Skip to content

Commit c5a93b8

Browse files
author
Evan Lezar
committed
Merge branch 'cherry-pick-wsl2' into 'release-1.13'
Backport CDI fixes for 1.13.3 release See merge request nvidia/container-toolkit/container-toolkit!429
2 parents 9b7904e + cc06766 commit c5a93b8

File tree

25 files changed

+1203
-241
lines changed

25 files changed

+1203
-241
lines changed

CHANGELOG.md

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,12 @@
22

33
## v1.13.3
44

5+
* Generate CDI specification files with `644` permissions to allow rootless applications (e.g. podman).
6+
* Fix bug causing incorrect nvidia-smi symlink to be created on WSL2 systems with multiple driver roots.
7+
* Fix bug when using driver versions that do not include a patch component in their version number.
8+
* Skip additional modifications in CDI mode.
9+
* Fix loading of kernel modules and creation of device nodes in containerized use cases.
10+
511
* [toolkit-container] Allow same envars for all runtime configs
612

713
## v1.13.2

cmd/nvidia-ctk/cdi/generate/generate.go

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -251,6 +251,7 @@ func (m command) generateSpec(cfg *config) (spec.Interface, error) {
251251
spec.WithDeviceSpecs(deviceSpecs),
252252
spec.WithEdits(*commonEdits.ContainerEdits),
253253
spec.WithFormat(cfg.format),
254+
spec.WithPermissions(0644),
254255
)
255256
}
256257

cmd/nvidia-ctk/system/create-dev-char-symlinks/all.go

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -28,14 +28,14 @@ import (
2828

2929
type allPossible struct {
3030
logger *logrus.Logger
31-
driverRoot string
31+
devRoot string
3232
deviceMajors devices.Devices
3333
migCaps nvcaps.MigCaps
3434
}
3535

3636
// newAllPossible returns a new allPossible device node lister.
3737
// This lister lists all possible device nodes for NVIDIA GPUs, control devices, and capability devices.
38-
func newAllPossible(logger *logrus.Logger, driverRoot string) (nodeLister, error) {
38+
func newAllPossible(logger *logrus.Logger, devRoot string) (nodeLister, error) {
3939
deviceMajors, err := devices.GetNVIDIADevices()
4040
if err != nil {
4141
return nil, fmt.Errorf("failed reading device majors: %v", err)
@@ -61,7 +61,7 @@ func newAllPossible(logger *logrus.Logger, driverRoot string) (nodeLister, error
6161

6262
l := allPossible{
6363
logger: logger,
64-
driverRoot: driverRoot,
64+
devRoot: devRoot,
6565
deviceMajors: deviceMajors,
6666
migCaps: migCaps,
6767
}
@@ -72,15 +72,15 @@ func newAllPossible(logger *logrus.Logger, driverRoot string) (nodeLister, error
7272
// DeviceNodes returns a list of all possible device nodes for NVIDIA GPUs, control devices, and capability devices.
7373
func (m allPossible) DeviceNodes() ([]deviceNode, error) {
7474
gpus, err := nvpci.NewFrom(
75-
filepath.Join(m.driverRoot, nvpci.PCIDevicesRoot),
75+
filepath.Join(m.devRoot, nvpci.PCIDevicesRoot),
7676
).GetGPUs()
7777
if err != nil {
7878
return nil, fmt.Errorf("failed to get GPU information: %v", err)
7979
}
8080

8181
count := len(gpus)
8282
if count == 0 {
83-
m.logger.Infof("No NVIDIA devices found in %s", m.driverRoot)
83+
m.logger.Infof("No NVIDIA devices found in %s", m.devRoot)
8484
return nil, nil
8585
}
8686

@@ -179,7 +179,7 @@ func (m allPossible) newDeviceNode(deviceName devices.Name, path string, minor i
179179
major, _ := m.deviceMajors.Get(deviceName)
180180

181181
return deviceNode{
182-
path: filepath.Join(m.driverRoot, path),
182+
path: filepath.Join(m.devRoot, path),
183183
major: uint32(major),
184184
minor: uint32(minor),
185185
}

cmd/nvidia-ctk/system/create-dev-char-symlinks/create-dev-char-symlinks.go

Lines changed: 31 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -24,7 +24,8 @@ import (
2424
"strings"
2525
"syscall"
2626

27-
"github.com/NVIDIA/nvidia-container-toolkit/internal/system"
27+
"github.com/NVIDIA/nvidia-container-toolkit/internal/system/nvdevices"
28+
"github.com/NVIDIA/nvidia-container-toolkit/internal/system/nvmodules"
2829
"github.com/fsnotify/fsnotify"
2930
"github.com/sirupsen/logrus"
3031
"github.com/urfave/cli/v2"
@@ -216,6 +217,7 @@ type linkCreator struct {
216217
logger *logrus.Logger
217218
lister nodeLister
218219
driverRoot string
220+
devRoot string
219221
devCharPath string
220222
dryRun bool
221223
createAll bool
@@ -243,6 +245,9 @@ func NewSymlinkCreator(opts ...Option) (Creator, error) {
243245
if c.driverRoot == "" {
244246
c.driverRoot = "/"
245247
}
248+
if c.devRoot == "" {
249+
c.devRoot = "/"
250+
}
246251
if c.devCharPath == "" {
247252
c.devCharPath = defaultDevCharPath
248253
}
@@ -252,13 +257,13 @@ func NewSymlinkCreator(opts ...Option) (Creator, error) {
252257
}
253258

254259
if c.createAll {
255-
lister, err := newAllPossible(c.logger, c.driverRoot)
260+
lister, err := newAllPossible(c.logger, c.devRoot)
256261
if err != nil {
257262
return nil, fmt.Errorf("failed to create all possible device lister: %v", err)
258263
}
259264
c.lister = lister
260265
} else {
261-
c.lister = existing{c.logger, c.driverRoot}
266+
c.lister = existing{c.logger, c.devRoot}
262267
}
263268
return c, nil
264269
}
@@ -268,36 +273,48 @@ func (m linkCreator) setup() error {
268273
return nil
269274
}
270275

271-
s, err := system.New(
272-
system.WithLogger(m.logger),
273-
system.WithDryRun(m.dryRun),
274-
)
275-
if err != nil {
276-
return err
277-
}
278-
279276
if m.loadKernelModules {
280-
if err := s.LoadNVIDIAKernelModules(); err != nil {
277+
modules := nvmodules.New(
278+
nvmodules.WithLogger(m.logger),
279+
nvmodules.WithDryRun(m.dryRun),
280+
nvmodules.WithRoot(m.driverRoot),
281+
)
282+
if err := modules.LoadAll(); err != nil {
281283
return fmt.Errorf("failed to load NVIDIA kernel modules: %v", err)
282284
}
283285
}
284286

285287
if m.createDeviceNodes {
286-
if err := s.CreateNVIDIAControlDeviceNodesAt(m.driverRoot); err != nil {
288+
devices, err := nvdevices.New(
289+
nvdevices.WithLogger(m.logger),
290+
nvdevices.WithDryRun(m.dryRun),
291+
nvdevices.WithDevRoot(m.devRoot),
292+
)
293+
if err != nil {
294+
return err
295+
}
296+
if err := devices.CreateNVIDIAControlDevices(); err != nil {
287297
return fmt.Errorf("failed to create NVIDIA device nodes: %v", err)
288298
}
289299
}
290-
291300
return nil
292301
}
293302

294303
// WithDriverRoot sets the driver root path.
304+
// This is the path in which kernel modules must be loaded.
295305
func WithDriverRoot(root string) Option {
296306
return func(c *linkCreator) {
297307
c.driverRoot = root
298308
}
299309
}
300310

311+
// WithDevRoot sets the root path for the /dev directory.
312+
func WithDevRoot(root string) Option {
313+
return func(c *linkCreator) {
314+
c.devRoot = root
315+
}
316+
}
317+
301318
// WithDevCharPath sets the path at which the symlinks will be created.
302319
func WithDevCharPath(path string) Option {
303320
return func(c *linkCreator) {

cmd/nvidia-ctk/system/create-dev-char-symlinks/existing.go

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -30,16 +30,16 @@ type nodeLister interface {
3030
}
3131

3232
type existing struct {
33-
logger *logrus.Logger
34-
driverRoot string
33+
logger *logrus.Logger
34+
devRoot string
3535
}
3636

3737
// DeviceNodes returns a list of NVIDIA device nodes in the specified root.
3838
// The nvidia-nvswitch* and nvidia-nvlink devices are excluded.
3939
func (m existing) DeviceNodes() ([]deviceNode, error) {
4040
locator := lookup.NewCharDeviceLocator(
4141
lookup.WithLogger(m.logger),
42-
lookup.WithRoot(m.driverRoot),
42+
lookup.WithRoot(m.devRoot),
4343
lookup.WithOptional(true),
4444
)
4545

@@ -54,7 +54,7 @@ func (m existing) DeviceNodes() ([]deviceNode, error) {
5454
}
5555

5656
if len(devices) == 0 && len(capDevices) == 0 {
57-
m.logger.Infof("No NVIDIA devices found in %s", m.driverRoot)
57+
m.logger.Infof("No NVIDIA devices found in %s", m.devRoot)
5858
return nil, nil
5959
}
6060

cmd/nvidia-ctk/system/create-device-nodes/create-device-nodes.go

Lines changed: 21 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -19,7 +19,8 @@ package createdevicenodes
1919
import (
2020
"fmt"
2121

22-
"github.com/NVIDIA/nvidia-container-toolkit/internal/system"
22+
"github.com/NVIDIA/nvidia-container-toolkit/internal/system/nvdevices"
23+
"github.com/NVIDIA/nvidia-container-toolkit/internal/system/nvmodules"
2324
"github.com/sirupsen/logrus"
2425
"github.com/urfave/cli/v2"
2526
)
@@ -96,19 +97,29 @@ func (m command) validateFlags(r *cli.Context, opts *options) error {
9697
}
9798

9899
func (m command) run(c *cli.Context, opts *options) error {
99-
s, err := system.New(
100-
system.WithLogger(m.logger),
101-
system.WithDryRun(opts.dryRun),
102-
system.WithLoadKernelModules(opts.loadKernelModules),
103-
)
104-
if err != nil {
105-
return fmt.Errorf("failed to create library: %v", err)
100+
if opts.loadKernelModules {
101+
modules := nvmodules.New(
102+
nvmodules.WithLogger(m.logger),
103+
nvmodules.WithDryRun(opts.dryRun),
104+
nvmodules.WithRoot(opts.driverRoot),
105+
)
106+
if err := modules.LoadAll(); err != nil {
107+
return fmt.Errorf("failed to load NVIDIA kernel modules: %v", err)
108+
}
106109
}
107110

108111
if opts.control {
112+
devices, err := nvdevices.New(
113+
nvdevices.WithLogger(m.logger),
114+
nvdevices.WithDryRun(opts.dryRun),
115+
nvdevices.WithDevRoot(opts.driverRoot),
116+
)
117+
if err != nil {
118+
return err
119+
}
109120
m.logger.Infof("Creating control device nodes at %s", opts.driverRoot)
110-
if err := s.CreateNVIDIAControlDeviceNodesAt(opts.driverRoot); err != nil {
111-
return fmt.Errorf("failed to create control device nodes: %v", err)
121+
if err := devices.CreateNVIDIAControlDevices(); err != nil {
122+
return fmt.Errorf("failed to create NVIDIA control device nodes: %v", err)
112123
}
113124
}
114125
return nil

internal/discover/graphics.go

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -271,7 +271,7 @@ func newXorgDiscoverer(logger *logrus.Logger, driverRoot string, nvidiaCTKPath s
271271
libCudaPaths, err := cuda.New(
272272
cuda.WithLogger(logger),
273273
cuda.WithDriverRoot(driverRoot),
274-
).Locate(".*.*.*")
274+
).Locate(".*.*")
275275
if err != nil {
276276
return nil, fmt.Errorf("failed to locate libcuda.so: %v", err)
277277
}

internal/runtime/runtime_factory.go

Lines changed: 8 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -61,10 +61,15 @@ func newNVIDIAContainerRuntime(logger *logrus.Logger, cfg *config.Config, argv [
6161

6262
// newSpecModifier is a factory method that creates constructs an OCI spec modifer based on the provided config.
6363
func newSpecModifier(logger *logrus.Logger, cfg *config.Config, ociSpec oci.Spec, argv []string) (oci.SpecModifier, error) {
64-
modeModifier, err := newModeModifier(logger, cfg, ociSpec, argv)
64+
mode := info.ResolveAutoMode(logger, cfg.NVIDIAContainerRuntimeConfig.Mode)
65+
modeModifier, err := newModeModifier(logger, mode, cfg, ociSpec, argv)
6566
if err != nil {
6667
return nil, err
6768
}
69+
// For CDI mode we make no additional modifications.
70+
if mode == "cdi" {
71+
return modeModifier, nil
72+
}
6873

6974
graphicsModifier, err := modifier.NewGraphicsModifier(logger, cfg, ociSpec)
7075
if err != nil {
@@ -96,8 +101,8 @@ func newSpecModifier(logger *logrus.Logger, cfg *config.Config, ociSpec oci.Spec
96101
return modifiers, nil
97102
}
98103

99-
func newModeModifier(logger *logrus.Logger, cfg *config.Config, ociSpec oci.Spec, argv []string) (oci.SpecModifier, error) {
100-
switch info.ResolveAutoMode(logger, cfg.NVIDIAContainerRuntimeConfig.Mode) {
104+
func newModeModifier(logger *logrus.Logger, mode string, cfg *config.Config, ociSpec oci.Spec, argv []string) (oci.SpecModifier, error) {
105+
switch mode {
101106
case "legacy":
102107
return modifier.NewStableRuntimeModifier(logger, cfg.NVIDIAContainerRuntimeHookConfig.Path), nil
103108
case "csv":

0 commit comments

Comments
 (0)