0dad38e02e
Captures live config state of nullstone Purpur 1.21.11 server: - docker-compose.yml (itzg/minecraft-server image, MODRINTH_PROJECTS + PLUGINS lists) - All plugin configs under live-server/plugins/ (no DBs, no jars, no world data) - Server core: bukkit.yml, spigot.yml, purpur.yml, paper-global.yml, paper-world-defaults.yml, server.properties Excluded via .gitignore: - World data (world/, world_nether/, world_the_end/, auth_limbo/) - Sensitive: AuthMe DB (password hashes), Lands DB, CoreProtect DB, Essentials userdata - Jars (auto-fetched), logs, caches, .paper-remapped
51 lines
2.1 KiB
Markdown
51 lines
2.1 KiB
Markdown
# Permissions / userns-remap quirk
|
|
|
|
## Symptom
|
|
|
|
Container restart-loops on first boot. Logs show:
|
|
|
|
```
|
|
[init] Changing ownership of /data to 1000 ...
|
|
chown: changing ownership of '/data/server.properties': Operation not permitted
|
|
[init] Running as uid=1000 gid=1000 with /data as 'drwxrwxr-x 4 65534 65534 4096 /data'
|
|
/image/scripts/start-configuration: line 87: /data/.rcon-cli.env: Permission denied
|
|
```
|
|
|
|
Files appear inside container as `65534:65534` (nobody:nogroup) even though host shows them as `1000:1000`.
|
|
|
|
## Cause
|
|
|
|
Docker daemon on this host runs with **userns-remap** enabled (`/etc/docker/daemon.json` has `"userns-remap": "default"`). This maps container UID 1000 → some host UID like 100000+1000=`101000`. Files owned by host UID 1000 (`user:user`) are *not* owned by the remapped container UID, so:
|
|
|
|
- They show up as `nobody:nogroup` (65534) inside the container.
|
|
- Container can read world-readable files but can't write unless dir is `o+w`.
|
|
- Init script can't `chown` (host kernel blocks it — bind mount, foreign UID).
|
|
|
|
## Fix
|
|
|
|
```bash
|
|
chmod -R 777 /opt/docker/minecraft
|
|
```
|
|
|
|
This is what the original deployment used (verified: old dir was `drwxrwxrwx 12 user user`). Container can now write logs, world data, plugin configs, RCON env file.
|
|
|
|
## Why not disable userns-remap?
|
|
|
|
It's a host-wide hardening setting, kept for the rest of the Docker stack on nullstone. Per-container userns override (`userns_mode: "host"` in compose) is possible but defeats the security benefit for this container.
|
|
|
|
## Alternative: named volume
|
|
|
|
```yaml
|
|
volumes:
|
|
- mc-data:/data
|
|
volumes:
|
|
mc-data:
|
|
```
|
|
|
|
Docker creates the volume owned by the remapped UID directly, no chmod needed. Trade-off: harder to inspect/edit configs from host (must `docker cp` or bind-mount inspect).
|
|
|
|
This repo sticks with bind mount + `chmod 777` for operator ergonomics.
|
|
|
|
## Cosmetic chown spam (non-blocking)
|
|
|
|
Even with `chmod 777`, init still logs `chown: ... Operation not permitted` for every file. Server starts and runs fine — kernel just won't let init re-claim ownership across the userns boundary. Ignore.
|