Make minio package support legacy MD5 checksum (#23768) (#23770)

Backport #23768 (no source code conflict, only some unrelated
docs/test-ini conflicts)

Some storages like:

 * https://developers.cloudflare.com/r2/api/s3/api/
 * https://www.backblaze.com/b2/docs/s3_compatible_api.html

They do not support "x-amz-checksum-algorithm" header

But minio recently uses that header with CRC32C by default. So we have
to tell minio to use legacy MD5 checksum.

cmd/migrate_storage.go

@@ -78,6 +78,15 @@ var CmdMigrateStorage = cli.Command{
 			Name:  "minio-use-ssl",
 			Usage: "Enable SSL for minio",
 		},
+		cli.BoolFlag{
+			Name:  "minio-insecure-skip-verify",
+			Usage: "Skip SSL verification",
+		},
+		cli.StringFlag{
+			Name:  "minio-checksum-algorithm",
+			Value: "",
+			Usage: "Minio checksum algorithm (default/md5)",
+		},
 	},
 }
 
@@ -175,6 +184,8 @@ func runMigrateStorage(ctx *cli.Context) error {
 				Location:           ctx.String("minio-location"),
 				BasePath:           ctx.String("minio-base-path"),
 				UseSSL:             ctx.Bool("minio-use-ssl"),
+				InsecureSkipVerify: ctx.Bool("minio-insecure-skip-verify"),
+				ChecksumAlgorithm:  ctx.String("minio-checksum-algorithm"),
 			})
 	default:
 		return fmt.Errorf("unsupported storage type: %s", ctx.String("storage"))

custom/conf/app.example.ini

@@ -1874,6 +1874,9 @@ ROUTER = console
 ;;
 ;; Minio skip SSL verification available when STORAGE_TYPE is `minio`
 ;MINIO_INSECURE_SKIP_VERIFY = false
+;;
+;; Minio checksum algorithm: default (for MinIO or AWS S3) or md5 (for Cloudflare or Backblaze)
+;MINIO_CHECKSUM_ALGORITHM = default
 
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

docs/content/doc/administration/config-cheat-sheet.en-us.md

@@ -855,6 +855,7 @@ Default templates for project boards:
 - `MINIO_BASE_PATH`: **attachments/**: Minio base path on the bucket only available when STORAGE_TYPE is `minio`
 - `MINIO_USE_SSL`: **false**: Minio enabled ssl only available when STORAGE_TYPE is `minio`
 - `MINIO_INSECURE_SKIP_VERIFY`: **false**: Minio skip SSL verification available when STORAGE_TYPE is `minio`
+- `MINIO_CHECKSUM_ALGORITHM`: **default**: Minio checksum algorithm: `default` (for MinIO or AWS S3) or `md5` (for Cloudflare or Backblaze)
 
 ## Log (`log`)
 

modules/lfs/content_store.go

@@ -59,15 +59,22 @@ func (s *ContentStore) Put(pointer Pointer, r io.Reader) error {
 		return err
 	}
 
-	// This shouldn't happen but it is sensible to test
-	if written != pointer.Size {
-		if err := s.Delete(p); err != nil {
-			log.Error("Cleaning the LFS OID[%s] failed: %v", pointer.Oid, err)
-		}
-		return ErrSizeMismatch
+	// check again whether there is any error during the Save operation
+	// because some errors might be ignored by the Reader's caller
+	if wrappedRd.lastError != nil && !errors.Is(wrappedRd.lastError, io.EOF) {
+		err = wrappedRd.lastError
+	} else if written != pointer.Size {
+		err = ErrSizeMismatch
 	}
 
-	return nil
+	// if the upload failed, try to delete the file
+	if err != nil {
+		if errDel := s.Delete(p); errDel != nil {
+			log.Error("Cleaning the LFS OID[%s] failed: %v", pointer.Oid, errDel)
+		}
+	}
+
+	return err
 }
 
 // Exists returns true if the object exists in the content store.
@@ -108,6 +115,17 @@ type hashingReader struct {
 	expectedSize int64
 	hash         hash.Hash
 	expectedHash string
+	lastError    error
+}
+
+// recordError records the last error during the Save operation
+// Some callers of the Reader doesn't respect the returned "err"
+// For example, MinIO's Put will ignore errors if the written size could equal to expected size
+// So we must remember the error by ourselves,
+// and later check again whether ErrSizeMismatch or ErrHashMismatch occurs during the Save operation
+func (r *hashingReader) recordError(err error) error {
+	r.lastError = err
+	return err
 }
 
 func (r *hashingReader) Read(b []byte) (int, error) {
@@ -117,22 +135,22 @@ func (r *hashingReader) Read(b []byte) (int, error) {
 		r.currentSize += int64(n)
 		wn, werr := r.hash.Write(b[:n])
 		if wn != n || werr != nil {
-			return n, werr
+			return n, r.recordError(werr)
 		}
 	}
 
-	if err != nil && err == io.EOF {
+	if errors.Is(err, io.EOF) || r.currentSize >= r.expectedSize {
 		if r.currentSize != r.expectedSize {
-			return n, ErrSizeMismatch
+			return n, r.recordError(ErrSizeMismatch)
 		}
 
 		shaStr := hex.EncodeToString(r.hash.Sum(nil))
 		if shaStr != r.expectedHash {
-			return n, ErrHashMismatch
+			return n, r.recordError(ErrHashMismatch)
 		}
 	}
 
-	return n, err
+	return n, r.recordError(err)
 }
 
 func newHashingReader(expectedSize int64, expectedHash string, reader io.Reader) *hashingReader {

modules/setting/storage.go

@@ -42,6 +42,7 @@ func getStorage(rootCfg ConfigProvider, name, typ string, targetSec *ini.Section
 	sec.Key("MINIO_LOCATION").MustString("us-east-1")
 	sec.Key("MINIO_USE_SSL").MustBool(false)
 	sec.Key("MINIO_INSECURE_SKIP_VERIFY").MustBool(false)
+	sec.Key("MINIO_CHECKSUM_ALGORITHM").MustString("default")
 
 	if targetSec == nil {
 		targetSec, _ = rootCfg.NewSection(name)

modules/storage/minio.go

@@ -6,6 +6,7 @@ package storage
 import (
 	"context"
 	"crypto/tls"
+	"fmt"
 	"io"
 	"net/http"
 	"net/url"
@@ -52,10 +53,12 @@ type MinioStorageConfig struct {
 	BasePath           string `ini:"MINIO_BASE_PATH"`
 	UseSSL             bool   `ini:"MINIO_USE_SSL"`
 	InsecureSkipVerify bool   `ini:"MINIO_INSECURE_SKIP_VERIFY"`
+	ChecksumAlgorithm  string `ini:"MINIO_CHECKSUM_ALGORITHM"`
 }
 
 // MinioStorage returns a minio bucket storage
 type MinioStorage struct {
+	cfg      *MinioStorageConfig
 	ctx      context.Context
 	client   *minio.Client
 	bucket   string
@@ -90,6 +93,10 @@ func NewMinioStorage(ctx context.Context, cfg interface{}) (ObjectStorage, error
 	}
 	config := configInterface.(MinioStorageConfig)
 
+	if config.ChecksumAlgorithm != "" && config.ChecksumAlgorithm != "default" && config.ChecksumAlgorithm != "md5" {
+		return nil, fmt.Errorf("invalid minio checksum algorithm: %s", config.ChecksumAlgorithm)
+	}
+
 	log.Info("Creating Minio storage at %s:%s with base path %s", config.Endpoint, config.Bucket, config.BasePath)
 
 	minioClient, err := minio.New(config.Endpoint, &minio.Options{
@@ -112,6 +119,7 @@ func NewMinioStorage(ctx context.Context, cfg interface{}) (ObjectStorage, error
 	}
 
 	return &MinioStorage{
+		cfg:      &config,
 		ctx:      ctx,
 		client:   minioClient,
 		bucket:   config.Bucket,
@@ -123,7 +131,7 @@ func (m *MinioStorage) buildMinioPath(p string) string {
 	return strings.TrimPrefix(path.Join(m.basePath, path.Clean("/" + strings.ReplaceAll(p, "\\", "/"))[1:]), "/")
 }
 
-// Open open a file
+// Open opens a file
 func (m *MinioStorage) Open(path string) (Object, error) {
 	opts := minio.GetObjectOptions{}
 	object, err := m.client.GetObject(m.ctx, m.bucket, m.buildMinioPath(path), opts)
@@ -133,7 +141,7 @@ func (m *MinioStorage) Open(path string) (Object, error) {
 	return &minioObject{object}, nil
 }
 
-// Save save a file to minio
+// Save saves a file to minio
 func (m *MinioStorage) Save(path string, r io.Reader, size int64) (int64, error) {
 	uploadInfo, err := m.client.PutObject(
 		m.ctx,
@@ -141,7 +149,14 @@ func (m *MinioStorage) Save(path string, r io.Reader, size int64) (int64, error)
 		m.buildMinioPath(path),
 		r,
 		size,
-		minio.PutObjectOptions{ContentType: "application/octet-stream"},
+		minio.PutObjectOptions{
+			ContentType: "application/octet-stream",
+			// some storages like:
+			// * https://developers.cloudflare.com/r2/api/s3/api/
+			// * https://www.backblaze.com/b2/docs/s3_compatible_api.html
+			// do not support "x-amz-checksum-algorithm" header, so use legacy MD5 checksum
+			SendContentMd5: m.cfg.ChecksumAlgorithm == "md5",
+		},
 	)
 	if err != nil {
 		return 0, convertMinioErr(err)

tests/mysql.ini.tmpl

@@ -76,6 +76,7 @@ MINIO_SECRET_ACCESS_KEY = 12345678
 MINIO_BUCKET = gitea
 MINIO_LOCATION = us-east-1
 MINIO_USE_SSL = false
+MINIO_CHECKSUM_ALGORITHM = md5
 
 [mailer]
 ENABLED = true