[client] Add NB_SKIP_SOCKET_MARK & fix crash instead of returing an error (#2899)

* dialer: fix crash instead of returning error

* add NB_SKIP_SOCKET_MARK

client/internal/routemanager/systemops/systemops_linux.go

@@ -55,7 +55,7 @@ type ruleParams struct {
 
 // isLegacy determines whether to use the legacy routing setup
 func isLegacy() bool {
-	return os.Getenv("NB_USE_LEGACY_ROUTING") == "true" || nbnet.CustomRoutingDisabled()
+	return os.Getenv("NB_USE_LEGACY_ROUTING") == "true" || nbnet.CustomRoutingDisabled() || os.Getenv(nbnet.EnvSkipSocketMark) == "true"
 }
 
 // setIsLegacy sets the legacy routing setup

util/grpc/dialer.go

@@ -3,6 +3,9 @@ package grpc
 import (
 	"context"
 	"crypto/tls"
+	"fmt"
+	"google.golang.org/grpc/codes"
+	"google.golang.org/grpc/status"
 	"net"
 	"os/user"
 	"runtime"
@@ -23,20 +26,22 @@ func WithCustomDialer() grpc.DialOption {
 		if runtime.GOOS == "linux" {
 			currentUser, err := user.Current()
 			if err != nil {
-				log.Fatalf("failed to get current user: %v", err)
+				return nil, status.Errorf(codes.FailedPrecondition, "failed to get current user: %v", err)
 			}
 
 			// the custom dialer requires root permissions which are not required for use cases run as non-root
 			if currentUser.Uid != "0" {
+				log.Debug("Not running as root, using standard dialer")
 				dialer := &net.Dialer{}
 				return dialer.DialContext(ctx, "tcp", addr)
 			}
 		}
 
+		log.Debug("Using nbnet.NewDialer()")
 		conn, err := nbnet.NewDialer().DialContext(ctx, "tcp", addr)
 		if err != nil {
 			log.Errorf("Failed to dial: %s", err)
-			return nil, err
+			return nil, fmt.Errorf("nbnet.NewDialer().DialContext: %w", err)
 		}
 		return conn, nil
 	})

util/net/dialer_nonios.go

@@ -69,7 +69,7 @@ func (d *Dialer) DialContext(ctx context.Context, network, address string) (net.
 
 	conn, err := d.Dialer.DialContext(ctx, network, address)
 	if err != nil {
-		return nil, fmt.Errorf("dial: %w", err)
+		return nil, fmt.Errorf("d.Dialer.DialContext: %w", err)
 	}
 
 	// Wrap the connection in Conn to handle Close with hooks

util/net/net_linux.go

@@ -4,9 +4,14 @@ package net
 
 import (
 	"fmt"
+	"os"
 	"syscall"
+
+	log "github.com/sirupsen/logrus"
 )
 
+const EnvSkipSocketMark = "NB_SKIP_SOCKET_MARK"
+
 // SetSocketMark sets the SO_MARK option on the given socket connection
 func SetSocketMark(conn syscall.Conn) error {
 	sysconn, err := conn.SyscallConn()
@@ -36,6 +41,13 @@ func SetRawSocketMark(conn syscall.RawConn) error {
 
 func SetSocketOpt(fd int) error {
 	if CustomRoutingDisabled() {
+		log.Infof("Custom routing is disabled, skipping SO_MARK")
+		return nil
+	}
+
+	// Check for the new environment variable
+	if skipSocketMark := os.Getenv(EnvSkipSocketMark); skipSocketMark == "true" {
+		log.Info("NB_SKIP_SOCKET_MARK is set to true, skipping SO_MARK")
 		return nil
 	}