding("severity", "id", "title", "evidence")Add a finding (info, low, medium, high, critical)
h.note("text")Add a note to the host

Example

From scripts/example.rhai:

#![allow(unused)]
fn main() {
fn on_host(h) {
    if h.has_port(23) {
        h.finding("high", "policy-telnet", "Telnet is not allowed by policy", "port 23 open on " + h.ip);
    }
    if h.has_port(3389) && !h.has_port(443) {
        h.finding("low", "policy-rdp", "RDP exposed without a web front end", "port 3389 open");
    }
    h.set("zone", if h.ip.starts_with("10.") { "internal" } else { "other" });
}
}

Sandbox

  • Scripts cannot touch files, the network or processes. eval, modules and sleep() are disabled.
  • Operations, sizes, recursion depth and wall time are bounded: 250 ms per host across all scripts. Scripts that would start after the budget is spent are skipped with a note.
  • A failing script adds a note to the host and never stops the scan.

Scanning engine

Discovery methods

  • TCP connect (default): Tokio, IPv4 and IPv6, no driver and no special rights.
  • ICMP: Windows ICMP APIs; on Linux and macOS an unprivileged datagram socket when the OS allows it, else a raw socket (root). --tcp-only skips it.
  • ARP / neighbor lookup (--arp): Windows SendARP, independently of ICMP, local IPv4 network only. On Linux and macOS the neighbor table is used (see Platforms).
  • SYN (--fast --syn): optional, needs Npcap/libpcap; see SYN engine.
  • ARP sweep (--fast --arp-sweep): optional, needs Npcap/libpcap; see ARP sweep.

A timeout never proves that a device is offline or a port closed.

Rate control

Safe Mode uses a token bucket: one probe per 100 ms or the configured spacing, never faster. Outside Safe Mode there is no limiter unless you set --rate PPS (desktop: “Rate cap”).

With --syn --rate N the SYN I/O thread takes one token per transmitted SYN, retransmits included, from the same bucket as every other probe. A probe’s reply window starts when it is actually sent, so probes queued behind the cap are never reported as unanswered. Under a rate cap SYNs yield to ICMP, ARP and service probes. The RST that closes a half-open connection is a reply and is not rate limited.

Concurrency

  • Ping probes of a host run in parallel, and so do its port probes (--parallel, default 64; at most 4 in Safe Mode).
  • Hosts: --workers (default 1024, max 4096); Safe Mode allows at most eight active hosts.
  • A global cap of 4,096 in-flight probes (TCP connect and SYN) and 256 blocking native calls bounds resources. The permit is taken before a probe task is spawned.
  • Scripts, DNS, ARP and ICMP calls hold their blocking slot until the native call really ends, including after a timeout or Stop.

Retries and timeouts

Ports that stay silent are retried (--retries, default 1). Each retry doubles the timeout up to the port timeout, for SYN as for TCP connect. The adaptive timeout from measured ping still sets the first attempt.

Adaptive TCP connect timing

--adaptive-tcp (opt-in, also usable in Safe Mode; named like --adaptive-syn, since --timeout already adapts to the ping RTT by default):

  • Each host keeps its own RFC 6298 estimate (the same estimator as --adaptive-syn) from its completed connects. An accepted connect (SYN-ACK) and a refused one (RST) are both samples; a timeout or a local error is not.
  • A probe’s first connect waits SRTT + 4·RTTVAR, clamped between the minimum timeout (100 ms) and the port timeout, and never longer than the wait the scan would use without the flag. Retries still double up to the port timeout, so a slow answer is found on a retry.
  • Before the first sample exists a host’s first three ports are probed alone, so the remaining ports already get the measured timeout: a fast LAN host does not wait the full port timeout on filtered ports. A host that answers nothing waits at most 250 ms extra, once.
  • It only shortens waits. Rate limits are unchanged: every attempt, retries included, still takes a gate token (Safe Mode: 100 ms spacing, 8 hosts, 4 parallel probes), and nothing is sent faster.
  • The SYN engine ignores it (use --adaptive-syn); when SYN falls back to TCP connect it applies.
  • Trade-off: a closed port that answers later than the measured RTT is reported as unanswered, not closed. Windows reports a refused connect only after about 2 s (it retransmits the SYN), so with this flag closed ports on a host that also has open ports usually show as unanswered there; open ports are unaffected. Use it on Linux/macOS scanners, or where distinguishing closed from filtered does not matter.

Ports

--top-ports N or top:N in the port field adds the N most common TCP ports (1-100). The ordering is approximate, recalled from public scanner statistics, and not generated from any third-party services file.

Targets

Ranges are expanded lazily. The CLI accepts up to 16,777,216 addresses (a /8); above 65,536 only responding hosts are kept and saved. The desktop keeps the 65,536 limit.

“Skip network/broadcast addresses”:

  • CIDR blocks of /30 or shorter skip only their own network and broadcast address (10.0.0.255 and 10.0.1.0 inside 10.0.0.0/23 are scanned).
  • /31 and /32 are never trimmed.
  • Dash ranges keep skipping every x.x.x.0 and x.x.x.255.

A diff of an old (0.4) /23-or-larger snapshot against a new one of the same text can therefore report the inner .0/.255 addresses as new.

SYN engine

--fast --syn --interface ID --source-ip IP --source-mac MAC. Source values must match that interface.

  • One Npcap/libpcap handle, opened non-promiscuous with the kernel filter tcp and dst host <source-ip>, is owned by one I/O thread that sends and receives.
  • Replies are matched on target, ports and a hashed sequence cookie.
  • Routed IPv4 targets are scanned too: the next hop comes from the OS routing table (Windows GetBestRoute, Linux longest-prefix match in /proc/net/route, macOS/BSD route -n get). An on-link target is sent to its own MAC, a routed one to the gateway MAC, resolved once per gateway. The route must leave through the capture interface.
  • A failed lookup, a silent gateway or another interface falls back to TCP connect with a note “SYN unavailable; TCP connect fallback: …”. Driver or configuration errors trigger the same explicit fallback; a missing reply is reported as unanswered.
  • IPv4 only, untagged frames; no IPv6 SYN.
  • After Stop no queued SYN or retransmit leaves.

Packet construction uses pnet_packet; Npcap/libpcap is loaded at run time.

Adaptive SYN timing

--adaptive-syn (opt-in, only with SYN, so never in Safe Mode):

  • A target’s reply window is SRTT + max(1 ms, 4·RTTVAR) from its own samples (RFC 6298, Karn’s rule), clamped between the minimum and the port timeout. Targets without samples keep the usual window, and a fast neighbour never shortens a slow target’s window.
  • An answer after a retransmit records an upper bound, so a target that slowed down gets longer windows again.
  • New probes are limited by an AIMD congestion window (start 10, max 4,096); an answer to a retransmitted probe halves it, at most once per reply window. A probe with no answer at all counts as completed, not as loss, so firewalled ranges are not held at 10 in flight.
  • Every SYN still takes a --rate token.

ARP sweep

--arp-sweep (opt-in, not in Safe Mode, needs Npcap/libpcap):

  • A second capture handle with the filter arp broadcasts who-has requests for on-link IPv4 targets (never the scanner’s own address), at most 256 unanswered at a time, one --rate token each.
  • Requests repeat up to --retries times (at most 3 requests per target), each waiting the ping timeout clamped to 100-300 ms.
  • A reply gives the MAC and proves the host is up now; a silent target gets no further per-host lookup.
  • Each host task submits its own address, so throughput is also bounded by --workers (a silent address holds its worker for wait × requests).
  • Routed and IPv6 targets, Safe Mode, and a missing driver, interface, source IPv4 or source MAC use the per-host OS lookup instead (host note “ARP sweep unavailable; per-host lookup: …”).

Traceroute

--trace, or network-scope trace IP. --trace-method icmp|tcp|udp implies --trace; the default is ICMP.

  • ICMP probes all TTLs in parallel. IPv6 targets use ICMPv6 echo with an increasing hop limit (Windows Icmp6SendEcho2, Linux/macOS raw ICMPv6 socket, root).
  • TCP sends TTL-limited SYNs through the capture interface (same settings as --syn) to --trace-port or the host’s first open port. OT ports (including 2222) and printer ports are never used; without an allowed open port ICMP is used. Routed targets go via the gateway MAC.
  • UDP sends datagrams to ports 33435-33498 (33434 + TTL) and reads answers from a raw ICMP socket (administrator/root).
  • TCP and UDP send TTLs in waves of three and stop after the wave in which the target answered, one --rate token per probe. A silent target (filtered port) receives one probe per remaining TTL up to --max-hops.
  • Safe Mode, IPv6 targets, no eligible port and local failures fall back to ICMP with a host note.
  • network-scope trace takes --fast --trace-method udp, or --fast --trace-method tcp --trace-port N with the SYN interface options (--rate applies). Without --fast only ICMP is allowed.
  • Replies are matched by a per-trace ICMP id and the quoted destination; echo matching checks the sequence (and the id where the OS keeps it).

network-scope topology ID --format ascii|dot and the Topology window draw the merged path map.

Platforms

  • ICMP and neighbor lookup work on Linux and macOS as well.
  • libpcap is loaded from libpcap.A.dylib/libpcap.dylib (macOS) or libpcap.so.1/libpcap.so/libpcap.so.0.8 (Linux).
  • MAC lookup is skipped for IPv4 targets that are not on a local subnet (on-link routes from /proc/net/route; without it, macOS treats every target as on-link).
  • On Linux a cached /proc/net/arp entry gives the MAC, but the host only counts as responding if ip neigh reports it REACHABLE or the entry appeared after the scanner’s one UDP/9 datagram. STALE entries do not prove a host is up, so on a rescan a device that answers neither ICMP nor TCP may be reported as not responding (use --arp-sweep). The miss path is one poke, one 250 ms wait and one re-read.

Benchmark (indicative only)

One machine, one day: a Windows 11 host scanning a Windows VM and its /24 over VMware NAT. nmap 7.92 (nmap -sT -Pn -n --unprivileged -T4 --max-retries 1 --max-rtt-timeout 1500ms) against Network Scope (scan --fast --tcp-only --timeout 1500). Both TCP connect scans, no Npcap.

ScenarionmapNetwork ScopeSame result?
1 host, ports 1-1000 (target drops most ports)46.7 s (one run)49.0 s (one run)yes, 7 open ports each
/24, ports 22,80,44538.4 s and 40.7 s3.3 s and 3.4 snot compared: nmap -Pn lists every address as up, so only our run’s 2 responding hosts were checked

On one host whose firewall silently drops most ports, both tools are limited by the timeout and are about equal. The /24 gap comes from how each handles mostly empty address space under these settings, not from a faster packet path; other timing or host-discovery options may close it. Single-digit runs, one network, no warm-up control: treat the numbers as a sanity check. Reproduce with the commands above. The numbers predate 0.5.

Source layout

core (scheduler/rate gate), scan (targets and model), syn, arp, service, tls, fingerprint, udp, osfp, oui, devices, trace, topology, risk, report, nmap, monitor, notify, script, probes, capture, passive, otpassive, pcapfile, ot (including ot/opcua), platform, store, ui (CLI/TUI) and gui (egui). All of them are crate-private modules of the library (src/lib.rs), which exports only run (CLI/TUI/GUI dispatch) and the fuzzing entry points (fuzz). main.rs (GUI subsystem on Windows) and bin/network-scope-cli.rs (console twin) just call network_scope::run.

Detection

All detection is opt-in: Settings checkboxes in the desktop, or --services, --udp, --os, --trace, --deep on the CLI. Guesses are heuristics: verify before acting on them.

Services (--services)

Every open TCP port gets a banner wait plus data-driven probes from data/probes.json: SSH, FTP, SMTP, POP3, IMAP, MySQL/MariaDB, VNC, Telnet, Redis, memcached, MQTT, PostgreSQL, SQL Server, MongoDB, RDP, LDAP, DNS, RTSP.

  • HTTP reports status, server, title, redirect and auth realm. The Host header carries :port for ports other than 80/443 and brackets IPv6 literals.
  • Industrial ports (102, 502, 4840, 44818, 47808 and the like) are never probed generically; they stay with the identification-only OT probes.
  • Printer ports 9100-9107 (raw) and 515 (LPD) only get a passive banner wait: nothing is ever sent, and user probes for those ports are ignored.
  • 2222/tcp is listen-only: only an SSH greeting leads to the SSH algorithm exchange; anything else is treated as industrial (EtherNet/IP) and gets the ot-reachable finding.
  • RDP: when the server selects NLA, a second connection offers only standard RDP security to learn whether NLA is actually required (“NLA required”, or “… also accepted (NLA not required)”, which raises rdp-no-nla).
  • LDAP: the anonymous-bind finding needs a BindResponse with resultCode 0; “anonymous bind refused” is reported without a finding.

Extending the probe database

Put a probes.json in %LOCALAPPDATA%/NetworkScope or pass --probes FILE. A probe with the same name replaces the built-in one; new names are added (a name can move between TCP and UDP). The file format and rule syntax are described in the _comment field of data/probes.json: fields name, proto (tcp|udp), ports, send, send_hex, wait_ms, rules[{re, service, product, version, info}]; rules are regular expressions over the raw reply bytes, $1..$9 expand capture groups, and the first matching rule wins.

network-scope probes lists and validates TCP and UDP probes; a bad regex or hex string stops the scan with an error. --probes applies to UDP-only scans too.

TLS

  • Certificate subject, CN, SAN, issuer, validity, days left, key and signature algorithm, and self-signed status (name equality plus AKI == SKI when both are present; the signature is not verified). Certificates are inventory data and are never validated.
  • The negotiated version and an enumeration of the accepted versions (SSLv3 to TLS 1.3) and cipher suites by raw ClientHello, so legacy and weak suites that rustls refuses to speak are still found.
  • Per version the budget is one general hello, up to three weakness-class hellos (broken, 3DES, static RSA) and general enumeration with what is left, so weak suites are found first. When the budget runs out the strong-suite list is cut (tls.<port>.suites_truncated, suites_truncated_versions).
  • TLS 1.3 suites: with Safe Mode off, servers that negotiated TLS 1.3 get TLS 1.3-only hellos (x25519 key share with random bytes; nothing is derived, only the ServerHello is read) until all five suites are known or refused, at most min(rounds, 6) hellos. Safe Mode sends none and reports only the negotiated version. Keys: tls.<port>.tls13_suites and a TLS1.3: part in ciphers; TLS_AES_128_CCM_8_SHA256 is listed as weak without a finding.
  • not_yet_valid (finding tls-cert-not-yet-valid, Medium) and cert_error (certificate not parseable, handshake facts kept).
  • tls.<port>.ja4s is a JA4S-style server fingerprint (FoxIO format, ALPN part 00 because our hellos offer none) of the first full-offer hello the server accepted. It is comparable between Network Scope scans, and comparable with other tools only for a similar client hello; it has not been cross-checked against the reference implementation.

SSH

ssh.<port>.hassh is HASSHServer (MD5 of kex;enc_s2c;mac_s2c;comp_s2c from the server KEXINIT); ssh.<port>.hassh_algorithms holds the hashed string. MD5 and SHA-256 are implemented in-crate and checked against the RFC 1321 / FIPS 180 test vectors.

SMB

  • The SMB2 NEGOTIATE offers dialects 2.0.2 to 3.1.1 with the 3.1.1 negotiate contexts (preauth integrity, encryption, compression, signing), so the server reports its own.
  • On the same connection a credential-less SESSION_SETUP carrying an NTLMSSP NEGOTIATE asks for the NTLM challenge. The exchange stops there: no AUTHENTICATE, so no logon attempt and no failed-logon event. Safe Mode skips the SESSION_SETUP, so os_version and the NetBIOS/DNS names are only collected with Safe Mode off.
  • A second connection sends an SMB1 negotiate offering only “NT LM 0.12” to detect SMBv1 (smb.<port>.smb1 = accepted/refused). smb1-enabled and smb-signing are reported independently.
  • Identity keys smb.<port>.: dialect, guid (server GUID), capabilities, encryption (negotiated cipher, supported, not supported, not supported (SMB 2.x)), compression (3.1.1), signing_algorithms, preauth_integrity, clock_skew (coarse: in sync (< 2 min), +5 min, …), boot_time (old Windows), auth_mechs (Kerberos/NTLM/NEGOEX), os_version (NTLM major.minor.build; Samba reports 6.1.0), netbios_name, netbios_domain, dns_name, dns_domain, dns_forest.
  • Findings smb-no-encryption (Info) and cve-cve-2020-0796 (High, version hint: only for dialect 3.1.1 with compression and NTLM build 18362/18363; the patch level is not visible remotely, so patched hosts are reported too).

Checked once against Windows 11, not against Samba.

UDP (--udp, --udp-ports)

Built-in handlers: DNS (version.bind), NTP client request, SNMP get of sysDescr (community public only, v2c then v1), SSDP unicast M-SEARCH, mDNS service enumeration, NetBIOS node status (also yields the MAC), and DHCPINFORM (needs local UDP port 68).

probes.json entries with "proto": "udp" are sent as one datagram to listed ports without a built-in handler, instead of an empty datagram. Built in: SQL Server Browser (1434), rpcbind NULL call (111), memcached version (11211), SIP OPTIONS (5060).

The UDP scan never sends anything to industrial or printer ports: the TCP OT list plus 2222, EtherNet/IP 2221-2223, BACnet/IP 47808-47823, Profinet 34962-34964, DNP3 20000, FF HSE 1089-1091, MELSEC 5006/5007, CODESYS 1740-1743, EtherCAT 34980, VxWorks WDB 17185, printer 515 and 9100-9107. Such ports in --udp-ports are skipped with a host note, and probes listing them are rejected when loading. BACnet is identified only by --ot.

A reply means open, an ICMP port-unreachable means closed, silence is “open or filtered” and is not reported as open.

OS, manufacturer and device type (--os)

  • The OS guess is a scored rule set over the IP TTL, the SYN-ACK window and TCP option order (only with --syn), and service evidence: SSH, HTTP and SNMP banners, the SMB NTLM version (Windows release and build, strong; build 0 counted as Samba/Linux, weighted lightly), Dropbear (embedded), Cisco/MikroTik SSH banners (network device). It reports a family and a confidence, never a version.
  • CPE and fingerprint keys (.cpe, .hassh, .ja4s) are not evidence for the OS or the device type.
  • The manufacturer comes from the embedded IEEE registry (MA-L, MA-M and MA-S); locally administered/randomised MACs are marked instead of looked up. EtherNet/IP vendor names come from the ODVA vendor list. See Data sources.
  • The device type is a second scored guess with its evidence listed.

SYN-ACK fingerprinting needs the SYN engine, so TCP-connect scans get TTL and service evidence only.

Devices

network-scope devices ID (and the Devices window) merges active results and passive observations into one record per MAC, or per IP when no MAC is known.

  • Routed hosts without a MAC that report the same SMB server GUID and NetBIOS name are one device (the GUID alone is not enough: cloned VMs share it). They keep the device id of the lowest address.
  • SMB NetBIOS/DNS names are added to the host names.

Device timelines (devices --history) are described in Inventory, diffs and reports.

Passive evidence (listen)

network-scope listen and the desktop Passive window decode traffic that is already on the wire; nothing is transmitted. Each observation is attached to the sender’s MAC (and IP when the frame has one) and merged into the device record (devices ID, Devices window) next to the active evidence.

ProtocolWhat is readWhat it tells about the device
ARPMAC and IPaddress binding, IP conflicts
LLDPchassis, port, name, description, management IPv4, system capabilitiesname; capabilities bridge, router, WLAN AP, repeater give a network device, telephone gives a VoIP phone
CDPdevice ID, platform, software version, port, native VLAN, duplex, capabilities, management/interface IPv4 (802.3 LLC/SNAP to 01:00:0C:CC:CC:CC)name; router, switch, bridge, repeater give a network device, the phone capability or an “IP Phone” platform gives a VoIP phone
mDNSrecords, <host>.local name (only when the A/AAAA address is the packet source), service.<type> for known service types, model= of _device-info._tcpservice types give a type hint (see below); Apple model identifiers give Apple/media
NetBIOSname-service replies, and name registration/refresh broadcasts (workstation <00> / server <20> name, workgroup)host name; a weak Windows hint
SSDP/UPnPSERVER, ST, NT, USN, LOCATION of NOTIFY and search repliesIGD or MiniUPnPd gives a router, MediaRenderer/DIAL/Roku/Sonos a media device, printers and cameras by device type, Microsoft-Windows a Windows host
WS-DiscoveryHello, ProbeMatches and ResolveMatches on UDP 3702: types, XAddrs, endpoint address (Probe, Resolve and Bye are ignored)PrintDeviceType/ScanDeviceType give a printer, NetworkVideoTransmitter a camera, Computer a weak Windows hint (Samba’s wsdd also publishes it)
DHCPclient messages: option 55 request list, 60 vendor class, 12 host name; server replies as beforeos_hint and a type hint (Windows 10/11, macOS, iOS, Android, ISC dhclient, IP phones, printers)
Profinet DCPname, vendor, IDs, IP settingsindustrial controller

Device-type hints are stored in the observation as hint.<category> (strong) and hint_weak.<category> (weak) fields with the reason, and scored like active evidence: a strong hint (40 points) is enough for a device type, a weak one (20) only adds to other evidence, and a weak hint beside a strong one of the same category is ignored. When a passive hint names a specific kind of device, a plain LLDP/CDP announcement no longer counts as a network-device hint. The reason text is never used for keyword matching.

mDNS service types used as hints: _ipp, _ipps, _printer, _pdl-datastream (printer); _airplay, _raop, _googlecast, _sonos (media); _hap, _matter, _hue, _esphomelib (IoT); _companion-link, _apple-mobdev2 (Apple); _airport (network); _axis-video (camera); weak: _scanner, _rtsp, _workstation, _smb, _adisk, _afpovertcp, _spotify-connect, _mqtt.

The fingerprint tables (DHCP request lists and vendor classes, default host names such as DESKTOP-xxxxxxx and NPIxxxxxx, service-type lists) are heuristics written for this project and kept small. They are checked against hand-built frames only, not against captures from real devices, and a device can announce services it does not offer. Treat the result as a hint. Observations are bounded: the number of distinct service.* fields per device comes from a closed list, XML is parsed with DTDs rejected and a size and node limit, and every parser ignores truncated or malformed input.

Not covered: WS-Discovery device metadata (a TCP request to port 5357 would be needed), the NetBIOS browser (UDP 138) host announcements, LLDP-MED, and DHCPv6/DHCP option 82 fingerprints.

Not implemented

  • JARM (10 byte-exact ClientHellos per port): an inexact implementation would not match public JARM databases, and it costs too many connections for Safe Mode.
  • SNMPv3.
  • x509 signature verification for self-signed detection.

Findings and CVE hints

Findings are derived from what a scan collected, so run it with --services (or --deep). They work on stored snapshots and never send traffic.

network-scope-cli findings ID --min-severity low

The desktop shows them in the Findings window (sortable, filterable by severity and text). Severities: info, low, medium, high, critical. Every finding has a page in the Finding reference that explains it in plain language, states exactly what the tool looked at and how to fix it.

Checks

  • Certificates: expired, not yet valid, soon expiring, self-signed, weak signature hashes, short RSA keys.
  • TLS: SSLv3, TLS 1.0/1.1; RC4/DES/3DES/export/anonymous suites; missing forward secrecy.
  • SSH: weak and legacy algorithms (including Terrapin, CVE-2023-48795), SSH protocol 1.
  • Cleartext and unauthenticated services: Telnet, FTP, VNC, cleartext mail, unauthenticated Redis/memcached/MQTT, HTTP Basic over plain HTTP, exposed Docker API, public SNMP.
  • Windows services: RDP without NLA, SMBv1, unsigned SMB, SMB without encryption, LDAP anonymous bind.
  • Industrial ports reachable from the scanner (ot-reachable).
  • PHP end of life, date-aware (table PHP_EOL in risk.rs; branches not in the table are never flagged).

Some findings carry an optional reference: a link to official documentation on the problem and how to fix it. Currently smb1-enabled and smb-signing (Microsoft Learn), rdp-no-nla (Microsoft Learn), telnet-open (CISA) and the weak-TLS rules tls-sslv3, tls-deprecated-protocol, tls-broken-cipher, tls-3des, tls-no-pfs, tls-weak-signature and tls-weak-key (Mozilla Server Side TLS). The link appears in the HTML report, in a Reference column of the Markdown summary, as reference in the JSON and JSONL exports (omitted when a finding has none) and in the Findings window. Only HTTPS links are shown. Exports written before this field existed still load.

CVE hints

CVE entries are version-based hints only. Distributions backport fixes without changing the version string, so confirm with the vendor advisory or an authenticated check. Network Scope is not a vulnerability scanner.

The rules live in data/cves.json (built in). A user file merges like probes.json: %LOCALAPPDATA%/NetworkScope/cves.json (or $XDG_CONFIG_HOME/NetworkScope/cves.json), or --cves FILE on any command, or the CVE rules file field in the GUI Settings (with a Browse button; the path is kept with the preferences and, like --cves FILE, replaces the default user file). A rule or CPE entry with the same id replaces the built-in one. In the GUI the file is loaded at start-up and when the field is changed; if it cannot be read or parsed, the error is shown under the field and the previous rules stay active.

Rule fields:

FieldMeaning
idRule id
productLower-case substring of the detected product (svc.<port>.product or the HTTP Server header name)
excludeOptional product substrings the rule does not apply to
min / maxOptional inclusive / exclusive dotted versions with optional pN (8.5p1)
severityinfo, low, medium, high, critical
cveList of CVE-YYYY-NNNN; the finding id is cve-<first cve>
titleFinding title

A broken file stops the command with a message naming the file and rule. A broken user file found at start-up is reported (CLI: stderr, desktop: error line) and the built-in rules are used.

Known exploited and EPSS (offline feeds)

Findings that carry CVE ids can be enriched with two public data sets, so the list shows what is being exploited right now instead of only what is severe on paper:

  • CISA KEV, the Known Exploited Vulnerabilities catalog: CVEs with evidence of exploitation in the wild.
  • FIRST EPSS, the Exploit Prediction Scoring System: a daily probability (0 to 1) that a CVE is exploited in the next 30 days, with a percentile.

Network Scope never downloads them. You fetch the two files yourself, on a machine that is allowed to, and import them from disk:

curl -L -o kev.json https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json
curl -L -o epss_scores-current.csv.gz https://epss.empiricalsecurity.com/epss_scores-current.csv.gz
gunzip epss_scores-current.csv.gz        # the importer reads plain CSV only
network-scope-cli feeds import --kev kev.json --epss epss_scores-current.csv
network-scope-cli feeds status

Either option may be given alone; the part not imported keeps its previous content. Repeat the import to refresh (EPSS changes daily, KEV several times a week). A file that cannot be parsed is rejected and the cache stays as it was. Gzip is not read directly because the project avoids extra dependencies; the importer detects a .gz file and says so.

Cache. The import writes feeds.json next to the user cves.json: %LOCALAPPDATA%/NetworkScope/feeds.json, or $XDG_CONFIG_HOME/NetworkScope/feeds.json; --feeds FILE selects another file for any command. It is a global lookup table, not part of a scan, so it is a file and not part of the SQLite inventory (--db is per inventory and would otherwise need one copy per database). It holds the KEV entries (date added, ransomware use) and about 300,000 EPSS scores kept with five decimals, roughly 10 MB. Saved scans are not modified: the data is looked up when findings are computed, so older scans also show current feed data.

What changes.

  • findings, export --format json and the other JSON output gain the optional fields kev, kev_added, kev_ransomware, epss and epss_percentile (highest values over the finding’s CVEs) on findings that match. Without an imported cache these fields are absent and the output is identical to before; exports written earlier still load.
  • Findings are ranked by in KEV first, then severity, then EPSS, then host, port and id. Without a cache the order is the usual severity order. findings ID --sort severity restores the usual order even with feeds; --sort exploited is the default.
  • The HTML report shows a “Known exploited” badge (with the date CISA added it and “Ransomware use” where KEV says so) and “EPSS NN %” next to the CVE links.

Limits. KEV and EPSS describe a CVE, not your host: a version-based hint that matches a known exploited CVE is still only a hint (see above). EPSS is a statistical estimate, not a prediction for one device. The desktop Findings window does not show these fields yet. Importing the NVD feed is a separate roadmap item and is not part of this.

Sources and terms: KEV is published by the US government (CISA); EPSS is published free of charge by the EPSS Special Interest Group at FIRST. See data/README.md. The EPSS download address was checked on FIRST’s data page (first.org/epss/data) on 2026-11-19.

CPE names

cpe entries (id, exact lower-case products, template with {version}, {base}, {update}) produce svc.<port>.cpe such as cpe:2.3:a:openbsd:openssh:7.4:p1:*:*:*:*:*:*; distribution suffixes are dropped. The CPE vendor/product names were written from memory of the NVD dictionary and have not been checked against it; spot-check them before relying on them.

Contributing rules

New CVE rules are welcome. Keep them version-based, cite the CVE ids, and add a test for the version boundaries. See CONTRIBUTING.md.

Finding reference

Every finding that Network Scope reports has a page here that explains it in plain language: what it means, why it matters, exactly what the tool looked at, when it can be wrong and how to fix it. The finding id (shown in the Findings window, in reports and in exports) is the name of the page.

The pages describe what the scan can observe. A finding is evidence that something deserves a look, not a verdict: several rules work from banners and version text, and the tool never exploits anything. Severities are the ones the program assigns (info, low, medium, high, critical).

Two families have one page each because their ids are built from other data: version-based CVE hints (ids cve-cve-YYYY-NNNN, see Version-based CVE hint) and findings raised by user scripts (ids script-<id>, see Finding raised by a user script). The built-in SMBGhost rule cve-cve-2020-0796 is a special case with its own page.

Hosts imported from Nmap XML go through the same rules by service name and version, so findings can also appear for imported data that this tool did not probe itself.

For how findings are produced and filtered, see Findings and CVE hints.

TLS and certificates

IdSeverityTitlePage
tls-cert-expiredHighTLS certificate has expiredtls-cert-expired
tls-cert-expiringMediumTLS certificate expires within 30 daystls-cert-expiring
tls-cert-not-yet-validMediumTLS certificate is not yet validtls-cert-not-yet-valid
tls-self-signedLowSelf-signed TLS certificatetls-self-signed
tls-weak-signatureMediumCertificate signed with a weak hashtls-weak-signature
tls-weak-keyHighCertificate uses a short RSA keytls-weak-key
tls-sslv3HighSSLv3 is enabledtls-sslv3
tls-deprecated-protocolMediumDeprecated TLS 1.0/1.1 is enabledtls-deprecated-protocol
tls-broken-cipherHighBroken TLS cipher suites are acceptedtls-broken-cipher
tls-3desMedium3DES cipher suites are acceptedtls-3des
tls-no-pfsLowCipher suites without forward secrecy are acceptedtls-no-pfs

SSH

IdSeverityTitlePage
ssh-weak-algorithmsMediumSSH server offers weak algorithmsssh-weak-algorithms
ssh-legacy-algorithmsLowSSH server offers legacy algorithmsssh-legacy-algorithms
ssh-protocol-1HighSSH protocol 1 in usessh-protocol-1

Cleartext and unauthenticated services

IdSeverityTitlePage
telnet-openHighTelnet exposes credentials in clear texttelnet-open
ftp-cleartextMediumFTP sends credentials in clear textftp-cleartext
vnc-exposedMediumVNC is reachablevnc-exposed
mail-cleartextLowMail retrieval without TLSmail-cleartext
redis-noauthCriticalRedis answers without authenticationredis-noauth
memcached-openHighmemcached answers without authenticationmemcached-open
mqtt-anonymousHighMQTT broker accepts anonymous clientsmqtt-anonymous
ldap-anonymousLowLDAP answers an anonymous bindldap-anonymous
postgres-no-tlsLowPostgreSQL does not offer TLSpostgres-no-tls

Windows and file sharing

IdSeverityTitlePage
rdp-no-nlaMedium or HighRDP uses legacy security without TLS or NLA / RDP does not require Network Level Authenticationrdp-no-nla
smb1-enabledHighSMBv1 is enabledsmb1-enabled
smb-signingMediumSMB signing is not requiredsmb-signing
smb-no-encryptionInfoSMB server does not offer encryptionsmb-no-encryption
cve-cve-2020-0796HighSMBv3 compression on a Windows build affected by SMBGhostcve-cve-2020-0796

Web services

IdSeverityTitlePage
http-basic-cleartextMediumHTTP Basic authentication over an unencrypted connectionhttp-basic-cleartext
docker-apiCriticalDocker Engine API is exposeddocker-api
heartbleedCriticalServer header names a Heartbleed-vulnerable OpenSSLheartbleed
php-eolLowPHP release is no longer supportedphp-eol

Other network services

IdSeverityTitlePage
snmp-publicHighSNMP answers the community string ‘public’snmp-public
netbios-nsLowNetBIOS name service is exposednetbios-ns
upnp-exposedLowUPnP/SSDP is answeringupnp-exposed

Industrial (OT)

IdSeverityTitlePage
ot-reachableMediumIndustrial protocol port is reachable from the scannerot-reachable
ot-control-trafficMedium or HighControl traffic observed on an industrial protocolot-control-traffic

Generic finding families

IdSeverityTitlePage
cve-cve-YYYY-NNNN, for example cve-cve-2021-41773As set by the rule (Info to Critical)Version-based CVE hintcve-hints
script-<id>, where <id> is chosen by the scriptAs set by the scriptFinding raised by a user scriptscript-findings

TLS certificate has expired

Finding idtls-cert-expired
SeverityHigh
Applies toTCP ports that speak TLS

What it means

The server presented a certificate for encrypted connections (HTTPS, mail over TLS, a device’s web interface) whose validity period has ended. A certificate is a time-limited identity card for the service; after the end date, clients can no longer rely on it.

Why it matters

Browsers and other clients show a warning or refuse to connect. People learn to click through such warnings, which makes a real attack harder to notice, and automated clients (monitoring, backups, APIs) fail outright. An expired certificate also tends to mean that nobody is looking after the service.

How it is detected

During service inspection (--services or --deep) the tool completes a TLS handshake, reads the certificate the server sends and compares its “not after” date with the clock of the machine running the scan. If the scan machine’s time is later than that date, the finding is raised. The evidence line shows the “not after” date.

The tool connects by IP address and sends no server name, so a host that serves several sites may present a default certificate instead of the one a browser would see.

Possible false positives

  • The clock of the scanning machine is wrong (the check uses that clock, not an external time source).
  • A name-based virtual host or load balancer returned a default or leftover certificate to a client that sends no server name, while real clients get a valid one.
  • The port serves a certificate that no client actually uses (an old default certificate on a management port).

How to fix

  • Renew the certificate with the issuer (public CA, your internal CA, or the device’s own certificate tool), then install it where the service reads it.
  • Web servers: after replacing the files, reload the service (nginx -s reload, systemctl reload apache2, or rebind the certificate in IIS Manager under Site Bindings).
  • Automate renewal for public names with an ACME client; see the Let’s Encrypt documentation.
  • Check the clock of the device if a freshly issued certificate is still reported as expired.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

TLS certificate expires within 30 days

Finding idtls-cert-expiring
SeverityMedium
Applies toTCP ports that speak TLS

What it means

The server’s certificate is still valid but will expire within the next 30 days. After that date clients will treat the service as untrusted.

Why it matters

It is a warning ahead of an outage: when the date passes, browsers and automated clients start failing. Renewing early costs little; renewing during an incident does not.

How it is detected

The tool reads the certificate’s “not after” date during the TLS handshake and computes the whole days left (rounded down) against the scanning machine’s clock. The finding is raised when the certificate has not expired and 30 days or fewer are left; an already expired certificate gets tls-cert-expired instead. The evidence shows the days left and the date.

Possible false positives

  • Certificates that are renewed automatically in the last weeks of their life show up here until the renewal job runs.
  • The scanning machine’s clock is ahead of real time.
  • A default certificate returned because the tool sends no server name (see tls-cert-expired).

How to fix

  • Renew and install the new certificate before the date shown, then reload the service.
  • If you use an ACME client, check that its renewal job runs and that the service is reloaded afterwards.
  • Put the expiry date in a calendar or monitoring system for certificates that are renewed by hand (devices, appliances).
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

TLS certificate is not yet valid

Finding idtls-cert-not-yet-valid
SeverityMedium
Applies toTCP ports that speak TLS

What it means

The certificate’s start date lies in the future. Clients reject it, just like an expired one, until that date arrives.

Why it matters

Connections fail or show certificate errors. The usual cause is a wrong clock: a device that lost its time while it generated its own certificate, or a certificate issued with a wrong start date.

How it is detected

The tool reads the “not before” date of the certificate during the TLS handshake. If it is later than the clock of the scanning machine, the finding is raised. The evidence shows the “not before” date.

Possible false positives

  • The scanning machine’s clock is behind real time.
  • The certificate was issued a moment ago and the clocks of issuer and scanner differ by a few minutes.

How to fix

  • Check the time source of the device (NTP) and correct its clock.
  • Reissue the certificate with a correct validity period if it was generated while the clock was wrong. Self-generated device certificates are usually recreated from the device’s certificate settings; some devices need a factory reset or vendor guidance.
  • Check the clock of the scanning machine as well before assuming a device problem.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Self-signed TLS certificate

Finding idtls-self-signed
SeverityLow
Applies toTCP ports that speak TLS

What it means

The certificate was signed by itself instead of by a certificate authority (CA) that clients trust. Such a certificate encrypts traffic, but a client has no independent way to tell whether it is talking to the genuine server.

Why it matters

Clients either show warnings or are configured to accept any certificate, and in that case an attacker in the network path can impersonate the server without being noticed. Self-signed certificates are common on devices and test systems, which is why the severity is low.

How it is detected

The tool compares the certificate’s issuer with its subject. If they are identical and, when both key identifiers are present, the authority key identifier names the certificate’s own key, the certificate counts as self-signed. The signature itself is not verified.

The tool does not check trust: a certificate issued by a private or unknown CA is not reported here.

Possible false positives

  • Expected on lab systems and on devices that ship with their own certificate (printers, switches, controllers).
  • A certificate that is pinned by clients on purpose.

How to fix

  • Where clients must verify the server, install a certificate from a CA those clients trust: a public CA (for example through ACME) for public names, or your organisation’s internal CA for internal names.
  • Where a self-signed certificate is accepted, record the fingerprint and let clients pin it instead of disabling verification.
  • Many devices only offer a certificate import in their web interface or need a newer firmware; check the vendor’s documentation.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Certificate signed with a weak hash

Finding idtls-weak-signature
SeverityMedium
Applies toTCP ports that speak TLS

What it means

The certificate was signed with an outdated hash function (MD5 or SHA-1). These functions are no longer considered safe for signatures, because attackers can craft two documents with the same hash.

Why it matters

A weak signature makes it feasible, in principle, to forge a certificate that looks genuine. Modern clients reject such certificates, so users also see errors.

How it is detected

The tool reads the signature algorithm of the certificate the server presents. It raises the finding for md5WithRSA, sha1WithRSA and ecdsa-with-SHA1. Other algorithms, including less common weak ones that the tool does not know by name, are not flagged. The evidence shows the algorithm.

Possible false positives

  • Only the certificate the server presents first is judged; the chain behind it is not examined.

How to fix

  • Reissue the certificate with SHA-256 or stronger (for RSA keys sha256WithRSAEncryption, or an ECDSA key with SHA-256).
  • With OpenSSL a new request is created with openssl req -new -sha256 -key server.key -out server.csr; most CAs and device interfaces create the request for you.
  • Replace the certificate on the service and reload it.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Certificate uses a short RSA key

Finding idtls-weak-key
SeverityHigh
Applies toTCP ports that speak TLS

What it means

The certificate’s RSA key is shorter than 2048 bits. Short keys can be broken with far less effort than current ones.

Why it matters

Whoever recovers the private key can decrypt recorded traffic that used RSA key exchange and can impersonate the server. 1024-bit RSA is considered within reach of well-funded attackers.

How it is detected

The tool reads the public key of the certificate the server presents. If it is an RSA key with fewer than 2048 bits, the finding is raised and the evidence shows the size (for example RSA 1024). Other key types (elliptic-curve keys) are not judged.

Possible false positives

  • Rarely: the certificate is not the one real clients are served (see tls-cert-expired on server names).

How to fix

  • Generate a new key pair (never reuse the old key): RSA 2048 or larger, or an elliptic-curve key such as P-256, for example openssl genpkey -algorithm EC -pkeyopt ec_paramgen_curve:P-256.
  • Request a new certificate for the new key from your CA and install it.
  • Older devices may not support long keys or elliptic curves; check the firmware release notes.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

SSLv3 is enabled

Finding idtls-sslv3
SeverityHigh
Applies toTCP ports that speak TLS

What it means

The server still accepts connections with SSL version 3, a protocol from 1996 that has been broken for years. It was replaced by TLS.

Why it matters

SSLv3 has design flaws, the best known being POODLE (CVE-2014-3566), which can let an attacker decrypt parts of an encrypted session under the right conditions. The finding carries that CVE as a hint; the risk comes from the protocol being enabled, not from a specific exploit being confirmed.

How it is detected

The tool sends its own TLS handshakes that offer only SSLv3 and records whether the server completes one. This works even though modern TLS libraries refuse SSLv3. The finding is raised when SSLv3 was accepted; the evidence is the tool’s list of weak protocol and cipher observations for the port.

The number of handshakes per port is limited (fewer in Safe Mode), so a protocol that was not seen is not proof that it is off.

Possible false positives

  • A device in front of the real server (load balancer, appliance) answers for it.

How to fix

  • Disable SSLv3 and keep TLS 1.2 and 1.3. Typical settings:
  • nginx: ssl_protocols TLSv1.2 TLSv1.3;, then nginx -t and reload. See the ngx_http_ssl_module documentation.
  • Apache httpd: SSLProtocol -all +TLSv1.2 +TLSv1.3 (mod_ssl), then reload.
  • Windows / IIS: control the protocols through the Schannel registry settings (see the Microsoft page below); a reboot is required.
  • A starting configuration for common servers is generated at ssl-config.mozilla.org.
  • Appliances and devices: use the TLS or protocol setting in the management interface, or update the firmware.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Deprecated TLS 1.0/1.1 is enabled

Finding idtls-deprecated-protocol
SeverityMedium
Applies toTCP ports that speak TLS

What it means

The server still accepts TLS 1.0 or TLS 1.1. Both were formally retired by the standards body in 2021 because they depend on outdated cryptography and have no safe modern cipher suites.

Why it matters

An attacker who can force a client to use an old version gets a weaker connection to attack. Compliance frameworks and current browsers no longer accept these versions.

How it is detected

The tool offers TLS 1.0 and 1.1 in separate handshakes and records whether the server completes them. The finding is raised if either was accepted. The evidence is the list of protocol versions the server accepted on that port.

Handshakes per port are limited, so a version that was not seen is not proof that it is off.

Possible false positives

  • A compatibility setting that is needed on purpose for very old clients (old industrial panels, scanners, Windows XP). That is a risk to accept knowingly, not a measurement error.

How to fix

  • Disable TLS 1.0 and 1.1 and leave TLS 1.2 and 1.3 enabled:
  • nginx: ssl_protocols TLSv1.2 TLSv1.3;, then nginx -t and reload. See the ngx_http_ssl_module documentation.
  • Apache httpd: SSLProtocol -all +TLSv1.2 +TLSv1.3 (mod_ssl), then reload.
  • Windows / IIS: control the protocols through the Schannel registry settings (see the Microsoft page below); a reboot is required.
  • A starting configuration for common servers is generated at ssl-config.mozilla.org.
  • Before switching off, check which clients still use the old versions (server logs) so that nothing breaks unannounced.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Broken TLS cipher suites are accepted

Finding idtls-broken-cipher
SeverityHigh
Applies toTCP ports that speak TLS

What it means

The server agrees to use encryption methods that are broken or provide no protection at all: RC4, single DES, “export-grade” ciphers from the 1990s, anonymous key exchange (no server authentication) or NULL ciphers (no encryption).

Why it matters

Traffic protected with these suites can be decrypted or altered by an attacker with modest resources, or is not encrypted at all.

How it is detected

For each protocol version the server accepts, the tool offers it only suites from the weak classes and records whether the server picks one. Suites are matched by their weakness label: RC4, single DES, export-grade, anonymous or NULL (no encryption). The finding is raised if any such suite was accepted; the evidence lists all weaknesses the tool recorded for the port.

If a broken suite is found, the weaker-but-different tls-3des finding is not raised for the same port. The number of handshakes is limited (fewer in Safe Mode), so suites can be missed.

Possible false positives

  • A device in front of the real server answers for it.

How to fix

  • Remove the listed suites from the server configuration. Use a maintained template rather than editing suite lists by hand:
  • nginx / Apache: set the cipher list to the Mozilla generator’s output (ssl_ciphers / SSLCipherSuite).
  • Windows / IIS: disable the cipher in the Schannel registry settings, or use the group policy for the SSL cipher suite order.
  • Appliances and devices: update the firmware or use the cipher setting in the management interface.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

3DES cipher suites are accepted

Finding idtls-3des
SeverityMedium
Applies toTCP ports that speak TLS

What it means

The server accepts Triple-DES (3DES) cipher suites. 3DES uses a small block size, which makes it possible, in long sessions, to recover parts of the plaintext.

Why it matters

The weakness is known as Sweet32 (CVE-2016-2183). Exploiting it needs a long-lived connection and a lot of traffic, which keeps the practical risk moderate; it is nevertheless on all deprecation lists.

How it is detected

The tool offers the server only 3DES suites and records whether it picks one. The finding is raised if it does and no suite from the broken classes (RC4, DES, export, anonymous, NULL) was found on the same port; in that case tls-broken-cipher is reported instead. The evidence lists the weaknesses recorded for the port. The CVE is attached as a hint.

Possible false positives

  • Kept on purpose for very old clients.

How to fix

  • Remove 3DES suites from the cipher list:
  • nginx / Apache: use the cipher list produced by the Mozilla generator.
  • Windows / IIS: disable Triple DES 168 in the Schannel registry settings.
  • Appliances and devices: update the firmware or use the cipher setting in the management interface.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Cipher suites without forward secrecy are accepted

Finding idtls-no-pfs
SeverityLow
Applies toTCP ports that speak TLS

What it means

The server accepts suites that use the server’s long-term RSA key to protect the session key. If that key is ever stolen, all recorded traffic from the past can be decrypted. “Forward secrecy” (ephemeral key exchange, ECDHE) prevents that.

Why it matters

Low in practice today, but an easy hardening step. An attacker who records traffic now and obtains the key later (by compromise, legal request or a future weakness) can read everything.

How it is detected

The tool offers the server only static-RSA suites (names such as RSA_AES_128_GCM_SHA256) and records whether it picks one. The finding is raised if it does. The evidence text is fixed: “static RSA key exchange offered”. TLS 1.3 always has forward secrecy and does not trigger this finding.

Possible false positives

  • Needed by very old clients that cannot use ECDHE.

How to fix

  • Prefer ECDHE suites and drop static-RSA suites from the cipher list; the Mozilla generator’s output already does this.
  • Enable TLS 1.3 where the software supports it.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

SSH server offers weak algorithms

Finding idssh-weak-algorithms
SeverityMedium
Applies toTCP ports running SSH

What it means

The SSH server advertises encryption or key-exchange methods that are considered broken, for example 1024-bit Diffie-Hellman with SHA-1, DSA host keys, RC4/3DES/Blowfish ciphers or MD5-based integrity checks. A client may end up using one of them.

Why it matters

Weak methods can allow an attacker in the network path to read or alter an SSH session. When the Terrapin hint is shown, an attacker who can intercept the connection may be able to remove messages at the start of a session (CVE-2023-48795).

How it is detected

The SSH server sends its list of supported algorithms (the KEXINIT message) to every client before any login. The tool reads that list; it does not authenticate. Entries are flagged as weak when they match:

  • key exchange diffie-hellman-group1-sha1;
  • host key ssh-dss;
  • ciphers containing arcfour, 3des, des-cbc, blowfish or cast128, or none;
  • MACs containing md5 or -96, starting with umac-64, or none;
  • the Terrapin condition: ChaCha20-Poly1305, or a CBC cipher together with an encrypt-then-MAC (-etm) MAC, is offered while the server does not list the strict key exchange marker (kex-strict-s-v00@openssh.com).

The evidence lists the matches. CVE-2023-48795 is attached only when the Terrapin condition applies. This is a statement about what the server offers, not about what a particular client negotiates.

Possible false positives

  • A server that offers a weak method last, for old clients only, and never selects it for modern ones. It is still reachable by a client that asks for it.
  • Terrapin: the condition is derived from the offered algorithms and the strict key exchange marker. Whether an attack is possible also depends on the client; the finding is a hint to update, not a proof.

How to fix

  • OpenSSH (Linux, BSD, macOS, Windows): limit the algorithms in sshd_config (KexAlgorithms, Ciphers, MACs, HostKeyAlgorithms) and reload sshd. sshd -T prints the effective values. On Windows the file is %ProgramData%\ssh\sshd_config.
  • Terrapin: update to OpenSSH 9.6 or later, or to a vendor release that includes strict key exchange, on both server and clients.
  • Network devices and appliances: look for an SSH cipher or algorithm setting in the vendor’s documentation, or update the firmware.
  • Check first which clients and automation still need the old methods, so that nothing breaks unannounced.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

SSH server offers legacy algorithms

Finding idssh-legacy-algorithms
SeverityLow
Applies toTCP ports running SSH

What it means

The SSH server also offers older methods that are no longer recommended but are not broken outright: SHA-1 based key exchange or signatures, and CBC-mode ciphers.

Why it matters

These methods have known weaknesses or are being phased out. They are listed so that you can plan their removal; the practical risk is lower than for the weak methods in ssh-weak-algorithms.

How it is detected

The tool reads the algorithm list the server sends before login (KEXINIT, no authentication) and flags: key exchange diffie-hellman-group14-sha1 and diffie-hellman-group-exchange-sha1; the host key type ssh-rsa (RSA with SHA-1 signatures); and ciphers whose name ends in -cbc that are not already counted as weak. The evidence lists the matches.

Possible false positives

  • Many servers still list diffie-hellman-group14-sha1 or CBC ciphers for compatibility with old clients. It is a recommendation, not an incident.

How to fix

  • OpenSSH: remove the entries with KexAlgorithms, Ciphers and HostKeyAlgorithms in sshd_config (for example prefix a list with - to remove entries from the defaults), then reload sshd.
  • Network devices and appliances: use the vendor’s SSH settings or update the firmware.
  • Check which clients still need the old methods first.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

SSH protocol 1 in use

Finding idssh-protocol-1
SeverityHigh
Applies toTCP ports running SSH

What it means

The server announces that it speaks version 1 of the SSH protocol. Version 1 has serious design flaws and was replaced by version 2 more than twenty years ago.

Why it matters

SSH protocol 1 is open to attacks that allow session hijacking and data injection. Current software does not support it at all, so a server that announces it is very old or heavily outdated.

How it is detected

The first line the server sends (its banner, for example SSH-1.5-...) names the protocol version. The tool stores it as “protocol 1.5” or similar and raises the finding when the service description contains protocol 1. but not protocol 1.99. The value 1.99 means “supports both 1 and 2” and is not flagged. The evidence is the service description.

Possible false positives

  • The rule matches the text of the service description and does not check that the service is SSH. A banner of another service that contains protocol 1. could in theory trigger it.
  • A server announcing 1.99 can still allow version 1; it is not reported.

How to fix

  • Use protocol 2 only. Modern OpenSSH has no protocol 1 at all (removed in OpenSSH 7.6); upgrade the software.
  • For old appliances: update the firmware, or replace the device. If neither is possible, restrict access to a management network and plan the replacement.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Telnet exposes credentials in clear text

Finding idtelnet-open
SeverityHigh
Applies toTCP ports running Telnet

What it means

A Telnet service is reachable. Telnet sends everything, including the user name and password, without encryption.

Why it matters

Anyone who can observe the network path (a compromised device, a mirrored switch port, an attacker on the same segment) can read the credentials and the whole session, and take over the device with them. Telnet also has no protection against tampering.

How it is detected

Service detection identifies Telnet from the data the server sends right after the connection is made: the protocol’s option negotiation bytes. Only an identified Telnet service counts. A port that is merely open on 23 and returned nothing is shown as “unidentified (port suggests telnet)” and is not reported here. Services imported from Nmap XML count if Nmap named them telnet. The evidence is the service description.

Possible false positives

  • A service that uses Telnet-style negotiation bytes for something that is not a login shell.
  • Not a false positive, but not caught: a Telnet service that sends nothing until the client speaks is not identified.

How to fix

  • Disable the Telnet service and use SSH instead. Linux: systemctl disable --now telnet.socket (or the matching service name) and install an SSH server. Windows: remove the optional feature TelnetServer.
  • Network devices and appliances: turn Telnet off in the management configuration and allow SSH only (on IOS-style devices transport input ssh on the VTY lines).
  • If Telnet cannot be disabled, restrict it to a dedicated management network with access lists and plan the replacement.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

FTP sends credentials in clear text

Finding idftp-cleartext
SeverityMedium
Applies toTCP ports running FTP

What it means

An FTP server is reachable. Plain FTP transmits user names, passwords and file contents without encryption.

Why it matters

Anyone who can observe the traffic can read the credentials and the transferred files, and can reuse the credentials elsewhere if they are shared with other systems.

How it is detected

The tool identifies FTP from the server’s greeting on a plain connection: a line starting with 220 that is not recognised as SMTP (known FTP servers such as vsFTPd, ProFTPD, Pure-FTPd, FileZilla Server and Microsoft FTP Service are named; any other 220 greeting is called FTP). It does not log in, does not test anonymous access and does not check whether the server also offers encryption (AUTH TLS). The evidence is the service description.

Possible false positives

  • A server that supports explicit FTPS and refuses logins without it shows the same greeting. The tool cannot tell the difference.
  • A mail server whose greeting starts with 220 but does not contain SMTP or ESMTP can be misidentified as FTP.

How to fix

  • Use SFTP (over SSH) or FTPS and disable plain FTP logins.
  • vsftpd: ssl_enable=YES, force_local_logins_ssl=YES, force_local_data_ssl=YES.
  • IIS FTP: set the SSL policy for the FTP site to require SSL connections.
  • Devices (cameras, printers, cell gateways): choose SFTP/FTPS in their upload settings if available, otherwise keep the FTP server on an isolated segment.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

VNC is reachable

Finding idvnc-exposed
SeverityMedium
Applies toTCP ports running VNC

What it means

A VNC remote-desktop server answers on the network. VNC gives full control of the screen, keyboard and mouse of the machine.

Why it matters

VNC is often protected only by a short password, sometimes by none, and frequently does not encrypt the session. Reachable VNC servers are a standard target for password guessing.

How it is detected

The tool identifies VNC from the server’s version greeting (RFB 003.008 and similar). It does not test which authentication method the server uses and does not try passwords. The evidence is the service description, so it says “reachable”, not “unprotected”.

Possible false positives

  • A VNC server with strong authentication and encryption on a controlled network is still reported; the finding is about exposure.

How to fix

  • Do not expose VNC directly. Bind it to localhost and reach it through an SSH tunnel or VPN, or restrict the port in the firewall to the administrators’ addresses.
  • Use strong passwords and enable the encryption options of your VNC software.
  • Embedded and industrial HMIs: follow the vendor’s remote-access guidance; plan changes in a maintenance window, because a changed password affects every operator station.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Mail retrieval without TLS

Finding idmail-cleartext
SeverityLow
Applies toTCP ports running POP3 or IMAP

What it means

A POP3 or IMAP mail server answers on a plain, unencrypted connection. Mail clients connecting this way may send their password in clear text.

Why it matters

Passwords and mail content can be read by anyone on the network path. Mail passwords are often reused for other accounts.

How it is detected

The tool identifies POP3 (+OK greeting) or IMAP (* OK greeting) on a plain connection. Servers on TLS ports such as 993 and 995 are identified as TLS services and do not trigger the finding. The tool does not check whether the server offers or enforces STARTTLS. The evidence is the service description.

Possible false positives

  • A server that requires STARTTLS before login (IMAP LOGINDISABLED, POP3 STLS enforcement) is reported although credentials are never sent in clear text.

How to fix

  • Require TLS: offer only IMAPS (993) and POP3S (995), or enforce STARTTLS before authentication.
  • Dovecot: ssl = required and disable_plaintext_auth = yes.
  • Switch off the plain ports in the firewall once clients are migrated.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Redis answers without authentication

Finding idredis-noauth
SeverityCritical
Applies toTCP port 6379

What it means

A Redis database answered a command without asking for a password. Anyone who can reach the port can read and change its data.

Why it matters

An open Redis lets an attacker read all stored data (sessions, tokens, caches) and, through its administrative commands, frequently write files on the server and run code. Redis servers without authentication that are reachable from the Internet are attacked routinely.

How it is detected

On port 6379 the tool sends one harmless command, INFO server, and reads the reply. If the reply contains a redis_version: line, the server answered without authentication and the finding is raised; the evidence is the service description with the version. A reply that asks for authentication (NOAUTH/DENIED) is not flagged. The tool does not write anything, does not run administrative commands and does not probe other ports.

Possible false positives

  • A honeypot or a deliberately open cache on an isolated network.
  • Redis 6 and later can limit what an unauthenticated default user may do, and still answer INFO.

How to fix

  • Require authentication: configure ACL users (Redis 6 and later) or requirepass, with a long random password.
  • Bind to a private interface (bind 127.0.0.1 or the internal address) and keep protected-mode yes; do not publish the port.
  • Use TLS if clients connect over a network, and limit access in the firewall.
  • Restart the service, then test with redis-cli that a command without credentials is refused.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

memcached answers without authentication

Finding idmemcached-open
SeverityHigh
Applies toTCP port 11211

What it means

A memcached cache server answered a command without asking for credentials. Memcached has no login by default, so anyone who can reach the port can read and overwrite cached data.

Why it matters

Cached data (sessions, user data, database results) can be stolen or manipulated. Exposed memcached servers have also been abused as amplifiers for denial-of-service attacks.

How it is detected

On port 11211 the tool sends the version command and reads the reply. A VERSION x.y.z answer raises the finding; the evidence is the service description. Only TCP is considered by this finding. The tool does not read or write cache entries.

Possible false positives

  • A cache on a private network that is intended to be open to a small number of application servers. It is still better to restrict it.

How to fix

  • Listen only on the interfaces that need it (-l 127.0.0.1 or the private address) and disable UDP (-U 0).
  • Restrict the port in the firewall to the application servers.
  • Use SASL authentication where the clients support it.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

MQTT broker accepts anonymous clients

Finding idmqtt-anonymous
SeverityHigh
Applies toTCP port 1883

What it means

An MQTT message broker (common in building automation, IoT and industrial gateways) accepts a connection without a user name or password.

Why it matters

Anonymous clients may be able to listen to all messages and publish their own, for example to send commands to devices. Whether they can depends on the broker’s topic rules, which the tool does not test.

How it is detected

On port 1883 the tool sends an MQTT 3.1.1 CONNECT packet without credentials and with an empty client identifier, then reads the acknowledgement. A “connection accepted” result raises the finding; the evidence is the service description. The tool disconnects without subscribing or publishing. Brokers on the TLS port (8883) are not covered.

Possible false positives

  • A broker that accepts the connection but denies every topic to anonymous clients through access rules. The tool cannot see those rules.

How to fix

  • Require authentication and per-topic access rules. Mosquitto: allow_anonymous false, a password_file and an ACL file.
  • Offer TLS (port 8883), ideally with client certificates, and close the plain port.
  • Industrial and building gateways: publishers and subscribers must be given credentials at the same time; schedule the change in a window and follow the vendor’s documentation.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

LDAP answers an anonymous bind

Finding idldap-anonymous
SeverityLow
Applies toTCP ports 389 and 3268 (LDAP and global catalog)

What it means

A directory server (LDAP, for example Active Directory or OpenLDAP) accepts a connection without a user name or password.

Why it matters

Depending on the directory’s access rules, anonymous users may be able to list users, groups and computers, which helps attackers prepare attacks. Many directories allow limited anonymous access by design.

How it is detected

On ports 389 and 3268 the tool sends an LDAP version 3 bind request with an empty name and an empty simple password and reads the result. A “success” result raises the finding; the evidence is the service description. The tool does not search the directory, so it does not know what an anonymous client could read.

Possible false positives

  • Directories commonly allow anonymous reading of their basic server description (rootDSE) and nothing else.

How to fix

  • Disable anonymous binds unless an application needs them, and restrict what anonymous users can read. OpenLDAP: disallow bind_anon and require authc.
  • Prefer LDAPS or StartTLS for all client connections.
  • In Active Directory, check that the directory settings for anonymous access have not been loosened beyond the defaults.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

PostgreSQL does not offer TLS

Finding idpostgres-no-tls
SeverityLow
Applies toTCP port 5432

What it means

The PostgreSQL database server answers on its network port but does not support encrypted connections. Clients connect in clear text, including passwords (depending on the authentication method) and query results.

Why it matters

Data and credentials can be read on the network path. The risk is higher when the database is reached over untrusted networks.

How it is detected

On port 5432 the tool sends PostgreSQL’s SSL request (the first step of a connection) and reads the single-character answer. N means TLS is not supported and raises the finding; S (supported) is fine. No login is attempted.

Possible false positives

  • A database that is reached only over a local socket or a trusted private link, but also listens on the network.

How to fix

  • Enable TLS: set ssl = on with ssl_cert_file and ssl_key_file in postgresql.conf and reload.
  • Force encrypted connections for remote clients with hostssl lines (and no host lines) in pg_hba.conf.
  • Restrict the listen addresses and the firewall to the application servers.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

RDP uses legacy security without TLS or NLA

Finding idrdp-no-nla
Other titleRDP does not require Network Level Authentication
SeverityMedium or High
Applies toTCP ports running Remote Desktop (RDP)

What it means

The Remote Desktop service accepts connections that do not require Network Level Authentication (NLA). With NLA, a user must authenticate before the server builds the remote session. Without it, anyone who can connect gets as far as the sign-in screen and can use the server’s resources before logging in.

Why it matters

Without NLA, the server is more exposed to password guessing and to denial of service, and it exposes more code to unauthenticated clients. The oldest RDP mode (“standard RDP security”) has weaker encryption and was the setting that the BlueKeep vulnerability (CVE-2019-0708) affected on older Windows versions.

How it is detected

The tool sends a read-only RDP connection request that offers TLS and NLA, and no login is attempted. The reply tells which security method the server selects. If the server selects NLA, a second request follows that offers only the oldest method (“standard RDP security”); the reply shows whether the server really requires NLA or would also accept TLS-only or standard security. The result is the service description. Two outcomes are reported:

  • High: the description contains “standard RDP security”: the server selected the oldest method, or accepted it in the second request. The CVE-2019-0708 hint is attached.
  • Medium: the description contains “TLS security” (and not the legacy text): the server selected TLS instead of NLA, or accepted TLS-only in the second request.

NLA counts as required only when the second request is refused (“NLA required”). The evidence is the service description.

Possible false positives

  • In the High case the title reads “legacy security without TLS or NLA”, but the same finding is raised when the server prefers NLA and merely also accepts the legacy method (“NLA (CredSSP); standard RDP security also accepted”). Read the evidence text to see which case applies.
  • The BlueKeep CVE hint is attached from the configuration alone. BlueKeep affects older Windows versions (for example Windows 7 and Windows Server 2008 R2) that lack the 2019 security updates; the tool cannot see the patch level.

How to fix

  • Windows: in System Properties, Remote, choose “Allow connections only from computers running Remote Desktop with Network Level Authentication”. Group policy: Computer Configuration, Administrative Templates, Windows Components, Remote Desktop Services, Remote Desktop Session Host, Security: enable “Require user authentication for remote connections by using Network Level Authentication” and set “Require use of specific security layer for remote (RDP) connections” to SSL.
  • Install the Windows security updates (the 2019 BlueKeep fixes for older systems) and plan the replacement of systems that no longer receive updates.
  • xrdp (Linux): set security_layer=tls in xrdp.ini.
  • Do not expose RDP to the Internet; use a VPN or a Remote Desktop Gateway and restrict the port in the firewall.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • Scan the host again: with NLA required the finding goes away; with TLS only, High becomes Medium.

References

All findings

SMBv1 is enabled

Finding idsmb1-enabled
SeverityHigh
Applies toTCP port 445 (SMB)

What it means

The server still speaks SMB version 1, the original Windows file-sharing protocol from the 1980s and 90s. It has been replaced by SMB 2 and 3.

Why it matters

SMBv1 contains design weaknesses and was the entry point of the 2017 WannaCry and NotPetya outbreaks (the EternalBlue flaw, MS17-010, CVE-2017-0144). Microsoft recommends removing it everywhere.

How it is detected

The tool sends a separate, read-only SMB1 negotiate request that offers only the dialect NT LM 0.12. If the server accepts it, or if the first reply was already SMB1, the finding is raised. Modern servers answer SMB2 requests even when SMB1 is still on, so only this separate request shows it. The evidence is the service description, extended with “SMB1 negotiate (NT LM 0.12) accepted” where applicable.

CVE-2017-0144 is attached as a hint because SMBv1 is enabled. The tool cannot see whether the MS17-010 patch is installed.

Possible false positives

  • The CVE hint does not mean the host is vulnerable: a patched system with SMB1 enabled still gets it, and this tool only sees that SMB1 is on.

How to fix

  • Windows 10/11 and Server: Set-SmbServerConfiguration -EnableSMB1Protocol $false, and remove the feature with Disable-WindowsOptionalFeature -Online -FeatureName SMB1Protocol. Check first which clients or devices still use SMB1 (the SMB audit setting AuditSmb1Access logs them).
  • Samba: server min protocol = SMB2 in smb.conf.
  • NAS and other appliances: turn SMB1 off in the file-sharing settings or update the firmware.
  • Install MS17-010 (March 2017) on systems that cannot disable SMB1 yet, and isolate them. This reduces the risk but does not clear the finding, because SMB1 stays on.
  • Old scanners, multifunction printers and legacy machines may need an update or replacement; in industrial setups clear the change with the vendor and plan it in a maintenance window.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • Scan the host again: the finding goes away once SMB1 no longer answers.

References

All findings

SMB signing is not required

Finding idsmb-signing
SeverityMedium
Applies toTCP port 445 (SMB)

What it means

The file-sharing server does not insist that messages are digitally signed. Signing proves that a message has not been altered on the way and that it comes from the right party.

Why it matters

Without required signing, an attacker in the network path can relay authentication to the server (“SMB relay”) and act with the victim’s rights. Requiring signing is the usual protection against it.

How it is detected

The tool sends an SMB2 negotiate request and reads the signing flags from the reply. “Signing disabled” or “signing enabled, not required” raises the finding; “signing required” does not. The evidence is the service description.

Possible false positives

  • Windows servers other than domain controllers do not require signing by default, so on a Windows network this finding is common and expected until a policy requires it.

How to fix

  • Windows: group policy “Microsoft network server: Digitally sign communications (always)” set to Enabled; or Set-SmbServerConfiguration -RequireSecuritySignature $true. Test first, because old clients may fail to connect and signing costs some performance.
  • Samba: server signing = mandatory.
  • NAS and appliances: look for an SMB signing option in the file-sharing settings.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

SMB server does not offer encryption

Finding idsmb-no-encryption
SeverityInfo
Applies toTCP port 445 (SMB)

What it means

The file-sharing server cannot encrypt its traffic. File names and file contents cross the network readable by anyone who can capture them.

Why it matters

Only relevant for sensitive data or untrusted networks. It is reported as information, not as a flaw: SMB encryption is an optional feature.

How it is detected

From the SMB2 negotiate reply the tool records the encryption support: SMB 3.1.1 with no cipher announced, SMB 3.0 or 3.0.2 without the encryption capability, or SMB 2.x. The finding is raised when the value begins with “not supported”. The evidence shows the dialect and the encryption value.

Possible false positives

  • Encryption is normally optional per server or per share, so a server that supports it but has it switched off looks fine; and the reverse is not detected.

How to fix

  • Windows: enable encryption for a share with Set-SmbShare -Name SHARE -EncryptData $true, or for the whole server with Set-SmbServerConfiguration -EncryptData $true. Old clients (SMB 2.x) cannot connect to an encrypted share.
  • Samba: server smb encrypt = required (or desired).
  • Update systems that only offer SMB 2.x.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

SMBv3 compression on a Windows build affected by SMBGhost

Finding idcve-cve-2020-0796
SeverityHigh
Applies toTCP port 445 (SMB), scans with Safe Mode off

What it means

The server uses SMB 3.1.1 with compression on a Windows 10 or Windows Server build (version 1903 or 1909) that was affected by SMBGhost (CVE-2020-0796), a flaw that allowed remote code execution through a crafted compressed SMB message. Microsoft fixed it in March 2020.

Why it matters

On an unpatched system the flaw can be exploited over the network without logging in. The finding is a hint to check the patch level, because the scan cannot see it.

How it is detected

The finding requires all of these: the SMB2 negotiate reply selects dialect 3.1.1; it announces a compression algorithm; and the NTLM challenge identifies the Windows build as 10.0.18362 or 10.0.18363 (Windows 10 or Server 1903/1909). The NTLM challenge is only requested when Safe Mode is off (a session setup message without credentials is sent), so in Safe Mode this finding is never raised. The evidence names the dialect, the compression and the build.

This rule is built into the program, separately from the version rules in cves.json. The patch level (KB4551762) is not visible to a remote scan, so patched and unpatched systems look the same.

Possible false positives

  • Hosts that have installed the March 2020 update or later still show the same dialect, compression and build, and are reported. Treat the finding as “check the patch level”, not as proof of vulnerability.

How to fix

  • Install the Microsoft update KB4551762 (March 2020) or any later cumulative update.
  • Workaround from Microsoft’s advisory: disable SMBv3 compression on the server: Set-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Services\LanmanServer\Parameters" DisableCompression -Type DWORD -Value 1 -Force; no restart is required.
  • Windows 10 1903 and 1909 are out of support: upgrade to a supported release.
  • Block TCP 445 at the network edge and between segments where file sharing is not needed.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • The finding goes away when the Windows build is no longer 1903/1909 or SMB compression is off. Installing the update alone does not change what a scan sees, so record the patch level separately.

References

All findings

HTTP Basic authentication over an unencrypted connection

Finding idhttp-basic-cleartext
SeverityMedium
Applies toTCP ports running plain HTTP

What it means

A web server asks for a user name and password using HTTP Basic authentication on a connection that is not encrypted. Basic authentication sends the credentials in a trivially reversible form with every request.

Why it matters

Anyone who can observe the traffic can read the user name and password. Web interfaces of devices are a common place for this.

How it is detected

The tool requests the start page over plain HTTP (not HTTPS) and reads the WWW-Authenticate response header. If the header begins with Basic (any capitalisation), the finding is raised; the evidence is that header value, usually including the realm name. Credentials are never sent.

Possible false positives

  • A server that redirects to HTTPS before it asks for authentication would not send the header on this request and is not reported.

How to fix

  • Serve the interface over HTTPS and redirect or disable plain HTTP. nginx: listen 443 ssl;; Apache: a VirtualHost on 443 with mod_ssl; IIS: add an HTTPS binding and require SSL.
  • Devices (cameras, printers, switches, controllers): turn on “HTTPS only” in the management settings, or update the firmware if it lacks HTTPS; otherwise keep the interface on an isolated management network.
  • Change the passwords that were exposed.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Docker Engine API is exposed

Finding iddocker-api
SeverityCritical
Applies toTCP ports serving HTTP or HTTPS

What it means

A web endpoint identifies itself as the Docker Engine. The Docker API, if reachable without protection, lets anyone start containers on the host.

Why it matters

Access to the Docker API is effectively root access to the host: an attacker can start a container that mounts the host’s file system and read or change anything, or install persistent backdoors.

How it is detected

The tool reads the HTTP Server header of the response. If it starts with Docker/ (the Engine API announces itself as Docker/<version> (<os>)), the finding is raised, and the evidence is the header value. The tool does not call any API function and does not check whether authentication is required.

Possible false positives

  • An API that requires TLS client certificates would normally refuse the connection before the header is seen, so such a daemon is usually not reported. A TLS endpoint without client authentication is reported.
  • A proxy or tool that imitates the Docker header.

How to fix

  • Do not publish the Docker socket on TCP without protection. Remove tcp://0.0.0.0:2375 from the daemon’s hosts setting (or -H flag) and use the local Unix socket, or SSH (docker context with an ssh:// endpoint).
  • If TCP access is needed: enable TLS with client certificate verification (--tlsverify), and limit the port in the firewall to the administrators’ addresses.
  • Check the host for unknown containers and images, and rotate secrets that containers had access to, if the port was reachable from untrusted networks.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Server header names a Heartbleed-vulnerable OpenSSL

Finding idheartbleed
SeverityCritical
Applies toTCP ports serving HTTP or HTTPS

What it means

The web server’s own description of itself mentions an OpenSSL version that was affected by Heartbleed (CVE-2014-0160), a 2014 flaw that let remote clients read chunks of the server’s memory, including private keys and passwords.

Why it matters

If the flaw is present, anyone can steal secrets from the server without leaving traces. A critical severity is assigned because of the impact if the version is genuine; the tool only sees the version text.

How it is detected

The tool reads the HTTP Server header. The finding is raised when it contains OpenSSL/1.0.1, with no letter or a lower-case letter up to f (1.0.1 to 1.0.1f). The evidence shows the version found. Version 1.0.1g and later are not reported. No Heartbleed request is sent; this is a version-based hint only.

Possible false positives

  • Distributions backported the fix without changing the version string (for example builds labelled 1.0.1e-fips that contain the patch).
  • The header may be a build-time text that does not match the library that is actually linked at run time, or may have been edited.
  • OpenSSL builds with the heartbeat feature disabled are not affected.

How to fix

  • Check the real version: openssl version -a on the host and the package changelog of your distribution (the Debian, Red Hat and Ubuntu security trackers list backports).
  • Upgrade OpenSSL to a current supported release (1.0.1 has been out of support since 2016) and restart every service that uses it.
  • If the server was running an affected version while reachable: reissue and replace its certificates, and rotate passwords and keys that passed through it.
  • Hide the version with ServerTokens Prod (Apache) or server_tokens off (nginx) if you wish, but that only hides it; it does not fix anything.
  • Embedded and industrial devices: ask the vendor for a firmware with a fixed OpenSSL and schedule it in a maintenance window.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After an upgrade the Server header no longer names OpenSSL 1.0.1 and the finding goes away. A backported fix behind an unchanged header leaves the finding in place; keep the result of your check as the record.

References

All findings

PHP release is no longer supported

Finding idphp-eol
SeverityLow
Applies toTCP ports serving HTTP or HTTPS

What it means

The web server announces a PHP version whose upstream support has ended. Releases that are out of support no longer get security fixes from the PHP project.

Why it matters

Known flaws in old PHP versions stay unfixed, and attackers scan for such servers.

How it is detected

The tool reads the X-Powered-By: PHP/x.y.z response header and checks the branch (for example 8.1) against a table of end-of-life dates built into the program (PHP_EOL in risk.rs, last reviewed 2026-11). The finding is raised when the branch’s date lies before today’s date, or when the branch is older than the oldest entry in the table. Branches that are not in the table, for example newer ones, are never reported. The evidence shows the version and the end-of-life date. “Today” is the date when the findings are computed, so analysing an old scan later can raise this finding.

Possible false positives

  • Distributions and vendors with extended support (for example enterprise Linux or paid long-term support) keep fixing old PHP versions while the header still shows the upstream version number.
  • For branches older than the table, the evidence gives the oldest table date (“before …”) rather than the true end-of-life date.
  • Servers that hide the header (expose_php = Off) are not reported at all.

How to fix

  • Upgrade to a supported PHP branch (see php.net/supported-versions) after testing the application; or use the distribution’s supported package stream.
  • If an old version must stay: confirm that your vendor still backports fixes, and keep the server isolated.
  • Set expose_php = Off in php.ini to stop announcing the version; this reduces information leakage but does not make the version safe.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • The finding goes away when the header shows a supported PHP branch or no version. A vendor-supported old branch keeps showing the old version and stays flagged.

References

All findings

SNMP answers the community string ‘public’

Finding idsnmp-public
SeverityHigh
Applies toUDP port 161

What it means

A device answers SNMP (the monitoring protocol of network devices) to the default password “public”. SNMP versions 1 and 2c use that “community string” as their only secret.

Why it matters

Anyone who can reach the port can read information about the device: system description, interfaces, addresses, running software, neighbours, often more. If a write community is also left at its default, devices can be reconfigured.

How it is detected

On UDP 161 the tool sends an SNMP GET request for the system description (sysDescr) with the community public, alternating between SNMP v2c and v1 on repeated tries. If the device returns the text successfully, the finding is raised and the evidence is “SNMP sysDescr: …” with the returned text. An answer with an error status (“community may differ”) is not reported. The tool reads one value and does not walk the device or try other community strings or write operations.

Possible false positives

  • A device that is intentionally monitored with the read community public inside a protected management network.

How to fix

  • Use SNMPv3 with authentication and encryption (authPriv), and disable v1 and v2c where possible.
  • If v2c must stay: change the community string to a long random value and restrict the source addresses to the monitoring system. Net-SNMP: rocommunity NAME SOURCE in snmpd.conf; Windows: the SNMP Service, Security tab.
  • Switches, UPS units, printers, controllers: use the vendor’s SNMP settings. Update the monitoring system at the same time, otherwise monitoring stops; plan it in a maintenance window.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

NetBIOS name service is exposed

Finding idnetbios-ns
SeverityLow
Applies toUDP port 137

What it means

The device answers NetBIOS name queries, an old Windows name-resolution service. The answer reveals the computer name, workgroup or domain and often the MAC address.

Why it matters

Names and domains help an attacker to map the network, and the old name resolution can be spoofed to capture credentials in local networks.

How it is detected

On UDP 137 the tool sends a NetBIOS node status query and records the reply. If the port answers, the finding is raised and the evidence is the decoded name list.

Possible false positives

  • Environments that still depend on NetBIOS for name resolution of old clients.

How to fix

  • Windows: disable NetBIOS over TCP/IP in the adapter’s advanced TCP/IP settings (WINS tab), or through the DHCP server’s NetBIOS option for all clients.
  • Samba: disable netbios = yes.
  • Block UDP 137 between network segments.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

UPnP/SSDP is answering

Finding idupnp-exposed
SeverityLow
Applies toUDP port 1900

What it means

The device answers discovery requests for UPnP (Universal Plug and Play) on port 1900. UPnP lets programs and devices find each other and, on routers, open ports in the firewall automatically.

Why it matters

UPnP discovery reveals what the device is and where its description is served, and a UPnP service on a router can be used by malware inside the network to open ports to the Internet. Devices that answer from the Internet can be abused for traffic amplification.

How it is detected

On UDP 1900 the tool sends a single unicast SSDP search request (M-SEARCH). If the device replies, the finding is raised, and the evidence is the decoded reply (server text, device type and description URL). It says that the device answers discovery, not that a firewall-opening UPnP function is switched on.

Possible false positives

  • Many televisions, media players, printers and routers answer SSDP on a home or office network as part of normal operation.

How to fix

  • Routers and gateways: turn off UPnP in the router settings unless you need it, and make sure UDP 1900 is not reachable from the Internet.
  • Other devices: disable UPnP/DLNA/media-sharing options that you do not use, or block UDP 1900 between network segments.
  • Industrial and embedded devices (OT): settings like this can often only be changed with the vendor’s tools or a firmware update. Follow the vendor’s guidance, try the change on a spare unit first, and apply it in an agreed change window with the plant operators informed; a changed setting can interrupt communication with a running process.
  • After the change, scan the host again to confirm that the finding is gone.

References

  • None.

All findings

Industrial protocol port is reachable from the scanner

Finding idot-reachable
SeverityMedium
Applies toTCP ports of industrial protocols

What it means

A TCP port that is normally used by industrial control protocols is open and reachable from the place where the scan ran. Industrial protocols were designed for closed networks and usually have no login or encryption.

Why it matters

Whoever can reach such a port can often read and sometimes change the process data and programs of a controller without any password. That is why industrial devices belong in a separate, controlled network.

How it is detected

The finding is based on the port number only. It is raised when one of these TCP ports is open: 102, 502, 789, 1911, 1962, 2404, 4840, 9600, 18245, 20000, 34962, 34963, 34964, 44818, 47808 or 48898, or 2222 unless the service greets as SSH (2222 is also a common SSH port). These ports are used by protocols such as S7comm, Modbus/TCP, IEC 60870-5-104, OPC UA, DNP3, PROFINET, EtherNet/IP and BACnet. The evidence reads like TCP 502 open.

Generic service detection never sends probes to these ports (to avoid disturbing fragile equipment), and this finding does not use the result of the separate --ot identification, so it does not say what really runs there. UDP ports are not part of this finding.

Possible false positives

  • Another application on a port that is also used by an industrial protocol (several of the ports are used by ordinary software too), or a simulator or test setup.
  • The finding says that the port is reachable from the scanner, which is intended information when the scan ran from an office network.

How to fix

  • Segment the network. Industrial devices belong in a control network; allow access from the office or Internet side only through a firewall or a DMZ with explicit rules for named engineering stations.
  • Restrict the port with firewall rules or the device’s own access lists to the systems that need it.
  • Follow the vendor’s guidance for each device and make firewall or configuration changes only in an agreed maintenance window, with the plant operators informed. Test on a spare device if available.
  • Prefer passive discovery (ot-listen on a mirror port) over active scanning on live control networks; see Safe Mode and OT safety.
  • After the change, scan the host again to confirm that the finding is gone.

References

All findings

Control traffic observed on an industrial protocol

Finding idot-control-traffic
SeverityMedium or High
Applies toPassively captured traffic (no port)

What it means

While listening to network traffic, the tool saw commands that change something in an industrial device, such as writing a value, loading a program, or stopping or restarting a controller. The finding names the device that received them and the machines that sent them.

Why it matters

Such commands are normal when an engineer or the control system sends them, and dangerous when someone else does. A write or a stop command can change a process or halt it. The finding helps to check that only the expected stations send them.

How it is detected

This finding comes from passive discovery (ot-listen on a live capture or a capture file, stored in the snapshot); it is never created by sending anything. The tool decodes Modbus/TCP, S7comm, DNP3, IEC 60870-5-104, EtherNet/IP and BACnet/IP and counts the functions that change state (“control traffic”) per server device, with the clients that sent them. OPC UA is recognised, but its writes are not classed as control traffic.

Examples of what counts as control traffic: Modbus write functions and the Restart Communications and Force Listen Only Mode diagnostics; S7 Write Var, program download and PLC Control/Stop; DNP3 Write, Select, Operate, Direct Operate, restart and Stop Application; IEC 104 commands, setpoints, clock synchronisation and reset process; EtherNet/IP Set Attribute, Reset and tag writes; BACnet WriteProperty, DeviceCommunicationControl, ReinitializeDevice and AtomicWriteFile.

The severity is High if the text of the observed functions contains “Stop”, “Download”, “Restart” or “PLC Control”; otherwise Medium. The evidence reads like “Modbus/TCP server: control traffic observed (Write Single Register x2) from 10.0.0.5”. The finding has no port.

Possible false positives

  • Routine writes from the real SCADA or engineering station are exactly what this finding shows; it does not say the traffic is unauthorised. Many sites will see it on every scan of a live plant.
  • A capture only shows the traffic that passes the capture point, and the decoders were tested with synthetic traffic only (see OT / industrial).
  • The severity depends on the wording of the function names. Some state-changing commands, for example a CIP Reset, a BACnet ReinitializeDevice or a DNP3 Direct Operate, are rated Medium.

How to fix

  • Check the clients named in the evidence against the list of authorised engineering and control stations. Unknown sources deserve an investigation, with the plant operator.
  • Restrict who may send commands: firewall or industrial-firewall rules between zones, and the access protection of the controller (password-protected programming access, keyswitch/mode settings) as described by the vendor.
  • Segment and monitor the control network and keep a baseline of expected writers (see ISA/IEC 62443 zones and conduits).
  • Make changes on controllers and network equipment only with the vendor’s guidance and in agreed change windows; never test commands on a running process.

References

All findings

Version-based CVE hint

Finding idcve-cve-YYYY-NNNN, for example cve-cve-2021-41773
SeverityAs set by the rule (Info to Critical)
Applies toAny service whose product and version were detected

What it means

The detected product and version match a rule from the CVE list that ships with the program (or your own list), which names a known vulnerability in that version range. The finding id is cve- followed by the first CVE id of the rule in lower case, which is why it reads cve-cve-.... The title is the one given by the rule.

Why it matters

Old versions of software often contain flaws with public exploits. The finding points you to the version that should be checked. It is a hint, not a confirmed vulnerability.

How it is detected

For each open port, the tool takes the product and version from the service banner (and, for web servers, the Server header) and compares them with the rules in data/cves.json (plus a user file, see Findings and CVE hints). A rule matches when its product text occurs in the detected product, the product is not on the rule’s exclusion list, and the version lies in the rule’s range. Each CVE is reported once per port. Severity and title come from the rule. The evidence is “product version” followed by a note that this is a version-based hint only.

The SMBGhost hint (cve-cve-2020-0796) and the Heartbleed hint (heartbleed) are separate, built-in rules with their own pages.

Possible false positives

  • Backports. Linux distributions and vendors fix flaws without changing the version number, so a patched system can still be flagged. The version text alone cannot settle this.
  • Version strings in banners can be missing, edited or from a different component.
  • The rule may not apply to the way the software is configured (a vulnerable module that is not loaded).

How to fix

  • Check the vendor advisory for the CVE and your product (the NVD entry links to it).
  • On Linux look at the package changelog (apt changelog PACKAGE, rpm -q --changelog PACKAGE) or the distribution’s security tracker to see whether the fix is included in your package version.
  • Update to a fixed release, or apply the vendor’s mitigation, and re-scan.
  • Embedded and industrial devices: ask the vendor for a fixed firmware and a statement whether your model is affected; apply in a planned change window.
  • The finding goes away when the detected version leaves the rule’s range. If you confirmed that the vendor backported the fix, the version text does not change and the finding stays; keep the result of your check as the record.

References

All findings

Finding raised by a user script

Finding idscript-<id>, where <id> is chosen by the script
SeverityAs set by the script
Applies toHosts processed by a user script

What it means

A script that you or your organisation added to Network Scope produced this finding. The program itself has no opinion about it: the id, severity, title and evidence come from the script.

Why it matters

Scripts express site policies such as “Telnet is not allowed” or “this host must not run a web server”. The importance of a finding therefore depends on that policy.

How it is detected

A script calls h.finding("severity", "id", "title", "evidence"). The result is stored with the host and reported with the id script- plus the id given by the script, no port, the severity given (an unknown severity becomes info), and the advice “Raised by a user script”. A | character in the title is replaced by /. See the Scripting page for the script interface.

Possible false positives

  • Depends entirely on the script’s logic.

How to fix

  • Follow the policy that the script’s author encoded. The script’s source (and its comments) is the place to find out what it checks and why.
  • To change or remove the finding, edit the script.

References

All findings

OT / industrial

Network Scope identifies industrial devices in two ways: active identification with single read-only requests on standard ports, and passive discovery from captured traffic, which sends nothing. Read Safe Mode and OT safety before using either on a production network.

Active identification (--ot)

All OT operations are identification-only, even when Safe Mode is off. OT identification probes standard ports only, and only for hosts that responded to something (ARP, ICMP, any TCP answer or UDP).

In the desktop, select Use industrial ports and enable nonresponders to find ICMP-silent TCP services. The CLI does both when --ot is used without --ports (ports 80,443,102,502,4840,44818).

ProtocolPortWhat is sentWhat is read
Modbus/TCP502One Read Device Identification request (function 43/14), configurable unit ID (--unit)Vendor, product, firmware when returned; additional object pages are marked partial
S7comm102COTP connection, communication setup, read-only SZL 0x0011/0x001C, configurable rack/slot (--rack, --slot)Order code, CPU type, firmware, serial when advertised
EtherNet/IP44818TCP List IdentityVendor ID (with ODVA vendor name), product, serial, revision
BACnet/IP47808/udpOne unicast Who-Is per IPv4 target (responding hosts, plus silent hosts when nonresponders are scanned)Device instance
OPC UA4840UA Binary HEL/OpenSecureChannel, GetEndpoints and FindServers without a session, CloseSecureChannelNames, application/product URIs, security policies, certificate DER as hex

Details and limits:

  • S7: segmented SZLs, denied access and unsupported devices return explicit errors. No S7plus, authentication, program/block access or control operations.
  • 2222/tcp: only an SSH greeting leads to SSH probing; anything else is treated as industrial (EtherNet/IP) and gets the ot-reachable finding.
  • BACnet/IP: waits the port timeout clamped to 200-2000 ms; ICMP port unreachable means “no BACnet” without an error. Silent hosts get this Who-Is only when nonresponders are scanned (the CLI does with --ot), so a device that answers only BACnet/IP is found; each silent address costs at most that wait. No broadcast and no property writes. IPv6 BACnet and forwarded/router NPDU forms are not supported.
  • OPC UA: only unsecured sessionless discovery is supported; servers denying it produce errors. The product URI is not a verified manufacturer identity. Discovery certificates are inventory data, not trusted certificates. Message sizes, chunk counts and parser recursion are bounded.
  • Profinet DCP: dcp --interface ID --source-mac MAC (or the desktop’s DCP Identify button) sends one read-only Identify multicast request. It never assigns IPs or names and never changes a device.

Generic service probes and the UDP scan never touch industrial ports; see Detection.

Passive OT discovery

network-scope-cli ot-listen --pcap plant.pcapng --progress
network-scope-cli ot-listen --interface ID --seconds 600
network-scope-cli ot-assets ID --format csv --output assets.csv

Passive mode (ot-listen) reads traffic and never transmits. A live capture goes through a wrapper with no send call, and a test fails the build if the passive modules contain socket or send code.

  • Capture files are parsed in pure Rust: pcap and pcapng with Ethernet, Linux cooked (SLL/SLL2, e.g. tcpdump -i any) and raw IP link types, pcapng if_tsresol (nanosecond captures), FCS stripping. Other link types fail with a message naming the type. --progress prints percent, frames and frames/s on stderr.
  • Recognised protocols: Modbus/TCP, S7comm, DNP3, IEC 60870-5-104, EtherNet/IP (List Identity, CIP services including Unconnected Send / Multiple Service Packet unwrapping; object-specific services decoded by path), BACnet/IP and OPC UA.
  • TCP reassembly: messages split across TCP segments are reassembled in order for Modbus, S7, IEC 104, EtherNet/IP and OPC UA (16 KiB per flow, 1,024 flows, 4 MiB total). A missing segment means that message is decoded only as far as it was captured. DNP3 is still per segment. --no-reassembly turns it off. On those ports the packets counter counts messages, not segments.
  • Control traffic (writes, program download, PLC stop, operate and restart commands, setpoints, WriteProperty, ReinitializeDevice, …) is counted per server and client and raised as ot-control-traffic findings.
  • Asset inventory: ot-assets ID lists address, vendor, product, firmware, serial, protocols, roles, first/last seen and control traffic, as JSON or CSV. Each capture is saved to the inventory as ot-passive:….

Limits: IPv4 only, standard ports only, decoders validated with synthetic frames only, not with real industrial traffic. A SPAN/mirror port or tap is needed to see more than your own segment.

General passive discovery

listen --interface ID --seconds N (and the desktop’s Passive section) decodes mDNS, SSDP/UPnP, LLDP, CDP, NetBIOS, WS-Discovery, DHCP and ARP without sending discovery packets. MAC-keyed observations retain devices that advertise no IP. What each protocol contributes to device identification is in Detection. Switch and PLC neighbours that announce themselves with LLDP or CDP (name, port, platform, software, management address) appear as observations with those fields.

  • Passive IPv4/IPv6 UDP decoding supports ordinary unfragmented packets; IPv6 extension headers and reassembly are not implemented. Ethernet VLAN tags are supported.
  • Requests to broadcast or multicast destinations record only the client.
  • Observations are bounded to 10,000; a full capture queue or the cap marks the capture partial and collection continues with what still fits. The capture thread tolerates transient read errors.
  • Passive and DCP records are kept alongside the scan inventory but do not imply open TCP ports.

Protocol references

OPC UA mappings, Modbus specification, Npcap API.

Safe Mode and OT safety

Safe Mode

Safe Mode is enabled by default in the desktop and on the CLI. --fast (or unchecking Safe Mode) turns it off.

Safe Mode (default)--fast
Active hosts at onceat most 8--workers (default 1024, max 4096)
Spacing between application probesat least 100 ms (token bucket)none, unless --rate PPS
Port probes per host in parallelat most 4--parallel (default 64)
SYN scan, ARP sweepdisabledopt-in
TracerouteICMP onlyICMP, TCP, UDP
SMB NTLM SESSION_SETUPnot sentsent (no AUTHENTICATE)
TLS 1.3 suite enumerationnot sentsent
OT operationsidentification onlyidentification only

In the desktop the disabled controls are greyed out. Safe Mode is a conservative traffic profile, not a guarantee for every PLC.

Rules that always apply

These hold with Safe Mode on or off:

  • All OT protocol operations are read-only identification: no writes, no program or block access, no control commands, no authentication. See OT / industrial for exactly what is sent.
  • OT identification runs only on standard ports. TCP probes go only to hosts with that port open; a silent host gets at most one BACnet Who-Is, and only when nonresponders are scanned.
  • Generic service probes never touch industrial ports (102, 502, 4840, 44818, 47808 and similar).
  • Printer ports 9100-9107 and 515 only get a passive banner wait; nothing is ever sent.
  • 2222/tcp is listen-only unless it greets with SSH.
  • The UDP scan never sends to industrial or printer ports; such ports in --udp-ports are skipped with a host note and probes listing them are rejected.
  • TCP traceroute never uses OT or printer ports.
  • Passive modes (listen, ot-listen, pcap import) send nothing. A test fails the build if the passive modules contain socket or send code.
  • Nothing scans automatically; every scan starts with an explicit Start or command.

The read-only promise is a type, not a convention

The active OT code (src/ot.rs, src/ot/) cannot send a write or control request, because such a request has no value in the type system:

  • Every OT request is built by ot::readonly from an allow-list enum: ModbusFunction, S7Szl, EnipCommand, BacnetService, OpcUaMessage. The request types (ModbusRequest, S7Request, EnipRequest, BacnetRequest, OpcUaRequest) have private fields, so the enums are the only way in.
  • The only functions that put OT bytes on a socket, OtStream::send and OtUdp::send, accept nothing but a ReadOnlyRequest. That trait is sealed (only ot::readonly implements it) and is not implemented for &[u8]. The rest of the OT code holds sockets only through these wrappers, which can read but not write; a test fails the build if ot.rs or ot/opcua.rs use a raw socket or write_all.
  • Sending anything else needs a new enum variant, which is a reviewed code change, and the tests fail until the allow-list in the test is changed deliberately too.

What the allow-lists contain (exactly what is sent today):

ProtocolAllowed
Modbus/TCPfunction 0x2B with MEI type 0x0E (Read Device Identification); not the other MEI types (for example 0x0D CANopen), not any write or diagnostic function
S7commCOTP connection request, Setup communication (job function 0xF0), Userdata CPU functions / Read SZL for SZL 0x0011 and 0x001C; no Read/Write Var, PLC control or stop, download, upload or block access
EtherNet/IPencapsulation command 0x0063 (List Identity); no SendRRData, SendUnitData or sessions
BACnet/IPunicast Who-Is (unconfirmed service 8); no confirmed requests (ReadProperty, WriteProperty, ReinitializeDevice) and no broadcast BVLC function
OPC UAHEL, OpenSecureChannel (node 446), GetEndpoints (428), FindServers (422), CloseSecureChannel (452), SecurityPolicy None; no CreateSession, Read, Write or Call

Tests (cargo test) check this for every possible byte: all 256 function codes (and all 65,536 MEI/group/subfunction pairs, EtherNet/IP commands and OPC UA node IDs) are accepted by the constructors exactly when they are on the allow-list, every encoded request decodes back to an allow-listed service, changing any discriminator byte of an encoded request to any other value is rejected, and the encoded bytes equal the literals used before the types existed. compile_fail doctests in ot::readonly show that a Modbus write, an S7 stop and a raw &[u8] send do not compile.

Limits of the guarantee: it covers the active OT requests listed above. The Profinet DCP Identify frame (dcp, built in passive.rs) and the generic service probes are separate code paths with their own tests and deny lists, and the type does not say anything about what a device does with a read-only request.

Recommendations for plant networks

  • Read-only traffic can still affect fragile equipment. Older controllers, serial gateways and devices with small TCP stacks may slow down, drop connections or fault under any active traffic.
  • Prefer passive discovery (ot-listen on a SPAN port or an existing capture) on production OT networks.
  • Coordinate with operations before an active scan, scan during a maintenance window, and start with a small target range.
  • Keep Safe Mode on. Do not use --fast, --syn or --arp-sweep against control networks.
  • Know how to stop: Stop in the desktop, or Q/Escape in the TUI, saves what was collected.

See also Responsible use.

Lab and testing

Automated tests

cargo test --locked
cargo clippy --all-targets --locked -- -D warnings
cargo fmt --check

The tests exercise real loopback networking and protocol fixtures, SQLite round trips, scan changes, parser bounds, report escaping, and the SYN, ARP and TCP-traceroute I/O loops against a fake wire. A source-level test fails the build if the passive capture modules contain socket or send code.

CI (.github/workflows/ci.yml) runs format, clippy with -D warnings, the tests and the lab check below on Windows, Linux and macOS, uploads both binaries as build artifacts, and builds the Windows installer. On Linux it also runs the accuracy benchmark and fails when an accuracy metric drops below the baseline. Tagged releases are built by release.yml.

Fuzzing

The parsers of untrusted input (capture files, Ethernet frames for the passive listener and the OT tracker with TCP reassembly, TLS records and certificates, SMB, NTLM, SSH, HTTP and RDP replies, service banners, Modbus, EtherNet/IP, BACnet, S7 and OPC UA replies, the UDP probe replies, ARP, SYN-ACK/RST and TCP options, ICMP/ICMPv6 ping and traceroute replies, nmap XML and the user probes.json and cves.json) have cargo-fuzz targets in fuzz/, one per parser, with seed inputs in fuzz/corpus/. The seeds also run as a unit test, so every input that once crashed a parser stays a regression test. ClusterFuzzLite (.github/workflows/cflite.yml) fuzzes pull requests for 5 minutes and all targets weekly for 30 minutes. How to run a target locally is in CONTRIBUTING.md.

Lab simulators

lab/simulators.py (Python standard library only) emulates deliberately weak services: an FTP banner of vsFTPd 2.3.4, Telnet, SSH offering weak algorithms, Apache 2.4.49 with Basic auth, open Redis, Modbus/TCP, EtherNet/IP and BACnet/IP identity. They are fake: they only produce what a scanner observes and never execute anything received from the network.

python lab/simulators.py --bind 127.0.0.1 --offset 20000   # unprivileged ports 20021, 20022, ...

End-to-end check

cargo build --release --locked
python lab/verify.py target/release/network-scope        # .exe on Windows

verify.py starts the simulators on loopback, scans them with the built binary and asserts the expected identities and findings (exit code 0 = all passed). CI runs it on Windows, Linux and macOS.

Accuracy benchmark

lab/benchmark.py measures how accurate a scan is against a lab where every host and port is known. The ground truth is lab/ground_truth.json: six simulated hosts on loopback addresses 127.0.10.1 to 127.0.10.6 (a weak Unix server, a PLC with Modbus, EtherNet/IP and BACnet, a web server, a Debian web server, a Redis server, an IIS web server) and three addresses where nothing runs. The benchmark starts the simulators, scans every address with --fast --deep --ot --tcp-only, compares the JSON result with the ground truth and writes benchmark.json and benchmark.md.

cargo build --release --locked
python -I lab/benchmark.py --cli /absolute/path/to/target/release/network-scope-cli --check

Only loopback addresses (127.0.0.0/8) are ever contacted; a ground truth with any other address is refused. Before starting, the benchmark checks that nothing already listens on the lab ports (for example a web server on a wildcard address) and that every simulator accepts connections; a dirty machine exits with code 2, which is a lab error, not a regression. Port 80 is not used because Windows machines often run IIS there. On Linux the lab needs the low ports (sudo sysctl -w net.ipv4.ip_unprivileged_port_start=0, and port 22 must be free).

MetricMeaning
Host recall / precisionReal hosts found / real hosts; real hosts / hosts reported with an open port or identity (an absent address reported that way counts against precision)
Port recallOpen ports found / open ports; also split into TCP and UDP (udp_open)
Port precisionReported open ports that are real / reported open ports
Service accuracyServices named correctly / services served (svc.<port>, or the OT identity fields for Modbus, EtherNet/IP and BACnet)
Product accuracyProduct and version correct / products with a known product (for example vsFTPd 2.3.4, Apache 2.4.49)
Device type, OS accuracyCorrect / hosts with a known device type, or a known OS (named by their banners)
Wall timeReported, not gated

--check exits with code 1 when an accuracy metric is below lab/benchmark_baseline.json (one entry per platform). CI runs it on Linux after the lab check, writes the table to the job summary and uploads benchmark.json as the benchmark-Linux artifact. Each release attaches network-scope-<version>-benchmark.json and .md, measured with the Linux release binary. To raise the baseline after an improvement (or to record a new platform), run --update-baseline and review the diff.

Numbers of the last local run (Windows 11, debug build of the 0.5.1 sources, 2026-11-19; the first run on Linux in CI is what the Linux baseline will be recorded from):

MetricResultFound
Host recall100.0%6/6
Host precision100.0%6/6
Port recall91.7%11/12
TCP port recall100.0%11/11
UDP port recall0.0%0/1
Port precision100.0%11/11
Service accuracy100.0%12/12
Product accuracy100.0%8/8
Device type accuracy100.0%4/4
OS accuracy100.0%2/2

What the numbers do not say:

  • The one miss is real: the BACnet simulator is identified through --ot (BACnet.device_id), but UDP 47808 is not listed in udp_open, which only reports the default UDP probe ports (DNS, DHCP, NTP, NetBIOS, SNMP, SSDP, mDNS). Nothing in the lab serves those, so the UDP check is a single port.
  • The lab is small and cooperative: every service answers its own protocol cleanly, nothing is firewalled, filtered, slow or lossy, and no real network or operating system is involved. 100% here means “no regression on these cases”, not “100% in the field”. Loopback also hides ICMP: every 127.x.x.x address answers ping, so the gated run uses --tcp-only; with ICMP discovery the absent addresses would be reported as live hosts (a lab artefact, not a scanner bug). The OS also answers every port of every 127.x address with a RST, so the absent addresses respond to TCP too; host precision therefore counts only hosts reported with an open port or an identity.
  • Only OS and device types that the simulators’ banners name are in the ground truth (Debian, IIS, a Redis server, a PLC). TTL-only OS guessing is not measured, and the simulators on 127.0.10.1 to 127.0.10.3 carry no OS evidence, so the tool correctly names none.
  • Only loopback addresses are used, so ARP, SYN scanning and routing are not covered.

Docker lab

lab/docker-compose.yml runs the simulators (plus a real Redis) on an internal Docker network 172.28.0.0/24 with no outside access:

cd lab
docker compose up -d --build
network-scope scan 172.28.0.0/24 --deep --ot --script ../scripts

The Docker compose file has not been run (no Docker on the development machine); verify.py has, on Windows. The lab has no S7, DNP3, IEC 104 or OPC UA simulator.

What has been tested on real hardware

Verified on real hardware (2026-11-18, one Windows 11 machine with Npcap, Wi-Fi adapter): a Safe Mode scan of a real /24 (78 responding hosts, 187 s), an on-link SYN scan through Npcap (open and closed ports correct, no fallback), a passive listen (ARP and LLDP decoded), and network-scope-cli in cmd.exe and PowerShell (both wait; exit code 0 on success, 1 on a bad option).

What has not been tested

Be aware of these gaps before relying on a feature:

  • Npcap/libpcap features not yet run on a real interface: SYN to routed targets, the ARP sweep, TCP traceroute, live OT capture, DCP; and everything with libpcap on Linux/macOS.
  • UDP traceroute with administrator/root rights; ICMPv6 hop-limit expiry from a real router on Windows.
  • Linux/macOS glue (next_hop, neighbor states, raw ICMP/ICMPv6, UDP hop): not compiled on the development machine (CI is the first compile) and not exercised on a real network.
  • The GUI binary’s console attach when started from a shell (use network-scope-cli for scripting).
  • The Inno Setup installer script (never compiled with iscc yet) and the CI, release and documentation workflows (not run yet). That includes the accuracy benchmark steps in ci.yml and release.yml; the benchmark itself has only been run on Windows, so the Linux baseline is not recorded yet.
  • The ClusterFuzzLite build (.clusterfuzzlite/) and workflow (cflite.yml): not run yet. The fuzz targets themselves were built and run locally on Windows (cargo-fuzz, AddressSanitizer), every target for at least 3 minutes, without a crash.
  • Webhook/syslog delivery against real receivers; SMB against Samba; TLS 1.3 enumeration against OpenSSL/Schannel; JA4S against the reference implementation; CPE names against the NVD dictionary.
  • OT decoders and reassembly against real industrial traffic (synthetic frames only); industrial-device interoperability in general.

No industrial production network was scanned during development. Reports from real-world use, especially from OT networks, are very welcome; see the protocol support issue template.

Screenshots

cargo run --release -- --screenshot screenshot.png renders the desktop to a PNG without scanning or modifying preferences or the inventory. NETWORK_SCOPE_SCREENSHOT_STATE=main|findings|devices|settings|ot|script renders a larger window with sample data in that view (append -small for the default 800x450 size, then -fast for Safe Mode off, e.g. main-small-fast).

Data sources

Network Scope embeds four data files at build time. Their provenance and licensing are documented in data/README.md.

FileContentSource
data/oui.tsvMAC address prefix → organization nameIEEE Registration Authority public registries (MA-L, MA-M, MA-S)
data/cip_vendors.tsvEtherNet/IP (CIP) vendor ID → vendor nameODVA CIP vendor ID list, as extracted from a public table (see data/README.md)
data/probes.jsonService probes and match rulesOriginal to Network Scope
data/cves.jsonVersion-based CVE hints and CPE templatesOriginal to Network Scope

Imported feeds

CISA KEV and FIRST EPSS are not embedded. You download them and import them with feeds import; the cache is feeds.json next to the user cves.json (or --feeds FILE). See Findings and CVE hints for the download addresses and what changes.

User overrides

probes.json and cves.json can be extended without rebuilding:

  • %LOCALAPPDATA%/NetworkScope/probes.json or --probes FILE: entries with the same name replace built-in probes, new names are added. See Detection.
  • %LOCALAPPDATA%/NetworkScope/cves.json (or $XDG_CONFIG_HOME/NetworkScope/cves.json) or --cves FILE: entries with the same id replace built-in rules. See Findings and CVE hints.

Limits of the data

  • The MAC vendor list reflects the registry on its retrieval date; newer assignments are unknown. Locally administered (randomised) MACs are marked as such instead of looked up.
  • CPE vendor/product names in cves.json were written from memory of the NVD dictionary and have not been checked against it.
  • The top-ports ordering is approximate, recalled from public scanner statistics, and not generated from any third-party services file.

Third-party libraries and their licenses are listed in THIRD-PARTY-NOTICES.md.

FAQ and troubleshooting

Do I need administrator rights or Npcap?

No, not for normal use. TCP connect scanning, service detection, ICMP/ARP through the OS, UDP probes, inventory, reports, monitoring and pcap file analysis all work without them. SYN scanning, the ARP sweep, TCP traceroute, live capture and DCP need Npcap (Windows) or libpcap plus capture rights; UDP traceroute needs administrator/root. See Installation.

The shell does not wait for network-scope.exe and $LASTEXITCODE is wrong

network-scope.exe is a GUI-subsystem program. Use network-scope-cli.exe for scripts: it is a console build of the same program. If you must use the GUI binary, run it with Start-Process -Wait -NoNewWindow or pipe it to Out-Host.

Windows SmartScreen warns about the download

The binaries and installer are not code-signed. Verify the file against SHA256SUMS from the release page.

“SYN unavailable; TCP connect fallback: …” in the host notes

The SYN engine could not be used: Npcap/libpcap missing, no capture rights, wrong --interface/--source-ip/--source-mac, a silent gateway, or the route leaves through another interface. The scan continued with TCP connect. Check network-scope-cli interfaces and make sure the source values belong to that interface. SYN is never used in Safe Mode; add --fast.

“ARP sweep unavailable; per-host lookup: …”

Same causes as above, or the target is routed or IPv6. The ARP sweep only works for on-link IPv4 targets.

The scan is slow

Safe Mode is on by default (8 hosts, 100 ms between application probes). On networks without fragile equipment use --fast. Silent, firewalled addresses are limited by --timeout and --retries; lower them if appropriate.

A host I know is up is reported as not responding

A timeout never proves a device is offline. The host may drop ICMP and have none of the scanned ports open. Every CLI scan also checks the ports of nonresponders; add more --ports or --top-ports, or --arp / --arp-sweep on the local network. On Linux a STALE neighbor entry does not count as proof that a host is up.

A device is missing from the OT results

OT identification runs on standard ports of hosts that answered something. In the desktop, enable Use industrial ports and scanning of nonresponders; with nonresponders enabled, silent hosts also get the single BACnet Who-Is, so devices that answer only BACnet/IP are found too. Passive discovery only sees the traffic that reaches your interface; use a SPAN/mirror port or tap.

A CVE finding is wrong; the host is patched

CVE entries are version-based hints. Distributions backport fixes without changing version strings. Confirm with the vendor advisory or an authenticated check. A user cves.json entry with the same id replaces the built-in rule, for example with a narrower version range or an exclude list.

The diff shows changes I do not care about

Use --ignore with exact keys, globs (http.*.title) or suffixes (.server). *.days_left, *.clock_skew and *.boot_time are always ignored.

My webhook uses HTTPS

--webhook https://… works. The server certificate is checked against the operating system’s certificate store, and verification cannot be turned off. If delivery fails with “certificate … rejected”, the certificate is not issued by a CA the system trusts, has expired, or does not match the host name (or IP address) in the URL. Install your private CA in the system certificate store, or use the receiver’s real host name instead of an IP address. See Alert delivery.

Where are my settings and the database?

%LOCALAPPDATA%\NetworkScope on Windows: preferences, inventory.sqlite, and optional user probes.json / cves.json.

Reporting a bug

Open an issue with the bug report template. Redact IP addresses, MAC addresses and host names from logs and screenshots. Security issues go through private vulnerability reporting, see SECURITY.md.

Changelog

The full changelog, in Keep a Changelog format, is maintained in the repository: CHANGELOG.md.

Release downloads and per-release notes: GitHub Releases.

0.5.0 at a glance

  • Console binary network-scope-cli for scripts on Windows.
  • SYN to routed IPv4 targets, adaptive SYN timing, ARP sweep, TCP/UDP/ICMPv6 traceroute.
  • SMB 3.1.1 negotiate contexts and NTLM challenge, TLS 1.3 suite enumeration, JA4S-style and HASSHServer fingerprints.
  • CVE hints from data/cves.json with user overrides, CPE names; UDP payload probes.
  • Diff formats with new/resolved/unconfirmed findings; device history; Markdown and JSON Lines exports.
  • Monitor alert sinks (webhook, syslog) with de-duplication.
  • OT TCP reassembly, more pcap link types, passive OT section in the GUI.

Read the “Behaviour changes”, “Not done” and “Untested” sections of the changelog before upgrading.