mirror of
https://github.com/sqlmapproject/sqlmap.git
synced 2024-11-27 03:53:48 +03:00
150 lines
4.4 KiB
Python
150 lines
4.4 KiB
Python
#!/usr/bin/env python
|
|
|
|
"""
|
|
Copyright (c) 2006-2012 sqlmap developers (http://sqlmap.org/)
|
|
See the file 'doc/COPYING' for copying permission
|
|
"""
|
|
|
|
import cookielib
|
|
import re
|
|
import socket
|
|
import sys
|
|
import urllib
|
|
import urllib2
|
|
import ConfigParser
|
|
|
|
from operator import itemgetter
|
|
|
|
def main():
|
|
|
|
TIMEOUT = 10
|
|
CONFIG_FILE = 'sqlharvest.cfg'
|
|
TABLES_FILE = 'tables.txt'
|
|
USER_AGENT = 'Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1; AskTB5.3)'
|
|
SEARCH_URL = 'http://www.google.com/m?source=mobileproducts&dc=gorganic'
|
|
MAX_FILE_SIZE = 2*1024*1024 # if a result (.sql) file for downloading is more than 2MB in size just skip it
|
|
QUERY = 'CREATE TABLE ext:sql'
|
|
REGEX_URLS = r';u=([^"]+)'
|
|
REGEX_RESULT = r'CREATE TABLE\s*(/\*.*\*/)?\s*(IF NOT EXISTS)?\s*(?P<result>[^\(;]+)'
|
|
|
|
tables = dict()
|
|
refiles = re.compile(REGEX_URLS)
|
|
retables = re.compile(REGEX_RESULT, re.I)
|
|
|
|
cookies = cookielib.CookieJar()
|
|
cookie_processor = urllib2.HTTPCookieProcessor(cookies)
|
|
opener = urllib2.build_opener(cookie_processor)
|
|
opener.addheaders = [('User-Agent', USER_AGENT)]
|
|
|
|
conn = opener.open(SEARCH_URL)
|
|
page = conn.read() #set initial cookie values
|
|
|
|
config = ConfigParser.ConfigParser()
|
|
config.read(CONFIG_FILE)
|
|
if not config.has_section('options'):
|
|
config.add_section('options')
|
|
|
|
if not config.has_option('options', 'index'):
|
|
config.set('options', 'index', '0')
|
|
|
|
i = int(config.get('options', 'index'))
|
|
|
|
try:
|
|
f = open(TABLES_FILE, 'r')
|
|
for line in f.xreadlines():
|
|
if len(line) > 0 and ',' in line:
|
|
temp = line.split(',')
|
|
tables[temp[0]] = int(temp[1])
|
|
f.close()
|
|
except:
|
|
pass
|
|
|
|
socket.setdefaulttimeout(TIMEOUT)
|
|
|
|
files, oldFiles = None, None
|
|
try:
|
|
while True:
|
|
abort = False
|
|
oldFiles = files
|
|
files = []
|
|
|
|
try:
|
|
conn = opener.open('%s&q=%s&start=%d&sa=N' % (SEARCH_URL, QUERY.replace(' ', '+'), i*10))
|
|
page = conn.read()
|
|
for match in refiles.finditer(page):
|
|
files.append(urllib.unquote(match.group(1)))
|
|
if len(files) >= 10: break
|
|
abort = (files == oldFiles)
|
|
|
|
except KeyboardInterrupt:
|
|
raise
|
|
|
|
except Exception, msg:
|
|
print msg
|
|
|
|
if abort:
|
|
break
|
|
|
|
sys.stdout.write("\n---------------\n")
|
|
sys.stdout.write("Result page #%d\n" % (i+1))
|
|
sys.stdout.write("---------------\n")
|
|
|
|
for sqlfile in files:
|
|
print sqlfile
|
|
try:
|
|
req = urllib2.Request(sqlfile)
|
|
response = urllib2.urlopen(req)
|
|
|
|
if response.headers.has_key('Content-Length'):
|
|
if int(response.headers.get('Content-Length')) > MAX_FILE_SIZE:
|
|
continue
|
|
|
|
page = response.read()
|
|
found = False
|
|
counter = 0
|
|
|
|
for match in retables.finditer(page):
|
|
counter += 1
|
|
table = match.group("result").strip().strip("`").strip("\"").strip("'").replace('"."', ".").replace("].[", ".").strip('[').strip(']')
|
|
|
|
if table and '>' not in table and '<' not in table and '--' not in table and ' ' not in table:
|
|
found = True
|
|
sys.stdout.write('*')
|
|
|
|
if table in tables:
|
|
tables[table] += 1
|
|
else:
|
|
tables[table] = 1
|
|
if found:
|
|
sys.stdout.write("\n")
|
|
|
|
except KeyboardInterrupt:
|
|
raise
|
|
|
|
except Exception, msg:
|
|
print msg
|
|
|
|
else:
|
|
i += 1
|
|
|
|
except KeyboardInterrupt:
|
|
pass
|
|
|
|
finally:
|
|
f = open(TABLES_FILE, 'w+')
|
|
|
|
tables = sorted(tables.items(), key=itemgetter(1), reverse=True)
|
|
|
|
for table, count in tables:
|
|
f.write("%s,%d\n" % (table, count))
|
|
|
|
f.close()
|
|
config.set('options', 'index', str(i+1))
|
|
|
|
f = open(CONFIG_FILE, 'w+')
|
|
config.write(f)
|
|
f.close()
|
|
|
|
if __name__ == "__main__":
|
|
main()
|