Experimental Discord bot written in Python
選択できるのは25トピックまでです。 トピックは、先頭が英数字で、英数字とダッシュ('-')を使用した35文字以内のものにしてください。

crosspostcog.py 15KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346
  1. """
  2. Cog for detecting spam messages posted in multiple channels.
  3. """
  4. import re
  5. from datetime import datetime, timedelta, timezone
  6. from typing import Optional
  7. from discord import Member, Message, utils as discordutils, TextChannel
  8. from discord.ext.commands import Cog
  9. from config import CONFIG
  10. from rocketbot.cogs.basecog import BaseCog, BotMessage, BotMessageReaction, CogSetting
  11. from rocketbot.collections import AgeBoundList, AgeBoundDict
  12. from rocketbot.storage import Storage
  13. from rocketbot.utils import str_from_timedelta
  14. class SpamContext:
  15. """
  16. Data about a set of duplicate messages from a user.
  17. """
  18. def __init__(self, member: Member) -> None:
  19. self.member: Member = member
  20. self.age: datetime = datetime.now()
  21. self.bot_message: Optional[BotMessage] = None
  22. self.is_kicked: bool = False
  23. self.is_banned: bool = False
  24. self.is_autobanned: bool = False
  25. self.spam_messages: set[Message] = set()
  26. self.deleted_messages: set[Message] = set()
  27. self.unique_channels: set[TextChannel] = set()
  28. self.duplicate_count: int = 0
  29. class CrossPostCog(BaseCog, name='Crosspost Detection'):
  30. """
  31. Detects a user posting in multiple channels in a short period
  32. of time: a common pattern for spammers.
  33. These used to be identical text, but more recent attacks have had small
  34. variations, such as different imgur URLs. It's reasonable to treat
  35. posting in many channels in a short period as suspicious on its own,
  36. regardless of whether they are identical.
  37. Repeated posts in the same channel aren't currently detected, as this can
  38. often be for a reason or due to trying a failed post when connectivity is
  39. poor. Minimum message length can be enforced for detection.
  40. """
  41. SETTING_ENABLED = CogSetting('enabled', bool,
  42. brief='crosspost detection',
  43. description='Whether crosspost detection is enabled.')
  44. SETTING_WARN_COUNT = CogSetting('warncount', int,
  45. brief='number of messages to trigger a warning',
  46. description='The number of unique channels messages are ' + \
  47. 'posted in by the same user to trigger a mod warning. The ' + \
  48. 'messages need not be identical (see dupewarncount).',
  49. usage='<count:int>',
  50. min_value=2)
  51. SETTING_DUPE_WARN_COUNT = CogSetting('dupewarncount', int,
  52. brief='number of identical messages to trigger a warning',
  53. description='The number of unique channels identical messages are ' + \
  54. 'posted in by the same user to trigger a mod warning.',
  55. usage='<count:int>',
  56. min_value=2)
  57. SETTING_BAN_COUNT = CogSetting('bancount', int,
  58. brief='number of messages to trigger a ban',
  59. description='The number of unique channels messages are ' + \
  60. 'posted in by the same user to trigger an automatic ban. The ' + \
  61. 'messages need not be identical (see dupebancount). Set ' + \
  62. 'to a large value to effectively disable, e.g. 9999.',
  63. usage='<count:int>',
  64. min_value=2)
  65. SETTING_DUPE_BAN_COUNT = CogSetting('dupebancount', int,
  66. brief='number of identical messages to trigger a ban',
  67. description='The number of unique channels identical messages are ' + \
  68. 'posted in by the same user to trigger an automatic ban. Set ' + \
  69. 'to a large value to effectively disable, e.g. 9999.',
  70. usage='<count:int>',
  71. min_value=2)
  72. SETTING_MIN_LENGTH = CogSetting('minlength', int,
  73. brief='minimum message length',
  74. description='The minimum number of characters in a message to be ' + \
  75. 'checked for duplicates. This can help ignore common short ' + \
  76. 'messages like "lol" or a single emoji.',
  77. usage='<character_count:int>',
  78. min_value=1)
  79. SETTING_TIMESPAN = CogSetting('timespan', timedelta,
  80. brief='time window to look for dupe messages',
  81. description='The number of seconds of message history to look at ' + \
  82. 'when looking for duplicates. Shorter values are preferred, ' + \
  83. 'both to detect bots and avoid excessive memory usage.',
  84. usage='<seconds:int>',
  85. min_value=timedelta(seconds=1))
  86. STATE_KEY_RECENT_MESSAGES = "CrossPostCog.recent_messages"
  87. STATE_KEY_SPAM_CONTEXT = "CrossPostCog.spam_context"
  88. def __init__(self, bot):
  89. super().__init__(
  90. bot,
  91. config_prefix='crosspost',
  92. short_description='Manages crosspost detection and handling.',
  93. )
  94. self.add_setting(CrossPostCog.SETTING_ENABLED)
  95. self.add_setting(CrossPostCog.SETTING_WARN_COUNT)
  96. self.add_setting(CrossPostCog.SETTING_DUPE_WARN_COUNT)
  97. self.add_setting(CrossPostCog.SETTING_BAN_COUNT)
  98. self.add_setting(CrossPostCog.SETTING_DUPE_BAN_COUNT)
  99. self.add_setting(CrossPostCog.SETTING_MIN_LENGTH)
  100. self.add_setting(CrossPostCog.SETTING_TIMESPAN)
  101. self.max_spam_contexts = 12
  102. async def __record_message(self, message: Message) -> None:
  103. if message.channel.permissions_for(message.author).ban_members:
  104. # User exempt from spam detection
  105. self.__trace("User exempt from crosspost checks")
  106. return
  107. def compute_message_hash(m: Message) -> int:
  108. to_hash = m.content
  109. # URLs sometimes differ per spam message, so simplify them
  110. url_regex = r'http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\(\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+'
  111. to_hash = re.sub(url_regex, '<url>', to_hash)
  112. # Add attachment metadata
  113. for attachment in m.attachments:
  114. to_hash += f'\n[[ATT: ct={attachment.content_type} s={attachment.size} w={attachment.width} h={attachment.height}]]'
  115. h = hash(to_hash)
  116. self.__trace(f"Message hash for {m.id} is {h}")
  117. return h
  118. min_length = self.get_guild_setting(message.guild, self.SETTING_MIN_LENGTH)
  119. if len(message.attachments) == 0 and len(message.content) < min_length:
  120. # Message too short to count towards spam total
  121. self.__trace(f"Message len {len(message.content)} < {min_length}")
  122. return
  123. # Get config
  124. max_age = timedelta(seconds=self.get_guild_setting(message.guild, self.SETTING_TIMESPAN))
  125. warn_count: int = self.get_guild_setting(message.guild, self.SETTING_WARN_COUNT)
  126. dupe_warn_count: int = self.get_guild_setting(message.guild, self.SETTING_DUPE_WARN_COUNT)
  127. # Record message
  128. recent_messages: AgeBoundList[Message, datetime, timedelta] = Storage.get_state_value(message.guild, self.STATE_KEY_RECENT_MESSAGES)
  129. if recent_messages is None:
  130. recent_messages = AgeBoundList(max_age, lambda index, message : message.created_at)
  131. Storage.set_state_value(message.guild, self.STATE_KEY_RECENT_MESSAGES, recent_messages)
  132. recent_messages.max_age = max_age
  133. recent_messages.append(message)
  134. self.__trace(f"Recent messages now length {len(recent_messages)}")
  135. # Get all recent messages by user
  136. member_messages = [m for m in recent_messages if m.author.id == message.author.id]
  137. message_count = len(member_messages)
  138. self.__trace(f"Found {message_count} messages for {message.author.name}")
  139. if message_count < warn_count and message_count < dupe_warn_count:
  140. self.__trace(f"Bailing because message count {message_count} < warn count {warn_count} and < dupe warn count {dupe_warn_count}")
  141. return
  142. # Look for identical(ish) messages and unique channels
  143. hash_to_channels: dict[int, set[TextChannel]] = {}
  144. unique_channels: set[TextChannel] = set()
  145. max_duplicate_count = 0
  146. for m in member_messages:
  147. message_hash = compute_message_hash(m)
  148. dupe_message_channels: set[TextChannel] = hash_to_channels.get(message_hash)
  149. if dupe_message_channels is None:
  150. dupe_message_channels = set()
  151. hash_to_channels[message_hash] = dupe_message_channels
  152. dupe_message_channels.add(m.channel)
  153. unique_channels.add(m.channel)
  154. max_duplicate_count = max(max_duplicate_count, len(dupe_message_channels))
  155. channel_count = len(unique_channels)
  156. self.__trace(f"Found {len(hash_to_channels)} unique messages, {channel_count} unique channels, {max_duplicate_count} duplicated messages")
  157. if channel_count < warn_count and max_duplicate_count < dupe_warn_count:
  158. self.__trace(f"Bailing because channels {channel_count} < warn count {warn_count} and max dupes {max_duplicate_count} < dupe warn count {dupe_warn_count}")
  159. return
  160. # This person is a problem
  161. spam_lookup: AgeBoundDict[str, SpamContext, datetime, timedelta] = Storage.get_state_value(message.guild, self.STATE_KEY_SPAM_CONTEXT)
  162. if spam_lookup is None:
  163. spam_lookup = AgeBoundDict(
  164. max_age,
  165. lambda key, context : context.age)
  166. Storage.set_state_value(message.guild, self.STATE_KEY_SPAM_CONTEXT, spam_lookup)
  167. key = f'{message.author.id}'
  168. context = spam_lookup.get(key)
  169. if context is not None and message.created_at - context.age > max_age:
  170. context = None
  171. if context is None:
  172. context = SpamContext(message.author)
  173. spam_lookup[key] = context
  174. self.log(message.guild,
  175. f'\u0007{message.author.name} ({message.author.id}) ' + \
  176. f'posted messages in {channel_count} channels.')
  177. context.age = message.created_at
  178. context.duplicate_count = max_duplicate_count
  179. context.spam_messages.update(member_messages)
  180. context.unique_channels.update(unique_channels)
  181. await self.__update_from_context(context)
  182. async def __update_from_context(self, context: SpamContext):
  183. ban_count = self.get_guild_setting(context.member.guild, self.SETTING_BAN_COUNT)
  184. dupe_ban_count = self.get_guild_setting(context.member.guild, self.SETTING_DUPE_BAN_COUNT)
  185. channel_count = len(context.unique_channels)
  186. if channel_count >= ban_count or context.duplicate_count >= dupe_ban_count:
  187. if not context.is_banned:
  188. max_age = timedelta(seconds=self.get_guild_setting(context.member.guild, self.SETTING_TIMESPAN))
  189. max_age_str = str_from_timedelta(max_age)
  190. await context.member.ban(
  191. reason=f'Rocketbot: Posted in {channel_count} channels within {max_age_str} ' + \
  192. f'({context.duplicate_count} identical). Banned by {self.bot.user.name}.',
  193. delete_message_days=1)
  194. context.is_kicked = True
  195. context.is_banned = True
  196. context.is_autobanned = True
  197. context.deleted_messages |= context.spam_messages
  198. self.__log_ban(context, self.bot.user.name)
  199. else:
  200. # Already banned. Nothing to update in the message.
  201. return
  202. await self.__update_message_from_context(context)
  203. async def __update_message_from_context(self, context: SpamContext) -> None:
  204. first_spam_message: Message = sorted(list(context.spam_messages), key=lambda m: m.created_at)[0]
  205. spam_count = len(context.spam_messages)
  206. channel_count = len(context.unique_channels)
  207. deleted_count = len(context.deleted_messages)
  208. duplicate_count = context.duplicate_count
  209. max_age = timedelta(seconds=self.get_guild_setting(context.member.guild, self.SETTING_TIMESPAN))
  210. max_age_str = str_from_timedelta(max_age)
  211. message = context.bot_message
  212. if message is None:
  213. message_type: int = BotMessage.TYPE_INFO if self.was_warned_recently(context.member) \
  214. else BotMessage.TYPE_MOD_WARNING
  215. message = BotMessage(context.member.guild, '', message_type, context)
  216. message.quote = discordutils.remove_markdown(first_spam_message.clean_content)
  217. self.record_warning(context.member)
  218. if context.is_autobanned:
  219. text = f'User {context.member.mention} auto banned for ' + \
  220. f'posting messages in {channel_count} channels within {max_age_str} ' + \
  221. f'({duplicate_count} identical). Messages from past 24 hours deleted.'
  222. await message.set_reactions([])
  223. await message.set_text(text)
  224. else:
  225. body: str = f'User {context.member.mention} posted '
  226. if duplicate_count == channel_count:
  227. body += f'identical messages in {channel_count} channels within {max_age_str} .'
  228. elif duplicate_count == 1:
  229. body += f'**different** messages in {channel_count} channels within ' + \
  230. f'{max_age_str}. (Showing first one).'
  231. else:
  232. body += f'messages in {channel_count} channels within {max_age_str} ' + \
  233. f'({duplicate_count} are identical, showing first one).'
  234. max_links = 10
  235. for msg in sorted(list(context.spam_messages), key=lambda m: m.created_at)[:max_links]:
  236. body += f'\n- {msg.jump_url}'
  237. if len(context.spam_messages) > max_links:
  238. body += f'\n- ...{len(context.spam_messages) - max_links} more...'
  239. await message.set_text(body)
  240. await message.set_reactions(BotMessageReaction.standard_set(
  241. did_delete = deleted_count >= spam_count,
  242. message_count = spam_count,
  243. did_kick = context.is_kicked,
  244. did_ban = context.is_banned))
  245. if context.bot_message is None:
  246. await self.post_message(message)
  247. context.bot_message = message
  248. async def on_mod_react(self,
  249. bot_message: BotMessage,
  250. reaction: BotMessageReaction,
  251. reacted_by: Member) -> None:
  252. context: SpamContext = bot_message.context
  253. if context is None:
  254. return
  255. channel_count = len(context.unique_channels)
  256. if reaction.emoji == CONFIG['trash_emoji']:
  257. for message in context.spam_messages - context.deleted_messages:
  258. await message.delete()
  259. context.deleted_messages.add(message)
  260. await self.__update_from_context(context)
  261. self.__log_deletion(context, reacted_by.name)
  262. elif reaction.emoji == CONFIG['kick_emoji']:
  263. await context.member.kick(
  264. reason=f'Rocketbot: Posted messages in {channel_count} ' + \
  265. f'channels. Kicked by {reacted_by.name}.')
  266. context.is_kicked = True
  267. await self.__update_from_context(context)
  268. self.__log_kick(context, reacted_by.name)
  269. elif reaction.emoji == CONFIG['ban_emoji']:
  270. await context.member.ban(
  271. reason=f'Rocketbot: Posted messages in {channel_count} ' + \
  272. f'channels. Banned by {reacted_by.name}.',
  273. delete_message_days=1)
  274. context.deleted_messages |= context.spam_messages
  275. context.is_kicked = True
  276. context.is_banned = True
  277. await self.__update_from_context(context)
  278. self.__log_ban(context, reacted_by.name)
  279. def __log_deletion(self, context: SpamContext, by_who: str) -> None:
  280. max_age = timedelta(seconds=self.get_guild_setting(context.member.guild, self.SETTING_TIMESPAN))
  281. max_age_str = str_from_timedelta(max_age)
  282. channel_count = len(context.unique_channels)
  283. duplicate_count = context.duplicate_count
  284. self.log(context.member.guild,
  285. f'{context.member.name} ({context.member.id}) posted ' + \
  286. f'messages in {channel_count} channels withint {max_age_str} ' + \
  287. f'({duplicate_count} identical). Deleted by {by_who}.')
  288. def __log_kick(self, context: SpamContext, by_who: str) -> None:
  289. max_age = timedelta(seconds=self.get_guild_setting(context.member.guild, self.SETTING_TIMESPAN))
  290. max_age_str = str_from_timedelta(max_age)
  291. channel_count = len(context.unique_channels)
  292. duplicate_count = context.duplicate_count
  293. self.log(context.member.guild,
  294. f'{context.member.name} ({context.member.id}) posted ' + \
  295. f'messages in {channel_count} channels within {max_age_str} ' + \
  296. f'({duplicate_count} identical). Kicked by {by_who}.')
  297. def __log_ban(self, context: SpamContext, by_who: str) -> None:
  298. max_age = timedelta(seconds=self.get_guild_setting(context.member.guild, self.SETTING_TIMESPAN))
  299. max_age_str = str_from_timedelta(max_age)
  300. channel_count = len(context.unique_channels)
  301. duplicate_count = context.duplicate_count
  302. self.log(context.member.guild,
  303. f'{context.member.name} ({context.member.id}) posted ' + \
  304. f'messages in {channel_count} channels within {max_age_str} ' + \
  305. f'({duplicate_count} identical). Banned by {by_who}.')
  306. def __trace(self, message):
  307. # print(message)
  308. pass
  309. @Cog.listener()
  310. async def on_message(self, message: Message):
  311. """Event handler"""
  312. if message.author is None or \
  313. message.author.bot or \
  314. message.channel is None or \
  315. message.guild is None:
  316. return
  317. if not self.get_guild_setting(message.guild, self.SETTING_ENABLED):
  318. return
  319. self.__trace("--ON MESSAGE--")
  320. await self.__record_message(message)