推荐一款让所有终端程序轻松支持 SOCKS5 代理的神器 graftcp

Posted by Mike on 2020-05-17

你是否经常有在终端下通过一些实用程序进行工作的需求呢,比如:Git 等。但是由于这些终端程序默认并不支持 Socks 5 代理或 HTTP 代理,在访问一些需要科学上网的网络服务时速度或许会非常的慢。

通常我们的解决方法就是使用环境变量 export ALL_PROXY=socks5://proxyAddress:port 或者 export http_proxy=http://proxyAddress:port 给所有终端程序配置一个全局代理,这样做虽然有效但并不是最佳解决方案。

今天就给大家介绍一款神器 graftcpgraftcp 可以把任何指定的终端程序的 TCP 连接重定向到 SOCKS5 或 HTTP 代理,并且不会影响其它的终端程序。是不是很好的解决了你的痛点呢?

简介

graftcp 可以把任何指定程序(应用程序、脚本、shell 等)的 TCP 连接重定向到 SOCKS5 或 HTTP 代理。

对比 tsocksproxychainsproxychains-nggraftcp 并不使用 LD_PRELOAD 技巧来劫持共享库的 connect()、getaddrinfo()
等系列函数达到重定向目的,这种方法只对使用动态链接编译的程序有效,对于静态链接编译出来的程序,例如默认选项编译的 Go 程序proxychains-ng 就无效了graftcp 使用 ptrace(2) 系统调用跟踪或修改任意指定程序的 connect 信息,对任何程序都有效。工作原理后面将会解释。

项目地址:https://github.com/hmgle/graftcp

安装

graftcp 在 Linux 系统内运行。 graftcp-local 使用 Go 编写, Go 环境是必需的。

1
2
3
$ git clone https://github.com/hmgle/graftcp.git
$ cd graftcp
$ make

make 执行完后,即可运行 graftcp-local/graftcp-local./graftcp。你也可以把它们都安装进系统:

1
$ sudo make install

之后 graftcp-local 会随着系统启动而自动运行。

用法参数

graftcp-local:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
$ graftcp-local/graftcp-local -h
Usage of graftcp-local/graftcp-local:
-config string
Path to the configuration file
-http_proxy string
http proxy address, e.g.: 127.0.0.1:8080
-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")
-select_proxy_mode string
Set the mode for select a proxy [auto | random | only_http_proxy | only_socks5] (default "auto")
-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:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
$ 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

使用示例

假设你正在运行默认地址 “localhost:1080” 的 SOCKS5 代理,首先启动 graftcp-local

1
$ graftcp-local/graftcp-local

通过 graftcp 安装来自 golang.org 的 Go 包:

1
$ ./graftcp go get -v golang.org/x/net/proxy

通过 graftcp 打开 Chromium / Chrome / Firefox 浏览器,网页的所有请求都会重定向到 SOCKS5 代理:

1
$ ./graftcp chromium-browser

通过 graftcp 启动 Bash / Zsh / Fish,在这个新开的 shell 里面执行的任何新命令产生的 TCP 连接都会重定向到 SOCKS5 代理:

1
2
% ./graftcp bash
$ wget https://www.google.com

demo

工作原理

要达到重定向一个 app 发起的的 TCP 连接到其他目标地址并且该 app 本身对此毫无感知的目的,大概需要这些条件:

  • fork(2) 一个新进程,通过 execve(2) 启动该 app,并使用 ptrace(2) 进行跟踪,在 app 执行每一次 TCP 连接前,捕获并拦截这次 connect(2) 系统调用,获取目标地址的参数,并通过管道传给 graftcp-local
  • 修改这次 connect(2) 系统调用的目标地址参数为 graftcp-local 的地址,然后恢复执行被中断的系统调用。返回成功后,这个程序以为自己连的是原始的地址,但其实连的是 graftcp-local 的地址。这个就叫“移花接木”。
  • graftcp-local 根据连接信息和目标地址信息,与 SOCKS5 proxy 建立连接,把 app 的请求的数据重定向到 SOCKS5 proxy。

这里可能有个疑问:既然可以修改任何系统调用的参数,那么通过修改 app 的 write(2) / send(2) 的参数,直接往 buffer 里面附加原始目标地址信息给 graftcp-local 不是更简单吗?答案是这无法做到。如果直接往运行在子进程的被跟踪程序的 buffer 添加信息,可能会造成缓冲区溢出,造成程序崩溃或者覆盖了其他数据。
另外,execve(2) 会分离所有的共享内存,所以也不能通过共享内存的方式让被跟踪的 app 的 write buffer 携带更多的数据,因此这里采用管道方式给 graftcp-local 传递原始的目标地址信息。

简单的流程如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
+---------------+             +---------+         +--------+         +------+
| graftcp | dest host | | | | | |
| (tracer) +---PIPE----->| | | | | |
| ^ | info | | | | | |
| | ptrace | | | | | | |
| v | | | | | | |
| +---------+ | | | | | | |
| | | | connect | | connect | | connect | |
| | +--------------->| graftcp +-------->| SOCKS5 +-------->| dest |
| | | | | -local | | or | | host |
| | app | | req | | req | HTTP | req | |
| |(tracee) +--------------->| +-------->| proxy +-------->| |
| | | | | | | | | |
| | | | resp | | resp | | resp | |
| | |<---------------+ |<--------+ |<--------+ |
| +---------+ | | | | | | |
+---------------+ +---------+ +--------+ +------+

常见问题解答及技巧

有哪些重定向 TCP 连接的方式?

主要有: 全局式、设置环境变量式和仅针对程序(或进程)式。

全局式:比如使用 iptables + RedSocks 可以把系统符合一定规则的流量转换为 SOCKS5 流量。这种方式的优点是全局有效;缺点是所有满足该规则的流量都被重定向了,影响范围较大。

设置环境变量方式:一些程序启动时会读取 proxy 相关的环境变量来决定是否将自己的数据转换为对应代理协议的流量,比如 curl读取 http_proxy, ftp_proxy, all_proxy 环境变量并根据请求 scheme 来决定转换为哪种代理流量。这种方法只有程序本身实现了转换的功能才有效,局限性较大。

仅针对程序方式: 这种方式可以仅针对特定的程序执行重定向,比如 tsocksproxychains。如前面提到,它们之前都是使用 LD_PRELOAD 劫持动态库方式实现,对 Go 之类默认静态链接编译的程序就无效了。graftcp 改进了这一点,能够重定向任何程序的 TCP 连接。

如果应用程序连接的目标地址是本机,使用 graftcp 会把该连接重定向到 SOCKS5 代理吗?

不会。默认会忽略目标地址为本地的连接,如果想重定向所有地址的话,可以使用 -n选项。如果想忽略更多的地址,可以把它们加入黑名单 IP 文件;如果想仅重定向某些 IP 地址,可以把这些地址加入白名单 IP 文件。使用 graftcp --help 获取设置参数。

我的 DNS 请求受到污染,graftcp 会处理 DNS 请求吗?

不会。graftcp 目前仅处理 TCP 连接。建议使用 dnscrypt-proxyChinaDNS 等方式解决 DNS 污染问题。

clone(2) 参数有个叫 CLONE_UNTRACED 的标志位,可以避免让父进程跟踪到自己,graftcp 是如何做到强制跟踪的?

graftcp 在子进程调用 clone(2) 之前会把它拦截,清除这个 CLONE_UNTRACED 标志位,所以被跟踪的子进程最终还是难逃被跟踪的命运。另外,这个 CLONE_UNTRACED 标志位本意是给内核使用的,普通程序不应该去设置它。

Linux 提供了一种限制被 ptrace(2) 跟踪的方法:设置 /proc/sys/kernel/yama/ptrace_scope 的值,若 ptrace(2) 失效,请检查该值是否被修改过。

支持 macOS 吗?

不。macOS 的 ptrace(2) 是个半残品。不过理论上参考 DTrace那一套也能实现,见issue 12。或许有兴趣的同学可以趟下这趟浑水。

参考文档

  1. https://www.google.com

  2. https://github.com/hmgle/graftcp