Merge pull request #10851 from Luap99/service-reaper

podman service reaper

cmd/podman/system/service_abi.go

@@ -12,6 +12,7 @@ import (
 	api "github.com/containers/podman/v3/pkg/api/server"
 	"github.com/containers/podman/v3/pkg/domain/entities"
 	"github.com/containers/podman/v3/pkg/domain/infra"
+	"github.com/containers/podman/v3/pkg/servicereaper"
 	"github.com/containers/podman/v3/pkg/util"
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
@@ -71,6 +72,8 @@ func restService(opts entities.ServiceOptions, flags *pflag.FlagSet, cfg *entiti
 		return err
 	}
 
+	servicereaper.Start()
+
 	infra.StartWatcher(rt)
 	server, err := api.NewServerWithSettings(rt, listener, api.Options{Timeout: opts.Timeout, CorsHeaders: opts.CorsHeaders})
 	if err != nil {

libpod/networking_slirp4netns.go

@@ -18,6 +18,7 @@ import (
 
 	"github.com/containers/podman/v3/pkg/errorhandling"
 	"github.com/containers/podman/v3/pkg/rootlessport"
+	"github.com/containers/podman/v3/pkg/servicereaper"
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
 )
@@ -299,6 +300,7 @@ func (r *Runtime) setupSlirp4netns(ctr *Container) error {
 		return errors.Wrapf(err, "failed to start slirp4netns process")
 	}
 	defer func() {
+		servicereaper.AddPID(cmd.Process.Pid)
 		if err := cmd.Process.Release(); err != nil {
 			logrus.Errorf("unable to release command process: %q", err)
 		}
@@ -514,6 +516,7 @@ outer:
 		return errors.Wrapf(err, "failed to start rootlessport process")
 	}
 	defer func() {
+		servicereaper.AddPID(cmd.Process.Pid)
 		if err := cmd.Process.Release(); err != nil {
 			logrus.Errorf("unable to release rootlessport process: %q", err)
 		}

pkg/servicereaper/service.go

@@ -0,0 +1,64 @@
+//+build linux
+
+package servicereaper
+
+import (
+	"os"
+	"os/signal"
+	"sync"
+	"syscall"
+
+	"github.com/sirupsen/logrus"
+)
+
+type service struct {
+	pidMap map[int]bool
+	mutex  *sync.Mutex
+}
+
+var s = service{
+	pidMap: map[int]bool{},
+	mutex:  &sync.Mutex{},
+}
+
+func AddPID(pid int) {
+	s.mutex.Lock()
+	s.pidMap[pid] = true
+	s.mutex.Unlock()
+}
+
+func Start() {
+	// create signal channel and only wait for SIGCHLD
+	sigc := make(chan os.Signal, 1)
+	signal.Notify(sigc, syscall.SIGCHLD)
+	// wait and reap in an  extra goroutine
+	go reaper(sigc)
+}
+
+func reaper(sigc chan os.Signal) {
+	for {
+		// block until we receive SIGCHLD
+		<-sigc
+		s.mutex.Lock()
+		for pid := range s.pidMap {
+			var status syscall.WaitStatus
+			waitpid, err := syscall.Wait4(pid, &status, syscall.WNOHANG, nil)
+			if err != nil {
+				// do not log error for ECHILD
+				if err != syscall.ECHILD {
+					logrus.Warnf("wait for pid %d failed: %v ", pid, err)
+				}
+				delete(s.pidMap, pid)
+				continue
+			}
+			// if pid == 0 nothing happened
+			if waitpid == 0 {
+				continue
+			}
+			if status.Exited() {
+				delete(s.pidMap, pid)
+			}
+		}
+		s.mutex.Unlock()
+	}
+}