import os
import sys
import re
import time
import requests
import random
import argparse
from functools import partial
from colored import fg, bg, attr
from multiprocessing.dummy import Pool
TOKENS_FILE = os.path.dirname(os.path.realpath(__file__))+'/.tokens'
def githubApiSearchCode( search, page ):
headers = {"Authorization":"token "+random.choice(t_tokens)}
url = 'https://api.github.com/search/code?s=indexed&type=Code&o=desc&q=' + search + '&page=' + str(page)
try:
r = requests.get( url, headers=headers, timeout=5 )
json = r.json()
return json
except Exception as e:
print( "%s[-] error occurred: %s%s" % (fg('red'),e,attr(0)) )
return False
def getRawUrl( result ):
raw_url = result['html_url'];
raw_url = raw_url.replace( 'https://github.com/', 'https://raw.githubusercontent.com/' )
raw_url = raw_url.replace( '/blob/', '/' )
return raw_url;
def readCode( regexp, source, result ):
url = getRawUrl( result )
code = doGetCode( url )
if code:
matches = re.findall( regexp, code )
if matches:
for sub in matches:
sub = sub[0].replace('2F','').lower().strip()
if len(sub) and not sub in t_history:
t_history.append( sub )
sys.stdout.write( "%s" % sub )
if source:
sys.stdout.write( "\t-> %s" % result['html_url'] )
sys.stdout.write( "\n" )
def doGetCode( url ):
try:
r = requests.get( url, timeout=5 )
except Exception as e:
sys.stdout.write( "%s[-] error occurred: %s%s\n" % (fg('red'),e,attr(0)) )
return False
return r.text
parser = argparse.ArgumentParser()
parser.add_argument( "-t","--token",help="auth token (required)" )
parser.add_argument( "-d","--domain",help="domain you are looking for (required)" )
parser.add_argument( "-e","--extend",help="also look for <dummy>example.com", action="store_true" )
parser.add_argument( "-s","--source",help="display first url where subdomains are found", action="store_true" )
parser.parse_args()
args = parser.parse_args()
t_tokens = []
if args.token:
t_tokens = args.token.split(',')
else:
if os.path.isfile(TOKENS_FILE):
fp = open(TOKENS_FILE,'r')
t_tokens = fp.read().split("\n")
fp.close()
if not len(t_tokens):
parser.error( 'auth token is missing' )
if args.source:
_source = True
else:
_source = False
if args.domain:
_domain = args.domain
else:
parser.error( 'domain is missing' )
t_history = []
page = 1
_search = '"' + _domain + '"'
import tldextract
t_host_parse = tldextract.extract( _domain )
_search = '"' + t_host_parse.domain + '"'
if args.extend:
_regexp = r'([0-9a-z_\-\.]+\.([0-9a-z_\-]+)?'+t_host_parse.domain+'([0-9a-z_\-\.]+)?\.[a-z]{1,5})'
else:
_regexp = r'(([0-9a-zA-Z_\-\.]+)\.' + _domain.replace('.','\.')+')'
while True:
time.sleep( 1 )
t_json = githubApiSearchCode( _search, page )
page = page + 1
if not t_json or 'documentation_url' in t_json or not 'items' in t_json or not len(t_json['items']):
break
pool = Pool( 30 )
pool.map( partial(readCode,_regexp,_source), t_json['items'] )
pool.close()
pool.join()