You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
702 lines
32 KiB
702 lines
32 KiB
# -*- coding: utf-8 -*-
|
|
#
|
|
# Electrum - lightweight Bitcoin client
|
|
# Copyright (C) 2018 The Electrum developers
|
|
#
|
|
# Permission is hereby granted, free of charge, to any person
|
|
# obtaining a copy of this software and associated documentation files
|
|
# (the "Software"), to deal in the Software without restriction,
|
|
# including without limitation the rights to use, copy, modify, merge,
|
|
# publish, distribute, sublicense, and/or sell copies of the Software,
|
|
# and to permit persons to whom the Software is furnished to do so,
|
|
# subject to the following conditions:
|
|
#
|
|
# The above copyright notice and this permission notice shall be
|
|
# included in all copies or substantial portions of the Software.
|
|
#
|
|
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
|
|
# EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
|
# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
|
|
# NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
|
|
# BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
|
|
# ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
|
|
# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
# SOFTWARE.
|
|
|
|
from datetime import datetime
|
|
import time
|
|
import random
|
|
import queue
|
|
import os
|
|
import json
|
|
import threading
|
|
import concurrent
|
|
from collections import defaultdict
|
|
from typing import Sequence, List, Tuple, Optional, Dict, NamedTuple, TYPE_CHECKING, Set
|
|
import binascii
|
|
import base64
|
|
|
|
from sqlalchemy import Column, ForeignKey, Integer, String, Boolean
|
|
from sqlalchemy.orm.query import Query
|
|
from sqlalchemy.ext.declarative import declarative_base
|
|
from sqlalchemy.sql import not_, or_
|
|
from .sql_db import SqlDB, sql
|
|
|
|
from . import constants
|
|
from .util import bh2u, profiler, get_headers_dir, bfh, is_ip_address, list_enabled_bits, print_msg
|
|
from .logging import Logger
|
|
from .storage import JsonDB
|
|
from .lnverifier import LNChannelVerifier, verify_sig_for_channel_update
|
|
from .crypto import sha256d
|
|
from . import ecc
|
|
from .lnutil import (LN_GLOBAL_FEATURES_KNOWN_SET, LNPeerAddr, NUM_MAX_EDGES_IN_PAYMENT_PATH,
|
|
NotFoundChanAnnouncementForUpdate)
|
|
from .lnmsg import encode_msg
|
|
|
|
if TYPE_CHECKING:
|
|
from .lnchannel import Channel
|
|
from .network import Network
|
|
|
|
class UnknownEvenFeatureBits(Exception): pass
|
|
class NoChannelPolicy(Exception):
|
|
def __init__(self, short_channel_id: bytes):
|
|
super().__init__(f'cannot find channel policy for short_channel_id: {bh2u(short_channel_id)}')
|
|
|
|
def validate_features(features : int):
|
|
enabled_features = list_enabled_bits(features)
|
|
for fbit in enabled_features:
|
|
if (1 << fbit) not in LN_GLOBAL_FEATURES_KNOWN_SET and fbit % 2 == 0:
|
|
raise UnknownEvenFeatureBits()
|
|
|
|
Base = declarative_base()
|
|
|
|
FLAG_DISABLE = 1 << 1
|
|
FLAG_DIRECTION = 1 << 0
|
|
|
|
class ChannelInfo(Base):
|
|
__tablename__ = 'channel_info'
|
|
short_channel_id = Column(String(64), primary_key=True)
|
|
node1_id = Column(String(66), ForeignKey('node_info.node_id'), nullable=False)
|
|
node2_id = Column(String(66), ForeignKey('node_info.node_id'), nullable=False)
|
|
capacity_sat = Column(Integer)
|
|
msg_payload_hex = Column(String(1024), nullable=False)
|
|
trusted = Column(Boolean, nullable=False)
|
|
|
|
@staticmethod
|
|
def from_msg(payload):
|
|
features = int.from_bytes(payload['features'], 'big')
|
|
validate_features(features)
|
|
channel_id = payload['short_channel_id'].hex()
|
|
node_id_1 = payload['node_id_1'].hex()
|
|
node_id_2 = payload['node_id_2'].hex()
|
|
assert list(sorted([node_id_1, node_id_2])) == [node_id_1, node_id_2]
|
|
msg_payload_hex = encode_msg('channel_announcement', **payload).hex()
|
|
capacity_sat = None
|
|
return ChannelInfo(short_channel_id = channel_id, node1_id = node_id_1,
|
|
node2_id = node_id_2, capacity_sat = capacity_sat, msg_payload_hex = msg_payload_hex,
|
|
trusted = False)
|
|
|
|
@property
|
|
def msg_payload(self):
|
|
return bytes.fromhex(self.msg_payload_hex)
|
|
|
|
|
|
class Policy(Base):
|
|
__tablename__ = 'policy'
|
|
start_node = Column(String(66), ForeignKey('node_info.node_id'), primary_key=True)
|
|
short_channel_id = Column(String(64), ForeignKey('channel_info.short_channel_id'), primary_key=True)
|
|
cltv_expiry_delta = Column(Integer, nullable=False)
|
|
htlc_minimum_msat = Column(Integer, nullable=False)
|
|
htlc_maximum_msat = Column(Integer)
|
|
fee_base_msat = Column(Integer, nullable=False)
|
|
fee_proportional_millionths = Column(Integer, nullable=False)
|
|
channel_flags = Column(Integer, nullable=False)
|
|
timestamp = Column(Integer, nullable=False)
|
|
|
|
@staticmethod
|
|
def from_msg(payload, start_node, short_channel_id):
|
|
cltv_expiry_delta = payload['cltv_expiry_delta']
|
|
htlc_minimum_msat = payload['htlc_minimum_msat']
|
|
fee_base_msat = payload['fee_base_msat']
|
|
fee_proportional_millionths = payload['fee_proportional_millionths']
|
|
channel_flags = payload['channel_flags']
|
|
timestamp = payload['timestamp']
|
|
htlc_maximum_msat = payload.get('htlc_maximum_msat') # optional
|
|
|
|
cltv_expiry_delta = int.from_bytes(cltv_expiry_delta, "big")
|
|
htlc_minimum_msat = int.from_bytes(htlc_minimum_msat, "big")
|
|
htlc_maximum_msat = int.from_bytes(htlc_maximum_msat, "big") if htlc_maximum_msat else None
|
|
fee_base_msat = int.from_bytes(fee_base_msat, "big")
|
|
fee_proportional_millionths = int.from_bytes(fee_proportional_millionths, "big")
|
|
channel_flags = int.from_bytes(channel_flags, "big")
|
|
timestamp = int.from_bytes(timestamp, "big")
|
|
|
|
return Policy(start_node=start_node,
|
|
short_channel_id=short_channel_id,
|
|
cltv_expiry_delta=cltv_expiry_delta,
|
|
htlc_minimum_msat=htlc_minimum_msat,
|
|
fee_base_msat=fee_base_msat,
|
|
fee_proportional_millionths=fee_proportional_millionths,
|
|
channel_flags=channel_flags,
|
|
timestamp=timestamp,
|
|
htlc_maximum_msat=htlc_maximum_msat)
|
|
|
|
def is_disabled(self):
|
|
return self.channel_flags & FLAG_DISABLE
|
|
|
|
class NodeInfo(Base):
|
|
__tablename__ = 'node_info'
|
|
node_id = Column(String(66), primary_key=True, sqlite_on_conflict_primary_key='REPLACE')
|
|
features = Column(Integer, nullable=False)
|
|
timestamp = Column(Integer, nullable=False)
|
|
alias = Column(String(64), nullable=False)
|
|
|
|
@staticmethod
|
|
def from_msg(payload):
|
|
node_id = payload['node_id'].hex()
|
|
features = int.from_bytes(payload['features'], "big")
|
|
validate_features(features)
|
|
addresses = NodeInfo.parse_addresses_field(payload['addresses'])
|
|
alias = payload['alias'].rstrip(b'\x00').hex()
|
|
timestamp = int.from_bytes(payload['timestamp'], "big")
|
|
return NodeInfo(node_id=node_id, features=features, timestamp=timestamp, alias=alias), [
|
|
Address(host=host, port=port, node_id=node_id, last_connected_date=None) for host, port in addresses]
|
|
|
|
@staticmethod
|
|
def parse_addresses_field(addresses_field):
|
|
buf = addresses_field
|
|
def read(n):
|
|
nonlocal buf
|
|
data, buf = buf[0:n], buf[n:]
|
|
return data
|
|
addresses = []
|
|
while buf:
|
|
atype = ord(read(1))
|
|
if atype == 0:
|
|
pass
|
|
elif atype == 1: # IPv4
|
|
ipv4_addr = '.'.join(map(lambda x: '%d' % x, read(4)))
|
|
port = int.from_bytes(read(2), 'big')
|
|
if is_ip_address(ipv4_addr) and port != 0:
|
|
addresses.append((ipv4_addr, port))
|
|
elif atype == 2: # IPv6
|
|
ipv6_addr = b':'.join([binascii.hexlify(read(2)) for i in range(8)])
|
|
ipv6_addr = ipv6_addr.decode('ascii')
|
|
port = int.from_bytes(read(2), 'big')
|
|
if is_ip_address(ipv6_addr) and port != 0:
|
|
addresses.append((ipv6_addr, port))
|
|
elif atype == 3: # onion v2
|
|
host = base64.b32encode(read(10)) + b'.onion'
|
|
host = host.decode('ascii').lower()
|
|
port = int.from_bytes(read(2), 'big')
|
|
addresses.append((host, port))
|
|
elif atype == 4: # onion v3
|
|
host = base64.b32encode(read(35)) + b'.onion'
|
|
host = host.decode('ascii').lower()
|
|
port = int.from_bytes(read(2), 'big')
|
|
addresses.append((host, port))
|
|
else:
|
|
# unknown address type
|
|
# we don't know how long it is -> have to escape
|
|
# if there are other addresses we could have parsed later, they are lost.
|
|
break
|
|
return addresses
|
|
|
|
class Address(Base):
|
|
__tablename__ = 'address'
|
|
node_id = Column(String(66), ForeignKey('node_info.node_id'), primary_key=True)
|
|
host = Column(String(256), primary_key=True)
|
|
port = Column(Integer, primary_key=True)
|
|
last_connected_date = Column(Integer(), nullable=True)
|
|
|
|
|
|
|
|
class ChannelDB(SqlDB):
|
|
|
|
NUM_MAX_RECENT_PEERS = 20
|
|
|
|
def __init__(self, network: 'Network'):
|
|
path = os.path.join(get_headers_dir(network.config), 'channel_db')
|
|
super().__init__(network, path, Base)
|
|
self.num_nodes = 0
|
|
self.num_channels = 0
|
|
self._channel_updates_for_private_channels = {} # type: Dict[Tuple[bytes, bytes], dict]
|
|
self.ca_verifier = LNChannelVerifier(network, self)
|
|
self.update_counts()
|
|
|
|
@sql
|
|
def update_counts(self):
|
|
self._update_counts()
|
|
|
|
def _update_counts(self):
|
|
self.num_channels = self.DBSession.query(ChannelInfo).count()
|
|
self.num_policies = self.DBSession.query(Policy).count()
|
|
self.num_nodes = self.DBSession.query(NodeInfo).count()
|
|
self.logger.info(f'update counts {self.num_channels} {self.num_policies}')
|
|
|
|
@sql
|
|
def add_recent_peer(self, peer: LNPeerAddr):
|
|
now = int(time.time())
|
|
node_id = peer.pubkey.hex()
|
|
addr = self.DBSession.query(Address).filter_by(node_id=node_id, host=peer.host, port=peer.port).one_or_none()
|
|
if addr:
|
|
addr.last_connected_date = now
|
|
else:
|
|
addr = Address(node_id=node_id, host=peer.host, port=peer.port, last_connected_date=now)
|
|
self.DBSession.add(addr)
|
|
self.DBSession.commit()
|
|
|
|
@sql
|
|
def get_200_randomly_sorted_nodes_not_in(self, node_ids_bytes):
|
|
unshuffled = self.DBSession \
|
|
.query(NodeInfo) \
|
|
.filter(not_(NodeInfo.node_id.in_(x.hex() for x in node_ids_bytes))) \
|
|
.limit(200) \
|
|
.all()
|
|
return random.sample(unshuffled, len(unshuffled))
|
|
|
|
@sql
|
|
def nodes_get(self, node_id):
|
|
return self.DBSession \
|
|
.query(NodeInfo) \
|
|
.filter_by(node_id = node_id.hex()) \
|
|
.one_or_none()
|
|
|
|
@sql
|
|
def get_last_good_address(self, node_id) -> Optional[LNPeerAddr]:
|
|
r = self.DBSession.query(Address).filter_by(node_id=node_id.hex()).order_by(Address.last_connected_date.desc()).all()
|
|
if not r:
|
|
return None
|
|
addr = r[0]
|
|
return LNPeerAddr(addr.host, addr.port, bytes.fromhex(addr.node_id))
|
|
|
|
@sql
|
|
def get_recent_peers(self):
|
|
r = self.DBSession.query(Address).filter(Address.last_connected_date.isnot(None)).order_by(Address.last_connected_date.desc()).limit(self.NUM_MAX_RECENT_PEERS).all()
|
|
return [LNPeerAddr(x.host, x.port, bytes.fromhex(x.node_id)) for x in r]
|
|
|
|
@sql
|
|
def missing_short_chan_ids(self) -> Set[int]:
|
|
expr = not_(Policy.short_channel_id.in_(self.DBSession.query(ChannelInfo.short_channel_id)))
|
|
chan_ids_from_policy = set(x[0] for x in self.DBSession.query(Policy.short_channel_id).filter(expr).all())
|
|
if chan_ids_from_policy:
|
|
return chan_ids_from_policy
|
|
return set()
|
|
|
|
@sql
|
|
def add_verified_channel_info(self, short_id, capacity):
|
|
# called from lnchannelverifier
|
|
channel_info = self.DBSession.query(ChannelInfo).filter_by(short_channel_id = short_id.hex()).one_or_none()
|
|
channel_info.trusted = True
|
|
channel_info.capacity = capacity
|
|
self.DBSession.commit()
|
|
|
|
@sql
|
|
#@profiler
|
|
def on_channel_announcement(self, msg_payloads, trusted=False):
|
|
if type(msg_payloads) is dict:
|
|
msg_payloads = [msg_payloads]
|
|
new_channels = {}
|
|
for msg in msg_payloads:
|
|
short_channel_id = bh2u(msg['short_channel_id'])
|
|
if self.DBSession.query(ChannelInfo).filter_by(short_channel_id=short_channel_id).count():
|
|
continue
|
|
if constants.net.rev_genesis_bytes() != msg['chain_hash']:
|
|
self.logger.info("ChanAnn has unexpected chain_hash {}".format(bh2u(msg['chain_hash'])))
|
|
continue
|
|
try:
|
|
channel_info = ChannelInfo.from_msg(msg)
|
|
except UnknownEvenFeatureBits:
|
|
self.logger.info("unknown feature bits")
|
|
continue
|
|
channel_info.trusted = trusted
|
|
new_channels[short_channel_id] = channel_info
|
|
if not trusted:
|
|
self.ca_verifier.add_new_channel_info(channel_info.short_channel_id, channel_info.msg_payload)
|
|
for channel_info in new_channels.values():
|
|
self.DBSession.add(channel_info)
|
|
self.DBSession.commit()
|
|
#self.logger.info('on_channel_announcement: %d/%d'%(len(new_channels), len(msg_payloads)))
|
|
self._update_counts()
|
|
self.network.trigger_callback('ln_status')
|
|
|
|
@sql
|
|
def get_last_timestamp(self):
|
|
return self._get_last_timestamp()
|
|
|
|
def _get_last_timestamp(self):
|
|
from sqlalchemy.sql import func
|
|
r = self.DBSession.query(func.max(Policy.timestamp).label('max_timestamp')).one()
|
|
return r.max_timestamp or 0
|
|
|
|
@sql
|
|
#@profiler
|
|
def on_channel_update(self, msg_payloads, trusted=False):
|
|
if type(msg_payloads) is dict:
|
|
msg_payloads = [msg_payloads]
|
|
short_channel_ids = [msg_payload['short_channel_id'].hex() for msg_payload in msg_payloads]
|
|
channel_infos_list = self.DBSession.query(ChannelInfo).filter(ChannelInfo.short_channel_id.in_(short_channel_ids)).all()
|
|
channel_infos = {bfh(x.short_channel_id): x for x in channel_infos_list}
|
|
new_policies = {}
|
|
for msg_payload in msg_payloads:
|
|
short_channel_id = msg_payload['short_channel_id']
|
|
if constants.net.rev_genesis_bytes() != msg_payload['chain_hash']:
|
|
continue
|
|
channel_info = channel_infos.get(short_channel_id)
|
|
if not channel_info:
|
|
continue
|
|
flags = int.from_bytes(msg_payload['channel_flags'], 'big')
|
|
direction = flags & FLAG_DIRECTION
|
|
node_id = channel_info.node1_id if direction == 0 else channel_info.node2_id
|
|
if not trusted and not verify_sig_for_channel_update(msg_payload, bytes.fromhex(node_id)):
|
|
continue
|
|
short_channel_id = channel_info.short_channel_id
|
|
new_policy = Policy.from_msg(msg_payload, node_id, short_channel_id)
|
|
#self.logger.info(f'on_channel_update {datetime.fromtimestamp(new_policy.timestamp).ctime()}')
|
|
old_policy = self.DBSession.query(Policy).filter_by(short_channel_id=short_channel_id, start_node=node_id).one_or_none()
|
|
if old_policy:
|
|
if old_policy.timestamp >= new_policy.timestamp:
|
|
continue
|
|
self.DBSession.delete(old_policy)
|
|
p = new_policies.get((short_channel_id, node_id))
|
|
if p and p.timestamp >= new_policy.timestamp:
|
|
continue
|
|
new_policies[(short_channel_id, node_id)] = new_policy
|
|
# commit pending removals
|
|
self.DBSession.commit()
|
|
# add and commit new policies
|
|
for new_policy in new_policies.values():
|
|
self.DBSession.add(new_policy)
|
|
self.DBSession.commit()
|
|
if new_policies:
|
|
self.logger.info(f'on_channel_update: {len(new_policies)}/{len(msg_payloads)}')
|
|
self.logger.info(f'last timestamp: {datetime.fromtimestamp(self._get_last_timestamp()).ctime()}')
|
|
self._update_counts()
|
|
|
|
@sql
|
|
#@profiler
|
|
def on_node_announcement(self, msg_payloads):
|
|
if type(msg_payloads) is dict:
|
|
msg_payloads = [msg_payloads]
|
|
old_addr = None
|
|
new_nodes = {}
|
|
new_addresses = {}
|
|
for msg_payload in msg_payloads:
|
|
pubkey = msg_payload['node_id']
|
|
signature = msg_payload['signature']
|
|
h = sha256d(msg_payload['raw'][66:])
|
|
if not ecc.verify_signature(pubkey, signature, h):
|
|
continue
|
|
try:
|
|
node_info, node_addresses = NodeInfo.from_msg(msg_payload)
|
|
except UnknownEvenFeatureBits:
|
|
continue
|
|
#self.logger.info(f'received node announcement from {datetime.fromtimestamp(node_info.timestamp).ctime()}')
|
|
node_id = node_info.node_id
|
|
# Ignore node if it has no associated channel (DoS protection)
|
|
expr = or_(ChannelInfo.node1_id==node_id, ChannelInfo.node2_id==node_id)
|
|
if self.DBSession.query(ChannelInfo.short_channel_id).filter(expr).count() == 0:
|
|
#self.logger.info('ignoring orphan node_announcement')
|
|
continue
|
|
node = self.DBSession.query(NodeInfo).filter_by(node_id=node_id).one_or_none()
|
|
if node and node.timestamp >= node_info.timestamp:
|
|
continue
|
|
node = new_nodes.get(node_id)
|
|
if node and node.timestamp >= node_info.timestamp:
|
|
continue
|
|
new_nodes[node_id] = node_info
|
|
for addr in node_addresses:
|
|
new_addresses[(addr.node_id,addr.host,addr.port)] = addr
|
|
#self.logger.info("on_node_announcement: %d/%d"%(len(new_nodes), len(msg_payloads)))
|
|
for node_info in new_nodes.values():
|
|
self.DBSession.add(node_info)
|
|
for new_addr in new_addresses.values():
|
|
old_addr = self.DBSession.query(Address).filter_by(node_id=new_addr.node_id, host=new_addr.host, port=new_addr.port).one_or_none()
|
|
if not old_addr:
|
|
self.DBSession.add(new_addr)
|
|
self.DBSession.commit()
|
|
self._update_counts()
|
|
self.network.trigger_callback('ln_status')
|
|
|
|
def get_routing_policy_for_channel(self, start_node_id: bytes,
|
|
short_channel_id: bytes) -> Optional[bytes]:
|
|
if not start_node_id or not short_channel_id: return None
|
|
channel_info = self.get_channel_info(short_channel_id)
|
|
if channel_info is not None:
|
|
return self.get_policy_for_node(short_channel_id, start_node_id)
|
|
msg = self._channel_updates_for_private_channels.get((start_node_id, short_channel_id))
|
|
if not msg:
|
|
return None
|
|
return Policy.from_msg(msg, None, short_channel_id) # won't actually be written to DB
|
|
|
|
def add_channel_update_for_private_channel(self, msg_payload: dict, start_node_id: bytes):
|
|
if not verify_sig_for_channel_update(msg_payload, start_node_id):
|
|
return # ignore
|
|
short_channel_id = msg_payload['short_channel_id']
|
|
self._channel_updates_for_private_channels[(start_node_id, short_channel_id)] = msg_payload
|
|
|
|
@sql
|
|
def remove_channel(self, short_channel_id):
|
|
r = self.DBSession.query(ChannelInfo).filter_by(short_channel_id = short_channel_id.hex()).one_or_none()
|
|
if not r:
|
|
return
|
|
self.DBSession.delete(r)
|
|
self.DBSession.commit()
|
|
|
|
def print_graph(self, full_ids=False):
|
|
# used for debugging.
|
|
# FIXME there is a race here - iterables could change size from another thread
|
|
def other_node_id(node_id, channel_id):
|
|
channel_info = self.get_channel_info(channel_id)
|
|
if node_id == channel_info.node1_id:
|
|
other = channel_info.node2_id
|
|
else:
|
|
other = channel_info.node1_id
|
|
return other if full_ids else other[-4:]
|
|
|
|
print_msg('nodes')
|
|
for node in self.DBSession.query(NodeInfo).all():
|
|
print_msg(node)
|
|
|
|
print_msg('channels')
|
|
for channel_info in self.DBSession.query(ChannelInfo).all():
|
|
short_channel_id = channel_info.short_channel_id
|
|
node1 = channel_info.node1_id
|
|
node2 = channel_info.node2_id
|
|
direction1 = self.get_policy_for_node(channel_info, node1) is not None
|
|
direction2 = self.get_policy_for_node(channel_info, node2) is not None
|
|
if direction1 and direction2:
|
|
direction = 'both'
|
|
elif direction1:
|
|
direction = 'forward'
|
|
elif direction2:
|
|
direction = 'backward'
|
|
else:
|
|
direction = 'none'
|
|
print_msg('{}: {}, {}, {}'
|
|
.format(bh2u(short_channel_id),
|
|
bh2u(node1) if full_ids else bh2u(node1[-4:]),
|
|
bh2u(node2) if full_ids else bh2u(node2[-4:]),
|
|
direction))
|
|
|
|
|
|
@sql
|
|
def get_node_addresses(self, node_info):
|
|
return self.DBSession.query(Address).join(NodeInfo).filter_by(node_id = node_info.node_id).all()
|
|
|
|
@sql
|
|
@profiler
|
|
def load_data(self):
|
|
r = self.DBSession.query(ChannelInfo).all()
|
|
self._channels = dict([(bfh(x.short_channel_id), x) for x in r])
|
|
r = self.DBSession.query(Policy).filter_by().all()
|
|
self._policies = dict([((bfh(x.start_node), bfh(x.short_channel_id)), x) for x in r])
|
|
self._channels_for_node = defaultdict(set)
|
|
for channel_info in self._channels.values():
|
|
self._channels_for_node[bfh(channel_info.node1_id)].add(bfh(channel_info.short_channel_id))
|
|
self._channels_for_node[bfh(channel_info.node2_id)].add(bfh(channel_info.short_channel_id))
|
|
self.logger.info(f'load data {len(self._channels)} {len(self._policies)} {len(self._channels_for_node)}')
|
|
|
|
def get_policy_for_node(self, short_channel_id: bytes, node_id: bytes) -> Optional['Policy']:
|
|
return self._policies.get((node_id, short_channel_id))
|
|
|
|
def get_channel_info(self, channel_id: bytes):
|
|
return self._channels.get(channel_id)
|
|
|
|
def get_channels_for_node(self, node_id):
|
|
"""Returns the set of channels that have node_id as one of the endpoints."""
|
|
return self._channels_for_node.get(node_id)
|
|
|
|
|
|
|
|
class RouteEdge(NamedTuple("RouteEdge", [('node_id', bytes),
|
|
('short_channel_id', bytes),
|
|
('fee_base_msat', int),
|
|
('fee_proportional_millionths', int),
|
|
('cltv_expiry_delta', int)])):
|
|
"""if you travel through short_channel_id, you will reach node_id"""
|
|
|
|
def fee_for_edge(self, amount_msat: int) -> int:
|
|
return self.fee_base_msat \
|
|
+ (amount_msat * self.fee_proportional_millionths // 1_000_000)
|
|
|
|
@classmethod
|
|
def from_channel_policy(cls, channel_policy: 'Policy',
|
|
short_channel_id: bytes, end_node: bytes) -> 'RouteEdge':
|
|
assert type(short_channel_id) is bytes
|
|
assert type(end_node) is bytes
|
|
return RouteEdge(end_node,
|
|
short_channel_id,
|
|
channel_policy.fee_base_msat,
|
|
channel_policy.fee_proportional_millionths,
|
|
channel_policy.cltv_expiry_delta)
|
|
|
|
def is_sane_to_use(self, amount_msat: int) -> bool:
|
|
# TODO revise ad-hoc heuristics
|
|
# cltv cannot be more than 2 weeks
|
|
if self.cltv_expiry_delta > 14 * 144: return False
|
|
total_fee = self.fee_for_edge(amount_msat)
|
|
# fees below 50 sat are fine
|
|
if total_fee > 50_000:
|
|
# fee cannot be higher than amt
|
|
if total_fee > amount_msat: return False
|
|
# fee cannot be higher than 5000 sat
|
|
if total_fee > 5_000_000: return False
|
|
# unless amt is tiny, fee cannot be more than 10%
|
|
if amount_msat > 1_000_000 and total_fee > amount_msat/10: return False
|
|
return True
|
|
|
|
|
|
def is_route_sane_to_use(route: List[RouteEdge], invoice_amount_msat: int, min_final_cltv_expiry: int) -> bool:
|
|
"""Run some sanity checks on the whole route, before attempting to use it.
|
|
called when we are paying; so e.g. lower cltv is better
|
|
"""
|
|
if len(route) > NUM_MAX_EDGES_IN_PAYMENT_PATH:
|
|
return False
|
|
amt = invoice_amount_msat
|
|
cltv = min_final_cltv_expiry
|
|
for route_edge in reversed(route[1:]):
|
|
if not route_edge.is_sane_to_use(amt): return False
|
|
amt += route_edge.fee_for_edge(amt)
|
|
cltv += route_edge.cltv_expiry_delta
|
|
total_fee = amt - invoice_amount_msat
|
|
# TODO revise ad-hoc heuristics
|
|
# cltv cannot be more than 2 months
|
|
if cltv > 60 * 144: return False
|
|
# fees below 50 sat are fine
|
|
if total_fee > 50_000:
|
|
# fee cannot be higher than amt
|
|
if total_fee > invoice_amount_msat: return False
|
|
# fee cannot be higher than 5000 sat
|
|
if total_fee > 5_000_000: return False
|
|
# unless amt is tiny, fee cannot be more than 10%
|
|
if invoice_amount_msat > 1_000_000 and total_fee > invoice_amount_msat/10: return False
|
|
return True
|
|
|
|
|
|
class LNPathFinder(Logger):
|
|
|
|
def __init__(self, channel_db: ChannelDB):
|
|
Logger.__init__(self)
|
|
self.channel_db = channel_db
|
|
self.blacklist = set()
|
|
|
|
def _edge_cost(self, short_channel_id: bytes, start_node: bytes, end_node: bytes,
|
|
payment_amt_msat: int, ignore_costs=False) -> Tuple[float, int]:
|
|
"""Heuristic cost of going through a channel.
|
|
Returns (heuristic_cost, fee_for_edge_msat).
|
|
"""
|
|
channel_info = self.channel_db.get_channel_info(short_channel_id) # type: ChannelInfo
|
|
if channel_info is None:
|
|
return float('inf'), 0
|
|
channel_policy = self.channel_db.get_policy_for_node(short_channel_id, start_node)
|
|
if channel_policy is None:
|
|
return float('inf'), 0
|
|
if channel_policy.is_disabled(): return float('inf'), 0
|
|
route_edge = RouteEdge.from_channel_policy(channel_policy, short_channel_id, end_node)
|
|
if payment_amt_msat < channel_policy.htlc_minimum_msat:
|
|
return float('inf'), 0 # payment amount too little
|
|
if channel_info.capacity_sat is not None and \
|
|
payment_amt_msat // 1000 > channel_info.capacity_sat:
|
|
return float('inf'), 0 # payment amount too large
|
|
if channel_policy.htlc_maximum_msat is not None and \
|
|
payment_amt_msat > channel_policy.htlc_maximum_msat:
|
|
return float('inf'), 0 # payment amount too large
|
|
if not route_edge.is_sane_to_use(payment_amt_msat):
|
|
return float('inf'), 0 # thanks but no thanks
|
|
fee_msat = route_edge.fee_for_edge(payment_amt_msat) if not ignore_costs else 0
|
|
# TODO revise
|
|
# paying 10 more satoshis ~ waiting one more block
|
|
fee_cost = fee_msat / 1000 / 10
|
|
cltv_cost = route_edge.cltv_expiry_delta if not ignore_costs else 0
|
|
return cltv_cost + fee_cost + 1, fee_msat
|
|
|
|
@profiler
|
|
def find_path_for_payment(self, nodeA: bytes, nodeB: bytes,
|
|
invoice_amount_msat: int,
|
|
my_channels: List['Channel']=None) -> Sequence[Tuple[bytes, bytes]]:
|
|
"""Return a path from nodeA to nodeB.
|
|
|
|
Returns a list of (node_id, short_channel_id) representing a path.
|
|
To get from node ret[n][0] to ret[n+1][0], use channel ret[n+1][1];
|
|
i.e. an element reads as, "to get to node_id, travel through short_channel_id"
|
|
"""
|
|
self.channel_db.load_data()
|
|
assert type(nodeA) is bytes
|
|
assert type(nodeB) is bytes
|
|
assert type(invoice_amount_msat) is int
|
|
if my_channels is None: my_channels = []
|
|
my_channels = {chan.short_channel_id: chan for chan in my_channels}
|
|
|
|
# FIXME paths cannot be longer than 21 edges (onion packet)...
|
|
|
|
# run Dijkstra
|
|
# The search is run in the REVERSE direction, from nodeB to nodeA,
|
|
# to properly calculate compound routing fees.
|
|
distance_from_start = defaultdict(lambda: float('inf'))
|
|
distance_from_start[nodeB] = 0
|
|
prev_node = {}
|
|
nodes_to_explore = queue.PriorityQueue()
|
|
nodes_to_explore.put((0, invoice_amount_msat, nodeB)) # order of fields (in tuple) matters!
|
|
|
|
def inspect_edge():
|
|
if edge_channel_id in my_channels:
|
|
if edge_startnode == nodeA: # payment outgoing, on our channel
|
|
if not my_channels[edge_channel_id].can_pay(amount_msat):
|
|
return
|
|
else: # payment incoming, on our channel. (funny business, cycle weirdness)
|
|
assert edge_endnode == nodeA, (bh2u(edge_startnode), bh2u(edge_endnode))
|
|
pass # TODO?
|
|
edge_cost, fee_for_edge_msat = self._edge_cost(edge_channel_id,
|
|
start_node=edge_startnode,
|
|
end_node=edge_endnode,
|
|
payment_amt_msat=amount_msat,
|
|
ignore_costs=(edge_startnode == nodeA))
|
|
alt_dist_to_neighbour = distance_from_start[edge_endnode] + edge_cost
|
|
if alt_dist_to_neighbour < distance_from_start[edge_startnode]:
|
|
distance_from_start[edge_startnode] = alt_dist_to_neighbour
|
|
prev_node[edge_startnode] = edge_endnode, edge_channel_id
|
|
amount_to_forward_msat = amount_msat + fee_for_edge_msat
|
|
nodes_to_explore.put((alt_dist_to_neighbour, amount_to_forward_msat, edge_startnode))
|
|
|
|
# main loop of search
|
|
while nodes_to_explore.qsize() > 0:
|
|
dist_to_edge_endnode, amount_msat, edge_endnode = nodes_to_explore.get()
|
|
if edge_endnode == nodeA:
|
|
break
|
|
if dist_to_edge_endnode != distance_from_start[edge_endnode]:
|
|
# queue.PriorityQueue does not implement decrease_priority,
|
|
# so instead of decreasing priorities, we add items again into the queue.
|
|
# so there are duplicates in the queue, that we discard now:
|
|
continue
|
|
for edge_channel_id in self.channel_db.get_channels_for_node(edge_endnode):
|
|
assert type(edge_channel_id) is bytes
|
|
if edge_channel_id in self.blacklist: continue
|
|
channel_info = self.channel_db.get_channel_info(edge_channel_id)
|
|
edge_startnode = bfh(channel_info.node2_id) if bfh(channel_info.node1_id) == edge_endnode else bfh(channel_info.node1_id)
|
|
inspect_edge()
|
|
else:
|
|
return None # no path found
|
|
|
|
# backtrack from search_end (nodeA) to search_start (nodeB)
|
|
edge_startnode = nodeA
|
|
path = []
|
|
while edge_startnode != nodeB:
|
|
edge_endnode, edge_taken = prev_node[edge_startnode]
|
|
path += [(edge_endnode, edge_taken)]
|
|
edge_startnode = edge_endnode
|
|
return path
|
|
|
|
def create_route_from_path(self, path, from_node_id: bytes) -> List[RouteEdge]:
|
|
assert type(from_node_id) is bytes
|
|
if path is None:
|
|
raise Exception('cannot create route from None path')
|
|
route = []
|
|
prev_node_id = from_node_id
|
|
for node_id, short_channel_id in path:
|
|
channel_policy = self.channel_db.get_routing_policy_for_channel(prev_node_id, short_channel_id)
|
|
if channel_policy is None:
|
|
raise NoChannelPolicy(short_channel_id)
|
|
route.append(RouteEdge.from_channel_policy(channel_policy, short_channel_id, node_id))
|
|
prev_node_id = node_id
|
|
return route
|
|
|