Merge: nixos/postgresql: set up sandboxing (#344925)

nixos/doc/manual/release-notes/rl-2411.section.md

@@ -790,6 +790,11 @@
 
 - `iproute2` now has libbpf support.
 
+- `postgresql` is now [hardened by default](#module-services-postgres-hardening) using the common `systemd` settings for that.
+
+  If you use extensions that are not packaged in nixpkgs, please review whether it still works
+  with the current settings and adjust accordingly if needed.
+
 - `nix.channel.enable = false` no longer implies `nix.settings.nix-path = []`.
   Since Nix 2.13, a `nix-path` set in `nix.conf` cannot be overridden by the `NIX_PATH` configuration variable.
 

nixos/modules/services/databases/postgresql.md

@@ -364,6 +364,24 @@ postgresql.withJIT.pname
 
 evaluates to `"foobar"`.
 
+## Service hardening {#module-services-postgres-hardening}
+
+The service created by the [`postgresql`-module](#opt-services.postgresql.enable) uses
+several common hardening options from `systemd`, most notably:
+
+* Memory pages must not be both writable and executable (this only applies to non-JIT setups).
+* A system call filter (see {manpage}`systemd.exec(5)` for details on `@system-service`).
+* A stricter default UMask (`0027`).
+* Only sockets of type `AF_INET`/`AF_INET6`/`AF_NETLINK`/`AF_UNIX` allowed.
+* Restricted filesystem access (private `/tmp`, most of the file-system hierachy is mounted read-only, only process directories in `/proc` that are owned by the same user).
+
+The NixOS module also contains necessary adjustments for extensions from `nixpkgs`
+if these are enabled. If an extension or a postgresql feature from `nixpkgs` breaks
+with hardening, it's considered a bug.
+
+When using extensions that are not packaged in `nixpkgs`, hardening adjustments may
+become necessary.
+
 ## Notable differences to upstream {#module-services-postgres-upstream-deviation}
 
 - To avoid circular dependencies between default and -dev outputs, the output of the `pg_config` system view has been removed.

nixos/modules/services/databases/postgresql.nix

@@ -623,7 +623,46 @@ in
             TimeoutSec = 120;
 
             ExecStart = "${postgresql}/bin/postgres";
+
+            # Hardening
+            CapabilityBoundingSet = [ "" ];
+            DevicePolicy = "closed";
+            PrivateTmp = true;
+            ProtectHome = true;
+            ProtectSystem = "strict";
+            MemoryDenyWriteExecute = lib.mkDefault (cfg.settings.jit == "off");
+            NoNewPrivileges = true;
+            LockPersonality = true;
+            PrivateDevices = true;
+            PrivateMounts = true;
+            ProcSubset = "pid";
+            ProtectClock = true;
+            ProtectControlGroups = true;
+            ProtectHostname = true;
+            ProtectKernelLogs = true;
+            ProtectKernelModules = true;
+            ProtectKernelTunables = true;
+            ProtectProc = "invisible";
+            RemoveIPC = true;
+            RestrictAddressFamilies = [
+              "AF_INET"
+              "AF_INET6"
+              "AF_NETLINK" # used for network interface enumeration
+              "AF_UNIX"
+            ];
+            RestrictNamespaces = true;
+            RestrictRealtime = true;
+            RestrictSUIDSGID = true;
+            SystemCallArchitectures = "native";
+            SystemCallFilter = [
+              "@system-service"
+              "~@privileged @resources"
+            ];
+            UMask = if groupAccessAvailable then "0027" else "0077";
           }
+          (mkIf (cfg.dataDir != "/var/lib/postgresql") {
+            ReadWritePaths = [ cfg.dataDir ];
+          })
           (mkIf (cfg.dataDir == "/var/lib/postgresql/${cfg.package.psqlSchema}") {
             StateDirectory = "postgresql postgresql/${cfg.package.psqlSchema}";
             StateDirectoryMode = if groupAccessAvailable then "0750" else "0700";

nixos/tests/postgresql-wal-receiver.nix

@@ -22,8 +22,8 @@ let
       replicationUser = "wal_receiver_user";
       replicationSlot = "wal_receiver_slot";
       replicationConn = "postgresql://${replicationUser}@localhost";
-      baseBackupDir = "/tmp/pg_basebackup";
-      walBackupDir = "/tmp/pg_wal";
+      baseBackupDir = "/var/cache/wals/pg_basebackup";
+      walBackupDir = "/var/cache/wals/pg_wal";
 
       recoveryFile = pkgs.writeTextDir "recovery.signal" "";
 
@@ -32,6 +32,10 @@ let
       meta.maintainers = with lib.maintainers; [ pacien ];
 
       nodes.machine = { ... }: {
+        systemd.tmpfiles.rules = [
+          "d /var/cache/wals 0750 postgres postgres - -"
+        ];
+
         services.postgresql = {
           package = pkg;
           enable = true;
@@ -60,6 +64,7 @@ let
         # This is only to speedup test, it isn't time racing. Service is set to autorestart always,
         # default 60sec is fine for real system, but is too much for a test
         systemd.services.postgresql-wal-receiver-main.serviceConfig.RestartSec = lib.mkForce 5;
+        systemd.services.postgresql.serviceConfig.ReadWritePaths = [ "/var/cache/wals" ];
       };
 
       testScript = ''

nixos/tests/postgresql.nix

@@ -126,6 +126,8 @@ let
       with subtest("Initdb works"):
           machine.succeed("sudo -u postgres initdb -D /tmp/testpostgres2")
 
+      machine.log(machine.execute("systemd-analyze security postgresql.service | grep -v ✓")[1])
+
       machine.shutdown()
     '';