restic/internal/pack/pack.go
Michael Eischer 120ccc8754 repository: Rework blob saving to use an async pack uploader
Previously, SaveAndEncrypt would assemble blobs into packs and either
return immediately if the pack is not yet full or upload the pack file
otherwise. The upload will block the current goroutine until it
finishes.

Now, the upload is done using separate goroutines. This requires changes
to the error handling. As uploads are no longer tied to a SaveAndEncrypt
call, failed uploads are signaled using an errgroup.

To count the uploaded amount of data, the pack header overhead is no
longer returned by `packer.Finalize` but rather by
`packer.HeaderOverhead`. This helper method is necessary to continue
returning the pack header overhead directly to the responsible call to
`repository.SaveBlob`. Without the method this would not be possible,
as packs are finalized asynchronously.
2022-07-02 22:42:34 +02:00

378 lines
9.9 KiB
Go

package pack
import (
"context"
"encoding/binary"
"fmt"
"io"
"sync"
"github.com/restic/restic/internal/debug"
"github.com/restic/restic/internal/errors"
"github.com/restic/restic/internal/restic"
"github.com/restic/restic/internal/crypto"
)
// Packer is used to create a new Pack.
type Packer struct {
blobs []restic.Blob
bytes uint
k *crypto.Key
wr io.Writer
m sync.Mutex
}
// NewPacker returns a new Packer that can be used to pack blobs together.
func NewPacker(k *crypto.Key, wr io.Writer) *Packer {
return &Packer{k: k, wr: wr}
}
// Add saves the data read from rd as a new blob to the packer. Returned is the
// number of bytes written to the pack plus the pack header entry size.
func (p *Packer) Add(t restic.BlobType, id restic.ID, data []byte, uncompressedLength int) (int, error) {
p.m.Lock()
defer p.m.Unlock()
c := restic.Blob{BlobHandle: restic.BlobHandle{Type: t, ID: id}}
n, err := p.wr.Write(data)
c.Length = uint(n)
c.Offset = p.bytes
c.UncompressedLength = uint(uncompressedLength)
p.bytes += uint(n)
p.blobs = append(p.blobs, c)
n += CalculateEntrySize(c)
return n, errors.Wrap(err, "Write")
}
var entrySize = uint(binary.Size(restic.BlobType(0)) + 2*headerLengthSize + len(restic.ID{}))
var plainEntrySize = uint(binary.Size(restic.BlobType(0)) + headerLengthSize + len(restic.ID{}))
// headerEntry describes the format of header entries. It serves only as
// documentation.
type headerEntry struct {
Type uint8
Length uint32
ID restic.ID
}
// compressedHeaderEntry describes the format of header entries for compressed blobs.
// It serves only as documentation.
type compressedHeaderEntry struct {
Type uint8
Length uint32
UncompressedLength uint32
ID restic.ID
}
// Finalize writes the header for all added blobs and finalizes the pack.
func (p *Packer) Finalize() error {
p.m.Lock()
defer p.m.Unlock()
header, err := p.makeHeader()
if err != nil {
return err
}
encryptedHeader := make([]byte, 0, restic.CiphertextLength(len(header)))
nonce := crypto.NewRandomNonce()
encryptedHeader = append(encryptedHeader, nonce...)
encryptedHeader = p.k.Seal(encryptedHeader, nonce, header, nil)
// append the header
n, err := p.wr.Write(encryptedHeader)
if err != nil {
return errors.Wrap(err, "Write")
}
hdrBytes := len(encryptedHeader)
if n != hdrBytes {
return errors.New("wrong number of bytes written")
}
// write length
err = binary.Write(p.wr, binary.LittleEndian, uint32(hdrBytes))
if err != nil {
return errors.Wrap(err, "binary.Write")
}
p.bytes += uint(hdrBytes + binary.Size(uint32(0)))
return nil
}
// HeaderOverhead returns an estimate of the number of bytes written by a call to Finalize.
func (p *Packer) HeaderOverhead() int {
return restic.CiphertextLength(0) + binary.Size(uint32(0))
}
// makeHeader constructs the header for p.
func (p *Packer) makeHeader() ([]byte, error) {
buf := make([]byte, 0, len(p.blobs)*int(entrySize))
for _, b := range p.blobs {
switch {
case b.Type == restic.DataBlob && b.UncompressedLength == 0:
buf = append(buf, 0)
case b.Type == restic.TreeBlob && b.UncompressedLength == 0:
buf = append(buf, 1)
case b.Type == restic.DataBlob && b.UncompressedLength != 0:
buf = append(buf, 2)
case b.Type == restic.TreeBlob && b.UncompressedLength != 0:
buf = append(buf, 3)
default:
return nil, errors.Errorf("invalid blob type %v", b.Type)
}
var lenLE [4]byte
binary.LittleEndian.PutUint32(lenLE[:], uint32(b.Length))
buf = append(buf, lenLE[:]...)
if b.UncompressedLength != 0 {
binary.LittleEndian.PutUint32(lenLE[:], uint32(b.UncompressedLength))
buf = append(buf, lenLE[:]...)
}
buf = append(buf, b.ID[:]...)
}
return buf, nil
}
// Size returns the number of bytes written so far.
func (p *Packer) Size() uint {
p.m.Lock()
defer p.m.Unlock()
return p.bytes
}
// Count returns the number of blobs in this packer.
func (p *Packer) Count() int {
p.m.Lock()
defer p.m.Unlock()
return len(p.blobs)
}
// Blobs returns the slice of blobs that have been written.
func (p *Packer) Blobs() []restic.Blob {
p.m.Lock()
defer p.m.Unlock()
return p.blobs
}
func (p *Packer) String() string {
return fmt.Sprintf("<Packer %d blobs, %d bytes>", len(p.blobs), p.bytes)
}
var (
// we require at least one entry in the header, and one blob for a pack file
minFileSize = plainEntrySize + crypto.Extension + uint(headerLengthSize)
)
const (
// size of the header-length field at the end of the file; it is a uint32
headerLengthSize = 4
// headerSize is the header's constant overhead (independent of #entries)
headerSize = headerLengthSize + crypto.Extension
// MaxHeaderSize is the max size of header including header-length field
MaxHeaderSize = 16*1024*1024 + headerLengthSize
// number of header enries to download as part of header-length request
eagerEntries = 15
)
// readRecords reads up to bufsize bytes from the underlying ReaderAt, returning
// the raw header, the total number of bytes in the header, and any error.
// If the header contains fewer than bufsize bytes, the header is truncated to
// the appropriate size.
func readRecords(rd io.ReaderAt, size int64, bufsize int) ([]byte, int, error) {
if bufsize > int(size) {
bufsize = int(size)
}
b := make([]byte, bufsize)
off := size - int64(bufsize)
if _, err := rd.ReadAt(b, off); err != nil {
return nil, 0, err
}
hlen := binary.LittleEndian.Uint32(b[len(b)-headerLengthSize:])
b = b[:len(b)-headerLengthSize]
debug.Log("header length: %v", hlen)
var err error
switch {
case hlen == 0:
err = InvalidFileError{Message: "header length is zero"}
case hlen < crypto.Extension:
err = InvalidFileError{Message: "header length is too small"}
case int64(hlen) > size-int64(headerLengthSize):
err = InvalidFileError{Message: "header is larger than file"}
case int64(hlen) > MaxHeaderSize-int64(headerLengthSize):
err = InvalidFileError{Message: "header is larger than maxHeaderSize"}
}
if err != nil {
return nil, 0, errors.Wrap(err, "readHeader")
}
total := int(hlen + headerLengthSize)
if total < bufsize {
// truncate to the beginning of the pack header
b = b[len(b)-int(hlen):]
}
return b, total, nil
}
// readHeader reads the header at the end of rd. size is the length of the
// whole data accessible in rd.
func readHeader(rd io.ReaderAt, size int64) ([]byte, error) {
debug.Log("size: %v", size)
if size < int64(minFileSize) {
err := InvalidFileError{Message: "file is too small"}
return nil, errors.Wrap(err, "readHeader")
}
// assuming extra request is significantly slower than extra bytes download,
// eagerly download eagerEntries header entries as part of header-length request.
// only make second request if actual number of entries is greater than eagerEntries
eagerSize := eagerEntries*int(entrySize) + headerSize
b, c, err := readRecords(rd, size, eagerSize)
if err != nil {
return nil, err
}
if c <= eagerSize {
// eager read sufficed, return what we got
return b, nil
}
b, _, err = readRecords(rd, size, c)
if err != nil {
return nil, err
}
return b, nil
}
// InvalidFileError is return when a file is found that is not a pack file.
type InvalidFileError struct {
Message string
}
func (e InvalidFileError) Error() string {
return e.Message
}
// List returns the list of entries found in a pack file and the length of the
// header (including header size and crypto overhead)
func List(k *crypto.Key, rd io.ReaderAt, size int64) (entries []restic.Blob, hdrSize uint32, err error) {
buf, err := readHeader(rd, size)
if err != nil {
return nil, 0, err
}
if len(buf) < restic.CiphertextLength(0) {
return nil, 0, errors.New("invalid header, too small")
}
hdrSize = headerLengthSize + uint32(len(buf))
nonce, buf := buf[:k.NonceSize()], buf[k.NonceSize():]
buf, err = k.Open(buf[:0], nonce, buf, nil)
if err != nil {
return nil, 0, err
}
// might over allocate a bit if all blobs have EntrySize but only by a few percent
entries = make([]restic.Blob, 0, uint(len(buf))/plainEntrySize)
pos := uint(0)
for len(buf) > 0 {
entry, headerSize, err := parseHeaderEntry(buf)
if err != nil {
return nil, 0, err
}
entry.Offset = pos
entries = append(entries, entry)
pos += entry.Length
buf = buf[headerSize:]
}
return entries, hdrSize, nil
}
func parseHeaderEntry(p []byte) (b restic.Blob, size uint, err error) {
l := uint(len(p))
size = plainEntrySize
if l < plainEntrySize {
err = errors.Errorf("parseHeaderEntry: buffer of size %d too short", len(p))
return b, size, err
}
tpe := p[0]
switch tpe {
case 0, 2:
b.Type = restic.DataBlob
case 1, 3:
b.Type = restic.TreeBlob
default:
return b, size, errors.Errorf("invalid type %d", tpe)
}
b.Length = uint(binary.LittleEndian.Uint32(p[1:5]))
p = p[5:]
if tpe == 2 || tpe == 3 {
size = entrySize
if l < entrySize {
err = errors.Errorf("parseHeaderEntry: buffer of size %d too short", len(p))
return b, size, err
}
b.UncompressedLength = uint(binary.LittleEndian.Uint32(p[0:4]))
p = p[4:]
}
copy(b.ID[:], p[:])
return b, size, nil
}
func CalculateEntrySize(blob restic.Blob) int {
if blob.UncompressedLength != 0 {
return int(entrySize)
}
return int(plainEntrySize)
}
func CalculateHeaderSize(blobs []restic.Blob) int {
size := headerSize
for _, blob := range blobs {
size += CalculateEntrySize(blob)
}
return size
}
// Size returns the size of all packs computed by index information.
// If onlyHdr is set to true, only the size of the header is returned
// Note that this function only gives correct sizes, if there are no
// duplicates in the index.
func Size(ctx context.Context, mi restic.MasterIndex, onlyHdr bool) map[restic.ID]int64 {
packSize := make(map[restic.ID]int64)
for blob := range mi.Each(ctx) {
size, ok := packSize[blob.PackID]
if !ok {
size = headerSize
}
if !onlyHdr {
size += int64(blob.Length)
}
packSize[blob.PackID] = size + int64(CalculateEntrySize(blob.Blob))
}
return packSize
}