| Network Working Group | T. Dierks |
| Internet-Draft | Independent |
| Obsoletes: 3268, 4346, 4366, 5246 (if | E. Rescorla |
| approved) | RTFM, Inc. |
| Updates: 4492 (if approved) | January 03, 2015 |
| Intended status: Standards Track | |
| Expires: July 7, 2015 |
The Transport Layer Security (TLS) Protocol Version 1.3
draft-ietf-tls-tls13-04
This document specifies Version 1.3 of the Transport Layer Security (TLS) protocol. The TLS protocol provides communications security over the Internet. The protocol allows client/server applications to communicate in a way that is designed to prevent eavesdropping, tampering, or message forgery.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on July 7, 2015.
Copyright (c) 2015 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
This document may contain material from IETF Documents or IETF Contributions published or made publicly available before November 10, 2008. The person(s) controlling the copyright in some of this material may not have granted the IETF Trust the right to allow modifications of such material outside the IETF Standards Process. Without obtaining an adequate license from the person(s) controlling the copyright in such materials, this document may not be modified outside the IETF Standards Process, and derivative works of it may not be created outside the IETF Standards Process, except to format it for publication as an RFC or to translate it into languages other than English.
DISCLAIMER: This is a WIP draft of TLS 1.3 and has not yet seen significant security analysis.
RFC EDITOR: PLEASE REMOVE THE FOLLOWING PARAGRAPH The source for this draft is maintained in GitHub. Suggested changes should be submitted as pull requests at https://github.com/tlswg/tls13-spec. Instructions are on that page as well. Editorial changes can be managed in GitHub, but any substantive change should be discussed on the TLS mailing list.
The primary goal of the TLS protocol is to provide privacy and data integrity between two communicating applications. The protocol is composed of two layers: the TLS Record Protocol and the TLS Handshake Protocol. At the lowest level, layered on top of some reliable transport protocol (e.g., TCP [RFC0793]), is the TLS Record Protocol. The TLS Record Protocol provides connection security that has two basic properties:
The TLS Record Protocol is used for encapsulation of various higher- level protocols. One such encapsulated protocol, the TLS Handshake Protocol, allows the server and client to authenticate each other and to negotiate an encryption algorithm and cryptographic keys before the application protocol transmits or receives its first byte of data. The TLS Handshake Protocol provides connection security that has three basic properties:
One advantage of TLS is that it is application protocol independent. Higher-level protocols can layer on top of the TLS protocol transparently. The TLS standard, however, does not specify how protocols add security with TLS; the decisions on how to initiate TLS handshaking and how to interpret the authentication certificates exchanged are left to the judgment of the designers and implementors of protocols that run on top of TLS.
The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “MAY”, and “OPTIONAL” in this document are to be interpreted as described in RFC 2119 [RFC2119].
draft-04
draft-03
draft-02
The goals of the TLS protocol, in order of priority, are as follows:
This document and the TLS protocol itself are based on the SSL 3.0 Protocol Specification as published by Netscape. The differences between this protocol and SSL 3.0 are not dramatic, but they are significant enough that the various versions of TLS and SSL 3.0 do not interoperate (although each protocol incorporates a mechanism by which an implementation can back down to prior versions). This document is intended primarily for readers who will be implementing the protocol and for those doing cryptographic analysis of it. The specification has been written with this in mind, and it is intended to reflect the needs of those two groups. For that reason, many of the algorithm-dependent data structures and rules are included in the body of the text (as opposed to in an appendix), providing easier access to them.
This document is not intended to supply any details of service definition or of interface definition, although it does cover select areas of policy as they are required for the maintenance of solid security.
This document deals with the formatting of data in an external representation. The following very basic and somewhat casually defined presentation syntax will be used. The syntax draws from several sources in its structure. Although it resembles the programming language “C” in its syntax and XDR [RFC4506] in both its syntax and intent, it would be risky to draw too many parallels. The purpose of this presentation language is to document TLS only; it has no general application beyond that particular goal.
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 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.
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 vector (single-dimensioned array) is a stream of homogeneous data elements. 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]; /* 3 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 4.4). On the other hand, 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 even multiple of the length of a single element (for example, a 17-byte vector of uint16 would be illegal).
opaque mandatory<300..400>;
/* length field is 2 bytes, cannot be empty */
uint16 longer<0..800>;
/* zero to 400 16-bit unsigned integers */
The basic numeric data type is an unsigned byte (uint8). All larger numeric data types are formed from fixed-length series of bytes concatenated as described in Section 4.1 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 stored in network byte (big-endian) order; the uint32 represented by the hex bytes 01 02 03 04 is equivalent to the decimal value 16909060.
Note that in some cases (e.g., DH parameters) it is necessary to represent integers as opaque vectors. In such cases, they are represented as unsigned integers (i.e., leading zero octets are not required even if the most significant bit is set).
An additional sparse data type is available called enum. A field of type enum can only assume the values declared in the definition. 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;
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.
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 */
For enumerateds that are never converted to external representation, the numerical information may be omitted.
enum { low, medium, high } Amount;
Structure types may be constructed from primitive types for convenience. Each specification declares a new, unique type. The syntax for definition is much like that of C.
struct {
T1 f1;
T2 f2;
...
Tn fn;
} [[T]];
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. Structure definitions may be embedded.
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. There must be a case arm for every element of the enumeration declared in the select. Case arms have limited fall-through: if two case arms follow in immediate succession with no fields in between, then they both contain the same fields. Thus, in the example below, “orange” and “banana” both contain V2. Note that this is a new piece of syntax in TLS 1.2.
The body of the variant structure may be given a label for reference. 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;
case e2: Te2;
case e3: case e4: Te3;
....
case en: Ten;
} [[fv]];
} [[Tv]];
For example:
enum { apple, orange, banana } VariantTag;
struct {
uint16 number;
opaque string<0..10>; /* variable length */
} V1;
struct {
uint32 number;
opaque string[10]; /* fixed length */
} V2;
struct {
select (VariantTag) { /* value of selector is implicit */
case apple:
V1; /* VariantBody, tag = apple */
case orange:
case banana:
V2; /* VariantBody, tag = orange or banana */
} variant_body; /* optional label on variant */
} VariantRecord;
The two cryptographic operations — digital signing, and authenticated encryption with additional data (AEAD) — are designated digitally-signed, and aead-ciphered, respectively. A field’s cryptographic processing is specified by prepending an appropriate key word designation before the field’s type specification. Cryptographic keys are implied by the current session state (see Section 6.1).
A digitally-signed element is encoded as a struct DigitallySigned:
struct {
SignatureAndHashAlgorithm algorithm;
opaque signature<0..2^16-1>;
} DigitallySigned;
The algorithm field specifies the algorithm used (see Section 7.3.2.5.1 for the definition of this field). Note that the algorithm field was introduced in TLS 1.2, and is not in earlier versions. The signature is a digital signature using those algorithms over the contents of the element. The contents themselves do not appear on the wire but are simply calculated. The length of the signature is specified by the signing algorithm and key.
In previous versions of TLS, the ServerKeyExchange format meant that attackers can obtain a signature of a message with a chosen, 32-byte prefix. Because TLS 1.3 servers are likely to also implement prior versions, the contents of the element always start with 64 bytes of octet 32 in order to clear that chosen-prefix.
Following that padding is a NUL-terminated context string in order to disambiguate signatures for different purposes. The context string will be specified whenever a digitally-signed element is used.
Finally, the specified contents of the digitally-signed structure follow the NUL at the end of the context string. (See the example at the end of this section.)
In RSA signing, the opaque vector contains the signature generated using the RSASSA-PKCS1-v1_5 signature scheme defined in [RFC3447]. As discussed in [RFC3447], the DigestInfo MUST be DER-encoded [X680] [X690]. For hash algorithms without parameters (which includes SHA-1), the DigestInfo.AlgorithmIdentifier.parameters field MUST be NULL, but implementations MUST accept both without parameters and with NULL parameters. Note that earlier versions of TLS used a different RSA signature scheme that did not include a DigestInfo encoding.
In DSA, the 20 bytes of the SHA-1 hash are run directly through the Digital Signing Algorithm with no additional hashing. This produces two values, r and s. The DSA signature is an opaque vector, as above, the contents of which are the DER encoding of:
Dss-Sig-Value ::= SEQUENCE {
r INTEGER,
s INTEGER
}
Note: In current terminology, DSA refers to the Digital Signature Algorithm and DSS refers to the NIST standard. In the original SSL and TLS specs, “DSS” was used universally. This document uses “DSA” to refer to the algorithm, “DSS” to refer to the standard, and it uses “DSS” in the code point definitions for historical continuity.
All ECDSA computations MUST be performed according to ANSI X9.62 [X962] or its successors. Data to be signed/verified is hashed, and the result run directly through the ECDSA algorithm with no additional hashing. The default hash function is SHA-1 [SHS]. However, an alternative hash function, such as one of the new SHA hash functions specified in FIPS 180-2 may be used instead if the certificate containing the EC public key explicitly requires use of another hash function. (The mechanism for specifying the required hash function has not been standardized, but this provision anticipates such standardization and obviates the need to update this document in response. Future PKIX RFCs may choose, for example, to specify the hash function to be used with a public key in the parameters field of subjectPublicKeyInfo.) [[OPEN ISSUE: This needs updating per 4492-bis https://github.com/tlswg/tls13-spec/issues/59]]
In AEAD encryption, the plaintext is simultaneously encrypted and integrity protected. The input may be of any length, and aead-ciphered output is generally larger than the input in order to accommodate the integrity check value.
In the following example
struct {
uint8 field1;
uint8 field2;
digitally-signed opaque {
uint8 field3<0..255>;
uint8 field4;
};
} UserType;
Assume that the context string for the signature was specified as “Example”. The input for the signature/hash algorithm would be:
2020202020202020202020202020202020202020202020202020202020202020 2020202020202020202020202020202020202020202020202020202020202020 4578616d706c6500
followed by the encoding of the inner struct (field3 and field4).
The length of the structure, in bytes, would be equal to two bytes for field1 and field2, plus two bytes for the signature and hash algorithm, plus two bytes for the length of the signature, plus the length of the output of the signing algorithm. The length of the signature is known because the algorithm and key used for the signing are known prior to encoding or decoding this structure.
Typed constants can be defined for purposes of specification by declaring a symbol of the desired type and assigning values to it.
Under-specified types (opaque, variable-length vectors, and structures that contain opaque) cannot be assigned values. No fields of a multi-element structure or vector may be elided.
For example:
struct {
uint8 f1;
uint8 f2;
} Example1;
Example1 ex1 = {1, 4}; /* assigns f1 = 1, f2 = 4 */
A construction is required to do expansion of secrets into blocks of data for the purposes of key generation or validation. This pseudorandom function (PRF) takes as input a secret, a seed, and an identifying label and produces an output of arbitrary length.
In this section, we define one PRF, based on HMAC [RFC2104]. This PRF with the SHA-256 hash function is used for all cipher suites defined in this document and in TLS documents published prior to this document when TLS 1.2 is negotiated. New cipher suites MUST explicitly specify a PRF and, in general, SHOULD use the TLS PRF with SHA-256 or a stronger standard hash function.
First, we define a data expansion function, P_hash(secret, data), that uses a single hash function to expand a secret and seed into an arbitrary quantity of output:
P_hash(secret, seed) = HMAC_hash(secret, A(1) + seed) +
HMAC_hash(secret, A(2) + seed) +
HMAC_hash(secret, A(3) + seed) + ...
where + indicates concatenation.
A() is defined as:
A(0) = seed A(i) = HMAC_hash(secret, A(i-1))
P_hash can be iterated as many times as necessary to produce the required quantity of data. For example, if P_SHA256 is being used to create 80 bytes of data, it will have to be iterated three times (through A(3)), creating 96 bytes of output data; the last 16 bytes of the final iteration will then be discarded, leaving 80 bytes of output data.
TLS’s PRF is created by applying P_hash to the secret as:
PRF(secret, label, seed) = P_<hash>(secret, label + seed)
The label is an ASCII string. It should be included in the exact form it is given without a length byte or trailing null character. For example, the label “slithy toves” would be processed by hashing the following bytes:
73 6C 69 74 68 79 20 74 6F 76 65 73
The TLS Record Protocol is a layered protocol. At each layer, messages may include fields for length, description, and content. The Record Protocol takes messages to be transmitted, fragments the data into manageable blocks, protects the records, and transmits the result. Received data is decrypted and verified, reassembled, and then delivered to higher-level clients.
Four protocols that use the record protocol are described in this document: the handshake protocol, the alert protocol, the change cipher spec protocol, and the application data protocol. In order to allow extension of the TLS protocol, additional record content types can be supported by the record protocol. New record content type values are assigned by IANA in the TLS Content Type Registry as described in Section 12.
Implementations MUST NOT send record types not defined in this document unless negotiated by some extension. If a TLS implementation receives an unexpected record type, it MUST send an unexpected_message alert.
Any protocol designed for use over TLS must be carefully designed to deal with all possible attacks against it. As a practical matter, this means that the protocol designer must be aware of what security properties TLS does and does not provide and cannot safely rely on the latter.
Note in particular that type and length of a record are not protected by encryption. If this information is itself sensitive, application designers may wish to take steps (padding, cover traffic) to minimize information leakage.
A TLS connection state is the operating environment of the TLS Record Protocol. It specifies a record protection algorithm and its parameters as well as the record protection keys and IVs for the connection in both the read and the write directions. The security parameters are be set by the TLS Handshake Protocol, which also determines when new cryptographic keys are installed and used for record protection. The initial current state always specifies that records are not protected.
The security parameters for a TLS Connection read and write state are set by providing the following values:
These parameters are defined in the presentation language as:
enum { server, client } ConnectionEnd;
enum { tls_prf_sha256 } PRFAlgorithm;
enum { aes_gcm } RecordProtAlgorithm;
/* The algorithms specified in PRFAlgorithm and
RecordProtAlgorithm may be added to. */
struct {
ConnectionEnd entity;
PRFAlgorithm prf_algorithm;
RecordProtAlgorithm record_prot_algorithm;
uint8 enc_key_length;
uint8 block_length;
uint8 fixed_iv_length;
uint8 record_iv_length;
opaque hs_master_secret[48];
opaque master_secret[48];
opaque client_random[32];
opaque server_random[32];
} SecurityParameters;
The record layer will use the security parameters to generate the following four items (some of which are not required by all ciphers, and are thus empty):
client write key server write key client write IV server write IV
The client write parameters are used by the server when receiving and processing records and vice versa. The algorithm used for generating these items from the security parameters is described in Section 6.3
Once the security parameters have been set and the keys have been generated, the connection states can be instantiated by making them the current states. These current states MUST be updated for each record processed. Each connection state includes the following elements:
The TLS record layer receives uninterpreted data from higher layers in non-empty blocks of arbitrary size.
The record layer fragments information blocks into TLSPlaintext records carrying data in chunks of 2^14 bytes or less. Client message boundaries are not preserved in the record layer (i.e., multiple client messages of the same ContentType MAY be coalesced into a single TLSPlaintext record, or a single message MAY be fragmented across several records).
struct {
uint8 major;
uint8 minor;
} ProtocolVersion;
enum {
reserved(20), alert(21), handshake(22),
application_data(23), (255)
} ContentType;
struct {
ContentType type;
ProtocolVersion version;
uint16 length;
opaque fragment[TLSPlaintext.length];
} TLSPlaintext;
Implementations MUST NOT send zero-length fragments of Handshake or Alert types. Zero-length fragments of Application data MAY be sent as they are potentially useful as a traffic analysis countermeasure.
The record protection functions translate a TLSPlaintext structure into a TLSCiphertext. The deprotection functions reverse the process. In TLS 1.3 as opposed to previous versions of TLS, all ciphers are modelled as “Authenticated Encryption with Additional Data” (AEAD) [RFC5116]. AEAD functions provide a unified encryption and authentication operation which turns plaintext into authenticated ciphertext and back again.
AEAD ciphers 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.
struct {
ContentType type;
ProtocolVersion version;
uint16 length;
opaque nonce_explicit[SecurityParameters.record_iv_length];
aead-ciphered struct {
opaque content[TLSPlaintext.length];
} fragment;
} TLSCiphertext;
Each AEAD cipher suite MUST specify how the nonce supplied to the AEAD operation is constructed, and what is the length of the TLSCiphertext.nonce_explicit part. In many cases, it is appropriate to use the partially implicit nonce technique described in Section 3.2.1 of [RFC5116]; with record_iv_length being the length of the explicit part. In this case, the implicit part SHOULD be derived from key_block as client_write_iv and server_write_iv (as described in Section 6.3), and the explicit part is included in GenericAEAEDCipher.nonce_explicit.
The plaintext is the TLSPlaintext.fragment.
The additional authenticated data, which we denote as additional_data, is defined as follows:
additional_data = seq_num + TLSPlaintext.type +
TLSPlaintext.version
where “+” denotes concatenation.
Note: In versions of TLS prior to 1.3, the additional_data included a length field. This presents a problem for cipher constructions with data-dependent padding (such as CBC). TLS 1.3 removes the length field and relies on the AEAD cipher to provide integrity for the length of the data.
The AEAD output consists of the ciphertext output by the AEAD encryption operation. The length will generally be larger than TLSPlaintext.length, but by an amount that varies with the AEAD cipher. Since the ciphers might incorporate padding, the amount of overhead could vary with different TLSPlaintext.length values. Each AEAD cipher MUST NOT produce an expansion of greater than 1024 bytes. Symbolically,
AEADEncrypted = AEAD-Encrypt(write_key, nonce, plaintext,
additional_data)
[[OPEN ISSUE: Reduce these values? https://github.com/tlswg/tls13-spec/issues/55]]
In order to decrypt and verify, the cipher takes as input the key, nonce, the “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. That is:
TLSPlaintext.fragment = AEAD-Decrypt(write_key, nonce,
AEADEncrypted,
additional_data)
If the decryption fails, a fatal bad_record_mac alert MUST be generated.
As a special case, we define the NULL_NULL AEAD cipher which is simply the identity operation and thus provides no security. This cipher MUST ONLY be used with the initial TLS_NULL_WITH_NULL_NULL cipher suite.
[[OPEN ISSUE: This needs to be revised. See https://github.com/tlswg/tls13-spec/issues/5]] The Record Protocol requires an algorithm to generate keys required by the current connection state (see Appendix A.6) from the security parameters provided by the handshake protocol.
The master secret is expanded into a sequence of secure bytes, which is then split to a client write encryption key and a server write encryption key. Each of these is generated from the byte sequence in that order. Unused values are empty. Some ciphers may additionally require a client write IV and a server write IV.
When keys are generated, the then current master secret (MS) is used as an entropy source. For handshake records, this means the hs_master_secret. For application data records, this means the regular master_secret.
To generate the key material, compute
key_block = PRF(MS,
"key expansion",
SecurityParameters.server_random +
SecurityParameters.client_random);
where MS is the relevant master secret. The PRF is computed enough times to generate the necessary amount of data for the key_block, which is then partitioned as follows:
client_write_key[SecurityParameters.enc_key_length] server_write_key[SecurityParameters.enc_key_length] client_write_IV[SecurityParameters.fixed_iv_length] server_write_IV[SecurityParameters.fixed_iv_length]
Currently, the client_write_IV and server_write_IV are only generated for implicit nonce techniques as described in Section 3.2.1 of [RFC5116].
TLS has three subprotocols that are used to allow peers to agree upon security parameters for the record layer, to authenticate themselves, to instantiate negotiated security parameters, and to report error conditions to each other.
The Handshake Protocol is responsible for negotiating a session, which consists of the following items:
These items are then used to create security parameters for use by the record layer when protecting application data. Many connections can be instantiated using the same session through the resumption feature of the TLS Handshake Protocol.
One of the content types supported by the TLS record layer is the alert type. Alert messages convey the severity of the message (warning or fatal) and a description of the alert. Alert messages with a level of fatal result in the immediate termination of the connection. In this case, other connections corresponding to the session may continue, but the session identifier MUST be invalidated, preventing the failed session from being used to establish new connections. Like other messages, alert messages are encrypted as specified by the current connection state.
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),
user_canceled(90),
no_renegotiation(100),
unsupported_extension(110),
(255)
} AlertDescription;
struct {
AlertLevel level;
AlertDescription description;
} Alert;
The client and the server must share knowledge that the connection is ending in order to avoid a truncation attack. Either party may initiate the exchange of closing messages.
Either party may initiate a close by sending a close_notify alert. Any data received after a closure alert is ignored.
Unless some other fatal alert has been transmitted, each party is required to send a close_notify alert before closing the write side of the connection. The other party MUST respond with a close_notify alert of its own and close down the connection immediately, discarding any pending writes. It is not required for the initiator of the close to wait for the responding close_notify alert before closing the read side of the connection.
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 the responding close_notify alert before indicating to the application layer that the TLS connection has ended. If the application protocol will not transfer any additional data, but will only close the underlying transport connection, then the implementation MAY choose to close the transport without waiting for the responding close_notify. 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 a connection reliably delivers pending data before destroying the transport.
Error handling in the TLS Handshake protocol is very simple. When an error is detected, the detecting party sends a message to the other party. Upon transmission or receipt of a fatal alert message, both parties immediately close the connection. Servers and clients MUST forget any session-identifiers, keys, and secrets associated with a failed connection. Thus, any connection terminated with a fatal alert MUST NOT be resumed.
Whenever an implementation encounters a condition which is defined as a fatal alert, it MUST send the appropriate alert prior to closing the connection. For all errors where an alert level is not explicitly specified, the sending party MAY determine at its discretion whether to treat this as a fatal error or not. If the implementation chooses to send an alert but intends to close the connection immediately afterwards, it MUST send that alert at the fatal alert level.
If an alert with a level of warning is sent and received, generally the connection can continue normally. If the receiving party decides not to proceed with the connection (e.g., after having received a no_renegotiation alert that it is not willing to accept), it SHOULD send a fatal alert to terminate the connection. Given this, the sending party cannot, in general, know how the receiving party will behave. Therefore, warning alerts are not very useful when the sending party wants to continue the connection, and thus are sometimes omitted. For example, if a peer decides to accept an expired certificate (perhaps after confirming this with the user) and wants to continue the connection, it would not generally send a certificate_expired alert.
The following error alerts are defined:
New Alert values are assigned by IANA as described in Section 12.
The cryptographic parameters of the session state are produced by the TLS Handshake Protocol, which operates on top of the TLS record layer. When a TLS client and server first start communicating, they agree on a protocol version, select cryptographic algorithms, optionally authenticate each other, and use public-key encryption techniques to generate shared secrets.
The TLS Handshake Protocol involves the following steps:
Note that higher layers should not be overly reliant on whether TLS always negotiates the strongest possible connection between two peers. There are a number of ways in which a man-in-the-middle attacker can attempt to make two entities drop down to the least secure method they support. The protocol has been designed to minimize this risk, but there are still attacks available: for example, an attacker could block access to the port a secure service runs on, or attempt to get the peers to negotiate an unauthenticated connection. The fundamental rule is that higher levels must be cognizant of what their security requirements are and never transmit information over a channel less secure than what they require. The TLS protocol is secure in that any cipher suite offers its promised level of security: if you negotiate AES-GCM [GCM] with a 1024-bit DHE key exchange with a host whose certificate you have verified, you can expect to be that secure.
These goals are achieved by the handshake protocol, which can be summarized as follows: The client sends a ClientHello message which contains a random nonce (ClientHello.random), its preferences for Protocol Version, Cipher Suite, and a variety of extensions. In the same flight, it sends a ClientKeyShare message which contains its share of the parameters for key agreement for some set of expected server parameters (DHE/ECDHE groups, etc.).
If the client has provided a ClientKeyShare with an appropriate set of keying material, the server responds to the ClientHello with a ServerHello message. The ServerHello contains the server’s nonce (ServerHello.random), the server’s choice of the Protocol Version, Session ID and Cipher Suite, and the server’s response to the extensions the client offered.
The server can then generate its own keying material share and send a ServerKeyShare message which contains its share of the parameters for the key agreement. The server can now compute the shared secret (the premaster secret). At this point, the server starts encrypting all remaining handshake traffic with the negotiated cipher suite using a key derived from the premaster secret (via the “handshake master secret”). The remainder of the server’s handshake messages will be encrypted using that key.
Following these messages, the server will send an EncryptedExtensions message which contains a response to any client’s extensions which are not necessary to establish the Cipher Suite. The server will then send its certificate in a Certificate message if it is to be authenticated. The server may optionally request a certificate from the client by sending a CertificateRequest message at this point. Finally, if the server is authenticated, it will send a CertificateVerify message which provides a signature over the entire handshake up to this point. This serves both to authenticate the server and to establish the integrity of the negotiation. Finally, the server sends a Finished message which includes an integrity check over the handshake keyed by the shared secret and demonstrates that the server and client have agreed upon the same keys. [[TODO: If the server is not requesting client authentication, it MAY start sending application data following the Finished, though the server has no way of knowing who will be receiving the data. Add this.]]
Once the client receives the ServerKeyShare, it can also compute the premaster secret and decrypt the server’s remaining handshake messages. The client generates its own sending keys based on the premaster secret and will encrypt the remainder of its handshake messages using those keys and the newly established cipher suite. If the server has sent a CertificateRequest message, the client MUST send the Certificate message, though it may contain zero certificates. If the client has sent a certificate, a digitally-signed CertificateVerify message is sent to explicitly verify possession of the private key in the certificate. Finally, the client sends the Finished message.
At this point, the handshake is complete, and the client and server may exchange application layer data, which is protected using a new set of keys derived from both the premaster secret and the handshake transcript (see [I-D.ietf-tls-session-hash] for the security rationale for this.)
Application data MUST NOT be sent prior to the Finished message. [[TODO: can we make this clearer and more clearly match the text above about server-side False Start.]] Client Server
ClientHello
ClientKeyShare -------->
ServerHello
ServerKeyShare
{EncryptedExtensions*}
{Certificate*}
{CertificateRequest*}
{CertificateVerify*}
<-------- {Finished}
{Certificate*}
{CertificateVerify*}
{Finished} -------->
[Application Data] <-------> [Application Data]
Figure 1. Message flow for a full handshake
* Indicates optional or situation-dependent messages that are not always sent.
{} Indicates messages protected using keys derived from the handshake master secret.
[] Indicates messages protected using keys derived from the master secret.
If the client has not provided an appropriate ClientKeyShare (e.g. it includes only DHE or ECDHE groups unacceptable or unsupported by the server), the server corrects the mismatch with a HelloRetryRequest and the client will need to restart the handshake with an appropriate ClientKeyShare, as shown in Figure 2:
Client Server
ClientHello
ClientKeyShare -------->
<-------- HelloRetryRequest
ClientHello
ClientKeyShare -------->
ServerHello
ServerKeyShare
{EncryptedExtensions*}
{Certificate*}
{CertificateRequest*}
{CertificateVerify*}
<-------- {Finished}
{Certificate*}
{CertificateVerify*}
{Finished} -------->
[Application Data] <-------> [Application Data]
Figure 2. Message flow for a full handshake with mismatched parameters
[[OPEN ISSUE: Should we restart the handshake hash? https://github.com/tlswg/tls13-spec/issues/104.]] [[OPEN ISSUE: We need to make sure that this flow doesn’t introduce downgrade issues. Potential options include continuing the handshake hashes (as long as clients don’t change their opinion of the server’s capabilities with aborted handshakes) and requiring the client to send the same ClientHello (as is currently done) and then checking you get the same negotiated parameters.]]
If no common cryptographic parameters can be negotiated, the server will send a fatal alert.
When the client and server decide to resume a previous session or duplicate an existing session (instead of negotiating new security parameters), the message flow is as follows:
The client sends a ClientHello using the Session ID of the session to be resumed. The server then checks its session cache for a match. If a match is found, and the server is willing to re-establish the connection under the specified session state, it will send a ServerHello with the same Session ID value. At this point, both client and server MUST proceed directly to sending Finished messages, which are protected using handshake keys as described above, computed using resumption premaster secret created in the first handshake as the premaster secret. Once the re-establishment is complete, the client and server MAY begin to exchange application layer data, which is protected using the application secrets (See flow chart below.) If a Session ID match is not found, the server generates a new session ID, and the TLS client and server perform a full handshake.
Client Server
ClientHello
ClientKeyExhange -------->
ServerHello
<-------- {Finished}
{Finished} -------->
[Application Data] <-------> [Application Data]
Figure 3. Message flow for an abbreviated handshake
The contents and significance of each message will be presented in detail in the following sections.
The TLS Handshake Protocol is one of the defined higher-level clients of the TLS Record Protocol. This protocol is used to negotiate the secure attributes of a session. Handshake messages are supplied to the TLS record layer, where they are encapsulated within one or more TLSPlaintext structures, which are processed and transmitted as specified by the current active session state.
enum {
reserved(0), client_hello(1), server_hello(2),
client_key_share(5), hello_retry_request(6),
server_key_share(7), certificate(11), reserved(12),
certificate_request(13), certificate_verify(15),
reserved(16), finished(20), (255)
} HandshakeType;
struct {
HandshakeType msg_type; /* handshake type */
uint24 length; /* bytes in message */
select (HandshakeType) {
case client_hello: ClientHello;
case client_key_share: ClientKeyShare;
case server_hello: ServerHello;
case hello_retry_request: HelloRetryRequest;
case server_key_share: ServerKeyShare;
case certificate: Certificate;
case certificate_request: CertificateRequest;
case certificate_verify: CertificateVerify;
case finished: Finished;
} body;
} Handshake;
The handshake protocol messages are presented below in the order they MUST be sent; sending handshake messages in an unexpected order results in a fatal error. Unneeded handshake messages can be omitted, however.
New handshake message types are assigned by IANA as described in Section 12.
The hello phase messages are used to exchange security enhancement capabilities between the client and server. When a new session begins, the record layer’s connection state AEAD algorithm is initialized to NULL_NULL. The current connection state is used for renegotiation messages.
When this message will be sent:
Structure of this message:
struct {
opaque random_bytes[32];
} Random;
Note: Versions of TLS prior to TLS 1.3 used the top 32 bits of the Random value to encode the time since the UNIX epoch.
Note: The ClientHello message includes a variable-length session identifier. If not empty, the value identifies a session between the same client and server whose security parameters the client wishes to reuse. The session identifier MAY be from an earlier connection, this connection, or from another currently active connection. The second option is useful if the client only wishes to update the random structures and derived values of a connection, and the third option makes it possible to establish several independent secure connections without repeating the full handshake protocol. These independent connections may occur sequentially or simultaneously; a SessionID becomes valid when the handshake negotiating it completes with the exchange of Finished messages and persists until it is removed due to aging or because a fatal error was encountered on a connection associated with the session. The actual contents of the SessionID are defined by the server.
opaque SessionID<0..32>;
Warning: Because the SessionID is transmitted without confidentiality or integrity protection, servers MUST NOT place confidential information in session identifiers or let the contents of fake session identifiers cause any breach of security. (Note that the content of the handshake as a whole, including the SessionID, is protected by the Finished messages exchanged at the end of the handshake.)
The cipher suite list, passed from the client to the server in the ClientHello message, contains the combinations of cryptographic algorithms supported by the client in order of the client’s preference (favorite choice first). Each cipher suite defines a key exchange algorithm, a record protection algorithm (including secret key length) and a PRF. The server will select a cipher suite or, if no acceptable choices are presented, return a handshake failure alert and close the connection. If the list contains cipher suites the server does not recognize, support, or wish to use, the server MUST ignore those cipher suites, and process the remaining ones as usual.
uint8 CipherSuite[2]; /* Cryptographic suite selector */
enum { null(0), (255) } CompressionMethod;
struct {
ProtocolVersion client_version;
Random random;
SessionID session_id;
CipherSuite cipher_suites<2..2^16-2>;
CompressionMethod compression_methods<1..2^8-1>;
select (extensions_present) {
case false:
struct {};
case true:
Extension extensions<0..2^16-1>;
};
} ClientHello;
TLS allows extensions to follow the compression_methods field in an extensions block. The presence of extensions can be detected by determining whether there are bytes following the compression_methods at the end of the ClientHello. Note that this method of detecting optional data differs from the normal TLS method of having a variable-length field, but it is used for compatibility with TLS before extensions were defined.
In the event that a client requests additional functionality using extensions, and this functionality is not supplied by the server, the client MAY abort the handshake. A server MUST accept ClientHello messages both with and without the extensions field, and (as for all other messages) it MUST check that the amount of data in the message precisely matches one of these formats; if not, then it MUST send a fatal “decode_error” alert.
After sending the ClientHello message, the client waits for a ServerHello or HelloRetryRequest message.
When this message will be sent:
Meaning of this message:
Structure of this message:
struct {
NamedGroup group;
opaque key_exchange<1..2^16-1>;
} ClientKeyShareOffer;
struct {
ClientKeyShareOffer offers<0..2^16-1>;
} ClientKeyShare;
Clients may offer an arbitrary number of ClientKeyShareOffer values, each representing a single set of key agreement parameters; for instance a client might offer shares for several elliptic curves or multiple integer DH groups. The shares for each ClientKeyShareOffer MUST by generated independently. Clients MUST NOT offer multiple ClientKeyShareOffers for the same parameters. It is explicitly permitted to send an empty ClientKeyShare message, as this is used to elicit the server’s parameters if the client has no useful information. [TODO: Recommendation about what the client offers. Presumably which integer DH groups and which curves.] [TODO: Work out how this interacts with PSK and SRP.]
Diffie-Hellman parameters for both clients and servers are encoded in the opaque key_exchange field of the ClientKeyShareOffer or ServerKeyShare structures. The opaque value contains the Diffie-Hellman public value (dh_Y = g^X mod p), encoded as a big-endian integer.
opaque dh_Y<1..2^16-1>;
ECDHE parameters for both clients and servers are encoded in the opaque key_exchange field of the ClientKeyShareOffer or ServerKeyShare structures. The opaque value conveys the Elliptic Curve Diffie-Hellman public value (ecdh_Y) represented as a byte string ECPoint.point.
opaque point <1..2^8-1>;
Although X9.62 supports multiple point formats, any given curve MUST specify only a single point format. All curves currently specified in this document MUST only be used with the uncompressed point format.
Note: Versions of TLS prior to 1.3 permitted point negotiation; TLS 1.3 removes this feature in favor of a single point format for each curve.
[[OPEN ISSUE: We will need to adjust the compressed/uncompressed point issue if we have new curves that don’t need point compression. This depends on the CFRG’s recommendations. The expectation is that future curves will come with defined point formats and that existing curves conform to X9.62.]]
When this message will be sent:
Structure of this message:
struct {
ProtocolVersion server_version;
Random random;
SessionID session_id;
CipherSuite cipher_suite;
select (extensions_present) {
case false:
struct {};
case true:
Extension extensions<0..2^16-1>;
};
} ServerHello;
The presence of extensions can be detected by determining whether there are bytes following the cipher_suite field at the end of the ServerHello.
When this message will be sent:
Structure of this message:
struct {
ProtocolVersion server_version;
CipherSuite cipher_suite;
NamedGroup selected_group;
Extension extensions<0..2^16-1>;
} HelloRetryRequest;
[[OPEN ISSUE: Merge in DTLS Cookies?]]
The “server_version”, “cipher_suite” and “extensions” fields have the same meanings as their corresponding values in the ServerHello. The server SHOULD send only the extensions necessary for the client to generate a correct ClientHello/ClientKeyShare pair.
Upon receipt of a HelloRetryRequest, the client MUST send a new ClientHello/ClientKeyShare pair to the server. The ClientKeyShare MUST contain both the groups in the original ClientKeyShare as well as a ClientKeyShareOffer consistent with the “selected_group” field. I.e., it MUST be a superset of the previous ClientKeyShareOffer.
Upon re-sending the ClientHello/ClientKeyShare and receiving the server’s ServerHello/ServerKeyShare, the client MUST verify that the selected ciphersuite and NamedGroup match that supplied in the HelloRetryRequest.
The extension format is:
struct {
ExtensionType extension_type;
opaque extension_data<0..2^16-1>;
} Extension;
enum {
signature_algorithms(13), early_data(TBD), (65535)
} ExtensionType;
Here:
The initial set of extensions is defined in a companion document [TLSEXT]. The list of extension types is maintained by IANA as described in Section 12.
An extension type MUST NOT appear in the ServerHello unless the same extension type appeared in the corresponding ClientHello. If a client receives an extension type in ServerHello that it did not request in the associated ClientHello, it MUST abort the handshake with an unsupported_extension fatal alert.
Nonetheless, “server-oriented” extensions may be provided in the future within this framework. Such an extension (say, of type x) would require the client to first send an extension of type x in a ClientHello with empty extension_data to indicate that it supports the extension type. In this case, the client is offering the capability to understand the extension type, and the server is taking the client up on its offer.
When multiple extensions of different types are present in the ClientHello or ServerHello messages, the extensions MAY appear in any order. There MUST NOT be more than one extension of the same type.
Finally, note that extensions can be sent both when starting a new session and when requesting session resumption. Indeed, a client that requests session resumption does not in general know whether the server will accept this request, and therefore it SHOULD send the same extensions as it would send if it were not attempting resumption.
In general, the specification of each extension type needs to describe the effect of the extension both during full handshake and session resumption. Most current TLS extensions are relevant only when a session is initiated: when an older session is resumed, the server does not process these extensions in Client Hello, and does not include them in Server Hello. However, some extensions may specify different behavior during session resumption.
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:
The client uses the “signature_algorithms” extension to indicate to the server which signature/hash algorithm pairs may be used in digital signatures. The “extension_data” field of this extension contains a “supported_signature_algorithms” value.
enum {
none(0), md5(1), sha1(2), sha224(3), sha256(4), sha384(5),
sha512(6), (255)
} HashAlgorithm;
enum { anonymous(0), rsa(1), dsa(2), ecdsa(3), (255) }
SignatureAlgorithm;
struct {
HashAlgorithm hash;
SignatureAlgorithm signature;
} SignatureAndHashAlgorithm;
SignatureAndHashAlgorithm
supported_signature_algorithms<2..2^16-2>;
Each SignatureAndHashAlgorithm value lists a single hash/signature pair that the client is willing to verify. The values are indicated in descending order of preference.
Note: Because not all signature algorithms and hash algorithms may be accepted by an implementation (e.g., DSA with SHA-1, but not SHA-256), algorithms here are listed in pairs.
The semantics of this extension are somewhat complicated because the cipher suite indicates permissible signature algorithms but not hash algorithms. Section 7.3.5 and Section 7.3.3 describe the appropriate rules.
If the client supports only the default hash and signature algorithms (listed in this section), it MAY omit the signature_algorithms extension. If the client does not support the default algorithms, or supports other hash and signature algorithms (and it is willing to use them for verifying messages sent by the server, i.e., server certificates and server key share), it MUST send the signature_algorithms extension, listing the algorithms it is willing to accept.
If the client does not send the signature_algorithms extension, the server MUST do the following:
Note: this is a change from TLS 1.1 where there are no explicit rules, but as a practical matter one can assume that the peer supports MD5 and SHA-1.
Note: this extension is not meaningful for TLS versions prior to 1.2. Clients MUST NOT offer it if they are offering prior versions. However, even if clients do offer it, the rules specified in [TLSEXT] require servers to ignore extensions they do not understand.
Servers MUST NOT send this extension. TLS servers MUST support receiving this extension.
When performing session resumption, this extension is not included in Server Hello, and the server ignores the extension in Client Hello (if present).
When sent by the client, the “supported_groups” extension indicates the named groups which the client supports, ordered from most preferred to least preferred.
Note: In versions of TLS prior to TLS 1.3, this extension was named “elliptic curves” and only contained elliptic curve groups. See [RFC4492] and [I-D.ietf-tls-negotiated-ff-dhe].
The “extension_data” field of this extension SHALL contain a “NamedGroupList” value:
enum {
// Elliptic Curve Groups.
sect163k1 (1), sect163r1 (2), sect163r2 (3),
sect193r1 (4), sect193r2 (5), sect233k1 (6),
sect233r1 (7), sect239k1 (8), sect283k1 (9),
sect283r1 (10), sect409k1 (11), sect409r1 (12),
sect571k1 (13), sect571r1 (14), secp160k1 (15),
secp160r1 (16), secp160r2 (17), secp192k1 (18),
secp192r1 (19), secp224k1 (20), secp224r1 (21),
secp256k1 (22), secp256r1 (23), secp384r1 (24),
secp521r1 (25),
// Finite Field Groups.
ffdhe2432(256), ffdhe3072(257), ffdhe4096(258),
ffdhe6144(259), ffdhe8192(260),
// Reserved Code Points.
reserved (0xFE00..0xFEFF),
reserved(0xFF01),
reserved(0xFF02),
(0xFFFF)
} NamedGroup;
struct {
NamedGroup named_group_list<1..2^16-1>
} NamedGroupList;
sect163k1, etc: Indicates support of the corresponding named curve The named curves defined here are those specified in SEC 2 [13]. Note that many of these curves are also recommended in ANSI X9.62 [X962] and FIPS 186-2 [DSS]. Values 0xFE00 through 0xFEFF are reserved for private use. Values 0xFF01 and 0xFF02 were used in previous versions of TLS but MUST NOT be offered by TLS 1.3 implementations. [[OPEN ISSUE: Triage curve list.]]
ffdhe2432, etc: Indicates support of the corresponding finite field group, defined in [I-D.ietf-tls-negotiated-ff-dhe]
Items in named_curve_list are ordered according to the client’s preferences (favorite choice first).
As an example, a client that only supports secp192r1 (aka NIST P-192; value 19 = 0x0013) and secp224r1 (aka NIST P-224; value 21 = 0x0015) and prefers to use secp192r1 would include a TLS extension consisting of the following octets. Note that the first two octets indicate the extension type (Supported Group Extension):
00 0A 00 06 00 04 00 13 00 15
The client MUST supply a “named_groups” extension containing at least one group for each key exchange algorithm (currently DHE and ECDHE) for which it offers a cipher suite. If the client does not supply a “named_groups” extension with a compatible group, the server MUST NOT negotiate a cipher suite of the relevant type. For instance, if a client supplies only ECDHE groups, the server MUST NOT negotiate finite field Diffie-Hellman. If no acceptable group can be selected across all cipher suites, then the server MUST generate a fatal “handshake_failure” alert.
NOTE: A server participating in an ECDHE-ECDSA key exchange may use different curves for (i) the ECDSA key in its certificate, and (ii) the ephemeral ECDH key in the ServerKeyExchange message. The server must consider the supported groups in both cases.
[[TODO: IANA Considerations.]]
TLS versions before 1.3 have a strict message ordering and do not permit additional messages to follow the ClientHello. The EarlyData extension allows TLS messages which would otherwise be sent as separate records to be instead inserted in the ClientHello. The extension simply contains the TLS records which would otherwise have been included in the client’s first flight.
struct {
TLSCipherText messages<5 .. 2^24-1>;
} EarlyDataExtension;
Extra messages for the client’s first flight MAY either be transmitted standalone or sent as EarlyData. However, when a client does not know whether TLS 1.3 can be negotiated – e.g., because the server may support a prior version of TLS or because of network intermediaries – it SHOULD use the EarlyData extension. If the EarlyData extension is used, then clients MUST NOT send any messages other than the ClientHello in their initial flight.
Any data included in EarlyData is not integrated into the handshake hashes directly. E.g., if the ClientKeyShare is included in EarlyData, then the handshake hashes consist of ClientHello + ServerHello, etc. However, because the ClientKeyShare is in a ClientHello extension, it is still hashed transitively. This procedure guarantees that the Finished message covers these messages even if they are ultimately ignored by the server (e.g., because it is sent to a TLS 1.2 server). TLS 1.3 servers MUST understand messages sent in EarlyData, and aside from hashing them differently, MUST treat them as if they had been sent immediately after the ClientHello.
Servers MUST NOT send the EarlyData extension. Negotiating TLS 1.3 serves as acknowledgement that it was processed as described above.
[[OPEN ISSUE: This is a fairly general mechanism which is possibly overkill in the 1-RTT case, where it would potentially be more attractive to just have a “ClientKeyShare” extension. However, for the 0-RTT case we will want to send the Certificate, CertificateVerify, and application data, so a more general extension seems appropriate at least until we have determined we don’t need it for 0-RTT.]]
When this message will be sent:
Meaning of this message:
Structure of this message:
struct {
NamedGroup group;
opaque key_exchange<1..2^16-1>;
} ServerKeyShare;
When this message will be sent:
Meaning of this message:
Structure of this message:
struct {
Extension extensions<0..2^16-1>;
} EncryptedExtensions;
When this message will be sent:
Meaning of this message:
Structure of this message:
opaque ASN1Cert<1..2^24-1>;
struct {
ASN1Cert certificate_list<0..2^24-1>;
} Certificate;
The same message type and structure will be used for the client’s response to a certificate request message. Note that a client MAY send no certificates if it does not have an appropriate certificate to send in response to the server’s authentication request.
Note: PKCS #7 [PKCS7] is not used as the format for the certificate vector because PKCS #6 [PKCS6] extended certificates are not used. Also, PKCS #7 defines a SET rather than a SEQUENCE, making the task of parsing the list more difficult.
The following rules apply to the certificates sent by the server:
Key Exchange Alg. Certificate Key Type
DHE_RSA RSA public key; the certificate MUST allow the
ECDHE_RSA key to be used for signing (the
digitalSignature bit MUST be set if the key
usage extension is present) with the signature
scheme and hash algorithm that will be employed
in the server key exchange message.
Note: ECDHE_RSA is defined in [RFC4492].
DHE_DSS DSA public key; the certificate MUST allow the
key to be used for signing with the hash
algorithm that will be employed in the server
key exchange message.
ECDHE_ECDSA ECDSA-capable public key; the certificate MUST
allow the key to be used for signing with the
hash algorithm that will be employed in the
server key exchange message. The public key
MUST use a curve and point format supported by
the client, as described in [RFC4492].
If the client provided a “signature_algorithms” extension, then all certificates provided by the server MUST be signed by a hash/signature algorithm pair that appears in that extension. Note that this implies that a certificate containing a key for one signature algorithm MAY be signed using a different signature algorithm (for instance, an RSA key signed with a DSA key). This is a departure from TLS 1.1, which required that the algorithms be the same.
If the server has multiple certificates, it chooses one of them based on the above-mentioned criteria (in addition to other criteria, such as transport layer endpoint, local configuration and preferences, etc.). If the server has a single certificate, it SHOULD attempt to validate that it meets these criteria.
Note that there are certificates that use algorithms and/or algorithm combinations that cannot be currently used with TLS. For example, a certificate with RSASSA-PSS signature key (id-RSASSA-PSS OID in SubjectPublicKeyInfo) cannot be used because TLS defines no corresponding signature algorithm.
As cipher suites that specify new key exchange methods are specified for the TLS protocol, they will imply the certificate format and the required encoded keying information.
When this message will be sent:
Structure of this message:
enum {
rsa_sign(1), dss_sign(2), rsa_fixed_dh(3), dss_fixed_dh(4),
rsa_ephemeral_dh_RESERVED(5), dss_ephemeral_dh_RESERVED(6),
fortezza_dms_RESERVED(20), (255)
} ClientCertificateType;
opaque DistinguishedName<1..2^16-1>;
struct {
ClientCertificateType certificate_types<1..2^8-1>;
SignatureAndHashAlgorithm
supported_signature_algorithms<2..2^16-2>;
DistinguishedName certificate_authorities<0..2^16-1>;
} CertificateRequest;
rsa_sign a certificate containing an RSA key
dss_sign a certificate containing a DSA key
rsa_fixed_dh a certificate containing a static DH key.
dss_fixed_dh a certificate containing a static DH key
The interaction of the certificate_types and supported_signature_algorithms fields is somewhat complicated. certificate_types has been present in TLS since SSLv3, but was somewhat underspecified. Much of its functionality is superseded by supported_signature_algorithms. The following rules apply:
New ClientCertificateType values are assigned by IANA as described in Section 12.
Note: Values listed as RESERVED may not be used. They were used in SSLv3.
Note: It is a fatal handshake_failure alert for an anonymous server to request client authentication.
When this message will be sent:
Structure of this message:
struct {
digitally-signed struct {
opaque handshake_messages_hash[hash_length];
}
} CertificateVerify;
When this message will be sent:
Meaning of this message:
Structure of this message:
struct {
opaque verify_data[verify_data_length];
} Finished;
verify_data
PRF(hs_master_secret, finished_label, Hash(handshake_messages))
[0..verify_data_length-1];
finished_label
For Finished messages sent by the client, the string
"client finished". For Finished messages sent by the server,
the string "server finished".
The value handshake_messages includes all handshake messages starting at ClientHello up to, but not including, this Finished message. This may be different from handshake_messages in Section 7.3.7 or Section 7.3.10. Also, the handshake_messages for the Finished message sent by the client will be different from that for the Finished message sent by the server, because the one that is sent second will include the prior one.
Note: Alerts and any other record types are not handshake messages and are not included in the hash computations. Also, HelloRequest messages are omitted from handshake hashes.
When this message will be sent:
Meaning of this message:
In particular:
Client Cert. Type Certificate Key Type
rsa_sign RSA public key; the certificate MUST allow the
key to be used for signing with the signature
scheme and hash algorithm that will be
employed in the certificate verify message.
dss_sign DSA public key; the certificate MUST allow the
key to be used for signing with the hash
algorithm that will be employed in the
certificate verify message.
ecdsa_sign ECDSA-capable public key; the certificate MUST
allow the key to be used for signing with the
hash algorithm that will be employed in the
certificate verify message; the public key
MUST use a curve and point format supported by
the server.
rsa_fixed_dh Diffie-Hellman public key; MUST use the same
dss_fixed_dh parameters as server's key.
rsa_fixed_ecdh ECDH-capable public key; MUST use the
ecdsa_fixed_ecdh same curve as the server's key, and MUST use a
point format supported by the server.
Note that, as with the server certificate, there are certificates that use algorithms/algorithm combinations that cannot be currently used with TLS.
When this message will be sent:
In order to begin connection protection, the TLS Record Protocol requires specification of a suite of algorithms, a master secret, and the client and server random values. The authentication, key agreement, and record protection algorithms are determined by the cipher_suite selected by the server and revealed in the ServerHello message. The random values are exchanged in the hello messages. All that remains is to calculate the master secret.
The pre_master_secret is used to generate a series of master secret values, as shown in the following diagram and described below.
Premaster Secret <---------+
| |
PRF |
| |
v |
Handshake <-PRF- Handshake |
Traffic Keys Master Secret |
| | Via
| | Session
+----------+----------+ | Cache
| | |
PRF PRF |
| | |
v v |
Application <-PRF- Master Resumption |
Traffic Keys Secret Premaster --+
Secret
First, as soon as the ClientKeyShare and ServerKeyShare messages have been exchanged, the client and server each use the unauthenticated key shares to generate a master secret which is used for the protection of the remaining handshake records. Specifically, they generate:
hs_master_secret = PRF(pre_master_secret, "handshake master secret",
session_hash)
[0..47];
During resumption, the premaster secret is initialized with the “resumption premaster secret”, rather than using the values from the ClientKeyShare/ServerKeyShare exchange.
This master secret value is used to generate the record protection keys used for the handshake, as described in Section 6.3. It is also used with TLS Exporters [RFC5705].
Once the hs_master_secret has been computed, the premaster secret SHOULD be deleted from memory.
Once the last non-Finished message has been sent, the client and server then compute the master secret which will be used for the remainder of the session:
master_secret = PRF(hs_master_secret, "extended master secret",
session_hash)
[0..47];
If the server does not request client authentication, the master secret can be computed at the time that the server sends its Finished, thus allowing the server to send traffic on its first flight (see [TODO] for security considerations on this practice.) If the server requests client authentication, this secret can be computed after the client’s Certificate and CertificateVerify have been sent, or, if the client refuses client authentication, after the client’s empty Certificate message has been sent.
For full handshakes, each side also derives a new secret which will be used as the premaster_secret for future resumptions of the newly established session. This is computed as:
resumption_premaster_secret = PRF(hs_master_secret,
"resumption premaster secret",
session_hash)
[0..47];
The session_hash value is a running hash of the handshake as defined in Section 8.1.1. Thus, the hs_master_secret is generated using a different session_hash from the other two secrets.
All master secrets are always exactly 48 bytes in length. The length of the premaster secret will vary depending on key exchange method.
When a handshake takes place, we define
session_hash = Hash(handshake_messages)
where “handshake_messages” refers to all handshake messages sent or received, starting at client hello up to the present time, with the exception of the Finished message, including the type and length fields of the handshake messages. This is the concatenation of all the exchanged Handshake structures.
For concreteness, at the point where the handshake master secret is derived, the session hash includes the ClientHello, ClientKeyShare, ServerHello, and ServerKeyShare, and HelloRetryRequest (if any) (though see [https://github.com/tlswg/tls13-spec/issues/104]). At the point where the master secret is derived, it includes every handshake message, with the exception of the Finished messages. Note that if client authentication is not used, then the session hash is complete at the point when the server has sent its first flight. Otherwise, it is only complete when the client has sent its first flight, as it covers the client’s Certificate and CertificateVerify.
A conventional Diffie-Hellman computation is performed. The negotiated key (Z) is used as the pre_master_secret, and is converted into the master_secret, as specified above. Leading bytes of Z that contain all zero bits are stripped before it is used as the pre_master_secret.
Note: Diffie-Hellman parameters are specified by the server and may be either ephemeral or contained within the server’s certificate.
All ECDH calculations (including parameter and key generation as well as the shared secret calculation) are performed according to [6] using the ECKAS-DH1 scheme with the identity map as key derivation function (KDF), so that the premaster secret is the x-coordinate of the ECDH shared secret elliptic curve point represented as an octet string. Note that this octet string (Z in IEEE 1363 terminology) as output by FE2OSP, the Field Element to Octet String Conversion Primitive, has constant length for any given field; leading zeros found in this octet string MUST NOT be truncated.
(Note that this use of the identity KDF is a technicality. The complete picture is that ECDH is employed with a non-trivial KDF because TLS does not directly use the premaster secret for anything other than for computing the master secret.)
In the absence of an application profile standard specifying otherwise, a TLS-compliant application MUST implement the cipher suite TODO:Needs to be selected. (See Appendix A.5 for the definition).
Application data messages are carried by the record layer and are fragmented and encrypted based on the current connection state. The messages are treated as transparent data to the record layer.
Security issues are discussed throughout this memo, especially in Appendices D, E, and F.
[[TODO: Update https://github.com/tlswg/tls13-spec/issues/62]]
This document uses several registries that were originally created in [RFC4346]. IANA has updated these to reference this document. The registries and their allocation policies (unchanged from [RFC4346]) are listed below.
This document also uses a registry originally created in [RFC4366]. IANA has updated it to reference this document. The registry and its allocation policy (unchanged from [RFC4366]) is listed below:
This document also uses two registries originally created in [RFC4492]. IANA [should update/has updated] it to reference this document. The registries and their allocation policies are listed below.
In addition, this document defines two new registries to be maintained by IANA:
This section describes protocol types and constants.
[[TODO: Clean this up to match the in-text description.]]
struct {
uint8 major;
uint8 minor;
} ProtocolVersion;
ProtocolVersion version = { 3, 4 }; /* TLS v1.3*/
enum {
change_cipher_spec(20), alert(21), handshake(22),
application_data(23), (255)
} ContentType;
struct {
ContentType type;
ProtocolVersion version;
uint16 length;
opaque fragment[TLSPlaintext.length];
} TLSPlaintext;
struct {
ContentType type;
ProtocolVersion version;
uint16 length;
opaque nonce_explicit[SecurityParameters.record_iv_length];
aead-ciphered struct {
opaque content[TLSPlaintext.length];
} fragment;
} TLSCiphertext;
struct {
enum { change_cipher_spec(1), (255) } type;
} ChangeCipherSpec;
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),
user_canceled(90),
no_renegotiation(100),
unsupported_extension(110), /* new */
(255)
} AlertDescription;
struct {
AlertLevel level;
AlertDescription description;
} Alert;
enum {
reserved(0), client_hello(1), server_hello(2),
client_key_share(5), hello_retry_request(6),
server_key_share(7), certificate(11), reserved(12),
certificate_request(13), certificate_verify(15),
reserved(16), finished(20), (255)
} HandshakeType;
struct {
HandshakeType msg_type;
uint24 length;
select (HandshakeType) {
case hello_request: HelloRequest;
case client_hello: ClientHello;
case server_hello: ServerHello;
case hello_retry_request: HelloRetryRequest;
case certificate: Certificate;
case server_key_share: ServerKeyShare;
case certificate_request: CertificateRequest;
case server_hello_done: ServerHelloDone;
case certificate_verify: CertificateVerify;
case client_key_share: ClientKeyShare;
case finished: Finished;
} body;
} Handshake;
struct { } HelloRequest;
struct {
opaque random_bytes[32];
} Random;
opaque SessionID<0..32>;
uint8 CipherSuite[2];
enum { null(0), (255) } CompressionMethod;
struct {
ProtocolVersion client_version;
Random random;
SessionID session_id;
CipherSuite cipher_suites<2..2^16-2>;
CompressionMethod compression_methods<1..2^8-1>;
select (extensions_present) {
case false:
struct {};
case true:
Extension extensions<0..2^16-1>;
};
} ClientHello;
struct {
ProtocolVersion server_version;
Random random;
SessionID session_id;
CipherSuite cipher_suite;
select (extensions_present) {
case false:
struct {};
case true:
Extension extensions<0..2^16-1>;
};
} ServerHello;
struct {
ExtensionType extension_type;
opaque extension_data<0..2^16-1>;
} Extension;
enum {
signature_algorithms(13), (65535)
} ExtensionType;
enum{
none(0), md5(1), sha1(2), sha224(3), sha256(4), sha384(5),
sha512(6), (255)
} HashAlgorithm;
enum {
anonymous(0), rsa(1), dsa(2), ecdsa(3), (255)
} SignatureAlgorithm;
struct {
HashAlgorithm hash;
SignatureAlgorithm signature;
} SignatureAndHashAlgorithm;
SignatureAndHashAlgorithm
supported_signature_algorithms<2..2^16-2>;
opaque ASN1Cert<2^24-1>;
struct {
ASN1Cert certificate_list<0..2^24-1>;
} Certificate;
struct {
NamedGroup group;
opaque key_exchange<1..2^16-1>;
} ServerKeyShare;
enum {
rsa_sign(1), dss_sign(2), rsa_fixed_dh(3), dss_fixed_dh(4),
rsa_ephemeral_dh_RESERVED(5), dss_ephemeral_dh_RESERVED(6),
fortezza_dms_RESERVED(20),
(255)
} ClientCertificateType;
opaque DistinguishedName<1..2^16-1>;
struct {
ClientCertificateType certificate_types<1..2^8-1>;
DistinguishedName certificate_authorities<0..2^16-1>;
} CertificateRequest;
struct { } ServerHelloDone;
struct {
ClientKeyShareOffer offers<0..2^16-1>;
} ClientKeyShare;
struct {
NamedGroup group;
opaque key_exchange<1..2^16-1>;
} ClientKeyShareOffer;
struct {
digitally-signed struct {
opaque handshake_messages_hash[hash_length];
}
} CertificateVerify;
The context string for the signature is “TLS 1.3, client CertificateVerify”.
struct {
opaque verify_data[verify_data_length];
} Finished;
The following values define the cipher suite codes used in the ClientHello and ServerHello messages.
A cipher suite defines a cipher specification supported in TLS Version 1.2.
TLS_NULL_WITH_NULL_NULL is specified and is the initial state of a TLS connection during the first handshake on that channel, but MUST NOT be negotiated, as it provides no more protection than an unsecured connection.
CipherSuite TLS_NULL_WITH_NULL_NULL = { 0x00,0x00 };
The following cipher suite definitions, defined in {{RFC5288}, are used for server-authenticated (and optionally client-authenticated) Diffie-Hellman. DHE denotes ephemeral Diffie-Hellman, where the Diffie-Hellman parameters are signed by a signature-capable certificate, which has been signed by the CA. The signing algorithm used by the server is specified after the DHE component of the CipherSuite name. The server can request any signature-capable certificate from the client for client authentication.
CipherSuite TLS_RSA_WITH_AES_128_GCM_SHA256 = {0x00,0x9C}
CipherSuite TLS_RSA_WITH_AES_256_GCM_SHA384 = {0x00,0x9D}
CipherSuite TLS_DHE_RSA_WITH_AES_128_GCM_SHA256 = {0x00,0x9E}
CipherSuite TLS_DHE_RSA_WITH_AES_256_GCM_SHA384 = {0x00,0x9F}
CipherSuite TLS_DHE_DSS_WITH_AES_128_GCM_SHA256 = {0x00,0xA2}
CipherSuite TLS_DHE_DSS_WITH_AES_256_GCM_SHA384 = {0x00,0xA3}
The following cipher suite definitions, defined in {{RFC5289}, are used for server-authenticated (and optionally client-authenticated) Elliptic Curve Diffie-Hellman. ECDHE denotes ephemeral Diffie-Hellman, where the Diffie-Hellman parameters are signed by a signature-capable certificate, which has been signed by the CA. The signing algorithm used by the server is specified after the DHE component of the CipherSuite name. The server can request any signature-capable certificate from the client for client authentication.
CipherSuite TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256 = {0xC0,0x2B};
CipherSuite TLS_ECDHE_ECDSA_WITH_AES_256_GCM_SHA384 = {0xC0,0x2C};
CipherSuite TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256 = {0xC0,0x2F};
CipherSuite TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384 = {0xC0,0x30};
The following ciphers, defined in [RFC5288], are used for completely anonymous Diffie-Hellman communications in which neither party is authenticated. Note that this mode is vulnerable to man-in-the-middle attacks. Using this mode therefore is of limited use: These cipher suites MUST NOT be used by TLS 1.2 implementations unless the application layer has specifically requested to allow anonymous key exchange. (Anonymous key exchange may sometimes be acceptable, for example, to support opportunistic encryption when no set-up for authentication is in place, or when TLS is used as part of more complex security protocols that have other means to ensure authentication.)
CipherSuite TLS_DH_anon_WITH_AES_128_GCM_SHA256 = {0x00,0xA6}
CipherSuite TLS_DH_anon_WITH_AES_256_GCM_SHA384 = {0x00,0xA7}
[[TODO: Add all the defined AEAD ciphers. This currently only lists GCM. https://github.com/tlswg/tls13-spec/issues/53]] Note that using non-anonymous key exchange without actually verifying the key exchange is essentially equivalent to anonymous key exchange, and the same precautions apply. While non-anonymous key exchange will generally involve a higher computational and communicational cost than anonymous key exchange, it may be in the interest of interoperability not to disable non-anonymous key exchange when the application layer is allowing anonymous key exchange.
The PRFs SHALL be as follows:
o For cipher suites ending with _SHA256, the PRF is the TLS PRF with SHA-256 as the hash function.
o For cipher suites ending with _SHA384, the PRF is the TLS PRF with SHA-384 as the hash function.
New cipher suite values are been assigned by IANA as described in Section 12.
Note: The cipher suite values { 0x00, 0x1C } and { 0x00, 0x1D } are reserved to avoid collision with Fortezza-based cipher suites in SSL 3.
These security parameters are determined by the TLS Handshake Protocol and provided as parameters to the TLS record layer in order to initialize a connection state. SecurityParameters includes:
enum { null(0), (255) } CompressionMethod;
enum { server, client } ConnectionEnd;
enum { tls_prf_sha256 } PRFAlgorithm;
enum { aes_gcm } RecordProtAlgorithm;
/* Other values may be added to the algorithms specified in
PRFAlgorithm and RecordProtAlgorithm */
struct {
ConnectionEnd entity;
PRFAlgorithm prf_algorithm;
RecordProtAlgorithm record_prot_algorithm;
uint8 enc_key_length;
uint8 block_length;
uint8 fixed_iv_length;
uint8 record_iv_length;
opaque master_secret[48];
opaque client_random[32];
opaque server_random[32];
} SecurityParameters;
RFC 4492 [RFC4492] adds Elliptic Curve cipher suites to TLS. This document changes some of the structures used in that document. This section details the required changes for implementors of both RFC 4492 and TLS 1.2. Implementors of TLS 1.2 who are not implementing RFC 4492 do not need to read this section.
This document adds a “signature_algorithm” field to the digitally- signed element in order to identify the signature and digest algorithms used to create a signature. This change applies to digital signatures formed using ECDSA as well, thus allowing ECDSA signatures to be used with digest algorithms other than SHA-1, provided such use is compatible with the certificate and any restrictions imposed by future revisions of [RFC3280].
As described in Section 7.3.5 and Section 7.3.9, the restrictions on the signature algorithms used to sign certificates are no longer tied to the cipher suite (when used by the server) or the ClientCertificateType (when used by the client). Thus, the restrictions on the algorithm used to sign certificates specified in Sections 2 and 3 of RFC 4492 are also relaxed. As in this document, the restrictions on the keys in the end-entity certificate remain.
Cipher Suite Key Record
Exchange Protection PRF
TLS_NULL_WITH_NULL_NULL NULL NULL_NULL N/A
TLS_DHE_RSA_WITH_AES_128_GCM_SHA256 DHE_RSA AES_128_GCM SHA256
TLS_DHE_RSA_WITH_AES_256_GCM_SHA384 DHE_RSA AES_256_GCM SHA384
TLS_DHE_DSS_WITH_AES_128_GCM_SHA256 DHE_DSS AES_128_GCM SHA256
TLS_DHE_DSS_WITH_AES_256_GCM_SHA384 DHE_DSS AES_256_GCM SHA384
TLS_DH_anon_WITH_AES_128_GCM_SHA256 DH_anon AES_128_GCM SHA256
TLS_DH_anon_WITH_AES_256_GCM_SHA384 DH_anon AES_128_GCM SHA384
Key Implicit IV Explicit IV
Cipher Material Size Size
------------ -------- ---------- -----------
NULL 0 0 0
AES_128_GCM 16 4 8
AES_256_GCM 32 4 8
The TLS protocol cannot prevent many common security mistakes. This section provides several recommendations to assist implementors.
TLS requires a cryptographically secure pseudorandom number generator (PRNG). Care must be taken in designing and seeding PRNGs. PRNGs based on secure hash operations, most notably SHA-1, are acceptable, but cannot provide more security than the size of the random number generator state.
To estimate the amount of seed material being produced, add the number of bits of unpredictable information in each seed byte. For example, keystroke timing values taken from a PC compatible’s 18.2 Hz timer provide 1 or 2 secure bits each, even though the total size of the counter value is 16 bits or more. Seeding a 128-bit PRNG would thus require approximately 100 such timer values.
[RFC4086] provides guidance on the generation of random values.
Implementations are responsible for verifying the integrity of certificates and should generally support certificate revocation messages. Certificates should always be verified to ensure proper signing by a trusted Certificate Authority (CA). The selection and addition of trusted CAs should be done very carefully. Users should be able to view information about the certificate and root CA.
TLS supports a range of key sizes and security levels, including some that provide no or minimal security. A proper implementation will probably not support many cipher suites. For instance, anonymous Diffie-Hellman is strongly discouraged because it cannot prevent man- in-the-middle attacks. Applications should also enforce minimum and maximum key sizes. For example, certificate chains containing 512- bit RSA keys or signatures are not appropriate for high-security applications.
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:
Cryptographic details:
[[TODO: Revise backward compatibility section for TLS 1.3. https://github.com/tlswg/tls13-spec/issues/54]] Since there are various versions of TLS (1.0, 1.1, 1.2, and any future versions) and SSL (2.0 and 3.0), means are needed to negotiate the specific protocol version to use. The TLS protocol provides a built-in mechanism for version negotiation so as not to bother other protocol components with the complexities of version selection.
TLS versions 1.0, 1.1, and 1.2, and SSL 3.0 are very similar, and use compatible ClientHello messages; thus, supporting all of them is relatively easy. Similarly, servers can easily handle clients trying to use future versions of TLS as long as the ClientHello format remains compatible, and the client supports the highest protocol version available in the server.
A TLS 1.3 client who wishes to negotiate with such older servers will send a normal TLS 1.3 ClientHello, containing { 3, 4 } (TLS 1.3) in ClientHello.client_version. If the server does not support this version, 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.
If the version chosen by the server is not supported by the client (or not acceptable), the client MUST send a “protocol_version” alert message and close the connection.
If a TLS server receives a ClientHello containing a version number greater than the highest version supported by the server, it MUST reply according to the highest version supported by the server.
A TLS server can also receive a ClientHello containing a version number smaller than the highest supported version. If the server wishes to negotiate with old clients, it will proceed as appropriate for the highest version supported by the server that is not greater than ClientHello.client_version. For example, if the server supports TLS 1.0, 1.1, and 1.2, and client_version is TLS 1.0, the server will proceed with a TLS 1.0 ServerHello. If server supports (or is willing to use) only versions greater than client_version, it MUST send a “protocol_version” alert message and close the connection.
Whenever a client already knows the highest protocol version known to a server (for example, when resuming a session), it SHOULD initiate the connection in that native protocol.
Note: some server implementations are known to implement version negotiation incorrectly. For example, there are buggy TLS 1.0 servers that simply close the connection when the client offers a version newer than TLS 1.0. Also, it is known that some servers will refuse the connection if any TLS extensions are included in ClientHello. Interoperability with such buggy servers is a complex topic beyond the scope of this document, and may require multiple connection attempts by the client.
Earlier versions of the TLS specification were not fully clear on what the record layer version number (TLSPlaintext.version) should contain when sending ClientHello (i.e., before it is known which version of the protocol will be employed). Thus, TLS servers compliant with this specification MUST accept any value {03,XX} as the record layer version number for ClientHello.
TLS clients that wish to negotiate with older servers MAY send any value {03,XX} as the record layer version number. Typical values would be {03,00}, the lowest version number supported by the client, and the value of ClientHello.client_version. No single value will guarantee interoperability with all old servers, but this is a complex topic beyond the scope of this document.
TLS 1.2 clients that wish to support SSL 2.0 servers MUST send version 2.0 CLIENT-HELLO messages defined in [SSL2]. The message MUST contain the same version number as would be used for ordinary ClientHello, and MUST encode the supported TLS cipher suites in the CIPHER-SPECS-DATA field as described below.
Warning: The ability to send version 2.0 CLIENT-HELLO messages will be phased out with all due haste, since the newer ClientHello format provides better mechanisms for moving to newer versions and negotiating extensions. TLS 1.2 clients SHOULD NOT support SSL 2.0.
However, even TLS servers that do not support SSL 2.0 MAY accept version 2.0 CLIENT-HELLO messages. The message is presented below in sufficient detail for TLS server implementors; the true definition is still assumed to be [SSL2].
For negotiation purposes, 2.0 CLIENT-HELLO is interpreted the same way as a ClientHello with a “null” compression method and no extensions. Note that this message MUST be sent directly on the wire, not wrapped as a TLS record. For the purposes of calculating Finished and CertificateVerify, the msg_length field is not considered to be a part of the handshake message.
uint8 V2CipherSpec[3];
struct {
uint16 msg_length;
uint8 msg_type;
Version version;
uint16 cipher_spec_length;
uint16 session_id_length;
uint16 challenge_length;
V2CipherSpec cipher_specs[V2ClientHello.cipher_spec_length];
opaque session_id[V2ClientHello.session_id_length];
opaque challenge[V2ClientHello.challenge_length;
} V2ClientHello;
Note: Requests to resume a TLS session MUST use a TLS client hello.
When TLS clients fall back to Version 2.0 compatibility mode, they MUST use special PKCS#1 block formatting. This is done so that TLS servers will reject Version 2.0 sessions with TLS-capable clients.
When a client negotiates SSL 2.0 but also supports TLS, it MUST set the right-hand (least-significant) 8 random bytes of the PKCS padding (not including the terminal null of the padding) for the RSA encryption of the ENCRYPTED-KEY-DATA field of the CLIENT-MASTER-KEY to 0x03 (the other padding bytes are random).
When a TLS-capable server negotiates SSL 2.0 it SHOULD, after decrypting the ENCRYPTED-KEY-DATA field, check that these 8 padding bytes are 0x03. If they are not, the server SHOULD generate a random value for SECRET-KEY-DATA, and continue the handshake (which will eventually fail since the keys will not match). Note that reporting the error situation to the client could make the server vulnerable to attacks described in [BLEI].
The TLS protocol is designed to establish a secure connection between a client and a server communicating over an insecure channel. This document makes several traditional assumptions, including that attackers have substantial computational resources and cannot obtain secret information from sources outside the protocol. Attackers are assumed to have the ability to capture, modify, delete, replay, and otherwise tamper with messages sent over the communication channel. This appendix outlines how TLS has been designed to resist a variety of attacks.
The handshake protocol is responsible for selecting a cipher spec and generating a master secret, which together comprise the primary cryptographic parameters associated with a secure session. The handshake protocol can also optionally authenticate parties who have certificates signed by a trusted certificate authority.
TLS supports three authentication modes: authentication of both parties, server authentication with an unauthenticated client, and total anonymity. Whenever the server is authenticated, the channel is secure against man-in-the-middle attacks, but completely anonymous sessions are inherently vulnerable to such attacks. Anonymous servers cannot authenticate clients. If the server is authenticated, its certificate message must provide a valid certificate chain leading to an acceptable certificate authority. Similarly, authenticated clients must supply an acceptable certificate to the server. Each party is responsible for verifying that the other’s certificate is valid and has not expired or been revoked.
The general goal of the key exchange process is to create a pre_master_secret known to the communicating parties and not to attackers. The pre_master_secret will be used to generate the master_secret (see Section 8.1). The master_secret is required to generate the Finished messages and record protection keys (see Section 7.3.8 and Section 6.3). By sending a correct Finished message, parties thus prove that they know the correct pre_master_secret.
Completely anonymous sessions can be established using Diffie-Hellman for key exchange. The server’s public parameters are contained in the server key share message, and the client’s are sent in the client key share message. Eavesdroppers who do not know the private values should not be able to find the Diffie-Hellman result (i.e., the pre_master_secret).
Warning: Completely anonymous connections only provide protection against passive eavesdropping. Unless an independent tamper-proof channel is used to verify that the Finished messages were not replaced by an attacker, server authentication is required in environments where active man-in-the-middle attacks are a concern.
When Diffie-Hellman key exchange is used, the client and server use the client key exchange and server key exchange messages to send temporary Diffie-Hellman parameters. The signature in the certificate verify message (if present) covers the entire handshake up to that point and thus attests the certificate holder’s desire to use the the ephemeral DHE keys.
Peers SHOULD validate each other’s public key Y (dh_Ys offered by the server or DH_Yc offered by the client) by ensuring that 1 < Y < p-1. This simple check ensures that the remote peer is properly behaved and isn’t forcing the local system into a small subgroup.
Additionally, using a fresh key for each handshake provides Perfect Forward Secrecy. Implementations SHOULD generate a new X for each handshake when using DHE cipher suites.
Because TLS includes substantial improvements over SSL Version 2.0, attackers may try to make TLS-capable clients and servers fall back to Version 2.0. This attack can occur if (and only if) two TLS- capable parties use an SSL 2.0 handshake.
Although the solution using non-random PKCS #1 block type 2 message padding is inelegant, it provides a reasonably secure way for Version 3.0 servers to detect the attack. This solution is not secure against attackers who can brute-force the key and substitute a new ENCRYPTED-KEY-DATA message containing the same key (but with normal padding) before the application-specified wait threshold has expired. Altering the padding of the least-significant 8 bytes of the PKCS padding does not impact security for the size of the signed hashes and RSA key lengths used in the protocol, since this is essentially equivalent to increasing the input block size by 8 bytes.
An attacker might try to influence the handshake exchange to make the parties select different encryption algorithms than they would normally choose.
For this attack, an attacker must actively change one or more handshake messages. If this occurs, the client and server will compute different values for the handshake message hashes. As a result, the parties will not accept each others’ Finished messages. Without the master_secret, the attacker cannot repair the Finished messages, so the attack will be discovered.
When a connection is established by resuming a session, new ClientHello.random and ServerHello.random values are hashed with the session’s master_secret. Provided that the master_secret has not been compromised and that the secure hash operations used to produce the record protection kayes are secure, the connection should be secure and effectively independent from previous connections. Attackers cannot use known keys to compromise the master_secret without breaking the secure hash operations.
Sessions cannot be resumed unless both the client and server agree. If either party suspects that the session may have been compromised, or that certificates may have expired or been revoked, it should force a full handshake. An upper limit of 24 hours is suggested for session ID lifetimes, since an attacker who obtains a master_secret may be able to impersonate the compromised party until the corresponding session ID is retired. Applications that may be run in relatively insecure environments should not write session IDs to stable storage.
The master_secret is hashed with the ClientHello.random and ServerHello.random to produce unique record protection secrets for each connection.
Outgoing data is protected using an AEAD algorithm before transmission. The authentication data includes the sequence number, message type, message length, and the message contents. The message type field is necessary to ensure that messages intended for one TLS record layer client are not redirected to another. The sequence number ensures that attempts to delete or reorder messages will be detected. Since sequence numbers are 64 bits long, they should never overflow. Messages from one party cannot be inserted into the other’s output, since they use independent keys.
TLS is susceptible to a number of denial-of-service (DoS) attacks. In particular, an attacker who initiates a large number of TCP connections can cause a server to consume large amounts of CPU doing asymmetric crypto operations. However, because TLS is generally used over TCP, it is difficult for the attacker to hide his point of origin if proper TCP SYN randomization is used [RFC1948] by the TCP stack.
Because TLS runs over TCP, it is also susceptible to a number of DoS attacks on individual connections. In particular, attackers can forge RSTs, thereby terminating connections, or forge partial TLS records, thereby causing the connection to stall. These attacks cannot in general be defended against by a TCP-using protocol. Implementors or users who are concerned with this class of attack should use IPsec AH [RFC4302] or ESP [RFC4303].
For TLS to be able to provide a secure connection, both the client and server systems, keys, and applications must be secure. In addition, the implementation must be free of security errors.
The system is only as strong as the weakest key exchange and authentication algorithm supported, and only trustworthy cryptographic functions should be used. Short public keys and anonymous servers should be used with great caution. Implementations and users must be careful when deciding which certificates and certificate authorities are acceptable; a dishonest certificate authority can do tremendous damage.
The discussion list for the IETF TLS working group is located at the e-mail address tls@ietf.org. Information on the group and information on how to subscribe to the list is at https://www1.ietf.org/mailman/listinfo/tls
Archives of the list can be found at: http://www.ietf.org/mail-archive/web/tls/current/index.html
Christopher Allen (co-editor of TLS 1.0) Alacrity Ventures ChristopherA@AlacrityManagement.com Martin Abadi University of California, Santa Cruz abadi@cs.ucsc.edu Karthikeyan Bhargavan (co-author of [I-D.ietf-tls-session-hash]) INRIA karthikeyan.bhargavan@inria.fr Steven M. Bellovin Columbia University smb@cs.columbia.edu Simon Blake-Wilson (co-author of RFC4492) BCI sblakewilson@bcisse.com Nelson Bolyard Sun Microsystems, Inc. nelson@bolyard.com (co-author of RFC4492) Ran Canetti IBM canetti@watson.ibm.com Pete Chown Skygate Technology Ltd pc@skygate.co.uk Antoine Delignat-Lavaud (co-author of [I-D.ietf-tls-session-hash]) INRIA antoine.delignat-lavaud@inria.fr Taher Elgamal taher@securify.com Securify Pasi Eronen pasi.eronen@nokia.com Nokia Anil Gangolli anil@busybuddha.org Vipul Gupta (co-author of RFC4492) Sun Microsystems Laboratories vipul.gupta@sun.com Kipp Hickman Chris Hawk (co-author of RFC4492) Corriente Networks LLC chris@corriente.net Alfred Hoenes David Hopwood Independent Consultant david.hopwood@blueyonder.co.uk Daniel Kahn Gillmor ACLU dkg@fifthhorseman.net Phil Karlton (co-author of SSLv3) Paul Kocher (co-author of SSLv3) Cryptography Research paul@cryptography.com Hugo Krawczyk IBM hugo@ee.technion.ac.il Adam Langley (co-author of [I-D.ietf-tls-session-hash]) Google agl@google.com Ilari Liusvaara ilari.liusvaara@elisanet.fi Jan Mikkelsen Transactionware janm@transactionware.com Bodo Moeller (co-author of RFC4492) Google bodo@openssl.org Magnus Nystrom RSA Security magnus@rsasecurity.com Alfredo Pironti (co-author of [I-D.ietf-tls-session-hash]) INRIA alfredo.pironti@inria.fr Marsh Ray (co-author of [I-D.ietf-tls-session-hash]) Microsoft maray@microsoft.com Robert Relyea Netscape Communications relyea@netscape.com Jim Roskind Netscape Communications jar@netscape.com Michael Sabin Dan Simon Microsoft, Inc. dansimon@microsoft.com Martin Thomson Mozilla mt@mozilla.com Tom Weinstein Tim Wright Vodafone timothy.wright@vodafone.com