You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
112 lines
3.7 KiB
112 lines
3.7 KiB
# -*- coding: utf-8 -*-
|
|
# Author: kelvinBen
|
|
# Github: https://github.com/kelvinBen/AppInfoScanner
|
|
|
|
import re
|
|
import os
|
|
import config
|
|
import threading
|
|
import libs.core as cores
|
|
|
|
class ParsesThreads(threading.Thread):
|
|
|
|
def __init__(self,threadID,name,file_queue,all,result_dict,types):
|
|
threading.Thread.__init__(self)
|
|
self.file_queue = file_queue
|
|
self.name = name
|
|
self.threadID = threadID
|
|
self.result_list = []
|
|
self.all = all
|
|
self.result_dict=result_dict
|
|
self.types = types
|
|
|
|
def __regular_parse__(self):
|
|
while True:
|
|
if self.file_queue.empty():
|
|
break
|
|
|
|
file_path = self.file_queue.get(timeout = 5)
|
|
scan_str = ("Scan file : %s" % file_path)
|
|
print(scan_str)
|
|
|
|
if self.types == "iOS":
|
|
self.__get_string_by_iOS__(file_path)
|
|
else:
|
|
self.__get_string_by_file__(file_path)
|
|
|
|
result_set = set(self.result_list)
|
|
if len(result_set) !=0:
|
|
self.result_dict[file_path] = result_set
|
|
|
|
def __get_string_by_iOS__(self,file_path):
|
|
output_path = cores.output_path
|
|
strings_path = cores.strings_path
|
|
temp = os.path.join(output_path,"temp.txt")
|
|
cmd_str = ("%s %s > %s") % (strings_path,file_path,temp)
|
|
if os.system(cmd_str) == 0:
|
|
with open(temp,"r",encoding='utf-8',errors='ignore') as f:
|
|
lines = f.readlines()
|
|
for line in lines:
|
|
self.__parse_string__(line)
|
|
|
|
def __get_string_by_file__(self,file_path):
|
|
with open(file_path,"r",encoding="utf8",errors='ignore') as f :
|
|
file_content = f.read()
|
|
# 获取到所有的字符串
|
|
pattern = re.compile(r'\"(.*?)\"')
|
|
results = pattern.findall(file_content)
|
|
|
|
# 遍历所有的字符串
|
|
for result in set(results):
|
|
self.__parse_string__(result)
|
|
|
|
def __parse_string__(self,result):
|
|
# 通过正则筛选需要过滤的字符串
|
|
for filter_str in config.filter_strs:
|
|
filter_str_pat = re.compile(filter_str)
|
|
filter_resl = filter_str_pat.findall(result)
|
|
# print(result,filter_resl)
|
|
# 过滤掉未搜索到的内容
|
|
if len(filter_resl)!=0:
|
|
# 提取第一个结果
|
|
resl_str = filter_resl[0]
|
|
# 过滤
|
|
if self.__filter__(resl_str) == 0:
|
|
continue
|
|
|
|
self.threadLock.acquire()
|
|
self.result_list.append(resl_str)
|
|
self.threadLock.release()
|
|
continue
|
|
|
|
def __filter__(self,resl_str):
|
|
return_flag = 1
|
|
print(resl_str)
|
|
resl_str = resl_str.replace("\r","").replace("\n","").replace(" ","")
|
|
if len(resl_str) == 0:
|
|
return 0
|
|
|
|
# 目前流通的域名中加上协议头最短长度为11位
|
|
if len(resl_str) <= 10:
|
|
return 0
|
|
|
|
# 单独处理https或者http开头的字符串
|
|
# http_list =["https","https://","https:","http","http://","https:",]
|
|
# for filte in http_list:
|
|
# if filte == resl_str:
|
|
# return 0
|
|
|
|
for filte in config.filter_no:
|
|
resl_str = resl_str.replace(filte,"")
|
|
if len(resl_str) == 0:
|
|
return_flag = 0
|
|
continue
|
|
|
|
if re.match(filte,resl_str):
|
|
return_flag = 0
|
|
continue
|
|
return return_flag
|
|
|
|
def run(self):
|
|
self.threadLock = threading.Lock()
|
|
self.__regular_parse__()
|
|
|