itsdrike.com/content/posts/git-credentials.md

27 KiB

title date tags sources changelog
Managing (multiple) git credentials 2022-07-27
programming
git
<https://docs.github.com/en/get-started/getting-started-with-git/caching-your-github-credentials-in-git>
<https://docs.github.com/en/authentication/connecting-to-github-with-ssh>
<https://www.onwebsecurity.com/configuration/git-on-windows-location-of-global-configuration-file.html>
<https://security.stackexchange.com/questions/90077/ssh-key-ed25519-vs-rsa>
<https://www.shellhacks.com/git-config-username-password-store-credentials/>
<https://git-scm.com/book/en/v2/Git-Tools-Credential-Storage>
<https://engineeringfordatascience.com/posts/how_to_manage_multiple_git_accounts_on_the_same_machine/>
<https://git-scm.com/docs/gitcredentials>
<https://www.baeldung.com/ops/git-configure-credentials>
<https://www.freecodecamp.org/news/manage-multiple-github-accounts-the-ssh-way-2dadc30ccaca/>
<https://blog.bitsrc.io/how-to-use-multiple-git-accounts-378ead121235>
<https://www.freecodecamp.org/news/the-ultimate-guide-to-ssh-setting-up-ssh-keys/>
<https://www.atlassian.com/git/tutorials/setting-up-a-repository/git-config>
2023-01-30
Add note about disabling commit signing
Add alternative command for copying on wayland
Fix typos and text wrapping

Many people often find initially setting up their git user a bit unclear, especially when it comes to managing multiple git users on a single machine. But even managing credentials for just a single user can be quite complicated without looking into it a bit deeper. Git provides a lot of different options for credential storage, and picking one can be hard without knowing the pros and cons of that option.

Even if you already have your git set up, I'd still recommend at least looking at the possible options git has for credential storage, find the method you're using and make sure it's actually secure enough for your purposes. But looking over the other methods may be helpful too, as there may be a better option to what you're using which you didn't even know about.

Adding user into global configuration

Let's first look at the simple single local git user setup:

git config --global user.name ItsDrike
git config --global user.email itsdrike@example.com
git config --global user.signingkey B014E761034AF742  # Signing key isn't required

So what do these commands actually do? It's really quite simple. The git program has a single global configuration file located in ~/.config/git/config (or in %APPDATA%/.gitconfig for Windows). Once this git config command runs, it actually just writes the settings into the config file using the TOML format:

[user]
    name = ItsDrike
    email = itsdrike@example.com
    signingkey = B014E761034AF742

Adding user into local configuration

A thing you may have noticed in the section above is the use of --global flag when running the git config commands. This flag means that this configuration will be stored into the global configuration that applies for every git repository. But we can also define a configuration for a single repository, by using the --local flag instead. Local configurations are actually also the default option, so we can even omit the flag entirely.

This local configuration will then be stored in the .git folder of your project, specifically in .git/config. The settings set in this configuration will take precedence over the global ones, meaning you can have some default git user, and some other user for a single specific local project.

git config --local user.name ItsDrike
git config --local user.email itsdrike@example.com
git config --local user.signingkey B014E761034AF742  # Signing key isn't required

{{< notice tip >}} If your global git config has commit signing enabled by default, but you don't want to sign commits for the locally configured account, you can disable it with:

git config --local commit.gpgsign false

{{< /notice >}}

Git credentials

User configuration is one thing, but there's another important part of account configuration to consider, that is storing the credentials. Even though you don't technically need to store the credentials, since git can just ask you to enter them each time you clone a private repo, or push into a repo, it's a huge annoyance to have to do this each time. So instead, we can use one of the below methods to store the credentials with git for longer.

Credentials in remote-url

The most basic way of specifying credentials is to just provide them via HTTPS. You can do this in more ways, but let's first take a look at the most straight-forward method, which is to store them into the remote URL directly:

# While clonning:
git clone https://<USERNAME>:<PASSWORD>@github.com/path/to/repo.git
# After initialized repo without any added remote:
git remote add origin
# On an already clonned repository without the credentials:
git remote set-url origin https://<USERNAME>:<PASSWORD>@github.com/path/to/repo.git

Since this method requires you to specify these credentials for every repository individually, it's easily usable with multiple accounts, but it's also still quite annoying since you'll need to set the credentials with each new repo.

{{< notice note >}} The password here is generally meant to be a user password for the git hosting provider site, however many platforms do also have support for "Personal Access Tokens", which are a safer, because they're limited in what they can do with your account (for example they may only allow you to pull/push code, but not to change the account's email). {{< /notice >}}

{{< notice warning >}} This method stores your credentials in the project's git config file in .git/config. Since this is a simple URL to one of the proejcts remotes, it will just be stored in this config file in plaintext without any form of encryption.

Bear this in mind when giving someone access to the project directory, your credentials will be present in that directory! {{< /notice >}}

Git credential contexts

To avoid some repetition, git supports configuring per context credentials. You can configure a specific git context to use a specific username:

git config --global credential.https://github.com.username <USERNAME>

Alternatively, we can directly edit the global git configuration:

[credential "https://github.com"]
	username = <USERNAME>

[credential "https://gitlab.com"]
	username = <USERNAME2>

[credential "https://gitlab.work_company.com"]
	username = <USERNAME2>

Each credential context is defined by a URL. This context will then be used to look up specific configuration. For example if we're accessing https://github.com/ItsDrike/itsdrike.com, git looks into the config file to see if a section matches this context. It will consider the two a match, if the context matches on both the protocols (http/https), and then on the host portion (github.com/gitlab.com/...). It can also optionally check the paths too, if they are present (/ItsDrike/itsdrike.com)

{{< notice note >}} Git matches the hosts directly, without considering if they come from the same domain, so if subdomain differs, it will not register as a match. For example, for context of https://gitlab.work_company.com/user/repo.git, it wouldn't match a configuration section for https://work_company.com, since work_company.com != gitlab.work_company.com.

The paths are also matched exactly (if they're included), so for the example context from above, we would not get a match on a config section with https://gitlab.work_company.com/user, only on https://gitlab.work_company.com/user/repo.git (in addition to the config entry without path https://gitlab.work_company.com). {{< /notice >}}

This does sound like a great option for multi-account usage, however the issue with this approach is that these credential contexts can only be used to store usernames, they don't support storing passwords, and you'll instead be prompted to enter your password each time. But it does save you from re-typing the username each time.

{{< notice info >}} The username will be stored in git's global config file in plaintext, making it potentially unsafe if you're worried about leaking your username (not password) for the git hosting provider.

If you're using the global configuration, this generally shouldn't be a big concern, since the username won't actually be in the project file unlike with the remote-urls. However if you share a machine with multiple people, you may want to consider securing your global configuration file (~/.config/git/config) using your filesystem's permission controls to prevent others from reading it.

If you're defining contexts in local project's config though, you should be aware that the username will be present in .git/config, and sharing this project with others may leak it. {{< /notice >}}

Git credential helpers

If you want to avoid both username and password repetition, and to have a safer way of storing your credentials, you can use git's "credential helpers". These allow you to store your data in multiple ways, and even integrate with 3rd party systems like password keychains. These credential helpers still use the same form of sending credentials, which is to send them over HTTPS.

Out of the box, git provides 2 credential helpers:

  • Cache: credentials stored in RAM memory for short durations
  • Store: credentials stored indefinitely on disk

Store credential helper

To configure the store credential helper, you can run:

git config --global credential.helper store

By default, this file will be stored in ~/.git-credentials, but this path can be changed. I'd suggest using ~/.config/git/git-credentials to avoid clutter in your home directory. To change the file, you can use the file option:

git config --global credentials.helper 'store --file=/full/path/to/git-credentials'

Once the helper is configured, you will first still get asked for your username and password, and only after that first time you enter them will the get cached into this credentials file.

{{< notice info >}}

The credentials file will cache the data in this format:

https://<USERNAME>:<PASSWORD>@github.com

Which is indeed a plaintext format, however the file will be protected with your file system permissions, and access should be limited to you (as the user who owns the file). And since this file should live somewhere outside of the project's directory, the project can be safely shared with others without worrying about leakage. {{< /notice >}}

Cache credential helper

To configure the cache credential helper, you can run:

git config --global credential.helper cache

A timeout option can also be provided, allowing us to define how long should the credentials be kept in memory in seconds.

git config --global credential.helper 'cache --timeout=86400'

The cache credential helper will never write your credential data to disk, although credentials are accessible using Unix sockets. These sockets are protected using file permissions that are limited to the user who stored them though, so even in multi-user machine, generally speaking, they are secure.

Custom credential helpers

Apart from these default options, you can also use custom helpers. These allow us to do more sophisticated credential management by delegating to 3rd party applications and services.

A commonly used external credential helper is for example the Git Credential Manager (GCM). GCM can even handle things like 2 factor authentication, or using OAuth2.

If you want to, you can even write your own custom credential helper to handle your exact needs, in which case I'd recommend going over git's official documentation about the credential helper system here, where they go over this system in depth, including some examples of a basic custom provider.

SSH Keys

Most modern git servers also provide a way to access their repositories using SSH keys rather than username and password over HTTPS. This approach is significantly better, since guessing SSH keys is generally much harder, and they can easily be revoked. They also generally aren't nowhere near as powerful as full user passwords, so even if they are compromised, the attacker would only have a limited access.

SSH uses public-private key pair, which means you will need to give out the public key over to the git hosting platform, and keep the private part on your machine for authentication. Using the public key, the server will then be able to safely verify that your connection is valid, without even actually knowing the key. This means that even if the git hosting server has leaked your stored SSH key, it would be useless without the private key on your machine.

The main downside to using SSH is that it uses non-standard ports. This may mean hitting the firewall some networks or proxies, making communication with the remote server impossible.

Generating an SSH key

To generate an SSH key, you can use ssh-keygen command line utility. Generating keys should always be done independently from the git hosting provider, since they don't shouldn't need to see your private key at any point.

The command for this key generation looks like this:

ssh-keygen -t ed25519 -C "<COMMENT>"
  • The -C flag allows you to specify a comment, which you can use to specify what this key will be used for. If you don't need a comment, you can also omit this flag.
  • The -t flag specifies the key type. The default type for SSH keys is rsa, however I'd suggest using ed25519 which is considered safer and more performant than RSA keys. If you will decide to use rsa, make sure to use a key size of at least 2048 bits, but for better security, but ideally you should try to use a key size of 4096.

After running this command, you will be asked to specify a file where this key should be stored. You will probably want to use some meaningful name, so that you can easily find it later. I'd recommend storing the keys in ~/.ssh/git, so you can have all of your git ssh keys grouped together and separated from SSH keys for actual machines or other things.

{{< notice info >}} Make sure to add the ~/.ssh (or C:\Users\your_username\.ssh for Windows) prefix to your filename, so the key is correctly added to the .ssh folder. You should keep your keys in this folder, since it is already protected by the filesystem from reading by other users. {{< /notice >}}

Once you select a file name, you will be asked to set a passphrase. You can opt to leave this empty by pressing enter without entering anything. Going with a passphrase protected key is safer, however it will also mean you will need to type your password each time, which may be annoying. However there is a way to cache this passphrase with SSH agent, which you can read more about in the GitHub's docs. Using passphrase is significantly better for your system's security, since it means that even if the private key got leaked somehow, it would be pretty much useless without the passphrase, which the attacker likely wouldn't have.

This will then generate two keys: a public key, denoted by the file extension .pub and a private key, with no file extension.

Add public key to your hosting provider's account

Now that you've create a public and private SSH key pair, you will need to let your git hosting provider know about it. It is important that you only give the public key (file with .pub extension) to your provider, and not your private key.

Instructions on how to add the public SSH key will differ for each platform, here are some links to documentations for the most commonly used platforms:

{{< notice tip >}} The documentation may tell you to use pbcopy or some other command line tool to copy the SSH key contents to your clipboard. For example:

pbcopy < ~/.ssh/id_ed25519.pub

However, if you are having trouble with this command, you can use xsel --clipboard --input < ~/.ssh/id_ed25519.pub instead, or wl-copy < ~/.ssh/id_ed25519 on wayland. You can also just simply open up the public key file in any editor of your choosing, and copy the entire file contents with Ctrl+C (if you're one of the weird people that use windows, this is your only option). {{< /notice >}}

Test if it works

After adding the public key to your git hosting provider, you can verify that everything went well and the SSH key is recognized. To run this test, you can simply issue this command (should work on both Unix and Windows systems):

ssh -T git@github.com -i ~/.ssh/id_ed25519

Running this command should produce a welcome message informing you that the connection works.

If you are unsuccessful, you can run the command in verbose mode in order to get more details on why your connection was not established.

ssh -Tvvv git@github.com -i ~/.ssh/id_ed25519

SSH Configuration file

To meaningfully use your key, you'll want register some specific host name for your key, so you won't need to use the -i flag. You can do this by editing (or creating) ~/.ssh/config file (or C:\Users\your_username\.ssh\config for Windows).

An example configuration file with multiple git accounts:

# Personal GitHub account
HOST github.com
    HostName github.com
    User git
    IdentityFile ~/.ssh/git/personal_gh

# Personal GitLab account
HOST gitlab.com
    HostName gitlab.com
    User git
    IdentityFile ~/.ssh/git/personal_gl

# Work GitHub account
HOST work.github.com
    HostName github.com
    User git
    IdentityFile ~/.ssh/git/work_gh

When you have multiple accounts with the same HostName (same git hosting provider), you will need to specify a unique Host name.

To then make sure this configuration works, you can run another test command, but this time without specifying the key file explicitly, as it should now be getting picked up from the settings:

ssh -T git@github.com
ssh -T git@gitlab.com
ssh -T git@work.github.com
# If you've specified the `User git` in your config file, you can even omit the username here:
ssh -T github.com
ssh -T gitlab.com
ssh -T work.github.com

Using the SSH keys

No let's finally get to actually using these keys in your repositories. Doing this can be pretty straight-forward, as it is very similar to the first method of handling credentials which I've talked about, being storing the credentials in the remote-url. However this time, instead of using the actual credentials, and therefore making the project directory unsafe to share, as it contains your password in plaintext, it will actually only contain the HOST name you've set in your config, without leaking any keys.

The commands to set this up are very similar, however instead of https://<USERNAME>:<PASSWORD>@github.com, we now use git@HOST:

# While clonning:
git clone git@github.com/user/repo.git
# After initialized repo without any added remote:
git remote add origin git@gitlab.com/user/repo.git
# On an already clonned repository without the credentials:
git remote set-url origin git@work.github.com/user/repo.git

This method does have the same disadvantage as the with the credentials passed directly into the remote-urls, which is that you will need to do something extra for every repository where you need the credentials, but unlike with remote-urls, it does not introduce security issues with storing the credentials in plaintext, and you also don't need to remember the username or the password, instead you just need to know the host you set in your ssh config.

Which method to use for credentials

Generally, using SSH keys is the safest approach, but it can also be a bit annoying since it requires you to specify the SSH host for each repository in it's remote url. For that reason, the approach that I would recommend is using git's credential helper system to store your credentials instead.

However if you will go with this method, make sure that you're using a personal access token instead of the actual account's password, to limit the permissions an attacker would gain in case your credentials were leaked.

If your git hosting platform doesn't provide access tokens, this method becomes a lot more dangerous to use, since if an attacker would somehow obtain the credentials file from your system, they would be able to gain full access to your account on that git host platform. That's why in that case, you should really consider using SSH keys instead, even if it's a bit less convenient, as they can be easily revoked and only allow limited access, just like personal access tokens.

Tackling credentials for multiple accounts

Credentials for differing hosts

When it comes to managing multiple accounts, this gets a bit more tricky. But if each of your accounts lives on a different domain/host, you can still use credential helpers without any issues, since it can handle multiple credentials for multiple websites out of the box. If you're using the file credential helper, this would result in the git-credentials file looking like this:

https://<USERNAME>:<PASSWORD>@github.com
https://<USERNAME2>:<PASSWORD2>@gitlab.com

With that, whenever you'd try to pull/push with the remote url, git will go through this file in order, searching for the first matching host. So for example when using a remote url belonging to github.com domain, the first line would apply, while if your remote url belongs to gitlab.com, the second line would apply. This means that if your accounts are from different providers, you can avoid the hassle of doing anything more complicated.

However if you have more accounts on a single host, you will need to somehow let git know what to do.

Using credential contexts

The good news is that even with same domains, you can actually still use the git credentials as your default method, and use git credential contexts to find a username. With that, even if you're using the same host, git will know to look for a specific username in the credentials file now, which should be sufficient distinction to match any amount of different credentials.

However the issue with git contexts is that they need to match the path component exactly, so even though you can configure git to use different contexts for different repositories in your global config, you can't configure it to use a certain context for a partial match on path, so you'd need to specify each repository which should use custom credentials into your global git configuration, which is not great.

Instead, you should use the local git configuration of each project and specify a git context with the username you want to use for that project. That way, you won't need to keep config for every non-default project in your global config, and yet still use the same file credential helper to store all of your credentials in a single place.

git config --local credential.https://github.com.username <USERNAME>

{{< notice info >}} Once again, this will store the credential context into the local project's git configuration (in .git/config), which is using plaintext, which means you might end up leaking your username (not password), if you give someone access to this project's directory.

The actual password will however be completely safe, as it should only be present in the git-credentials file, which should be located elsewhere, and configured from the global git config. So this only affects you if you want to keep your username for that git hosting provider private too. If you do, you will need to keep this fact in mind when sharing project files, or use a different method. {{< /notice >}}

Using different credentials file

The alternative to using credential contexts with your plaintext stored username would be using multiple git-credentials files, and simply overriding the credential helper system in the local config, setting a different file for the store credential helper. This could for example look like this:

git config credentials.helper 'store --file=/home/user/.config/git-credentials-work'

With this approach, you can have your credentials kept in multiple separate credential files, and just mention the path to the file you need for each project.

Security-wise, this method is better because your username will be kept outside of the project in the referenced git credential file, which should be secured by the file system's permissions to prevent reads from other users. However practicality-wise, it may be a bit more inconvenient to type and even to remember the path to each credential file.

SSH keys instead

The thing you may have noticed about all of these methods is that you'll generally need to do some extra work for all repositories that require non-default credentials. So even though relying on git's file credential helper is convenient for the default case, extending it to non-default cases will always require doing some extra configuration.

This extra configuration is inevitable, which is why I'd suggest going with SSH keys instead, which are pretty much equally as annoying, requiring you to do something extra for each non-default project (specifying them in the remote URL). However as I've already explained, they're pretty much the most secure way to handle credentials. So instead of doing some extra work just to configure a less secure method, you might as well do an equal amount of work and use the more secure way with SSH keys.

The only disadvantage to this method is then the use of non-standard ports, which some networks might end up blocking, making connection to the server [pretty much]({{< ref "posts/escaping-isolated-network#port-22-is-blocked" >}}) unreachable from those networks.

Make convenience aliases

If you really dislike the idea of all of this repetition, I'd suggest making short-hands for whichever method you ended up picking, in the form of git aliases (you can also use shell aliases though). Git supports defining aliases through it's configuration file, where you can use the [alias] section for them.

[alias]
# Clone the repository with the SSH host prefixed
work-clone="!sh -c 'git clone git@work.github.com:$1'"
# Make current repository use the work git credentials file
make-work="config --local credentials.helper 'store --file=/path/to/work/credentials'"
# Set the username for credentials to your work account, so it can find it in default git credentials
use-work-uname="config --local credential.https://github.com.username my-work-username"

To then use these aliases, you can simply execute them as you would any other git command:

git work-clone ItsDrike/itsdrike.com
git make-work
git user-work-uname