mirror of
https://github.com/TeamPGM/PagerMaid_Plugins_Pyro.git
synced 2024-11-24 19:27:36 +00:00
95 lines
3.2 KiB
Python
95 lines
3.2 KiB
Python
import time
|
|
import hashlib
|
|
from random import randint
|
|
|
|
from pagermaid.listener import listener
|
|
from pagermaid.enums import Message
|
|
from pagermaid.services import client as requests
|
|
from pagermaid.utils import pip_install
|
|
|
|
pip_install("tld")
|
|
|
|
from tld import get_fld
|
|
|
|
|
|
async def post_data(path, data, content, token):
|
|
url = "https://hlwicpfwc.miit.gov.cn/icpproject_query/api/"
|
|
client_ip = f"{str(randint(1, 254))}.{str(randint(1, 254))}.{str(randint(1, 254))}.{str(randint(1, 254))}"
|
|
headers = {
|
|
"Content-Type": content,
|
|
"Origin": "https://beian.miit.gov.cn/",
|
|
"Referer": "https://beian.miit.gov.cn/",
|
|
"token": token,
|
|
"Client-IP": client_ip,
|
|
"X-Forwarded-For": client_ip,
|
|
}
|
|
return await requests.post(url + path, data=data, headers=headers)
|
|
|
|
|
|
async def icp_search(domain):
|
|
md5 = hashlib.md5()
|
|
timestamp = int(time.time())
|
|
auth_key = f"testtest{timestamp}"
|
|
md5.update(auth_key.encode("utf-8"))
|
|
auth_key = md5.hexdigest()
|
|
token = await post_data(
|
|
"auth",
|
|
f"authKey={auth_key}&timeStamp={timestamp}",
|
|
"application/x-www-form-urlencoded;charset=UTF-8",
|
|
"0",
|
|
)
|
|
token = token.json()
|
|
if token.get("code", None) == 200:
|
|
token = token["params"]["bussiness"]
|
|
else:
|
|
return {"isBeian": False, "msg": "获取token失败"}
|
|
|
|
query = await post_data(
|
|
"icpAbbreviateInfo/queryByCondition",
|
|
'{"pageNum":"","pageSize":"","unitName":"%s"}' % (domain),
|
|
"application/json;charset=UTF-8",
|
|
token,
|
|
)
|
|
query = query.json()
|
|
if query.get("code", None) != 200:
|
|
return {"isBeian": False, "msg": "查询失败"}
|
|
icp_list = query["params"]["list"]
|
|
if len(icp_list) <= 0:
|
|
return {"isBeian": False, "msg": "成功"}
|
|
return {"isBeian": True, "msg": "成功", "data": icp_list[0]}
|
|
|
|
|
|
@listener(command="icp", parameters="[域名]", description="查询域名是否已备案")
|
|
async def icp_bei_an(message: Message):
|
|
url = None
|
|
if message.arguments:
|
|
url = message.arguments
|
|
elif message.reply_to_message:
|
|
url = message.reply_to_message.text
|
|
if not url:
|
|
return await message.edit("请输入或回复一个域名或链接。")
|
|
try:
|
|
url = get_fld(url, fix_protocol=True)
|
|
message: Message = await message.edit(f"查询中...{url}")
|
|
except Exception:
|
|
return await message.edit("出错了呜呜呜 ~ 无效的参数。")
|
|
try:
|
|
data = await icp_search(url)
|
|
except Exception as e:
|
|
return await message.edit(f"出错了呜呜呜 ~ 查询失败。{e}")
|
|
|
|
if data.get("isBeian", False):
|
|
data = data.get("data", {})
|
|
await message.edit(
|
|
f"域名: {url}\n"
|
|
f"主体: {data.get('unitName', '')}\n"
|
|
f"备案时间: {data.get('updateRecordTime', '')}\n"
|
|
f"备案类型: {data.get('natureName', '')}\n"
|
|
f"备案号: {data.get('serviceLicence', '')}\n"
|
|
f"是否限制访问: {data.get('limitAccess', '')}"
|
|
)
|
|
elif data.get("msg", "") == "成功":
|
|
await message.edit(f"域名 {url} 未备案!")
|
|
else:
|
|
await message.edit("出错了呜呜呜 ~ " + data.get("msg", "") + "。")
|