Rooba
/
proxy.py
mirror of https://github.com/abhinavsingh/proxy.py.git
1
1
Fork 0
Code Issues Packages Projects Releases Wiki Activity
proxy.py/README.md

89 KiB
Raw Blame History

Proxy.Py

PyPi Monthly Docker Pulls No Dependencies Gitter License

Tested With MacOS, Ubuntu, Windows, Android, Android Emulator, iOS, iOS Simulator Android, Android Emulator iOS, iOS Simulator

pypi version Python 3.x Checked with mypy

doc codecov lib

Contributions Welcome Need Help Sponsored by Jaxl Innovations Private Limited

Table of Contents

Features

  • Fast & Scalable

    • Scale up by using all available cores on the system

    • Threadless executions using asyncio

    • Made to handle tens-of-thousands connections / sec

      # On Macbook Pro 2019 / 2.4 GHz 8-Core Intel Core i9 / 32 GB RAM
❯ ./helper/benchmark.sh
  CONCURRENCY: 100 workers, TOTAL REQUESTS: 100000 req

  Summary:
    Success rate:	1.0000
    Total:	2.5489 secs
    Slowest:	0.0443 secs
    Fastest:	0.0006 secs
    Average:	0.0025 secs
    Requests/sec:	39232.6572

    Total data:	1.81 MiB
    Size/request:	19 B
    Size/sec:	727.95 KiB

  Response time histogram:
    0.001 [5006]  |■■■■■
    0.001 [19740] |■■■■■■■■■■■■■■■■■■■■■
    0.002 [29701] |■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
    0.002 [21278] |■■■■■■■■■■■■■■■■■■■■■■
    0.003 [15376] |■■■■■■■■■■■■■■■■
    0.004 [6644]  |■■■■■■■
    0.004 [1609]  |■
    0.005 [434]   |
    0.006 [83]    |
    0.006 [29]    |
    0.007 [100]   |

  Latency distribution:
    10% in 0.0014 secs
    25% in 0.0018 secs
    50% in 0.0023 secs
    75% in 0.0030 secs
    90% in 0.0036 secs
    95% in 0.0040 secs
    99% in 0.0047 secs

  Details (average, fastest, slowest):
    DNS+dialup:	0.0025 secs, 0.0015 secs, 0.0030 secs
    DNS-lookup:	0.0000 secs, 0.0000 secs, 0.0001 secs

  Status code distribution:
    [200] 100000 responses

      Consult Threads vs Threadless and Threadless Remote vs Local Execution Mode to control number of CPU cores utilized.

      See Benchmark for more details and for how to run benchmarks locally.

  • Lightweight

    • Uses only ~5-20 MB RAM
      • No memory leaks
      • Start once and forget, no restarts required
    • Compressed containers size is only ~25 MB
    • No external dependency other than standard Python library

  • Programmable

    • Customize proxy behavior using Proxy Server Plugins. Example:
      • --plugins proxy.plugin.ProxyPoolPlugin
    • Enable builtin Web Server. Example:
      • --enable-web-server --plugins proxy.plugin.WebServerPlugin
    • Enable builtin Reverse Proxy Server. Example:
      • --enable-reverse-proxy --plugins proxy.plugin.ReverseProxyPlugin
    • Plugin API is currently in development phase. Expect breaking changes. See Deploying proxy.py in production on how to ensure reliability across code changes.

  • Can listen on multiple addresses and ports

    • Use --hostnames flag to provide additional addresses
    • Use --ports flag to provide additional ports
    • Optionally, use --port flag to override default port 8899
    • Capable of serving multiple protocols over the same port

  • Real-time Dashboard

  • Secure

  • Private

    • Protection against DNS based traffic blockers
    • Browse with malware and adult content protection enabled
    • See DNS-over-HTTPS

  • Man-In-The-Middle

    • Can decrypt TLS traffic between clients and upstream servers
    • See TLS Interception

  • Supported http protocols for proxy requests

    • http(s)
      • http1
      • http1.1 with pipeline
    • http2
    • websockets

  • Support for HAProxy Protocol

    • See --enable-proxy-protocol flag

  • Static file server support

    • See --enable-static-server and --static-server-dir flags

  • Optimized for large file uploads and downloads

    • See --client-recvbuf-size, --server-recvbuf-size, --max-sendbuf-size flags

  • IPv4 and IPv6 support

    • See --hostname flag

  • Unix domain socket support

    • See --unix-socket-path flag

  • Basic authentication support

    • See --basic-auth flag

  • PAC (Proxy Auto-configuration) support

    • See --pac-file and --pac-file-url-path flags

Install

Consult Deploying proxy.py in production when deploying production grade applications using proxy.py.

Using PIP

Stable Version with PIP

Install from PyPi

❯ pip install --upgrade proxy.py

or from GitHub master branch

❯ pip install git+https://github.com/abhinavsingh/proxy.py.git@master

Development Version with PIP

❯ pip install git+https://github.com/abhinavsingh/proxy.py.git@develop

Using Docker

Multi-platform containers are available via:

  • Docker Hub
    • latest tag points to last stable release
    • docker pull abhinavsingh/proxy.py:latest
  • GitHub container registry (GHCR)
    • latest tag points to last develop release
    • docker pull ghcr.io/abhinavsingh/proxy.py:latest

Stable version container releases are available for following platforms:

  • linux/386
  • linux/amd64
  • linux/arm/v6
  • linux/arm/v7
  • linux/arm64/v8
  • linux/ppc64le
  • linux/s390x

Stable Version from Docker Hub

Run proxy.py latest container:

❯ docker run -it -p 8899:8899 --rm abhinavsingh/proxy.py:latest

Docker daemon will automatically pull the matching platform image. To run specific target platform container on multi-platform supported servers:

❯ docker run -it -p 8899:8899 --rm --platform linux/arm64/v8 abhinavsingh/proxy.py:latest

Development Version from GHCR

Run proxy.py container from cutting edge code in the develop branch:

❯ docker run -it -p 8899:8899 --rm ghcr.io/abhinavsingh/proxy.py:latest

Build Development Version Locally

❯ git clone https://github.com/abhinavsingh/proxy.py.git
❯ cd proxy.py && make container
❯ docker run -it -p 8899:8899 --rm abhinavsingh/proxy.py:latest

WARNING docker image is currently broken on macOS due to incompatibility with vpnkit.

Using HomeBrew

Updated formulae for HomeBrew are maintained in develop branch under the helper/homebrew directory.

  • stable formulae installs the package from master branch.
  • develop formulae installs the package from develop branch.

Stable Version with HomeBrew

❯ brew install https://raw.githubusercontent.com/abhinavsingh/proxy.py/develop/helper/homebrew/stable/proxy.rb

Development Version with HomeBrew

❯ brew install https://raw.githubusercontent.com/abhinavsingh/proxy.py/develop/helper/homebrew/develop/proxy.rb

Start proxy.py

From command line when installed using PIP

When proxy.py is installed using pip, an executable named proxy is placed under your $PATH.

Run it

Simply type proxy on command line to start with default configuration.

❯ proxy
...[redacted]... - Loaded plugin proxy.http.proxy.HttpProxyPlugin
...[redacted]... - Started 8 threadless workers
...[redacted]... - Started 8 acceptors
...[redacted]... - Listening on 127.0.0.1:8899

Understanding logs

Things to notice from above logs:

  • Loaded plugin

    • proxy.py will load proxy.http.proxy.HttpProxyPlugin by default
    • As name suggests, this core plugin adds http(s) proxy server capabilities to proxy.py instance

  • Started N threadless workers

    • By default, proxy.py will start as many worker processes as there are CPU cores on the machine
    • Use --num-workers flag to customize number of worker processes
    • See Threads vs Threadless to understand how to control execution mode

  • Started N acceptors

    • By default, proxy.py will start as many acceptor processes as there are CPU cores on the machine
    • Use --num-acceptors flag to customize number of acceptor processes
    • See High Level Architecture to understand relationship between acceptors and workers

  • Started server on ::1:8899

    • By default, proxy.py listens on IPv6 ::1, which is equivalent of IPv4 127.0.0.1
    • If you want to access proxy.py from external host, use --hostname :: or --hostname 0.0.0.0 or bind to any other interface available on your machine.
    • See CustomNetworkInterface for how to customize proxy.py public IP seen by upstream servers.

  • Port 8899

    • Use --port flag to customize default TCP port.

Enable DEBUG logging

All the logs above are INFO level logs, default --log-level for proxy.py

Lets start proxy.py with DEBUG level logging:

❯ proxy --log-level d
...[redacted]... - Open file descriptor soft limit set to 1024
...[redacted]... - Loaded plugin proxy.http_proxy.HttpProxyPlugin
...[redacted]... - Started 8 workers
...[redacted]... - Started server on ::1:8899

You can use single letter to customize log level. Example:

  • d = DEBUG
  • i = INFO
  • w = WARNING
  • e = ERROR
  • c = CRITICAL

As we can see from the above logs, before starting up:

  • proxy.py tried to set open file limit ulimit on the system
  • Default value for --open-file-limit used is 1024
  • --open-file-limit flag is a no-op on Windows operating systems

See flags for full list of available configuration options.

From command line using repo source

If you are trying to run proxy.py from source code, there is no binary file named proxy in the source code.

To start proxy.py from source code follow these instructions:

  • Clone repo

    ❯ git clone https://github.com/abhinavsingh/proxy.py.git
❯ cd proxy.py

  • Create a Python 3 virtual env

    ❯ python3 -m venv venv
❯ source venv/bin/activate

  • Install deps

    ❯ make lib-dep

  • Generate proxy/common/_scm_version.py

    NOTE: Following step is not necessary for editable installs.

    This file writes SCM detected version to proxy/common/_scm_version.py file.

    ❯ ./write-scm-version.sh

  • Optionally, run tests

    ❯ make

  • Run proxy.py

    ❯ python -m proxy

See Plugin Developer and Contributor Guide if you plan to work with proxy.py source code.

Docker image

Customize startup flags

By default docker binary is started with IPv4 networking flags:

--hostname 0.0.0.0 --port 8899

You can override flag from command line when starting the docker container. For example, to check proxy.py version within the docker container, run:

❯ docker run -it \
    -p 8899:8899 \
    --rm abhinavsingh/proxy.py:latest \
    -v

Plugin Examples

  • See plugin module for full code.
  • All the bundled plugin examples also works with https traffic
  • Plugin examples are also bundled with Docker image.

HTTP Proxy Plugins

ShortLinkPlugin

Add support for short links in your favorite browsers / applications.

Shortlink Plugin

Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.ShortLinkPlugin

Now you can speed up your daily browsing experience by visiting your favorite website using single character domain names :). This works across all browsers.

Following short links are enabled by default:

Short Link Destination URL
a/ amazon.com
i/ instagram.com
l/ linkedin.com
f/ facebook.com
g/ google.com
t/ twitter.com
w/ web.whatsapp.com
y/ youtube.com
proxy/ localhost:8899

ModifyPostDataPlugin

Modifies POST request body before sending request to upstream server.

Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.ModifyPostDataPlugin

By default plugin replaces POST body content with hard-coded b'{"key": "modified"}' and enforced Content-Type: application/json.

Verify the same using curl -x localhost:8899 -d '{"key": "value"}' http://httpbin.org/post

{
  "args": {},
  "data": "{\"key\": \"modified\"}",
  "files": {},
  "form": {},
  "headers": {
    "Accept": "*/*",
    "Content-Length": "19",
    "Content-Type": "application/json",
    "Host": "httpbin.org",
    "User-Agent": "curl/7.54.0"
  },
  "json": {
    "key": "modified"
  },
  "origin": "1.2.3.4, 5.6.7.8",
  "url": "https://httpbin.org/post"
}

Note following from the response above:

  1. POST data was modified "data": "{\"key\": \"modified\"}". Original curl command data was {"key": "value"}.
  2. Our curl command did not add any Content-Type header, but our plugin did add one "Content-Type": "application/json". Same can also be verified by looking at json field in the output above: 
    "json": {
 "key": "modified"
},
  3. Our plugin also added a Content-Length header to match length of modified body.

MockRestApiPlugin

Mock responses for your server REST API. Use to test and develop client side applications without need of an actual upstream REST API server.

Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.ProposedRestApiPlugin

Verify mock API response using curl -x localhost:8899 http://api.example.com/v1/users/

{"count": 2, "next": null, "previous": null, "results": [{"email": "you@example.com", "groups": [], "url": "api.example.com/v1/users/1/", "username": "admin"}, {"email": "someone@example.com", "groups": [], "url": "api.example.com/v1/users/2/", "username": "admin"}]}

Verify the same by inspecting proxy.py logs:

... [redacted] ... - access_log:1210 - ::1:64792 - GET None:None/v1/users/ - None None - 0 byte

Access log shows None:None as server ip:port. None simply means that the server connection was never made, since response was returned by our plugin.

Now modify ProposedRestApiPlugin to returns REST API mock responses as expected by your clients.

RedirectToCustomServerPlugin

Redirects all incoming http requests to custom web server. By default, it redirects client requests to inbuilt web server, also running on 8899 port.

Start proxy.py and enable inbuilt web server:

❯ proxy \
    --enable-web-server \
    --plugins proxy.plugin.RedirectToCustomServerPlugin

Verify using curl -v -x localhost:8899 http://google.com

... [redacted] ...
< HTTP/1.1 404 NOT FOUND
< Server: proxy.py v1.0.0
< Connection: Close
<
* Closing connection 0

Above 404 response was returned from proxy.py web server.

Verify the same by inspecting the logs for proxy.py. Along with the proxy request log, you must also see a http web server request log.

... [redacted] ... - access_log:1241 - ::1:49525 - GET /
... [redacted] ... - access_log:1157 - ::1:49524 - GET localhost:8899/ - 404 NOT FOUND - 70 bytes

FilterByUpstreamHostPlugin

Drops traffic by inspecting upstream host. By default, plugin drops traffic for facebook.com and www.facebok.com.

Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.FilterByUpstreamHostPlugin

Verify using curl -v -x localhost:8899 http://facebook.com:

... [redacted] ...
< HTTP/1.1 418 I'm a tea pot
< Proxy-agent: proxy.py v1.0.0
* no chunk, no close, no size. Assume close to signal end
<
* Closing connection 0

Above 418 I'm a tea pot is sent by our plugin.

Verify the same by inspecting logs for proxy.py:

... [redacted] ... - handle_readables:1347 - HttpProtocolException type raised
Traceback (most recent call last):
... [redacted] ...
... [redacted] ... - access_log:1157 - ::1:49911 - GET None:None/ - None None - 0 bytes

CacheResponsesPlugin

Caches Upstream Server Responses.

Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.CacheResponsesPlugin

You may also use the --cache-requests flag to enable request packet caching for inspection.

Verify using curl -v -x localhost:8899 http://httpbin.org/get:

... [redacted] ...
< HTTP/1.1 200 OK
< Access-Control-Allow-Credentials: true
< Access-Control-Allow-Origin: *
< Content-Type: application/json
< Date: Wed, 25 Sep 2019 02:24:25 GMT
< Referrer-Policy: no-referrer-when-downgrade
< Server: nginx
< X-Content-Type-Options: nosniff
< X-Frame-Options: DENY
< X-XSS-Protection: 1; mode=block
< Content-Length: 202
< Connection: keep-alive
<
{
  "args": {},
  "headers": {
    "Accept": "*/*",
    "Host": "httpbin.org",
    "User-Agent": "curl/7.54.0"
  },
  "origin": "1.2.3.4, 5.6.7.8",
  "url": "https://httpbin.org/get"
}
* Connection #0 to host localhost left intact

Get path to the cache file from proxy.py logs:

... [redacted] ... - GET httpbin.org:80/get - 200 OK - 556 bytes
... [redacted] ... - Cached response at /var/folders/k9/x93q0_xn1ls9zy76m2mf2k_00000gn/T/httpbin.org-1569378301.407512.txt

Verify contents of the cache file cat /path/to/your/cache/httpbin.org.txt

HTTP/1.1 200 OK
Access-Control-Allow-Credentials: true
Access-Control-Allow-Origin: *
Content-Type: application/json
Date: Wed, 25 Sep 2019 02:24:25 GMT
Referrer-Policy: no-referrer-when-downgrade
Server: nginx
X-Content-Type-Options: nosniff
X-Frame-Options: DENY
X-XSS-Protection: 1; mode=block
Content-Length: 202
Connection: keep-alive

{
  "args": {},
  "headers": {
    "Accept": "*/*",
    "Host": "httpbin.org",
    "User-Agent": "curl/7.54.0"
  },
  "origin": "1.2.3.4, 5.6.7.8",
  "url": "https://httpbin.org/get"
}

CacheByResponseType

CacheResponsesPlugin plugin can also automatically cache responses by content-type. To try this, you must be running under TLS Interception mode and then pass --cache-by-content-type flag. Example:

❯ proxy \
    --plugins proxy.plugin.CacheResponsesPlugin \
    --cache-by-content-type \
    --ca-key-file ca-key.pem \
    --ca-cert-file ca-cert.pem \
    --ca-signing-key ca-signing-key.pem

Make a few requests to the proxy server and you shall see data under ~/.proxy/cache directory.

You should see 2 folders:

  • content: Contains parsed jpg, css, js, html, pdf etc by content type
  • responses: Contains raw responses as received (of-course decrypted because of interception)

ManInTheMiddlePlugin

Modifies upstream server responses.

Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.ManInTheMiddlePlugin

Verify using curl -v -x localhost:8899 http://google.com:

... [redacted] ...
< HTTP/1.1 200 OK
< Content-Length: 28
<
* Connection #0 to host localhost left intact
Hello from man in the middle

Response body Hello from man in the middle is sent by our plugin.

ProxyPoolPlugin

Forward incoming proxy requests to a set of upstream proxy servers.

Let's start 2 upstream proxies first. To simulate upstream proxies, start proxy.py on port 9000 and 9001

❯ proxy --port 9000

❯ proxy --port 9001

Now, start proxy.py with ProxyPoolPlugin (on default 8899 port), pointing to our upstream proxies at 9000 and 9001 port.

❯ proxy \
    --plugins proxy.plugin.ProxyPoolPlugin \
    --proxy-pool localhost:9000 \
    --proxy-pool localhost:9001

Make a curl request via 8899 proxy:

curl -v -x localhost:8899 http://httpbin.org/get

Verify that 8899 proxy forwards requests to upstream proxies by checking respective logs.

If an upstream proxy require credentials, pass them as arguments. Example:

--proxy-pool user:pass@upstream.proxy:port

FilterByClientIpPlugin

Reject traffic from specific IP addresses. By default this plugin blocks traffic from 127.0.0.1 and ::1.

Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.FilterByClientIpPlugin

Send a request using curl -v -x localhost:8899 http://google.com:

... [redacted] ...
> Proxy-Connection: Keep-Alive
>
< HTTP/1.1 418 I'm a tea pot
< Connection: close
<
* Closing connection 0

Modify plugin to your taste e.g. Allow specific IP addresses only.

ModifyChunkResponsePlugin

This plugin demonstrate how to modify chunked encoded responses. In able to do so, this plugin uses proxy.py core to parse the chunked encoded response. Then we reconstruct the response using custom hard-coded chunks, ignoring original chunks received from upstream server.

Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.ModifyChunkResponsePlugin

Verify using curl -v -x localhost:8899 http://httpbin.org/stream/5:

... [redacted] ...
modify
chunk
response
plugin
* Connection #0 to host localhost left intact
* Closing connection 0

Modify ModifyChunkResponsePlugin to your taste. Example, instead of sending hard-coded chunks, parse and modify the original JSON chunks received from the upstream server.

CloudflareDnsResolverPlugin

This plugin uses Cloudflare hosted DNS-over-HTTPS API (json).

DoH mandates a HTTP2 compliant client. Unfortunately proxy.py does not provide that yet, so we use a dependency. Install it:

❯ pip install "httpx[http2]"

Now start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.CloudflareDnsResolverPlugin

By default, CloudflareDnsResolverPlugin runs in security mode and provides malware protection. Use --cloudflare-dns-mode family to also enable adult content protection too.

CustomDnsResolverPlugin

This plugin demonstrate how to use a custom DNS resolution implementation with proxy.py. This example plugin currently uses Python's in-built resolution mechanism. Customize code to your taste. Example, query your custom DNS server, implement DoH or other mechanisms.

Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.CustomDnsResolverPlugin

CustomNetworkInterface

HttpProxyBasePlugin.resolve_dns callback can also be used to configure network interface which must be used as the source_address for connection to the upstream server.

See this thread for more details.

PS: There is no plugin named, but CustomDnsResolverPlugin can be easily customized according to your needs.

ProgramNamePlugin

Attempts to resolve program (application) name for proxy requests originating from the local machine. If identified, client IP in the access logs is replaced with program name.

Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.ProgramNamePlugin

Make a request using curl:

❯ curl -v -x localhost:8899 https://httpbin.org/get

You must see log lines like this:

... [redacted] ... - [I] server.access_log:419 - curl:58096 - CONNECT httpbin.org:443 - 6010 bytes - 1824.62ms

Notice curl in-place of ::1 or 127.0.0.1 as client IP.

WARNING If ProgramNamePlugin does not work reliably on your operating system, kindly contribute by sending a pull request and/or open an issue. Thank you!!!

HTTP Web Server Plugins

Web Server Route

Demonstrates inbuilt web server routing using plugin.

Start proxy.py as:

❯ proxy --enable-web-server \
    --plugins proxy.plugin.WebServerPlugin

Verify using curl -v localhost:8899/http-route-example, should return:

HTTP route response

Reverse Proxy Plugins

Extends in-built Web Server to add Reverse Proxy capabilities.

Reverse Proxy

Start proxy.py as:

❯ proxy --enable-reverse-proxy \
    --plugins proxy.plugin.ReverseProxyPlugin

With default configuration, ReverseProxyPlugin plugin is equivalent to following Nginx config:

location /get {
    proxy_pass http://httpbin.org/get
}

Verify using curl -v localhost:8899/get:

{
  "args": {},
  "headers": {
    "Accept": "*/*",
    "Host": "localhost",
    "User-Agent": "curl/7.64.1"
  },
  "origin": "1.2.3.4, 5.6.7.8",
  "url": "https://localhost/get"
}

Plugin Ordering

When using multiple plugins, depending upon plugin functionality, it might be worth considering the order in which plugins are passed on the command line.

Plugins are called in the same order as they are passed. Example, say we are using both FilterByUpstreamHostPlugin and RedirectToCustomServerPlugin. Idea is to drop all incoming http requests for facebook.com and www.facebook.com and redirect other http requests to our inbuilt web server.

Hence, in this scenario it is important to use FilterByUpstreamHostPlugin before RedirectToCustomServerPlugin. If we enable RedirectToCustomServerPlugin before FilterByUpstreamHostPlugin, facebook requests will also get redirected to inbuilt web server, instead of being dropped.

End-to-End Encryption

By default, proxy.py uses http protocol for communication with clients e.g. curl, browser. For enabling end-to-end encrypting using tls / https first generate certificates. Checkout the repository and run:

make https-certificates

Start proxy.py as:

❯ proxy \
    --cert-file https-cert.pem \
    --key-file https-key.pem

Verify using curl -x https://localhost:8899 --proxy-cacert https-cert.pem https://httpbin.org/get:

{
  "args": {},
  "headers": {
    "Accept": "*/*",
    "Host": "httpbin.org",
    "User-Agent": "curl/7.54.0"
  },
  "origin": "1.2.3.4, 5.6.7.8",
  "url": "https://httpbin.org/get"
}

If you want to avoid passing --proxy-cacert flag, also consider signing generated SSL certificates. Example:

First, generate CA certificates:

make ca-certificates

Then, sign SSL certificate:

make sign-https-certificates

Now restart the server with --cert-file https-signed-cert.pem flag. Note that you must also trust generated ca-cert.pem in your system keychain.

TLS Interception

By default, proxy.py will not decrypt https traffic between client and server. To enable TLS interception first generate root CA certificates:

❯ make ca-certificates

Lets also enable CacheResponsePlugin so that we can verify decrypted response from the server. Start proxy.py as:

❯ proxy \
    --plugins proxy.plugin.CacheResponsesPlugin \
    --ca-key-file ca-key.pem \
    --ca-cert-file ca-cert.pem \
    --ca-signing-key-file ca-signing-key.pem

NOTE Also provide explicit CA bundle path needed for validation of peer certificates. See --ca-file flag.

Verify TLS interception using curl

❯ curl -v -x localhost:8899 --cacert ca-cert.pem https://httpbin.org/get

*  issuer: C=US; ST=CA; L=SanFrancisco; O=proxy.py; OU=CA; CN=Proxy PY CA; emailAddress=proxyca@mailserver.com
*  SSL certificate verify ok.
> GET /get HTTP/1.1
... [redacted] ...
< Connection: keep-alive
<
{
  "args": {},
  "headers": {
    "Accept": "*/*",
    "Host": "httpbin.org",
    "User-Agent": "curl/7.54.0"
  },
  "origin": "1.2.3.4, 5.6.7.8",
  "url": "https://httpbin.org/get"
}

The issuer line confirms that response was intercepted.

Also verify the contents of cached response file. Get path to the cache file from proxy.py logs.

❯ cat /path/to/your/tmp/directory/httpbin.org-1569452863.924174.txt

HTTP/1.1 200 OK
Access-Control-Allow-Credentials: true
Access-Control-Allow-Origin: *
Content-Type: application/json
Date: Wed, 25 Sep 2019 23:07:05 GMT
Referrer-Policy: no-referrer-when-downgrade
Server: nginx
X-Content-Type-Options: nosniff
X-Frame-Options: DENY
X-XSS-Protection: 1; mode=block
Content-Length: 202
Connection: keep-alive

{
  "args": {},
  "headers": {
    "Accept": "*/*",
    "Host": "httpbin.org",
    "User-Agent": "curl/7.54.0"
  },
  "origin": "1.2.3.4, 5.6.7.8",
  "url": "https://httpbin.org/get"
}

Viola!!! If you remove CA flags, encrypted data will be found in the cached file instead of plain text.

Now use CA flags with other plugin examples to see them work with https traffic.

TLS Interception With Docker

Important notes about TLS Interception with Docker container:

  • Since v2.2.0, proxy.py docker container also ships with openssl. This allows proxy.py to generate certificates on the fly for TLS Interception.

  • For security reasons, proxy.py docker container does not ship with CA certificates.

Here is how to start a proxy.py docker container with TLS Interception:

  1. Generate CA certificates on host computer

    ❯ make ca-certificates

  2. Copy all generated certificates into a separate directory. We'll later mount this directory into our docker container

    ❯ mkdir /tmp/ca-certificates
❯ cp ca-cert.pem ca-key.pem ca-signing-key.pem /tmp/ca-certificates

  3. Start docker container

    ❯ docker run -it --rm \
    -v /tmp/ca-certificates:/tmp/ca-certificates \
    -p 8899:8899 \
    abhinavsingh/proxy.py:latest \
    --hostname 0.0.0.0 \
    --plugins proxy.plugin.CacheResponsesPlugin \
    --ca-key-file /tmp/ca-certificates/ca-key.pem \
    --ca-cert-file /tmp/ca-certificates/ca-cert.pem \
    --ca-signing-key /tmp/ca-certificates/ca-signing-key.pem
    • -v /tmp/ca-certificates:/tmp/ca-certificates flag mounts our CA certificate directory in container environment
    • --plugins proxy.plugin.CacheResponsesPlugin enables CacheResponsesPlugin so that we can inspect intercepted traffic
    • --ca-* flags enable TLS Interception.

  4. From another terminal, try TLS Interception using curl. You can omit --cacert flag if CA certificate is already trusted by the system.

    ❯ curl -v \
    --cacert ca-cert.pem \
    -x 127.0.0.1:8899 \
    https://httpbin.org/get

  5. Verify issuer field from response headers.

    * Server certificate:
*  subject: CN=httpbin.org; C=NA; ST=Unavailable; L=Unavailable; O=Unavailable; OU=Unavailable
*  start date: Jun 17 09:26:57 2020 GMT
*  expire date: Jun 17 09:26:57 2022 GMT
*  subjectAltName: host "httpbin.org" matched cert's "httpbin.org"
*  issuer: CN=example.com
*  SSL certificate verify ok.

  6. Back on docker terminal, copy response dump path logs.

    ...[redacted]... [I] access_log:338 - 172.17.0.1:56498 - CONNECT httpbin.org:443 - 1031 bytes - 1216.70 ms
...[redacted]... [I] close:49 - Cached response at /tmp/httpbin.org-ae1a927d064e4ab386ea319eb38fe251.txt

  7. In another terminal, cat the response dump:

    ❯ docker exec -it $(docker ps | grep proxy.py | awk '{ print $1 }') cat /tmp/httpbin.org-ae1a927d064e4ab386ea319eb38fe251.txt
HTTP/1.1 200 OK
...[redacted]...
{
  ...[redacted]...,
  "url": "http://httpbin.org/get"
}

Proxy Over SSH Tunnel

This is a WIP and may not work as documented

Requires paramiko to work.

See requirements-tunnel.txt

Proxy Remote Requests Locally

                        |
+------------+          |            +----------+
|   LOCAL    |          |            |  REMOTE  |
|   HOST     | <== SSH ==== :8900 == |  SERVER  |
+------------+          |            +----------+
:8899 proxy.py          |
                        |
                     FIREWALL
                  (allow tcp/22)

What

Proxy HTTP(s) requests made on a remote server through proxy.py server running on localhost.

How

  • Requested remote port is forwarded over the SSH connection.
  • proxy.py running on the localhost handles and responds to remote proxy requests.

Requirements

  1. localhost MUST have SSH access to the remote server
  2. remote server MUST be configured to proxy HTTP(s) requests through the forwarded port number e.g. :8900.
    • remote and localhost ports CAN be same e.g. :8899.
    • :8900 is chosen in ascii art for differentiation purposes.

Try it

Start proxy.py as:

❯ # On localhost
❯ proxy --enable-tunnel \
    --tunnel-username username \
    --tunnel-hostname ip.address.or.domain.name \
    --tunnel-port 22 \
    --tunnel-remote-port 8899 \
    --tunnel-ssh-key /path/to/ssh/private.key \
    --tunnel-ssh-key-passphrase XXXXX
...[redacted]... [I] listener.setup:97 - Listening on 127.0.0.1:8899
...[redacted]... [I] pool.setup:106 - Started 16 acceptors in threadless (local) mode
...[redacted]... [I] transport._log:1873 - Connected (version 2.0, client OpenSSH_7.6p1)
...[redacted]... [I] transport._log:1873 - Authentication (publickey) successful!
...[redacted]... [I] listener.setup:116 - SSH connection established to ip.address.or.domain.name:22...
...[redacted]... [I] listener.start_port_forward:91 - :8899 forwarding successful...

Make a HTTP proxy request on remote server and verify that response contains public IP address of localhost as origin:

❯ # On remote
❯ curl -x 127.0.0.1:8899 http://httpbin.org/get
{
  "args": {},
  "headers": {
    "Accept": "*/*",
    "Host": "httpbin.org",
    "User-Agent": "curl/7.54.0"
  },
  "origin": "x.x.x.x, y.y.y.y",
  "url": "https://httpbin.org/get"
}

Also, verify that proxy.py logs on localhost contains remote IP as client IP.

access_log:328 - remote:52067 - GET httpbin.org:80

Proxy Local Requests Remotely

                        |
+------------+          |     +----------+
|   LOCAL    |          |     |  REMOTE  |
|   HOST     | === SSH =====> |  SERVER  |
+------------+          |     +----------+
                        |     :8899 proxy.py
                        |
                    FIREWALL
                 (allow tcp/22)

Not planned.

If you have a valid use case, kindly open an issue. You are always welcome to send contributions via pull-requests to add this functionality :)

To proxy local requests remotely, make use of Proxy Pool Plugin.

Embed proxy.py

Blocking Mode

Start proxy.py in embedded mode with default configuration by using proxy.main method. Example:

import proxy

if __name__ == '__main__':
  proxy.main()

Customize startup flags by passing them as kwargs:

import ipaddress
import proxy

if __name__ == '__main__':
  proxy.main(
    hostname=ipaddress.IPv6Address('::1'),
    port=8899
  )

Note that:

  1. main is equivalent to starting proxy.py from command line.
  2. main does not accept any args (only kwargs).
  3. main will automatically consume any available sys.argv as args.
  4. main will block until proxy.py shuts down.

Non-blocking Mode

Start proxy.py in non-blocking embedded mode with default configuration by using Proxy context manager: Example:

import proxy

if __name__ == '__main__':
  with proxy.Proxy() as p:
    # Uncomment the line below and
    # implement your app your logic here
    proxy.sleep_loop()

Note that:

  1. Proxy is similar to main, except Proxy will not block.
  2. Internally, Proxy is a context manager which will start proxy.py when called and will shut it down once the scope ends.
  3. Unlike main, startup flags with Proxy can also be customized by using args and kwargs. e.g. Proxy(['--port', '8899']) or by using passing flags as kwargs e.g. Proxy(port=8899).
  4. Unlike main, Proxy will not inspect sys.argv.

Ephemeral Port

Use --port=0 to bind proxy.py on a random port allocated by the kernel.

In embedded mode, you can access this port. Example:

import proxy

if __name__ == '__main__':
  with proxy.Proxy() as p:
    print(p.flags.port)
    proxy.sleep_loop()

flags.port will give you access to the random port allocated by the kernel.

Loading Plugins

Users can use --plugins flag multiple times to load multiple plugins. See Unable to load plugins if you are running into issues.

When using in embedded mode, you have a few more options. Example:

  1. Provide a fully-qualified name of the plugin class as bytes to the proxy.main method or proxy.Proxy context manager.
  2. Provide type instance of the plugin class. This is especially useful if you plan to define plugins at runtime.

Example, load a single plugin using --plugins flag:

import proxy

if __name__ == '__main__':
  proxy.main(plugins=['proxy.plugin.CacheResponsesPlugin'])

For simplicity, you can also pass the list of plugins as a keyword argument to proxy.main or the Proxy constructor.

Example:

import proxy
from proxy.plugin import FilterByUpstreamHostPlugin

if __name__ == '__main__':
  proxy.main(plugins=[
    b'proxy.plugin.CacheResponsesPlugin',
    FilterByUpstreamHostPlugin,
  ])

Unit testing with proxy.py

proxy.TestCase

To setup and tear down proxy.py for your Python unittest classes, simply use proxy.TestCase instead of unittest.TestCase. Example:

import proxy

class TestProxyPyEmbedded(proxy.TestCase):

    def test_my_application_with_proxy(self) -> None:
        self.assertTrue(True)

Note that:

  1. proxy.TestCase overrides unittest.TestCase.run() method to setup and tear down proxy.py.
  2. proxy.py server will listen on a random available port on the system. This random port is available as self.PROXY.flags.port within your test cases.
  3. Only a single acceptor and worker is started by default (--num-workers 1 --num-acceptors 1) for faster setup and tear down.
  4. Most importantly, proxy.TestCase also ensures proxy.py server is up and running before proceeding with execution of tests. By default, proxy.TestCase will wait for 10 seconds for proxy.py server to start, upon failure a TimeoutError exception will be raised.

Override startup flags

To override default startup flags, define a PROXY_PY_STARTUP_FLAGS variable in your test class. Example:

class TestProxyPyEmbedded(TestCase):

    PROXY_PY_STARTUP_FLAGS = [
        '--num-workers', '2',
        '--num-acceptors', '1',
        '--enable-web-server',
    ]

    def test_my_application_with_proxy(self) -> None:
        self.assertTrue(True)

See test_embed.py for full working example.

With unittest.TestCase

If for some reasons you are unable to directly use proxy.TestCase, then simply override unittest.TestCase.run yourself to setup and tear down proxy.py. Example:

import unittest
import proxy


class TestProxyPyEmbedded(unittest.TestCase):

    def test_my_application_with_proxy(self) -> None:
        self.assertTrue(True)

    def run(self, result: Optional[unittest.TestResult] = None) -> Any:
        with proxy.start([
                '--num-workers', '1',
                '--num-acceptors', '1',
                '--port', '... random port ...']):
            super().run(result)

or simply setup / tear down proxy.py within setUpClass and teardownClass class methods.

Utilities

TCP Sockets

new_socket_connection

Attempts to create an IPv4 connection, then IPv6 and finally a dual stack connection to provided address.

>>> conn = new_socket_connection(('httpbin.org', 80))
>>> ...[ use connection ]...
>>> conn.close()

socket_connection

socket_connection is a convenient decorator + context manager around new_socket_connection which ensures conn.close is implicit.

As a context manager:

>>> with socket_connection(('httpbin.org', 80)) as conn:
>>>   ... [ use connection ] ...

As a decorator:

>>> @socket_connection(('httpbin.org', 80))
>>> def my_api_call(conn, *args, **kwargs):
>>>   ... [ use connection ] ...

HTTP Client

build_http_request

  • Generate HTTP GET request

    >>> build_http_request(b'GET', b'/')
b'GET / HTTP/1.1\r\n\r\n'

  • Generate HTTP GET request with headers

    >>> build_http_request(b'GET', b'/', conn_close=True)
b'GET / HTTP/1.1\r\nConnection: close\r\n\r\n'

  • Generate HTTP POST request with headers and body

    >>> import json
>>> build_http_request(b'POST', b'/form',
        headers={b'Content-type': b'application/json'},
        body=proxy.bytes_(json.dumps({'email': 'hello@world.com'})))
    b'POST /form HTTP/1.1\r\nContent-type: application/json\r\n\r\n{"email": "hello@world.com"}'

build_http_response

build_http_response(
    status_code: int,
    protocol_version: bytes = HTTP_1_1,
    reason: Optional[bytes] = None,
    headers: Optional[Dict[bytes, bytes]] = None,
    body: Optional[bytes] = None) -> bytes

PKI

API Usage

  • gen_private_key

    gen_private_key(
    key_path: str,
    password: str,
    bits: int = 2048,
    timeout: int = 10) -> bool

  • gen_public_key

    gen_public_key(
    public_key_path: str,
    private_key_path: str,
    private_key_password: str,
    subject: str,
    alt_subj_names: Optional[List[str]] = None,
    extended_key_usage: Optional[str] = None,
    validity_in_days: int = 365,
    timeout: int = 10) -> bool

  • remove_passphrase

    remove_passphrase(
    key_in_path: str,
    password: str,
    key_out_path: str,
    timeout: int = 10) -> bool

  • gen_csr

    gen_csr(
    csr_path: str,
    key_path: str,
    password: str,
    crt_path: str,
    timeout: int = 10) -> bool

  • sign_csr

    sign_csr(
    csr_path: str,
    crt_path: str,
    ca_key_path: str,
    ca_key_password: str,
    ca_crt_path: str,
    serial: str,
    alt_subj_names: Optional[List[str]] = None,
    extended_key_usage: Optional[str] = None,
    validity_in_days: int = 365,
    timeout: int = 10) -> bool

See pki.py and test_pki.py for usage examples.

CLI Usage

Use proxy.common.pki module for:

  1. Generation of public and private keys
  2. Generating CSR requests
  3. Signing CSR requests using custom CA.
❯ python -m proxy.common.pki -h
usage: pki.py [-h] [--password PASSWORD] [--private-key-path PRIVATE_KEY_PATH] [--public-key-path PUBLIC_KEY_PATH]
              [--subject SUBJECT] [--csr-path CSR_PATH] [--crt-path CRT_PATH] [--hostname HOSTNAME] [--openssl OPENSSL]
              action

proxy.py v2.4.4rc2.dev12+gdc06ea4 : PKI Utility

positional arguments:
  action                Valid actions: remove_passphrase, gen_private_key, gen_public_key, gen_csr, sign_csr

options:
  -h, --help            show this help message and exit
  --password PASSWORD   Password to use for encryption. Default: proxy.py
  --private-key-path PRIVATE_KEY_PATH
                        Private key path
  --public-key-path PUBLIC_KEY_PATH
                        Public key path
  --subject SUBJECT     Subject to use for public key generation. Default: /CN=localhost
  --csr-path CSR_PATH   CSR file path. Use with gen_csr and sign_csr action.
  --crt-path CRT_PATH   Signed certificate path. Use with sign_csr action.
  --hostname HOSTNAME   Alternative subject names to use during CSR signing.
  --openssl OPENSSL     Path to openssl binary. By default, we assume openssl is in your PATH

Internal Documentation

Read The Doc

make lib-doc

pydoc

Code is well documented. Grab the source code and run:

pydoc3 proxy

pyreverse

Generate class level hierarchy UML diagrams for in-depth analysis:

make lib-pyreverse

Run Dashboard

Dashboard is currently under development and not yet bundled with pip packages. To run dashboard, you must checkout the source.

Dashboard is written in Typescript and SCSS, so let's build it first using:

❯ make dashboard

Also build the embedded Chrome DevTools if you plan on using it:

❯ make devtools

Now start proxy.py with dashboard plugin and by overriding root directory for static server:

❯ proxy --enable-dashboard --static-server-dir dashboard/public
...[redacted]... - Loaded plugin proxy.http.server.HttpWebServerPlugin
...[redacted]... - Loaded plugin proxy.dashboard.dashboard.ProxyDashboard
...[redacted]... - Loaded plugin proxy.dashboard.inspect_traffic.InspectTrafficPlugin
...[redacted]... - Loaded plugin proxy.http.inspector.DevtoolsProtocolPlugin
...[redacted]... - Loaded plugin proxy.http.proxy.HttpProxyPlugin
...[redacted]... - Listening on ::1:8899
...[redacted]... - Core Event enabled

Currently, enabling dashboard will also enable all the dashboard plugins.

Visit dashboard:

❯ open http://localhost:8899/dashboard/

Inspect Traffic

This is a WIP and may not work as documented

Wait for embedded Chrome Dev Console to load. Currently, detail about all traffic flowing through proxy.py is pushed to the Inspect Traffic tab. However, received payloads are not yet integrated with the embedded developer console.

Current functionality can be verified by opening the Dev Console of dashboard and inspecting the websocket connection that dashboard established with the proxy.py server.

Proxy.Py Dashboard Inspect Traffic

Chrome DevTools Protocol

For scenarios where you want direct access to Chrome DevTools protocol websocket endpoint, start proxy.py as:

❯ proxy --enable-devtools --enable-events

Now point your CDT instance to ws://localhost:8899/devtools.

Frequently Asked Questions

Deploying proxy.py in production

Listed below are a few strategies for using proxy.py in your private/production/corporate projects.

What not to do?

You MUST avoid forking the repository "just" to put your plugin code in proxy/plugin directory. Forking is recommended workflow for project contributors, NOT for project users.

  • Instead, use one of the suggested approaches from below.
  • Then load your plugins using --plugin, --plugins flags or plugin kwargs.
  • See skeleton app for example standalone project using proxy.py.

Via Requirements

It is highly recommended that you use proxy.py via requirements.txt or similar dependency management setups. This will allow you to take advantages of regular performance updates, bug fixes, security patches and other improvements happening in the proxy.py ecosystem. Example:

  1. Use --pre option to depend upon last pre-release

    ❯ pip install proxy.py --pre

    Pre-releases are similar to depending upon develop branch code, just that pre-releases may not point to the HEAD. This could happen because pre-releases are NOT made available on PyPi after every PR merge.

  2. Use TestPyPi with --pre option to depend upon develop branch code

    ❯ pip install -i https://test.pypi.org/simple/ proxy.py --pre

    A pre-release is made available on TestPyPi after every PR merge.

  3. Use last stable release code

    As usual, simply use:

    ❯ pip install proxy.py

Via Docker Container

If you are into deploying containers, then simply build your image from base proxy.py container images.

  1. Use GHCR to build from develop branch code:

    FROM ghcr.io/abhinavsingh/proxy.py:latest as base

    PS: I use GHCR latest for several production level projects

  2. Use DockerHub to build from last stable release code:

    FROM abhinavsingh/proxy.py:latest as base

PS: IMHO, container based strategy is the best approach and the only strategy that I use myself.

Integrate your CI/CD with proxy.py

Hey, but you keep making breaking changes in the develop branch.

I hear you. And hence, for your production grade applications, you MUST integrate application CI/CD with proxy.py. You must make sure that your application builds and passes its tests for every PR merge into the proxy.py upstream repo.

If your application repository is public, in certain scenarios, PR authors may send patch PRs for all dependents to maintain backward incompatibility and green CI/CD.

CI/CD integration ensure your app continues to build with latest proxy.py code. Depending upon where you host your code, use the strategy listed below:

  • GitHub

    TBD

  • Google Cloud Build

    TBD

  • AWS

    TBD

  • Azure

    TBD

  • Others

    TBD

At some stage, we'll deprecate master branch segregation and simply maintain a develop branch. As dependents can maintain stability via CI/CD integrations. Currently, it's hard for a production grade project to blindly depend upon develop branch.

Stable vs Develop

  • master branch contains latest stable code and is available via PyPi repository and Docker containers via docker.io and ghcr.io registries.

    Issues reported for stable releases are considered with top-priority. However, currently we don't back port fixes into older releases. Example, if you reported an issue in v2.3.1, but current master branch now contains v2.4.0rc1. Then, the fix will land in v2.4.0rc2.

  • develop branch contains cutting edge changes

    Development branch is kept stable (most of the times). But, if you want 100% reliability and serving users in production environment, ALWAYS use the stable version.

Release Schedule

A vX.Y.ZrcN pull request is created once a month which merges develop → master. Find below how code flows from a pull request to the next stable release.

  1. Development release is deployed from develop → test.pypi.org after every pull request merge

  2. Alpha release is deployed from develop → pypi.org before merging the vX.Y.Z.rcN pull request from develop → master branch. There can be multiple alpha releases made before merging the rc pull request

  3. Beta release is deployed from master → pypi.org. Beta releases are made in preparation of rc releases and can be skipped if unnecessary

  4. Release candidate is deployed from master → pypi.org. Release candidates are always made available before final stable release

  5. Stable release is deployed from master → pypi.org

Threads vs Threadless

v1.x

proxy.py used to spawn new threads for handling client requests.

v2.0+

proxy.py added support for threadless execution of client requests using asyncio.

v2.4.0+

Threadless execution was turned ON by default for Python 3.8+ on mac and linux environments.

proxy.py threadless execution has been reported safe on these environments by our users. If you are running into trouble, fallback to threaded mode using --threaded flag.

For windows and Python < 3.8, you can still try out threadless mode by starting proxy.py with --threadless flag.

If threadless works for you, consider sending a PR by editing _env_threadless_compliant method in the proxy/common/constants.py file.

Threadless Remote vs Local execution mode

Original threadless implementation used remote execution mode. This is also depicted under High level architecture as ASCII art.

Under remote execution mode, acceptors delegate incoming client connection processing to a remote worker process. By default, acceptors delegate connections in round-robin fashion. Worker processing the request may or may not be running on the same CPU core as the acceptor. This architecture scales well for high throughput, but results in spawning two process per CPU core.

Example, if there are N-CPUs on the machine, by default, N acceptors and N worker processes are started. You can tune number of processes using --num-acceptors and --num-workers flag. You might want more workers than acceptors or vice versa depending upon your use case.

In v2.4.x, local execution mode was added, mainly to reduce number of processes spawned by default. This model serves well for day-to-day single user use cases and for developer testing scenarios. Under local execution mode, acceptors delegate client connections to a companion thread, instead of a remote process. local execution mode ensure CPU affinity, unlike in the remote mode where acceptor and worker might be running on different CPU cores.

--local-executor 1 was made default in v2.4.x series. Under local execution mode, --num-workers flag has no effect, as no remote workers are started.

To use remote execution mode, use --local-executor 0 flag. Then use --num-workers to tune number of worker processes.

SyntaxError: invalid syntax

proxy.py is strictly typed and uses Python typing annotations. Example:

>>> my_strings : List[str] = []
>>> #############^^^^^^^^^#####

Hence a Python version that understands typing annotations is required. Make sure you are using Python 3.6+.

Verify the version before running proxy.py:

❯ python --version

All typing annotations can be replaced with comment-only annotations. Example:

>>> my_strings = [] # List[str]
>>> ################^^^^^^^^^^^

It will enable proxy.py to run on Python pre-3.6, even on 2.7. However, as all future versions of Python will support typing annotations, this has not been considered.

Unable to load plugins

Make sure plugin modules are discoverable by adding them to PYTHONPATH. Example:

PYTHONPATH=/path/to/my/app proxy --plugins my_app.proxyPlugin

...[redacted]... - Loaded plugin proxy.HttpProxyPlugin
...[redacted]... - Loaded plugin my_app.proxyPlugin

OR, simply pass fully-qualified path as parameter, e.g.

proxy --plugins /path/to/my/app/my_app.proxyPlugin

Here is a quick working example:

  • Contents of /tmp/plug folder
╰─ ls -1 /tmp/plug                                                                                                                       ─╯
my_plugin.py
  • Custom MyPlugin class
╰─ cat /tmp/plug/my_plugin.py                                                                                                            ─╯
from proxy.http.proxy import HttpProxyBasePlugin


class MyPlugin(HttpProxyBasePlugin):
  pass

This is an empty plugin for demonstrating external plugin usage. You must implement necessary methods to make your plugins work for real traffic

  • Start proxy.py with MyPlugin
╰─ PYTHONPATH=/tmp/plug proxy --plugin my_plugin.MyPlugin                                                                      ─╯
...[redacted]... - Loaded plugin proxy.http.proxy.HttpProxyPlugin
...[redacted]... - Loaded plugin my_plugin.MyPlugin
...[redacted]... - Listening on ::1:8899

Unable to connect with proxy.py from remote host

Make sure proxy.py is listening on correct network interface. Try following flags:

  • For IPv6 --hostname ::
  • For IPv4 --hostname 0.0.0.0

Basic auth not working with a browser

Most likely it's a browser integration issue with system keychain.

  • First verify that basic auth is working using curl

    curl -v -x username:password@localhost:8899 https://httpbin.org/get

  • See this thread for further details.

Docker image not working on macOS

It's a compatibility issue with vpnkit.

See moby/vpnkit exhausts docker resources and Connection refused: The proxy could not connect for some background.

GCE log viewer integration for proxy.py

A starter fluentd.conf template is available.

  1. Copy this configuration file as proxy.py.conf under /etc/google-fluentd/config.d/

  2. Update path field to log file path as used with --log-file flag. By default /tmp/proxy.log path is tailed.

  3. Reload google-fluentd:

    sudo service google-fluentd restart

Now proxy.py logs can be browsed using GCE log viewer.

ValueError: filedescriptor out of range in select

proxy.py is made to handle thousands of connections per second without any socket leaks.

  1. Make use of --open-file-limit flag to customize ulimit -n.
  2. Make sure to adjust --backlog flag for higher concurrency.

If nothing helps, open an issue with requests per second sent and output of following debug script:

❯ ./helper/monitor_open_files.sh <proxy-py-pid>

None:None in access logs

Sometimes you may see None:None in access logs. It simply means that an upstream server connection was never established i.e. upstream_host=None, upstream_port=None.

There can be several reasons for no upstream connection, few obvious ones include:

  1. Client established a connection but never completed the request.
  2. A plugin returned a response prematurely, avoiding connection to upstream server.

OSError when wrapping client for TLS Interception

With TLS Interception on, you might occasionally see following exceptions:

2021-11-06 23:33:34,540 - pid:91032 [E] server.intercept:678 - OSError when wrapping client
Traceback (most recent call last):
  ...[redacted]...
  ...[redacted]...
  ...[redacted]...
ssl.SSLError: [SSL: TLSV1_ALERT_UNKNOWN_CA] tlsv1 alert unknown ca (_ssl.c:997)
...[redacted]... - CONNECT oauth2.googleapis.com:443 - 0 bytes - 272.08 ms

Some clients can throw TLSV1_ALERT_UNKNOWN_CA if they cannot verify the certificate of the server because it is signed by an unknown issuer CA. Which is the case when we are doing TLS interception. This can be for a variety of reasons e.g. certificate pinning etc.

Another exception you might see is CERTIFICATE_VERIFY_FAILED:

2021-11-06 23:36:02,002 - pid:91033 [E] handler.handle_readables:293 - Exception while receiving from client connection <socket.socket fd=28, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 8899), raddr=('127.0.0.1', 51961)> with reason SSLCertVerificationError(1, '[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: self signed certificate in certificate chain (_ssl.c:997)')
Traceback (most recent call last):
  ...[redacted]...
  ...[redacted]...
  ...[redacted]...
ssl.SSLCertVerificationError: [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: self signed certificate in certificate chain (_ssl.c:997)
...[redacted]... - CONNECT init.push.apple.com:443 - 0 bytes - 892.99 ms

In future, we might support serving original HTTPS content for such clients while still performing TLS interception in the background. This will keep the clients happy without impacting our ability to TLS intercept. Unfortunately, this feature is currently not available.

Another example with SSLEOFError exception:

2021-11-06 23:46:40,446 - pid:91034 [E] server.intercept:678 - OSError when wrapping client
Traceback (most recent call last):
  ...[redacted]...
  ...[redacted]...
  ...[redacted]...
ssl.SSLEOFError: EOF occurred in violation of protocol (_ssl.c:997)
...[redacted]... - CONNECT stock.adobe.io:443 - 0 bytes - 685.32 ms

Plugin Developer and Contributor Guide

High level architecture

                        +-------------+
                        |             |
                        |  Proxy([])  |
                        |             |
                        +------+------+
                               |
                               |
                   +-----------v--------------+
                   |                          |
                   |    AcceptorPool(...)     |
                   |                          |
                   +------------+-------------+
                                |
+-----------------+             |           +-----------------+
|                 |             |           |                 |
|   Acceptor(..)  <-------------+----------->  Acceptor(..)   |
|                 |                         |                 |
+---+-------------+                         +---------+-------+
    |                                                 |
    |                                                 |
    |    +------++------++------++------++------+     |
    |    |      ||      ||      ||      ||      |     |
    +---->      ||      ||      ||      ||      <-----+
         |      ||      ||      ||      ||      |
         +------++------++------++------++------+
                Threadless Worker Processes

proxy.py is made with performance in mind. By default, proxy.py will try to utilize all available CPU cores to it for accepting new client connections. This is achieved by starting AcceptorPool which listens on configured server port. Then, AcceptorPool starts Acceptor processes (--num-acceptors) to accept incoming client connections. Alongside, if --threadless is enabled, ThreadlessPool is setup which starts Threadless processes (--num-workers) to handle the incoming client connections.

Each Acceptor process delegates the accepted client connection to a threadless process via Work class. Currently, HttpProtocolHandler is the default work class.

HttpProtocolHandler simply assumes that incoming clients will follow HTTP specification. Specific HTTP proxy and HTTP server implementations are written as plugins of HttpProtocolHandler.

See documentation of HttpProtocolHandlerPlugin for available lifecycle hooks. Use HttpProtocolHandlerPlugin to add new features for http(s) clients. Example, See HttpWebServerPlugin.

Everything is a plugin

Within proxy.py everything is a plugin.

  • We enabled proxy server plugins using --plugins flag. Proxy server HttpProxyPlugin is a plugin of HttpProtocolHandler. Further, Proxy server allows plugin through HttpProxyBasePlugin specification.

  • All the proxy server plugin examples were implementing HttpProxyBasePlugin. See documentation of HttpProxyBasePlugin for available lifecycle hooks. Use HttpProxyBasePlugin to modify behavior of http(s) proxy protocol between client and upstream server. Example, FilterByUpstreamHostPlugin.

  • We also enabled inbuilt web server using --enable-web-server. Web server HttpWebServerPlugin is a plugin of HttpProtocolHandler and implements HttpProtocolHandlerPlugin specification.

  • There also is a --disable-http-proxy flag. It disables inbuilt proxy server. Use this flag with --enable-web-server flag to run proxy.py as a programmable http(s) server.

Managing states for your stateless plugins

Plugin class instances are created per-request. Most importantly, plugin instances are created within CPU core context where the request was received.

For above reason, global variables in your plugins may work as expected. Your plugin code by design must be stateless.

To manage global states, you have a couple of options:

  1. Make use of Python's multiprocessing safe data structures
  2. Make use of proxy.py in-built eventing mechanism

Passing processing context between plugins

Sometimes, a plugin may need to pass additional context to other plugins after them in the processing chain. Example, this additional context can also be dumped as part of access logs.

To pass processing context, make use of plugin's on_access_log method. See how Program Name plugin modifies default client_ip key in the context and updates it to detected program name.

As a result, when we enable Program Name Plugin, we see local client program name instead of IP address in the access logs.

Development Guide

Setup Local Environment

Contributors must start proxy.py from source to verify and develop new features / fixes.

See Run proxy.py from command line using repo source for details.

WARNING On macOS you must install Python using pyenv, as Python installed via homebrew tends to be problematic. See linked thread for more details.

Setup Git Hooks

Pre-commit hook ensures tests are passing.

  1. cd /path/to/proxy.py
  2. ln -s $(PWD)/git-pre-commit .git/hooks/pre-commit

Pre-push hook ensures lint and tests are passing.

  1. cd /path/to/proxy.py
  2. ln -s $(PWD)/git-pre-push .git/hooks/pre-push

Sending a Pull Request

Every pull request is tested using GitHub actions.

See GitHub workflow for list of tests.

Projects Using Proxy.Py

Some of the projects using proxy.py

  1. ray-project
  2. aio-libs
  3. wifipumpkin3
  4. MerossIot
  5. pyshorteners
  6. Slack API
  7. ibeam
  8. PyPaperBot

For full list see used by

Benchmarks

See Benchmark directory on how to run benchmark comparisons with other OSS web servers.

To run standalone benchmark for proxy.py, use the following command from repo root:

❯ ./helper/benchmark.sh

Flags

❯ proxy -h
usage: -m [-h] [--tunnel-hostname TUNNEL_HOSTNAME] [--tunnel-port TUNNEL_PORT]
          [--tunnel-username TUNNEL_USERNAME]
          [--tunnel-ssh-key TUNNEL_SSH_KEY]
          [--tunnel-ssh-key-passphrase TUNNEL_SSH_KEY_PASSPHRASE]
          [--tunnel-remote-port TUNNEL_REMOTE_PORT] [--threadless]
          [--threaded] [--num-workers NUM_WORKERS] [--enable-events]
          [--local-executor LOCAL_EXECUTOR] [--backlog BACKLOG]
          [--hostname HOSTNAME] [--hostnames HOSTNAMES [HOSTNAMES ...]]
          [--port PORT] [--ports PORTS [PORTS ...]] [--port-file PORT_FILE]
          [--unix-socket-path UNIX_SOCKET_PATH]
          [--num-acceptors NUM_ACCEPTORS] [--version] [--log-level LOG_LEVEL]
          [--log-file LOG_FILE] [--log-format LOG_FORMAT]
          [--open-file-limit OPEN_FILE_LIMIT]
          [--plugins PLUGINS [PLUGINS ...]] [--enable-dashboard]
          [--basic-auth BASIC_AUTH] [--enable-ssh-tunnel]
          [--work-klass WORK_KLASS] [--pid-file PID_FILE] [--openssl OPENSSL]
          [--enable-proxy-protocol] [--enable-conn-pool] [--key-file KEY_FILE]
          [--cert-file CERT_FILE] [--client-recvbuf-size CLIENT_RECVBUF_SIZE]
          [--server-recvbuf-size SERVER_RECVBUF_SIZE]
          [--max-sendbuf-size MAX_SENDBUF_SIZE] [--timeout TIMEOUT]
          [--disable-http-proxy] [--disable-headers DISABLE_HEADERS]
          [--ca-key-file CA_KEY_FILE] [--ca-cert-dir CA_CERT_DIR]
          [--ca-cert-file CA_CERT_FILE] [--ca-file CA_FILE]
          [--ca-signing-key-file CA_SIGNING_KEY_FILE]
          [--auth-plugin AUTH_PLUGIN] [--cache-requests]
          [--cache-by-content-type] [--cache-dir CACHE_DIR]
          [--proxy-pool PROXY_POOL] [--enable-web-server]
          [--enable-static-server] [--static-server-dir STATIC_SERVER_DIR]
          [--min-compression-length MIN_COMPRESSION_LENGTH]
          [--enable-reverse-proxy] [--pac-file PAC_FILE]
          [--pac-file-url-path PAC_FILE_URL_PATH]
          [--cloudflare-dns-mode CLOUDFLARE_DNS_MODE]
          [--filtered-upstream-hosts FILTERED_UPSTREAM_HOSTS]
          [--filtered-client-ips-mode FILTERED_CLIENT_IPS_MODE]
          [--filtered-client-ips FILTERED_CLIENT_IPS]
          [--filtered-url-regex-config FILTERED_URL_REGEX_CONFIG]

proxy.py v2.4.4rc6.dev11+gac1f05d7.d20240413

options:
  -h, --help            show this help message and exit
  --tunnel-hostname TUNNEL_HOSTNAME
                        Default: None. Remote hostname or IP address to which
                        SSH tunnel will be established.
  --tunnel-port TUNNEL_PORT
                        Default: 22. SSH port of the remote host.
  --tunnel-username TUNNEL_USERNAME
                        Default: None. Username to use for establishing SSH
                        tunnel.
  --tunnel-ssh-key TUNNEL_SSH_KEY
                        Default: None. Private key path in pem format
  --tunnel-ssh-key-passphrase TUNNEL_SSH_KEY_PASSPHRASE
                        Default: None. Private key passphrase
  --tunnel-remote-port TUNNEL_REMOTE_PORT
                        Default: 8899. Remote port which will be forwarded
                        locally for proxy.
  --threadless          Default: True. Enabled by default on Python 3.8+ (mac,
                        linux). When disabled a new thread is spawned to
                        handle each client connection.
  --threaded            Default: False. Disabled by default on Python < 3.8
                        and windows. When enabled a new thread is spawned to
                        handle each client connection.
  --num-workers NUM_WORKERS
                        Defaults to number of CPU cores.
  --enable-events       Default: False. Enables core to dispatch lifecycle
                        events. Plugins can be used to subscribe for core
                        events.
  --local-executor LOCAL_EXECUTOR
                        Default: 1. Enabled by default. Use 0 to disable. When
                        enabled acceptors will make use of local (same
                        process) executor instead of distributing load across
                        remote (other process) executors. Enable this option
                        to achieve CPU affinity between acceptors and
                        executors, instead of using underlying OS kernel
                        scheduling algorithm.
  --backlog BACKLOG     Default: 100. Maximum number of pending connections to
                        proxy server.
  --hostname HOSTNAME   Default: 127.0.0.1. Server IP address.
  --hostnames HOSTNAMES [HOSTNAMES ...]
                        Default: None. Additional IP addresses to listen on.
  --port PORT           Default: 8899. Server port. To listen on more ports,
                        pass them using --ports flag.
  --ports PORTS [PORTS ...]
                        Default: None. Additional ports to listen on.
  --port-file PORT_FILE
                        Default: None. Save server port numbers. Useful when
                        using --port=0 ephemeral mode.
  --unix-socket-path UNIX_SOCKET_PATH
                        Default: None. Unix socket path to use. When provided
                        --host and --port flags are ignored
  --num-acceptors NUM_ACCEPTORS
                        Defaults to number of CPU cores.
  --version, -v         Prints proxy.py version.
  --log-level LOG_LEVEL
                        Valid options: DEBUG, INFO (default), WARNING, ERROR,
                        CRITICAL. Both upper and lowercase values are allowed.
                        You may also simply use the leading character e.g.
                        --log-level d
  --log-file LOG_FILE   Default: sys.stdout. Log file destination.
  --log-format LOG_FORMAT
                        Log format for Python logger.
  --open-file-limit OPEN_FILE_LIMIT
                        Default: 1024. Maximum number of files (TCP
                        connections) that proxy.py can open concurrently.
  --plugins PLUGINS [PLUGINS ...]
                        Comma separated plugins. You may use --plugins flag
                        multiple times.
  --enable-dashboard    Default: False. Enables proxy.py dashboard.
  --basic-auth BASIC_AUTH
                        Default: No authentication. Specify colon separated
                        user:password to enable basic authentication.
  --enable-ssh-tunnel   Default: False. Enable SSH tunnel.
  --work-klass WORK_KLASS
                        Default: proxy.http.HttpProtocolHandler. Work klass to
                        use for work execution.
  --pid-file PID_FILE   Default: None. Save "parent" process ID to a file.
  --openssl OPENSSL     Default: openssl. Path to openssl binary. By default,
                        assumption is that openssl is in your PATH.
  --enable-proxy-protocol
                        Default: False. If used, will enable proxy protocol.
                        Only version 1 is currently supported.
  --enable-conn-pool    Default: False. (WIP) Enable upstream connection
                        pooling.
  --key-file KEY_FILE   Default: None. Server key file to enable end-to-end
                        TLS encryption with clients. If used, must also pass
                        --cert-file.
  --cert-file CERT_FILE
                        Default: None. Server certificate to enable end-to-end
                        TLS encryption with clients. If used, must also pass
                        --key-file.
  --client-recvbuf-size CLIENT_RECVBUF_SIZE
                        Default: 128 KB. Maximum amount of data received from
                        the client in a single recv() operation.
  --server-recvbuf-size SERVER_RECVBUF_SIZE
                        Default: 128 KB. Maximum amount of data received from
                        the server in a single recv() operation.
  --max-sendbuf-size MAX_SENDBUF_SIZE
                        Default: 64 KB. Maximum amount of data to flush in a
                        single send() operation.
  --timeout TIMEOUT     Default: 10.0. Number of seconds after which an
                        inactive connection must be dropped. Inactivity is
                        defined by no data sent or received by the client.
  --disable-http-proxy  Default: False. Whether to disable
                        proxy.HttpProxyPlugin.
  --disable-headers DISABLE_HEADERS
                        Default: None. Comma separated list of headers to
                        remove before dispatching client request to upstream
                        server.
  --ca-key-file CA_KEY_FILE
                        Default: None. CA key to use for signing dynamically
                        generated HTTPS certificates. If used, must also pass
                        --ca-cert-file and --ca-signing-key-file
  --ca-cert-dir CA_CERT_DIR
                        Default: ~/.proxy/certificates. Directory to store
                        dynamically generated certificates. Also see --ca-key-
                        file, --ca-cert-file and --ca-signing-key-file
  --ca-cert-file CA_CERT_FILE
                        Default: None. Signing certificate to use for signing
                        dynamically generated HTTPS certificates. If used,
                        must also pass --ca-key-file and --ca-signing-key-file
  --ca-file CA_FILE     Default: /Users/abhinavsingh/Dev/proxy.py/.venv3118/li
                        b/python3.11/site-packages/certifi/cacert.pem. Provide
                        path to custom CA bundle for peer certificate
                        verification
  --ca-signing-key-file CA_SIGNING_KEY_FILE
                        Default: None. CA signing key to use for dynamic
                        generation of HTTPS certificates. If used, must also
                        pass --ca-key-file and --ca-cert-file
  --auth-plugin AUTH_PLUGIN
                        Default: proxy.http.proxy.auth.AuthPlugin. Auth plugin
                        to use instead of default basic auth plugin.
  --cache-requests      Default: False. Whether to also write request packets
                        in the cache file.
  --cache-by-content-type
                        Default: False. Whether to extract content by type
                        from responses. Extracted content type is written to
                        the cache directory e.g. video.mp4.
  --cache-dir CACHE_DIR
                        Default: /Users/abhinavsingh/.proxy/cache. Flag only
                        applicable when cache plugin is used with on-disk
                        storage.
  --proxy-pool PROXY_POOL
                        List of upstream proxies to use in the pool
  --enable-web-server   Default: False. Whether to enable
                        proxy.HttpWebServerPlugin.
  --enable-static-server
                        Default: False. Enable inbuilt static file server.
                        Optionally, also use --static-server-dir to serve
                        static content from custom directory. By default,
                        static file server serves out of installed proxy.py
                        python module folder.
  --static-server-dir STATIC_SERVER_DIR
                        Default: "public" folder in directory where proxy.py
                        is placed. This option is only applicable when static
                        server is also enabled. See --enable-static-server.
  --min-compression-length MIN_COMPRESSION_LENGTH
                        Default: 20 bytes. Sets the minimum length of a
                        response that will be compressed (gzipped).
  --enable-reverse-proxy
                        Default: False. Whether to enable reverse proxy core.
  --pac-file PAC_FILE   A file (Proxy Auto Configuration) or string to serve
                        when the server receives a direct file request. Using
                        this option enables proxy.HttpWebServerPlugin.
  --pac-file-url-path PAC_FILE_URL_PATH
                        Default: /. Web server path to serve the PAC file.
  --cloudflare-dns-mode CLOUDFLARE_DNS_MODE
                        Default: security. Either "security" (for malware
                        protection) or "family" (for malware and adult content
                        protection)
  --filtered-upstream-hosts FILTERED_UPSTREAM_HOSTS
                        Default: Blocks Facebook. Comma separated list of IPv4
                        and IPv6 addresses.
  --filtered-client-ips-mode FILTERED_CLIENT_IPS_MODE
                        Default: blacklist. Can be either "whitelist"
                        (restrict access to specific IPs)or "blacklist" (allow
                        everything except specific IPs).
  --filtered-client-ips FILTERED_CLIENT_IPS
                        Default: 127.0.0.1,::1. Comma separated list of IPv4
                        and IPv6 addresses.
  --filtered-url-regex-config FILTERED_URL_REGEX_CONFIG
                        Default: No config. Comma separated list of IPv4 and
                        IPv6 addresses.

Proxy.py not working? Report at:
https://github.com/abhinavsingh/proxy.py/issues/new