nwaku/waku/v1/protocol/waku_protocol.nim

671 lines
24 KiB
Nim

#
# Waku
# (c) Copyright 2018-2019
# Status Research & Development GmbH
#
# Licensed under either of
# Apache License, version 2.0, (LICENSE-APACHEv2)
# MIT license (LICENSE-MIT)
#
## Waku
## *******
##
## Waku is a fork of Whisper.
##
## Waku is a gossip protocol that synchronizes a set of messages across nodes
## with attention given to sender and recipient anonymitiy. Messages are
## categorized by a topic and stay alive in the network based on a time-to-live
## measured in seconds. Spam prevention is based on proof-of-work, where large
## or long-lived messages must spend more work.
##
## Implementation should be according to Waku specification defined here:
## https://github.com/vacp2p/specs/blob/master/waku/waku.md
##
## Example usage
## ----------
## First an `EthereumNode` needs to be created, either with all capabilities set
## or with specifically the Waku capability set.
## The latter can be done like this:
##
## .. code-block::nim
## var node = newEthereumNode(keypair, address, netId, nil,
## addAllCapabilities = false)
## node.addCapability Waku
##
## Now calls such as ``postMessage`` and ``subscribeFilter`` can be done.
## However, they only make real sense after ``connectToNetwork`` was started. As
## else there will be no peers to send and receive messages from.
import
options, tables, times, chronos, chronicles, metrics,
eth/[keys, async_utils, p2p], eth/p2p/rlpx_protocols/whisper/whisper_types,
eth/trie/trie_defs
export
whisper_types
logScope:
topics = "waku"
const
defaultQueueCapacity = 2048
wakuVersion* = 1 ## Waku version.
wakuVersionStr* = $wakuVersion ## Waku version.
defaultMinPow* = 0.2'f64 ## The default minimum PoW requirement for this node.
defaultMaxMsgSize* = 1024'u32 * 1024'u32 ## The current default and max
## message size. This can never be larger than the maximum RLPx message size.
messageInterval* = chronos.milliseconds(300) ## Interval at which messages are
## send to peers, in ms.
pruneInterval* = chronos.milliseconds(1000) ## Interval at which message
## queue is pruned, in ms.
topicInterestMax = 10000
type
WakuConfig* = object
powRequirement*: float64
bloom*: Option[Bloom]
isLightNode*: bool
maxMsgSize*: uint32
confirmationsEnabled*: bool
rateLimits*: Option[RateLimits]
topics*: Option[seq[Topic]]
Accounting* = ref object
sent*: uint
received*: uint
WakuPeer = ref object
initialized: bool # when successfully completed the handshake
powRequirement*: float64
bloom*: Bloom
isLightNode*: bool
trusted*: bool
topics*: Option[seq[Topic]]
received: HashSet[Hash]
accounting*: Accounting
P2PRequestHandler* = proc(peer: Peer, envelope: Envelope)
{.gcsafe, raises: [Defect].}
EnvReceivedHandler* = proc(envelope: Envelope) {.gcsafe, raises: [Defect].}
WakuNetwork = ref object
queue*: ref Queue
filters*: Filters
config*: WakuConfig
p2pRequestHandler*: P2PRequestHandler
envReceivedHandler*: EnvReceivedHandler
RateLimits* = object
# TODO: uint or specifically uint32?
limitIp*: uint
limitPeerId*: uint
limitTopic*: uint
StatusOptions* = object
powRequirement*: Option[(float64)]
bloomFilter*: Option[Bloom]
lightNode*: Option[bool]
confirmationsEnabled*: Option[bool]
rateLimits*: Option[RateLimits]
topicInterest*: Option[seq[Topic]]
KeyKind* = enum
powRequirementKey,
bloomFilterKey,
lightNodeKey,
confirmationsEnabledKey,
rateLimitsKey,
topicInterestKey
template countSomeFields*(x: StatusOptions): int =
var count = 0
for f in fields(x):
if f.isSome():
inc count
count
proc append*(rlpWriter: var RlpWriter, value: StatusOptions) =
var list = initRlpList(countSomeFields(value))
if value.powRequirement.isSome():
list.append((powRequirementKey, cast[uint64](value.powRequirement.get())))
if value.bloomFilter.isSome():
list.append((bloomFilterKey, @(value.bloomFilter.get())))
if value.lightNode.isSome():
list.append((lightNodeKey, value.lightNode.get()))
if value.confirmationsEnabled.isSome():
list.append((confirmationsEnabledKey, value.confirmationsEnabled.get()))
if value.rateLimits.isSome():
list.append((rateLimitsKey, value.rateLimits.get()))
if value.topicInterest.isSome():
list.append((topicInterestKey, value.topicInterest.get()))
let bytes = list.finish()
rlpWriter.append(rlpFromBytes(bytes))
proc read*(rlp: var Rlp, T: typedesc[StatusOptions]): T =
if not rlp.isList():
raise newException(RlpTypeMismatch,
"List expected, but the source RLP is not a list.")
let sz = rlp.listLen()
# We already know that we are working with a list
doAssert rlp.enterList()
for i in 0 ..< sz:
rlp.tryEnterList()
var k: KeyKind
try:
k = rlp.read(KeyKind)
except RlpTypeMismatch:
# skip unknown keys and their value
rlp.skipElem()
rlp.skipElem()
continue
case k
of powRequirementKey:
let pow = rlp.read(uint64)
result.powRequirement = some(cast[float64](pow))
of bloomFilterKey:
let bloom = rlp.read(seq[byte])
if bloom.len != bloomSize:
raise newException(RlpTypeMismatch, "Bloomfilter size mismatch")
var bloomFilter: Bloom
bloomFilter.bytesCopy(bloom)
result.bloomFilter = some(bloomFilter)
of lightNodeKey:
result.lightNode = some(rlp.read(bool))
of confirmationsEnabledKey:
result.confirmationsEnabled = some(rlp.read(bool))
of rateLimitsKey:
result.rateLimits = some(rlp.read(RateLimits))
of topicInterestKey:
result.topicInterest = some(rlp.read(seq[Topic]))
proc allowed*(msg: Message, config: WakuConfig): bool =
# Check max msg size, already happens in RLPx but there is a specific waku
# max msg size which should always be < RLPx max msg size
if msg.size > config.maxMsgSize:
envelopes_dropped.inc(labelValues = ["too_large"])
warn "Message size too large", size = msg.size
return false
if msg.pow < config.powRequirement:
envelopes_dropped.inc(labelValues = ["low_pow"])
warn "Message PoW too low", pow = msg.pow, minPow = config.powRequirement
return false
if config.topics.isSome():
if msg.env.topic notin config.topics.get():
envelopes_dropped.inc(labelValues = ["topic_mismatch"])
warn "Message topic does not match Waku topic list"
return false
else:
if config.bloom.isSome() and not bloomFilterMatch(config.bloom.get(), msg.bloom):
envelopes_dropped.inc(labelValues = ["bloom_filter_mismatch"])
warn "Message does not match node bloom filter"
return false
return true
proc run(peer: Peer) {.gcsafe, async, raises: [Defect].}
proc run(node: EthereumNode, network: WakuNetwork)
{.gcsafe, async, raises: [Defect].}
proc initProtocolState*(network: WakuNetwork, node: EthereumNode) {.gcsafe.} =
new(network.queue)
network.queue[] = initQueue(defaultQueueCapacity)
network.filters = initTable[string, Filter]()
network.config.bloom = some(fullBloom())
network.config.powRequirement = defaultMinPow
network.config.isLightNode = false
# RateLimits and confirmations are not yet implemented so we set confirmations
# to false and we don't pass RateLimits at all.
network.config.confirmationsEnabled = false
network.config.rateLimits = none(RateLimits)
network.config.maxMsgSize = defaultMaxMsgSize
network.config.topics = none(seq[Topic])
asyncSpawn node.run(network)
p2pProtocol Waku(version = wakuVersion,
rlpxName = "waku",
peerState = WakuPeer,
networkState = WakuNetwork):
onPeerConnected do (peer: Peer):
trace "onPeerConnected Waku"
let
wakuNet = peer.networkState
wakuPeer = peer.state
let options = StatusOptions(
powRequirement: some(wakuNet.config.powRequirement),
bloomFilter: wakuNet.config.bloom,
lightNode: some(wakuNet.config.isLightNode),
confirmationsEnabled: some(wakuNet.config.confirmationsEnabled),
rateLimits: wakuNet.config.rateLimits,
topicInterest: wakuNet.config.topics)
let m = await peer.status(options,
timeout = chronos.milliseconds(5000))
wakuPeer.powRequirement = m.options.powRequirement.get(defaultMinPow)
wakuPeer.bloom = m.options.bloomFilter.get(fullBloom())
wakuPeer.isLightNode = m.options.lightNode.get(false)
if wakuPeer.isLightNode and wakuNet.config.isLightNode:
# No sense in connecting two light nodes so we disconnect
raise newException(UselessPeerError, "Two light nodes connected")
wakuPeer.topics = m.options.topicInterest
if wakuPeer.topics.isSome():
if wakuPeer.topics.get().len > topicInterestMax:
raise newException(UselessPeerError, "Topic-interest is too large")
if wakuNet.config.topics.isSome():
raise newException(UselessPeerError,
"Two Waku nodes with topic-interest connected")
wakuPeer.received.init()
wakuPeer.trusted = false
wakuPeer.accounting = Accounting(sent: 0, received: 0)
wakuPeer.initialized = true
# No timer based queue processing for a light node.
if not wakuNet.config.isLightNode:
asyncSpawn peer.run()
debug "Waku peer initialized", peer
handshake:
proc status(peer: Peer, options: StatusOptions)
proc messages(peer: Peer, envelopes: openarray[Envelope]) =
if not peer.state.initialized:
warn "Handshake not completed yet, discarding messages"
return
for envelope in envelopes:
# check if expired or in future, or ttl not 0
if not envelope.valid():
warn "Expired or future timed envelope", peer
# disconnect from peers sending bad envelopes
# await peer.disconnect(SubprotocolReason)
continue
peer.state.accounting.received += 1
let msg = initMessage(envelope)
if not msg.allowed(peer.networkState.config):
# disconnect from peers sending bad envelopes
# await peer.disconnect(SubprotocolReason)
continue
# This peer send this message thus should not receive it again.
# If this peer has the message in the `received` set already, this means
# it was either already received here from this peer or send to this peer.
# Either way it will be in our queue already (and the peer should know
# this) and this peer is sending duplicates.
# Note: geth does not check if a peer has send a message to them before
# broadcasting this message. This too is seen here as a duplicate message
# (see above comment). If we want to seperate these cases (e.g. when peer
# rating), then we have to add a "peer.state.send" HashSet.
# Note: it could also be a race between the arrival of a message send by
# this node to a peer and that same message arriving from that peer (after
# it was received from another peer) here.
if peer.state.received.containsOrIncl(msg.hash):
envelopes_dropped.inc(labelValues = ["duplicate"])
trace "Peer sending duplicate messages", peer, hash = $msg.hash
# await peer.disconnect(SubprotocolReason)
continue
# This can still be a duplicate message, but from another peer than
# the peer who send the message.
if peer.networkState.queue[].add(msg):
# notify filters of this message
peer.networkState.filters.notify(msg)
# trigger handler on received envelope, if registered
if not peer.networkState.envReceivedHandler.isNil():
peer.networkState.envReceivedHandler(envelope)
nextID 22
proc statusOptions(peer: Peer, options: StatusOptions) =
if not peer.state.initialized:
warn "Handshake not completed yet, discarding statusOptions"
return
if options.topicInterest.isSome():
peer.state.topics = options.topicInterest
elif options.bloomFilter.isSome():
peer.state.bloom = options.bloomFilter.get()
peer.state.topics = none(seq[Topic])
if options.powRequirement.isSome():
peer.state.powRequirement = options.powRequirement.get()
if options.lightNode.isSome():
peer.state.isLightNode = options.lightNode.get()
nextID 126
proc p2pRequest(peer: Peer, envelope: Envelope) =
if not peer.networkState.p2pRequestHandler.isNil():
peer.networkState.p2pRequestHandler(peer, envelope)
proc p2pMessage(peer: Peer, envelopes: openarray[Envelope]) =
if peer.state.trusted:
# when trusted we can bypass any checks on envelope
for envelope in envelopes:
let msg = Message(env: envelope, isP2P: true)
peer.networkState.filters.notify(msg)
# Following message IDs are not part of EIP-627, but are added and used by
# the Status application, we ignore them for now.
nextID 11
proc batchAcknowledged(peer: Peer) = discard
proc messageResponse(peer: Peer) = discard
nextID 123
requestResponse:
proc p2pSyncRequest(peer: Peer) = discard
proc p2pSyncResponse(peer: Peer) = discard
proc p2pRequestComplete(peer: Peer, requestId: Hash, lastEnvelopeHash: Hash,
cursor: seq[byte]) = discard
# TODO:
# In the current specification the parameters are not wrapped in a regular
# envelope as is done for the P2P Request packet. If we could alter this in
# the spec it would be a cleaner separation between Waku and Mail server /
# client.
# Also, if a requestResponse block is used, a reqestId will automatically
# be added by the protocol DSL.
# However the requestResponse block in combination with p2pRequest cannot be
# used due to the unfortunate fact that the packet IDs are not consecutive,
# and nextID is not recognized in between these. The nextID behaviour could
# be fixed, however it would be cleaner if the specification could be
# changed to have these IDs to be consecutive.
# 'Runner' calls ---------------------------------------------------------------
proc processQueue(peer: Peer) {.raises: [Defect].} =
# Send to peer all valid and previously not send envelopes in the queue.
var
envelopes: seq[Envelope] = @[]
wakuPeer = peer.state(Waku)
wakuNet = peer.networkState(Waku)
for message in wakuNet.queue.items:
if wakuPeer.received.contains(message.hash):
# trace "message was already send to peer", hash = $message.hash, peer
continue
if message.pow < wakuPeer.powRequirement:
trace "Message PoW too low for peer", pow = message.pow,
powReq = wakuPeer.powRequirement
continue
if wakuPeer.topics.isSome():
if message.env.topic notin wakuPeer.topics.get():
trace "Message does not match topics list"
continue
else:
if not bloomFilterMatch(wakuPeer.bloom, message.bloom):
trace "Message does not match peer bloom filter"
continue
trace "Adding envelope"
envelopes.add(message.env)
wakuPeer.accounting.sent += 1
wakuPeer.received.incl(message.hash)
if envelopes.len() > 0:
trace "Sending envelopes", amount=envelopes.len
# Ignore failure of sending messages, this could occur when the connection
# gets dropped
traceAsyncErrors peer.messages(envelopes)
proc run(peer: Peer) {.async, raises: [Defect].} =
while peer.connectionState notin {Disconnecting, Disconnected}:
peer.processQueue()
await sleepAsync(messageInterval)
proc pruneReceived(node: EthereumNode) {.raises: [].} =
if node.peerPool != nil: # XXX: a bit dirty to need to check for this here ...
var wakuNet = node.protocolState(Waku)
for peer in node.protocolPeers(Waku):
if not peer.initialized:
continue
# NOTE: Perhaps alter the queue prune call to keep track of a HashSet
# of pruned messages (as these should be smaller), and diff this with
# the received sets.
peer.received = intersection(peer.received, wakuNet.queue.itemHashes)
proc run(node: EthereumNode, network: WakuNetwork) {.async, raises: [Defect].} =
while true:
# prune message queue every second
# TTL unit is in seconds, so this should be sufficient?
network.queue[].prune()
# pruning the received sets is not necessary for correct workings
# but simply from keeping the sets growing indefinitely
node.pruneReceived()
await sleepAsync(pruneInterval)
# Private EthereumNode calls ---------------------------------------------------
proc sendP2PMessage(node: EthereumNode, peerId: NodeId,
envelopes: openarray[Envelope]): bool =
for peer in node.peers(Waku):
if peer.remote.id == peerId:
let f = peer.p2pMessage(envelopes)
# Can't make p2pMessage not raise so this is the "best" option I can think
# of instead of using asyncSpawn and still keeping the call not async.
f.callback = proc(data: pointer) {.gcsafe, raises: [Defect].} =
if f.failed:
warn "P2PMessage send failed", msg = f.readError.msg
return true
proc queueMessage(node: EthereumNode, msg: Message): bool =
var wakuNet = node.protocolState(Waku)
# We have to do the same checks here as in the messages proc not to leak
# any information that the message originates from this node.
if not msg.allowed(wakuNet.config):
return false
trace "Adding message to queue", hash = $msg.hash
if wakuNet.queue[].add(msg):
# Also notify our own filters of the message we are sending,
# e.g. msg from local Dapp to Dapp
wakuNet.filters.notify(msg)
return true
# Public EthereumNode calls ----------------------------------------------------
proc postMessage*(node: EthereumNode, pubKey = none[PublicKey](),
symKey = none[SymKey](), src = none[PrivateKey](),
ttl: uint32, topic: Topic, payload: seq[byte],
padding = none[seq[byte]](), powTime = 1'f,
powTarget = defaultMinPow,
targetPeer = none[NodeId]()): bool =
## Post a message on the message queue which will be processed at the
## next `messageInterval`.
##
## NOTE: This call allows a post without encryption. If encryption is
## mandatory it should be enforced a layer up
let payload = encode(node.rng[], Payload(
payload: payload, src: src, dst: pubKey, symKey: symKey, padding: padding))
if payload.isSome():
var env = Envelope(expiry:epochTime().uint32 + ttl,
ttl: ttl, topic: topic, data: payload.get(), nonce: 0)
# Allow lightnode to post only direct p2p messages
if targetPeer.isSome():
return node.sendP2PMessage(targetPeer.get(), [env])
else:
# non direct p2p message can not have ttl of 0
if env.ttl == 0:
return false
var msg = initMessage(env, powCalc = false)
# XXX: make this non blocking or not?
# In its current blocking state, it could be noticed by a peer that no
# messages are send for a while, and thus that mining PoW is done, and
# that next messages contains a message originated from this peer
# zah: It would be hard to execute this in a background thread at the
# moment. We'll need a way to send custom "tasks" to the async message
# loop (e.g. AD2 support for AsyncChannels).
if not msg.sealEnvelope(powTime, powTarget):
return false
# need to check expiry after mining PoW
if not msg.env.valid():
return false
result = node.queueMessage(msg)
# Allows light nodes to post via untrusted messages packet.
# Queue gets processed immediatly as the node sends only its own messages,
# so the privacy ship has already sailed anyhow.
# TODO:
# - Could be still a concern in terms of efficiency, if multiple messages
# need to be send.
# - For Waku Mode, the checks in processQueue are rather useless as the
# idea is to connect only to 1 node? Also refactor in that case.
if node.protocolState(Waku).config.isLightNode:
for peer in node.peers(Waku):
peer.processQueue()
else:
error "Encoding of payload failed"
return false
proc subscribeFilter*(node: EthereumNode, filter: Filter,
handler:FilterMsgHandler = nil): string =
## Initiate a filter for incoming/outgoing messages. Messages can be
## retrieved with the `getFilterMessages` call or with a provided
## `FilterMsgHandler`.
##
## NOTE: This call allows for a filter without decryption. If encryption is
## mandatory it should be enforced a layer up.
return subscribeFilter(
node.rng[], node.protocolState(Waku).filters, filter, handler)
proc unsubscribeFilter*(node: EthereumNode, filterId: string): bool =
## Remove a previously subscribed filter.
var filter: Filter
return node.protocolState(Waku).filters.take(filterId, filter)
proc getFilterMessages*(node: EthereumNode, filterId: string): seq[ReceivedMessage] =
## Get all the messages currently in the filter queue. This will reset the
## filter message queue.
return node.protocolState(Waku).filters.getFilterMessages(filterId)
proc filtersToBloom*(node: EthereumNode): Bloom =
## Returns the bloom filter of all topics of all subscribed filters.
return node.protocolState(Waku).filters.toBloom()
proc setPowRequirement*(node: EthereumNode, powReq: float64) {.async.} =
## Sets the PoW requirement for this node, will also send
## this new PoW requirement to all connected peers.
##
## Failures when sending messages to peers will not be reported.
# NOTE: do we need a tolerance of old PoW for some time?
node.protocolState(Waku).config.powRequirement = powReq
var futures: seq[Future[void]] = @[]
let list = StatusOptions(powRequirement: some(powReq))
for peer in node.peers(Waku):
futures.add(peer.statusOptions(list))
# Exceptions from sendMsg will not be raised
await allFutures(futures)
proc setBloomFilter*(node: EthereumNode, bloom: Bloom) {.async.} =
## Sets the bloom filter for this node, will also send
## this new bloom filter to all connected peers.
##
## Failures when sending messages to peers will not be reported.
# NOTE: do we need a tolerance of old bloom filter for some time?
node.protocolState(Waku).config.bloom = some(bloom)
# reset topics
node.protocolState(Waku).config.topics = none(seq[Topic])
var futures: seq[Future[void]] = @[]
let list = StatusOptions(bloomFilter: some(bloom))
for peer in node.peers(Waku):
futures.add(peer.statusOptions(list))
# Exceptions from sendMsg will not be raised
await allFutures(futures)
proc setTopicInterest*(node: EthereumNode, topics: seq[Topic]):
Future[bool] {.async.} =
if topics.len > topicInterestMax:
return false
node.protocolState(Waku).config.topics = some(topics)
var futures: seq[Future[void]] = @[]
let list = StatusOptions(topicInterest: some(topics))
for peer in node.peers(Waku):
futures.add(peer.statusOptions(list))
# Exceptions from sendMsg will not be raised
await allFutures(futures)
return true
proc setMaxMessageSize*(node: EthereumNode, size: uint32): bool =
## Set the maximum allowed message size.
## Can not be set higher than ``defaultMaxMsgSize``.
if size > defaultMaxMsgSize:
warn "size > defaultMaxMsgSize"
return false
node.protocolState(Waku).config.maxMsgSize = size
return true
proc setPeerTrusted*(node: EthereumNode, peerId: NodeId): bool =
## Set a connected peer as trusted.
for peer in node.peers(Waku):
if peer.remote.id == peerId:
peer.state(Waku).trusted = true
return true
proc setLightNode*(node: EthereumNode, isLightNode: bool) {.async.} =
## Set this node as a Waku light node.
node.protocolState(Waku).config.isLightNode = isLightNode
# TODO: Add starting/stopping of `processQueue` loop depending on value of isLightNode.
var futures: seq[Future[void]] = @[]
let list = StatusOptions(lightNode: some(isLightNode))
for peer in node.peers(Waku):
futures.add(peer.statusOptions(list))
# Exceptions from sendMsg will not be raised
await allFutures(futures)
proc configureWaku*(node: EthereumNode, config: WakuConfig) =
## Apply a Waku configuration.
##
## NOTE: Should be run before connection is made with peers as some
## of the settings are only communicated at peer handshake.
node.protocolState(Waku).config = config
proc registerP2PRequestHandler*(node: EthereumNode,
customHandler: P2PRequestHandler) =
node.protocolState(Waku).p2pRequestHandler = customHandler
proc registerEnvReceivedHandler*(node: EthereumNode,
customHandler: EnvReceivedHandler) =
node.protocolState(Waku).envReceivedHandler = customHandler
proc resetMessageQueue*(node: EthereumNode) =
## Full reset of the message queue.
##
## NOTE: Not something that should be run in normal circumstances.
node.protocolState(Waku).queue[] = initQueue(defaultQueueCapacity)