Merge pull request 'Passer wgautomesh en prod' (#9) from wgautomesh into main

Reviewed-on: #9

cluster/prod/cluster.nix

@@ -13,7 +13,6 @@
       site_name = "neptune";
       publicKey = "VvXT0fPDfWsHxumZqVShpS33dJQAdpJ1E79ZbCBJP34=";
       IP = "10.83.1.1";
-      lan_endpoint = "192.168.1.31:33799";
       endpoint = "77.207.15.215:33731";
     }
     {
@@ -21,7 +20,6 @@
       site_name = "neptune";
       publicKey = "goTkBJGmzrGDOAjUcdH9G0JekipqSMoaYQdB6IHnzi0=";
       IP = "10.83.1.2";
-      lan_endpoint = "192.168.1.32:33799";
       endpoint = "77.207.15.215:33732";
     }
     {
@@ -29,7 +27,6 @@
       site_name = "neptune";
       publicKey = "oZDAb8LoLW87ktUHyFFec0VaIar97bqq47mGbdVqJ0U=";
       IP = "10.83.1.3";
-      lan_endpoint = "192.168.1.33:33799";
       endpoint = "77.207.15.215:33733";
     }
     {
@@ -37,7 +34,6 @@
       site_name = "orion";
       publicKey = "EtRoWBYCdjqgXX0L+uWLg8KxNfIK8k9OTh30tL19bXU=";
       IP = "10.83.2.1";
-      lan_endpoint = "192.168.1.11:33799";
       endpoint = "82.66.80.201:33731";
     }
     {
@@ -45,7 +41,6 @@
       site_name = "orion";
       publicKey = "HbLC938mysadMSOxWgq8+qrv+dBKzPP/43OMJp/3phA=";
       IP = "10.83.2.2";
-      lan_endpoint = "192.168.1.12:33799";
       endpoint = "82.66.80.201:33732";
     }
     {
@@ -53,7 +48,6 @@
       site_name = "orion";
       publicKey = "e1C8jgTj9eD20ywG08G1FQZ+Js3wMK/msDUE1wO3l1Y=";
       IP = "10.83.2.3";
-      lan_endpoint = "192.168.1.13:33799";
       endpoint = "82.66.80.201:33733";
     }
     {
@@ -61,7 +55,6 @@
       site_name = "bespin";
       publicKey = "bIjxey/VhBgVrLa0FxN/KISOt2XFmQeSh1MPivUq9gg=";
       IP = "10.83.3.1";
-      lan_endpoint = "192.168.5.117:33799";
       endpoint = "109.136.55.235:33731";
     }
     {
@@ -69,7 +62,6 @@
       site_name = "bespin";
       publicKey = "pUIKv8UBl586O7DBrHBsb9BgNU7WlYQ2r2RSNkD+JAQ=";
       IP = "10.83.3.2";
-      lan_endpoint = "192.168.5.134:33799";
       endpoint = "109.136.55.235:33732";
     }
     {
@@ -77,7 +69,6 @@
       site_name = "bespin";
       publicKey = "VBmpo15iIJP7250NAsF+ryhZc3j+8TZFnE1Djvn5TXI=";
       IP = "10.83.3.3";
-      lan_endpoint = "192.168.5.116:33799";
       endpoint = "109.136.55.235:33733";
     }
     {
@@ -85,7 +76,6 @@
       site_name = "scorpio";
       publicKey = "Sm9cmNZ/BfWVPFflMO+fuyiera4r203b/dKhHTQmBFg=";
       IP = "10.83.4.1";
-      lan_endpoint = "192.168.1.41:33799";
       endpoint = "82.65.41.110:33741";
     }
   ];

cluster/staging/cluster.nix

@@ -13,7 +13,6 @@
       site_name = "neptune";
       publicKey = "7Nm7pMmyS7Nts1MB+loyD8u84ODxHPTkDu+uqQR6yDk=";
       IP = "10.14.1.2";
-      lan_endpoint = "192.168.1.22:33799";
       endpoint = "77.207.15.215:33722";
     }
     {
@@ -21,7 +20,6 @@
       site_name = "neptune";
       publicKey = "lABn/axzD1jkFulX8c+K3B3CbKXORlIMDDoe8sQVxhs=";
       IP = "10.14.1.3";
-      lan_endpoint = "192.168.1.23:33799";
       endpoint = "77.207.15.215:33723";
     }
     {
@@ -29,7 +27,6 @@
       site_name = "jupiter";
       publicKey = "smBQYUS60JDkNoqkTT7TgbpqFiM43005fcrT6472llI=";
       IP = "10.14.2.33";
-      lan_endpoint = "192.168.1.33:33799";
       endpoint = "82.64.238.84:33733";
     }
     {
@@ -37,18 +34,17 @@
       site_name = "corrin";
       publicKey = "m9rLf+233X1VColmeVrM/xfDGro5W6Gk5N0zqcf32WY=";
       IP = "10.14.3.1";
-      lan_endpoint = "192.168.1.25:33799";
+      #endpoint = "82.120.233.78:33721";
-      endpoint = "82.120.233.78:33721";
     }
     {
       hostname = "df-pw5";
       site_name = "bespin";
       publicKey = "XLOYoMXF+PO4jcgfSVAk+thh4VmWx0wzWnb0xs08G1s=";
       IP = "10.14.4.1";
-      lan_endpoint = "192.168.5.130:33799";
       endpoint = "bitfrost.fiber.shirokumo.net:33734";
     }
   ];
+  services.wgautomesh.logLevel = "debug";
 
   # Bootstrap IPs for Consul cluster,
   # these are IPs on the Wireguard overlay

cluster/staging/known_hosts

@@ -8,3 +8,4 @@ piranha.polyno.me,2a01:cb05:8984:3c00:223:24ff:feb0:ea82 ssh-ed25519 AAAAC3NzaC1
 2001:910:1204:1::21 ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIPXTUrXRFhudJBESCqjHCOttzqYPyIzpPOMkI8+SwLRx
 2a01:e0a:5e4:1d0:223:24ff:feaf:fdec ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIAsZas74RT6lCZwuUOPR23nPdbSdpWORyAmRgjoiMVHK
 df-pw5.machine.deuxfleurs.fr ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIK/dJIxioCkfeehxeGiZR7qquYGoqEH/YrRJ/ukEcaLH
+10.14.3.1 ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIJnpO6zpLWsyyugOoOj+2bUow9TUrcWgURFGGaoyu+co

cluster/staging/ssh_config

@@ -10,7 +10,9 @@ Host origan
 	HostName origan.df.trinity.fr.eu.org
 
 Host piranha
-	HostName piranha.polyno.me
+	ProxyJump caribou.machine.deuxfleurs.fr
+	HostName 10.14.3.1
+	#HostName piranha.polyno.me
 
 Host df-pw5
 	HostName df-pw5.machine.deuxfleurs.fr

deploy_nixos

@@ -7,6 +7,10 @@ copy cluster/$CLUSTER/cluster.nix /etc/nixos/cluster.nix
 copy cluster/$CLUSTER/node/$NIXHOST.nix /etc/nixos/node.nix
 copy cluster/$CLUSTER/node/$NIXHOST.site.nix /etc/nixos/site.nix
 
+cmd mkdir -p /var/lib/wgautomesh
+write_pass deuxfleurs/cluster/$CLUSTER/wgautomesh_gossip_secret /var/lib/wgautomesh/gossip_secret
+copy nix/wgautomesh.nix /etc/nixos/wgautomesh.nix
+
 if [ "$CLUSTER" = "staging" ]; then
 	copy nix/nomad-driver-nix2.nix /etc/nixos/nomad-driver-nix2.nix
 fi
@@ -14,6 +18,9 @@ fi
 if [ "$CLUSTER" = "prod" ]; then
 	cmd nixos-rebuild boot
 	message "-------------------------------------------------------------------------------------"
+	message "NIXOS CHANGES THAT WOULD BE APPLIED UPON SWITCH OR REBOOT:"
+	cmd nixos-rebuild dry-activate
+	message "-------------------------------------------------------------------------------------"
 	message "New NixOS configuration hasn't been applied, to avoid disturbing production services."
 	message "Please apply the following procedure to node '$NIXHOST':"
 	message "1. Drain node in Nomad so that all jobs are relocated elsewhere"

doc/ports

@@ -16,6 +16,7 @@ ports so that we can avoid conflicts when adding services.
 587     postfix submission
 993     dovecot imaps
 1337    dovecot zauthoentication
+1666    wgautomesh
 1728    nix-serve
 3719    grafana
 3900    garage S3 (prod)

nix/configuration.nix

@@ -84,5 +84,8 @@ SystemMaxUse=1G
       dns = [ "172.17.0.1" ];
     })}";
   };
+
+  nix.gc.automatic = true;
+  nix.gc.options = "--delete-older-than 30d";
 }
 

nix/deuxfleurs.nix

@@ -28,12 +28,8 @@ in
         };
         endpoint = mkOption {
           type = nullOr str;
-          description = "Wireguard endpoint on the public Internet";
-        };
-        lan_endpoint = mkOption {
-          type = nullOr str;
-          description = "Wireguard endpoint for nodes in the same site";
           default = null;
+          description = "Wireguard endpoint on the public Internet";
         };
       };
     };
@@ -134,6 +130,10 @@ in
     };
   };
 
+  imports = [
+      ./wgautomesh.nix
+    ];
+
   config =
     let node_meta = {
         "site" = cfg.site_name;
@@ -147,6 +147,7 @@ in
         else {});
     in
   {
+
     # Configure admin accounts on all nodes
     users.users = builtins.mapAttrs (name: publicKeys: {
       isNormalUser = true;
@@ -233,20 +234,43 @@ in
     services.resolved.enable = false;
 
     # Configure Wireguard VPN between all nodes
-    systemd.services."wg-quick-wg0".after = [ "unbound.service" ];
+    networking.wireguard.interfaces.wg0 = {
-    networking.wg-quick.interfaces.wg0 = {
+      ips = [ "${cfg.cluster_ip}/16" ];
-      address = [ "${cfg.cluster_ip}/16" ];
       listenPort = cfg.wireguard_port;
       privateKeyFile = "/var/lib/deuxfleurs/wireguard-keys/private";
       mtu = 1420;
-      peers = map ({ publicKey, endpoint, IP, site_name, lan_endpoint, ... }: {
+    };
-        publicKey = publicKey;
+    services.wgautomesh = {
-        allowedIPs = [ "${IP}/32" ];
+      enable = true;
-        endpoint = if site_name != null && site_name == cfg.site_name && lan_endpoint != null
+      interface = "wg0";
-                   then lan_endpoint else endpoint;
+      gossipPort = 1666;
-        persistentKeepalive = 25;
+      gossipSecretFile = "/var/lib/wgautomesh/gossip_secret";
+      persistFile = "/var/lib/wgautomesh/state";
+      upnpForwardPublicPort = 
+        let
+          us = filter ({ hostname, ...}: hostname == config.networking.hostName) cfg.cluster_nodes;
+        in
+        if length us > 0 && (head us).endpoint != null then
+          strings.toInt (lists.last (split ":" (head us).endpoint))
+        else null;
+      peers = map ({ publicKey, endpoint, IP, ... }: {
+        address = IP;
+        pubkey = publicKey;
+        endpoint = endpoint;
       }) cfg.cluster_nodes;
     };
+    # Old code for wg-quick, we can use this as a fallback if we fail to make wgautomesh work
+    # systemd.services."wg-quick-wg0".after = [ "unbound.service" ];
+    # networking.wg-quick.interfaces.wg0 = {
+    #   address = [ "${cfg.cluster_ip}/16" ];
+    #   listenPort = cfg.wireguard_port;
+    #   privateKeyFile = "/var/lib/deuxfleurs/wireguard-keys/private";
+    #   mtu = 1420;
+    #   peers = map ({ publicKey, endpoint, IP, ... }: {
+    #     inherit publicKey endpoint;
+    #     allowedIPs = [ "${IP}/32" ];
+    #     persistentKeepalive = 25;
+    # };
 
     system.activationScripts.generate_df_wg_key = ''
       if [ ! -f /var/lib/deuxfleurs/wireguard-keys/private ]; then
@@ -303,7 +327,7 @@ in
 
     services.nomad.enable = true;
     systemd.services.nomad.after = [ "wg-quick-wg0.service" ];
-    services.nomad.package = pkgs.nomad_1_3;
+    services.nomad.package = pkgs.nomad_1_4;
     services.nomad.extraPackages = [
       pkgs.glibc
       pkgs.zstd

nix/wgautomesh.nix

@@ -0,0 +1,129 @@
+let
+  wgautomesh = builtins.fetchTarball {
+    url = "https://git.deuxfleurs.fr/attachments/ce203833-1ae7-43d4-9bf4-b49b560c9f4b";
+    sha256 = "sha256:1kc990s7xkwff53vs6c3slg7ljsyr9xz1i13j61ivfj6djyh8rmj";
+  };
+in
+
+{ lib, config, pkgs, ... }:
+with lib;
+let 
+  cfg = config.services.wgautomesh;
+in
+  with builtins;
+  {
+    options.services.wgautomesh = {
+      enable = mkEnableOption "wgautomesh";
+      logLevel = mkOption {
+        type = types.enum [ "trace" "debug" "info" "warn" "error" ];
+        default = "info";
+        description = "wgautomesh log level (trace/debug/info/warn/error)";
+      };
+      interface = mkOption {
+        type = types.str;
+        description = "Wireguard interface to manage";
+      };
+      gossipPort = mkOption {
+        type = types.port;
+        description = "wgautomesh gossip port";
+      };
+      gossipSecretFile = mkOption {
+        type = types.nullOr types.str;
+        description = "File containing the gossip secret encryption key";
+      };
+      persistFile = mkOption {
+        type = types.nullOr types.str;
+        description = "Path where to persist known peer addresses";
+      };
+      lanDiscovery = mkOption {
+        type = types.bool;
+        default = true;
+        description = "Enable discovery using LAN broadcast";
+      };
+      openFirewall = mkOption {
+        type = types.bool;
+        default = true;
+        description = "Automatically open gossip port in firewall";
+      };
+      upnpForwardPublicPort = mkOption {
+        type = types.nullOr types.port;
+        default = null;
+        description = "Public port number to try to redirect to this machine using UPnP IGD";
+      };
+      peers = mkOption {
+        type = types.listOf (types.submodule {
+          options = {
+            pubkey = mkOption {
+              type = types.str;
+              description = "Wireguard public key";
+            };
+            address = mkOption {
+              type = types.str;
+              description = "Wireguard peer address";
+            };
+            endpoint = mkOption {
+              type = types.nullOr types.str;
+              description = "bootstrap endpoint";
+            };
+          };
+        });
+        description = "wgautomesh peer list";
+      };
+    };
+
+    config = mkIf cfg.enable ( 
+    let
+      peerDefs = map (peer:
+        let endpointDef = if peer.endpoint == null then ""
+        else ''endpoint = "${peer.endpoint}"'';
+        in
+        ''
+          [[peers]]
+          pubkey = "${peer.pubkey}"
+          address = "${peer.address}"
+          ${endpointDef}
+        '') cfg.peers;
+      extraDefs = (if cfg.lanDiscovery then ["lan_discovery = true"] else [])
+        ++ (if (cfg.gossipSecretFile != null)
+          then [''gossip_secret_file = "${cfg.gossipSecretFile}"''] else [])
+        ++ (if (cfg.persistFile != null)
+          then [''persist_file = "${cfg.persistFile}"''] else [])
+        ++ (if (cfg.upnpForwardPublicPort != null)
+          then [''upnp_forward_external_port = ${toString cfg.upnpForwardPublicPort}''] else []);
+      configfile = pkgs.writeText "wgautomesh.toml" ''
+        interface = "${cfg.interface}"
+        gossip_port = ${toString cfg.gossipPort}
+        ${concatStringsSep "\n" extraDefs}
+
+        ${concatStringsSep "\n" peerDefs}
+        '';
+    in {
+      systemd.services.wgautomesh = {
+        enable = true;
+        path = [ pkgs.wireguard-tools ];
+        environment = {
+          RUST_LOG = "wgautomesh=${cfg.logLevel}";
+        };
+        description = "wgautomesh";
+        serviceConfig = {
+          Type = "simple";
+
+          ExecStart = "${wgautomesh}/bin/wgautomesh ${configfile}";
+          Restart = "always";
+          RestartSec = "30";
+
+          ExecStartPre = [ "+${pkgs.coreutils}/bin/chown wgautomesh /var/lib/wgautomesh/gossip_secret" ];
+
+          DynamicUser = true;
+          User = "wgautomesh";
+          StateDirectory = "wgautomesh";
+          StateDirectoryMode = "0700";
+          AmbientCapabilities = "CAP_NET_ADMIN";
+          CapabilityBoundingSet = "CAP_NET_ADMIN";
+        };
+        wantedBy = [ "multi-user.target" ];
+      };
+      networking.firewall.allowedUDPPorts = mkIf cfg.openFirewall [ cfg.gossipPort ];
+    });
+  }
+