Unusual Process Connection to Docker or Containerd Socket
Description
Detects a process connecting to a container runtime Unix socket (containerd or Docker) that is not a known legitimate runtime component. Direct access to the container runtime socket allows an attacker to create, exec into, or manipulate containers without going through the Kubernetes API server, bypassing RBAC, admission webhooks, pod security standards, and Kubernetes audit logging entirely.
Query · kuery
host.os.type:"linux" and
event.category:"network" and
event.action:"connected-to" and network.direction:"egress" and
destination.address:("/run/containerd/containerd.sock" or "/var/run/containerd/containerd.sock" or "/var/run/docker.sock" or "/run/docker.sock") and
process.executable:(* and not
("/usr/bin/kubelet" or
"/usr/local/bin/kubelet" or
"/usr/bin/containerd" or
"/usr/sbin/containerd" or
"/usr/bin/containerd-shim" or
"/usr/bin/containerd-shim-runc-v2" or
"/usr/local/bin/containerd-shim-runc-v2" or
"/usr/bin/dockerd" or
"/usr/sbin/dockerd" or
/var/lib/*/usr/bin/dockerd or
"/usr/bin/docker-proxy")
)
Implementation guide
This rule requires Auditd Manager (or Auditbeat) process and network events where Unix socket paths populate
destination.address (or equivalent ECS mapping from your pipeline).
Auditd Manager: network and socket visibility
Enable auditing of socket-related activity so event.category:network and event.action:connected-to (or your
pipeline’s equivalent) are emitted for connect to Unix sockets. Example audit rules to extend as needed:
# 64-bit connect (required for socket connection telemetry)
-a always,exit -F arch=b64 -S connect -k netconn
# 32-bit (if applicable)
-a always,exit -F arch=b32 -S connect -k netconn
After deployment, confirm in Discover that events for connections to
/var/run/docker.sock, /run/docker.sock, or containerd socket paths include process.executable and
destination.address fields used by this rule.
For more details on the integration refer to the Auditd Manager documentation.
Known false positives
- Custom container tooling, CI agents, or monitoring may connect to docker.sock or containerd.sock from non-standard paths after relocation or bind mounts. Tune by process.executable or user.name when noise is high.
Analyst notes
Investigating Unusual Process Connection to Docker or Containerd Socket
Review the initiating process executable, user, and parent chain. Confirm whether the socket path is the host default or a bind-mounted path inside a container. Pivot on the same host for subsequent container creation, image pulls, or credential access.
Possible investigation steps
- Map
process.executable,process.args,process.titleanduser.idto an identity and session (SSH, cron, web shell). - Check file permissions on the socket path and whether the workload should have access at all.
- Correlate with process and authentication telemetry before and after the connection.
False positive analysis
- Vendor agents that wrap docker or containerd CLIs from non-standard install locations may match; add explicit exclusions for known binaries.
Response and remediation
- If malicious, isolate the host, revoke credentials, inspect for rogue containers and persistence, and restrict socket permissions to trusted groups only.