Skip to main content

Temporary, well scoped credentials for pushing to GitHub

Project description

gh-scoped-creds

Provide finely scoped push access to GitHub from a JupyterHub or HPC system.

Goals

  1. Allow users on a JupyterHub to grant push access to only specific repositories rather than all the repositories they have access to.
  2. Do not store long-term credentials (like personal access tokens or ssh-keys) on disk, as they may get archived / fall into the wrong hands in the future.
  3. Allow GitHub organization admins visibility and control over what repos users can push to from remote systems (like JupyterHub or a shared cluster), where other admins of the remote system might be able to access the files of users with push access to repos. This has serious implications for supply chain security, as credentials might be stolen or lost and serious vulnerabilities be pushed to the repo.

These goals are accomplished by:

  1. Creating a GitHub app specific to the remote service (JupyterHub, HPC cluster, etc). Users and GitHub organization admins can then provide fine grained, repo level access to this GitHub app - Users can only push to repos that have the app installed.
  2. A commandline tool (gh-scoped-creds) that lets specific users authorize push access to the selected repositories temporarily - a token that expires after 8 hours.
  3. An IPython Magic (%ghscopedcreds) that provides a convenient wrapper to call gh-scoped-creds from inside a Jupyter Notebook.

In the future, an optional web app might also be provided to aid in authentication.

Installation

You can install gh-scoped-creds from PyPI.

pip install gh-scoped-creds

GitHub App configuration

  1. Create a GitHub app for use by the service (JupyterHub, HPC cluster, etc). You can either create it under your personal account, or preferably under a GitHub organization account (Go to Settings -> Developer Settings -> GitHub Apps -> New GitHub app from the organization's GitHub page).

  2. Give it a descriptive name and description, as your users will see this when they authenticate. Provide a link to a descriptive page explaining your service (if you are using a JupyterHub, this could be just your JupyterHub URL).

  3. Select 'Enable Device Flow', as we rely on the device flow authentication method.

  4. Disable webhooks (uncheck the 'Active' checkbox under 'Webhooks'). All other textboxes can be left empty.

  5. Under 'Repository permissions', select 'Read & write' for 'Contents'. This will provide users authenticating via the app just enough permissions to push and pull from repositories.

  6. Under 'Where can this GitHub App be installed?', select 'Any account'. This will enable users to push to their own user repositories or other organization repositaries, rather than just the repos of the user or organization owning this GitHub app.

  7. Save the Client ID provided in the information page of the app. You'll need this in the client. Save the Public link as well, as users will need to use this to grant access to particular repositories.

Client configuration

  1. gh-scoped-creds will need to know the "Client ID" of the created GitHub app to perform authentication. This can be either set with the environment variable GH_SCOPED_CREDS_CLIENT_ID, or be passed in as a commandline parameter --client-id to the gh-scoped-creds script when users use it to authenticate.

  2. Users will need to explicitly grant access to the repos they want to push to from this particular JupyterHub or HPC system (see Usage section below). Admins can set the GH_SCOPED_CREDS_APP_URL environment variable to inform users where to go to manage this access - gh-scoped-creds will print this out after authentication.

Usage

Grant access to the GitHub app

Users will first need to go to the public page of the GitHub app, and 'Install' the app on their account and in organizations with repos they want to push to. We highly recommend allowing access only to selected repositories, and explicitly select the repositories this hosted service (JupyterHub, HPC cluster, etc) should be able to push to. You can modify this list afterwards, to make sure you only grant the required permissions.

Given the common usage pattern where you are only pushing to a limited set of repositories from a particular hosted service, this should hopefully not be too cumborsome.

Admins can set the GH_SCOPED_CREDS_APP_URL to the Public Link of the GitHub app to make it easier for users to know where to go.

Authenticate to GitHub

The hosted service must have gh-scoped-creds installed.

  1. Open a terminal, and type gh-scoped-creds. If you're in a Python based Jupyter Notebook, you can also do:

    import gh_scoped_creds
    
    %ghscopedcreds
    

    This will offer to open the page in a new window for you, and conveniently copy the code you need to paste in the new window too.

    If you're on a non-containerized system (like a HPC), you must also specify the path to put the credentials files in explicitly with --git-credentials-path. The same path must be used in the gitconfig configuration mentioned earlier.

  2. It should give you a link to go to, and a code to input into the web page when that link is opened. Open the link, enter the code there (the page accepts pasting, so you can copy it from the output).

  3. Grant access to the device in the web page, and you're done!

Authentication is valid for 8 hours, and once it expires, this process will need to be repeated. In the future, we might have a web app or other process to make this less painful. However, keeping the length of this session limited drastically helps with security too.

Alternatives

  1. Create an ssh key specifically for the hosted service (JupyterHub, HPC cluster, etc) and add it to your GitHub account. If the key doesn't have a passphrase, this is very insecure - anyone who can exfiltrate your key once can keep it and use it whenever they wish. Even with a passphrase, the key can still be exfiltrated and passphrase stolen when used. There's also no way to restrict which repositories this can push to, which is a big issue.

  2. Create a Personal Access Token and use that. This is a little more insecure than the ssh key, as it can be used to make requests on your behalf too after being stolen! There is also no way to restrict which repositories you can push to.

  3. Create a GitHub deploy key for each repository you want to push to, for each hosted service you want to push from. While this lets you control which repos this ssh key can access, it is still stored long term at risk and can be exfiltrated.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

gh-scoped-creds-4.0.tar.gz (6.9 kB view details)

Uploaded Source

Built Distribution

gh_scoped_creds-4.0-py3-none-any.whl (7.4 kB view details)

Uploaded Python 3

File details

Details for the file gh-scoped-creds-4.0.tar.gz.

File metadata

  • Download URL: gh-scoped-creds-4.0.tar.gz
  • Upload date:
  • Size: 6.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/3.7.1 importlib_metadata/4.10.1 pkginfo/1.8.2 requests/2.27.1 requests-toolbelt/0.9.1 tqdm/4.62.3 CPython/3.9.12

File hashes

Hashes for gh-scoped-creds-4.0.tar.gz
Algorithm Hash digest
SHA256 23316842a135a14d2e0df7a99e2710d10590e04635f1536c351572d2bfff11ef
MD5 24672c065a94abe4126a172cc1e947dd
BLAKE2b-256 be25860f907e2140375d1bf09250e5b3f31d811118cf72417c077e749c30d165

See more details on using hashes here.

File details

Details for the file gh_scoped_creds-4.0-py3-none-any.whl.

File metadata

  • Download URL: gh_scoped_creds-4.0-py3-none-any.whl
  • Upload date:
  • Size: 7.4 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/3.7.1 importlib_metadata/4.10.1 pkginfo/1.8.2 requests/2.27.1 requests-toolbelt/0.9.1 tqdm/4.62.3 CPython/3.9.12

File hashes

Hashes for gh_scoped_creds-4.0-py3-none-any.whl
Algorithm Hash digest
SHA256 72ea8316d3686c6198fa0a7f4a0a17f7e0bd52bcab3c2e5f3032d3d820c1860d
MD5 7c3776834a12d412c7fe80227eb74d41
BLAKE2b-256 36933725788dae2a978b545df83f84b2ba7f9c057869f9274333d6faa9e99a8b

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page