Files

122 lines
5.6 KiB
Nix

{pkgs, ...}: let
# Caddy's canonical certificate storage (XDG data dir). Renewals land here,
# owned caddy:caddy 0600 — stalwart cannot read it directly, so we copy it
# into a distribution dir stalwart's service (user "stalwart") can reach.
caddyCertDir = "/var/lib/caddy/.local/share/caddy/certificates/acme-v02.api.letsencrypt.org-directory/mail.severijnse.eu";
# Distribution dir read by stalwart (cert 0644, private key regranted to the
# "stalwart" group by systemd.services.stalwart-cert-perm).
distCertDir = "/var/lib/caddy/certificates/acme-v02.api.letsencrypt.org-directory/mail.severijnse.eu";
zoneFile = "/var/lib/coredns/zones/severijnse.eu.db";
# Records the SPKI hash applied at the last restart of stalwart, so cert
# renewals trigger exactly one restart and unchanged certs never do.
stateFile = "/var/lib/tlsa-update/.last-spki";
syncScript = pkgs.writeShellScript "tlsa-update" ''
set -euo pipefail
OPENSSL="${pkgs.openssl}/bin/openssl"
SRC_CERT="${caddyCertDir}/mail.severijnse.eu.crt"
SRC_KEY="${caddyCertDir}/mail.severijnse.eu.key"
DST_CERT="${distCertDir}/mail.severijnse.eu.crt"
DST_KEY="${distCertDir}/mail.severijnse.eu.key"
# Caddy has not obtained the certificate yet (e.g. first boot before HTTP-01).
if [ ! -f "$SRC_CERT" ] || [ ! -f "$SRC_KEY" ]; then
echo "tlsa-update: certificate not found at $caddyCertDir, skipping" >&2
exit 0
fi
# 1) Propagate Caddy's renewed certificate into the distribution dir stalwart
# reads. Caddy stores certs 0600 caddy:caddy. The cert is world-readable;
# the private key is restricted to root (0640) and stalwart-cert-perm
# regrants it to the "stalwart" group so it is not exposed to other local
# users.
install -D -m 0644 "$SRC_CERT" "$DST_CERT"
install -D -m 0640 "$SRC_KEY" "$DST_KEY"
# Stalwart reads the key as user "stalwart" via %{file:...}%; regrant the
# group immediately so every sync leaves it readable (0640 root:stalwart)
# and webadmin config reload never fails with EACCES.
chgrp stalwart "$DST_KEY"
# 2) TLSA 3 1 1 = SHA-256 of the certificate's SubjectPublicKeyInfo (SPKI),
# NOT the whole certificate. Matching type 1 = SHA-256 of the SPKI DER.
HEX=$("$OPENSSL" x509 -in "$DST_CERT" -noout -pubkey 2>/dev/null | "$OPENSSL" pkey -pubin -outform DER 2>/dev/null | "$OPENSSL" dgst -sha256 | cut -d' ' -f2)
# 3) Update the CoreDNS zone. coredns.service regenerates this file from the
# Nix store on every start, so this unit (which is partOf coredns and runs
# after it) re-applies the correct TLSA after each rebuild/restart.
if [ -f "${zoneFile}" ]; then
sed -i -E "s/^(_25\._tcp\.mail.*TLSA 3 1 1).*/\1 $HEX/" "${zoneFile}"
sed -i -E "s/^(_465\._tcp\.mail.*TLSA 3 1 1).*/\1 $HEX/" "${zoneFile}"
sed -i -E "s/^(_993\._tcp\.mail.*TLSA 3 1 1).*/\1 $HEX/" "${zoneFile}"
chown coredns:coredns "${zoneFile}"
chmod 0640 "${zoneFile}"
fi
# 4) Reload services so the changes take effect immediately.
systemctl reload coredns.service || true
# Stalwart reads its TLS certs ($certDir) via %{file:...}% placeholders at
# startup only; there is no signal-based reload (management API reload needs
# admin credentials we must not store). Restart it, but only when the cert
# actually changed (SPKI hash differs from the last applied run), so the
# daily timer and inotify events for unchanged certs do not drop connections.
# NOTE: stalwart.service `requires` this unit, so the restart must be issued
# with --no-block (async): a synchronous restart waits for stalwart to come
# back up, which in turn waits for this unit to finish — a deadlock. The state
# file is updated BEFORE the restart so the tlsa-update run that stalwart's
# required-activation re-triggers sees a matching hash and exits immediately.
if [ "$(cat "${stateFile}" 2>/dev/null || true)" != "$HEX" ]; then
printf '%s\n' "$HEX" > "${stateFile}"
# Restarting stalwart re-runs its Requires=tlsa-update dependency; guard
# with is-active so a boot-time run never races stalwart's initial start.
if systemctl is-active --quiet stalwart.service 2>/dev/null; then
systemctl --no-block restart stalwart.service || true
fi
fi
echo "tlsa-update: TLSA set to $HEX"
'';
in {
systemd = {
# Ensure the distribution dir exists (Caddy does not write here).
tmpfiles.rules = [
"d ${distCertDir} 0755 root root - -"
"d /var/lib/tlsa-update 0755 root root - -"
];
services.tlsa-update = {
description = "Sync Caddy TLS certificate to mail server and update DANE/TLSA records";
after = ["caddy.service" "coredns.service"];
partOf = ["coredns.service"];
wantedBy = ["multi-user.target"];
path = with pkgs; [openssl coreutils gnused systemd];
serviceConfig = {
Type = "oneshot";
ExecStart = "${syncScript}";
User = "root";
Group = "root";
};
};
# Fire as soon as Caddy rewrites the certificate on renewal (the atomic rewrite
# changes the directory mtime), eliminating the up-to-24h DANE drift window.
paths.tlsa-update = {
description = "Watch Caddy certificate directory for renewal";
wantedBy = ["paths.target"];
pathConfig = {
PathModified = [caddyCertDir];
Unit = "tlsa-update.service";
};
};
# Fallback in case a renewal event is missed (e.g. inotify overflow).
timers.tlsa-update = {
wantedBy = ["timers.target"];
timerConfig = {
OnCalendar = "daily";
Persistent = true;
};
};
};
}