An experimental Telegram server written from scratch in Python. Development chat: linked group to @ChameleonGram.
- WebK gets stuck on
sendCode()
. (note to self: inspect the MTProto workers inchrome://inspect/#workers
) - Multiple sessions handling for:
Give correctmsg_id
/seq_no
according to the Telegram specification -
A Websocket proxy for Telegram Web (WebZ / WebK). A work in progress temporary implementation is intools/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 withpython -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.
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()
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.
$ git clone https://github.com/DavideGalilei/piltover
$ cd piltover
Follow instructions at: https://python-poetry.org/docs/#installation
$ 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.
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.
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(
prime16), int(
exponent, 16))
- Replace those values, (optional: delete the rest of the keys)
- The key is the server fingerprint, the value is formed by this
expression:
- 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 withreturn (ip,
4430)
, instead of ports 80/443
- Every ip should become
- Install in development mode with
python3 -m pip install -e .
- Ready to use, run the server and check if
test.py
works
git clone --depth=1 https://github.com/LonamiWebs/telethon
- Edit these variables:
https://github.com/LonamiWebs/Telethon/blob/2007c83c9e1b1c85e60e4eca8e8651fcb120ee88/telethon/client/telegrambaseclient.py#L21-L24
-
DEFAULT_DC_ID = 2 DEFAULT_IPV4_IP = '127.0.0.1' DEFAULT_IPV6_IP = '2001:67c:4e8:f002::a' DEFAULT_PORT = 4430
- Just make sure that the default dc is 2, the ipv4 is localhost, and the default port is 4430. We don't really use ipv6 anyway...
-
- Add the rsa public key:
- Edit this file: https://github.com/LonamiWebs/Telethon/blob/2007c83c9e1b1c85e60e4eca8e8651fcb120ee88/telethon/crypto/rsa.py#L85
- Ideally, delete all the existing keys
- Take your server's public key from the
data/secrets/pubkey.asc
file, and add it there withadd_key("""
key here""", old=False)
- Install in development mode with
python3 -m pip install -e .
- Ready to use, run the server and check if
telethon_test.py
works
- Edit
this file:
- As always, replace every ip with
127.0.0.1
(localhost), and every port with4430
- 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.
- As always, replace every ip with
- 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
- 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 runningip -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.
- Replace every ip with your local ip address. It can be
- 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
)
- Remove the existing rsa keys, and replace them with your own, taken from the
- 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.
- Clone repo and install dependencies:
-
$ git clone https://github.com/morethanwords/tweb $ cd tweb $ npm i -g pnpm $ pnpm install
-
- Edit the values in this file:
https://github.com/morethanwords/tweb/blob/f2827d9c19616a560346bd1662665ca30dc54668/src/lib/mtproto/dcConfigurator.ts#L50
- Change
const chosenServer = `wss://...`
to: -
const chosenServer = `ws://127.0.0.1:3000/proxy`;
- Change every datacenter ip and port below, respectively to
127.0.0.1
(localhost) and3000
(websocket proxy port) https://github.com/morethanwords/tweb/blob/f2827d9c19616a560346bd1662665ca30dc54668/src/lib/mtproto/dcConfigurator.ts#L58-L70
- Change
- Edit the values in this file:
https://github.com/morethanwords/tweb/blob/f2827d9c19616a560346bd1662665ca30dc54668/src/lib/mtproto/rsaKeysManager.ts#L69-L78
- Change the
modulus
to the lowercase string ofprime
obtained previously
- Change the
- Run the websocket proxy from piltover
-
$ poetry run python tools/websocket_proxy.py
-
- Run with
npm start
- Wait some time for the app to compile
- Open the app in your browser (usually
https://0.0.0.0:8080/
)
- #TODO: WebZ instructions
- #TODO: the client is currently under active development and refactoring, so I will wait until a working version is released
- #TODO: add instructions. I haven't figured out how it should be done yet.
- 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
: Abridged0xeeeeeeee
: Intermediate0xdddddddd
: Padded Intermediate[length: 4 bytes][0x00000000]
: TCP Full, distinguishable by the emptyseq_no
(0x00000000
)[presumably random bytes]
: Usually and Obfuscated transport- To distinguish between
TCP Full
andObfuscated
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 usesjinja2
to generate theapi_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'sServer
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
- An authorization process starts, done by the
- 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...
- Client sends
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 :)
List of other server implementations I found:
- https://github.com/teamgram/teamgram-server
- https://github.com/aykutalparslan/Telegram-Server, moved to https://github.com/aykutalparslan/Ferrite
- https://github.com/loyldg/mytelegram
- https://github.com/nebula-chat/telegramd (now gone, probably moved to teamgram: https://github.com/nebula-chat/chatengine)
Various applications similar to Telegram (probably using a custom MTProto backend):
- https://nebula.chat/
- https://potato.im/
- https://icq.com/ (not sure about this one, but the clients are a copycat of Telegram's)