
CamSniff is a shell-based reconnaissance toolkit that builds reproducible, schema-versioned inventories of IP cameras and network video endpoints for authorized security assessments.
| Tool | John0n1/CamSniff — automated IP camera and network video stream reconnaissance toolkit (Shell, MIT) |
| Category | Network reconnaissance / asset inventory for video endpoints |
| Primary Use | Mapping authorized networks to enumerate RTSP, ONVIF, SSDP, mDNS, HLS, RTMP, SRT, and WebRTC video services and produce structured discovery.json evidence |
| Safe Use | Authorized penetration tests, internal asset audits, and defensive network inventories on networks you are explicitly permitted to assess; the README's own caution restricts use to authorized systems |
| Telemetry Note | Generates visible Nmap/Masscan scan traffic, RTSP OPTIONS/DESCRIBE requests, ONVIF probes, and short TShark captures — all easily logged by IDS/IPS and flow monitoring; artifacts land in 0700 run directories with credentials.json and thumbnails |
CamSniff from John0n1 is a Shell-scripted reconnaissance toolkit, currently at version 2.3.0 and carrying an MIT license with roughly 151 stars, that folds the messy problem of IP camera discovery into a single reproducible pipeline. Its central thesis is explicitly stated in the README: a common open port is not proof of a particular vendor or device type. Instead of pattern-matching on port 554 and calling it a day, the tool correlates active scans, local service discovery, short packet captures, protocol probes, and a vendor catalogue into structured evidence with confidence scoring. That framing alone distinguishes it from the flood of camera "scanner" scripts that saturate this niche.
Architecturally, a run flows through several phases. Nmap — optionally fronted by Masscan — establishes the port baseline, while SSDP, mDNS, and a bounded TShark capture gather passive and near-passive context. Focused UDP, HTTP, and ONVIF probes then follow, feeding an evidence correlation and confidence-scoring engine. The README describes this in a flow diagram, and two details reveal careful engineering: every discovered address is checked against the original target list before it can become a candidate or receive follow-up probes, and TShark request traffic is interpreted directionally so client devices are never mistakenly promoted as camera servers. Both are classic false-positive controls rarely seen in this category.
Protocol coverage is unusually broad. CamSniff correlates RTSP, HTTP, ONVIF, SSDP, mDNS, HLS, RTMP, SRT, WebRTC, and optionally CoAP signals. Critically, RTSP verification is bounded and non-intrusive: OPTIONS and DESCRIBE requests parse SDP media tracks without authenticating or starting playback. ONVIF enumeration runs under a fixed six-request budget, pulling identity, services, capabilities, scopes, network interfaces, and system time — but only what the device exposes unauthenticated. This is reconnaissance designed to characterize, not to consume streams.
Version 2.3.0 is described as the release that turns the discovery pipeline into an evidence-driven identification engine. Smart mode (--smart) now plans follow-up probes from the same confidence model used in reports, and the fingerprinting layer moved to declarative YAML vendor profiles under data/vendors/. These correlate HTTP, ONVIF, RTSP, SSDP, path, and OUI evidence without treating shared ports as identity, and the contribution guidelines explicitly discourage broad credential lists or unverified vendor and CVE claims. Negative evidence also factors into scoring — the discovery schema shows hosts carrying both positive_score and negative_score fields, so a host can accumulate evidence against a camera classification.
The mode system maps noise levels to authorization realities. Six profiles — stealth+, stealth, medium, aggressive, war, and nuke — trade port coverage, Masscan usage, credential caps, and probe windows. stealth+ is a minimal 90-second check with a 16-credential cap for the lowest-volume targeted validation, medium is the default balanced audit, and nuke scans ports 1–65535 over a 240-second window, which the README confines to isolated labs or explicitly scoped full audits. The naming of the top tiers is theatrical, but the table's intent column shows deliberate restraint in how each profile is positioned.
Credential checking exists but is opt-out-first in spirit: the documented "safe first run" recipe passes --skip-credentials and uses stealth mode against an explicit --targets file containing authorized IPv4 addresses and CIDRs. When credential attempts are enabled, they are bounded by the mode's cap, validated against actual media rather than accepting any HTTP response, and thumbnail capture requires a syntactically valid video stream — the troubleshooting section notes a 200 response alone is insufficient. From a defender's perspective, all of this is noisy: Nmap and Masscan traffic, RTSP method requests, ONVIF SOAP probes, and SSDP DESCRIBE fetches are all readily observable in flow data and IDS rules.
Output hygiene is where CamSniff shows unusual operational maturity. Runs write schema-versioned discovery.json, credentials.json, a redacted Markdown/HTML report, a vendor catalogue snapshot in paths.json, validated thumbnails/, and per-phase logs/. Run directories are created 0700 with an owner-only umask, rendered reports redact passwords and URL userinfo, and --encrypt-results produces an age/GPG encrypted archive — with the sensible warning to verify the archive before deleting plaintext. The discovery schema (version 2) represents TCP and UDP transports explicitly, recording per-evidence kind, source, weight, reason, and polarity, which makes the output auditable rather than a black-box verdict.
Integration is a differentiator for teams running continuous asset intelligence: --extra ivre sets up and synchronizes discovery and credential results into IVRE, the open-source network recon framework, letting CamSniff act as a camera-focused feeder for a broader asset database. There is also an upstream intelligence sync mechanism with provenance tracking, review-only candidates, and automated update PRs for vendor data — a governance model for fingerprint freshness that most comparable tools completely lack.
Packaging is genuinely production-grade. GitHub Actions workflows independently build Debian and Arch packages, the repo's PKGBUILD packages the current checkout for release CI, and make test, make lint (shellcheck plus Bash syntax), make arch-verify, and make build gate development. The regression suite is described as covering target containment, SSDP URL scope, TShark CSV parsing, Masscan contracts, TCP/UDP preservation, report redaction, profile confidence, and credential/media probing behavior — that list reads like a catalogue of the exact bugs that plague ad-hoc scanner scripts.
Installation is standard and non-weaponized: sudo apt install ./camsniff_2.3.0_all.deb on Debian-family systems, or git clone https://github.com/John0n1/CamSniff.git followed by sudo ./bin/camsniff --mode medium. Missing dependencies can be installed on first run, with --skip-install deferring to OS-managed packages; notably, the optional coap-client is no longer built from unpinned network source during ordinary scans, an apparent supply-chain hardening decision.
Where does this fit in an authorized workflow? As a scoping and inventory instrument for internal camera estates, physical security audits, and penetration tests where video infrastructure is in scope, CamSniff replaces a stack of manual nmap + curl + ffmpeg incantations with something reproducible and evidence-annotated. Watch items: credentials.json and captured thumbnails are sensitive by the project's own admission, default-interface targeting (used when --targets is omitted) should be confirmed against your authorization boundary, and higher modes generate exactly the traffic volumes their names imply. Used within scope, it is one of the more disciplined entries in the camera-recon space.
John0n1/CamSniff.Educational analysis for authorized security professionals. Use only in controlled, authorized environments.
Related coverage
0 comentários:
Post a Comment
Note: Only a member of this blog may post a comment.