diff --git a/.gitignore b/.gitignore index f263abb..699fae5 100644 --- a/.gitignore +++ b/.gitignore @@ -22,3 +22,8 @@ export/ # NB: config/config.json IS committed on purpose — it's the v0.16 bootstrap # config and is secret-free (DB password comes from $STALWART_DB_PASSWORD via # the EnvironmentVariable secret type). Don't add it here. + +# Editor swap / backup files +*.swp +*.swo +*~ diff --git a/LESSONS.md b/LESSONS.md new file mode 100644 index 0000000..b283f3d --- /dev/null +++ b/LESSONS.md @@ -0,0 +1,96 @@ +# tailwart — lessons learned + +Hard-won notes from bringing the mail edge up. Each entry is **symptom → cause → +fix**, ordered roughly by how long it cost. Read this before re-debugging. + +## 1. Postgres startup race ate cert/setting writes + +**Symptom:** TLS certs (manual import *and* ACME) would validate but never +persist — Stalwart kept serving its `rcgen` self-signed fallback. Logs showed +`Failed to create tables: error connecting to server` on most boots. + +**Cause:** Stalwart shares the `ts-stalwart` sidecar's netns. Its `depends_on` +only waited for the sidecar's *own* health (`/healthz` = "tailscaled up"), which +flips green **before** the tailnet route to Postgres (`the-record-prod:5432`) is +usable. Stalwart started into that gap, failed the DB connect, and any write in +that window — including a freshly obtained cert — was silently lost. + +**Fix:** the sidecar healthcheck now also requires Postgres to be reachable +(`nc -z … 5432`), so `depends_on: service_healthy` can't release Stalwart into +the race. See `docker-compose.yml`. First clean boot after this: zero PG errors, +4 live connections immediately. + +## 2. DNS-01 was blocked by a dead Spaceship API key + +**Symptom:** `Failed to set DNS RRSet: Unauthorized` on every record; no cert +issued; no `_acme-challenge` TXT ever set. + +**Cause:** the cert design is ACME **DNS-01** via the **Spaceship** provider +(bundled in caddy/lego). The stored API key was invalid (recovery debris from an +earlier config attempt). Note `STALWART_ACME_PROVIDER` / `STALWART_ACME_TOKEN` +in `.env` are **empty and not even passed through by compose** — the provider + +secret are entered in the **admin UI** (stored in the DB), not via env. + +**Gotcha:** secret fields render **blank** in the Stalwart admin even when set +(the S3 secret behaves identically). A blank field is *not* evidence it's unset. + +**Fix / how to verify a key directly (egresses the box's WAN IP, same as +Stalwart):** +```bash +curl -i 'https://spaceship.dev/api/v1/dns/records/?take=5&skip=0' \ + -H 'X-Api-Key: KEY' -H 'X-Api-Secret: SECRET' +# 401 application.unauthorized = bad key/secret or IP-restricted +# 200 = good +``` +A fresh Spaceship key fixed it. + +## 3. Stalwart's auto-ban vs PROXY protocol (the "8080 mystery") + +**Symptom:** the edge box could relay mail fine but could **not** reach +Stalwart's `:8080` admin — connections accept then immediately close. Looked like +"tagged devices rejected, user phone works." + +**Cause:** Stalwart's fail2ban checks the **proxied client IP** (from the PROXY +header) on the mail listeners, but the **raw connection IP** on the non-proxied +admin listener. A banned edge-box IP therefore still relays mail (ban checked +against the header IP) while direct `→:8080` is dropped (checked against the box +IP). Malformed probing of the mail ports **re-arms** the ban. + +**Fix:** add `100.64.0.0/10` (and the box's WAN IP, which appears as the proxied +client when you hit the box's own public hostname) to the fail2ban allow-list. +Bans are in-memory — a Stalwart restart flushes them. **Don't rapid-poll the mail +ports** to test. + +## 4. The wildcard request *required* DNS-01 (why HTTP-01 was a dead end) + +With "Additional Hostnames" left empty, Stalwart requests a **wildcard** +(`*.`). Wildcards can **only** be issued via DNS-01 — HTTP-01 literally +cannot satisfy them. We burned time on an HTTP-01 + Caddy-challenge-forwarding +detour before realizing DNS-01 was the intended (and only viable) path. One +wildcard cert then covers `mail`, `mta-sts`, `autoconfig`, `autodiscover`, etc. + +## 5. `:443` web endpoints need SNI pass-through, not L7 proxy + +MTA-STS / autoconfig / autodiscover serve over **:443**. You cannot L7 +`reverse_proxy` them through Caddy, because the **CAA** record pins issuance to +Stalwart's ACME account — Caddy can't get its own cert for those names. Stalwart +holds the wildcard, so the edge **passes TLS through** by SNI. See +`caddy/README.md` → "The HTTP side". Needed `tcp:443` added to the +`reverse-proxy → stalwart` ACL grant. + +## 6. The sidecar is ephemeral — never hardcode its tailnet IP + +`ts-stalwart` runs with `?ephemeral=true`, so its tailnet IP **changes on +re-registration** (an ACL re-sync did this mid-debug: `100.112.26.122 → +100.79.87.80`). Everything must use the MagicDNS name +`stalwart.tail7b1641.ts.net`. A hardcoded IP will mysteriously go +`Network is unreachable`. + +## 7. Don't trust crt.sh for rate-limit checks + +crt.sh was flaky/empty all session. To gauge Let's Encrypt's weekly +duplicate-cert limit, use **certspotter** instead: +`https://api.certspotter.com/v1/issuances?domain=&include_subdomains=true`. +Also: LE limits are dimensioned — **failed validations** are hourly (5/hr/host, +the one a retry storm trips), **issued duplicates** are weekly (5/wk). A renewal +task hammering every 10 min trips the hourly one; consolidate to a single task. diff --git a/acl-snippet.hujson b/acl-snippet.hujson index 31ad71f..32b60e9 100644 --- a/acl-snippet.hujson +++ b/acl-snippet.hujson @@ -12,10 +12,13 @@ // 3) grants — the edge proxy (tag:reverse-proxy) reaches the mailbox ports. // 8080 is the JMAP/admin HTTP tier (fronted by the main L7 Caddy). +// 443 is Stalwart's HTTPS web listener; the edge L4-proxies the public +// mta-sts/autoconfig/autodiscover SNIs to it (Stalwart terminates TLS with +// its wildcard cert). PROXY protocol v2, same as the mail ports. { "src": ["tag:reverse-proxy"], "dst": ["tag:stalwart"], - "ip": ["tcp:25", "tcp:465", "tcp:587", "tcp:143", "tcp:993", "tcp:8080"], + "ip": ["tcp:25", "tcp:465", "tcp:587", "tcp:143", "tcp:993", "tcp:443", "tcp:8080"], }, // 4) admin console (not this file): assign tag:stalwart to the same OAuth diff --git a/caddy/README.md b/caddy/README.md index 749e5a8..e8c4221 100644 --- a/caddy/README.md +++ b/caddy/README.md @@ -34,17 +34,43 @@ binary. To add plugins, append `&p=` to can't read `.env`; this is the one spot the MagicDNS name is hardcoded — same trade-off as pgAdmin's `servers.json`.) -## The HTTP side (JMAP / autoconfig / admin) is separate +## The HTTP side (MTA-STS / autoconfig / autodiscover) — `:443` SNI fan-out -That part *is* ordinary layer 7. Don't put it here if this box already runs the -main Caddy on :443 — you'll collide. Instead add a vhost to the existing Caddy: +Stalwart publishes DNS that points public HTTPS names at this edge: +`mta-sts.`, `autoconfig.`, `autodiscover.`. They serve the MTA-STS +policy and mail-client autoconfig over **:443** — so the edge has to handle +`:443` too, which is where a naive setup collides with a box that already runs a +web Caddy. + +The fix is **not** an L7 `reverse_proxy` (terminate at Caddy). You can't: the +domain's **CAA** record pins issuance to Stalwart's ACME account +(`accounturi=…`), so Caddy can't obtain its own cert for `*.`. Stalwart +already holds the wildcard. So we **pass TLS through** to it. + +The `web` server in `caddy.json` owns `:443` and fans out by SNI: + +- `mta-sts` / `autoconfig` / `autodiscover.` → `stalwart:443` + (pass-through; Stalwart terminates with its wildcard cert — **no** proxy + protocol on `:443`, unlike the mail ports). +- every other SNI → `127.0.0.1:8443`, the box's own web Caddy. + +For that fallback to exist, move the web Caddy's HTTPS off `:443`: ```caddyfile -mail.infinidim.net { - reverse_proxy stalwart.tail7b1641.ts.net:8080 +{ + https_port 8443 # web vhosts now listen here; the L4 :443 forwards to them } + +your-web-site.example { reverse_proxy … } ``` +HTTP→HTTPS redirects still resolve to `:443` correctly. A **mail-only** edge (no +web vhosts on the box) omits the `web` server entirely — keep just the mail +ports above. + +> Note: `tag:reverse-proxy → tag:stalwart` must also grant **`tcp:443`** in the +> Tailscale ACL (see `../acl-snippet.hujson`), on top of the mail ports. + ## Prerequisites on the host running this - Joined to the tailnet, tagged `tag:reverse-proxy` (so the ACL lets it reach diff --git a/caddy/caddy.json b/caddy/caddy.json index dec141b..52a5930 100644 --- a/caddy/caddy.json +++ b/caddy/caddy.json @@ -45,6 +45,30 @@ "proxy_protocol": "v2", "upstreams": [{ "dial": ["stalwart.tail7b1641.ts.net:993"] }] }]}] + }, + "web": { + "//": "SNI fan-out on the public :443. Stalwart's HTTPS web endpoints", + "//2": "(MTA-STS policy, autoconfig, autodiscover) pass through to Stalwart,", + "//3": "which terminates TLS with its wildcard cert — NO proxy_protocol here,", + "//4": "unlike the mail ports above. Every other SNI falls to the box's own", + "//5": "web Caddy on :8443 (set `https_port 8443` there). A mail-only standalone", + "//6": "edge omits this server. See README — 'The HTTP side'.", + "listen": [":443"], + "routes": [ + { + "match": [{ "tls": { "sni": ["mta-sts.infinidim.net", "autoconfig.infinidim.net", "autodiscover.infinidim.net"] } }], + "handle": [{ + "handler": "proxy", + "upstreams": [{ "dial": ["stalwart.tail7b1641.ts.net:443"] }] + }] + }, + { + "handle": [{ + "handler": "proxy", + "upstreams": [{ "dial": ["127.0.0.1:8443"] }] + }] + } + ] } } } diff --git a/docker-compose.yml b/docker-compose.yml index 9223627..e5c9ec2 100644 --- a/docker-compose.yml +++ b/docker-compose.yml @@ -31,9 +31,14 @@ services: - NET_ADMIN - NET_RAW healthcheck: - test: ["CMD", "wget", "-qO-", "http://127.0.0.1:9002/healthz"] + # Healthy only when BOTH the tailnet link is up AND Postgres is reachable + # over it. The stalwart service gates on this (depends_on: service_healthy), + # so it can no longer start into the race where it tries the DB before the + # tailnet route exists — which logged "Failed to create tables" and dropped + # in-flight cert/setting writes (e.g. lost the ACME cert on 2026-06-10). + test: ["CMD-SHELL", "wget -qO- http://127.0.0.1:9002/healthz && nc -z -w3 ${DB_MAGIC_NAME}.${TS_TAILNET} 5432"] interval: 10s - timeout: 5s + timeout: 8s retries: 6 start_period: 30s restart: unless-stopped