Skip to content

jborean93/pykrb5

Repository files navigation

Python Kerberos 5 Library

Test workflow PyPI version License

This library provides Python functions that wraps the Kerberos 5 C API. Due to the complex nature of this API it is highly recommended to use something like python-gssapi which exposes the Kerberos authentication details through GSSAPI.

Requirements

  • An implementation of the Kerberos 5 API - including the header files
  • A C compiler, such as GCC
  • Python 3.6+

Note: macOS includes their own implementation of Heimdal and a compiler isn't needed on that platform if installing from the wheel.

Installation

Simply run:

pip install krb5

To install from source run the following:

git clone https://github.com/jborean93/pykrb5.git
python -m pip install build
python -m build
pip install dist/krb5-*.whl

Compiling the code should automatically pick up the proper paths for the KRB5 headers and locations. If further customisation is needed, the following environment variables can be set when building the wheel:

  • KRB5_KRB5CONFIG
    • The path to krb5-config to use for detecting the Kerberos library to link to
    • The compiler and linker args are derived from what this function outputs
    • Defaults to whatever krb5-config is on the PATH
    • FreeBSD will default to /usr/local/bin/krb5-config instead of /usr/bin/krb5-config
  • KRB5_MAIN_LIB
    • The path to the libkrb5 shared library used to check if any of the optional functions are available
  • KRB5_COMPILER_ARGS
    • Compiler flags to use when compiling the extensions
    • Defaults to the output of krb5-config --cflags krb5 if not set
  • KRB5_LINKER_ARGS
    • Linker flags to use when compiling the extensions
    • Defaults to the output of krb5-config --libs krb5 if not set
  • KRB5_SKIP_MODULE_CHECK
    • Skips the checks used to detect if optional functions are available - will treat them all as available
    • This is only really useful when building the sdist as no implementation provides all these functions
  • KRB5_CYTHON_TRACING
    • Used to generate the Cython extensions with line tracing for coverage collection
  • KRB5_MACOS_HEIMDAL_DIR
    • Used when compiling on macOS to point to the Heimdal install directory
    • Used to find the Heimdal header files as macOS does not include, or provide a way to obtain, these header files for their Heimdal framework
    • Defaults to {git_root}/build_helpers/heimdal

Development

To run the tests or make changes to this repo run the following:

git clone https://github.com/jborean93/pykrb5.git
pip install -r requirements-dev.txt
pre-commit install

python -m pip install -e .

# Can compile the krb5 extensions on an adhoc basis
# python setup.py build_ext --inplace

From there an editor like VSCode can be used to make changes and run the test suite. To recompile the Cython files after a change run the build_ext --inplace command.

Structure

This library is merely a wrapper around the Kerberos 5 APIs. The functions under the krb5 namespace match the KRB5 API specification but with the krb5_ prefix remove. For example the krb5_init_context function is called through krb5.init_context(). Errors are raised as a Krb5Error which contains the message as formatted by the KRB5 implementation and the error code for that error. Some of the structures returned by these functions are represented by a Python class and are freed once they are deallocated once all references to that object is removed. Some classes expose an addr property that returns the raw pointer address of the structure it is wrapping. This is so the structure can be used in other libraries like python-gssapi but great care must be taken that nothing else frees the structure as that could cause a segmentation fault.

Not all the functions exposed in this library are available on every KRB5 API implementation. To check if a function is available run the following:

import krb5

if not hasattr(krb5, "kt_dup"):
    raise Exception("Current implementation does not support krb5_kt_dup")

There may also be some difference in behaviour, error codes, error messages, between te different implementations. It is up to the caller to paper over these differences when required.