listing: improve listing of encrypted objects (#14667)

This commit improves the listing of encrypted objects:
 - Use `etag.Format` and `etag.Decrypt`
 - Detect SSE-S3 single-part objects in a single iteration
 - Fix batch size to `250`
 - Pass request context to `DecryptAll` to not waste resources
   when a client cancels the operation.

Signed-off-by: Andreas Auernhammer <hi@aead.dev>
This commit is contained in:
Andreas Auernhammer 2022-04-04 20:42:03 +02:00 committed by GitHub
parent d4251b2545
commit 6b1c62133d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
7 changed files with 87 additions and 102 deletions

View File

@ -19,12 +19,12 @@ package cmd
import ( import (
"context" "context"
"fmt"
"net/http" "net/http"
"strconv" "strconv"
"strings" "strings"
"github.com/gorilla/mux" "github.com/gorilla/mux"
"github.com/minio/minio/internal/kms"
"github.com/minio/minio/internal/logger" "github.com/minio/minio/internal/logger"
"github.com/minio/pkg/bucket/policy" "github.com/minio/pkg/bucket/policy"
@ -100,7 +100,8 @@ func (api objectAPIHandlers) ListObjectVersionsHandler(w http.ResponseWriter, r
return return
} }
if err = DecryptETags(ctx, GlobalKMS, listObjectVersionsInfo.Objects, kms.BatchSize()); err != nil { if err = DecryptETags(ctx, GlobalKMS, listObjectVersionsInfo.Objects); err != nil {
logger.LogIf(ctx, fmt.Errorf("Failed to decrypt ETag: %v", err)) // TODO(aead): Remove once we are confident that decryption does not fail accidentially
writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL) writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL)
return return
} }
@ -165,7 +166,8 @@ func (api objectAPIHandlers) ListObjectsV2MHandler(w http.ResponseWriter, r *htt
return return
} }
if err = DecryptETags(ctx, GlobalKMS, listObjectsV2Info.Objects, kms.BatchSize()); err != nil { if err = DecryptETags(ctx, GlobalKMS, listObjectsV2Info.Objects); err != nil {
logger.LogIf(ctx, fmt.Errorf("Failed to decrypt ETag: %v", err)) // TODO(aead): Remove once we are confident that decryption does not fail accidentially
writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL) writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL)
return return
} }
@ -243,7 +245,8 @@ func (api objectAPIHandlers) ListObjectsV2Handler(w http.ResponseWriter, r *http
return return
} }
if err = DecryptETags(ctx, GlobalKMS, listObjectsV2Info.Objects, kms.BatchSize()); err != nil { if err = DecryptETags(ctx, GlobalKMS, listObjectsV2Info.Objects); err != nil {
logger.LogIf(ctx, fmt.Errorf("Failed to decrypt ETag: %v", err)) // TODO(aead): Remove once we are confident that decryption does not fail accidentially
writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL) writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL)
return return
} }
@ -343,7 +346,8 @@ func (api objectAPIHandlers) ListObjectsV1Handler(w http.ResponseWriter, r *http
return return
} }
if err = DecryptETags(ctx, GlobalKMS, listObjectsInfo.Objects, kms.BatchSize()); err != nil { if err = DecryptETags(ctx, GlobalKMS, listObjectsInfo.Objects); err != nil {
logger.LogIf(ctx, fmt.Errorf("Failed to decrypt ETag: %v", err)) // TODO(aead): Remove once we are confident that decryption does not fail accidentially
writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL) writeErrorResponse(ctx, w, toAPIError(ctx, err), r.URL)
return return
} }

View File

@ -96,102 +96,104 @@ func kmsKeyIDFromMetadata(metadata map[string]string) string {
return ARNPrefix + kmsID return ARNPrefix + kmsID
} }
// DecryptETags dectypts all ObjectInfo ETags, if encrypted, using the KMS. // DecryptETags decryptes the ETag of all ObjectInfos using the KMS.
func DecryptETags(ctx context.Context, KMS kms.KMS, objects []ObjectInfo, batchSize int) error { //
// It adjusts the size of all encrypted objects since encrypted
// objects are slightly larger due to encryption overhead.
// Further, it decrypts all single-part SSE-S3 encrypted objects
// and formats ETags of SSE-C / SSE-KMS encrypted objects to
// be AWS S3 compliant.
//
// DecryptETags uses a KMS bulk decryption API, if available, which
// is more efficient than decrypting ETags sequentually.
func DecryptETags(ctx context.Context, KMS kms.KMS, objects []ObjectInfo) error {
const BatchSize = 250 // We process the objects in batches - 250 is a reasonable default.
var ( var (
metadata = make([]map[string]string, 0, batchSize) metadata = make([]map[string]string, 0, BatchSize)
buckets = make([]string, 0, batchSize) buckets = make([]string, 0, BatchSize)
names = make([]string, 0, batchSize) names = make([]string, 0, BatchSize)
) )
for len(objects) > 0 { for len(objects) > 0 {
var N int N := BatchSize
if len(objects) < batchSize { if len(objects) < BatchSize {
N = len(objects) N = len(objects)
} else {
N = batchSize
} }
batch := objects[:N]
// We have to conntect the KMS only if there is at least // We have to decrypt only ETags of SSE-S3 single-part
// one SSE-S3 single-part object. SSE-C and SSE-KMS objects // objects.
// don't return the plaintext MD5 ETag and the ETag of // Therefore, we remember which objects (there index)
// SSE-S3 multipart objects is not encrypted. // in the current batch are single-part SSE-S3 objects.
// Therefore, we can skip the expensive KMS calls whenever metadata = metadata[:0:N]
// there is no single-part SSE-S3 object entirely. buckets = buckets[:0:N]
var containsSSES3SinglePart bool names = names[:0:N]
for _, object := range objects[:N] { SSES3SinglePartObjects := make(map[int]bool)
for i, object := range batch {
if kind, ok := crypto.IsEncrypted(object.UserDefined); ok && kind == crypto.S3 && !crypto.IsMultiPart(object.UserDefined) { if kind, ok := crypto.IsEncrypted(object.UserDefined); ok && kind == crypto.S3 && !crypto.IsMultiPart(object.UserDefined) {
containsSSES3SinglePart = true SSES3SinglePartObjects[i] = true
break
metadata = append(metadata, object.UserDefined)
buckets = append(buckets, object.Bucket)
names = append(names, object.Name)
} }
} }
if !containsSSES3SinglePart {
for i := range objects[:N] { // If there are no SSE-S3 single-part objects
size, err := objects[i].GetActualSize() // we can skip the decryption process. However,
// we still have to adjust the size and ETag
// of SSE-C and SSE-KMS objects.
if len(SSES3SinglePartObjects) == 0 {
for i := range batch {
size, err := batch[i].GetActualSize()
if err != nil { if err != nil {
return err return err
} }
objects[i].Size = size batch[i].Size = size
objects[i].ETag = objects[i].GetActualETag(nil)
if _, ok := crypto.IsEncrypted(batch[i].UserDefined); ok {
ETag, err := etag.Parse(batch[i].ETag)
if err != nil {
return err
}
batch[i].ETag = ETag.Format().String()
}
} }
objects = objects[N:] objects = objects[N:]
continue continue
} }
// Now, there are some SSE-S3 single-part objects. // There is at least one SSE-S3 single-part object.
// We only request the decryption keys for them. // For all SSE-S3 single-part objects we have to
// We don't want to get the decryption keys for multipart // fetch their decryption keys. We do this using
// or non-SSE-S3 objects. // a Bulk-Decryption API call, if available.
// keys, err := crypto.S3.UnsealObjectKeys(ctx, KMS, metadata, buckets, names)
// Therefore, we keep a map of indicies to remember which
// object was an SSE-S3 single-part object.
// Then we request the decryption keys for these objects.
// Finally, we decrypt the ETags of these objects using
// the decryption keys.
// However, we must also adjust the size and ETags of all
// objects (not just the SSE-S3 single part objects).
// For example, the ETag of SSE-KMS objects are random values
// and the size of an SSE-KMS object must be adjusted as well.
SSES3Objects := make(map[int]bool, 10)
metadata = metadata[:0:N]
buckets = buckets[:0:N]
names = names[:0:N]
for i, object := range objects[:N] {
if kind, ok := crypto.IsEncrypted(object.UserDefined); ok && kind == crypto.S3 && !crypto.IsMultiPart(object.UserDefined) {
metadata = append(metadata, object.UserDefined)
buckets = append(buckets, object.Bucket)
names = append(names, object.Name)
SSES3Objects[i] = true
}
}
keys, err := crypto.S3.UnsealObjectKeys(KMS, metadata, buckets, names)
if err != nil { if err != nil {
return err return err
} }
var keyIndex int
for i := range objects[:N] { // Now, we have to decrypt the ETags of SSE-S3 single-part
size, err := objects[i].GetActualSize() // objects and adjust the size and ETags of all encrypted
// objects.
for i := range batch {
size, err := batch[i].GetActualSize()
if err != nil { if err != nil {
return err return err
} }
objects[i].Size = size batch[i].Size = size
if !SSES3Objects[i] { if _, ok := crypto.IsEncrypted(batch[i].UserDefined); ok {
objects[i].ETag = objects[i].GetActualETag(nil) ETag, err := etag.Parse(batch[i].ETag)
} else {
ETag, err := etag.Parse(objects[i].ETag)
if err != nil { if err != nil {
return err return err
} }
if ETag.IsEncrypted() { if SSES3SinglePartObjects[i] && ETag.IsEncrypted() {
tag, err := keys[keyIndex].UnsealETag(ETag) ETag, err = etag.Decrypt(keys[0][:], ETag)
if err != nil { if err != nil {
return err return err
} }
ETag = etag.ETag(tag) keys = keys[1:]
objects[i].ETag = ETag.String()
} }
keyIndex++ batch[i].ETag = ETag.Format().String()
} }
} }
objects = objects[N:] objects = objects[N:]

View File

@ -386,13 +386,13 @@ func getHostFromSrv(records []dns.SrvRecord) (host string) {
} }
// IsCompressed returns true if the object is marked as compressed. // IsCompressed returns true if the object is marked as compressed.
func (o ObjectInfo) IsCompressed() bool { func (o *ObjectInfo) IsCompressed() bool {
_, ok := o.UserDefined[ReservedMetadataPrefix+"compression"] _, ok := o.UserDefined[ReservedMetadataPrefix+"compression"]
return ok return ok
} }
// IsCompressedOK returns whether the object is compressed and can be decompressed. // IsCompressedOK returns whether the object is compressed and can be decompressed.
func (o ObjectInfo) IsCompressedOK() (bool, error) { func (o *ObjectInfo) IsCompressedOK() (bool, error) {
scheme, ok := o.UserDefined[ReservedMetadataPrefix+"compression"] scheme, ok := o.UserDefined[ReservedMetadataPrefix+"compression"]
if !ok { if !ok {
return false, nil return false, nil
@ -404,17 +404,8 @@ func (o ObjectInfo) IsCompressedOK() (bool, error) {
return true, fmt.Errorf("unknown compression scheme: %s", scheme) return true, fmt.Errorf("unknown compression scheme: %s", scheme)
} }
// GetActualETag - returns the actual etag of the stored object
// decrypts SSE objects.
func (o ObjectInfo) GetActualETag(h http.Header) string {
if _, ok := crypto.IsEncrypted(o.UserDefined); !ok {
return o.ETag
}
return getDecryptedETag(h, o, false)
}
// GetActualSize - returns the actual size of the stored object // GetActualSize - returns the actual size of the stored object
func (o ObjectInfo) GetActualSize() (int64, error) { func (o *ObjectInfo) GetActualSize() (int64, error) {
if o.IsCompressed() { if o.IsCompressed() {
sizeStr, ok := o.UserDefined[ReservedMetadataPrefix+"actual-size"] sizeStr, ok := o.UserDefined[ReservedMetadataPrefix+"actual-size"]
if !ok { if !ok {

View File

@ -92,7 +92,7 @@ func (s3 sses3) UnsealObjectKey(KMS kms.KMS, metadata map[string]string, bucket,
// keys. // keys.
// //
// The metadata, buckets and objects slices must have the same length. // The metadata, buckets and objects slices must have the same length.
func (s3 sses3) UnsealObjectKeys(KMS kms.KMS, metadata []map[string]string, buckets, objects []string) ([]ObjectKey, error) { func (s3 sses3) UnsealObjectKeys(ctx context.Context, KMS kms.KMS, metadata []map[string]string, buckets, objects []string) ([]ObjectKey, error) {
if KMS == nil { if KMS == nil {
return nil, Errorf("KMS not configured") return nil, Errorf("KMS not configured")
} }
@ -124,7 +124,7 @@ func (s3 sses3) UnsealObjectKeys(KMS kms.KMS, metadata []map[string]string, buck
for i := range buckets { for i := range buckets {
contexts = append(contexts, kms.Context{buckets[i]: path.Join(buckets[i], objects[i])}) contexts = append(contexts, kms.Context{buckets[i]: path.Join(buckets[i], objects[i])})
} }
unsealKeys, err := KMS.DecryptAll(keyIDs[0], kmsKeys, contexts) unsealKeys, err := KMS.DecryptAll(ctx, keyIDs[0], kmsKeys, contexts)
if err != nil { if err != nil {
return nil, err return nil, err
} }

View File

@ -162,7 +162,7 @@ func (c *kesClient) DecryptKey(keyID string, ciphertext []byte, ctx Context) ([]
return c.client.Decrypt(context.Background(), keyID, ciphertext, ctxBytes) return c.client.Decrypt(context.Background(), keyID, ciphertext, ctxBytes)
} }
func (c *kesClient) DecryptAll(keyID string, ciphertexts [][]byte, contexts []Context) ([][]byte, error) { func (c *kesClient) DecryptAll(ctx context.Context, keyID string, ciphertexts [][]byte, contexts []Context) ([][]byte, error) {
if c.bulkAvailable { if c.bulkAvailable {
CCPs := make([]kes.CCP, 0, len(ciphertexts)) CCPs := make([]kes.CCP, 0, len(ciphertexts))
for i := range ciphertexts { for i := range ciphertexts {
@ -175,7 +175,7 @@ func (c *kesClient) DecryptAll(keyID string, ciphertexts [][]byte, contexts []Co
Context: bCtx, Context: bCtx,
}) })
} }
PCPs, err := c.client.DecryptAll(context.Background(), keyID, CCPs...) PCPs, err := c.client.DecryptAll(ctx, keyID, CCPs...)
if err != nil { if err != nil {
return nil, err return nil, err
} }

View File

@ -18,12 +18,11 @@
package kms package kms
import ( import (
"context"
"encoding" "encoding"
"encoding/json" "encoding/json"
"strconv"
jsoniter "github.com/json-iterator/go" jsoniter "github.com/json-iterator/go"
"github.com/minio/pkg/env"
) )
// KMS is the generic interface that abstracts over // KMS is the generic interface that abstracts over
@ -57,19 +56,7 @@ type KMS interface {
// DecryptAll decrypts all ciphertexts with the key referenced // DecryptAll decrypts all ciphertexts with the key referenced
// by the key ID. The contexts must match the context value // by the key ID. The contexts must match the context value
// used to generate the ciphertexts. // used to generate the ciphertexts.
DecryptAll(keyID string, ciphertext [][]byte, context []Context) ([][]byte, error) DecryptAll(ctx context.Context, keyID string, ciphertext [][]byte, context []Context) ([][]byte, error)
}
// BatchSize returns the size of the batches that should be used during
// KES bulk decryption API calls.
func BatchSize() int {
const DefaultBatchSize = 500
v := env.Get("MINIO_KMS_KES_BULK_API_BATCH_SIZE", strconv.Itoa(DefaultBatchSize))
n, err := strconv.Atoi(v)
if err != nil {
return DefaultBatchSize
}
return n
} }
// Status describes the current state of a KMS. // Status describes the current state of a KMS.

View File

@ -18,6 +18,7 @@
package kms package kms
import ( import (
"context"
"crypto/aes" "crypto/aes"
"crypto/cipher" "crypto/cipher"
"crypto/hmac" "crypto/hmac"
@ -224,7 +225,7 @@ func (kms secretKey) DecryptKey(keyID string, ciphertext []byte, context Context
return plaintext, nil return plaintext, nil
} }
func (kms secretKey) DecryptAll(keyID string, ciphertexts [][]byte, contexts []Context) ([][]byte, error) { func (kms secretKey) DecryptAll(_ context.Context, keyID string, ciphertexts [][]byte, contexts []Context) ([][]byte, error) {
plaintexts := make([][]byte, 0, len(ciphertexts)) plaintexts := make([][]byte, 0, len(ciphertexts))
for i := range ciphertexts { for i := range ciphertexts {
plaintext, err := kms.DecryptKey(keyID, ciphertexts[i], contexts[i]) plaintext, err := kms.DecryptKey(keyID, ciphertexts[i], contexts[i])