123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272 |
- import ipaddress
- from random import randint, uniform
- from os import urandom
- from datetime import datetime
- from calendar import timegm
- from lea import Lea
- platforms = {"win7", "win10", "winxp", "win8.1", "macos", "linux", "win8", "winvista", "winnt", "win2000"}
- platform_probability = {"win7": 48.43, "win10": 27.99, "winxp": 6.07, "win8.1": 6.07, "macos": 5.94, "linux": 3.38,
- "win8": 1.35, "winvista": 0.46, "winnt": 0.31}
- x86_nops = {b'\x90', b'\xfc', b'\xfd', b'\xf8', b'\xf9', b'\xf5', b'\x9b'}
- x86_pseudo_nops = {b'\x97', b'\x96', b'\x95', b'\x93', b'\x92', b'\x91', b'\x99', b'\x4d', b'\x48', b'\x47', b'\x4f',
- b'\x40', b'\x41', b'\x37', b'\x3f', b'\x27', b'\x2f', b'\x46', b'\x4e', b'\x98', b'\x9f', b'\x4a',
- b'\x44', b'\x42', b'\x43', b'\x49', b'\x4b', b'\x45', b'\x4c', b'\x60', b'\x0e', b'\x1e', b'\x50',
- b'\x55', b'\x53', b'\x51', b'\x57', b'\x52', b'\x06', b'\x56', b'\x54', b'\x16', b'\x58', b'\x5d',
- b'\x5b', b'\x59', b'\x5f', b'\x5a', b'\x5e', b'\xd6'}
- forbidden_chars = [b'\x00', b'\x0a', b'\x0d']
- def update_timestamp(timestamp, pps, delay=0):
- """
- Calculates the next timestamp to be used based on the packet per second rate (pps) and the maximum delay.
- :return: Timestamp to be used for the next packet.
- """
- if delay == 0:
- # Calculate request timestamp
- # To imitate the bursty behavior of traffic
- randomdelay = Lea.fromValFreqsDict({1 / pps: 70, 2 / pps: 20, 5 / pps: 7, 10 / pps: 3})
- return timestamp + uniform(1 / pps, randomdelay.random())
- else:
- # Calculate reply timestamp
- randomdelay = Lea.fromValFreqsDict({2 * delay: 70, 3 * delay: 20, 5 * delay: 7, 10 * delay: 3})
- return timestamp + uniform(1 / pps + delay, 1 / pps + randomdelay.random())
- def get_interval_pps(complement_interval_pps, timestamp):
- """
- Gets the packet rate (pps) for a specific time interval.
- :param complement_interval_pps: an array of tuples (the last timestamp in the interval, the packet rate in the
- corresponding interval).
- :param timestamp: the timestamp at which the packet rate is required.
- :return: the corresponding packet rate (pps) .
- """
- for row in complement_interval_pps:
- if timestamp<=row[0]:
- return row[1]
- return complement_interval_pps[-1][1] # in case the timstamp > capture max timestamp
- def get_nth_random_element(*element_list):
- """
- Returns the n-th element of every list from an arbitrary number of given lists.
- For example, list1 contains IP addresses, list 2 contains MAC addresses. Use of this function ensures that
- the n-th IP address uses always the n-th MAC address.
- :param element_list: An arbitrary number of lists.
- :return: A tuple of the n-th element of every list.
- """
- range_max = min([len(x) for x in element_list])
- if range_max > 0: range_max -= 1
- n = randint(0, range_max)
- return tuple(x[n] for x in element_list)
- def index_increment(number: int, max: int):
- if number + 1 < max:
- return number + 1
- else:
- return 0
- def get_rnd_os():
- """
- Chooses random platform over an operating system probability distribution
- :return: random platform as string
- """
- os_dist = Lea.fromValFreqsDict(platform_probability)
- return os_dist.random()
- def check_platform(platform: str):
- """
- Checks if the given platform is currently supported
- if not exits with error
- :param platform: the platform, which should be validated
- """
- if platform not in platforms:
- print("\nERROR: Invalid platform: " + platform + "." +
- "\n Please select one of the following platforms: ", platforms)
- exit(1)
- def get_ip_range(start_ip: str, end_ip: str):
- """
- Generates a list of IPs of a given range. If the start_ip is greater than the end_ip, the reverse range is generated
- :param start_ip: the start_ip of the desired IP-range
- :param end_ip: the end_ip of the desired IP-range
- :return: a list of all IPs in the desired IP-range, including start-/end_ip
- """
- start = ipaddress.ip_address(start_ip)
- end = ipaddress.ip_address(end_ip)
- ips = []
- if start < end:
- while start <= end:
- ips.append(start.exploded)
- start = start+1
- elif start > end:
- while start >= end:
- ips.append(start.exploded)
- start = start-1
- else:
- ips.append(start_ip)
- return ips
- def generate_source_port_from_platform(platform: str, previousPort=0):
- """
- Generates the next source port according to the TCP-port-selection strategy of the given platform
- :param platform: the platform for which to generate source ports
- :param previousPort: the previously used/generated source port. Must be 0 if no port was generated before
- :return: the next source port for the given platform
- """
- check_platform(platform)
- if platform in {"winnt", "winxp", "win2000"}:
- if (previousPort == 0) or (previousPort + 1 > 5000):
- return randint(1024, 5000)
- else:
- return previousPort + 1
- elif platform == "linux":
- return randint(32768, 61000)
- else:
- if (previousPort == 0) or (previousPort + 1 > 65535):
- return randint(49152, 65535)
- else:
- return previousPort + 1
- def get_filetime_format(timestamp):
- """
- Converts a timestamp into MS FILETIME format
- :param timestamp: a timestamp in seconds
- :return: MS FILETIME timestamp
- """
- boot_datetime = datetime.fromtimestamp(timestamp)
- if boot_datetime.tzinfo is None or boot_datetime.tzinfo.utcoffset(boot_datetime) is None:
- boot_datetime = boot_datetime.replace(tzinfo=boot_datetime.tzname())
- boot_filetime = 116444736000000000 + (timegm(boot_datetime.timetuple()) * 10000000)
- return boot_filetime + (boot_datetime.microsecond * 10)
- def get_rnd_boot_time(timestamp, platform="winxp"):
- """
- Generates a random boot time based on a given timestamp and operating system
- :param timestamp: a timestamp in seconds
- :param platform: a platform as string as specified in check_platform above. default is winxp. this param is optional
- :return: timestamp of random boot time in seconds since EPOCH
- """
- check_platform(platform)
- if platform is "linux":
- uptime_in_days = Lea.fromValFreqsDict({3: 50, 7: 25, 14: 12.5, 31: 6.25, 92: 3.125, 183: 1.5625,
- 365: 0.78125, 1461: 0.390625, 2922: 0.390625})
- elif platform is "macos":
- uptime_in_days = Lea.fromValFreqsDict({7: 50, 14: 25, 31: 12.5, 92: 6.25, 183: 3.125, 365: 3.076171875,
- 1461: 0.048828125})
- else:
- uptime_in_days = Lea.fromValFreqsDict({3: 50, 7: 25, 14: 12.5, 31: 6.25, 92: 3.125, 183: 1.5625,
- 365: 0.78125, 1461: 0.78125})
- timestamp -= randint(0, uptime_in_days.random()*86400)
- return timestamp
- def get_rnd_x86_nop(count=1, side_effect_free=False, char_filter=set()):
- """
- Generates a specified number of x86 single-byte (pseudo-)NOPs
- :param count: The number of bytes to generate
- :param side_effect_free: Determines whether NOPs with side-effects (to registers or the stack) are allowed
- :param char_filter: A set of bytes which are forbidden to generate
- :return: Random x86 NOP bytestring
- """
- result = b''
- nops = x86_nops.copy()
- if not side_effect_free:
- nops |= x86_pseudo_nops.copy()
- if not isinstance(char_filter, set):
- char_filter = set(char_filter)
- nops = list(nops-char_filter)
- for i in range(0, count):
- result += nops[randint(0, len(nops) - 1)]
- return result
- def get_rnd_bytes(count=1, ignore=None):
- """
- Generates a specified number of random bytes while excluding unwanted bytes
- :param count: Number of wanted bytes
- :param ignore: The bytes, which should be ignored, as an array
- :return: Random bytestring
- """
- if ignore is None:
- ignore = []
- result = b''
- for i in range(0, count):
- char = urandom(1)
- while char in ignore:
- char = urandom(1)
- result += char
- return result
- def get_bytes_from_file(filepath):
- """
- Converts the content of a file into its byte representation
- The content of the file can either be a string or hexadecimal numbers/bytes (e.g. shellcode)
- The file must have the keyword "str" or "hex" in its first line to specify the rest of the content
- If the content is hex, whitespaces, backslashes, "x", quotation marks and "+" are removed
- Example for a hexadecimal input file:
- hex
- "abcd ef \xff10\ff 'xaa' x \ ab"
- Output: b'\xab\xcd\xef\xff\x10\xff\xaa\xab'
- :param filepath: The path of the file from which to get the bytes
- :return: The bytes of the file (either a byte representation of a string or the bytes contained in the file)
- """
- try:
- file = open(filepath)
- result_bytes = b''
- header = file.readline().strip()
- content = file.read()
- if header == "hex":
- content = content.replace(" ", "").replace("\n", "").replace("\\", "").replace("x", "").replace("\"", "")\
- .replace("'", "").replace("+", "").replace("\r", "")
- try:
- result_bytes = bytes.fromhex(content)
- except ValueError:
- print("\nERROR: Content of file is not all hexadecimal.")
- file.close()
- exit(1)
- elif header == "str":
- result_bytes = content.strip().encode()
- else:
- print("\nERROR: Invalid header found: " + header + ". Try 'hex' or 'str' followed by endline instead.")
- file.close()
- exit(1)
- for forbidden_char in forbidden_chars:
- if forbidden_char in result_bytes:
- print("\nERROR: Forbidden character found in payload: ", forbidden_char)
- file.close()
- exit(1)
- file.close()
- return result_bytes
- except FileNotFoundError:
- print("\nERROR: File not found: ", filepath)
- exit(1)
|