The SSL Protocol

The SSL protocol has been submitted to the W3O working group on security for consideration as part of a general security approach for the Web, and we are actively working within the W3O and with many of its member entities on establishing open security standards for the net.

This protocol specification was last revised on November 29th, 1994. Recent changes include:


Experimental                                             Kipp E.B. Hickman
Request For Comments: XXXX                   Netscape Communications Corp.
                                              Last Update: Nov. 29th, 1994

The SSL Protocol

Status of this Memo

This RFC specifies a security protocol for the Internet community, and requests discussion and suggestions for improvements. Distribution of this memo is unlimited.

Abstract

This document specifies the Secure Sockets Layer (SSL) protocol, a security protocol that provides privacy over the Internet. The protocol allows client/server applications to communicate in a way that cannot be eavesdropped. Server's are always authenticated and clients are optionally authenticated.

Motivation

The SSL Protocol is designed to provide privacy between two communicating applications (a client and a server). Second, the protocol is designed to authenticate the server, and optionally the client. SSL requires a reliable transport protocol (e.g. TCP) for data transmission and reception.

The advantage of the SSL Protocol is that it is application protocol independent. A "higher level" application protocol (e.g. HTTP, FTP, TELNET, etc.) can layer on top of the SSL Protocol transparently. The SSL Protocol can negotiate an encryption algorithm and session key as well as authenticate a server before the application protocol transmits or receives its first byte of data. All of the application protocol data is transmitted encrypted, ensuring privacy.

1. SSL Record Protocol Specification

1.1 SSL Record Header Format

In SSL, all data sent is encapsulated in a record, an object which is composed of a header and some non-zero amount of data. Each record header contains a two or three byte length code. If the most significant bit is set in the first byte of the record length code then the record has no padding and the total header length will be 2 bytes, otherwise the record has padding and the total header length will be 3 bytes. The record header is transmitted before the data portion of the record.

Note that in the long header case (3 bytes total), the second most significant bit in the first byte has special meaning. When zero, the record being sent is a data record. When one, the record being sent is a security escape (there are currently no examples of security escapes; this is reserved for future versions of the protocol). In either case, the length code describes how much data is in the record.

The record length code does not include the number of bytes consumed by the record header (2 or 3). For the 2 byte header, the record length is computed as follows:

RECORD-LENGTH = ((byte[0] & 0x7f) << 8)) | byte[1];
When the 3 byte header is used, the record length is computed as follows:

RECORD-LENGTH = ((byte[0] & 0x3f) << 8)) | byte[1];
IS-ESCAPE = (byte[0] & 0x40) != 0;
PADDING = byte[2];
The record header defines a value called PADDING. The PADDING value specifies how many bytes of data were appended to the original record by the sender. The padding data is used to make the record length be a multiple of the block ciphers block size when a block cipher is used for encryption.

The sender of a "padded" record appends the padding data to the end of its normal data and then encrypts the total amount (which is now a multiple of the block cipher's block size). The actual value of the padding data is unimportant, but the encrypted form of it must be transmitted for the receiver to properly decrypt the record. Once the total amount being transmitted is known the header can be properly constructed with the PADDING value set appropriately.

The receiver of a padded record decrypts the entire record data (sans record length and the optional padding) to get the clear data, then subtracts the PADDING value from the RECORD-LENGTH to determine the final RECORD-LENGTH. The clear form of the padding data must be discarded.

1.2 SSL Record Data Format

The data portion of an SSL record is composed of three components:

MAC-DATA[MAC-SIZE]
ACTUAL-DATA[N]
PADDING-DATA[PADDING]
ACTUAL-DATA is the actual data being transmitted (the message payload). PADDING-DATA is the padding data sent when a block cipher is used and padding is needed. Finally, MAC-DATA is the Message Authentication Code.

When SSL records are sent in the clear, no cipher is used. Consequently the amount of PADDING-DATA will be zero and the amount of MAC-DATA will be zero. When encryption is in effect, the PADDING-DATA will be a function of the cipher block size. The MAC-DATA is a function of the CIPHER-CHOICE (more about that later).

The MAC-DATA is computed as follows:

MAC-DATA = HASH[ SECRET, ACTUAL-DATA, PADDING-DATA, SEQUENCE-NUMBER ]
The SECRET value is a function of which party is sending the message. If the client is sending the message then the SECRET is the CLIENT-WRITE-KEY (the server will use the SERVER-READ-KEY to verify the MAC). If the client is receiving the message then the SECRET is the CLIENT-READ-KEY (the server will use the SERVER-WRITE-KEY to generate the MAC).

The SEQUENCE-NUMBER is a counter which is incremented by both the sender and the receiver. For each transmission direction, a pair of counters is kept (one by the sender, one by the receiver). Every time a message is sent by a sender the counter is incremented. Sequence numbers are 32 bit unsigned quantities and must wrap to zero after incrementing past 0xFFFFFFFF.

The receiver of a message uses the expected value of the sequence number as input into the MAC HASH function (the HASH function is chosen from the CIPHER-CHOICE). The computed MAC-DATA must agree bit for bit with the transmitted MAC-DATA. If the comparision is not identity then the record is considered damaged, and it is to be treated as if an "I/O Error" had occurred (i.e. an unrecoverable error is asserted and the connection is closed).

A final consistency check is done when a block cipher is used and the protocol is using encryption. The amount of data present in a record (RECORD-LENGTH))must be a multiple of the cipher's block size. If the received record is not a multiple of the cipher's block size then the record is considered damaged, and it is to be treated as if an "I/O Error" had occurred (i.e. an unrecoverable error is asserted and the connection is closed).

The SSL Record Layer is used for all SSL communications, including handshake messages, security escapes and application data transfers. The SSL Record Layer is used by both the client and the server at all times.

For a six byte header, the maximum record length is 32767 bytes. For the seven byte header, the maximum record length is 16383 bytes. The SSL Handshake Protocol messages are constrained to fit in a single SSL Record Protocol record. Application protocol messages are allowed to consume multiple SSL Record Protocol record's.

Before the first record is sent using SSL the sender and receiver send/receive sequence numbers to zero.

2. SSL Handshake Protocol Specification

2.1 SSL Handshake Protocol Flow

The SSL Handshake Protocol has two major phases. The first phase is used to establish private communications. The second phase is used for client authentication.

Phase 1

The first phase is the initial connection phase where both parties communicate their "hello" messages. The client initiates the conversation by sending the CLIENT-HELLO message. The server receives the CLIENT-HELLO message and processes it responding with the SERVER-HELLO message.

At this point both the client and server have enough information to know whether or not a new master key is needed. When a new master key is not needed, both the client and the server proceed immediately to phase 2.

When a new master key is needed, the SERVER-HELLO message will contain enough information for the client to generate it. This includes the servers signed certificate (more about that later), a list of bulk cipher specifications (see below), and a connection-id (a connection-id is a randomly generated value generated by the server that is used by the client and server during a single connection). The client generates the master key and responds with a CLIENT-MASTER-KEY message (or an ERROR message if the server information indicates that the client and server cannot agree on a bulk cipher).

It should be noted here that each SSL endpoint uses a pair of ciphers per connection (for a total of four ciphers). At each endpoint, one cipher is used for outgoing communications, and one is used for incoming communications. When the client or server generate a session key, they actually generate two keys, the SERVER-READ-KEY (also known as the CLIENT-WRITE-KEY) and the SERVER-WRITE-KEY (also known as the CLIENT-READ-KEY). The master key is used by the client and server to generate the various session keys (more about that later).

Finally, the server sends a SERVER-VERIFY message to the client after the master key has been determined. This final step authenticates the server, because only a server which has the appropriate public key can know the master key.

Phase 2

The second phase is the authentication phase. The server has already been authenticated by the client in the first phase, so this phase is primarily used to authenticate the client. In a typical scenario, the server will require something from the client and send a request. The client will answer in the positive if it has the needed information, or send an ERROR message if it does not. This protocol specification does not define the semantics of an ERROR response to a server request (e.g., an implementation can ignore the error, close the connection, etc. and still conform to this specification).

When a party is done authenticating the other party, it sends its finished message. For the client, the CLIENT-FINISHED message contains the encrypted form of the CONNECTION-ID for the server to verify. If the verification fails, the server sends an ERROR message.

Once a party has sent its finished message it must continue to listen to its peers messages until it too receives a finished message. Once a party has both sent a finished message and received its peers finished message, the SSL handshake protocol is done. At this point the application protocol begins to operate (Note: the application protocol continues to be layered on the SSL Record Protocol).

2.2 Typical Protocol Message Flow

The following sequences define several typical protocol message flows for the SSL Handshake Protocol. In these examples we have two principals in the conversation: the client and the server. We use a notation commonly found in the literature [10]. When something is enclosed in curly braces "{something}key" then the something has been encrypted using "key".

2.2.1 Assuming no session-identifier

client-hello         C -> S: challenge, cipher_specs
server-hello         S -> C: connection-id,server_certificate,cipher_specs
client-master-key    C -> S: {master_key}server_public_key
client-finish        C -> S: {connection-id}client_write_key
server-verify        S -> C: {challenge}server_write_key
server-finish        S -> C: {new_session_id}server_write_key

2.2.2 Assuming a session-identifier was found by both client & server

client-hello         C -> S: challenge, session_id, cipher_specs
server-hello         S -> C: connection-id, session_id_hit
client-finish        C -> S: {connection-id}client_write_key
server-verify        S -> C: {challenge}server_write_key
server-finish        S -> C: {session_id}server_write_key

2.2.3 Assuming a session-identifier was used and client authentication is used

client-hello         C -> S: challenge, session_id, cipher_specs
server-hello         S -> C: connection-id, session_id_hit
client-finish        C -> S: {connection-id}client_write_key
server-verify        S -> C: {challenge}server_write_key
request-certificate  S -> C: {auth_type,challenge'}server_write_key
client-certificate   C -> S: {cert_type,client_cert,
			      response_data}client_write_key
server-finish        S -> C: {session_id}server_write_key
In this last exchange, the response_data is a function of the auth_type.

2.3 Errors

Error handling in the SSL connection protocol is very simple. When an error is detected, the detecting party sends a message to the other party. Errors that are not recoverable cause the client and server to abort the secure connection. Servers and client are required to "forget" any session-identifiers associated with a failing connection.

The SSL Handshake Protocol defines the following errors:

NO-CIPHER-ERROR
This error is returned by the client to the server when it cannot find a cipher or key size that it supports that is also supported by the server. This error is not recoverable.

NO-CERTIFICATE-ERROR
When a REQUEST-CERTIFICATE message is sent, this error may be returned if the receiving party has no certificate to reply with. This error is recoverable.

BAD-CERTIFICATE-ERROR
This error is returned when a certificate is deemed bad by the receiving party. Bad means that either the signature of the certificate was bad or that the values in the certificate were inappropriate (e.g. a name in the certificate did not match the expected name). This error is recoverable.

UNSUPPORTED-CERTIFICATE-TYPE-ERROR
This error is returned when a client/server receives a certificate type that it can't support. This error is recoverable.

2.4 SSL Handshake Protocol Messages

The SSL Handshake Protocol messages are encapsulated in the SSL Record Protocol and are composed of two parts: a single byte message type code, and some data. The client and server exchange messages until both ends have sent their "finished" message, indicating that they are satisfied with the SSL Handshake Protocol conversation. While one end may be finished, the other may not, therefore the finished end must continue to receive SSL Handshake Protocol messages until it too receives a "finished" message.

After the pair of session keys has been determined by each party, the message bodies are encrypted using it. For the client, this happens after it verifies the session-identifier or creates a new session key and has sent it to the server. For the server, this happens after the session-identifier is found to be good, or the server receives the client's session key message.

The following notation is used for SSLHP messages:

    char MSG-EXAMPLE
    char FIELD1
    char FIELD2
    char THING-MSB
    char THING-LSB
    char THING-DATA[(MSB<<8)|LSB];
    ...
This notation defines the data in the protocol message, including the message type code. The order is presented top to bottom, with the top most element being transmitted first, and the bottom most element transferred last.

For the "THING-DATA" entry, the MSB and LSB values are actually THING-MSB and THING-LSB (respectively) and define the number of bytes of data actually present in the message. For example, if THING-MSB were zero and THING-LSB were 8 then the THING-DATA array would be exactly 8 bytes long. This shorthand is used below.

Length codes are unsigned values, and when the MSB and LSB are combined the result is an unsigned value. Unless otherwise specified lengths values are "length in bytes".

2.5 Client Only Protocol Messages

There are several messages that are only generated by clients. These messages are never generated by correctly functioning servers. A client receiving such a message closes the connection to the server and returns an error status to the application through some unspecified mechanism.

CLIENT-HELLO (Phase 1; Sent in the clear)

    char MSG-CLIENT-HELLO
    char CLIENT-VERSION-MSB
    char CLIENT-VERSION-LSB
    char CIPHER-SPECS-LENGTH-MSB
    char CIPHER-SPECS-LENGTH-LSB
    char SESSION-ID-LENGTH-MSB
    char SESSION-ID-LENGTH-LSB
    char CHALLENGE-LENGTH-MSB
    char CHALLENGE-LENGTH-LSB
    char CIPHER-SPECS-DATA[(MSB<<8)|LSB]
    char SESSION-ID-DATA[(MSB<<8)|LSB]
    char CHALLENGE-DATA[(MSB<<8)|LSB]
When a client first connects to a server it is required to send the CLIENT-HELLO message. The server is expecting this message from the client as its first message. It is an error for a client to send anything else as its first message.

The client sends to the server its SSL version, its cipher specs (see below), some challenge data, and the session-identifier data. The session-identifier data is only sent if the client found a session-identifier in its cache for the server, and the SESSION-ID-LENGTH will be non-zero. When there is no session-identifier for the server SESSION-ID-LENGTH must be zero. The challenge data is used to authenticate the server. After the client and server agree on a pair of session keys, the server returns a SERVER-VERIFY message with the encrypted form of the CHALLENGE-DATA.

Also note that the server will not send its SERVER-HELLO message until it has received the CLIENT-HELLO message. This is done so that the server can indicate the status of the client's session-identifier back to the client in the servers first message (i.e. to increase protocol efficiency and reduce the number of round trips required).

The server examines the CLIENT-HELLO message and will verify that it can support the client version and one of the client cipher specs. The server can optionally edit the cipher specs, removing any entries it doesn't choose to support. The edited version will be returned in the SERVER-HELLO message if the session-identifier is not in the servers cache.

The CIPHER-SPECS-LENGTH must be greater than zero and a multiple of 3. The SESSION-ID-LENGTH must either be zero or 16. The CHALLENGE-LENGTH must be greater than or equal to 16 and less than or equal to 32.

This message must be the first message sent by the client to the server. After the message is sent the client waits for a SERVER-HELLO message. Any other message returned by the server (other than ERROR) is disallowed.

CLIENT-MASTER-KEY (Phase 1; Sent primarily in the clear)

    char MSG-CLIENT-MASTER-KEY
    char CIPHER-CHOICE
    char KEY-LENGTH-MSB
    char KEY-LENGTH-LSB
    char CLEAR-KEY-LENGTH-MSB
    char CLEAR-KEY-LENGTH-LSB
    char ENCRYPTED-KEY-LENGTH-MSB
    char ENCRYPTED-KEY-LENGTH-LSB
    char KEY-ARG-LENGTH-MSB
    char KEY-ARG-LENGTH-LSB
    char CLEAR-KEY-DATA[MSB<<8|LSB]
    char ENCRYPTED-KEY-DATA[MSB<<8|LSB]
    char KEY-ARG-DATA[MSB<<8|LSB]
The client sends this message when it has determined a master key for the server to use. Note that when a session-identifier has been agreed upon, this message is not sent.

The CIPHER-CHOICE field indicates which cipher was chosen from the server's CIPHER-SPECS. The value is an index into the CIPHER-SPECS array sent by the server to the client in the SERVER-HELLO message (recall that each element of the array is a 3 byte value known as a CIPHER-KIND). The KEY-LENGTH fields (MSB and LSB) when combined yield the actual size of the key to use (in bits). KEY-LENGTH must be a multiple of 8. The server must verify that the CIPHER-CHOICE value is legal, and if not send an ERROR and close the connection.

The CLEAR-KEY-DATA contains the clear portion of the MASTER-KEY. The ENCRYPTED-KEY-DATA contains the secret portions of the MASTER-KEY, encrypted using the servers public key. The encryption block is formatted using block type 2 from PKCS#1 [5]. The data portion of the block is formatted as follows:

    char MASTER-KEY[SECRET-LENGTH]
SECRET-LENGTH is the number of bytes of each session key that is being transmitted encrypted. It is derived from the CIPHER-CHOICE value and can be computed by subtracting the CLEAR-KEY-LENGTH * 8 from the KEY-LENGTH. It is an error if the difference does not equal the value defined by the CIPHER-CHOICE. It is also an error if the KEY-LENGTH doesn't match the length specified by the CIPHER-CHOICE. Finally, it is an error if CLEAR-KEY-LENGTH is non-zero and the CIPHER-CHOICE specifies a a CIPHER-KIND that is not an export cipher.

If the key algorithm needs an argument (for example, DES's initialization vector) then the KEY-ARG-LENGTH fields will be non-zero and the KEY-ARG-DATA will contain the relevant data.

The client produces its session keys as follows:

CLIENT-READ-KEY = HASH[ MASTER-KEY, "0", CHALLENGE, CONNECTION-ID ]
CLIENT-WRITE-KEY = HASH[ MASTER-KEY, "1", CHALLENGE, CONNECTION-ID ]
Where the HASH function is chosen from the CIPHER-CHOICE. Conversion of the hash value into a key is cipher type specific. The following rules are used:

SSL_CK_RC4_WITH_MD5, SSL_CK_RC4_EXPORT40_WITH_MD5
The MD5 digest is used as a 128 bit key.
SSL_CK_RC2_CBC_WITH_MD5, SSL_CK_RC2_CBC_EXPORT40_WITH_MD5
The MD5 digest is used as a 128 bit key.
SSL_CK_IDEA_CBC_WITH_MD5
The MD5 digest is used as a 128 bit key.

The CLIENT-MASTER-KEY message must be sent after the CLIENT-HELLO message and before the CLIENT-FINISHED message. The CLIENT-MASTER-KEY message must be sent if the SERVER-HELLO message contains a SESSION-ID-HIT value of 0.

CLIENT-CERTIFICATE (Phase 2; Sent encrypted)

    char MSG-CLIENT-CERTIFICATE
    char CERTIFICATE-TYPE
    char CERTIFICATE-LENGTH-MSB
    char CERTIFICATE-LENGTH-LSB
    char RESPONSE-LENGTH-MSB
    char RESPONSE-LENGTH-LSB
    char CERTIFICATE-DATA[MSB<<8|LSB]
    char RESPONSE-DATA[MSB<<8|LSB]
This message is sent by one an SSL client in response to a server REQUEST-CERTIFICATE message. The CERTIFICATE-DATA contains data defined by the CERTIFICATE-TYPE value. An ERROR message is sent with error code NO-CERTIFICATE-ERROR when this request cannot be answered properly (e.g. the receiver of the message has no registered certificate).

CERTIFICATE-TYPE is one of:

SSL_X509_CERTIFICATE
The CERTIFICATE-DATA contains an X.509 (1988) [3] signed certificate.

The RESPONSE-DATA contains the authentication response data. This data is a function of the AUTHENTICATION-TYPE value sent by the server.

When AUTHENTICATION-TYPE is SSL_AT_MD5_WITH_RSA_ENCRYPTION then the RESPONSE-DATA contains a digital signature of the following components (in the order shown):

  • the CLIENT-READ-KEY
  • the CLIENT-WRITE-KEY
  • the CERTIFICATE-CHALLENGE-DATA (from the REQUEST-CERTIFICATE message)
  • the server's signed certificate (from the SERVER-HELLO message)

The digital signature is constructed using MD5 and then encrypted using the clients private key, formatted according to PKCS#1's digital signature standard [5]. The server authenticates the client by verifying the digital signature using standard techniques. Note that other digest functions are supported. Either a new AUTHENTICATION-TYPE can be added, or the algorithm-id in the digital signature can be changed.

This message must be sent by the client only in response to a REQUEST-CERTIFICATE message.

CLIENT-FINISHED (Phase 2; Sent encrypted)

    char MSG-CLIENT-FINISHED
    char CONNECTION-ID[N-1]
The client sends this message when it is satisfied with the server. Note that the client must continue to listen for server messages until it receives a SERVER-FINISHED message. The CONNECTION-ID data is the original connection-identifier the server sent with its SERVER-HELLO message, encrypted using the agreed upon session key.

"N" is the number of bytes in the message that was sent, so "N-1" is the number of bytes in the message without the message header byte.

For version 2 of the protocol, the client must send this message after it has received the SERVER-HELLO message. If the SERVER-HELLO message SESSION-ID-HIT flag is non-zero then the CLIENT-FINISHED message is sent immediately, otherwise the CLIENT-FINISHED message is sent after the CLIENT-MASTER-KEY message.

2.6 Server Only Protocol Messages

There are several messages that are only generated by servers. The messages are never generated by correctly functioning clients.

SERVER-HELLO (Phase 1; Sent in the clear)

    char MSG-SERVER-HELLO
    char SESSION-ID-HIT
    char CERTIFICATE-TYPE
    char SERVER-VERSION-MSB
    char SERVER-VERSION-LSB
    char CERTIFICATE-LENGTH-MSB
    char CERTIFICATE-LENGTH-LSB
    char CIPHER-SPECS-LENGTH-MSB
    char CIPHER-SPECS-LENGTH-LSB
    char CONNECTION-ID-LENGTH-MSB
    char CONNECTION-ID-LENGTH-LSB
    char CERTIFICATE-DATA[MSB<<8|LSB]
    char CIPHER-SPECS-DATA[MSB<<8|LSB]
    char CONNECTION-ID-DATA[MSB<<8|LSB]
The server sends this message after receiving the clients CLIENT-HELLO message. The server returns the SESSION-ID-HIT flag indicating whether or not the received session-identifier is known by the server (i.e. in the servers session-identifier cache). The SESSION-ID-HIT flag will be non-zero if the client sent the server a session-identifier (in the CLIENT-HELLO message with SESSION-ID-LENGTH != 0) and the server found the client's session-identifier in its cache. If the SESSION-ID-HIT flag is non-zero then the CERTIFICATE-TYPE, CERTIFICATE-LENGTH and CIPHER-SPECS-LENGTH fields will be zero.

The CERTIFICATE-TYPE value, when non-zero, has one of the values described above (see the information on the CLIENT-CERTIFICATE message).

When the SESSION-ID-HIT flag is zero, the server packages up its certificate, its cipher specs and a connection-id to send to the client. Using this information the client can generate a session key and return it to the server with the CLIENT-MASTER-KEY message.

When the SESSION-ID-HIT flag is non-zero, both the server and the client compute a new pair of session keys for the current session derived from the MASTER-KEY that was exchanged when the SESSION-ID was created. The SERVER-READ-KEY and SERVER-WRITE-KEY are derived from the original MASTER-KEY keys as follows (the hash function used is chosen from the CIPHER-CHOICE):

SERVER-READ-KEY = HASH[ MASTER-KEY, "1", CHALLENGE, CONNECTION-ID ]
SERVER-WRITE-KEY = HASH[ MASTER-KEY, "0", CHALLENGE, CONNECTION-ID ]
Recall that the MASTER-KEY is given to the server in the CLIENT-MASTER-KEY message. The CHALLENGE is given to the server by the client in the CLIENT-HELLO message. The CONNECTION-ID is given to the client by the server in the SERVER-HELLO message. This makes the resulting cipher keys a function of the original session and the current session. Note that the master key is never directly used to encrypt data, and therefore cannot be easily disovered.

The CONNECTION-ID-DATA is a string of randomly generated bytes used by the server and client at various points in the protocol. The CLIENT-FINISHED message contains an encrypted version of the CONNECTION-ID-DATA.

The CIPHER-SPECS-DATA define a cipher type and key length (in bits) that the receiving end supports. Each SESSION-CIPHER-SPEC is 3 bytes long and looks like this:

    char CIPHER-KIND
    char KEY-BITS-MSB
    char KEY-BITS-LSB
Where CIPHER-KIND is one of:

  • SSL_CK_RC4_WITH_MD5
  • SSL_CK_RC4_EXPORT40_WITH_MD5
  • SSL_CK_RC2_CBC_WITH_MD5
  • SSL_CK_RC2_CBC_EXPORT40_WITH_MD5
  • SSL_CK_IDEA_CBC_WITH_MD5

This list is not exhaustive and may be changed in the future.

The SSL_CK_RC4_EXPORT40_WITH_MD5 cipher is an RC4 cipher where some of the session key is sent in the clear and the rest is sent encrypted (exactly 40 bits of it). MD5 is used as the hash function for production of MAC's and session key's. This cipher type is provided to support "export" versions (i.e. versions of the protocol that can be distributed outside of the United States) of the client or server.

An exportable implementation of the SSL Handshake Protocol will have secret key lengths restricted to 40 bits. For non-export implementations key lengths can be more generous (we recommend the full 128 bits). It is permissible for the client and server to have a non-intersecting set of stream ciphers. This, simply put, means they cannot communicate.

Version 2 of the SSL Handshake Protocol defines the SSL_CK_RC4_WITH_MD5 to have a key length of 128 bits. The SSL_CK_RC4_EXPORT40_WITH_MD5 also has a key length of 128 bits. However, only 40 of the bits are secret (the other 88 bits are sent in the clear by the client to the server).

The SERVER-HELLO message is sent after the server receives the CLIENT-HELLO message, and before the server sends the SERVER-VERIFY message.

SERVER-VERIFY (Phase 1; Sent encrypted)

    char MSG-SERVER-VERIFY
    char CHALLENGE-DATA[N-1]
The server sends this message after a pair of session keys (SERVER-READ-KEY and SERVER-WRITE-KEY) have been agreed upon either by a session-identifier or by explicit specification with the CLIENT-MASTER-KEY message. The message contains an encrypted copy of the CHALLENGE-DATA sent by the client in the CLIENT-HELLO message.

"N" is the number of bytes in the message that was sent, so "N-1" is the number of bytes in the CHALLENGE-DATA without the message header byte.

This message is used to verify the server as follows. A legitimate server will have the private key that corresponds to the public key contained in the server certificate that was transmitted in the SERVER-HELLO message. Accordingly, the legitimate server will be able to extract and reconstruct the pair of session keys (SERVER-READ-KEY and SERVER-WRITE-KEY). Finally, only a server that has done the extraction and decryption properly can correctly encrypt the CHALLENGE-DATA. This, in essence, "proves" that the server has the private key that goes with the public key in the servers certificate.

The CHALLENGE-DATA must be the exact same length as originally sent by the client in the CLIENT-HELLO message. Its value must match exactly the value sent in the clear by the client in the CLIENT-HELLO message. The client must decrypt this message and compare the value received with the value sent, and only if the values are identical is the server to be "trusted". If the lengths do not match or the value doesn't match then the connection is to be closed by the client.

This message must be sent by the server to the client after either detecting a session-identifier hit (and replying with a SERVER-HELLO message with SESSION-ID-HIT not equal to zero) or when the server receives the CLIENT-MASTER-KEY message. This message must be sent before any Phase 2 messages or a SEVER-FINISHED message.

SERVER-FINISHED (Phase 2; Sent encrypted)

    char MSG-SERVER-FINISHED
    char SESSION-ID-DATA[N-1]
The server sends this message when it is satisfied with the clients security handshake and is ready to proceed with transmission/reception of the higher level protocols data. The SESSION-ID-DATA is used by the client and the server at this time to add entries to their respective session-identifier caches. The session-identifier caches must contain a copy of the MASTER-KEY sent in the CLIENT-MASTER-KEY message as the master key is used for all subsequent session key generation.

"N" is the number of bytes in the message that was sent, so "N-1" is the number of bytes in the SESSION-ID-DATA without the message header byte.

This message must be sent after the SERVER-VERIFY message.

REQUEST-CERTIFICATE (Phase 2; Sent encrypted)

    char MSG-REQUEST-CERTIFICATE
    char AUTHENTICATION-TYPE
    char CERTIFICATE-CHALLENGE-DATA[N-2]
A server may issue this request at any time during the second phase of the connection handshake, asking for the client's certificate. The client responds with a CLIENT-CERTIFICATE message immediately if it has one, or an ERROR message (with error code NO-CERTIFICATE-ERROR) if it doesn't. The CERTIFICATE-CHALLENGE-DATA is a short byte string (whose length is greater than or equal to 16 bytes and less than or equal to 32 bytes) that the client will use to respond to this message.

The AUTHENTICATION-TYPE value is used to choose a particular means of authenticating the client. The following types are defined:

  • SSL_AT_MD5_WITH_RSA_ENCRYPTION

The SSL_AT_MD5_WITH_RSA_ENCRYPTION type requires that the client construct an MD5 message digest using information as described above in the section on the CLIENT-CERTIFICATE message. Once the digest is created, the client encrypts it using its private key (formatted according to the digital signature standard defined in PKCS#1). The server authenticates the client when it receives the CLIENT-CERTIFICATE message.

This message may be sent after a SERVER-VERIFY message and before a SERVER-FINISHED message.

2.7 Client/Server Protocol Messages

These messages are generated by both the client and the server.

ERROR (Sent clear or encrypted)

    char MSG-ERROR
    char ERROR-CODE-MSB
    char ERROR-CODE-LSB
This message is sent when an error is detected. After the message is sent, the sending party shuts the connection down. The receiving party records the error and then shuts its connection down.

This message is sent in the clear if an error occurs during session key negotiation. After a session key has been agreed upon, errors are sent encrypted like all other messages.


Appendix A: ASN.1 Syntax For Certificates

Certificates are used by SSL to authenticate servers and clients. SSL Certificates are based largely on the X.509 [3] certificates. An X.509 certificate contains the following information (in ASN.1 [1] notation):

X.509-Certificate ::= SEQUENCE {
    certificateInfo CertificateInfo,
    signatureAlgorithm AlgorithmIdentifier,
    signature BIT STRING
}

CertificateInfo ::= SEQUENCE {
    version [0] Version DEFAULT v1988,
    serialNumber CertificateSerialNumber,
    signature AlgorithmIdentifier,
    issuer Name,
    validity Validity,
    subject Name,
    subjectPublicKeyInfo SubjectPublicKeyInfo
}

Version ::= INTEGER { v1988(0) }

CertificateSerialNumber ::= INTEGER

Validity ::= SEQUENCE {
    notBefore UTCTime,
    notAfter UTCTime
}

SubjectPublicKeyInfo ::= SEQUENCE {
    algorithm AlgorithmIdentifier,
    subjectPublicKey BIT STRING
}

AlgorithmIdentifier ::= SEQUENCE {
    algorithm OBJECT IDENTIFIER,
    parameters ANY DEFINED BY ALGORITHM OPTIONAL
}
For SSL's purposes we restrict the values of some of the X.509 fields:

Certificates are validated using a few straightforward steps. First, the signature on the certificate is checked and if invalid, the certificate is invalid (either a transmission error or an attempted forgery occurred). Next, the CertificateInfo::issuer field is verified to be an issuer that the application trusts (using an unspecified mechanism). The CertificateInfo::validity field is checked against the current date and verified.

Finally, the CertificateInfo::subject field is checked. This check is optional and depends on the level of trust required by the application using SSL.

Appendix B: Attribute Types and Object Identifiers

SSL uses a subset of the X.520 selected attribute types as well as a few specific object identifiers. Future revisions of the SSL protocol may include support for more attribute types and more object identifiers.

B.1 Selected attribute types

commonName { attributeType 3 }
The common name contained in the distinguished name contained within a certificate issuer or certificate subject.

countryName { attributeType 6 }
The country name contained in the distinguished name contained within a certificate issuer or certificate subject.

localityName { attributeType 7 }
The locality name contained in the distinguished name contained within a certificate issuer or certificate subject.

stateOrProvinceName { attributeType 8 }
The state or province name contained in the distinguished name contained within a certificate issuer or certificate subject.

organizationName { attributeType 10 }
The organization name contained in the distinguished name contained within a certificate issuer or certificate subject.

organizationalUnitName { attributeType 11 }
The organizational unit name contained in the distinguished name contained within a certificate issuer or certificate subject.

B.2 Object identifiers

md2withRSAEncryption { ... pkcs(1) 1 2 }
The object identifier for digital signatures that use both MD2 and RSA encryption. Used by SSL for certificate signature verification.

md5withRSAEncryption { ... pkcs(1) 1 4 }
The object identifier for digital signatures that use both MD5 and RSA encryption. Used by SSL for certificate signature verification.

rc4 { ... rsadsi(113549) 3 4 }
The RC4 symmetric stream cipher algorithm used by SSL for bulk encryption.

Appendix C: Protocol Constant Values

This section describes various protocol constants. A special value needs mentioning - the IANA reserved port number for "https" (HTTP using SSL). IANA has reserved port number 443 (decimal) for "https".

C.1 Protocol Version Codes

#define SSL_CLIENT_VERSION			0x0002
#define SSL_SERVER_VERSION			0x0002

C.2 Protocol Message Codes

The following values define the message codes that are used by version 2 of the SSL Handshake Protocol.

#define SSL_MT_ERROR				0
#define SSL_MT_CLIENT_HELLO			1
#define SSL_MT_CLIENT_MASTER_KEY		2
#define SSL_MT_CLIENT_FINISHED			3
#define SSL_MT_SERVER_HELLO			4
#define SSL_MT_SERVER_VERIFY			5
#define SSL_MT_SERVER_FINISHED			6
#define SSL_MT_REQUEST_CERTIFICATE		7
#define SSL_MT_CLIENT_CERTIFICATE		8

C.3 Error Message Codes

The following values define the error codes used by the ERROR message.

#define SSL_PE_NO_CIPHER			0x0001
#define SSL_PE_NO_CERTIFICATE			0x0002
#define SSL_PE_BAD_CERTIFICATE			0x0004
#define SSL_PE_UNSUPPORTED_CERTIFICATE_TYPE	0x0006

C.4 Cipher Kind Values

The following values define the CIPHER-KIND codes used in the CLIENT-HELLO and SERVER-HELLO messages.

#define SSL_CK_RC4_WITH_MD5			0x01
#define SSL_CK_RC4_EXPORT40_WITH_MD5		0x02
#define SSL_CK_RC2_CBC_WITH_MD5			0x03
#define SSL_CK_RC2_CBC_EXPORT40_WITH_MD5	0x04
#define SSL_CK_IDEA_CBC_WITH_MD5		0x05

C.5 Certificate Type Codes

The following values define the certificate type codes used in the SERVER-HELLO and CLIENT-CERTIFICATE messages.

#define SSL_CT_X509_CERTIFICATE			0x01

C.6 Authentication Type Codes

The following values define the authentication type codes used in the REQUEST-CERTIFICATE message.

#define SSL_AT_MD5_WITH_RSA_ENCRYPTION		0x01

C.7 Upper/Lower Bounds

The following values define upper/lower bounds for various protocol parameters.

#define SSL_MAX_MASTER_KEY_LENGTH_IN_BITS	128
#define SSL_MAX_SESSION_ID_LENGTH_IN_BYTES	16
#define SSL_MIN_RSA_MODULUS_LENGTH_IN_BYTES	64
#define SSL_MAX_RECORD_LENGTH_2_BYTE_HEADER	32767
#define SSL_MAX_RECORD_LENGTH_3_BYTE_HEADER	16383

C.8 Recommendations

Because protocols have to be implemented to be of value, we recommend the following values for various operational parameters. This is only a recommendation, and not a strict requirement for conformance to the protocol.

Session-identifier Cache Timeout

Session-identifiers are kept in SSL clients and SSL servers. Session-identifiers should have a lifetime that serves their purpose (namely, reducing the number of expensive public key operations for a single client/server pairing). Consequently, we recommend a maximum session-identifier cache timeout value of 100 seconds. Given a server that can perform N private key operations per second, this reduces the server load for a particular client by a factor of 100.

Appendix D: Attacks

In this section we attempt to describe various attacks that might be used against the SSL protocol. This list is not guaranteed to be exhaustive. SSL was defined to thwart these attacks.

D.1 Cracking Ciphers

SSL depends on several cryptographic technologies. RSA Public Key encryption [5] is used for the exchange of the session key and client/server authentication. Various cryptographic algorithms are used for the session cipher. If successful cryptographic attacks are made against these technologies then SSL is no longer secure.

Attacks against a specific communications session can be made by recording the session, and then spending some large number of compute cycles to crack either the session key or the RSA public key until the communication can be seen in the clear. This is easier than cracking the cryptographic technologies for all cases, but it is still possible. Note that SSL tries to make the cost of such of an attack greater than the benefits gained from a successful attack, thus making it a waste of money/time to perform such an attack.

There have been many books [9] and papers [10] written on cryptography. This document does not attempt to reference them all.

D.2 Clear Text Attack

A clear text attack is done when the attacker has an idea of what kind of message is being sent using encryption. The attacker can generate a data base whose keys are the encrypted value of the known text (or clear text), and whose values are the session cipher key (we call this a "dictionary"). Once this data base is constructed, a simple lookup function identifies the session key that goes with a particular encrypted value. Once the session key is known, the entire message stream can be decrypted. Custom hardware can be used to make this cost effective and very fast.

Because of the very nature of SSL clear text attacks are possible. For example, the most common byte string sent by an HTTP client application to an HTTP server is "GET". SSL attempts to address this attack by using large session cipher keys. First, the client generates a key which is larger than allowed by export, and sends some of it in the clear to the server (this is allowed by United States government export rules). The clear portion of the key concatenated with the secret portion make a key which is very large (for RC4, exactly 128 bits).

That way that this "defeats" a clear text attack is by making the amount of custom hardware needed prohibitively large. Every bit added to the length of the session cipher key increases the dictionary size by a factor of 2. By using a 128 bit session cipher key length the size of the dictionary required is beyond the ability of anyone to fabricate (it would require more atoms to construct than exist in the entire universe). Even if a smaller dictionary is to be used, it must first be generated using the clear key bits. This is a time consumptive process and also eliminates many possible custom hardware architectures (e.g. static prom arrays).

The second way that SSL attacks this problem is by using large key lengths when permissible (e.g. in the non-export version). Large key sizes require larger dictionaries (just one more bit of key size doubles the size of the dictionary). SSL attempts to use keys that are 128 bits in length.

Note that the consequence of the SSL defense is that a brute force attack becomes the cheapest way to attack the key. Brute force attacks have well known space/time tradeoffs and so it becomes possible to define a cost of the attack. For the 128 bit secret key, the known cost is essentially infinite. For the 40 bit secret key, the cost is much smaller, but still outside the range of the "random hacker".

D.3 Replay

The replay attack is simple. A bad-guy records a communication session between a client and server. Later, it reconnects to the server, and plays back the previously recorded client messages.

SSL defeats this attack using a "nonce" (the connection-id) which is "unique" to the connection. In theory the bad-guy cannot predict the nonce in advance as it is based on a set of random events outside the bad-guys control, and therefore the bad-guy cannot respond properly to server requests.

A bad-guy with large resources can record many sessions between a client and a server, and attempt to choose the right session based on the nonce the server sends initially in its SERVER-HELLO message. However, SSL nonces are at least 128 bits long, so a bad-guy would need to record approximately 2^64 nonces to even have a 50% chance of choosing the right session. This number is sufficiently large that one cannot economically construct a device to record 2^64 messages, and therefore the odds are overwhelmingly against the replay attack ever being successful.

D.4 The Man In The Middle

The man in the middle attack works by having three people in a communications session: the client, the server, and the bad guy. The bad guy sits between the client and the server on the network and intercepts traffic that the client sends to the server, and traffic that the server sends to the client.

The man in the middle operates by pretending to be the real server to the client. With SSL this attack is impossible because of the usage of server certificates. During the security connection handshake the server is required to provide a certificate that is signed by a certificate authority. Contained in the certificate is the server's public key as well as its name and the name of the certificate issuer. The client verifies the certificate by first checking the signature and then verifying that the name of the issuer is somebody that the client trusts.

In addition, the server must encrypt something with the private key that goes with the public key mentioned in the certificate. This in essence is a single pass "challenge response" mechanism. Only a server that has both the certificate and the private key can respond properly to the challenge.

If the man in the middle provides a phony certificate, then the signature check will fail. If the certificate provided by the bad guy is legitimate, but for the bad guy instead of for the real server, then the signature will pass but the name check will fail (note that the man in the middle cannot forge certificates without discovering a certificate authority's private key).

Finally, if the bad guy provides the real server's certificate then the signature check will pass and the name check will pass. However, because the bad guy does not have the real server's private key, the bad guy cannot properly encode the response to the challenge code, and this check will fail.

In the unlikely case that a bad guy happens to guess the response code to the challenge, the bad guy still cannot decrypt the session key and therefore cannot examine the encrypted data.

Appendix E: Terms

Application Protocol
An application protocol is a protocol that normally layers directly on top of TCP/IP. For example: HTTP, TELNET, FTP, and SMTP.

Authentication
Authentication is the ability of one entity to determine the identity of another entity. Identity is defined by this document to mean the binding between a public key and a name and the implicit ownership of the corresponding private key.

Bulk Cipher
This term is used to describe a cryptographic technique with certain performance properties. Bulk ciphers are used when large quantities of data are to be encrypted/decrypted in a timely manner. Examples include RC2, RC4, and IDEA.

Client
In this document client refers to the application entity that is initiates a connection to a server.

CLIENT-READ-KEY
The session key that the client uses to initialize the client read cipher. This key has the same value as the SERVER-WRITE-KEY.

CLIENT-WRITE-KEY
The session key that the client uses to initialize the client write cipher. This key has the same value as the SERVER-READ-KEY.

MASTER-KEY
The master key that the client and server use for all session key generation. The CLIENT-READ-KEY, CLIENT-WRITE-KEY, SERVER-READ-KEY and SERVER-WRITE-KEY are generated from the MASTER-KEY.

MD2
MD2 [8] is a hashing function that converts an arbitrarily long data stream into a digest of fixed size. This function predates MD5 [7] which is viewed as a more robust hash function [9].

MD5
MD5 [7] is a hashing function that converts an arbitrarily long data stream into a digest of fixed size. The function has certain properties that make it useful for security, the most important of which is it's inability to be reversed.

Nonce
A randomly generated value used to defeat "playback" attacks. One party randomly generates a nonce and sends it to the other party. The receiver encrypts it using the agreed upon secret key and returns it to the sender. Because the nonce was randomly generated by the sender this defeats playback attacks because the replayer can't know in advance the nonce the sender will generate. The receiver denies connections that do not have the correctly encrypted nonce.

Non-repudiable Information Exchange
When two entities exchange information it is sometimes valuable to have a record of the communication that is non-repudiable. Neither party can then deny that the information exchange occurred. Version 2 of the SSL protocol does not support Non-repudiable information exchange.

Public Key Encryption
Public key encryption is a technique that leverages asymmetric ciphers. A public key system consists of two keys: a public key and a private key. Messages encrypted with the public key can only be decrypted with the associated private key. Conversely, messages encrypted with the private key can only be decrypted with the public key. Public key encryption tends to be extremely compute intensive and so is not suitable as a bulk cipher.

Privacy
Privacy is the ability of two entities to communicate without fear of eavesdropping. Privacy is often implemented by encrypting the communications stream between the two entities.

RC2, RC4
Proprietary bulk ciphers invented by RSA (There is no good reference to these as they are unpublished works; however, see [9]). RC2 is block cipher and RC4 is a stream cipher.

Server
The server is the application entity that responds to requests for connections from clients. The server is passive, waiting for requests from clients.

Session cipher
A session cipher is a "bulk" cipher that is capable of encrypting or decrypting arbitrarily large amounts of data. Session ciphers are used primarily for performance reasons. The session ciphers used by this protocol are symmetric. Symmetric ciphers have the property of using a single key for encryption and decryption.

Session identifier
A session identifier is a random value generated by a client that identifies itself to a particular server. The session identifier can be thought of as a handle that both parties use to access a recorded secret key (in our case a session key). If both parties remember the session identifier then the implication is that the secret key is already known and need not be negotiated.

Session key
The key to the session cipher. In SSL there are four keys that are called session keys: CLIENT-READ-KEY, CLIENT-WRITE-KEY, SERVER-READ-KEY, and SERVER-WRITE-KEY.

SERVER-READ-KEY
The session key that the server uses to initialize the server read cipher. This key has the same value as the CLIENT-WRITE-KEY.

SERVER-WRITE-KEY
The session key that the server uses to initialize the server write cipher. This key has the same value as the CLIENT-READ-KEY.

Symmetric Cipher
A symmetric cipher has the property that the same key can be used for decryption and encryption. An asymmetric cipher does not have this behavior. Some examples of symmetric ciphers: IDEA, RC2, RC4.

References

[1] CCITT. Recommendation X.208: "Specification of Abstract Syntax Notation One (ASN.1). 1988.

[2] CCITT. Recommendation X.209: "Specification of Basic Encoding Rules for Abstract Syntax Notation One (ASN.1). 1988.

[3] CCITT. Recommendation X.509: "The Directory - Authentication Framework". 1988.

[4] CCITT. Recommendation X.520: "The Directory - Selected Attribute Types". 1988.

[5] RSA Laboratories. PKCS #1: RSA Encryption Standard, Version 1.5, November 1993.

[6] RSA Laboratories. PKCS #6: Extended-Certificate Syntax Standard, Version 1.5, November 1993.

[7] R. Rivest. RFC 1321: The MD5 Message Digest Algorithm. April 1992.

[8] R. Rivest. RFC 1319: The MD2 Message Digest Algorithm. April 1992.

[9] B. Schneier. Applied Cryptography: Protocols, Algorithms, and Source Code in C, Published by John Wiley & Sons, Inc. 1994.

[10] M. Abadi and R. Needham. Prudent engineering practice for cryptographic protocols. 1994.

Patent Statement

This version of the SSL protocol relies on the use of patented public key encryption technology for authentication and encryption. The Internet Standards Process as defined in RFC 1310 requires a written statement from the Patent holder that a license will be made available to applicants under reasonable terms and conditions prior to approving a specification as a Proposed, Draft or Internet Standard.

The Massachusetts Institute of Technology and the Board of Trustees of the Leland Stanford Junior University have granted Public Key Partners (PKP) exclusive sub-licensing rights to the following patents issued in the United States, and all of their corresponding foreign patents:

Cryptographic Apparatus and Method
("Diffie-Hellman")............................... No. 4,200,770 

Public Key Cryptographic Apparatus and Method ("Hellman-Merkle").................... No. 4,218,582

Cryptographic Communications System and Method ("RSA")................................... No. 4,405,829

Exponential Cryptographic Apparatus and Method ("Hellman-Pohlig").................... No. 4,424,414

These patents are stated by PKP to cover all known methods of practicing the art of Public Key encryption, including the variations collectively known as El Gamal.

Public Key Partners has provided written assurance to the Internet Society that parties will be able to obtain, under reasonable, nondiscriminatory terms, the right to use the technology covered by these patents. This assurance is documented in RFC 1170 titled "Public Key Standards and Licenses". A copy of the written assurance dated April 20, 1990, may be obtained from the Internet Assigned Number Authority (IANA).

The Internet Society, Internet Architecture Board, Internet Engineering Steering Group and the Corporation for National Research Initiatives take no position on the validity or scope of the patents and patent applications, nor on the appropriateness of the terms of the assurance. The Internet Society and other groups mentioned above have not made any determination as to any other intellectual property rights which may apply to the practice of this standard. Any further consideration of these matters is the user's own responsibility.

Security Considerations

This entire document is about security.

Author's Address

Kipp E.B. Hickman
Netscape Communications Corp.
650 Castro St.
Mountain View, CA 94024
kipp@mcom.com