feat(kanidm): add Kanidm service configuration and documentation

systems/jeeves/kanidm.md

@@ -0,0 +1,96 @@
+# Kanidm on Jeeves
+
+Kanidm is served at **https://auth.tmmworkshop.com**. Portal terminates public
+HTTPS and forwards requests over Tailscale to Jeeves on port 8443, verifying
+Jeeves's TLS certificate. Only the Tailscale interface exposes that backend port.
+
+The service module is automatically imported by `services/default.nix`.
+`services/kanidm.nix` pins Kanidm to the 1.11 release series; review the
+[upgrade guide](https://kanidm.github.io/kanidm/stable/server_updates.html)
+before changing that series. The flake lock determines the patch release.
+
+## Before deployment
+
+1. Point `auth.tmmworkshop.com` at Portal's public address, using the same DNS
+   arrangement as `gitea.tmmworkshop.com`. Port 80 must reach Portal for ACME.
+2. Enable [HTTPS certificates in Tailscale](https://tailscale.com/kb/1153/enabling-https).
+   Jeeves must be able to obtain a certificate for `jeeves.taile39c31.ts.net`.
+3. Allow Portal to reach Jeeves on TCP 8443 in your Tailscale policy.
+
+The trusted proxy address in `services/kanidm.nix` is Portal's current Tailscale
+IPv4 address, `100.80.102.33`. Update it if Portal is re-enrolled with a new address.
+HAProxy connects over IPv4 and replaces incoming `X-Forwarded-For` headers.
+
+## Deploy
+
+Ensure the new Nix file is tracked by Git before using this Git-backed flake.
+From this repository, using the existing SSH aliases:
+
+```sh
+nixos-rebuild switch --flake .#jeeves --target-host jeeves --sudo
+nixos-rebuild switch --flake .#portal-1 --target-host portal-1 --sudo
+```
+
+Deploy Jeeves first, then Portal. Portal manages the public Let's Encrypt
+certificate. `kanidm-tls.service` on Jeeves obtains the backend certificate and
+checks for renewal daily, restarting Kanidm only when the certificate changes.
+Private keys are generated at runtime and never stored in this repository or
+the Nix store.
+
+On Jeeves, check startup:
+
+```sh
+sudo systemctl status kanidm-tls.service kanidm.service
+sudo journalctl -u kanidm-tls.service -u kanidm.service -n 100
+```
+
+If the initial certificate request fails, fix the Tailscale HTTPS configuration
+and run `sudo systemctl restart kanidm.service`; its dependency retries the
+certificate request. A successful oneshot certificate service normally appears
+as inactive after finishing.
+
+After both hosts are deployed:
+
+```sh
+curl --fail https://auth.tmmworkshop.com/status
+```
+
+## First login
+
+On Jeeves, recover the initial administrator passwords interactively:
+
+```sh
+sudo kanidmd recover-account admin
+sudo kanidmd recover-account idm_admin
+kanidm login --name idm_admin
+kanidm person create richie "Richie"
+kanidm person credential create-reset-token richie
+```
+
+Store the recovered passwords securely. Follow the reset URL to enroll your
+personal account's passkey. `admin` manages the Kanidm server; `idm_admin`
+manages people, groups, and application integrations. Creating `richie` does
+not automatically grant administrator rights. The client is installed and
+configured to use `https://auth.tmmworkshop.com`.
+
+Register each website as its own OAuth2/OIDC client and each backend identity
+as a service account, then grant application scopes through groups. This
+deployment does not create application clients or change existing sites' login
+settings. See the [OAuth2 guide](https://kanidm.github.io/kanidm/stable/integrations/oauth2.html)
+and [service-account guide](https://kanidm.github.io/kanidm/stable/accounts/service_accounts.html).
+
+## Storage and recovery
+
+- Database: `/var/lib/kanidm/kanidm.db` on Jeeves's existing ZFS `/var` filesystem.
+- Online backups: `/var/lib/kanidm/backups`, daily at 02:30, retaining 14.
+- Backend certificate and key: `/var/lib/kanidm-tls`, readable by Kanidm.
+- Server configuration: `/etc/kanidm/server.toml`.
+
+The existing `root_pool/var` snapshot policy covers the database and backups.
+Copy backups off Jeeves as part of disaster recovery; local snapshots do not
+protect against loss of the host. Follow Kanidm's
+[backup and restore instructions](https://kanidm.github.io/kanidm/stable/backup_and_restore.html).
+
+The configured identity domain and origin are both `auth.tmmworkshop.com`.
+Changing them after enrollment can affect passkeys and application integrations;
+follow the [domain rename guide](https://kanidm.github.io/kanidm/stable/domain_rename.html).

systems/jeeves/networking/firewall.nix

@@ -8,6 +8,7 @@ let
     grafana = 3000;
     jellyfin_http = 8096;
     jellyswarrm = 3100;
+    kanidm = 8443;
     nix_binary_cache = 5000;
     ollama = 11434;
     open_webui = 8080;
@@ -28,6 +29,7 @@ in
       ports.gems
       ports.jellyfin_http
       ports.jellyswarrm
+      ports.kanidm
       ports.nix_binary_cache
       ports.share
     ];

systems/jeeves/services/kanidm.nix

@@ -0,0 +1,101 @@
+{ config, pkgs, ... }:
+let
+  certificateDir = "/var/lib/kanidm-tls";
+in
+{
+  services.kanidm = {
+    # Pin the minor release: Kanidm upgrades require following its upgrade guide.
+    package = pkgs.kanidm_1_11;
+    server = {
+      enable = true;
+      settings = {
+        domain = "auth.tmmworkshop.com";
+        origin = "https://auth.tmmworkshop.com";
+        bindaddress = "0.0.0.0:8443";
+        db_fs_type = "zfs";
+        tls_chain = "${certificateDir}/certificate.pem";
+        tls_key = "${certificateDir}/private-key.pem";
+
+        # Only Portal may supply the original client address. HAProxy replaces
+        # X-Forwarded-For, so clients cannot supply a forged address themselves.
+        http_client_address_info.x-forward-for = [ "100.80.102.33" ];
+
+        # /var/lib/kanidm is on root_pool/var, which is already snapshotted.
+        online_backup = {
+          schedule = "30 2 * * *";
+          versions = 14;
+        };
+      };
+    };
+    client = {
+      enable = true;
+      settings.uri = "https://auth.tmmworkshop.com";
+    };
+  };
+
+  systemd = {
+    services = {
+      kanidm = {
+        requires = [ "kanidm-tls.service" ];
+        after = [ "kanidm-tls.service" ];
+      };
+
+      # Kanidm requires TLS even behind HAProxy. Use Jeeves's Tailscale
+      # certificate so Portal can verify both the CA and the backend hostname.
+      # HTTPS certificates must be enabled in the Tailscale admin console.
+      kanidm-tls = {
+        description = "Obtain and renew Kanidm's backend TLS certificate";
+        requires = [ "tailscaled.service" ];
+        wants = [ "tailscaled-autoconnect.service" ];
+        after = [
+          "tailscaled.service"
+          "tailscaled-autoconnect.service"
+        ];
+        path = [
+          config.services.tailscale.package
+          pkgs.coreutils
+          pkgs.systemd
+        ];
+        serviceConfig = {
+          Type = "oneshot";
+          User = "root";
+          Group = "kanidm";
+          StateDirectory = "kanidm-tls";
+          StateDirectoryMode = "0750";
+          RuntimeDirectory = "kanidm-tls";
+          RuntimeDirectoryMode = "0700";
+          UMask = "0077";
+          TimeoutStartSec = "2min";
+        };
+        script = ''
+          set -euo pipefail
+          tailscale cert --min-validity 720h \
+            --cert-file /run/kanidm-tls/certificate.pem \
+            --key-file /run/kanidm-tls/private-key.pem \
+            jeeves.taile39c31.ts.net
+
+          if ! cmp -s /run/kanidm-tls/certificate.pem ${certificateDir}/certificate.pem \
+            || ! cmp -s /run/kanidm-tls/private-key.pem ${certificateDir}/private-key.pem; then
+            install -m 0640 -o root -g kanidm /run/kanidm-tls/certificate.pem ${certificateDir}/certificate.pem
+            install -m 0640 -o root -g kanidm /run/kanidm-tls/private-key.pem ${certificateDir}/private-key.pem
+
+            # The NixOS service bind-mounts the certificate files, so restart to
+            # pick up renewals. Do not wait on a service that depends on this one.
+            if systemctl is-active --quiet kanidm.service; then
+              systemctl --no-block try-restart kanidm.service
+            fi
+          fi
+        '';
+      };
+    };
+
+    timers.kanidm-tls = {
+      wantedBy = [ "timers.target" ];
+      timerConfig = {
+        OnCalendar = "daily";
+        Persistent = true;
+        RandomizedDelaySec = "1h";
+      };
+    };
+  };
+}

systems/portal-1/haproxy/haproxy.cfg

@@ -32,7 +32,7 @@ defaults
 #Application Setup
 frontend ContentSwitching
   bind [::]:80 v4v6
-  bind [::]:443 v4v6 ssl crt /var/lib/acme/audiobookshelf.tmmworkshop.com/full.pem crt /var/lib/acme/cache.tmmworkshop.com/full.pem crt /var/lib/acme/gems.tmmworkshop.com/full.pem crt /var/lib/acme/jellyfin.tmmworkshop.com/full.pem crt /var/lib/acme/share.tmmworkshop.com/full.pem crt /var/lib/acme/gitea.tmmworkshop.com/full.pem crt
+  bind [::]:443 v4v6 ssl crt /var/lib/acme/audiobookshelf.tmmworkshop.com/full.pem crt /var/lib/acme/auth.tmmworkshop.com/full.pem crt /var/lib/acme/cache.tmmworkshop.com/full.pem crt /var/lib/acme/gems.tmmworkshop.com/full.pem crt /var/lib/acme/jellyfin.tmmworkshop.com/full.pem crt /var/lib/acme/share.tmmworkshop.com/full.pem crt /var/lib/acme/gitea.tmmworkshop.com/full.pem
   mode  http
 
   # ACME challenge routing (must be first)
@@ -40,6 +40,7 @@ frontend ContentSwitching
 
   # Host ACLs (defined early so rate-limiting can scope to a single vhost)
   acl host_audiobookshelf  hdr(host) -i audiobookshelf.tmmworkshop.com
+  acl host_auth  hdr(host) -i auth.tmmworkshop.com
   acl host_cache  hdr(host) -i cache.tmmworkshop.com
   acl host_jellyfin  hdr(host) -i jellyfin.tmmworkshop.com
   acl host_share  hdr(host) -i share.tmmworkshop.com
@@ -111,6 +112,7 @@ frontend ContentSwitching
 
   use_backend acme_challenge if is_acme
   use_backend audiobookshelf_nodes if host_audiobookshelf
+  use_backend kanidm if host_auth
   use_backend cache_nodes  if host_cache
   use_backend jellyfin if host_jellyfin
   use_backend share_nodes  if host_share
@@ -141,6 +143,15 @@ backend audiobookshelf_nodes
   mode http
   server server jeeves:8000
 
+backend kanidm
+  mode http
+  http-request set-header X-Forwarded-For %[src]
+  http-request set-header X-Forwarded-Proto https
+  option httpchk
+  http-check send meth GET uri /status ver HTTP/1.1 hdr Host auth.tmmworkshop.com
+  http-check expect status 200
+  server kanidm jeeves.taile39c31.ts.net:8443 ssl verify required ca-file /etc/ssl/certs/ca-certificates.crt verifyhost jeeves.taile39c31.ts.net sni str(jeeves.taile39c31.ts.net) check check-sni jeeves.taile39c31.ts.net resolve-prefer ipv4
+
 backend cache_nodes
   mode http
   server server jeeves:5000