restic/internal/pack/pack.go

408 lines
11 KiB
Go
Raw Normal View History

2015-04-26 13:36:49 +00:00
package pack
import (
"bytes"
"context"
2015-04-26 13:36:49 +00:00
"encoding/binary"
"fmt"
"io"
"sync"
2017-07-23 12:21:03 +00:00
"github.com/restic/restic/internal/debug"
"github.com/restic/restic/internal/errors"
2017-07-24 15:42:25 +00:00
"github.com/restic/restic/internal/restic"
2017-07-23 12:21:03 +00:00
"github.com/restic/restic/internal/crypto"
2015-04-26 13:36:49 +00:00
)
// Packer is used to create a new Pack.
type Packer struct {
2016-08-31 20:39:36 +00:00
blobs []restic.Blob
2015-04-26 13:36:49 +00:00
bytes uint
k *crypto.Key
wr io.Writer
2015-04-26 13:36:49 +00:00
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}
2015-04-26 13:36:49 +00:00
}
// 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.
2022-02-13 16:24:09 +00:00
func (p *Packer) Add(t restic.BlobType, id restic.ID, data []byte, uncompressedLength int) (int, error) {
2015-04-26 13:36:49 +00:00
p.m.Lock()
defer p.m.Unlock()
2020-11-05 20:52:34 +00:00
c := restic.Blob{BlobHandle: restic.BlobHandle{Type: t, ID: id}}
2015-04-26 13:36:49 +00:00
n, err := p.wr.Write(data)
c.Length = uint(n)
2015-04-26 13:36:49 +00:00
c.Offset = p.bytes
2022-02-13 16:24:09 +00:00
c.UncompressedLength = uint(uncompressedLength)
2015-04-26 13:36:49 +00:00
p.bytes += uint(n)
p.blobs = append(p.blobs, c)
n += CalculateEntrySize(c)
2015-04-26 13:36:49 +00:00
2016-08-29 20:16:58 +00:00
return n, errors.Wrap(err, "Write")
2015-04-26 13:36:49 +00:00
}
2022-02-13 16:24:09 +00:00
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{}))
2015-04-26 13:36:49 +00:00
// headerEntry describes the format of header entries. It serves only as
// documentation.
2015-04-26 13:36:49 +00:00
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
2015-04-26 13:36:49 +00:00
}
// Finalize writes the header for all added blobs and finalizes the pack.
func (p *Packer) Finalize() error {
2015-04-26 13:36:49 +00:00
p.m.Lock()
defer p.m.Unlock()
header, err := makeHeader(p.blobs)
2015-04-29 22:36:36 +00:00
if err != nil {
return err
2015-04-26 13:36:49 +00:00
}
2022-06-12 12:48:30 +00:00
encryptedHeader := make([]byte, 0, crypto.CiphertextLength(len(header)))
2017-10-29 10:33:57 +00:00
nonce := crypto.NewRandomNonce()
encryptedHeader = append(encryptedHeader, nonce...)
encryptedHeader = p.k.Seal(encryptedHeader, nonce, header, nil)
encryptedHeader = binary.LittleEndian.AppendUint32(encryptedHeader, uint32(len(encryptedHeader)))
if err := verifyHeader(p.k, encryptedHeader, p.blobs); err != nil {
//nolint:revive // ignore linter warnings about error message spelling
return fmt.Errorf("Detected data corruption while writing pack-file header: %w\nCorrupted data is either caused by hardware issues or software bugs. Please open an issue at https://github.com/restic/restic/issues/new/choose for further troubleshooting.", err)
}
2015-04-29 22:41:11 +00:00
2016-01-24 18:30:14 +00:00
// append the header
n, err := p.wr.Write(encryptedHeader)
2015-04-26 13:36:49 +00:00
if err != nil {
return errors.Wrap(err, "Write")
2016-01-24 18:30:14 +00:00
}
if n != len(encryptedHeader) {
return errors.New("wrong number of bytes written")
2015-04-26 13:36:49 +00:00
}
p.bytes += uint(len(encryptedHeader))
return nil
}
2015-04-26 13:36:49 +00:00
func verifyHeader(k *crypto.Key, header []byte, expected []restic.Blob) error {
// do not offer a way to skip the pack header verification, as pack headers are usually small enough
// to not result in a significant performance impact
decoded, hdrSize, err := List(k, bytes.NewReader(header), int64(len(header)))
2015-04-26 13:36:49 +00:00
if err != nil {
return fmt.Errorf("header decoding failed: %w", err)
}
if hdrSize != uint32(len(header)) {
return fmt.Errorf("unexpected header size %v instead of %v", hdrSize, len(header))
}
if len(decoded) != len(expected) {
return fmt.Errorf("pack header size mismatch")
}
for i := 0; i < len(decoded); i++ {
if decoded[i] != expected[i] {
return fmt.Errorf("pack header entry mismatch got %v instead of %v", decoded[i], expected[i])
}
2015-04-26 13:36:49 +00:00
}
return nil
}
// HeaderOverhead returns an estimate of the number of bytes written by a call to Finalize.
func (p *Packer) HeaderOverhead() int {
2022-06-12 12:48:30 +00:00
return crypto.CiphertextLength(0) + binary.Size(uint32(0))
2015-04-26 13:36:49 +00:00
}
// makeHeader constructs the header for p.
func makeHeader(blobs []restic.Blob) ([]byte, error) {
buf := make([]byte, 0, len(blobs)*int(entrySize))
2015-04-29 22:36:36 +00:00
for _, b := range blobs {
2022-02-13 16:24:09 +00:00
switch {
case b.Type == restic.DataBlob && b.UncompressedLength == 0:
buf = append(buf, 0)
2022-02-13 16:24:09 +00:00
case b.Type == restic.TreeBlob && b.UncompressedLength == 0:
buf = append(buf, 1)
2022-02-13 16:24:09 +00:00
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)
2015-04-29 22:36:36 +00:00
}
var lenLE [4]byte
binary.LittleEndian.PutUint32(lenLE[:], uint32(b.Length))
buf = append(buf, lenLE[:]...)
2022-02-13 16:24:09 +00:00
if b.UncompressedLength != 0 {
binary.LittleEndian.PutUint32(lenLE[:], uint32(b.UncompressedLength))
buf = append(buf, lenLE[:]...)
}
buf = append(buf, b.ID[:]...)
2015-04-29 22:36:36 +00:00
}
return buf, nil
2015-04-29 22:36:36 +00:00
}
2015-04-26 13:36:49 +00:00
// 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)
}
2022-04-30 22:05:20 +00:00
// HeaderFull returns true if the pack header is full.
func (p *Packer) HeaderFull() bool {
p.m.Lock()
defer p.m.Unlock()
return headerSize+uint(len(p.blobs)+1)*entrySize > MaxHeaderSize
}
2015-04-26 13:36:49 +00:00
// Blobs returns the slice of blobs that have been written.
2016-08-31 20:39:36 +00:00
func (p *Packer) Blobs() []restic.Blob {
2015-04-26 13:36:49 +00:00
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
2022-02-13 16:24:09 +00:00
minFileSize = plainEntrySize + crypto.Extension + uint(headerLengthSize)
)
const (
2020-05-01 20:56:34 +00:00
// 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
2020-05-01 20:56:34 +00:00
// MaxHeaderSize is the max size of header including header-length field
MaxHeaderSize = 16*1024*1024 + headerLengthSize
2023-12-06 12:11:55 +00:00
// number of header entries to download as part of header-length request
eagerEntries = 15
)
2022-02-13 16:24:09 +00:00
// 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
2018-02-22 17:37:10 +00:00
// the appropriate size.
2022-02-13 16:24:09 +00:00
func readRecords(rd io.ReaderAt, size int64, bufsize int) ([]byte, int, error) {
if bufsize > int(size) {
bufsize = int(size)
}
2016-08-25 19:51:07 +00:00
b := make([]byte, bufsize)
off := size - int64(bufsize)
if _, err := rd.ReadAt(b, off); err != nil {
return nil, 0, err
}
2018-02-22 17:37:10 +00:00
hlen := binary.LittleEndian.Uint32(b[len(b)-headerLengthSize:])
b = b[:len(b)-headerLengthSize]
2018-02-22 17:37:10 +00:00
debug.Log("header length: %v", hlen)
var err error
switch {
2018-02-22 17:37:10 +00:00
case hlen == 0:
err = InvalidFileError{Message: "header length is zero"}
2018-02-22 17:37:10 +00:00
case hlen < crypto.Extension:
err = InvalidFileError{Message: "header length is too small"}
2018-02-22 17:37:10 +00:00
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")
}
2022-02-13 16:24:09 +00:00
total := int(hlen + headerLengthSize)
if total < bufsize {
2018-02-22 17:37:10 +00:00
// truncate to the beginning of the pack header
b = b[len(b)-int(hlen):]
}
2017-06-08 18:40:12 +00:00
2018-02-22 17:37:10 +00:00
return b, total, nil
}
2016-08-25 19:51:07 +00:00
// 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) {
2017-06-08 19:04:07 +00:00
debug.Log("size: %v", size)
2017-06-08 18:40:12 +00:00
if size < int64(minFileSize) {
2017-06-08 19:04:07 +00:00
err := InvalidFileError{Message: "file is too small"}
2017-06-08 18:40:12 +00:00
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
2022-02-13 16:24:09 +00:00
b, c, err := readRecords(rd, size, eagerSize)
if err != nil {
2016-08-25 19:51:07 +00:00
return nil, err
}
2022-02-13 16:24:09 +00:00
if c <= eagerSize {
// eager read sufficed, return what we got
return b, nil
2016-08-23 20:21:29 +00:00
}
b, _, err = readRecords(rd, size, c)
if err != nil {
return nil, err
}
return b, nil
2016-08-25 19:51:07 +00:00
}
2017-06-08 18:40:12 +00:00
// 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
}
2020-11-16 03:03:45 +00:00
// 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) {
2016-08-25 19:51:07 +00:00
buf, err := readHeader(rd, size)
if err != nil {
2020-11-16 03:03:45 +00:00
return nil, 0, err
2016-08-25 19:51:07 +00:00
}
2015-04-26 13:36:49 +00:00
2022-06-12 12:48:30 +00:00
if len(buf) < crypto.CiphertextLength(0) {
2020-11-16 03:03:45 +00:00
return nil, 0, errors.New("invalid header, too small")
2017-10-29 10:33:57 +00:00
}
2020-11-16 03:03:45 +00:00
hdrSize = headerLengthSize + uint32(len(buf))
2017-10-29 10:33:57 +00:00
nonce, buf := buf[:k.NonceSize()], buf[k.NonceSize():]
buf, err = k.Open(buf[:0], nonce, buf, nil)
2015-04-26 13:36:49 +00:00
if err != nil {
2020-11-16 03:03:45 +00:00
return nil, 0, err
2015-04-26 13:36:49 +00:00
}
2022-02-13 16:24:09 +00:00
// 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)
2017-01-15 14:27:58 +00:00
pos := uint(0)
for len(buf) > 0 {
2022-02-13 16:24:09 +00:00
entry, headerSize, err := parseHeaderEntry(buf)
if err != nil {
2020-11-16 03:03:45 +00:00
return nil, 0, err
}
entry.Offset = pos
entries = append(entries, entry)
pos += entry.Length
2022-02-13 16:24:09 +00:00
buf = buf[headerSize:]
2015-04-26 13:36:49 +00:00
}
2020-11-16 03:03:45 +00:00
return entries, hdrSize, nil
2015-04-26 13:36:49 +00:00
}
2020-05-01 20:56:34 +00:00
2022-02-13 16:24:09 +00:00
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))
2022-02-13 16:24:09 +00:00
return b, size, err
}
2022-02-13 16:24:09 +00:00
tpe := p[0]
2022-02-13 16:24:09 +00:00
switch tpe {
case 0, 2:
b.Type = restic.DataBlob
2022-02-13 16:24:09 +00:00
case 1, 3:
b.Type = restic.TreeBlob
default:
2022-02-13 16:24:09 +00:00
return b, size, errors.Errorf("invalid type %d", tpe)
}
b.Length = uint(binary.LittleEndian.Uint32(p[1:5]))
2022-02-13 16:24:09 +00:00
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:]
}
2022-02-13 16:24:09 +00:00
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
2022-02-13 16:24:09 +00:00
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)
mi.Each(ctx, func(blob restic.PackedBlob) {
size, ok := packSize[blob.PackID]
if !ok {
size = headerSize
}
if !onlyHdr {
size += int64(blob.Length)
}
2022-02-13 16:24:09 +00:00
packSize[blob.PackID] = size + int64(CalculateEntrySize(blob.Blob))
})
return packSize
}