Merge branch 'CNT-4285/add-runtime-hook-path' into 'main'

Add nvidia-contianer-runtime-hook.path config option

See merge request nvidia/container-toolkit/container-toolkit!401

internal/config/config.go

@@ -21,13 +21,19 @@ import (
 	"io"
 	"os"
 	"path"
+	"path/filepath"
 
+	"github.com/NVIDIA/nvidia-container-toolkit/internal/lookup"
 	"github.com/pelletier/go-toml"
+	"github.com/sirupsen/logrus"
 )
 
 const (
 	configOverride = "XDG_CONFIG_HOME"
 	configFilePath = "nvidia-container-runtime/config.toml"
+
+	nvidiaContainerRuntimeHookExecutable  = "nvidia-container-runtime-hook"
+	nvidiaContainerRuntimeHookDefaultPath = "/usr/bin/nvidia-container-runtime-hook"
 )
 
 var (
@@ -124,3 +130,41 @@ func getDefaultConfig() *Config {
 
 	return &c
 }
+
+// ResolveNVIDIAContainerRuntimeHookPath resolves the path the nvidia-container-runtime-hook binary.
+func ResolveNVIDIAContainerRuntimeHookPath(logger *logrus.Logger, nvidiaContainerRuntimeHookPath string) string {
+	return resolveWithDefault(
+		logger,
+		"NVIDIA Container Runtime Hook",
+		nvidiaContainerRuntimeHookPath,
+		nvidiaContainerRuntimeHookDefaultPath,
+	)
+}
+
+// resolveWithDefault resolves the path to the specified binary.
+// If an absolute path is specified, it is used directly without searching for the binary.
+// If the binary cannot be found in the path, the specified default is used instead.
+func resolveWithDefault(logger *logrus.Logger, label string, path string, defaultPath string) string {
+	if filepath.IsAbs(path) {
+		logger.Debugf("Using specified %v path %v", label, path)
+		return path
+	}
+
+	if path == "" {
+		path = filepath.Base(defaultPath)
+	}
+	logger.Debugf("Locating %v as %v", label, path)
+	lookup := lookup.NewExecutableLocator(logger, "")
+
+	resolvedPath := defaultPath
+	targets, err := lookup.Locate(path)
+	if err != nil {
+		logger.Warnf("Failed to locate %v: %v", path, err)
+	} else {
+		logger.Debugf("Found %v candidates: %v", path, targets)
+		resolvedPath = targets[0]
+	}
+	logger.Debugf("Using %v path %v", label, path)
+
+	return resolvedPath
+}

internal/config/config_test.go

@@ -76,6 +76,9 @@ func TestGetConfig(t *testing.T) {
 						},
 					},
 				},
+				NVIDIAContainerRuntimeHookConfig: RuntimeHookConfig{
+					Path: "nvidia-container-runtime-hook",
+				},
 				NVIDIACTKConfig: CTKConfig{
 					Path: "nvidia-ctk",
 				},
@@ -95,6 +98,7 @@ func TestGetConfig(t *testing.T) {
 				"nvidia-container-runtime.modes.cdi.default-kind = \"example.vendor.com/device\"",
 				"nvidia-container-runtime.modes.cdi.annotation-prefixes = [\"cdi.k8s.io/\", \"example.vendor.com/\",]",
 				"nvidia-container-runtime.modes.csv.mount-spec-path = \"/not/etc/nvidia-container-runtime/host-files-for-container.d\"",
+				"nvidia-container-runtime-hook.path = \"/foo/bar/nvidia-container-runtime-hook\"",
 				"nvidia-ctk.path = \"/foo/bar/nvidia-ctk\"",
 			},
 			expectedConfig: &Config{
@@ -120,6 +124,9 @@ func TestGetConfig(t *testing.T) {
 						},
 					},
 				},
+				NVIDIAContainerRuntimeHookConfig: RuntimeHookConfig{
+					Path: "/foo/bar/nvidia-container-runtime-hook",
+				},
 				NVIDIACTKConfig: CTKConfig{
 					Path: "/foo/bar/nvidia-ctk",
 				},
@@ -143,6 +150,8 @@ func TestGetConfig(t *testing.T) {
 				"annotation-prefixes = [\"cdi.k8s.io/\", \"example.vendor.com/\",]",
 				"[nvidia-container-runtime.modes.csv]",
 				"mount-spec-path = \"/not/etc/nvidia-container-runtime/host-files-for-container.d\"",
+				"[nvidia-container-runtime-hook]",
+				"path = \"/foo/bar/nvidia-container-runtime-hook\"",
 				"[nvidia-ctk]",
 				"path = \"/foo/bar/nvidia-ctk\"",
 			},
@@ -169,6 +178,9 @@ func TestGetConfig(t *testing.T) {
 						},
 					},
 				},
+				NVIDIAContainerRuntimeHookConfig: RuntimeHookConfig{
+					Path: "/foo/bar/nvidia-container-runtime-hook",
+				},
 				NVIDIACTKConfig: CTKConfig{
 					Path: "/foo/bar/nvidia-ctk",
 				},

internal/config/hook.go

@@ -24,6 +24,9 @@ import (
 
 // RuntimeHookConfig stores the config options for the NVIDIA Container Runtime
 type RuntimeHookConfig struct {
+	// Path specifies the path to the NVIDIA Container Runtime hook binary.
+	// If an executable name is specified, this will be resolved in the path.
+	Path string `toml:"path"`
 	// SkipModeDetection disables the mode check for the runtime hook.
 	SkipModeDetection bool `toml:"skip-mode-detection"`
 }
@@ -55,6 +58,7 @@ func getRuntimeHookConfigFrom(toml *toml.Tree) (*RuntimeHookConfig, error) {
 // GetDefaultRuntimeHookConfig defines the default values for the config
 func GetDefaultRuntimeHookConfig() *RuntimeHookConfig {
 	c := RuntimeHookConfig{
+		Path:              NVIDIAContainerRuntimeHookExecutable,
 		SkipModeDetection: false,
 	}
 

internal/modifier/stable.go

@@ -17,10 +17,8 @@
 package modifier
 
 import (
-	"fmt"
+	"path/filepath"
 
-	"github.com/NVIDIA/nvidia-container-toolkit/internal/config"
-	"github.com/NVIDIA/nvidia-container-toolkit/internal/lookup"
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/oci"
 	"github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/sirupsen/logrus"
@@ -28,8 +26,11 @@ import (
 
 // NewStableRuntimeModifier creates an OCI spec modifier that inserts the NVIDIA Container Runtime Hook into an OCI
 // spec. The specified logger is used to capture log output.
-func NewStableRuntimeModifier(logger *logrus.Logger) oci.SpecModifier {
-	m := stableRuntimeModifier{logger: logger}
+func NewStableRuntimeModifier(logger *logrus.Logger, nvidiaContainerRuntimeHookPath string) oci.SpecModifier {
+	m := stableRuntimeModifier{
+		logger:                         logger,
+		nvidiaContainerRuntimeHookPath: nvidiaContainerRuntimeHookPath,
+	}
 
 	return &m
 }
@@ -37,7 +38,8 @@ func NewStableRuntimeModifier(logger *logrus.Logger) oci.SpecModifier {
 // stableRuntimeModifier modifies an OCI spec inplace, inserting the nvidia-container-runtime-hook as a
 // prestart hook. If the hook is already present, no modification is made.
 type stableRuntimeModifier struct {
-	logger *logrus.Logger
+	logger                         *logrus.Logger
+	nvidiaContainerRuntimeHookPath string
 }
 
 // Modify applies the required modification to the incoming OCI spec, inserting the nvidia-container-runtime-hook
@@ -53,18 +55,9 @@ func (m stableRuntimeModifier) Modify(spec *specs.Spec) error {
 		}
 	}
 
-	// We create a locator and look for the NVIDIA Container Runtime Hook in the path.
-	candidates, err := lookup.NewExecutableLocator(m.logger, "").Locate(config.NVIDIAContainerRuntimeHookExecutable)
-	if err != nil {
-		return fmt.Errorf("failed to locate NVIDIA Container Runtime Hook: %v", err)
-	}
-	path := candidates[0]
-	if len(candidates) > 1 {
-		m.logger.Debugf("Using %v from multiple NVIDIA Container Runtime Hook candidates: %v", path, candidates)
-	}
-
+	path := m.nvidiaContainerRuntimeHookPath
 	m.logger.Infof("Using prestart hook path: %v", path)
-	args := []string{path}
+	args := []string{filepath.Base(path)}
 	if spec.Hooks == nil {
 		spec.Hooks = &specs.Hooks{}
 	}

internal/modifier/stable_test.go

@@ -79,7 +79,7 @@ func TestAddHookModifier(t *testing.T) {
 					Prestart: []specs.Hook{
 						{
 							Path: testHookPath,
-							Args: []string{testHookPath, "prestart"},
+							Args: []string{"nvidia-container-runtime-hook", "prestart"},
 						},
 					},
 				},
@@ -95,7 +95,7 @@ func TestAddHookModifier(t *testing.T) {
 					Prestart: []specs.Hook{
 						{
 							Path: testHookPath,
-							Args: []string{testHookPath, "prestart"},
+							Args: []string{"nvidia-container-runtime-hook", "prestart"},
 						},
 					},
 				},
@@ -141,7 +141,7 @@ func TestAddHookModifier(t *testing.T) {
 						},
 						{
 							Path: testHookPath,
-							Args: []string{testHookPath, "prestart"},
+							Args: []string{"nvidia-container-runtime-hook", "prestart"},
 						},
 					},
 				},
@@ -154,7 +154,7 @@ func TestAddHookModifier(t *testing.T) {
 
 		t.Run(tc.description, func(t *testing.T) {
 
-			m := NewStableRuntimeModifier(logger)
+			m := NewStableRuntimeModifier(logger, testHookPath)
 
 			err := m.Modify(&tc.spec)
 			if tc.expectedError != nil {

internal/runtime/runtime.go

@@ -63,6 +63,10 @@ func (r rt) Run(argv []string) (rerr error) {
 		r.logger.Reset()
 	}()
 
+	// We apply some config updates here to ensure that the config is valid in
+	// all cases.
+	cfg.NVIDIAContainerRuntimeHookConfig.Path = config.ResolveNVIDIAContainerRuntimeHookPath(r.logger.Logger, cfg.NVIDIAContainerRuntimeHookConfig.Path)
+
 	// Print the config to the output.
 	configJSON, err := json.MarshalIndent(cfg, "", "  ")
 	if err == nil {

internal/runtime/runtime_factory.go

@@ -99,7 +99,7 @@ func newSpecModifier(logger *logrus.Logger, cfg *config.Config, ociSpec oci.Spec
 func newModeModifier(logger *logrus.Logger, cfg *config.Config, ociSpec oci.Spec, argv []string) (oci.SpecModifier, error) {
 	switch info.ResolveAutoMode(logger, cfg.NVIDIAContainerRuntimeConfig.Mode) {
 	case "legacy":
-		return modifier.NewStableRuntimeModifier(logger), nil
+		return modifier.NewStableRuntimeModifier(logger, cfg.NVIDIAContainerRuntimeHookConfig.Path), nil
 	case "csv":
 		return modifier.NewCSVModifier(logger, cfg, ociSpec)
 	case "cdi":