fix runc events error in cgroup v2

Signed-off-by: lifubang <[email protected]>

Author: lifubang

libcontainer/container_linux.go

@@ -648,6 +648,13 @@ func (c *linuxContainer) NotifyOOM() (<-chan struct{}, error) {
 	if c.config.RootlessCgroups {
 		logrus.Warn("getting OOM notifications may fail if you don't have the full access to cgroups")
 	}
+	if cgroups.IsCgroup2UnifiedMode() {
+		path, err := c.cgroupManager.GetUnifiedPath()
+		if err != nil {
+			return nil, err
+		}
+		return notifyOnOOMV2(path)
+	}
 	return notifyOnOOM(c.cgroupManager.GetPaths())
 }
 

libcontainer/notify_linux_v2.go

@@ -0,0 +1,127 @@
+// +build linux
+
+package libcontainer
+
+import (
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"unsafe"
+
+	"github.com/sirupsen/logrus"
+	"golang.org/x/sys/unix"
+)
+
+func getOOMCount(path string) (int, error) {
+	if _, err := os.Lstat(path); err != nil {
+		return 0, err
+	}
+
+	content, err := ioutil.ReadFile(path)
+	if err != nil {
+		return 0, err
+	}
+
+	oomCount := 0
+	lines := strings.Split(string(content), "\n")
+	for _, line := range lines {
+		arr := strings.Split(line, " ")
+		if len(arr) == 2 && (arr[0] == "oom" || arr[0] == "oom_kill") {
+			count, err := strconv.Atoi(arr[1])
+			if err == nil && count > oomCount {
+				oomCount = count
+			}
+		}
+	}
+	return oomCount, nil
+}
+
+func getPopulated(path string) (int, error) {
+	if _, err := os.Lstat(path); err != nil {
+		return 0, err
+	}
+
+	content, err := ioutil.ReadFile(path)
+	if err != nil {
+		return 0, err
+	}
+	lines := strings.Split(string(content), "\n")
+	for _, line := range lines {
+		arr := strings.Split(line, " ")
+		if len(arr) == 2 && arr[0] == "populated" {
+			return strconv.Atoi(arr[1])
+		}
+	}
+	return 0, nil
+}
+
+func registerMemoryEventV2(cgDir string, evName string, cgEvName string) (<-chan struct{}, error) {
+	eventControlPath := filepath.Join(cgDir, evName)
+	cgEvPath := filepath.Join(cgDir, cgEvName)
+	fd, err := unix.InotifyInit()
+	if err != nil {
+		return nil, err
+	}
+	// watching oom kill
+	_, err = unix.InotifyAddWatch(fd, eventControlPath, unix.IN_MODIFY)
+	if err != nil {
+		unix.Close(fd)
+		return nil, err
+	}
+	// Because no `unix.IN_DELETE|unix.IN_DELETE_SELF` event for cgroup file system, so watching all process exited
+	_, err = unix.InotifyAddWatch(fd, cgEvPath, unix.IN_MODIFY)
+	if err != nil {
+		unix.Close(fd)
+		return nil, err
+	}
+	ch := make(chan struct{})
+	go func() {
+		var (
+			buffer [unix.SizeofInotifyEvent + unix.PathMax]byte
+			offset uint32
+		)
+		defer func() {
+			unix.Close(fd)
+			close(ch)
+		}()
+
+		for {
+			n, err := unix.Read(fd, buffer[:])
+			if err != nil {
+				logrus.Warn(err)
+				return
+			}
+			offset = 0
+			for offset <= uint32(n-unix.SizeofInotifyEvent) {
+				rawEvent := (*unix.InotifyEvent)(unsafe.Pointer(&buffer[offset]))
+				bytes := (*[unix.PathMax]byte)(unsafe.Pointer(&buffer[unix.SizeofInotifyEvent]))
+				eventName := strings.TrimRight(string(bytes[0:rawEvent.Len]), "\000")
+				offset += unix.SizeofInotifyEvent + uint32(rawEvent.Len)
+				if rawEvent.Mask&unix.IN_MODIFY == unix.IN_MODIFY {
+					switch eventName {
+					case evName:
+						oom, err := getOOMCount(eventControlPath)
+						if err != nil || oom > 0 {
+							ch <- struct{}{}
+							return
+						}
+					case cgEvName:
+						pids, err := getPopulated(cgEvPath)
+						if err != nil || pids == 0 {
+							return
+						}
+					}
+				}
+			}
+		}
+	}()
+	return ch, nil
+}
+
+// notifyOnOOMV2 returns channel on which you can expect event about OOM,
+// if process died without OOM this channel will be closed.
+func notifyOnOOMV2(path string) (<-chan struct{}, error) {
+	return registerMemoryEventV2(path, "memory.events", "cgroup.events")
+}

tests/integration/events.bats

@@ -13,8 +13,8 @@ function teardown() {
 
 @test "events --stats" {
   # XXX: currently cgroups require root containers.
-  # TODO: support cgroup v2 memory.events
-  requires root cgroups_v1
+  requires root
+  init_cgroup_paths
 
   # run busybox detached
   runc run -d --console-socket $CONSOLE_SOCKET test_busybox
@@ -29,8 +29,8 @@ function teardown() {
 
 @test "events --interval default " {
   # XXX: currently cgroups require root containers.
-  # TODO: support cgroup v2 memory.events
-  requires root cgroups_v1
+  requires root
+  init_cgroup_paths
 
   # run busybox detached
   runc run -d --console-socket $CONSOLE_SOCKET test_busybox
@@ -57,8 +57,8 @@ function teardown() {
 
 @test "events --interval 1s " {
   # XXX: currently cgroups require root containers.
-  # TODO: support cgroup v2 memory.events
-  requires root cgroups_v1
+  requires root
+  init_cgroup_paths
 
   # run busybox detached
   runc run -d --console-socket $CONSOLE_SOCKET test_busybox
@@ -84,8 +84,8 @@ function teardown() {
 
 @test "events --interval 100ms " {
   # XXX: currently cgroups require root containers.
-  # TODO: support cgroup v2 memory.events
-  requires root cgroups_v1
+  requires root
+  init_cgroup_paths
 
   # run busybox detached
   runc run -d --console-socket $CONSOLE_SOCKET test_busybox