This action
- starts the
ssh-agent
, - exports the
SSH_AUTH_SOCK
environment variable, and - loads one or several private SSH key into the agent.
It should work in all GitHub Actions virtual environments, including container-based workflows.
Windows and Docker support is, however, somewhat new. Since we have little feedback from the field, things might not run so smooth for you as we'd hope. If Windows and/or Docker-based workflows work well for you, leave a π at webfactory#17.
Also, using multiple GitHub deployment keys is supported; keys are mapped to repositories by using SSH key comments (see below).
When running a GitHub Action workflow to stage your project, run tests or build images, you might need to fetch additional libraries or vendors from private repositories.
GitHub Actions only have access to the repository they run for. So, in order to access additional private repositories, create an SSH key with sufficient access privileges. Then, use this action to make the key available with ssh-agent
on the Action worker node. Once this has been set up, git clone
commands using ssh
URLs will just work. Also, running ssh
commands to connect to other servers will be able to use the key.
- Generate a new SSH key with sufficient access privileges. For security reasons, don't use your personal SSH key but set up a dedicated one for use in GitHub Actions. See below for a few hints if you are unsure about this step.
- Make sure you don't have a passphrase set on the private key.
- Add the public SSH key to the private repository you are pulling from during the Github Action as a 'Deploy Key'.
- Add the private SSH key to the repository triggering the Github Action:
- In your repository, go to the Settings > Secrets menu and create a new secret. In this example, we'll call it
SSH_PRIVATE_KEY
. - Put the contents of the private SSH key file into the contents field.
- This key should start with
-----BEGIN ... PRIVATE KEY-----
, consist of many lines and ends with-----END ... PRIVATE KEY-----
.
- In your repository, go to the Settings > Secrets menu and create a new secret. In this example, we'll call it
- In your workflow definition file, add the following step. Preferably this would be rather on top, near the
actions/checkout@v4
line.
# .github/workflows/my-workflow.yml
jobs:
my_job:
...
steps:
- uses: actions/checkout@v4
# Make sure the @v0.9.0 matches the current version of the action
- uses: webfactory/[email protected]
with:
ssh-private-key: ${{ secrets.SSH_PRIVATE_KEY }}
# ... other steps
- If, for some reason, you need to change the location of the SSH agent socket, you can use the
ssh-auth-sock
input to provide a path.
There are cases where you might need to use multiple keys. For example, "deploy keys" might be limited to a single repository, so you'll need several of them.
You can set up different keys as different secrets and pass them all to the action like so:
# ... contents as before
- uses: webfactory/[email protected]
with:
ssh-private-key: |
${{ secrets.FIRST_KEY }}
${{ secrets.NEXT_KEY }}
${{ secrets.ANOTHER_KEY }}
The ssh-agent
will load all of the keys and try each one in order when establishing SSH connections.
There's one caveat, though: SSH servers may abort the connection attempt after a number of mismatching keys have been presented. So if, for example, you have six different keys loaded into the ssh-agent
, but the server aborts after five unknown keys, the last key (which might be the right one) will never even be tried. But when you're using GitHub Deploy Keys, read on!
When using Github deploy keys, GitHub servers will accept the first known key. But since deploy keys are scoped to a single repository, this might not be the key needed to access a particular repository. Thus, you will get the error message fatal: Could not read from remote repository. Please make sure you have the correct access rights and the repository exists.
if the wrong key/repository combination is tried.
To support picking the right key in this use case, this action scans key comments and will set up extra Git and SSH configuration to make things work.
- When creating the deploy key for a repository like
[email protected]:owner/repo.git
orhttps://github.com/owner/repo
, put that URL into the key comment. (Hint: Tryssh-keygen ... -C "[email protected]:owner/repo.git"
.) - After keys have been added to the agent, this action will scan the key comments.
- For key comments containing such URLs, a Git config setting is written that uses
url.<base>.insteadof
. It will redirectgit
requests to URLs starting with eitherhttps://github.com/owner/repo
or[email protected]:owner/repo
to a fake hostname/URL like[email protected]...:owner/repo
. - An SSH configuration section is generated that applies to the fake hostname. It will map the SSH connection back to
github.com
, while at the same time pointing SSH to a file containing the appropriate key's public part. That will make SSH use the right key when connecting to GitHub.com.
The following inputs can be used to control the action's behavior:
ssh-private-key
: Required. Use this to provide the key(s) to load as GitHub Actions secrets.ssh-auth-sock
: Can be used to control where the SSH agent socket will be placed. Ultimately affects the$SSH_AUTH_SOCK
environment variable.log-public-key
: Set this tofalse
if you want to suppress logging of public key information. To simplify debugging and since it contains public key information only, this is turned on by default.ssh-agent-cmd
: Optional. Use this to specify a custom location for thessh-agent
binary.ssh-add-cmd
: Optional. Use this to specify a custom location for thessh-add
binary.git-cmd
: Optional. Use this to specify a custom location for thegit
binary.
The action exports the SSH_AUTH_SOCK
and SSH_AGENT_PID
environment variables through the Github Actions core module.
The $SSH_AUTH_SOCK
is used by several applications like git or rsync to connect to the SSH authentication agent.
The $SSH_AGENT_PID
contains the process id of the agent. This is used to kill the agent in post job action.
Since each job runs in a fresh instance of the virtual environment, the SSH key will only be available in the job where this action has been referenced. You can, of course, add the action in multiple jobs or even workflows. All instances can use the same SSH_PRIVATE_KEY
secret.
If the private key is not in the PEM
format, you will see an Error loading key "(stdin)": invalid format
message.
Use ssh-keygen -p -f path/to/your/key -m pem
to convert your key file to PEM
, but be sure to make a backup of the file first π.
If you know that your favorite tool or platform of choice requires extra tweaks or has some caveats when running with SSH, feel free to open a PR to amend this section here.
If you are using this action on container-based workflows, make sure the container has the necessary SSH binaries or package(s) installed.
When you are building Docker images with docker build
or docker compose build
and need to provide the SSH keys to the build, don't forget to pass --ssh default=${{ env.SSH_AUTH_SOCK }}
on the command line to pass the SSH agent socket through. See the Docker documentation for more information on this option.
If you are using the docker/build-push-action
, you can do so by adding the following config.
- name: Build and push
id: docker_build
uses: docker/build-push-action@v2
with:
ssh: |
default=${{ env.SSH_AUTH_SOCK }}
Make sure not to miss the next section, though.
When you pass the SSH agent socket to the Docker build environment and want to use multiple GitHub deploy keys, you need to copy the Git and SSH configuration files to the build environment as well. This is necessary in addition to forwarding the SSH agent socket into the build process. The config files are required so that Git can pick the right one from your deployment keys.
This requires an additional step in the workflow file after the ssh-agent
step and before the Docker build step. You also need two additional lines in the Dockerfile
to actually copy the configs.
The following example will:
- collect the necessary Git and SSH configuration files in a directory that must be part of the Docker build context so that...
- ... the files can be copied into the Docker image (or an intermediate build stage).
Workflow:
- name: ssh-agent setup
...
- name: Collect Git and SSH config files in a directory that is part of the Docker build context
run: |
mkdir root-config
cp -r ~/.gitconfig ~/.ssh root-config/
- name: Docker build
# build-push-action | docker [compose] build | etc.
...
Dockerfile:
# Copy the two files in place and fix different path/locations inside the Docker image
COPY root-config /root/
RUN sed 's|/home/runner|/root|g' -i.bak /root/.ssh/config
Keep in mind that the resulting Docker image now might contain these customized Git and SSH configuration files! Your private SSH keys are never written to files anywhere, just loaded into the SSH agent and forwarded into the container. The config files might, however, give away details about your build or development process and contain the names and URLs of your (private) repositories. You might want to use a multi-staged build to make sure these files do not end up in the final image.
If you still get the error message: fatal: Could not read from remote repository. Please make sure you have the correct access rights and the repository exists.
, you most likely forgot one of the steps above.
If you are using private repositories in your dependencies like this:
stuff = { git = "ssh://[email protected]/myorg/stuff.git", branch = "main" }
... you will need to change a configuration in the workflow for Windows machines in order to make cargo able to clone private repositories.
There are 2 ways you can achieve this:
- Add this step once in your job before any cargo command:
- name: Update cargo config to use Git CLI
run: Set-Content -Path $env:USERPROFILE\.cargo\config.toml "[net]`ngit-fetch-with-cli = true"
This will configure Cargo to use the Git CLI as explained in the Cargo's documentation.
- Alternatively you can set it to the environment variables for the entire workflow:
env:
CARGO_NET_GIT_FETCH_WITH_CLI: true
xcodebuild
by default uses Xcode's built-in Git tooling. If you want to use GitHub Deploy Keys as supported by this action, however, that version of Git will lack the necessary URL remapping. In this case, pass -scmProvider system
to the xcodebuild
command, as mentioned in Apple's documentation.
The following items are not issues, but beyond what this Action is supposed to do.
When using ssh
to connect from the GitHub Action worker node to another machine, you can forward the SSH Agent socket and use your private key on the other (remote) machine. However, this Action will not configure known_hosts
or other SSH settings on the remote machine for you.
This Action is designed to pass the SSH key directly into ssh-agent
; that is, the key is available in memory on the GitHub Action worker node, but never written to disk. As a consequence, you cannot pass the key as a build argument or a mounted file into Docker containers that you build or run on the worker node. You can, however, mount the ssh-agent
Unix socket into a Docker container that you run, set up the SSH_AUTH_SOCK
env var and then use SSH from within the container (see webfactory#11).
If you want to use ssh-keyscan
to add additional hosts (that you own/know) to the known_hosts
file, you can do so with a single shell line in your Action definition. You don't really need this Action to do this for you.
As a side note, using ssh-keyscan
without proper key verification is susceptible to man-in-the-middle attacks. You might prefer putting your known SSH host key in your own Action files to add it to the known_hosts
file. The SSH host key is not secret and can safely be committed into the repo.
In order to create a new SSH key, run ssh-keygen -t ed25519 -a 100 -f path/to/keyfile
, as suggested in this blog post.
If you need to work with some older server software and need RSA keys, try ssh-keygen -t rsa -b 4096 -o -f path/to/keyfile
instead.
Both commands will prompt you for a key passphrase and save the key in path/to/keyfile
.
In general, having a passphrase is a good thing, since it will keep the key encrypted on your disk. When using the key with this action, however, you need to make sure you don't
specify a passphrase: The key must be usable without reading the passphrase from input. Since the key itself is stored using GitHub's "Secret" feature, it should be fairly safe anyway.
To actually grant the SSH key access, you can β on GitHub β use at least two ways:
-
Deploy keys can be added to individual GitHub repositories. They can give read and/or write access to the particular repository. When pulling a lot of dependencies, however, you'll end up adding the key in many places. Rotating the key probably becomes difficult. The deploy key needs to be added to the private repository that is being fetched as a private dependency.
-
A machine user can be used for more fine-grained permissions management and have access to multiple repositories with just one instance of the key being registered. It will, however, count against your number of users on paid GitHub plans.
As a note to my future self, in order to work on this repo:
-
Clone it
-
Run
yarn install
to fetch dependencies -
hack hack hack
-
node index.js
. Inputs are passed throughINPUT_
env vars with their names uppercased.On *nix use:
env "INPUT_SSH-PRIVATE-KEY=\`cat file\`" node index.js
On Windows (cmd):
set /P INPUT_SSH-PRIVATE-KEY=< file node index.js
On Windows (PowerShell):
${env:INPUT_SSH-PRIVATE-KEY} = (Get-Content .\test-keys -Raw); node index.js node index.js
-
Run
npm run build
to updatedist/*
, which holds the files actually run -
Read https://help.github.com/en/articles/creating-a-javascript-action if unsure.
-
Maybe update the README example when publishing a new version.
This action was written by webfactory GmbH, Bonn, Germany. We're a software development agency with a focus on PHP (mostly Symfony). If you're a developer looking for new challenges, we'd like to hear from you!
Copyright 2019 β 2023 webfactory GmbH, Bonn. Code released under the MIT license.