fix(openclaw): use named volume for home dir to fix crash loop

[yasinBursali]

What

Replace the OpenClaw bind mount ./data/openclaw/home:/home/node/.openclaw with a named Docker volume (openclaw-home).

Why

Docker creates bind mount directories as root:root on the host. The container's node user cannot write to /home/node/.openclaw/openclaw.json, so inject-token.js Part 1 fails silently. The gateway then starts without allowedOrigins, and since --bind lan requires explicit origins (OpenClaw 2026.3.8+), the gateway crash-loops.

How

Named volumes are initialized by Docker with the image's directory ownership (node:node), so inject-token.js can write the patched config with allowedOrigins and dangerouslyAllowHostHeaderOriginFallback. The workspace bind mount (./config/openclaw/workspace) is preserved as an overlay on the named volume.

Testing

• docker compose config validates cleanly
• Manual: fresh install, verify OpenClaw starts without crash loop
• Manual: verify Control UI loads at http://localhost:7860
• Verify: docker exec dream-openclaw ls -la /home/node/.openclaw/ shows node ownership

Review

Critique Guardian: APPROVED WITH WARNINGS

• Installer phase 06 still writes to orphaned host path (tracked: Installer phase 06 writes OpenClaw config to orphaned host path after named volume migration yasinBursali/DreamServer#258)
• Upgrade path: existing session data in ./data/openclaw/home/ becomes orphaned (sessions are ephemeral)

Platform Impact

• macOS: Named volumes work identically on Docker Desktop
• Linux: Named volumes work identically on Docker Engine
• Windows/WSL2: Named volumes work identically on Docker Desktop

[yasinBursali]

Update: Fix for gateway.controlUi in merged config

Testing confirmed the named volume alone doesn't fix the crash — the volume is still root-owned, and Part 3's merged config (which the gateway actually reads) was missing the gateway.controlUi section entirely.

New commit adds gateway patching to Part 3 of inject-token.js:

• allowedOrigins with http://localhost:<port> and http://127.0.0.1:<port>
• dangerouslyAllowHostHeaderOriginFallback: true
• allowInsecureAuth: true
• dangerouslyDisableDeviceAuth: true

This ensures the merged config at /tmp/openclaw-config.json always has the correct gateway settings, regardless of whether Part 1's write to ~/.openclaw/openclaw.json succeeds.

Also passes OPENCLAW_EXTERNAL_PORT to the container so inject-token.js can build correct origin URLs.

[yasinBursali]

Update: Fix for root-owned volume (Update 3 findings)

The gateway reads ~/.openclaw/openclaw.json at startup BEFORE loading $OPENCLAW_CONFIG. Even though Part 3 now correctly patches the merged config, the gateway applies its default security policy first from the home-dir config — which doesn't exist because the volume is root-owned.

New commit: Run entrypoint as user: "0:0" (root), chown -R 1000:1000 /home/node/.openclaw before running inject-token.js, then exec into docker-entrypoint.sh for the gateway. The chown is a no-op on subsequent starts once ownership is set.

This should resolve both the permission denied and the allowedOrigins issues together.