AIL-framework/bin/packages/Term.py

517 lines
18 KiB
Python
Executable File

#!/usr/bin/env python3
# -*-coding:UTF-8 -*
import os
import re
import sys
import time
import uuid
import redis
import datetime
from collections import defaultdict
from nltk.tokenize import RegexpTokenizer
from textblob import TextBlob
sys.path.append(os.path.join(os.environ['AIL_BIN'], 'lib/'))
import ConfigLoader
import Tracker
from flask import escape
import Date
import Item
config_loader = ConfigLoader.ConfigLoader()
r_serv_term = config_loader.get_redis_conn("ARDB_Tracker")
config_loader = None
email_regex = r'[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,6}'
email_regex = re.compile(email_regex)
special_characters = set('[<>~!?@#$%^&*|()_-+={}":;,.\'\n\r\t]/\\')
special_characters.add('\\s')
# NLTK tokenizer
tokenizer = RegexpTokenizer('[\&\~\:\;\,\.\(\)\{\}\|\[\]\\\\/\-/\=\'\"\%\$\?\@\+\#\_\^\<\>\!\*\n\r\t\s]+',
gaps=True, discard_empty=True)
def is_valid_uuid_v4(UUID):
UUID = UUID.replace('-', '')
try:
uuid_test = uuid.UUID(hex=UUID, version=4)
return uuid_test.hex == UUID
except:
return False
# # TODO: use new package => duplicate fct
def is_in_role(user_id, role):
if r_serv_db.sismember('user_role:{}'.format(role), user_id):
return True
else:
return False
def check_term_uuid_valid_access(term_uuid, user_id):
if not is_valid_uuid_v4(term_uuid):
return ({"status": "error", "reason": "Invalid uuid"}, 400)
level = r_serv_term.hget('tracker:{}'.format(term_uuid), 'level')
if not level:
return ({"status": "error", "reason": "Unknown uuid"}, 404)
if level == 0:
if r_serv_term.hget('tracker:{}'.format(term_uuid), 'user_id') != user_id:
if not is_in_role(user_id, 'admin'):
return ({"status": "error", "reason": "Unknown uuid"}, 404)
return None
def is_valid_mail(email):
result = email_regex.match(email)
if result:
return True
else:
return False
def verify_mail_list(mail_list):
for mail in mail_list:
if not is_valid_mail(mail):
return ({'status': 'error', 'reason': 'Invalid email', 'value': mail}, 400)
return None
def is_valid_regex(term_regex):
try:
re.compile(term_regex)
return True
except:
return False
def get_text_word_frequency(item_content, filtering=True):
item_content = item_content.lower()
words_dict = defaultdict(int)
if filtering:
blob = TextBlob(item_content , tokenizer=tokenizer)
else:
blob = TextBlob(item_content)
for word in blob.tokens:
words_dict[word] += 1
return words_dict
# # TODO: create all tracked words
def get_tracked_words_list():
return list(r_serv_term.smembers('all:tracker:word'))
def get_set_tracked_words_list():
set_list = r_serv_term.smembers('all:tracker:set')
all_set_list = []
for elem in set_list:
res = elem.split(';')
num_words = int(res[1])
ter_set = res[0].split(',')
all_set_list.append((ter_set, num_words, elem))
return all_set_list
def get_regex_tracked_words_dict():
regex_list = r_serv_term.smembers('all:tracker:regex')
dict_tracked_regex = {}
for regex in regex_list:
dict_tracked_regex[regex] = re.compile(regex)
return dict_tracked_regex
def get_tracked_term_list_item(term_uuid, date_from, date_to):
all_item_id = []
if date_from and date_to:
for date in r_serv_term.zrangebyscore('tracker:stat:{}'.format(term_uuid), int(date_from), int(date_to)):
all_item_id = all_item_id + list(r_serv_term.smembers('tracker:item:{}:{}'.format(term_uuid, date)))
return all_item_id
def is_term_tracked_in_global_level(term, term_type):
res = r_serv_term.smembers('all:tracker_uuid:{}:{}'.format(term_type, term))
if res:
for elem_uuid in res:
if r_serv_term.hget('tracker:{}'.format(elem_uuid), 'level')=='1':
return True
return False
def is_term_tracked_in_user_level(term, term_type, user_id):
res = r_serv_term.smembers('user:tracker:{}'.format(user_id))
if res:
for elem_uuid in res:
if r_serv_term.hget('tracker:{}'.format(elem_uuid), 'tracked')== term:
if r_serv_term.hget('tracker:{}'.format(elem_uuid), 'type')== term_type:
return True
return False
def parse_json_term_to_add(dict_input, user_id):
term = dict_input.get('term', None)
if not term:
return ({"status": "error", "reason": "Term not provided"}, 400)
term_type = dict_input.get('type', None)
if not term_type:
return ({"status": "error", "reason": "Term type not provided"}, 400)
nb_words = dict_input.get('nb_words', 1)
description = dict_input.get('description', '')
description = escape(description)
res = parse_tracked_term_to_add(term , term_type, nb_words=nb_words)
if res[1]!=200:
return res
term = res[0]['term']
term_type = res[0]['type']
tags = dict_input.get('tags', [])
mails = dict_input.get('mails', [])
res = verify_mail_list(mails)
if res:
return res
## TODO: add dashboard key
level = dict_input.get('level', 1)
try:
level = int(level)
if level not in range(0, 1):
level = 1
except:
level = 1
# check if term already tracked in global
if level==1:
if is_term_tracked_in_global_level(term, term_type):
return ({"status": "error", "reason": "Term already tracked"}, 409)
else:
if is_term_tracked_in_user_level(term, term_type, user_id):
return ({"status": "error", "reason": "Term already tracked"}, 409)
term_uuid = add_tracked_term(term , term_type, user_id, level, tags, mails, description)
return ({'term': term, 'type': term_type, 'uuid': term_uuid}, 200)
def parse_tracked_term_to_add(term , term_type, nb_words=1):
if term_type=='regex':
if not is_valid_regex(term):
return ({"status": "error", "reason": "Invalid regex"}, 400)
elif term_type=='word' or term_type=='set':
# force lowercase
term = term.lower()
word_set = set(term)
set_inter = word_set.intersection(special_characters)
if set_inter:
return ({"status": "error", "reason": "special character not allowed", "message": "Please use a regex or remove all special characters"}, 400)
words = term.split()
# not a word
if term_type=='word' and len(words)>1:
term_type = 'set'
# ouput format: term1,term2,term3;2
if term_type=='set':
try:
nb_words = int(nb_words)
except:
nb_words = 1
if nb_words==0:
nb_words = 1
words_set = set(words)
words_set = sorted(words_set)
term = ",".join(words_set)
term = "{};{}".format(term, nb_words)
if nb_words > len(words_set):
nb_words = len(words_set)
elif term_type=='yara_custom':
if not Tracker.is_valid_yara_rule(term):
return ({"status": "error", "reason": "Invalid custom Yara Rule"}, 400)
elif term_type=='yara_default':
if not Tracker.is_valid_default_yara_rule(term):
return ({"status": "error", "reason": "The Yara Rule doesn't exist"}, 400)
else:
return ({"status": "error", "reason": "Incorrect type"}, 400)
return ({"status": "success", "term": term, "type": term_type}, 200)
def add_tracked_term(term , term_type, user_id, level, tags, mails, description, dashboard=0):
term_uuid = str(uuid.uuid4())
# YARA
if term_type == 'yara_custom' or term_type == 'yara_default':
term = Tracker.save_yara_rule(term_type, term, tracker_uuid=term_uuid)
term_type = 'yara'
# create metadata
r_serv_term.hset('tracker:{}'.format(term_uuid), 'tracked',term) # # TODO: use hash
r_serv_term.hset('tracker:{}'.format(term_uuid), 'type', term_type)
r_serv_term.hset('tracker:{}'.format(term_uuid), 'date', datetime.date.today().strftime("%Y%m%d"))
r_serv_term.hset('tracker:{}'.format(term_uuid), 'user_id', user_id)
r_serv_term.hset('tracker:{}'.format(term_uuid), 'level', level)
r_serv_term.hset('tracker:{}'.format(term_uuid), 'dashboard', dashboard)
if description:
r_serv_term.hset('tracker:{}'.format(term_uuid), 'description', description)
# create all term set
r_serv_term.sadd('all:tracker:{}'.format(term_type), term)
# create term - uuid map
r_serv_term.sadd('all:tracker_uuid:{}:{}'.format(term_type, term), term_uuid)
# add display level set
if level == 0: # user only
r_serv_term.sadd('user:tracker:{}'.format(user_id), term_uuid)
r_serv_term.sadd('user:tracker:{}:{}'.format(user_id, term_type), term_uuid)
elif level == 1: # global
r_serv_term.sadd('global:tracker', term_uuid)
r_serv_term.sadd('global:tracker:{}'.format(term_type), term_uuid)
# create term tags list
for tag in tags:
r_serv_term.sadd('tracker:tags:{}'.format(term_uuid), escape(tag) )
# create term tags mail notification list
for mail in mails:
r_serv_term.sadd('tracker:mail:{}'.format(term_uuid), escape(mail) )
# toggle refresh module tracker list/set
r_serv_term.set('tracker:refresh:{}'.format(term_type), time.time())
return term_uuid
def parse_tracked_term_to_delete(dict_input, user_id):
term_uuid = dict_input.get("uuid", None)
res = check_term_uuid_valid_access(term_uuid, user_id)
if res:
return res
delete_term(term_uuid)
return ({"uuid": term_uuid}, 200)
def delete_term(term_uuid):
term = r_serv_term.hget('tracker:{}'.format(term_uuid), 'tracked')
term_type = r_serv_term.hget('tracker:{}'.format(term_uuid), 'type')
level = r_serv_term.hget('tracker:{}'.format(term_uuid), 'level')
r_serv_term.srem('all:tracker_uuid:{}:{}'.format(term_type, term), term_uuid)
# Term not tracked by other users
if not r_serv_term.exists('all:tracker_uuid:{}:{}'.format(term_type, term)):
r_serv_term.srem('all:tracker:{}'.format(term_type), term)
# toggle refresh module tracker list/set
r_serv_term.set('tracker:refresh:{}'.format(term_type), time.time())
if level == '0': # user only
user_id = term_type = r_serv_term.hget('tracker:{}'.format(term_uuid), 'user_id')
r_serv_term.srem('user:tracker:{}'.format(user_id), term_uuid)
r_serv_term.srem('user:tracker:{}:{}'.format(user_id, term_type), term_uuid)
elif level == '1': # global
r_serv_term.srem('global:tracker', term_uuid)
r_serv_term.srem('global:tracker:{}'.format(term_type), term_uuid)
# delete metatadata
r_serv_term.delete('tracker:{}'.format(term_uuid))
# remove tags
r_serv_term.delete('tracker:tags:{}'.format(term_uuid))
# remove mails
r_serv_term.delete('tracker:mail:{}'.format(term_uuid))
# remove item set
all_item_date = r_serv_term.zrange('tracker:stat:{}'.format(term_uuid), 0, -1)
for date in all_item_date:
r_serv_term.delete('tracker:item:{}:{}'.format(term_uuid, date))
r_serv_term.delete('tracker:stat:{}'.format(term_uuid))
if term_type == 'yara':
# # TODO:
pass
def replace_tracker_description(term_uuid, description):
description = escape(description)
r_serv_term.hset('tracker:{}'.format(term_uuid), 'description', description)
def replace_tracked_term_tags(term_uuid, tags):
r_serv_term.delete('tracker:tags:{}'.format(term_uuid))
for tag in tags:
tag = escape(tag)
r_serv_term.sadd('tracker:tags:{}'.format(term_uuid), tag)
def replace_tracked_term_mails(term_uuid, mails):
res = verify_mail_list(mails)
if res:
return res
else:
r_serv_term.delete('tracker:mail:{}'.format(term_uuid))
for mail in mails:
mail = escape(mail)
r_serv_term.sadd('tracker:mail:{}'.format(term_uuid), mail)
def get_term_uuid_list(term, term_type):
return list(r_serv_term.smembers('all:tracker_uuid:{}:{}'.format(term_type, term)))
def get_term_tags(term_uuid):
return list(r_serv_term.smembers('tracker:tags:{}'.format(term_uuid)))
def get_term_mails(term_uuid):
return list(r_serv_term.smembers('tracker:mail:{}'.format(term_uuid)))
def add_tracked_item(term_uuid, item_id, item_date):
# track item
r_serv_term.sadd('tracker:item:{}:{}'.format(term_uuid, item_date), item_id)
# track nb item by date
r_serv_term.zadd('tracker:stat:{}'.format(term_uuid), item_date, int(item_date))
def create_token_statistics(item_date, word, nb):
r_serv_term.zincrby('stat_token_per_item_by_day:{}'.format(item_date), word, 1)
r_serv_term.zincrby('stat_token_total_by_day:{}'.format(item_date), word, nb)
r_serv_term.sadd('stat_token_history', item_date)
def delete_token_statistics_by_date(item_date):
r_serv_term.delete('stat_token_per_item_by_day:{}'.format(item_date))
r_serv_term.delete('stat_token_total_by_day:{}'.format(item_date))
r_serv_term.srem('stat_token_history', item_date)
def get_all_token_stat_history():
return r_serv_term.smembers('stat_token_history')
def get_tracked_term_last_updated_by_type(term_type):
epoch_update = r_serv_term.get('tracker:refresh:{}'.format(term_type))
if not epoch_update:
epoch_update = 0
return float(epoch_update)
def parse_get_tracker_term_item(dict_input, user_id):
term_uuid = dict_input.get('uuid', None)
res = check_term_uuid_valid_access(term_uuid, user_id)
if res:
return res
date_from = dict_input.get('date_from', None)
date_to = dict_input.get('date_to', None)
if date_from is None:
date_from = get_tracked_term_first_seen(term_uuid)
if date_from:
date_from = date_from[0]
if date_to is None:
date_to = date_from
if date_from > date_to:
date_from = date_to
all_item_id = get_tracked_term_list_item(term_uuid, date_from, date_to)
all_item_id = Item.get_item_list_desc(all_item_id)
res_dict = {}
res_dict['uuid'] = term_uuid
res_dict['date_from'] = date_from
res_dict['date_to'] = date_to
res_dict['items'] = all_item_id
return (res_dict, 200)
def get_tracked_term_first_seen(term_uuid):
res = r_serv_term.zrange('tracker:stat:{}'.format(term_uuid), 0, 0)
if res:
return res[0]
else:
return None
def get_tracked_term_last_seen(term_uuid):
res = r_serv_term.zrevrange('tracker:stat:{}'.format(term_uuid), 0, 0)
if res:
return res[0]
else:
return None
def get_term_metedata(term_uuid, user_id=False, description=False, level=False, tags=False, mails=False, sparkline=False):
dict_uuid = {}
dict_uuid['term'] = r_serv_term.hget('tracker:{}'.format(term_uuid), 'tracked')
dict_uuid['type'] = r_serv_term.hget('tracker:{}'.format(term_uuid), 'type')
dict_uuid['date'] = r_serv_term.hget('tracker:{}'.format(term_uuid), 'date')
dict_uuid['description'] = r_serv_term.hget('tracker:{}'.format(term_uuid), 'description')
dict_uuid['first_seen'] = get_tracked_term_first_seen(term_uuid)
dict_uuid['last_seen'] = get_tracked_term_last_seen(term_uuid)
if user_id:
dict_uuid['user_id'] = r_serv_term.hget('tracker:{}'.format(term_uuid), 'user_id')
if level:
dict_uuid['level'] = r_serv_term.hget('tracker:{}'.format(term_uuid), 'level')
if mails:
dict_uuid['mails'] = get_list_trackeed_term_mails(term_uuid)
if tags:
dict_uuid['tags'] = get_list_trackeed_term_tags(term_uuid)
if sparkline:
dict_uuid['sparkline'] = get_tracked_term_sparkline(term_uuid)
dict_uuid['uuid'] = term_uuid
return dict_uuid
def get_tracked_term_sparkline(tracker_uuid, num_day=6):
date_range_sparkline = Date.get_date_range(num_day)
sparklines_value = []
for date_day in date_range_sparkline:
nb_seen_this_day = r_serv_term.scard('tracker:item:{}:{}'.format(tracker_uuid, date_day))
if nb_seen_this_day is None:
nb_seen_this_day = 0
sparklines_value.append(int(nb_seen_this_day))
return sparklines_value
def get_list_tracked_term_stats_by_day(list_tracker_uuid, num_day=31, date_from=None, date_to=None):
if date_from and date_to:
date_range = Date.substract_date(date_from, date_to)
else:
date_range = Date.get_date_range(num_day)
list_tracker_stats = []
for tracker_uuid in list_tracker_uuid:
dict_tracker_data = []
tracker = r_serv_term.hget('tracker:{}'.format(tracker_uuid), 'tracked')
for date_day in date_range:
nb_seen_this_day = r_serv_term.scard('tracker:item:{}:{}'.format(tracker_uuid, date_day))
if nb_seen_this_day is None:
nb_seen_this_day = 0
dict_tracker_data.append({"date": date_day,"value": int(nb_seen_this_day)})
list_tracker_stats.append({"name": tracker,"Data": dict_tracker_data})
return list_tracker_stats
def get_list_trackeed_term_tags(term_uuid):
res = r_serv_term.smembers('tracker:tags:{}'.format(term_uuid))
if res:
return list(res)
else:
return []
def get_list_trackeed_term_mails(term_uuid):
res = r_serv_term.smembers('tracker:mail:{}'.format(term_uuid))
if res:
return list(res)
else:
return []
def get_user_tracked_term_uuid(user_id, filter_type=None):
if filter_type:
return list(r_serv_term.smembers('user:tracker:{}:{}'.format(user_id,filter_type)))
else:
return list(r_serv_term.smembers('user:tracker:{}'.format(user_id)))
def get_global_tracked_term_uuid(filter_type=None):
if filter_type:
return list(r_serv_term.smembers('global:tracker:{}'.format(filter_type)))
else:
return list(r_serv_term.smembers('global:tracker'))
def get_all_user_tracked_terms(user_id, filter_type=None):
all_user_term = []
all_user_term_uuid = get_user_tracked_term_uuid(user_id, filter_type=filter_type)
for term_uuid in all_user_term_uuid:
all_user_term.append(get_term_metedata(term_uuid, tags=True, mails=True, sparkline=True))
return all_user_term
def get_all_global_tracked_terms(filter_type=None):
all_user_term = []
all_user_term_uuid = get_global_tracked_term_uuid(filter_type=filter_type)
for term_uuid in all_user_term_uuid:
all_user_term.append(get_term_metedata(term_uuid, user_id=True, tags=True, mails=True, sparkline=True))
return all_user_term