cri-o/server/server.go
Alexander Larsson 7bb957bf75 Implement non-terminal attach
We use a SOCK_SEQPACKET socket for the attach unix domain socket, which
means the kernel will ensure that the reading side only ever get the
data from one write operation. We use this for frameing, where the
first byte is the pipe that the next bytes are for. We have to make sure
that all reads from the socket are using at least the same size of buffer
as the write side, because otherwise the extra data in the message
will be dropped.

This also adds a stdin pipe for the container, similar to the ones we
use for stdout/err, because we need a way for an attached client
to write to stdin, even if not using a tty.

This fixes https://github.com/kubernetes-incubator/cri-o/issues/569

Signed-off-by: Alexander Larsson <alexl@redhat.com>
2017-06-14 22:59:50 +02:00

716 lines
20 KiB
Go

package server
import (
"encoding/json"
"fmt"
"io/ioutil"
"net"
"os"
"path/filepath"
"sync"
"time"
"github.com/Sirupsen/logrus"
"github.com/containers/image/types"
sstorage "github.com/containers/storage"
"github.com/docker/docker/pkg/ioutils"
"github.com/docker/docker/pkg/registrar"
"github.com/docker/docker/pkg/truncindex"
"github.com/kubernetes-incubator/cri-o/oci"
"github.com/kubernetes-incubator/cri-o/pkg/annotations"
"github.com/kubernetes-incubator/cri-o/pkg/ocicni"
"github.com/kubernetes-incubator/cri-o/pkg/storage"
"github.com/kubernetes-incubator/cri-o/server/apparmor"
"github.com/kubernetes-incubator/cri-o/server/seccomp"
rspec "github.com/opencontainers/runtime-spec/specs-go"
"github.com/opencontainers/selinux/go-selinux/label"
knet "k8s.io/apimachinery/pkg/util/net"
pb "k8s.io/kubernetes/pkg/kubelet/api/v1alpha1/runtime"
"k8s.io/kubernetes/pkg/kubelet/server/streaming"
)
const (
runtimeAPIVersion = "v1alpha1"
shutdownFile = "/var/lib/crio/crio.shutdown"
)
func isTrue(annotaton string) bool {
return annotaton == "true"
}
// streamService implements streaming.Runtime.
type streamService struct {
runtimeServer *Server // needed by Exec() endpoint
streamServer streaming.Server
streaming.Runtime
}
// Server implements the RuntimeService and ImageService
type Server struct {
config Config
runtime *oci.Runtime
store sstorage.Store
storageImageServer storage.ImageServer
storageRuntimeServer storage.RuntimeServer
stateLock sync.Mutex
updateLock sync.RWMutex
state *serverState
netPlugin ocicni.CNIPlugin
podNameIndex *registrar.Registrar
podIDIndex *truncindex.TruncIndex
ctrNameIndex *registrar.Registrar
ctrIDIndex *truncindex.TruncIndex
imageContext *types.SystemContext
seccompEnabled bool
seccompProfile seccomp.Seccomp
appArmorEnabled bool
appArmorProfile string
stream streamService
}
// GetExec returns exec stream request
func (s *Server) GetExec(req *pb.ExecRequest) (*pb.ExecResponse, error) {
return s.stream.streamServer.GetExec(req)
}
// GetAttach returns attach stream request
func (s *Server) GetAttach(req *pb.AttachRequest) (*pb.AttachResponse, error) {
return s.stream.streamServer.GetAttach(req)
}
// GetPortForward returns port forward stream request
func (s *Server) GetPortForward(req *pb.PortForwardRequest) (*pb.PortForwardResponse, error) {
return s.stream.streamServer.GetPortForward(req)
}
func (s *Server) loadContainer(id string) error {
config, err := s.store.FromContainerDirectory(id, "config.json")
if err != nil {
return err
}
var m rspec.Spec
if err = json.Unmarshal(config, &m); err != nil {
return err
}
labels := make(map[string]string)
if err = json.Unmarshal([]byte(m.Annotations[annotations.Labels]), &labels); err != nil {
return err
}
name := m.Annotations[annotations.Name]
name, err = s.reserveContainerName(id, name)
if err != nil {
return err
}
defer func() {
if err != nil {
s.releaseContainerName(name)
}
}()
var metadata pb.ContainerMetadata
if err = json.Unmarshal([]byte(m.Annotations[annotations.Metadata]), &metadata); err != nil {
return err
}
sb := s.getSandbox(m.Annotations[annotations.SandboxID])
if sb == nil {
return fmt.Errorf("could not get sandbox with id %s, skipping", m.Annotations[annotations.SandboxID])
}
tty := isTrue(m.Annotations[annotations.TTY])
stdin := isTrue(m.Annotations[annotations.Stdin])
stdinOnce := isTrue(m.Annotations[annotations.StdinOnce])
containerPath, err := s.store.ContainerRunDirectory(id)
if err != nil {
return err
}
containerDir, err := s.store.ContainerDirectory(id)
if err != nil {
return err
}
var img *pb.ImageSpec
image, ok := m.Annotations[annotations.Image]
if ok {
img = &pb.ImageSpec{
Image: image,
}
}
kubeAnnotations := make(map[string]string)
if err = json.Unmarshal([]byte(m.Annotations[annotations.Annotations]), &kubeAnnotations); err != nil {
return err
}
created, err := time.Parse(time.RFC3339Nano, m.Annotations[annotations.Created])
if err != nil {
return err
}
ctr, err := oci.NewContainer(id, name, containerPath, m.Annotations[annotations.LogPath], sb.netNs(), labels, kubeAnnotations, img, &metadata, sb.id, tty, stdin, stdinOnce, sb.privileged, containerDir, created, m.Annotations["org.opencontainers.image.stopSignal"])
if err != nil {
return err
}
s.containerStateFromDisk(ctr)
s.addContainer(ctr)
return s.ctrIDIndex.Add(id)
}
func (s *Server) containerStateFromDisk(c *oci.Container) error {
if err := c.FromDisk(); err != nil {
return err
}
// ignore errors, this is a best effort to have up-to-date info about
// a given container before its state gets stored
s.runtime.UpdateStatus(c)
return nil
}
func (s *Server) containerStateToDisk(c *oci.Container) error {
// ignore errors, this is a best effort to have up-to-date info about
// a given container before its state gets stored
s.runtime.UpdateStatus(c)
jsonSource, err := ioutils.NewAtomicFileWriter(c.StatePath(), 0644)
if err != nil {
return err
}
defer jsonSource.Close()
enc := json.NewEncoder(jsonSource)
return enc.Encode(s.runtime.ContainerStatus(c))
}
func configNetNsPath(spec rspec.Spec) (string, error) {
for _, ns := range spec.Linux.Namespaces {
if ns.Type != rspec.NetworkNamespace {
continue
}
if ns.Path == "" {
return "", fmt.Errorf("empty networking namespace")
}
return ns.Path, nil
}
return "", fmt.Errorf("missing networking namespace")
}
func (s *Server) loadSandbox(id string) error {
config, err := s.store.FromContainerDirectory(id, "config.json")
if err != nil {
return err
}
var m rspec.Spec
if err = json.Unmarshal(config, &m); err != nil {
return err
}
labels := make(map[string]string)
if err = json.Unmarshal([]byte(m.Annotations[annotations.Labels]), &labels); err != nil {
return err
}
name := m.Annotations[annotations.Name]
name, err = s.reservePodName(id, name)
if err != nil {
return err
}
defer func() {
if err != nil {
s.releasePodName(name)
}
}()
var metadata pb.PodSandboxMetadata
if err = json.Unmarshal([]byte(m.Annotations[annotations.Metadata]), &metadata); err != nil {
return err
}
processLabel, mountLabel, err := label.InitLabels(label.DupSecOpt(m.Process.SelinuxLabel))
if err != nil {
return err
}
kubeAnnotations := make(map[string]string)
if err = json.Unmarshal([]byte(m.Annotations[annotations.Annotations]), &kubeAnnotations); err != nil {
return err
}
privileged := isTrue(m.Annotations[annotations.PrivilegedRuntime])
sb := &sandbox{
id: id,
name: name,
kubeName: m.Annotations[annotations.KubeName],
logDir: filepath.Dir(m.Annotations[annotations.LogPath]),
labels: labels,
containers: oci.NewMemoryStore(),
processLabel: processLabel,
mountLabel: mountLabel,
annotations: kubeAnnotations,
metadata: &metadata,
shmPath: m.Annotations[annotations.ShmPath],
privileged: privileged,
resolvPath: m.Annotations[annotations.ResolvPath],
}
// We add a netNS only if we can load a permanent one.
// Otherwise, the sandbox will live in the host namespace.
netNsPath, err := configNetNsPath(m)
if err == nil {
netNS, nsErr := netNsGet(netNsPath, sb.name)
// If we can't load the networking namespace
// because it's closed, we just set the sb netns
// pointer to nil. Otherwise we return an error.
if nsErr != nil && nsErr != errSandboxClosedNetNS {
return nsErr
}
sb.netns = netNS
}
s.addSandbox(sb)
defer func() {
if err != nil {
s.removeSandbox(sb.id)
}
}()
sandboxPath, err := s.store.ContainerRunDirectory(id)
if err != nil {
return err
}
sandboxDir, err := s.store.ContainerDirectory(id)
if err != nil {
return err
}
cname, err := s.reserveContainerName(m.Annotations[annotations.ContainerID], m.Annotations[annotations.ContainerName])
if err != nil {
return err
}
defer func() {
if err != nil {
s.releaseContainerName(cname)
}
}()
created, err := time.Parse(time.RFC3339Nano, m.Annotations[annotations.Created])
if err != nil {
return err
}
scontainer, err := oci.NewContainer(m.Annotations[annotations.ContainerID], cname, sandboxPath, m.Annotations[annotations.LogPath], sb.netNs(), labels, kubeAnnotations, nil, nil, id, false, false, false, privileged, sandboxDir, created, m.Annotations["org.opencontainers.image.stopSignal"])
if err != nil {
return err
}
s.containerStateFromDisk(scontainer)
if err = label.ReserveLabel(processLabel); err != nil {
return err
}
sb.infraContainer = scontainer
if err = s.ctrIDIndex.Add(scontainer.ID()); err != nil {
return err
}
if err = s.podIDIndex.Add(id); err != nil {
return err
}
return nil
}
func (s *Server) restore() {
containers, err := s.store.Containers()
if err != nil && !os.IsNotExist(err) {
logrus.Warnf("could not read containers and sandboxes: %v", err)
}
pods := map[string]*storage.RuntimeContainerMetadata{}
podContainers := map[string]*storage.RuntimeContainerMetadata{}
for _, container := range containers {
metadata, err2 := s.storageRuntimeServer.GetContainerMetadata(container.ID)
if err2 != nil {
logrus.Warnf("error parsing metadata for %s: %v, ignoring", container.ID, err2)
continue
}
if metadata.Pod {
pods[container.ID] = &metadata
} else {
podContainers[container.ID] = &metadata
}
}
for containerID, metadata := range pods {
if err = s.loadSandbox(containerID); err != nil {
logrus.Warnf("could not restore sandbox %s container %s: %v", metadata.PodID, containerID, err)
}
}
for containerID := range podContainers {
if err := s.loadContainer(containerID); err != nil {
logrus.Warnf("could not restore container %s: %v", containerID, err)
}
}
}
// Update makes changes to the server's state (lists of pods and containers) to
// reflect the list of pods and containers that are stored on disk, possibly
// having been modified by other parties
func (s *Server) Update() {
logrus.Debugf("updating sandbox and container information")
if err := s.update(); err != nil {
logrus.Errorf("error updating sandbox and container information: %v", err)
}
}
func (s *Server) update() error {
s.updateLock.Lock()
defer s.updateLock.Unlock()
containers, err := s.store.Containers()
if err != nil && !os.IsNotExist(err) {
logrus.Warnf("could not read containers and sandboxes: %v", err)
return err
}
newPods := map[string]*storage.RuntimeContainerMetadata{}
oldPods := map[string]string{}
removedPods := map[string]string{}
newPodContainers := map[string]*storage.RuntimeContainerMetadata{}
oldPodContainers := map[string]string{}
removedPodContainers := map[string]string{}
for _, container := range containers {
if s.hasSandbox(container.ID) {
// FIXME: do we need to reload/update any info about the sandbox?
oldPods[container.ID] = container.ID
oldPodContainers[container.ID] = container.ID
continue
}
if s.getContainer(container.ID) != nil {
// FIXME: do we need to reload/update any info about the container?
oldPodContainers[container.ID] = container.ID
continue
}
// not previously known, so figure out what it is
metadata, err2 := s.storageRuntimeServer.GetContainerMetadata(container.ID)
if err2 != nil {
logrus.Errorf("error parsing metadata for %s: %v, ignoring", container.ID, err2)
continue
}
if metadata.Pod {
newPods[container.ID] = &metadata
} else {
newPodContainers[container.ID] = &metadata
}
}
s.ctrIDIndex.Iterate(func(id string) {
if _, ok := oldPodContainers[id]; !ok {
// this container's ID wasn't in the updated list -> removed
removedPodContainers[id] = id
}
})
for removedPodContainer := range removedPodContainers {
// forget this container
c := s.getContainer(removedPodContainer)
if c == nil {
logrus.Warnf("bad state when getting container removed %+v", removedPodContainer)
continue
}
s.releaseContainerName(c.Name())
s.removeContainer(c)
if err = s.ctrIDIndex.Delete(c.ID()); err != nil {
return err
}
logrus.Debugf("forgetting removed pod container %s", c.ID())
}
s.podIDIndex.Iterate(func(id string) {
if _, ok := oldPods[id]; !ok {
// this pod's ID wasn't in the updated list -> removed
removedPods[id] = id
}
})
for removedPod := range removedPods {
// forget this pod
sb := s.getSandbox(removedPod)
if sb == nil {
logrus.Warnf("bad state when getting pod to remove %+v", removedPod)
continue
}
podInfraContainer := sb.infraContainer
s.releaseContainerName(podInfraContainer.Name())
s.removeContainer(podInfraContainer)
if err = s.ctrIDIndex.Delete(podInfraContainer.ID()); err != nil {
return err
}
sb.infraContainer = nil
s.releasePodName(sb.name)
s.removeSandbox(sb.id)
if err = s.podIDIndex.Delete(sb.id); err != nil {
return err
}
logrus.Debugf("forgetting removed pod %s", sb.id)
}
for sandboxID := range newPods {
// load this pod
if err = s.loadSandbox(sandboxID); err != nil {
logrus.Warnf("could not load new pod sandbox %s: %v, ignoring", sandboxID, err)
} else {
logrus.Debugf("loaded new pod sandbox %s", sandboxID, err)
}
}
for containerID := range newPodContainers {
// load this container
if err = s.loadContainer(containerID); err != nil {
logrus.Warnf("could not load new sandbox container %s: %v, ignoring", containerID, err)
} else {
logrus.Debugf("loaded new pod container %s", containerID, err)
}
}
return nil
}
func (s *Server) reservePodName(id, name string) (string, error) {
if err := s.podNameIndex.Reserve(name, id); err != nil {
if err == registrar.ErrNameReserved {
id, err := s.podNameIndex.Get(name)
if err != nil {
logrus.Warnf("conflict, pod name %q already reserved", name)
return "", err
}
return "", fmt.Errorf("conflict, name %q already reserved for pod %q", name, id)
}
return "", fmt.Errorf("error reserving pod name %q", name)
}
return name, nil
}
func (s *Server) releasePodName(name string) {
s.podNameIndex.Release(name)
}
func (s *Server) reserveContainerName(id, name string) (string, error) {
if err := s.ctrNameIndex.Reserve(name, id); err != nil {
if err == registrar.ErrNameReserved {
id, err := s.ctrNameIndex.Get(name)
if err != nil {
logrus.Warnf("conflict, ctr name %q already reserved", name)
return "", err
}
return "", fmt.Errorf("conflict, name %q already reserved for ctr %q", name, id)
}
return "", fmt.Errorf("error reserving ctr name %s", name)
}
return name, nil
}
func (s *Server) releaseContainerName(name string) {
s.ctrNameIndex.Release(name)
}
// cleanupSandboxesOnShutdown Remove all running Sandboxes on system shutdown
func (s *Server) cleanupSandboxesOnShutdown() {
_, err := os.Stat(shutdownFile)
if err == nil || !os.IsNotExist(err) {
logrus.Debugf("shutting down all sandboxes, on shutdown")
s.StopAllPodSandboxes()
err = os.Remove(shutdownFile)
if err != nil {
logrus.Warnf("Failed to remove %q", shutdownFile)
}
}
}
// Shutdown attempts to shut down the server's storage cleanly
func (s *Server) Shutdown() error {
// why do this on clean shutdown! we want containers left running when ocid
// is down for whatever reason no?!
// notice this won't trigger just on system halt but also on normal
// ocid.service restart!!!
s.cleanupSandboxesOnShutdown()
_, err := s.store.Shutdown(false)
return err
}
// New creates a new Server with options provided
func New(config *Config) (*Server, error) {
store, err := sstorage.GetStore(sstorage.StoreOptions{
RunRoot: config.RunRoot,
GraphRoot: config.Root,
GraphDriverName: config.Storage,
GraphDriverOptions: config.StorageOptions,
})
if err != nil {
return nil, err
}
imageService, err := storage.GetImageService(store, config.DefaultTransport, config.InsecureRegistries)
if err != nil {
return nil, err
}
storageRuntimeService := storage.GetRuntimeService(imageService, config.PauseImage)
if err != nil {
return nil, err
}
if err := os.MkdirAll("/var/run/crio", 0755); err != nil {
return nil, err
}
r, err := oci.New(config.Runtime, config.RuntimeHostPrivileged, config.Conmon, config.ConmonEnv, config.CgroupManager)
if err != nil {
return nil, err
}
sandboxes := make(map[string]*sandbox)
containers := oci.NewMemoryStore()
netPlugin, err := ocicni.InitCNI(config.NetworkDir, config.PluginDir)
if err != nil {
return nil, err
}
s := &Server{
runtime: r,
store: store,
storageImageServer: imageService,
storageRuntimeServer: storageRuntimeService,
netPlugin: netPlugin,
config: *config,
state: &serverState{
sandboxes: sandboxes,
containers: containers,
},
seccompEnabled: seccomp.IsEnabled(),
appArmorEnabled: apparmor.IsEnabled(),
appArmorProfile: config.ApparmorProfile,
}
if s.seccompEnabled {
seccompProfile, fileErr := ioutil.ReadFile(config.SeccompProfile)
if fileErr != nil {
return nil, fmt.Errorf("opening seccomp profile (%s) failed: %v", config.SeccompProfile, fileErr)
}
var seccompConfig seccomp.Seccomp
if jsonErr := json.Unmarshal(seccompProfile, &seccompConfig); jsonErr != nil {
return nil, fmt.Errorf("decoding seccomp profile failed: %v", jsonErr)
}
s.seccompProfile = seccompConfig
}
if s.appArmorEnabled && s.appArmorProfile == apparmor.DefaultApparmorProfile {
if apparmorErr := apparmor.EnsureDefaultApparmorProfile(); apparmorErr != nil {
return nil, fmt.Errorf("ensuring the default apparmor profile is installed failed: %v", apparmorErr)
}
}
s.podIDIndex = truncindex.NewTruncIndex([]string{})
s.podNameIndex = registrar.NewRegistrar()
s.ctrIDIndex = truncindex.NewTruncIndex([]string{})
s.ctrNameIndex = registrar.NewRegistrar()
s.imageContext = &types.SystemContext{
SignaturePolicyPath: config.ImageConfig.SignaturePolicyPath,
}
s.restore()
s.cleanupSandboxesOnShutdown()
bindAddress := net.ParseIP(config.StreamAddress)
if bindAddress == nil {
bindAddress, err = knet.ChooseBindAddress(net.IP{0, 0, 0, 0})
if err != nil {
return nil, err
}
}
_, err = net.LookupPort("tcp", config.StreamPort)
if err != nil {
return nil, err
}
// Prepare streaming server
streamServerConfig := streaming.DefaultConfig
streamServerConfig.Addr = net.JoinHostPort(bindAddress.String(), config.StreamPort)
s.stream.runtimeServer = s
s.stream.streamServer, err = streaming.NewServer(streamServerConfig, s.stream)
if err != nil {
return nil, fmt.Errorf("unable to create streaming server")
}
// TODO: Is it should be started somewhere else?
go func() {
s.stream.streamServer.Start(true)
}()
logrus.Debugf("sandboxes: %v", s.state.sandboxes)
logrus.Debugf("containers: %v", s.state.containers)
return s, nil
}
type serverState struct {
sandboxes map[string]*sandbox
containers oci.ContainerStorer
}
func (s *Server) addSandbox(sb *sandbox) {
s.stateLock.Lock()
s.state.sandboxes[sb.id] = sb
s.stateLock.Unlock()
}
func (s *Server) getSandbox(id string) *sandbox {
s.stateLock.Lock()
sb := s.state.sandboxes[id]
s.stateLock.Unlock()
return sb
}
func (s *Server) hasSandbox(id string) bool {
s.stateLock.Lock()
_, ok := s.state.sandboxes[id]
s.stateLock.Unlock()
return ok
}
func (s *Server) removeSandbox(id string) {
s.stateLock.Lock()
delete(s.state.sandboxes, id)
s.stateLock.Unlock()
}
func (s *Server) addContainer(c *oci.Container) {
s.stateLock.Lock()
sandbox := s.state.sandboxes[c.Sandbox()]
// TODO(runcom): handle !ok above!!! otherwise it panics!
sandbox.addContainer(c)
s.state.containers.Add(c.ID(), c)
s.stateLock.Unlock()
}
func (s *Server) getContainer(id string) *oci.Container {
s.stateLock.Lock()
c := s.state.containers.Get(id)
s.stateLock.Unlock()
return c
}
// GetSandboxContainer returns the infra container for a given sandbox
func (s *Server) GetSandboxContainer(id string) *oci.Container {
sb := s.getSandbox(id)
return sb.infraContainer
}
// GetContainer returns a container by its ID
func (s *Server) GetContainer(id string) *oci.Container {
return s.getContainer(id)
}
func (s *Server) removeContainer(c *oci.Container) {
s.stateLock.Lock()
sandbox := s.state.sandboxes[c.Sandbox()]
sandbox.removeContainer(c)
s.state.containers.Delete(c.ID())
s.stateLock.Unlock()
}