mirror of
https://github.com/NeoCloud/NeoNetwork
synced 2024-11-16 13:50:41 +08:00
373 lines
12 KiB
Python
Executable file
373 lines
12 KiB
Python
Executable file
#!/usr/bin/env python3
|
|
import argparse
|
|
import json
|
|
import re
|
|
import time
|
|
from collections import defaultdict
|
|
from contextlib import redirect_stdout
|
|
from io import StringIO
|
|
from ipaddress import IPv4Network, IPv6Network, ip_network
|
|
from itertools import combinations
|
|
from pathlib import Path
|
|
|
|
import netaddr
|
|
import toml
|
|
from tabulate import tabulate
|
|
|
|
NEO_NETWORK_POOL = [ip_network("10.127.0.0/16"), ip_network("fd10:127::/32")]
|
|
|
|
|
|
def pick(entity: dict, fields: [str], **kwargs: dict):
|
|
new_entity = {}
|
|
for field in fields:
|
|
new_entity[field] = entity.get(field)
|
|
for old_field, new_field in kwargs.items():
|
|
new_entity[new_field] = entity.get(old_field)
|
|
return new_entity
|
|
|
|
|
|
def is_neo_network(address):
|
|
return any(
|
|
address.version == neo.version and address.subnet_of(neo)
|
|
for neo in NEO_NETWORK_POOL
|
|
)
|
|
|
|
|
|
def is_neo_network_asn(asn: int):
|
|
return 4201270000 <= asn <= 4201279999
|
|
|
|
|
|
def is_dn42_asn(asn: int):
|
|
return 4242420000 <= asn <= 4242429999
|
|
|
|
|
|
def name_to_nic_hdl(name):
|
|
r, num = re.subn(r"[^0-9A-Z]", "-", name.upper())
|
|
_r = len(r.replace("-", ""))
|
|
assert _r >= 3 # has at least 3 effective chars
|
|
assert r[0] != "-" # starts with [0-9A-Z]
|
|
assert num < _r # not too many subs
|
|
return r
|
|
|
|
|
|
def iter_toml_file(path: str):
|
|
for item in Path(path).iterdir():
|
|
if not item.is_file() or item.suffix != ".toml":
|
|
continue
|
|
yield item, toml.loads(item.read_text())
|
|
|
|
|
|
def load_entities():
|
|
for item, entity in iter_toml_file("entity"):
|
|
yield item.stem, entity
|
|
|
|
|
|
def load_asn(entities: dict):
|
|
for item, entity in iter_toml_file("asn"):
|
|
asn = int(item.stem.lstrip("AS"))
|
|
entity["source"] = (
|
|
is_neo_network_asn(asn)
|
|
and "NeoNetwork"
|
|
or is_dn42_asn(asn)
|
|
and "DN42"
|
|
or entity.get("source")
|
|
)
|
|
assert entity["owner"] in entities
|
|
assert entity["source"] in ["NeoNetwork", "DN42", "Internet"]
|
|
yield asn, entity
|
|
|
|
|
|
def node_to_asn(orignal_asn_set: set):
|
|
node_table = dict()
|
|
for _, entities in iter_toml_file("node"):
|
|
mapping = {name: entity["asn"] for (name, entity) in entities.items()}
|
|
asn_set = set(mapping.values())
|
|
assert orignal_asn_set & asn_set == asn_set
|
|
node_table.update(mapping)
|
|
return node_table
|
|
|
|
|
|
def assert_peer(nodes: set):
|
|
for item, entities in iter_toml_file("peer"):
|
|
peers = set(entities["to-peer"])
|
|
assert item.stem in nodes
|
|
assert nodes & peers == peers
|
|
|
|
|
|
def route_to_roa(asn_table: dict):
|
|
def make_route():
|
|
for item, entity in iter_toml_file("route"):
|
|
asn = int(item.stem.lstrip("AS"))
|
|
for prefix, fields in entity.items():
|
|
if fields["type"] not in ("loopback", "subnet"):
|
|
continue
|
|
fields["asn"] = asn
|
|
fields["prefix"] = ip_network(prefix, strict=True)
|
|
supernet = fields.get("supernet")
|
|
fields["supernet"] = (
|
|
ip_network(supernet, strict=True) if supernet else None
|
|
)
|
|
assert fields["name"]
|
|
assert is_neo_network(fields["prefix"])
|
|
assert not fields["supernet"] or is_neo_network(fields["supernet"])
|
|
yield pick(fields, ["asn", "name", "type", "prefix", "supernet"])
|
|
|
|
entities = sorted(make_route(), key=lambda item: item["asn"])
|
|
prefixes = [item["prefix"] for item in entities]
|
|
for net1, net2 in combinations(
|
|
sorted(entities, key=lambda net: net["prefix"].prefixlen), 2
|
|
):
|
|
if not net1["prefix"].overlaps(net2["prefix"]):
|
|
continue
|
|
assert net1["prefix"] != net2["prefix"]
|
|
assert net1["prefix"].supernet_of(net2["prefix"])
|
|
s1net, s2net = (net1["supernet"], net2["supernet"])
|
|
assert s2net # please include supernet = <cidr> in your route
|
|
# if net1(the bigger net) has a supernet s1net, then s1net and net1
|
|
# will be checked or must have been checked, same for net2
|
|
assert not s1net or s1net in prefixes # net1.supernet is garbage
|
|
assert s2net == net1["prefix"] or s2net in prefixes # net2.supernet is garbage
|
|
return entities
|
|
|
|
|
|
def prehandle_roa(asn_table: dict, args):
|
|
roa = route_to_roa(asn_table)
|
|
max_prefixlen = IPv4Network(0).max_prefixlen
|
|
roa4 = filter(lambda item: isinstance(item["prefix"], IPv4Network), roa)
|
|
roa6 = filter(lambda item: isinstance(item["prefix"], IPv6Network), roa)
|
|
if args.ipv4:
|
|
roa6 = []
|
|
elif args.ipv6:
|
|
roa4 = []
|
|
roa4 = [
|
|
r
|
|
for r in roa4
|
|
if r["prefix"].prefixlen <= args.max or r["prefix"].prefixlen == max_prefixlen
|
|
]
|
|
roa6 = [r for r in roa6 if r["prefix"].prefixlen <= args.max6]
|
|
for r in roa4:
|
|
r["maxLength"] = args.max
|
|
if r["prefix"].prefixlen == max_prefixlen:
|
|
r["maxLength"] = max_prefixlen
|
|
for r in roa6:
|
|
r["maxLength"] = args.max6
|
|
for r in (*roa4, *roa6):
|
|
r["prefix"] = r["prefix"].with_prefixlen
|
|
return roa4, roa6
|
|
|
|
|
|
def make_export(roa4, roa6):
|
|
def modify_entity(entity):
|
|
entity["nic_hdl"] = name_to_nic_hdl(entity["name"])
|
|
return entity
|
|
|
|
def filter_route(records, asn):
|
|
return [
|
|
pick(roa, ["prefix", "maxLength"], name="netname")
|
|
for roa in records
|
|
if roa["asn"] == asn
|
|
]
|
|
|
|
entities = dict(load_entities())
|
|
asn_list = [
|
|
{
|
|
"asn": asn,
|
|
"owner": asn_info["owner"],
|
|
"name": asn_info["name"],
|
|
"source": asn_info["source"],
|
|
"description": asn_info.get("description"),
|
|
"routes": {
|
|
"ipv4": filter_route(roa4, asn),
|
|
"ipv6": filter_route(roa6, asn),
|
|
},
|
|
}
|
|
for asn, asn_info in load_asn(entities)
|
|
]
|
|
|
|
current = int(time.time())
|
|
output = {
|
|
"metadata": {"generated": current, "valid": current + 14 * 86400},
|
|
"people": {
|
|
owner: {
|
|
"info": modify_entity(entity),
|
|
"asns": list(filter(lambda item: item["owner"] == owner, asn_list)),
|
|
}
|
|
for owner, entity in entities.items()
|
|
},
|
|
}
|
|
return json.dumps(output, indent=2)
|
|
|
|
|
|
def make_json(roa4, roa6):
|
|
current = int(time.time())
|
|
output = {
|
|
"metadata": {
|
|
"counts": len(roa4) + len(roa6),
|
|
"generated": current,
|
|
"valid": current + 14 * 86400,
|
|
},
|
|
"roas": [
|
|
{"asn": "AS%d" % roa["asn"], **pick(roa, ["prefix", "maxLength"])}
|
|
for roa in (*roa4, *roa6)
|
|
],
|
|
}
|
|
return json.dumps(output, indent=2)
|
|
|
|
|
|
def make_rfc8416(roa4, roa6):
|
|
output = {
|
|
"slurmVersion": 1,
|
|
"validationOutputFilters": {"prefixFilters": [], "bgpsecFilters": []},
|
|
"locallyAddedAssertions": {
|
|
"bgpsecAssertions": [],
|
|
"prefixAssertions": [
|
|
pick(
|
|
roa, ["asn", "prefix"], maxLength="maxPrefixLength", name="comment",
|
|
)
|
|
for roa in (*roa4, *roa6)
|
|
],
|
|
},
|
|
}
|
|
return json.dumps(output, indent=2)
|
|
|
|
|
|
def make_roa_records(roa4, roa6):
|
|
records = [
|
|
"route {asn} max {prefix} as {maxLength};".format_map(roa)
|
|
for roa in (*roa4, *roa6)
|
|
]
|
|
return "\n".join(["# NeoNetwork ROA tool", "", *records])
|
|
|
|
|
|
def make_summary():
|
|
entities = dict(load_entities())
|
|
asn_table = dict(load_asn(entities))
|
|
node_table = node_to_asn(set(asn_table.keys()))
|
|
stream = StringIO()
|
|
with redirect_stdout(stream):
|
|
print("Entity table:")
|
|
entity_table = tabulate(
|
|
(
|
|
(
|
|
entity["name"],
|
|
entity.get("contact", {}).get("email"),
|
|
entity.get("contact", {}).get("telegram"),
|
|
)
|
|
for entity in entities.values()
|
|
),
|
|
headers=["Name", "Email", "Telegram"],
|
|
tablefmt="presto",
|
|
)
|
|
print(entity_table)
|
|
print()
|
|
print("AS table:")
|
|
as_table = tabulate(
|
|
(
|
|
(entity["source"], "AS{}".format(asn), entity["owner"], entity["name"])
|
|
for asn, entity in sorted(asn_table.items(), key=lambda item: item[0])
|
|
),
|
|
headers=["Source", "ASN", "Owner", "Name"],
|
|
tablefmt="presto",
|
|
)
|
|
print(as_table)
|
|
print()
|
|
print("Node table:")
|
|
node_table = tabulate(
|
|
(
|
|
("AS{}".format(asn), name)
|
|
for name, asn in sorted(node_table.items(), key=lambda item: item[1])
|
|
),
|
|
headers=["ASN", "Name"],
|
|
tablefmt="presto",
|
|
)
|
|
print(node_table)
|
|
print()
|
|
print("Peer table:")
|
|
peer_table = tabulate(
|
|
(
|
|
(item.stem, downstream)
|
|
for item, entity in iter_toml_file("peer")
|
|
for downstream in entity["to-peer"]
|
|
),
|
|
headers=["Upstream", "Downstream"],
|
|
tablefmt="presto",
|
|
colalign=("right",),
|
|
)
|
|
print(peer_table)
|
|
print()
|
|
print("Route table:")
|
|
route_table = tabulate(
|
|
(
|
|
(
|
|
"AS{asn}".format_map(entity),
|
|
entity["name"],
|
|
entity["type"],
|
|
entity["prefix"] or "",
|
|
entity["supernet"] or "",
|
|
)
|
|
for entity in route_to_roa(asn_table)
|
|
),
|
|
headers=["ASN", "Name", "Type", "Prefix", "Supernet"],
|
|
tablefmt="presto",
|
|
)
|
|
print(route_table)
|
|
print()
|
|
print("Used CIDR Range:")
|
|
prefixes = netaddr.cidr_merge(
|
|
netaddr.IPNetwork(str(entity["prefix"]))
|
|
for entity in route_to_roa(asn_table)
|
|
)
|
|
for prefix in prefixes:
|
|
print(prefix)
|
|
return stream.getvalue()
|
|
|
|
|
|
def main(args):
|
|
entities = dict(load_entities())
|
|
asn_table = dict(load_asn(entities))
|
|
node_table = node_to_asn(set(asn_table.keys()))
|
|
assert_peer(set(node_table.keys()))
|
|
roa4, roa6 = prehandle_roa(asn_table, args)
|
|
if args.export:
|
|
return make_export(roa4, roa6)
|
|
elif args.json:
|
|
return make_json(roa4, roa6)
|
|
elif args.rfc8416:
|
|
return make_rfc8416(roa4, roa6)
|
|
elif args.summary:
|
|
return make_summary()
|
|
else:
|
|
return make_roa_records(roa4, roa6)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
parser = argparse.ArgumentParser(description="NeoNetwork ROA tool")
|
|
parser.add_argument(
|
|
"-m", "--max", type=int, default=29, help="set ipv4 max prefix length"
|
|
)
|
|
parser.add_argument(
|
|
"-M", "--max6", type=int, default=64, help="set ipv6 max prefix length"
|
|
)
|
|
parser.add_argument("-j", "--json", action="store_true", help="output json")
|
|
parser.add_argument("-r", "--rfc8416", action="store_true", help="output rfc8416")
|
|
parser.add_argument("-s", "--summary", action="store_true", help="output summary")
|
|
parser.add_argument("-o", "--output", default="", help="write output to file")
|
|
parser.add_argument("-4", "--ipv4", action="store_true", help="print ipv4 only")
|
|
parser.add_argument("-6", "--ipv6", action="store_true", help="print ipv6 only")
|
|
parser.add_argument(
|
|
"-e", "--export", action="store_true", help="export registry to json"
|
|
)
|
|
args = parser.parse_args()
|
|
if (
|
|
args.max < 0
|
|
or args.max6 < 0
|
|
or args.max > IPv4Network(0).max_prefixlen
|
|
or args.max6 > IPv6Network(0).max_prefixlen
|
|
):
|
|
parser.error("check your max prefix length")
|
|
output = main(args)
|
|
if not args.output or args.output == "-":
|
|
print(output)
|
|
elif output:
|
|
Path(args.output).write_text(output)
|
|
print("written to", args.output)
|