A seemingly common problem that people encounter is how to handle all of your users authorized_keys file.
People struggle over management, ensuring that users only have specific keys in the authorized_keys file or even a method for expiring keys. A centralized key management system could help provide all of this functionality with a little scripting.
One piece of functionality overlooked in OpenSSH is the AuthorizedKeysCommand configuration keyword. This configuration allows you to specify a command that will run during login to retrieve a users public key file from a remote source and perform validation just as if the authorized_keys file was local.
Here is an example directory structure for a set of users with SSH public keys that can be shared out via a web server:
users/
├── dave
│ └── keys
├── matt
│ └── keys
├── nathen
│ └── keys
└── paul
└── keys
Each of these keys files might look like:
ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQCt8lGmfZ0fxPz/66JlNg9CmZNaLsJ/TDrYnpBpiWWeuoLxP1tEbDiutApVOkjjQszBQV6CgvG3PeBYYAcJxUTRKhY8dUUbsAvVK3SRVwpr8jhtcohYgRE4V9/xPnwilDAfd9TymCMvM/mBpauQCyL40SImFQMJl5aBAhBiy6zyWx6WeDTzJ4+ZGUTmwFFyaWzzIqIZXWe1QiM98rfzle0mYM8KSKdTuGEf0EmY63MbMl3PQ61ms/qkR3fnKWpGF+EsigS0NgT6nBYoOZm5nFtrB2WM8nixyD5v82Z6yA6+O2SfLxtzJ6OcowtwtitrcZrAZdcNIwOAX1T7G4qcFEFn [email protected]
ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQCZ0N1dcto3td7j5/7UPCE2XlhDaCOZTlYtCgNifJygM5GNAG97JcChnnoYbdmiEM+dFMs7Jk6fS/WzG0Q0Ypu3rQ9AzzeUEMbhrFB90f28JsfUtgnkYuUF+1dNDGZn1fhYMlNwwyIt5s0KSS18iJNU6ZrSTudk9v1gyBM+Sxz97YMg2RiiGpCajPHzZbj2AwMl52MjT8ZDCGLt2qFo+w4u4BNQdtAA+zs/GiwgFbdGHM2HR1VxmII61LpvyyeuRkRwxN1ak3R7FcPMmYNhC9cvzbnvpmVcXxwXChI/9ceOm6DODCgHl9YeOgngoe5gEtZHnqtOWZWao8cFfd4wcEEN [email protected]
If I were to serve out the users/ directory using a webserver such as nginx, I might be able to request the keys for the user matt such as:
curl -sf https://keyserver.example.org/users/matt/keys
The AuthorizedKeysCommand expects an executable that takes a single argument, which is the username to retrieve the keys for. An example executable may look like:
#!/bin/bash
curl -sf https://keyserver.example.org/users/$1/keysName this file something like /usr/local/bin/userkeys.sh and make it executable: chmod a+x /usr/local/bin/userkeys.sh
Now add the following to your /etc/sshd/sshd_config file:
AuthorizedKeysCommand /usr/local/bin/userkeys.sh
AuthorizedKeysCommandUser nobody
Most operating systems have a nobody user, but you can replace that user with any non-root user that is different from the user running OpenSSH on the server. This should preferably be a user not already in use by another daemon.
Now, when a user logs in, userkeys.sh will be executed and if there are keys for that user they will be returned by our simple script.
Now all you need to do is manage that users/ directory via something like git and you are good to go.
If all your users are on GitHub, you can even have them use GitHub for the storage location of their SSH public keys, and you can replace the URL with https://github.com/$1.keys.
Although, if you usernames don't match GitHub, then you would have to maintain a lookup table that may get complicated.
This also works for GitHub Enterprise too, which if your company uses it could solve the username issues.
What would be the implications of adding hostname to the path? The use case is where you want to restrict a user's access to a specific set of machines?