podman/libpod/container_internal.go

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"
)

const (
	// name of the directory holding the artifacts
	artifactsDir = "artifacts"
)

// 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(containerState)

	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
}