mtg

Bullshit-free MTPROTO proxy for Telegram

Rationale

There are several available proxies for Telegram MTPROTO available. Here are the most notable:

Almost all of them follow the way how official proxy was build. This includes support of multiple secrets, support of promoted channels etc.

mtg is an implementation in golang which is intended to be:

Lightweight It has to consume as less resources as possible but not by losing maintainability.
Easily deployable I strongly believe that Telegram proxies should follow the way of ShadowSocks: promoted channels is a strange way of doing business I suppose. I think the only viable way is to have a proxy with minimum configuration which should work everywhere.
Single secret I think that multiple secrets solves no problems and just complexify software. I also believe that in case of throwout proxies, this feature is useless luxury.
Minimum docker image size Official image is less than 3 megabytes. Literally.
No management WebUI This is an implementation of simple lightweight proxy. I won't do that.

This proxy supports 2 modes of work: direct connection to Telegram and promoted channel mode. If you do not need promoted channels, I would recommend you to go with direct mode: this is way more robust.

To run proxy in direct mode, all you need to do is just provide a secret. If you do not provide ADTag as a second parameter, promoted channels mode won't be activated.

To get promoted channel, please contact @MTProxybot and provide generated adtag as a second parameter.

Source code organization

There are 2 main branches:

master branch contains potentially unstable features
stable branch contains stable version. Usually you want to use this branch.

How to build

make

If you want to build for another platform:

make crosscompile

If you want to build Docker image (called mtg):

make docker

Docker image

Docker follows the same policy as the source code organization:

latest mirrors the master branch
stable mirrors the stable branch
tags are for tagged releases

docker pull nineseconds/mtg:latest

docker pull nineseconds/mtg:stable

docker pull nineseconds/mtg:0.10

Ansible role

You can find unofficial Ansible role for mtg here: https://github.com/rlex/ansible-role-mtg Also, there is another project on Ansible Galaxy: https://galaxy.ansible.com/ivansible/lin_mtproxy

Configuration

Basically, to run this tool you need to configure as less as possible.

First, you need to generate a secret:

openssl rand -hex 16

head -c 512 /dev/urandom | md5sum | cut -f 1 -d ' '

Secure mode

tl;dr - use secret mode for all new installation of proxy; only clients with dd-secrets will be able to connect. This mode abuses attempts to DPI MTPROTO traffic.

Secure mode is not the best name and of course, it creates a lot of confusion. To explain what it means, we need to tell you some bits on dd-secrets.

MTPROTO proxy protocol requires 16-byte secret. You usually propagate it as a 32 characters hexadecimal string like 282831900f371ca182feb0e4e1e1aeef (if you decode this string to bytes, you will get a real secret which is used in the protocol). Everything went quite good until the moment when developers found an evidence that protocol is quite weak to DPI and some enthusiasts even created simple proofs of concepts on detecting MTPROTO traffic.

Telegram team has introduced a patch called dd-secrets. If you have a secret 282831900f371ca182feb0e4e1e1aeef then your dd-secret is dd282831900f371ca182feb0e4e1e1aeef. That is, you just add dd prefix to the secret, prepend it with dd. In that case, original secret 282831900f371ca182feb0e4e1e1aeef is used but client and server start to act a little bit different: they start to add random noise to the packets so they can't be detected by their length. In order to keep backward compatibility, all proxies a quite liberal to the secrets to use: if the client uses plain secret, without dd prefix, they fall back to the normal behavior. If dd-secret is used (proxy can extract this information on the handshake), then more secured, the hardened behavior is used.

Yes, it can look like a hack but it is as it is.

Now going back to the secure mode: if you do not pass -s flag to the mtg, then it checks what mode is requested by the client. If the client uses plain secret, without dd prefix, then proxy falls back to the original behavior and do not play with paddings. If dd-secret is used and client demands this mode, then proxy start to add that random noise to the packets. But if you pass -s, then only clients with dd-secrets can connect. How to migrate existing clients then? If a client is new enough, you can just prepend the secret with dd string in the settings. If it is an old guy, then nothing to do, sorry.

Why this mode matters? We do not have evidence but there is quite a big suspicion that some ISPs start to filter MTPROTO traffic. If they detect the IP address which acts as a proxy, they block it and no clients can use this proxy. This is an attempt to prevent such a situation.

General rule of thumb: with all new installation of proxies I would advise to go with secure mode by default. But please do remember that it means that clients, which do not pass dd-prefix to their secrets, will not be able to connect. Secure mode works only with dd-prefixes!

Oneliners to generate such secrets:

echo dd$(openssl rand -hex 16)

echo dd$(head -c 512 /dev/urandom | md5sum | cut -f 1 -d ' ')

Antireplay cache

In order to prevent replay attacks, we have internal storage of first frames messages for connected clients. These frames are generated randomly by design and we have negligible possibility of duplication (probability is 1/(2^64)) but it could be quite effective in order to prevent replays.

Environment variables

It is possible to configure this tool using environment variables. You can configure any flag but not secret or adtag. Here is the list of supported environment variables:

Environment variable	Corresponding flags	Default value	Description
`MTG_DEBUG`	`-d`, `--debug`	`false`	Run in debug mode. Usually, you need to run in this mode only if you develop this tool or its maintainer is asking you to provide logs with such verbosity.
`MTG_VERBOSE`	`-v`, `--verbose`	`false`	Run in verbose mode. This is way less chatty than debug mode.
`MTG_IP`	`-b`, `--bind-ip`	`127.0.0.1`	Which IP should we bind to. As usual, `0.0.0.0` means that we want to listen on all interfaces. Also, 4 zeroes will bind to both IPv4 and IPv6.
`MTG_PORT`	`-p`, `--bind-port`	`3128`	Which port should we bind to (listen on).
`MTG_IPV4`	`-4`, `--public-ipv4`	Autodetect	IPv4 address of this proxy. This is required if you NAT your proxy or run it in a docker container. In that case, you absolutely need to specify public IPv4 address of the proxy, otherwise either URLs will be broken or proxy could not access Telegram middle proxies.
`MTG_IPV4_PORT`	`--public-ipv4-port`	Value of `--bind-port`	Which port should be public of IPv4 interface. This affects only generated links and should be changed only if you NAT your proxy or run it in a docker container.
`MTG_IPV6`	`-6`, `--public-ipv6`	Autodetect	IPv6 address of this proxy. This is required if you NAT your proxy or run it in a docker container. In that case, you absolutely need to specify public IPv6 address of the proxy, otherwise either URLs will be broken or proxy could not access Telegram middle proxies.
`MTG_IPV6_PORT`	`--public-ipv6-port`	Value of `--bind-port`	Which port should be public of IPv6 interface. This affects only generated links and should be changed only if you NAT your proxy or run it in a docker container.
`MTG_STATS_IP`	`-t`, `--stats-ip`	`127.0.0.1`	Which IP should we bind the internal statistics HTTP server.
`MTG_STATS_PORT`	`-q`, `--stats-port`	`3129`	Which port should we bind the internal statistics HTTP server.
`MTG_STATSD_IP`	`--statsd-ip`		IP/host addresses of statsd service. No defaults, by defaults we do not send anything there.
`MTG_STATSD_PORT`	`--statsd-port`	`8125`	Which port should we use to work with statsd.
`MTG_STATSD_NETWORK`	`--statsd-network`	`udp`	Which protocol should we use to work with statsd. Possible options are `udp` and `tcp`.
`MTG_STATSD_PREFIX`	`--statsd-prefix`	`mtg`	Which bucket prefix we should use. For example, if you set `mtg`, then metric `traffic.ingress` would be send as `mtg.traffic.ingress`.
`MTG_STATSD_TAGS_FORMAT`	`--statsd-tags-format`		Which tags format we should use. By default, we are using default vanilla statsd tags format but if you want to send directly to InfluxDB or Datadog, please specify it there. Possible options are `influxdb` and `datadog`.
`MTG_STATSD_TAGS`	`--statsd-tags`		Which tags should we send to statsd with our metrics. Please specify them as `key=value` pairs.
`MTG_PROMETHEUS_PREFIX`	`--prometheus-prefix`	`mtg`	Which namespace should be used for prometheus metrics.
`MTG_BUFFER_WRITE`	`-w`, `--write-buffer`	`65536`	The size of TCP write buffer in bytes. Write buffer is the buffer for messages which are going from client to Telegram.
`MTG_BUFFER_READ`	`-r`, `--read-buffer`	`131072`	The size of TCP read buffer in bytes. Read buffer is the buffer for messages from Telegram to client.
`MTG_SECURE_ONLY`	`-s`, `--secure-only`	`false`	Support only clients with secure mode (i.e only clients with dd-secrets).
`MTG_ANTIREPLAY_MAXSIZE`	`anti-replay-max-size`	`128`	Max size of antireplay cache in megabytes.
`MTG_ANTIREPLAY_EVICTIONTIME`	`anti-replay-eviction-time`	`168h`	Eviction time for antireplay cache entries.

Usually you want to modify only read/write buffer sizes. If you feel that proxy is slow, try to increase both sizes giving more priority to read buffer.

Unfortunately, MTPROTO proxy protocol does not allow us to use splice or any other neat tricks how to eliminate the need of copying data into userspace.

How to run the tool

Now run the tool:

mtg <secret>

How to run the tool with ADTag:

mtg <secret> <adtag>

This tool will listen on port 3128 by default with the given secret.

One-line runner

docker run --name mtg --restart=unless-stopped -p 3128:3128 -p 127.0.0.1:3129:3129 -d nineseconds/mtg:stable $(openssl rand -hex 16)

or in secret mode:

docker run --name mtg --restart=unless-stopped -p 3128:3128 -p 127.0.0.1:3129:3129 -d nineseconds/mtg:stable dd$(openssl rand -hex 16)

You will have this tool up and running on port 3128. Now curl localhost:3129 to get tg:// links or do docker logs mtg. Also, port 3129 will show you some statistics if you are interested in.

Also, you can use run-mtg.sh script

statsd integration

mtg provides an integration with statsd, you can enable it with command line interface. To enable it, you have to provide IP address of statsd service.

Out of the box, mtg supports 2 additional dialects: InfluxDB and Datadog.

All metrics are gauges. Here is the list of metrics and their meaning:

Metric name	Unit	Description
`connections.abridged.ipv4`	number	The number of active abridged IPv4 connections
`connections.abridged.ipv6`	number	The number of active abridged IPv6 connections
`connections.intermediate.ipv4`	number	The number of active intermediate IPv4 connections
`connections.intermediate.ipv6`	number	The number of active intermediate IPv6 connections
`connections.secure.ipv4`	number	The number of active secure intermediate IPv4 connections
`connections.secure.ipv6`	number	The number of active secure intermediate IPv6 connections
`crashes`	number	An amount of crashes in client handlers
`traffic.ingress`	bytes	Ingress traffic from the start of application (incoming)
`traffic.egress`	bytes	Egress traffic from the start of application (outgoing)
`speed.ingress`	bytes/s	Ingress bandwidth of the latest second (incoming traffic)
`speed.egress`	bytes/s	Egress bandwidth of the latest second (outgoing traffic)

All metrics are prefixed with given prefix. Default prefix is mtg. With such prefix metric name traffic.ingress, for example, would be mtg.traffic.ingress.

Prometheus integration

Prometheus integration comes out of the box, you do not need to setup anything special. Prometheus scrape endpoint lives on the same IP/port where generic stats service (http://${MTG_STATS_IP}:${MTG_STATS_PORT}) but on /prometheus path. So, if you access http stats service as curl http://localhost:3129/, then your prometheus endpoint is curl http://localhost:3129/prometheus/.

ayanamist/mtg

mtg