mirror of
https://github.com/Grasscutters/mitmproxy.git
synced 2024-11-27 02:24:18 +00:00
258 lines
7.8 KiB
Python
258 lines
7.8 KiB
Python
"""
|
|
This script serializes the entire traffic dump, including websocket traffic,
|
|
as JSON, and either sends it to a URL or writes to a file. The serialization
|
|
format is optimized for Elasticsearch; the script can be used to send all
|
|
captured traffic to Elasticsearch directly.
|
|
|
|
Usage:
|
|
|
|
mitmproxy
|
|
--mode reverse:http://example.com/
|
|
-s examples/complex/jsondump.py
|
|
|
|
Configuration:
|
|
|
|
Send to a URL:
|
|
|
|
cat > ~/.mitmproxy/config.yaml <<EOF
|
|
dump_destination: "https://elastic.search.local/my-index/my-type"
|
|
# Optional Basic auth:
|
|
dump_username: "never-gonna-give-you-up"
|
|
dump_password: "never-gonna-let-you-down"
|
|
# Optional base64 encoding of content fields
|
|
# to store as binary fields in Elasticsearch:
|
|
dump_encodecontent: true
|
|
EOF
|
|
|
|
Dump to a local file:
|
|
|
|
cat > ~/.mitmproxy/config.yaml <<EOF
|
|
dump_destination: "/user/rastley/output.log"
|
|
EOF
|
|
"""
|
|
from threading import Lock, Thread
|
|
from queue import Queue
|
|
import base64
|
|
import json
|
|
import requests
|
|
|
|
from mitmproxy import ctx
|
|
|
|
FILE_WORKERS = 1
|
|
HTTP_WORKERS = 10
|
|
|
|
|
|
class JSONDumper:
|
|
"""
|
|
JSONDumper performs JSON serialization and some extra processing
|
|
for out-of-the-box Elasticsearch support, and then either writes
|
|
the result to a file or sends it to a URL.
|
|
"""
|
|
def __init__(self):
|
|
self.outfile = None
|
|
self.transformations = None
|
|
self.encode = None
|
|
self.url = None
|
|
self.lock = None
|
|
self.auth = None
|
|
self.queue = Queue()
|
|
|
|
def done(self):
|
|
self.queue.join()
|
|
if self.outfile:
|
|
self.outfile.close()
|
|
|
|
fields = {
|
|
'timestamp': (
|
|
('error', 'timestamp'),
|
|
|
|
('request', 'timestamp_start'),
|
|
('request', 'timestamp_end'),
|
|
|
|
('response', 'timestamp_start'),
|
|
('response', 'timestamp_end'),
|
|
|
|
('client_conn', 'timestamp_start'),
|
|
('client_conn', 'timestamp_end'),
|
|
('client_conn', 'timestamp_tls_setup'),
|
|
|
|
('server_conn', 'timestamp_start'),
|
|
('server_conn', 'timestamp_end'),
|
|
('server_conn', 'timestamp_tls_setup'),
|
|
('server_conn', 'timestamp_tcp_setup'),
|
|
),
|
|
'ip': (
|
|
('server_conn', 'source_address'),
|
|
('server_conn', 'ip_address'),
|
|
('server_conn', 'address'),
|
|
('client_conn', 'address'),
|
|
),
|
|
'ws_messages': (
|
|
('messages', ),
|
|
),
|
|
'headers': (
|
|
('request', 'headers'),
|
|
('response', 'headers'),
|
|
),
|
|
'content': (
|
|
('request', 'content'),
|
|
('response', 'content'),
|
|
),
|
|
}
|
|
|
|
def _init_transformations(self):
|
|
self.transformations = [
|
|
{
|
|
'fields': self.fields['headers'],
|
|
'func': dict,
|
|
},
|
|
{
|
|
'fields': self.fields['timestamp'],
|
|
'func': lambda t: int(t * 1000),
|
|
},
|
|
{
|
|
'fields': self.fields['ip'],
|
|
'func': lambda addr: {
|
|
'host': addr[0].replace('::ffff:', ''),
|
|
'port': addr[1],
|
|
},
|
|
},
|
|
{
|
|
'fields': self.fields['ws_messages'],
|
|
'func': lambda ms: [{
|
|
'type': m[0],
|
|
'from_client': m[1],
|
|
'content': base64.b64encode(bytes(m[2], 'utf-8')) if self.encode else m[2],
|
|
'timestamp': int(m[3] * 1000),
|
|
} for m in ms],
|
|
}
|
|
]
|
|
|
|
if self.encode:
|
|
self.transformations.append({
|
|
'fields': self.fields['content'],
|
|
'func': base64.b64encode,
|
|
})
|
|
|
|
@staticmethod
|
|
def transform_field(obj, path, func):
|
|
"""
|
|
Apply a transformation function `func` to a value
|
|
under the specified `path` in the `obj` dictionary.
|
|
"""
|
|
for key in path[:-1]:
|
|
if not (key in obj and obj[key]):
|
|
return
|
|
obj = obj[key]
|
|
if path[-1] in obj and obj[path[-1]]:
|
|
obj[path[-1]] = func(obj[path[-1]])
|
|
|
|
@classmethod
|
|
def convert_to_strings(cls, obj):
|
|
"""
|
|
Recursively convert all list/dict elements of type `bytes` into strings.
|
|
"""
|
|
if isinstance(obj, dict):
|
|
return {cls.convert_to_strings(key): cls.convert_to_strings(value)
|
|
for key, value in obj.items()}
|
|
elif isinstance(obj, list) or isinstance(obj, tuple):
|
|
return [cls.convert_to_strings(element) for element in obj]
|
|
elif isinstance(obj, bytes):
|
|
return str(obj)[2:-1]
|
|
return obj
|
|
|
|
def worker(self):
|
|
while True:
|
|
frame = self.queue.get()
|
|
self.dump(frame)
|
|
self.queue.task_done()
|
|
|
|
def dump(self, frame):
|
|
"""
|
|
Transform and dump (write / send) a data frame.
|
|
"""
|
|
for tfm in self.transformations:
|
|
for field in tfm['fields']:
|
|
self.transform_field(frame, field, tfm['func'])
|
|
frame = self.convert_to_strings(frame)
|
|
|
|
if self.outfile:
|
|
self.lock.acquire()
|
|
self.outfile.write(json.dumps(frame) + "\n")
|
|
self.lock.release()
|
|
else:
|
|
requests.post(self.url, json=frame, auth=(self.auth or None))
|
|
|
|
@staticmethod
|
|
def load(loader):
|
|
"""
|
|
Extra options to be specified in `~/.mitmproxy/config.yaml`.
|
|
"""
|
|
loader.add_option('dump_encodecontent', bool, False,
|
|
'Encode content as base64.')
|
|
loader.add_option('dump_destination', str, 'jsondump.out',
|
|
'Output destination: path to a file or URL.')
|
|
loader.add_option('dump_username', str, '',
|
|
'Basic auth username for URL destinations.')
|
|
loader.add_option('dump_password', str, '',
|
|
'Basic auth password for URL destinations.')
|
|
|
|
def configure(self, _):
|
|
"""
|
|
Determine the destination type and path, initialize the output
|
|
transformation rules.
|
|
"""
|
|
self.encode = ctx.options.dump_encodecontent
|
|
|
|
if ctx.options.dump_destination.startswith('http'):
|
|
self.outfile = None
|
|
self.url = ctx.options.dump_destination
|
|
ctx.log.info('Sending all data frames to %s' % self.url)
|
|
if ctx.options.dump_username and ctx.options.dump_password:
|
|
self.auth = (ctx.options.dump_username, ctx.options.dump_password)
|
|
ctx.log.info('HTTP Basic auth enabled.')
|
|
else:
|
|
self.outfile = open(ctx.options.dump_destination, 'a')
|
|
self.url = None
|
|
self.lock = Lock()
|
|
ctx.log.info('Writing all data frames to %s' % ctx.options.dump_destination)
|
|
|
|
self._init_transformations()
|
|
|
|
for i in range(FILE_WORKERS if self.outfile else HTTP_WORKERS):
|
|
t = Thread(target=self.worker)
|
|
t.daemon = True
|
|
t.start()
|
|
|
|
def response(self, flow):
|
|
"""
|
|
Dump request/response pairs.
|
|
"""
|
|
self.queue.put(flow.get_state())
|
|
|
|
def error(self, flow):
|
|
"""
|
|
Dump errors.
|
|
"""
|
|
self.queue.put(flow.get_state())
|
|
|
|
def websocket_end(self, flow):
|
|
"""
|
|
Dump websocket messages once the connection ends.
|
|
|
|
Alternatively, you can replace `websocket_end` with
|
|
`websocket_message` if you want the messages to be
|
|
dumped one at a time with full metadata. Warning:
|
|
this takes up _a lot_ of space.
|
|
"""
|
|
self.queue.put(flow.get_state())
|
|
|
|
def websocket_error(self, flow):
|
|
"""
|
|
Dump websocket errors.
|
|
"""
|
|
self.queue.put(flow.get_state())
|
|
|
|
|
|
addons = [JSONDumper()] # pylint: disable=invalid-name
|