xl: Fix ReadFile to keep the order always for reading the data back. (#1339)

Also fixes a stackoverflow bug in namespace locking.

object-api-multipart.go

@@ -297,6 +297,9 @@ func (o objectAPI) NewMultipartUpload(bucket, object string) (string, *probe.Err
 			// uploadIDPath doesn't exist, so create empty file to reserve the name
 			var w io.WriteCloser
 			if w, e = o.storage.CreateFile(minioMetaVolume, uploadIDPath); e == nil {
+				// Just write some data for erasure code, rather than zero bytes.
+				w.Write([]byte(uploadID))
+				// Close the writer.
 				if e = w.Close(); e != nil {
 					return "", probe.NewError(e)
 				}

xl-v1-createfile.go

@@ -55,6 +55,9 @@ func (xl XL) writeErasure(volume, path string, reader *io.PipeReader) {
 	var sha512Writers = make([]hash.Hash, len(xl.storageDisks))
 	var metadataWriters = make([]io.WriteCloser, len(xl.storageDisks))
 
+	// Save additional erasureMetadata.
+	modTime := time.Now().UTC()
+
 	// Initialize storage disks, get all the writers and corresponding
 	// metadata writers.
 	for index, disk := range xl.storageDisks {
@@ -131,9 +134,6 @@ func (xl XL) writeErasure(volume, path string, reader *io.PipeReader) {
 		}
 	}
 
-	// Save additional erasureMetadata.
-	modTime := time.Now().UTC()
-
 	// Initialize metadata map, save all erasure related metadata.
 	metadata := make(map[string]string)
 	metadata["version"] = minioVersion

xl-v1-namespace.go

@@ -27,39 +27,38 @@ type nameSpaceParam struct {
 // nameSpaceLock - provides primitives for locking critical namespace regions.
 type nameSpaceLock struct {
 	rwMutex *sync.RWMutex
-	rcount  uint
-	wcount  uint
+	count   uint
 }
 
 func (nsLock *nameSpaceLock) InUse() bool {
-	return nsLock.rcount != 0 || nsLock.wcount != 0
+	return nsLock.count != 0
 }
 
 // Lock acquires write lock and increments the namespace counter.
 func (nsLock *nameSpaceLock) Lock() {
-	nsLock.Lock()
-	nsLock.wcount++
+	nsLock.rwMutex.Lock()
+	nsLock.count++
 }
 
 // Unlock releases write lock and decrements the namespace counter.
 func (nsLock *nameSpaceLock) Unlock() {
-	nsLock.Unlock()
-	if nsLock.wcount != 0 {
-		nsLock.wcount--
+	nsLock.rwMutex.Unlock()
+	if nsLock.count != 0 {
+		nsLock.count--
 	}
 }
 
 // RLock acquires read lock and increments the namespace counter.
 func (nsLock *nameSpaceLock) RLock() {
-	nsLock.RLock()
-	nsLock.rcount++
+	nsLock.rwMutex.RLock()
+	nsLock.count++
 }
 
 // RUnlock release read lock and decrements the namespace counter.
 func (nsLock *nameSpaceLock) RUnlock() {
-	nsLock.RUnlock()
-	if nsLock.rcount != 0 {
-		nsLock.rcount--
+	nsLock.rwMutex.RUnlock()
+	if nsLock.count != 0 {
+		nsLock.count--
 	}
 }
 
@@ -67,7 +66,6 @@ func (nsLock *nameSpaceLock) RUnlock() {
 func newNSLock() *nameSpaceLock {
 	return &nameSpaceLock{
 		rwMutex: &sync.RWMutex{},
-		rcount:  0,
-		wcount:  0,
+		count:   0,
 	}
 }

xl-v1-readfile.go

@@ -44,38 +44,37 @@ func getEncodedBlockLen(inputLen, dataBlocks int) (curBlockSize int) {
 	return
 }
 
-func (xl XL) getMetaDataFileVersions(volume, path string) (diskVersionMap map[StorageAPI]int64) {
+func (xl XL) getMetaFileVersionMap(volume, path string) (diskFileVersionMap map[int]int64) {
 	metadataFilePath := slashpath.Join(path, metadataFile)
-	// set offset to 0 to read entire file
+	// Set offset to 0 to read entire file.
 	offset := int64(0)
 	metadata := make(map[string]string)
 
-	// read meta data from all disks
-	for _, disk := range xl.storageDisks {
-		diskVersionMap[disk] = -1
+	// Allocate disk index format map - do not use maps directly without allocating.
+	diskFileVersionMap = make(map[int]int64)
+
+	// TODO - all errors should be logged here.
 
-		if metadataReader, err := disk.ReadFile(volume, metadataFilePath, offset); err != nil {
-			// error reading meta data file
-			// TODO: log it
+	// Read meta data from all disks
+	for index, disk := range xl.storageDisks {
+		diskFileVersionMap[index] = -1
+
+		metadataReader, err := disk.ReadFile(volume, metadataFilePath, offset)
+		if err != nil {
 			continue
-		} else if err := json.NewDecoder(metadataReader).Decode(&metadata); err != nil {
-			// error in parsing json
-			// TODO: log it
+		} else if err = json.NewDecoder(metadataReader).Decode(&metadata); err != nil {
 			continue
 		} else if _, ok := metadata["file.version"]; !ok {
-			// missing "file.version" is completely valid
-			diskVersionMap[disk] = 0
-			continue
-		} else if fileVersion, err := strconv.ParseInt(metadata["file.version"], 10, 64); err != nil {
-			// version is not a number
-			// TODO: log it
+			diskFileVersionMap[index] = 0
+		}
+		// Convert string to integer.
+		fileVersion, err := strconv.ParseInt(metadata["file.version"], 10, 64)
+		if err != nil {
 			continue
-		} else {
-			diskVersionMap[disk] = fileVersion
 		}
+		diskFileVersionMap[index] = fileVersion
 	}
-
-	return
+	return diskFileVersionMap
 }
 
 type quorumDisk struct {
@@ -83,24 +82,28 @@ type quorumDisk struct {
 	index int
 }
 
+// getReadFileQuorumDisks - get the current quorum disks.
 func (xl XL) getReadFileQuorumDisks(volume, path string) (quorumDisks []quorumDisk) {
-	diskVersionMap := xl.getMetaDataFileVersions(volume, path)
+	diskVersionMap := xl.getMetaFileVersionMap(volume, path)
 	higherVersion := int64(0)
-	i := 0
-	for disk, version := range diskVersionMap {
-		if version > higherVersion {
-			higherVersion = version
-			quorumDisks = []quorumDisk{{disk, i}}
-		} else if version == higherVersion {
-			quorumDisks = append(quorumDisks, quorumDisk{disk, i})
+	for diskIndex, formatVersion := range diskVersionMap {
+		if formatVersion > higherVersion {
+			higherVersion = formatVersion
+			quorumDisks = []quorumDisk{{
+				disk:  xl.storageDisks[diskIndex],
+				index: diskIndex,
+			}}
+		} else if formatVersion == higherVersion {
+			quorumDisks = append(quorumDisks, quorumDisk{
+				disk:  xl.storageDisks[diskIndex],
+				index: diskIndex,
+			})
 		}
-
-		i++
 	}
-
 	return
 }
 
+// getFileSize - extract file size from metadata.
 func (xl XL) getFileSize(volume, path string, disk StorageAPI) (size int64, err error) {
 	metadataFilePath := slashpath.Join(path, metadataFile)
 	// set offset to 0 to read entire file
@@ -133,10 +136,10 @@ func (xl XL) ReadFile(volume, path string, offset int64) (io.ReadCloser, error)
 		return nil, errInvalidArgument
 	}
 
-	// Acquire a read lock. - TODO - disable this due to stack overflow bug.
-	// readLock := true
-	// xl.lockNS(volume, path, readLock)
-	// defer xl.unlockNS(volume, path, readLock)
+	// Acquire a read lock.
+	readLock := true
+	xl.lockNS(volume, path, readLock)
+	defer xl.unlockNS(volume, path, readLock)
 
 	// Check read quorum.
 	quorumDisks := xl.getReadFileQuorumDisks(volume, path)
@@ -151,26 +154,30 @@ func (xl XL) ReadFile(volume, path string, offset int64) (io.ReadCloser, error)
 	}
 	totalBlocks := xl.DataBlocks + xl.ParityBlocks // Total blocks.
 
-	readers := []io.ReadCloser{}
+	readers := make([]io.ReadCloser, len(quorumDisks))
 	readFileError := 0
-	i := 0
 	for _, quorumDisk := range quorumDisks {
 		erasurePart := slashpath.Join(path, fmt.Sprintf("part.%d", quorumDisk.index))
 		var erasuredPartReader io.ReadCloser
 		if erasuredPartReader, err = quorumDisk.disk.ReadFile(volume, erasurePart, offset); err != nil {
-			// we can safely allow ReadFile errors up to len(quorumDisks) - xl.readQuorum
+			// We can safely allow ReadFile errors up to len(quorumDisks) - xl.readQuorum
 			// otherwise return failure
 			if readFileError < len(quorumDisks)-xl.readQuorum {
+				// Set the reader to 'nil' to be able to reconstruct later.
+				readers[quorumDisk.index] = nil
 				readFileError++
 				continue
 			}
-
-			// TODO: handle currently available io.Reader in readers variable
-			return nil, err
+			// Control reaches here we do not have quorum
+			// anymore. Close all the readers.
+			for _, reader := range readers {
+				if reader != nil {
+					reader.Close()
+				}
+			}
+			return nil, errReadQuorum
 		}
-
-		readers[i] = erasuredPartReader
-		i++
+		readers[quorumDisk.index] = erasuredPartReader
 	}
 
 	// Initialize pipe.