containerd/core/unpack/unpacker.go
Saket Jajoo d7f83034c2
Fix the race condition during GC of snapshots when client retries
When an upstream client (e.g. kubelet) stops or restarts, the CRI
connection to the containerd gets interrupted which is treated as a
cancellation of context which subsequently cancels an ongoing operation,
including an image pull. This generally gets followed by containerd's
GC routine that tries to delete the prepared snapshots for the image
layer(s) corresponding to the image in the pull operation that got
cancelled. However, if the upstream client immediately retries (or
starts a new) image pull operation, containerd initiates a new image
pull and starts unpacking the image layers into snapshots. This may
create a race condition: the GC routine (corresponding to the failed
image pull operation) trying to clean up the same snapshot that the new
image pull operation is preparing, thus leading to the "parent snapshot
does not exist: not found" error.

Race Condition Scenario:
Assume an image consisting of 2 layers (L1 and L2, L1 being the bottom
layer) that are supposed to get unpacked into snapshots S1 and S2
respectively.

During an image pull operation, containerd unpacks(L1) which involves
Stat()'ing the chainID. This Stat() fails as the chainID does not
exist and Prepare(L1) gets called. Once S1 gets prepared, containerd
processes L2 - unpack(L2) which again involves Stat()'ing the chainID
which fails as the chainID for S2 does not exist which results in the
call to Prepare(L2). However, if the image pull operation gets
cancelled before Prepare(L2) is called, then the GC routine tries to
clean up S1.

When the image pull operation is retried by the upstream client,
containerd follows the same series of operations. unpack(L1) gets
called which then calls Stat(chainID) for L1. However, this time,
Stat(L1) succedes as S1 already exists (from the previous image pull
operation) and thus containerd goes to the next iteration to
unpack(L2). Now, GC cleans up S1 and when Prepare(L2) gets called, it
returns back the "parent snapshot does not exist: not found" error.

Fix:
Removing the "Stat() + early return" fixes the race condition. Now
during the image pull operation corresponding to the client retry,
although the chainID (for L1) already exists, containerd does not
return early and goes on to Prepare(L1). Since L1 is already prepared,
it adds a new lease to S1 and then returns `ErrAlreadyExists`. This
new lease prevents GC from cleaning up S1 when containerd processes
L2 (unpack(L2) -> Prepare(L2)).

Fixes: https://github.com/containerd/containerd/issues/3787

Signed-off-by: Saket Jajoo <saketjajoo@google.com>
2024-10-02 22:10:15 +00:00

545 lines
14 KiB
Go

/*
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 unpack
import (
"context"
"crypto/rand"
"encoding/base64"
"encoding/json"
"errors"
"fmt"
"strconv"
"sync"
"sync/atomic"
"time"
"github.com/containerd/containerd/v2/core/content"
"github.com/containerd/containerd/v2/core/diff"
"github.com/containerd/containerd/v2/core/images"
"github.com/containerd/containerd/v2/core/mount"
"github.com/containerd/containerd/v2/core/snapshots"
"github.com/containerd/containerd/v2/internal/cleanup"
"github.com/containerd/containerd/v2/internal/kmutex"
"github.com/containerd/containerd/v2/pkg/labels"
"github.com/containerd/containerd/v2/pkg/tracing"
"github.com/containerd/errdefs"
"github.com/containerd/log"
"github.com/containerd/platforms"
"github.com/opencontainers/go-digest"
"github.com/opencontainers/image-spec/identity"
ocispec "github.com/opencontainers/image-spec/specs-go/v1"
"golang.org/x/sync/errgroup"
"golang.org/x/sync/semaphore"
)
const (
labelSnapshotRef = "containerd.io/snapshot.ref"
unpackSpanPrefix = "pkg.unpack.unpacker"
)
// Result returns information about the unpacks which were completed.
type Result struct {
Unpacks int
}
type unpackerConfig struct {
platforms []*Platform
content content.Store
limiter *semaphore.Weighted
duplicationSuppressor kmutex.KeyedLocker
}
// Platform represents a platform-specific unpack configuration which includes
// the platform matcher as well as snapshotter and applier.
type Platform struct {
Platform platforms.Matcher
SnapshotterKey string
Snapshotter snapshots.Snapshotter
SnapshotOpts []snapshots.Opt
Applier diff.Applier
ApplyOpts []diff.ApplyOpt
}
type UnpackerOpt func(*unpackerConfig) error
func WithUnpackPlatform(u Platform) UnpackerOpt {
return UnpackerOpt(func(c *unpackerConfig) error {
if u.Platform == nil {
u.Platform = platforms.All
}
if u.Snapshotter == nil {
return fmt.Errorf("snapshotter must be provided to unpack")
}
if u.SnapshotterKey == "" {
if s, ok := u.Snapshotter.(fmt.Stringer); ok {
u.SnapshotterKey = s.String()
} else {
u.SnapshotterKey = "unknown"
}
}
if u.Applier == nil {
return fmt.Errorf("applier must be provided to unpack")
}
c.platforms = append(c.platforms, &u)
return nil
})
}
func WithLimiter(l *semaphore.Weighted) UnpackerOpt {
return UnpackerOpt(func(c *unpackerConfig) error {
c.limiter = l
return nil
})
}
func WithDuplicationSuppressor(d kmutex.KeyedLocker) UnpackerOpt {
return UnpackerOpt(func(c *unpackerConfig) error {
c.duplicationSuppressor = d
return nil
})
}
// Unpacker unpacks images by hooking into the image handler process.
// Unpacks happen in the backgrounds and waited on to complete.
type Unpacker struct {
unpackerConfig
unpacks int32
ctx context.Context
eg *errgroup.Group
}
// NewUnpacker creates a new instance of the unpacker which can be used to wrap an
// image handler and unpack in parallel to handling. The unpacker will handle
// calling the block handlers when they are needed by the unpack process.
func NewUnpacker(ctx context.Context, cs content.Store, opts ...UnpackerOpt) (*Unpacker, error) {
eg, ctx := errgroup.WithContext(ctx)
u := &Unpacker{
unpackerConfig: unpackerConfig{
content: cs,
duplicationSuppressor: kmutex.NewNoop(),
},
ctx: ctx,
eg: eg,
}
for _, opt := range opts {
if err := opt(&u.unpackerConfig); err != nil {
return nil, err
}
}
if len(u.platforms) == 0 {
return nil, fmt.Errorf("no unpack platforms defined: %w", errdefs.ErrInvalidArgument)
}
return u, nil
}
// Unpack wraps an image handler to filter out blob handling and scheduling them
// during the unpack process. When an image config is encountered, the unpack
// process will be started in a goroutine.
func (u *Unpacker) Unpack(h images.Handler) images.Handler {
var (
lock sync.Mutex
layers = map[digest.Digest][]ocispec.Descriptor{}
)
return images.HandlerFunc(func(ctx context.Context, desc ocispec.Descriptor) ([]ocispec.Descriptor, error) {
ctx, span := tracing.StartSpan(ctx, tracing.Name(unpackSpanPrefix, "UnpackHandler"))
defer span.End()
span.SetAttributes(
tracing.Attribute("descriptor.media.type", desc.MediaType),
tracing.Attribute("descriptor.digest", desc.Digest.String()))
unlock, err := u.lockBlobDescriptor(ctx, desc)
if err != nil {
return nil, err
}
children, err := h.Handle(ctx, desc)
unlock()
if err != nil {
return children, err
}
if images.IsManifestType(desc.MediaType) {
var nonLayers []ocispec.Descriptor
var manifestLayers []ocispec.Descriptor
// Split layers from non-layers, layers will be handled after
// the config
for i, child := range children {
span.SetAttributes(
tracing.Attribute("descriptor.child."+strconv.Itoa(i), []string{child.MediaType, child.Digest.String()}),
)
if images.IsLayerType(child.MediaType) {
manifestLayers = append(manifestLayers, child)
} else {
nonLayers = append(nonLayers, child)
}
}
lock.Lock()
for _, nl := range nonLayers {
layers[nl.Digest] = manifestLayers
}
lock.Unlock()
children = nonLayers
} else if images.IsConfigType(desc.MediaType) {
lock.Lock()
l := layers[desc.Digest]
lock.Unlock()
if len(l) > 0 {
u.eg.Go(func() error {
return u.unpack(h, desc, l)
})
}
}
return children, nil
})
}
// Wait waits for any ongoing unpack processes to complete then will return
// the result.
func (u *Unpacker) Wait() (Result, error) {
if err := u.eg.Wait(); err != nil {
return Result{}, err
}
return Result{
Unpacks: int(u.unpacks),
}, nil
}
func (u *Unpacker) unpack(
h images.Handler,
config ocispec.Descriptor,
layers []ocispec.Descriptor,
) error {
ctx := u.ctx
ctx, layerSpan := tracing.StartSpan(ctx, tracing.Name(unpackSpanPrefix, "unpack"))
defer layerSpan.End()
unpackStart := time.Now()
p, err := content.ReadBlob(ctx, u.content, config)
if err != nil {
return err
}
var i ocispec.Image
if err := json.Unmarshal(p, &i); err != nil {
return fmt.Errorf("unmarshal image config: %w", err)
}
diffIDs := i.RootFS.DiffIDs
if len(layers) != len(diffIDs) {
return fmt.Errorf("number of layers and diffIDs don't match: %d != %d", len(layers), len(diffIDs))
}
// TODO: Support multiple unpacks rather than just first match
var unpack *Platform
imgPlatform := platforms.Normalize(i.Platform)
for _, up := range u.platforms {
if up.Platform.Match(imgPlatform) {
unpack = up
break
}
}
if unpack == nil {
log.G(ctx).WithField("image", config.Digest).WithField("platform", platforms.Format(imgPlatform)).Debugf("unpacker does not support platform, only fetching layers")
return u.fetch(ctx, h, layers, nil)
}
atomic.AddInt32(&u.unpacks, 1)
var (
sn = unpack.Snapshotter
a = unpack.Applier
cs = u.content
chain []digest.Digest
fetchOffset int
fetchC []chan struct{}
fetchErr chan error
)
// If there is an early return, ensure any ongoing
// fetches get their context cancelled
ctx, cancel := context.WithCancel(ctx)
defer cancel()
doUnpackFn := func(i int, desc ocispec.Descriptor) error {
parent := identity.ChainID(chain)
chain = append(chain, diffIDs[i])
chainID := identity.ChainID(chain).String()
unlock, err := u.lockSnChainID(ctx, chainID, unpack.SnapshotterKey)
if err != nil {
return err
}
defer unlock()
// inherits annotations which are provided as snapshot labels.
snapshotLabels := snapshots.FilterInheritedLabels(desc.Annotations)
if snapshotLabels == nil {
snapshotLabels = make(map[string]string)
}
snapshotLabels[labelSnapshotRef] = chainID
var (
key string
mounts []mount.Mount
opts = append(unpack.SnapshotOpts, snapshots.WithLabels(snapshotLabels))
)
for try := 1; try <= 3; try++ {
// Prepare snapshot with from parent, label as root
key = fmt.Sprintf(snapshots.UnpackKeyFormat, uniquePart(), chainID)
mounts, err = sn.Prepare(ctx, key, parent.String(), opts...)
if err != nil {
if errdefs.IsAlreadyExists(err) {
if _, err := sn.Stat(ctx, chainID); err != nil {
if !errdefs.IsNotFound(err) {
return fmt.Errorf("failed to stat snapshot %s: %w", chainID, err)
}
// Try again, this should be rare, log it
log.G(ctx).WithField("key", key).WithField("chainid", chainID).Debug("extraction snapshot already exists, chain id not found")
} else {
// no need to handle, snapshot now found with chain id
return nil
}
} else {
return fmt.Errorf("failed to prepare extraction snapshot %q: %w", key, err)
}
} else {
break
}
}
if err != nil {
return fmt.Errorf("unable to prepare extraction snapshot: %w", err)
}
// Abort the snapshot if commit does not happen
abort := func(ctx context.Context) {
if err := sn.Remove(ctx, key); err != nil {
log.G(ctx).WithError(err).Errorf("failed to cleanup %q", key)
}
}
if fetchErr == nil {
fetchErr = make(chan error, 1)
fetchOffset = i
fetchC = make([]chan struct{}, len(layers)-fetchOffset)
for i := range fetchC {
fetchC[i] = make(chan struct{})
}
go func(i int) {
err := u.fetch(ctx, h, layers[i:], fetchC)
if err != nil {
fetchErr <- err
}
close(fetchErr)
}(i)
}
select {
case <-ctx.Done():
cleanup.Do(ctx, abort)
return ctx.Err()
case err := <-fetchErr:
if err != nil {
cleanup.Do(ctx, abort)
return err
}
case <-fetchC[i-fetchOffset]:
}
diff, err := a.Apply(ctx, desc, mounts, unpack.ApplyOpts...)
if err != nil {
cleanup.Do(ctx, abort)
return fmt.Errorf("failed to extract layer %s: %w", diffIDs[i], err)
}
if diff.Digest != diffIDs[i] {
cleanup.Do(ctx, abort)
return fmt.Errorf("wrong diff id calculated on extraction %q", diffIDs[i])
}
if err = sn.Commit(ctx, chainID, key, opts...); err != nil {
cleanup.Do(ctx, abort)
if errdefs.IsAlreadyExists(err) {
return nil
}
return fmt.Errorf("failed to commit snapshot %s: %w", key, err)
}
// Set the uncompressed label after the uncompressed
// digest has been verified through apply.
cinfo := content.Info{
Digest: desc.Digest,
Labels: map[string]string{
labels.LabelUncompressed: diff.Digest.String(),
},
}
if _, err := cs.Update(ctx, cinfo, "labels."+labels.LabelUncompressed); err != nil {
return err
}
return nil
}
for i, desc := range layers {
_, layerSpan := tracing.StartSpan(ctx, tracing.Name(unpackSpanPrefix, "unpackLayer"))
unpackLayerStart := time.Now()
layerSpan.SetAttributes(
tracing.Attribute("layer.media.type", desc.MediaType),
tracing.Attribute("layer.media.size", desc.Size),
tracing.Attribute("layer.media.digest", desc.Digest.String()),
)
if err := doUnpackFn(i, desc); err != nil {
layerSpan.SetStatus(err)
layerSpan.End()
return err
}
layerSpan.End()
log.G(ctx).WithFields(log.Fields{
"layer": desc.Digest,
"duration": time.Since(unpackLayerStart),
}).Debug("layer unpacked")
}
chainID := identity.ChainID(chain).String()
cinfo := content.Info{
Digest: config.Digest,
Labels: map[string]string{
fmt.Sprintf("containerd.io/gc.ref.snapshot.%s", unpack.SnapshotterKey): chainID,
},
}
_, err = cs.Update(ctx, cinfo, fmt.Sprintf("labels.containerd.io/gc.ref.snapshot.%s", unpack.SnapshotterKey))
if err != nil {
return err
}
log.G(ctx).WithFields(log.Fields{
"config": config.Digest,
"chainID": chainID,
"duration": time.Since(unpackStart),
}).Debug("image unpacked")
return nil
}
func (u *Unpacker) fetch(ctx context.Context, h images.Handler, layers []ocispec.Descriptor, done []chan struct{}) error {
eg, ctx2 := errgroup.WithContext(ctx)
for i, desc := range layers {
ctx2, layerSpan := tracing.StartSpan(ctx2, tracing.Name(unpackSpanPrefix, "fetchLayer"))
layerSpan.SetAttributes(
tracing.Attribute("layer.media.type", desc.MediaType),
tracing.Attribute("layer.media.size", desc.Size),
tracing.Attribute("layer.media.digest", desc.Digest.String()),
)
desc := desc
var ch chan struct{}
if done != nil {
ch = done[i]
}
if err := u.acquire(ctx); err != nil {
return err
}
eg.Go(func() error {
defer layerSpan.End()
unlock, err := u.lockBlobDescriptor(ctx2, desc)
if err != nil {
u.release()
return err
}
_, err = h.Handle(ctx2, desc)
unlock()
u.release()
if err != nil && !errors.Is(err, images.ErrSkipDesc) {
return err
}
if ch != nil {
close(ch)
}
return nil
})
}
return eg.Wait()
}
func (u *Unpacker) acquire(ctx context.Context) error {
if u.limiter == nil {
return nil
}
return u.limiter.Acquire(ctx, 1)
}
func (u *Unpacker) release() {
if u.limiter == nil {
return
}
u.limiter.Release(1)
}
func (u *Unpacker) lockSnChainID(ctx context.Context, chainID, snapshotter string) (func(), error) {
key := u.makeChainIDKeyWithSnapshotter(chainID, snapshotter)
if err := u.duplicationSuppressor.Lock(ctx, key); err != nil {
return nil, err
}
return func() {
u.duplicationSuppressor.Unlock(key)
}, nil
}
func (u *Unpacker) lockBlobDescriptor(ctx context.Context, desc ocispec.Descriptor) (func(), error) {
key := u.makeBlobDescriptorKey(desc)
if err := u.duplicationSuppressor.Lock(ctx, key); err != nil {
return nil, err
}
return func() {
u.duplicationSuppressor.Unlock(key)
}, nil
}
func (u *Unpacker) makeChainIDKeyWithSnapshotter(chainID, snapshotter string) string {
return fmt.Sprintf("sn://%s/%v", snapshotter, chainID)
}
func (u *Unpacker) makeBlobDescriptorKey(desc ocispec.Descriptor) string {
return fmt.Sprintf("blob://%v", desc.Digest)
}
func uniquePart() string {
t := time.Now()
var b [3]byte
// Ignore read failures, just decreases uniqueness
rand.Read(b[:])
return fmt.Sprintf("%d-%s", t.Nanosecond(), base64.URLEncoding.EncodeToString(b[:]))
}