mirror of
https://github.com/MarshalX/telegram-crawler.git
synced 2024-12-12 09:51:06 +01:00
436 lines
16 KiB
Python
436 lines
16 KiB
Python
import asyncio
|
|
import logging
|
|
import os
|
|
import platform
|
|
import re
|
|
import shutil
|
|
import sys
|
|
import zipfile
|
|
import hashlib
|
|
from asyncio.exceptions import TimeoutError
|
|
from string import punctuation, whitespace
|
|
from time import time
|
|
from typing import List
|
|
|
|
import aiofiles
|
|
import aiohttp
|
|
from aiohttp import ClientConnectorError, ServerDisconnectedError
|
|
|
|
PROTOCOL = 'https://'
|
|
ILLEGAL_PATH_CHARS = punctuation.replace('.', '') + whitespace
|
|
|
|
DYNAMIC_PART_MOCK = 'telegram-crawler'
|
|
|
|
INPUT_FILENAME = os.environ.get('INPUT_FILENAME', 'tracked_links.txt')
|
|
OUTPUT_FOLDER = os.environ.get('OUTPUT_FOLDER', 'data/')
|
|
|
|
TRANSLATIONS_EN_CATEGORY_URL_REGEX = r'/en/[a-z_]+/[a-z_]+/$'
|
|
|
|
PAGE_GENERATION_TIME_REGEX = r'<!-- page generated in .+ -->'
|
|
PAGE_API_HASH_REGEX = r'\?hash=[a-z0-9]+'
|
|
PAGE_API_HASH_TEMPLATE = f'?hash={DYNAMIC_PART_MOCK}'
|
|
PASSPORT_SSID_REGEX = r'passport_ssid=[a-z0-9]+_[a-z0-9]+_[a-z0-9]+'
|
|
PASSPORT_SSID_TEMPLATE = f'passport_ssid={DYNAMIC_PART_MOCK}'
|
|
NONCE_REGEX = r'"nonce":"[a-z0-9]+_[a-z0-9]+_[a-z0-9]+'
|
|
NONCE_TEMPLATE = f'"nonce":"{DYNAMIC_PART_MOCK}'
|
|
PROXY_CONFIG_SUB_NET_REGEX = r'\d+\.\d+:8888;'
|
|
PROXY_CONFIG_SUB_NET_TEMPLATE = 'X.X:8888;'
|
|
TRANSLATE_SUGGESTION_REGEX = r'<div class="tr-value-suggestion">(.?)+</div>'
|
|
SPARKLE_SIG_REGEX = r';sig=(.*?);'
|
|
SPARKLE_SE_REGEX = r';se=(.*?);'
|
|
SPARKLE_SIG_TEMPLATE = f';sig={DYNAMIC_PART_MOCK};'
|
|
SPARKLE_SE_TEMPLATE = f';se={DYNAMIC_PART_MOCK};'
|
|
|
|
# unsecure but so simple
|
|
CONNECTOR = aiohttp.TCPConnector(ssl=False)
|
|
TIMEOUT = aiohttp.ClientTimeout(total=10)
|
|
|
|
logging.basicConfig(format='%(message)s', level=logging.INFO)
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
def get_hash(data: bytes) -> str:
|
|
return hashlib.sha256(data).hexdigest()
|
|
|
|
|
|
async def download_file(url, path, session):
|
|
async with session.get(url) as response:
|
|
if response.status != 200:
|
|
return
|
|
|
|
content = await response.read()
|
|
|
|
async with aiofiles.open(path, mode='wb') as f:
|
|
await f.write(content)
|
|
|
|
|
|
async def get_download_link_of_latest_appcenter_release(parameterized_url: str, session: aiohttp.ClientSession):
|
|
api_base = 'https://install.appcenter.ms/api/v0.1'
|
|
base_url = f'{api_base}/{parameterized_url}'
|
|
|
|
async def make_req(url):
|
|
async with session.get(url) as response:
|
|
if response.status != 200:
|
|
return
|
|
|
|
return await response.json(encoding='UTF-8')
|
|
|
|
json = await make_req(f'{base_url}/public_releases')
|
|
if json and json[0]:
|
|
latest_id = json[0]['id']
|
|
else:
|
|
return
|
|
|
|
json = await make_req(f'{base_url}/releases/{latest_id}')
|
|
if json:
|
|
return json['download_url']
|
|
|
|
return None
|
|
|
|
|
|
async def track_additional_files(
|
|
files_to_track: List[str], input_dir_name: str, output_dir_name: str, encoding='utf-8', save_hash_only=False
|
|
):
|
|
kwargs = {'mode': 'r', 'encoding': encoding}
|
|
if save_hash_only:
|
|
kwargs['mode'] = 'rb'
|
|
del kwargs['encoding']
|
|
|
|
for file in files_to_track:
|
|
async with aiofiles.open(os.path.join(input_dir_name, file), **kwargs) as r_file:
|
|
content = await r_file.read()
|
|
|
|
if save_hash_only:
|
|
content = get_hash(content)
|
|
else:
|
|
content = re.sub(r'id=".*"', 'id="tgcrawl"', content)
|
|
content = re.sub(r'name="APKTOOL_DUMMY_.*" id', 'name="tgcrawl" id', content)
|
|
|
|
filename = os.path.join(OUTPUT_FOLDER, output_dir_name, file)
|
|
os.makedirs(os.path.dirname(filename), exist_ok=True)
|
|
async with aiofiles.open(filename, 'w', encoding='utf-8') as w_file:
|
|
await w_file.write(content)
|
|
|
|
|
|
async def download_telegram_macos_beta_and_extract_resources(session: aiohttp.ClientSession):
|
|
parameterized_url = 'apps/keepcoder/telegram-swift/distribution_groups/public'
|
|
download_url = await get_download_link_of_latest_appcenter_release(parameterized_url, session)
|
|
|
|
if not download_url:
|
|
return
|
|
|
|
crawled_data_folder = 'telegram-beta-macos'
|
|
client_folder_name = 'macos'
|
|
client_archive_name = 'macos.zip'
|
|
|
|
assets_output_dir = 'macos_assets'
|
|
assets_filename = 'Assets.car'
|
|
assets_extractor = 'acextract'
|
|
tool_archive_name = f'{assets_extractor}.zip'
|
|
|
|
tool_download_url = 'https://github.com/bartoszj/acextract/releases/download/2.2/acextract.zip'
|
|
|
|
if 'darwin' not in platform.system().lower():
|
|
await download_file(download_url, client_archive_name, session)
|
|
else:
|
|
await asyncio.gather(
|
|
download_file(download_url, client_archive_name, session),
|
|
download_file(tool_download_url, tool_archive_name, session),
|
|
)
|
|
|
|
# synced
|
|
with zipfile.ZipFile(client_archive_name, 'r') as f:
|
|
f.extractall(client_folder_name)
|
|
|
|
resources_path = 'Telegram.app/Contents/Resources'
|
|
files_to_track = [
|
|
f'{resources_path}/en.lproj/Localizable.strings',
|
|
]
|
|
await track_additional_files(files_to_track, client_folder_name, crawled_data_folder, 'utf-16')
|
|
|
|
_, _, hash_of_files_to_track = next(os.walk(f'{client_folder_name}/{resources_path}'))
|
|
hash_of_files_to_track = [f'{resources_path}/{i}' for i in hash_of_files_to_track]
|
|
await track_additional_files(hash_of_files_to_track, client_folder_name, crawled_data_folder, save_hash_only=True)
|
|
|
|
def cleanup1():
|
|
os.path.isdir(client_folder_name) and shutil.rmtree(client_folder_name)
|
|
os.remove(client_archive_name)
|
|
|
|
# .car crawler works only in macOS
|
|
if 'darwin' not in platform.system().lower():
|
|
cleanup1()
|
|
return
|
|
|
|
# synced
|
|
with zipfile.ZipFile(tool_archive_name, 'r') as f:
|
|
f.extractall(assets_extractor)
|
|
|
|
path_to_car = os.path.join(client_folder_name, resources_path, assets_filename)
|
|
path_to_extractor = os.path.join(assets_extractor, assets_extractor)
|
|
await (await asyncio.create_subprocess_exec('chmod', '+x', path_to_extractor)).communicate()
|
|
process = await asyncio.create_subprocess_exec(path_to_extractor, '-i', path_to_car, '-o', assets_output_dir)
|
|
await process.communicate()
|
|
|
|
def cleanup2():
|
|
cleanup1()
|
|
os.path.isdir(assets_output_dir) and shutil.rmtree(assets_output_dir)
|
|
os.path.isdir(assets_extractor) and shutil.rmtree(assets_extractor)
|
|
os.remove(tool_archive_name)
|
|
|
|
if process.returncode != 0:
|
|
cleanup2()
|
|
return
|
|
|
|
_, _, hash_of_files_to_track = next(os.walk(assets_output_dir))
|
|
await track_additional_files(
|
|
hash_of_files_to_track,
|
|
assets_output_dir,
|
|
os.path.join(crawled_data_folder, assets_filename),
|
|
save_hash_only=True
|
|
)
|
|
|
|
cleanup2()
|
|
|
|
|
|
async def download_telegram_android_beta_and_extract_resources(session: aiohttp.ClientSession):
|
|
parameterized_url = 'apps/drklo-2kb-ghpo/telegram-beta-2/distribution_groups/all-users-of-telegram-beta-2'
|
|
download_url = await get_download_link_of_latest_appcenter_release(parameterized_url, session)
|
|
|
|
if not download_url:
|
|
return
|
|
|
|
await asyncio.gather(
|
|
download_file('https://bitbucket.org/iBotPeaches/apktool/downloads/apktool_2.6.1.jar', 'tool.apk', session),
|
|
download_file(download_url, 'android.apk', session),
|
|
)
|
|
|
|
def cleanup():
|
|
os.path.isdir('android') and shutil.rmtree('android')
|
|
os.remove('tool.apk')
|
|
os.remove('android.apk')
|
|
|
|
process = await asyncio.create_subprocess_exec('java', '-jar', 'tool.apk', 'd', '-s', '-f', 'android.apk')
|
|
await process.communicate()
|
|
|
|
if process.returncode != 0:
|
|
cleanup()
|
|
return
|
|
|
|
files_to_track = [
|
|
'res/values/strings.xml',
|
|
'res/values/public.xml'
|
|
]
|
|
await track_additional_files(files_to_track, 'android', 'telegram-beta-android')
|
|
|
|
cleanup()
|
|
|
|
|
|
async def collect_translations_paginated_content(url: str, session: aiohttp.ClientSession) -> str:
|
|
headers = {'X-Requested-With': 'XMLHttpRequest'}
|
|
content = list()
|
|
|
|
async def _get_page(offset: int):
|
|
logger.info(f'Url: {url}, offset: {offset}')
|
|
data = {'offset': offset, 'more': 1}
|
|
|
|
try:
|
|
new_offset = None
|
|
async with session.post(
|
|
f'{PROTOCOL}{url}', data=data, headers=headers, allow_redirects=False, timeout=TIMEOUT
|
|
) as response:
|
|
if response.status != 200:
|
|
logger.debug(f'Resend cuz {response.status}')
|
|
new_offset = offset
|
|
else:
|
|
json = await response.json(encoding='UTF-8')
|
|
if 'more_html' in json and json['more_html']:
|
|
content.append(json['more_html'])
|
|
new_offset = offset + 200
|
|
|
|
new_offset and await _get_page(new_offset)
|
|
except (TimeoutError, ClientConnectorError):
|
|
logger.warning(f'Client or timeout error. Retrying {url}; offset {offset}')
|
|
await _get_page(offset)
|
|
|
|
await _get_page(0)
|
|
|
|
return '\n'.join(content)
|
|
|
|
|
|
async def track_mtproto_configs():
|
|
import json
|
|
from pyrogram import Client
|
|
from pyrogram.raw import functions
|
|
from pyrogram.raw.types import InputStickerSetAnimatedEmoji
|
|
|
|
app = Client(
|
|
os.environ['TELEGRAM_SESSION'],
|
|
api_id=int(os.environ['TELEGRAM_API_ID']),
|
|
api_hash=os.environ['TELEGRAM_API_HASH'],
|
|
)
|
|
await app.start()
|
|
|
|
configs = {
|
|
'GetConfig': await app.send(functions.help.GetConfig()),
|
|
'GetCdnConfig': await app.send(functions.help.GetCdnConfig()),
|
|
# 'GetInviteText': await app.send(functions.help.GetInviteText()),
|
|
# 'GetSupport': await app.send(functions.help.GetSupport()),
|
|
# 'GetSupportName': await app.send(functions.help.GetSupportName()),
|
|
# 'GetPassportConfig': await app.send(functions.help.GetPassportConfig(hash=0)),
|
|
'GetCountriesList': await app.send(functions.help.GetCountriesList(lang_code='en', hash=0)),
|
|
'GetAppConfig': await app.send(functions.help.GetAppConfig()),
|
|
# 'GetAppUpdate': await app.send(functions.help.GetAppUpdate(source='')),
|
|
'AnimatedEmoji': await app.send(
|
|
functions.messages.GetStickerSet(stickerset=InputStickerSetAnimatedEmoji(), hash=0)
|
|
),
|
|
}
|
|
|
|
keys_to_hide = {'access_hash', 'autologin_token', 'file_reference', 'file_reference_base64'}
|
|
|
|
def rem_rec(config):
|
|
if not isinstance(config, dict):
|
|
return
|
|
|
|
for key, value in config.items():
|
|
if isinstance(value, dict):
|
|
rem_rec(value)
|
|
elif isinstance(value, list):
|
|
for item in value:
|
|
rem_rec(item)
|
|
elif key == 'key' and value in keys_to_hide:
|
|
config['value']['value'] = 'crawler'
|
|
elif key in keys_to_hide:
|
|
config[key] = 'crawler'
|
|
|
|
for config_name in {'GetAppConfig', 'AnimatedEmoji'}:
|
|
configs[config_name] = json.loads(str(configs[config_name]))
|
|
rem_rec(configs[config_name])
|
|
configs[config_name] = json.dumps(configs[config_name], indent=4)
|
|
|
|
configs['GetConfig'].date = 0
|
|
configs['GetConfig'].expires = 0
|
|
configs['GetConfig'].dc_options = []
|
|
|
|
output_dir_name = 'telegram-mtproto'
|
|
for file, content in configs.items():
|
|
filename = os.path.join(OUTPUT_FOLDER, output_dir_name, f'{file}.json')
|
|
os.makedirs(os.path.dirname(filename), exist_ok=True)
|
|
async with aiofiles.open(filename, 'w', encoding='utf-8') as w_file:
|
|
await w_file.write(str(content))
|
|
|
|
await app.stop()
|
|
|
|
|
|
def is_hashable_only_content_type(content_type) -> bool:
|
|
hashable_only_content_types = (
|
|
'png',
|
|
'jpeg',
|
|
'x-icon',
|
|
'gif',
|
|
'mp4',
|
|
'webm',
|
|
'application/zip',
|
|
)
|
|
|
|
for hashable_only_content_type in hashable_only_content_types:
|
|
if hashable_only_content_type in content_type:
|
|
return True
|
|
|
|
return False
|
|
|
|
|
|
class RetryError(Exception):
|
|
...
|
|
|
|
|
|
async def crawl(url: str, session: aiohttp.ClientSession):
|
|
try:
|
|
# f*ck this shit. I believe it's temp solution
|
|
if 'css/telegram.css' in url:
|
|
return
|
|
|
|
logger.info(f'Process {url}')
|
|
async with session.get(f'{PROTOCOL}{url}', allow_redirects=False, timeout=TIMEOUT) as response:
|
|
if response.status // 100 == 5:
|
|
msg = f'Error 5XX. Retrying {url}'
|
|
logger.warning(msg)
|
|
raise RetryError(msg)
|
|
|
|
if response.status not in {200, 304}:
|
|
if response.status != 302:
|
|
content = await response.text()
|
|
logger.debug(f'Skip {url} because status code == {response.status}. Content: {content}')
|
|
return
|
|
|
|
# bypass external slashes and so on
|
|
url_parts = [p for p in url.split('/') if p not in ILLEGAL_PATH_CHARS]
|
|
|
|
is_hashable_only = is_hashable_only_content_type(response.content_type)
|
|
# amazing dirt for media files like
|
|
# telegram.org/file/811140591/1/q7zZHjgES6s/9d121a89ffb0015837
|
|
# with response content type HTML instead of image. Shame on you
|
|
# sometimes it returns correct type. noice load balancing
|
|
is_sucking_file = '/file/' in url and 'text' in response.content_type
|
|
|
|
# handle pure domains and html pages without ext in url
|
|
ext = '.html' if '.' not in url_parts[-1] or len(url_parts) == 1 else ''
|
|
|
|
# I don't add ext by content type for images and so on cuz TG servers sucks.
|
|
# Some servers do not return correct content type. Some servers do...
|
|
if is_hashable_only or is_sucking_file:
|
|
ext = ''
|
|
|
|
filename = OUTPUT_FOLDER + '/'.join(url_parts) + ext
|
|
os.makedirs(os.path.dirname(filename), exist_ok=True)
|
|
|
|
if is_sucking_file or is_hashable_only:
|
|
content = await response.read()
|
|
async with aiofiles.open(filename, 'w', encoding='utf-8') as f:
|
|
await f.write(get_hash(content))
|
|
return
|
|
|
|
content = await response.text(encoding='UTF-8')
|
|
if re.search(TRANSLATIONS_EN_CATEGORY_URL_REGEX, url):
|
|
content = await collect_translations_paginated_content(url, session)
|
|
|
|
content = re.sub(PAGE_GENERATION_TIME_REGEX, '', content)
|
|
content = re.sub(PAGE_API_HASH_REGEX, PAGE_API_HASH_TEMPLATE, content)
|
|
content = re.sub(PASSPORT_SSID_REGEX, PASSPORT_SSID_TEMPLATE, content)
|
|
content = re.sub(NONCE_REGEX, NONCE_TEMPLATE, content)
|
|
content = re.sub(PROXY_CONFIG_SUB_NET_REGEX, PROXY_CONFIG_SUB_NET_TEMPLATE, content)
|
|
content = re.sub(TRANSLATE_SUGGESTION_REGEX, '', content)
|
|
content = re.sub(SPARKLE_SIG_REGEX, SPARKLE_SIG_TEMPLATE, content)
|
|
content = re.sub(SPARKLE_SE_REGEX, SPARKLE_SE_TEMPLATE, content)
|
|
|
|
async with aiofiles.open(filename, 'w', encoding='utf-8') as f:
|
|
logger.info(f'Write to {filename}')
|
|
await f.write(content)
|
|
except (ServerDisconnectedError, TimeoutError, ClientConnectorError):
|
|
logger.warning(f'Client or timeout error. Retrying {url}')
|
|
await crawl(url, session)
|
|
|
|
|
|
async def start(url_list: set[str], mode: int):
|
|
async with aiohttp.ClientSession(connector=CONNECTOR) as session:
|
|
mode == 0 and await asyncio.gather(
|
|
*[crawl(url, session) for url in url_list],
|
|
download_telegram_android_beta_and_extract_resources(session),
|
|
download_telegram_macos_beta_and_extract_resources(session),
|
|
track_mtproto_configs(),
|
|
)
|
|
mode == 1 and await asyncio.gather(*[crawl(url, session) for url in url_list])
|
|
mode == 2 and await download_telegram_android_beta_and_extract_resources(session)
|
|
mode == 3 and await download_telegram_macos_beta_and_extract_resources(session)
|
|
mode == 4 and await track_mtproto_configs()
|
|
|
|
|
|
if __name__ == '__main__':
|
|
run_mode = int(sys.argv[1]) if len(sys.argv) > 1 else 0
|
|
|
|
with open(INPUT_FILENAME, 'r') as f:
|
|
tracked_urls = set([l.replace('\n', '') for l in f.readlines()])
|
|
|
|
start_time = time()
|
|
logger.info(f'Start crawling content of {len(tracked_urls)} tracked urls...')
|
|
asyncio.get_event_loop().run_until_complete(start(tracked_urls, run_mode))
|
|
logger.info(f'Stop crawling content in mode {run_mode}. {time() - start_time} sec.')
|