Skip to content
/ piltover Public

An experimental Telegram Server implementation in Python

License

MIT license
60 stars 6 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

DavideGalilei/piltover

BranchesTags

Repository files navigation

piltover 🐳

An experimental Telegram server written from scratch in Python. Development chat: linked group to @ChameleonGram.

TODO

  • WebK gets stuck on sendCode(). (note to self: inspect the MTProto workers in chrome://inspect/#workers)
  • Multiple sessions handling for: Give correct msg_id/seq_no according to the Telegram specification
  • A Websocket proxy for Telegram Web (WebZ / WebK). A work in progress temporary implementation is in tools/websocket_proxy.js
  • Updates handling: pts, qts, etc.
  • Refactor the TL de/serialization module since the code is messy (e.g. make custom boxed types for list/int/str/bytes).
  • Refactor the server authorize() method.
  • Support multiple server keys to automatically switch to RSA_PAD for official clients, whilst keeping clients like Pyrogram/Telethon working with the old method. Currently handled manually in server.py: old = False
  • Support TL from multiple layers, and layer-based handlers. Add fallbacks eventually.
  • Add a tests/ directory with patched assertions from client libraries.
  • Use custom exceptions instead of Python assertions: assert statements are disabled with python -O, leading to missing important checks.
  • Add missing security checks, e.g., checking of g_a/g_b.
  • Refactor piltover/__main__.py, and use a database for auth keys/messages/users/updates (probably with SQLAlchemy and alambic due to reliable database migrations).
  • MTProxy support maybe? Obfuscation is already implemented, so why not?
  • HTTP/UDP support? Probably Telegram itself forgot those also exist.
  • Switch to hypercorn for the tcp server maybe?
  • Improve the README.

Purpose

This project is currently not meant to be used to host custom Telegram instances, as most security measures are currently barely in place. For now, it can be used by MTProto clients developers to understand why their code fails, whereas Telegram just closes the connection with a -404 error code.

That being said, it is planned in future to make it usable for most basic Telegram featues, including but not limited to, sending and receiving text and media messages, media, search.

This can be really useful for bots developers that would like to have a testing sandbox that doesn't ratelimit their bots.

The server is meant to be used as a library, providing 100% control of every answer

  • TODO: allow the user to override authorize()

Example

An example quick-start (incomplete) code would look like this:

import asyncio
from piltover.server import Server, Client, Request
from piltover.utils import gen_keys

async def main():
    pilt = Server(server_keys=gen_keys())
    # Running on localhost
    # Port: 4430

    @pilt.on_message("ping")
    async def pong(client: Client, request: Request):
        print("Received ping:", request.obj)

        return {
            "_": "pong",
            "msg_id": request.msg_id,
            "ping_id": request.obj.ping_id,
        }

    await pilt.serve()

asyncio.run(main())
$ poetry install --no-root
$ poetry run python -m piltover
# Server running on 127.0.0.1:4430...

Of course, this minimal setup is far from complete, and will only work for auth key generation and pings.

Development setup

General steps

1. Clone the repo:

$ git clone https://github.com/DavideGalilei/piltover
$ cd piltover

2. Install poetry

Follow instructions at: https://python-poetry.org/docs/#installation

3. Initial setup

$ poetry install --no-root
$ poetry run python tools/gen_tl.py update
$ poetry run python -m piltover

Now wait until it loads correctly and fire a Ctrl-C to stop the process.

You should see a line looking like this at the beginning

2023-11-05 19:52:31.171 | INFO     | __main__:main:49 - Pubkey fingerprint: -6bff292cf4837025 (9400d6d30b7c8fdb)

Get the fingerprint hex string and save it for later (some clients need it). In this case, the unsigned fingerprint is 9400d6d30b7c8fdb, but only for this example. Do not reuse this key fingerprint, as it will be different in your setup.

4. Extract public key number and exponent

At this point, two files should have been generated in your directory. Namely, data/secrets/privkey.asc and data/secrets/pubkey.asc. Keep in mind that some clients might need the PKCS1 public key in the normal ascii format.

Some others like pyrogram, do not have a RSA key parser and hardcode the number/exponent. To extract it, you can use this command:

$ grep -v -- - data/secrets/pubkey.asc | tr -d \\n | base64 -d | openssl asn1parse -inform DER -i

An example output would look like this:

  0:d=0  hl=4 l= 266 cons: SEQUENCE          
  4:d=1  hl=4 l= 257 prim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
265:d=1  hl=2 l=   3 prim:  INTEGER           :010001

Note the exponent (010001) and the prime number: (C3AE94...B575D1). Save those values for later.

Pyrogram

  • git clone --depth=1 https://github.com/pyrogram/pyrogram
  • Edit this dictionary: https://github.com/pyrogram/pyrogram/blob/b19764d5dc9e2d59a4ccbb7f520f78505800656b/pyrogram/crypto/rsa.py#L33
    • The key is the server fingerprint, the value is formed by this expression: PublicKey(int( prime 16), int( exponent , 16))
    • Replace those values, (optional: delete the rest of the keys)
  • Edit the datacenters ips in this file:
    • Every ip should become "127.0.0.1" (localhost)
    • In the DataCenter.__new__ method below, replace every return with return (ip, 4430 ), instead of ports 80/443
  • Install in development mode with python3 -m pip install -e .
  • Ready to use, run the server and check if test.py works

Telethon

Telegram Desktop

  • Edit this file:
    • As always, replace every ip with 127.0.0.1 (localhost), and every port with 4430
    • Remove the existing rsa keys, and replace them with your own, taken from the data/secrets/pubkey.asc file on your piltover folder. Important: check the newlines thoroughly and make sure they are there, or it won't work.
  • Build the program, ideally with GitHub Actions
  • Put the executable in a folder, e.g. tdesk
  • Since we don't save the auth keys, you should delete the leftover files from tdesktop at every run. You can do this conveniently and run your custom TDesktop with this command:
$ rm -rf tdata/ DebugLogs/ log.txt && c && ./Telegram

Telegram for Android

  • Clone the repo (optional: checkout the commit I used)
$ git clone https://github.com/DrKLO/Telegram.git
$ cd Telegram
$ # Optional: checkout the commit I used
$ git checkout 5bc1c3dce0e9108615c784a565051e54246fe0cb
  • Edit this file: https://github.com/DrKLO/Telegram/blob/5bc1c3dce0e9108615c784a565051e54246fe0cb/TMessagesProj/jni/tgnet/ConnectionsManager.cpp#L1779-L1839
    • Replace every ip with your local ip address. It can be 127.0.0.1 (localhost) only in the case you're running the app with an emulator on the same machine the server is running. Otherwise, change it with e.g. 192.168.1.35 (the LAN ip address of your machine, which you can obtain by running ip -c a).
    • Replace every port with 4430
    • Either delete or change IPv6 addresses to ::1 (localhost). This will break the IPv6 support, but if you know what you're doing, you can change it to your ipv6 address.
    • In DataCenters like DC2 there are multiple ip addresses, you should delete the extra ones and change only the first one.
  • Edit this file: https://github.com/DrKLO/Telegram/blob/5bc1c3dce0e9108615c784a565051e54246fe0cb/TMessagesProj/jni/tgnet/Handshake.cpp#L366-L385
    • Remove the existing rsa keys, and replace them with your own, taken from the data/secrets/pubkey.asc file on your piltover folder. Important: check the newlines thoroughly and make sure they are there, or it won't work. This took me way too much debugging time to realize that the missing newlines was the cause of the app crashes.
    • Change the unsigned fingerprint with the one you got from the server logs (e.g. 9400d6d30b7c8fdb)
  • If you haven't already, check the official repo for the build instructions. Generally, remember to edit TMessagesProj/src/main/java/org/telegram/messenger/BuildVars.java
  • Build the app, and see if it works.

Telegram WebK

Telegram WebZ

  • #TODO: WebZ instructions

Nimgram

  • #TODO: the client is currently under active development and refactoring, so I will wait until a working version is released

Telegram X/TDLib

  • #TODO: add instructions. I haven't figured out how it should be done yet.

Make a pull request if you want to add instructions for your own client.

How it works

  • The client connects with TCP sockets to the server (websockets for web clients)
  • The first bytes sent within a new connection determine the used transport
    • 0xef: Abridged
    • 0xeeeeeeee: Intermediate
    • 0xdddddddd: Padded Intermediate
    • [length: 4 bytes][0x00000000]: TCP Full, distinguishable by the empty seq_no (0x00000000)
    • [presumably random bytes]: Usually and Obfuscated transport
    • To distinguish between TCP Full and Obfuscated transports, a buffered reader is needed, to allow for peeking the stream without consuming it.
  • Type Language (TL) Data Serialization
    • In piltover, the TL de/serialization is JIT (Just In Time), allowing for an easy json-like interface at the cost of slow type checking at runtime (#TODO: do something about this) without complex code-generation parsers
    • The TL parser (tools/gen_tl.py) utility uses jinja2 to generate the api_tl.py / mtproto_tl.py files from the official TDesktop repo. (#TODO retrieve as much old schema layers for multi-layer support)
  • Authorization Key generation
    • An authorization process starts, done by the authorize() method of the piltover's Server class.
    • Generate random prime numbers for pq decomposition, a proof of work to avoid clients' DoS to the server
    • Either use an old algorithm or RSA_PAD to encrypt the inner data payload
    • The server checks the stuff it needs to check, the client too
    • If everything went correctly, we are authorized
    • It is worth noting that every auth key has its own id (the 8 lower order bytes of SHA1(auth_key))
    • Apart from the auth key id, every session has its own arbitrary (client provided) session_id, bound to the auth key. #TODO: Piltover doesn't currently check this value
  • Sign in / sign up process
    • Client sends invokeWithLayer(initConnection(getConfig(...)))
    • Client signs in with number / sms
    • Run the server and see the logs to find out more...

Why

One day, my Telegram account stopped working properly due to an internal server error originating from a supposedly corrupted message I forwarded. Every time the client tried to fetch new messages from private chats, it would face a [500 STORE_INVALID_OBJECT_TYPE] error. Hopefully, the bug was fixed in ~1/2 days after being reported, but the fact that it happened at all motivated me enough to try building my own server. In several days, I managed to make it kinda work :)

Miscellaneous

List of other server implementations I found:

Various applications similar to Telegram (probably using a custom MTProto backend):

About

An experimental Telegram Server implementation in Python

Topics

python telegram server messaging selfhosted self-hosted python3 mtproto telegram-server mtproto-server

Resources

Readme

License

MIT license
Activity

Stars

60 stars

Watchers

8 watching

Forks

6 forks
Report repository

Releases

No releases published

Contributors 3

  •  
  •  
  •  

Languages