Add support for a platform override file

This change adds support for reading the detected
platform (if set to `auto`) from a platform override
file. This allows system administrators to explicitly
select a detected platform for tooling such as the
nvidia-container-toolkit, the k8s-device-plugin, and
k8s-dra-driver-gpu.

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Author: elezar

pkg/nvlib/info/resolver.go

@@ -16,6 +16,13 @@
 
 package info
 
+import (
+	"bufio"
+	"fmt"
+	"os"
+	"strings"
+)
+
 // Platform represents a supported plaform.
 type Platform string
 
@@ -39,6 +46,13 @@ func (p platformResolver) ResolvePlatform() Platform {
 		return p.platform
 	}
 
+	override, reason := getPlaformOverride()
+	if override != "" {
+		p.logger.Debugf("Using platform-override %q", override)
+		return Platform(override)
+	}
+	p.logger.Debugf("No platform-override detected: %v", reason)
+
 	hasDXCore, reason := p.propertyExtractor.HasDXCore()
 	p.logger.Debugf("Is WSL-based system? %v: %v", hasDXCore, reason)
 
@@ -62,3 +76,33 @@ func (p platformResolver) ResolvePlatform() Platform {
 		return PlatformUnknown
 	}
 }
+
+// getPlatformOverride checks the system for a platform override file.
+// This allows system administrators to force the detection of a specific
+// platform.
+//
+// The first non-empty and non-comment line (starting with #) in the file is
+// returned.
+//
+// Note that no checks are performed for a valid platform value.
+var getPlaformOverride = func() (string, string) {
+	platformOverrideFile, err := os.Open("/etc/nvidia-container-toolkit/platform-override")
+	if os.IsNotExist(err) {
+		return "", "platform-override file does not exist"
+	}
+	if err != nil {
+		return "", fmt.Errorf("failed to open platform-override file: %w", err).Error()
+	}
+	defer platformOverrideFile.Close()
+
+	scanner := bufio.NewScanner(platformOverrideFile)
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+		if line == "" || strings.HasPrefix(line, "#") {
+			continue
+		}
+		return line, "read from platform-override file"
+	}
+
+	return "", "empty platform-override file"
+}

pkg/nvlib/info/resolver_test.go

@@ -26,6 +26,7 @@ import (
 func TestResolvePlatform(t *testing.T) {
 	testCases := []struct {
 		platform           string
+		platformOverride   string
 		hasTegraFiles      bool
 		hasDXCore          bool
 		hasNVML            bool
@@ -82,10 +83,16 @@ func TestResolvePlatform(t *testing.T) {
 			hasDXCore: true,
 			expected:  "not-auto",
 		},
+		{
+			platform:         "auto",
+			platformOverride: "overridden",
+			expected:         "overridden",
+		},
 	}
 
 	for i, tc := range testCases {
 		t.Run(fmt.Sprintf("test case %d", i), func(t *testing.T) {
+			defer setGetPlatformOverrideForTest(tc.platformOverride)()
 			l := New(
 				WithPropertyExtractor(&PropertyExtractorMock{
 					HasDXCoreFunc: func() (bool, string) {
@@ -108,3 +115,16 @@ func TestResolvePlatform(t *testing.T) {
 		})
 	}
 }
+
+// setGetPlatformOverrideForTest overrides the distribution IDs that would normally be read from the /etc/os-release file.
+func setGetPlatformOverrideForTest(override string) func() {
+	original := getPlaformOverride
+
+	getPlaformOverride = func() (string, string) {
+		return override, "overridden for test"
+	}
+
+	return func() {
+		getPlaformOverride = original
+	}
+}