mirror of https://github.com/vacp2p/nim-libp2p.git
269 lines
7.8 KiB
Nim
269 lines
7.8 KiB
Nim
# Nim-LibP2P
|
|
# Copyright (c) 2023 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.
|
|
|
|
## To enable dump of all incoming and outgoing unencrypted messages you need
|
|
## to compile project with ``-d:libp2p_dump`` compile-time option. When this
|
|
## option enabled ``nim-libp2p`` will create dumps of unencrypted messages for
|
|
## every peer libp2p communicates.
|
|
##
|
|
## Every file is created with name "<PeerId>.pbcap". One file represents
|
|
## all the communication with peer which identified by ``PeerId``.
|
|
##
|
|
## File can have multiple protobuf encoded messages of this format:
|
|
##
|
|
## Protobuf's message fields:
|
|
## 1. SeqID: optional uint64
|
|
## 2. Timestamp: required uint64
|
|
## 3. Type: optional uint64
|
|
## 4. Direction: required uint64
|
|
## 5. LocalAddress: optional bytes
|
|
## 6. RemoteAddress: optional bytes
|
|
## 7. Message: required bytes
|
|
import os
|
|
import nimcrypto/utils, stew/endians2
|
|
import protobuf/minprotobuf, stream/connection, protocols/secure/secure,
|
|
multiaddress, peerid, varint, muxers/mplex/coder
|
|
|
|
from times import getTime, toUnix, fromUnix, nanosecond, format, Time,
|
|
NanosecondRange, initTime
|
|
from strutils import toHex, repeat
|
|
export peerid, multiaddress
|
|
|
|
type
|
|
FlowDirection* = enum
|
|
Outgoing, Incoming
|
|
|
|
ProtoMessage* = object
|
|
timestamp*: uint64
|
|
direction*: FlowDirection
|
|
message*: seq[byte]
|
|
seqID*: Opt[uint64]
|
|
mtype*: Opt[uint64]
|
|
local*: Opt[MultiAddress]
|
|
remote*: Opt[MultiAddress]
|
|
|
|
const
|
|
libp2p_dump_dir* {.strdefine.} = "nim-libp2p"
|
|
## default directory where all the dumps will be stored, if the path
|
|
## relative it will be created in home directory. You can overload this path
|
|
## using ``-d:libp2p_dump_dir=<otherpath>``.
|
|
|
|
proc getTimestamp(): uint64 =
|
|
## This procedure is present because `stdlib.times` missing it.
|
|
let time = getTime()
|
|
uint64(time.toUnix() * 1_000_000_000 + time.nanosecond())
|
|
|
|
proc getTimedate(value: uint64): string =
|
|
## This procedure is workaround for `stdlib.times` to just convert
|
|
## nanoseconds' timestamp to DateTime object.
|
|
let time = initTime(int64(value div 1_000_000_000), value mod 1_000_000_000)
|
|
time.format("yyyy-MM-dd HH:mm:ss'.'fffzzz")
|
|
|
|
proc dumpMessage*(conn: SecureConn, direction: FlowDirection,
|
|
data: openArray[byte]) =
|
|
## Store unencrypted message ``data`` to dump file, all the metadata will be
|
|
## extracted from ``conn`` instance.
|
|
var pb = initProtoBuffer(options = {WithVarintLength})
|
|
pb.write(2, getTimestamp())
|
|
pb.write(4, uint64(direction))
|
|
conn.observedAddr.withValue(oaddr):
|
|
pb.write(6, oaddr)
|
|
pb.write(7, data)
|
|
pb.finish()
|
|
|
|
let dirName =
|
|
if isAbsolute(libp2p_dump_dir):
|
|
libp2p_dump_dir
|
|
else:
|
|
getHomeDir() / libp2p_dump_dir
|
|
|
|
let fileName = $(conn.peerId) & ".pbcap"
|
|
|
|
# This is debugging procedure so it should not generate any exceptions,
|
|
# and we going to return at every possible OS error.
|
|
if not(dirExists(dirName)):
|
|
try:
|
|
createDir(dirName)
|
|
except CatchableError:
|
|
return
|
|
|
|
let pathName = dirName / fileName
|
|
var handle: File
|
|
try:
|
|
if open(handle, pathName, fmAppend):
|
|
discard writeBuffer(handle, addr pb.buffer[pb.offset], pb.getLen())
|
|
finally:
|
|
close(handle)
|
|
|
|
proc decodeDumpMessage*(data: openArray[byte]): Opt[ProtoMessage] =
|
|
## Decode protobuf's message ProtoMessage from array of bytes ``data``.
|
|
var
|
|
pb = initProtoBuffer(data)
|
|
value: uint64
|
|
ma1, ma2: MultiAddress
|
|
pmsg: ProtoMessage
|
|
|
|
let
|
|
r2 = pb.getField(2, pmsg.timestamp)
|
|
r4 = pb.getField(4, value)
|
|
r7 = pb.getField(7, pmsg.message)
|
|
if not r2.get(false) or not r4.get(false) or not r7.get(false):
|
|
return Opt.none(ProtoMessage)
|
|
|
|
# `case` statement could not work here with an error "selector must be of an
|
|
# ordinal type, float or string"
|
|
pmsg.direction =
|
|
if value == uint64(Outgoing):
|
|
Outgoing
|
|
elif value == uint64(Incoming):
|
|
Incoming
|
|
else:
|
|
return Opt.none(ProtoMessage)
|
|
|
|
let r1 = pb.getField(1, value)
|
|
if r1.get(false):
|
|
pmsg.seqID = Opt.some(value)
|
|
|
|
let r3 = pb.getField(3, value)
|
|
if r3.get(false):
|
|
pmsg.mtype = Opt.some(value)
|
|
|
|
let
|
|
r5 = pb.getField(5, ma1)
|
|
r6 = pb.getField(6, ma2)
|
|
if r5.get(false):
|
|
pmsg.local = Opt.some(ma1)
|
|
if r6.get(false):
|
|
pmsg.remote = Opt.some(ma2)
|
|
|
|
Opt.some(pmsg)
|
|
|
|
iterator messages*(data: seq[byte]): Opt[ProtoMessage] =
|
|
## Iterate over sequence of bytes and decode all the ``ProtoMessage``
|
|
## messages we found.
|
|
var value: uint64
|
|
var size: int
|
|
var offset = 0
|
|
while offset < len(data):
|
|
value = 0
|
|
size = 0
|
|
let res = PB.getUVarint(data.toOpenArray(offset, len(data) - 1),
|
|
size, value)
|
|
if res.isOk():
|
|
if (value > 0'u64) and (value < uint64(len(data) - offset)):
|
|
offset += size
|
|
yield decodeDumpMessage(data.toOpenArray(offset,
|
|
offset + int(value) - 1))
|
|
# value is previously checked to be less then len(data) which is `int`.
|
|
offset += int(value)
|
|
else:
|
|
break
|
|
else:
|
|
break
|
|
|
|
proc dumpHex*(pbytes: openArray[byte], groupBy = 1, ascii = true): string =
|
|
## Get hexadecimal dump of memory for array ``pbytes``.
|
|
var res = ""
|
|
var offset = 0
|
|
var ascii = ""
|
|
|
|
while offset < len(pbytes):
|
|
if (offset mod 16) == 0:
|
|
res = res & toHex(uint64(offset)) & ": "
|
|
|
|
for k in 0 ..< groupBy:
|
|
let ch = pbytes[offset + k]
|
|
ascii.add(if ord(ch) > 31 and ord(ch) < 127: char(ch) else: '.')
|
|
|
|
let item =
|
|
case groupBy:
|
|
of 1:
|
|
toHex(pbytes[offset])
|
|
of 2:
|
|
toHex(uint16.fromBytes(pbytes.toOpenArray(offset, len(pbytes) - 1)))
|
|
of 4:
|
|
toHex(uint32.fromBytes(pbytes.toOpenArray(offset, len(pbytes) - 1)))
|
|
of 8:
|
|
toHex(uint64.fromBytes(pbytes.toOpenArray(offset, len(pbytes) - 1)))
|
|
else:
|
|
""
|
|
res.add(item)
|
|
res.add(" ")
|
|
offset = offset + groupBy
|
|
|
|
if (offset mod 16) == 0:
|
|
res.add(" ")
|
|
res.add(ascii)
|
|
ascii.setLen(0)
|
|
res.add("\p")
|
|
|
|
if (offset mod 16) != 0:
|
|
let spacesCount = ((16 - (offset mod 16)) div groupBy) *
|
|
(groupBy * 2 + 1) + 1
|
|
res = res & repeat(' ', spacesCount)
|
|
res = res & ascii
|
|
|
|
res.add("\p")
|
|
res
|
|
|
|
proc mplexMessage*(data: seq[byte]): string =
|
|
var value = 0'u64
|
|
var size = 0
|
|
let res = PB.getUVarint(data, size, value)
|
|
if res.isOk():
|
|
if size < len(data) and data[size] == 0x00'u8:
|
|
let header = cast[MessageType](value and 0x07'u64)
|
|
let ident = (value shr 3)
|
|
"mplex: [" & $header & ", ident = " & $ident & "] "
|
|
else:
|
|
""
|
|
else:
|
|
""
|
|
|
|
proc toString*(msg: ProtoMessage, dump = true): string =
|
|
## Convert message ``msg`` to its string representation.
|
|
## If ``dump`` is ``true`` (default) full hexadecimal dump with ascii will be
|
|
## used, otherwise just a simple hexadecimal string will be printed.
|
|
var res = getTimedate(msg.timestamp)
|
|
let direction =
|
|
case msg.direction
|
|
of Incoming:
|
|
" << "
|
|
of Outgoing:
|
|
" >> "
|
|
let address =
|
|
block:
|
|
let local = block:
|
|
msg.local.withValue(loc): "[" & $loc & "]"
|
|
else: "[LOCAL]"
|
|
let remote = block:
|
|
msg.remote.withValue(rem): "[" & $rem & "]"
|
|
else: "[REMOTE]"
|
|
local & direction & remote
|
|
let seqid = block:
|
|
msg.seqID.withValue(seqid): "seqID = " & $seqid & " "
|
|
else: ""
|
|
let mtype = block:
|
|
msg.mtype.withValue(typ): "type = " & $typ & " "
|
|
else: ""
|
|
res.add(" ")
|
|
res.add(address)
|
|
res.add(" ")
|
|
res.add(mtype)
|
|
res.add(seqid)
|
|
res.add(mplexMessage(msg.message))
|
|
res.add(" ")
|
|
res.add("\p")
|
|
if dump:
|
|
res.add(dumpHex(msg.message))
|
|
else:
|
|
res.add(utils.toHex(msg.message))
|
|
res.add("\p")
|
|
res
|