mirror of https://github.com/d3cim/block.git
309 lines
9.0 KiB
Python
309 lines
9.0 KiB
Python
#! /usr/bin/env python3
|
|
|
|
# run with python3 generate-domains-blocklist.py > list.txt.tmp && mv -f list.txt.tmp list
|
|
|
|
import argparse
|
|
import re
|
|
import sys
|
|
|
|
try:
|
|
import urllib2 as urllib
|
|
|
|
URLLIB_NEW = False
|
|
except (ImportError, ModuleNotFoundError):
|
|
import urllib.request as urllib
|
|
from urllib.request import Request
|
|
|
|
URLLIB_NEW = True
|
|
|
|
|
|
def parse_time_restricted_list(content):
|
|
rx_comment = re.compile(r"^(#|$)")
|
|
rx_inline_comment = re.compile(r"\s*#\s*[a-z0-9-].*$")
|
|
rx_trusted = re.compile(r"^([*a-z0-9.-]+)\s*(@\S+)?$")
|
|
rx_timed = re.compile(r".+\s*(@\S+)?$")
|
|
|
|
names = set()
|
|
time_restrictions = {}
|
|
rx_set = [rx_trusted]
|
|
for line in content.splitlines():
|
|
line = str.lower(str.strip(line))
|
|
if rx_comment.match(line):
|
|
continue
|
|
line = rx_inline_comment.sub("", line)
|
|
for rx in rx_set:
|
|
matches = rx.match(line)
|
|
if not matches:
|
|
continue
|
|
name = matches.group(1)
|
|
names.add(name)
|
|
time_restriction = matches.group(2)
|
|
if time_restriction:
|
|
time_restrictions[name] = time_restriction
|
|
return names, time_restrictions
|
|
|
|
|
|
def parse_trusted_list(content):
|
|
names, _time_restrictions = parse_time_restricted_list(content)
|
|
time_restrictions = {}
|
|
return names, time_restrictions
|
|
|
|
|
|
def parse_list(content, trusted=False):
|
|
rx_comment = re.compile(r"^(#|$)")
|
|
rx_inline_comment = re.compile(r"\s*#\s*[a-z0-9-].*$")
|
|
rx_u = re.compile(
|
|
r"^@*\|\|([a-z0-9][a-z0-9.-]*[.][a-z]{2,})\^?(\$(popup|third-party))?$")
|
|
rx_l = re.compile(r"^([a-z0-9][a-z0-9.-]*[.][a-z]{2,})$")
|
|
rx_h = re.compile(
|
|
r"^[0-9]{1,3}[.][0-9]{1,3}[.][0-9]{1,3}[.][0-9]{1,3}\s+([a-z0-9][a-z0-9.-]*[.][a-z]{2,})$"
|
|
)
|
|
rx_mdl = re.compile(r'^"[^"]+","([a-z0-9][a-z0-9.-]*[.][a-z]{2,})",')
|
|
rx_b = re.compile(r"^([a-z0-9][a-z0-9.-]*[.][a-z]{2,}),.+,[0-9: /-]+,")
|
|
rx_dq = re.compile(r"^address=/([a-z0-9][a-z0-9.-]*[.][a-z]{2,})/.")
|
|
|
|
if trusted:
|
|
return parse_trusted_list(content)
|
|
|
|
names = set()
|
|
time_restrictions = {}
|
|
rx_set = [rx_u, rx_l, rx_h, rx_mdl, rx_b, rx_dq]
|
|
for line in content.splitlines():
|
|
line = str.lower(str.strip(line))
|
|
if rx_comment.match(line):
|
|
continue
|
|
line = rx_inline_comment.sub("", line)
|
|
for rx in rx_set:
|
|
matches = rx.match(line)
|
|
if not matches:
|
|
continue
|
|
name = matches.group(1)
|
|
names.add(name)
|
|
return names, time_restrictions
|
|
|
|
|
|
# basic check if the line contains any regex specific char
|
|
def is_regex(line):
|
|
regex_chars = "*[]?}{"
|
|
return any(char in line for char in regex_chars)
|
|
|
|
|
|
def parse_regex(names):
|
|
regexes = set()
|
|
for line in names:
|
|
# skip lines without regex characters:
|
|
if not is_regex(line):
|
|
continue
|
|
# convert to python regex:
|
|
line=line.replace(".", "\.")
|
|
line=line.replace("*", ".*")
|
|
line = "^"+line+"$"
|
|
# check if resulting regex is valid:
|
|
try:
|
|
if re.compile(line):
|
|
regexes.add(line)
|
|
except re.error:
|
|
sys.stderr.write("Invalid regex: {} [{}]\n".format(line, re.error))
|
|
continue
|
|
return regexes
|
|
|
|
|
|
def print_restricted_name(name, time_restrictions):
|
|
if name in time_restrictions:
|
|
print("{}\t{}".format(name, time_restrictions[name]))
|
|
else:
|
|
print(
|
|
"# ignored: [{}] was in the time-restricted list, "
|
|
"but without a time restriction label".format(name)
|
|
)
|
|
|
|
|
|
def load_from_url(url):
|
|
sys.stderr.write("Loading data from [{}]\n".format(url))
|
|
req = urllib.Request(url=url, headers={"User-Agent": "dnscrypt-proxy"})
|
|
trusted = False
|
|
|
|
if URLLIB_NEW:
|
|
req_type = req.type
|
|
else:
|
|
req_type = req.get_type()
|
|
if req_type == "file":
|
|
trusted = True
|
|
|
|
response = None
|
|
try:
|
|
response = urllib.urlopen(req, timeout=int(args.timeout))
|
|
except urllib.URLError as err:
|
|
raise Exception("[{}] could not be loaded: {}\n".format(url, err))
|
|
if trusted is False and response.getcode() != 200:
|
|
raise Exception("[{}] returned HTTP code {}\n".format(
|
|
url, response.getcode()))
|
|
content = response.read()
|
|
if URLLIB_NEW:
|
|
content = content.decode("utf-8", errors="replace")
|
|
|
|
return (content, trusted)
|
|
|
|
|
|
def name_cmp(name):
|
|
parts = name.split(".")
|
|
parts.reverse()
|
|
return str.join(".", parts)
|
|
|
|
|
|
def has_suffix(names, name):
|
|
parts = str.split(name, ".")
|
|
while parts:
|
|
parts = parts[1:]
|
|
if str.join(".", parts) in names:
|
|
return True
|
|
|
|
return False
|
|
|
|
|
|
# check if a line matches with any of the collected regexes:
|
|
def covered_by_regex(line, regexes):
|
|
|
|
# only check lines that aren't regexes themselves:
|
|
if not is_regex(line):
|
|
for regex in regexes:
|
|
if re.match(regex, line):
|
|
return True
|
|
|
|
return False
|
|
|
|
|
|
def allowlist_from_url(url):
|
|
if not url:
|
|
return set()
|
|
content, trusted = load_from_url(url)
|
|
|
|
names, _time_restrictions = parse_list(content, trusted)
|
|
return names
|
|
|
|
|
|
def blocklists_from_config_file(
|
|
file, allowlist, time_restricted_url, ignore_retrieval_failure
|
|
):
|
|
blocklists = {}
|
|
allowed_names = set()
|
|
all_regexes = set()
|
|
all_names = set()
|
|
unique_names = set()
|
|
|
|
# Load conf & blocklists
|
|
with open(file) as fd:
|
|
for line in fd:
|
|
line = str.strip(line)
|
|
if str.startswith(line, "#") or line == "":
|
|
continue
|
|
url = line
|
|
try:
|
|
content, trusted = load_from_url(url)
|
|
names, _time_restrictions = parse_list(content, trusted)
|
|
blocklists[url] = names
|
|
all_names |= names
|
|
all_regexes |= parse_regex(names)
|
|
|
|
except Exception as e:
|
|
sys.stderr.write(str(e))
|
|
if not ignore_retrieval_failure:
|
|
exit(1)
|
|
|
|
# Time-based blocklist
|
|
if time_restricted_url and not re.match(r"^[a-z0-9]+:", time_restricted_url):
|
|
time_restricted_url = "file:" + time_restricted_url
|
|
|
|
if time_restricted_url:
|
|
time_restricted_content, _trusted = load_from_url(time_restricted_url)
|
|
time_restricted_names, time_restrictions = parse_time_restricted_list(
|
|
time_restricted_content
|
|
)
|
|
|
|
if time_restricted_names:
|
|
print("########## Time-based blocklist ##########\n")
|
|
for name in time_restricted_names:
|
|
print_restricted_name(name, time_restrictions)
|
|
|
|
# Time restricted names should be allowed, or they could be always blocked
|
|
allowed_names |= time_restricted_names
|
|
|
|
# Allowed list
|
|
if allowlist and not re.match(r"^[a-z0-9]+:", allowlist):
|
|
allowlist = "file:" + allowlist
|
|
|
|
allowed_names |= allowlist_from_url(allowlist)
|
|
|
|
# Process blocklists
|
|
for url, names in blocklists.items():
|
|
print("\n\n########## Blocklist from {} ##########\n".format(url))
|
|
ignored, allowed = 0, 0
|
|
list_names = list()
|
|
for name in names:
|
|
if has_suffix(all_names, name) or name in unique_names or covered_by_regex(name, all_regexes):
|
|
ignored = ignored + 1
|
|
elif has_suffix(allowed_names, name) or name in allowed_names:
|
|
allowed = allowed + 1
|
|
else:
|
|
list_names.append(name)
|
|
unique_names.add(name)
|
|
|
|
list_names.sort(key=name_cmp)
|
|
if ignored:
|
|
print("# Ignored duplicates: {}\n".format(ignored))
|
|
if allowed:
|
|
print("# Ignored entries due to the allowlist: {}\n".format(allowed))
|
|
for name in list_names:
|
|
print(name)
|
|
|
|
|
|
argp = argparse.ArgumentParser(
|
|
description="Create a unified blocklist from a set of local and remote files"
|
|
)
|
|
argp.add_argument(
|
|
"-c",
|
|
"--config",
|
|
default="domains-blocklist.conf",
|
|
help="file containing blocklist sources",
|
|
)
|
|
argp.add_argument(
|
|
"-w",
|
|
"--whitelist",
|
|
help="Deprecated. Please use -a or --allowlist",
|
|
)
|
|
argp.add_argument(
|
|
"-a",
|
|
"--allowlist",
|
|
default="domains-allowlist.txt",
|
|
help="file containing a set of names to exclude from the blocklist",
|
|
)
|
|
argp.add_argument(
|
|
"-r",
|
|
"--time-restricted",
|
|
default="domains-time-restricted.txt",
|
|
help="file containing a set of names to be time restricted",
|
|
)
|
|
argp.add_argument(
|
|
"-i",
|
|
"--ignore-retrieval-failure",
|
|
action="store_true",
|
|
help="generate list even if some urls couldn't be retrieved",
|
|
)
|
|
argp.add_argument("-t", "--timeout", default=30, help="URL open timeout")
|
|
|
|
args = argp.parse_args()
|
|
|
|
whitelist = args.whitelist
|
|
if whitelist:
|
|
print('Use of -w WHITELIST has been removed. Please use -a ALLOWLIST instead.')
|
|
exit(1)
|
|
|
|
conf = args.config
|
|
allowlist = args.allowlist
|
|
time_restricted = args.time_restricted
|
|
ignore_retrieval_failure = args.ignore_retrieval_failure
|
|
|
|
blocklists_from_config_file(
|
|
conf, allowlist, time_restricted, ignore_retrieval_failure)
|
|
|