feat(netflow): автоматизировать настройку туннеля коллектора

deploy/netflow-agent/agent.py

@@ -0,0 +1,367 @@
+"""Local, single-interface WireGuard controller. Never executes wg-quick/hooks."""
+import base64
+import binascii
+import http.client
+import http.server
+import ipaddress
+import json
+import os
+from pathlib import Path
+import re
+import socket
+import socketserver
+import subprocess
+import sys
+import tempfile
+import threading
+
+IFACE = "wg-flow"
+CONFIG = Path(os.environ.get("FLOW_AGENT_CONFIG_DIR", "/etc/wireguard"))
+SOCKET = os.environ.get("FLOW_AGENT_SOCKET", "/run/mm-netflow-host/agent.sock")
+LOCK = threading.RLock()
+MAX_BODY = 262144
+
+
+class AgentError(Exception):
+    pass
+
+
+def run(*args, input=None):
+    try:
+        result = subprocess.run(args, input=input, text=True, capture_output=True, timeout=10)
+    except (OSError, subprocess.TimeoutExpired):
+        raise AgentError("Команда WireGuard/IP недоступна или превысила время ожидания") from None
+    if result.returncode:
+        # stderr/stdout can contain keys and must never reach an HTTP response/log.
+        raise AgentError("Операция WireGuard/IP не выполнена")
+    return result.stdout.strip()
+
+
+def key(value):
+    try:
+        if not isinstance(value, str) or len(value) != 44 or len(base64.b64decode(value, validate=True)) != 32:
+            raise ValueError()
+    except (ValueError, binascii.Error):
+        raise AgentError("Некорректный ключ WireGuard") from None
+    return value
+
+
+def port(value):
+    if type(value) is not int or not 1 <= value <= 65535:
+        raise AgentError("Некорректный UDP-порт")
+    return value
+
+
+def endpoint(value):
+    if not isinstance(value, str) or not re.fullmatch(r"(?:\[[0-9a-fA-F:]+\]|[a-zA-Z0-9._-]+):[0-9]{1,5}", value):
+        raise AgentError("Некорректный endpoint")
+    port(int(value.rsplit(":", 1)[1]))
+    return value
+
+
+def parse_conf(text):
+    interface, peers, section = {}, [], None
+    allowed = {"interface": {"PrivateKey", "Address", "ListenPort", "MTU"},
+               "peer": {"PublicKey", "PresharedKey", "AllowedIPs", "Endpoint", "PersistentKeepalive"}}
+    for raw in text.splitlines():
+        line = raw.split("#", 1)[0].strip()
+        if not line:
+            continue
+        if line == "[Interface]" and section is None:
+            section, current = "interface", interface
+        elif line == "[Peer]" and section is not None:
+            section, current = "peer", {}
+            peers.append(current)
+        else:
+            name, sep, value = line.partition("=")
+            name, value = name.strip(), value.strip()
+            if not sep or section is None or name not in allowed[section] or name in current or not value:
+                raise AgentError("Конфигурация wg-flow содержит неподдерживаемые или повторные параметры; требуется ручная проверка")
+            current[name] = value
+    if not interface or not {"PrivateKey", "Address", "ListenPort"} <= interface.keys():
+        raise AgentError("Неполная конфигурация wg-flow")
+    key(interface["PrivateKey"])
+    try:
+        ipaddress.IPv4Interface(interface["Address"])
+        port(int(interface["ListenPort"]))
+        if "MTU" in interface and not 576 <= int(interface["MTU"]) <= 9000:
+            raise ValueError()
+        for peer in peers:
+            key(peer.get("PublicKey"))
+            if "PresharedKey" in peer:
+                key(peer["PresharedKey"])
+            if "Endpoint" in peer:
+                endpoint(peer["Endpoint"])
+            if peer.get("AllowedIPs"):
+                networks(peer["AllowedIPs"].split(","))
+            if "PersistentKeepalive" in peer and not 0 <= int(peer["PersistentKeepalive"]) <= 65535:
+                raise ValueError()
+    except (ValueError, TypeError):
+        raise AgentError("Некорректные параметры wg-flow") from None
+    if len({p["PublicKey"] for p in peers}) != len(peers):
+        raise AgentError("Повторный публичный ключ пира")
+    return interface, peers
+
+
+def networks(values):
+    if not isinstance(values, list) or not values:
+        raise AgentError("AllowedIPs должны содержать адреса")
+    try:
+        return [ipaddress.ip_network(v.strip(), strict=False) for v in values]
+    except (ValueError, AttributeError):
+        raise AgentError("Некорректные AllowedIPs") from None
+
+
+def validate(raw):
+    fields = {"collectorIp", "prefix", "flowListenPort", "wgListenPort", "hostPublicKey", "hostPrivateKey", "peers"}
+    if not isinstance(raw, dict) or set(raw) != fields:
+        raise AgentError("Некорректный снимок NetFlow")
+    try:
+        collector = ipaddress.IPv4Address(raw["collectorIp"])
+        prefix = ipaddress.IPv4Network(raw["prefix"], strict=False)
+        if prefix.prefixlen != 24 or collector not in prefix or collector in (prefix.network_address, prefix.broadcast_address):
+            raise ValueError()
+    except (ValueError, TypeError):
+        raise AgentError("Требуется IPv4 overlay /24 и адрес коллектора внутри него") from None
+    port(raw["flowListenPort"])
+    port(raw["wgListenPort"])
+    key(raw["hostPublicKey"])
+    key(raw["hostPrivateKey"])
+    if run("wg", "pubkey", input=raw["hostPrivateKey"] + "\n") != raw["hostPublicKey"]:
+        raise AgentError("Публичный и приватный ключи MM не соответствуют друг другу")
+    binding = json.loads((CONFIG / "binding.json").read_text())
+    for name in ("collectorIp", "prefix", "flowListenPort", "wgListenPort"):
+        if binding.get(name) != raw[name]:
+            raise AgentError("Адрес, префикс или порт отличаются от установки агента; повторите установщик")
+    if not isinstance(raw["peers"], list) or len(raw["peers"]) > 253:
+        raise AgentError("Некорректный список пиров")
+    peers, ids, keys = [], set(), set()
+    for peer in raw["peers"]:
+        if not isinstance(peer, dict) or not {"serverId", "publicKey", "allowedIps"} <= peer.keys() or set(peer) - {"serverId", "publicKey", "allowedIps", "endpoint"}:
+            raise AgentError("Некорректный пир")
+        peer_key = key(peer["publicKey"])
+        if type(peer["serverId"]) is not int or peer["serverId"] <= 0 or peer["serverId"] in ids or peer_key in keys or peer_key == raw["hostPublicKey"]:
+            raise AgentError("Повторный или некорректный пир")
+        ids.add(peer["serverId"])
+        keys.add(peer_key)
+        nets = networks(peer["allowedIps"])
+        if len(nets) != 1 or nets[0].version != 4 or nets[0].prefixlen != 32 or nets[0].network_address not in prefix or nets[0].network_address in (collector, prefix.network_address, prefix.broadcast_address):
+            raise AgentError("Пиру разрешён только собственный IPv4 /32 внутри overlay")
+        data = {"PublicKey": peer_key, "AllowedIPs": str(nets[0])}
+        if peer.get("endpoint"):
+            data.update(Endpoint=endpoint(peer["endpoint"]), PersistentKeepalive="25")
+        peers.append(data)
+    return peers
+
+
+def merge_peers(existing, desired):
+    merged = {peer["PublicKey"]: dict(peer) for peer in existing}
+    for peer in desired:
+        public = peer["PublicKey"]
+        # Replacing a PSK would silently break a previously working tunnel.
+        if merged.get(public, {}).get("PresharedKey", "A" * 43 + "=") != "A" * 43 + "=":
+            raise AgentError("Управляемый пир использует preshared key; требуется ручная проверка")
+        merged[public] = dict(peer)
+    rows = list(merged.values())
+    wanted = {p["PublicKey"] for p in desired}
+    for i, left in enumerate(rows):
+        for right in rows[i + 1:]:
+            if left["PublicKey"] not in wanted and right["PublicKey"] not in wanted:
+                continue
+            if not left.get("AllowedIPs") or not right.get("AllowedIPs"):
+                continue
+            if any(a.version == b.version and a.overlaps(b) for a in networks(left["AllowedIPs"].split(",")) for b in networks(right["AllowedIPs"].split(","))):
+                raise AgentError("AllowedIPs управляемого пира пересекаются с другим пиром")
+    return rows
+
+
+def render(interface, peers, native=False):
+    rows = ["[Interface]"]
+    rows += [f"{k} = {v}" for k, v in interface.items() if not native or k not in ("Address", "MTU")]
+    for peer in peers:
+        rows += ["", "[Peer]"] + [f"{k} = {v}" for k, v in peer.items() if k != "AllowedIPs" or v]
+    return "\n".join(rows) + "\n"
+
+
+def atomic_write(path, text):
+    fd, name = tempfile.mkstemp(prefix=".flow-", dir=path.parent)
+    try:
+        with os.fdopen(fd, "w") as stream:
+            stream.write(text)
+            stream.flush()
+            os.fsync(stream.fileno())
+        os.replace(name, path)
+        if os.name == "posix":
+            directory = os.open(path.parent, os.O_RDONLY)
+            try:
+                os.fsync(directory)
+            finally:
+                os.close(directory)
+    finally:
+        if os.path.exists(name):
+            os.unlink(name)
+
+
+def sync(text):
+    # wg requires a file. mkstemp creates it with mode 0600.
+    fd, name = tempfile.mkstemp(prefix="flow-sync-")
+    try:
+        with os.fdopen(fd, "w") as stream:
+            stream.write(text)
+        run("wg", "syncconf", IFACE, name)
+    finally:
+        os.unlink(name)
+
+
+def live():
+    dump = run("wg", "show", IFACE, "dump").splitlines()
+    if not dump or len(dump[0].split("\t")) != 4:
+        raise AgentError("Интерфейс wg-flow недоступен")
+    interface = dump[0].split("\t")
+    peers, raw_peers = [], []
+    for line in dump[1:]:
+        cols = line.split("\t")
+        if len(cols) != 8:
+            raise AgentError("Некорректный ответ WireGuard")
+        public, psk, ep, ips, handshake, rx, tx, keepalive = cols
+        peers.append({"publicKey": public, "allowedIps": ips.split(",") if ips != "(none)" else [],
+                      "lastHandshakeAt": int(handshake) or None, "receivedBytes": int(rx), "sentBytes": int(tx)})
+        item = {"PublicKey": public, "AllowedIPs": ips if ips != "(none)" else ""}
+        if psk != "(none)":
+            item["PresharedKey"] = psk
+        if ep != "(none)":
+            item["Endpoint"] = ep
+        item["PersistentKeepalive"] = keepalive
+        raw_peers.append(item)
+    addresses = json.loads(run("ip", "-j", "-4", "address", "show", "dev", IFACE))
+    cidrs = [f"{a['local']}/{a['prefixlen']}" for row in addresses for a in row.get("addr_info", []) if a.get("family") == "inet"]
+    return {"interfaceUp": any("UP" in row.get("flags", []) for row in addresses), "publicKey": interface[1], "listenPort": int(interface[2]), "addresses": cidrs, "peers": peers}, raw_peers
+
+
+def check(raw, apply=False):
+    desired = validate(raw)
+    path = CONFIG / "wg-flow.conf"
+    previous = path.read_text()
+    interface, stored = parse_conf(previous)
+    expected_address = f"{raw['collectorIp']}/{raw['prefix'].split('/')[1]}"
+    if run("wg", "pubkey", input=interface["PrivateKey"] + "\n") != raw["hostPublicKey"] or interface["Address"] != expected_address or int(interface["ListenPort"]) != raw["wgListenPort"]:
+        raise AgentError("Ключ, адрес или порт существующего wg-flow отличаются от MM; требуется ручная проверка")
+    status, runtime = live()
+    if not status["interfaceUp"]:
+        raise AgentError("Интерфейс wg-flow выключен; проверьте wg-quick на хосте")
+    if status["publicKey"] != raw["hostPublicKey"] or expected_address not in status["addresses"] or status["listenPort"] != raw["wgListenPort"]:
+        raise AgentError("Работающий wg-flow отличается от конфигурации; требуется ручная проверка")
+    # Preserve both persistent and runtime-only third-party peers. Detect conflicts before writing.
+    combined = {p["PublicKey"]: p for p in runtime}
+    combined.update({p["PublicKey"]: p for p in stored})
+    merged = merge_peers(list(combined.values()), desired)
+    runtime_merged = merge_peers(runtime, desired)
+    # Do not reset a third-party peer's runtime settings to an older persistent file.
+    # Preserve roaming endpoints of managed peers unless the persisted endpoint was edited.
+    stored_by_key = {p["PublicKey"]: p for p in stored}
+    runtime_by_key = {p["PublicKey"]: p for p in runtime}
+    desired_by_key = {p["PublicKey"]: p for p in desired}
+    for peer in runtime_merged:
+        public = peer["PublicKey"]
+        if public in desired_by_key and stored_by_key.get(public, {}).get("Endpoint") == desired_by_key[public].get("Endpoint"):
+            if runtime_by_key.get(public, {}).get("Endpoint"):
+                peer["Endpoint"] = runtime_by_key[public]["Endpoint"]
+    def matches(actual, persistent=False):
+        by_key = {p["PublicKey"]: p for p in actual}
+        return all(p["PublicKey"] in by_key and all(by_key[p["PublicKey"]].get(k) == v for k, v in p.items() if persistent or k != "Endpoint") for p in desired)
+    matched = matches(stored, persistent=True) and matches(runtime)
+    if apply:
+        rollback = run("wg", "showconf", IFACE)
+        try:
+            atomic_write(path, render(interface, merged))
+            # Keep the running interface header verbatim (including a possible fwmark).
+            header = rollback.split("[Peer]", 1)[0].rstrip()
+            peer_config = render({}, runtime_merged, native=True).removeprefix("[Interface]\n")
+            sync(header + "\n" + peer_config)
+            status, after = live()
+            if not matches(after):
+                raise AgentError("Применение пиров не подтверждено")
+        except Exception:
+            failures = []
+            for action in (lambda: atomic_write(path, previous), lambda: sync(rollback)):
+                try:
+                    action()
+                except Exception:
+                    failures.append(True)
+            raise AgentError("Применение не удалось; " + ("восстановление требует ручной проверки" if failures else "предыдущая конфигурация восстановлена")) from None
+        matched = True
+    return {**status, "configMatches": matched}
+
+
+class Handler(http.server.BaseHTTPRequestHandler):
+    def log_message(self, *_):
+        pass
+
+    def handle_request(self):
+        try:
+            with LOCK:
+                if self.command == "GET" and self.path == "/status":
+                    result, _ = live()
+                    result["configMatches"] = None
+                elif self.command == "POST" and self.path in ("/check", "/apply"):
+                    size = int(self.headers.get("Content-Length", "0"))
+                    if not 0 < size <= MAX_BODY or self.headers.get("Transfer-Encoding"):
+                        raise AgentError("Некорректный размер запроса")
+                    raw = json.loads(self.rfile.read(size))
+                    result = check(raw, apply=self.path == "/apply")
+                else:
+                    self.respond(404, {"error": "Операция не поддерживается"})
+                    return
+            self.respond(200, result)
+        except AgentError as error:
+            self.respond(409, {"error": str(error)})
+        except Exception:
+            self.respond(503, {"error": "Агент NetFlow не готов; проверьте конфигурацию и интерфейс хоста"})
+
+    def setup(self):
+        super().setup()
+        self.connection.settimeout(15)
+
+    def respond(self, code, body):
+        data = json.dumps(body, ensure_ascii=False).encode()
+        self.send_response(code)
+        self.send_header("Content-Type", "application/json; charset=utf-8")
+        self.send_header("Content-Length", str(len(data)))
+        self.end_headers()
+        self.wfile.write(data)
+
+    do_GET = handle_request
+    do_POST = handle_request
+
+
+class Server(socketserver.ThreadingMixIn, getattr(socketserver, "UnixStreamServer", socketserver.TCPServer)):
+    daemon_threads = True
+
+
+class UnixConnection(http.client.HTTPConnection):
+    def connect(self):
+        self.sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
+        self.sock.settimeout(3)
+        self.sock.connect(SOCKET)
+
+
+def main():
+    if not hasattr(socketserver, "UnixStreamServer"):
+        raise SystemExit("NetFlow agent requires Linux Unix sockets")
+    if "--healthcheck" in sys.argv:
+        client = UnixConnection("localhost")
+        client.request("GET", "/status")
+        response = client.getresponse()
+        sys.exit(0 if response.status == 200 and json.loads(response.read()).get("interfaceUp") else 1)
+    os.umask(0o077)
+    Path(SOCKET).parent.mkdir(parents=True, exist_ok=True)
+    if os.path.lexists(SOCKET):
+        Path(SOCKET).unlink()
+    with Server(SOCKET, Handler) as server:
+        os.chmod(SOCKET, 0o660)
+        print("NetFlow agent: Unix API ready", flush=True)
+        server.serve_forever()
+
+
+if __name__ == "__main__":
+    main()