285 lines
8.6 KiB
Nim
285 lines
8.6 KiB
Nim
# Nim-LibP2P
|
|
# Copyright (c) 2022 Status Research & Development GmbH
|
|
# Licensed under either of
|
|
# * Apache License, version 2.0, ([LICENSE-APACHE](LICENSE-APACHE))
|
|
# * MIT license ([LICENSE-MIT](LICENSE-MIT))
|
|
# at your option.
|
|
# This file may not be copied, modified, or distributed except according to
|
|
# those terms.
|
|
|
|
## This module implementes CID (Content IDentifier).
|
|
|
|
{.push raises: [Defect].}
|
|
|
|
import tables, hashes
|
|
import multibase, multicodec, multihash, vbuffer, varint
|
|
import stew/[base58, results]
|
|
|
|
export results
|
|
|
|
type
|
|
CidError* {.pure.} = enum
|
|
Error, Incorrect, Unsupported, Overrun
|
|
|
|
CidVersion* = enum
|
|
CIDvIncorrect, CIDv0, CIDv1, CIDvReserved
|
|
|
|
Cid* = object
|
|
cidver*: CidVersion
|
|
mcodec*: MultiCodec
|
|
hpos*: int
|
|
data*: VBuffer
|
|
|
|
const
|
|
ContentIdsList = [
|
|
multiCodec("raw"),
|
|
multiCodec("dag-pb"),
|
|
multiCodec("dag-cbor"),
|
|
multiCodec("dag-json"),
|
|
multiCodec("git-raw"),
|
|
multiCodec("eth-block"),
|
|
multiCodec("eth-block-list"),
|
|
multiCodec("eth-tx-trie"),
|
|
multiCodec("eth-tx"),
|
|
multiCodec("eth-tx-receipt-trie"),
|
|
multiCodec("eth-tx-receipt"),
|
|
multiCodec("eth-state-trie"),
|
|
multiCodec("eth-account-snapshot"),
|
|
multiCodec("eth-storage-trie"),
|
|
multiCodec("bitcoin-block"),
|
|
multiCodec("bitcoin-tx"),
|
|
multiCodec("zcash-block"),
|
|
multiCodec("zcash-tx"),
|
|
multiCodec("stellar-block"),
|
|
multiCodec("stellar-tx"),
|
|
multiCodec("decred-block"),
|
|
multiCodec("decred-tx"),
|
|
multiCodec("dash-block"),
|
|
multiCodec("dash-tx"),
|
|
multiCodec("torrent-info"),
|
|
multiCodec("torrent-file"),
|
|
multiCodec("ed25519-pub")
|
|
]
|
|
|
|
proc initCidCodeTable(): Table[int, MultiCodec] {.compileTime.} =
|
|
for item in ContentIdsList:
|
|
result[int(item)] = item
|
|
|
|
const
|
|
CodeContentIds = initCidCodeTable()
|
|
|
|
template orError*(exp: untyped, err: untyped): untyped =
|
|
(exp.mapErr do (_: auto) -> auto: err)
|
|
|
|
proc decode(data: openArray[byte]): Result[Cid, CidError] =
|
|
if len(data) == 34 and data[0] == 0x12'u8 and data[1] == 0x20'u8:
|
|
ok(Cid(
|
|
cidver: CIDv0,
|
|
mcodec: multiCodec("dag-pb"),
|
|
hpos: 0,
|
|
data: initVBuffer(data)))
|
|
else:
|
|
var version, codec: uint64
|
|
var res, offset: int
|
|
var vb = initVBuffer(data)
|
|
if vb.isEmpty():
|
|
err(CidError.Incorrect)
|
|
else:
|
|
res = vb.readVarint(version)
|
|
if res == -1:
|
|
err(CidError.Incorrect)
|
|
else:
|
|
offset += res
|
|
if version != 1'u64:
|
|
err(CidError.Incorrect)
|
|
else:
|
|
res = vb.readVarint(codec)
|
|
if res == -1:
|
|
err(CidError.Incorrect)
|
|
else:
|
|
offset += res
|
|
var mcodec = CodeContentIds.getOrDefault(cast[int](codec),
|
|
InvalidMultiCodec)
|
|
if mcodec == InvalidMultiCodec:
|
|
err(CidError.Incorrect)
|
|
else:
|
|
if not MultiHash.validate(vb.buffer.toOpenArray(vb.offset,
|
|
vb.buffer.high)):
|
|
err(CidError.Incorrect)
|
|
else:
|
|
vb.finish()
|
|
ok(Cid(
|
|
cidver: CIDv1,
|
|
mcodec: mcodec,
|
|
hpos: offset,
|
|
data: vb))
|
|
|
|
proc decode(data: openArray[char]): Result[Cid, CidError] =
|
|
var buffer: seq[byte]
|
|
var plen = 0
|
|
if len(data) < 2:
|
|
return err(CidError.Incorrect)
|
|
if len(data) == 46:
|
|
if data[0] == 'Q' and data[1] == 'm':
|
|
buffer = newSeq[byte](BTCBase58.decodedLength(len(data)))
|
|
if BTCBase58.decode(data, buffer, plen) != Base58Status.Success:
|
|
return err(CidError.Incorrect)
|
|
buffer.setLen(plen)
|
|
if len(buffer) == 0:
|
|
let length = MultiBase.decodedLength(data[0], len(data))
|
|
if length == -1:
|
|
return err(CidError.Incorrect)
|
|
buffer = newSeq[byte](length)
|
|
if MultiBase.decode(data, buffer, plen) != MultiBaseStatus.Success:
|
|
return err(CidError.Incorrect)
|
|
buffer.setLen(plen)
|
|
if buffer[0] == 0x12'u8:
|
|
return err(CidError.Incorrect)
|
|
decode(buffer)
|
|
|
|
proc validate*(ctype: typedesc[Cid], data: openArray[byte]): bool =
|
|
## Returns ``true`` is data has valid binary CID representation.
|
|
var version, codec: uint64
|
|
var res: VarintResult[void]
|
|
if len(data) < 2:
|
|
return false
|
|
let last = data.high
|
|
if len(data) == 34:
|
|
if data[0] == 0x12'u8 and data[1] == 0x20'u8:
|
|
return true
|
|
var offset = 0
|
|
var length = 0
|
|
res = LP.getUVarint(data.toOpenArray(offset, last), length, version)
|
|
if res.isErr():
|
|
return false
|
|
if version != 1'u64:
|
|
return false
|
|
offset += length
|
|
if offset >= len(data):
|
|
return false
|
|
res = LP.getUVarint(data.toOpenArray(offset, last), length, codec)
|
|
if res.isErr():
|
|
return false
|
|
var mcodec = CodeContentIds.getOrDefault(cast[int](codec), InvalidMultiCodec)
|
|
if mcodec == InvalidMultiCodec:
|
|
return false
|
|
if not MultiHash.validate(data.toOpenArray(offset, last)):
|
|
return false
|
|
result = true
|
|
|
|
proc mhash*(cid: Cid): Result[MultiHash, CidError] =
|
|
## Returns MultiHash part of CID.
|
|
if cid.cidver notin {CIDv0, CIDv1}:
|
|
err(CidError.Incorrect)
|
|
else:
|
|
MultiHash.init(cid.data.buffer.toOpenArray(cid.hpos, cid.data.high)).orError(CidError.Incorrect)
|
|
|
|
proc contentType*(cid: Cid): Result[MultiCodec, CidError] =
|
|
## Returns content type part of CID
|
|
if cid.cidver notin {CIDv0, CIDv1}:
|
|
err(CidError.Incorrect)
|
|
else:
|
|
ok(cid.mcodec)
|
|
|
|
proc version*(cid: Cid): CidVersion =
|
|
## Returns CID version
|
|
result = cid.cidver
|
|
|
|
proc init*[T: char|byte](ctype: typedesc[Cid], data: openArray[T]): Result[Cid, CidError] =
|
|
## Create new content identifier using array of bytes or string ``data``.
|
|
decode(data)
|
|
|
|
proc init*(ctype: typedesc[Cid], version: CidVersion, content: MultiCodec,
|
|
hash: MultiHash): Result[Cid, CidError] =
|
|
## Create new content identifier using content type ``content`` and
|
|
## MultiHash ``hash`` using version ``version``.
|
|
##
|
|
## To create ``CIDv0`` you need to use:
|
|
## Cid.init(CIDv0, multiCodec("dag-pb"), MultiHash.digest("sha2-256", data))
|
|
##
|
|
## All other encodings and hashes are not supported by CIDv0.
|
|
|
|
var res: Cid
|
|
res.cidver = version
|
|
|
|
if version == CIDv0:
|
|
if content != multiCodec("dag-pb"):
|
|
return err(CidError.Unsupported)
|
|
res.data = initVBuffer()
|
|
if hash.mcodec != multiCodec("sha2-256"):
|
|
return err(CidError.Unsupported)
|
|
res.mcodec = content
|
|
res.data.write(hash)
|
|
res.data.finish()
|
|
return ok(res)
|
|
elif version == CIDv1:
|
|
let mcodec = CodeContentIds.getOrDefault(cast[int](content),
|
|
InvalidMultiCodec)
|
|
if mcodec == InvalidMultiCodec:
|
|
return err(CidError.Incorrect)
|
|
res.mcodec = mcodec
|
|
res.data = initVBuffer()
|
|
res.data.writeVarint(cast[uint64](1))
|
|
res.data.write(mcodec)
|
|
res.hpos = len(res.data.buffer)
|
|
res.data.write(hash)
|
|
res.data.finish()
|
|
return ok(res)
|
|
else:
|
|
return err(CidError.Unsupported)
|
|
|
|
proc `==`*(a: Cid, b: Cid): bool =
|
|
## Compares content identifiers ``a`` and ``b``, returns ``true`` if hashes
|
|
## are equal, ``false`` otherwise.
|
|
if a.mcodec == b.mcodec:
|
|
var ah, bh: MultiHash
|
|
if MultiHash.decode(
|
|
a.data.buffer.toOpenArray(a.hpos, a.data.high), ah).isErr:
|
|
return false
|
|
if MultiHash.decode(
|
|
b.data.buffer.toOpenArray(b.hpos, b.data.high), bh).isErr:
|
|
return false
|
|
result = (ah == bh)
|
|
|
|
proc base58*(cid: Cid): string =
|
|
## Get BASE58 encoded string representation of content identifier ``cid``.
|
|
result = BTCBase58.encode(cid.data.buffer)
|
|
|
|
proc hex*(cid: Cid): string =
|
|
## Get hexadecimal string representation of content identifier ``cid``.
|
|
result = $(cid.data)
|
|
|
|
proc repr*(cid: Cid): string =
|
|
## Get string representation of content identifier ``cid``.
|
|
result = $(cid.cidver)
|
|
result.add("/")
|
|
result.add($(cid.mcodec))
|
|
result.add("/")
|
|
result.add($(cid.mhash()))
|
|
|
|
proc write*(vb: var VBuffer, cid: Cid) {.inline.} =
|
|
## Write CID value ``cid`` to buffer ``vb``.
|
|
vb.writeArray(cid.data.buffer)
|
|
|
|
proc encode*(mbtype: typedesc[MultiBase], encoding: string,
|
|
cid: Cid): string {.inline.} =
|
|
## Get MultiBase encoded representation of ``cid`` using encoding
|
|
## ``encoding``.
|
|
result = MultiBase.encode(encoding, cid.data.buffer).tryGet()
|
|
|
|
proc hash*(cid: Cid): Hash {.inline.} =
|
|
hash(cid.data.buffer)
|
|
|
|
proc `$`*(cid: Cid): string =
|
|
## Return official string representation of content identifier ``cid``.
|
|
if cid.cidver == CIDv0:
|
|
BTCBase58.encode(cid.data.buffer)
|
|
elif cid.cidver == CIDv1:
|
|
let res = MultiBase.encode("base58btc", cid.data.buffer)
|
|
if res.isOk():
|
|
res.get()
|
|
else:
|
|
""
|
|
else:
|
|
""
|