containerd/snapshots/devmapper/snapshotter.go

// +build linux

/*
   Copyright The containerd Authors.

   Licensed under the Apache License, Version 2.0 (the "License");
   you may not use this file except in compliance with the License.
   You may obtain a copy of the License at

       http://www.apache.org/licenses/LICENSE-2.0

   Unless required by applicable law or agreed to in writing, software
   distributed under the License is distributed on an "AS IS" BASIS,
   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
   See the License for the specific language governing permissions and
   limitations under the License.
*/

package devmapper

import (
	"context"
	"fmt"
	"os"
	"os/exec"
	"path/filepath"
	"strings"
	"sync"

	"github.com/containerd/containerd/errdefs"
	"github.com/containerd/containerd/log"
	"github.com/containerd/containerd/mount"
	"github.com/containerd/containerd/snapshots"
	"github.com/containerd/containerd/snapshots/devmapper/dmsetup"
	"github.com/containerd/containerd/snapshots/storage"
	"github.com/hashicorp/go-multierror"
	"github.com/pkg/errors"
	"github.com/sirupsen/logrus"
)

const (
	metadataFileName = "metadata.db"
	fsTypeExt4       = "ext4"
)

type closeFunc func() error

// Snapshotter implements containerd's snapshotter (https://godoc.org/github.com/containerd/containerd/snapshots#Snapshotter)
// based on Linux device-mapper targets.
type Snapshotter struct {
	store     *storage.MetaStore
	pool      *PoolDevice
	config    *Config
	cleanupFn []closeFunc
	closeOnce sync.Once
}

// NewSnapshotter creates new device mapper snapshotter.
// Internally it creates thin-pool device (or reloads if it's already exists) and
// initializes a database file for metadata.
func NewSnapshotter(ctx context.Context, config *Config) (*Snapshotter, error) {
	// Make sure snapshotter configuration valid before running
	if err := config.parse(); err != nil {
		return nil, err
	}

	if err := config.Validate(); err != nil {
		return nil, err
	}

	var cleanupFn []closeFunc

	if err := os.MkdirAll(config.RootPath, 0750); err != nil && !os.IsExist(err) {
		return nil, errors.Wrapf(err, "failed to create root directory: %s", config.RootPath)
	}

	store, err := storage.NewMetaStore(filepath.Join(config.RootPath, metadataFileName))
	if err != nil {
		return nil, errors.Wrap(err, "failed to create metastore")
	}

	cleanupFn = append(cleanupFn, store.Close)

	poolDevice, err := NewPoolDevice(ctx, config)
	if err != nil {
		return nil, err
	}

	cleanupFn = append(cleanupFn, poolDevice.Close)

	return &Snapshotter{
		store:     store,
		config:    config,
		pool:      poolDevice,
		cleanupFn: cleanupFn,
	}, nil
}

// Stat returns the info for an active or committed snapshot from store
func (s *Snapshotter) Stat(ctx context.Context, key string) (snapshots.Info, error) {
	log.G(ctx).WithField("key", key).Debug("stat")

	var (
		info snapshots.Info
		err  error
	)

	err = s.withTransaction(ctx, false, func(ctx context.Context) error {
		_, info, _, err = storage.GetInfo(ctx, key)
		return err
	})

	return info, err
}

// Update updates an existing snapshot info's data
func (s *Snapshotter) Update(ctx context.Context, info snapshots.Info, fieldpaths ...string) (snapshots.Info, error) {
	log.G(ctx).Debugf("update: %s", strings.Join(fieldpaths, ", "))

	var err error
	err = s.withTransaction(ctx, true, func(ctx context.Context) error {
		info, err = storage.UpdateInfo(ctx, info, fieldpaths...)
		return err
	})

	return info, err
}

// Usage returns the resource usage of an active or committed snapshot excluding the usage of parent snapshots.
func (s *Snapshotter) Usage(ctx context.Context, key string) (snapshots.Usage, error) {
	log.G(ctx).WithField("key", key).Debug("usage")

	var (
		id    string
		err   error
		info  snapshots.Info
		usage snapshots.Usage
	)

	err = s.withTransaction(ctx, false, func(ctx context.Context) error {
		id, info, usage, err = storage.GetInfo(ctx, key)
		if err != nil {
			return err
		}

		if info.Kind == snapshots.KindActive {
			deviceName := s.getDeviceName(id)
			usage.Size, err = s.pool.GetUsage(deviceName)
			if err != nil {
				return err
			}
		}

		if info.Parent != "" {
			// GetInfo returns total number of bytes used by a snapshot (including parent).
			// So subtract parent usage in order to get delta consumed by layer itself.
			_, _, parentUsage, err := storage.GetInfo(ctx, info.Parent)
			if err != nil {
				return err
			}

			usage.Size -= parentUsage.Size
		}

		return err
	})

	return usage, err
}

// Mounts return the list of mounts for the active or view snapshot
func (s *Snapshotter) Mounts(ctx context.Context, key string) ([]mount.Mount, error) {
	log.G(ctx).WithField("key", key).Debug("mounts")

	var (
		snap storage.Snapshot
		err  error
	)

	err = s.withTransaction(ctx, false, func(ctx context.Context) error {
		snap, err = storage.GetSnapshot(ctx, key)
		return err
	})

	return s.buildMounts(snap), nil
}

// Prepare creates thin device for an active snapshot identified by key
func (s *Snapshotter) Prepare(ctx context.Context, key, parent string, opts ...snapshots.Opt) ([]mount.Mount, error) {
	log.G(ctx).WithFields(logrus.Fields{"key": key, "parent": parent}).Debug("prepare")

	var (
		mounts []mount.Mount
		err    error
	)

	err = s.withTransaction(ctx, true, func(ctx context.Context) error {
		mounts, err = s.createSnapshot(ctx, snapshots.KindActive, key, parent, opts...)
		return err
	})

	return mounts, err
}

// View creates readonly thin device for the given snapshot key
func (s *Snapshotter) View(ctx context.Context, key, parent string, opts ...snapshots.Opt) ([]mount.Mount, error) {
	log.G(ctx).WithFields(logrus.Fields{"key": key, "parent": parent}).Debug("prepare")

	var (
		mounts []mount.Mount
		err    error
	)

	err = s.withTransaction(ctx, true, func(ctx context.Context) error {
		mounts, err = s.createSnapshot(ctx, snapshots.KindView, key, parent, opts...)
		return err
	})

	return mounts, err
}

// Commit marks an active snapshot as committed in meta store.
// Block device unmount operation captures snapshot changes by itself, so no
// additional actions needed within Commit operation.
func (s *Snapshotter) Commit(ctx context.Context, name, key string, opts ...snapshots.Opt) error {
	log.G(ctx).WithFields(logrus.Fields{"name": name, "key": key}).Debug("commit")

	return s.withTransaction(ctx, true, func(ctx context.Context) error {
		id, _, _, err := storage.GetInfo(ctx, key)
		if err != nil {
			return err
		}

		deviceName := s.getDeviceName(id)
		size, err := s.pool.GetUsage(deviceName)
		if err != nil {
			return err
		}

		usage := snapshots.Usage{
			Size: size,
		}

		_, err = storage.CommitActive(ctx, key, name, usage, opts...)
		if err != nil {
			return err
		}

		// After committed, the snapshot device will not be directly
		// used anymore. We'd better deactivate it to make it *invisible*
		// in userspace, so that tools like LVM2 and fdisk cannot touch it,
		// and avoid useless IOs on it.
		//
		// Before deactivation, we need to flush the outstanding IO by suspend.
		// Afterward, we resume it again to prevent a race window which may cause
		// a process IO hang. See the issue below for details:
		//   (https://github.com/containerd/containerd/issues/4234)
		err = s.pool.SuspendDevice(ctx, deviceName)
		if err != nil {
			return err
		}

		err = s.pool.ResumeDevice(ctx, deviceName)
		if err != nil {
			return err
		}

		return s.pool.DeactivateDevice(ctx, deviceName, true, false)
	})
}

// Remove removes thin device and snapshot metadata by key
func (s *Snapshotter) Remove(ctx context.Context, key string) error {
	log.G(ctx).WithField("key", key).Debug("remove")

	return s.withTransaction(ctx, true, func(ctx context.Context) error {
		return s.removeDevice(ctx, key)
	})
}

func (s *Snapshotter) removeDevice(ctx context.Context, key string) error {
	snapID, _, err := storage.Remove(ctx, key)
	if err != nil {
		return err
	}

	deviceName := s.getDeviceName(snapID)
	if !s.config.AsyncRemove {
		if err := s.pool.RemoveDevice(ctx, deviceName); err != nil {
			log.G(ctx).WithError(err).Errorf("failed to remove device")
			// Tell snapshot GC continue to collect other snapshots.
			// Otherwise, one snapshot collection failure will stop
			// the GC, and all snapshots won't be collected even though
			// having no relationship with the failed one.
			return errdefs.ErrFailedPrecondition
		}
	} else {
		// The asynchronous cleanup will do the real device remove work.
		log.G(ctx).WithField("device", deviceName).Debug("async remove")
		if err := s.pool.MarkDeviceState(ctx, deviceName, Removed); err != nil {
			log.G(ctx).WithError(err).Errorf("failed to mark device as removed")
			return err
		}
	}

	return nil
}

// Walk iterates through all metadata Info for the stored snapshots and calls the provided function for each.
func (s *Snapshotter) Walk(ctx context.Context, fn snapshots.WalkFunc, fs ...string) error {
	log.G(ctx).Debug("walk")
	return s.withTransaction(ctx, false, func(ctx context.Context) error {
		return storage.WalkInfo(ctx, fn, fs...)
	})
}

// ResetPool deactivates and deletes all thin devices in thin-pool.
// Used for cleaning pool after benchmarking.
func (s *Snapshotter) ResetPool(ctx context.Context) error {
	names, err := s.pool.metadata.GetDeviceNames(ctx)
	if err != nil {
		return err
	}

	var result *multierror.Error
	for _, name := range names {
		if err := s.pool.RemoveDevice(ctx, name); err != nil {
			result = multierror.Append(result, err)
		}
	}

	return result.ErrorOrNil()
}

// Close releases devmapper snapshotter resources.
// All subsequent Close calls will be ignored.
func (s *Snapshotter) Close() error {
	log.L.Debug("close")

	var result *multierror.Error
	s.closeOnce.Do(func() {
		for _, fn := range s.cleanupFn {
			if err := fn(); err != nil {
				result = multierror.Append(result, err)
			}
		}
	})

	return result.ErrorOrNil()
}

func (s *Snapshotter) createSnapshot(ctx context.Context, kind snapshots.Kind, key, parent string, opts ...snapshots.Opt) ([]mount.Mount, error) {
	snap, err := storage.CreateSnapshot(ctx, kind, key, parent, opts...)
	if err != nil {
		return nil, err
	}

	if len(snap.ParentIDs) == 0 {
		deviceName := s.getDeviceName(snap.ID)
		log.G(ctx).Debugf("creating new thin device '%s'", deviceName)

		err := s.pool.CreateThinDevice(ctx, deviceName, s.config.BaseImageSizeBytes)
		if err != nil {
			log.G(ctx).WithError(err).Errorf("failed to create thin device for snapshot %s", snap.ID)
			return nil, err
		}

		if err := s.mkfs(ctx, deviceName); err != nil {
			// Rollback thin device creation if mkfs failed
			return nil, multierror.Append(err,
				s.pool.RemoveDevice(ctx, deviceName))
		}
	} else {
		parentDeviceName := s.getDeviceName(snap.ParentIDs[0])
		snapDeviceName := s.getDeviceName(snap.ID)
		log.G(ctx).Debugf("creating snapshot device '%s' from '%s'", snapDeviceName, parentDeviceName)

		err := s.pool.CreateSnapshotDevice(ctx, parentDeviceName, snapDeviceName, s.config.BaseImageSizeBytes)
		if err != nil {
			log.G(ctx).WithError(err).Errorf("failed to create snapshot device from parent %s", parentDeviceName)
			return nil, err
		}
	}

	mounts := s.buildMounts(snap)

	// Remove default directories not expected by the container image
	_ = mount.WithTempMount(ctx, mounts, func(root string) error {
		return os.Remove(filepath.Join(root, "lost+found"))
	})

	return mounts, nil
}

// mkfs creates ext4 filesystem on the given devmapper device
func (s *Snapshotter) mkfs(ctx context.Context, deviceName string) error {
	args := []string{
		"-E",
		// We don't want any zeroing in advance when running mkfs on thin devices (see "man mkfs.ext4")
		"nodiscard,lazy_itable_init=0,lazy_journal_init=0",
		dmsetup.GetFullDevicePath(deviceName),
	}

	log.G(ctx).Debugf("mkfs.ext4 %s", strings.Join(args, " "))
	output, err := exec.Command("mkfs.ext4", args...).CombinedOutput()
	if err != nil {
		log.G(ctx).WithError(err).Errorf("failed to write fs:\n%s", string(output))
		return err
	}

	log.G(ctx).Debugf("mkfs:\n%s", string(output))
	return nil
}

func (s *Snapshotter) getDeviceName(snapID string) string {
	// Add pool name as prefix to avoid collisions with devices from other pools
	return fmt.Sprintf("%s-snap-%s", s.config.PoolName, snapID)
}

func (s *Snapshotter) getDevicePath(snap storage.Snapshot) string {
	name := s.getDeviceName(snap.ID)
	return dmsetup.GetFullDevicePath(name)
}

func (s *Snapshotter) buildMounts(snap storage.Snapshot) []mount.Mount {
	var options []string

	if snap.Kind != snapshots.KindActive {
		options = append(options, "ro")
	}

	mounts := []mount.Mount{
		{
			Source:  s.getDevicePath(snap),
			Type:    fsTypeExt4,
			Options: options,
		},
	}

	return mounts
}

// withTransaction wraps fn callback with containerd's meta store transaction.
// If callback returns an error or transaction is not writable, database transaction will be discarded.
func (s *Snapshotter) withTransaction(ctx context.Context, writable bool, fn func(ctx context.Context) error) error {
	ctx, trans, err := s.store.TransactionContext(ctx, writable)
	if err != nil {
		return err
	}

	var result *multierror.Error

	err = fn(ctx)
	if err != nil {
		result = multierror.Append(result, err)
	}

	// Always rollback if transaction is not writable
	if err != nil || !writable {
		if terr := trans.Rollback(); terr != nil {
			log.G(ctx).WithError(terr).Error("failed to rollback transaction")
			result = multierror.Append(result, errors.Wrap(terr, "rollback failed"))
		}
	} else {
		if terr := trans.Commit(); terr != nil {
			log.G(ctx).WithError(terr).Error("failed to commit transaction")
			result = multierror.Append(result, errors.Wrap(terr, "commit failed"))
		}
	}

	if err := result.ErrorOrNil(); err != nil {
		log.G(ctx).WithError(err).Debug("snapshotter error")

		// Unwrap if just one error
		if len(result.Errors) == 1 {
			return result.Errors[0]
		}

		return err
	}

	return nil
}

func (s *Snapshotter) Cleanup(ctx context.Context) error {
	var removedDevices []*DeviceInfo

	if !s.config.AsyncRemove {
		return nil
	}

	if err := s.pool.WalkDevices(ctx, func(info *DeviceInfo) error {
		if info.State == Removed {
			removedDevices = append(removedDevices, info)
		}
		return nil
	}); err != nil {
		log.G(ctx).WithError(err).Errorf("failed to query devices from metastore")
		return err
	}

	var result *multierror.Error
	for _, dev := range removedDevices {
		log.G(ctx).WithField("device", dev.Name).Debug("cleanup device")
		if err := s.pool.RemoveDevice(ctx, dev.Name); err != nil {
			log.G(ctx).WithField("device", dev.Name).Error("failed to cleanup device")
			result = multierror.Append(result, err)
		} else {
			log.G(ctx).WithField("device", dev.Name).Debug("cleanuped device")
		}
	}

	return result.ErrorOrNil()
}