TLS/SSL concepts

TLS/SSL is a set of protocols that rely on a public key infrastructure (PKI) to enable secure communication between a client and a server. For most common cases, each server must have a private key.

Private keys can be generated in multiple ways. The example below illustrates use of the OpenSSL command-line interface to generate a 2048-bit RSA private key:

openssl genrsa -out ryans-key.pem 2048

With TLS/SSL, all servers (and some clients) must have a certificate. Certificates are public keys that correspond to a private key, and that are digitally signed either by a Certificate Authority or by the owner of the private key (such certificates are referred to as "self-signed"). The first step to obtaining a certificate is to create a Certificate Signing Request (CSR) file.

The OpenSSL command-line interface can be used to generate a CSR for a private key:

openssl req -new -sha256 -key ryans-key.pem -out ryans-csr.pem

Once the CSR file is generated, it can either be sent to a Certificate Authority for signing or used to generate a self-signed certificate.

Creating a self-signed certificate using the OpenSSL command-line interface is illustrated in the example below:

openssl x509 -req -in ryans-csr.pem -signkey ryans-key.pem -out ryans-cert.pem

Once the certificate is generated, it can be used to generate a .pfx or .p12 file:

openssl pkcs12 -export -in ryans-cert.pem -inkey ryans-key.pem \
      -certfile ca-cert.pem -out ryans.pfx

Where:

• in: is the signed certificate
• inkey: is the associated private key
• certfile: is a concatenation of all Certificate Authority (CA) certs into a single file, e.g. cat ca1-cert.pem ca2-cert.pem > ca-cert.pem

The term forward secrecy or perfect forward secrecy describes a feature of key-agreement (i.e., key-exchange) methods. That is, the server and client keys are used to negotiate new temporary keys that are used specifically and only for the current communication session. Practically, this means that even if the server's private key is compromised, communication can only be decrypted by eavesdroppers if the attacker manages to obtain the key-pair specifically generated for the session.

Perfect forward secrecy is achieved by randomly generating a key pair for key-agreement on every TLS/SSL handshake (in contrast to using the same key for all sessions). Methods implementing this technique are called "ephemeral".

Currently two methods are commonly used to achieve perfect forward secrecy (note the character "E" appended to the traditional abbreviations):

• ECDHE: An ephemeral version of the Elliptic Curve Diffie-Hellman key-agreement protocol.
• DHE: An ephemeral version of the Diffie-Hellman key-agreement protocol.

Perfect forward secrecy using ECDHE is enabled by default. The ecdhCurve option can be used when creating a TLS server to customize the list of supported ECDH curves for TLSv1.2 and below, and the list of supported TLS groups for TLSv1.3. See tls.createServer() for more info.

DHE is disabled by default but can be enabled alongside ECDHE by setting the dhparam option to 'auto'. Custom DHE parameters are also supported but discouraged in favor of automatically selected, well-known parameters.

Perfect forward secrecy was optional up to TLSv1.2. As of TLSv1.3, (EC)DHE is always used (with the exception of PSK-only connections).

ALPN (Application-Layer Protocol Negotiation Extension) and SNI (Server Name Indication) are TLS handshake extensions:

• ALPN: Allows the use of one TLS server for multiple protocols (HTTP, HTTP/2)
• SNI: Allows the use of one TLS server for multiple hostnames with different certificates.

TLS-PSK support is available as an alternative to normal certificate-based authentication. It uses a pre-shared key instead of certificates to authenticate a TLS connection, providing mutual authentication. TLS-PSK and public key infrastructure are not mutually exclusive. Clients and servers can accommodate both, choosing either of them during the normal cipher negotiation step.

TLS-PSK is only a good choice where means exist to securely share a key with every connecting machine, so it does not replace the public key infrastructure (PKI) for the majority of TLS uses. The TLS-PSK implementation in OpenSSL has seen many security flaws in recent years, mostly because it is used only by a minority of applications. Please consider all alternative solutions before switching to PSK ciphers. Upon generating PSK it is of critical importance to use sufficient entropy as discussed in RFC 4086. Deriving a shared secret from a password or other low-entropy sources is not secure.

PSK ciphers are disabled by default, and using TLS-PSK thus requires explicitly specifying a cipher suite with the ciphers option. The list of available ciphers can be retrieved via openssl ciphers -v 'PSK'. All TLS 1.3 ciphers are eligible for PSK and can be retrieved via openssl ciphers -v -s -tls1_3 -psk. On the client connection, a custom checkServerIdentity should be passed because the default one will fail in the absence of a certificate.

According to the RFC 4279, PSK identities up to 128 bytes in length and PSKs up to 64 bytes in length must be supported. As of OpenSSL 1.1.0 maximum identity size is 128 bytes, and maximum PSK length is 256 bytes.

The current implementation doesn't support asynchronous PSK callbacks due to the limitations of the underlying OpenSSL API.

To use TLS-PSK, client and server must specify the pskCallback option, a function that returns the PSK to use (which must be compatible with the selected cipher's digest).

It will be called first on the client:

Then on the server:

A return value of null stops the negotiation process and sends an unknown_psk_identity alert message to the other party. If the server wishes to hide the fact that the PSK identity was not known, the callback must provide some random data as psk to make the connection fail with decrypt_error before negotiation is finished.

The TLS protocol allows clients to renegotiate certain aspects of the TLS session. Unfortunately, session renegotiation requires a disproportionate amount of server-side resources, making it a potential vector for denial-of-service attacks.

To mitigate the risk, renegotiation is limited to three times every ten minutes. An 'error' event is emitted on the tls.TLSSocket instance when this threshold is exceeded. The limits are configurable:

The default renegotiation limits should not be modified without a full understanding of the implications and risks.

TLSv1.3 does not support renegotiation.

Establishing a TLS session can be relatively slow. The process can be sped up by saving and later reusing the session state. There are several mechanisms to do so, discussed here from oldest to newest (and preferred).

Servers generate a unique ID for new connections and send it to the client. Clients and servers save the session state. When reconnecting, clients send the ID of their saved session state and if the server also has the state for that ID, it can agree to use it. Otherwise, the server will create a new session. See RFC 2246 for more information, page 23 and 30.

Resumption using session identifiers is supported by most web browsers when making HTTPS requests.

For Node.js, clients wait for the 'session' event to get the session data, and provide the data to the session option of a subsequent tls.connect() to reuse the session. Servers must implement handlers for the 'newSession' and 'resumeSession' events to save and restore the session data using the session ID as the lookup key to reuse sessions. To reuse sessions across load balancers or cluster workers, servers must use a shared session cache (such as Redis) in their session handlers.

The servers encrypt the entire session state and send it to the client as a "ticket". When reconnecting, the state is sent to the server in the initial connection. This mechanism avoids the need for a server-side session cache. If the server doesn't use the ticket, for any reason (failure to decrypt it, it's too old, etc.), it will create a new session and send a new ticket. See RFC 5077 for more information.

Resumption using session tickets is becoming commonly supported by many web browsers when making HTTPS requests.

For Node.js, clients use the same APIs for resumption with session identifiers as for resumption with session tickets. For debugging, if tls.TLSSocket.getTLSTicket() returns a value, the session data contains a ticket, otherwise it contains client-side session state.

With TLSv1.3, be aware that multiple tickets may be sent by the server, resulting in multiple 'session' events, see 'session' for more information.

Single process servers need no specific implementation to use session tickets. To use session tickets across server restarts or load balancers, servers must all have the same ticket keys. There are three 16-byte keys internally, but the tls API exposes them as a single 48-byte buffer for convenience.

It's possible to get the ticket keys by calling server.getTicketKeys() on one server instance and then distribute them, but it is more reasonable to securely generate 48 bytes of secure random data and set them with the ticketKeys option of tls.createServer(). The keys should be regularly regenerated and server's keys can be reset with server.setTicketKeys().

Session ticket keys are cryptographic keys, and they must be stored securely. With TLS 1.2 and below, if they are compromised all sessions that used tickets encrypted with them can be decrypted. They should not be stored on disk, and they should be regenerated regularly.

If clients advertise support for tickets, the server will send them. The server can disable tickets by supplying require('node:constants').SSL_OP_NO_TICKET in secureOptions.

Both session identifiers and session tickets timeout, causing the server to create new sessions. The timeout can be configured with the sessionTimeout option of tls.createServer().

For all the mechanisms, when resumption fails, servers will create new sessions. Since failing to resume the session does not cause TLS/HTTPS connection failures, it is easy to not notice unnecessarily poor TLS performance. The OpenSSL CLI can be used to verify that servers are resuming sessions. Use the -reconnect option to openssl s_client, for example:

openssl s_client -connect localhost:443 -reconnect

Read through the debug output. The first connection should say "New", for example:

New, TLSv1.2, Cipher is ECDHE-RSA-AES128-GCM-SHA256

Subsequent connections should say "Reused", for example:

Reused, TLSv1.2, Cipher is ECDHE-RSA-AES128-GCM-SHA256