The Transport Layer Security (TLS) Protocol Version 1.3

1.1. Conventions and Terminology

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶

The following terms are used:¶

client: The endpoint initiating the TLS connection.¶

connection: A transport-layer connection between two endpoints.¶

endpoint: Either the client or server of the connection.¶

handshake: An initial negotiation between client and server that establishes the parameters of their subsequent interactions within TLS.¶

peer: An endpoint. When discussing a particular endpoint, "peer" refers to the endpoint that is not the primary subject of discussion.¶

receiver: An endpoint that is receiving records.¶

sender: An endpoint that is transmitting records.¶

server: The endpoint that did not initiate the TLS connection.¶

1.2. Relationship to RFC 8446

TLS 1.3 was originally specified in [RFC8446]. This document is solely an editorial update. It contains updated text in areas which were found to be unclear as well as other editorial improvements. In addition, it removes the use of the term "master" as applied to secrets in favor of the term "main" or shorter names where no term was neccessary.¶

1.3. Major Differences from TLS 1.2

The following is a list of the major functional differences between TLS 1.2 and TLS 1.3. It is not intended to be exhaustive, and there are many minor differences.¶

• The list of supported symmetric encryption algorithms has been pruned of all algorithms that are considered legacy. Those that remain are all Authenticated Encryption with Associated Data (AEAD) algorithms. The cipher suite concept has been changed to separate the authentication and key exchange mechanisms from the record protection algorithm (including secret key length) and a hash to be used with both the key derivation function and handshake message authentication code (MAC).¶
• A zero round-trip time (0-RTT) mode was added, saving a round trip at connection setup for some application data, at the cost of certain security properties.¶
• Static RSA and Diffie-Hellman cipher suites have been removed; all public-key based key exchange mechanisms now provide forward secrecy.¶
• All handshake messages after the ServerHello are now encrypted. The newly introduced EncryptedExtensions message allows various extensions previously sent in the clear in the ServerHello to also enjoy confidentiality protection.¶
• The key derivation function has been redesigned. The new design allows easier analysis by cryptographers due to their improved key separation properties. The HMAC-based Extract-and-Expand Key Derivation Function (HKDF) is used as an underlying primitive.¶
• The handshake state machine has been significantly restructured to be more consistent and to remove superfluous messages such as ChangeCipherSpec (except when needed for middlebox compatibility).¶
• Elliptic curve algorithms are now in the base spec, and new signature algorithms, such as EdDSA, are included. TLS 1.3 removed point format negotiation in favor of a single point format for each curve.¶
• Other cryptographic improvements were made, including changing the RSA padding to use the RSA Probabilistic Signature Scheme (RSASSA-PSS), and the removal of compression, the Digital Signature Algorithm (DSA), and custom Ephemeral Diffie-Hellman (DHE) groups.¶
• The TLS 1.2 version negotiation mechanism has been deprecated in favor of a version list in an extension. This increases compatibility with existing servers that incorrectly implemented version negotiation.¶
• Session resumption with and without server-side state as well as the PSK-based cipher suites of earlier TLS versions have been replaced by a single new PSK exchange.¶
• References have been updated to point to the updated versions of RFCs, as appropriate (e.g., RFC 5280 rather than RFC 3280).¶

1.4. Updates Affecting TLS 1.2

This document defines several changes that optionally affect implementations of TLS 1.2, including those which do not also support TLS 1.3:¶

• A version downgrade protection mechanism is described in Section 4.1.3.¶
• RSASSA-PSS signature schemes are defined in Section 4.2.3.¶
• The "supported_versions" ClientHello extension can be used to negotiate the version of TLS to use, in preference to the legacy_version field of the ClientHello.¶
• The "signature_algorithms_cert" extension allows a client to indicate which signature algorithms it can validate in X.509 certificates.¶
• The term "master" as applied to secrets has been removed, and the "extended_master_secret" extension [RFC7627] has been renamed to "extended_main_secret".¶

Additionally, this document clarifies some compliance requirements for earlier versions of TLS; see Section 9.3.¶

2.1. Incorrect DHE Share

If the client has not provided a sufficient "key_share" extension (e.g., it includes only DHE or ECDHE groups unacceptable to or unsupported by the server), the server corrects the mismatch with a HelloRetryRequest and the client needs to restart the handshake with an appropriate "key_share" extension, as shown in Figure 2. If no common cryptographic parameters can be negotiated, the server MUST abort the handshake with an appropriate alert.¶

         Client                                               Server

         ClientHello
         + key_share             -------->
                                                   HelloRetryRequest
                                 <--------               + key_share
         ClientHello
         + key_share             -------->
                                                         ServerHello
                                                         + key_share
                                               {EncryptedExtensions}
                                               {CertificateRequest*}
                                                      {Certificate*}
                                                {CertificateVerify*}
                                                          {Finished}
                                 <--------       [Application Data*]
         {Certificate*}
         {CertificateVerify*}
         {Finished}              -------->
         [Application Data]      <------->        [Application Data]

Note: The handshake transcript incorporates the initial ClientHello/HelloRetryRequest exchange; it is not reset with the new ClientHello.¶

TLS also allows several optimized variants of the basic handshake, as described in the following sections.¶

2.2. Resumption and Pre-Shared Key (PSK)

Although TLS PSKs can be established externally, PSKs can also be established in a previous connection and then used to establish a new connection ("session resumption" or "resuming" with a PSK). Once a handshake has completed, the server can send the client a PSK identity that corresponds to a unique key derived from the initial handshake (see Section 4.6.1). The client can then use that PSK identity in future handshakes to negotiate the use of the associated PSK. If the server accepts the PSK, then the security context of the new connection is cryptographically tied to the original connection and the key derived from the initial handshake is used to bootstrap the cryptographic state instead of a full handshake. In TLS 1.2 and below, this functionality was provided by "session IDs" and "session tickets" [RFC5077]. Both mechanisms are obsoleted in TLS 1.3.¶

PSKs can be used with (EC)DHE key exchange in order to provide forward secrecy in combination with shared keys, or can be used alone, at the cost of losing forward secrecy for the application data.¶

Figure 3 shows a pair of handshakes in which the first handshake establishes a PSK and the second handshake uses it:¶

       Client                                               Server

Initial Handshake:
       ClientHello
       + key_share               -------->
                                                       ServerHello
                                                       + key_share
                                             {EncryptedExtensions}
                                             {CertificateRequest*}
                                                    {Certificate*}
                                              {CertificateVerify*}
                                                        {Finished}
                                 <--------     [Application Data*]
       {Certificate*}
       {CertificateVerify*}
       {Finished}                -------->
                                 <--------      [NewSessionTicket]
       [Application Data]        <------->      [Application Data]


Subsequent Handshake:
       ClientHello
       + key_share*
       + pre_shared_key          -------->
                                                       ServerHello
                                                  + pre_shared_key
                                                      + key_share*
                                             {EncryptedExtensions}
                                                        {Finished}
                                 <--------     [Application Data*]
       {Finished}                -------->
       [Application Data]        <------->      [Application Data]

As the server is authenticating via a PSK, it does not send a Certificate or a CertificateVerify message. When a client offers resumption via a PSK, it SHOULD also supply a "key_share" extension to the server to allow the server to decline resumption and fall back to a full handshake, if needed. The server responds with a "pre_shared_key" extension to negotiate the use of PSK key establishment and can (as shown here) respond with a "key_share" extension to do (EC)DHE key establishment, thus providing forward secrecy.¶

When PSKs are provisioned externally, the PSK identity and the KDF hash algorithm to be used with the PSK MUST also be provisioned.¶

Note:

When using an externally provisioned pre-shared secret, a critical consideration is using sufficient entropy during the key generation, as discussed in [RFC4086]. Deriving a shared secret from a password or other low-entropy sources is not secure. A low-entropy secret, or password, is subject to dictionary attacks based on the PSK binder. The specified PSK authentication is not a strong password-based authenticated key exchange even when used with Diffie-Hellman key establishment. Specifically, it does not prevent an attacker that can observe the handshake from performing a brute-force attack on the password/pre-shared key.¶

2.3. 0-RTT Data

When clients and servers share a PSK (either obtained externally or via a previous handshake), TLS 1.3 allows clients to send data on the first flight ("early data"). The client uses the PSK to authenticate the server and to encrypt the early data.¶

As shown in Figure 4, the 0-RTT data is just added to the 1-RTT handshake in the first flight. The rest of the handshake uses the same messages as for a 1-RTT handshake with PSK resumption.¶

         Client                                               Server

         ClientHello
         + early_data
         + key_share*
         + psk_key_exchange_modes
         + pre_shared_key
         (Application Data*)     -------->
                                                         ServerHello
                                                    + pre_shared_key
                                                        + key_share*
                                               {EncryptedExtensions}
                                                       + early_data*
                                                          {Finished}
                                 <--------       [Application Data*]
         (EndOfEarlyData)
         {Finished}              -------->
         [Application Data]      <------->        [Application Data]

               +  Indicates noteworthy extensions sent in the
                  previously noted message.

               *  Indicates optional or situation-dependent
                  messages/extensions that are not always sent.

               () Indicates messages protected using keys
                  derived from a client_early_traffic_secret.

               {} Indicates messages protected using keys
                  derived from a [sender]_handshake_traffic_secret.

               [] Indicates messages protected using keys
                  derived from [sender]_application_traffic_secret_N.

IMPORTANT NOTE: The security properties for 0-RTT data are weaker than those for other kinds of TLS data. Specifically:¶

1. The protocol does not provide any forward secrecy guarantees for this data. The server's behavior determines what forward secrecy guarantees, if any, apply (see Section 8.1). This behavior is not communicated to the client as part of the protocol. Therefore, absent out-of-band knowledge of the server's behavior, the client should assume that this data is not forward secret.¶
2. There are no guarantees of non-replay between connections. Protection against replay for ordinary TLS 1.3 1-RTT data is provided via the server's Random value, but 0-RTT data does not depend on the ServerHello and therefore has weaker guarantees. This is especially relevant if the data is authenticated either with TLS client authentication or inside the application protocol. The same warnings apply to any use of the early_exporter_secret.¶

0-RTT data cannot be duplicated within a connection (i.e., the server will not process the same data twice for the same connection), and an attacker will not be able to make 0-RTT data appear to be 1-RTT data (because it is protected with different keys). Appendix F.5 contains a description of potential attacks, and Section 8 describes mechanisms which the server can use to limit the impact of replay.¶

3.1. Basic Block Size

The representation of all data items is explicitly specified. The basic data block size is one byte (i.e., 8 bits). Multiple-byte data items are concatenations of bytes, from left to right, from top to bottom. From the byte stream, a multi-byte item (a numeric in the following example) is formed (using C notation) by:¶

   value = (byte[0] << 8*(n-1)) | (byte[1] << 8*(n-2)) |
           ... | byte[n-1];

This byte ordering for multi-byte values is the commonplace network byte order or big-endian format.¶

3.2. Miscellaneous

Comments begin with "/*" and end with "*/".¶

Optional components are denoted by enclosing them in "[[ ]]" (double brackets).¶

Single-byte entities containing uninterpreted data are of type opaque.¶

A type alias T' for an existing type T is defined by:¶

   T T';

3.3. Numbers

The basic numeric data type is an unsigned byte (uint8). All larger numeric data types are constructed from a fixed-length series of bytes concatenated as described in Section 3.1 and are also unsigned. The following numeric types are predefined.¶

   uint8 uint16[2];
   uint8 uint24[3];
   uint8 uint32[4];
   uint8 uint64[8];

All values, here and elsewhere in the specification, are transmitted in network byte (big-endian) order; the uint32 represented by the hex bytes 01 02 03 04 is equivalent to the decimal value 16909060.¶

3.4. Vectors

A vector (single-dimensioned array) is a stream of homogeneous data elements. For presentation purposes, this specification refers to vectors as lists. The size of the vector may be specified at documentation time or left unspecified until runtime. In either case, the length declares the number of bytes, not the number of elements, in the vector. The syntax for specifying a new type, T', that is a fixed-length vector of type T is¶

   T T'[n];

Here, T' occupies n bytes in the data stream, where n is a multiple of the size of T. The length of the vector is not included in the encoded stream.¶

In the following example, Datum is defined to be three consecutive bytes that the protocol does not interpret, while Data is three consecutive Datum, consuming a total of nine bytes.¶

   opaque Datum[3];      /* three uninterpreted bytes */
   Datum Data[9];        /* three consecutive 3-byte vectors */

Variable-length vectors are defined by specifying a subrange of legal lengths, inclusively, using the notation <floor..ceiling>. When these are encoded, the actual length precedes the vector's contents in the byte stream. The length will be in the form of a number consuming as many bytes as required to hold the vector's specified maximum (ceiling) length. A variable-length vector with an actual length field of zero is referred to as an empty vector.¶

   T T'<floor..ceiling>;

In the following example, "mandatory" is a vector that must contain between 300 and 400 bytes of type opaque. It can never be empty. The actual length field consumes two bytes, a uint16, which is sufficient to represent the value 400 (see Section 3.3). Similarly, "longer" can represent up to 800 bytes of data, or 400 uint16 elements, and it may be empty. Its encoding will include a two-byte actual length field prepended to the vector. The length of an encoded vector must be an exact multiple of the length of a single element (e.g., a 17-byte vector of uint16 would be illegal).¶

   opaque mandatory<300..400>;
         /* length field is two bytes, cannot be empty */
   uint16 longer<0..800>;
         /* zero to 400 16-bit unsigned integers */

3.5. Enumerateds

An additional sparse data type, called "enum" or "enumerated", is available. Each definition is a different type. Only enumerateds of the same type may be assigned or compared. Every element of an enumerated must be assigned a value, as demonstrated in the following example. Since the elements of the enumerated are not ordered, they can be assigned any unique value, in any order.¶

   enum { e1(v1), e2(v2), ... , en(vn) [[, (n)]] } Te;

Future extensions or additions to the protocol may define new values. Implementations need to be able to parse and ignore unknown values unless the definition of the field states otherwise.¶

An enumerated occupies as much space in the byte stream as would its maximal defined ordinal value. The following definition would cause one byte to be used to carry fields of type Color.¶

   enum { red(3), blue(5), white(7) } Color;

One may optionally specify a value without its associated tag to force the width definition without defining a superfluous element.¶

In the following example, Taste will consume two bytes in the data stream but can only assume the values 1, 2, or 4 in the current version of the protocol.¶

   enum { sweet(1), sour(2), bitter(4), (32000) } Taste;

The names of the elements of an enumeration are scoped within the defined type. In the first example, a fully qualified reference to the second element of the enumeration would be Color.blue. Such qualification is not required if the target of the assignment is well specified.¶

   Color color = Color.blue;     /* overspecified, legal */
   Color color = blue;           /* correct, type implicit */

The names assigned to enumerateds do not need to be unique. The numerical value can describe a range over which the same name applies. The value includes the minimum and maximum inclusive values in that range, separated by two period characters. This is principally useful for reserving regions of the space.¶

   enum { sad(0), meh(1..254), happy(255) } Mood;

3.6. Constructed Types

Structure types may be constructed from primitive types for convenience. Each specification declares a new, unique type. The syntax used for definitions is much like that of C.¶

   struct {
       T1 f1;
       T2 f2;
       ...
       Tn fn;
   } T;

Fixed- and variable-length list (vector) fields are allowed using the standard list syntax. Structures V1 and V2 in the variants example (Section 3.8) demonstrate this.¶

The fields within a structure may be qualified using the type's name, with a syntax much like that available for enumerateds. For example, T.f2 refers to the second field of the previous declaration.¶

3.7. Constants

Fields and variables may be assigned a fixed value using "=", as in:¶

   struct {
       T1 f1 = 8;  /* T.f1 must always be 8 */
       T2 f2;
   } T;

3.8. Variants

Defined structures may have variants based on some knowledge that is available within the environment. The selector must be an enumerated type that defines the possible variants the structure defines. Each arm of the select (below) specifies the type of that variant's field and an optional field label. The mechanism by which the variant is selected at runtime is not prescribed by the presentation language.¶

   struct {
       T1 f1;
       T2 f2;
       ....
       Tn fn;
       select (E) {
           case e1: Te1 [[fe1]];
           case e2: Te2 [[fe2]];
           ....
           case en: Ten [[fen]];
       };
   } Tv;

For example:¶

   enum { apple(0), orange(1) } VariantTag;

   struct {
       uint16 number;
       opaque string<0..10>; /* variable length */
   } V1;

   struct {
       uint32 number;
       opaque string[10];    /* fixed length */
   } V2;

   struct {
       VariantTag type;
       select (VariantRecord.type) {
           case apple:  V1;
           case orange: V2;
       };
   } VariantRecord;

4.1. Key Exchange Messages

The key exchange messages are used to determine the security capabilities of the client and the server and to establish shared secrets, including the traffic keys used to protect the rest of the handshake and the data.¶

   uint16 ProtocolVersion;
   opaque Random[32];

   uint8 CipherSuite[2];    /* Cryptographic suite selector */

   struct {
       ProtocolVersion legacy_version = 0x0303;    /* TLS v1.2 */
       Random random;
       opaque legacy_session_id<0..32>;
       CipherSuite cipher_suites<2..2^16-2>;
       opaque legacy_compression_methods<1..2^8-1>;
       Extension extensions<8..2^16-1>;
   } ClientHello;

   struct {
       ProtocolVersion legacy_version = 0x0303;    /* TLS v1.2 */
       Random random;
       opaque legacy_session_id_echo<0..32>;
       CipherSuite cipher_suite;
       uint8 legacy_compression_method = 0;
       Extension extensions<6..2^16-1>;
   } ServerHello;

  CF 21 AD 74 E5 9A 61 11 BE 1D 8C 02 1E 65 B8 91
  C2 A2 11 16 7A BB 8C 5E 07 9E 09 E2 C8 A8 33 9C

  44 4F 57 4E 47 52 44 01

  44 4F 57 4E 47 52 44 00

4.2. Extensions

A number of TLS messages contain tag-length-value encoded extensions structures.¶

   struct {
       ExtensionType extension_type;
       opaque extension_data<0..2^16-1>;
   } Extension;

   enum {
       server_name(0),                             /* RFC 6066 */
       max_fragment_length(1),                     /* RFC 6066 */
       status_request(5),                          /* RFC 6066 */
       supported_groups(10),                       /* RFC 8422, 7919 */
       signature_algorithms(13),                   /* RFC 8446 */
       use_srtp(14),                               /* RFC 5764 */
       heartbeat(15),                              /* RFC 6520 */
       application_layer_protocol_negotiation(16), /* RFC 7301 */
       signed_certificate_timestamp(18),           /* RFC 6962 */
       client_certificate_type(19),                /* RFC 7250 */
       server_certificate_type(20),                /* RFC 7250 */
       padding(21),                                /* RFC 7685 */
       pre_shared_key(41),                         /* RFC 8446 */
       early_data(42),                             /* RFC 8446 */
       supported_versions(43),                     /* RFC 8446 */
       cookie(44),                                 /* RFC 8446 */
       psk_key_exchange_modes(45),                 /* RFC 8446 */
       certificate_authorities(47),                /* RFC 8446 */
       oid_filters(48),                            /* RFC 8446 */
       post_handshake_auth(49),                    /* RFC 8446 */
       signature_algorithms_cert(50),              /* RFC 8446 */
       key_share(51),                              /* RFC 8446 */
       (65535)
   } ExtensionType;

Here:¶

• "extension_type" identifies the particular extension type.¶
• "extension_data" contains information specific to the particular extension type.¶

The contents of the "extension_data" field are typically defined by an extension-specific structure defined in the TLS presentation language. Unless otherwise specified, trailing data is forbidden. That is, senders MUST NOT include data after the structure in the "extension_data" field. When processing an extension, receivers MUST abort the handshake with a "decode_error" alert if there is data left over after parsing the structure. This does not apply if the receiver does not implement or is configured to ignore an extension.¶

The list of extension types is maintained by IANA as described in Section 11.¶

Extensions are generally structured in a request/response fashion, though some extensions are just requests with no corresponding response (i.e., indications). The client sends its extension requests in the ClientHello message, and the server sends its extension responses in the ServerHello, EncryptedExtensions, HelloRetryRequest, and Certificate messages. The server sends extension requests in the CertificateRequest message which a client MAY respond to with a Certificate message. The server MAY also send unsolicited extensions in the NewSessionTicket, though the client does not respond directly to these.¶

Implementations MUST NOT send extension responses if the remote endpoint did not send the corresponding extension requests, with the exception of the "cookie" extension in the HelloRetryRequest. Upon receiving such an extension, an endpoint MUST abort the handshake with an "unsupported_extension" alert.¶

The table below indicates the messages where a given extension may appear, using the following notation: CH (ClientHello), SH (ServerHello), EE (EncryptedExtensions), CT (Certificate), CR (CertificateRequest), NST (NewSessionTicket), and HRR (HelloRetryRequest). If an implementation receives an extension which it recognizes and which is not specified for the message in which it appears, it MUST abort the handshake with an "illegal_parameter" alert.¶

When multiple extensions of different types are present, the extensions MAY appear in any order, with the exception of "pre_shared_key" (Section 4.2.11) which MUST be the last extension in the ClientHello (but can appear anywhere in the ServerHello extensions block). There MUST NOT be more than one extension of the same type in a given extension block.¶

In TLS 1.3, unlike TLS 1.2, extensions are negotiated for each handshake even when in resumption-PSK mode. However, 0-RTT parameters are those negotiated in the previous handshake; mismatches may require rejecting 0-RTT (see Section 4.2.10).¶

There are subtle (and not so subtle) interactions that may occur in this protocol between new features and existing features which may result in a significant reduction in overall security. The following considerations should be taken into account when designing new extensions:¶

• Some cases where a server does not agree to an extension are error conditions (e.g., the handshake cannot continue), and some are simply refusals to support particular features. In general, error alerts should be used for the former and a field in the server extension response for the latter.¶
• Extensions should, as far as possible, be designed to prevent any attack that forces use (or non-use) of a particular feature by manipulation of handshake messages. This principle should be followed regardless of whether the feature is believed to cause a security problem. Often the fact that the extension fields are included in the inputs to the Finished message hashes will be sufficient, but extreme care is needed when the extension changes the meaning of messages sent in the handshake phase. Designers and implementors should be aware of the fact that until the handshake has been authenticated, active attackers can modify messages and insert, remove, or replace extensions.¶

   struct {
       select (Handshake.msg_type) {
           case client_hello:
                ProtocolVersion versions<2..254>;

           case server_hello: /* and HelloRetryRequest */
                ProtocolVersion selected_version;
       };
   } SupportedVersions;

   struct {
       opaque cookie<1..2^16-1>;
   } Cookie;

   enum {
       /* RSASSA-PKCS1-v1_5 algorithms */
       rsa_pkcs1_sha256(0x0401),
       rsa_pkcs1_sha384(0x0501),
       rsa_pkcs1_sha512(0x0601),

       /* ECDSA algorithms */
       ecdsa_secp256r1_sha256(0x0403),
       ecdsa_secp384r1_sha384(0x0503),
       ecdsa_secp521r1_sha512(0x0603),

       /* RSASSA-PSS algorithms with public key OID rsaEncryption */
       rsa_pss_rsae_sha256(0x0804),
       rsa_pss_rsae_sha384(0x0805),
       rsa_pss_rsae_sha512(0x0806),

       /* EdDSA algorithms */
       ed25519(0x0807),
       ed448(0x0808),

       /* RSASSA-PSS algorithms with public key OID RSASSA-PSS */
       rsa_pss_pss_sha256(0x0809),
       rsa_pss_pss_sha384(0x080a),
       rsa_pss_pss_sha512(0x080b),

       /* Legacy algorithms */
       rsa_pkcs1_sha1(0x0201),
       ecdsa_sha1(0x0203),

       /* Reserved Code Points */
       private_use(0xFE00..0xFFFF),
       (0xFFFF)
   } SignatureScheme;

   struct {
       SignatureScheme supported_signature_algorithms<2..2^16-2>;
   } SignatureSchemeList;

   opaque DistinguishedName<1..2^16-1>;

   struct {
       DistinguishedName authorities<3..2^16-1>;
   } CertificateAuthoritiesExtension;

   struct {
       opaque certificate_extension_oid<1..2^8-1>;
       opaque certificate_extension_values<0..2^16-1>;
   } OIDFilter;

   struct {
       OIDFilter filters<0..2^16-1>;
   } OIDFilterExtension;

   struct {} PostHandshakeAuth;

   enum {

       /* Elliptic Curve Groups (ECDHE) */
       secp256r1(0x0017), secp384r1(0x0018), secp521r1(0x0019),
       x25519(0x001D), x448(0x001E),

       /* Finite Field Groups (DHE) */
       ffdhe2048(0x0100), ffdhe3072(0x0101), ffdhe4096(0x0102),
       ffdhe6144(0x0103), ffdhe8192(0x0104),

       /* Reserved Code Points */
       ffdhe_private_use(0x01FC..0x01FF),
       ecdhe_private_use(0xFE00..0xFEFF),
       (0xFFFF)
   } NamedGroup;

   struct {
       NamedGroup named_group_list<2..2^16-1>;
   } NamedGroupList;

   struct {
       NamedGroup group;
       opaque key_exchange<1..2^16-1>;
   } KeyShareEntry;

   struct {
       KeyShareEntry client_shares<0..2^16-1>;
   } KeyShareClientHello;

   struct {
       NamedGroup selected_group;
   } KeyShareHelloRetryRequest;

   struct {
       KeyShareEntry server_share;
   } KeyShareServerHello;

   struct {
       uint8 legacy_form = 4;
       opaque X[coordinate_length];
       opaque Y[coordinate_length];
   } UncompressedPointRepresentation;

   enum { psk_ke(0), psk_dhe_ke(1), (255) } PskKeyExchangeMode;

   struct {
       PskKeyExchangeMode ke_modes<1..255>;
   } PskKeyExchangeModes;

   struct {} Empty;

   struct {
       select (Handshake.msg_type) {
           case new_session_ticket:   uint32 max_early_data_size;
           case client_hello:         Empty;
           case encrypted_extensions: Empty;
       };
   } EarlyDataIndication;

   struct {
       opaque identity<1..2^16-1>;
       uint32 obfuscated_ticket_age;
   } PskIdentity;

   opaque PskBinderEntry<32..255>;

   struct {
       PskIdentity identities<7..2^16-1>;
       PskBinderEntry binders<33..2^16-1>;
   } OfferedPsks;

   struct {
       select (Handshake.msg_type) {
           case client_hello: OfferedPsks;
           case server_hello: uint16 selected_identity;
       };
   } PreSharedKeyExtension;

   Transcript-Hash(Truncate(ClientHello1))

   Transcript-Hash(ClientHello1,
                   HelloRetryRequest,
                   Truncate(ClientHello2))

4.3. Server Parameters

The next two messages from the server, EncryptedExtensions and CertificateRequest, contain information from the server that determines the rest of the handshake. These messages are encrypted with keys derived from the server_handshake_traffic_secret.¶

   struct {
       Extension extensions<0..2^16-1>;
   } EncryptedExtensions;

   struct {
       opaque certificate_request_context<0..2^8-1>;
       Extension extensions<0..2^16-1>;
   } CertificateRequest;

4.4. Authentication Messages

As discussed in Section 2, TLS generally uses a common set of messages for authentication, key confirmation, and handshake integrity: Certificate, CertificateVerify, and Finished. (The PSK binders also perform key confirmation, in a similar fashion.) These three messages are always sent as the last messages in their handshake flight. The Certificate and CertificateVerify messages are only sent under certain circumstances, as defined below. The Finished message is always sent as part of the Authentication Block. These messages are encrypted under keys derived from the [sender]_handshake_traffic_secret.¶

The computations for the Authentication messages all uniformly take the following inputs:¶

• The certificate and signing key to be used.¶
• A Handshake Context consisting of the list of messages to be included in the transcript hash.¶
• A Base Key to be used to compute a MAC key.¶

Based on these inputs, the messages then contain:¶

Certificate

The certificate to be used for authentication, and any supporting certificates in the chain. Note that certificate-based client authentication is not available in PSK handshake flows (including 0-RTT).¶

CertificateVerify:

A signature over the value Transcript-Hash(Handshake Context, Certificate)¶

Finished:

A MAC over the value Transcript-Hash(Handshake Context, Certificate, CertificateVerify) using a MAC key derived from the Base Key.¶

The following table defines the Handshake Context and MAC Base Key for each scenario:¶

 Transcript-Hash(M1, M2, ... Mn) = Hash(M1 || M2 || ... || Mn)

 Transcript-Hash(ClientHello1, HelloRetryRequest, ... Mn) =
     Hash(message_hash ||        /* Handshake type */
          00 00 Hash.length  ||   /* Handshake message length (bytes) */
          Hash(ClientHello1) ||  /* Hash of ClientHello1 */
          HelloRetryRequest  || ... || Mn)

   enum {
       X509(0),
       RawPublicKey(2),
       (255)
   } CertificateType;

   struct {
       select (certificate_type) {
           case RawPublicKey:
             /* From RFC 7250 ASN.1_subjectPublicKeyInfo */
             opaque ASN1_subjectPublicKeyInfo<1..2^24-1>;

           case X509:
             opaque cert_data<1..2^24-1>;
       };
       Extension extensions<0..2^16-1>;
   } CertificateEntry;

   struct {
       opaque certificate_request_context<0..2^8-1>;
       CertificateEntry certificate_list<0..2^24-1>;
   } Certificate;

   struct {
       SignatureScheme algorithm;
       opaque signature<0..2^16-1>;
   } CertificateVerify;

   Transcript-Hash(Handshake Context, Certificate)

   2020202020202020202020202020202020202020202020202020202020202020
   2020202020202020202020202020202020202020202020202020202020202020
   544c5320312e332c207365727665722043657274696669636174655665726966
   79
   00
   0101010101010101010101010101010101010101010101010101010101010101

finished_key =
    HKDF-Expand-Label(BaseKey, "finished", "", Hash.length)

   struct {
       opaque verify_data[Hash.length];
   } Finished;

   verify_data =
       HMAC(finished_key,
            Transcript-Hash(Handshake Context,
                            Certificate*, CertificateVerify*))

   * Only included if present.

4.5. End of Early Data

   struct {} EndOfEarlyData;

If the server sent an "early_data" extension in EncryptedExtensions, the client MUST send an EndOfEarlyData message after receiving the server Finished. If the server does not send an "early_data" extension in EncryptedExtensions, then the client MUST NOT send an EndOfEarlyData message. This message indicates that all 0-RTT application_data messages, if any, have been transmitted and that the following records are protected under handshake traffic keys. Servers MUST NOT send this message, and clients receiving it MUST terminate the connection with an "unexpected_message" alert. This message is encrypted under keys derived from the client_early_traffic_secret.¶

4.6. Post-Handshake Messages

TLS also allows other messages to be sent after the main handshake. These messages use a handshake content type and are encrypted under the appropriate application traffic key.¶

   struct {
       uint32 ticket_lifetime;
       uint32 ticket_age_add;
       opaque ticket_nonce<0..255>;
       opaque ticket<1..2^16-1>;
       Extension extensions<0..2^16-1>;
   } NewSessionTicket;

    HKDF-Expand-Label(resumption_secret,
                      "resumption", ticket_nonce, Hash.length)

   enum {
       update_not_requested(0), update_requested(1), (255)
   } KeyUpdateRequest;

   struct {
       KeyUpdateRequest request_update;
   } KeyUpdate;

5.1. Record Layer

The record layer fragments information blocks into TLSPlaintext records carrying data in chunks of 2^14 bytes or less. Message boundaries are handled differently depending on the underlying ContentType. Any future content types MUST specify appropriate rules. Note that these rules are stricter than what was enforced in TLS 1.2.¶

Handshake messages MAY be coalesced into a single TLSPlaintext record or fragmented across several records, provided that:¶

• Handshake messages MUST NOT be interleaved with other record types. That is, if a handshake message is split over two or more records, there MUST NOT be any other records between them.¶
• Handshake messages MUST NOT span key changes. Implementations MUST verify that all messages immediately preceding a key change align with a record boundary; if not, then they MUST terminate the connection with an "unexpected_message" alert. Because the ClientHello, EndOfEarlyData, ServerHello, Finished, and KeyUpdate messages can immediately precede a key change, implementations MUST send these messages in alignment with a record boundary.¶

Implementations MUST NOT send zero-length fragments of Handshake types, even if those fragments contain padding.¶

Alert messages (Section 6) MUST NOT be fragmented across records, and multiple alert messages MUST NOT be coalesced into a single TLSPlaintext record. In other words, a record with an Alert type MUST contain exactly one message.¶

Application Data messages contain data that is opaque to TLS. Application Data messages are always protected. Zero-length fragments of Application Data MAY be sent, as they are potentially useful as a traffic analysis countermeasure. Application Data fragments MAY be split across multiple records or coalesced into a single record.¶

   enum {
       invalid(0),
       change_cipher_spec(20),
       alert(21),
       handshake(22),
       application_data(23),
       (255)
   } ContentType;

   struct {
       ContentType type;
       ProtocolVersion legacy_record_version;
       uint16 length;
       opaque fragment[TLSPlaintext.length];
   } TLSPlaintext;

type:

The higher-level protocol used to process the enclosed fragment.¶

legacy_record_version:

MUST be set to 0x0303 for all records generated by a TLS 1.3 implementation other than an initial ClientHello (i.e., one not generated after a HelloRetryRequest), where it MAY also be 0x0301 for compatibility purposes. This field is deprecated and MUST be ignored for all purposes. Previous versions of TLS would use other values in this field under some circumstances.¶

length:

The length (in bytes) of the following TLSPlaintext.fragment. The length MUST NOT exceed 2^14 bytes. An endpoint that receives a record that exceeds this length MUST terminate the connection with a "record_overflow" alert.¶

fragment

The data being transmitted. This value is transparent and is treated as an independent block to be dealt with by the higher-level protocol specified by the type field.¶

This document describes TLS 1.3, which uses the version 0x0304. This version value is historical, deriving from the use of 0x0301 for TLS 1.0 and 0x0300 for SSL 3.0. In order to maximize backward compatibility, a record containing an initial ClientHello SHOULD have version 0x0301 (reflecting TLS 1.0) and a record containing a second ClientHello or a ServerHello MUST have version 0x0303 (reflecting TLS 1.2). When negotiating prior versions of TLS, endpoints follow the procedure and requirements provided in Appendix E.¶

When record protection has not yet been engaged, TLSPlaintext structures are written directly onto the wire. Once record protection has started, TLSPlaintext records are protected and sent as described in the following section. Note that Application Data records MUST NOT be written to the wire unprotected (see Section 2 for details).¶

5.2. Record Payload Protection

The record protection functions translate a TLSPlaintext structure into a TLSCiphertext structure. The deprotection functions reverse the process. In TLS 1.3, as opposed to previous versions of TLS, all ciphers are modeled as "Authenticated Encryption with Associated Data" (AEAD) [RFC5116]. AEAD functions provide a unified encryption and authentication operation which turns plaintext into authenticated ciphertext and back again. Each encrypted record consists of a plaintext header followed by an encrypted body, which itself contains a type and optional padding.¶

   struct {
       opaque content[TLSPlaintext.length];
       ContentType type;
       uint8 zeros[length_of_padding];
   } TLSInnerPlaintext;

   struct {
       ContentType opaque_type = application_data; /* 23 */
       ProtocolVersion legacy_record_version = 0x0303; /* TLS v1.2 */
       uint16 length;
       opaque encrypted_record[TLSCiphertext.length];
   } TLSCiphertext;

content:

The TLSPlaintext.fragment value, containing the byte encoding of a handshake or an alert message, or the raw bytes of the application's data to send.¶

type:

The TLSPlaintext.type value containing the content type of the record.¶

zeros:

An arbitrary-length run of zero-valued bytes may appear in the cleartext after the type field. This provides an opportunity for senders to pad any TLS record by a chosen amount as long as the total stays within record size limits. See Section 5.4 for more details.¶

opaque_type:

The outer opaque_type field of a TLSCiphertext record is always set to the value 23 (application_data) for outward compatibility with middleboxes accustomed to parsing previous versions of TLS. The actual content type of the record is found in TLSInnerPlaintext.type after decryption.¶

legacy_record_version:

The legacy_record_version field is always 0x0303. TLS 1.3 TLSCiphertexts are not generated until after TLS 1.3 has been negotiated, so there are no historical compatibility concerns where other values might be received. Note that the handshake protocol, including the ClientHello and ServerHello messages, authenticates the protocol version, so this value is redundant.¶

length:

The length (in bytes) of the following TLSCiphertext.encrypted_record, which is the sum of the lengths of the content and the padding, plus one for the inner content type, plus any expansion added by the AEAD algorithm. The length MUST NOT exceed 2^14 + 256 bytes. An endpoint that receives a record that exceeds this length MUST terminate the connection with a "record_overflow" alert.¶

encrypted_record:

The AEAD-encrypted form of the serialized TLSInnerPlaintext structure.¶

AEAD algorithms take as input a single key, a nonce, a plaintext, and "additional data" to be included in the authentication check, as described in Section 2.1 of [RFC5116]. The key is either the client_write_key or the server_write_key, the nonce is derived from the sequence number and the client_write_iv or server_write_iv (see Section 5.3), and the additional data input is the record header. I.e.,¶

   additional_data = TLSCiphertext.opaque_type ||
                     TLSCiphertext.legacy_record_version ||
                     TLSCiphertext.length

The plaintext input to the AEAD algorithm is the encoded TLSInnerPlaintext structure. Derivation of traffic keys is defined in Section 7.3.¶

The AEAD output consists of the ciphertext output from the AEAD encryption operation. The length of the plaintext is greater than the corresponding TLSPlaintext.length due to the inclusion of TLSInnerPlaintext.type and any padding supplied by the sender. The length of the AEAD output will generally be larger than the plaintext, but by an amount that varies with the AEAD algorithm. Since the ciphers might incorporate padding, the amount of overhead could vary with different lengths of plaintext. Symbolically,¶

   AEADEncrypted =
       AEAD-Encrypt(write_key, nonce, additional_data, plaintext)

The encrypted_record field of TLSCiphertext is set to AEADEncrypted.¶

In order to decrypt and verify, the cipher takes as input the key, nonce, additional data, and the AEADEncrypted value. The output is either the plaintext or an error indicating that the decryption failed. There is no separate integrity check. Symbolically,¶

   plaintext of encrypted_record =
       AEAD-Decrypt(peer_write_key, nonce, additional_data, AEADEncrypted)

If the decryption fails, the receiver MUST terminate the connection with a "bad_record_mac" alert.¶

An AEAD algorithm used in TLS 1.3 MUST NOT produce an expansion greater than 255 octets. An endpoint that receives a record from its peer with TLSCiphertext.length larger than 2^14 + 256 octets MUST terminate the connection with a "record_overflow" alert. This limit is derived from the maximum TLSInnerPlaintext length of 2^14 octets + 1 octet for ContentType + the maximum AEAD expansion of 255 octets.¶

5.3. Per-Record Nonce

A 64-bit sequence number is maintained separately for reading and writing records. The appropriate sequence number is incremented by one after reading or writing each record. Each sequence number is set to zero at the beginning of a connection and whenever the key is changed; the first record transmitted under a particular traffic key MUST use sequence number 0.¶

Because the size of sequence numbers is 64-bit, they should not wrap. If a TLS implementation would need to wrap a sequence number, it MUST either rekey (Section 4.6.3) or terminate the connection.¶

Each AEAD algorithm will specify a range of possible lengths for the per-record nonce, from N_MIN bytes to N_MAX bytes of input [RFC5116]. The length of the TLS per-record nonce (iv_length) is set to the larger of 8 bytes and N_MIN for the AEAD algorithm (see [RFC5116], Section 4). An AEAD algorithm where N_MAX is less than 8 bytes MUST NOT be used with TLS. The per-record nonce for the AEAD construction is formed as follows:¶

1. The 64-bit record sequence number is encoded in network byte order and padded to the left with zeros to iv_length.¶
2. The padded sequence number is XORed with either the static client_write_iv or server_write_iv (depending on the role).¶

The resulting quantity (of length iv_length) is used as the per-record nonce.¶

Note: This is a different construction from that in TLS 1.2, which specified a partially explicit nonce.¶

5.4. Record Padding

All encrypted TLS records can be padded to inflate the size of the TLSCiphertext. This allows the sender to hide the size of the traffic from an observer.¶

When generating a TLSCiphertext record, implementations MAY choose to pad. An unpadded record is just a record with a padding length of zero. Padding is a string of zero-valued bytes appended to the ContentType field before encryption. Implementations MUST set the padding octets to all zeros before encrypting.¶

Application Data records may contain a zero-length TLSInnerPlaintext.content if the sender desires. This permits generation of plausibly sized cover traffic in contexts where the presence or absence of activity may be sensitive. Implementations MUST NOT send Handshake and Alert records that have a zero-length TLSInnerPlaintext.content; if such a message is received, the receiving implementation MUST terminate the connection with an "unexpected_message" alert.¶

The padding sent is automatically verified by the record protection mechanism; upon successful decryption of a TLSCiphertext.encrypted_record, the receiving implementation scans the field from the end toward the beginning until it finds a non-zero octet. This non-zero octet is the content type of the message. This padding scheme was selected because it allows padding of any encrypted TLS record by an arbitrary size (from zero up to TLS record size limits) without introducing new content types. The design also enforces all-zero padding octets, which allows for quick detection of padding errors.¶

Implementations MUST limit their scanning to the cleartext returned from the AEAD decryption. If a receiving implementation does not find a non-zero octet in the cleartext, it MUST terminate the connection with an "unexpected_message" alert.¶

The presence of padding does not change the overall record size limitations: the full encoded TLSInnerPlaintext MUST NOT exceed 2^14 + 1 octets. If the maximum fragment length is reduced -- as for example by the record_size_limit extension from [RFC8449] -- then the reduced limit applies to the full plaintext, including the content type and padding.¶

Selecting a padding policy that suggests when and how much to pad is a complex topic and is beyond the scope of this specification. If the application-layer protocol on top of TLS has its own padding, it may be preferable to pad Application Data TLS records within the application layer. Padding for encrypted Handshake or Alert records must still be handled at the TLS layer, though. Later documents may define padding selection algorithms or define a padding policy request mechanism through TLS extensions or some other means.¶

5.5. Limits on Key Usage

There are cryptographic limits on the amount of plaintext which can be safely encrypted under a given set of keys. [AEAD-LIMITS] provides an analysis of these limits under the assumption that the underlying primitive (AES or ChaCha20) has no weaknesses. Implementations SHOULD do a key update as described in Section 4.6.3 prior to reaching these limits.¶

For AES-GCM, up to 2^24.5 full-size records (about 24 million) may be encrypted on a given connection while keeping a safety margin of approximately 2^-57 for Authenticated Encryption (AE) security. For ChaCha20/Poly1305, the record sequence number would wrap before the safety limit is reached.¶

6.1. Closure Alerts

The client and the server must share knowledge that the connection is ending in order to avoid a truncation attack.¶

close_notify:

This alert notifies the recipient that the sender will not send any more messages on this connection. Any data received after a closure alert has been received MUST be ignored.¶

user_canceled:

This alert notifies the recipient that the sender is canceling the handshake for some reason unrelated to a protocol failure. If a user cancels an operation after the handshake is complete, just closing the connection by sending a "close_notify" is more appropriate. This alert SHOULD be followed by a "close_notify". This alert generally has AlertLevel=warning.¶

Either party MAY initiate a close of its write side of the connection by sending a "close_notify" alert. Any data received after a closure alert has been received MUST be ignored. If a transport-level close is received prior to a "close_notify", the receiver cannot know that all the data that was sent has been received.¶

Each party MUST send a "close_notify" alert before closing its write side of the connection, unless it has already sent some error alert. This does not have any effect on its read side of the connection. Note that this is a change from versions of TLS prior to TLS 1.3 in which implementations were required to react to a "close_notify" by discarding pending writes and sending an immediate "close_notify" alert of their own. That previous requirement could cause truncation in the read side. Both parties need not wait to receive a "close_notify" alert before closing their read side of the connection, though doing so would introduce the possibility of truncation.¶

If the application protocol using TLS provides that any data may be carried over the underlying transport after the TLS connection is closed, the TLS implementation MUST receive a "close_notify" alert before indicating end-of-data to the application layer. No part of this standard should be taken to dictate the manner in which a usage profile for TLS manages its data transport, including when connections are opened or closed.¶

Note: It is assumed that closing the write side of a connection reliably delivers pending data before destroying the transport.¶

6.2. Error Alerts

Error handling in TLS is very simple. When an error is detected, the detecting party sends a message to its peer. Upon transmission or receipt of a fatal alert message, both parties MUST immediately close the connection.¶

Whenever an implementation encounters a fatal error condition, it SHOULD send an appropriate fatal alert and MUST close the connection without sending or receiving any additional data. Throughout this specification, when the phrases "terminate the connection" and "abort the handshake" are used without a specific alert it means that the implementation SHOULD send the alert indicated by the descriptions below. The phrases "terminate the connection with an X alert" and "abort the handshake with an X alert" mean that the implementation MUST send alert X if it sends any alert. All alerts defined below in this section, as well as all unknown alerts, are universally considered fatal as of TLS 1.3 (see Section 6). The implementation SHOULD provide a way to facilitate logging the sending and receiving of alerts.¶

The following error alerts are defined:¶

unexpected_message:

An inappropriate message (e.g., the wrong handshake message, premature Application Data, etc.) was received. This alert should never be observed in communication between proper implementations.¶

bad_record_mac:

This alert is returned if a record is received which cannot be deprotected. Because AEAD algorithms combine decryption and verification, and also to avoid side-channel attacks, this alert is used for all deprotection failures. This alert should never be observed in communication between proper implementations, except when messages were corrupted in the network.¶

record_overflow:

A TLSCiphertext record was received that had a length more than 2^14 + 256 bytes, or a record decrypted to a TLSPlaintext record with more than 2^14 bytes (or some other negotiated limit). This alert should never be observed in communication between proper implementations, except when messages were corrupted in the network.¶

handshake_failure:

Receipt of a "handshake_failure" alert message indicates that the sender was unable to negotiate an acceptable set of security parameters given the options available.¶

bad_certificate:

A certificate was corrupt, contained signatures that did not verify correctly, etc.¶

unsupported_certificate:

A certificate was of an unsupported type.¶

certificate_revoked:

A certificate was revoked by its signer.¶

certificate_expired:

A certificate has expired or is not currently valid.¶

certificate_unknown:

Some other (unspecified) issue arose in processing the certificate, rendering it unacceptable.¶

illegal_parameter:

A field in the handshake was incorrect or inconsistent with other fields. This alert is used for errors which conform to the formal protocol syntax but are otherwise incorrect.¶

unknown_ca:

A valid certificate chain or partial chain was received, but the certificate was not accepted because the CA certificate could not be located or could not be matched with a known trust anchor.¶

access_denied:

A valid certificate or PSK was received, but when access control was applied, the sender decided not to proceed with negotiation.¶

decode_error:

A message could not be decoded because some field was out of the specified range or the length of the message was incorrect. This alert is used for errors where the message does not conform to the formal protocol syntax. This alert should never be observed in communication between proper implementations, except when messages were corrupted in the network.¶

decrypt_error:

A handshake (not record layer) cryptographic operation failed, including being unable to correctly verify a signature or validate a Finished message or a PSK binder.¶

protocol_version:

The protocol version the peer has attempted to negotiate is recognized but not supported (see Appendix E).¶

insufficient_security:

Returned instead of "handshake_failure" when a negotiation has failed specifically because the server requires parameters more secure than those supported by the client.¶

internal_error:

An internal error unrelated to the peer or the correctness of the protocol (such as a memory allocation failure) makes it impossible to continue.¶

inappropriate_fallback:

Sent by a server in response to an invalid connection retry attempt from a client (see [RFC7507]).¶

missing_extension:

Sent by endpoints that receive a handshake message not containing an extension that is mandatory to send for the offered TLS version or other negotiated parameters.¶

unsupported_extension:

Sent by endpoints receiving any handshake message containing an extension known to be prohibited for inclusion in the given handshake message, or including any extensions in a ServerHello or Certificate not first offered in the corresponding ClientHello or CertificateRequest.¶

unrecognized_name:

Sent by servers when no server exists identified by the name provided by the client via the "server_name" extension (see [RFC6066]).¶

bad_certificate_status_response:

Sent by clients when an invalid or unacceptable OCSP response is provided by the server via the "status_request" extension (see [RFC6066]).¶

unknown_psk_identity:

Sent by servers when PSK key establishment is desired but no acceptable PSK identity is provided by the client. Sending this alert is OPTIONAL; servers MAY instead choose to send a "decrypt_error" alert to merely indicate an invalid PSK identity.¶

certificate_required:

Sent by servers when a client certificate is desired but none was provided by the client.¶

no_application_protocol:

Sent by servers when a client "application_layer_protocol_negotiation" extension advertises only protocols that the server does not support (see [RFC7301]).¶

New Alert values are assigned by IANA as described in Section 11.¶

7.1. Key Schedule

The key derivation process makes use of the HKDF-Extract and HKDF-Expand functions as defined for HKDF [RFC5869], as well as the functions defined below:¶

    HKDF-Expand-Label(Secret, Label, Context, Length) =
         HKDF-Expand(Secret, HkdfLabel, Length)

    Where HkdfLabel is specified as:

    struct {
        uint16 length = Length;
        opaque label<7..255> = "tls13 " + Label;
        opaque context<0..255> = Context;
    } HkdfLabel;

    Derive-Secret(Secret, Label, Messages) =
         HKDF-Expand-Label(Secret, Label,
                           Transcript-Hash(Messages), Hash.length)

The Hash function used by Transcript-Hash and HKDF is the cipher suite hash algorithm. Hash.length is its output length in bytes. Messages is the concatenation of the indicated handshake messages, including the handshake message type and length fields, but not including record layer headers. Note that in some cases a zero-length Context (indicated by "") is passed to HKDF-Expand-Label. The labels specified in this document are all ASCII strings and do not include a trailing NUL byte.¶

Note: With common hash functions, any label longer than 12 characters requires an additional iteration of the hash function to compute. The labels in this specification have all been chosen to fit within this limit.¶

Keys are derived from two input secrets using the HKDF-Extract and Derive-Secret functions. The general pattern for adding a new secret is to use HKDF-Extract with the Salt being the current secret state and the Input Keying Material (IKM) being the new secret to be added. In this version of TLS 1.3, the two input secrets are:¶

• PSK (a pre-shared key established externally or derived from the resumption_secret value from a previous connection)¶
• (EC)DHE shared secret (Section 7.4)¶

This produces a full key derivation schedule shown in the diagram below. In this diagram, the following formatting conventions apply:¶

• HKDF-Extract is drawn as taking the Salt argument from the top and the IKM argument from the left, with its output to the bottom and the name of the output on the right.¶
• Derive-Secret's Secret argument is indicated by the incoming arrow. For instance, the Early Secret is the Secret for generating the client_early_traffic_secret.¶
• "0" indicates a string of Hash.length bytes set to zero.¶

Note: the key derivation labels use the string "master" even though the values are referred to as "main" secrets. This mismatch is a result of renaming the values while retaining compatibility.¶

[[OPEN ISSUE: Replace the strings with hex value?]]¶

                 0
                 |
                 v
   PSK ->  HKDF-Extract = Early Secret
                 |
                 +-----> Derive-Secret(.,
                 |                     "ext binder" |
                 |                     "res binder",
                 |                     "")
                 |                     = binder_key
                 |
                 +-----> Derive-Secret(., "c e traffic",
                 |                     ClientHello)
                 |                     = client_early_traffic_secret
                 |
                 +-----> Derive-Secret(., "e exp master",
                 |                     ClientHello)
                 |                     = early_exporter_secret
                 v
           Derive-Secret(., "derived", "")
                 |
                 v
(EC)DHE -> HKDF-Extract = Handshake Secret
                 |
                 +-----> Derive-Secret(., "c hs traffic",
                 |                     ClientHello...ServerHello)
                 |                     = client_handshake_traffic_secret
                 |
                 +-----> Derive-Secret(., "s hs traffic",
                 |                     ClientHello...ServerHello)
                 |                     = server_handshake_traffic_secret
                 v
           Derive-Secret(., "derived", "")
                 |
                 v
      0 -> HKDF-Extract = Main Secret
                 |
                 +-----> Derive-Secret(., "c ap traffic",
                 |                     ClientHello...server Finished)
                 |                     = client_application_traffic_secret_0
                 |
                 +-----> Derive-Secret(., "s ap traffic",
                 |                     ClientHello...server Finished)
                 |                     = server_application_traffic_secret_0
                 |
                 +-----> Derive-Secret(., "exp master",
                 |                     ClientHello...server Finished)
                 |                     = exporter_secret
                 |
                 +-----> Derive-Secret(., "res master",
                                       ClientHello...client Finished)
                                       = resumption_secret

The general pattern here is that the secrets shown down the left side of the diagram are just raw entropy without context, whereas the secrets down the right side include Handshake Context and therefore can be used to derive working keys without additional context. Note that the different calls to Derive-Secret may take different Messages arguments, even with the same secret. In a 0-RTT exchange, Derive-Secret is called with four distinct transcripts; in a 1-RTT-only exchange, it is called with three distinct transcripts.¶

If a given secret is not available, then the 0-value consisting of a string of Hash.length bytes set to zeros is used. Note that this does not mean skipping rounds, so if PSK is not in use, Early Secret will still be HKDF-Extract(0, 0). For the computation of the binder_key, the label is "ext binder" for external PSKs (those provisioned outside of TLS) and "res binder" for resumption PSKs (those provisioned as the resumption secret of a previous handshake). The different labels prevent the substitution of one type of PSK for the other.¶

There are multiple potential Early Secret values, depending on which PSK the server ultimately selects. The client will need to compute one for each potential PSK; if no PSK is selected, it will then need to compute the Early Secret corresponding to the zero PSK.¶

Once all the values which are to be derived from a given secret have been computed, that secret SHOULD be erased.¶

7.2. Updating Traffic Secrets

Once the handshake is complete, it is possible for either side to update its sending traffic keys using the KeyUpdate handshake message defined in Section 4.6.3. The next generation of traffic keys is computed by generating client_/server_application_traffic_secret_N+1 from client_/server_application_traffic_secret_N as described in this section and then re-deriving the traffic keys as described in Section 7.3.¶

The next-generation application_traffic_secret is computed as:¶

    application_traffic_secret_N+1 =
        HKDF-Expand-Label(application_traffic_secret_N,
                          "traffic upd", "", Hash.length)

Once client_/server_application_traffic_secret_N+1 and its associated traffic keys have been computed, implementations SHOULD delete client_/server_application_traffic_secret_N and its associated traffic keys.¶

7.3. Traffic Key Calculation

The traffic keying material is generated from the following input values:¶

• A secret value¶
• A purpose value indicating the specific value being generated¶
• The length of the key being generated¶

The traffic keying material is generated from an input traffic secret value using:¶

    [sender]_write_key = HKDF-Expand-Label(Secret, "key", "", key_length)
    [sender]_write_iv  = HKDF-Expand-Label(Secret, "iv", "", iv_length)

[sender] denotes the sending side. The value of Secret for each record type is shown in the table below.¶

All the traffic keying material is recomputed whenever the underlying Secret changes (e.g., when changing from the handshake to Application Data keys or upon a key update).¶

7.4. (EC)DHE Shared Secret Calculation

7.5. Exporters

[RFC5705] defines keying material exporters for TLS in terms of the TLS pseudorandom function (PRF). This document replaces the PRF with HKDF, thus requiring a new construction. The exporter interface remains the same.¶

The exporter value is computed as:¶

TLS-Exporter(label, context_value, key_length) =
    HKDF-Expand-Label(Derive-Secret(Secret, label, ""),
                      "exporter", Hash(context_value), key_length)

Where Secret is either the early_exporter_secret or the exporter_secret. Implementations MUST use the exporter_secret unless explicitly specified by the application. The early_exporter_secret is defined for use in settings where an exporter is needed for 0-RTT data. A separate interface for the early exporter is RECOMMENDED; this avoids the exporter user accidentally using an early exporter when a regular one is desired or vice versa.¶

If no context is provided, the context_value is zero length. Consequently, providing no context computes the same value as providing an empty context. This is a change from previous versions of TLS where an empty context produced a different output than an absent context. As of this document's publication, no allocated exporter label is used both with and without a context. Future specifications MUST NOT define a use of exporters that permit both an empty context and no context with the same label. New uses of exporters SHOULD provide a context in all exporter computations, though the value could be empty.¶

Requirements for the format of exporter labels are defined in Section 4 of [RFC5705].¶

8.1. Single-Use Tickets

The simplest form of anti-replay defense is for the server to only allow each session ticket to be used once. For instance, the server can maintain a database of all outstanding valid tickets, deleting each ticket from the database as it is used. If an unknown ticket is provided, the server would then fall back to a full handshake.¶

If the tickets are not self-contained but rather are database keys, and the corresponding PSKs are deleted upon use, then connections established using PSKs enjoy not only anti-replay protection, but also forward secrecy once all copies of the PSK from the database entry have been deleted. This mechanism also improves security for PSK usage when PSK is used without (EC)DHE.¶

Because this mechanism requires sharing the session database between server nodes in environments with multiple distributed servers, it may be hard to achieve high rates of successful PSK 0-RTT connections when compared to self-encrypted tickets. Unlike session databases, session tickets can successfully do PSK-based session establishment even without consistent storage, though when 0-RTT is allowed they still require consistent storage for anti-replay of 0-RTT data, as detailed in the following section.¶

8.2. Client Hello Recording

An alternative form of anti-replay is to record a unique value derived from the ClientHello (generally either the random value or the PSK binder) and reject duplicates. Recording all ClientHellos causes state to grow without bound, but a server can instead record ClientHellos within a given time window and use the "obfuscated_ticket_age" to ensure that tickets aren't reused outside that window.¶

In order to implement this, when a ClientHello is received, the server first verifies the PSK binder as described in Section 4.2.11. It then computes the expected_arrival_time as described in the next section and rejects 0-RTT if it is outside the recording window, falling back to the 1-RTT handshake.¶

If the expected_arrival_time is in the window, then the server checks to see if it has recorded a matching ClientHello. If one is found, it either aborts the handshake with an "illegal_parameter" alert or accepts the PSK but rejects 0-RTT. If no matching ClientHello is found, then it accepts 0-RTT and then stores the ClientHello for as long as the expected_arrival_time is inside the window. Servers MAY also implement data stores with false positives, such as Bloom filters, in which case they MUST respond to apparent replay by rejecting 0-RTT but MUST NOT abort the handshake.¶

The server MUST derive the storage key only from validated sections of the ClientHello. If the ClientHello contains multiple PSK identities, then an attacker can create multiple ClientHellos with different binder values for the less-preferred identity on the assumption that the server will not verify it (as recommended by Section 4.2.11). I.e., if the client sends PSKs A and B but the server prefers A, then the attacker can change the binder for B without affecting the binder for A. If the binder for B is part of the storage key, then this ClientHello will not appear as a duplicate, which will cause the ClientHello to be accepted, and may cause side effects such as replay cache pollution, although any 0-RTT data will not be decryptable because it will use different keys. If the validated binder or the ClientHello.random is used as the storage key, then this attack is not possible.¶

Because this mechanism does not require storing all outstanding tickets, it may be easier to implement in distributed systems with high rates of resumption and 0-RTT, at the cost of potentially weaker anti-replay defense because of the difficulty of reliably storing and retrieving the received ClientHello messages. In many such systems, it is impractical to have globally consistent storage of all the received ClientHellos. In this case, the best anti-replay protection is provided by having a single storage zone be authoritative for a given ticket and refusing 0-RTT for that ticket in any other zone. This approach prevents simple replay by the attacker because only one zone will accept 0-RTT data. A weaker design is to implement separate storage for each zone but allow 0-RTT in any zone. This approach limits the number of replays to once per zone. Application message duplication of course remains possible with either design.¶

When implementations are freshly started, they SHOULD reject 0-RTT as long as any portion of their recording window overlaps the startup time. Otherwise, they run the risk of accepting replays which were originally sent during that period.¶

Note: If the client's clock is running much faster than the server's, then a ClientHello may be received that is outside the window in the future, in which case it might be accepted for 1-RTT, causing a client retry, and then acceptable later for 0-RTT. This is another variant of the second form of attack described in Section 8.¶

8.3. Freshness Checks

Because the ClientHello indicates the time at which the client sent it, it is possible to efficiently determine whether a ClientHello was likely sent reasonably recently and only accept 0-RTT for such a ClientHello, otherwise falling back to a 1-RTT handshake. This is necessary for the ClientHello storage mechanism described in Section 8.2 because otherwise the server needs to store an unlimited number of ClientHellos, and is a useful optimization for self-contained single-use tickets because it allows efficient rejection of ClientHellos which cannot be used for 0-RTT.¶

In order to implement this mechanism, a server needs to store the time that the server generated the session ticket, offset by an estimate of the round-trip time between client and server. I.e.,¶

    adjusted_creation_time = creation_time + estimated_RTT

This value can be encoded in the ticket, thus avoiding the need to keep state for each outstanding ticket. The server can determine the client's view of the age of the ticket by subtracting the ticket's "ticket_age_add" value from the "obfuscated_ticket_age" parameter in the client's "pre_shared_key" extension. The server can determine the expected_arrival_time of the ClientHello as:¶

    expected_arrival_time = adjusted_creation_time + clients_ticket_age

When a new ClientHello is received, the expected_arrival_time is then compared against the current server wall clock time and if they differ by more than a certain amount, 0-RTT is rejected, though the 1-RTT handshake can be allowed to complete.¶

There are several potential sources of error that might cause mismatches between the expected_arrival_time and the measured time. Variations in client and server clock rates are likely to be minimal, though potentially the absolute times may be off by large values. Network propagation delays are the most likely causes of a mismatch in legitimate values for elapsed time. Both the NewSessionTicket and ClientHello messages might be retransmitted and therefore delayed, which might be hidden by TCP. For clients on the Internet, this implies windows on the order of ten seconds to account for errors in clocks and variations in measurements; other deployment scenarios may have different needs. Clock skew distributions are not symmetric, so the optimal tradeoff may involve an asymmetric range of permissible mismatch values.¶

Note that freshness checking alone is not sufficient to prevent replays because it does not detect them during the error window, which -- depending on bandwidth and system capacity -- could include billions of replays in real-world settings. In addition, this freshness checking is only done at the time the ClientHello is received, and not when subsequent early Application Data records are received. After early data is accepted, records may continue to be streamed to the server over a longer time period.¶

9.1. Mandatory-to-Implement Cipher Suites

In the absence of an application profile standard specifying otherwise:¶

A TLS-compliant application MUST implement the TLS_AES_128_GCM_SHA256 [GCM] cipher suite and SHOULD implement the TLS_AES_256_GCM_SHA384 [GCM] and TLS_CHACHA20_POLY1305_SHA256 [RFC8439] cipher suites (see Appendix B.4).¶

A TLS-compliant application MUST support digital signatures with rsa_pkcs1_sha256 (for certificates), rsa_pss_rsae_sha256 (for CertificateVerify and certificates), and ecdsa_secp256r1_sha256. A TLS-compliant application MUST support key exchange with secp256r1 (NIST P-256) and SHOULD support key exchange with X25519 [RFC7748].¶

9.2. Mandatory-to-Implement Extensions

In the absence of an application profile standard specifying otherwise, a TLS-compliant application MUST implement the following TLS extensions:¶

• Supported Versions ("supported_versions"; Section 4.2.1)¶
• Cookie ("cookie"; Section 4.2.2)¶
• Signature Algorithms ("signature_algorithms"; Section 4.2.3)¶
• Signature Algorithms Certificate ("signature_algorithms_cert"; Section 4.2.3)¶
• Negotiated Groups ("supported_groups"; Section 4.2.7)¶
• Key Share ("key_share"; Section 4.2.8)¶
• Server Name Indication ("server_name"; Section 3 of [RFC6066])¶

All implementations MUST send and use these extensions when offering applicable features:¶

• "supported_versions" is REQUIRED for all ClientHello, ServerHello, and HelloRetryRequest messages.¶
• "signature_algorithms" is REQUIRED for certificate authentication.¶
• "supported_groups" is REQUIRED for ClientHello messages using DHE or ECDHE key exchange.¶
• "key_share" is REQUIRED for DHE or ECDHE key exchange.¶
• "pre_shared_key" is REQUIRED for PSK key agreement.¶
• "psk_key_exchange_modes" is REQUIRED for PSK key agreement.¶

A client is considered to be attempting to negotiate using this specification if the ClientHello contains a "supported_versions" extension with 0x0304 contained in its body. Such a ClientHello message MUST meet the following requirements:¶

• If not containing a "pre_shared_key" extension, it MUST contain both a "signature_algorithms" extension and a "supported_groups" extension.¶
• If containing a "supported_groups" extension, it MUST also contain a "key_share" extension, and vice versa. An empty KeyShare.client_shares list is permitted.¶

Servers receiving a ClientHello which does not conform to these requirements MUST abort the handshake with a "missing_extension" alert.¶

Additionally, all implementations MUST support the use of the "server_name" extension with applications capable of using it. Servers MAY require clients to send a valid "server_name" extension. Servers requiring this extension SHOULD respond to a ClientHello lacking a "server_name" extension by terminating the connection with a "missing_extension" alert.¶

9.3. Protocol Invariants

This section describes invariants that TLS endpoints and middleboxes MUST follow. It also applies to earlier versions of TLS.¶

TLS is designed to be securely and compatibly extensible. Newer clients or servers, when communicating with newer peers, should negotiate the most preferred common parameters. The TLS handshake provides downgrade protection: Middleboxes passing traffic between a newer client and newer server without terminating TLS should be unable to influence the handshake (see Appendix F.1). At the same time, deployments update at different rates, so a newer client or server MAY continue to support older parameters, which would allow it to interoperate with older endpoints.¶

For this to work, implementations MUST correctly handle extensible fields:¶

• A client sending a ClientHello MUST support all parameters advertised in it. Otherwise, the server may fail to interoperate by selecting one of those parameters.¶
• A server receiving a ClientHello MUST correctly ignore all unrecognized cipher suites, extensions, and other parameters. Otherwise, it may fail to interoperate with newer clients. In TLS 1.3, a client receiving a CertificateRequest or NewSessionTicket MUST also ignore all unrecognized extensions.¶

• A middlebox which terminates a TLS connection MUST behave as a compliant TLS server (to the original client), including having a certificate which the client is willing to accept, and also as a compliant TLS client (to the original server), including verifying the original server's certificate. In particular, it MUST generate its own ClientHello containing only parameters it understands, and it MUST generate a fresh ServerHello random value, rather than forwarding the endpoint's value.¶

  Note that TLS's protocol requirements and security analysis only apply to the two connections separately. Safely deploying a TLS terminator requires additional security considerations which are beyond the scope of this document.¶

• A middlebox which forwards ClientHello parameters it does not understand MUST NOT process any messages beyond that ClientHello. It MUST forward all subsequent traffic unmodified. Otherwise, it may fail to interoperate with newer clients and servers.¶

  Forwarded ClientHellos may contain advertisements for features not supported by the middlebox, so the response may include future TLS additions the middlebox does not recognize. These additions MAY change any message beyond the ClientHello arbitrarily. In particular, the values sent in the ServerHello might change, the ServerHello format might change, and the TLSCiphertext format might change.¶

The design of TLS 1.3 was constrained by widely deployed non-compliant TLS middleboxes (see Appendix E.4); however, it does not relax the invariants. Those middleboxes continue to be non-compliant.¶

A.1. Client

                           START <----+
            Send ClientHello |        | Recv HelloRetryRequest
       [K_send = early data] |        |
                             v        |
        /                 WAIT_SH ----+
        |                    | Recv ServerHello
        |                    | K_recv = handshake
    Can |                    V
   send |                 WAIT_EE
  early |                    | Recv EncryptedExtensions
   data |           +--------+--------+
        |     Using |                 | Using certificate
        |       PSK |                 v
        |           |            WAIT_CERT_CR
        |           |        Recv |       | Recv CertificateRequest
        |           | Certificate |       v
        |           |             |    WAIT_CERT
        |           |             |       | Recv Certificate
        |           |             v       v
        |           |              WAIT_CV
        |           |                 | Recv CertificateVerify
        |           +> WAIT_FINISHED <+
        |                  | Recv Finished
        \                  | [Send EndOfEarlyData]
                           | K_send = handshake
                           | [Send Certificate [+ CertificateVerify]]
 Can send                  | Send Finished
 app data   -->            | K_send = K_recv = application
 after here                v
                       CONNECTED

Note that with the transitions as shown above, clients may send alerts that derive from post-ServerHello messages in the clear or with the early data keys. If clients need to send such alerts, they SHOULD first rekey to the handshake keys if possible.¶

A.2. Server

                             START <-----+
              Recv ClientHello |         | Send HelloRetryRequest
                               v         |
                            RECVD_CH ----+
                               | Select parameters
                               v
                            NEGOTIATED
                               | Send ServerHello
                               | K_send = handshake
                               | Send EncryptedExtensions
                               | [Send CertificateRequest]
Can send                       | [Send Certificate + CertificateVerify]
app data                       | Send Finished
after   -->                    | K_send = application
here                  +--------+--------+
             No 0-RTT |                 | 0-RTT
                      |                 |
  K_recv = handshake  |                 | K_recv = early data
[Skip decrypt errors] |    +------> WAIT_EOED -+
                      |    |       Recv |      | Recv EndOfEarlyData
                      |    | early data |      | K_recv = handshake
                      |    +------------+      |
                      |                        |
                      +> WAIT_FLIGHT2 <--------+
                               |
                      +--------+--------+
              No auth |                 | Cert-based client auth
                      |                 |
                      |                 v
                      |             WAIT_CERT
                      |        Recv |       | Recv Certificate
                      |       empty |       v
                      | Certificate |    WAIT_CV
                      |             |       | Recv
                      |             v       | CertificateVerify
                      +-> WAIT_FINISHED <---+
                               | Recv Finished
                               | K_recv = application
                               v
                           CONNECTED

B.1. Record Layer

   enum {
       invalid(0),
       change_cipher_spec(20),
       alert(21),
       handshake(22),
       application_data(23),
       (255)
   } ContentType;

   struct {
       ContentType type;
       ProtocolVersion legacy_record_version;
       uint16 length;
       opaque fragment[TLSPlaintext.length];
   } TLSPlaintext;

   struct {
       opaque content[TLSPlaintext.length];
       ContentType type;
       uint8 zeros[length_of_padding];
   } TLSInnerPlaintext;

   struct {
       ContentType opaque_type = application_data; /* 23 */
       ProtocolVersion legacy_record_version = 0x0303; /* TLS v1.2 */
       uint16 length;
       opaque encrypted_record[TLSCiphertext.length];
   } TLSCiphertext;

B.2. Alert Messages

   enum { warning(1), fatal(2), (255) } AlertLevel;

   enum {
       close_notify(0),
       unexpected_message(10),
       bad_record_mac(20),
       decryption_failed_RESERVED(21),
       record_overflow(22),
       decompression_failure_RESERVED(30),
       handshake_failure(40),
       no_certificate_RESERVED(41),
       bad_certificate(42),
       unsupported_certificate(43),
       certificate_revoked(44),
       certificate_expired(45),
       certificate_unknown(46),
       illegal_parameter(47),
       unknown_ca(48),
       access_denied(49),
       decode_error(50),
       decrypt_error(51),
       export_restriction_RESERVED(60),
       protocol_version(70),
       insufficient_security(71),
       internal_error(80),
       inappropriate_fallback(86),
       user_canceled(90),
       no_renegotiation_RESERVED(100),
       missing_extension(109),
       unsupported_extension(110),
       certificate_unobtainable_RESERVED(111),
       unrecognized_name(112),
       bad_certificate_status_response(113),
       bad_certificate_hash_value_RESERVED(114),
       unknown_psk_identity(115),
       certificate_required(116),
       no_application_protocol(120),
       (255)
   } AlertDescription;

   struct {
       AlertLevel level;
       AlertDescription description;
   } Alert;

B.3. Handshake Protocol

   enum {
       hello_request_RESERVED(0),
       client_hello(1),
       server_hello(2),
       hello_verify_request_RESERVED(3),
       new_session_ticket(4),
       end_of_early_data(5),
       hello_retry_request_RESERVED(6),
       encrypted_extensions(8),
       certificate(11),
       server_key_exchange_RESERVED(12),
       certificate_request(13),
       server_hello_done_RESERVED(14),
       certificate_verify(15),
       client_key_exchange_RESERVED(16),
       finished(20),
       certificate_url_RESERVED(21),
       certificate_status_RESERVED(22),
       supplemental_data_RESERVED(23),
       key_update(24),
       message_hash(254),
       (255)
   } HandshakeType;

   struct {
       HandshakeType msg_type;    /* handshake type */
       uint24 length;             /* remaining bytes in message */
       select (Handshake.msg_type) {
           case client_hello:          ClientHello;
           case server_hello:          ServerHello;
           case end_of_early_data:     EndOfEarlyData;
           case encrypted_extensions:  EncryptedExtensions;
           case certificate_request:   CertificateRequest;
           case certificate:           Certificate;
           case certificate_verify:    CertificateVerify;
           case finished:              Finished;
           case new_session_ticket:    NewSessionTicket;
           case key_update:            KeyUpdate;
       };
   } Handshake;

   uint16 ProtocolVersion;
   opaque Random[32];

   uint8 CipherSuite[2];    /* Cryptographic suite selector */

   struct {
       ProtocolVersion legacy_version = 0x0303;    /* TLS v1.2 */
       Random random;
       opaque legacy_session_id<0..32>;
       CipherSuite cipher_suites<2..2^16-2>;
       opaque legacy_compression_methods<1..2^8-1>;
       Extension extensions<8..2^16-1>;
   } ClientHello;

   struct {
       ProtocolVersion legacy_version = 0x0303;    /* TLS v1.2 */
       Random random;
       opaque legacy_session_id_echo<0..32>;
       CipherSuite cipher_suite;
       uint8 legacy_compression_method = 0;
       Extension extensions<6..2^16-1>;
   } ServerHello;

   struct {
       ExtensionType extension_type;
       opaque extension_data<0..2^16-1>;
   } Extension;

   enum {
       server_name(0),                             /* RFC 6066 */
       max_fragment_length(1),                     /* RFC 6066 */
       status_request(5),                          /* RFC 6066 */
       supported_groups(10),                       /* RFC 8422, 7919 */
       signature_algorithms(13),                   /* RFC 8446 */
       use_srtp(14),                               /* RFC 5764 */
       heartbeat(15),                              /* RFC 6520 */
       application_layer_protocol_negotiation(16), /* RFC 7301 */
       signed_certificate_timestamp(18),           /* RFC 6962 */
       client_certificate_type(19),                /* RFC 7250 */
       server_certificate_type(20),                /* RFC 7250 */
       padding(21),                                /* RFC 7685 */
       pre_shared_key(41),                         /* RFC 8446 */
       early_data(42),                             /* RFC 8446 */
       supported_versions(43),                     /* RFC 8446 */
       cookie(44),                                 /* RFC 8446 */
       psk_key_exchange_modes(45),                 /* RFC 8446 */
       certificate_authorities(47),                /* RFC 8446 */
       oid_filters(48),                            /* RFC 8446 */
       post_handshake_auth(49),                    /* RFC 8446 */
       signature_algorithms_cert(50),              /* RFC 8446 */
       key_share(51),                              /* RFC 8446 */
       (65535)
   } ExtensionType;

   struct {
       NamedGroup group;
       opaque key_exchange<1..2^16-1>;
   } KeyShareEntry;

   struct {
       KeyShareEntry client_shares<0..2^16-1>;
   } KeyShareClientHello;

   struct {
       NamedGroup selected_group;
   } KeyShareHelloRetryRequest;

   struct {
       KeyShareEntry server_share;
   } KeyShareServerHello;

   struct {
       uint8 legacy_form = 4;
       opaque X[coordinate_length];
       opaque Y[coordinate_length];
   } UncompressedPointRepresentation;

   enum { psk_ke(0), psk_dhe_ke(1), (255) } PskKeyExchangeMode;

   struct {
       PskKeyExchangeMode ke_modes<1..255>;
   } PskKeyExchangeModes;

   struct {} Empty;

   struct {
       select (Handshake.msg_type) {
           case new_session_ticket:   uint32 max_early_data_size;
           case client_hello:         Empty;
           case encrypted_extensions: Empty;
       };
   } EarlyDataIndication;

   struct {
       opaque identity<1..2^16-1>;
       uint32 obfuscated_ticket_age;
   } PskIdentity;

   opaque PskBinderEntry<32..255>;

   struct {
       PskIdentity identities<7..2^16-1>;
       PskBinderEntry binders<33..2^16-1>;
   } OfferedPsks;

   struct {
       select (Handshake.msg_type) {
           case client_hello: OfferedPsks;
           case server_hello: uint16 selected_identity;
       };
   } PreSharedKeyExtension;

   struct {
       select (Handshake.msg_type) {
           case client_hello:
                ProtocolVersion versions<2..254>;

           case server_hello: /* and HelloRetryRequest */
                ProtocolVersion selected_version;
       };
   } SupportedVersions;

   struct {
       opaque cookie<1..2^16-1>;
   } Cookie;

   enum {
       /* RSASSA-PKCS1-v1_5 algorithms */
       rsa_pkcs1_sha256(0x0401),
       rsa_pkcs1_sha384(0x0501),
       rsa_pkcs1_sha512(0x0601),

       /* ECDSA algorithms */
       ecdsa_secp256r1_sha256(0x0403),
       ecdsa_secp384r1_sha384(0x0503),
       ecdsa_secp521r1_sha512(0x0603),

       /* RSASSA-PSS algorithms with public key OID rsaEncryption */
       rsa_pss_rsae_sha256(0x0804),
       rsa_pss_rsae_sha384(0x0805),
       rsa_pss_rsae_sha512(0x0806),

       /* EdDSA algorithms */
       ed25519(0x0807),
       ed448(0x0808),

       /* RSASSA-PSS algorithms with public key OID RSASSA-PSS */
       rsa_pss_pss_sha256(0x0809),
       rsa_pss_pss_sha384(0x080a),
       rsa_pss_pss_sha512(0x080b),

       /* Legacy algorithms */
       rsa_pkcs1_sha1(0x0201),
       ecdsa_sha1(0x0203),

       /* Reserved Code Points */
       obsolete_RESERVED(0x0000..0x0200),
       dsa_sha1_RESERVED(0x0202),
       obsolete_RESERVED(0x0204..0x0400),
       dsa_sha256_RESERVED(0x0402),
       obsolete_RESERVED(0x0404..0x0500),
       dsa_sha384_RESERVED(0x0502),
       obsolete_RESERVED(0x0504..0x0600),
       dsa_sha512_RESERVED(0x0602),
       obsolete_RESERVED(0x0604..0x06FF),
       private_use(0xFE00..0xFFFF),
       (0xFFFF)
   } SignatureScheme;

   struct {
       SignatureScheme supported_signature_algorithms<2..2^16-2>;
   } SignatureSchemeList;

   enum {
       unallocated_RESERVED(0x0000),

       /* Elliptic Curve Groups (ECDHE) */
       obsolete_RESERVED(0x0001..0x0016),
       secp256r1(0x0017), secp384r1(0x0018), secp521r1(0x0019),
       obsolete_RESERVED(0x001A..0x001C),
       x25519(0x001D), x448(0x001E),

       /* Finite Field Groups (DHE) */
       ffdhe2048(0x0100), ffdhe3072(0x0101), ffdhe4096(0x0102),
       ffdhe6144(0x0103), ffdhe8192(0x0104),

       /* Reserved Code Points */
       ffdhe_private_use(0x01FC..0x01FF),
       ecdhe_private_use(0xFE00..0xFEFF),
       obsolete_RESERVED(0xFF01..0xFF02),
       (0xFFFF)
   } NamedGroup;

   struct {
       NamedGroup named_group_list<2..2^16-1>;
   } NamedGroupList;

   opaque DistinguishedName<1..2^16-1>;

   struct {
       DistinguishedName authorities<3..2^16-1>;
   } CertificateAuthoritiesExtension;

   struct {
       opaque certificate_extension_oid<1..2^8-1>;
       opaque certificate_extension_values<0..2^16-1>;
   } OIDFilter;

   struct {
       OIDFilter filters<0..2^16-1>;
   } OIDFilterExtension;

   struct {} PostHandshakeAuth;

   struct {
       Extension extensions<0..2^16-1>;
   } EncryptedExtensions;

   struct {
       opaque certificate_request_context<0..2^8-1>;
       Extension extensions<0..2^16-1>;
   } CertificateRequest;

   enum {
       X509(0),
       OpenPGP_RESERVED(1),
       RawPublicKey(2),
       (255)
   } CertificateType;

   struct {
       select (certificate_type) {
           case RawPublicKey:
             /* From RFC 7250 ASN.1_subjectPublicKeyInfo */
             opaque ASN1_subjectPublicKeyInfo<1..2^24-1>;

           case X509:
             opaque cert_data<1..2^24-1>;
       };
       Extension extensions<0..2^16-1>;
   } CertificateEntry;

   struct {
       opaque certificate_request_context<0..2^8-1>;
       CertificateEntry certificate_list<0..2^24-1>;
   } Certificate;

   struct {
       SignatureScheme algorithm;
       opaque signature<0..2^16-1>;
   } CertificateVerify;

   struct {
       opaque verify_data[Hash.length];
   } Finished;

   struct {
       uint32 ticket_lifetime;
       uint32 ticket_age_add;
       opaque ticket_nonce<0..255>;
       opaque ticket<1..2^16-1>;
       Extension extensions<0..2^16-1>;
   } NewSessionTicket;

   struct {} EndOfEarlyData;

   enum {
       update_not_requested(0), update_requested(1), (255)
   } KeyUpdateRequest;

   struct {
       KeyUpdateRequest request_update;
   } KeyUpdate;

B.4. Cipher Suites

A cipher suite defines the pair of the AEAD algorithm and hash algorithm to be used with HKDF. Cipher suite names follow the naming convention:¶

   CipherSuite TLS_AEAD_HASH = VALUE;

This specification defines the following cipher suites for use with TLS 1.3.¶

The corresponding AEAD algorithms AEAD_AES_128_GCM, AEAD_AES_256_GCM, and AEAD_AES_128_CCM are defined in [RFC5116]. AEAD_CHACHA20_POLY1305 is defined in [RFC8439]. AEAD_AES_128_CCM_8 is defined in [RFC6655]. The corresponding hash algorithms are defined in [SHS].¶

Although TLS 1.3 uses the same cipher suite space as previous versions of TLS, TLS 1.3 cipher suites are defined differently, only specifying the symmetric ciphers, and cannot be used for TLS 1.2. Similarly, cipher suites for TLS 1.2 and lower cannot be used with TLS 1.3.¶

New cipher suite values are assigned by IANA as described in Section 11.¶

C.1. Random Number Generation and Seeding

TLS requires a cryptographically secure pseudorandom number generator (CSPRNG). In most cases, the operating system provides an appropriate facility such as /dev/urandom, which should be used absent other (e.g., performance) concerns. It is RECOMMENDED to use an existing CSPRNG implementation in preference to crafting a new one. Many adequate cryptographic libraries are already available under favorable license terms. Should those prove unsatisfactory, [RFC4086] provides guidance on the generation of random values.¶

TLS uses random values (1) in public protocol fields such as the public Random values in the ClientHello and ServerHello and (2) to generate keying material. With a properly functioning CSPRNG, this does not present a security problem, as it is not feasible to determine the CSPRNG state from its output. However, with a broken CSPRNG, it may be possible for an attacker to use the public output to determine the CSPRNG internal state and thereby predict the keying material, as documented in [CHECKOWAY] and [DSA-1571-1].¶

Implementations can provide extra security against this form of attack by using separate CSPRNGs to generate public and private values.¶

[RFC8937] describes a way way for security protocol implementations to augment their (pseudo)random number generators using a long-term private key and a deterministic signature function. This improves randomness from broken or otherwise subverted random number generators.¶

C.2. Certificates and Authentication

Implementations are responsible for verifying the integrity of certificates and should generally support certificate revocation messages. Absent a specific indication from an application profile, certificates should always be verified to ensure proper signing by a trusted certificate authority (CA). The selection and addition of trust anchors should be done very carefully. Users should be able to view information about the certificate and trust anchor. Applications SHOULD also enforce minimum and maximum key sizes. For example, certification paths containing keys or signatures weaker than 2048-bit RSA or 224-bit ECDSA are not appropriate for secure applications.¶

C.3. Implementation Pitfalls

Implementation experience has shown that certain parts of earlier TLS specifications are not easy to understand and have been a source of interoperability and security problems. Many of these areas have been clarified in this document but this appendix contains a short list of the most important things that require special attention from implementors.¶

TLS protocol issues:¶

• Do you correctly handle handshake messages that are fragmented to multiple TLS records (see Section 5.1)? Do you correctly handle corner cases like a ClientHello that is split into several small fragments? Do you fragment handshake messages that exceed the maximum fragment size? In particular, the Certificate and CertificateRequest handshake messages can be large enough to require fragmentation. Certificate compression as defined in [RFC8879] can be used to reduce the risk of fragmentation.¶
• Do you ignore the TLS record layer version number in all unencrypted TLS records (see Appendix E)?¶
• Have you ensured that all support for SSL, RC4, EXPORT ciphers, and MD5 (via the "signature_algorithms" extension) is completely removed from all possible configurations that support TLS 1.3 or later, and that attempts to use these obsolete capabilities fail correctly? (see Appendix E)?¶
• Do you handle TLS extensions in ClientHellos correctly, including unknown extensions?¶
• When the server has requested a client certificate but no suitable certificate is available, do you correctly send an empty Certificate message, instead of omitting the whole message (see Section 4.4.2)?¶
• When processing the plaintext fragment produced by AEAD-Decrypt and scanning from the end for the ContentType, do you avoid scanning past the start of the cleartext in the event that the peer has sent a malformed plaintext of all zeros?¶
• Do you properly ignore unrecognized cipher suites (Section 4.1.2), hello extensions (Section 4.2), named groups (Section 4.2.7), key shares (Section 4.2.8), supported versions (Section 4.2.1), and signature algorithms (Section 4.2.3) in the ClientHello?¶
• As a server, do you send a HelloRetryRequest to clients which support a compatible (EC)DHE group but do not predict it in the "key_share" extension? As a client, do you correctly handle a HelloRetryRequest from the server?¶

Cryptographic details:¶

• What countermeasures do you use to prevent timing attacks [TIMING]?¶
• When using Diffie-Hellman key exchange, do you correctly preserve leading zero bytes in the negotiated key (see Section 7.4.1)?¶
• Does your TLS client check that the Diffie-Hellman parameters sent by the server are acceptable (see Section 4.2.8.1)?¶
• Do you use a strong and, most importantly, properly seeded random number generator (see Appendix C.1) when generating Diffie-Hellman private values, the ECDSA "k" parameter, and other security-critical values? It is RECOMMENDED that implementations implement "deterministic ECDSA" as specified in [RFC6979]. Note that purely deterministic ECC signatures such as deterministic ECDSA and EdDSA may be vulnerable to certain side-channel and fault injection attacks in easily accessible IoT devices.¶
• Do you zero-pad Diffie-Hellman public key values and shared secrets to the group size (see Section 4.2.8.1 and Section 7.4.1)?¶
• Do you verify signatures after making them, to protect against RSA-CRT key leaks [FW15]?¶

C.4. Client Tracking Prevention

Clients SHOULD NOT reuse a ticket for multiple connections. Reuse of a ticket allows passive observers to correlate different connections. Servers that issue tickets SHOULD offer at least as many tickets as the number of connections that a client might use; for example, a web browser using HTTP/1.1 [RFC7230] might open six connections to a server. Servers SHOULD issue new tickets with every connection. This ensures that clients are always able to use a new ticket when creating a new connection.¶

Offering a ticket to a server additionally allows the server to correlate different connections. This is possible independent of ticket reuse. Client applications SHOULD NOT offer tickets across connections that are meant to be uncorrelated. For example, [FETCH] defines network partition keys to separate cache lookups in web browsers.¶

C.5. Unauthenticated Operation

Previous versions of TLS offered explicitly unauthenticated cipher suites based on anonymous Diffie-Hellman. These modes have been deprecated in TLS 1.3. However, it is still possible to negotiate parameters that do not provide verifiable server authentication by several methods, including:¶

• Raw public keys [RFC7250].¶
• Using a public key contained in a certificate but without validation of the certificate chain or any of its contents.¶

Either technique used alone is vulnerable to man-in-the-middle attacks and therefore unsafe for general use. However, it is also possible to bind such connections to an external authentication mechanism via out-of-band validation of the server's public key, trust on first use, or a mechanism such as channel bindings (though the channel bindings described in [RFC5929] are not defined for TLS 1.3). If no such mechanism is used, then the connection has no protection against active man-in-the-middle attack; applications MUST NOT use TLS in such a way absent explicit configuration or a specific application profile.¶

E.1. Negotiating with an Older Server

A TLS 1.3 client who wishes to negotiate with servers that do not support TLS 1.3 will send a normal TLS 1.3 ClientHello containing 0x0303 (TLS 1.2) in ClientHello.legacy_version but with the correct version(s) in the "supported_versions" extension. If the server does not support TLS 1.3, it will respond with a ServerHello containing an older version number. If the client agrees to use this version, the negotiation will proceed as appropriate for the negotiated protocol. A client using a ticket for resumption SHOULD initiate the connection using the version that was previously negotiated.¶

Note that 0-RTT data is not compatible with older servers and SHOULD NOT be sent absent knowledge that the server supports TLS 1.3. See Appendix E.3.¶

If the version chosen by the server is not supported by the client (or is not acceptable), the client MUST abort the handshake with a "protocol_version" alert.¶

Some legacy server implementations are known to not implement the TLS specification properly and might abort connections upon encountering TLS extensions or versions which they are not aware of. Interoperability with buggy servers is a complex topic beyond the scope of this document. Multiple connection attempts may be required in order to negotiate a backward-compatible connection; however, this practice is vulnerable to downgrade attacks and is NOT RECOMMENDED.¶

E.2. Negotiating with an Older Client

A TLS server can also receive a ClientHello indicating a version number smaller than its highest supported version. If the "supported_versions" extension is present, the server MUST negotiate using that extension as described in Section 4.2.1. If the "supported_versions" extension is not present, the server MUST negotiate the minimum of ClientHello.legacy_version and TLS 1.2. For example, if the server supports TLS 1.0, 1.1, and 1.2, and legacy_version is TLS 1.0, the server will proceed with a TLS 1.0 ServerHello. If the "supported_versions" extension is absent and the server only supports versions greater than ClientHello.legacy_version, the server MUST abort the handshake with a "protocol_version" alert.¶

Note that earlier versions of TLS did not clearly specify the record layer version number value in all cases (TLSPlaintext.legacy_record_version). Servers will receive various TLS 1.x versions in this field, but its value MUST always be ignored.¶

E.3. 0-RTT Backward Compatibility

0-RTT data is not compatible with older servers. An older server will respond to the ClientHello with an older ServerHello, but it will not correctly skip the 0-RTT data and will fail to complete the handshake. This can cause issues when a client attempts to use 0-RTT, particularly against multi-server deployments. For example, a deployment could deploy TLS 1.3 gradually with some servers implementing TLS 1.3 and some implementing TLS 1.2, or a TLS 1.3 deployment could be downgraded to TLS 1.2.¶

A client that attempts to send 0-RTT data MUST fail a connection if it receives a ServerHello with TLS 1.2 or older. It can then retry the connection with 0-RTT disabled. To avoid a downgrade attack, the client SHOULD NOT disable TLS 1.3, only 0-RTT.¶

To avoid this error condition, multi-server deployments SHOULD ensure a uniform and stable deployment of TLS 1.3 without 0-RTT prior to enabling 0-RTT.¶

E.4. Middlebox Compatibility Mode

Field measurements [Ben17a] [Ben17b] [Res17a] [Res17b] have found that a significant number of middleboxes misbehave when a TLS client/server pair negotiates TLS 1.3. Implementations can increase the chance of making connections through those middleboxes by making the TLS 1.3 handshake look more like a TLS 1.2 handshake:¶

• The client always provides a non-empty session ID in the ClientHello, as described in the legacy_session_id section of Section 4.1.2.¶
• If not offering early data, the client sends a dummy change_cipher_spec record (see the third paragraph of Section 5) immediately before its second flight. This may either be before its second ClientHello or before its encrypted handshake flight. If offering early data, the record is placed immediately after the first ClientHello.¶
• The server sends a dummy change_cipher_spec record immediately after its first handshake message. This may either be after a ServerHello or a HelloRetryRequest.¶

When put together, these changes make the TLS 1.3 handshake resemble TLS 1.2 session resumption, which improves the chance of successfully connecting through middleboxes. This "compatibility mode" is partially negotiated: the client can opt to provide a session ID or not, and the server has to echo it. Either side can send change_cipher_spec at any time during the handshake, as they must be ignored by the peer, but if the client sends a non-empty session ID, the server MUST send the change_cipher_spec as described in this appendix.¶

E.5. Security Restrictions Related to Backward Compatibility

Implementations negotiating the use of older versions of TLS SHOULD prefer forward secret and AEAD cipher suites, when available.¶

The security of RC4 cipher suites is considered insufficient for the reasons cited in [RFC7465]. Implementations MUST NOT offer or negotiate RC4 cipher suites for any version of TLS for any reason.¶

Old versions of TLS permitted the use of very low strength ciphers. Ciphers with a strength less than 112 bits MUST NOT be offered or negotiated for any version of TLS for any reason.¶

The security of SSL 3.0 [RFC6101] is considered insufficient for the reasons enumerated in [RFC7568], and it MUST NOT be negotiated for any reason.¶

The security of SSL 2.0 [SSL2] is considered insufficient for the reasons enumerated in [RFC6176], and it MUST NOT be negotiated for any reason.¶

Implementations MUST NOT send an SSL version 2.0 compatible CLIENT-HELLO. Implementations MUST NOT negotiate TLS 1.3 or later using an SSL version 2.0 compatible CLIENT-HELLO. Implementations are NOT RECOMMENDED to accept an SSL version 2.0 compatible CLIENT-HELLO in order to negotiate older versions of TLS.¶

Implementations MUST NOT send a ClientHello.legacy_version or ServerHello.legacy_version set to 0x0300 or less. Any endpoint receiving a Hello message with ClientHello.legacy_version or ServerHello.legacy_version set to 0x0300 MUST abort the handshake with a "protocol_version" alert.¶

Implementations MUST NOT send any records with a version less than 0x0300. Implementations SHOULD NOT accept any records with a version less than 0x0300 (but may inadvertently do so if the record version number is ignored completely).¶

Implementations MUST NOT use the Truncated HMAC extension, defined in Section 7 of [RFC6066], as it is not applicable to AEAD algorithms and has been shown to be insecure in some scenarios.¶

F.1. Handshake

The TLS handshake is an Authenticated Key Exchange (AKE) protocol which is intended to provide both one-way authenticated (server-only) and mutually authenticated (client and server) functionality. At the completion of the handshake, each side outputs its view of the following values:¶

• A set of "session keys" (the various secrets derived from the main secret) from which can be derived a set of working keys.¶
• A set of cryptographic parameters (algorithms, etc.).¶
• The identities of the communicating parties.¶

We assume the attacker to be an active network attacker, which means it has complete control over the network used to communicate between the parties [RFC3552]. Even under these conditions, the handshake should provide the properties listed below. Note that these properties are not necessarily independent, but reflect the protocol consumers' needs.¶

Establishing the same session keys:

The handshake needs to output the same set of session keys on both sides of the handshake, provided that it completes successfully on each endpoint (see [CK01]; Definition 1, part 1).¶

Secrecy of the session keys:

The shared session keys should be known only to the communicating parties and not to the attacker (see [CK01]; Definition 1, part 2). Note that in a unilaterally authenticated connection, the attacker can establish its own session keys with the server, but those session keys are distinct from those established by the client.¶

Peer Authentication:

The client's view of the peer identity should reflect the server's identity. If the client is authenticated, the server's view of the peer identity should match the client's identity.¶

Uniqueness of the session keys:

Any two distinct handshakes should produce distinct, unrelated session keys. Individual session keys produced by a handshake should also be distinct and independent.¶

Downgrade Protection:

The cryptographic parameters should be the same on both sides and should be the same as if the peers had been communicating in the absence of an attack (see [BBFGKZ16]; Definitions 8 and 9).¶

Forward secret with respect to long-term keys:

If the long-term keying material (in this case the signature keys in certificate-based authentication modes or the external/resumption PSK in PSK with (EC)DHE modes) is compromised after the handshake is complete, this does not compromise the security of the session key (see [DOW92]), as long as the session key itself has been erased. The forward secrecy property is not satisfied when PSK is used in the "psk_ke" PskKeyExchangeMode.¶

Key Compromise Impersonation (KCI) resistance:

In a mutually authenticated connection with certificates, compromising the long-term secret of one actor should not break that actor's authentication of their peer in the given connection (see [HGFS15]). For example, if a client's signature key is compromised, it should not be possible to impersonate arbitrary servers to that client in subsequent handshakes.¶

Protection of endpoint identities:

The server's identity (certificate) should be protected against passive attackers. The client's identity (certificate) should be protected against both passive and active attackers. This property does not hold for cipher suites without confidentiality; while this specification does not define any such cipher suites, other documents may do so.¶

Informally, the signature-based modes of TLS 1.3 provide for the establishment of a unique, secret, shared key established by an (EC)DHE key exchange and authenticated by the server's signature over the handshake transcript, as well as tied to the server's identity by a MAC. If the client is authenticated by a certificate, it also signs over the handshake transcript and provides a MAC tied to both identities. [SIGMA] describes the design and analysis of this type of key exchange protocol. If fresh (EC)DHE keys are used for each connection, then the output keys are forward secret.¶

The external PSK and resumption PSK bootstrap from a long-term shared secret into a unique per-connection set of short-term session keys. This secret may have been established in a previous handshake. If PSK with (EC)DHE key establishment is used, these session keys will also be forward secret. The resumption PSK has been designed so that the resumption secret computed by connection N and needed to form connection N+1 is separate from the traffic keys used by connection N, thus providing forward secrecy between the connections. In addition, if multiple tickets are established on the same connection, they are associated with different keys, so compromise of the PSK associated with one ticket does not lead to the compromise of connections established with PSKs associated with other tickets. This property is most interesting if tickets are stored in a database (and so can be deleted) rather than if they are self-encrypted.¶

The PSK binder value forms a binding between a PSK and the current handshake, as well as between the session where the PSK was established and the current session. This binding transitively includes the original handshake transcript, because that transcript is digested into the values which produce the resumption secret. This requires that both the KDF used to produce the resumption secret and the MAC used to compute the binder be collision resistant. See Appendix F.1.1 for more on this. Note: The binder does not cover the binder values from other PSKs, though they are included in the Finished MAC.¶

Note: TLS does not currently permit the server to send a certificate_request message in non-certificate-based handshakes (e.g., PSK). If this restriction were to be relaxed in future, the client's signature would not cover the server's certificate directly. However, if the PSK was established through a NewSessionTicket, the client's signature would transitively cover the server's certificate through the PSK binder. [PSK-FINISHED] describes a concrete attack on constructions that do not bind to the server's certificate (see also [Kraw16]). It is unsafe to use certificate-based client authentication when the client might potentially share the same PSK/key-id pair with two different endpoints. Implementations MUST NOT combine external PSKs with certificate-based authentication of either the client or server. Future specifications MAY provide an extension to permit this.¶

If an exporter is used, then it produces values which are unique and secret (because they are generated from a unique session key). Exporters computed with different labels and contexts are computationally independent, so it is not feasible to compute one from another or the session secret from the exported value. Note: Exporters can produce arbitrary-length values; if exporters are to be used as channel bindings, the exported value MUST be large enough to provide collision resistance. The exporters provided in TLS 1.3 are derived from the same Handshake Contexts as the early traffic keys and the application traffic keys, respectively, and thus have similar security properties. Note that they do not include the client's certificate; future applications which wish to bind to the client's certificate may need to define a new exporter that includes the full handshake transcript.¶

For all handshake modes, the Finished MAC (and, where present, the signature) prevents downgrade attacks. In addition, the use of certain bytes in the random nonces as described in Section 4.1.3 allows the detection of downgrade to previous TLS versions. See [BBFGKZ16] for more details on TLS 1.3 and downgrade.¶

As soon as the client and the server have exchanged enough information to establish shared keys, the remainder of the handshake is encrypted, thus providing protection against passive attackers, even if the computed shared key is not authenticated. Because the server authenticates before the client, the client can ensure that if it authenticates to the server, it only reveals its identity to an authenticated server. Note that implementations must use the provided record-padding mechanism during the handshake to avoid leaking information about the identities due to length. The client's proposed PSK identities are not encrypted, nor is the one that the server selects.¶

F.2. Record Layer

The record layer depends on the handshake producing strong traffic secrets which can be used to derive bidirectional encryption keys and nonces. Assuming that is true, and the keys are used for no more data than indicated in Section 5.5, then the record layer should provide the following guarantees:¶

Confidentiality:

An attacker should not be able to determine the plaintext contents of a given record.¶

Integrity:

An attacker should not be able to craft a new record which is different from an existing record which will be accepted by the receiver.¶

Order protection/non-replayability:

An attacker should not be able to cause the receiver to accept a record which it has already accepted or cause the receiver to accept record N+1 without having first processed record N.¶

Length concealment:

Given a record with a given external length, the attacker should not be able to determine the amount of the record that is content versus padding.¶

Forward secrecy after key change:

If the traffic key update mechanism described in Section 4.6.3 has been used and the previous generation key is deleted, an attacker who compromises the endpoint should not be able to decrypt traffic encrypted with the old key.¶

Informally, TLS 1.3 provides these properties by AEAD-protecting the plaintext with a strong key. AEAD encryption [RFC5116] provides confidentiality and integrity for the data. Non-replayability is provided by using a separate nonce for each record, with the nonce being derived from the record sequence number (Section 5.3), with the sequence number being maintained independently at both sides; thus records which are delivered out of order result in AEAD deprotection failures. In order to prevent mass cryptanalysis when the same plaintext is repeatedly encrypted by different users under the same key (as is commonly the case for HTTP), the nonce is formed by mixing the sequence number with a secret per-connection initialization vector derived along with the traffic keys. See [BT16] for analysis of this construction.¶

The rekeying technique in TLS 1.3 (see Section 7.2) follows the construction of the serial generator as discussed in [REKEY], which shows that rekeying can allow keys to be used for a larger number of encryptions than without rekeying. This relies on the security of the HKDF-Expand-Label function as a pseudorandom function (PRF). In addition, as long as this function is truly one way, it is not possible to compute traffic keys from prior to a key change (forward secrecy).¶

TLS does not provide security for data which is communicated on a connection after a traffic secret of that connection is compromised. That is, TLS does not provide post-compromise security/future secrecy/backward secrecy with respect to the traffic secret. Indeed, an attacker who learns a traffic secret can compute all future traffic secrets on that connection. Systems which want such guarantees need to do a fresh handshake and establish a new connection with an (EC)DHE exchange.¶

F.3. Traffic Analysis

TLS is susceptible to a variety of traffic analysis attacks based on observing the length and timing of encrypted packets [CLINIC] [HCJC16]. This is particularly easy when there is a small set of possible messages to be distinguished, such as for a video server hosting a fixed corpus of content, but still provides usable information even in more complicated scenarios.¶

TLS does not provide any specific defenses against this form of attack but does include a padding mechanism for use by applications: The plaintext protected by the AEAD function consists of content plus variable-length padding, which allows the application to produce arbitrary-length encrypted records as well as padding-only cover traffic to conceal the difference between periods of transmission and periods of silence. Because the padding is encrypted alongside the actual content, an attacker cannot directly determine the length of the padding, but may be able to measure it indirectly by the use of timing channels exposed during record processing (i.e., seeing how long it takes to process a record or trickling in records to see which ones elicit a response from the server). In general, it is not known how to remove all of these channels because even a constant-time padding removal function will likely feed the content into data-dependent functions. At minimum, a fully constant-time server or client would require close cooperation with the application-layer protocol implementation, including making that higher-level protocol constant time.¶

Note: Robust traffic analysis defenses will likely lead to inferior performance due to delays in transmitting packets and increased traffic volume.¶

F.4. Side Channel Attacks

In general, TLS does not have specific defenses against side-channel attacks (i.e., those which attack the communications via secondary channels such as timing), leaving those to the implementation of the relevant cryptographic primitives. However, certain features of TLS are designed to make it easier to write side-channel resistant code:¶

• Unlike previous versions of TLS which used a composite MAC-then-encrypt structure, TLS 1.3 only uses AEAD algorithms, allowing implementations to use self-contained constant-time implementations of those primitives.¶
• TLS uses a uniform "bad_record_mac" alert for all decryption errors, which is intended to prevent an attacker from gaining piecewise insight into portions of the message. Additional resistance is provided by terminating the connection on such errors; a new connection will have different cryptographic material, preventing attacks against the cryptographic primitives that require multiple trials.¶

Information leakage through side channels can occur at layers above TLS, in application protocols and the applications that use them. Resistance to side-channel attacks depends on applications and application protocols separately ensuring that confidential information is not inadvertently leaked.¶

F.5. Replay Attacks on 0-RTT

Replayable 0-RTT data presents a number of security threats to TLS-using applications, unless those applications are specifically engineered to be safe under replay (minimally, this means idempotent, but in many cases may also require other stronger conditions, such as constant-time response). Potential attacks include:¶

• Duplication of actions which cause side effects (e.g., purchasing an item or transferring money) to be duplicated, thus harming the site or the user.¶
• Attackers can store and replay 0-RTT messages in order to reorder them with respect to other messages (e.g., moving a delete to after a create).¶
• Exploiting cache timing behavior to discover the content of 0-RTT messages by replaying a 0-RTT message to a different cache node and then using a separate connection to measure request latency, to see if the two requests address the same resource.¶

If data can be replayed a large number of times, additional attacks become possible, such as making repeated measurements of the speed of cryptographic operations. In addition, they may be able to overload rate-limiting systems. For a further description of these attacks, see [Mac17].¶

Ultimately, servers have the responsibility to protect themselves against attacks employing 0-RTT data replication. The mechanisms described in Section 8 are intended to prevent replay at the TLS layer but do not provide complete protection against receiving multiple copies of client data. TLS 1.3 falls back to the 1-RTT handshake when the server does not have any information about the client, e.g., because it is in a different cluster which does not share state or because the ticket has been deleted as described in Section 8.1. If the application-layer protocol retransmits data in this setting, then it is possible for an attacker to induce message duplication by sending the ClientHello to both the original cluster (which processes the data immediately) and another cluster which will fall back to 1-RTT and process the data upon application-layer replay. The scale of this attack is limited by the client's willingness to retry transactions and therefore only allows a limited amount of duplication, with each copy appearing as a new connection at the server.¶

If implemented correctly, the mechanisms described in Section 8.1 and Section 8.2 prevent a replayed ClientHello and its associated 0-RTT data from being accepted multiple times by any cluster with consistent state; for servers which limit the use of 0-RTT to one cluster for a single ticket, then a given ClientHello and its associated 0-RTT data will only be accepted once. However, if state is not completely consistent, then an attacker might be able to have multiple copies of the data be accepted during the replication window. Because clients do not know the exact details of server behavior, they MUST NOT send messages in early data which are not safe to have replayed and which they would not be willing to retry across multiple 1-RTT connections.¶

Application protocols MUST NOT use 0-RTT data without a profile that defines its use. That profile needs to identify which messages or interactions are safe to use with 0-RTT and how to handle the situation when the server rejects 0-RTT and falls back to 1-RTT.¶

In addition, to avoid accidental misuse, TLS implementations MUST NOT enable 0-RTT (either sending or accepting) unless specifically requested by the application and MUST NOT automatically resend 0-RTT data if it is rejected by the server unless instructed by the application. Server-side applications may wish to implement special processing for 0-RTT data for some kinds of application traffic (e.g., abort the connection, request that data be resent at the application layer, or delay processing until the handshake completes). In order to allow applications to implement this kind of processing, TLS implementations MUST provide a way for the application to determine if the handshake has completed.¶

F.6. PSK Identity Exposure

Because implementations respond to an invalid PSK binder by aborting the handshake, it may be possible for an attacker to verify whether a given PSK identity is valid. Specifically, if a server accepts both external-PSK and certificate-based handshakes, a valid PSK identity will result in a failed handshake, whereas an invalid identity will just be skipped and result in a successful certificate handshake. Servers which solely support PSK handshakes may be able to resist this form of attack by treating the cases where there is no valid PSK identity and where there is an identity but it has an invalid binder identically.¶

F.7. Sharing PSKs

TLS 1.3 takes a conservative approach to PSKs by binding them to a specific KDF. By contrast, TLS 1.2 allows PSKs to be used with any hash function and the TLS 1.2 PRF. Thus, any PSK which is used with both TLS 1.2 and TLS 1.3 must be used with only one hash in TLS 1.3, which is less than optimal if users want to provision a single PSK. The constructions in TLS 1.2 and TLS 1.3 are different, although they are both based on HMAC. While there is no known way in which the same PSK might produce related output in both versions, only limited analysis has been done. Implementations can ensure safety from cross-protocol related output by not reusing PSKs between TLS 1.3 and TLS 1.2.¶

F.8. Attacks on Static RSA

Although TLS 1.3 does not use RSA key transport and so is not directly susceptible to Bleichenbacher-type attacks [Blei98]if TLS 1.3 servers also support static RSA in the context of previous versions of TLS, then it may be possible to impersonate the server for TLS 1.3 connections [JSS15]. TLS 1.3 implementations can prevent this attack by disabling support for static RSA across all versions of TLS. In principle, implementations might also be able to separate certificates with different keyUsage bits for static RSA decryption and RSA signature, but this technique relies on clients refusing to accept signatures using keys in certificates that do not have the digitalSignature bit set, and many clients do not enforce this restriction.¶