ON-15180: Read sysfs to check for NICs

Rather than rely on an external tool (lshw) which requires extra
dependencies in container images and doesn't provide an simple way of
determining the vendor of a nic, it would be better to search through
sysfs directly (since that is basically all lshw is doing). This
approach also allows us to easily get other information from sysfs (such
as numa nodes).

Author: tcrawley-xilinx

deviceplugin.Dockerfile

@@ -22,7 +22,6 @@ COPY LICENSE /app/LICENSE
 RUN CGO_ENABLED=0 make device-plugin-build worker-build
 
 FROM registry.access.redhat.com/ubi8/ubi-minimal:8.9
-RUN microdnf install -y lshw-B.02.19.2 && microdnf clean all
 COPY --from=builder /app/bin/onload-device-plugin /app/bin/onload-worker /usr/bin/
 COPY --from=builder /app/LICENSE /licenses/LICENSE
 USER 1001

pkg/deviceplugin/manager.go

@@ -93,7 +93,6 @@ func NewNicManager(
 // Initialises the set of devices to advertise to kubernetes
 func (manager *NicManager) initDevices() {
 	manager.devices = []*pluginapi.Device{}
-	fmt.Println(manager.interfaces)
 	for i := 0; i < manager.config.MaxPodsPerNode; i++ {
 		name := fmt.Sprintf("sfc-%v", i)
 		device := &pluginapi.Device{

pkg/deviceplugin/nic.go

@@ -3,69 +3,67 @@
 package deviceplugin
 
 import (
+	"errors"
 	"os"
-	"os/exec"
-	"regexp"
+	"path"
 	"strings"
 
 	"github.com/golang/glog"
 )
 
-// Takes the output from lshw and returns the device name for each solarflare
-// device.
-func parseOutput(output string) []string {
-	// "lshw -short -class network" sample output:
-	// H/W path            Device     Class          Description
-	// =========================================================
-	// /0/100/1b/0         enp2s0f0   network        XtremeScale SFC9250 10/25/40/50/100G Ethernet Controller
-	// /0/100/1b/0.1       enp2s0f1   network        XtremeScale SFC9250 10/25/40/50/100G Ethernet Controller
-	// /0/100/1c.1/0       eno1       network        NetXtreme BCM5720 Gigabit Ethernet PCIe
-	// /0/100/1c.1/0.1     eno2       network        NetXtreme BCM5720 Gigabit Ethernet PCIe
-
-	lines := strings.Split(output, "\n")
+const (
+	sysClassNetPath  = "/sys/class/net/"
+	solarflareVendor = "0x1924"
+)
 
-	var interfaces []string
+func isSFCNic(devicePath string) bool {
+	deviceDir, err := os.Stat(path.Join(devicePath, "device"))
+	if errors.Is(err, os.ErrNotExist) {
+		// Not a physical device, so won't have the "vendor" file
+		return false
+	} else if err != nil {
+		glog.Errorf("Failed to stat %s (%v)", devicePath, err)
+		return false
+	}
+	if !deviceDir.IsDir() {
+		return false
+	}
 
-	// Assume that we are running as root, if not then we would have to skip
-	// an additional line at the start of the output
-	skip_lines := 2
-	end_lines := 1
-	if os.Geteuid() != 0 {
-		skip_lines = 3
-		end_lines = 2
+	data, err := os.ReadFile(path.Join(devicePath, "device", "vendor"))
+	if errors.Is(err, os.ErrNotExist) {
+		// File doesn't exist but that is fine
+		return false
+	} else if err != nil {
+		glog.Errorf("Error reading %s (%v)",
+			path.Join(devicePath, "device", "vendor"), err)
+		return false
 	}
 
-	for _, line := range lines[skip_lines : len(lines)-end_lines] {
-		// This regex makes the assumption that all interface names only
-		// contain either lowercase letters or numbers. If that is not true,
-		// then this should be updated to reflect that.
-		r := regexp.MustCompile("([a-z0-9]+) *network *.*SFC")
-		out := r.FindStringSubmatch(line)
-		if out != nil {
-			// It is safe to access out[1] here since the return value of
-			// FindStringSubmatch is an array where the first value is the
-			// whole string and any subsequent values are the submatches.
-			// In this case since there is a submatch that should match the
-			// device name if FindStringSubmatch returns non-nil then there
-			// will be at least 2 elements in the return array.
-			interfaces = append(interfaces, out[1])
+	vendor := strings.TrimSuffix(string(data), "\n")
+	return vendor == solarflareVendor
+}
+
+func readSysFiles() ([]string, error) {
+	infos, err := os.ReadDir(sysClassNetPath)
+	if err != nil {
+		glog.Errorf("Error reading %s (%v)", sysClassNetPath, err)
+		return []string{}, err
+	}
+	interfaces := []string{}
+	for _, info := range infos {
+		if isSFCNic(path.Join(sysClassNetPath, info.Name())) {
+			interfaces = append(interfaces, info.Name())
 		}
 	}
-	return interfaces
+	return interfaces, nil
 }
 
 // Returns a list of the Solarflare interfaces present on the node
 func queryNics() ([]string, error) {
-	// Depending on what information we are looking for in the output I think it
-	// is quite tempting to retrieve the information in a json format, then
-	// parse this using golang's built-in features.
-	bytes, err := exec.Command("lshw", "-short", "-class", "network").CombinedOutput()
-	output := string(bytes)
+	interfaces, err := readSysFiles()
 	if err != nil {
-		glog.Error(output)
-		glog.Errorf("error while listing sfc devices : %v", err)
-		return nil, err
+		glog.Errorf("Failed to list interfaces (%v)", err)
+		return []string{}, err
 	}
-	interfaces := parseOutput(output)
 	return interfaces, nil
 }