Enabling TLS
Last updated
Was this helpful?
Last updated
Was this helpful?
The NATS server uses modern TLS semantics to encrypt client, route, and monitoring connections. Server configuration revolves around a tls map, which has the following properties:
cert_file
TLS certificate file.
key_file
TLS certificate key file.
ca_file
cipher_suites
When set, only the specified TLS cipher suites will be allowed. Values must match the golang version used to build the server.
curve_preferences
List of TLS cipher curves to use in order.
insecure
Skip certificate verification. This only applies to outgoing connections, NOT incoming client connections. NOT Recommended
min_version
Minimum TLS version. Default is "1.2".
timeout
verify
verify_and_map
verify_cert_and_check_known_urls
pinned_certs
List of hex-encoded SHA256 of DER encoded public key fingerprints. When present, during the TLS handshake, the provided certificate's fingerprint is required to be present in the list or the connection is closed. This sequence of commands generates an entry for a provided certificate: `openssl x509 -noout -pubkey -in
openssl pkey -pubin -outform DER
openssl dgst -sha256`.
The simplest configuration:
Or by using :
Notice that the log indicates that the client connections will be required to use TLS. If you run the server in Debug mode with -D or -DV, the logs will show the cipher suite selection for each connected client:
When a tls section is specified at the root of the configuration, it also affects the monitoring port if https_port option is specified. Other sections such as cluster can specify a tls block.
As of NATS v2.10.4
Some environments prefer having clients' TLS connections be initiated right away, that is, not having any traffic sent in clear text. It was possible to by-pass this using a websocket connection. However, if a websocket connection is not desired, the server can be configured to perform a TLS handshake before sending the INFO protocol message.
Only clients that implement an equivalent option would be able to connect if the server runs with this option enabled.
The configuration would look something like this:
However, the parameter can be set to auto or a Golang time duration (e.g. 250ms) to fallback to the original behavior. This is intended for deployments where it is known that not all clients have been upgraded to a client library providing the TLS-first handshake option.
After the delay has elapsed without receiving the TLS handshake from the client, the server reverts to sending the INFO protocol so that older clients can connect. Clients that do connect with the "TLS first" option will be marked as such in the monitoring's Connz page/result. It will allow the administrator to keep track of applications still needing to upgrade.
The configuration would be similar to:
With the above value, the fallback delay used by the server is 50 milliseconds.
The duration can be explicitly set, say 300 milliseconds:
It is understood that any configuration other than "true" will result in the server sending the INFO protocol after the elapsed amount of time without the client initiating the TLS handshake. Therefore, for administrators who do not want any data transmitted in plain text, the value must be set to "true" only. It will require applications to be updated to a library that provides the option, which may or may not be readily available.
The timeout setting enables you to control the amount of time that a client is allowed to upgrade its connection to tls. If your clients are experiencing disconnects during TLS handshake, you'll want to increase the value, however, if you do be aware that an extended timeout exposes your server to attacks where a client doesn't upgrade to TLS and thus consumes resources. Conversely, if you reduce the TLS timeout too much, you are likely to experience handshake errors.
The ca_file file should contain one or more Certificate Authorities in PEM format, in a bundle. This is a common format.
When a certificate is issued, it is often accompanied by a copy of the intermediate certificate used to issue it. This is useful for validating that certificate. It is not necessarily a good choice as the only CA suitable for use in verifying other certificates a server may see.
Do consider though that organizations issuing certificates will change the intermediate they use. For instance, a CA might issue intermediates in pairs, with an active and a standby, and reserve the right to switch to the standby without notice. You probably would want to trust both of those for the ca_file directive, to be prepared for such a day, and then after the first CA has been compromised you can remove it. This way the roll from one CA to another will not break your NATS server deployment.
Explaining Public key infrastructure, Certificate Authorities (CA) and x509 certificates fall well outside the scope of this document. So does an explanation on how to obtain a properly trusted certificates.
If anybody outside your organization needs to connect, get certs from a public certificate authority. Think carefully about revocation and cycling times, as well as automation, when picking a CA. If arbitrary applications inside your organization need to connect, use a cert from your in-house CA. If only resources inside a specific environment need to connect, that environment might have its own dedicated automatic CA, eg in Kubernetes clusters, so use that.
Only for testing purposes does it make sense to generate self-signed certificates, even your own CA. This is a short guide on how to do just that and what to watch out for.
DO NOT USE these certificates in production!!!
As they should, these are not trusted by the system your server or clients are running on.
One option is to specify the CA in every client you are using. In case you make use of verify, verify_and_map or verify_cert_and_check_known_urls you need to specify ca_file in the server. If you are having a more complex setup involving cluster, gateways or leaf nodes, ca_file needs to be present in tls maps used to connect to the server with self-signed certificates. While this works for server and libraries from the NATS ecosystem, you will experience issues when connecting with other tools such as your Browser.
Another option is to configure your system's trust store to include self-signed certificate(s). Which trust store needs to be configured depends on what you are testing.
This may be your OS for server and certain clients.
The runtime environment for other clients like Java, Python or Node.js.
Your browser for monitoring endpoints and websockets.
Please check your system's documentation on how to trust a particular self-signed certificate.
Another common problem is failed identity validation. The IP or DNS name to connect to needs to match a Subject Alternative Name (SAN) inside the certificate. Meaning, if a client/browser/server connect via tls to 127.0.0.1, the server needs to present a certificate with a SAN containing the IP 127.0.0.1 or the connection will be closed with a handshake error.
When verify_cert_and_check_known_urls is specified, Subject Alternative Name (SAN) DNS records are necessary. In order to successfully connect there must be an overlap between the DNS records provided as part of the certificate and the urls configured. If you dynamically grow your cluster and use a new certificate, this route or gateway the server connects to will have to be reconfigured to include an url for the new server. Only then can the new server connect. If the DNS record is a wildcard, matching according to rfc6125 will be performed. Using certificates with a wildcard Subject Alternative Name (SAN) and configuration with url(s) that would match are a way to keep the flexibility of dynamic cluster growth without configuration changes in other clusters.
When generating your certificate you need to make sure to include the right purpose for which you want to use the certificate. This is encoded in key usage and extended key usage. The necessary values for key usage depend on the ciphers used. Digital Signature and Key Encipherment are an interoperable choice.
With respect to NATS the relevant values for extended key usage are:
TLS WWW server authentication - To authenticate as server for incoming connections. A NATS server will need a certificate containing this.
TLS WWW client authentication - To authenticate as client for outgoing connections. Only needed when connecting to a server where verify, verify_and_map or verify_cert_and_check_known_urls are specified. In these cases, a NATS client will need a certificate with this value.
Leaf node connections can be configured with verify as well. Then the connecting NATS server will have to present a certificate with this value too. Certificates containing both values are an option.
Cluster connections always have verify enabled. Which server acts as client and server comes down to timing and therefore can't be individually configured. Certificates containing both values are a must.
Gateway connections always have verify enabled. Unlike cluster outgoing connections can specify a separate cert. Certificates containing both values are an option that reduce configuration.
Note that it's common practice for non-web protocols to use the TLS WWW authentication fields, as a matter of history those have become embedded as generic options.
The simplest way to generate a CA as well as client and server certificates is mkcert. This zero config tool generates and installs the CA into your local system trust store(s) and makes providing SAN straight forward. Check its documentation for installation and your system's trust store. Here is a simple example:
Generate a CA as well as a certificate, valid for server authentication by localhost and the IP ::1(-cert-file and -key-file overwrite default file names). Then start a NATS server using the generated certificate.
Now you should be able to access the monitoring endpoint https://localhost:8222 with your browser. https://127.0.0.1:8222 however should result in an error as 127.0.0.1 is not listed as SAN. You will not be able to establish a connection from another computer either. For that to work you have to provide appropriate DNS and/or IP SAN(s)
To generate certificates that work with verify and cluster/gateway/leaf_nodes provide the -client option. It will cause the appropriate key usage for client authentication to be added. This example also adds a SAN email for usage as user name in verify_and_map.
Please note:
That client refers to connecting process, not necessarily a NATS client.
mkcert -clientwill generate a certificate with key usage suitable for client and server authentication.
Examples in this document make use of the certificates generated so far. To simplify examples using the CA certificate, copy rootCA.pem into the same folder where the certificates were generated. To obtain the CA certificate's location use this command:
Once you are done testing, remove the CA from your local system trust store(s).
Alternatively, you can also use openssl to generate certificates. This tool allows a lot more customization of the generated certificates. It is more complex and does not manage installation into the system trust store(s).
However, for inspecting certificates it is quite handy. To inspect the certificates from the above example execute these commands:
Using a TLS-terminating reverse proxy with NATS requires some specific configuration on the server. In a typical proxy scenario, the client to proxy communication is secured and the proxy to server is insecure. This causes a "mismatch" because the server appears to be insecure but the client is told to connect securely. To fix this, the server must be configured as "tls available". This is done via an empty tls block and the allow_non_tls flag.
Once this is configured, your client can connect to the proxy with normal (language specific) tls configuration. Please make sure you are using the appropriate version of your language specific client.
nats.go
v1.31.0
nats.js
2024.1.2
nats.java
2.18.0
nats.rs
0.33
nats.net.v2
2.0.0
nats.net (v1)
1.1.5
See: https://github.com/nats-io/nats.js/issues/369
See: https://github.com/nats-io/nats.rs/blob/main/async-nats/src/connector.rs
See: https://github.com/nats-io/nats.net.v1/tree/main/src/Samples/TlsVariationsExample
TLS . When not present, default to the system trust store.
TLS handshake in fractional seconds. Default set to 2 seconds.
If true, require and client certificates. To support use by Browser, this option does not apply to monitoring.
If true, require and verify client certificates and certificate values for authentication purposes. Does not apply to monitoring either.
Only settable in a non client context where verify: true is the default (/). The incoming connections certificate's X509v3 Subject Alternative Name DNS entries will be matched against all urls in the configuration context that contains this tls map. If a match is found, the connection is accepted and rejected otherwise. Meaning for gateways we will match all DNS entries in the certificate against all gateway urls. For cluster, we will match against all route urls. As a consequence of this, dynamic cluster growth may require config changes in other clusters where this flag is true. DNS name checking is performed according to . Only the full wildcard * is supported for the left most label. This would be one way to keep cluster growth flexible.
Client connections follow the model where, when a TCP connection is created to the server, the server will immediately send an in clear text. This INFO protocol provides metadata, including whether the server requires a secure connection.