feat: Implement CheckAndMountFilesystems for required filesystem checks in DNS and TCP retrans plugins

Author: Iti Agrawal

pkg/plugin/common/filesystem_linux.go

@@ -0,0 +1,67 @@
+// Copyright (c) Microsoft Corporation.
+// Licensed under the MIT license.
+
+//go:build linux
+// +build linux
+
+package common
+
+import (
+	"github.com/microsoft/retina/pkg/log"
+	"go.uber.org/zap"
+	"golang.org/x/sys/unix"
+)
+
+// CheckAndMountFilesystems checks if required filesystems are mounted.
+// Returns an error if any required filesystem is not available.
+// This helps prevent os.Exit() calls from dependencies that expect these filesystems.
+func CheckAndMountFilesystems(l *log.ZapLogger) error {
+	filesystems := []struct {
+		name     string
+		paths    []string
+		magic    int64
+		required bool // if true, return error if not available
+	}{
+		{
+			name:     "bpf",
+			paths:    []string{"/sys/fs/bpf"},
+			magic:    unix.BPF_FS_MAGIC,
+			required: false, // bpffs is less critical
+		},
+		{
+			name:     "debugfs",
+			paths:    []string{"/sys/kernel/debug"},
+			magic:    unix.DEBUGFS_MAGIC,
+			required: true,
+		},
+		{
+			name:     "tracefs",
+			paths:    []string{"/sys/kernel/tracing", "/sys/kernel/debug/tracing"},
+			magic:    unix.TRACEFS_MAGIC,
+			required: true,
+		},
+	}
+
+	var firstError error
+filesystemLoop:
+	for _, fs := range filesystems {
+		var statfs unix.Statfs_t
+		for _, path := range fs.paths {
+			if err := unix.Statfs(path, &statfs); err != nil {
+				l.Debug("statfs returned error", zap.String("fs", fs.name), zap.String("path", path), zap.Error(err))
+				continue
+			}
+			if statfs.Type == fs.magic {
+				l.Debug("Filesystem already mounted", zap.String("fs", fs.name), zap.String("path", path))
+				continue filesystemLoop
+			}
+		}
+
+		// Filesystem not found or not mounted
+		l.Error("Filesystem not mounted", zap.String("fs", fs.name), zap.Strings("paths", fs.paths))
+		if fs.required && firstError == nil {
+			firstError = unix.ENOENT
+		}
+	}
+	return firstError
+}

pkg/plugin/dns/dns_linux.go

@@ -6,6 +6,7 @@ package dns
 
 import (
 	"context"
+	"fmt"
 	"net"
 	"os"
 
@@ -47,8 +48,21 @@ func (d *dns) Compile(ctx context.Context) error {
 }
 
 func (d *dns) Init() error {
-	// Create tracer. In this case no parameters are passed.
-	err := host.Init(host.Config{})
+	// Check and mount filesystems before calling host.Init to avoid os.Exit()
+	if err := common.CheckAndMountFilesystems(d.l); err != nil {
+		d.l.Error("Required filesystems not available for DNS plugin", zap.Error(err))
+		// Return error to let retina decide whether to continue without DNS plugin
+		// or fail the entire agent initialization
+		return fmt.Errorf("required filesystems not available: %w", err)
+	}
+
+	// Filesystems are available, safe to call host.Init()
+	if err := host.Init(host.Config{}); err != nil {
+		d.l.Error("Host initialization failed", zap.Error(err))
+		return fmt.Errorf("host initialization failed: %w", err)
+	}
+
+	// Create tracer
 	tracer, err := tracer.NewTracer()
 	if err != nil {
 		d.l.Error("Failed to create tracer", zap.Error(err))

pkg/plugin/tcpretrans/tcpretrans_linux.go

@@ -19,6 +19,7 @@ import (
 	kcfg "github.com/microsoft/retina/pkg/config"
 	"github.com/microsoft/retina/pkg/enricher"
 	"github.com/microsoft/retina/pkg/log"
+	"github.com/microsoft/retina/pkg/plugin/common"
 	"github.com/microsoft/retina/pkg/plugin/registry"
 	"github.com/microsoft/retina/pkg/utils"
 	"go.uber.org/zap"
@@ -53,6 +54,14 @@ func (t *tcpretrans) Init() error {
 		t.l.Warn("tcpretrans will not init because pod level is disabled")
 		return nil
 	}
+
+	if err := common.CheckAndMountFilesystems(t.l); err != nil {
+		t.l.Error("Required filesystems not available for tcpretrans plugin", zap.Error(err))
+		// Return error to let retina decide whether to continue without tcpretrans plugin
+		// or fail the entire agent initialization
+		return fmt.Errorf("required filesystems not available: %w", err)
+	}
+
 	// Create tracer. In this case no parameters are passed.
 	if err := host.Init(host.Config{}); err != nil {
 		t.l.Error("failed to init host", zap.Error(err))