Add authenticated host capabilities endpoint

cmd/api/api/capabilities.go

@@ -0,0 +1,224 @@
+package api
+
+import (
+	"context"
+	"runtime"
+	"sync"
+
+	"github.com/kernel/hypeman/lib/hypervisor"
+	"github.com/kernel/hypeman/lib/images"
+	"github.com/kernel/hypeman/lib/logger"
+	"github.com/kernel/hypeman/lib/network"
+	"github.com/kernel/hypeman/lib/oapi"
+)
+
+// Server-level feature IDs: API surfaces this server exposes regardless of
+// which runtime backs an instance. Per-runtime feature IDs are owned by
+// lib/hypervisor (hypervisor.Capabilities.FeatureIDs) so adding a runtime
+// capability never requires touching this handler.
+const (
+	featureInstances        = "instances"
+	featureImages           = "images"
+	featureBuilds           = "builds"
+	featureVolumes          = "volumes"
+	featureIngress          = "ingress"
+	featureExec             = "exec"
+	featureLogs             = "logs"
+	featureDevices          = "devices"
+	featureRosettaEmulation = "rosetta-emulation"
+)
+
+// apiVersion is the API contract version from the embedded OpenAPI document.
+// The decoded spec is cached: decoding it per request is needlessly expensive.
+var apiVersion = sync.OnceValue(func() string {
+	spec, err := oapi.GetSwagger()
+	if err != nil || spec.Info == nil {
+		return "unknown"
+	}
+	return spec.Info.Version
+})
+
+// defaultHypervisorProvider is the narrow accessor this handler needs from
+// the instance manager. The concrete instances manager implements it; it is
+// type-asserted rather than added to instances.Manager so alternate Manager
+// implementations (mocks, wrappers) compiled against the public module keep
+// building without a new method.
+type defaultHypervisorProvider interface {
+	DefaultHypervisor() hypervisor.Type
+}
+
+// GetCapabilities reports host, runtime, network, and image capabilities.
+func (s *ApiService) GetCapabilities(ctx context.Context, _ oapi.GetCapabilitiesRequestObject) (oapi.GetCapabilitiesResponseObject, error) {
+	log := logger.FromContext(ctx)
+
+	// Resolve the default runtime the way launches do. Prefer the manager's
+	// own effective default via the optional accessor; when the manager does
+	// not expose it (a wrapper embedding instances.Manager hides the concrete
+	// manager's extra method), fall back to the configured default the manager
+	// was constructed from — launches still route through the wrapped manager,
+	// so a hardcoded fallback would misreport a Firecracker/QEMU default as
+	// cloud-hypervisor. Only an empty (unconfigured) value normalizes to the
+	// compile-time default, mirroring lib/instances.NewManagerWithConfigE.
+	defaultRuntime := hypervisor.Type(s.Config.Hypervisor.Default)
+	if defaultRuntime == "" {
+		defaultRuntime = hypervisor.TypeCloudHypervisor
+	}
+	if p, ok := s.InstanceManager.(defaultHypervisorProvider); ok {
+		defaultRuntime = p.DefaultHypervisor()
+	}
+
+	// The capability registry is platform-gated at registration time, so its
+	// contents are exactly the runtimes this build supports on this host —
+	// including ones added after this handler was written. Capabilities and
+	// launch prerequisites are resolved per request, so configuration applied
+	// after init (e.g. a pinned cloud-hypervisor version) and host state
+	// (e.g. an installed QEMU binary) are reflected without a restart.
+	registered := hypervisor.RegisteredRuntimes()
+	runtimes := make([]oapi.CapabilitiesRuntime, 0, len(registered))
+	defaultAvailable := false
+	for _, rt := range registered {
+		available := rt.Available()
+		if rt.Type == defaultRuntime {
+			defaultAvailable = available
+		}
+		if !available {
+			log.WarnContext(ctx, "runtime is registered but missing launch prerequisites",
+				"runtime", string(rt.Type), "error", rt.LaunchErr)
+		}
+		runtimes = append(runtimes, oapi.CapabilitiesRuntime{
+			Name:      string(rt.Type),
+			Available: available,
+			Features:  rt.Capabilities.FeatureIDs(),
+		})
+	}
+	if !defaultAvailable {
+		// Ordinary launches use the default runtime and will fail on this
+		// host; surface that in logs as well as in the response.
+		log.WarnContext(ctx, "configured default runtime is not available on this host",
+			"runtime", string(defaultRuntime), "host_os", runtime.GOOS, "host_arch", runtime.GOARCH)
+	}
+
+	networkCaps, err := s.networkCapabilities(ctx)
+	if err != nil {
+		log.ErrorContext(ctx, "failed to resolve network capabilities", "error", err)
+		return oapi.GetCapabilities500JSONResponse{
+			Code:    "internal_error",
+			Message: "failed to resolve network capabilities",
+		}, nil
+	}
+
+	emulation := emulationAvailable(runtime.GOOS, runtime.GOARCH, rosettaInstalled())
+
+	resp := oapi.Capabilities{
+		Server: oapi.CapabilitiesServer{
+			Version:    s.Config.Version,
+			ApiVersion: apiVersion(),
+		},
+		Host: oapi.CapabilitiesHost{
+			Os:   runtime.GOOS,
+			Arch: runtime.GOARCH,
+		},
+		DefaultRuntime: oapi.CapabilitiesDefaultRuntime{
+			Name:      string(defaultRuntime),
+			Available: defaultAvailable,
+		},
+		Runtimes: runtimes,
+		Network:  *networkCaps,
+		Images: oapi.CapabilitiesImages{
+			Platforms:       imagePlatforms(runtime.GOARCH, emulation),
+			DefaultPlatform: images.HostPlatformString(),
+		},
+		Features: serverFeatures(runtime.GOOS, emulation),
+	}
+
+	return oapi.GetCapabilities200JSONResponse(resp), nil
+}
+
+// networkCapabilities resolves the guest networking model and the
+// guest-visible host gateway from the network manager's effective default
+// network. Gateway and subnet are omitted (not serialized as empty strings)
+// when no default network has been resolved.
+func (s *ApiService) networkCapabilities(ctx context.Context) (*oapi.CapabilitiesNetwork, error) {
+	caps := &oapi.CapabilitiesNetwork{
+		Model:        oapiNetworkModel(network.NetworkModel()),
+		GuestToGuest: false,
+	}
+	if s.NetworkManager == nil {
+		return caps, nil
+	}
+	nw, err := s.NetworkManager.EffectiveDefaultNetwork()
+	if err != nil {
+		return nil, err
+	}
+	if nw == nil {
+		return caps, nil
+	}
+	if nw.Gateway != "" {
+		gateway := nw.Gateway
+		caps.Gateway = &gateway
+	}
+	if nw.Subnet != "" {
+		subnet := nw.Subnet
+		caps.Subnet = &subnet
+	}
+	caps.GuestToGuest = network.GuestToGuestEnabled(nw)
+	return caps, nil
+}
+
+// oapiNetworkModel maps the network package's typed model onto the API enum,
+// keeping lib/network free of oapi dependencies.
+func oapiNetworkModel(m network.Model) oapi.CapabilitiesNetworkModel {
+	switch m {
+	case network.ModelBridge:
+		return oapi.Bridge
+	case network.ModelNAT:
+		return oapi.Nat
+	}
+	// A new network.Model must also be added to the OpenAPI enum; surface it
+	// verbatim rather than misreporting it as a known model.
+	return oapi.CapabilitiesNetworkModel(m)
+}
+
+// emulationAvailable reports whether the host can boot images built for the
+// other CPU architecture right now. Only Apple Silicon macOS hosts qualify
+// (vz with Rosetta), and only when the Rosetta availability probe — the same
+// Virtualization.framework check the vz-shim enforces at launch — reports it
+// installed. Platform eligibility alone (darwin/arm64) is deliberately not
+// enough: a macOS 11/12 host or one without Rosetta installed would advertise
+// launches that lib/hypervisor/vz rejects.
+func emulationAvailable(goos, goarch string, rosettaInstalled bool) bool {
+	return goos == "darwin" && goarch == "arm64" && rosettaInstalled
+}
+
+// imagePlatforms returns the image platforms (os/arch) the host can run: the
+// host-native Linux guest platform, plus Rosetta-emulated linux/amd64 on
+// Apple Silicon with Rosetta installed.
+func imagePlatforms(goarch string, emulation bool) []string {
+	platforms := []string{"linux/" + goarch}
+	if emulation {
+		platforms = append(platforms, "linux/amd64")
+	}
+	return platforms
+}
+
+// serverFeatures builds the server-level feature ID list: API surfaces that
+// are always present plus host-platform conditionals.
+func serverFeatures(goos string, emulation bool) []string {
+	features := []string{
+		featureInstances,
+		featureImages,
+		featureBuilds,
+		featureVolumes,
+		featureIngress,
+		featureExec,
+		featureLogs,
+	}
+	// Device (GPU/PCI) passthrough management is only available on Linux hosts.
+	if goos == "linux" {
+		features = append(features, featureDevices)
+	}
+	if emulation {
+		features = append(features, featureRosettaEmulation)
+	}
+	return features
+}

cmd/api/api/capabilities_darwin_test.go

@@ -0,0 +1,57 @@
+//go:build darwin
+
+package api
+
+import (
+	"runtime"
+	"slices"
+	"testing"
+
+	"github.com/kernel/hypeman/lib/hypervisor"
+	"github.com/stretchr/testify/require"
+)
+
+// TestGetCapabilitiesRosettaTracksProbe pins that Rosetta emulation reporting
+// follows the live Virtualization.framework availability probe on this host
+// — the same check the vz-shim enforces at launch — rather than treating
+// every Apple Silicon host as emulation-capable: a host without Rosetta
+// installed (or on macOS < 13) must advertise neither the rosetta-emulation
+// feature nor the linux/amd64 image platform.
+func TestGetCapabilitiesRosettaTracksProbe(t *testing.T) {
+	t.Parallel()
+	if runtime.GOARCH != "arm64" {
+		t.Skipf("rosetta emulation exists only on Apple Silicon (GOARCH=%s)", runtime.GOARCH)
+	}
+	svc := newTestService(t)
+	svc.NetworkManager = &stubCapabilitiesNetworkManager{}
+
+	caps := getCapabilities(t, svc)
+
+	want := rosettaInstalled()
+	require.Equal(t, want, slices.Contains(caps.Features, "rosetta-emulation"),
+		"rosetta-emulation feature must track the launch-path availability probe")
+	require.Equal(t, want, slices.Contains(caps.Images.Platforms, "linux/amd64"),
+		"linux/amd64 image platform must track the launch-path availability probe")
+}
+
+// TestRegisteredRuntimesDarwin pins the macOS registration boundary: only vz
+// can genuinely launch VMs on macOS (the Linux backends require KVM and
+// kernel AF_VSOCK), so the capability registry contains exactly vz even
+// though the cloud-hypervisor and qemu packages are linked into the binary.
+func TestRegisteredRuntimesDarwin(t *testing.T) {
+	t.Parallel()
+
+	registered := hypervisor.RegisteredRuntimes()
+	require.Len(t, registered, 1)
+	require.Equal(t, hypervisor.TypeVZ, registered[0].Type)
+
+	for _, linuxOnly := range []hypervisor.Type{
+		hypervisor.TypeCloudHypervisor,
+		hypervisor.TypeFirecracker,
+		hypervisor.TypeQEMU,
+		hypervisor.TypeQEMUMicroVM,
+	} {
+		_, ok := hypervisor.CapabilitiesForType(linuxOnly)
+		require.False(t, ok, "%s must not register capabilities on macOS", linuxOnly)
+	}
+}

cmd/api/api/capabilities_linux_test.go

@@ -0,0 +1,33 @@
+//go:build linux
+
+package api
+
+import (
+	"runtime"
+	"testing"
+
+	"github.com/kernel/hypeman/lib/hypervisor"
+	"github.com/stretchr/testify/require"
+)
+
+// TestRegisteredRuntimesLinux pins the Linux registration boundary: the
+// capability registry contains exactly the runtimes launchable on this host,
+// in deterministic sorted order. qemu-microvm appears only where its x86
+// board exists, and vz never registers off macOS.
+func TestRegisteredRuntimesLinux(t *testing.T) {
+	t.Parallel()
+
+	expected := []string{"cloud-hypervisor", "firecracker", "qemu"}
+	if runtime.GOARCH == "amd64" {
+		expected = append(expected, "qemu-microvm")
+	}
+
+	names := make([]string, 0, len(expected))
+	for _, rt := range hypervisor.RegisteredRuntimes() {
+		names = append(names, string(rt.Type))
+	}
+	require.Equal(t, expected, names)
+
+	_, vzRegistered := hypervisor.CapabilitiesForType(hypervisor.TypeVZ)
+	require.False(t, vzRegistered, "vz must not register capabilities on Linux")
+}

cmd/api/api/capabilities_rosetta_darwin_arm64.go

@@ -0,0 +1,18 @@
+//go:build darwin && arm64
+
+package api
+
+import "github.com/Code-Hex/vz/v3"
+
+// rosettaInstalled reports whether Rosetta translation for Linux guests is
+// installed and usable right now, using the same Virtualization.framework
+// probe the vz-shim enforces when a launch requests Rosetta
+// (cmd/vz-shim/rosetta_arm64.go): NotInstalled (Rosetta missing) and
+// NotSupported (macOS < 13) both fail launches, so capability reporting must
+// not advertise emulation in either state. Evaluated per request, so
+// installing Rosetta (softwareupdate --install-rosetta) is reflected without
+// a restart. cmd/api already links Virtualization.framework on macOS
+// (checkHypervisorAccess), so this adds no build or runtime requirement.
+var rosettaInstalled = func() bool {
+	return vz.LinuxRosettaDirectoryShareAvailability() == vz.LinuxRosettaAvailabilityInstalled
+}

cmd/api/api/capabilities_rosetta_other.go

@@ -0,0 +1,8 @@
+//go:build !(darwin && arm64)
+
+package api
+
+// rosettaInstalled is never true off Apple Silicon macOS: Rosetta emulation
+// exists only under vz. Kept as a var with the same shape as the
+// darwin/arm64 probe so the handler code is identical on every platform.
+var rosettaInstalled = func() bool { return false }