mirror of
https://github.com/mgeeky/Penetration-Testing-Tools.git
synced 2024-11-24 19:41:36 +01:00
511 lines
18 KiB
Python
511 lines
18 KiB
Python
#!/usr/bin/python3
|
|
#
|
|
# A script that enumerates Imports and Exports of PE files and prints them according to search criterias.
|
|
#
|
|
# Let's the user find imported/exported symbols matching criterias such as:
|
|
# - symbol being import or export
|
|
# - symbol matching name
|
|
# - symbol NOT matching name
|
|
# - module matching name
|
|
# - module NOT matching name
|
|
#
|
|
# Mariusz Banach / mgeeky, '21
|
|
# <mb [at] binary-offensive.com>
|
|
#
|
|
|
|
import os
|
|
import re
|
|
import sys
|
|
import glob
|
|
import json
|
|
import time
|
|
import signal
|
|
import pprint
|
|
import pefile
|
|
import tabulate
|
|
import platform
|
|
import textwrap
|
|
import argparse
|
|
import tempfile
|
|
import subprocess
|
|
import multiprocessing
|
|
|
|
from datetime import datetime
|
|
from itertools import product
|
|
from multiprocessing import Pool, Queue, Process, freeze_support, Manager, Lock
|
|
|
|
|
|
DEFAULT_COLUMN_SORTED = 'filename'
|
|
|
|
headers = [
|
|
'filename',
|
|
'symbol type',
|
|
'module',
|
|
'symbol',
|
|
'file size',
|
|
'path',
|
|
]
|
|
|
|
symbol_idx = headers.index('symbol')
|
|
|
|
class Logger:
|
|
colors_map = {
|
|
'red': 31,
|
|
'green': 32,
|
|
'yellow': 33,
|
|
'blue': 34,
|
|
'magenta': 35,
|
|
'cyan': 36,
|
|
'white': 37,
|
|
'grey': 38,
|
|
}
|
|
|
|
colors_dict = {
|
|
'error': colors_map['red'],
|
|
'trace': colors_map['magenta'],
|
|
'info ': colors_map['green'],
|
|
'debug': colors_map['grey'],
|
|
'other': colors_map['grey'],
|
|
}
|
|
|
|
@staticmethod
|
|
def with_color(c, s):
|
|
return "\x1b[%dm%s\x1b[0m" % (c, s)
|
|
|
|
@staticmethod
|
|
def end_color(s):
|
|
return "%s\x1b[0m" % (s)
|
|
|
|
@staticmethod
|
|
def colored(args, txt, col):
|
|
if not args.color:
|
|
return txt
|
|
|
|
return Logger.with_color(Logger.colors_map[col], txt)
|
|
|
|
def out(x):
|
|
sys.stderr.write(x + '\n')
|
|
|
|
def verbose(args, x):
|
|
if args.verbose:
|
|
sys.stderr.write('[verbose] ' + x + '\n')
|
|
|
|
def collectImports(args, mod):
|
|
imports = []
|
|
if not hasattr(mod, 'DIRECTORY_ENTRY_IMPORT') or not mod.DIRECTORY_ENTRY_IMPORT:
|
|
return imports
|
|
|
|
try:
|
|
for entry in mod.DIRECTORY_ENTRY_IMPORT:
|
|
module = entry.dll.decode('utf-8').lower()
|
|
for func in entry.imports:
|
|
if not func.name:
|
|
continue
|
|
func = func.name.decode()
|
|
imports.append(('import', module, func))
|
|
|
|
except Exception as e:
|
|
verbose(args, f'Exception occured while collecting PE imports: {e}')
|
|
|
|
return imports
|
|
|
|
def collectExports(args, filename, mod):
|
|
exports = []
|
|
if not hasattr(mod, 'DIRECTORY_ENTRY_EXPORT') or not mod.DIRECTORY_ENTRY_EXPORT:
|
|
return exports
|
|
|
|
try:
|
|
for entry in mod.DIRECTORY_ENTRY_EXPORT.symbols:
|
|
if not entry.name:
|
|
continue
|
|
|
|
func = entry.name.decode()
|
|
exports.append(('export', os.path.basename(filename), func))
|
|
|
|
except Exception as e:
|
|
verbose(args, f'Exception occured while collecting PE exports: {e}')
|
|
|
|
return exports
|
|
|
|
def verifyCriterias(args, regexes, infos, uniqueSymbols):
|
|
if args.unique and infos['symbol'] in uniqueSymbols:
|
|
verbose(args, f'(-) Skipping symbol {infos["module"]}.{infos["symbol"]} because it is not unique in our results.')
|
|
return False
|
|
|
|
if args.imports and infos['symbol type'] != 'import':
|
|
verbose(args, f'(-) Skipping symbol {infos["module"]}.{infos["symbol"]} because it was not an import.')
|
|
return False
|
|
|
|
if args.exports and infos['symbol type'] != 'export':
|
|
verbose(args, f'(-) Skipping symbol {infos["module"]}.{infos["symbol"]} because it was not an export.')
|
|
return False
|
|
|
|
regexesVerified = sum([len(v) for k, v in regexes.items()])
|
|
|
|
regexes_name = len(regexes['name'])
|
|
regexes_not_name = len(regexes['not-name'])
|
|
regexes_module = len(regexes['module'])
|
|
regexes_not_module = len(regexes['not-module'])
|
|
|
|
for name, rex in regexes['not-name']:
|
|
match = rex.search(infos['symbol'])
|
|
if match:
|
|
matched = match.group(1)
|
|
infos['symbol'] = infos['symbol'].replace(matched, Logger.colored(args, matched, 'red'))
|
|
verbose(args, f'(-) Skipping symbol {infos["module"]}.{infos["symbol"]} as it DID satisfy not-name ({name}) regex.')
|
|
return False
|
|
|
|
if regexes_not_module+regexes_module+regexes_name == 0:
|
|
verbose(args, f'(+) Symbol {infos["module"]}.{infos["symbol"]} satisfied all criterias.')
|
|
return True
|
|
|
|
for name, rex in regexes['not-module']:
|
|
match = rex.search(infos['module'])
|
|
if match:
|
|
matched = match.group(1)
|
|
infos['module'] = infos['module'].replace(matched, Logger.colored(args, matched, 'red'))
|
|
verbose(args, f'(-) Skipping symbol\'s module {infos["module"]}.{infos["symbol"]} as it DID satisfy not-module ({name}) regex.')
|
|
return False
|
|
|
|
if regexes_module+regexes_name == 0:
|
|
verbose(args, f'(+) Symbol {infos["module"]}.{infos["symbol"]} satisfied all criterias.')
|
|
return True
|
|
|
|
satisifed = False
|
|
carryOn = False
|
|
|
|
if len(regexes['module']) > 0:
|
|
for name, rex in regexes['module']:
|
|
match = rex.search(infos['module'])
|
|
if match:
|
|
matched = match.group(1)
|
|
infos['module'] = infos['module'].replace(matched, Logger.colored(args, matched, 'green'))
|
|
verbose(args, f'(+) Symbol\'s module {infos["module"]}.{infos["symbol"]} satisfied module ({name}) regex.')
|
|
carryOn = True
|
|
break
|
|
else:
|
|
carryOn = True
|
|
|
|
if regexes_name == 0:
|
|
verbose(args, f'(+) Symbol {infos["module"]}.{infos["symbol"]} satisfied all criterias.')
|
|
return True
|
|
|
|
if carryOn:
|
|
for name, rex in regexes['name']:
|
|
match = rex.search(infos['symbol'])
|
|
if match:
|
|
matched = match.group(1)
|
|
infos['symbol'] = infos['symbol'].replace(matched, Logger.colored(args, matched, 'green'))
|
|
verbose(args, f'(+) Symbol {infos["module"]}.{infos["symbol"]} satisfied name ({name}) regex.')
|
|
satisifed = True
|
|
break
|
|
|
|
if regexesVerified == 0 or satisifed:
|
|
verbose(args, f'(+) Symbol {infos["module"]}.{infos["symbol"]} satisfied all criterias.')
|
|
return True
|
|
else:
|
|
verbose(args, f'(-) Skipping symbol {infos["module"]}.{infos["symbol"]} as it DID NOT satisfy all criterias.')
|
|
return False
|
|
|
|
def processFileWorker(arguments):
|
|
out = None
|
|
try:
|
|
(args, regexes, path, results, uniqueSymbols, filesProcessed, symbolsProcessed) = arguments
|
|
out = processFile(args, regexes, path, results, uniqueSymbols, filesProcessed, symbolsProcessed)
|
|
|
|
except (KeyboardInterrupt, SystemExit) as e:
|
|
out(e)
|
|
|
|
return out
|
|
|
|
def processFile(args, regexes, path, results, uniqueSymbols, filesProcessed, symbolsProcessed):
|
|
verbose(args, 'Processing file: ' + path)
|
|
|
|
mod = None
|
|
|
|
try:
|
|
mod = pefile.PE(path, fast_load = True)
|
|
mod.parse_data_directories()
|
|
|
|
except:
|
|
mod.close()
|
|
return
|
|
|
|
imports = collectImports(args, mod)
|
|
exports = collectExports(args, os.path.basename(path), mod)
|
|
symbols = imports + exports
|
|
|
|
mod.close()
|
|
once = False
|
|
|
|
for (symbolType, symbolModule, symbolName) in symbols:
|
|
infos = {
|
|
'path' : path,
|
|
'filename' : os.path.basename(path),
|
|
'file size' : os.path.getsize(path),
|
|
'symbol type' : symbolType,
|
|
'symbol' : symbolName,
|
|
'module' : symbolModule,
|
|
}
|
|
|
|
if not once:
|
|
assert len(infos.keys()) == len(headers), "headers and infos.keys() mismatch"
|
|
assert list(infos.keys()).sort() == list(headers).sort(), "headers and infos.keys() mismatch while sorted"
|
|
once = True
|
|
|
|
if args.format == 'text':
|
|
appendRow = verifyCriterias(args, regexes, infos, uniqueSymbols)
|
|
|
|
if args.color:
|
|
if infos['symbol type'] == 'import':
|
|
infos['symbol type'] = Logger.colored(args, infos['symbol type'], 'cyan')
|
|
else:
|
|
infos['symbol type'] = Logger.colored(args, infos['symbol type'], 'yellow')
|
|
|
|
if appendRow:
|
|
row = []
|
|
MaxWidth = 40
|
|
|
|
for h in headers:
|
|
obj = None
|
|
|
|
if type(infos[h]) == set or type(infos[h]) == list or type(infos[h]) == tuple:
|
|
obj = ', '.join(infos[h])
|
|
else:
|
|
obj = infos[h]
|
|
|
|
if type(obj) == str and len(obj) > MaxWidth:
|
|
if h == 'path':
|
|
obj = '\n'.join(textwrap.wrap(obj, width = 2 * MaxWidth))
|
|
else:
|
|
obj = '\n'.join(textwrap.wrap(obj, width = MaxWidth))
|
|
|
|
row.append(obj)
|
|
|
|
results.append(row)
|
|
uniqueSymbols.append(symbolName)
|
|
|
|
#verbose(args, 'Processed results:\n' + pprint.pformat(infos))
|
|
|
|
else:
|
|
verbose(args, f'Symbol {symbolModule}.{symbolName} did not met filter criterias.')
|
|
|
|
elif args.format == 'json':
|
|
appendRow = verifyCriterias(args, regexes, infos, uniqueSymbols)
|
|
|
|
if args.color:
|
|
if infos['symbol type'] == 'import':
|
|
infos['symbol type'] = Logger.colored(args, infos['symbol type'], 'cyan')
|
|
else:
|
|
infos['symbol type'] = Logger.colored(args, infos['symbol type'], 'yellow')
|
|
|
|
if appendRow:
|
|
results.append(infos)
|
|
uniqueSymbols.append(symbolName)
|
|
|
|
#verbose(args, 'Processed results:\n' + pprint.pformat(infos))
|
|
|
|
else:
|
|
verbose(args, f'Symbol {symbolModule}.{symbolName} did not met filter criterias.')
|
|
|
|
filesProcessed.value += 1
|
|
symbolsProcessed.value += len(symbols)
|
|
|
|
def trap_handler(signum, frame):
|
|
out('[-] CTRL-C pressed. Wait a minute until all processes wrap up or manually terminate python\'s child processes tree.')
|
|
|
|
def init_worker():
|
|
signal.signal(signal.SIGINT, trap_handler)
|
|
|
|
def processDir(args, regexes, path, results, uniqueSymbols, filesProcessed, symbolsProcessed):
|
|
filePaths = []
|
|
|
|
out('[.] Building list of files to process...')
|
|
for file in glob.glob(os.path.join(path, '**'), recursive=args.recurse):
|
|
try:
|
|
if len(args.extension) > 0:
|
|
skip = True
|
|
for ext in args.extension:
|
|
if file.lower().endswith(f'.{ext}'):
|
|
skip = False
|
|
break
|
|
if skip:
|
|
verbose(args, f'[-] Skipping file as it not matched extension ({ext}): {file}')
|
|
continue
|
|
|
|
if os.path.isfile(file):
|
|
looks_like_pe = False
|
|
with open(file, 'rb') as f:
|
|
mz = f.read(2)
|
|
if len(mz) == 2:
|
|
looks_like_pe = (mz[0] == ord('M') and mz[1] == ord('Z')) or (mz[1] == ord('M') and mz[0] == ord('Z'))
|
|
|
|
if looks_like_pe: filePaths.append(file)
|
|
|
|
except Exception as e:
|
|
verbose(args, f'[-] Could not open file: ({file}). Exception: {e}')
|
|
continue
|
|
|
|
cpu_count = multiprocessing.cpu_count()
|
|
|
|
pool = Pool(cpu_count, initializer=init_worker)
|
|
|
|
try:
|
|
arguments = [[args, regexes, _path, results, uniqueSymbols, filesProcessed, symbolsProcessed] for _path in filePaths]
|
|
|
|
out(f'[.] Scanning {Logger.colored(args, len(filePaths), "yellow")} files...')
|
|
if len(filePaths) > 5000:
|
|
out(f'[.] Be patient that\'s gonna take a long while...')
|
|
|
|
res = pool.map(processFileWorker, arguments)
|
|
|
|
except KeyboardInterrupt:
|
|
out(f'[-] User interrupted the scan after {Logger.colored(args, filesProcessed.value, "red")} files.')
|
|
pool.terminate()
|
|
pool.join()
|
|
|
|
def opts(argv):
|
|
|
|
params = argparse.ArgumentParser(
|
|
prog = argv[0],
|
|
usage='%(prog)s [options] <path>'
|
|
)
|
|
|
|
params.add_argument('path', help = 'Path to a PE file or directory.')
|
|
params.add_argument('-r', '--recurse', action='store_true', help='If <path> is a directory, perform recursive scan.')
|
|
params.add_argument('-v', '--verbose', action='store_true', help='Verbose mode.')
|
|
params.add_argument('-f', '--format', choices=['text', 'json'], default='text', help='Output format. Text or JSON.')
|
|
params.add_argument('-E', '--extension', default=[], action='append', help='Extensions of files to scan. By default will scan all files. Can be repeated: -E exe -E dll')
|
|
params.add_argument('-o', '--output', metavar='PATH', help='Write output to file.')
|
|
params.add_argument('-C', '--color', default=False, action='store_true', help='Add colors to text output. May uglify table text output')
|
|
|
|
sorting = params.add_argument_group('Output sorting')
|
|
sorting.add_argument('-u', '--unique', action='store_true', help = 'Return unique symbols only. The first symbol with a name that occurs in results, will be returned.')
|
|
sorting.add_argument('-d', '--descending', action='store_true', help = 'Sort in descending order instead of default of descending.')
|
|
sorting.add_argument('-c', '--column', default=DEFAULT_COLUMN_SORTED, choices=headers, metavar='COLUMN', help = 'Sort by this column name. Default: filename. Available columns: "' + '", "'.join(headers) + '"')
|
|
sorting.add_argument('-n', '--first', type=int, default=0, metavar='NUM', help='Show only first N results, as specified in this paremeter. By default will show all candidates.')
|
|
|
|
filters = params.add_argument_group('Output filtering')
|
|
filters.add_argument('-i', '--imports', action='store_true', help = 'Filter only Imports.')
|
|
filters.add_argument('-e', '--exports', action='store_true', help = 'Filter only Exports.')
|
|
filters.add_argument('-s', '--name', action='append', default=[], help = 'Search for symbols with name matching this regular expression. Can be repeated, case insensitive')
|
|
filters.add_argument('-S', '--not-name', action='append', default=[], help = 'Search for symbols with name NOT matching this regular expression.')
|
|
filters.add_argument('-m', '--module', action='append', default=[], help = 'Search for symbols exported in/imported from this module matching regular expression.')
|
|
filters.add_argument('-M', '--not-module', action='append', default=[], help = 'Search for symbols NOT exported in/NOT imported from this module matching regular expression.')
|
|
|
|
args = params.parse_args()
|
|
|
|
if args.imports and args.exports:
|
|
out('[!] --imports and --exports are mutually exclusive. Pick only one of them!')
|
|
sys.exit(1)
|
|
|
|
accomodate_rex = lambda x: f'({x})'
|
|
|
|
regexes = {
|
|
'name': [],
|
|
'not-name': [],
|
|
'module': [],
|
|
'not-module': []
|
|
}
|
|
|
|
for name in args.name:
|
|
regexes['name'].append((name, re.compile(accomodate_rex(name), re.I)))
|
|
|
|
for not_name in args.not_name:
|
|
regexes['not-name'].append((not_name, re.compile(accomodate_rex(not_name), re.I)))
|
|
|
|
for module in args.module:
|
|
regexes['module'].append((module, re.compile(accomodate_rex(module), re.I)))
|
|
|
|
for not_module in args.not_module:
|
|
regexes['not-module'].append((not_module, re.compile(accomodate_rex(not_module), re.I)))
|
|
|
|
for i in range(len(args.extension)):
|
|
args.extension[i] = args.extension[i].lower()
|
|
if args.extension[i].startswith('.'):
|
|
args.extension[i] = args.extension[i][1:]
|
|
|
|
return args, regexes
|
|
|
|
def main():
|
|
results = Manager().list()
|
|
uniqueSymbols = Manager().list()
|
|
filesProcessed = Manager().Value('i', 0)
|
|
symbolsProcessed = Manager().Value('i', 0)
|
|
|
|
out('''
|
|
:: findSymbols.py - Finds PE Import/Exports based on supplied filters.
|
|
|
|
Mariusz Banach / mgeeky, '21
|
|
<mb [at] binary-offensive.com>
|
|
''')
|
|
|
|
args, regexes = opts(sys.argv)
|
|
|
|
is_wow64 = (platform.architecture()[0] == '32bit' and 'ProgramFiles(x86)' in os.environ)
|
|
|
|
start_time = datetime.now()
|
|
try:
|
|
if '\\system32\\' in args.path.lower() and is_wow64:
|
|
verbose(args, 'Redirecting input path from System32 to SysNative as we run from 32bit Python.')
|
|
args.path = args.path.lower().replace('\\system32\\', '\\SysNative\\')
|
|
|
|
if os.path.isdir(args.path):
|
|
processDir(args, regexes, args.path, results, uniqueSymbols, filesProcessed, symbolsProcessed)
|
|
|
|
else:
|
|
if not os.path.isfile(args.path):
|
|
out(f'[!] Input file does not exist! Path: {args.path}')
|
|
sys.exit(1)
|
|
|
|
processFile(args, regexes, args.path, results, uniqueSymbols, filesProcessed, symbolsProcessed)
|
|
|
|
except KeyboardInterrupt:
|
|
out(f'[-] User interrupted the scan.')
|
|
|
|
time_elapsed = datetime.now() - start_time
|
|
|
|
if args.format == 'json':
|
|
resultsList = list(results)
|
|
dumped = str(json.dumps(resultsList, indent=4))
|
|
|
|
if args.output:
|
|
with open(args.output, 'w') as f:
|
|
f.write(dumped)
|
|
else:
|
|
print('\n' + dumped)
|
|
else:
|
|
resultsList = list(results)
|
|
if len(resultsList) > 0:
|
|
|
|
idx = headers.index(args.column)
|
|
|
|
resultsList.sort(key = lambda x: x[idx], reverse = args.descending)
|
|
headers[idx] = '▼ ' + headers[idx] if args.descending else '▲ ' + headers[idx]
|
|
|
|
if args.first > 0:
|
|
for i in range(len(resultsList) - args.first):
|
|
resultsList.pop()
|
|
|
|
table = tabulate.tabulate(resultsList, headers=['#',] + headers, showindex='always', tablefmt='pretty')
|
|
|
|
if args.output:
|
|
with open(args.output, 'w', encoding='utf-8') as f:
|
|
f.write(str(table))
|
|
else:
|
|
print('\n' + table)
|
|
|
|
if args.first > 0:
|
|
out(f'\n[+] Found {Logger.colored(args, len(resultsList), "green")} symbols meeting all the criterias (but shown only first {Logger.colored(args, args.first, "magenta")} ones).\n')
|
|
else:
|
|
out(f'\n[+] Found {Logger.colored(args, len(resultsList), "green")} symbols meeting all the criterias.\n')
|
|
|
|
else:
|
|
out(f'[-] Did not find symbols meeting specified criterias.')
|
|
|
|
out(f'[.] Processed {Logger.colored(args, filesProcessed.value, "green")} files and {Logger.colored(args, symbolsProcessed.value, "green")} symbols.')
|
|
out('[.] Time elapsed: {}'.format(Logger.colored(args, time_elapsed, "magenta")))
|
|
|
|
if __name__ == '__main__':
|
|
freeze_support()
|
|
main() |