Standard image incompatible with rootless Docker and security-hardened containers

[tecio-first]

Summary

The standard triliumnext/trilium image's entrypoint (start-docker.sh) requires root privileges that are unavailable in rootless Docker environments with security hardening (read_only, cap_drop: ALL, no-new-privileges). The referenced docker-compose.rootless.yml uses triliumnext/notes:rootless, but this tag does not exist on Docker Hub for either triliumnext/notes or triliumnext/trilium.

Environment

• Trilium version: v0.101.3
• Docker: 29.2.1 (rootless mode)
• Docker Compose: v5.0.2
• OS: Ubuntu 24.04.4 LTS (kernel 6.8.0-100-generic)

Problem

The start-docker.sh entrypoint runs:

chown -R node:node /home/node
exec su -c "node ./main.cjs" node

This requires:

CAP_CHOWN, CAP_DAC_OVERRIDE for chown
CAP_SETUID, CAP_SETGID for su
A writable filesystem at /home/node
no-new-privileges must be disabled (required for su)
In a security-hardened setup (read_only: true, cap_drop: ALL, security_opt: no-new-privileges), the container fails with:

chown: /home/node/trilium-data: Operation not permitted
chown: /home/node: Read-only file system
su: can't set groups: Operation not permitted

Workaround

We bypass the entrypoint entirely by setting user: "1000:1000"
and running node ./main.cjs directly. A one-shot init container
fixes volume ownership. This preserves full security hardening:

  trilium-init:
    image: alpine:${ALPINE_TAG:-3.23.3}
    container_name: trilium-init
    command: ["chown", "1000:1000", "/data"]
    volumes:
      - trilium-data:/data
    restart: "no"
    read_only: true
    mem_limit: 32M
    memswap_limit: 32M
    cpus: 0.25
    pids_limit: 8
    cap_drop:
      - ALL
    cap_add:
      - CHOWN
    security_opt:
      - no-new-privileges

  trilium:
    image: triliumnext/trilium:${TRILIUM_TAG:-v0.101.3}
    container_name: trilium
    user: "1000:1000"
    command: ["node", "./main.cjs"]
    restart: unless-stopped
    read_only: true
    mem_limit: 512M
    memswap_limit: 512M
    cpus: 1.0
    pids_limit: 128
    cap_drop:
      - ALL
    security_opt:
      - no-new-privileges
    depends_on:
      trilium-init:
        condition: service_completed_successfully
      traefik:
        condition: service_healthy
    environment:
      - TZ=${TZ:-UTC}
      - TRILIUM_DATA_DIR=/home/node/trilium-data
      - TRILIUM_TRUSTED_REVERSE_PROXY=true
    volumes:
      - trilium-data:/home/node/trilium-data
    tmpfs:
      - /tmp:noexec,nosuid,nodev,size=64M
    healthcheck:
      test: ["CMD", "nc", "-z", "127.0.0.1", "8080"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 15s
    labels:
      - "traefik.enable=true"
      - "traefik.http.routers.trilium.rule=Host(`${TRILIUM_SUBDOMAIN:-hz-notes}.${BASE_DOMAIN:-hz.tecio.xyz}`)"
      - "traefik.http.routers.trilium.entrypoints=websecure"
      - "traefik.http.routers.trilium.tls.certresolver=myresolver"
      - "traefik.http.services.trilium.loadbalancer.server.port=8080"
      - "traefik.http.routers.trilium.middlewares=crowdsec-bouncer@file,authentik@file,security-headers@file,rate-limit@file"
    networks:
      - traefik-network

Suggestions

Publish the rootless image: The docker-compose.rootless.yml in this repo references triliumnext/notes:rootless, but this tag does not exist on Docker Hub. Publishing it would solve this cleanly.

Use gosu instead of su: gosu is compatible with no-new-privileges and doesn't require CAP_SETUID/CAP_SETGID in the same way. Many official Docker images (postgres, redis) use this pattern.

Skip chown when already correct: The entrypoint could check ownership before running chown, avoiding the need for CAP_CHOWN/CAP_DAC_OVERRIDE when the volume is already owned by the right user.