octoleo/restic
2
2
Fork 0
mirror of https://github.com/octoleo/restic.git synced 2024-11-22 12:55:18 +00:00
Code Issues Releases Wiki Activity

Add packer, update Design.md

Browse Source
This commit is contained in:
Alexander Neumann 2015-04-26 15:36:49 +02:00
parent 66b81a2a8b
commit 8f3b91ba32
3 changed files with 450 additions and 51 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

120
doc/Design.md
View File

@ -64,6 +64,10 @@ The basic layout of a sample restic repository is shown below:
/tmp/restic-repo
├── data
│ ├── 21
│ │ └── 2159dd48f8a24f33c307b750592773f8b71ff8d11452132a7b2e2a6a01611be1
│ ├── 32
│ │ └── 32ea976bc30771cebad8285cd99120ac8786f9ffd42141d452458089985043a5
│ ├── 59
│ │ └── 59fe4bcde59bd6222eba87795e35a90d82cd2f138a27b6835032b7b58173a426
│ ├── 73
@ -71,25 +75,14 @@ The basic layout of a sample restic repository is shown below:
│ [...]
├── id
├── index
│ └── c38f5fb68307c6a3e3aa945d556e325dc38f5fb68307c6a3e3aa945d556e325d
│ ├── c38f5fb68307c6a3e3aa945d556e325dc38f5fb68307c6a3e3aa945d556e325d
│ └── ca171b1b7394d90d330b265d90f506f9984043b342525f019788f97e745c71fd
├── keys
│ └── b02de829beeb3c01a63e6b25cbd421a98fef144f03b9a02e46eff9e2ca3f0bd7
├── locks
├── snapshots
│ └── 22a5af1bdc6e616f8a29579458c49627e01b32210d09adb288d1ecda7c5711ec
├── tmp
├── trees
│ ├── 21
│ │ └── 2159dd48f8a24f33c307b750592773f8b71ff8d11452132a7b2e2a6a01611be1
│ ├── 32
│ │ └── 32ea976bc30771cebad8285cd99120ac8786f9ffd42141d452458089985043a5
│ ├── 95
│ │ └── 95f75feb05a7cc73e328b2efa668b1ea68f65fece55a93bc65aff6cd0bcfeefc
│ ├── b8
│ │ └── b8138ab08a4722596ac89c917827358da4672eac68e3c03a8115b88dbf4bfb59
│ ├── e0
│ │ └── e01150928f7ad24befd6ec15b087de1b9e0f92edabd8e5cabb3317f8b20ad044
│ [...]
└── version
A repository can be initialized with the `restic init` command, e.g.:
@ -99,39 +92,47 @@ A repository can be initialized with the `restic init` command, e.g.:
Pack Format
-----------
All files in the repository except Key, Tree and Data files just contain raw
data, stored as `IV || Ciphertext || MAC`. Tree and Data files may contain
several Blobs of data. The format is described in the following.
All files in the repository except Key and Data files just contain raw data,
stored as `IV || Ciphertext || MAC`. Data files may contain one or more Blobs
of data. The format is described in the following.
A Pack starts with a nonce and a header, the header describes the content and
is encrypted and signed. The Pack's structure is as follows:
The Pack's structure is as follows:
NONCE || Header_Length ||
IV_Header || Ciphertext_Header || MAC_Header ||
IV_Blob_1 || Ciphertext_Blob_1 || MAC_Blob_1 ||
[...]
IV_Blob_n || Ciphertext_Blob_n || MAC_Blob_n ||
MAC
EncryptedBlob1 || ... || EncryptedBlobN || EncryptedHeader || Header_Length
`NONCE` consists of 16 bytes and `Header_Length` is a four byte integer in
little-endian encoding.
At the end of the Pack is a header, which describes the content and is
encrypted and signed. `Header_Length` is the length of the encrypted header
encoded as a is a four byte integer in little-endian encoding.
All the parts (`Ciphertext_Header`, `Ciphertext_Blob1` etc.) are signed and
encrypted independently. In addition, the complete pack is signed using
`NONCE`. This enables repository reorganisation without having to touch the
encrypted Blobs. In addition it also allows efficient indexing, for only the
header needs to be read in order to find out which Blobs are contained in the
Pack. Since the header is signed, authenticity of the header can be checked
without having to read the complete Pack.
All the blobs (`EncryptedBlob1`, `EncryptedBlobN` etc.) are signed and
encrypted independently. This enables repository reorganisation without having
to touch the encrypted Blobs. In addition it also allows efficient indexing,
for only the header needs to be read in order to find out which Blobs are
contained in the Pack. Since the header is signed, authenticity of the header
can be checked without having to read the complete Pack.
After decryption, a Pack's header consists of the following elements:
Length(IV_Blob_1+Ciphertext_Blob1+MAC_Blob_1) || Hash(Plaintext_Blob_1) ||
Type_Blob1 || Length(EncryptedBlob1) || Hash(Plaintext_Blob1) ||
[...]
Length(IV_Blob_n+Ciphertext_Blob_n+MAC_Blob_n) || Hash(Plaintext_Blob_n) ||
Type_BlobN || Length(EncryptedBlobN) || Hash(Plaintext_Blobn) ||
This is enough to calculate the offsets for all the Blobs in the Pack. Length
is the length of a Blob as a four byte integer in little-endian format.
is the length of a Blob as a four byte integer in little-endian format. The
type field is a one byte field and labels the content of a blob according to
the following table:
Type | Meaning
-----|---------
0 | data
1 | tree
All other types are invalid, more types may be added in the future.
For reconstructing the index or parsing a pack without an index, first the last
four bytes must be read in order to find the length of the header. Afterwards,
the header can be read and parsed, which yields all plaintext hashes, types,
offsets and lengths of all included blobs.
Indexing
--------
@ -139,23 +140,40 @@ Indexing
Index files contain information about Data and Tree Blobs and the Packs they
are contained in and store this information in the repository. When the local
cached index is not accessible any more, the index files can be downloaded and
used to reconstruct the index. The index Blobs are encrypted and signed like
Data and Tree Blobs, so the outer structure is `IV || Ciphertext || MAC` again.
The plaintext consists of a JSON document like the following:
used to reconstruct the index. The files are encrypted and signed like Data and
Tree Blobs, so the outer structure is `IV || Ciphertext || MAC` again. The
plaintext consists of a JSON document like the following:
[
{
"id": "73d04e6125cf3c28a299cc2f3cca3b78ceac396e4fcf9575e34536b26782413c",
"blobs": [
"3ec79977ef0cf5de7b08cd12b874cd0f62bbaf7f07f3497a5b1bbcc8cb39b1ce",
"9ccb846e60d90d4eb915848add7aa7ea1e4bbabfc60e573db9f7bfb2789afbae",
"d3dc577b4ffd38cc4b32122cabf8655a0223ed22edfd93b353dc0c3f2b0fdf66"
]
}
]
[ {
"id": "73d04e6125cf3c28a299cc2f3cca3b78ceac396e4fcf9575e34536b26782413c",
"blobs": [
{
"id": "3ec79977ef0cf5de7b08cd12b874cd0f62bbaf7f07f3497a5b1bbcc8cb39b1ce",
"type": "data",
"offset": 0,
"length": 25
},{
"id": "9ccb846e60d90d4eb915848add7aa7ea1e4bbabfc60e573db9f7bfb2789afbae",
"type": "tree",
"offset": 38,
"length": 100
},
{
"id": "d3dc577b4ffd38cc4b32122cabf8655a0223ed22edfd93b353dc0c3f2b0fdf66",
"type": "data",
"offset": 150,
"length": 123
}
]
} ]
This JSON document lists all the Blobs with contents. In this example, the Pack
`73d04e61` contains three Blobs, the plaintext hashes are listed afterwards.
This JSON document lists Blobs with contents. In this example, the Pack
`73d04e61` contains two data Blobs and one Tree blob, the plaintext hashes are
listed afterwards.
There may be an arbitrary number of index files, containing information on
non-disjoint sets of Packs. The number of packs described in a single file is
chosen so that the file size is kep below 8 MiB.
Keys, Encryption and MAC
------------------------

270
pack/pack.go Normal file
View File

@ -0,0 +1,270 @@
package pack
import (
"crypto/sha256"
"encoding/binary"
"errors"
"fmt"
"io"
"sync"
"github.com/restic/restic/backend"
"github.com/restic/restic/crypto"
)
type BlobType uint8
const (
Data BlobType = 0
Tree = 1
)
func (t BlobType) MarshalJSON() ([]byte, error) {
switch t {
case Data:
return []byte(`"data"`), nil
case Tree:
return []byte(`"tree"`), nil
}
return nil, errors.New("unknown blob type")
}
func (t *BlobType) UnmarshalJSON(buf []byte) error {
switch string(buf) {
case `"data"`:
*t = Data
case `"tree"`:
*t = Tree
default:
return errors.New("unknown blob type")
}
return nil
}
// Blob is a blob within a pack.
type Blob struct {
Type BlobType
Length uint32
ID backend.ID
Offset uint
}
// GetReader returns an io.Reader for the blob entry e.
func (e Blob) GetReader(rd io.ReadSeeker) (io.Reader, error) {
// seek to the correct location
_, err := rd.Seek(int64(e.Offset), 0)
if err != nil {
return nil, err
}
return io.LimitReader(rd, int64(e.Length)), nil
}
// Packer is used to create a new Pack.
type Packer struct {
blobs []Blob
bytes uint
k *crypto.Key
wr io.Writer
hw *backend.HashingWriter
m sync.Mutex
}
// NewPacker returns a new Packer that can be used to pack blobs
// together.
func NewPacker(k *crypto.Key, w io.Writer) *Packer {
return &Packer{k: k, wr: w, hw: backend.NewHashingWriter(w, sha256.New())}
}
// Add saves the data read from rd as a new blob to the packer. Returned is the
// number of bytes written to the pack.
func (p *Packer) Add(t BlobType, id backend.ID, rd io.Reader) (int64, error) {
p.m.Lock()
defer p.m.Unlock()
c := Blob{Type: t, ID: id}
n, err := io.Copy(p.hw, rd)
c.Length = uint32(n)
c.Offset = p.bytes
p.bytes += uint(n)
p.blobs = append(p.blobs, c)
return n, err
}
var entrySize = binary.Size(BlobType(0)) + binary.Size(uint32(0)) + backend.IDSize
// headerEntry is used with encoding/binary to read and write header entries
type headerEntry struct {
Type BlobType
Length uint32
ID [backend.IDSize]byte
}
// Finalize writes the header for all added blobs and finalizes the pack.
// Returned are the complete number of bytes written, including the header.
// After Finalize() has finished, the ID of this pack can be obtained by
// calling ID().
func (p *Packer) Finalize() (int64, error) {
p.m.Lock()
defer p.m.Unlock()
// n tracks the number of bytes written for the header
var n int64 = int64(p.bytes)
// create writer to encrypt header
wr := crypto.EncryptTo(p.k, p.hw)
// write header
for _, b := range p.blobs {
entry := headerEntry{
Type: b.Type,
Length: b.Length,
}
copy(entry.ID[:], b.ID)
err := binary.Write(wr, binary.LittleEndian, entry)
if err != nil {
return int64(n), err
}
n += int64(entrySize)
}
// finalize encrypted header
err := wr.Close()
if err != nil {
return int64(n), err
}
// account for crypto overhead
n += crypto.Extension
// write length
err = binary.Write(p.hw, binary.LittleEndian, uint32(len(p.blobs)*entrySize+crypto.Extension))
if err != nil {
return int64(n), err
}
n += int64(binary.Size(uint32(0)))
p.bytes = uint(n)
return n, nil
}
// ID returns the ID of all data written so far.
func (p *Packer) ID() backend.ID {
p.m.Lock()
defer p.m.Unlock()
return p.hw.Sum(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() []Blob {
p.m.Lock()
defer p.m.Unlock()
return p.blobs
}
// Writer returns the underlying writer.
func (p *Packer) Writer() io.Writer {
return p.wr
}
func (p *Packer) String() string {
return fmt.Sprintf("<Packer %d blobs, %d bytes>", len(p.blobs), p.bytes)
}
// Unpacker is used to read individual blobs from a pack.
type Unpacker struct {
rd io.ReadSeeker
Entries []Blob
k *crypto.Key
}
// NewUnpacker returns a pointer to Unpacker which can be used to read
// individual Blobs from a pack.
func NewUnpacker(k *crypto.Key, entries []Blob, rd io.ReadSeeker) (*Unpacker, error) {
var err error
ls := binary.Size(uint32(0))
// reset to the end to read header length
_, err = rd.Seek(-int64(ls), 2)
if err != nil {
return nil, fmt.Errorf("seeking to read header length failed: %v", err)
}
// read length
var l uint32
err = binary.Read(rd, binary.LittleEndian, &l)
if err != nil {
return nil, fmt.Errorf("reading header length failed: %v", err)
}
// reset to the beginning of the header
_, err = rd.Seek(-int64(ls)-int64(l), 2)
if err != nil {
return nil, fmt.Errorf("seeking to read header length failed: %v", err)
}
// read header
hrd, err := crypto.DecryptFrom(k, io.LimitReader(rd, int64(l)))
if err != nil {
return nil, err
}
if entries == nil {
pos := uint(0)
for {
e := headerEntry{}
err = binary.Read(hrd, binary.LittleEndian, &e)
if err == io.EOF {
break
}
if err != nil {
return nil, err
}
entries = append(entries, Blob{
Type: e.Type,
Length: e.Length,
ID: e.ID[:],
Offset: pos,
})
pos += uint(e.Length)
}
}
p := &Unpacker{
rd: rd,
k: k,
Entries: entries,
}
return p, nil
}

111
pack/pack_test.go Normal file
View File

@ -0,0 +1,111 @@
package pack_test
import (
"bytes"
"crypto/rand"
"crypto/sha256"
"encoding/binary"
"encoding/json"
"io"
"io/ioutil"
"testing"
"github.com/restic/restic/backend"
"github.com/restic/restic/crypto"
"github.com/restic/restic/pack"
. "github.com/restic/restic/test"
)
var lengths = []int{23, 31650, 25860, 10928, 13769, 19862, 5211, 127, 13690, 30231}
// var lengths = []int{200}
func TestCreatePack(t *testing.T) {
type Buf struct {
data []byte
id backend.ID
}
bufs := []Buf{}
for _, l := range lengths {
b := make([]byte, l)
_, err := io.ReadFull(rand.Reader, b)
OK(t, err)
h := sha256.Sum256(b)
bufs = append(bufs, Buf{data: b, id: h[:]})
}
file := bytes.NewBuffer(nil)
// create random keys
k := crypto.NewKey()
// pack blobs
p := pack.NewPacker(k, file)
for _, b := range bufs {
p.Add(pack.Tree, b.id, bytes.NewReader(b.data))
}
// write file
n, err := p.Finalize()
OK(t, err)
written := 0
// data
for _, l := range lengths {
written += l
}
// header length
written += binary.Size(uint32(0))
// header
written += len(lengths) * (binary.Size(pack.BlobType(0)) + binary.Size(uint32(0)) + backend.IDSize)
// header crypto
written += crypto.Extension
// check length
Equals(t, int64(written), n)
Equals(t, uint(written), p.Size())
// read and parse it again
rd := bytes.NewReader(file.Bytes())
np, err := pack.NewUnpacker(k, nil, rd)
OK(t, err)
Equals(t, len(np.Entries), len(bufs))
for i, b := range bufs {
e := np.Entries[i]
Equals(t, b.id, e.ID)
brd, err := e.GetReader(rd)
OK(t, err)
data, err := ioutil.ReadAll(brd)
OK(t, err)
Assert(t, bytes.Equal(b.data, data),
"data for blob %v doesn't match", i)
}
}
var blobTypeJson = []struct {
t pack.BlobType
res string
}{
{pack.Data, `"data"`},
{pack.Tree, `"tree"`},
}
func TestBlobTypeJSON(t *testing.T) {
for _, test := range blobTypeJson {
// test serialize
buf, err := json.Marshal(test.t)
OK(t, err)
Equals(t, test.res, string(buf))
// test unserialize
var v pack.BlobType
err = json.Unmarshal([]byte(test.res), &v)
OK(t, err)
Equals(t, test.t, v)
}
}