Merge pull request #1134 from crosbymichael/shim-cgroup

Allow shim to be placed inside cgroup

container_linux_test.go

@@ -3,10 +3,12 @@
 package containerd
 
 import (
+	"context"
 	"syscall"
 	"testing"
 
 	"github.com/containerd/cgroups"
+	"github.com/containerd/containerd/linux/runcopts"
 	specs "github.com/opencontainers/runtime-spec/specs-go"
 )
 
@@ -98,3 +100,78 @@ func TestContainerUpdate(t *testing.T) {
 
 	<-statusC
 }
+
+func TestShimInCgroup(t *testing.T) {
+	client, err := newClient(t, address)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer client.Close()
+
+	var (
+		ctx, cancel = testContext()
+		id          = t.Name()
+	)
+	defer cancel()
+
+	image, err := client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Error(err)
+		return
+	}
+	spec, err := GenerateSpec(WithImageConfig(ctx, image), WithProcessArgs("sleep", "30"))
+	if err != nil {
+		t.Error(err)
+		return
+	}
+	container, err := client.NewContainer(ctx, id, WithSpec(spec), WithNewSnapshot(id, image))
+	if err != nil {
+		t.Error(err)
+		return
+	}
+	defer container.Delete(ctx, WithSnapshotCleanup)
+	// create a cgroup for the shim to use
+	path := "/containerd/shim"
+	cg, err := cgroups.New(cgroups.V1, cgroups.StaticPath(path), &specs.LinuxResources{})
+	if err != nil {
+		t.Error(err)
+		return
+	}
+	defer cg.Delete()
+
+	task, err := container.NewTask(ctx, empty(), func(_ context.Context, client *Client, r *TaskInfo) error {
+		r.Options = &runcopts.CreateOptions{
+			ShimCgroup: path,
+		}
+		return nil
+	})
+	if err != nil {
+		t.Error(err)
+		return
+	}
+	defer task.Delete(ctx)
+
+	statusC := make(chan uint32, 1)
+	go func() {
+		status, err := task.Wait(ctx)
+		if err != nil {
+			t.Error(err)
+		}
+		statusC <- status
+	}()
+	// check to see if the shim is inside the cgroup
+	processes, err := cg.Processes(cgroups.Devices, false)
+	if err != nil {
+		t.Error(err)
+		return
+	}
+	if len(processes) == 0 {
+		t.Errorf("created cgroup should have atleast one process inside: %d", len(processes))
+	}
+	if err := task.Kill(ctx, syscall.SIGKILL); err != nil {
+		t.Error(err)
+		return
+	}
+
+	<-statusC
+}

linux/bundle.go

@@ -9,7 +9,10 @@ import (
 	"os"
 	"path/filepath"
 
+	"github.com/containerd/containerd/linux/runcopts"
 	client "github.com/containerd/containerd/linux/shim"
+	"github.com/containerd/containerd/runtime"
+	"github.com/containerd/containerd/typeurl"
 )
 
 func loadBundle(path, namespace string) *bundle {
@@ -56,15 +59,24 @@ type bundle struct {
 }
 
 // NewShim connects to the shim managing the bundle and tasks
-func (b *bundle) NewShim(ctx context.Context, binary, grpcAddress string, remote, debug bool) (*client.Client, error) {
+func (b *bundle) NewShim(ctx context.Context, binary, grpcAddress string, remote, debug bool, createOpts runtime.CreateOpts) (*client.Client, error) {
 	opt := client.WithStart(binary, grpcAddress, debug)
 	if !remote {
 		opt = client.WithLocal
 	}
+	var options runcopts.CreateOptions
+	if createOpts.Options != nil {
+		v, err := typeurl.UnmarshalAny(createOpts.Options)
+		if err != nil {
+			return nil, err
+		}
+		options = *v.(*runcopts.CreateOptions)
+	}
 	return client.New(ctx, client.Config{
 		Address:    b.shimAddress(),
 		Path:       b.path,
 		Namespace:  b.namespace,
+		CgroupPath: options.ShimCgroup,
 	}, opt)
 }
 

linux/runcopts/runc.pb.go

@@ -54,6 +54,7 @@ type CreateOptions struct {
 	EmptyNamespaces     []string `protobuf:"bytes,6,rep,name=empty_namespaces,json=emptyNamespaces" json:"empty_namespaces,omitempty"`
 	CgroupsMode         string   `protobuf:"bytes,7,opt,name=cgroups_mode,json=cgroupsMode,proto3" json:"cgroups_mode,omitempty"`
 	NoNewKeyring        bool     `protobuf:"varint,8,opt,name=no_new_keyring,json=noNewKeyring,proto3" json:"no_new_keyring,omitempty"`
+	ShimCgroup          string   `protobuf:"bytes,9,opt,name=shim_cgroup,json=shimCgroup,proto3" json:"shim_cgroup,omitempty"`
 }
 
 func (m *CreateOptions) Reset()                    { *m = CreateOptions{} }
@@ -205,6 +206,12 @@ func (m *CreateOptions) MarshalTo(dAtA []byte) (int, error) {
 		}
 		i++
 	}
+	if len(m.ShimCgroup) > 0 {
+		dAtA[i] = 0x4a
+		i++
+		i = encodeVarintRunc(dAtA, i, uint64(len(m.ShimCgroup)))
+		i += copy(dAtA[i:], m.ShimCgroup)
+	}
 	return i, nil
 }
 
@@ -369,6 +376,10 @@ func (m *CreateOptions) Size() (n int) {
 	if m.NoNewKeyring {
 		n += 2
 	}
+	l = len(m.ShimCgroup)
+	if l > 0 {
+		n += 1 + l + sovRunc(uint64(l))
+	}
 	return n
 }
 
@@ -440,6 +451,7 @@ func (this *CreateOptions) String() string {
 		`EmptyNamespaces:` + fmt.Sprintf("%v", this.EmptyNamespaces) + `,`,
 		`CgroupsMode:` + fmt.Sprintf("%v", this.CgroupsMode) + `,`,
 		`NoNewKeyring:` + fmt.Sprintf("%v", this.NoNewKeyring) + `,`,
+		`ShimCgroup:` + fmt.Sprintf("%v", this.ShimCgroup) + `,`,
 		`}`,
 	}, "")
 	return s
@@ -783,6 +795,35 @@ func (m *CreateOptions) Unmarshal(dAtA []byte) error {
 				}
 			}
 			m.NoNewKeyring = bool(v != 0)
+		case 9:
+			if wireType != 2 {
+				return fmt.Errorf("proto: wrong wireType = %d for field ShimCgroup", wireType)
+			}
+			var stringLen uint64
+			for shift := uint(0); ; shift += 7 {
+				if shift >= 64 {
+					return ErrIntOverflowRunc
+				}
+				if iNdEx >= l {
+					return io.ErrUnexpectedEOF
+				}
+				b := dAtA[iNdEx]
+				iNdEx++
+				stringLen |= (uint64(b) & 0x7F) << shift
+				if b < 0x80 {
+					break
+				}
+			}
+			intStringLen := int(stringLen)
+			if intStringLen < 0 {
+				return ErrInvalidLengthRunc
+			}
+			postIndex := iNdEx + intStringLen
+			if postIndex > l {
+				return io.ErrUnexpectedEOF
+			}
+			m.ShimCgroup = string(dAtA[iNdEx:postIndex])
+			iNdEx = postIndex
 		default:
 			iNdEx = preIndex
 			skippy, err := skipRunc(dAtA[iNdEx:])
@@ -1122,32 +1163,33 @@ func init() {
 }
 
 var fileDescriptorRunc = []byte{
-	// 428 bytes of a gzipped FileDescriptorProto
+	// 438 bytes of a gzipped FileDescriptorProto
-	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0xdc, 0x93, 0x31, 0x6f, 0xd4, 0x4c,
+	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0xdc, 0x93, 0xb1, 0x6f, 0xd4, 0x30,
-	0x10, 0x86, 0xe3, 0x24, 0x5f, 0xe2, 0xdb, 0xcb, 0xe5, 0x83, 0x85, 0x48, 0x26, 0x08, 0x2b, 0x9c,
+	0x18, 0xc5, 0x9b, 0xb6, 0xb4, 0x89, 0xaf, 0x57, 0xc0, 0x50, 0x29, 0x14, 0x11, 0xca, 0x09, 0xa4,
-	0x40, 0x0a, 0xcd, 0x9d, 0x04, 0x0d, 0x82, 0x8e, 0x2b, 0x81, 0x10, 0x0c, 0x34, 0x34, 0x2b, 0x67,
+	0xb2, 0xdc, 0x49, 0xb0, 0x20, 0xd8, 0xb8, 0x11, 0x28, 0x25, 0xc0, 0xc2, 0x62, 0xa5, 0xbe, 0x8f,
-	0x6f, 0xf0, 0xad, 0xce, 0x9e, 0x59, 0xed, 0x8e, 0x89, 0xaf, 0xe3, 0x2f, 0xf1, 0x2f, 0x52, 0x52,
+	0xc4, 0xba, 0xe4, 0xfb, 0x2c, 0xdb, 0xa1, 0xb9, 0x8d, 0x3f, 0xaf, 0x23, 0x62, 0x62, 0xa4, 0xf9,
-	0x52, 0x92, 0xfb, 0x23, 0x20, 0xaf, 0xe3, 0x40, 0x4b, 0x4b, 0xf7, 0xce, 0x33, 0x8f, 0xc6, 0xd2,
+	0x47, 0x40, 0x71, 0x2e, 0x85, 0x95, 0x95, 0xed, 0xf9, 0xf7, 0x9e, 0x9e, 0xa5, 0x27, 0x7d, 0xec,
-	0x2b, 0xaf, 0x78, 0x56, 0x18, 0x5e, 0xd4, 0x67, 0x13, 0x4d, 0xd5, 0x54, 0x13, 0x72, 0x6e, 0x10,
+	0x79, 0xae, 0x5c, 0x51, 0x9f, 0x4d, 0x25, 0x55, 0x33, 0x49, 0xe8, 0x32, 0x85, 0x60, 0x16, 0x7f,
-	0xdc, 0xfc, 0xcf, 0x58, 0x1a, 0xac, 0x9b, 0xa9, 0xab, 0x51, 0x93, 0x65, 0x1f, 0xc2, 0xc4, 0x3a,
+	0xcb, 0x52, 0x61, 0xdd, 0xcc, 0x4c, 0x8d, 0x92, 0xb4, 0xb3, 0x5e, 0x4c, 0xb5, 0x21, 0x47, 0xfc,
-	0x62, 0x92, 0x07, 0xbf, 0xad, 0x49, 0xb0, 0x26, 0xed, 0xf2, 0xf0, 0x76, 0x41, 0x05, 0x05, 0x63,
+	0xe0, 0x4f, 0x6a, 0xea, 0x53, 0xd3, 0xce, 0x3c, 0xbc, 0x9d, 0x53, 0x4e, 0x3e, 0x31, 0xeb, 0x54,
-	0xda, 0xa6, 0x4e, 0x1e, 0xbf, 0x15, 0xc3, 0xac, 0x46, 0xfd, 0xc6, 0xb2, 0x21, 0xf4, 0xf2, 0xae,
+	0x1f, 0x9e, 0xbc, 0x63, 0xa3, 0xb4, 0x46, 0xf9, 0x56, 0x3b, 0x45, 0x68, 0xf9, 0x5d, 0x16, 0x49,
-	0x18, 0x68, 0x67, 0x6a, 0x65, 0x73, 0x5e, 0x24, 0xd1, 0x51, 0x74, 0x3c, 0xc8, 0xe2, 0x16, 0x9c,
+	0xa3, 0x6a, 0xa1, 0x33, 0x57, 0xc4, 0xc1, 0x51, 0x70, 0x1c, 0xa5, 0x61, 0x07, 0x4e, 0x33, 0x57,
-	0xe6, 0xbc, 0x90, 0x0f, 0xc5, 0xbe, 0x5f, 0x79, 0x86, 0x6a, 0xae, 0x74, 0xe1, 0xa8, 0xb6, 0xc9,
+	0xf0, 0x47, 0x6c, 0xdf, 0xae, 0xac, 0x83, 0x6a, 0x21, 0x64, 0x6e, 0xa8, 0xd6, 0xf1, 0xa6, 0x4f,
-	0x66, 0x30, 0x46, 0x57, 0x74, 0x16, 0xe0, 0xf8, 0xeb, 0xa6, 0x18, 0xcd, 0x1c, 0xe4, 0x0c, 0xfd,
+	0x8c, 0xd7, 0x74, 0xee, 0xe1, 0xe4, 0xfb, 0x26, 0x1b, 0xcf, 0x0d, 0x64, 0x0e, 0x86, 0xd6, 0x09,
-	0xd5, 0xb1, 0x18, 0x21, 0x29, 0x6b, 0x3e, 0x13, 0x2b, 0x47, 0xc4, 0xe1, 0x72, 0x9c, 0x0d, 0x91,
+	0x1b, 0x23, 0x09, 0xad, 0xbe, 0x90, 0x13, 0x86, 0xc8, 0xf9, 0xe6, 0x30, 0x1d, 0x21, 0x9d, 0x76,
-	0x4e, 0x5b, 0x96, 0x11, 0xb1, 0xbc, 0x23, 0x62, 0xb2, 0x80, 0x8a, 0x75, 0x77, 0x36, 0xce, 0x76,
+	0x2c, 0x25, 0x72, 0xfc, 0x0e, 0x0b, 0x49, 0x03, 0x0a, 0x27, 0xfb, 0xda, 0x30, 0xdd, 0xed, 0xde,
-	0xdb, 0xf9, 0xbd, 0xb6, 0xf2, 0xb1, 0x38, 0x80, 0x86, 0xc1, 0x61, 0x5e, 0xaa, 0x1a, 0x4d, 0xa3,
+	0x1f, 0xa4, 0xe6, 0x4f, 0xd8, 0x01, 0x34, 0x0e, 0x0c, 0x66, 0xa5, 0xa8, 0x51, 0x35, 0xc2, 0x92,
-	0x3c, 0xe9, 0x25, 0xb0, 0x4f, 0xb6, 0x82, 0x77, 0xab, 0x5f, 0x7e, 0x40, 0xd3, 0xbc, 0xeb, 0x56,
+	0x5c, 0x82, 0xb3, 0xf1, 0x96, 0xcf, 0xdd, 0x1a, 0xcc, 0x8f, 0xa8, 0x9a, 0xf7, 0xbd, 0xc5, 0x0f,
-	0xf2, 0x50, 0xc4, 0x0c, 0xae, 0x32, 0x98, 0x97, 0xc9, 0x76, 0xd0, 0xae, 0x67, 0x79, 0x4f, 0x88,
+	0x59, 0xe8, 0xc0, 0x54, 0x0a, 0xb3, 0x32, 0xde, 0xf6, 0xb1, 0xab, 0x37, 0xbf, 0xc7, 0xd8, 0x67,
-	0x4f, 0xa6, 0x04, 0x55, 0x92, 0x5e, 0xfa, 0xe4, 0xbf, 0xb0, 0x1d, 0xb4, 0xe4, 0x55, 0x0b, 0xe4,
+	0x55, 0x82, 0x28, 0x49, 0x2e, 0x6d, 0x7c, 0xcd, 0xbb, 0x51, 0x47, 0x5e, 0x77, 0x80, 0x3f, 0x66,
-	0x23, 0x71, 0x03, 0x2a, 0xcb, 0x2b, 0x85, 0x79, 0x05, 0xde, 0xe6, 0x1a, 0x7c, 0xb2, 0x73, 0xb4,
+	0x37, 0xa0, 0xd2, 0x6e, 0x25, 0x30, 0xab, 0xc0, 0xea, 0x4c, 0x82, 0x8d, 0x77, 0x8e, 0xb6, 0x8e,
-	0x75, 0x3c, 0xc8, 0xfe, 0x0f, 0xfc, 0xe4, 0x1a, 0xcb, 0xfb, 0x62, 0xaf, 0x6b, 0xc2, 0xab, 0x8a,
+	0xa3, 0xf4, 0xba, 0xe7, 0x27, 0x57, 0x98, 0x3f, 0x60, 0x7b, 0xfd, 0x12, 0x56, 0x54, 0xb4, 0x80,
-	0xe6, 0x90, 0xec, 0x86, 0x3e, 0x86, 0x57, 0xec, 0x35, 0xcd, 0x41, 0x3e, 0x10, 0xfb, 0x48, 0x0a,
+	0x78, 0xd7, 0xef, 0x31, 0x5a, 0xb3, 0x37, 0xb4, 0x00, 0xfe, 0x90, 0xed, 0x23, 0x09, 0x84, 0x73,
-	0xe1, 0x5c, 0x2d, 0x61, 0xe5, 0x0c, 0x16, 0x49, 0x1c, 0x3e, 0xb8, 0x87, 0x74, 0x02, 0xe7, 0x2f,
+	0xb1, 0x84, 0x95, 0x51, 0x98, 0xc7, 0xa1, 0xff, 0x70, 0x0f, 0xe9, 0x04, 0xce, 0x5f, 0xf5, 0x8c,
-	0x3b, 0x36, 0xfe, 0x19, 0x89, 0x9b, 0xb3, 0x05, 0xe8, 0xa5, 0x25, 0x83, 0xdc, 0xf7, 0x26, 0xc5,
+	0xdf, 0x67, 0x23, 0x5b, 0xa8, 0x6a, 0xd8, 0x35, 0xf2, 0x3d, 0xac, 0x43, 0xeb, 0x51, 0x7f, 0x05,
-	0x36, 0x34, 0xa6, 0xaf, 0x2b, 0xe4, 0x7f, 0xb5, 0xa7, 0x17, 0xd9, 0xc5, 0x65, 0xba, 0xf1, 0xfd,
+	0xec, 0xe6, 0xbc, 0x00, 0xb9, 0xd4, 0xa4, 0xd0, 0x0d, 0xc3, 0x72, 0xb6, 0x0d, 0x8d, 0x1a, 0xf6,
-	0x32, 0xdd, 0xf8, 0xb2, 0x4e, 0xa3, 0x8b, 0x75, 0x1a, 0x7d, 0x5b, 0xa7, 0xd1, 0x8f, 0x75, 0x1a,
+	0xf4, 0xfa, 0x7f, 0x1d, 0xf2, 0x65, 0x7a, 0x71, 0x99, 0x6c, 0xfc, 0xb8, 0x4c, 0x36, 0xbe, 0xb6,
-	0x7d, 0x7c, 0xfa, 0x97, 0x6f, 0xe1, 0x79, 0x1f, 0xce, 0x76, 0xc2, 0x3f, 0xfe, 0xe4, 0x57, 0x00,
+	0x49, 0x70, 0xd1, 0x26, 0xc1, 0xb7, 0x36, 0x09, 0x7e, 0xb6, 0x49, 0xf0, 0xe9, 0xd9, 0x3f, 0x1e,
-	0x00, 0x00, 0xff, 0xff, 0x9d, 0xd8, 0x49, 0xfe, 0x4e, 0x03, 0x00, 0x00,
+	0xcb, 0x8b, 0x41, 0x9c, 0xed, 0xf8, 0x23, 0x78, 0xfa, 0x3b, 0x00, 0x00, 0xff, 0xff, 0x00, 0x19,
+	0xba, 0x8f, 0x6f, 0x03, 0x00, 0x00,
 }

linux/runcopts/runc.proto

@@ -20,6 +20,7 @@ message CreateOptions {
 	repeated string empty_namespaces = 6;
 	string cgroups_mode = 7;
 	bool no_new_keyring = 8;
+	string shim_cgroup = 9;
 }
 
 message CheckpointOptions {

linux/runtime.go

@@ -139,7 +139,7 @@ func (r *Runtime) Create(ctx context.Context, id string, opts runtime.CreateOpts
 			bundle.Delete()
 		}
 	}()
-	s, err := bundle.NewShim(ctx, r.shim, r.address, r.remote, r.shimDebug)
+	s, err := bundle.NewShim(ctx, r.shim, r.address, r.remote, r.shimDebug, opts)
 	if err != nil {
 		return nil, err
 	}

linux/shim/client.go

@@ -28,7 +28,7 @@ type ClientOpt func(context.Context, Config) (shim.ShimClient, io.Closer, error)
 
 // WithStart executes a new shim process
 func WithStart(binary, address string, debug bool) ClientOpt {
-	return func(ctx context.Context, config Config) (shim.ShimClient, io.Closer, error) {
+	return func(ctx context.Context, config Config) (_ shim.ShimClient, _ io.Closer, err error) {
 		socket, err := newSocket(config)
 		if err != nil {
 			return nil, nil, err
@@ -44,16 +44,36 @@ func WithStart(binary, address string, debug bool) ClientOpt {
 		if err := reaper.Default.Start(cmd); err != nil {
 			return nil, nil, errors.Wrapf(err, "failed to start shim")
 		}
+		defer func() {
+			if err != nil {
+				terminate(cmd)
+			}
+		}()
 		log.G(ctx).WithFields(logrus.Fields{
 			"pid":     cmd.Process.Pid,
 			"address": config.Address,
 			"debug":   debug,
 		}).Infof("shim %s started", binary)
+		// set shim in cgroup if it is provided
+		if config.CgroupPath != "" {
+			if err := setCgroup(ctx, config, cmd); err != nil {
+				return nil, nil, err
+			}
+		}
 		if err = sys.SetOOMScore(cmd.Process.Pid, sys.OOMScoreMaxKillable); err != nil {
 			return nil, nil, errors.Wrap(err, "failed to set OOM Score on shim")
 		}
-		return WithConnect(ctx, config)
+		c, clo, err := WithConnect(ctx, config)
+		if err != nil {
+			return nil, nil, errors.Wrap(err, "failed to connect")
 		}
+		return c, clo, nil
+	}
+}
+
+func terminate(cmd *exec.Cmd) {
+	cmd.Process.Kill()
+	reaper.Default.Wait(cmd)
 }
 
 func newCommand(binary, address string, debug bool, config Config, socket *os.File) *exec.Cmd {
@@ -141,6 +161,7 @@ type Config struct {
 	Address    string
 	Path       string
 	Namespace  string
+	CgroupPath string
 }
 
 // New returns a new shim client

linux/shim/client_linux.go

@@ -2,9 +2,35 @@
 
 package shim
 
-import "syscall"
+import (
+	"context"
+	"os/exec"
+	"syscall"
+
+	"github.com/containerd/cgroups"
+	"github.com/containerd/containerd/log"
+	"github.com/pkg/errors"
+	"github.com/sirupsen/logrus"
+)
 
 var atter = syscall.SysProcAttr{
 	Cloneflags: syscall.CLONE_NEWNS,
 	Setpgid:    true,
 }
+
+func setCgroup(ctx context.Context, config Config, cmd *exec.Cmd) error {
+	cg, err := cgroups.Load(cgroups.V1, cgroups.StaticPath(config.CgroupPath))
+	if err != nil {
+		return errors.Wrapf(err, "failed to load cgroup %s", config.CgroupPath)
+	}
+	if err := cg.Add(cgroups.Process{
+		Pid: cmd.Process.Pid,
+	}); err != nil {
+		return errors.Wrapf(err, "failed to join cgroup %s", config.CgroupPath)
+	}
+	log.G(ctx).WithFields(logrus.Fields{
+		"pid":     cmd.Process.Pid,
+		"address": config.Address,
+	}).Infof("shim placed in cgroup %s", config.CgroupPath)
+	return nil
+}

linux/shim/client_unix.go

@@ -2,8 +2,16 @@
 
 package shim
 
-import "syscall"
+import (
+	"context"
+	"os/exec"
+	"syscall"
+)
 
 var atter = syscall.SysProcAttr{
 	Setpgid: true,
 }
+
+func setCgroup(ctx context.Context, config Config, cmd *exec.Cmd) error {
+	return nil
+}