RFC 8120
Mutual Authentication Protocol for HTTP
Pages: 53
Experimental
Experimental
Part 2 of 3 – Pages 15 to 35
4. Messages
In this section, we define the six kinds of messages in the authentication protocol, along with the formats and requirements of the headers for each type of message. To determine under what circumstances each message is expected to be sent, see Sections 10 and 11. In the descriptions below, the types of allowable values for each header parameter are shown in parentheses after each parameter name. The "algorithm-determined" type means that the acceptable value for the parameter is one of the types defined in Section 3 and is
determined by the value of the "algorithm" parameter. The parameters
marked "mandatory" SHALL be contained in the message. The parameters
marked "non-mandatory" MAY be either contained in the message or
omitted from it. Each parameter SHALL appear in each header exactly
once at most.
All credentials and challenges MAY contain any parameters not
explicitly specified in the following sections. Recipients that
do not understand such parameters MUST silently ignore them.
However, all credentials and challenges MUST meet the following
criteria:
o For responses, the parameters "reason", any "ks#" (where "#"
stands for any decimal integer), and "vks" are mutually exclusive;
any challenges MUST NOT contain two or more parameters among them.
They MUST NOT contain any "kc#" or "vkc" parameters.
o For requests, the parameters "kc#" (where "#" stands for any
decimal integer) and "vkc" are mutually exclusive; any challenges
MUST NOT contain two or more parameters among them. They MUST NOT
contain any "ks#" or "vks" parameters.
Every message defined in this section contains a "version" field to
detect any future revisions of the protocol that are incompatible.
Implementations of the protocol described in this specification MUST
always send a token "1" to represent the version number. Recipients
MUST reject messages that contain any other value for the version,
unless another specification defines specific behavior for that
version.
4.1. 401-INIT and 401-STALE
Every 401-INIT or 401-STALE message SHALL be a valid HTTP 401
(Unauthorized) status message (or some other 4xx status message, if
appropriate) containing one and only one (hereafter not explicitly
noted) WWW-Authenticate header containing a "reason" parameter in the
challenge. The challenge SHALL contain all of the parameters marked
"mandatory" below and MAY contain those marked "non-mandatory".
version:
(mandatory extensive-token) should be the token "1".
algorithm:
(mandatory extensive-token) specifies the authentication algorithm
to be used. The value MUST be one of the tokens specified in
[RFC8121] or another supplemental specification.
validation:
(mandatory extensive-token) specifies the method of host
validation. The value MUST be one of the tokens described in
Section 7 or the tokens specified in another supplemental
specification.
auth-scope:
(non-mandatory string) specifies the authentication scope, i.e.,
the set of hosts for which the authentication credentials are
valid. It MUST be one of the strings described in Section 5. If
the value is omitted, it is assumed to be the "single-server type"
domain as described in Section 5.
realm:
(mandatory string) is a string representing the name of the
authentication realm inside the authentication scope. As
specified in [RFC7235], this value MUST always be sent in the
quoted-string form, and an encoding as specified in [RFC5987]
MUST NOT be used.
The realm value sent from the server SHOULD be an ASCII string.
Clients MAY treat any non-ASCII value received in this field as a
binary blob, an NFC-normalized UTF-8 string ("NFC" stands for
"Normalization Form C"), or an error.
reason:
(mandatory extensive-token) SHALL be an extensive-token that
describes the possible reason for the failed authentication or
authorization. Both servers and clients SHALL understand and
support the following three tokens:
* initial: Authentication was not attempted because there was no
Authorization header in the corresponding request.
* stale-session: The provided sid in the request was either
unknown to the server or expired in the server.
* auth-failed: The authentication trial failed for some reason,
possibly because of a bad authentication credential.
Implementations MAY support the following tokens or any
extensive-tokens defined outside of this specification. If
clients receive any unknown tokens, they SHOULD treat them as if
they were "auth-failed" or "initial".
* reauth-needed: The server-side application requires a new
authentication trial, regardless of the current status.
* invalid-parameters: The server did not attempt authentication
because some parameters were not acceptable.
* internal-error: The server did not attempt authentication
because there are some problems on the server side.
* user-unknown: This is a special case of auth-failed; it
suggests that the provided username is invalid. Due to
security implications, the use of this parameter is
NOT RECOMMENDED, except for special-purpose applications where
it would make sense to do so.
* invalid-credential: This is another special case of
auth-failed; it suggests that the provided username was valid
but authentication still failed. For security reasons, the use
of this parameter is NOT RECOMMENDED.
* authz-failed: Authentication was successful, but access to the
specified resource is not authorized to the specific
authenticated user. (It might be used along with either a
401 (Unauthorized) or 403 (Forbidden) status code to indicate
that the authentication result is one of the existing reasons
for the failed authorization.)
It is RECOMMENDED that the reason for failure be recorded to some
type of diagnostic log, shown to the client user immediately, or
both. It will be helpful to find out later whether the reason for
the failure is technical or caused by user error.
The algorithm specified in this header will determine the types
(among those defined in Section 3) and the values for K_c1, K_s1,
VK_c, and VK_s.
Among these messages, any messages with the "reason" parameter value
"stale-session" will be called "401-STALE" messages hereafter,
because these messages have a special meaning in the protocol flow.
Messages with any other "reason" parameters will be called "401-INIT"
messages.
4.2. req-KEX-C1
Every req-KEX-C1 message SHALL be a valid HTTP request message containing an Authorization header with a credential containing a "kc1" parameter. The credential SHALL contain the parameters with the following names: version: (mandatory, extensive-token) should be the token "1". algorithm, validation, auth-scope, realm: MUST be the same values as those received from the server. user: (mandatory, string) is the UTF-8 encoded name of the user. The string SHOULD be prepared according to the method presented in Section 9. kc1: (mandatory, algorithm-determined) is the client-side key exchange value K_c1, which is specified by the algorithm that is used.4.3. 401-KEX-S1
Every 401-KEX-S1 message SHALL be a valid HTTP 401 (Unauthorized) status response message containing a WWW-Authenticate header with a challenge containing a "ks1" parameter. The challenge SHALL contain the parameters with the following names: version: (mandatory, extensive-token) should be the token "1". algorithm, validation, auth-scope, realm: MUST be the same values as those received from the client. sid: (mandatory, hex-fixed-number) MUST be a session identifier, which is a random integer. The sid SHOULD have uniqueness of at least 80 bits or the square of the maximum estimated transactions concurrently available in the session table, whichever is larger. See Section 6 for more details. ks1: (mandatory, algorithm-determined) is the server-side key exchange value K_s1, which is specified by the algorithm.
nc-max:
(mandatory, integer) is the maximum value of nonce numbers that
the server accepts.
nc-window:
(mandatory, integer) is the number of available nonce number slots
that the server will accept. It is RECOMMENDED that the value of
the "nc-window" parameter be 128 or more.
time:
(mandatory, integer) represents the suggested time (in seconds)
that the client can reuse the session represented by the sid. It
is RECOMMENDED that the time be set to at least 60 (seconds).
However, the server is not required to guarantee that the session
represented by the sid will be available (e.g., alive, usable) for
the time specified in this parameter.
path:
(non-mandatory, string) specifies to which path in the URI space
the same authentication is expected to be applied. The value is a
space-separated list of URIs, in the same format as that specified
in the "domain" parameter [RFC7616] for Digest authentications.
All path elements contained in the "path" parameter MUST be inside
the specified auth-scope; if not, clients SHOULD ignore such
elements. For better performance, it is important that clients
recognize and use this parameter.
4.4. req-VFY-C
Every req-VFY-C message SHALL be a valid HTTP request message
containing an Authorization header with a credential containing a
"vkc" parameter.
The parameters contained in the header are as follows:
version:
(mandatory, extensive-token) should be the token "1".
algorithm, validation, auth-scope, realm:
MUST be the same values as those received from the server for the
session.
sid:
(mandatory, hex-fixed-number) MUST be one of the sid values that
was received from the server for the same authentication realm.
nc:
(mandatory, integer) is a nonce request number that is unique
among the requests sharing the same sid. The values of the nonce
numbers SHOULD satisfy the properties outlined in Section 6.
vkc:
(mandatory, algorithm-determined) is the client-side
authentication verification value VK_c, which is specified by the
algorithm.
4.5. 200-VFY-S
Every 200-VFY-S message SHALL be a valid HTTP message that does not
have a 401 (Unauthorized) status code and SHALL contain an
Authentication-Info header with a "vks" parameter.
The parameters contained in the header are as follows:
version:
(mandatory, extensive-token) should be the token "1".
sid:
(mandatory, hex-fixed-number) MUST be the value received from the
client.
vks:
(mandatory, algorithm-determined) is the server-side
authentication verification value VK_s, which is specified by the
algorithm.
The header MUST be sent before the content body; it MUST NOT be sent
in the trailer of a chunked-encoded response. If a "100 (Continue)"
[RFC7231] response is sent from the server, the Authentication-Info
header SHOULD be included in that response instead of the final
response.
5. Authentication Realms
In this protocol, an authentication realm is defined as a set of
resources (URIs) for which the same set of usernames and passwords is
valid. If the server requests authentication for an authentication
realm for which the client is already authenticated, the client will
automatically perform the authentication using the already-known
credentials. However, for different authentication realms, clients
MUST NOT automatically reuse usernames and passwords for another
realm.
As is the case for the Basic and Digest access authentication protocols, the Mutual authentication protocol supports multiple, separate protection spaces to be set up inside each host. Furthermore, the protocol allows a single authentication realm to span several hosts within the same Internet domain. Each authentication realm is defined and distinguished by the triple of an authentication algorithm, an authentication scope, and a "realm" parameter. However, it is NOT RECOMMENDED that server operators use the same pair of an authentication scope and a realm with different authentication algorithms. The "realm" parameter is a string as defined in Section 4. Authentication scopes are described in the remainder of this section. An authentication scope specifies the range of hosts spanned by the authentication realm. In this protocol, it MUST be one of the following kinds of strings: o Single-server type: A string in the format "<scheme>://<host>" or "<scheme>://<host>:<port>", where <scheme>, <host>, and <port> are the corresponding URI parts of the request URI. If the default port (i.e., 80 for HTTP and 443 for HTTPS) is used for the underlying HTTP communications, the port part MUST be omitted, regardless of whether it was present in the request URI. In all other cases, the port part MUST be present, and it MUST NOT contain leading zeros. Use this format when authentication is only valid for a specific protocol (such as HTTPS). This format is equivalent to the ASCII serialization of a Web origin, as presented in Section 6.2 of [RFC6454]. o Single-host type: The "host" part of the requested URI. This is the default value. Authentication realms within this kind of authentication scope will span several protocols (e.g., HTTP and HTTPS) and ports but will not span different hosts. o Wildcard-domain type: A string in the format "*.<domain-postfix>", where <domain-postfix> is either the host part of the requested URI or any domain in which the requested host is included (this means that the specification "*.example.com" is valid for all of hosts "www.example.com", "web.example.com", "www.sales.example.com", and "example.com"). The domain-postfix sent by the servers MUST be equal to or included in a valid Internet domain assigned to a specific organization; if clients know, via some means such as a blacklist for HTTP cookies [RFC6265], that the specified domain is not to be assigned to any specific organization (e.g., "*.com" or "*.jp"), it is RECOMMENDED that clients reject the authentication request.
In the above specifications, every "scheme", "host", and "domain"
MUST be in lower case, and any internationalized domain names beyond
the ASCII character set SHALL be represented in the way they are sent
in the underlying HTTP protocol, represented in lowercase characters,
i.e., these domain names SHALL be in the form of LDH ("letters,
digits, hyphen") labels as defined in the Internationalized Domain
Names for Applications (IDNA) specification [RFC5890]. A "port" MUST
be given in shortest unsigned decimal number notation. Not obeying
these requirements will cause valid authentication attempts to fail.
5.1. Resolving Ambiguities
In the above definitions of authentication scopes, several scopes may
overlap each other. If a client has already been authenticated to
several realms applicable to the same server, the client may have
multiple lists of the "path" parameters received with the
"401-KEX-S1" message (see Section 4). If these path lists have any
overlap, a single URI may belong to multiple possible candidate
realms to which the client can be authenticated. In such cases,
clients face an ambiguous choice regarding which credentials to send
for a new request (see Steps 3 and 4 of the decision procedure
presented in Section 10).
In such cases, a client MAY freely send requests that belong to any
of these candidate realms, or it MAY simply send an unauthenticated
request and see for which realm the server requests an
authentication. It is RECOMMENDED that server operators provide
properly configured "path" parameters (more precisely, disjoint path
sets for each realm) for clients so that such ambiguities will not
occur.
The following procedure is one possible tactic for resolving
ambiguities in such cases:
o If the client has previously sent a request to the same URI and it
remembers the authentication realm requested by the 401-INIT
message at that time, use that realm.
o In other cases, use one of the authentication realms representing
the most-specific authentication scopes. The list of possible
domain specifications shown above is given from most specific to
least specific.
If there are several choices with different wildcard-domain
specifications, the one that has the longest domain-postfix has
priority over those with shorter domain-postfixes.
o If there are realms with the same authentication scope, there is
no defined priority; the client MAY choose any one of the possible
choices.
6. Session Management
In the Mutual authentication protocol, a session represented by
an sid is set up using four messages (first request, 401-INIT,
req-KEX-C1, and 401-KEX-S1), after which a session secret (z)
associated with the session is established. After mutually
establishing a session secret, this session, along with the secret,
can be used for one or more requests for resources protected by the
same realm on the same server. Note that session management is only
an inside detail of the protocol and usually not visible to normal
users. If a session expires, the client and server SHOULD
automatically re-establish another session without informing
the user.
Sessions and session identifiers are local to each server (defined by
scheme, host, and port), even if an authentication scope covers
multiple servers; clients MUST establish separate sessions for each
port of a host to be accessed. Furthermore, sessions and identifiers
are also local to each authentication realm, even if they are
provided by the same server. The same session identifiers provided
either from different servers or for different realms MUST be treated
as being independent of each other.
The server SHOULD accept at least one req-VFY-C request for each
session if the request reaches the server in a time window specified
by the "timeout" parameter in the 401-KEX-S1 message and if there are
no emergent reasons (such as flooding attacks) to forget the session.
After that, the server MAY discard any session at any time and MAY
send 401-STALE messages for any further req-VFY-C requests received
for that session.
The client MAY send two or more requests using a single session
specified by the sid. However, for all such requests, each value of
the nonce number (in the "nc" parameter) MUST satisfy the following
conditions:
o It is a natural number.
o The same nonce number was not sent within the same session.
o It is not larger than the nc-max value that was sent from the
server in the session represented by the sid.
o It is larger than (largest-nc - nc-window), where largest-nc is
the largest value of nc that was previously sent in the session
and nc-window is the value of the "nc-window" parameter that was
received from the server for the session.
The last condition allows servers to reject any nonce numbers that
are "significantly" smaller than the "current" value (defined by the
value of nc-window) of the nonce number used in the session involved.
In other words, servers MAY treat such nonce numbers as "already
received". This restriction enables servers to implement
duplicate-nonce detection in a constant amount of memory for each
session.
Servers MUST check for duplication of the received nonce numbers, and
if any duplication is detected, the server MUST discard the session
and respond with a 401-STALE message, as outlined in Section 11. The
server MAY also reject other invalid nonce numbers (such as those
above the nc-max limit) by sending a 401-STALE message.
For example, assume that the nc-window value of the current session
is 128 and nc-max is 400, and that the client has already used the
following nonce numbers: {1-120, 122, 124, 130-238, 255-360,
363-372}. The nonce number that can then be used for the next
request is a number from the following set: {245-254, 361, 362,
373-400}. The values {0, 121, 123, 125-129, 239-244} MAY be rejected
by the server because they are not above the current "window limit"
(244 = 372 - 128).
Typically, clients can ensure the above property by using a
monotonically increasing integer counter that counts from zero up to
the value of nc-max.
The values of the nonce numbers and any nonce-related values MUST
always be treated as natural numbers within an infinite range.
Implementations that use fixed-width integer representations,
fixed-precision floating-point numbers, or similar representations
SHOULD NOT reject any larger values that overflow such representative
limits and MUST NOT silently truncate them using any modulus-like
rounding operation (e.g., by mod 2^32). Instead, the whole protocol
is carefully designed so that recipients MAY replace any such
overflowing values (e.g., 2^80) with some reasonably large maximum
representative integer (e.g., 2^31 - 1 or others).
7. Host Validation Methods
The "validation method" specifies a method to "relate" (or "bind") the mutual authentication processed by this protocol with other authentications already performed in the underlying layers and to prevent man-in-the-middle attacks. It determines the value vh that is an input to the authentication protocols. When HTTPS or another possible secure transport is used, this corresponds to the idea of "channel binding" as described in [RFC5929]. Even when HTTP is used, similar, but somewhat limited, "binding" is performed to prevent a malicious server from trying to authenticate itself to another server as a valid user by forwarding the received credentials. The valid tokens for the "validation" parameter and corresponding values of vh are as follows: host: hostname validation. The value vh will be the ASCII string in the following format: "<scheme>://<host>:<port>", where <scheme>, <host>, and <port> are the URI components corresponding to the server-side resource currently being accessed. The scheme and host are in lower case, and the port is listed in shortest decimal notation. Even if the request URI does not have a port part, vh will include the default port number. tls-server-end-point: TLS endpoint (certificate) validation. The value vh will be the octet string of the hash value of the server's public key certificate used in the underlying TLS [RFC5246] connection, processed as specified in Section 4.1 of [RFC5929]. tls-unique: TLS shared-key validation. The value vh will be the channel-binding material derived from the Finished messages, as defined in Section 3.1 of [RFC5929]. (Note: See Section 7.2 for some security-related notes regarding this validation method.) If HTTP is used on a non-encrypted channel (TCP and the Stream Control Transmission Protocol (SCTP), for example), the validation type MUST be "host". If HTTP/TLS [RFC2818] (HTTPS) is used with a server certificate, the validation type MUST be "tls-server-end-point". If HTTP/TLS is used with an anonymous Diffie-Hellman key exchange, the validation type MUST be "tls-unique" (see the note below).
If the validation type "tls-server-end-point" is used, the server
certificate provided in the TLS connection MUST be verified at least
to make sure that the server actually owns the corresponding private
key. (Note: This verification is automatic in some RSA-based key
exchanges but is NOT automatic in Diffie-Hellman-based key exchanges
with separate exchanges for server verification.)
Clients MUST validate this parameter upon receipt of 401-INIT
messages.
Note: The protocol defines two variants of validation on the TLS
connections. The "tls-unique" method is technically more secure.
However, there are some situations where "tls-server-end-point" is
preferable:
o When TLS accelerating proxies are used. In this case, it is
difficult for the authenticating server to acquire the TLS key
information that is used between the client and the proxy. This
is not the case for client-side "tunneling" proxies using the HTTP
CONNECT method.
o When a black-box implementation of the TLS protocol is used on
either peer.
7.1. Applicability Notes
When the client is a Web browser with any scripting capabilities
(support of dynamic contents), the underlying TLS channel used with
HTTP/TLS MUST provide server identity verification. This means that
(1) anonymous Diffie-Hellman key exchange cipher suites MUST NOT be
used and (2) verification of the server certificate provided by the
server MUST be performed. This is to prevent loading identity-
unauthenticated scripts or dynamic contents, which are referenced
from the authenticated page.
For other systems, when the underlying TLS channel used with HTTP/TLS
does not perform server identity verification, the client SHOULD
ensure that all responses are validated using the Mutual
authentication protocol, regardless of the existence of 401-INIT
responses.
7.2. Notes on "tls-unique"
As described in the interoperability note in Section 3.1 of [RFC5929], the "tls-unique" verification value will be changed by possible TLS renegotiation, causing an interoperability problem. TLS renegotiations are used in several HTTPS server implementations for enforcing some security properties (such as cryptographic strength) for some specific responses. If an implementation supports the "tls-unique" verification method, the following precautions SHOULD be taken: o Both peers must be aware that the vh values used for vkc (in req-VFY-C messages) and vks (in 200-VFY-S messages) may be different. These values MUST be retrieved from underlying TLS libraries each time they are used. o After calculating the values vh and vkc to send a req-VFY-C request, clients SHOULD NOT initiate TLS renegotiation until the end of the corresponding response header is received. An exception is that clients can and SHOULD perform TLS renegotiation as a response to the server's request for TLS renegotiation, before receipt of the beginning of the response header. Also, implementers MUST take care of session resumption attacks regarding "tls-unique" channel-binding mechanisms and master secrets. As a mitigation, the TLS extension defined in [RFC7627] SHOULD be used when "tls-unique" host verification is to be used.8. Authentication Extensions
It is RECOMMENDED that interactive clients (e.g., Web browsers) supporting this protocol support non-mandatory authentication and the Authentication-Control header defined in [RFC8053], except for the "auth-style" parameter. This specification also proposes (but does not mandate) that the default "auth-style" be "non-modal". Web applications SHOULD, however, consider the security impacts of the behavior of clients that do not support these headers. Authentication-initializing messages with the Optional-WWW-Authenticate header are used only where the 401-INIT response is valid. It will not replace other 401-type messages such as 401-STALE and 401-KEX-S1. That is, the "reason" field of such a message MUST be "initial" (or any extensive-tokens NOT defined in Section 4.1).
9. String Preparation
For interoperability reasons, it is important that usernames and passwords used in this protocol be binary-comparable, regardless of the user's input methods and/or environments. To ensure this, the following preparation SHOULD be performed: o Usernames received from users SHOULD be prepared using the "UsernameCasePreserved" profile defined in Section 3.3 of [RFC7613]. o Passwords received from users SHOULD be prepared using the "OpaqueString" profile defined in Section 4.2 of [RFC7613]. In both cases, it is the sender's duty to correctly prepare the character strings. If any non-prepared character string is received from the other peer of the communication, the behavior of its recipient is not defined; the recipient MAY either accept or reject such input. Server applications SHOULD also prepare usernames and passwords accordingly upon registration of user credentials. In addition, binary-based "interfaces" of implementations MAY require and assume that the string is already prepared accordingly; when a string is already stored as a binary Unicode string form, implementations MAY omit preparation and Unicode normalization (performing UTF-8 encoding only) before using it. When a string is already stored as an octet blob, implementations MAY send it as is.10. Decision Procedure for Clients
10.1. General Principles and Requirements
To securely implement the protocol, the client must be careful about accepting the authenticated responses from the server. This also holds true for the reception of a "normal response" (a response that does not contain mutual-authentication-related headers) from HTTP servers.
Per typical HTTP authentication, a single user-level request may
result in the exchange of two or more HTTP requests and responses in
sequence. The following normative rules MUST be followed by the
clients implementing this protocol:
o Any kind of "normal response" MUST only be accepted for the very
first request in the sequence. Any "normal response" returned for
the second or subsequent requests in the sequence SHALL be
considered invalid.
o By the same principle, if any response is related to an
authentication realm that is different from that of the client's
request (for example, a 401-INIT message requesting authentication
on another realm), it MUST only be accepted for the very first
request in the sequence. Such a response returned for a second or
subsequent request in the sequence SHALL be considered invalid.
o A req-KEX-C1 message MAY be sent as either an initial request or a
response to a 401-INIT or 401-STALE message. However, to avoid
infinite loops of messages, the req-KEX-C1 message SHOULD NOT be
sent more than once in the sequence for a single authentication
realm. A 401-KEX-S1 response MUST be accepted only when the
corresponding request is req-KEX-C1.
o A req-VFY-C message MAY be sent if there is a valid session secret
shared between the client and the server, as established by
req-KEX-C1 and 401-KEX-S1 messages. If any response with a
401 status code is returned for such a message, the corresponding
session secret SHOULD be discarded as unusable.
In particular, upon the reception of a 401-STALE response, the
client SHOULD try to establish a new session by sending a
req-KEX-C1 message, but only once within the request/response
sequence.
o A 200-VFY-S message MUST be accepted only as a response to a
req-VFY-C message and nothing else. The VK_s values of such
response messages MUST always be checked against the correct
value, and if it is incorrect, the whole response SHOULD be
considered invalid.
The final status of the client request following the message exchange
sequence shall be determined as follows:
o AUTH-SUCCEED: A 200-VFY-S message with the correct VK_s value was
returned in response to the req-VFY-C request in the sequence.
o AUTH-REQUIRED: Two cases exist:
* A 401-INIT message was returned from the server, and the client
does not know how to authenticate to the given authentication
realm.
* A 401-INIT response was returned for a req-VFY-C (or
req-KEX-C1) message, which means that the user-supplied
authentication credentials were not accepted.
o UNAUTHENTICATED: A "normal response" is returned for an initial
request of any kind in the sequence.
Any kind of response (including a "normal response") other than those
explicitly allowed in the above rules SHOULD be interpreted as a
fatal communication error. In such cases, the clients MUST NOT
process any data (the response body and other content-related
headers) sent from the server. However, to handle exceptional error
cases, clients MAY accept a message without an Authentication-Info
header if it has a Server Error (5xx) status code. In such cases,
they SHOULD be careful about processing the body of the content
(ignoring it is still RECOMMENDED, as it may possibly be forged by
intermediate attackers), and the client will then have a status of
"UNAUTHENTICATED".
If a request is a sub-request for a resource included in another
resource (e.g., embedded images, style sheets, frames), clients MAY
treat an AUTH-REQUESTED status the same way they would treat an
UNAUTHENTICATED status. In other words, the client MAY ignore the
server's request to start authentication with new credentials via
sub-requests.
10.2. State Machine for the Client (Informative)
The following state machine describes the possible request-response
sequences derived from the above normative rules. If implementers
are not quite sure of the security consequences of the above rules,
we strongly advise that the decision procedure below be followed. In
particular, clients SHOULD NOT accept "normal responses" unless
explicitly allowed in the rules. The labels in the steps below are
for informational purposes only. Action entries within each step are
checked in top-to-bottom order, and the first clause satisfied is to
be followed.
Step 1 (step_new_request):
If the client software needs to access a new Web resource, check
to see whether the resource is expected to be inside some
authentication realm for which the user has already been
authenticated via the Mutual authentication scheme. If yes,
go to Step 2. Otherwise, go to Step 5.
Step 2:
Check to see whether there is an available sid for the expected
authentication realm. If there is one, go to Step 3. Otherwise,
go to Step 4.
Step 3 (step_send_vfy_1):
Send a req-VFY-C request.
* If a 401-INIT message is received with a different
authentication realm than expected, go to Step 6.
* If a 401-STALE message is received, go to Step 9.
* If a 401-INIT message is received, go to Step 13.
* If a 200-VFY-S message is received, go to Step 14.
* If a "normal response" is received, go to Step 11.
Step 4 (step_send_kex1_1):
Send a req-KEX-C1 request.
* If a 401-INIT message is received with a different
authentication realm than expected, go to Step 6.
* If a 401-KEX-S1 message is received, go to Step 10.
* If a 401-INIT message is received with the same authentication
realm, go to Step 13 (see Note 1).
* If a "normal response" is received, go to Step 11.
Step 5 (step_send_normal_1):
Send a request without any mutual-authentication headers.
* If a 401-INIT message is received, go to Step 6.
* If a "normal response" is received, go to Step 11.
Step 6 (step_rcvd_init):
Check to see whether the user's password for the requested
authentication realm is known. If yes, go to Step 7. Otherwise,
go to Step 12.
Step 7:
Check to see whether there is an available sid for the expected
authentication realm. If there is one, go to Step 8. Otherwise,
go to Step 9.
Step 8 (step_send_vfy):
Send a req-VFY-C request.
* If a 401-STALE message is received, go to Step 9.
* If a 401-INIT message is received, go to Step 13.
* If a 200-VFY-S message is received, go to Step 14.
Step 9 (step_send_kex1):
Send a req-KEX-C1 request.
* If a 401-KEX-S1 message is received, go to Step 10.
* If a 401-INIT message is received, go to Step 13 (see Note 1).
Step 10 (step_rcvd_kex1):
Send a req-VFY-C request.
* If a 401-INIT message is received, go to Step 13.
* If a 200-VFY-S message is received, go to Step 14.
Step 11 (step_rcvd_normal):
The requested resource is out of the authenticated area. The
client will be in the "UNAUTHENTICATED" status. If the response
contains a request for authentication other than Mutual
authentication, it MAY be handled normally.
Step 12 (step_rcvd_init_unknown):
The requested resource requires Mutual authentication, and the
user is not yet authenticated. The client will be in the
"AUTH-REQUESTED" status; it is RECOMMENDED that the client
process the content sent from the server and ask the user for a
username and password. When those are supplied by the user,
go to Step 9.
Step 13 (step_rcvd_init_failed):
The authentication failed for some reason, possibly because the
password or username is invalid for the authenticated resource.
Forget the user-provided credentials for the authentication
realm, and go to Step 12.
Step 14 (step_rcvd_vfy):
The received message is the 200-VFY-S message, which always
contains a "vks" field. Check the validity of the received VK_s
value. If it is equal to the expected value, then the mutual
authentication succeeded. The client will be in the
"AUTH-SUCCEED" status.
An unexpected value is interpreted as a fatal communication
error.
If a user explicitly asks to log out (via the user interface),
the client MUST forget the user's password, go to Step 5, and
reload the current resource without an authentication header.
Note 1: These transitions MAY be accepted by clients, but it is
NOT RECOMMENDED that servers initiate them.
Figure 5 shows an informative diagram of the client state.
=========== -(11)------------ NEW REQUEST ( UNAUTHENTICATED ) =========== ----------------- | ^ normal v | response +(1)-------------------+ NO +(5)----------+ | The requested URI |--------------------------->| send normal | | known to be auth'ed? | | request | +----------------------+ +-------------+ YES | 401-INIT 401-INIT| | with a different realm | | -----------------------------------. | | / v v | | -(12)------------ NO +(6)--------+ | | ( AUTH-REQUESTED )<------| user/pass | | | ----------------- | known? | | | +-----------+ | | |YES v | v +(2)--------+ | +(7)--------+ | session | | | session | NO NO /| available?| | | available?|\ / +-----------+ | +-----------+ | / |YES | |YES | | | /| | | | v / | 401- 401- v | | +(3)--------+ | INIT --(13)---------- INIT +(8)--------+ | | | send |--+----->/ AUTH-REQUESTED \<-------| send | | | /| req-VFY-C | | \forget password / | req-VFY-C | | \/ +-----------+ / ---------------- /+-----------+ | /\ \ \/ ^ 401-INIT | |401- | | ------ \/\ 401-STALE | | | STALE / | \ /\ -----------------+--------------+---. | / | | / \ | | | | / | v / | 401- | 401- | v v v | +(4)--------+ | KEX-S1 +(10)-------+ KEX-S1 | +(9)--------+ | | send |-|--------->| send |<-------+-| send | | --| req-KEX-C1| | | req-VFY-C | | | req-KEX-C1| |/ +-----------+ | +-----------+ | +-----------+ | |200-VFY-S | 200-VFY-S| ^ |normal | |200-VFY-S / | |response | v / ================== v \ -(14)--------- / USER/PASS INPUTTED -(11)------------ ------->( AUTH-SUCCEED )<-- ================== ( UNAUTHENTICATED ) -------------- ----------------- Figure 5: State Diagram for Clients
(next page on part 3)