last changes

Signed-off-by: Shahriyar Jalayeri <[email protected]>

pkg/pillar/hypervisor/kvm.go

@@ -1443,7 +1443,8 @@ func requestvTPMLaunch(id uuid.UUID, wp *types.WatchdogParam, timeoutSeconds uin
 	}
 
 	// One last time, check SWTPM is not dead right after launch.
-	if !utils.IsProcAlive(pid) {
+	swtpmCtrlSock := fmt.Sprintf(types.SwtpmCtrlSocketPath, id.String())
+	if !utils.IsUnixDomSockAlive(swtpmCtrlSock) {
 		return fmt.Errorf("SWTPM (pid: %d) is dead", pid)
 	}
 

pkg/pillar/utils/proc.go

@@ -5,6 +5,7 @@ package utils
 
 import (
 	"fmt"
+	"net"
 	"os"
 	"strconv"
 	"strings"
@@ -13,6 +14,7 @@ import (
 
 	"github.com/lf-edge/eve/pkg/pillar/base"
 	"github.com/lf-edge/eve/pkg/pillar/pubsub"
+	log "github.com/sirupsen/logrus"
 )
 
 // WatchdogKicker is used in some proc functions that have a timeout,
@@ -110,7 +112,34 @@ func GetPidFromFileTimeout(pidFile string, timeoutSeconds uint, wk *WatchdogKick
 	}
 }
 
-// IsProcAlive checks if a process is alive or not.
+// IsUnixDomSockAlive checks if a unix domain socket is alive.
+func IsUnixDomSockAlive(sock string) bool {
+	conn, err := net.Dial("unix", sock)
+	if err != nil {
+		log.Errorf("failed to dial socket %s, %v", sock, err)
+		return false
+	}
+	rawConn, err := conn.(syscall.Conn).SyscallConn()
+	if err != nil {
+		log.Errorf("failed to get raw connection %v", err)
+		return false
+	}
+
+	var peerAlive bool
+	err = rawConn.Control(func(fd uintptr) {
+		_, err := syscall.Getpeername(int(fd))
+		peerAlive = (err == nil)
+	})
+
+	if err != nil {
+		log.Errorf("failed to get peer name %v", err)
+		return false
+	}
+
+	return peerAlive
+}
+
+// IsProcAlive checks current proc namespace to see process is alive or not.
 func IsProcAlive(pid int) bool {
 	err := syscall.Kill(pid, syscall.Signal(0))
 	if err != nil {
@@ -122,7 +151,7 @@ func IsProcAlive(pid int) bool {
 	return true
 }
 
-// IsProcAliveTimeout checks if a process is alive for a given timeout.
+// IsProcAliveTimeout checks current proc namespace to see process is alive or not for a given timeout.
 func IsProcAliveTimeout(pid int, timeoutSeconds uint, wk *WatchdogKicker) bool {
 	startTime := time.Now()
 	for {