// Data buffer to decode must be of the (M/K)*Size given in the constructor.
// The error list must contain M-K values, corresponding to the shards
// with errors (eg. [0, 2, 4, 6]).
// Cache stores the decode matrices in a trie, enabling a faster decode
// with a memory tradeoff.
// The returned decoded data is the orignal data of length Size
func (c *Code) Decode(encoded []byte, errList []byte, cache bool) (recovered []byte) {
if len(encoded) != c.M*c.ShardLength {
panic("Data to decode is not the proper size")
}
if len(errList) > c.M-c.K {
panic("Too many errors, cannot decode")
}
if len(errList) == 0 {
recovered = append(recovered, encoded[:c.K*c.ShardLength]...)
} else {
node := c.getDecode(errList, cache)
for i := 0; i < c.K; i++ {
recovered = append(recovered, encoded[(int(node.decodeIndex[i])*c.ShardLength):int(node.decodeIndex[i]+1)*c.ShardLength]...)
}
decoded := make([]byte, c.M*c.ShardLength)
C.ec_encode_data(C.int(c.ShardLength), C.int(c.K), C.int(c.M), (*C.uchar)(&node.galoisTables[0]), (*C.uchar)(&recovered[0]), (*C.uchar)(&decoded[0]))
copy(recovered, encoded)
for i, err := range errList {
if int(err) < c.K {
copy(recovered[int(err)*c.ShardLength:int(err+1)*c.ShardLength], decoded[i*c.ShardLength:(i+1)*c.ShardLength])
}
}
}
return recovered
}
// Encode erasure codes a block of data in "k" data blocks and "m" parity blocks.
// Output is [k+m][]blocks of data and parity slices.
func (e *Erasure) Encode(inputData []byte) (encodedBlocks [][]byte, err error) {
e.mutex.Lock()
defer e.mutex.Unlock()
k := int(e.params.K) // "k" data blocks
m := int(e.params.M) // "m" parity blocks
n := k + m // "n" total encoded blocks
// Length of a single encoded chunk.
// Total number of encoded chunks = "k" data + "m" parity blocks
encodedBlockLen := GetEncodedBlockLen(len(inputData), uint8(k))
// Length of total number of "k" data chunks
encodedDataBlocksLen := encodedBlockLen * k
// Length of extra padding required for the data blocks.
encodedDataBlocksPadLen := encodedDataBlocksLen - len(inputData)
// Extend inputData buffer to accommodate coded data blocks if necesssary
if encodedDataBlocksPadLen > 0 {
padding := make([]byte, encodedDataBlocksPadLen)
// Expand with new padded blocks to the byte array
inputData = append(inputData, padding...)
}
// Extend inputData buffer to accommodate coded parity blocks
{ // Local Scope
encodedParityBlocksLen := encodedBlockLen * m
parityBlocks := make([]byte, encodedParityBlocksLen)
inputData = append(inputData, parityBlocks...)
}
// Allocate memory to the "encoded blocks" return buffer
encodedBlocks = make([][]byte, n) // Return buffer
// Neccessary to bridge Go to the C world. C requires 2D arry of pointers to
// byte array. "encodedBlocks" is a 2D slice.
pointersToEncodedBlock := make([]*byte, n) // Pointers to encoded blocks.
// Copy data block slices to encoded block buffer
for i := 0; i < k; i++ {
encodedBlocks[i] = inputData[i*encodedBlockLen : (i+1)*encodedBlockLen]
pointersToEncodedBlock[i] = &encodedBlocks[i][0]
}
// Copy erasure block slices to encoded block buffer
for i := k; i < n; i++ {
encodedBlocks[i] = make([]byte, encodedBlockLen)
pointersToEncodedBlock[i] = &encodedBlocks[i][0]
}
// Erasure code the data into K data blocks and M parity
// blocks. Only the parity blocks are filled. Data blocks remain
// intact.
C.ec_encode_data(C.int(encodedBlockLen), C.int(k), C.int(m), e.encodeTbls,
(**C.uchar)(unsafe.Pointer(&pointersToEncodedBlock[:k][0])), // Pointers to data blocks
(**C.uchar)(unsafe.Pointer(&pointersToEncodedBlock[k:][0]))) // Pointers to parity blocks
return encodedBlocks, nil
}
// The data buffer to encode must be of the length Size given in the constructor.
// The returned encoded buffer is (M-K)*Shard length, since the first Size bytes
// of the encoded data is just the original data due to the identity matrix.
func (c *Code) Encode(data []byte) (encoded []byte) {
if len(data) != c.K*c.ShardLength {
panic("Data to encode is not the proper size")
}
// Since the first k rows of the encode matrix is actually the identity matrix
// we only need to encode the last m-k shards of the data and append
// them to the original data
encoded = make([]byte, (c.M-c.K)*(c.ShardLength))
C.ec_encode_data(C.int(c.ShardLength), C.int(c.K), C.int(c.M-c.K), (*C.uchar)(&c.galoisTables[0]), (*C.uchar)(&data[0]), (*C.uchar)(&encoded[0]))
// return append(data, encoded...)
return encoded
}
// EncodeStream erasure codes a block of data in "k" data blocks and "m" parity blocks.
// Output is [k+m][]blocks of data and parity slices.
func (e *Erasure) EncodeStream(data io.Reader, size int64) ([][]byte, []byte, error) {
k := int(e.params.K) // "k" data blocks
m := int(e.params.M) // "m" parity blocks
n := k + m // "n" total encoded blocks
// Length of a single encoded chunk.
// Total number of encoded chunks = "k" data + "m" parity blocks
encodedBlockLen := GetEncodedBlockLen(int(size), uint8(k))
// Length of total number of "n" data chunks
encodedDataBlocksLen := encodedBlockLen * n
// allocate byte array for encodedBlock length
inputData := make([]byte, size, encodedDataBlocksLen)
_, err := io.ReadFull(data, inputData)
if err != nil {
// do not check for io.ErrUnexpectedEOF, we know the right amount of size
// to be read if its a short read we need to throw error since reader could
// have been prematurely closed.
if err != io.EOF {
return nil, nil, err
}
}
// Allocate memory to the "encoded blocks" return buffer
encodedBlocks := make([][]byte, n) // Return buffer
// Neccessary to bridge Go to the C world. C requires 2D arry of pointers to
// byte array. "encodedBlocks" is a 2D slice.
pointersToEncodedBlock := make([]*byte, n) // Pointers to encoded blocks.
// Copy data block slices to encoded block buffer
for i := 0; i < n; i++ {
encodedBlocks[i] = inputData[i*encodedBlockLen : (i+1)*encodedBlockLen]
pointersToEncodedBlock[i] = &encodedBlocks[i][0]
}
// Erasure code the data into K data blocks and M parity
// blocks. Only the parity blocks are filled. Data blocks remain
// intact.
C.ec_encode_data(C.int(encodedBlockLen), C.int(k), C.int(m), e.encodeTbls,
(**C.uchar)(unsafe.Pointer(&pointersToEncodedBlock[:k][0])), // Pointers to data blocks
(**C.uchar)(unsafe.Pointer(&pointersToEncodedBlock[k:][0]))) // Pointers to parity blocks
return encodedBlocks, inputData[0:size], nil
}
// Decode decodes erasure coded blocks of data into its original
// form. Erasure coded data contains K data blocks and M parity
// blocks. Decode can withstand data loss up to any M number of blocks.
//
// "encodedDataBlocks" is an array of K data blocks and M parity
// blocks. Data blocks are position and order dependent. Missing blocks
// are set to "nil". There must be at least "K" number of data|parity
// blocks.
//
// "dataLen" is the length of original source data
func (e *Erasure) Decode(encodedDataBlocks [][]byte, dataLen int) (decodedData []byte, err error) {
e.mutex.Lock()
defer e.mutex.Unlock()
var source, target **C.uchar
k := int(e.params.K)
m := int(e.params.M)
n := k + m
// We need the data and parity blocks preserved in the same order. Missing blocks are set to nil.
if len(encodedDataBlocks) != n {
msg := fmt.Sprintf("Encoded data blocks slice must of length [%d]", n)
return nil, errors.New(msg)
}
// Length of a single encoded block
encodedBlockLen := GetEncodedBlockLen(dataLen, uint8(k))
// Keep track of errors per block.
missingEncodedBlocks := make([]int, n+1)
var missingEncodedBlocksCount int
// Check for the missing encoded blocks
for i := range encodedDataBlocks {
if encodedDataBlocks[i] == nil || len(encodedDataBlocks[i]) == 0 {
missingEncodedBlocks[missingEncodedBlocksCount] = i
missingEncodedBlocksCount++
}
}
// Cannot reconstruct original data. Need at least M number of data or parity blocks.
if missingEncodedBlocksCount > m {
return nil, fmt.Errorf("Cannot reconstruct original data. Need at least [%d] data or parity blocks", m)
}
// Convert from Go int slice to C int array
missingEncodedBlocksC := intSlice2CIntArray(missingEncodedBlocks[:missingEncodedBlocksCount])
// Allocate buffer for the missing blocks
for i := range encodedDataBlocks {
if encodedDataBlocks[i] == nil || len(encodedDataBlocks[i]) == 0 {
encodedDataBlocks[i] = make([]byte, encodedBlockLen)
}
}
// If not already initialized, recompute and cache
if e.decodeMatrix == nil || e.decodeTbls == nil || e.decodeIndex == nil {
var decodeMatrix, decodeTbls *C.uchar
var decodeIndex *C.uint32_t
C.minio_init_decoder(missingEncodedBlocksC, C.int(k), C.int(n), C.int(missingEncodedBlocksCount),
e.encodeMatrix, &decodeMatrix, &decodeTbls, &decodeIndex)
// cache this for future needs
e.decodeMatrix = decodeMatrix
e.decodeTbls = decodeTbls
e.decodeIndex = decodeIndex
}
// Make a slice of pointers to encoded blocks. Necessary to bridge to the C world.
pointers := make([]*byte, n)
for i := range encodedDataBlocks {
pointers[i] = &encodedDataBlocks[i][0]
}
// Get pointers to source "data" and target "parity" blocks from the output byte array.
ret := C.minio_get_source_target(C.int(missingEncodedBlocksCount), C.int(k), C.int(m), missingEncodedBlocksC,
e.decodeIndex, (**C.uchar)(unsafe.Pointer(&pointers[0])), &source, &target)
if int(ret) == -1 {
return nil, errors.New("Unable to decode data")
}
// Decode data
C.ec_encode_data(C.int(encodedBlockLen), C.int(k), C.int(missingEncodedBlocksCount), e.decodeTbls,
source, target)
// Allocate buffer to output buffer
decodedData = make([]byte, 0, encodedBlockLen*int(k))
for i := 0; i < int(k); i++ {
decodedData = append(decodedData, encodedDataBlocks[i]...)
}
return decodedData[:dataLen], nil
}
