SIP (Session Initiation Protocol) Usage of the Offer/Answer Model
Takuya Sawada
KDDI Corporation
3-10-10, Iidabashi, Chiyoda-ku, Tokyo, Japan
KDDI Corporation3-10-10, Iidabashi, Chiyoda-kuTokyoJapantu-sawada@kddi.comCisco Systems, Inc.1414 Massachusetts AvenueBoxboroughMA01719USApkyzivat@cisco.com
Transport
Sipping
The Session Initiation Protocol (SIP) utilizes the offer/answer
model to establish and update multimedia sessions using the Session
Description Protocol (SDP). The description of the offer/answer
model in SIP is dispersed across multiple RFCs. This document
summarizes all the current usages of the offer/answer model in SIP
communication.
SIP utilizes the offer/answer model to establish and update
sessions. The rules to govern the offer/answer behaviors in SIP are
described in the several RFCs. (,
, ,
and .)
The primary purpose of this document is to describe all forms of
SIP usage of the offer/answer model in one document to help the
readers to fully understand it. Also, this document tries to
incorporate the results of the discussions on the controversial
issues to avoid repeating the same discussions later.
This document is not intended to make normative changes. Rather, it
makes the remaining open issues clear and leaves them for further
study.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in
this document are to be interpreted as described in RFC 2119
.
This document only uses these key words when referencing normative
statements in existing RFCs.
The offer/answer model itself is independent from the higher layer
application protocols which utilize it. SIP is one of the
applications using the offer/answer model. defines the
offer/answer model, but does not specify which SIP messages should
convey an offer or an answer. This should be defined in the SIP
core and extensions RFCs.
In theory, any SIP message can include a session description in its
body. But a session description in a SIP message is not necessarily
an offer or an answer. Only certain session description usages that
conform to the rules described in standards-track RFCs can be
interpreted as an offer or an answer. The rules for how to handle
the offer/answer model are currently defined in several RFCs.
The offer/answer model defines a mechanism for update of sessions.
In SIP, a dialog is used to associate an offer/answer exchange with
the session which it is to update. In other words, only the
offer/answer exchange in the SIP dialog can update the session
which is managed by that dialog.
Currently, the rules on the offer/answer model are defined in
, ,
, and .
In these RFCs, only the six patterns shown in Table 1 are defined for
exchanging an offer and an answer with SIP messages.
Note that an offer/answer exchange initiated by an INVITE request
must follow exactly one of the patterns 1, 2, 3, 4. When an initial
INVITE causes multiple dialogs due to forking, an offer/answer
exchange is carried out independently in each distinct dialog. When
an INVITE request contains no offer, only pattern 2 or pattern 4
apply. 'The first reliable non-failure message' must have an offer
if there is no offer in the INVITE request. This means that UA
which receives the INVITE request without an offer must include an
offer in the first reliable response with 100rel extension. If no
reliable provisional response has been sent, the UAS must include
an offer when sending 2xx response.
In pattern 3, the first reliable provisional response may or may
not have an answer. When a reliable provisional response contains a
session description, and is the first to do so, then that session
description is the answer to the offer in the INVITE request. The
answer can not be updated, and a new offer can not be sent in a
subsequent reliable response for the same INVITE transaction.
In pattern 5, a PRACK request can contain an offer only if the
reliable response which it acknowledges contains an answer to the
previous offer/answer exchange.
NOTE: It is legal to have UPDATE/2xx exchanges without
offer/answer exchanges (pattern 6). However when re-INVITEs
are sent for non-offer/answer purposes, an offer/answer
exchange is required. In that case the prior SDP will
typically be repeated.
There may be ONLY ONE offer/answer negotiation in progress for a
single dialog at any point in time.
explains how to
ensure this. When an INVITE results in multiple dialogs each has a
separate offer/answer negotiation.
NOTE: This is when using a Content-Disposition of "session".
There may be a second offer/answer negotiation in progress
using a Content-Disposition of "early-session"
. That is
not addressed by this draft.
In Table 1, '1xx-rel' corresponds to the reliable provisional
response which contains the 100rel option defined in .
The 'Ini' column shows the ability to exchange the offer/answer to
initiate the session. 'Y' indicates that the pattern can be used in
the initial offer/answer exchange, while 'N' indicates that it can
not. Only the initial INVITE transaction can be used to exchange
the offer/answer to establish a multimedia session.
The 'Est' column shows the ability to update the established
session.
The 'Early' column indicates which patterns may be used to modify
the established session in an early dialog. There are two ways to
exchange a subsequent offer/answer in an early dialog.
It is not entirely clear how to reject an offer when it is
unacceptable, and some methods do not allow explicit rejection of
an offer. For each of the patterns in Table 1, Table 2 shows how to
reject an offer.
When a UA receives an INVITE request with an unacceptable offer, it
should respond with a 488 response, preferably with Warning header
field indicating the reason of the rejection, unless another
response code is more appropriate to reject it. (Pattern 1 and
Pattern 3)
When a UA receives an UPDATE request with an offer which it can not
accept, it should respond with a 488 response preferably with
Warning header field indicating the reason of the rejection, unless
another response code is more appropriate to reject it. (Pattern 6)
When a UA receives a PRACK request with an offer which it can not
accept, it may respond with a 200 response with a syntactically
correct session description. This may optionally be followed by an
UPDATE request to rearrange the session parameters if both ends
support the UPDATE method. Alternatively the UA may terminate the
dialog and send an error response to the INVITE request. The
validity and consequences of a 488 response to PRACK is an open
issue which is discussed in . (Pattern 5)
When a UA receives a response with an offer which it can not accept,
the UA does not have a way to reject it explicitly. Therefore, a UA
should respond to the offer with the correct session description
and rearrange the session parameters by initiating a new
offer/answer exchange, or alternatively terminate the session.
(Pattern 2 and Pattern 4) When initiating a new offer/answer, a UA
should take care not to cause an infinite offer/answer loop.
As previously stated, a session description in a SIP message is not
necessarily an offer or an answer. For example, SIP can use a
session description to describe capabilities apart from
offer/answer exchange. Examples of this are 200 OK responses for
OPTIONS and 488 responses for INVITE.
The INVITE method provides the basic procedure for offer/answer
exchange in SIP. Without the 100rel option, the rules are simple as
described in . If an INVITE request includes a session
description, pattern 1 is applied and if an INVITE request does not
include a session description, pattern 2 is applied.
With 100rel, pattern 3 and pattern 4 are added and this complicates
the rules. An INVITE request may cause multiple responses. Note
that even if both UAs support the 100rel extension, not all the
provisional responses may be sent reliably. Note also that a
reliable provisional response is allowed without a session
description if the UAS does not wish to send the answer yet. An
unreliable provisional response may include a session description
in the body if the UAS has not sent a reliable response, but its
session description is neither an offer nor an answer. All the
session descriptions in the unreliable responses to the INVITE
request must be identical to the answer which is included in the
reliable response. A session description in an unreliable response
that precedes a reliable response can be considered a "preview" of
the answer that will be coming, and hence may be treated like an
answer until the actual one arrives.
NOTE: This "preview" session description rule applies to a
single offer/answer exchange. In parallel offer/answer
exchanges (caused by forking) a UA may obviously receive a
different "preview" of an answer in each dialog. UAs are
expected to deal with this.
Although RFC 3261 says a UA should accept media once an INVITE with
an offer has been sent, in many cases, an answer (or, at least a
preview of it) is required in order for media to be accepted. Two
examples of why this might be required are:
To avoid receiving media from undesired sources, some User
Agents assume symmetric RTP will be used, ignore all incoming
media packets until an address/port has been received from the
other end, and then use that address/port to filter incoming
media packets.
In some networks, an intermediate node must authorize a media
stream before it can flow and requires a confirming answer to
the offer before doing so.
Therefore, a UAS should send an SDP answer reliably (if possible)
before it starts sending media. And, if neither the UAC nor the UAS
support 100rel, the UAS should send a preview of the answer before
it starts sending media.
When a UAC includes an SDP body in the INVITE request as an offer,
it expects the answer to be received with one of the reliable
responses. Other than that, no offer/answer exchanges can occur in
the messages within the INVITE transaction.
For example, in Figure 1, only the SDP in F6 is the answer. The SDP
in the non-reliable response (F2) is the preview of the answer and
must be the same as the answer in F6. Receiving F2, the UAC should
act as if it receives the answer. However, offer/answer exchange is
not completed yet and the UAC must not send a new offer until it
receives the same SDP in the first reliable response, which is the
real answer. After sending the SDP in F6, the UAS must prepare to
receive a new offer from the UAC with an UPDATE request or a PRACK
request.
The UAS does not include SDP in responses F9 and F12. However, the
UAC should prepare to receive SDP bodies in F9 and/or F12, and just
ignore them, to handle a peer that does not conform to the
recommended implementation.
When a UAC does not include an SDP body in the INVITE request, it
expects the offer to be received with the first reliable response.
The UAC will send the answer in the request to acknowledge the
response, i.e. PRACK or ACK request of the reliable response. Other
than that, no offer/answer exchanges can occur in the messages
within the INVITE transaction.
NOTE: The UAS should not include SDP in the responses F6 and
F9. However, the UAC should prepare to receive SDP bodies in
F6 and/or F9, and just ignore them to handle a peer that does
not conform to the recommended implementation.
Note that in the case that the UAC needs to prompt the user to
accept or reject the offer, the reliable provisional response with
SDP as an offer (pattern 4) can result in the retransmission until
the PRACK request can be sent. The UAC should take care to avoid
this situation when it sends the INVITE request without SDP.
When both UAs support the 100rel extension, they can update the
session in the early dialog once the first offer/answer exchange
has been completed.
From a UA sending an INVITE request:
A UA can send an UPDATE request with a new offer if both ends
support the UPDATE method. Note that if the UAS needs to prompt the
user to accept or reject the offer, the delay can result in
retransmission of the UPDATE request.
A UA can send a PRACK request with a new offer only when
acknowledging the reliable provisional response carrying the answer
to an offer in the INVITE request. Compared to using the UPDATE
method, using PRACK can reduce the number of messages exchanged
between the UAs. However, to avoid problems or delays caused by
PRACK offer rejection, the UA is recommended to send a PRACK
request only when it has strong reasons to expect the receiver will
accept it. For example, the procedure used in precondition
extension is a case where a PRACK request should be used for
updating the session status in an early dialog. Note also that if a
UAS needs to prompt the user to accept or reject the offer, the
delay can result in retransmission of the PRACK request.
From a UA receiving an INVITE request:
A UA can send an UPDATE request with a new offer if both ends
support the UPDATE method. A UAS can not send a new offer in the
reliable provisional response, so the UPDATE method is the only
method for a UAS to update an early session.
Both the re-INVITE and UPDATE methods can be used in an established
dialog to update the session.
The UPDATE method is simpler and can save at least one message
compared with the INVITE method. But both ends must support the
UPDATE method for it to be used.
The INVITE method needs at least three messages to complete but no
extensions are needed. Additionally, the INVITE method allows the
peer to take time to decide whether it will accept a session update
or not by sending provisional responses. That is, re-INVITE allows
the UAS to interact with the user at the peer, while UPDATE needs
to be answered automatically by the UAS. It is noted that re-INVITE
should be answered immediately unless such a user interaction is
needed. Otherwise, some 3pcc flows will break.
In , the following restrictions are defined with regard
to sending a new offer.
"At any time, either agent MAY generate a new offer that updates
the session. However, it MUST NOT generate a new offer if it
has received an offer which it has not yet answered or rejected.
It MUST NOT generate a new offer if it has generated a prior
offer for which it has not yet received an answer or a
rejection."
Assuming that the above rules are guaranteed, there seem to be two
possible 'exceptional' cases to be considered in SIP offer/answer
usage: the 'message crossing' case, and the 'glare' case. One of
the reasons why the usage of SIP methods to exchange offer/answer
needs to be carefully restricted in the RFCs is to ensure that the
UA can detect and handle appropriately the 'exceptional' cases to
avoid incompatible behavior.
When message packets cross in the transport network, an offer may
be received before the answer for the previous offer/answer
exchange, as shown in Figure 3. In such a case, UA A must detect
that the session description SDP-2 is not the answer to offer1.
Because of the restrictions on placement of offers and answers
(summarized in Table 1) there are a limited number of valid
exchanges of messages that may lead to this message crossing case.
These are enumerated in Table 3. (This table only shows messages
containing offers or answers. There could be other messages,
without session descriptions, which are not shown.)
There is a variant, shown in Figure 4, which is dependent on an
INVITE (Mx) that contains no offer. This case should be extremely
rare - it is easily avoided by delaying Mx until answer1 is
received. It adds another possibility to Table 3.
Table 3 shows that there are only two ambiguous cases when an
answer is expected and an arriving message M2 containing SDP could
be either the expected answer or an offer. These are a reliable 1xx
response to an INVITE, or an UPDATE.
When message M2 is an UPDATE request or a (re)INVITE request, then
message M1 must also have been an UPDATE or INVITE. There may have
been message crossing, or not. If not then it is a glare case.
Either way, the remedy is for UA A to reject message M2 with a 491
response with Retry-After header field.
When M2 is a reliable provisional response or a successful final
response, and M1 was an UPDATE, then SDP-2 cannot be the expected
answer1. In this case, since UA A can not reject offer2 in reliable
response M2, it is recommended that it wait for answer1 before
sending a PRACK request with the answer to offer2. Note that this
case only occurs when UA A, while waiting for an answer, sends an
INVITE request without session description.
When M2 is a PRACK request Table 3 shows that it cannot be an offer
out of order, so UA A may infer SDP-2 is an answer.
Table 4 summarizes the discussions above.
When both ends in a dialog send a new offer at nearly the same time,
as described in Figure 5, a UA may receive a new offer before it
receives the answer to the offer it sent. This case is usually
called a 'glare' case.
When offer2 is in an UPDATE request or (re-)INVITE request, it must
be rejected with a 491 response.
When offer2 is in a PRACK request (within the current rules, only
possible if offer1 is in an UPDATE request), the PRACK may be
accepted with 200 or may be rejected with a 491 response. A 491
response is valid to satisfy the offer/answer model but it may
delay the completion of the reliable response transfer mechanism or,
in worst case, may result in the failure to complete the SIP
transaction because there is no clear retry rule when a PRACK
request is rejected with a 491 response. To avoid this glare
condition, UA A should not send an offer if it has already sent a
reliable provisional response containing an answer to a previous
offer and has not received the corresponding PRACK request.
To avoid a glare condition involving an offer in a response, when
UA A has sent a (re)INVITE request without session description, it
should not send an offer until it has received an offer in a
reliable response to the (re)INVITE, and sent an answer to that
offer.
While and give some guidance, questions remain
about exactly what should be included in an offer or answer. This
is especially a problem when the common "hold" feature has been
activated, and when there is the potential for a multimedia call.
Details of behavior depend on the capabilities and state of the
User Agent. The kinds of recommendations that can be made are
limited by the model of device capabilities and state that is
presumed to exist.
This section focuses on a few key aspects of offers and answers
that have been identified as troublesome, and will consider other
aspects to be out of scope. This section considers:
choice of supported media types and formats to include and
exclude
hold and resume of media
The following are out of scope for this document:
NAT traversal and ICE
specific codecs and their parameters
the negotiation of secure media streams
grouping of media streams
preconditions
A UA should send an offer that indicates what it, and its user, are
interested in using/doing at that time, without regard for what the
other party in the call may have indicated previously. This is the
case even when the offer is sent in response to an INVITE or re-
INVITE that contains no offer. (However in the case of re-INVITE
the constraints of RFCs 3261 and 3264 must be observed.)
A UA should send an answer that includes as close an approximation
to what the UA and its user are interested in doing at that time,
while remaining consistent with the offer/answer rules of
and other RFCs.
NOTE: "at that time" is important. The device may permit the
user to configure which supported media are to be used by
default.
In some cases a UA may not have direct knowledge of what it is
interested in doing at a particular time. If it is an intermediary
it may be able to delegate the decision. In the worst case it may
apply a default, such as assuming it wants to use all of its
capabilities.
When a UAC sends an initial INVITE with an offer, it has complete
freedom to choose which media type(s) and media format(s) (payload
types in the case of RTP) it should include in the offer.
The media types may be all or a subset of the media the UAC is
capable of supporting, with the particular subset being determined
by the design and configuration (e.g., via )
of the UAC combined with input from the user interface of the UAC.
The media formats may be all or a subset of the media formats the
UAC is capable of supporting for the corresponding media type, with
the particular subset being determined by the design and
configuration of the
UAC combined with input from the user interface of the UAC.
Including all supported media formats will maximize the possibility
that the other party will have a supported format in common. But
including many can result in an unacceptably large SDP body.
When a UAS has received an initial INVITE without an offer, it must
include an offer in the first reliable response to the INVITE. It
has largely the same options as when sending an initial INVITE with
an offer, but there are some differences. The choice may be
governed by both static (default) selections of media types as well
as dynamic selections made by a user via interaction with the
device while it is alerting.
NOTE: The offer may be sent in a provisional response, before
the user of the device has been alerted and had an opportunity
to select media options for the call. In this case the UAS
cannot include any call-specific options from the user of the
device. If there is a possibility that the user of the device
will wish to change what is offered before answering the call,
then special care should be taken. If PRACK and UPDATE are
supported by caller and callee then an initial offer can be
sent reliably, and changed with an UPDATE if the user desires
a change. If PRACK and UPDATE are not supported then the
initial offer cannot be changed until the call is fully
established. In that case either the offer should be delayed
until the 200 is sent, or else the offer should include the
minimum set of media the user is able to select.
When a UAS receives an initial INVITE with an offer, what media
lines the answer may contain is constrained by . The
answer must contain the same number of m-lines as the offer, and
they must contain the same media types. Each media line may be
accepted, by including a non-zero port number, or rejected by
including a zero port number in the answer. The media lines that
are accepted should typically be those that would have been offered
had the INVITE not contained an offer, excluding those not offered.
The media formats the answer may contain are constrained by
. For each accepted m-line in the answer, there must be at
least one media format in common with the corresponding m-line of
the offer. The UAS may also include other media formats it is able
to support at this time. However there is little benefit to
including added types.
If the UAS does not wish to indicate support for any of the media
types in a particular media line of the offer it must reject the
corresponding media line, by setting the port number to zero.
When a UAC has sent an initial INVITE without an offer, and then
receives a response with the first offer, it should answer in the
same way as a UAS receiving an initial INVITE with an offer.
The guidelines above ( and
through
)
apply, but constraints in must also be followed. The
following are of particular note because they have proven
troublesome:
The number of m-lines may not be reduced in a subsequent offer.
Previously rejected media streams must remain, or be reused to
offer the same or a different stream. (Section 6 of .)
In the o-line, only the version number may change, and if it
changes it must increment by one from the one previously sent as
an offer or answer. (Section 8 of .) If it doesn't
change then the entire SDP body must be identical to what was
previously sent as an offer or answer. Changing the o-line,
except version number value, during the session is an error case.
The behavior when receiving such a non-compliant offer/answer
SDP body is implementation dependent. If a UA needs to negotiate
a 'new' SDP session, it should use the INVITE/Replaces method.
In the case of RTP, the mapping from a particular dynamic
payload type number to a particular codec within that media
stream (m-line) must not change for the duration of the session.
(Section 8.3.2 of .)
NOTE: This may be impossible for a B2BUA to follow in some
cases (e.g. 3pcc transfer) if it does not terminate media.
When the new offer is sent in response to an offerless (re)INVITE,
all codecs supported by the UA are to be included, not just the
ones that were negotiated by previous offer/answer exchanges. The
same is true for media types - so if UA A initially offered audio
and video to UA B, and they end up with only audio, and UA B sends
an offerless (re)INVITE to UA A, A's resulting offer should re-
attempt video, by reusing the zeroed m-line used previously.
NOTE: The behavior above is recommended, but it is not always
achievable - for example in some interworking scenarios. Or,
the offerer may simply not have enough resources to offer
"everything" at that point. Even if the UAS is not able to
offer any other SDP that the one currently being used, it
should not reject the re-INVITE. Instead, it should generate
an offer with the currently used SDP with o- line unchanged.
specifies (non-normatively) that "hold" should be
indicated in an established session by sending a new offer
containing "a=sendonly" for each media stream to be held. An
answerer is then to respond with "a=recvonly" to acknowledge that
the hold request has been understood.
Note that the use of sendonly/recvonly is not limited to hold.
These may be used for other reasons, such as devices that are only
capable of sending or receiving. So receiving an offer with
"a=sendonly" must not be treated as a certain indication that the
offerer has placed the media stream on hold.
This model is based on an assumption that the UA initiating the
hold will want to play Music on Hold, which is not always the case.
A UA may, if desired, initiate hold by offering "a=inactive" if it
does not intend to transmit any media while in hold status.
The rules of constrain what may be in an answer when
the offer contains "sendonly", "recvonly", or "inactive" in an a=
line. But they do not constrain what must be in a subsequent offer.
The General Principle for Constructing Offers and Answers
() is important here.
The initiation of "hold" is a local
action. It should reflect the desired state of the UA. It then
affects what the UA includes in offers and answers until the local
state is reset.
The receipt of an offer containing "a=sendonly" or "a=inactive" and
the sending of a compatible answer should not change the desired
state of the recipient. However, a UA that has been "placed on
hold" may itself desire to initiate its own hold status, based on
local input.
If UA2 has previously been "placed on hold" by UA1, via receipt of
"a=sendonly", then it may initiate its own hold by sending a new
offer containing "a=sendonly" to UA1. Upon receipt of that, UA1
will answer with "a=inactive" because that is the only valid answer
that reflects its desire not to receive media.
Once in this state, to resume a two way exchange of media each side
must reset its local hold status. If UA1 is first to go off hold it
will then send an offer with "a=sendrecv". The UA2 will respond
with its desired state of "a=sendonly" because that is a permitted
response. When UA2 desires to also resume, it will send an offer
with "a=sendrecv". In this case, because UA1 has the same desire it
will respond with "a=sendrecv". In the same case, when UA2 receives
the offer with "a=sendrecv", if it has decided it wants to reset
its local hold but has not yet signaled the intent, it may send
"a=sendrecv" in the answer.
If UA2 has been "placed on hold" by UA1 via receipt of "a=inactive",
and subsequently wants to initiate its own hold, also using
"a=inactive", it need not send a new offer, since the only valid
response is "a=inactive" and that is already in effect. However,
its local desired state will now be either "inactive" or
"a=sendonly". This affects what it will send in future offers and
answers.
If a UA has occasion to send another offer in the session, without
any desire to change the hold status (e.g. in response to a re-
INVITE without an offer, or when sending a re-INVITE to refresh the
session timer) it should follow the General Principle for
Constructing Offers and Answers ().
If it previously
initiated a "hold" by sending "a=sendonly" or "a=inactive" then it
should offer that again. If it had not previously initiated "hold"
then it should offer "a=sendrecv", even if it had previously been
forced to answer something else. Without this behavior it is
possible to get "stuck on hold" in some cases, especially when a
third-party call controller is involved.
specifies that an agent MUST be capable of receiving
SDP with a connection address of 0.0.0.0, in which case it means
that neither RTP nor RTCP should be sent to the peer.
If a UA generates an answer to the offer received with c=0.0.0.0,
the direction attribute of the accepted media stream in the answer
must be based on direction attribute of the offered stream and
rules specified in RFC 3264 to form the a-line in the answer.
c=0.0.0.0 has no special meaning for the direction attribute of the
accepted stream in the answer.
This document clarifies the offer/answer usage in SIP and
summarizes the correct or recommended behaviors along with the
existing RFCs. To create any new normative behaviors beyond these
RFCs is not the intent of this document.
However, through the scrutiny of the offer/answer model in SIP,
some issues are found to be unresolved within the current set of
RFCs. Those remaining issues are described in this section mainly
for further study.
As stated in and
, it is recommended that an
offer not be sent in a PRACK request unless UAC has strong reasons
to assume the receiver will accept it. Even so, there may be cases
when the UAS has to reject the offer for some reason. The current
RFCs do not provide a way to reject the offer and at the same time
to indicate that the PRACK adequately acknowledged the reliable
response. It is unclear whether a non-200 response can still
indicate an acknowledgement of the reliable response.
Several ideas were presented to resolve this issue, such as sending
2xx PRACK response without SDP to reject the offer, or sending SDP
with a decreased version value in the o-line. Some of the
candidates may also be adapted as a way to reject an unacceptable
offer in a response. Anyway, those proposals violate the current
rules and lose backward compatibility to some extent (e.g. section
5 of ).
It is beyond the scope of this document and remains
for further study.
The 488 response is another proposed solution; however the validity
and consequences of a 488 response to PRACK is an open issue.
Because the 488 response may be sent by a proxy, the UAC cannot
assume the reliable transaction has been adequately acknowledged.
If a 488 response is received, the UAC should ensure acknowledgment
of the reliable response by sending a new PRACK with the offer
removed or modified based upon the received 488 response. If the
488 response is sent by UAS (open issue), it cannot assume that the
UAC thinks that the reliable transaction has been adequately
acknowledged even though the UAS may treat otherwise (open issue).
If a 488 response is sent by UAS, the UAC should accommodate
receiving the altered PRACK with higher CSeq without expecting it
to trigger a 481 response (open issue).
NOTE: Deprecation of the usage of offer in PRACK may be
another solution. As the precondition mechanism specification
explicitly shows a usage of sending offer in PRACK, its
deprecation could cause backward compatibility issues.
When a re-INVITE transaction fails, the dialog remains with the
session bound to it. The issue here is: what is the session status
if an offer/answer exchange has been completed (if a session
description has been sent in a reliable provisional response to the
re-INVITE request), or if subsequent offer/answer exchanges have
taken place (using UPDATE or PRACK transactions), before the re-
INVITE transaction is terminated with a final error response
(Figure 6). One option is to take those offer/answer exchanges not
committed yet and to make the session status rollback to the one
before re-INVITE transaction was initiated. Another option is to
take those exchanges committed and to keep the session status as it
is even after re-INVITE fails. There is no clear consensus on which
one is the correct behavior.
There are some cases where it is useful to exchange
offer(s)/answer(s) even before re-INVITE completes. The case of
adding a new media (like adding video to audio only session) which
requires permission from the peer through some user interaction is
one example. Precondition procedures can be another case which may
require several offer/answer exchanges in one re-INVITE transaction.
To make bad things worse, if a new offer from UAC and the final
response to re-INVITE are sent at nearly the same time, the UAS can
not know whether this new offer was sent before or after UAC
received the final failure response (Figure 7). Note that the ACK
request to the failure response is sent hop-by-hop basis, therefore
even after receiving the ACK request, UAS can not make sure that
UPDATE request was sent after the final response had been reached
to the other end.
Sending a new UPDATE request from UAC to synchronize the status
anytime after the re-INVITE fails may be a good option. This
solution, however, requires that the UPDATE method be supported by
both ends and needs care to avoid flapping when each end tries to
advertise their different views of the session status.
The proper handling of this issue is undefined by existing
standards. Resolution is beyond the scope of this document, and
will require a new normative document.
In RFC 3261, it is stated that when an INVITE is sent without an
offer, the first reliable response MUST contain an offer. There was
discussion on whether this rule can be loosened up. There is no
clear explanation why this restriction is defined. However, this
rule will be left as it is, unless the strong necessity to loosen
it up is raised in the future.
RFC 3264, section 8.4, contains procedures for putting a unicast
media stream on hold. Of particular note, it states:
"If the stream to be placed on hold was previously a recvonly
media stream, it is placed on hold by marking it inactive."
of the current
document makes a recommendation for
this case which conflicts with that, and explains why. Some
concerns have been raised that such a recommendation is invalid
because RFC 3264 is normative on this subject.
This document takes the position that Section 8.4 of RFC 3264 is
non-normative, and so may be overridden. It is further recommended
that RFC 3264 be revised to avoid the confusion.
This document recommends against the addition of new offer/answer
methods using SIP. However, it may be necessary to define new
offer/answer exchange methods as SIP extensions evolve. This
section recommends some things that should be taken into
considerations in that case.
New method definitions should define offer/answer usage explicitly
without any ambiguity.
New method definitions should define how to reject an offer where
possible.
New methods must keep backward compatibility.
New methods should take care of how to handle exceptional cases,
message crossing case and glare case.
This document has no actions for IANA.
There are not any security issues beyond the referenced RFCs.
The authors would like to thank Christer Holmberg, Rajeev Seth,
Nataraju A B, Byron Campen and Jonathan Rosenberg for their
thorough reviews and comments. Many of their suggestions and ideas
are incorporated to complete this document.