theHarvester/discovery/baidusearch.py

40 lines
1.1 KiB
Python
Raw Normal View History

from parsers import myparser
import time
import requests
from discovery.constants import *
class search_baidu:
def __init__(self, word, limit):
self.word = word
self.total_results = ""
self.server = "www.baidu.com"
self.hostname = "www.baidu.com"
self.limit = limit
self.counter = 0
def do_search(self):
url = 'http://' + self.server + "/s?wd=%40" + self.word + "&pn=" + str(self.counter) + "&oq=" + self.word
headers = {
'Host': self.hostname,
'User-agent': getUserAgent()
}
h = requests.get(url=url, headers=headers)
time.sleep(getDelay())
self.total_results += h.text
def process(self):
while self.counter <= self.limit and self.counter <= 1000:
self.do_search()
print("\tSearching " + str(self.counter) + " results...")
self.counter += 10
def get_emails(self):
rawres = myparser.parser(self.total_results, self.word)
return rawres.emails()
def get_hostnames(self):
rawres = myparser.parser(self.total_results, self.word)
return rawres.hostnames()