muonsocks

Introduction

This is an enhancement of rofl0r's excellent microsocks program with the following changes:

Support SOCKS4a clients
Support disabling outgoing ipv4 or ipv6
Support changing uid and chroot
Support binding to multiple ip/port tuples
Rewritten SOCKS5 parser that tolerates inputs split across multiple recv()
More performance from larger buffers and fewer poll invocations
Use TCP_NODELAY to lower latency impact
Use lock-free list rather than dynamic array for threads
Minimal memory allocations after init, and low heap fragmentation
Enhanced error handling

muonsocks fully supports SOCKS5 and SOCKS4a TCP proxying as a server.

It inherits the good design from microsocks, so the only real limits are set by the available RAM and file descriptor limits. OOM does not cause termination, and explicit memory allocation and heap fragmentation are minimized.

It is ~1000 LoC compared to microsocks's ~600 LoC, so it is not as minimal, but it is still a very small program (~27KiB dynamically linked to glibc on amd64).

Requirements

Linux or BSD system
GCC or Clang
GNU Make

Standard Usage

Compile and install muonsocks.

Build muonsocks: make
Install the muonsocks executable in a normal place. I suggest /usr/local/bin.

Set up the user account and chroot directory for muonsocks. Example:

$ su -
# umask 077
# groupadd muonsocks
# useradd -d /var/empty -s /sbin/nologin -g muonsocks muonsocks

Then the program can be run similarly to:

# muonsocks -u muonsocks -C /var/empty -4 -i 192.168.0.1 -i 10.0.0.1 -p 1080

Which would run a SOCKS5 server listening for requests on 192.168.0.1:1080 and 10.0.0.1:1080 that would only send outgoing IPv4 requests.

I suggest running muonsocks from a process supervisor such as s6. This will allow for reliable functioning in the case of unforseen or unrecoverable errors.

For full information on command line options, run:

$ muonsocks -?

History / Rationale

I previously used a SOCKS server that I wrote called nsocks. It used an event-driven model that attempted to use Linux's splice() to reduce kernel->userspace->kernel copies. However, the event-driven model works better for pure servers such as HTTP than for proxies; buffering is challenging to properly control in all cases for an event-driven proxy, but is natural with threads and careful use of blocking writes. splice() also gave marginal performance gains but made the program significantly more complex.

I ended up using the original microsocks when I became tired of trying to fix the many corner-cases in nsocks, and I expected to only use it temporarily; there were some features I needed, which I added, and I ported over some parts of nsocks that were well-tested such as the SOCKS4 support. Since nsocks was written in C++, I simply used C++ rather than porting my own code to C.

After relatively little work, my local version of microsocks (which I ended up calling muonsocks) worked trouble-free, and I ended up with no motivation to write another server. A few years later, I ended up porting back to C and making further improvements.

Downloads

Original microsocks README below

MicroSocks - multithreaded, small, efficient SOCKS5 server.

a SOCKS5 service that you can run on your remote boxes to tunnel connections through them, if for some reason SSH doesn't cut it for you.

It's very lightweight, and very light on resources too:

for every client, a thread with a stack size of 8KB is spawned. the main process basically doesn't consume any resources at all.

the only limits are the amount of file descriptors and the RAM.

It's also designed to be robust: it handles resource exhaustion gracefully by simply denying new connections, instead of calling abort() as most other programs do these days.

another plus is ease-of-use: no config file necessary, everything can be done from the command line and doesn't even need any parameters for quick setup.

History

This is the successor of "rocksocks5", and it was written with different goals in mind:

prefer usage of standard libc functions over homegrown ones
no artificial limits
do not aim for minimal binary size, but for minimal source code size, and maximal readability, reusability, and extensibility.

as a result of that, ipv4, dns, and ipv6 is supported out of the box and can use the same code, while rocksocks5 has several compile time defines to bring down the size of the resulting binary to extreme values like 10 KB static linked when only ipv4 support is enabled.

still, if optimized for size, this program when static linked against musl libc is not even 50 KB. that's easily usable even on the cheapest routers.

command line options

muonsocks -1 -i listenip -p port -U user -P password -b bindaddr

all arguments are optional. by default listenip is 0.0.0.0 and port 1080.

option -1 activates auth_once mode: once a specific ip address authed successfully with user/pass, it is added to a whitelist and may use the proxy without auth. this is handy for programs like firefox that don't support user/pass auth. for it to work you'd basically make one connection with another program that supports it, and then you can use firefox too.

niklata/muonsocks