diff --git a/Gemfile b/Gemfile index 613515628..15334678b 100644 --- a/Gemfile +++ b/Gemfile @@ -58,6 +58,7 @@ gem 'idn-ruby', require: 'idn' gem 'kaminari', '~> 1.1' gem 'link_header', '~> 0.0' gem 'mime-types', '~> 3.2', require: 'mime/types/columnar' +gem 'nilsimsa', git: 'https://github.com/witgo/nilsimsa', ref: 'fd184883048b922b176939f851338d0a4971a532' gem 'nokogiri', '~> 1.10' gem 'nsa', '~> 0.2' gem 'oj', '~> 3.7' diff --git a/Gemfile.lock b/Gemfile.lock index 340bbcdd8..c3198b7d9 100644 --- a/Gemfile.lock +++ b/Gemfile.lock @@ -12,6 +12,13 @@ GIT specs: http_parser.rb (0.6.1) +GIT + remote: https://github.com/witgo/nilsimsa + revision: fd184883048b922b176939f851338d0a4971a532 + ref: fd184883048b922b176939f851338d0a4971a532 + specs: + nilsimsa (1.1.2) + GEM remote: https://rubygems.org/ specs: @@ -704,6 +711,7 @@ DEPENDENCIES microformats (~> 4.1) mime-types (~> 3.2) net-ldap (~> 0.10) + nilsimsa! nokogiri (~> 1.10) nsa (~> 0.2) oj (~> 3.7) diff --git a/app/lib/activitypub/activity/create.rb b/app/lib/activitypub/activity/create.rb index 5849c20d7..56c24680a 100644 --- a/app/lib/activitypub/activity/create.rb +++ b/app/lib/activitypub/activity/create.rb @@ -41,6 +41,7 @@ class ActivityPub::Activity::Create < ActivityPub::Activity resolve_thread(@status) fetch_replies(@status) + check_for_spam distribute(@status) forward_for_reply if @status.distributable? end @@ -406,6 +407,18 @@ class ActivityPub::Activity::Create < ActivityPub::Activity Account.local.where(username: local_usernames).exists? end + def check_for_spam + spam_check = SpamCheck.new(@status) + + return if spam_check.skip? + + if spam_check.spam? + spam_check.flag! + else + spam_check.remember! + end + end + def forward_for_reply return unless @json['signature'].present? && reply_to_local? ActivityPub::RawDistributionWorker.perform_async(Oj.dump(@json), replied_to_status.account_id, [@account.preferred_inbox_url]) diff --git a/app/lib/spam_check.rb b/app/lib/spam_check.rb new file mode 100644 index 000000000..923d48a02 --- /dev/null +++ b/app/lib/spam_check.rb @@ -0,0 +1,169 @@ +# frozen_string_literal: true + +class SpamCheck + include Redisable + include ActionView::Helpers::TextHelper + + NILSIMSA_COMPARE_THRESHOLD = 95 + NILSIMSA_MIN_SIZE = 10 + EXPIRE_SET_AFTER = 1.week.seconds + + def initialize(status) + @account = status.account + @status = status + end + + def skip? + already_flagged? || trusted? || no_unsolicited_mentions? || solicited_reply? + end + + def spam? + if insufficient_data? + false + elsif nilsimsa? + any_other_digest?('nilsimsa') { |_, other_digest| nilsimsa_compare_value(digest, other_digest) >= NILSIMSA_COMPARE_THRESHOLD } + else + any_other_digest?('md5') { |_, other_digest| other_digest == digest } + end + end + + def flag! + auto_silence_account! + auto_report_status! + end + + def remember! + # The scores in sorted sets don't actually have enough bits to hold an exact + # value of our snowflake IDs, so we use it only for its ordering property. To + # get the correct status ID back, we have to save it in the string value + + redis.zadd(redis_key, @status.id, digest_with_algorithm) + redis.zremrangebyrank(redis_key, '0', '-10') + redis.expire(redis_key, EXPIRE_SET_AFTER) + end + + def reset! + redis.del(redis_key) + end + + def hashable_text + return @hashable_text if defined?(@hashable_text) + + @hashable_text = @status.text + @hashable_text = remove_mentions(@hashable_text) + @hashable_text = strip_tags(@hashable_text) unless @status.local? + @hashable_text = normalize_unicode(@status.spoiler_text + ' ' + @hashable_text) + @hashable_text = remove_whitespace(@hashable_text) + end + + def insufficient_data? + hashable_text.blank? + end + + def digest + @digest ||= begin + if nilsimsa? + Nilsimsa.new(hashable_text).hexdigest + else + Digest::MD5.hexdigest(hashable_text) + end + end + end + + def digest_with_algorithm + if nilsimsa? + ['nilsimsa', digest, @status.id].join(':') + else + ['md5', digest, @status.id].join(':') + end + end + + private + + def remove_mentions(text) + return text.gsub(Account::MENTION_RE, '') if @status.local? + + Nokogiri::HTML.fragment(text).tap do |html| + mentions = @status.mentions.map { |mention| ActivityPub::TagManager.instance.url_for(mention.account) } + + html.traverse do |element| + element.unlink if element.name == 'a' && mentions.include?(element['href']) + end + end.to_s + end + + def normalize_unicode(text) + text.unicode_normalize(:nfkc).downcase + end + + def remove_whitespace(text) + text.gsub(/\s+/, ' ').strip + end + + def auto_silence_account! + @account.silence! + end + + def auto_report_status! + status_ids = Status.where(visibility: %i(public unlisted)).where(id: matching_status_ids).pluck(:id) + [@status.id] if @status.distributable? + ReportService.new.call(Account.representative, @account, status_ids: status_ids, comment: I18n.t('spam_check.spam_detected_and_silenced')) + end + + def already_flagged? + @account.silenced? + end + + def trusted? + @account.trust_level > Account::TRUST_LEVELS[:untrusted] + end + + def no_unsolicited_mentions? + @status.mentions.all? { |mention| mention.silent? || (!@account.local? && !mention.account.local?) || mention.account.following?(@account) } + end + + def solicited_reply? + !@status.thread.nil? && @status.thread.mentions.where(account: @account).exists? + end + + def nilsimsa_compare_value(first, second) + first = [first].pack('H*') + second = [second].pack('H*') + bits = 0 + + 0.upto(31) do |i| + bits += Nilsimsa::POPC[255 & (first[i].ord ^ second[i].ord)].ord + end + + 128 - bits # -128 <= Nilsimsa Compare Value <= 128 + end + + def nilsimsa? + hashable_text.size > NILSIMSA_MIN_SIZE + end + + def other_digests + redis.zrange(redis_key, 0, -1) + end + + def any_other_digest?(filter_algorithm) + other_digests.any? do |record| + algorithm, other_digest, status_id = record.split(':') + + next unless algorithm == filter_algorithm + + yield algorithm, other_digest, status_id + end + end + + def matching_status_ids + if nilsimsa? + other_digests.select { |record| record.start_with?('nilsimsa') && nilsimsa_compare_value(digest, record.split(':')[1]) >= NILSIMSA_COMPARE_THRESHOLD }.map { |record| record.split(':')[2] }.compact + else + other_digests.select { |record| record.start_with?('md5') && record.split(':')[1] == digest }.map { |record| record.split(':')[2] }.compact + end + end + + def redis_key + @redis_key ||= "spam_check:#{@account.id}" + end +end diff --git a/app/models/account.rb b/app/models/account.rb index d6772eb98..a22b7fd7c 100644 --- a/app/models/account.rb +++ b/app/models/account.rb @@ -45,6 +45,7 @@ # also_known_as :string is an Array # silenced_at :datetime # suspended_at :datetime +# trust_level :integer # class Account < ApplicationRecord @@ -62,6 +63,11 @@ class Account < ApplicationRecord include AccountCounters include DomainNormalizable + TRUST_LEVELS = { + untrusted: 0, + trusted: 1, + }.freeze + enum protocol: [:ostatus, :activitypub] validates :username, presence: true @@ -163,6 +169,10 @@ class Account < ApplicationRecord last_webfingered_at.nil? || last_webfingered_at <= 1.day.ago end + def trust_level + self[:trust_level] || 0 + end + def refresh! ResolveAccountService.new.call(acct) unless local? end @@ -171,21 +181,19 @@ class Account < ApplicationRecord silenced_at.present? end - def silence!(date = nil) - date ||= Time.now.utc + def silence!(date = Time.now.utc) update!(silenced_at: date) end def unsilence! - update!(silenced_at: nil) + update!(silenced_at: nil, trust_level: trust_level == TRUST_LEVELS[:untrusted] ? TRUST_LEVELS[:trusted] : trust_level) end def suspended? suspended_at.present? end - def suspend!(date = nil) - date ||= Time.now.utc + def suspend!(date = Time.now.utc) transaction do user&.disable! if local? update!(suspended_at: date) diff --git a/app/services/remove_status_service.rb b/app/services/remove_status_service.rb index 6311971ff..a69fce8b8 100644 --- a/app/services/remove_status_service.rb +++ b/app/services/remove_status_service.rb @@ -23,6 +23,7 @@ class RemoveStatusService < BaseService remove_from_hashtags remove_from_public remove_from_media if status.media_attachments.any? + remove_from_spam_check @status.destroy! else @@ -142,6 +143,10 @@ class RemoveStatusService < BaseService redis.publish('timeline:public:local:media', @payload) if @status.local? end + def remove_from_spam_check + redis.zremrangebyscore("spam_check:#{@status.account_id}", @status.id, @status.id) + end + def lock_options { redis: Redis.current, key: "distribute:#{@status.id}" } end diff --git a/config/locales/en.yml b/config/locales/en.yml index 00b7d1dbe..89251ad40 100644 --- a/config/locales/en.yml +++ b/config/locales/en.yml @@ -875,6 +875,8 @@ en: profile: Profile relationships: Follows and followers two_factor_authentication: Two-factor Auth + spam_check: + spam_detected_and_silenced: This is an automated report. Spam has been detected and the sender has been silenced automatically. If this is a mistake, please unsilence the account. statuses: attached: description: 'Attached: %{attached}' diff --git a/db/migrate/20190701022101_add_trust_level_to_accounts.rb b/db/migrate/20190701022101_add_trust_level_to_accounts.rb new file mode 100644 index 000000000..917486d2e --- /dev/null +++ b/db/migrate/20190701022101_add_trust_level_to_accounts.rb @@ -0,0 +1,5 @@ +class AddTrustLevelToAccounts < ActiveRecord::Migration[5.2] + def change + add_column :accounts, :trust_level, :integer + end +end diff --git a/db/schema.rb b/db/schema.rb index 2e38fb1f2..c7b6b9be6 100644 --- a/db/schema.rb +++ b/db/schema.rb @@ -148,6 +148,7 @@ ActiveRecord::Schema.define(version: 2019_07_06_233204) do t.string "also_known_as", array: true t.datetime "silenced_at" t.datetime "suspended_at" + t.integer "trust_level" t.index "(((setweight(to_tsvector('simple'::regconfig, (display_name)::text), 'A'::\"char\") || setweight(to_tsvector('simple'::regconfig, (username)::text), 'B'::\"char\")) || setweight(to_tsvector('simple'::regconfig, (COALESCE(domain, ''::character varying))::text), 'C'::\"char\")))", name: "search_index", using: :gin t.index "lower((username)::text), lower((domain)::text)", name: "index_accounts_on_username_and_domain_lower", unique: true t.index ["moved_to_account_id"], name: "index_accounts_on_moved_to_account_id" diff --git a/spec/lib/spam_check_spec.rb b/spec/lib/spam_check_spec.rb new file mode 100644 index 000000000..c722dc642 --- /dev/null +++ b/spec/lib/spam_check_spec.rb @@ -0,0 +1,160 @@ +require 'rails_helper' + +RSpec.describe SpamCheck do + let!(:sender) { Fabricate(:account) } + let!(:alice) { Fabricate(:account, username: 'alice') } + let!(:bob) { Fabricate(:account, username: 'bob') } + + def status_with_html(text, options = {}) + status = PostStatusService.new.call(sender, { text: text }.merge(options)) + status.update_columns(text: Formatter.instance.format(status), local: false) + status + end + + describe '#hashable_text' do + it 'removes mentions from HTML for remote statuses' do + status = status_with_html('@alice Hello') + expect(described_class.new(status).hashable_text).to eq 'hello' + end + + it 'removes mentions from text for local statuses' do + status = PostStatusService.new.call(alice, text: "Hey @#{sender.username}, how are you?") + expect(described_class.new(status).hashable_text).to eq 'hey , how are you?' + end + end + + describe '#insufficient_data?' do + it 'returns true when there is no text' do + status = status_with_html('@alice') + expect(described_class.new(status).insufficient_data?).to be true + end + + it 'returns false when there is text' do + status = status_with_html('@alice h') + expect(described_class.new(status).insufficient_data?).to be false + end + end + + describe '#digest' do + it 'returns a string' do + status = status_with_html('@alice Hello world') + expect(described_class.new(status).digest).to be_a String + end + end + + describe '#spam?' do + it 'returns false for a unique status' do + status = status_with_html('@alice Hello') + expect(described_class.new(status).spam?).to be false + end + + it 'returns false for different statuses to the same recipient' do + status1 = status_with_html('@alice Hello') + described_class.new(status1).remember! + status2 = status_with_html('@alice Are you available to talk?') + expect(described_class.new(status2).spam?).to be false + end + + it 'returns false for statuses with different content warnings' do + status1 = status_with_html('@alice Are you available to talk?') + described_class.new(status1).remember! + status2 = status_with_html('@alice Are you available to talk?', spoiler_text: 'This is a completely different matter than what I was talking about previously, I swear!') + expect(described_class.new(status2).spam?).to be false + end + + it 'returns false for different statuses to different recipients' do + status1 = status_with_html('@alice How is it going?') + described_class.new(status1).remember! + status2 = status_with_html('@bob Are you okay?') + expect(described_class.new(status2).spam?).to be false + end + + it 'returns false for very short different statuses to different recipients' do + status1 = status_with_html('@alice 🙄') + described_class.new(status1).remember! + status2 = status_with_html('@bob Huh?') + expect(described_class.new(status2).spam?).to be false + end + + it 'returns false for statuses with no text' do + status1 = status_with_html('@alice') + described_class.new(status1).remember! + status2 = status_with_html('@bob') + expect(described_class.new(status2).spam?).to be false + end + + it 'returns true for duplicate statuses to the same recipient' do + status1 = status_with_html('@alice Hello') + described_class.new(status1).remember! + status2 = status_with_html('@alice Hello') + expect(described_class.new(status2).spam?).to be true + end + + it 'returns true for duplicate statuses to different recipients' do + status1 = status_with_html('@alice Hello') + described_class.new(status1).remember! + status2 = status_with_html('@bob Hello') + expect(described_class.new(status2).spam?).to be true + end + + it 'returns true for nearly identical statuses with random numbers' do + source_text = 'Sodium, atomic number 11, was first isolated by Humphry Davy in 1807. A chemical component of salt, he named it Na in honor of the saltiest region on earth, North America.' + status1 = status_with_html('@alice ' + source_text + ' 1234') + described_class.new(status1).remember! + status2 = status_with_html('@bob ' + source_text + ' 9568') + expect(described_class.new(status2).spam?).to be true + end + end + + describe '#skip?' do + it 'returns true when the sender is already silenced' do + status = status_with_html('@alice Hello') + sender.silence! + expect(described_class.new(status).skip?).to be true + end + + it 'returns true when the mentioned person follows the sender' do + status = status_with_html('@alice Hello') + alice.follow!(sender) + expect(described_class.new(status).skip?).to be true + end + + it 'returns false when even one mentioned person doesn\'t follow the sender' do + status = status_with_html('@alice @bob Hello') + alice.follow!(sender) + expect(described_class.new(status).skip?).to be false + end + + it 'returns true when the sender is replying to a status that mentions the sender' do + parent = PostStatusService.new.call(alice, text: "Hey @#{sender.username}, how are you?") + status = status_with_html('@alice @bob Hello', thread: parent) + expect(described_class.new(status).skip?).to be true + end + end + + describe '#remember!' do + pending + end + + describe '#flag!' do + let!(:status1) { status_with_html('@alice General Kenobi you are a bold one') } + let!(:status2) { status_with_html('@alice @bob General Kenobi, you are a bold one') } + + before do + described_class.new(status1).remember! + described_class.new(status2).flag! + end + + it 'silences the account' do + expect(sender.silenced?).to be true + end + + it 'creates a report about the account' do + expect(sender.targeted_reports.unresolved.count).to eq 1 + end + + it 'attaches both matching statuses to the report' do + expect(sender.targeted_reports.first.status_ids).to include(status1.id, status2.id) + end + end +end