SIP Caller Preferences: Caller Preferences for the Session Initiation Protocol (SIP)

Network Working Group                                       J. Rosenberg
Request for Comments: 3841                                   dynamicsoft
Category: Standards Track                                 H. Schulzrinne
                                                     Columbia University
                                                              P. Kyzivat
                                                           Cisco Systems
                                                             August 2004


      Caller Preferences for the Session Initiation Protocol (SIP)

Status of this Memo

   This document specifies an Internet standards track protocol for the
   Internet community, and requests discussion and suggestions for
   improvements.  Please refer to the current edition of the "Internet
   Official Protocol Standards" (STD 1) for the standardization state
   and status of this protocol.  Distribution of this memo is unlimited.

Copyright Notice

   Copyright (C) The Internet Society (2004).

Abstract

   This document describes a set of extensions to the Session Initiation
   Protocol (SIP) which allow a caller to express preferences about
   request handling in servers.  These preferences include the ability
   to select which Uniform Resource Identifiers (URI) a request gets
   routed to, and to specify certain request handling directives in
   proxies and redirect servers.  It does so by defining three new
   request header fields, Accept-Contact, Reject-Contact, and Request-
   Disposition, which specify the caller's preferences.

Table of Contents

   1.  Introduction . . . . . . . . . . . . . . . . . . . . . . . . .  2
   2.  Terminology  . . . . . . . . . . . . . . . . . . . . . . . . .  4
   3.  Definitions  . . . . . . . . . . . . . . . . . . . . . . . . .  4
   4.  Overview of Operation  . . . . . . . . . . . . . . . . . . . .  4
   5.  UAC Behavior . . . . . . . . . . . . . . . . . . . . . . . . .  5
       5.1.  Request Handling Preferences . . . . . . . . . . . . . .  6
       5.2.  Feature Set Preferences  . . . . . . . . . . . . . . . .  6
   6.  UAS Behavior . . . . . . . . . . . . . . . . . . . . . . . . .  8
   7.  Proxy Behavior . . . . . . . . . . . . . . . . . . . . . . . .  9
       7.1.  Request-Disposition Processing . . . . . . . . . . . . .  9
       7.2.  Preference and Capability Matching . . . . . . . . . . .  9
             7.2.1. Extracting Explicit Preferences . . . . . . . . . 10
             7.2.2. Extracting Implicit Preferences . . . . . . . . . 10
                    7.2.2.1. Methods. . . . . . . . . . . . . . . . . 10
                    7.2.2.2. Event Packages . . . . . . . . . . . . . 11
             7.2.3. Constructing Contact Predicates . . . . . . . . . 11
             7.2.4. Matching. . . . . . . . . . . . . . . . . . . . . 12
             7.2.5. Example . . . . . . . . . . . . . . . . . . . . . 16
   8.  Mapping Feature Parameters to a Predicate. . . . . . . . . . . 17
   9.  Header Field Definitions . . . . . . . . . . . . . . . . . . . 19
       9.1.  Request Disposition  . . . . . . . . . . . . . . . . . . 20
       9.2.  Accept-Contact and Reject-Contact Header Fields  . . . . 21
   10. Augmented BNF  . . . . . . . . . . . . . . . . . . . . . . . . 22
   11. Security Considerations  . . . . . . . . . . . . . . . . . . . 22
   12. IANA Considerations  . . . . . . . . . . . . . . . . . . . . . 23
   13. Acknowledgments  . . . . . . . . . . . . . . . . . . . . . . . 23
   14. References . . . . . . . . . . . . . . . . . . . . . . . . . . 24
       14.1. Normative References . . . . . . . . . . . . . . . . . . 24
       14.2. Informative References . . . . . . . . . . . . . . . . . 24
   15. Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . 25
   16. Full Copyright Statements. . . . . . . . . . . . . . . . . . . 26

1.  Introduction

   When a Session Initiation Protocol (SIP) [1] server receives a
   request, there are a number of decisions it can make regarding the
   processing of the request.  These include:

   o  whether to proxy or redirect the request

   o  which URIs to proxy or redirect to

   o  whether to fork or not

   o  whether to search recursively or not
   o  whether to search in parallel or sequentially

   The server can base these decisions on any local policy.  This policy
   can be statically configured, or can be based on execution of a
   program or database access.

   However, the administrator of the server is the not the only entity
   with an interest in request processing.  There are at least three
   parties which have an interest: (1) the administrator of the server,
   (2) the user that sent the request, and (3) the user to whom the
   request is directed.  The directives of the administrator are
   embedded in the policy of the server.  The preferences of the user to
   whom the request is directed (referred to as the callee, even though
   the request method may not be INVITE) can be expressed most easily
   through a script written in some type of scripting language, such as
   the Call Processing Language (CPL) [11].  However, no mechanism
   exists to incorporate the preferences of the user that sent the
   request (also referred to as the caller, even though the request
   method may not be INVITE).  For example, the caller might want to
   speak to a specific user, but wants to reach them only at work,
   because the call is a business call.  As another example, the caller
   might want to reach a user, but not their voicemail, since it is
   important that the caller talk to the called party.  In both of these
   examples, the caller's preference amounts to having a proxy make a
   particular routing choice based on the preferences of the caller.

   This extension allows the caller to have these preferences met.  It
   does so by specifying mechanisms by which a caller can provide
   preferences on processing of a request.  There are two types of
   preferences.  One of them, called request handling preferences, are
   encapsulated in the Request-Disposition header field.  They provide
   specific request handling directives for a server.  The other, called
   feature preferences, is present in the Accept-Contact and Reject-
   Contact header fields.  They allow the caller to provide a feature
   set [2] that expresses its preferences on the characteristics of the
   UA that is to be reached.  These are matched with a feature set
   provided by a UA to its registrar [3].  The extension is very general
   purpose, and not tied to a particular service.  Rather, it is a tool
   that can be used in the development of many services.

   One example of a service enabled by caller preferences is a "one
   number" service.  A user can have a single identity (their SIP URI)
   for all of their devices - their cell phone, PDA, work phone, home
   phone, and so on.  If the caller wants to reach the user at their
   business phone, they simply select "business phone" from a pull-down
   menu of options when calling that URI.  Users would no longer need to
   maintain and distribute separate identities for each device.

2.  Terminology

   In this document, the key words "MUST", "MUST NOT", "REQUIRED",
   "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY",
   and "OPTIONAL" are to be interpreted as described in BCP 14, RFC 2119
   [4] and indicate requirement levels for compliant implementations.

3.  Definitions

   Much of the terminology used in this specification is presented in
   [3].  This specification defines the following additional terms:

   Caller: Within the context of this specification, a caller refers to
      the user on whose behalf a UAC is operating.  It is not limited to
      a user whose UAC sends an INVITE request.

   Feature Preferences: Caller preferences that describe desired
      properties of a UA to which the request is to be routed.  Feature
      preferences can be made explicit with the Accept-Contact and
      Reject-Contact header fields.

   Request Handling Preferences: Caller preferences that describe
      desired request treatment at a server.  These preferences are
      carried in the Request-Disposition header field.

   Target Set: A target set is a set of candidate URIs to which a proxy
      or redirect server can send or redirect a request.  Frequently,
      target sets are obtained from a registration, but they need not
      be.

   Explicit Preference: A caller preference indicated explicitly in the
      Accept-Contact or Reject-Contact header fields.

   Implicit Preference: A caller preference that is implied through the
      presence of other aspects of a request.  For example, if the
      request method is INVITE, it represents an implicit caller
      preference to route the request to a UA that supports the INVITE
      method.

4.  Overview of Operation

   When a caller sends a request, it can optionally include new header
   fields which request certain handling at a server.  These preferences
   fall into two categories.  The first category, called request
   handling preferences, is carried in the Request-Disposition header
   field.  It describes specific behavior that is desired at a server.
   Request handling preferences include whether the caller wishes the
   server to proxy or redirect, and whether sequential or parallel
   search is desired.  These preferences can be applied at every proxy
   or redirect server on the call signaling path.

   The second category of preferences, called feature preferences, is
   carried in the Accept-Contact and Reject-Contact header fields.
   These header fields contain feature sets, represented by the same
   feature parameters that are used to indicate capabilities [3].  Here,
   the feature parameters represent the caller's preferences.  The
   Accept-Contact header field contains feature sets that describe UAs
   that the caller would like to reach.  The Reject-Contact header field
   contains feature sets which, if matched by a UA, imply that the
   request should not be routed to that UA.

   Proxies use the information in the Accept-Contact and Reject-Contact
   header fields to select amongst contacts in their target set.  When
   neither of those header fields are present, the proxy computes
   implicit preferences from the request.  These are caller preferences
   that are not explicitly placed into the request, but can be inferred
   from the presence of other message components.  As an example, if the
   request method is INVITE, this is an implicit preference to route the
   call to a UA that supports the INVITE method.

   Both request handling and feature preferences can appear in any
   request, not just INVITE.  However, they are only useful in requests
   where proxies need to determine a request target.  If the domain in
   the request URI is not owned by any proxies along the request path,
   those proxies will never access a location service, and therefore,
   never have the opportunity to apply the caller preferences.  This
   makes sense because typically, the request URI will identify a UAS
   for mid-dialog requests.  In those cases, the routing decisions were
   already made on the initial request, and it makes no sense to redo
   them for subsequent requests in the dialog.

5.  UAC Behavior

   A caller wishing to express preferences for a request includes
   Accept-Contact, Reject-Contact, or Request-Disposition header fields
   in the request, depending on their particular preferences.  No
   additional behavior is required after the request is sent.

   The Accept-Contact, Reject-Contact, and Request-Disposition header
   fields in an ACK for a non-2xx final response, or in a CANCEL
   request, MUST be equal to the values in the original request being
   acknowledged or cancelled.  This is to ensure proper operation
   through stateless proxies.

   If the UAC wants to determine whether servers along the path
   understand the header fields described in this specification, it
   includes a Proxy-Require header field with a value of "pref" [3] in
   its request.  If the request should fail with a 420 response code,
   the UAC knows that the extension is not supported.  In that case, it
   SHOULD retry, and may decide whether or not to use caller
   preferences.  A UA should only use Proxy-Require if knowledge about
   support is essential for handling of the request.  Note that, in any
   case, caller preferences can only be considered preferences - there
   is no guarantee that the requested service will be executed.  As
   such, inclusion of a Proxy-Require header field does not mean that
   the preferences will be executed, just that the caller preferences
   extension is understood by the proxies.

5.1.  Request Handling Preferences

   The Request-Disposition header field specifies caller preferences for
   how a server should process a request.  Its value is a list of
   tokens, each of which specifies a particular processing directive.

   The syntax of the header field can be found in Section 10, and the
   semantics of the directives are described in Section 9.1.

5.2.  Feature Set Preferences

   A UAC can indicate caller preferences for the capabilities of a UA
   that should be reached or not reached as a result of sending a SIP
   request.  To do that, it adds one or more Accept-Contact and Reject-
   Contact header field values.  Each header field value contains a set
   of feature parameters that define a feature set.  The syntax of the
   header field can be found in Section 10, and a discussion of their
   usage in Section 9.2.

   Each feature set is constructed as described in Section 5 of [3].
   The feature sets placed into these header fields MAY overlap; that
   is, a UA MAY indicate preferences for feature sets that match
   according to the matching algorithm of RFC 2533 [2].

   A UAC can express explicit preferences for the methods and event
   packages supported by a UA.  It is RECOMMENDED that a UA include a
   term in an Accept-Contact feature set with the "sip.methods" feature
   tag (note, however, that even though the name of this feature tag is
   sip.methods, it would be encoded into the Accept-Contact header field
   as just "methods"), whose value includes the method of the request.
   When a UA sends a SUBSCRIBE request, it is RECOMMENDED that a UA
   include a term in an Accept-Contact feature set with the "sip.events"
   feature tag, whose value includes the event package of the request.
   Whether these terms are placed into a new feature set, or whether
   they are included in each feature set, is at the discretion of the
   implementor.  In most cases, the right effect is achieved by
   including a term in each feature set.

   As an example, the following Accept-Contact header field expresses a
   desire to route a call to a mobile device, using feature parameters
   taken from [3]:

   Accept-Contact: *;mobility="mobile";methods="INVITE"

   The Reject-Contact header field allows the UAC to specify that a UA
   should not be contacted if it matches any of the values of the header
   field.  Each value of the Reject-Contact header field contains a "*",
   purely to align the syntax with guidelines for SIP extensions [12],
   and is parameterized by a set of feature parameters.  Any UA whose
   capabilities match the feature set described by the feature
   parameters matches the value.

   The Accept-Contact header field allows the UAC to specify that a UA
   should be contacted if it matches some or all of the values of the
   header field.  Each value of the Accept-Contact header field contains
   a "*", and is parameterized by a set of feature parameters.  Any UA
   whose capabilities match the feature set described by the feature
   parameters matches the value.  The precise behavior depends heavily
   on whether the "require" and "explicit" parameters are present.  When
   both of them are present, a proxy will only forward the request to
   contacts which have explicitly indicated that they support the
   desired feature set.  Any others are discarded.  As such, a UAC
   should only use "require" and "explicit" together when it wishes the
   call to fail unless a contact definitively matches.  It's possible
   that a UA supports a desired feature, but did not indicate it in its
   registration.  When a UAC uses both "explicit" and "require", such a
   contact would not be reached.  As a result, this combination is often
   not the one a UAC will want.

   When only "require" is present, it means that a contact will not be
   used if it doesn't match.  If it does match, or if it's not known
   whether it's a complete match, the contact is still used.  A UAC
   would use "require" alone when a non-matching contact is useless.
   This is common for services where the request simply can't be
   serviced without the necessary features.  An example is support for
   specific methods or event packages.  When only "require" is present,
   the proxy will also preferentially route the request to the UA which
   represents the "best" match.  Here, "best" means that the UA has
   explicitly indicated it supports more of the desired features than
   any other. Note, however, that this preferential routing will never
   override an ordering provided by the called party.  The preferential
   routing will only choose amongst contacts of equal q-value.

   When only "explicit" is present, it means that all contacts provided
   by the callee will be used.  However, if the contact isn't an
   explicit match, it is tried last amongst all other contacts with the
   same q-value.  The principle difference, therefore, between this
   configuration and the usage of both "require" and "explicit" is the
   fallback behavior for contacts that don't match explicitly.  Here,
   they are tried as a last resort.  If "require" is also present, they
   are never tried.

   Finally, if neither "require" nor "explicit" are present, it means
   that all contacts provided by the callee will be used.  However, if
   the contact doesn't match, it is tried last amongst all other
   contacts with the same q-value.  If it does match, the request is
   routed preferentially to the "best" match.  This is a common
   configuration for preferences that, if not honored, will still allow
   for a successful call, and the greater the match, the better.

6.  UAS Behavior

   When a UAS compliant to this specification receives a request whose
   request-URI corresponds to one of its registered contacts, it SHOULD
   apply the behavior described in Section 7.2 as if it were a proxy for
   the domain in the request-URI.  The UAS acts as if its location
   database contains a single request target for the request-URI.  That
   target is associated with a feature set.  The feature set is the same
   as the one placed in the registration of the URI in the request-URI.
   If a UA had registered against multiple separate addresses-of-record,
   and the contacts registered for each had different capabilities, it
   will have used a different URI in each registration, so it can
   determine which feature set to use.

   This processing occurs after the client authenticates and authorizes
   the request, but before the remainder of the general UAS processing
   described in Section 8.2.1 of RFC 3261.

   If, after performing this processing, there are no URI left in the
   target set, the UA SHOULD reject the request with a 480 response.  If
   there is a URI remaining (there was only one to begin with), the UA
   proceeds with request processing as per RFC 3261.

   Having a UAS perform the matching operations as if it were a proxy
   allows certain caller preferences to be honored, even if the proxy
   doesn't support the extension.

   A UAS SHOULD process any queue directive present in a Request-
   Disposition header field in the request.  All other directives MUST
   be ignored.

7.  Proxy Behavior

   Proxy behavior consists of two orthogonal sets of rules - one for
   processing the Request-Disposition header field, and one for
   processing the URI and feature set preferences in the Accept-Contact
   and Reject-Contact header fields.

   In addition to processing these headers, a proxy MAY add one if not
   present, or add a value to an existing header field, as if it were a
   UAC.  This is useful for a proxy to request processing in downstream
   proxies in the implementation of a feature.  However, a proxy MUST
   NOT modify or remove an existing header field value.  This is
   particularly important when S/MIME is used.  The message signature
   could include the caller preferences header fields, allowing the UAS
   to verify that, even though proxies may have added header fields, the
   original caller preferences were still present.

7.1.  Request-Disposition Processing

   If the request contains a Request-Disposition header field and it is
   the owner of the domain in the Request URI, the server SHOULD execute
   the directives as described in Section 9.1, unless it has local
   policy configured to direct it otherwise.

7.2.  Preference and Capability Matching

   A proxy compliant to this specification MUST NOT apply the
   preferences matching operation described here to a request unless it
   is the owner of the domain in the request URI, and accessing a
   location service that has capabilities associated with request
   targets.  However, if it is the owner of the domain, and accessing a
   location service that has capabilities associated with request
   targets, it SHOULD apply the processing described in this section.
   Typically, this is a proxy that is using a registration database to
   determine the request targets.  However, if a proxy knows about
   capabilities through some other means, it SHOULD apply the processing
   defined here as well.  If it does perform the processing, it MUST do
   so as described below.

   The processing is described through a conversion from the syntax
   described in this specification to RFC 2533 [2] syntax, followed by a
   matching operation and a sorting of resulting contact values.  The
   usage of RFC 2533 syntax as an intermediate step is not required; it
   only serves as a useful tool to describe the behavior required of the
   proxy.  A proxy can use any steps it likes, so long as the results
   are identical to the ones that would be achieved with the processing
   described here.

7.2.1.  Extracting Explicit Preferences

   The first step in proxy processing is to extract explicit
   preferences.  To do that, it looks for the Accept-Contact and
   Reject-Contact header fields.

   For each value of those header fields, it extracts the feature
   parameters.  These are the header field parameters whose name is
   "audio", "automata", "class", "duplex", "data", "control",
   "mobility", "description", "events", "priority", "methods",
   "extensions", "schemes", "application", "video", "language", "type",
   "isfocus", "actor", or "text", or whose name begins with a plus (+)
   [3].  The proxy converts all of those parameters to the syntax of RFC
   2533, based on the rules in Section 8.

   The result will be a set of feature set predicates in conjunctive
   normal form, each of which is associated with one of the two
   preference header fields.  If there was a req-parameter associated
   with a header field value in the Accept-Contact header field, the
   feature set predicate derived from that header field value is said to
   have its require flag set.  Similarly, if there was an explicit-param
   associated with a header field value in the Accept-Contact header
   field, the feature set predicate derived from that header field value
   is said to have its explicit flag set.

7.2.2.  Extracting Implicit Preferences

   If, and only if, the proxy did not find any explicit preferences in
   the request (because there was no Accept-Contact or Reject-Contact
   header field), the proxy extracts implicit preferences.  These
   preferences are ones implied by the presence of other information in
   the request.

   First, the proxy creates a conjunction with no terms.  This
   conjunction represents a feature set that will be associated with the
   Accept-Contact header field, as if it were included there.  Note that
   there is no modification of the message implied - only an association
   for the purposes of processing.  Furthermore, this feature set has
   its require flag set, but not its explicit flag.

   The proxy then adds terms to the conjunction for the two implicit
   preference types below.

7.2.2.1.  Methods

   One implicit preference is the method.  When a UAC sends a request
   with a specific method, it is an implicit preference to have the
   request routed only to UAs that support that method.  To support this
   implicit preference, the proxy adds a term to the conjunction of the
   following form:

   (sip.methods=[method of request])

7.2.2.2.  Event Packages

   For requests that establish a subscription [5], the Event header
   field is another expression of an implicit preference.  It expresses
   a desire for the request to be routed only to a server that supports
   the given event package.  To support this implicit preference, the
   proxy adds a term to the conjunction of the following form:

   (sip.events=[value of the Event header field])

7.2.3.  Constructing Contact Predicates

   The proxy then takes each URI in the target set (the set of URI it is
   going to proxy or redirect to), and obtains its capabilities as an
   RFC 2533 formatted feature set predicate.  This is called a contact
   predicate.  If the target URI was obtained through a registration,
   the proxy computes the contact predicate by extracting the feature
   parameters from the Contact header field [3] and then converting them
   to a feature predicate.  To extract the feature parameters, the proxy
   follows these steps:

   1. Create an initial, empty list of feature parameters.

   2. If the Contact URI parameters included the "audio", "automata",
      "class", "duplex", "data", "control", "mobility", "description",
      "events", "priority", "methods", "schemes", "application",
      "video", "actor", "language", "isfocus", "type", "extensions", or
      "text" parameters, those are copied into the list.

   3. If any Contact URI parameter name begins with a "+", it is copied
      into the list if the list does not already contain that name with
      the plus removed.  In other words, if the "video" feature
      parameter is in the list, the "+video" parameter would not be
      placed into the list.  This conflict should never arise if the
      client were compliant to [3], since it is illegal to use the +
      form for encoding of a feature tag in the base set.

   If the URI in the target set had no feature parameters, it is said to
   be immune to caller preference processing.  This means that the URI
   is removed from the target set temporarily, the caller preferences
   processing described below is executed, and then the URI is added
   back in.

   Assuming the URI has feature parameters, they are converted to RFC
   2533 syntax using the rules of Section 8.

   The resulting predicate is associated with a q-value.  If the contact
   predicate was learned through a REGISTER request, the q-value is
   equal to the q-value in the Contact header field parameter, else
   "1.0" if not specified.

   As an example, consider the following registered Contact header
   field:

     Contact: <sip:user@example.com>;audio;video;mobility="fixed";
         +sip.message="TRUE";other-param=66372;
         methods="INVITE,OPTIONS,BYE,CANCEL,ACK";schemes="sip,http"

   This would be converted into the following predicate:

      (& (sip.audio=TRUE)
         (sip.video=TRUE)
         (sip.mobility=fixed)
         (sip.message=TRUE)
         (| (sip.methods=INVITE) (sip.methods=OPTIONS) (sip.methods=BYE)
            (sip.methods=CANCEL) (sip.methods=ACK))
         (| (sip.schemes=sip) (sip.schemes=http)))

   Note that "other-param" was not considered a feature parameter, since
   it is neither a base tag nor did it begin with a leading +.

7.2.4.  Matching

   It is important to note that the proxy does not have to know anything
   about the meaning of the feature tags that it is comparing in order
   to perform the matching operation.  The rules for performing the
   comparison depend on syntactic hints present in the values of each
   feature tag.  For example, a predicate such as:

   (foo>=4)

   implies that the feature tag "foo" is a numeric value.  The matching
   rules in RFC 2533 only require an implementation to know whether the
   feature tag is a numeric, token, or quoted string (booleans can be
   treated as tokens).  Quoted strings are always matched using a case-
   sensitive matching operation.  Tokens are matched using case-
   insensitive matching.  These two cases are differentiated by the
   presence of angle brackets around the feature tag value.  When these
   brackets are present (i.e., ;+sip.foo="<value>"), it implies case
   sensitive string comparison.  When they are not present, (i.e.,
   (;+sip.bar="val"), it implies case insensitivity.  Numerics are
   matched using normal mathematical comparisons.

   First, the proxy applies the predicates associated with the Reject-
   Contact header field.

   For each contact predicate, each Reject-Contact predicate (that is,
   each predicate associated with the Reject-Contact header field) is
   examined.  If that Reject-Contact predicate contains a filter for a
   feature tag, and that feature tag is not present anywhere in the
   contact predicate, that Reject-Contact predicate is discarded for the
   processing of that contact predicate.  If the Reject-Contact
   predicate is not discarded, it is matched with the contact predicate
   using the matching operation of RFC 2533 [2].  If the result is a
   match, the URI corresponding to that contact predicate is discarded
   from the target set.

   The result is that Reject-Contact will only discard URIs where the UA
   has explicitly indicated support for the features that are not
   wanted.

   Next, the proxy applies the predicates associated with the Accept-
   Contact header field.  For each contact that remains in the target
   set, the proxy constructs a matching set, Ms.  Initially, this set
   contains all of the Accept-Contact predicates.  Each of those
   predicates is examined.  It is matched with the contact predicate
   using the matching operation of RFC 2533 [2].  If the result is not a
   match, and the Accept-Contact predicate had its require flag set, the
   URI corresponding to that contact predicate is discarded from the
   target set.  If the result is not a match, but the Accept-Contact
   predicate did not have its require flag set, that contact URI is not
   discarded from the target set, however, the Accept-Contact predicate
   is removed from the matching set for that contact.

   For each contact that remains in the target set, the proxy computes a
   score for that contact against each predicate in the contact's
   matching set.  Let the number of terms in the Accept-Contact
   predicate conjunction be equal to N.  Each term in that predicate
   contains a single feature tag.  If the contact predicate has a term
   containing that same feature tag, the score is incremented by 1/N.
   If the feature tag was not present in the contact predicate, the
   score remains unchanged.  Based on these rules, the score can range
   between zero and one.

                                                    T
                                              +----------> DROP Contact
                                              |
                                              |
                                             / \
                                            /   \
                                        T  /     \   F
                                    +---->/require\------> Set score=0
                                    |     \      /
                                    |      \    /
                                   / \      \  /
                                  /   \      \/
                       score<1   /     \
                      +-------> /explicit----> Score unchanged
                      |         \      /    F
                      |          \    /
                     / \          \  /
                    /   \          \/
    +--------+     /     \
 -->|Compute |--> /Score  \ --------> Score unchanged
    |  Score |    \      /  score=1
    +--------+     \    /
                    \  /
                     \/

   Figure 1: Applying the Score

   The require and explicit tags are then applied, resulting in
   potential modification of the score and the target set.  This process
   is summarized in Figure 1.  If the score for the contact predicate
   against that Accept-Contact predicate was less than one, the Accept-
   Contact predicate had an explicit tag, and if the predicate also had
   a require tag, the Contact URI corresponding to that contact
   predicate is dropped.  If, however, the predicate did not have a
   require tag, the score is set to zero.  If there was no explicit tag,
   the score is unchanged.

   The next step is to combine the scores and the q-values associated
   with the predicates in the matching set, to arrive at an overall
   caller preference, Qa.  For those URIs in the target set which
   remain, there will be a score which indicates its match against each
   Accept-Contact predicate in the matching set.  If there are M
   Accept-Contact predicates in the matching set, there will be M scores
   S1 through SM, for each contact.  The overall caller preference, Qa,
   is the arithmetic average of S1 through SM.

   At this point, any URIs that were removed from the target set because
   they were immune from caller preferences are added back in, and Qa
   for that URI is set to 1.0.

   The purpose of the caller preference Qa is to provide an ordering for
   any contacts remaining in the target set, if the callee has not
   provided an ordering.  To do this, the contacts remaining in the
   target set are sorted by the q-value provided by the callee.  Once
   sorted, they are grouped into equivalence classes, such that all
   contacts with the same q-value are in the same equivalence class.
   Within each equivalence class, the contacts are then ordered based on
   their values of Qa.  The result is an ordered list of contacts that
   is used by the proxy.

   If there were no URIs in the target set after the application of the
   processing in this section, and the caller preferences were based on
   implicit preferences (Section 7.2.2), the processing in this section
   is discarded, and the original target set, ordered by their original
   q-values, is used.

      This handles the case where implicit preferences for the method or
      event packages resulted in the elimination of all potential
      targets.  By going back to the original target set, those URIs
      will be tried, and result in the generation of a 405 or 489
      response.  The UAC can then use this information to try again, or
      report the error to the user.  Without reverting to the original
      target set, the UAC would see a 480 response, and have no
      knowledge of why their request failed.  Of course, the target set
      can also be empty after the application of explicit preferences.
      This will result in the generation of a 480 by the proxy.  This
      behavior is acceptable, and indeed, desirable in the case of
      explicit preferences.  When the caller makes an explicit
      preference, it is agreeing that its request might fail because of
      a preference mismatch.  One might try to return an error
      indicating the capabilities of the callee, so that the caller
      could perhaps try again.  However, doing so results in the leaking
      of potentially sensitive information to the caller without
      authorization from the callee, and therefore this specification
      does not provide a means for it.

   If a proxy server is recursing, it adds the Contact header fields
   returned in the redirect responses to the target set, and re-applies
   the caller preferences algorithm.

   If the server is redirecting, it returns all entries in the target
   set.  It assigns q-values to those entries so that the ordering is
   identical to the ordering determined by the processing above.
   However, it MUST NOT include the feature parameters for the entries
   in the target set.  If it did, the upstream proxy server would apply
   the same caller preferences once more, resulting in a double
   application of those preferences.  If the redirect server does wish
   to include the feature parameters in the Contact header field, it
   MUST redirect using the original target set and original q-values,
   before the application of caller preferences.

7.2.5.  Example

   Consider the following example, which is contrived but illustrative
   of the various components of the matching process.  There are five
   registered Contacts for sip:user@example.com.  They are:

   Contact: sip:u1@h.example.com;audio;video;methods="INVITE,BYE";q=0.2
   Contact: sip:u2@h.example.com;audio="FALSE";
     methods="INVITE";actor="msg-taker";q=0.2
   Contact: sip:u3@h.example.com;audio;actor="msg-taker";
     methods="INVITE";video;q=0.3
   Contact: sip:u4@h.example.com;audio;methods="INVITE,OPTIONS";q=0.2
   Contact: sip:u5@h.example.com;q=0.5

   An INVITE sent to sip:user@example.com contained the following caller
   preferences header fields:

   Reject-Contact: *;actor="msg-taker";video
   Accept-Contact: *;audio;require
   Accept-Contact: *;video;explicit
   Accept-Contact: *;methods="BYE";class="business";q=1.0

   There are no implicit preferences in this example, because explicit
   preferences are provided.

   The proxy first removes u5 from the target set, since it is immune
   from caller preferences processing.

   Next, the proxy processes the Reject-Contact header field.  It is a
   match for all four remaining contacts, but only an explicit match for
   u3.  That is because u3 is the only one that explicitly indicated
   support for video, and explicitly indicated it is a message taker.
   So, u3 gets discarded, and the others remain.

   Next, each of the remaining three contacts is compared against each
   of the three Accept-Contact predicates.  u1 is a match to all three,
   earning a score of 1.0 for the first two predicates, and 0.5 for the
   third (the methods feature tag was present in the contact predicate,
   but the class tag was not).  u2 doesn't match the first predicate.
   Because that predicate has a require tag, u2 is discarded.  u4
   matches the first predicate, earning a score of 1.0.  u4 matches the
   second predicate, but since the match is not explicit (the score is
   0.0, in fact), the score is set to zero (it was already zero, so
   nothing changes).  u4 does not match the third predicate.

   At this point, u1 and u4 remain.  u1 matched all three Accept-Contact
   predicates, so its matching set contains all three, with scores of 1,
   1, and 0.5.  u4 matches the first two predicates, with scores of 1.0
   and 0.0.  Qa for u1 is 0.83 and Qa for u4 is 0.5.  u5 is added back
   in with a Qa of 1.0.

   Next, the remaining contacts in the target set are sorted by q-value.
   u5 has a value of 0.5, u1 has a q-value of 0.2 and so does u4.  There
   are two equivalence classes.  The first has a q-value of 0.5, and
   consists of just u5.  Since there is only one member of the class,
   sorting within the class has no impact.  The second equivalence class
   has a q-value of 0.2.  Within that class, the two contacts, u1 and
   u4, are ordered based on their values of Qa.  u1 has a Qa of 0.83,
   and u4, a Qa of 0.5.  Thus, u1 comes first, followed by u4.  The
   resulting overall ordered set of contacts in the target set is u5,
   u1, and then u4.

8.  Mapping Feature Parameters to a Predicate

   Mapping between feature parameters and a feature set predicate,
   formatted according to the syntax of RFC 2533 [2], is trivial.  It is
   just the opposite of the process described in Section 5 of [3].

   Starting from a set of feature-param, the procedure is as follows.
   Construct a conjunction.  Each term in the conjunction derives from
   one feature-param.  If the feature-param has no value, it is
   equivalent, in terms of the processing which follows, as if it had a
   value of "TRUE".

   If the feature-param value is a tag-value-list, the element of the
   conjunction is a disjunction.  There is one term in the disjunction
   for each tag-value in the tag-value-list.

   Consider now the construction of a filter from a tag-value.  If the
   tag-value starts with an exclamation mark (!), the filter is of the
   form:

   (! <filter from remainder>)

   where "<filter from remainder>" refers to the filter that would be
   constructed from the tag-value if the exclamation mark had not been
   present.

   If the tag-value starts with an octothorpe (#), the filter is a
   numeric comparison.  The comparator is either =, >=, <=, or a range
   based on the next characters in the phrase.  If the next characters
   are =, >=, or <=, the filter is of the form:

   (name comparator compare-value)

   where name is the name of the feature parameter after it has been
   decoded (see below), and the comparator is either =, >=, or <=
   depending of the initial characters in the phrase.  If the remainder
   of the text in the tag-value after the equal contains a decimal point
   (implying a rational number), the decimal point is shifted right N
   times until it is an integer, I.  Compare-value above is then set to
   "I / 10**N", where 10**N is the result of computing the number 10 to
   the Nth power.

   If the value after the octothorpe is a number, the filter is a range.
   The format of the filter is:

      (name=<remainder>)

   where "name" is the feature-tag after it has been decoded (see
   below), and "<remainder>" is the remainder of the text in the tag-
   value after the #, with any decimal numbers converted to a rational
   form, and the colon replaced by a double dot (..).

   If the tag-value does not begin with an octothorpe (it is a token-
   nobang or boolean), the filter is of the form:

      (name=tag-value)

   where name is the feature-tag after it has been decoded (see below).

   If the feature-param contains a string-value (based on the fact that
   it begins with a left angle bracket ("<") and ends with a right angle
   bracket (">")), the filter is of the form:

      (name="qdtext")

   Note the explicit usage of quotes around the qdtext, which indicate
   that the value is a string.  In RFC 2533, strings are compared using
   case sensitive rules, and tokens are compared using case insensitive
   rules.

   Feature tags, as specified in RFC 2506 [13], cannot be directly
   represented as header field parameters in the Contact, Accept-
   Contact, and Reject-Contact header fields.  This is due to an
   inconsistency in the grammars, and in the need to differentiate
   feature parameters from parameters used by other extensions.  As
   such, feature tag values are encoded from RFC 2506 format to yield an
   enc-feature-tag, and then are decoded into RFC 2506 format.  The
   decoding process is simple.  If there is a leading plus (+) sign, it
   is removed.  Any exclamation point (!) is converted to a colon (:)
   and any single quote (') is converted to a forward slash (/).  If
   there was no leading plus sign, and the remainder of the encoded name
   was "audio", "automata", "class", "duplex", "data", "control",
   "mobility", "description", "events", "priority", "methods",
   "schemes", "application", "video", "actor", "isfocus", "extensions"
   or "text", the prefix "sip." is added to the remainder of the encoded
   name to compute the feature tag name.

   As an example, the Accept-Contact header field:

      Accept-Contact:*;mobility="fixed"
        ;events="!presence,message-summary"
        ;language="en,de";description="<PC>";+sip.newparam
        ;+rangeparam="#-4:+5.125"

   would be converted to the following feature predicate:

         (& (sip.mobility=fixed)
            (| (! (sip.events=presence)) (sip.events=message-summary))
            (| (language=en) (language=de))
            (sip.description="PC")
            (sip.newparam=TRUE)
            (rangeparam=-4..5125/1000))

9.  Header Field Definitions

   This specification defines three new header fields - Accept-Contact,
   Reject-Contact, and Request-Disposition.

   Figure 2 and Figure 3 are an extension of Tables 2 and 3 in RFC 3261
   [1] for the Accept-Contact, Reject-Contact, and Request-Disposition
   header fields.  The column "INF" is for the INFO method [6], "PRA" is
   for the PRACK method [7], "UPD" is for the UPDATE method [8], "SUB"
   is for the SUBSCRIBE method [5], "NOT" is for the NOTIFY method [5],
   "MSG" is for the MESSAGE method [9], and "REF" is for the REFER
   method [10].

   Header field          where  proxy  ACK BYE CAN INV OPT REG

   Accept-Contact          R      ar    o   o   o   o   o   -
   Reject-Contact          R      ar    o   o   o   o   o   -
   Request-Disposition     R      ar    o   o   o   o   o   o

   Figure 2: Accept-Contact, Reject-Contact, and Request-Disposition
             header fields

   Header field          where  proxy  PRA UPD SUB NOT INF MSG REF

   Accept-Contact          R      ar    o   o   o   o   o   o   o
   Reject-Contact          R      ar    o   o   o   o   o   o   o
   Request-Disposition     R      ar    o   o   o   o   o   o   o

   Figure 3: Accept-Contact, Reject-Contact, and Request-Disposition
             header fields

9.1.  Request Disposition

   The Request-Disposition header field specifies caller preferences for
   how a server should process a request.  Its value is a list of
   tokens, each of which specifies a particular directive.  Its syntax
   is specified in Section 10.  Note that a compact form, using the
   letter d, has been defined.  The directives are grouped into types.
   There can only be one directive of each type per request (e.g., you
   cannot have both "proxy" and "redirect" in the same Request-
   Disposition header field).

   When the caller specifies a directive, the server SHOULD honor that
   directive.

   The following types of directives are defined:

   proxy-directive: This type of directive indicates whether the caller
      would like each server to proxy ("proxy") or redirect
      ("redirect").

   cancel-directive: This type of directive indicates whether the caller
      would like each proxy server to send a CANCEL request downstream
      ("cancel") in response to a 200 OK from the downstream server
      (which is the normal mode of operation, making it redundant), or
      whether this function should be left to the caller ("no-cancel").
      If a proxy receives a request with this parameter set to "no-
      cancel", it SHOULD NOT CANCEL any outstanding branches upon
      receipt of a 2xx.  However, it would still send CANCEL on any
      outstanding branches upon receipt of a 6xx.

   fork-directive: This type of directive indicates whether a proxy
      should fork a request ("fork"), or proxy to only a single address
      ("no-fork").  If the server is requested not to fork, the server
      SHOULD proxy the request to the "best" address (generally the one
      with the highest q-value).  If there are multiple addresses with
      the highest q-value, the server chooses one based on its local
      policy.  The directive is ignored if "redirect" has been
      requested.

   recurse-directive: This type of directive indicates whether a proxy
      server receiving a 3xx response should send requests to the
      addresses listed in the response ("recurse"), or forward the list
      of addresses upstream towards the caller ("no-recurse").  The
      directive is ignored if "redirect" has been requested.

   parallel-directive: For a forking proxy server, this type of
      directive indicates whether the caller would like the proxy server
      to proxy the request to all known addresses at once ("parallel"),
      or go through them sequentially, contacting the next address only
      after it has received a non-2xx or non-6xx final response for the
      previous one ("sequential").  The directive is ignored if
      "redirect" has been requested.

   queue-directive: If the called party is temporarily unreachable,
      e.g., because it is in another call, the caller can indicate that
      it wants to have its call queued ("queue") or rejected immediately
      ("no-queue").  If the call is queued, the server returns "182
      Queued".  A queued call can be terminated as described in [1].

   Example:

      Request-Disposition: proxy, recurse, parallel

   The set of request disposition directives is not extensible on
   purpose.  This is to avoid a proliferation of new extensions to SIP
   that are "tunneled" through this header field.

9.2.  Accept-Contact and Reject-Contact Header Fields

   The syntax for these header fields is described in Section 10.  A
   compact form, with the letter a, has been defined for the Accept-
   Contact header field, and with the letter j for the Reject-Contact
   header field.

10.  Augmented BNF

   The BNF for the Request-Disposition header field is:

   Request-Disposition   =   ( "Request-Disposition" / "d" ) HCOLON
                             directive *(COMMA directive)
   directive             =   proxy-directive / cancel-directive /
                             fork-directive / recurse-directive /
                             parallel-directive / queue-directive
   proxy-directive       =  "proxy" / "redirect"
   cancel-directive      =  "cancel" / "no-cancel"
   fork-directive        =  "fork" / "no-fork"
   recurse-directive     =  "recurse" / "no-recurse"
   parallel-directive    =  "parallel" / "sequential"
   queue-directive       =  "queue" / "no-queue"

   The BNF for the Accept-Contact and Reject-Contact header fields is:

   Accept-Contact  =  ("Accept-Contact" / "a") HCOLON ac-value
                      *(COMMA ac-value)
   Reject-Contact  =  ("Reject-Contact" / "j") HCOLON rc-value
                      *(COMMA rc-value)
   ac-value        =  "*" *(SEMI ac-params)
   rc-value        =  "*" *(SEMI rc-params)
   ac-params       =  feature-param / req-param
                         / explicit-param / generic-param
                       ;;feature param from RFC 3840
                       ;;generic-param from RFC 3261
   rc-params       =  feature-param / generic-param
   req-param       =  "require"
   explicit-param  =  "explicit"

   Despite the BNF, there MUST NOT be more than one req-param or
   explicit-param in an ac-params.  Furthermore, there can only be one
   instance of any feature tag in feature-param.

11.  Security Considerations

   The presence of caller preferences in a request has an effect on the
   ways in which the request is handled at a server.  As a result,
   requests with caller preferences SHOULD be integrity-protected with
   the sips mechanism specified in RFC 3261, Section 26.

   Processing of caller preferences requires set operations and searches
   which can require some amount of computation.  This enables a DOS
   attack whereby a user can send requests with substantial numbers of
   caller preferences, in the hopes of overloading the server.  To
   counter this, servers SHOULD reject requests with too many rules.  A
   reasonable number is around 20.

12.  IANA Considerations

   This specification registers three new SIP header fields, according
   to the process of RFC 3261 [1].

   The following is the registration for the Accept-Contact header
   field:

   RFC Number: RFC 3841

   Header Field Name: Accept-Contact

   Compact Form: a

   The following is the registration for the Reject-Contact header
   field:

   RFC Number: RFC 3841

   Header Field Name: Reject-Contact

   Compact Form: j

   The following is the registration for the Request-Disposition header
   field:

   RFC Number: RFC 3841

   Header Field Name: Request-Disposition

   Compact Form: d

13.  Acknowledgments

   The initial set of media feature tags used by this specification were
   influenced by Scott Petrack's CMA design.  Jonathan Lennox, Bob
   Penfield, Ben Campbell, Mary Barnes, Rohan Mahy, and John Hearty
   provided helpful comments.  Graham Klyne provided assistance on the
   usage of RFC 2533.

14.  References

14.1.  Normative References

   [1]   Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, A.,
         Peterson, J., Sparks, R., Handley, M., and E. Schooler, "SIP:
         Session Initiation Protocol", RFC 3261, June 2002.

   [2]   Klyne, G., "A Syntax for Describing Media Feature Sets", RFC
         2533, March 1999.

   [3]   Rosenberg, J., Schulzrinne, J., and P. Kyzivat, "Indicating
         User Agent Capabilities in the Session Initiation Protocol
         (SIP)", RFC 3840, August 2004.

   [4]   Bradner, S., "Key words for use in RFCs to Indicate Requirement
         Levels", BCP 14, RFC 2119, March 1997.

   [5]   Roach, A.B., "Session Initiation Protocol (SIP)-Specific Event
         Notification", RFC 3265, June 2002.

   [6]   Donovan, S., "The SIP INFO Method", RFC 2976, October 2000.

   [7]   Rosenberg, J. and H. Schulzrinne, "Reliability of Provisional
         Responses in Session Initiation Protocol (SIP)", RFC 3262, June
         2002.

   [8]   Rosenberg, J., "The Session Initiation Protocol (SIP) UPDATE
         Method", RFC 3311, October 2002.

   [9]   Campbell, B., Ed., Rosenberg, J., Schulzrinne, H., Huitema, C.,
         and D. Gurle, "Session Initiation Protocol (SIP) Extension for
         Instant Messaging", RFC 3428, December 2002.

   [10]  Sparks, R., "The Session Initiation Protocol (SIP) Refer
         Method", RFC 3515, April 2003.

14.2.  Informative References

   [11]  Lennox, J. and H. Schulzrinne, "Call Processing Language
         Framework and Requirements", RFC 2824, May 2000.

   [12]  Rosenberg, J., "Guidelines for Authors of Extensions to the
         Session Initiation Protocol  (SIP)", Work in Progress, November
         2002.

   [13]  Holtman, K., Muntz, A., and T. Hardie, "Media Feature Tag
         Registration Procedure", BCP 31, RFC 2506, March 1999.

15.  Authors' Addresses

   Jonathan Rosenberg
   dynamicsoft
   600 Lanidex Plaza
   Parsippany, NJ  07054
   US

   Phone: +1 973 952-5000
   EMail: jdrosen@dynamicsoft.com
   URI:   http://www.jdrosen.net


   Henning Schulzrinne
   Columbia University
   M/S 0401
   1214 Amsterdam Ave.
   New York, NY  10027
   US

   EMail: schulzrinne@cs.columbia.edu
   URI:   http://www.cs.columbia.edu/~hgs


   Paul Kyzivat
   Cisco Systems
   1414 Massachusetts Avenue
   BXB500 C2-2
   Boxboro, MA  01719
   US

   EMail: pkyzivat@cisco.com

16.  Full Copyright Statement

   Copyright (C) The Internet Society (2004).  This document is subject
   to the rights, licenses and restrictions contained in BCP 78, and
   except as set forth therein, the authors retain all their rights.

   This document and the information contained herein are provided on an
   "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
   OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET
   ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED,
   INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE
   INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
   WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Intellectual Property

   The IETF takes no position regarding the validity or scope of any
   Intellectual Property Rights or other rights that might be claimed to
   pertain to the implementation or use of the technology described in
   this document or the extent to which any license under such rights
   might or might not be available; nor does it represent that it has
   made any independent effort to identify any such rights.  Information
   on the procedures with respect to rights in RFC documents can be
   found in BCP 78 and BCP 79.

   Copies of IPR disclosures made to the IETF Secretariat and any
   assurances of licenses to be made available, or the result of an
   attempt made to obtain a general license or permission for the use of
   such proprietary rights by implementers or users of this
   specification can be obtained from the IETF on-line IPR repository at
   http://www.ietf.org/ipr.

   The IETF invites any interested party to bring to its attention any
   copyrights, patents or patent applications, or other proprietary
   rights that may cover technology that may be required to implement
   this standard.  Please address the information to the IETF at ietf-
   ipr@ietf.org.

Acknowledgement

   Funding for the RFC Editor function is currently provided by the
   Internet Society.