stale pidfile handling

internal/app/wwclient/root.go

@@ -16,7 +16,7 @@ import (
 
 	"github.com/google/uuid"
 	"github.com/talos-systems/go-smbios/smbios"
-	"github.com/hpcng/warewulf/internal/pkg/util"
+	"github.com/hpcng/warewulf/internal/pkg/pidfile"
 	"github.com/hpcng/warewulf/internal/pkg/warewulfconf"
 	"github.com/hpcng/warewulf/internal/pkg/wwlog"
 	"github.com/spf13/cobra"
@@ -53,16 +53,12 @@ func CobraRunE(cmd *cobra.Command, args []string) error {
 		return err
 	}
 
-	if util.IsFile(PIDFile) {
+	pid, err := pidfile.Write(PIDFile)
+	if err != nil && pid == -1 {
+		wwlog.Printf(wwlog.WARN, "%v. starting new wwclient", err)
+	} else if err != nil && pid > 0 {
 		return errors.New("found pidfile " + PIDFile + " not starting")
 	}
-	p, err := os.OpenFile(PIDFile, os.O_RDWR|os.O_CREATE|os.O_TRUNC, 0644)
-	if err != nil {
-		return err
-	}
-	defer p.Close()
-
-	fmt.Fprintf(p, "%d", os.Getpid())
 
 	if os.Args[0] == "/warewulf/bin/wwclient" {
 		err := os.Chdir("/")
@@ -137,20 +133,20 @@ func CobraRunE(cmd *cobra.Command, args []string) error {
 	wwid := strings.Split(wwid_tmp[1], " ")[0]
 
 	// listen on SIGHUP
-	sigs := make(chan os.Signal)
+	sigs := make(chan os.Signal, 1)
 
 	signal.Notify(sigs, syscall.SIGHUP, syscall.SIGTERM, syscall.SIGINT)
 
 	go func() {
-		for sig := range sigs {
+		sig := <-sigs
-			switch sig {
+		switch sig {
-			case syscall.SIGHUP:
+		case syscall.SIGHUP:
-				log.Printf("Received SIGNAL: %s\n", sig)
+			log.Printf("Received SIGNAL: %s\n", sig)
-        updateSystem(conf.Ipaddr, conf.Warewulf.Port, wwid, tag, localUUID)
+      updateSystem(conf.Ipaddr, conf.Warewulf.Port, wwid, tag, localUUID)
-			case syscall.SIGTERM, syscall.SIGINT:
+		case syscall.SIGTERM, syscall.SIGINT:
-				cleanUp()
+			wwlog.Printf(wwlog.INFO, "termination wwclient!, %v", sig)
-				os.Exit(0)
+			cleanUp()
-			}
+			os.Exit(0)
 		}
 	}()
 
@@ -200,9 +196,8 @@ func updateSystem(ipaddr string, port int, wwid string, tag string, localUUID uu
 }
 
 func cleanUp() {
-	err := os.Remove(PIDFile)
+	err := pidfile.Remove(PIDFile)
 	if err != nil {
-		log.Printf("could not remove pidfile: %s\n", err)
+		wwlog.Printf(wwlog.ERROR, "could not remove pidfile: %s\n", err)
 	}
-
 }

internal/pkg/pidfile/pidfile.go

@@ -0,0 +1,84 @@
+package pidfile
+
+// based on original work found here, github.com/soellman/pidfile
+
+import (
+	"errors"
+	"fmt"
+	"io/ioutil"
+	"os"
+	"strconv"
+	"strings"
+	"syscall"
+)
+
+var (
+	ErrProcessRunning = errors.New("process is running")
+	ErrFileStale      = errors.New("pidfile exists but process is not running")
+	ErrFileInvalid    = errors.New("pidfile has invalid contents")
+)
+
+// Remove a pidfile
+func Remove(filename string) error {
+	return os.RemoveAll(filename)
+}
+
+// Write writes a pidfile, returning an error
+// if the process is already running or pidfile is orphaned
+func Write(filename string) (int, error) {
+	return WriteControl(filename, os.Getpid(), false)
+}
+
+func WriteControl(filename string, pid int, overwrite bool) (int, error) {
+	// Check for existing pid
+	oldpid, err := pidfileContents(filename)
+	if err != nil && !os.IsNotExist(err) {
+		return oldpid, err
+	}
+
+	// We have a pid
+	if err == nil {
+		if pidIsRunning(oldpid) {
+			return oldpid, ErrProcessRunning
+		}
+		if !overwrite {
+			return -1, ErrFileStale
+		}
+	}
+
+	// We're clear to (over)write the file
+	return pid, ioutil.WriteFile(filename, []byte(fmt.Sprintf("%d\n", pid)), 0644)
+}
+
+func pidfileContents(filename string) (int, error) {
+	contents, err := ioutil.ReadFile(filename)
+	if err != nil {
+		return 0, err
+	}
+
+	pid, err := strconv.Atoi(strings.TrimSpace(string(contents)))
+	if err != nil {
+		return 0, ErrFileInvalid
+	}
+
+	return pid, nil
+}
+
+func pidIsRunning(pid int) bool {
+	process, err := os.FindProcess(pid)
+	if err != nil {
+		return false
+	}
+
+	err = process.Signal(syscall.Signal(0))
+
+	if err != nil && err.Error() == "no such process" {
+		return false
+	}
+
+	if err != nil && err.Error() == "os: process already finished" {
+		return false
+	}
+
+	return true
+}