README.md

graftcp

English | 简体中文

Introduction

graftcp can redirect the TCP connection made by the given program [application, script, shell, etc.] to SOCKS5 proxy.

Compared with tsocks, proxychains or proxyChains-ng, graftcp is not using the LD_PRELOAD trick which only work for dynamically linked programs, e.g., applications builded by Go can not be hook by proxychains-ng. graftcp can trace or modify any given program's connect by ptrace(2), so it is workable for any program. The principle will be explained in this paragraph of how does it work.

Installation

Building graftcp-local requires Go installed.

git clone https://github.com/hmgle/graftcp.git
cd graftcp
make

After make finishes, you'll be able to use ./graftcp-local/graftcp-local and ./graftcp. Optionally, you can also install them to system:

sudo make install

Then graftcp-local will run automatically as the operating system starts.

Usage

graftcp-local:

$ graftcp-local/graftcp-local -h
Usage of graftcp-local/graftcp-local:
  -config string
        Path to the configuration file
  -listen string
        Listen address (default ":2233")
  -logfile string
        Write logs to file
  -loglevel value
        Log level (0-6) (default 1)
  -pipepath string
        Pipe path for graftcp to send address info (default "/tmp/graftcplocal.fifo")
  -service string
        Control the system service: ["start" "stop" "restart" "install" "uninstall"]
  -socks5 string
        SOCKS5 address (default "127.0.0.1:1080")
  -syslog
        Send logs to the local system logger (Eventlog on Windows, syslog on Unix)

graftcp:

$ graftcp -h
Usage: graftcp [options] prog [prog-args]

Options:
  -a --local-addr=<graftcp-local-IP-addr>
                    graftcp-local's IP address. Default: localhost
  -p --local-port=<graftcp-local-port>
                    Which port is graftcp-local listening? Default: 2233
  -f --local-fifo=<fifo-path>
                    Path of fifo to communicate with graftcp-local.
                    Default: /tmp/graftcplocal.fifo
  -b --blackip-file=<black-ip-file-path>
                    The IP in black-ip-file will connect direct
  -w --whiteip-file=<white-ip-file-path>
                    Only redirect the connect that destination ip in the
                    white-ip-file to SOCKS5
  -n --not-ignore-local
                    Connecting to local is not changed by default, this
                    option will redirect it to SOCKS5
  -h --help
                    Display this help and exit

Quick start

Assume you are running the SOCKS5 proxy with the default IP address: "localhost:1080". Start the graftcp-local first:

./graftcp-local/graftcp-local

Install the Go package from golang.org (now is blocked by the GFW) via graftcp:

./graftcp go get -v golang.org/x/net/proxy

Open Chromium / Chrome / Firefox browser via graftcp, then all the requests from this browser will redirect to the SOCKS5 proxy:

./graftcp chromium-browser

Launch Bash / Zsh / Fish via graftcp, then all the TCP traffic generated by the command in this shell will redirect to the SOCKS5 proxy:

% ./graftcp bash
$ wget https://www.google.com

How does it work?

To achieve the goal of redirecting the TCP connection of a app to another destination address and the app itself is not aware of it, these conditions are probably required:

• fork(2) a new process and trace it using ptrace(2), execv(2) to run the app. Every connect(2) syscall will be intercepted, then get the destination address argument and send it to graftcp-local via pipe.
• Modify the destination address argument of connect(2) to graftcp-local's address, and restart the stoped syscall. After the syscall returns successfully, the app thought it has connected the original destination address, but in face it connected the graftcp-local, so we named it "graft".
• graftcp-local establish a SOCKS5 connection based on the information of app's original destination address, then redirect the requests from the app to the SOCKS5 proxy.

Someone may have a question here: since we can modify the arguments of a syscall, modify the app's write(2) / send(2) buf argument, attach the original destination information to the write buffer, isn't it simpler? The answer is that cannot be done. Because attach data to the buffer of the tracked child process, it may case a buffer overflow, causing crash or overwrite other data.
In addition, as the execve(2) will detach and unmap all shared memory, we also cannot add extra data to the write buffer of traced app by sharing memory, so we send the original destination address via pipe.

The simple sketch is as follows:

+---------------+             +---------+         +--------+         +------+
|   graftcp     |  dest host  |         |         |        |         |      |
|   (tracer)    +---PIPE----->|         |         |        |         |      |
|      ^        |  info       |         |         |        |         |      |
|      | ptrace |             |         |         |        |         |      |
|      v        |             |         |         |        |         |      |
|  +---------+  |             |         |         |        |         |      |
|  |         |  |  connect    |         | connect |        | connect |      |
|  |         +--------------->| graftcp +-------->| socks5 +-------->| dest |
|  |         |  |             | -local  |         | proxy  |         | host |
|  |  app    |  |  req        |         |  req    |        |  req    |      |
|  |(tracee) +--------------->|         +-------->|        +-------->|      |
|  |         |  |             |         |         |        |         |      |
|  |         |  |  resp       |         |  resp   |        |  resp   |      |
|  |         |<---------------+         |<--------+        |<--------+      |
|  +---------+  |             |         |         |        |         |      |
+---------------+             +---------+         +--------+         +------+

FAQ and Tips

What are some ways to redirect TCP connections?

The main ones are: global way, environment variables setting way, and programs selection way.

Global way: e.g., use iptables + RedSocks to convert the system's traffic that match certain rules into SOCKS5 traffic. The pros is that it is globally effective; the cons is that all traffic that satisfies the rule is redirected, and the scope of influence is large.

Environment variable setting: some programs will read the proxy-related environment variables to determine whether to convert their own traffic to the corresponding proxy protocol traffic, such as curl will read http_proxy, ftp_proxy, all_proxy Environment variables and decide which proxy traffic to convert based on the request URL scheme. This way is effective only if the program itself implements the traffic conversion function, so it is very limited.

programs selection way: this way can only perform redirection for specified programs, such as tsocks or proxychains. As mentioned earlier, they were using the LD_PRELOAD hijacking dynamic library funciton, and the default static link compiled program such as Go is invalid. graftcp improves this by being able to redirect TCP connections from any program.

will graftcp redirect the connection to the SOCKS5 proxy if the target address is localhost?

No. By default, graftcp ignore the connections to localhost. If you want to redirect all addresses, you can use the -n option. If you want to ignore more addresses, you can add them to the blacklist IP file; if you want to redirect only certain IP addresses, you can add them to the whitelist IP file. Use graftcp --help to get more information.

I am suffering a DNS cache poisoning attack, does graftcp handle DNS requests?

No. graftcp currently only handles TCP connections. dnscrypt-proxy or ChinaDNS may help you.

The clone(2)'s argument has a flag CLONE_UNTRACED to avoid being traced, how does graftcp do forced tracing?

graftcp will intercept the clone(2) syscall, and clearing the CLONE_UNTRACED flag, so the tracked child process could not escape the fate of being tracked. In addition, this CLONE_UNTRACED flag is intended for the kernel, and user space program should not set it.

Linux provides a way to limit the ptrace(2): set the value of /proc/sys/kernel/yama/ptrace_scope. If ptrace(2) is invalid, check if the default value has been modified.

Does it support macOS?

No. macOS's ptrace(2) is useless. However, it can also be achieved theoretically by referringin DTrace. Anyone try it? ?

TODO

• ARM/Linux Support
• i386/Linux Support

Acknowledgements and References

• maybe, proxychains and proxyChains-ng for inspiration.
• strace.
• uthash
• uthash
• service
• dlog

LICENSE

Copyright © 2016, 2018 Hmgle [email protected]

Released under the terms of the GNU General Public License, version 3