Px is a HTTP(s) proxy server that allows applications to authenticate through an NTLM or Kerberos proxy server, typically used in corporate deployments, without having to deal with the actual handshake. Px leverages Windows SSPI or single sign-on and automatically authenticates using the currently logged in Windows user account. It is also possible to run Px on Windows, Linux and MacOS without single sign-on by configuring the domain, username and password to authenticate with.
Px uses libcurl and as a result supports all the authentication mechanisms supported by libcurl.
The whole point of Px is to help tools get through a typical corporate proxy. This means using a package manager to install Px might not always be feasible which is why Px offers two binary options:
-
If Python is already available, Px and all its dependencies can be easily installed by downloading the
wheelspackage for the target OS from the releases page. After extraction, Px and all dependencies can be installed withpip:python -m pip install px-proxy --no-index -f /path/to/wheels -
If Python is not available, get the latest compiled binary from the releases page instead. The Windows binary is built using Python Embedded and the Linux and OSX binaries are compiled with Nuitka and contain everything needed to run standalone.
If direct internet access is available along with Python, Px can be easily
installed using the Python package manager pip. This will download and install
Px as a Python module along with all dependencies:
python -m pip install px-proxy
On Windows, scoop can also be used to install Px:
scoop install px
Once installed, Px can be run as follows:
- Running
pxdirectly - In the background:
pythonw -m px - In the foreground in a console window:
python -m px - As a wrapped service using WinSW
Px requires libcurl and the Windows builds ship with a copy. On Linux, it is required to install libcurl using the package manager:
- RHEL:
yum install libcurl - Ubuntu:
apt install libcurl4 - Alpine:
apk add libcurl
Manually prepare and configure px to be able to run it and verify the connectivity. Download the executable WinSW-x64.exe from WinSW. We are using 2.12.0.
Place a minimal WinSW-x64.xml configuration file next to the executable.
<service>
<!-- ID of the service. It should be unique across the Windows system-->
<id>Px</id>
<!-- Display name of the service -->
<name>Px Service (powered by WinSW)</name>
<!-- Service description -->
<description>This service is a service created from a minimal configuration</description>
<!-- Path to the executable, which should be started -->
<executable>D:\px\px.exe</executable>
</service>
Run
WinSW-x64.exe install WinSW-x64.xml from elevated administrator cmd. You should now have a new service on your computer named Px Service (powered by WinSW).
You can run it from Services window or run WinSW-x64.exe start WinSW-x64.xml.
The latest Px version can be downloaded and installed from source via pip:
python -m pip install https://github.com/genotrance/px/archive/master.zip
Source can also be downloaded and installed:
-
Via git:
git clone https://github.com/genotrance/px -
Download ZIP:
https://github.com/genotrance/px/archive/master.zip
Once downloaded, Px can be installed as a standard Python module along with all dependencies :
python -m pip install .
NOTE: Source install methods will require internet access since Python will try to install Px dependencies from the internet. The binaries mentioned in the previous section could be used to bootstrap a source install.
NOTE: libcurl will need to be installed on Linux, as described earlier, using
the package manager. For Windows, download and
extract libcurl.dll and libcurl-x64.dll to $PATH.
Px can be run as a local Python script without installation. Download the source as described above, install all dependencies and then run Px:
pip install keyring netaddr psutil python-dotenv pyspnego quickjs
# Download/install libcurl
pythonw px.py # run in the background
python px.py # run in a console window
If Px has been installed to the Windows registry to start on boot, it should be uninstalled before removal:
python -m px --uninstall
Px can then be uninstalled using pip as follows:
python -m pip uninstall px-proxy
Px is available as a prebuilt Docker image.
Two images are posted - the default includes keyring and associated dependencies
whereas the mini version is smaller but will have to depend on PX_PASSWORD and
PX_CLIENT_PASSWORD for credentials.
The following Docker flags will be useful to configure and run Px:
--name px name container so it is easy to stop it
-d run in the background
--rm remove container on exit
--network host make Px directly accessible from host network
OR
-p 3128:3128 publish the port - Px needs to run in --gateway mode
-e PX_LOG=4 set environment variables to configure Px
-v /dir:/px mount a host directory with a px.ini or .env file to configure Px
OR
--mount source=/dir,target=/px
mount a volume if preferred
docker run ... genotrance/px --gateway --verbose
configure directly from the command line
Keyring credentials can be stored in a host folder and mounted into the container as follows:
-v /keyrings:/root/.local/share/keyrings
mount a local dir to store keyring info
OR
--mount source=/keyrings,target=/root/.local/share/keyrings
mount a volume if preferred
Credentials can be saved using the command line:
docker run ... genotrance/px --username=... --password
configure keyring directly from the command line
The mini version does not have keyring so credentials need to be set using environment variables:
-e PX_PASSWORD=... -e PX_CLIENT_PASSWORD=...
set environment variables to configure credentials
Px requires only one piece of information in order to function - the server
name and port of the proxy server. If not specified, Px will check Internet Options or environment variables for any proxy definitions. Without this, Px
will try to connect to sites directly.
The noproxy capability allows Px to connect to configured hosts directly,
bypassing the proxy altogether. This allows clients to connect to hosts within
the intranet without requiring additional configuration for each client or at
the proxy.
Configuration can be specified in multiple ways, listed in order of precedence:
- Command line flags
- Environment variables
- Variables in a dotenv file (.env)
- In the working directory
- In the Px directory
- Configuration file
px.ini- In the working directory
- In the Px directory
There are many configuration options to tweak - refer to the Usage
section or --help for details and syntax.
If SSPI is not available or not preferred, providing --username in domain\username
format allows Px to authenticate as that user. The corresponding password is
retrieved using Python keyring and needs to be setup in the appropriate OS
specific backend.
Credentials can be setup with the command line:
px --username=domain\username --password
If username is already defined with PX_USERNAME or in px.ini:
px --password
Information on keyring backends can be found here.
As an alternative, Px can also load credentials from the environment variable
PX_PASSWORD or a dotenv file. This is only recommended when keyring is not
available.
Credential Manager is the recommended backend for Windows and the password is stored as a 'Generic Credential' type with 'Px' as the network address name. Credential Manager can be accessed as follows:
Control Panel > User Accounts > Credential Manager > Windows Credentials
Or on the command line: `rundll32.exe keymgr.dll, KRShowKeyMgr`
Gnome Keyring or KWallet is used to store passwords on Linux.
For systems without a GUI (headless, docker), D-Bus can be started interactively:
dbus-run-session -- sh
If this needs to be done in a script:
export DBUS_SESSION_BUS_ADDRESS=`dbus-daemon --fork --config-file=/usr/share/dbus-1/session.conf --print-address`
Gnome Keyring can then be unlocked as follows:
echo 'somecredstorepass' | gnome-keyring-daemon --unlock
If the default SecretService keyring backend does not work, a third-party
backend might be
required. Simply install and configure one and keyring will use it. Remember
to specify the environment variables they require before starting Px.
This will not work for the Nuitka binaries so as a fallback, PX_PASSWORD can
be used instead to set credentials.
Px is useful to authenticate with the upstream proxy server on behalf of clients
but it can also authenticate the client that connects to it if needed. This can
be useful in gateway mode where remote clients should log in before accessing
the upstream proxy via Px. BASIC and DIGEST auth are supported, along with
NTLM and NEGOTIATE.
The client credentials can be different from the upstream proxy credentials or the same if preferred. SSPI is also supported on Windows and can be leveraged for only the client or upstream or both.
Client authentication is turned off by default and can be enabled using
--client-auth, PX_CLIENT_AUTH or px.ini. Setting the value to ANYSAFE is
recommended.
Similar to the upstream proxy, the client username can be configured with
--client-username, PX_CLIENT_USERNAME or px.ini The password can be setup
in keyring using PxClient as the network address name. PX_CLIENT_PASSWORD is
available for cases where keyring is not available.
SSPI is enabled by default on Windows and can be disabled with --client-nosspi,
PX_CLIENT_NOSSPI or in px.ini.
Client credentials can be setup in keyring with the command line:
px --client-username=domain\username --client-password
Px only supports one credential for the upstream proxy but can be configured to
support multiple client users when keyring is used. Each user should be added to
keyring with the PxClient network address.
Using an upstream proxy is not required so Px can also be used simply as an authenticating proxy for smaller setups.
The configuration file px.ini can be created or updated from the command line
using --save.
The binary distribution of Px runs in the background once started and can be
quit by running px --quit. When running in the foreground, use CTRL-C.
Px can also be setup to automatically run on startup on Windows with the
--install flag. This is done by adding an entry into the Window registry which
can be removed with --uninstall.
NOTE: Command line parameters passed with --install are not saved for use on
startup. The --save flag or manual editing of px.ini is required to provide
configuration to Px on startup.
NOTE: When using the Px binary distribution on Windows, run .\pythonw -m px --install
instead of using px.exe. This will avoid the popup console window on startup.
px [FLAGS]
python px.py [FLAGS]
python -m px [FLAGS]
Actions:
--save
Save configuration to file specified with --config or px.ini in working directory
Allows setting up Px config directly from command line
Values specified on CLI override any values in existing config file
Values not specified on CLI or config file are set to defaults
--install
Add Px to the Windows registry to run on startup
--uninstall
Remove Px from the Windows registry
--quit
Quit a running instance of Px
--restart
Quit a running instance of Px and start a new instance
--password | PX_PASSWORD
Collect and save password to default keyring. Username needs to be provided
via --username, PX_USERNAME or in the config file.
As an alternative, Px can also load credentials from the environment variable
`PX_PASSWORD` or a dotenv file.
--client-password | PX_CLIENT_PASSWORD
Collect and save password to default keyring. Username needs to be provided
via --client-username, PX_CLIENT_USERNAME or in the config file.
As an alternative, Px can also load credentials from the environment variable
`PX_CLIENT_PASSWORD` or a dotenv file.
--test=URL | --test
Test Px as configured with the URL specified. This can be used to confirm that
Px is configured correctly and is able to connect and authenticate with the
upstream proxy. If URL is skipped, Px runs multiple tests against httpbin.org.
Configuration:
--config= | PX_CONFIG=
Specify config file. Valid file path, default: px.ini in working directory
or script directory
--proxy= --server= | PX_SERVER= | proxy:server=
NTLM server(s) to connect through. IP:port, hostname:port
Multiple proxies can be specified comma separated. Px will iterate through
and use the one that works
--pac= | PX_PAC= | proxy:pac=
PAC file to use to connect
Use in place of --server if PAC file should be loaded from a URL or local
file. Relative paths will be relative to the Px script or binary
--pac_encoding= | PX_PAC_ENCODING= | proxy:pac_encoding=
PAC file encoding
Specify in case default 'utf-8' encoding does not work
--listen= | PX_LISTEN= | proxy:listen=
Network interface(s) to listen on. Comma separated, default: 127.0.0.1
--gateway and --hostonly override this to bind to all interfaces
--port= | PX_PORT= | proxy:port=
Port to run this proxy on - default: 3128
--gateway | PX_GATEWAY= | proxy:gateway=
Allow remote machines to use proxy. 0 or 1, default: 0
Overrides --listen and binds to all interfaces
--hostonly | PX_HOSTONLY= | proxy:hostonly=
Allow only local interfaces to use proxy. 0 or 1, default: 0
Px allows all IP addresses assigned to local interfaces to use the service.
This allows local apps as well as VM or container apps to use Px when in a
NAT config. Overrides --listen and binds to all interfaces, overrides the
default --allow rules
--allow= | PX_ALLOW= | proxy:allow=
Allow connection from specific subnets. Comma separated, default: *.*.*.*
Whitelist which IPs can use the proxy. --hostonly overrides any definitions
unless --gateway mode is also specified
127.0.0.1 - specific ip
192.168.0.* - wildcards
192.168.0.1-192.168.0.255 - ranges
192.168.0.1/24 - CIDR
--noproxy= | PX_NOPROXY= | proxy:noproxy=
Direct connect to specific subnets or domains like a regular proxy. Comma separated
Skip the NTLM proxy for connections to these hosts
127.0.0.1 - specific ip
192.168.0.* - wildcards
192.168.0.1-192.168.0.255 - ranges
192.168.0.1/24 - CIDR
example.com - domains
--useragent= | PX_USERAGENT= | proxy:useragent=
Override or send User-Agent header on client's behalf
--username= | PX_USERNAME= | proxy:username=
Authentication to use when SSPI is unavailable. Format is domain\username
Service name "Px" and this username are used to retrieve the password using
Python keyring if available.
--auth= | PX_AUTH= | proxy:auth=
Force instead of discovering upstream proxy type
By default, Px will attempt to discover the upstream proxy type. This
option can be used to force either NEGOTIATE, NTLM, DIGEST, BASIC or the
other libcurl supported upstream proxy types. See:
https://curl.se/libcurl/c/CURLOPT_HTTPAUTH.html
To control which methods are available during proxy detection:
Prefix NO to avoid method - e.g. NONTLM => ANY - NTLM
Prefix SAFENO to avoid method - e.g. SAFENONTLM => ANYSAFE - NTLM
Prefix ONLY to support only that method - e.g ONLYNTLM => ONLY + NTLM
Set to NONE to defer all authentication to the client. This allows multiple
instances of Px to be chained together to access an upstream proxy that is not
directly connected:
Client -> Auth Px -> no-Auth Px -> Upstream proxy
'Auth Px' cannot directly access upstream proxy but 'no-Auth Px' can
--client-username= | PX_CLIENT_USERNAME= | client:client_username=
Client authentication to use when SSPI is unavailable. Format is domain\username
Service name "PxClient" and this username are used to retrieve the password using
Python keyring if available.
--client-auth= | PX_CLIENT_AUTH= | client:client_auth=
Enable authentication for client connections. Comma separated, default: NONE
Mechanisms supported: NEGOTIATE, NTLM, DIGEST, BASIC
ANY = enable all supported mechanisms
ANYSAFE = enable all supported mechanisms except BASIC
NTLM = enable only NTLM, etc.
NONE = disable client authentication altogether (default)
--client-nosspi= | PX_CLIENT_NOSSPI= | client:client_nosspi=
Disable SSPI for client authentication on Windows. default: 0
Set to 1 to disable SSPI and use the configured username and password
--workers= | PX_WORKERS= | settings:workers=
Number of parallel workers (processes). Valid integer, default: 2
--threads= | PX_THREADS= | settings:threads=
Number of parallel threads per worker (process). Valid integer, default: 32
--idle= | PX_IDLE= | settings:idle=
Idle timeout in seconds for HTTP connect sessions. Valid integer, default: 30
--socktimeout= | PX_SOCKTIMEOUT= | settings:socktimeout=
Timeout in seconds for connections before giving up. Valid float, default: 20
--proxyreload= | PX_PROXYRELOAD= | settings:proxyreload=
Time interval in seconds before refreshing proxy info. Valid int, default: 60
Proxy info reloaded from manual proxy info defined in Internet Options
--foreground | PX_FOREGROUND= | settings:foreground=
Run in foreground when compiled or run with pythonw.exe. 0 or 1, default: 0
Px will attach to the console and write to it even though the prompt is
available for further commands. CTRL-C in the console will exit Px
--log= | PX_LOG= | settings:log=
Enable debug logging. default: 0
1 = Log to script dir [--debug]
2 = Log to working dir
3 = Log to working dir with unique filename [--uniqlog]
4 = Log to stdout [--verbose]. Implies --foreground
If Px crashes without logging, traceback is written to the working dir
Use proxyserver.com:80 and allow requests from localhost only:
px --proxy=proxyserver.com:80
Don't use any forward proxy at all, just log what's going on:
px --noproxy=0.0.0.0/0 --debug
Allow requests from localhost and all locally assigned IP addresses. This
is very useful for Docker for Windows and VMs in a NAT configuration because
all requests originate from the host's IP:
px --proxy=proxyserver.com:80 --hostonly
Allow requests from localhost, locally assigned IP addresses and the IPs
specified in the allow list outside the host:
px --proxy=proxyserver:80 --hostonly --gateway --allow=172.*.*.*
Allow requests from everywhere. Be careful, every client will use your login:
px --proxy=proxyserver.com:80 --gateway
NOTE: In Docker for Windows you need to set your proxy to
http://host.docker.internal:3128 or http://<your_ip>:3128 (or actual port
Px is listening to) in your containers and be aware of
docker/for-win#1380.
Workaround:
docker build --build-arg http_proxy=http://<your ip>:3128 --build-arg https_proxy=http://<your ip>:3128 -t containername ../dir/with/Dockerfile
NOTE: In WSL2 you can setup your proxy in /etc/profile as follows:
export http_proxy="http://$(tail -1 /etc/resolv.conf | cut -d' ' -f2):3128"
export https_proxy="http://$(tail -1 /etc/resolv.conf | cut -d' ' -f2):3128"
NOTE: When running MQTT over websockets, it will help to increase the idle
timeout to 120 seconds: --idle=120. The default value of 30 will cause the
websocket connection to disconnect since the default MQTT keepalive period
is 60 seconds.
Px doesn't have any GUI and runs completely in the background. It depends on the following Python packages:
Px also depends on libcurl for all outbound HTTP connections and proxy authentication.
Windows multiprocessing only works on Python 3.3+ since that's when support was added to share sockets across processes. On older versions of Python, Px will run multi-threaded but in a single process.
MacOSX socket sharing is not implemented at this time and is limited to running in a single process.
While it should mostly work, Px is not tested on MacOSX since there's no test environment available at this time to verify functionality. PRs are welcome to help fix any issues.
To build a self-sufficient executable that does not depend on the presence of
Python and dependency modules, both Nuitka and PyInstaller scripts are provided.
There is also a Python Embedded build that is preferable on Windows. Check out
python tools.py for more details.
Px is definitely a work in progress and any feedback or suggestions are welcome. It is hosted on GitHub with an MIT license so issues, forks and PRs are most appreciated. Join us on the discussion board, Gitter or Matrix to chat about Px.
Thank you to all contributors for their PRs and all issue submitters.
Px is based on code from all over the internet and acknowledges innumerable sources.