nixos/binfmt: Add option to use static emulators when available (#334859)

lib/systems/default.nix

@@ -277,25 +277,6 @@ let
       let
         selectEmulator = pkgs:
           let
-            qemu-user = pkgs.qemu.override {
-              smartcardSupport = false;
-              spiceSupport = false;
-              openGLSupport = false;
-              virglSupport = false;
-              vncSupport = false;
-              gtkSupport = false;
-              sdlSupport = false;
-              alsaSupport = false;
-              pulseSupport = false;
-              pipewireSupport = false;
-              jackSupport = false;
-              smbdSupport = false;
-              seccompSupport = false;
-              tpmSupport = false;
-              capstoneSupport = false;
-              enableDocs = false;
-              hostCpuTargets = [ "${final.qemuArch}-linux-user" ];
-            };
             wine = (pkgs.winePackagesFor "wine${toString final.parsed.cpu.bits}").minimal;
           in
           # Note: we guarantee that the return value is either `null` or a path
@@ -306,7 +287,7 @@ let
           else if final.isWindows
           then "${wine}/bin/wine${optionalString (final.parsed.cpu.bits == 64) "64"}"
           else if final.isLinux && pkgs.stdenv.hostPlatform.isLinux && final.qemuArch != null
-          then "${qemu-user}/bin/qemu-${final.qemuArch}"
+          then "${pkgs.qemu-user}/bin/qemu-${final.qemuArch}"
           else if final.isWasi
           then "${pkgs.wasmtime}/bin/wasmtime"
           else if final.isMmix
@@ -315,6 +296,10 @@ let
       in {
         emulatorAvailable = pkgs: (selectEmulator pkgs) != null;
 
+        # whether final.emulator pkgs.pkgsStatic works
+        staticEmulatorAvailable = pkgs: final.emulatorAvailable pkgs
+          && (final.isLinux || final.isWasi || final.isMmix);
+
         emulator = pkgs:
           if (final.emulatorAvailable pkgs)
           then selectEmulator pkgs

lib/tests/systems.nix

@@ -96,6 +96,7 @@ lib.runTests (
         canExecute = null;
         emulator = null;
         emulatorAvailable = null;
+        staticEmulatorAvailable = null;
         isCompatible = null;
       }?${platformAttrName};
     };

nixos/modules/system/boot/binfmt.nix

@@ -28,8 +28,6 @@ let
          ''
     else interpreter;
 
-  getEmulator = system: (lib.systems.elaborate { inherit system; }).emulator pkgs;
-  getQemuArch = system: (lib.systems.elaborate { inherit system; }).qemuArch;
 
   # Mapping of systems to “magicOrExtension” and “mask”. Mostly taken from:
   # - https://github.com/cleverca22/nixos-configs/blob/master/qemu.nix
@@ -280,28 +278,50 @@ in {
         '';
         type = types.listOf (types.enum (builtins.attrNames magics));
       };
+
+      preferStaticEmulators = mkOption {
+        default = false;
+        description = ''
+          Whether to use static emulators when available.
+
+          This enables the kernel to preload the emulator binaries when
+          the binfmt registrations are added, obviating the need to make
+          the emulator binaries available inside chroots and chroot-like
+          sandboxes.
+        '';
+        type = types.bool;
+      };
     };
   };
 
   config = {
+    assertions = lib.mapAttrsToList (name: reg: {
+      assertion = reg.fixBinary -> !reg.wrapInterpreterInShell;
+      message = "boot.binfmt.registrations.\"${name}\" cannot have fixBinary when the interpreter is invoked through a shell.";
+    }) cfg.registrations;
+
     boot.binfmt.registrations = builtins.listToAttrs (map (system: assert system != pkgs.stdenv.hostPlatform.system; {
       name = system;
       value = { config, ... }: let
-        interpreter = getEmulator system;
-        qemuArch = getQemuArch system;
+        elaborated = lib.systems.elaborate { inherit system; };
+        useStaticEmulator = cfg.preferStaticEmulators && elaborated.staticEmulatorAvailable pkgs;
+        interpreter = elaborated.emulator (if useStaticEmulator then pkgs.pkgsStatic else pkgs);
+
+        inherit (elaborated) qemuArch;
+        isQemu = "qemu-${qemuArch}" == baseNameOf interpreter;
 
-        preserveArgvZero = "qemu-${qemuArch}" == baseNameOf interpreter;
         interpreterReg = let
           wrapperName = "qemu-${qemuArch}-binfmt-P";
           wrapper = pkgs.wrapQemuBinfmtP wrapperName interpreter;
         in
-          if preserveArgvZero then "${wrapper}/bin/${wrapperName}"
+          if isQemu && !useStaticEmulator then "${wrapper}/bin/${wrapperName}"
           else interpreter;
       in ({
-        preserveArgvZero = mkDefault preserveArgvZero;
+        preserveArgvZero = mkDefault isQemu;
 
         interpreter = mkDefault interpreterReg;
-        wrapInterpreterInShell = mkDefault (!config.preserveArgvZero);
+        fixBinary = mkDefault useStaticEmulator;
+        wrapInterpreterInShell = mkDefault (!config.preserveArgvZero && !config.fixBinary);
         interpreterSandboxPath = mkDefault (dirOf (dirOf config.interpreter));
       } // (magics.${system} or (throw "Cannot create binfmt registration for system ${system}")));
     }) cfg.emulatedSystems);

nixos/tests/systemd-binfmt.nix

@@ -87,4 +87,29 @@ in {
       ).lower()
     '';
   };
+
+  chroot = makeTest {
+    name = "systemd-binfmt-chroot";
+    nodes.machine = { pkgs, lib, ... }: {
+      boot.binfmt.emulatedSystems = [
+        "aarch64-linux" "wasm32-wasi"
+      ];
+      boot.binfmt.preferStaticEmulators = true;
+
+      environment.systemPackages = [
+        (pkgs.writeShellScriptBin "test-chroot" ''
+          set -euo pipefail
+          mkdir -p /tmp/chroot
+          cp ${lib.getExe' pkgs.pkgsCross.aarch64-multiplatform.pkgsStatic.busybox "busybox"} /tmp/chroot/busybox
+          cp ${lib.getExe pkgs.pkgsCross.wasi32.yaml2json} /tmp/chroot/yaml2json # wasi binaries that build are hard to come by
+          chroot /tmp/chroot /busybox uname -m | grep aarch64
+          echo 42 | chroot /tmp/chroot /yaml2json | grep 42
+        '')
+      ];
+    };
+    testScript = ''
+      machine.start()
+      machine.succeed("test-chroot")
+    '';
+  };
 }

pkgs/applications/virtualization/qemu/default.nix

@@ -94,10 +94,9 @@ stdenv.mkDerivation (finalAttrs: {
     ++ lib.optionals stdenv.hostPlatform.isDarwin [ sigtool ]
     ++ lib.optionals (!userOnly) [ dtc ];
 
-  buildInputs = [ zlib glib pixman
-    vde2 lzo snappy libtasn1
-    gnutls nettle curl libslirp
-  ]
+  buildInputs = [ glib zlib ]
+    ++ lib.optionals (!minimal) [ dtc pixman vde2 lzo snappy libtasn1 gnutls nettle libslirp ]
+    ++ lib.optionals (!userOnly) [ curl ]
     ++ lib.optionals ncursesSupport [ ncurses ]
     ++ lib.optionals stdenv.hostPlatform.isDarwin [ CoreServices Cocoa Hypervisor Kernel rez setfile vmnet ]
     ++ lib.optionals seccompSupport [ libseccomp ]
@@ -112,8 +111,7 @@ stdenv.mkDerivation (finalAttrs: {
     ++ lib.optionals smartcardSupport [ libcacard ]
     ++ lib.optionals spiceSupport [ spice-protocol spice ]
     ++ lib.optionals usbredirSupport [ usbredir ]
-    ++ lib.optionals stdenv.hostPlatform.isLinux [ libcap_ng libcap attr ]
-    ++ lib.optionals (stdenv.hostPlatform.isLinux && !userOnly) [ libaio ]
+    ++ lib.optionals (stdenv.hostPlatform.isLinux && !userOnly) [ libcap_ng libcap attr libaio ]
     ++ lib.optionals xenSupport [ xen ]
     ++ lib.optionals cephSupport [ ceph ]
     ++ lib.optionals glusterfsSupport [ glusterfs libuuid ]
@@ -124,8 +122,7 @@ stdenv.mkDerivation (finalAttrs: {
     ++ lib.optionals smbdSupport [ samba ]
     ++ lib.optionals uringSupport [ liburing ]
     ++ lib.optionals canokeySupport [ canokey-qemu ]
-    ++ lib.optionals capstoneSupport [ capstone ]
-    ++ lib.optionals (!userOnly) [ dtc ];
+    ++ lib.optionals capstoneSupport [ capstone ];
 
   dontUseMesonConfigure = true; # meson's configurePhase isn't compatible with qemu build
   dontAddStaticConfigureFlags = true;