Merge pull request #802 from elezar/bump-release-v1.17.2

Bump release v1.17.2

CHANGELOG.md

@@ -1,5 +1,16 @@
 # NVIDIA Container Toolkit Changelog
 
+## v1.17.2
+- Fixed a bug where legacy images would set imex channels as `all`.
+
+## v1.17.1
+- Fixed a bug where specific symlinks existing in a container image could cause a container to fail to start.
+- Fixed a bug on Tegra-based systems where a container would fail to start.
+- Fixed a bug where the default container runtime config path was not properly set.
+
+### Changes in the Toolkit Container
+- Fallback to using a config file if the current runtime config can not be determined from the command line.
+
 ## v1.17.0
 - Promote v1.17.0-rc.2 to v1.17.0
 - Fix bug when using just-in-time CDI spec generation

cmd/nvidia-ctk/runtime/configure/configure.go

@@ -225,6 +225,17 @@ func (m command) validateFlags(c *cli.Context, config *config) error {
 		return fmt.Errorf("unrecognized Config Source: %v", config.configSource)
 	}
 
+	if config.configFilePath == "" {
+		switch config.runtime {
+		case "containerd":
+			config.configFilePath = defaultContainerdConfigFilePath
+		case "crio":
+			config.configFilePath = defaultCrioConfigFilePath
+		case "docker":
+			config.configFilePath = defaultDockerConfigFilePath
+		}
+	}
+
 	return nil
 }
 
@@ -241,9 +252,6 @@ func (m command) configureWrapper(c *cli.Context, config *config) error {
 
 // configureConfigFile updates the specified container engine config file to enable the NVIDIA runtime.
 func (m command) configureConfigFile(c *cli.Context, config *config) error {
-	configFilePath := config.resolveConfigFilePath()
-
-	var err error
 	configSource, err := config.resolveConfigSource()
 	if err != nil {
 		return err
@@ -254,19 +262,19 @@ func (m command) configureConfigFile(c *cli.Context, config *config) error {
 	case "containerd":
 		cfg, err = containerd.New(
 			containerd.WithLogger(m.logger),
-			containerd.WithPath(configFilePath),
+			containerd.WithPath(config.configFilePath),
 			containerd.WithConfigSource(configSource),
 		)
 	case "crio":
 		cfg, err = crio.New(
 			crio.WithLogger(m.logger),
-			crio.WithPath(configFilePath),
+			crio.WithPath(config.configFilePath),
 			crio.WithConfigSource(configSource),
 		)
 	case "docker":
 		cfg, err = docker.New(
 			docker.WithLogger(m.logger),
-			docker.WithPath(configFilePath),
+			docker.WithPath(config.configFilePath),
 		)
 	default:
 		err = fmt.Errorf("unrecognized runtime '%v'", config.runtime)
@@ -307,22 +315,6 @@ func (m command) configureConfigFile(c *cli.Context, config *config) error {
 	return nil
 }
 
-// resolveConfigFilePath returns the default config file path for the configured container engine
-func (c *config) resolveConfigFilePath() string {
-	if c.configFilePath != "" {
-		return c.configFilePath
-	}
-	switch c.runtime {
-	case "containerd":
-		return defaultContainerdConfigFilePath
-	case "crio":
-		return defaultCrioConfigFilePath
-	case "docker":
-		return defaultDockerConfigFilePath
-	}
-	return ""
-}
-
 // resolveConfigSource returns the default config source or the user provided config source
 func (c *config) resolveConfigSource() (toml.Loader, error) {
 	switch c.configSource {
@@ -351,7 +343,7 @@ func (c *config) getOutputConfigPath() string {
 	if c.dryRun {
 		return ""
 	}
-	return c.resolveConfigFilePath()
+	return c.configFilePath
 }
 
 // configureOCIHook creates and configures the OCI hook for the NVIDIA runtime

internal/config/image/cuda_image.go

@@ -292,7 +292,11 @@ func (i CUDA) CDIDevicesFromMounts() []string {
 
 // ImexChannelsFromEnvVar returns the list of IMEX channels requested for the image.
 func (i CUDA) ImexChannelsFromEnvVar() []string {
-	return i.DevicesFromEnvvars(EnvVarNvidiaImexChannels).List()
+	imexChannels := i.DevicesFromEnvvars(EnvVarNvidiaImexChannels).List()
+	if len(imexChannels) == 1 && imexChannels[0] == "all" {
+		return nil
+	}
+	return imexChannels
 }
 
 // ImexChannelsFromMounts returns the list of IMEX channels requested for the image.

internal/config/image/cuda_image_test.go

@@ -203,6 +203,37 @@ func TestGetVisibleDevicesFromMounts(t *testing.T) {
 	}
 }
 
+func TestImexChannelsFromEnvVar(t *testing.T) {
+	testCases := []struct {
+		description string
+		env         []string
+		expected    []string
+	}{
+		{
+			description: "no imex channels specified",
+		},
+		{
+			description: "imex channel specified",
+			env: []string{
+				"NVIDIA_IMEX_CHANNELS=3,4",
+			},
+			expected: []string{"3", "4"},
+		},
+	}
+
+	for _, tc := range testCases {
+		for id, baseEnvvars := range map[string][]string{"": nil, "legacy": {"CUDA_VERSION=1.2.3"}} {
+			t.Run(tc.description+id, func(t *testing.T) {
+				i, err := NewCUDAImageFromEnv(append(baseEnvvars, tc.env...))
+				require.NoError(t, err)
+
+				channels := i.ImexChannelsFromEnvVar()
+				require.EqualValues(t, tc.expected, channels)
+			})
+		}
+	}
+}
+
 func makeTestMounts(paths ...string) []specs.Mount {
 	var mounts []specs.Mount
 	for _, path := range paths {

tools/container/runtime/containerd/containerd.go

@@ -25,6 +25,7 @@ import (
 
 	"github.com/NVIDIA/nvidia-container-toolkit/pkg/config/engine"
 	"github.com/NVIDIA/nvidia-container-toolkit/pkg/config/engine/containerd"
+	"github.com/NVIDIA/nvidia-container-toolkit/pkg/config/toml"
 	"github.com/NVIDIA/nvidia-container-toolkit/tools/container"
 )
 
@@ -84,13 +85,7 @@ func Flags(opts *Options) []cli.Flag {
 func Setup(c *cli.Context, o *container.Options, co *Options) error {
 	log.Infof("Starting 'setup' for %v", c.App.Name)
 
-	cfg, err := containerd.New(
-		containerd.WithPath(o.Config),
-		containerd.WithConfigSource(containerd.CommandLineSource(o.HostRootMount)),
-		containerd.WithRuntimeType(co.runtimeType),
-		containerd.WithUseLegacyConfig(co.useLegacyConfig),
-		containerd.WithContainerAnnotations(co.containerAnnotationsFromCDIPrefixes()...),
-	)
+	cfg, err := getRuntimeConfig(o, co)
 	if err != nil {
 		return fmt.Errorf("unable to load config: %v", err)
 	}
@@ -114,13 +109,7 @@ func Setup(c *cli.Context, o *container.Options, co *Options) error {
 func Cleanup(c *cli.Context, o *container.Options, co *Options) error {
 	log.Infof("Starting 'cleanup' for %v", c.App.Name)
 
-	cfg, err := containerd.New(
-		containerd.WithPath(o.Config),
-		containerd.WithConfigSource(containerd.CommandLineSource(o.HostRootMount)),
-		containerd.WithRuntimeType(co.runtimeType),
-		containerd.WithUseLegacyConfig(co.useLegacyConfig),
-		containerd.WithContainerAnnotations(co.containerAnnotationsFromCDIPrefixes()...),
-	)
+	cfg, err := getRuntimeConfig(o, co)
 	if err != nil {
 		return fmt.Errorf("unable to load config: %v", err)
 	}
@@ -169,13 +158,24 @@ func (o *Options) runtimeConfigOverride() (map[string]interface{}, error) {
 }
 
 func GetLowlevelRuntimePaths(o *container.Options, co *Options) ([]string, error) {
-	cfg, err := containerd.New(
-		containerd.WithConfigSource(containerd.CommandLineSource(o.HostRootMount)),
-		containerd.WithRuntimeType(co.runtimeType),
-		containerd.WithUseLegacyConfig(co.useLegacyConfig),
-	)
+	cfg, err := getRuntimeConfig(o, co)
 	if err != nil {
 		return nil, fmt.Errorf("unable to load containerd config: %w", err)
 	}
 	return engine.GetBinaryPathsForRuntimes(cfg), nil
 }
+
+func getRuntimeConfig(o *container.Options, co *Options) (engine.Interface, error) {
+	return containerd.New(
+		containerd.WithPath(o.Config),
+		containerd.WithConfigSource(
+			toml.LoadFirst(
+				containerd.CommandLineSource(o.HostRootMount),
+				toml.FromFile(o.Config),
+			),
+		),
+		containerd.WithRuntimeType(co.runtimeType),
+		containerd.WithUseLegacyConfig(co.useLegacyConfig),
+		containerd.WithContainerAnnotations(co.containerAnnotationsFromCDIPrefixes()...),
+	)
+}

tools/container/runtime/crio/crio.go

@@ -28,6 +28,7 @@ import (
 	"github.com/NVIDIA/nvidia-container-toolkit/pkg/config/engine"
 	"github.com/NVIDIA/nvidia-container-toolkit/pkg/config/engine/crio"
 	"github.com/NVIDIA/nvidia-container-toolkit/pkg/config/ocihook"
+	"github.com/NVIDIA/nvidia-container-toolkit/pkg/config/toml"
 	"github.com/NVIDIA/nvidia-container-toolkit/tools/container"
 )
 
@@ -116,10 +117,7 @@ func setupHook(o *container.Options, co *Options) error {
 func setupConfig(o *container.Options) error {
 	log.Infof("Updating config file")
 
-	cfg, err := crio.New(
-		crio.WithPath(o.Config),
-		crio.WithConfigSource(crio.CommandLineSource(o.HostRootMount)),
-	)
+	cfg, err := getRuntimeConfig(o)
 	if err != nil {
 		return fmt.Errorf("unable to load config: %v", err)
 	}
@@ -168,10 +166,7 @@ func cleanupHook(co *Options) error {
 func cleanupConfig(o *container.Options) error {
 	log.Infof("Reverting config file modifications")
 
-	cfg, err := crio.New(
-		crio.WithPath(o.Config),
-		crio.WithConfigSource(crio.CommandLineSource(o.HostRootMount)),
-	)
+	cfg, err := getRuntimeConfig(o)
 	if err != nil {
 		return fmt.Errorf("unable to load config: %v", err)
 	}
@@ -195,11 +190,21 @@ func RestartCrio(o *container.Options) error {
 }
 
 func GetLowlevelRuntimePaths(o *container.Options) ([]string, error) {
-	cfg, err := crio.New(
-		crio.WithConfigSource(crio.CommandLineSource(o.HostRootMount)),
-	)
+	cfg, err := getRuntimeConfig(o)
 	if err != nil {
 		return nil, fmt.Errorf("unable to load crio config: %w", err)
 	}
 	return engine.GetBinaryPathsForRuntimes(cfg), nil
 }
+
+func getRuntimeConfig(o *container.Options) (engine.Interface, error) {
+	return crio.New(
+		crio.WithPath(o.Config),
+		crio.WithConfigSource(
+			toml.LoadFirst(
+				crio.CommandLineSource(o.HostRootMount),
+				toml.FromFile(o.Config),
+			),
+		),
+	)
+}

tools/container/runtime/docker/docker.go

@@ -45,9 +45,7 @@ func Flags(opts *Options) []cli.Flag {
 func Setup(c *cli.Context, o *container.Options) error {
 	log.Infof("Starting 'setup' for %v", c.App.Name)
 
-	cfg, err := docker.New(
-		docker.WithPath(o.Config),
-	)
+	cfg, err := getRuntimeConfig(o)
 	if err != nil {
 		return fmt.Errorf("unable to load config: %v", err)
 	}
@@ -71,9 +69,7 @@ func Setup(c *cli.Context, o *container.Options) error {
 func Cleanup(c *cli.Context, o *container.Options) error {
 	log.Infof("Starting 'cleanup' for %v", c.App.Name)
 
-	cfg, err := docker.New(
-		docker.WithPath(o.Config),
-	)
+	cfg, err := getRuntimeConfig(o)
 	if err != nil {
 		return fmt.Errorf("unable to load config: %v", err)
 	}
@@ -107,3 +103,9 @@ func GetLowlevelRuntimePaths(o *container.Options) ([]string, error) {
 	}
 	return engine.GetBinaryPathsForRuntimes(cfg), nil
 }
+
+func getRuntimeConfig(o *container.Options) (engine.Interface, error) {
+	return docker.New(
+		docker.WithPath(o.Config),
+	)
+}

versions.mk

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 LIB_NAME := nvidia-container-toolkit
-LIB_VERSION := 1.17.0
+LIB_VERSION := 1.17.2
 LIB_TAG :=
 
 # The package version is the combination of the library version and tag.