diff options
author | Matthew Heon <matthew.heon@gmail.com> | 2018-01-18 11:37:41 -0500 |
---|---|---|
committer | Matthew Heon <matthew.heon@gmail.com> | 2018-01-18 11:48:20 -0500 |
commit | 64d2190ec1a0a763962cf6e4c8f8420fa88cc3e5 (patch) | |
tree | 9d7925cc6bc0debc036608d1b63658bad9f755e4 | |
parent | 0d69ca6637b30a3370529b3e272f27f6fafdb0c3 (diff) | |
download | podman-64d2190ec1a0a763962cf6e4c8f8420fa88cc3e5.tar.gz podman-64d2190ec1a0a763962cf6e4c8f8420fa88cc3e5.tar.bz2 podman-64d2190ec1a0a763962cf6e4c8f8420fa88cc3e5.zip |
Split container.go into three files
Weighing in at ~1700 lines, container.go is just too big. Split
it into three files: core structs and accessors (container.go),
public API (container_api.go), and internal functions
(container_internal.go).
Signed-off-by: Matthew Heon <matthew.heon@gmail.com>
-rw-r--r-- | libpod/container.go | 1110 | ||||
-rw-r--r-- | libpod/container_api.go | 754 | ||||
-rw-r--r-- | libpod/container_internal.go | 382 |
3 files changed, 1136 insertions, 1110 deletions
diff --git a/libpod/container.go b/libpod/container.go index 71de63fd1..2c2377327 100644 --- a/libpod/container.go +++ b/libpod/container.go @@ -1,39 +1,18 @@ package libpod import ( - "encoding/json" "fmt" - "io" - "io/ioutil" "net" - "os" "path/filepath" - "syscall" "time" "github.com/containerd/cgroups" "github.com/containernetworking/plugins/pkg/ns" "github.com/containers/storage" - "github.com/containers/storage/pkg/archive" "github.com/cri-o/ocicni/pkg/ocicni" - "github.com/docker/docker/daemon/caps" - "github.com/docker/docker/pkg/mount" - "github.com/docker/docker/pkg/namesgenerator" - "github.com/docker/docker/pkg/stringid" - "github.com/docker/docker/pkg/term" - "github.com/mrunalp/fileutils" spec "github.com/opencontainers/runtime-spec/specs-go" - "github.com/opencontainers/runtime-tools/generate" - "github.com/opencontainers/selinux/go-selinux/label" "github.com/pkg/errors" - "github.com/projectatomic/libpod/libpod/driver" - crioAnnotations "github.com/projectatomic/libpod/pkg/annotations" - "github.com/projectatomic/libpod/pkg/chrootuser" - "github.com/sirupsen/logrus" "github.com/ulule/deepcopier" - "golang.org/x/sys/unix" - "k8s.io/apimachinery/pkg/util/wait" - "k8s.io/client-go/tools/remotecommand" ) // ContainerState represents the current state of a container @@ -131,7 +110,6 @@ type Container struct { // TODO fetch IP and Subnet Mask from networks once we have updated OCICNI // TODO enable pod support // TODO Add readonly support -// TODO add SHM size support // containerRuntimeInfo contains the current state of the container // It is stored on disk in a tmpfs and recreated on reboot @@ -380,63 +358,6 @@ func (c *Container) RuntimeName() string { return c.runtime.ociRuntime.name } -// rootFsSize gets the size of the container's root filesystem -// A container FS is split into two parts. The first is the top layer, a -// mutable layer, and the rest is the RootFS: the set of immutable layers -// that make up the image on which the container is based. -func (c *Container) rootFsSize() (int64, error) { - container, err := c.runtime.store.Container(c.ID()) - if err != nil { - return 0, err - } - - // Ignore the size of the top layer. The top layer is a mutable RW layer - // and is not considered a part of the rootfs - rwLayer, err := c.runtime.store.Layer(container.LayerID) - if err != nil { - return 0, err - } - layer, err := c.runtime.store.Layer(rwLayer.Parent) - if err != nil { - return 0, err - } - - size := int64(0) - for layer.Parent != "" { - layerSize, err := c.runtime.store.DiffSize(layer.Parent, layer.ID) - if err != nil { - return size, errors.Wrapf(err, "getting diffsize of layer %q and its parent %q", layer.ID, layer.Parent) - } - size += layerSize - layer, err = c.runtime.store.Layer(layer.Parent) - if err != nil { - return 0, err - } - } - // Get the size of the last layer. Has to be outside of the loop - // because the parent of the last layer is "", andlstore.Get("") - // will return an error. - layerSize, err := c.runtime.store.DiffSize(layer.Parent, layer.ID) - return size + layerSize, err -} - -// rwSize Gets the size of the mutable top layer of the container. -func (c *Container) rwSize() (int64, error) { - container, err := c.runtime.store.Container(c.ID()) - if err != nil { - return 0, err - } - - // Get the size of the top layer by calculating the size of the diff - // between the layer and its parent. The top layer of a container is - // the only RW layer, all others are immutable - layer, err := c.runtime.store.Layer(container.LayerID) - if err != nil { - return 0, err - } - return c.runtime.store.DiffSize(layer.Parent, layer.ID) -} - // LogPath returns the path to the container's log file // This file will only be present after Init() is called to create the container // in runc @@ -587,1043 +508,12 @@ func (c *Container) NamespacePath(ns LinuxNS) (string, error) { return fmt.Sprintf("/proc/%d/ns/%s", c.state.PID, ns.String()), nil } -// The path to the container's root filesystem - where the OCI spec will be -// placed, amongst other things -func (c *Container) bundlePath() string { - return c.config.StaticDir -} - -// The path to the container's logs file -func (c *Container) logPath() string { - return filepath.Join(c.config.StaticDir, "ctr.log") -} - -// Retrieves the path of the container's attach socket -func (c *Container) attachSocketPath() string { - return filepath.Join(c.runtime.ociRuntime.socketsDir, c.ID(), "attach") -} - -// Sync this container with on-disk state and runc status -// Should only be called with container lock held -// This function should suffice to ensure a container's state is accurate and -// it is valid for use. -func (c *Container) syncContainer() error { - if err := c.runtime.state.UpdateContainer(c); err != nil { - return err - } - // If runc knows about the container, update its status in runc - // And then save back to disk - if (c.state.State != ContainerStateUnknown) && - (c.state.State != ContainerStateConfigured) { - oldState := c.state.State - // TODO: optionally replace this with a stat for the exit file - if err := c.runtime.ociRuntime.updateContainerStatus(c); err != nil { - return err - } - // Only save back to DB if state changed - if c.state.State != oldState { - if err := c.save(); err != nil { - return err - } - } - } - - if !c.valid { - return errors.Wrapf(ErrCtrRemoved, "container %s is not valid", c.ID()) - } - - return nil -} - -// Make a new container -func newContainer(rspec *spec.Spec, lockDir string) (*Container, error) { - if rspec == nil { - return nil, errors.Wrapf(ErrInvalidArg, "must provide a valid runtime spec to create container") - } - - ctr := new(Container) - ctr.config = new(ContainerConfig) - ctr.state = new(containerRuntimeInfo) - - ctr.config.ID = stringid.GenerateNonCryptoID() - ctr.config.Name = namesgenerator.GetRandomName(0) - - ctr.config.Spec = new(spec.Spec) - deepcopier.Copy(rspec).To(ctr.config.Spec) - ctr.config.CreatedTime = time.Now() - - ctr.config.ShmSize = DefaultShmSize - ctr.config.CgroupParent = CgroupParent - - // Path our lock file will reside at - lockPath := filepath.Join(lockDir, ctr.config.ID) - // Grab a lockfile at the given path - lock, err := storage.GetLockfile(lockPath) - if err != nil { - return nil, errors.Wrapf(err, "error creating lockfile for new container") - } - ctr.lock = lock - - return ctr, nil -} - -// Create container root filesystem for use -func (c *Container) setupStorage() error { - if !c.valid { - return errors.Wrapf(ErrCtrRemoved, "container %s is not valid", c.ID()) - } - - if c.state.State != ContainerStateConfigured { - return errors.Wrapf(ErrCtrStateInvalid, "container %s must be in Configured state to have storage set up", c.ID()) - } - - // Need both an image ID and image name, plus a bool telling us whether to use the image configuration - if c.config.RootfsImageID == "" || c.config.RootfsImageName == "" { - return errors.Wrapf(ErrInvalidArg, "must provide image ID and image name to use an image") - } - - containerInfo, err := c.runtime.storageService.CreateContainerStorage(c.runtime.imageContext, c.config.RootfsImageName, c.config.RootfsImageID, c.config.Name, c.config.ID, c.config.MountLabel) - if err != nil { - return errors.Wrapf(err, "error creating container storage") - } - - c.config.StaticDir = containerInfo.Dir - c.state.RunDir = containerInfo.RunDir - - artifacts := filepath.Join(c.config.StaticDir, artifactsDir) - if err := os.MkdirAll(artifacts, 0755); err != nil { - return errors.Wrapf(err, "error creating artifacts directory %q", artifacts) - } - - return nil -} - -// Tear down a container's storage prior to removal -func (c *Container) teardownStorage() error { - if !c.valid { - return errors.Wrapf(ErrCtrRemoved, "container %s is not valid", c.ID()) - } - - if c.state.State == ContainerStateRunning || c.state.State == ContainerStatePaused { - return errors.Wrapf(ErrCtrStateInvalid, "cannot remove storage for container %s as it is running or paused", c.ID()) - } - - artifacts := filepath.Join(c.config.StaticDir, artifactsDir) - if err := os.RemoveAll(artifacts); err != nil { - return errors.Wrapf(err, "error removing artifacts %q", artifacts) - } - - if err := c.cleanupStorage(); err != nil { - return errors.Wrapf(err, "failed to cleanup container %s storage", c.ID()) - } - - if err := c.runtime.storageService.DeleteContainer(c.ID()); err != nil { - return errors.Wrapf(err, "error removing container %s root filesystem", c.ID()) - } - - return nil -} - -// Refresh refreshes the container's state after a restart -func (c *Container) refresh() error { - c.lock.Lock() - defer c.lock.Unlock() - - if !c.valid { - return errors.Wrapf(ErrCtrRemoved, "container %s is not valid - may have been removed", c.ID()) - } - - // We need to get the container's temporary directory from c/storage - // It was lost in the reboot and must be recreated - dir, err := c.runtime.storageService.GetRunDir(c.ID()) - if err != nil { - return errors.Wrapf(err, "error retrieving temporary directory for container %s", c.ID()) - } - c.state.RunDir = dir - - if err := c.runtime.state.SaveContainer(c); err != nil { - return errors.Wrapf(err, "error refreshing state for container %s", c.ID()) - } - - return nil -} - -// Init creates a container in the OCI runtime -func (c *Container) Init() (err error) { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - } - - if c.state.State != ContainerStateConfigured { - return errors.Wrapf(ErrCtrExists, "container %s has already been created in runtime", c.ID()) - } - - if err := c.mountStorage(); err != nil { - return err - } - defer func() { - if err != nil { - if err2 := c.cleanupStorage(); err2 != nil { - logrus.Errorf("Error cleaning up storage for container %s: %v", c.ID(), err2) - } - } - }() - - // Make a network namespace for the container - if c.config.CreateNetNS && c.state.NetNS == nil { - if err := c.runtime.createNetNS(c); err != nil { - return err - } - } - defer func() { - if err != nil { - if err2 := c.runtime.teardownNetNS(c); err2 != nil { - logrus.Errorf("Error tearing down network namespace for container %s: %v", c.ID(), err2) - } - } - }() - - // If the OCI spec already exists, we need to replace it - // Cannot guarantee some things, e.g. network namespaces, have the same - // paths - jsonPath := filepath.Join(c.bundlePath(), "config.json") - if _, err := os.Stat(jsonPath); err != nil { - if !os.IsNotExist(err) { - return errors.Wrapf(err, "error doing stat on container %s spec", c.ID()) - } - // The spec does not exist, we're fine - } else { - // The spec exists, need to remove it - if err := os.Remove(jsonPath); err != nil { - return errors.Wrapf(err, "error replacing runtime spec for container %s", c.ID()) - } - } - - // Copy /etc/resolv.conf to the container's rundir - resolvPath := "/etc/resolv.conf" - - // Check if the host system is using system resolve and if so - // copy its resolv.conf - _, err = os.Stat("/run/systemd/resolve/resolv.conf") - if err == nil { - resolvPath = "/run/systemd/resolve/resolv.conf" - } - runDirResolv, err := c.copyHostFileToRundir(resolvPath) - if err != nil { - return errors.Wrapf(err, "unable to copy resolv.conf to ", runDirResolv) - } - // Copy /etc/hosts to the container's rundir - runDirHosts, err := c.copyHostFileToRundir("/etc/hosts") - if err != nil { - return errors.Wrapf(err, "unable to copy /etc/hosts to ", runDirHosts) - } - - // Save OCI spec to disk - g := generate.NewFromSpec(c.config.Spec) - // If network namespace was requested, add it now - if c.config.CreateNetNS { - g.AddOrReplaceLinuxNamespace(spec.NetworkNamespace, c.state.NetNS.Path()) - } - // Remove default /etc/shm mount - g.RemoveMount("/dev/shm") - // Mount ShmDir from host into container - shmMnt := spec.Mount{ - Type: "bind", - Source: c.config.ShmDir, - Destination: "/dev/shm", - Options: []string{"rw", "bind"}, - } - g.AddMount(shmMnt) - // Bind mount resolv.conf - resolvMnt := spec.Mount{ - Type: "bind", - Source: runDirResolv, - Destination: "/etc/resolv.conf", - Options: []string{"rw", "bind"}, - } - g.AddMount(resolvMnt) - // Bind mount hosts - hostsMnt := spec.Mount{ - Type: "bind", - Source: runDirHosts, - Destination: "/etc/hosts", - Options: []string{"rw", "bind"}, - } - g.AddMount(hostsMnt) - - if c.config.User != "" { - if !c.state.Mounted { - return errors.Wrapf(ErrCtrStateInvalid, "container %s must be mounted in order to translate User field", c.ID()) - } - uid, gid, err := chrootuser.GetUser(c.state.Mountpoint, c.config.User) - if err != nil { - return err - } - // User and Group must go together - g.SetProcessUID(uid) - g.SetProcessGID(gid) - } - - // Add shared namespaces from other containers - if c.config.IPCNsCtr != "" { - ipcCtr, err := c.runtime.state.Container(c.config.IPCNsCtr) - if err != nil { - return err - } - - nsPath, err := ipcCtr.NamespacePath(IPCNS) - if err != nil { - return err - } - - if err := g.AddOrReplaceLinuxNamespace(spec.IPCNamespace, nsPath); err != nil { - return err - } - } - if c.config.MountNsCtr != "" { - mountCtr, err := c.runtime.state.Container(c.config.MountNsCtr) - if err != nil { - return err - } - - nsPath, err := mountCtr.NamespacePath(MountNS) - if err != nil { - return err - } - - if err := g.AddOrReplaceLinuxNamespace(spec.MountNamespace, nsPath); err != nil { - return err - } - } - if c.config.NetNsCtr != "" { - netCtr, err := c.runtime.state.Container(c.config.NetNsCtr) - if err != nil { - return err - } - - nsPath, err := netCtr.NamespacePath(NetNS) - if err != nil { - return err - } - - if err := g.AddOrReplaceLinuxNamespace(spec.NetworkNamespace, nsPath); err != nil { - return err - } - } - if c.config.PIDNsCtr != "" { - pidCtr, err := c.runtime.state.Container(c.config.PIDNsCtr) - if err != nil { - return err - } - - nsPath, err := pidCtr.NamespacePath(PIDNS) - if err != nil { - return err - } - - if err := g.AddOrReplaceLinuxNamespace(string(spec.PIDNamespace), nsPath); err != nil { - return err - } - } - if c.config.UserNsCtr != "" { - userCtr, err := c.runtime.state.Container(c.config.UserNsCtr) - if err != nil { - return err - } - - nsPath, err := userCtr.NamespacePath(UserNS) - if err != nil { - return err - } - - if err := g.AddOrReplaceLinuxNamespace(spec.UserNamespace, nsPath); err != nil { - return err - } - } - if c.config.UTSNsCtr != "" { - utsCtr, err := c.runtime.state.Container(c.config.UTSNsCtr) - if err != nil { - return err - } - - nsPath, err := utsCtr.NamespacePath(UTSNS) - if err != nil { - return err - } - - if err := g.AddOrReplaceLinuxNamespace(spec.UTSNamespace, nsPath); err != nil { - return err - } - } - if c.config.CgroupNsCtr != "" { - cgroupCtr, err := c.runtime.state.Container(c.config.CgroupNsCtr) - if err != nil { - return err - } - - nsPath, err := cgroupCtr.NamespacePath(CgroupNS) - if err != nil { - return err - } - - if err := g.AddOrReplaceLinuxNamespace(spec.CgroupNamespace, nsPath); err != nil { - return err - } - } - - c.runningSpec = g.Spec() - c.runningSpec.Root.Path = c.state.Mountpoint - c.runningSpec.Annotations[crioAnnotations.Created] = c.config.CreatedTime.Format(time.RFC3339Nano) - c.runningSpec.Annotations["org.opencontainers.image.stopSignal"] = fmt.Sprintf("%d", c.config.StopSignal) - - fileJSON, err := json.Marshal(c.runningSpec) - if err != nil { - return errors.Wrapf(err, "error exporting runtime spec for container %s to JSON", c.ID()) - } - if err := ioutil.WriteFile(jsonPath, fileJSON, 0644); err != nil { - return errors.Wrapf(err, "error writing runtime spec JSON to file for container %s", c.ID()) - } - - logrus.Debugf("Created OCI spec for container %s at %s", c.ID(), jsonPath) - - c.state.ConfigPath = jsonPath - - // With the spec complete, do an OCI create - // TODO set cgroup parent in a sane fashion - if err := c.runtime.ociRuntime.createContainer(c, CgroupParent); err != nil { - return err - } - - logrus.Debugf("Created container %s in runc", c.ID()) - - c.state.State = ContainerStateCreated - - return c.save() -} - -// Start starts a container -func (c *Container) Start() error { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - } - - // Container must be created or stopped to be started - if !(c.state.State == ContainerStateCreated || c.state.State == ContainerStateStopped) { - return errors.Wrapf(ErrCtrStateInvalid, "container %s must be in Created or Stopped state to be started", c.ID()) - } - - // Mount storage for the container - if err := c.mountStorage(); err != nil { - return err - } - - if err := c.runtime.ociRuntime.startContainer(c); err != nil { - return err - } - - logrus.Debugf("Started container %s", c.ID()) - - c.state.State = ContainerStateRunning - - return c.save() -} - -// Stop uses the container's stop signal (or SIGTERM if no signal was specified) -// to stop the container, and if it has not stopped after the given timeout (in -// seconds), uses SIGKILL to attempt to forcibly stop the container. -// If timeout is 0, SIGKILL will be used immediately -func (c *Container) Stop(timeout uint) error { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - } - - logrus.Debugf("Stopping ctr %s with timeout %d", c.ID(), timeout) - - if c.state.State == ContainerStateConfigured || - c.state.State == ContainerStateUnknown || - c.state.State == ContainerStatePaused { - return errors.Wrapf(ErrCtrStateInvalid, "can only stop created, running, or stopped containers") - } - - if err := c.runtime.ociRuntime.stopContainer(c, timeout); err != nil { - return err - } - - // Sync the container's state to pick up return code - if err := c.runtime.ociRuntime.updateContainerStatus(c); err != nil { - return err - } - - return c.cleanupStorage() -} - -// Kill sends a signal to a container -func (c *Container) Kill(signal uint) error { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - } - - if c.state.State != ContainerStateRunning { - return errors.Wrapf(ErrCtrStateInvalid, "can only kill running containers") - } - - return c.runtime.ociRuntime.killContainer(c, signal) -} - -// Exec starts a new process inside the container -func (c *Container) Exec(tty, privileged bool, env, cmd []string, user string) error { - var capList []string - - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - } - - conState := c.state.State - - if conState != ContainerStateRunning { - return errors.Errorf("cannot attach to container that is not running") - } - if privileged { - capList = caps.GetAllCapabilities() - } - globalOpts := runcGlobalOptions{ - log: c.LogPath(), - } - execOpts := runcExecOptions{ - capAdd: capList, - pidFile: filepath.Join(c.state.RunDir, fmt.Sprintf("%s-execpid", stringid.GenerateNonCryptoID()[:12])), - env: env, - user: user, - cwd: c.config.Spec.Process.Cwd, - tty: tty, - } - - return c.runtime.ociRuntime.execContainer(c, cmd, globalOpts, execOpts) -} - -// Attach attaches to a container -// Returns fully qualified URL of streaming server for the container -func (c *Container) Attach(noStdin bool, keys string, attached chan<- bool) error { - if !c.locked { - c.lock.Lock() - if err := c.syncContainer(); err != nil { - c.lock.Unlock() - return err - } - c.lock.Unlock() - } - - if c.state.State != ContainerStateCreated && - c.state.State != ContainerStateRunning { - return errors.Wrapf(ErrCtrStateInvalid, "can only attach to created or running containers") - } - - // Check the validity of the provided keys first - var err error - detachKeys := []byte{} - if len(keys) > 0 { - detachKeys, err = term.ToBytes(keys) - if err != nil { - return errors.Wrapf(err, "invalid detach keys") - } - } - - resize := make(chan remotecommand.TerminalSize) - defer close(resize) - - err = c.attachContainerSocket(resize, noStdin, detachKeys, attached) - return err -} - -// Mount mounts a container's filesystem on the host -// The path where the container has been mounted is returned -func (c *Container) Mount(label string) (string, error) { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return "", err - } - } - - // return mountpoint if container already mounted - if c.state.Mounted { - return c.state.Mountpoint, nil - } - - mountLabel := label - if label == "" { - mountLabel = c.config.MountLabel - } - mountPoint, err := c.runtime.store.Mount(c.ID(), mountLabel) - if err != nil { - return "", err - } - c.state.Mountpoint = mountPoint - c.state.Mounted = true - c.config.MountLabel = mountLabel - - if err := c.save(); err != nil { - return "", err - } - - return mountPoint, nil -} - -// Unmount unmounts a container's filesystem on the host -func (c *Container) Unmount() error { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - } - - if c.state.State == ContainerStateRunning || c.state.State == ContainerStatePaused { - return errors.Wrapf(ErrCtrStateInvalid, "cannot remove storage for container %s as it is running or paused", c.ID()) - } - - return c.cleanupStorage() -} - -// Pause pauses a container -func (c *Container) Pause() error { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - } - - if c.state.State == ContainerStatePaused { - return errors.Wrapf(ErrCtrStateInvalid, "%q is already paused", c.ID()) - } - if c.state.State != ContainerStateRunning && c.state.State != ContainerStateCreated { - return errors.Wrapf(ErrCtrStateInvalid, "%q is not running/created, can't pause", c.state.State) - } - if err := c.runtime.ociRuntime.pauseContainer(c); err != nil { - return err - } - - logrus.Debugf("Paused container %s", c.ID()) - - c.state.State = ContainerStatePaused - - return c.save() -} - -// Unpause unpauses a container -func (c *Container) Unpause() error { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - } - - if c.state.State != ContainerStatePaused { - return errors.Wrapf(ErrCtrStateInvalid, "%q is not paused, can't unpause", c.ID()) - } - if err := c.runtime.ociRuntime.unpauseContainer(c); err != nil { - return err - } - - logrus.Debugf("Unpaused container %s", c.ID()) - - c.state.State = ContainerStateRunning - - return c.save() -} - -// Export exports a container's root filesystem as a tar archive -// The archive will be saved as a file at the given path -func (c *Container) Export(path string) error { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - } - - return c.export(path) -} - -func (c *Container) export(path string) error { - mountPoint := c.state.Mountpoint - if !c.state.Mounted { - mount, err := c.runtime.store.Mount(c.ID(), c.config.MountLabel) - if err != nil { - return errors.Wrapf(err, "error mounting container %q", c.ID()) - } - mountPoint = mount - defer func() { - if err := c.runtime.store.Unmount(c.ID()); err != nil { - logrus.Errorf("error unmounting container %q: %v", c.ID(), err) - } - }() - } - - input, err := archive.Tar(mountPoint, archive.Uncompressed) - if err != nil { - return errors.Wrapf(err, "error reading container directory %q", c.ID()) - } - - outFile, err := os.Create(path) - if err != nil { - return errors.Wrapf(err, "error creating file %q", path) - } - defer outFile.Close() - - _, err = io.Copy(outFile, input) - return err -} - -// AddArtifact creates and writes to an artifact file for the container -func (c *Container) AddArtifact(name string, data []byte) error { - if !c.valid { - return ErrCtrRemoved - } - - return ioutil.WriteFile(c.getArtifactPath(name), data, 0740) -} - -// GetArtifact reads the specified artifact file from the container -func (c *Container) GetArtifact(name string) ([]byte, error) { - if !c.valid { - return nil, ErrCtrRemoved - } - - return ioutil.ReadFile(c.getArtifactPath(name)) -} - -// RemoveArtifact deletes the specified artifacts file -func (c *Container) RemoveArtifact(name string) error { - if !c.valid { - return ErrCtrRemoved - } - - return os.Remove(c.getArtifactPath(name)) -} - -func (c *Container) getArtifactPath(name string) string { - return filepath.Join(c.config.StaticDir, artifactsDir, name) -} - -// Inspect a container for low-level information -func (c *Container) Inspect(size bool) (*ContainerInspectData, error) { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return nil, err - } - } - - storeCtr, err := c.runtime.store.Container(c.ID()) - if err != nil { - return nil, errors.Wrapf(err, "error getting container from store %q", c.ID()) - } - layer, err := c.runtime.store.Layer(storeCtr.LayerID) - if err != nil { - return nil, errors.Wrapf(err, "error reading information about layer %q", storeCtr.LayerID) - } - driverData, err := driver.GetDriverData(c.runtime.store, layer.ID) - if err != nil { - return nil, errors.Wrapf(err, "error getting graph driver info %q", c.ID()) - } - - return c.getContainerInspectData(size, driverData) -} - -// Commit commits the changes between a container and its image, creating a new -// image -func (c *Container) Commit(pause bool, options CopyOptions) error { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - } - - if c.state.State == ContainerStateRunning && pause { - if err := c.runtime.ociRuntime.pauseContainer(c); err != nil { - return errors.Wrapf(err, "error pausing container %q", c.ID()) - } - defer func() { - if err := c.runtime.ociRuntime.unpauseContainer(c); err != nil { - logrus.Errorf("error unpausing container %q: %v", c.ID(), err) - } - }() - } - - tempFile, err := ioutil.TempFile(c.runtime.config.TmpDir, "podman-commit") - if err != nil { - return errors.Wrapf(err, "error creating temp file") - } - defer os.Remove(tempFile.Name()) - defer tempFile.Close() - - if err := c.export(tempFile.Name()); err != nil { - return err - } - return c.runtime.ImportImage(tempFile.Name(), options) -} - -// Wait blocks on a container to exit and returns its exit code -func (c *Container) Wait() (int32, error) { - if !c.valid { - return -1, ErrCtrRemoved - } - - err := wait.PollImmediateInfinite(1, - func() (bool, error) { - stopped, err := c.isStopped() - if err != nil { - return false, err - } - if !stopped { - return false, nil - } else { // nolint - return true, nil // nolint - } // nolint - }, - ) - if err != nil { - return 0, err - } - exitCode := c.state.ExitCode - return exitCode, nil -} - -func (c *Container) isStopped() (bool, error) { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - } - err := c.syncContainer() - if err != nil { - return true, err - } - return c.state.State == ContainerStateStopped, nil -} - -// save container state to the database -func (c *Container) save() error { - if err := c.runtime.state.SaveContainer(c); err != nil { - return errors.Wrapf(err, "error saving container %s state", c.ID()) - } - return nil -} - -// mountStorage sets up the container's root filesystem -// It mounts the image and any other requested mounts -// TODO: Add ability to override mount label so we can use this for Mount() too -// TODO: Can we use this for export? Copying SHM into the export might not be -// good -func (c *Container) mountStorage() (err error) { - // Container already mounted, nothing to do - if c.state.Mounted { - return nil - } - - // TODO: generalize this mount code so it will mount every mount in ctr.config.Mounts - - mounted, err := mount.Mounted(c.config.ShmDir) - if err != nil { - return errors.Wrapf(err, "unable to determine if %q is mounted", c.config.ShmDir) - } - - if !mounted { - shmOptions := fmt.Sprintf("mode=1777,size=%d", c.config.ShmSize) - if err := unix.Mount("shm", c.config.ShmDir, "tmpfs", unix.MS_NOEXEC|unix.MS_NOSUID|unix.MS_NODEV, - label.FormatMountLabel(shmOptions, c.config.MountLabel)); err != nil { - return errors.Wrapf(err, "failed to mount shm tmpfs %q", c.config.ShmDir) - } - } - - mountPoint, err := c.runtime.storageService.MountContainerImage(c.ID()) - if err != nil { - return errors.Wrapf(err, "error mounting storage for container %s", c.ID()) - } - c.state.Mounted = true - c.state.Mountpoint = mountPoint - - logrus.Debugf("Created root filesystem for container %s at %s", c.ID(), c.state.Mountpoint) - - defer func() { - if err != nil { - if err2 := c.cleanupStorage(); err2 != nil { - logrus.Errorf("Error unmounting storage for container %s: %v", c.ID(), err) - } - } - }() - - return c.save() -} - -// CleanupStorage unmounts all mount points in container and cleans up container storage -func (c *Container) CleanupStorage() error { - if !c.locked { - c.lock.Lock() - defer c.lock.Unlock() - if err := c.syncContainer(); err != nil { - return err - } - } - return c.cleanupStorage() -} - -// cleanupStorage unmounts and cleans up the container's root filesystem -func (c *Container) cleanupStorage() error { - if !c.state.Mounted { - // Already unmounted, do nothing - return nil - } - - for _, mount := range c.config.Mounts { - if err := unix.Unmount(mount, unix.MNT_DETACH); err != nil { - if err != syscall.EINVAL { - logrus.Warnf("container %s failed to unmount %s : %v", c.ID(), mount, err) - } - } - } - - // Also unmount storage - if err := c.runtime.storageService.UnmountContainerImage(c.ID()); err != nil { - return errors.Wrapf(err, "error unmounting container %s root filesystem", c.ID()) - } - - c.state.Mountpoint = "" - c.state.Mounted = false - - return c.save() -} - // CGroupPath returns a cgroups "path" for a given container. func (c *Container) CGroupPath() cgroups.Path { return cgroups.StaticPath(filepath.Join(c.config.CgroupParent, fmt.Sprintf("libpod-conmon-%s", c.ID()))) } -// copyHostFileToRundir copies the provided file to the runtimedir -func (c *Container) copyHostFileToRundir(sourcePath string) (string, error) { - destFileName := filepath.Join(c.state.RunDir, filepath.Base(sourcePath)) - if err := fileutils.CopyFile(sourcePath, destFileName); err != nil { - return "", err - } - // Relabel runDirResolv for the container - if err := label.Relabel(destFileName, c.config.MountLabel, false); err != nil { - return "", err - } - return destFileName, nil -} - // StopTimeout returns a stop timeout field for this container func (c *Container) StopTimeout() uint { return c.config.StopTimeout } - -// Batch starts a batch operation on the given container -// All commands in the passed function will execute under the same lock and -// without syncronyzing state after each operation -// This will result in substantial performance benefits when running numerous -// commands on the same container -// Note that the container passed into the Batch function cannot be removed -// during batched operations. runtime.RemoveContainer can only be called outside -// of Batch -// Any error returned by the given batch function will be returned unmodified by -// Batch -// As Batch normally disables updating the current state of the container, the -// Sync() function is provided to enable container state to be updated and -// checked within Batch. -func (c *Container) Batch(batchFunc func(*Container) error) error { - c.lock.Lock() - defer c.lock.Unlock() - - if err := c.syncContainer(); err != nil { - return err - } - - newCtr := new(Container) - newCtr.config = c.config - newCtr.state = c.state - newCtr.runtime = c.runtime - newCtr.lock = c.lock - newCtr.valid = true - - newCtr.locked = true - - if err := batchFunc(newCtr); err != nil { - return err - } - - newCtr.locked = false - - return c.save() -} - -// Sync updates the current state of the container, checking whether its state -// has changed -// Sync can only be used inside Batch() - otherwise, it will be done -// automatically. -// When called outside Batch(), Sync() is a no-op -func (c *Container) Sync() error { - if !c.locked { - return nil - } - - // If runc knows about the container, update its status in runc - // And then save back to disk - if (c.state.State != ContainerStateUnknown) && - (c.state.State != ContainerStateConfigured) { - oldState := c.state.State - // TODO: optionally replace this with a stat for the exit file - if err := c.runtime.ociRuntime.updateContainerStatus(c); err != nil { - return err - } - // Only save back to DB if state changed - if c.state.State != oldState { - if err := c.save(); err != nil { - return err - } - } - } - - return nil -} diff --git a/libpod/container_api.go b/libpod/container_api.go new file mode 100644 index 000000000..9a4f50079 --- /dev/null +++ b/libpod/container_api.go @@ -0,0 +1,754 @@ +package libpod + +import ( + "encoding/json" + "fmt" + "io/ioutil" + "os" + "path/filepath" + "time" + + "github.com/docker/docker/daemon/caps" + "github.com/docker/docker/pkg/stringid" + "github.com/docker/docker/pkg/term" + spec "github.com/opencontainers/runtime-spec/specs-go" + "github.com/opencontainers/runtime-tools/generate" + "github.com/pkg/errors" + "github.com/projectatomic/libpod/libpod/driver" + crioAnnotations "github.com/projectatomic/libpod/pkg/annotations" + "github.com/projectatomic/libpod/pkg/chrootuser" + "github.com/sirupsen/logrus" + "k8s.io/apimachinery/pkg/util/wait" + "k8s.io/client-go/tools/remotecommand" +) + +// Init creates a container in the OCI runtime +func (c *Container) Init() (err error) { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + } + + if c.state.State != ContainerStateConfigured { + return errors.Wrapf(ErrCtrExists, "container %s has already been created in runtime", c.ID()) + } + + if err := c.mountStorage(); err != nil { + return err + } + defer func() { + if err != nil { + if err2 := c.cleanupStorage(); err2 != nil { + logrus.Errorf("Error cleaning up storage for container %s: %v", c.ID(), err2) + } + } + }() + + // Make a network namespace for the container + if c.config.CreateNetNS && c.state.NetNS == nil { + if err := c.runtime.createNetNS(c); err != nil { + return err + } + } + defer func() { + if err != nil { + if err2 := c.runtime.teardownNetNS(c); err2 != nil { + logrus.Errorf("Error tearing down network namespace for container %s: %v", c.ID(), err2) + } + } + }() + + // If the OCI spec already exists, we need to replace it + // Cannot guarantee some things, e.g. network namespaces, have the same + // paths + jsonPath := filepath.Join(c.bundlePath(), "config.json") + if _, err := os.Stat(jsonPath); err != nil { + if !os.IsNotExist(err) { + return errors.Wrapf(err, "error doing stat on container %s spec", c.ID()) + } + // The spec does not exist, we're fine + } else { + // The spec exists, need to remove it + if err := os.Remove(jsonPath); err != nil { + return errors.Wrapf(err, "error replacing runtime spec for container %s", c.ID()) + } + } + + // Copy /etc/resolv.conf to the container's rundir + resolvPath := "/etc/resolv.conf" + + // Check if the host system is using system resolve and if so + // copy its resolv.conf + _, err = os.Stat("/run/systemd/resolve/resolv.conf") + if err == nil { + resolvPath = "/run/systemd/resolve/resolv.conf" + } + runDirResolv, err := c.copyHostFileToRundir(resolvPath) + if err != nil { + return errors.Wrapf(err, "unable to copy resolv.conf to ", runDirResolv) + } + // Copy /etc/hosts to the container's rundir + runDirHosts, err := c.copyHostFileToRundir("/etc/hosts") + if err != nil { + return errors.Wrapf(err, "unable to copy /etc/hosts to ", runDirHosts) + } + + // Save OCI spec to disk + g := generate.NewFromSpec(c.config.Spec) + // If network namespace was requested, add it now + if c.config.CreateNetNS { + g.AddOrReplaceLinuxNamespace(spec.NetworkNamespace, c.state.NetNS.Path()) + } + // Remove default /etc/shm mount + g.RemoveMount("/dev/shm") + // Mount ShmDir from host into container + shmMnt := spec.Mount{ + Type: "bind", + Source: c.config.ShmDir, + Destination: "/dev/shm", + Options: []string{"rw", "bind"}, + } + g.AddMount(shmMnt) + // Bind mount resolv.conf + resolvMnt := spec.Mount{ + Type: "bind", + Source: runDirResolv, + Destination: "/etc/resolv.conf", + Options: []string{"rw", "bind"}, + } + g.AddMount(resolvMnt) + // Bind mount hosts + hostsMnt := spec.Mount{ + Type: "bind", + Source: runDirHosts, + Destination: "/etc/hosts", + Options: []string{"rw", "bind"}, + } + g.AddMount(hostsMnt) + + if c.config.User != "" { + if !c.state.Mounted { + return errors.Wrapf(ErrCtrStateInvalid, "container %s must be mounted in order to translate User field", c.ID()) + } + uid, gid, err := chrootuser.GetUser(c.state.Mountpoint, c.config.User) + if err != nil { + return err + } + // User and Group must go together + g.SetProcessUID(uid) + g.SetProcessGID(gid) + } + + // Add shared namespaces from other containers + if c.config.IPCNsCtr != "" { + ipcCtr, err := c.runtime.state.Container(c.config.IPCNsCtr) + if err != nil { + return err + } + + nsPath, err := ipcCtr.NamespacePath(IPCNS) + if err != nil { + return err + } + + if err := g.AddOrReplaceLinuxNamespace(spec.IPCNamespace, nsPath); err != nil { + return err + } + } + if c.config.MountNsCtr != "" { + mountCtr, err := c.runtime.state.Container(c.config.MountNsCtr) + if err != nil { + return err + } + + nsPath, err := mountCtr.NamespacePath(MountNS) + if err != nil { + return err + } + + if err := g.AddOrReplaceLinuxNamespace(spec.MountNamespace, nsPath); err != nil { + return err + } + } + if c.config.NetNsCtr != "" { + netCtr, err := c.runtime.state.Container(c.config.NetNsCtr) + if err != nil { + return err + } + + nsPath, err := netCtr.NamespacePath(NetNS) + if err != nil { + return err + } + + if err := g.AddOrReplaceLinuxNamespace(spec.NetworkNamespace, nsPath); err != nil { + return err + } + } + if c.config.PIDNsCtr != "" { + pidCtr, err := c.runtime.state.Container(c.config.PIDNsCtr) + if err != nil { + return err + } + + nsPath, err := pidCtr.NamespacePath(PIDNS) + if err != nil { + return err + } + + if err := g.AddOrReplaceLinuxNamespace(string(spec.PIDNamespace), nsPath); err != nil { + return err + } + } + if c.config.UserNsCtr != "" { + userCtr, err := c.runtime.state.Container(c.config.UserNsCtr) + if err != nil { + return err + } + + nsPath, err := userCtr.NamespacePath(UserNS) + if err != nil { + return err + } + + if err := g.AddOrReplaceLinuxNamespace(spec.UserNamespace, nsPath); err != nil { + return err + } + } + if c.config.UTSNsCtr != "" { + utsCtr, err := c.runtime.state.Container(c.config.UTSNsCtr) + if err != nil { + return err + } + + nsPath, err := utsCtr.NamespacePath(UTSNS) + if err != nil { + return err + } + + if err := g.AddOrReplaceLinuxNamespace(spec.UTSNamespace, nsPath); err != nil { + return err + } + } + if c.config.CgroupNsCtr != "" { + cgroupCtr, err := c.runtime.state.Container(c.config.CgroupNsCtr) + if err != nil { + return err + } + + nsPath, err := cgroupCtr.NamespacePath(CgroupNS) + if err != nil { + return err + } + + if err := g.AddOrReplaceLinuxNamespace(spec.CgroupNamespace, nsPath); err != nil { + return err + } + } + + c.runningSpec = g.Spec() + c.runningSpec.Root.Path = c.state.Mountpoint + c.runningSpec.Annotations[crioAnnotations.Created] = c.config.CreatedTime.Format(time.RFC3339Nano) + c.runningSpec.Annotations["org.opencontainers.image.stopSignal"] = fmt.Sprintf("%d", c.config.StopSignal) + + fileJSON, err := json.Marshal(c.runningSpec) + if err != nil { + return errors.Wrapf(err, "error exporting runtime spec for container %s to JSON", c.ID()) + } + if err := ioutil.WriteFile(jsonPath, fileJSON, 0644); err != nil { + return errors.Wrapf(err, "error writing runtime spec JSON to file for container %s", c.ID()) + } + + logrus.Debugf("Created OCI spec for container %s at %s", c.ID(), jsonPath) + + c.state.ConfigPath = jsonPath + + // With the spec complete, do an OCI create + // TODO set cgroup parent in a sane fashion + if err := c.runtime.ociRuntime.createContainer(c, CgroupParent); err != nil { + return err + } + + logrus.Debugf("Created container %s in runc", c.ID()) + + c.state.State = ContainerStateCreated + + return c.save() +} + +// Start starts a container +func (c *Container) Start() error { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + } + + // Container must be created or stopped to be started + if !(c.state.State == ContainerStateCreated || c.state.State == ContainerStateStopped) { + return errors.Wrapf(ErrCtrStateInvalid, "container %s must be in Created or Stopped state to be started", c.ID()) + } + + // Mount storage for the container + if err := c.mountStorage(); err != nil { + return err + } + + if err := c.runtime.ociRuntime.startContainer(c); err != nil { + return err + } + + logrus.Debugf("Started container %s", c.ID()) + + c.state.State = ContainerStateRunning + + return c.save() +} + +// Stop uses the container's stop signal (or SIGTERM if no signal was specified) +// to stop the container, and if it has not stopped after the given timeout (in +// seconds), uses SIGKILL to attempt to forcibly stop the container. +// If timeout is 0, SIGKILL will be used immediately +func (c *Container) Stop(timeout uint) error { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + } + + logrus.Debugf("Stopping ctr %s with timeout %d", c.ID(), timeout) + + if c.state.State == ContainerStateConfigured || + c.state.State == ContainerStateUnknown || + c.state.State == ContainerStatePaused { + return errors.Wrapf(ErrCtrStateInvalid, "can only stop created, running, or stopped containers") + } + + if err := c.runtime.ociRuntime.stopContainer(c, timeout); err != nil { + return err + } + + // Sync the container's state to pick up return code + if err := c.runtime.ociRuntime.updateContainerStatus(c); err != nil { + return err + } + + return c.cleanupStorage() +} + +// Kill sends a signal to a container +func (c *Container) Kill(signal uint) error { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + } + + if c.state.State != ContainerStateRunning { + return errors.Wrapf(ErrCtrStateInvalid, "can only kill running containers") + } + + return c.runtime.ociRuntime.killContainer(c, signal) +} + +// Exec starts a new process inside the container +func (c *Container) Exec(tty, privileged bool, env, cmd []string, user string) error { + var capList []string + + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + } + + conState := c.state.State + + if conState != ContainerStateRunning { + return errors.Errorf("cannot attach to container that is not running") + } + if privileged { + capList = caps.GetAllCapabilities() + } + globalOpts := runcGlobalOptions{ + log: c.LogPath(), + } + execOpts := runcExecOptions{ + capAdd: capList, + pidFile: filepath.Join(c.state.RunDir, fmt.Sprintf("%s-execpid", stringid.GenerateNonCryptoID()[:12])), + env: env, + user: user, + cwd: c.config.Spec.Process.Cwd, + tty: tty, + } + + return c.runtime.ociRuntime.execContainer(c, cmd, globalOpts, execOpts) +} + +// Attach attaches to a container +// Returns fully qualified URL of streaming server for the container +func (c *Container) Attach(noStdin bool, keys string, attached chan<- bool) error { + if !c.locked { + c.lock.Lock() + if err := c.syncContainer(); err != nil { + c.lock.Unlock() + return err + } + c.lock.Unlock() + } + + if c.state.State != ContainerStateCreated && + c.state.State != ContainerStateRunning { + return errors.Wrapf(ErrCtrStateInvalid, "can only attach to created or running containers") + } + + // Check the validity of the provided keys first + var err error + detachKeys := []byte{} + if len(keys) > 0 { + detachKeys, err = term.ToBytes(keys) + if err != nil { + return errors.Wrapf(err, "invalid detach keys") + } + } + + resize := make(chan remotecommand.TerminalSize) + defer close(resize) + + err = c.attachContainerSocket(resize, noStdin, detachKeys, attached) + return err +} + +// Mount mounts a container's filesystem on the host +// The path where the container has been mounted is returned +func (c *Container) Mount(label string) (string, error) { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return "", err + } + } + + // return mountpoint if container already mounted + if c.state.Mounted { + return c.state.Mountpoint, nil + } + + mountLabel := label + if label == "" { + mountLabel = c.config.MountLabel + } + mountPoint, err := c.runtime.store.Mount(c.ID(), mountLabel) + if err != nil { + return "", err + } + c.state.Mountpoint = mountPoint + c.state.Mounted = true + c.config.MountLabel = mountLabel + + if err := c.save(); err != nil { + return "", err + } + + return mountPoint, nil +} + +// Unmount unmounts a container's filesystem on the host +func (c *Container) Unmount() error { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + } + + if c.state.State == ContainerStateRunning || c.state.State == ContainerStatePaused { + return errors.Wrapf(ErrCtrStateInvalid, "cannot remove storage for container %s as it is running or paused", c.ID()) + } + + return c.cleanupStorage() +} + +// Pause pauses a container +func (c *Container) Pause() error { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + } + + if c.state.State == ContainerStatePaused { + return errors.Wrapf(ErrCtrStateInvalid, "%q is already paused", c.ID()) + } + if c.state.State != ContainerStateRunning && c.state.State != ContainerStateCreated { + return errors.Wrapf(ErrCtrStateInvalid, "%q is not running/created, can't pause", c.state.State) + } + if err := c.runtime.ociRuntime.pauseContainer(c); err != nil { + return err + } + + logrus.Debugf("Paused container %s", c.ID()) + + c.state.State = ContainerStatePaused + + return c.save() +} + +// Unpause unpauses a container +func (c *Container) Unpause() error { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + } + + if c.state.State != ContainerStatePaused { + return errors.Wrapf(ErrCtrStateInvalid, "%q is not paused, can't unpause", c.ID()) + } + if err := c.runtime.ociRuntime.unpauseContainer(c); err != nil { + return err + } + + logrus.Debugf("Unpaused container %s", c.ID()) + + c.state.State = ContainerStateRunning + + return c.save() +} + +// Export exports a container's root filesystem as a tar archive +// The archive will be saved as a file at the given path +func (c *Container) Export(path string) error { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + } + + return c.export(path) +} + +// AddArtifact creates and writes to an artifact file for the container +func (c *Container) AddArtifact(name string, data []byte) error { + if !c.valid { + return ErrCtrRemoved + } + + return ioutil.WriteFile(c.getArtifactPath(name), data, 0740) +} + +// GetArtifact reads the specified artifact file from the container +func (c *Container) GetArtifact(name string) ([]byte, error) { + if !c.valid { + return nil, ErrCtrRemoved + } + + return ioutil.ReadFile(c.getArtifactPath(name)) +} + +// RemoveArtifact deletes the specified artifacts file +func (c *Container) RemoveArtifact(name string) error { + if !c.valid { + return ErrCtrRemoved + } + + return os.Remove(c.getArtifactPath(name)) +} + +// Inspect a container for low-level information +func (c *Container) Inspect(size bool) (*ContainerInspectData, error) { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return nil, err + } + } + + storeCtr, err := c.runtime.store.Container(c.ID()) + if err != nil { + return nil, errors.Wrapf(err, "error getting container from store %q", c.ID()) + } + layer, err := c.runtime.store.Layer(storeCtr.LayerID) + if err != nil { + return nil, errors.Wrapf(err, "error reading information about layer %q", storeCtr.LayerID) + } + driverData, err := driver.GetDriverData(c.runtime.store, layer.ID) + if err != nil { + return nil, errors.Wrapf(err, "error getting graph driver info %q", c.ID()) + } + + return c.getContainerInspectData(size, driverData) +} + +// Commit commits the changes between a container and its image, creating a new +// image +func (c *Container) Commit(pause bool, options CopyOptions) error { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + } + + if c.state.State == ContainerStateRunning && pause { + if err := c.runtime.ociRuntime.pauseContainer(c); err != nil { + return errors.Wrapf(err, "error pausing container %q", c.ID()) + } + defer func() { + if err := c.runtime.ociRuntime.unpauseContainer(c); err != nil { + logrus.Errorf("error unpausing container %q: %v", c.ID(), err) + } + }() + } + + tempFile, err := ioutil.TempFile(c.runtime.config.TmpDir, "podman-commit") + if err != nil { + return errors.Wrapf(err, "error creating temp file") + } + defer os.Remove(tempFile.Name()) + defer tempFile.Close() + + if err := c.export(tempFile.Name()); err != nil { + return err + } + return c.runtime.ImportImage(tempFile.Name(), options) +} + +// Wait blocks on a container to exit and returns its exit code +func (c *Container) Wait() (int32, error) { + if !c.valid { + return -1, ErrCtrRemoved + } + + err := wait.PollImmediateInfinite(1, + func() (bool, error) { + stopped, err := c.isStopped() + if err != nil { + return false, err + } + if !stopped { + return false, nil + } else { // nolint + return true, nil // nolint + } // nolint + }, + ) + if err != nil { + return 0, err + } + exitCode := c.state.ExitCode + return exitCode, nil +} + +// CleanupStorage unmounts all mount points in container and cleans up container storage +func (c *Container) CleanupStorage() error { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + if err := c.syncContainer(); err != nil { + return err + } + } + return c.cleanupStorage() +} + +// Batch starts a batch operation on the given container +// All commands in the passed function will execute under the same lock and +// without syncronyzing state after each operation +// This will result in substantial performance benefits when running numerous +// commands on the same container +// Note that the container passed into the Batch function cannot be removed +// during batched operations. runtime.RemoveContainer can only be called outside +// of Batch +// Any error returned by the given batch function will be returned unmodified by +// Batch +// As Batch normally disables updating the current state of the container, the +// Sync() function is provided to enable container state to be updated and +// checked within Batch. +func (c *Container) Batch(batchFunc func(*Container) error) error { + c.lock.Lock() + defer c.lock.Unlock() + + if err := c.syncContainer(); err != nil { + return err + } + + newCtr := new(Container) + newCtr.config = c.config + newCtr.state = c.state + newCtr.runtime = c.runtime + newCtr.lock = c.lock + newCtr.valid = true + + newCtr.locked = true + + if err := batchFunc(newCtr); err != nil { + return err + } + + newCtr.locked = false + + return c.save() +} + +// Sync updates the current state of the container, checking whether its state +// has changed +// Sync can only be used inside Batch() - otherwise, it will be done +// automatically. +// When called outside Batch(), Sync() is a no-op +func (c *Container) Sync() error { + if !c.locked { + return nil + } + + // If runc knows about the container, update its status in runc + // And then save back to disk + if (c.state.State != ContainerStateUnknown) && + (c.state.State != ContainerStateConfigured) { + oldState := c.state.State + // TODO: optionally replace this with a stat for the exit file + if err := c.runtime.ociRuntime.updateContainerStatus(c); err != nil { + return err + } + // Only save back to DB if state changed + if c.state.State != oldState { + if err := c.save(); err != nil { + return err + } + } + } + + return nil +} diff --git a/libpod/container_internal.go b/libpod/container_internal.go new file mode 100644 index 000000000..245246869 --- /dev/null +++ b/libpod/container_internal.go @@ -0,0 +1,382 @@ +package libpod + +import ( + "fmt" + "io" + "os" + "path/filepath" + "syscall" + "time" + + "github.com/containers/storage" + "github.com/containers/storage/pkg/archive" + "github.com/docker/docker/pkg/mount" + "github.com/docker/docker/pkg/namesgenerator" + "github.com/docker/docker/pkg/stringid" + "github.com/mrunalp/fileutils" + spec "github.com/opencontainers/runtime-spec/specs-go" + "github.com/opencontainers/selinux/go-selinux/label" + "github.com/pkg/errors" + "github.com/sirupsen/logrus" + "github.com/ulule/deepcopier" + "golang.org/x/sys/unix" +) + +// rootFsSize gets the size of the container's root filesystem +// A container FS is split into two parts. The first is the top layer, a +// mutable layer, and the rest is the RootFS: the set of immutable layers +// that make up the image on which the container is based. +func (c *Container) rootFsSize() (int64, error) { + container, err := c.runtime.store.Container(c.ID()) + if err != nil { + return 0, err + } + + // Ignore the size of the top layer. The top layer is a mutable RW layer + // and is not considered a part of the rootfs + rwLayer, err := c.runtime.store.Layer(container.LayerID) + if err != nil { + return 0, err + } + layer, err := c.runtime.store.Layer(rwLayer.Parent) + if err != nil { + return 0, err + } + + size := int64(0) + for layer.Parent != "" { + layerSize, err := c.runtime.store.DiffSize(layer.Parent, layer.ID) + if err != nil { + return size, errors.Wrapf(err, "getting diffsize of layer %q and its parent %q", layer.ID, layer.Parent) + } + size += layerSize + layer, err = c.runtime.store.Layer(layer.Parent) + if err != nil { + return 0, err + } + } + // Get the size of the last layer. Has to be outside of the loop + // because the parent of the last layer is "", andlstore.Get("") + // will return an error. + layerSize, err := c.runtime.store.DiffSize(layer.Parent, layer.ID) + return size + layerSize, err +} + +// rwSize Gets the size of the mutable top layer of the container. +func (c *Container) rwSize() (int64, error) { + container, err := c.runtime.store.Container(c.ID()) + if err != nil { + return 0, err + } + + // Get the size of the top layer by calculating the size of the diff + // between the layer and its parent. The top layer of a container is + // the only RW layer, all others are immutable + layer, err := c.runtime.store.Layer(container.LayerID) + if err != nil { + return 0, err + } + return c.runtime.store.DiffSize(layer.Parent, layer.ID) +} + +// The path to the container's root filesystem - where the OCI spec will be +// placed, amongst other things +func (c *Container) bundlePath() string { + return c.config.StaticDir +} + +// The path to the container's logs file +func (c *Container) logPath() string { + return filepath.Join(c.config.StaticDir, "ctr.log") +} + +// Retrieves the path of the container's attach socket +func (c *Container) attachSocketPath() string { + return filepath.Join(c.runtime.ociRuntime.socketsDir, c.ID(), "attach") +} + +// Sync this container with on-disk state and runc status +// Should only be called with container lock held +// This function should suffice to ensure a container's state is accurate and +// it is valid for use. +func (c *Container) syncContainer() error { + if err := c.runtime.state.UpdateContainer(c); err != nil { + return err + } + // If runc knows about the container, update its status in runc + // And then save back to disk + if (c.state.State != ContainerStateUnknown) && + (c.state.State != ContainerStateConfigured) { + oldState := c.state.State + // TODO: optionally replace this with a stat for the exit file + if err := c.runtime.ociRuntime.updateContainerStatus(c); err != nil { + return err + } + // Only save back to DB if state changed + if c.state.State != oldState { + if err := c.save(); err != nil { + return err + } + } + } + + if !c.valid { + return errors.Wrapf(ErrCtrRemoved, "container %s is not valid", c.ID()) + } + + return nil +} + +// Make a new container +func newContainer(rspec *spec.Spec, lockDir string) (*Container, error) { + if rspec == nil { + return nil, errors.Wrapf(ErrInvalidArg, "must provide a valid runtime spec to create container") + } + + ctr := new(Container) + ctr.config = new(ContainerConfig) + ctr.state = new(containerRuntimeInfo) + + ctr.config.ID = stringid.GenerateNonCryptoID() + ctr.config.Name = namesgenerator.GetRandomName(0) + + ctr.config.Spec = new(spec.Spec) + deepcopier.Copy(rspec).To(ctr.config.Spec) + ctr.config.CreatedTime = time.Now() + + ctr.config.ShmSize = DefaultShmSize + ctr.config.CgroupParent = CgroupParent + + // Path our lock file will reside at + lockPath := filepath.Join(lockDir, ctr.config.ID) + // Grab a lockfile at the given path + lock, err := storage.GetLockfile(lockPath) + if err != nil { + return nil, errors.Wrapf(err, "error creating lockfile for new container") + } + ctr.lock = lock + + return ctr, nil +} + +// Create container root filesystem for use +func (c *Container) setupStorage() error { + if !c.valid { + return errors.Wrapf(ErrCtrRemoved, "container %s is not valid", c.ID()) + } + + if c.state.State != ContainerStateConfigured { + return errors.Wrapf(ErrCtrStateInvalid, "container %s must be in Configured state to have storage set up", c.ID()) + } + + // Need both an image ID and image name, plus a bool telling us whether to use the image configuration + if c.config.RootfsImageID == "" || c.config.RootfsImageName == "" { + return errors.Wrapf(ErrInvalidArg, "must provide image ID and image name to use an image") + } + + containerInfo, err := c.runtime.storageService.CreateContainerStorage(c.runtime.imageContext, c.config.RootfsImageName, c.config.RootfsImageID, c.config.Name, c.config.ID, c.config.MountLabel) + if err != nil { + return errors.Wrapf(err, "error creating container storage") + } + + c.config.StaticDir = containerInfo.Dir + c.state.RunDir = containerInfo.RunDir + + artifacts := filepath.Join(c.config.StaticDir, artifactsDir) + if err := os.MkdirAll(artifacts, 0755); err != nil { + return errors.Wrapf(err, "error creating artifacts directory %q", artifacts) + } + + return nil +} + +// Tear down a container's storage prior to removal +func (c *Container) teardownStorage() error { + if !c.valid { + return errors.Wrapf(ErrCtrRemoved, "container %s is not valid", c.ID()) + } + + if c.state.State == ContainerStateRunning || c.state.State == ContainerStatePaused { + return errors.Wrapf(ErrCtrStateInvalid, "cannot remove storage for container %s as it is running or paused", c.ID()) + } + + artifacts := filepath.Join(c.config.StaticDir, artifactsDir) + if err := os.RemoveAll(artifacts); err != nil { + return errors.Wrapf(err, "error removing artifacts %q", artifacts) + } + + if err := c.cleanupStorage(); err != nil { + return errors.Wrapf(err, "failed to cleanup container %s storage", c.ID()) + } + + if err := c.runtime.storageService.DeleteContainer(c.ID()); err != nil { + return errors.Wrapf(err, "error removing container %s root filesystem", c.ID()) + } + + return nil +} + +// Refresh refreshes the container's state after a restart +func (c *Container) refresh() error { + c.lock.Lock() + defer c.lock.Unlock() + + if !c.valid { + return errors.Wrapf(ErrCtrRemoved, "container %s is not valid - may have been removed", c.ID()) + } + + // We need to get the container's temporary directory from c/storage + // It was lost in the reboot and must be recreated + dir, err := c.runtime.storageService.GetRunDir(c.ID()) + if err != nil { + return errors.Wrapf(err, "error retrieving temporary directory for container %s", c.ID()) + } + c.state.RunDir = dir + + if err := c.runtime.state.SaveContainer(c); err != nil { + return errors.Wrapf(err, "error refreshing state for container %s", c.ID()) + } + + return nil +} + +func (c *Container) export(path string) error { + mountPoint := c.state.Mountpoint + if !c.state.Mounted { + mount, err := c.runtime.store.Mount(c.ID(), c.config.MountLabel) + if err != nil { + return errors.Wrapf(err, "error mounting container %q", c.ID()) + } + mountPoint = mount + defer func() { + if err := c.runtime.store.Unmount(c.ID()); err != nil { + logrus.Errorf("error unmounting container %q: %v", c.ID(), err) + } + }() + } + + input, err := archive.Tar(mountPoint, archive.Uncompressed) + if err != nil { + return errors.Wrapf(err, "error reading container directory %q", c.ID()) + } + + outFile, err := os.Create(path) + if err != nil { + return errors.Wrapf(err, "error creating file %q", path) + } + defer outFile.Close() + + _, err = io.Copy(outFile, input) + return err +} + +// Get path of artifact with a given name for this container +func (c *Container) getArtifactPath(name string) string { + return filepath.Join(c.config.StaticDir, artifactsDir, name) +} + +// Used with Wait() to determine if a container has exited +func (c *Container) isStopped() (bool, error) { + if !c.locked { + c.lock.Lock() + defer c.lock.Unlock() + } + err := c.syncContainer() + if err != nil { + return true, err + } + return c.state.State == ContainerStateStopped, nil +} + +// save container state to the database +func (c *Container) save() error { + if err := c.runtime.state.SaveContainer(c); err != nil { + return errors.Wrapf(err, "error saving container %s state", c.ID()) + } + return nil +} + +// mountStorage sets up the container's root filesystem +// It mounts the image and any other requested mounts +// TODO: Add ability to override mount label so we can use this for Mount() too +// TODO: Can we use this for export? Copying SHM into the export might not be +// good +func (c *Container) mountStorage() (err error) { + // Container already mounted, nothing to do + if c.state.Mounted { + return nil + } + + // TODO: generalize this mount code so it will mount every mount in ctr.config.Mounts + + mounted, err := mount.Mounted(c.config.ShmDir) + if err != nil { + return errors.Wrapf(err, "unable to determine if %q is mounted", c.config.ShmDir) + } + + if !mounted { + shmOptions := fmt.Sprintf("mode=1777,size=%d", c.config.ShmSize) + if err := unix.Mount("shm", c.config.ShmDir, "tmpfs", unix.MS_NOEXEC|unix.MS_NOSUID|unix.MS_NODEV, + label.FormatMountLabel(shmOptions, c.config.MountLabel)); err != nil { + return errors.Wrapf(err, "failed to mount shm tmpfs %q", c.config.ShmDir) + } + } + + mountPoint, err := c.runtime.storageService.MountContainerImage(c.ID()) + if err != nil { + return errors.Wrapf(err, "error mounting storage for container %s", c.ID()) + } + c.state.Mounted = true + c.state.Mountpoint = mountPoint + + logrus.Debugf("Created root filesystem for container %s at %s", c.ID(), c.state.Mountpoint) + + defer func() { + if err != nil { + if err2 := c.cleanupStorage(); err2 != nil { + logrus.Errorf("Error unmounting storage for container %s: %v", c.ID(), err) + } + } + }() + + return c.save() +} + +// cleanupStorage unmounts and cleans up the container's root filesystem +func (c *Container) cleanupStorage() error { + if !c.state.Mounted { + // Already unmounted, do nothing + return nil + } + + for _, mount := range c.config.Mounts { + if err := unix.Unmount(mount, unix.MNT_DETACH); err != nil { + if err != syscall.EINVAL { + logrus.Warnf("container %s failed to unmount %s : %v", c.ID(), mount, err) + } + } + } + + // Also unmount storage + if err := c.runtime.storageService.UnmountContainerImage(c.ID()); err != nil { + return errors.Wrapf(err, "error unmounting container %s root filesystem", c.ID()) + } + + c.state.Mountpoint = "" + c.state.Mounted = false + + return c.save() +} + +// copyHostFileToRundir copies the provided file to the runtimedir +func (c *Container) copyHostFileToRundir(sourcePath string) (string, error) { + destFileName := filepath.Join(c.state.RunDir, filepath.Base(sourcePath)) + if err := fileutils.CopyFile(sourcePath, destFileName); err != nil { + return "", err + } + // Relabel runDirResolv for the container + if err := label.Relabel(destFileName, c.config.MountLabel, false); err != nil { + return "", err + } + return destFileName, nil +} |