Repository navigation
Description
#437 asked for TLS metadata without decryption, and named JA3/JA3S. fenrir now logs SNI, certificate subjects and issuers, and chain validation (ssl.log, x509.log), but not fingerprints: the stock zeek/zeek:9.0.0 image in stacks/sensor ships no JA3 or JA4 script. ADR-0068 and stacks/sensor/zeek/local.zeek both record the gap.
Why JA4, not JA3
JA3 hashes the ClientHello's extension order, and browsers have randomised that order since 2023, so one client produces many JA3 hashes. JA4 sorts before hashing. The Zeek implementation is FoxIO's package, zkg install zeek/foxio/ja4.
Check the licence before choosing what to load. JA4 itself (the TLS client fingerprint) is BSD-3. The rest of JA4+ (JA4S, JA4H, JA4X, JA4SSH and the others) is under the FoxIO License 1.1. That is likely fine for a home lab, but the decision should name it.
The decision this needs
Nothing in this repository builds an image. Every image is pulled and digest-pinned from a compose.yaml, and scripts/check_image_pins.py enforces that. A zkg package has to get into the container somehow:
- A derived image built here: a
build:instacks/sensor/compose.yaml,FROMthe pinned digest, pluszkg installat a pinned package version. This is the first locally built image in the estate.check_image_pins.py,check_compose_health.py --probeand Dependabot would each need to learn about it. - Install at start into a volume:
zkg installin an entrypoint wrapper. No new tooling, but what runs is whatever the package index served that day, which is the unpinned state the pinning rules exist to prevent. - A published image that already includes it. Use one only if a maintained, digest-pinnable one exists.
Option 1 looks right, but it changes repository policy, so it should be an ADR rather than a line in a compose file.
Done when
ssl.log(orja4*.log) onfenrircarries a JA4 for a guest's outbound TLS.{job="zeek"} | json | ja4 != ""returns lines in the lab's Loki.local.zeek's "JA4 fingerprints are not in the stock image" comment is removed, ADR-0068 gets a NOTE, and the changelog records the result.
Follows #437.
Activity
docs(adr): ADR-0069's JA4+ list, method set, loss fields and input cl…
docs(adr): ADR-0069, vendor the JA4 scripts into the sensor stack rat…
Gerrrt commentedon Oct 1, 2026
Reopened. It closed when #780 merged, but #780 only recorded the decision (ADR-0069) and said
Refs #776. The implementation hasn't landed: the vendoredja4-zeek-scriptswith LICENSE, LICENSE-JA4 and NOTICE, the read-only mount, the@load,scripts/vendor-ja4.sh, and the before/after loss measurement. The baseline is now about 0% after #782's GRO fix, rather than the 8.1% ADR-0069 first recorded.fix(sensor): vendor-ja4.sh requires the NOTICE; field list and #776's…
feat(sensor): vendor FoxIO's JA4+ scripts into stacks/sensor (#776) (#…
feat(ci): fail a PR whose close keywords sit in prose (#672) (#786)
docs(sensor): ADR-0069's baseline starts after the GRO fix, and the c…
docs(sensor): JA4+ is live on fenrir, and #776 closes (#776) (#926)