docker-postfix/scripts/email-anonymizer.py
Bojan Čekrlić c7c56d3ff1 Upd: Add new integration and unit tests
Test a few more functionalities in the image. Change how postfix
message IDs are detected. Message ID can be set by the client and
hence detecting them via regex is not the best way to go around it.

This fix will actually look at the log line and try to determine
if we're looking at the message ID or not.
2022-04-10 17:52:12 +02:00

357 lines
12 KiB
Python

#!/usr/bin/env python3
"""
Filter to anonyimize email addresses. It reads input line by line,
finds all emails in the input and masks them using given filter.
Big thanks to [Sergio Del Río Mayoral](https://github.com/sdelrio)
for the concept and the idea, although not a lot of the code went
into this commit in the end.
"""
import re
import logging
import typing
import json
import sys
import importlib
from urllib.parse import parse_qs
import hashlib
import hmac
logger = logging.getLogger(__name__)
# BIG FAT NOTICE on emails and regular expressions:
# If you're planning on using a regular expression to validate an email: don't. Emails
# are much more complex than you would imagine and most regular expressions will not
# cover all usecases. Newer RFCs even allow for international (read: UTF-8) email addresses.
# Most of your favourite programming languages will have a dedicated library for validating
# addresses.
#
# This pattern below, should, however match anything that remotely looks like an email.
# It is too broad, though, as it will match things which are not considered valid email
# addresses as well. But for our use case, that's OK and more than sufficient.
EMAIL_CATCH_ALL_PATTERN = '([^ "\\[\\]<]+|".+")@(\[([0-9]+\.[0-9]+\.[0-9]+\.[0-9]+|[A-Za-z0-9]+:.+)\]|([^ \\{}():;>]+(\.[^ \\{}():;]+)*))'
EMAIL_CATCH_ALL = re.compile(EMAIL_CATCH_ALL_PATTERN)
EMPTY_RESPONSE = json.dumps({})
"""A default filter, if none other is provided."""
DEFAULT_FILTER_CLASS: str = 'SmartFilter'
"""Map filter names to friendly names"""
FILTER_MAPPINGS = {
'default': DEFAULT_FILTER_CLASS,
'smart': 'SmartFilter',
'paranoid': 'ParanoidFilter',
'noop': 'Filter',
'hash': 'HashFilter',
}
"""
Helper function to check if the provided value is truthy or falsy.
Will throw an error on unknown values.
"""
def is_truthy(val: any, name: str) -> bool:
if val is None:
raise ValueError(f"Cannot evaluate empty (None) value for {name}")
if val == True:
return True
elif val == False:
return False
val = str(val).lower()
if val == 'y' or val == 'yes' or val == 't' or val == 'true' or val == '1':
return True
elif val == 'n' or val == 'no' or val == 'f' or val == 'false' or val == '0':
return False
raise ValueError(f"Cannot evaluate value '{val}' for {name}")
# -------------------------------------------------------------------------------- #
"""
Abstract base for all filters. Does nothing.
"""
class Filter():
MESSAGE_ID_LINE = "message-id="
MESSAGE_ID_LINE_LEN = len(MESSAGE_ID_LINE)
def init(self, args: 'dict[str, list[str]]') -> None:
pass
def is_message_id(self, match: re.match, msg: str) -> bool:
start = match.start()
email = match.group()
# Note that our regex will match thigs like "message-id=Issue1649523226559@postfix-mail.mail-system.svc.cluster.local"
# so we need to filter / check for these first
if email.startswith(self.MESSAGE_ID_LINE):
return True
if start >= self.MESSAGE_ID_LINE_LEN:
pos = start-1
while True:
char = msg[pos]
if char == '=':
break
elif char in '{<["\'':
pos = pos - 1
continue
return False
check = msg[pos-self.MESSAGE_ID_LINE_LEN+1:pos+1]
if check == self.MESSAGE_ID_LINE:
return True
return False
def replace(self, match: re.match, msg: str) -> str:
return match.group()
def processMessage(self, msg: str) -> typing.Optional[str]:
result = EMAIL_CATCH_ALL.sub(
lambda x: self.replace(x, msg), msg
)
return json.dumps({'msg': result}, ensure_ascii=False) if result != msg else EMPTY_RESPONSE
# -------------------------------------------------------------------------------- #
"""
This filter will take an educated guess at how to best mask the emails, specifically:
* It will leave the first and the last letter of the local part (if it's oly one letter, it will get repated)
* If the local part is in quotes, it will remove the quotes (Warning: if the email starts with a space, this might look weird in logs)
* It will replace all the letters inbetween with **ONE** asterisk
* It will replace everything but a TLD with a star
* Address-style domains will see the number replaced with stars
E.g.:
* `demo@example.org` -> `d*o@*******.org`
* `john.doe@example.solutions` -> `j*e@*******.solutions`
* `sa@localhost` -> `s*a@*********`
* `s@[192.168.8.10]` -> `s*s@[*.*.*.*]`
* `"multi....dot"@[IPv6:2001:db8:85a3:8d3:1319:8a2e:370:7348]` -> `m*t@[IPv6:*]`
This filter should still allow you to grep for lines relating to a specific email, if you know the original email, but you won't
be able to extract the original email back from the log.
"""
class SmartFilter(Filter):
mask_symbol: str = '*'
def init(self, args: 'dict[str, list[str]]') -> None:
if 'mask_symbol' in args and len(args['mask_symbol']) > 0:
self.mask_symbol = args['mask_symbol'][0]
super().init(args)
def mask_local(self, local: str) -> str:
if local[0] == '"' and local[-1] == '"':
return local[:2] + self.mask_symbol + local[-2:]
else:
return local[0] + self.mask_symbol + local[-1]
def mask_domain(self, domain: str) -> str:
if domain[0] == '[' and domain[-1] == ']': # Numerical domain
if ':' in domain[1:-1]:
left, right = domain.split(":", 1)
return left + ':' + (len(right)-1) * self.mask_symbol + ']'
else:
return '[' + self.mask_symbol + '.' + self.mask_symbol + '.' + self.mask_symbol + '.' + self.mask_symbol + ']'
elif '.' in domain: # Normal domain
s, tld = domain.rsplit('.', 1)
return len(s) * self.mask_symbol + '.' + tld
pass
else: # Local domain
return len(domain) * self.mask_symbol
def replace(self, match: re.match, msg: str) -> str:
email = match.group()
# Return the details unchanged if they look like Postfix message ID
if self.is_message_id(match, msg):
return email
# The "@" can show up in the local part, but shouldn't appear in the
# domain part (at least not that we know).
local, domain = email.rsplit("@", 1)
local = self.mask_local(local)
domain = self.mask_domain(domain)
return local + '@' + domain
# -------------------------------------------------------------------------------- #
"""
Paranoid filter is an extension of the smart filter. It will replace everything in the
local part with the `mask_symbol` and will leavel only top level domain (TLD) in the
domain part. Domain will be replaced by ONE occurence of the `mask_symbol`.
E.g.:
* `demo@example.org` -> `*@*.org`
* `john.doe@example.solutions` -> `*@*.solutions`
* `sa@localhost` -> `*@*`
* `s@[192.168.8.10]` -> `*@[*]`
* `"multi....dot"@[IPv6:2001:db8:85a3:8d3:1319:8a2e:370:7348]` -> `*@[IPv6:*]`
"""
class ParanoidFilter(SmartFilter):
def mask_local(self, local: str) -> str:
return self.mask_symbol
def mask_domain(self, domain: str) -> str:
if domain[0] == '[' and domain[-1] == ']': # Numerical domain
if ':' in domain[1:-1]:
left, right = domain.split(":", 1)
return left + ':*]'
else:
return '[*]'
elif '.' in domain: # Normal domain
s, tld = domain.rsplit('.', 1)
return self.mask_symbol + '.' + tld
pass
else: # Local domain
return self.mask_symbol
# -------------------------------------------------------------------------------- #
"""
HashFilter will replace the email with its hash. This should allow you to grep through
the logs by hashing the email address yourself, while retaining the anonymity.
HashFilter uses HMAC to hash addresses (SHA256), so make sure you inject your unique salt.
This way, the calculated hashes will be different accross organizations, though providing
complete anonymity.
Notice that for IP addresses hashes are not appropriate, the set of IP addresses is
limited and rainbow tables could be used to get back IP addresses. This is not possible
with emails.
IANAL, though, so your mileage my vary.
"""
class HashFilter(Filter):
prefix: str = '' # Prefix emails with this (set of) character(s) for easier grepping
suffix: str = '' # Suffix emails with this (set of) character(s) for easier grepping
case_sensitive: bool = True # Convert strings to lowercase if false
short_sha: bool = False # Cut the string to first 8 characters
split: bool = False # Split and anonymize local and domain part independently
def init(self, args: 'dict[str, list[str]]') -> None:
if 'salt' in args and len(args['salt']) > 0:
self.salt_encoded = args['salt'][0].encode()
else:
raise ValueError("You need to specify salt with this filter!")
if 'prefix' in args and len(args['prefix']) > 0:
self.prefix = args['prefix'][0]
if 'suffix' in args and len(args['suffix']) > 0:
self.suffix = args['suffix'][0]
if 'case_sensitive' in args and len(args['case_sensitive']) > 0:
self.case_sensitive = is_truthy(args['case_sensitive'][0], 'case_sensitive')
if 'short_sha' in args and len(args['short_sha']) > 0:
self.short_sha = is_truthy(args['short_sha'][0], 'short_sha')
if 'split' in args and len(args['split']) > 0:
self.split = is_truthy(args['split'][0], 'split')
super().init(args)
def replace(self, match: re.match, msg: str) -> str:
email = match.group()
# Return the details unchanged if they look like Postfix message ID
if self.is_message_id(match, msg):
return email
if not self.case_sensitive:
email = email.lower()
if self.split:
# The "@" can show up in the local part, but shouldn't appear in the
# domain part (at least not that we know).
local, domain = email.rsplit("@", 1)
local = hmac.new(self.salt_encoded, local.encode(), hashlib.sha256).hexdigest()
domain = hmac.new(self.salt_encoded, domain.encode(), hashlib.sha256).hexdigest()
if self.short_sha:
local = local[:8]
domain = domain[:8]
email = local + "@" + domain
else:
email = hmac.new(self.salt_encoded, email.encode(), hashlib.sha256).hexdigest()
if self.short_sha:
email = email[:8]
return self.prefix + email + self.suffix
# -------------------------------------------------------------------------------- #
def get_filter() -> Filter:
"""
Initialize the filter
This method will check your configuration and create a new filter
:return: Returns a specific implementation of the `Filter`
"""
opts: dict[str, list[str]] = {}
clazz: typing.Optional[str] = None
if len(sys.argv) > 1:
clazz = sys.argv[1].strip()
# opts = sys.argv[2:]
if '?' in clazz:
clazz, query_string = clazz.split("?", 1)
opts = parse_qs(query_string, keep_blank_values=True,strict_parsing=True)
if clazz.lower() in FILTER_MAPPINGS:
clazz = FILTER_MAPPINGS[clazz.lower()]
if clazz is None or clazz.strip() == '':
clazz = DEFAULT_FILTER_CLASS
logger.debug(f"Constructing new {clazz} filter.")
try:
if "." in clazz:
module_name, class_name = clazz.rsplit(".", 1)
filter_class = getattr(importlib.import_module(module_name), class_name)
filter_obj: Filter = filter_class()
else:
filter_class = getattr(sys.modules[__name__], clazz)
filter_obj: Filter = filter_class()
except Exception as e:
raise RuntimeError(f'Could not instatiate filter named "{clazz}"!') from e
try:
filter_obj.init(opts)
except Exception as e:
raise RuntimeError(f'Init of filter "{clazz}" with parameters {opts} failed!') from e
return filter_obj
def process(f: Filter) -> None:
while True:
message = sys.stdin.readline()
if message:
message = message[:-1] # Remove line feed
result = f.processMessage(message)
print(result)
sys.stdout.flush()
else:
# Empty line. stdin has been closed
break
process(get_filter())