NetExec/nxc/modules/groupmembership.py

98 lines
4.3 KiB
Python

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
from impacket.ldap import ldapasn1 as ldapasn1_impacket
from impacket.ldap import ldap as ldap_impacket
class NXCModule:
"""
Created as a contributtion from HackTheBox Academy team for CrackMapExec
Reference: https://academy.hackthebox.com/module/details/84
Module by @juliourena
Based on: https://github.com/juliourena/CrackMapExec/blob/master/cme/modules/get_description.py
"""
name = "groupmembership"
description = "Query the groups to which a user belongs."
supported_protocols = ["ldap"]
opsec_safe = True
multiple_hosts = True
def options(self, context, module_options):
"""
USER Choose a username to query group membership
"""
self.user = ""
if "USER" in module_options:
if module_options["USER"] == "":
context.log.fail("Invalid value for USER option!")
exit(1)
self.user = module_options["USER"]
else:
context.log.fail("Missing USER option, use --options to list available parameters")
exit(1)
def on_login(self, context, connection):
"""Concurrent. Required if on_admin_login is not present. This gets called on each authenticated connection"""
# Building the search filter
searchFilter = "(&(objectClass=user)(sAMAccountName={}))".format(self.user)
try:
context.log.debug("Search Filter=%s" % searchFilter)
resp = connection.ldapConnection.search(
searchFilter=searchFilter,
attributes=["memberOf", "primaryGroupID"],
sizeLimit=0,
)
except ldap_impacket.LDAPSearchError as e:
if e.getErrorString().find("sizeLimitExceeded") >= 0:
context.log.debug("sizeLimitExceeded exception caught, giving up and processing the data received")
# We reached the sizeLimit, process the answers we have already and that's it. Until we implement
# paged queries
resp = e.getAnswers()
pass
else:
context.log.debug(e)
return False
memberOf = []
primaryGroupID = ""
context.log.debug("Total of records returned %d" % len(resp))
for item in resp:
if isinstance(item, ldapasn1_impacket.SearchResultEntry) is not True:
continue
try:
for attribute in item["attributes"]:
if str(attribute["type"]) == "primaryGroupID":
primaryGroupID = attribute["vals"][0]
# Hardcode value for Domain Users primary Group ID 513
# For future improvement maybe we can query the primary ID value
# Reference: https://social.technet.microsoft.com/Forums/Azure/en-US/373febac-665c-494d-91f7-834541c74bee/cant-get-all-member-objects-from-domain-users-in-ldap?forum=winserverDS
if str(primaryGroupID) == "513":
memberOf.append("CN=Domain Users,CN=Users,DC=XXXXX,DC=XXX")
elif str(attribute["type"]) == "memberOf":
for group in attribute["vals"]:
if isinstance(group._value, bytes):
memberOf.append(str(group))
except Exception as e:
context.log.debug("Exception:", exc_info=True)
context.log.debug("Skipping item, cannot process due to error %s" % str(e))
pass
if len(memberOf) > 0:
context.log.success("User: {} is member of following groups: ".format(self.user))
for group in memberOf:
# Split the string on the "," character to get a list of the group name and parent group names
group_parts = group.split(",")
# The group name is the first element in the list, so we can extract it by taking the first element of the list
# and splitting it on the "=" character to get a list of the group name and its prefix (e.g., "CN")
group_name = group_parts[0].split("=")[1]
# print("Group name: %s" % group_name)
context.log.highlight("{}".format(group_name))