Autosigning certificate requests

Before Puppet agent nodes can retrieve their configuration catalogs, they require a signed certificate from the local Puppet certificate authority (CA). When using Puppet’s built-in CA instead of an external CA, agents submit a certificate signing request (CSR) to the CA to retrieve a signed certificate after it's available.

By default, these CSRs must be manually signed by an admin user, using either the puppetserver ca command or the Node requests page in the Puppet Enterprise console.

Alternatively, to speed up the process of bringing new agent nodes into the deployment, you can configure the CA to automatically sign certain CSRs.

CAUTION: Autosigning CSRs changes the nature of your deployment’s security, and you should understand the implications before configuring it. Each type of autosigning has its own security impact.

Disabling autosigning

By default, the autosign setting in the [server] section of the CA’s puppet.conf file is set to $confdir/autosign.conf. The basic autosigning functionality is enabled upon installation.

Depending on your installation method, there might not be an allowlist at that location after the Puppet Server is running:

  • Open source Puppet: autosign.conf doesn’t exist by default.

  • Monolithic Puppet Enterprise (PE) installations: All required services run on one server, and autosign.conf exists on the primary server, but by default it's empty because the primary server doesn’t need to add other servers to an allowlist.

  • Split PE installations: Services like PuppetDB can run on different servers, the autosign.conf exists on the CA server and contains an allowlist of other required hosts.

If the autosign.conf file is empty or doesn’t exist, the allowlist is effectively empty. The CA Puppet primary server doesn’t autosign any certificates until the the autosign setting’s path is configured, or until the default autosign.conf file is a non-executable allowlist file. This file must contain correctly formatted content or a custom policy executable that the Puppet user has permission to run.

To explicitly disable autosigning, set autosign = false in the [server] section of the CA Puppet primary server’s puppet.conf. This disables CA autosigning even if the autosign.conf file or a custom policy executable exists.

For more information about the autosign setting in puppet.conf, see the configuration reference.

Naïve autosigning

Naïve autosigning causes the CA to autosign all CSRs.

To enable naïve autosigning, set autosign = true in the [server] section of the CA Puppet primary server’s puppet.conf.

CAUTION: For security reasons, never use naïve autosigning in a production deployment. Naïve autosigning is suitable only for temporary test deployments that are incapable of serving catalogs containing sensitive information.

Basic autosigning (autosign.conf)

In basic autosigning, the CA uses a config file containing an allowlist of certificate names and domain name globs. When a CSR arrives, the requested certificate name is checked against the allowlist file. If the name is present, or covered by one of the domain name globs, the certificate is autosigned. If not, it's left for a manual review.

Enabling basic autosigning

The autosign.conf allowlist file’s location and contents are described in its documentation.

Puppet looks for autosign.conf at the path configured in the [autosign setting] within the [server] section of puppet.conf. The default path is $confdir/autosign.conf, and the default confdir path depends on your operating system. For more information, see the confdir documentation.

If the autosign.conf file pointed to by the autosign setting is a file that the Puppet user can execute, Puppet instead attempts to run it as a custom policy executable, even if it contains a valid autosign.conf allowlist.

Note: In open source Puppet, no autosign.conf file exists by default. In Puppet Enterprise, the file exists by default but might be empty. In both cases, the basic autosigning feature is technically enabled by default but doesn’t autosign any certificates because the allowlist is effectively empty.

The CA Puppet primary server therefore doesn’t autosign any certificates until the autosign.conf file contains a properly formatted allowlist or is a custom policy executable that the Puppet user has permission to run, or until the autosign setting is pointed at an allowlist file with properly formatted content or a custom policy executable that the Puppet user has permission to run.

Security implications of basic autosigning

Basic autosigning is insecure because any host can provide any certname when requesting a certificate. Use it only when you fully trust any computer capable of connecting to the Puppet primary server.

With basic autosigning enabled, an attacker who guesses an unused certname allowed by autosign.conf can obtain a signed agent certificate from the Puppet primary server. The attacker could then obtain a configuration catalog, which can contain sensitive information depending on your deployment’s Puppet code and node classification.

Policy-based autosigning

In policy-based autosigning, the CA runs an external policy executable every time it receives a CSR. This executable examines the CSR and tells the CA whether the certificate is approved for autosigning. If the executable approves, the certificate is autosigned; if not, it's left for manual review.

Enabling policy-based autosigning

To enable policy-based autosigning, set autosign = <policy executable file> in the [server] section of the CA Puppet primary server’s puppet.conf.

The policy executable file must be executable by the same user as the Puppet primary server. If not, it is treated as a certname allowlist file.

Custom policy executables

A custom policy executable can be written in any programming language; it just has to be executable in a *nix-like environment. The Puppet primary server passes it the certname of the request (as a command line argument) and the PEM-encoded CSR (on stdin), and expects a 0 (approved) or non-zero (rejected) exit code.

After it has the CSR, a policy executable can extract information from it and decide whether to approve the certificate for autosigning. This is useful when you are provisioning your nodes and are embedding additional information in the CSR.

If you aren’t embedding additional data, the CSR contains only the node’s certname and public key. This can still provide more flexibility and security than autosign.conf, as the executable can do things like query your provisioning system, CMDB, or cloud provider to make sure a node with that name was recently added.

Security implications of policy-based autosigning

Depending on how you manage the information the policy executable is using, policy-based autosigning can be fast and extremely secure.

For example:
  • If you embed a unique pre-shared key on each node you provision, and provide your policy executable with a database of these keys, your autosigning security is as good as your handling of the keys. As long as it’s impractical for an attacker to acquire a PSK, it's impractical for them to acquire a signed certificate.

  • If nodes running on a cloud service embed their instance UUIDs in their CSRs, and your executable queries the cloud provider’s API to check that a node's UUID exists in your account, your autosigning security is as good as the security of the cloud provider’s API. If an attacker can impersonate a legit user to the API and get a list of node UUIDs, or if they can create a rogue node in your account, they can acquire a signed certificate.

When designing your CSR data and signing policy, you must think things through carefully. If you can arrange reasonable end-to-end security for secret data on your nodes, you can configure a secure autosigning system.

Policy executable API

The API for policy executables is as follows.
Run environment
  • The executable runs one time for each incoming CSR.

  • It is executed by the Puppet primary server process and runs as the same user as the Puppet primary server.

  • The Puppet primary server process is blocked until the executable finishes running. We expect policy executables to finish in a timely fashion; if they do not, it’s possible for them to tie up all available Puppet primary server threads and deny service to other agents. If an executable needs to perform network requests or other potentially expensive operations, the author is in charge of implementing any necessary timeouts, possibly bailing and exiting non-zero in the event of failure.

  • The executable must allow a single command line argument. This argument is the Subject CN (certname) of the incoming CSR.

  • No other command line arguments should be provided.

  • The Puppet primary server should never fail to provide this argument.

  • The executable receives the entirety of the incoming CSR on its stdin stream. The CSR is encoded in pem format.

  • The stdin stream contains nothing but the complete CSR.

  • The Puppet primary server should never fail to provide the CSR on stdin.

Exit status
  • The executable must exit with a status of 0 if the certificate should be autosigned; it must exit with a non-zero status if it should not be autosigned.

  • The Puppet primary server treats all non-zero exit statuses as equivalent.

Stdout and stderr
  • Anything the executable emits on stdout or stderr is copied to the Puppet Server log output at the debug log level. Puppet otherwise ignores the executable’s output; only the exit code is considered significant.