stalwart: answer the setup wizard instead of handing it to the user

A new Stalwart drops you into a five-screen wizard — hostname, domain,
storage backend, directory, logging, DNS — before it will do anything.
LibrePortal already knows the two answers that matter and the rest have
sane defaults, so asking is asking a question we can answer ourselves.

v0.16 exposes those wizard fields as a `Bootstrap` singleton, so the whole
thing is one `update` applied through the Stalwart CLI. The CLI is not in
the server image (upstream split it into its own repo), but it publishes a
multi-arch container, so we borrow the server's network namespace and run
it there — nothing installed on the host, nothing to clean up, arm64 works.

Setup now also:

- generates DKIM keys (Ed25519 + RSA) with rotation left switched on, and
  requests a TLS certificate. That last one is easy to miss: Traefik only
  fronts the admin port, so 25/465/587/993 never see its certificate and
  clients would hit a self-signed one on 993.
- creates postmaster@<domain>. The generated zone points DMARC and TLS-RPT
  reports there and nothing was creating it, so those reports bounced.
- prints the record set read back from the server rather than composed
  here, so it includes the real DKIM public keys, MTA-STS, TLS-RPT and the
  SRV records clients autoconfigure from. This hook used to tell the user
  to go and fetch DKIM themselves; by that point the keys exist.

Optionally hands DNS to a provider API (Cloudflare/DigitalOcean/DeSEC),
which keeps the whole record set in sync and makes DKIM rotation safe to
leave on. Off by default: the token can write to your zone and lives in
the mail server's database.

Re-running is safe — provisioning is skipped once config.json exists, and
the plans use upsert so they reconcile rather than duplicate.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
librelad 2026-08-18 05:05:03 +01:00
parent 0eda3104c7
commit 7ed1cddd5c
3 changed files with 275 additions and 22 deletions

View File

@ -17,6 +17,175 @@ stalwart_install_message_data()
printf '%s %s' "${CFG_STALWART_ADMIN_USER:-admin}" "${CFG_STALWART_ADMIN_PASSWORD:-}"
}
# Run the Stalwart CLI against our own container.
#
# The CLI is deliberately NOT in the server image — upstream split it into its
# own repository so it ships and versions separately. It does publish a
# multi-arch container, which suits us better than a host binary: nothing to
# install, nothing to clean up on uninstall, and arm64 works for Pi installs.
#
# `--network container:stalwart-service` borrows the server's network namespace,
# so the CLI reaches it on localhost:8080 without us having to resolve the
# LibrePortal network name or expose the admin port to get at it.
stalwart_cli()
{
runFileOp docker run --rm -i --network "container:stalwart-service" \
-e STALWART_URL="http://localhost:8080" \
-e STALWART_USER="${CFG_STALWART_ADMIN_USER:-admin}" \
-e STALWART_PASSWORD="${CFG_STALWART_ADMIN_PASSWORD:-}" \
"ghcr.io/stalwartlabs/cli:${CFG_STALWART_CLI_VERSION:-1.0.12}" --no-color "$@"
}
# Wait for the admin HTTP listener. Used twice: once for the bootstrap listener
# before we configure anything, once for the real one after the restart.
stalwart_wait_http()
{
local probe="$1" tries="${2:-40}" i code
for ((i = 0; i < tries; i++)); do
code=$(runFileOp docker exec stalwart-service curl -fsS -o /dev/null -w '%{http_code}' \
--max-time 3 "http://localhost:8080/healthz/$probe" 2>/dev/null | tr -d '\r')
[[ "$code" == "200" ]] && return 0
sleep 2
done
return 1
}
# Fill in the first-run setup wizard instead of making the user do it.
#
# Out of the box Stalwart boots into "bootstrap mode" and waits for a human to
# answer five screens of questions in the WebUI: hostname, domain, which storage
# backend, which directory, where to log, how to handle DNS. LibrePortal already
# knows the answers to the ones that matter and the rest have sane defaults, so
# making the user answer them is asking a question we can answer ourselves.
#
# `Bootstrap` is a singleton object holding exactly those wizard fields, so the
# whole wizard is one `update`. Once it is set the server writes its config.json,
# provisions the domain, generates DKIM keys and leaves bootstrap mode — and the
# user's first sight of Stalwart is a configured mail server, not a form.
stalwart_install_provision()
{
# config.json only exists once setup has completed, which makes it the
# honest "is this already configured?" test. Re-running the installer over a
# working server must not re-answer its setup questions.
if runFileOp docker exec stalwart-service test -f /etc/stalwart/config.json 2>/dev/null; then
isNotice "Stalwart is already configured — leaving its existing settings alone."
return 0
fi
local mail_host="${host_setup:-}"
local mail_domain="${mail_host#*.}"
if [[ -z "$mail_host" || "$mail_host" != *.* ]]; then
isError "No mail hostname is configured, so Stalwart cannot be set up automatically."
isNotice " Set a domain in the general config, then open the admin console to"
isNotice " complete setup by hand."
return 1
fi
isNotice "Waiting for Stalwart's setup listener…"
if ! stalwart_wait_http live; then
isError "Stalwart did not open its setup listener — skipping automatic setup."
return 1
fi
# requestTlsCertificate matters more than it looks. Traefik only fronts the
# admin port; SMTP and IMAP (25/465/587/993) bypass it entirely, so Traefik's
# certificate never reaches a mail client. Without this Stalwart serves a
# self-signed cert on 993 and every mail client throws a warning.
#
# generateDkimKeys makes the server own DKIM: it creates an Ed25519 and an
# RSA key, publishes both, and rotates them on a schedule. Hand-managed DKIM
# keys are, in practice, keys that never get rotated.
isNotice "Configuring Stalwart for ${mail_domain}"
local plan
plan=$(printf '{"@type":"update","object":"Bootstrap","value":{"serverHostname":"%s","defaultDomain":"%s","generateDkimKeys":true,"requestTlsCertificate":true}}' \
"$mail_host" "$mail_domain")
if ! printf '%s\n' "$plan" | stalwart_cli apply --stdin >/dev/null 2>&1; then
isError "Automatic setup failed — open the admin console to finish it by hand."
return 1
fi
# Leaving bootstrap mode needs a restart: the server swaps its temporary
# setup listener for the real ones (SMTP, IMAP, submission) on the way back up.
isNotice "Restarting Stalwart to bring up the mail services…"
runFileOp docker restart stalwart-service >/dev/null 2>&1
if ! stalwart_wait_http ready 60; then
isError "Stalwart did not come back up after setup — check: docker logs stalwart-service"
return 1
fi
isSuccessful "Stalwart configured: ${mail_domain} added, DKIM keys generated."
stalwart_install_dns_provider
stalwart_install_first_mailbox
return 0
}
# Optionally hand DNS to the provider's API.
#
# This is the part that turns a page of records-to-paste into nothing at all:
# Stalwart writes MX, SPF, DKIM, DMARC, MTA-STS, TLS-RPT, SRV and CAA into the
# zone itself and keeps them in sync — including republishing DKIM records when
# it rotates the keys, which is the whole reason rotation is safe to automate.
#
# Off by default. It needs an API token with write access to the zone, stored in
# the mail server's database, which widens what a compromise of this box costs.
# Scope the token to the single zone if your provider supports it.
stalwart_install_dns_provider()
{
local provider="${CFG_STALWART_DNS_PROVIDER:-manual}"
[[ "$provider" == "manual" || -z "$provider" ]] && return 0
if [[ -z "${CFG_STALWART_DNS_API_TOKEN:-}" ]]; then
isError "CFG_STALWART_DNS_PROVIDER is set to '$provider' but no API token was given."
isNotice " Falling back to manual DNS — the records are printed below."
return 0
fi
local mail_domain="${host_setup#*.}"
# Verified against Cloudflare; DigitalOcean and DeSEC take the same
# description+secret shape. Providers needing more than a token (Route 53,
# Google Cloud DNS) are not wired up here — add them in the admin console.
local plan
plan=$(printf '{"@type":"upsert","object":"DnsServer","matchOn":["description"],"value":{"dns":{"@type":"%s","description":"LibrePortal managed DNS","secret":{"@type":"Value","secret":"%s"}}}}\n{"@type":"upsert","object":"Domain","matchOn":["name"],"value":{"dom":{"name":"%s","dnsManagement":{"@type":"Automatic","dnsServerId":"#dns"}}}}' \
"$provider" "$CFG_STALWART_DNS_API_TOKEN" "$mail_domain")
if printf '%s\n' "$plan" | stalwart_cli apply --stdin >/dev/null 2>&1; then
isSuccessful "DNS records will be published and kept in sync via $provider."
else
isError "Could not enable automatic DNS via $provider — check the API token."
isNotice " Manual DNS is still in effect; the records are printed below."
fi
}
# Create one real mailbox.
#
# Setup leaves you with admin@<domain>, which is an administrator account — using
# it as a day-to-day mailbox is the wrong habit to start someone on. More
# concretely: the generated zone file points DMARC and TLS-RPT reports at
# postmaster@<domain>, and nothing creates that address, so those reports would
# bounce. Creating it fixes a real gap and gives the user a mailbox to log into.
stalwart_install_first_mailbox()
{
local mailbox="${CFG_STALWART_FIRST_MAILBOX:-postmaster}"
[[ -z "$mailbox" ]] && return 0
local mail_domain="${host_setup#*.}"
local plan
# The domain is upserted by name purely to get a reference to it — it already
# exists, so this matches rather than creates. Credentials are a map keyed by
# slot, not a list. Stalwart enforces password strength, so the password has
# to be a generated one, not a short hand-picked string.
plan=$(printf '{"@type":"upsert","object":"Domain","matchOn":["name"],"value":{"dom":{"name":"%s"}}}\n{"@type":"upsert","object":"Account","matchOn":["name"],"value":{"acc":{"@type":"User","name":"%s","domainId":"#dom","description":"First mailbox","credentials":{"0":{"@type":"Password","secret":"%s"}}}}}' \
"$mail_domain" "$mailbox" "${CFG_STALWART_FIRST_MAILBOX_PASSWORD:-}")
if printf '%s\n' "$plan" | stalwart_cli apply --stdin >/dev/null 2>&1; then
isSuccessful "Created the first mailbox: ${mailbox}@${mail_domain}"
else
isError "Could not create ${mailbox}@${mail_domain} — add it in the admin console."
fi
}
stalwart_install_post_start()
{
local app_name="$1"
@ -34,7 +203,10 @@ stalwart_install_post_start()
admin_pair=$(tagsManagerGetTagContent "$compose_file" "PORTS_TAG_1" 2>/dev/null)
admin_port="${admin_pair%%:*}"
# ---- 1. Can this host even send mail? --------------------------------
# ---- 1. Answer the setup wizard on the user's behalf ------------------
stalwart_install_provision
# ---- 2. Can this host even send mail? --------------------------------
# Most cheap VPS providers block outbound 25 by default (and several only
# unblock on request). A blocked port 25 means no mail EVER leaves the box,
# and nothing in the WebUI would otherwise reveal it.
@ -48,7 +220,7 @@ stalwart_install_post_start()
isNotice " outbound 25, or mail will queue and never deliver."
fi
# ---- 2. Reverse DNS --------------------------------------------------
# ---- 3. Reverse DNS --------------------------------------------------
# Receiving servers check that the sending IP resolves back to a name. A
# generic provider PTR (e.g. static.1.2.3.4.provider.net) is a common reason
# for mail being junked, and it can only be fixed in the provider's panel.
@ -62,7 +234,7 @@ stalwart_install_post_start()
fi
fi
# ---- 3. Is the admin console actually there? -------------------------
# ---- 4. Is the admin console actually there? -------------------------
# Stalwart v0.16 does not ship the WebUI inside the Docker image: the admin
# console is a single-page app the server fetches from GitHub on first
# start. If this host had no outbound HTTPS at that moment the download
@ -87,29 +259,56 @@ stalwart_install_post_start()
isSuccessful "Admin console is being served (HTTP $admin_code)."
fi
# ---- 4. The records the admin must add themselves --------------------
# Printed with real values so they can be pasted at the registrar. DKIM is
# deliberately NOT guessed here: Stalwart generates the keypair on first
# run, and the public key must be copied from its admin UI.
# ---- 5. The records that still have to be published -------------------
# Read back from the server rather than composed here. Stalwart keeps the
# domain's full record set in `dnsZoneFile`, so this prints what it actually
# expects — MX, SPF, DMARC, MTA-STS, TLS-RPT, the SRV records clients use to
# autoconfigure, and crucially the real DKIM public keys. The old version of
# this hook had to tell the user to go and fetch DKIM themselves; by this
# point the keys exist, so there is nothing left to look up.
local mail_host="${host_setup:-your-mail-hostname}"
local mail_domain="${mail_host#*.}"
echo ""
isNotice "DNS records to add at your domain registrar:"
echo " MX @ 10 ${mail_host}"
echo " A ${mail_host} ${public_ip_v4:-<the IP of this server>}"
echo " TXT @ \"v=spf1 mx -all\""
echo " TXT _dmarc \"v=DMARC1; p=quarantine; rua=mailto:postmaster@${mail_domain}\""
echo " TXT <selector>._domainkey (DKIM — copy from Stalwart's admin UI once it has"
echo " generated the key; it is not known until then)"
echo ""
# ---- 5. Where to go next ---------------------------------------------
if [[ -n "$admin_port" ]]; then
isNotice "Finish setup in the admin interface:"
[[ -n "$public_ip_v4" ]] && echo " http://$public_ip_v4:$admin_port/"
if [[ "${CFG_STALWART_DNS_PROVIDER:-manual}" != "manual" && -n "${CFG_STALWART_DNS_API_TOKEN:-}" ]]; then
isNotice "DNS is managed automatically — no records to add by hand."
isNotice " Still set at your provider, because they are not in the zone:"
echo " A ${mail_host} ${public_ip_v4:-<the IP of this server>}"
echo " PTR ${public_ip_v4:-<the IP of this server>} ${mail_host}"
else
local domain_id zone
domain_id=$(stalwart_cli query Domain 2>/dev/null | awk 'NR==2{print $1}')
if [[ -n "$domain_id" ]]; then
zone=$(stalwart_cli get Domain "$domain_id" 2>/dev/null \
| sed -n '/DNS Zone File:/,$p' | sed 's/^ *DNS Zone File: *//')
fi
echo ""
isNotice "Sign in as '${CFG_STALWART_ADMIN_USER:-admin}' with the password shown below,"
isNotice "then add your domain and create mailboxes."
if [[ -n "$zone" ]]; then
isNotice "DNS records to add at your domain registrar:"
echo " A ${mail_host} ${public_ip_v4:-<the IP of this server>}"
echo ""
printf '%s\n' "$zone" | sed 's/^/ /'
else
# Setup did not complete, so there is no zone to read. Say what is
# needed rather than printing nothing.
isNotice "DNS records to add at your domain registrar:"
echo " MX @ 10 ${mail_host}"
echo " A ${mail_host} ${public_ip_v4:-<the IP of this server>}"
echo " TXT @ \"v=spf1 mx -all\""
echo " TXT _dmarc \"v=DMARC1; p=quarantine; rua=mailto:postmaster@${mail_domain}\""
isNotice " DKIM records are shown in the admin console under your domain."
fi
echo ""
fi
# ---- 6. Where to go next ---------------------------------------------
if [[ -n "$admin_port" ]]; then
isNotice "Admin console (server settings, domains, queue, reports):"
[[ -n "$public_ip_v4" ]] && echo " http://$public_ip_v4:$admin_port/admin"
isNotice "Mailbox settings for everyday users (password, filters, autoresponder):"
[[ -n "$public_ip_v4" ]] && echo " http://$public_ip_v4:$admin_port/account"
echo ""
isNotice "Sign in as '${CFG_STALWART_ADMIN_USER:-admin}' with the password shown below."
fi
isNotice "Until MX, PTR, SPF, DKIM and DMARC are all in place, expect delivery"

View File

@ -98,3 +98,47 @@ CFG_STALWART_PORT_9="stalwart-service|jmap|443:443|disabled|tcp|false|false|fals
CFG_STALWART_AUTH_PROFILE=user_password
CFG_STALWART_ADMIN_USER=admin
CFG_STALWART_ADMIN_PASSWORD=RANDOMIZEDPASSWORD1
#
# =============================================================================
# AUTOMATIC SETUP
# =============================================================================
# Out of the box Stalwart opens a five-screen setup wizard on first sign-in and
# waits for a human. LibrePortal already knows the answers that matter — the
# hostname and the domain — so the installer fills the wizard in instead, and
# the user's first sight of Stalwart is a configured mail server.
#
# CLI_VERSION = the version of the Stalwart CLI used to apply that setup.
# Pinned, not ':latest', for the same reason the server image is: this writes
# configuration, so an unattended jump to a new CLI is not something to
# discover during an install. The CLI is a separate upstream project from the
# server and is NOT included in the server image, which is why it is a
# container of its own rather than something bundled.
CFG_STALWART_CLI_VERSION=1.0.12
#
# FIRST_MAILBOX = local part of one real mailbox created during setup.
# Setup leaves you with admin@<domain>, which is an administrator account
# rather than somewhere to read mail. 'postmaster' is the default for a
# concrete reason: the generated zone file points DMARC and TLS-RPT reports at
# postmaster@<domain>, and if nothing creates it those reports bounce.
# Set empty to skip creating it.
CFG_STALWART_FIRST_MAILBOX=postmaster
CFG_STALWART_FIRST_MAILBOX_PASSWORD=RANDOMIZEDPASSWORD2
#
# DNS_PROVIDER = 'manual', or a provider name to let Stalwart publish DNS itself.
# On 'manual' the installer prints the full record set — including the real
# DKIM public keys — for you to paste at your registrar.
# Given a provider, Stalwart writes MX, SPF, DKIM, DMARC, MTA-STS, TLS-RPT,
# SRV and CAA into the zone through the provider's API and keeps them in sync.
# That also makes DKIM key rotation safe to leave switched on, since the server
# republishes the records itself — hand-managed DKIM keys are, in practice,
# keys nobody ever rotates.
#
# Wired up here: Cloudflare, DigitalOcean, DeSEC (token-only providers).
# Route 53 and Google Cloud DNS need more than a token — set those up in the
# admin console instead.
#
# The trade: the token is stored in the mail server's database and can write to
# your zone, so compromising this box means compromising your DNS. Scope the
# token to this one zone if your provider allows it.
CFG_STALWART_DNS_PROVIDER=manual
CFG_STALWART_DNS_API_TOKEN=

View File

@ -862,8 +862,13 @@ declare -gA LP_FN_MAP=(
[showInstructions]="menu/message/instructions.sh"
[sourceBackupLocations]="backup/locations/location_loader.sh"
[sshRemote]="network/ssh/ssh.sh"
[stalwart_cli]="stalwart/scripts/stalwart_install_hooks.sh"
[stalwart_install_dns_provider]="stalwart/scripts/stalwart_install_hooks.sh"
[stalwart_install_first_mailbox]="stalwart/scripts/stalwart_install_hooks.sh"
[stalwart_install_message_data]="stalwart/scripts/stalwart_install_hooks.sh"
[stalwart_install_post_start]="stalwart/scripts/stalwart_install_hooks.sh"
[stalwart_install_provision]="stalwart/scripts/stalwart_install_hooks.sh"
[stalwart_wait_http]="stalwart/scripts/stalwart_install_hooks.sh"
[stalwart_upgrade_admin_ui_code]="stalwart/scripts/stalwart_upgrade_hooks.sh"
[stalwart_upgrade_check_admin_ui]="stalwart/scripts/stalwart_upgrade_hooks.sh"
[stalwart_upgrade_verify]="stalwart/scripts/stalwart_upgrade_hooks.sh"
@ -1900,8 +1905,13 @@ declare -gA LP_FN_ROOT=(
[showInstructions]="scripts"
[sourceBackupLocations]="scripts"
[sshRemote]="scripts"
[stalwart_cli]="containers"
[stalwart_install_dns_provider]="containers"
[stalwart_install_first_mailbox]="containers"
[stalwart_install_message_data]="containers"
[stalwart_install_post_start]="containers"
[stalwart_install_provision]="containers"
[stalwart_wait_http]="containers"
[stalwart_upgrade_admin_ui_code]="containers"
[stalwart_upgrade_check_admin_ui]="containers"
[stalwart_upgrade_verify]="containers"