Skip to content

CL-0030: cap_add: SYSLOG — host kernel log via dmesg

Severity: HIGH (how it is derived)

The service adds SYSLOG, which lets it read the host's kernel log: boot messages, hardware and driver state, and on some hosts kernel addresses. Remove it and read the kernel log on the host.

Fix

Remove the capability. The kernel log belongs to the host, and so does reading it:

# Before
services:
  monitor:
    image: myapp:1.0
    cap_add:
      - SYSLOG

# After — collect kernel messages on the host, not from inside a container
services:
  monitor:
    image: myapp:1.0
  1. Read the buffer on the host instead:
journalctl -k -n 50        # the last 50 kernel messages

or run the log shipper on the host. A workload that needs kernel messages about its own device should be given that device explicitly — see CL-0016 — rather than the whole machine's log.

  1. Check the container can no longer read it:
docker compose exec monitor dmesg
# dmesg: klogctl: Operation not permitted

With SYSLOG the same command printed the host's whole kernel log (1,937 lines on the test host). Measured on Docker 29.7.2 with busybox:1.37; the line-count comparison is re-proven on every CI run (_cl0030_syslog), the klogctl wording is captured, not CI-asserted.

  1. Check the finding is gone:
compose-lint -q compose.yaml | grep CL-0030    # prints nothing once fixed

Tested on: Docker 29.7.2 with Compose 5.5.0, on a host without AppArmor (Yama ptrace_scope 1); the Fix steps above were re-run there in October 2026. Not tested: an AppArmor or SELinux host.

When to suppress

A host-level monitoring agent deliberately deployed as a container is the real case: it is doing the host's job, and reading the kernel log is the job. Such an agent usually announces itself with pid: host or a host bind mount as well. Suppress per service with a reason: naming the agent, rather than disabling the rule globally — the next service to ask for SYSLOG is unlikely to be a monitoring agent.

What it detects

cap_add entries naming a capability that reads host state:

Capability Grants Why it is not CRITICAL
SYSLOG syslog(2) against the kernel ring buffer, and the ability to read it regardless of kernel.dmesg_restrict Pure observation. It writes nothing, runs nothing, and breaks nothing — the loss is confidentiality alone

CAP_-prefixed and lowercase spellings are equivalent and are matched the same way: Docker treats CAP_SYSLOG and SYSLOG as the same capability.

Why it matters

The kernel ring buffer is one log for the whole machine. A container holding SYSLOG reads the host's boot messages, its hardware and driver state, its filesystem and network events — and on a host that leaves kernel.kptr_restrict at 0, kernel pointers along with them, which is the address-layout information a local exploit wants.

It reads what the host itself hides. On a host with kernel.dmesg_restrict=1 — the hardened default on most distributions — an unprivileged user on the host cannot read that buffer. A container granted this one capability can. The container is not a smaller trust domain than the host user in that comparison; it is a larger one.

The grant does not depend on the host being hardened. Docker's default seccomp profile admits syslog(2) only when CAP_SYSLOG or CAP_SYS_ADMIN is held, and the kernel then requires CAP_SYSLOG itself (with SYS_ADMIN alone dmesg still fails with klogctl: Operation not permitted). So a capless container reads nothing even where kernel.dmesg_restrict=0 would let any host user read everything. That was measured, not assumed, and it is why this rule does not carry a "depends on host configuration" caveat.

ATT&CK coverage

Remediating this finding contributes to mitigating the following MITRE ATT&CK techniques (pinned to ATT&CK v18). compose-lint is a static analyser, so this is mitigation coverage — it detects nothing at runtime.

Technique Tactic
T1613 Container and Resource Discovery Discovery

See also

  • CL-0028 — the same Direct × Host cell, spending its qualifier on integrity-only
  • CL-0029 — the same cell again, spending it on availability-only
  • CL-0006 — dropping capabilities wholesale, the control this rule's findings evade
  • Hardening guide, step 5 — where this rule fits when hardening a whole file

Severity derivation

Derivation (see severity model):

  • Baseline: A — the attacker already has code execution in this container, as the workload uid
  • Precondition: Direct — syslog(2) alone returns the buffer. No published technique, no second defect, and no other key in the file
  • Impact: Host — the kernel ring buffer is a single host-wide log; it is not namespaced, so what the container reads is the host's
  • Qualifier/modifier: read-only — the realised loss is disclosure. The capability grants no write to the buffer and no execution anywhere, so the tier drops one from CRITICAL
  • Derived: Direct × Host + read-only = HIGH
  • Shipped: HIGH
  • Scoping assumptions: none. The reach needs nothing else in the file, and — measured — does not depend on the host's kernel.dmesg_restrict either (see Evidence)
  • Evidence: _cl0030_syslog in scripts/validate_rule_premises.py asserts the grant against a live daemon: dmesg returns no lines at --cap-drop ALL and returns the host's buffer with --cap-add SYSLOG. The check compares line counts, never content — the host's kernel log is not something a CI log should carry. Measured on Docker 29.4.3: 0 lines capless against 1,967 with the capability, on a host where kernel.dmesg_restrict=1 blocks an unprivileged host user from reading the same buffer. The independence from that sysctl was measured rather than assumed: with kernel.dmesg_restrict=0 a capless container still read 0 lines, because Docker's own default seccomp profile admits the syslog(2) syscall only for CAP_SYSLOG. The reach is therefore gated by the capability on any host, hardened or not

References: - OWASP Docker Security Rule #3 - CIS Docker Benchmark 5.4 — Ensure that Linux kernel capabilities are restricted within containers

A reference means the rule relates to that source. It does not mean a file with no finding complies with it. The CIS, OWASP and ATT&CK mapping for every rule, and the tags SARIF output carries for it, are on Standards mapping.