108 lines
3.2 KiB
Python
Executable File
108 lines
3.2 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
# Generate the array of Certificate Transparency Log ID to description mappings
|
|
# for the TLS dissector.
|
|
#
|
|
# To update the TLS dissector source file, run this from the source directory:
|
|
#
|
|
# python3 tools/make-tls-ct-logids.py --update
|
|
#
|
|
|
|
import argparse
|
|
from base64 import b64decode, b64encode
|
|
import itertools
|
|
import requests
|
|
from hashlib import sha256
|
|
|
|
|
|
# Begin of comment, followed by the actual array definition
|
|
HEADER = "/* Generated by tools/make-tls-ct-logids.py\n"
|
|
# See also https://www.certificate-transparency.org/known-logs
|
|
CT_JSON_URL = 'https://www.gstatic.com/ct/log_list/v3/all_logs_list.json'
|
|
# File to be patched
|
|
SOURCE_FILE = "epan/dissectors/packet-tls-utils.c"
|
|
|
|
# Maximum elements per line in the value array. 11 is chosen because it results
|
|
# in output consistent with clang-format.
|
|
BYTES_PER_LINE = 11
|
|
|
|
|
|
def escape_c(s):
|
|
return s.replace('\\', '\\\\').replace('"', '\\"')
|
|
|
|
|
|
def byteshex(b):
|
|
return " ".join("0x%02x," % b for b in bytearray(b))
|
|
|
|
|
|
def process_json(obj, lastmod):
|
|
logs = list(itertools.chain(*[op['logs'] for op in obj['operators']]))
|
|
lines = HEADER
|
|
lines += " * Last-Modified %s, %s entries. */\n" % (lastmod, len(logs))
|
|
lines += "static const bytes_string ct_logids[] = {\n"
|
|
for entry in logs:
|
|
desc = entry["description"]
|
|
pubkey_der = b64decode(entry["key"])
|
|
key_id = sha256(pubkey_der).digest()
|
|
lines += ' { (const guint8[]){\n'
|
|
for offset in range(0, len(key_id), BYTES_PER_LINE):
|
|
lines += ' %s\n' % \
|
|
byteshex(key_id[offset:offset+BYTES_PER_LINE])
|
|
lines += ' },\n'
|
|
lines += ' %d, "%s" },\n' % (len(key_id), escape_c(desc))
|
|
lines += " { NULL, 0, NULL }\n"
|
|
lines += "};\n"
|
|
return lines
|
|
|
|
|
|
def parse_source():
|
|
"""
|
|
Reads the source file and tries to split it in the parts before, inside and
|
|
after the block.
|
|
"""
|
|
begin, block, end = '', '', ''
|
|
# Stages: 1 (before block), 2 (in block, skip), 3 (after block)
|
|
stage = 1
|
|
with open(SOURCE_FILE) as f:
|
|
for line in f:
|
|
if line == HEADER:
|
|
stage = 2 # Begin of block
|
|
if stage == 1:
|
|
begin += line
|
|
elif stage == 2:
|
|
block += line
|
|
if line.startswith('}'):
|
|
stage = 3 # End of block reached
|
|
elif stage == 3:
|
|
end += line
|
|
if stage != 3:
|
|
raise RuntimeError("Could not parse file (in stage %d)" % stage)
|
|
return begin, block, end
|
|
|
|
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument("--update", action="store_true",
|
|
help="Update %s as needed instead of writing to stdout" % SOURCE_FILE)
|
|
|
|
|
|
def main():
|
|
args = parser.parse_args()
|
|
r = requests.get(CT_JSON_URL)
|
|
code = process_json(r.json(), lastmod=r.headers['Last-Modified'])
|
|
|
|
if args.update:
|
|
begin, block, end = parse_source()
|
|
if block == code:
|
|
print("File is up-to-date")
|
|
else:
|
|
with open(SOURCE_FILE, "w") as f:
|
|
f.write(begin)
|
|
f.write(code)
|
|
f.write(end)
|
|
print("Updated %s" % SOURCE_FILE)
|
|
else:
|
|
print(code)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|