Tiny /bin/sh entrypoint shim that, if invoked as root, reads PUID/PGID
env vars (defaulting to 99/100 — Unraid's nobody:users), remaps the
in-image pad user, chowns /data, and execs the binary via su-exec.
If invoked as non-root (caller passed --user), it just execs directly
— caller knows what they want.
Solves the classic Unraid appdata-ownership-mismatch first-run failure
where the in-image pad user (uid 1000) couldn't write to a host volume
owned by nobody:users (uid 99, gid 100). Reusable on Synology / QNAP /
TrueNAS where the host's appdata user is similarly non-1000.
Behavior changes:
- Container starts as root (USER directive removed). Entrypoint drops
privileges via su-exec before exec'ing pad — standard PUID/PGID
pattern. Healthcheck adapts: root → su-exec to pad; non-root →
direct wget.
- chown -R is always-run (warn-and-continue on per-file failures). A
shallow stat-only check would silently break pad on a restored
backup with mixed-ownership inner files.
- Healthcheck start-period bumped 10s → 60s to absorb slow chown -R
on large attachment stores.
- Compose default 1000/1000 for backward compat with existing deploys
whose volumes were created under the previous USER pad image.
- Raw `docker run` defaults to 99/100 (Unraid convention).
Validation rejects PUID=0 / PGID=0 (would defeat the unprivileged-user
invariant), empty values, and non-numeric values with clear errors.
Goes through 11 rounds of codex pre-implementation design review,
catching:
- gid bug where groupmod alone leaves /etc/passwd's primary-gid stale
- compose $-interpolation gotcha (needs $$( ) not $())
- getent missing from default alpine BusyBox
- shell ${VAR:-} silently masking explicit empty values
- healthcheck running as root after USER drop
- su-exec failing for --user non-root pass-through
Part of PLAN-1166 (Pad on Unraid — Community Apps launch). Unblocks
TASK-1169 (XML template authoring).
xarmian