MMUSIC J. Rosenberg Internet-Draft Cisco Systems Expires: September 7, 2006 March 6, 2006 Interactive Connectivity Establishment (ICE): A Methodology for Network Address Translator (NAT) Traversal for Offer/Answer Protocols draft-ietf-mmusic-ice-07 Status of this Memo By submitting this Internet-Draft, each author represents that any applicable patent or other IPR claims of which he or she is aware have been or will be disclosed, and any of which he or she becomes aware will be disclosed, in accordance with Section 6 of BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt. The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. This Internet-Draft will expire on September 7, 2006. Copyright Notice Copyright (C) The Internet Society (2006). Abstract This document describes a protocol for Network Address Translator (NAT) traversal for multimedia session signaling protocols based on the offer/answer model, such as the Session Initiation Protocol (SIP). This protocol is called Interactive Connectivity Establishment (ICE). ICE makes use of the Simple Traversal of UDP through NAT (STUN), applying its binding discovery, connectivity check and relay usages. Rosenberg Expires September 7, 2006 [Page 1] Internet-Draft ICE March 2006 Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 4 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . 4 3. Overview of ICE . . . . . . . . . . . . . . . . . . . . . . 8 4. Sending the Initial Offer . . . . . . . . . . . . . . . . . 11 5. Receipt of the Offer and Generation of the Answer . . . . . 11 6. Processing the Answer . . . . . . . . . . . . . . . . . . . 12 7. Common Procedures . . . . . . . . . . . . . . . . . . . . . 12 7.1 Gathering Candidates . . . . . . . . . . . . . . . . . . . 12 7.2 Prioritizing the Candidates and Choosing an Active One . . 16 7.3 Encoding Candidates into SDP . . . . . . . . . . . . . . . 18 7.4 Forming Candidate Pairs . . . . . . . . . . . . . . . . . 21 7.5 Ordering the Candidate Pairs . . . . . . . . . . . . . . . 23 7.6 Performing the Connectivity Checks . . . . . . . . . . . . 26 7.7 Sending a Binding Request for Connectivity Checks . . . . 30 7.8 Receiving a Binding Request for Connectivity Checks . . . 31 7.9 Promoting a Candidate to Active . . . . . . . . . . . . . 33 7.10 Learning New Candidates from Connectivity Checks . . . . 34 7.10.1 On Receipt of a Binding Request . . . . . . . . . . 34 7.10.2 On Receipt of a Binding Response . . . . . . . . . . 38 7.11 Subsequent Offer/Answer Exchanges . . . . . . . . . . . 39 7.11.1 Sending of a Subsequent Offer . . . . . . . . . . . 40 7.11.2 Receiving the Offer and Sending an Answer . . . . . 42 7.11.3 Receiving the Answer . . . . . . . . . . . . . . . . 45 7.12 Binding Keepalives . . . . . . . . . . . . . . . . . . . 45 7.13 Sending Media . . . . . . . . . . . . . . . . . . . . . 46 8. Guidelines for Usage with SIP . . . . . . . . . . . . . . . 49 9. Interactions with Forking . . . . . . . . . . . . . . . . . 51 10. Interactions with Preconditions . . . . . . . . . . . . . . 51 11. Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 51 11.1 Basic Example . . . . . . . . . . . . . . . . . . . . . 53 11.2 Advanced Example . . . . . . . . . . . . . . . . . . . . 57 12. Grammar . . . . . . . . . . . . . . . . . . . . . . . . . . 77 13. Security Considerations . . . . . . . . . . . . . . . . . . 79 13.1 Attacks on Connectivity Checks . . . . . . . . . . . . . 79 13.2 Attacks on Address Gathering . . . . . . . . . . . . . . 81 13.3 Attacks on the Offer/Answer Exchanges . . . . . . . . . 82 13.4 Insider Attacks . . . . . . . . . . . . . . . . . . . . 82 13.4.1 The Voice Hammer Attack . . . . . . . . . . . . . . 82 13.4.2 STUN Amplification Attack . . . . . . . . . . . . . 83 14. IANA Considerations . . . . . . . . . . . . . . . . . . . . 83 14.1 candidate Attribute . . . . . . . . . . . . . . . . . . 83 14.2 remote-candidate Attribute . . . . . . . . . . . . . . . 84 14.3 ice-pwd Attribute . . . . . . . . . . . . . . . . . . . 84 15. IAB Considerations . . . . . . . . . . . . . . . . . . . . . 85 15.1 Problem Definition . . . . . . . . . . . . . . . . . . . 85 15.2 Exit Strategy . . . . . . . . . . . . . . . . . . . . . 86 Rosenberg Expires September 7, 2006 [Page 2] Internet-Draft ICE March 2006 15.3 Brittleness Introduced by ICE . . . . . . . . . . . . . 86 15.4 Requirements for a Long Term Solution . . . . . . . . . 87 15.5 Issues with Existing NAPT Boxes . . . . . . . . . . . . 87 16. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 88 17. References . . . . . . . . . . . . . . . . . . . . . . . . . 88 17.1 Normative References . . . . . . . . . . . . . . . . . . 88 17.2 Informative References . . . . . . . . . . . . . . . . . 89 Author's Address . . . . . . . . . . . . . . . . . . . . . . 91 Intellectual Property and Copyright Statements . . . . . . . 92 Rosenberg Expires September 7, 2006 [Page 3] Internet-Draft ICE March 2006 1. Introduction RFC 3264 [4] defines a two-phase exchange of Session Descrption Protocol (SDP) messages [5] for the purposes of establishment of multimedia sessions. This offer/answer mechanism is used by protocols such as the Session Initiation Protocol (SIP) [2]. Protocols using offer/answer are difficult to operate through Network Address Translators (NAT). Because their purpose is to establish a flow of media packets, they tend to carry IP addresses within their messages, which is known to be problematic through NAT [17]. The protocols also seek to create a media flow directly between participants, so that there is no application layer intermediary between them. This is done to reduce media latency, decrease packet loss, and reduce the operational costs of deploying the application. However, this is difficult to accomplish through NAT. A full treatment of the reasons for this is beyond the scope of this specification. Numerous solutions have been proposed for allowing these protocols to operate through NAT. These include Application Layer Gateways (ALGs), the Middlebox Control Protocol [19], Simple Traversal of UDP through NAT (STUN) [16] and its revision [13], the STUN Relay Usage [14], and Realm Specific IP [20] [21] along with session description extensions needed to make them work, such as the Session Description Protocol (SDP) [5] attribute for the Real Time Control Protocol (RTCP) [1]. Unfortunately, these techniques all have pros and cons which make each one optimal in some network topologies, but a poor choice in others. The result is that administrators and implementors are making assumptions about the topologies of the networks in which their solutions will be deployed. This introduces complexity and brittleness into the system. What is needed is a single solution which is flexible enough to work well in all situations. This specification provides that solution for media streams established by signaling protocols based on the offer-answer model. It is called Interactive Connectivity Establishment, or ICE. ICE makes use of STUN and its relay extension, commonly called TURN, but uses them in a specific methodology which avoids many of the pitfalls of using any one alone. 2. Terminology Several new terms are introduced in this specification: Rosenberg Expires September 7, 2006 [Page 4] Internet-Draft ICE March 2006 Agent: As defined in RFC 3264, an agent is the protocol implementation involved in the offer/answer exchange. There are two agents involved in an offer/answer exchange. Peer: From the perspective of one of the agents in a session, its peer is the other agent. Specifically, from the perspective of the offerer, the peer is the answerer. From the perspective of the answerer, the peer is the offerer. Transport Address: The combination of an IP address and port. Local Transport Address: A local transport address is a transport address that has been allocated from the operating system on the host. This includes transport addresses obtained through Virtual Private Networks (VPNs) and transport addresses obtained through Realm Specific IP (RSIP) [20] (which lives at the operating system level). Transport addresses are typically obtained by binding to an interface. m/c line: The media and connection lines in the SDP, which together hold the transport address used for the receipt of media. Derived Transport Address: A derived transport address is a transport address which is derived from a local transport address. The derived transport address is related to the associated local transport address in that packets sent to the derived transport address are received on the socket bound to its associated local transport address. Derived addresses are obtained using protocols like STUN, and more generally, any UNSAF protocol [22]. Reflexive Transport Address: As defined in [13], a transport address learned by a client which identifies that client as seen by another host on an IP network, typically a STUN server. When there is an intervening NAT between the client and the other host, the reflexive transport address represents the binding allocated to the client on the public side of the NAT. Reflexive transport addresses are learned from the MAPPED-ADDRESS attribute in STUN Binding Responses and Allocate Responses [14], and are a type of derived transport address. Server Reflexive Transport Address: A server reflexive transport address is a reflexive address that is reflected off of a server, distinct from the peer, whose address is configured or learned by the client prior to an offer/answer exchange. Rosenberg Expires September 7, 2006 [Page 5] Internet-Draft ICE March 2006 Peer Reflexive Transport Address: A peer reflexive transport address is a reflexive address that is reflected off of the peer. Peer reflexive transport addresses are learned by connectivity checks. Relayed Transport Address: A transport address that terminates on a server, and is forwarded towards the client. The STUN Allocate Request can be used to obtain a relayed transport address, for example. Associated Local Transport Address: When a peer sends a packet to a transport address, the associated local transport address is the local transport address at which those packets will actually arrive. For a local transport address, its associated local transport address is the same as the local transport address itself. For reflexive and relayed transport addresses, however, they are not the same. The associated local transport address is the one from which the reflexive or relayed transport was derived. Candidate: A sequence of transport addresses that form an atomic set for usage with a particular media session. Here, atomic means that all of transport addresses in the candidate need to work before the candidate will be used for actual media transport. In the case of RTP, there can be one or more transport addresses per candidate. In the most common case, there are two - one for RTP, and another for RTCP. If the agent doesn't use RTCP, there would be just one. If Generic Forward Error Correction (FEC) [18] is in use, there may be more than two. The transport addresses that compose a candidate are all of the same type - local, server reflexive, peer reflexive or relayed. Local Candidate: A candidate whose transport addresses are local transport addresses. Server Reflexive Candidate: A candidate whose transport addresses are server reflexive transport addresses. Peer Reflexive Candidate: A candidate whose transport addresses are peer reflexive transport addresses. Relayed Candidate: A candidate whose transport addresses are relayed transport addresses. Generating Candidate: The candidate from which a peer reflexive candidate is derived. Rosenberg Expires September 7, 2006 [Page 6] Internet-Draft ICE March 2006 Active Candidate: The candidate that is in use for exchange of media. This is the one that an agent places in the m/c line of an offer or answer. Candidate ID: An identifier for a candidate. Component: When a media stream, and as a consequence, its candidate, require several IP addresses and ports to work atomically, each of the constituent IP addresses and ports represents a component of that media stream. For example, RTP-based media streams typically have two components - one for RTP, and one for RTCP. Component ID: An integer, starting with one within each candidate and incrementing by one for each component, which identifies the component. Transport Address ID (tid): An identifier for a transport address, formed by concatenating the candidate ID with the component ID, separated by a "colon". Candidate Pair: The combination of a candidate from one agent along with a candidate from its peer. Native Candidate: From the perspective of each agent, the candidate in a candidate pair which represents a set of addresses obtained by that agent. Remote Candidate: From the perspective of each agent, the candidate in a candidate pair which represents the set of addresses obtained by that agents peer. Transport Address Pair: The combination of the transport address for one component of a candidate with the transport address of the same component for the matching candidate in a candidate pair. Transport Address Pair ID: An identifier for a transport address pair. Formed by concatenating the native transport address ID with the remote transport address ID, separated by a "colon". Matching Transport Address Pair: When a STUN Binding Request is received on a local transport address, the matching transport address pair is the transport address pair whose connectivity is being checked by that Binding Request. Candidate Pair Priority Ordering: An ordering of candidate pairs based on a combination of the qvalues of each candidate and the candidate IDs of each candidate. Rosenberg Expires September 7, 2006 [Page 7] Internet-Draft ICE March 2006 Candidate Pair Check Ordering: An ordering of candidate pairs that is similar to the candidate pair priority ordering, except that the active candidate appears at the top of the list, regardless of its priority. Transport Address Pair Check Ordering: An ordering of transport address pairs that determines the sequence of connectivity checks performed for the pairs. Transport Address Pair Count: The number of transport address pairs in a candidate pair. This is equal to the minimum of the number of transport addresses in the native candidate and the number of transport addresses in the remote candidate. 3. Overview of ICE ICE makes the fundamental assumption that clients exist in a network of segmented connectivity. This segmentation is the result of a number of addressing realms in which a client can simultaneously be connected. We use "realms" here in the broadest sense. A realm is defined purely by connectivity. Two clients are in the same realm if, when they exchange the addresses each has in that realm, they are able to send packets to each other. This includes IPv6 and IPv4 realms, which actually use different address spaces, in addition to private networks connected to the public Internet through NAT. The key assumption in ICE is that a client cannot know, apriori, which address realms it shares with any peer it may wish to communicate with. Therefore, in order to communicate, it has to try connecting to addresses in all of the realms. Rosenberg Expires September 7, 2006 [Page 8] Internet-Draft ICE March 2006 Agent A STUN Servers Agent B |(1) Gather Addresses | | |-------------------->| | |(2) Offer | | |------------------------------------------>| | |(3) Gather Addresses | | |<--------------------| |(4) Answer | | |<------------------------------------------| |(5) STUN Check | | |<------------------------------------------| |(6) STUN Check | | |------------------------------------------>| |(7) Media | | |<------------------------------------------| |(8) Media | | |------------------------------------------>| |(9) Offer | | |------------------------------------------>| |(10) Answer | | |<------------------------------------------| Figure 1 The basic flow of operation for ICE is shown in Figure 1. Before the offerer establishes a session, it obtains local transport addresses from its operating system on as many interfaces as it has access to. These interfaces can include IPv4 and IPv6 interfaces, in addition to Virtual Private Network (VPN) interfaces or ones associated with RSIP. It then obtains transport addresses for the media from each interface. Though ICE can support any type of transport protocol, this specification only defines mechanisms for UDP. In addition, the agent obtains server reflexive and relayed transport addresses. These are usually obtained through a single STUN Allocate request, which provides both. These requests are paced at a fixed rate in order to limit network load and avoid NAT overload. The local, server reflexive and relayed transport addresses are formed into candidates, each of which represents a possible set of transport addresses that might be viable for a media stream. Each candidate is listed in a set of a=candidate attributes in the offer. Each candidate is given a priority. Priority is a matter of local policy, but typically, lowest priority would be given to relayed transport addresses. Each candidate is also assigned a distinct ID, called a candidate ID. The agent will choose one of its candidates as its active candidate Rosenberg Expires September 7, 2006 [Page 9] Internet-Draft ICE March 2006 for inclusion in the connection and media lines in the offer. Media can be sent to this candidate immediately following its validation. Media can also be sent to a candidate that is not active but has been validated. Media is not sent without validation in order to avoid denial-of-service attacks. In particular, without ICE, an offerer can send an offer to another agent, and list the IP address and port of a target in the offer. If the agent is an automata that answers a call automatically, it will do so and then proceed to send media to the target. This provides substantial packet amplifications. ICE fixes this by requiring that an agent never send media packets unless it has sent a STUN message towards the target of the RTP packets, and received a reply from that target Section 7.13. The offer is then sent to the answerer. This specification does not address the issue of how the signaling messages themselves traverse NAT. It is assumed that signaling protocol specific mechanisms are used for that purpose. The answerer follows a similar process as the offerer followed; it obtains addresses from local interfaces, obtains derived transport addresses from those, and then groups them into candidates for inclusion in a=candidate attributes in the answer. It picks one candidate as its active candidate and places it into the m/c line in the answer. Once the offer/answer exchange has completed, both agents pair up the candidates, and then determine an ordered set of transport address pairs. This ordering is based primarily on the priority of the candidates, with the exception of the active candidate, whose addresses are at the top of the list. Both agents start at the top of this list, beginning a connectivity check for that transport address pair. At a fixed interval, checks for the next transport address on the list begin. This results in a pacing of the connectivity checks. These connectivity checks are performed through peer-to-peer STUN requests, sent from one agent to the other. In addition to pacing the checks out at regular intervals, the offerer will generate a connectivity check for a transport address pair when it receives one from its peer. As soon as the active candidate has been verified by the STUN checks, media can begin to flow. Once a higher priority candidate has been verified by the offerer, it ceases additional connectivity checks, begins using that candidate for media, and sends an updated offer which promotes this higher priority candidate to the m/c-line. That candidate is also listed in a=candidate attributes, resulting in periodic STUN keepalives through the duration of the media session. If an agent receives a STUN connectivity check with a new source IP address and port, or a response to such a check with a new IP address and port indicated in the MAPPED-ADDRESS attribute, this new address might be a viable candidate for the receipt of media. This happens Rosenberg Expires September 7, 2006 [Page 10] Internet-Draft ICE March 2006 when there is a NAT with an address dependent or address and port dependent mapping property [37] between the agents. In such a case, the agents algorithmically construct a new candidate. Like other candidates, connectivity checks begin for it, and if they succeed, its transport addresses can be used for receipt of media by promoting it to the m/c-line. The gathering of addresses and connectivity checks take time. As a consequence, in order to have minimal impact on the call setup time or post-pickup delay for SIP, these offer/answer exchanges and checks happen while the call is ringing. 4. Sending the Initial Offer When an agent wishes to begin a session by sending an initial offer, it starts by gathering transport addresses, as described in Section 7.1. This will produce a set of candidates, including local ones, server reflexive ones, and relayed ones. This process of gathering candidates can actually happen at any time before sending the initial offer. A agent can pre-gather transport addresses, using a user interface cue (such as picking up the phone, or entry into an address book) as a hint that communications is imminent. Doing so eliminates any additional perceivable call setup delays due to address gathering. When it comes time to offer communications, the agent determines a priority for each candidate and identifies the active candidate that will be used for receipt of media, as described in Section 7.2. The next step is to construct the offer message. For each media stream, it places its candidates into a=candidate attributes in the offer and puts its active candidate into the m/c line. The process for doing this is described in Section 7.3. The offer is then sent. 5. Receipt of the Offer and Generation of the Answer Upon receipt of the offer message, the agent checks if the offer contains any a=candidate attributes. If the offer does, the offerer supports ICE. In that case, it starts gathering candidates, as described in Section 7.1, and prioritizes them as described in Section 7.2. This processing is done immediately on receipt of the offer, to prepare for the case where the user should accept the call, or early media needs to be generated. By gathering candidates (and performing connectivity checks) while the user is being alerted to the request for communications, session establishment delays are reduced. Rosenberg Expires September 7, 2006 [Page 11] Internet-Draft ICE March 2006 The agent then constructs its answer, encoding its candidates into a=candidate attributes and including the active one in the m/c-line, as described in Section 7.3. The agent then forms candidate pairs as described in Section 7.4. These are ordered as described in Section 7.5. The agent then begins connectivity checks, as described in Section 7.6. It follows the logic in Section 7.10 on receipt of Binding Requests and responses to learn new candidates from the checks themselves. Transmission of media is performed according to the procedures in Section 7.13. 6. Processing the Answer There are two possible cases for processing of the answer. If the answerer did not support ICE, the answer will not contain any a=candidate attributes. As a result, the offerer knows that it cannot perform its connectivity checks. In this case, it proceeds with normal media processing as if ICE was not in use. The procedures for sending media, described in Section 7.13, MUST be followed however. If the answer contains candidates, it implies that the answerer supports ICE. The offerer then forms candidate pairs as described in Section 7.4. These are ordered as described in Section 7.5. The agent then begins connectivity checks, as described in Section 7.6. It follows the logic in Section 7.10 on receipt of Binding Requests and responses to learn new candidates from the checks themselves. Transmission of media is performed according to the procedures in Section 7.13. 7. Common Procedures This section discusses procedures that are common between offerer and answerer. 7.1 Gathering Candidates An agent gathers candidates when it believes that communications is imminent. For offerers, this occurs before sending an offer (Section 4). For answerers, it occurs before sending an answer (Section 5). Each candidate has one or more components, each of which is associated with a sequence number, starting at 1 for the first component of each candidate, and incrementing by 1 for each additional component within that candidate. These components Rosenberg Expires September 7, 2006 [Page 12] Internet-Draft ICE March 2006 represent a set of transport addresses for which connectivity must be validated. For a particular media stream, all of the candidates SHOULD have the same number of components. The number of components that are needed are a function of the type of media stream. All of the components in a candidate MUST be of the same type - server reflexive, relayed, or local, and obtained from the same server in the case of server reflexive or relayed candidates. For local candidates, each component MUST be obtained from the same interface. For traditional RTP-based media streams, it is RECOMMENDED that there be two components per candidate - one for RTP and one for RTCP. The component with the component ID of 1 MUST be RTP, and the one with component ID of 2 MUST be RTCP. If an agent doesn't implement RTCP, it SHOULD have a single component for the RTP stream (which will have a component ID of 1 by definition). Each component of a candidate has a single transport address. The first step is to gather local candidates. Local candidates are obtained by binding to ephemeral ports on an interface (physical or virtual, including VPN interfaces) on the host. The process for gathering local candidates depends on the transport protocol. Procedures are specified here for UDP. Extensions to ICE that define procedures for other transport protocols MUST specify how local transport addresses are gathered. For each UDP media stream the agent wishes to use, the agent SHOULD obtain a set of candidates (one for each interface) by binding to N ephemeral UDP ports on each interface, where N is the number of components needed for the candidate. For RTP, N is typically two. If a host has K local interfaces, this will result in K candidates for each UDP stream, requiring K*N local transport addresses. Once the agent has obtained local candidates, it obtains candidates with derived transport addresses. The process for gathering derived candidates depends on the transport protocol. Procedures are specified here for UDP. Extensions to ICE that define procedures for other transport protocols MUST specify how derived transport addresses are gathered. Agents which serve end users directly, such as softphones, hardphones, terminal adapters and so on, MUST implement the STUN Binding Discovery usage and SHOULD use it to obtain server reflexive candidates. These devices SHOULD implement the STUN Relay usage, and SHOULD use its Allocate request to obtain both server reflexive and relayed candidates. They MAY implement and MAY use other protocols that provide server reflexive or relayed transport addresses, such as TEREDO [33]. Rosenberg Expires September 7, 2006 [Page 13] Internet-Draft ICE March 2006 The requirement to use the relay Usage is at SHOULD strength to allow for provider variation. If it is not to be used, it is RECOMMENDED that it be implemented and just disabled through configuration, so that it can re-enabled through configuration if conditions change in the future. Agents which represent network servers under the control of a service provider, such as gateways to the telephone network, media servers, or conferencing servers that are targeted at deployment only in networks with public IP addresses MAY use the STUN Binding Discovery usage and relay usage, or other similar protocols to obtain candidates. Why would these types of endpoints even bother to implement ICE? The answer is that such an implementation greatly facilitates NAT traversal for clients that connect to it. The ability to process STUN connectivity checks allows for clients to obtain peer reflexive transport addresses that can be used by the network server to reach them without a relay, even through NATs with restrictive mapping and filtering policies. Furthermore, implementation of the STUN connectivity checks allows for NAT bindings along the way to be kept open. ICE also provides numerous security properties that are independent of NAT traversal, and would benefit any multimedia endpoint. See Section 13 for a discussion on these benefits. Obtaining derived candidates requires transmission of packets which have the effect of creating bindings on NAT devices between the client and the STUN servers. Experience has shown that many NAT devices have upper limits on the rate at which they will create new bindings. Furthermore, transmission of these packets on the network makes use of bandwidth and needs to be rate limited by the agent. As a consequence, a client SHOULD pace its STUN transactions, such that the start of each new transaction occurs at least Ta seconds after the start of the previous transaction. The value of Ta SHOULD be configurable, and SHOULD have a default of 50ms. Note that this pacing applies only to the start of a new transaction; pacing of retransmissions within a STUN transaction is governed by the retransmission rules defined by STUN. Derived candidates can be obtained from the STUN Binding Discovery usage or the STUN Relay usage. The latter is preferred since it will provide the client with both a server reflexive and a relayed transport address with a single transaction. It is possible that some STUN servers will only support the Relay usage or only the Binding Discovery usage, in which case a client might be configured with different servers depending on the usage. Rosenberg Expires September 7, 2006 [Page 14] Internet-Draft ICE March 2006 To obtain both server reflexive and relayed candidates using the STUN Relay Usage, the client takes a local UDP candidate, and for each configured STUN server, produces both candidates. It is anticipated that clients may have a multiplicity of STUN servers configured or discovered in network environments where there are multiple layers of NAT, and that layering is known to the provider of the client. To obtain these candidates, for each configured STUN server, the client initiates an Allocate Request transaction using the procedures of Section 8.1.2 of [14] from each transport address of a particular local candidate. The Allocate Response will provide the client with its server reflexive transport address in the MAPPED-ADDRESS attribute and its relayed transport address in the RELAY-ADDRESS attribute. Once the Allocate requests have given a client a relayed transport address for all transport addresses in a relayed candidate, there is no reason for a client to obtain further relayed candidates through the same STUN server. Thus, if there are other local candidates from which the client has not yet obtained relayed transport address, the client SHOULD NOT bother to obtain them. Instead, it SHOULD use the STUN Binding Discovery usage and obtain just server reflexive addresses from that STUN server. The order in which local candidates are tried against the STUN server to obtain relayed candidates is a matter of local policy. To obtain server reflexice candidates using the STUN Binding Discovery usage, the client takes a local UDP candidate, and for each configured STUN server, produces a server reflexive candidate. To produce the server reflexive candidate from the local candidate, it follows the procedures of Section XX of [13] for each local transport address in the local candidate. The Binding Response will provide the client with its server reflexive transport address in the MAPPED- ADDRESS attribute. If the client had K local candidates, this will produce S*K server reflexive candidates, where S is the number of STUN servers. Since a client will pace its STUN transactions (both Binding and Allocate requests) at a total rate of one new transaction every Ta seconds, it will take a certain amount of time to complete the address gathering phase. It is RECOMMENDED that implementations have a configurable upper bound on the total amount of time allotted to address gathering. Any transactions not completed at that point SHOULD be abandoned, but MAY continue and be used in an updated offer once they complete. A default value of 5s is RECOMMENDED. Since the total number of allocations that could be done (based on the number of STUN servers and local interfaces) might exceed this value, clients SHOULD prioritize their local candidates and STUN servers, performing transactions from the highest priority local candidates to the highest priority STUN servers first. A STUN server would typically be higher priority if it supports the STUN Relay Usage, Rosenberg Expires September 7, 2006 [Page 15] Internet-Draft ICE March 2006 since such a server provides two transport addresses with one transaction. Once the allocations are complete, any redundant candidates are discarded. Candidate A is redundant with candidate B if the transport addresses for each component of each component match, and each component of their associated local candidates match. For example, consider a set of candidates with a single component. One candidate is a local candidate, and its one component has a transport address of 10.0.1.1:4458. A reflexive transport address is derived from this local transport address, producing a 10.0.1.1:4458. These two candidates are identical, and also have identical associated local transport addresses, so they are redundant. However, in a more complicated case, consider a multi-homed host, with one interface at 192.168.1.1 and another at 10.0.1.1. The 192.168 network is natted, with its "public" side in another net-10 private network. The client obtains two local candidates, A and B, with transport addresses of 192.168.1.1:2376 and 10.0.1.1:7266 respectively. A server reflexive transport address is derived from A through a STUN query, and it happens to produce 10.0.1.1:7266. Call this candidate C. Candidate C is not redundant with candidate B, since they have different associated local transport addresses. 7.2 Prioritizing the Candidates and Choosing an Active One The prioritization process takes the set of candidates and associates each with a priority. This priority reflects the desire that the agent has to receive media at that candidate, and is assigned as a value from 0 to 1 (1 being most preferred). Priorities are ordinal, so that their significance is only meaningful relative to other candidates from that agent for a particular media stream. Candidates MAY have the same priority. However, it is RECOMMENDED that each candidate have a distinct priority. Doing so improves the efficiency of ICE. This specification makes no normative statements on how the prioritization is done. However, some useful guidelines are suggested on how such a prioritization can be determined. One criteria for choosing one candidate over another is whether or not that candidate involves the use of an intermediary. That is, if media is sent to that candidate, will the media first transit an intermediate server before being received. Relayed candidates are clearly one type of candidates that involve an intermediary. Another are local candidates associated with a VPN server. When media is transited through an intermediary, it can increase the latency between transmission and reception. It can increase the packet losses, because of the additional router hops that may be taken. It Rosenberg Expires September 7, 2006 [Page 16] Internet-Draft ICE March 2006 may increase the cost of providing service, since media will be routed in and right back out of an intermediary run by the provider. If these concerns are important, candidates with this property can be listed with lower priority. Another criteria for choosing one candidate over another is IP address family. ICE works with both IPv4 and IPv6. It therefore provides a transition mechanism that allows dual-stack hosts to prefer connectivity over IPv6, but to fall back to IPv4 in case the v6 networks are disconnected (due, for example, to a failure in a 6to4 relay) [25]. It can also help with hosts that have both a native IPv6 address and a 6to4 address. In such a case, higher priority could be afforded to the native v6 address, followed by the 6to4 address, followed by a native v4 address. This allows a site to obtain and begin using native v6 addresses immediately, yet still fallback to 6to4 addresses when communicating with agents in other sites that do not yet have native v6 connectivity. Another criteria for choosing one candidate over another is security. If a user is a telecommuter, and therefore connected to their corporate network and a local home network, they may prefer their voice traffic to be routed over the VPN in order to keep it on the corporate network when communicating within the enterprise, but use the local network when communicating with users outside of the enterprise. Another criteria for choosing one address over another is topological awareness. This is most useful for candidates that make use of relays. In those cases, if an agent has preconfigured or dynamically discovered knowledge of the topological proximity of the relays to itself, it can use that to select closer relays with higher priority. There may be transport-specific reasons for preferring one candidate over another. In such a case, specifications defining usage of ICE with other transport protocols SHOULD document such considerations. Once the candidates have been prioritized, one may be selected as the active one. This is the candidate that will be used for actual exchange of media if and when its validated, until a higher priority candidate is validated. The active candidate will also be used to receive media from ICE-unaware peers. As such, it is RECOMMENDED that one be chosen based on the likelihood of that candidate to work with the peer that is being contacted. Unfortunately, it is difficult to ascertain which candidate that might be. As an example, consider a user within an enterprise. To reach non-ICE capable agents within the enterprise, a local candidate has to be used, since the enterprise policies may prevent communication between elements using a relay on the public network. However, when communicating to Rosenberg Expires September 7, 2006 [Page 17] Internet-Draft ICE March 2006 peers outside of the enterprise, a relayed candidate from a publically accessible STUN server is needed. Indeed, the difficulty in picking just one address that will work is the whole problem that motivated the development of this specification in the first place. As such, it is RECOMMENDED that the active candidate be a relayed candidate from a STUN server providing public IP addresses in response to an Allocate request. Furthermore, ICE is only truly effective when it is supported on both sides of the session. It is therefore most prudent to deploy it to close-knit communities as a whole, rather than piecemeal. In the example above, this would mean that ICE would ideally be deployed completely within the enterprise, rather than just to parts of it. An additional consideration for selection of the active candidate is the switching of media stream destinations between the initial offer and the subsequent offer. If the active candidate pair in the initial offer is being validated, media will flow to that pair once it is validated. When the ICE checks complete and yield a higher priority candidate pair, media will begin to flow to it (there will also be an updated offer/answer exchange that changes the active candidate). This will result in a change in the destination of the media packets. This may also cause a different path for the media packets. That path might have different delay and jitter characteristics. As a consequence, the jitter buffers may see a glitch, causing possible media artifacts. If these issues are a concern, the initial offer MAY omit an active candidate. In such a case, an updated offer will need to be sent immediately when communicating with an ICE-unaware agent, setting an active candidate. There may be transport-specific reasons for selection of an active candidate. In such a case, specifications defining usage of ICE with other transport protocols SHOULD document such considerations. 7.3 Encoding Candidates into SDP For each candidate for a media stream, the agent includes a series of a=candidate attributes as media-level attributes, one for each component in the candidate. Each candidate has a unique identifier, called the candidate-id. The candidate-id MUST be chosen randomly and contain at least 24 bits of randomness (this does not mean that the candidate-id is 24 bits long; just that it has at least 24 bits of randomness). It is chosen only when the candidate is placed into the SDP for the first time; subsequent offers or answers within the same session containing that same candidate MUST use the same candidate-id used previously. 24 bits is sufficient because the candidate-id is not providing security (the much more random password is). It is needed only to prevent a possible simultaneous selection Rosenberg Expires September 7, 2006 [Page 18] Internet-Draft ICE March 2006 by two agents within a private network for the useful lifetime of the software or hardware. Each component of the candidate has an identifier, called the component-id. The component-id is a sequence number. For each candidate, it starts at one, and increments by one for each component. As discussed below, ICE will perform connectivity checks such that, between a pair of candidates, checks only occur between transport addresses with the same component-id. As a consequence, if one candidate has three components, and it is paired with a candidate that has two, there will only be two transport address pairs and two connectivity checks. ICE will work without a standardized mapping between the components of a media stream and the numerical value of the component-id. This allows ICE to be used with media streams with multiple components without development of standards around such a mapping. However, a specific mapping has been defined in this specification for RTP - component-id 1 corresponds to RTP, and component-id of 2 corresponds to RTCP. Like the candidate-id, the component-id is assigned at the time the candidate is first placed into the SDP; subsequent offers or answers within the same session containing that same candidate MUST use the same component-id used previously. The transport, addr and port of the a=candidate attribute (all defined in Section 12) are set to the transport protocol, unicast address and port of the tranport address. A Fully Qualified Domain Name (FQDN) for a host MAY be used in place of a unicast address. In that case, when receiving an offer or answer containing an FQDN in an a=candidate attribute, the FQDN is looked up in the DNS using an A or AAAA record, and the resulting IP address is used for the remainder of ICE processing. The qvalue is set to the priority of the candidate, and MUST be the same for all components of the candidate. All of the candidates share a password that is used for securing the STUN connectivity checks. This password MUST be chosen randomly with 128 bits of randomness (though it can be longer than 128 bits). This password is contained in the a=ice-pwd attribute, present as a session level attribute. A new password MUST be selected for each new session, and MUST be present with the same value in all subsequent offers and answers from the agent. The converse is true; if a new offer is generated as part of a new multimedia session, a new password MUST be used even if the transport address from a previous session was being recycled. The combination of candidate-id and component-id uniquely identify each transport address. As a consequence, each transport address has a unique identifier, called the tid. The tid is formed by Rosenberg Expires September 7, 2006 [Page 19] Internet-Draft ICE March 2006 concatenating the candidate-id with the component-id, separated by the colon (":"). The tid is not explicitly encoded in the SDP; it is derived from the candidate-id and component-id, which are present in the SDP. The usage of the colon as a separator allows the candidate-id and component-id to be extracted from the tid, since the colon is not a valid character for the candidate-id. The tid gets combined, through further concatenation, with the tid of a transport address from the remote candidate (separated again by another colon) to form the username that is placed in the STUN checks between the peers. This allows the STUN message to uniquely identify the pairing whose connectivity it is checking. The tid is needed as a unique identifier because the IP address within the candidate fails to provide that uniqueness as a consequence of NAT. Consider agents A, B, and C. A and B are within private enterprise 1, which is using 10.0.0.0/8. C is within private enterprise 2, which is also using 10.0.0.0/8. As it turns out, B and C both have IP address 10.0.1.1. A sends an offer to C. C, in its answer, provides A with its transport addresses. In this case, thats 10.0.1.1:8866 and 8877. As it turns out, B is in a session at that same time, and is also using 10.0.1.1:8866 and 8877. This means that B is prepared to accept STUN messages on those ports, just as C is. A will send a STUN request to 10.0.1.1:8866 and 8877. However, these do not go to C as expected. Instead, they go to B. If B just replied to them, A would believe it has connectivity to C, when in fact it has connectivity to a completely different user, B. To fix this, tid takes on the role of a unique identifier. C provides A with an identifier for its transport address, and A provides one to C. A concatenates these two identifiers (with a colon between) and uses the result as the username in its STUN query to 10.0.1.1:8866. This STUN query arrives at B. However, the username is unknown to B, and so the request is rejected. A treats the rejected STUN request as if there were no connectivity to C (which is actually true). Therefore, the error is avoided. An unfortunate consequence of the non-uniqueness of IP addresses is that, in the above example, B might not even be an ICE agent. It could be any host, and the port to which the STUN packet is directed could be any ephemeral port on that host. If there is an application listening on this socket for packets, and it is not prepared to handle malformed packets for whatever protocol is in use, the operation of that application could be affected. Fortunately, since the ports exchanged in SDP are ephemeral and ususally drawn from the dynamic or registered range, the odds are good that the port is not used to run a server on host B, but rather is the agent side of some protocol. This decreases the probability of hitting a port in-use, due to the transient nature of port usage in this range. However, Rosenberg Expires September 7, 2006 [Page 20] Internet-Draft ICE March 2006 the possibility of a problem does exist, and network deployers should be prepared for it. Note that this is not a problem specific to ICE; stray packets can arrive at a port at any time for any type of protocol, especially ones on the public Internet. As such, this requirement is just restating a general design guideline for Internet applications - be prepared for unknown packets on any port. The active candidate, if there is one, is placed into the m/c lines of the SDP. For RTP streams, this is done by placing the RTP address and port into the c and m lines in the SDP respectively. If the agent is utilizing RTCP, it MUST encode its address and port using the a=rtcp attribute as defined in RFC 3605 [1]. If RTCP is not in use, the agent MUST signal that using b=RS:0 and b=RR:0 as defined in RFC 3556 [6]. If there is no active candidate, the agent MUST include an a=inactive attribute. The RTP address and port in the m/c-line is inconsequential, since it won't be used. Encoding of candidates may involve transport protocol specific considerations. There are none for UDP. However, extensions that define usage of ICE with other transport protocols SHOULD specify any special encoding considerations. Once an offer or answer are sent, an agent MUST be prepared to receive both STUN and media packets on each candidate. As discussed in Section 7.13, media packets can be sent to a candidate prior to its promotion to active. 7.4 Forming Candidate Pairs Once the offer/answer exchange has completed, both agents will have a set of candidates for each media stream. Each agent forms a set of candidate pairs for each media stream by combining each of its candidates with each of the candidates of its peer. Candidates can be paired up only if their transport protocols are identical. If an offer/answer exchange took place for a session comprised of an audio and a video stream, and each agent had two candidates per media stream, there would be 8 candidate pairs, 4 for audio and 4 for video. One agent can offer two candidates for a media stream, and the answer can contain three candidates for the same media stream. In that case, there would be six candidate pairs. Each candidate has a number of components, each of which has a transport address. Within a candidate pair, the components themselves are paired up such that transport addresses with the same component ID are combined to form a transport address pair. Returning to the previous example, for each of the 8 candidate pairs, Rosenberg Expires September 7, 2006 [Page 21] Internet-Draft ICE March 2006 there would be two transport address pairs - one for RTP, and one for RTCP. If one candidate has more components than the other, those extra components will not be part of a transport address pair, won't be validated, and will effectively be treated as if they weren't included in the candidate pair in the first place. The relationship between a candidate, candidate pair, transport address, transport address pair and component are shown in Figure 2. This figure shows the relationships as seen by the agent that owns the candidate with candidate ID "L". This candidate has two components with transport addresses A and B respectively. This candidate is called the native candidate, since it is the one owned by the agent in question. The candidate owned by its peer is called the remote candidate. As the figure shows, there is a single candidate pair, and two components in each candidate. The native candidate has a candidate-id of "L", and the remote candidate has a candidate-id of "R". Since the two component-ids are 1 and 2, candidate "L" has two transport addresses with transport address IDs of "L:1" and "L:2" respectively. Similarly, candidate "R" has two transport addresses with transport address IDs of "R:1" and "R:2" respectively. Furthermore, each transport address pair is associated with an ID, the transport address pair ID. This ID is equal to the concatenation of the tid of the native transport address with the tid of the remote transport address, separated by a colon. This means that the identifiers are seen differenly for each agent. For the agent that owns candidate "L", there are two transport address pairs. One contains transport address "L:1" and "R:1", with a transport address pair ID of "L:1:R:1". The other contains transport address "L:2" and "R:2", with a transport address pair ID of "L:2:R:2". For the agent that owns candidate "R", the identifiers for these two transport address pairs are reversed; it would be "R:1:L:1" for the first one and "R:2:L:2" for the second. Rosenberg Expires September 7, 2006 [Page 22] Internet-Draft ICE March 2006 ............................................... . . . . . ............. ............. . . . tid=L:1 . . tid=R:1 . . . . -- . . -- . . component component. . | A|------------------------| C| . . id=1 id=1 . . -- . Transport . -- . . . . . Address . . . . . . Pair . . . . . . id=L:1:R:1 . . . . . . . . . . . . . . . . . tid=L:2 . . tid=R:2 . . component . . -- . . -- . . id=2 . . | B|------------------------| D| component . . -- . Transport . -- . . id=2 . . . Address . . . . . . Pair . . . . . . id=L:2:R:2 . . . . . . . . . . ............. ............. . . Native Remote . . Candidate Candidate . . id=L id=R . . . . . ............................................... Candidate Pair Figure 2 If a candidate pair was created as a consequence of an offer generated by an agent, then that agent is said to be the offerer of that candidate pair and all of its transport address pairs. Similarly, the other agent is said to be the answerer of that candidate pair and all of its transport address pairs. As a consequence, each agent has a particular role, either offerer or answerer, for each transport address pair. This role is important; when a candidate pair is to be promoted to active, the offerer is the one which performs the updated offer. 7.5 Ordering the Candidate Pairs For the same reason that the STUN transactions during address gathering are paced at a rate of Ta transactions per second, so too Rosenberg Expires September 7, 2006 [Page 23] Internet-Draft ICE March 2006 are the connectivity checks paced, also at a rate of Ta transactions per second. However, in order to rapidly converge on a valid candidate pair that is mutually desirable, the candidate pairs are ordered, and the checks start with the candidate pair at the top of the list. Rapid convergence of ICE depends on both the offerer and answerer coming to the same conclusion on the ordering of candidate pairs. Recall that when each candidate is encoded into SDP, it contains a qvalue between 1 and 0, with 1 being the highest priority. Peer reflexive candidates, learned through the procedures described in Section 7.10 also have a priority between 0 and 1. For each media stream, the native candidates are ordered based on their qvalues, with higher q-values coming first. Amongst candidates with the same qvalue, they are ordered based on candidate ID, using reverse lexicographic order, where C1 is placed before C2, if C2 precedes C1 lexicographically. Lexicographic order can be viewed as a numerical ordering where each "digit" is actually a number in numerical base 256, with the mapping of characters to numerical value being defined by their ASCII encoding. For example, the candidate with candidate ID agD is greater than the candidate with ID ad7, and both of those are greater than the candidate with ID zz. Consequently, if these three candidates had equal q-values, they would be ordered as agD, ad7, zz - reverse of their lexicographic order. The usage of a reverse lexicographic order is important; as discussed in Section 13, it allows peer-derived candidates to be preferred over native ones. The result of these ordering rules will be an ordered list of candidates. The first candidate in this list is given a sequence number of 1, the next is given a sequence number of 2, and so on. This same procedure is done for the remote candidates. The result is that each candidate pair has two sequence numbers, one for the native candidate, and one for the remote candidate. First, all of the candidate pairs for whom the smaller of the two sequence numbers equals 1 are taken first. Then, all of those for whom the smaller of the two sequence numbers equals 2 are taken next, and so on. Amongst those pairs that share the same value for their smaller sequence number, they are ordered by the larger of their two sequence numbers (smallest first). Amongst those pairs that share the same value for their smaller sequence number and the same value for their larger sequence number, the larger of the two candidate IDs in each pair are selected, and the pairs are lexicographically ordered in reverse by that candidate ID, largest first. As an example, consider two agents, A and B. One offers two Rosenberg Expires September 7, 2006 [Page 24] Internet-Draft ICE March 2006 candidates for a media stream with candidate IDs of "g9" and "88", with q-values of 1.0 and 0.8 respectively. The other answers with three candidates with candidate IDs of "h8", "65" and "kl", with q-values of 0.3, 0.2 and 0.1 respectively. The following table shows the rank ordering of the six candidate pairs. The column labeled "Max SN" is the larger of the two sequence numbers in the candidate pair, and "Min SN" is the minimum. The column labeled "Max Cand. ID" is the value of the larger of the two candidate IDs in the candidate pair. Order A A A B B B Max Cand. Cand. Cand. Cand. Cand. Cand. Max Min Cand. ID q-value SN ID q-value SN SN SN ID --------------------------------------------------------------------- 1 g9 1.0 1 h8 0.3 1 1 1 h8 2 88 0.8 2 h8 0.3 1 2 1 h8 3 g9 1.0 1 65 0.2 2 2 1 g9 4 g9 1.0 1 k1 0.1 3 3 1 k1 5 88 0.8 2 65 0.2 2 2 2 88 6 88 0.8 2 k1 0.1 3 3 2 k1 This ordering is then modified slightly by taking the candidate pair corresponding to the active candidate, if there is one, and promoting it to the top of the list. To find this candidate pair, the agent looks for candidate pairs whose native and remote transport addresses match the native and remote transport addresses in the m/c-line. It is possible that multiple candidates match; this happens in the case where an agent obtained the same derived transport address from different local transport addresses. In such a case, the agent should pick one of the matching candidates. Putting the active candidate at the top of the list allows it to be tested first. As discussed below, media is not sent until the corresponding candidate is verified, necessitating rapid verification of the active candidate. This modified ordering is called the candidate pair check ordering, since it reflects the order in which connectivity checks will be done. If there was no active candidate, the candidate pair check ordering and the candidate pair priority ordering will be identical. Within each candidate pair there will be a set of transport address pairs, one for each component ID. Those pairs are ordered by component ID. The result is an absolute ordering of all transport address pairs for a media stream, sorted first by the order of their candidate pairs (with the exception of the active candidate), followed by the order of their component IDs. This ordering is Rosenberg Expires September 7, 2006 [Page 25] Internet-Draft ICE March 2006 called the transport address pair check ordering. Ordering of candidates may involve transport protocol specific considerations. There are none for UDP. However, extensions that define usage of ICE with other transport protocols SHOULD specify any special ordering considerations. 7.6 Performing the Connectivity Checks Connectivity checks are a STUN usage defined in [13]. They are performed by sending peer-to-peer STUN Binding Requests. These checks result in a candidate progressing through a state machine that captures the progress of connectivity checks. The specific state machine and the procedures for the connectivity checks are specific to the transport protocol. This specification defines rules for UDP. Extensions to ICE that describe other transport protocols SHOULD describe the state machine and the procedures for connectivity checks. The set of states visited by the offerer and answerer are depicted graphically in Figure 4 | |Start | | V +------------+ | | | | | Waiting |----------------+ | | | | | | +------------+ | | | | Timer Ta | Get Req | --------. | ------- | Send Req Get Req | Send Res, V ------- | Send Req Get Res +------------+ Send Res, | ------- | | Re-Xmit | - | | Req | +---------------| Testing |-----------+ | | | | | | | | | | | | +------------+ | | | | | | Rosenberg Expires September 7, 2006 [Page 26] Internet-Draft ICE March 2006 | | Error | | | | ----- | | Timer Tr | | - | | -------- V V V V Send Req +------------+ +------------+ +------------+ +-----| | | | | | | | Recv- | | | | Send- | | | Valid |------->| Invalid |<-------| Valid | | | | | | | | +---->| | Error | | Error | | +------------+ ----- +------------+ ----- +------------+ | - ^ - | | | Error | | | ----- | | | - | | +------------+ | | | | | | | | | +-------------->| Valid |<-------------+ Get Req | | Get Res ------- | | ------- Send Res +------------+ - | ^ | | | | +-------+ Timer Tr -------- Send Req Figure 4 The state machine has six states - waiting, testing, Recv-Valid, Send-Valid, Valid and Invalid. Initially, all transport address pairs start in the waiting state. In this state, the agent waits for one of two events - a chance to send a Binding Request, or receipt of a Binding Request. Since there is an instance of the state machine for each transport address pair, Binding Requests and responses need to be matched to the specific state machine for which they apply. This is done by computing the matching transport address pair for each Binding Request. This is done by examining the USERNAME of the incoming Binding Request. The USERNAME directly contains the transport address pair ID. Requests that are sent by an agent as part of the processing described here encode the transport address pair in the Rosenberg Expires September 7, 2006 [Page 27] Internet-Draft ICE March 2006 USERNAME. Binding Responses are matched to their requests using the STUN transaction ID, and then mapped to the transport address pair from that. Every Ta seconds, the agent starts a new connectivity check for a transport address pair. The check is started for the first transport address pair in the transport address pair check ordered list (which will be part of the active candidate) that is in the Waiting state. The state machine for this transport address pair is moved to the Testing state, and the agent sends a connectivity check using a STUN Binding Request, as outlined in Section 7.7. Once a STUN connectivity check begins, the processing of the check follows the rules for STUN. Specifically, retransmits of STUN requests are done as specified in [13], and furthermore, if a transaction fails and needs to be retried, that retry can happen rapidly, as described below. It doesn't "count" against the rate limit of 1/Ta checks per second. In addition, the keepalives that are generated for a valid pair do not count against the rate limit either. The rate limit applies strictly to the start of connectivity checks for a transport address pair that has been newly signaled through an offer/answer exchange. In addition, if, while in the Waiting state, an agent receives a Binding Request matching that transport address pair, and this Binding Request generates a successful response, the transport address pair moves into the Send-Valid state, and the agent sends a connectivity check of its own using a STUN Binding Request, as outlined in Section 7.7. If the Binding Request didn't generate a success response, there is no change in state or generation of a Binding Request. If, while in the Testing state, the agent receives a successful response to its STUN request, the transport address pair moves into the Recv-Valid state. In this state, the agent knows that packets can flow in both directions. However, its peer agent doesn't yet know that; all it knows is that it has been able to receive a packet. Thus, in this state, the agent awaits receipt of the Binding Request sent by its peer, as the response to that request is what informs its peer that packets can flow in both directions. If, while in the Testing state, the agent receives a Binding Request matching that transport address pair, and this Binding Request generates a successful response, the transport address pair moves into the Send-Valid state. In addition, the agent retransmits a Binding Request for the transaction in progress. This helps speed up bidirectional connectivity verification when one agent is behind a symmetric NAT. If the Binding Request didn't generate a success response, there is no change in state or generation of a Binding Rosenberg Expires September 7, 2006 [Page 28] Internet-Draft ICE March 2006 Request. If, while in the Send-Valid state, the agent receives a successful response to its STUN request, the transport address pair moves to the Valid state. In this state, the agent knows that packets can flow in each direction. It also knows that its peer has sent it the STUN Request whose response will demonstrate to the peer that packets can flow in each direction. If, while in the Recv-Valid state, the agent receives a STUN Binding Request from its peer that results in a successful response, the transport address pair moves into the Valid state. Receipt of a request whose response was not a successful one does not result in a change in state. In any state, if the STUN transaction results in an error, the state machine moves into the invalid state. A STUN transaction produces an "error" based on the processing in Section 7.7, which indicates which STUN response codes constitute an error as far as ICE processing is concerned. If a transport address pair is in the Recv-Valid or Valid state, an agent MUST generate a new STUN Binding Request transaction every Tr seconds. This transaction ensures that NAT bindings for the transport address pair remain open while the candidate is under consideration. The transaction is performed as outlined in Section 7.7. These transactions can also be used to keep the NAT bindings alive when the candidate is promoted to active, as described in Section 7.12. Tr SHOULD be configurable, and SHOULD default to 15 seconds. If the transaction results in an error, the state machine moves to the invalid state. This happens in cases where the NAT bindings expire (e.g., due to binding timeouts or NAT failures). The candidate pair itself has a state, which is derived from the states of its transport address pairs. If at least one of the transport address pairs in a candidate pair is in the invalid state, the state of the candidate pair is considered to be invalid. If the candidate pair enters this state, an agent SHOULD move the state machines for all of the other transport address pairs in this candidate pair into the invalid state as well. This will ensure that connectivity checks never start for those transport address pairs. Furthermore, if checks are already in progress for one of those transport address pairs, the agent SHOULD cease them. If all of the transport address pairs making up the candidate pair are Valid, the candidate pair is considered valid. If all of the transport address pairs making up the candidate pair are either Valid or Recv-Valid, and at least one is Recv-Valid, the candidate pair is Rosenberg Expires September 7, 2006 [Page 29] Internet-Draft ICE March 2006 considered to be Recv-Valid. If all of the transport address pairs making up the candidate pair are either Valid or Send-Valid, and at least one is Send-Valid, the candidate pair is considered to be Send- Valid. If all of the transport address pairs in a candidate pair are in the Waiting state, the candidate pair is in the waiting state. If all of the transport address pairs in the candidate pair are either in the Waiting or Testing states, and at least one is in the Testing state, the state of the candidate pair is Testing. Otherwise, the state of the candidate pair is considered Indeterminate. A candidate itself also has a state. If a candidate is present in at least one valid candidate pair, that candidate is said to be valid. If all of the candidate pairs containing that candidate are invalid, the candidate itself is invalid. Otherwise, the candidate's state is Indeterminate. 7.7 Sending a Binding Request for Connectivity Checks An agent performs a connectivity check on a transport address pair by sending a STUN Binding Request from its native transport address, and sending it to the remote transport address. The meaning of "sending from its native transport address" depends on the type of transport protocol and the type of transport address (local, reflexive, or relayed). This specification defines the meaning for UDP. Specifications defining other transport protocols must define what this means for them. For UDP-based local transport addresses, sending from the local transport address has the meaning one would expect - the request is sent such that the source IP address and port equal that of the local transport address. For reflexive ransport addresses, it is sent by sending from the associated local transport address used to derive that reflesive address. For relayed transport addresses, it is sent by using STUN mechanisms to send the request through the STUN relay (using the Send request). Sending the request through the STUN relay server neccesarily requires that the request be sent from the client, using the local transport address used to derive the relayed transport address. The Binding Request sent by the agent MUST contain the USERNAME attribute. This attribute MUST be set to the transport address pair ID of the corresponding transport address pair as seen by its peer. Thus, for the first transport address pair in Figure 2, if the agent on the left sends the STUN Binding Request, the USERNAME will have the value R:1:L:1. If the agent on the right sends the STUN Binding Request, the USERNAME will have the value L:1:R:1. To be clear, the USERNAME that is used is NOT the one seen locally, but rather the one as seen by its peer. The request SHOULD contain the MESSAGE- Rosenberg Expires September 7, 2006 [Page 30] Internet-Draft ICE March 2006 INTEGRITY attribute, computed according to [13]. The key used as input to the HMAC is the password provided by the peer for this remote transport address. This password will be identical for all remote transport addresses for the same media stream. The STUN transaction will generate either a timeout, or a response. If the response is a 420, 500, or 401, the agent should try again as described in [13] (as mentioned above, it need not wait Ta seconds to try again). Either initially, or after such a retry, the STUN transaction might produce a non-recoverable failure response (error codes 400, 430, 431, or 600) or a failure result inapplicable to this usage of STUN and thus unrecoverable (432, 433). If this happens, an error event is generated into the state machine, and the transport address pair enters the invalid state. If the STUN transaction times out, the client SHOULD NOT retry. The only reason a retry might succeed is if there was severe packet loss during the duration of the check, or the answer was significantly delayed, also due to packet loss. However, STUN Binding Request transactions run for 9.5 seconds, which is well beyond the typical tolerance for a session establishment. The retries come with a penalty of additional traffic, which can be used to launch DoS attacks Section 13.4.2. The only reason to not follow the SHOULD NOT is if the agent has adjusted the STUN transaction timers to be more aggressive. If the Binding Response is a 200, the agent SHOULD check for the MESSAGE-INTEGRITY attribute and verify it, as discussed in [13]. Indeed, this check SHOULD be done for all responses. This will result in the response being discarded (eventually leading to a timeout), if the integrity check fails. 7.8 Receiving a Binding Request for Connectivity Checks As a result of providing a list of candidates in its offer or answer, an agent will receive STUN Binding Request messages. An agent MUST be prepared to receive STUN Binding Requests on each local transport address from the moment it sends an offer or answer that contains a candidate with that local transport address. Similarly, it MUST be prepared to receive STUN Binding Requests on a local transport address the moment it sends an offer or answer that contains a reflexive or relayed candidate derived from a local candidate with that local transport address. It can cease listening for STUN messages on that local transport address after sending an updated offer or answer which does not include any candidates with transport addresses that are equal to or derived from that local transport address. Rosenberg Expires September 7, 2006 [Page 31] Internet-Draft ICE March 2006 As discussed in [13], since the username and password for STUN requests are exchanged through another mechanism - here, ICE - the Shared Secret Request mechanism is not needed and need not be implemented by agents that provide the connectivity check usage. One of the candidates may be in use as the active candidate, or may become promoted to the active candidate in the next offer/answer exchange as a consequence of a successful validation. In either case, both media and STUN packets will be sent to the transport addresses comprising that candidate, causing both to receive on their associated local transport addresses. The agent MUST be able to disambiguate them. This is done trivially by looking for the STUN magic cookie as the value of the second 32-bit word in the packet. If present, it identifies a STUN packet. Processing of the Binding Request proceeds in two steps. The first is generation of the response, and the second ICE-specific processing. Generation of the response follows the general procedures of [13]. The USERNAME is considered valid if one of the candidate IDs sent in an offer or answer is a prefix of the USERNAME (this will always be the case, even for peer reflexive candidates). The password associated with that candidate ID is used to verify the MESSAGE-INTEGRITY attribute, if one was present in the request. If the USERNAME was not valid, the agent generates a 430. Otherwise, the success response will include the MAPPED-ADDRESS attribute, which is used for learning new candidates, as described in Section 7.10. The MAPPED-ADDRESS attribute is populated with the source IP address and port of the Binding Request. For Binding Requests received over relayed transport addresses, this MUST be the source IP address and port of the Binding Request when it arrived at the relay, prior to forwarding towards the agent. That source transport address will be present in the REMOTE-ADDRESS attribute of a STUN Data Indication message, if the Binding Request was delivered through a Data Indication. If the Binding Request was not encapsulated in a Data Indication, that source address is equal to the current active destination for the STUN relay session. The ICE processing involves changes to the state machine for a transport address pair. This processing cannot be done until the initial offer/answer exchange has completed. As a consequence, if the oferrer received a Binding Request that generated a success response, but had not yet received the answer to its offer, it waits for the answer, and when it arrives, then performs the ICE processing. The agent takes the entire contents of the USERNAME, and compares them against the transport address pair identifiers as seen by that agent for each transport address pair. If there is no match, nothing Rosenberg Expires September 7, 2006 [Page 32] Internet-Draft ICE March 2006 is done - this should never happen for compliant implementations. If there is a match, the resulting transport address pair is called the matching transport address pair. The state machine for the matching transport address pair is then updated based on the receipt of a STUN Binding Request, and the resulting actions described in Section 7.6 are undertaken. An agent will continue to receive periodic STUN connectivity checks on a local transport address as long as it had listed that transport address, or one derived from it, in an a=candidate attribute in its most recent offer or answer, the state machine for that transport address is in the Recv-Valid or Valid states, and the transport address is for UDP. Whether STUN keepalives are used for other transport protocols is defined by the specifications for that transport protocol. The agent processes any such transactions according to this section. It is possible that a transport address pair that was previously valid may become invalidated as a result of a subsequent failed STUN transaction. 7.9 Promoting a Candidate to Active As a consequence of the connectivity checks, each agent will change the states for each transport address pair, and consequently, for the candidate pairs. When a candidate pair becomes valid, and the agent is in the role of offerer for that candidate pair, the agent follows the logic in this section. The rules only apply to the offerer of a candidate pair in order to eliminate the possibility of both agents simultaneously offering an update to promote a candidate to active. If this candidate pair is the first one in the candidate pair priority ordered list, the agent SHOULD send an updated offer as described in Section 7.11.1. If this candidate pair is not the first on that list, but it is the first on the candidate pair check ordered list, it means that this candidate pair is the active one, and its connectivity has been verified. This is good news; the currently active candidate is working. Media can now flow as described in Section 7.13 (media will never flow prior to validation). However, no updated offer is sent at this time. If this candidate pair is not the first on the candidate pair priority ordered list or the candidate pair check ordered list, and the wait-state timer has not yet been set, the agent sets this timer to Tws seconds. Tws SHOULD be configurable, and SHOULD have a default of 100ms. This timer allows for a higher priority connectivity check to complete, in the event its STUN Binding Request was lost or delayed in the network. If, prior to the wait-state timer firing, another connectivity check completes and a candidate pair is validated, there is no need to reset or cancel the timer. Rosenberg Expires September 7, 2006 [Page 33] Internet-Draft ICE March 2006 Once the timer fires, the agent SHOULD issue an updated offer as described in Section 7.11.1. In addition, in order to speed up ICE processing, once the agent has determined the candidate that is to be promoted, it will send and receive media using that candidate in expectation of an updated offer. This is discussed in Section 7.13. 7.10 Learning New Candidates from Connectivity Checks ICE makes use of reflexive addresses, which are addresses that inform an agent of its transport address as seen by another host. An initial offer or answer generated by an agent includes server reflexive addresses, which are learned from a configured or discovered STUN server in the network. However, the connectivity checks themselves can inform an agent of reflexive addresses, and in particular, ones that are reflexive towards its peer. These are called peer reflexive candidates. A new peer reflexive candidate is typically observed when two agents are separated by a NAT with the address-dependent or address and port dependent mapping properties [37]. When the agent behind such a NAT sends a Binding Request to the other agent (assuming it is reachable), the NAT will create a new mapping for this Binding Request. Because STUN and the media packets are sent on the same port, regardless of the filtering properties of the NAT (whether endpoint independent, address dependent, or address and port dependent), this reflexive address can be used by the peer for sending STUN and media packets back towards the agent. To obtain and use these peer reflexive transport addresses, ICE agents perform additional processing on the receipt of STUN Binding Requests and responses, beyond the logic described in Section 7.7 and Section 7.8. This logic is described below. 7.10.1 On Receipt of a Binding Request When a STUN Binding Request is received which generates a success response, that Binding Request would have been associated with a matching transport address pair and corresponding candidate pair. The source IP and port of this Binding Request are compared to the IP address and port of the remote transport address in the matching transport address pair. Note that, in this case, we are comparing actual IP addresses and ports - not tids. In addition, if the Binding Request arrived through a relayed transport address, the source IP and port of this binding request used for the comparison are those in the Binding Request when it arrived at the relay, prior to forwarding towards the agent. That source transport address will be present in the REMOTE-ADDRESS attribute of a STUN Data Indication message, if the Binding Request were delivered through a Data Rosenberg Expires September 7, 2006 [Page 34] Internet-Draft ICE March 2006 Indication. If the Binding Request was not encapsulated in a Data Indication, that source address is equal to the current active destination for the STUN relay session. The comparison of the source IP and port of the Binding Request and the IP address and port of the remote transport address in the matching transport address pair may indicate inequality. In that case, the source IP and port of the Binding Request (and again, for relayed transport address, this refers to the source IP address and port of the packet when it arrived at the relay) are compared to the IP address and ports across the transport address pairs in *all* remote candidates. If there is still no match, it means that the source IP and port might represent another valid remote transport address - a peer derived one. To use it, that address needs to be associated with a candidate (called a peer-derived candidate). In this case, however, the candidate isn't signaled through an offer/answer exchange; it is constructed dynamically from information in the STUN request. Like all other candidates, the peer-derived candidate has a candidate ID. The candidate ID is derived from the candidate IDs of the matching candidate pair. In particular, the candidate ID is constructed by concatenating the remote candidate ID with the native candidate ID (without the colon). The password for the new candidate equals that of the remote candidate ID in the matching candidate pair. On receipt of a STUN Binding Request whose source IP and port don't match the transport address in any remote candidate, the agent constructs the candidate ID that represents the peer reflexive candidate, and checks to see if that candidate exists. It may already exist if it had been constructed as a consequence of a previous application of this logic on receipt of a Binding Request for a different transport address pair of the same candidate pair. If there is not yet a peer reflexive candidate with that candidate ID, the agent creates it, and assigns it the newly computed candidate ID. The priority of the peer-derived candidate MUST be set to the priority of its generating candidate - the remote candidate in the matching transport address pair. Note that, at this time, the peer derived candidate has no transport addresses in it. Newly created or not, the agent extracts the component ID from the matching transport address pair, and sees if a transport address with that same component ID exists in the peer reflexive candidate. If not (and it shouldn't), the agent adds a transport address to the peer reflexive candidate. This transport address is equal to the source IP address and port from the incoming STUN Binding Request (and in the case of a relayed transport address, the one seen by the relay). It is assigned the component ID equal to the component ID in Rosenberg Expires September 7, 2006 [Page 35] Internet-Draft ICE March 2006 the matching transport address pair. This transport address will have a tid, equal to the concatenation of the candidate ID for this new candidate, and the component ID, separated by a colon. The peer reflexive candidate becomes usable once the number of transport addresses in it equals the transport address pair count of the candidate pair from which it is derived. Initially, the peer reflexive candidate will start with a single transport address. More are added as the connectivity checks for the original candidate pair take place. Once the peer reflexive candidate becomes usable, it has to be paired up with native candidates. However, unlike the procedures of Section 7.5, which pair up each remote candidate with each native candidate, this peer reflexive candidate is only paired up with the native candidate from the candidate pair from which it was derived. This creates a new candidate pair, and a set of new transport address pairs. Recall that, for each candidate pair, one agent plays the role of offerer, and the other of answerer. For a peer-reflexive candidate, the role is identical to that of its generating candidate. Figure 5 provides a pictorial representation of the peer reflexive candidate (the one with id=RL) and its pairing with the native candidate with id L. The candidate with ID R is referred to as the generating candidate. The peer reflexive candidate is effectively an alternate for that generating candidate, but is only paired with a specific native candidate. Note that, for a particular generating candidate, there can be many peer derived candidates, up to one for each native candidate. Rosenberg Expires September 7, 2006 [Page 36] Internet-Draft ICE March 2006 ............. ............. . tid=L:1 . . tid=R:1 . component. -- . id=L:1:R:1 . -- .component id=1 . | A|-------------------------| C| . id=1 . -- -------+ . -- . . . | . . Generating . . | . . Candidate . tid=L:2 . | . tid=R:2 . component. -- . | id=L:2:R:2 . -- .component id=2 . | B|-------C-----------------| D| . id=2 . -- -----+ | . -- . .............| | ............. Native | | Remote Candidate | | Candidate id=L | | id=R | | | | ............. | | . tid=RL:1 . | | id=L:1:RL:1 . -- .component | +-----------------| C| . id=1 | . -- . | . . Peer Derived | . . Candidate | . tid=RL:2 . | id=L:2:RL:2 . -- .component +-------------------| D| . id=2 . -- . ............. Remote Candidate id=RL Figure 5 The new transport address pairs have a state machine associated with them. The state that is entered, and actions to take as a consequence, are specific to the transport protocol. For UDP, the procedures are defined here. Extensions that define processing for other transport protocols SHOULD describe the behavior. For UDP, the state machine enters the Send-Valid state. Effectively, the Binding Request just received "counts" as a validation in this direction, even though it was formally done for a different candidate pair. In addition, the agent SHOULD generate a Binding Request for each transport address in this new candidate pair, as described in Section 7.7. The transport address pairs are inserted into the ordered list of pairs based on the ordering described in Section 7.5 and processing follows the logic described in Section 7.6. Rosenberg Expires September 7, 2006 [Page 37] Internet-Draft ICE March 2006 7.10.2 On Receipt of a Binding Response The procedures on receipt of a Binding Response are nearly identical to those for receipt of a Binding Request as described above. When a successful STUN Binding Response is received, it will be associated with a matching transport address pair and corresponding candidate pair. This matching is done based on comparison of candidate IDs. The value of the MAPPED-ADDRESS attribute of the Binding Response are compared to the IP address and port of the native transport address in the matching transport address pair. Note that, in this case, we are comparing actual IP addresses and ports - not tids. These may not match if there was a NAT between the two agents. If they do not match, the value of the MAPPED-ADDRESS attribute of the Binding Response are compared to the IP address and ports across the transport address pairs in *all* native candidates. If there is still no match, it means that the MAPPED-ADDRESS might represent another valid native transport address. To use it, that address needs to be associated with a candidate. In this case, however, the candidate isn't signaled through an offer/ answer exchange; it is constructed dynamically from information in the STUN response. Such a candidate is called a peer reflexive candidate. Like all other candidates, the peer reflexive candidate has a candidate ID. The candidate ID is derived from the candidate IDs of the matching candidate pair. In particular, the candidate ID is constructed by concatenating the native candidate ID with the remote candidate ID (without the colon). The password for the new candidate equals that of the native candidate ID in the matching candidate pair. On receipt of a STUN Binding Response whose MAPPED-ADDRESS didn't match the transport address in any native candidate, the agent constructs the candidate ID that represents the peer reflexive candidate, and checks to see if that candidate exists. It may already exist if it had been constructed as a consequence of a previous application of this logic on receipt of a Binding Response for a different transport address pair of the same candidate pair. If there is not yet a peer derived candidate with that candidate ID, the agent creates it, and assigns it the newly computed candidate ID. The priority of the new candidate MUST be set to the priority of the generating candidate - the native candidate in the matching transport address pair. Note that, at this time, the peer derived candidate has no transport addresses in it. Newly created or not, the agent extracts the component ID from the matching transport address pair, and sees if a transport address with that same component ID exists in the peer reflexive candidate. If Rosenberg Expires September 7, 2006 [Page 38] Internet-Draft ICE March 2006 not (and it shouldn't), the agent adds a transport address to the peer reflexive candidate. This transport address is equal to the MAPPED-ADDRESS from the STUN Binding Response. It is assigned the component ID equal to the component ID in the matching transport address pair. This transport address will have a tid, equal to the concatenation of the candidate ID for this new candidate, and the component ID, separated by a colon. The peer-derived candidate becomes usable once the number of transport addresses in it equals the transport address pair count of candidate pair from which it is derived. Initially, the peer-derived candidate will start with a single transport address. More are added as the connectivity checks for the original candidate pair take place. Once the peer-derived candidate becomes usable, it has to be paired up with remote candidates. However, unlike the procedures of Section 7.5, which pair up each remote candidate with each native candidate, the peer-derived candidate is only paired up with the remote candidate from the matching candidate pair. This creates a new candidate pair, and a set of new transport address pairs. Recall that, for each candidate pair, one agent plays the role of offerer, and the other of answerer. For a peer-reflexive candidate, the role is identical to that of its generating candidate. The new transport address pairs have a state machine associated with them. The state that is entered, and actions to take as a consequence, are specific to the transport protocol. For UDP, the procedures are defined here. Extensions that define processing for other transport protocols SHOULD describe the behavior. For UDP, the state machine enters the Recv-Valid state. Effectively, the Binding Response just received "counts" as a validation in this direction, even though it was formally done for a different candidate pair. The transport address pairs are inserted into the ordered list of pairs based on the ordering described in Section 7.5, and processing follows the logic described in Section 7.6. 7.11 Subsequent Offer/Answer Exchanges An agent MAY issue an updated offer at any time. This updated offer may be sent for reasons having nothing to do with ICE processing (for example, the addition of a video stream in a multimedia session), or it may be due to a change in ICE-related parameters. For example, if an agent acquires a new candidate after the initial offer/answer exchange, it may seek to add it. However, agents SHOULD follow the logic described in Section 7.9 to determine when to send an updated offer as a consequence of promoting Rosenberg Expires September 7, 2006 [Page 39] Internet-Draft ICE March 2006 a candidate to active. If there are any aspects of this processing that are specific to the transport protocol, those SHOULD be called out in ICE extensions that define operation with other transport protocols. There are no additional considerations for UDP. 7.11.1 Sending of a Subsequent Offer The offer MAY contain a new active candidate in the m/c line. This candidate SHOULD be the native candidate from the highest candidate pair in the candidate pair priority ordered list whose state is Valid. If there are no candidate pairs in this state, the highest one whose state is Send-Valid or Recv-Valid SHOULD be used. If there are no candidate pairs in these states, the candidate pair that is most likely to work with this peer, as described in Section 7.2, SHOULD be used. The candidate is encoded into the m/c line in an updated offer as described in Section 7.3. If the candidate pair whose native candidate was encoded into the m/c-line was Valid, Send-Valid or Recv-Valid, the agent MUST include an a=remote-candidate attribute into the offer. This attribute MUST contain the candidate ID of the remote candidate in the candidate pair. It is used by the recipient of the offer in selecting its candidate for the answer. The meaning of a=candidate attributes within a subsequent offer have the same meaning as they do in an initial offer. They are a request for the peer to attempt (or continue to attempt if the candidate was provided previously) a connectivity check using STUN from each of its own candidates. When an updated offer is sent, there are several dispositions regarding the candidates: retained: A candidate is retained if the candidate ID for the candidate is included in the new offer, and matches the candidate ID for a candidate in the previous offer or answer from the agent. In this case, all of the information about the candidate - its qvalue and components, and the IP addresses, ports, and transport protocols of its components, MUST be the same as the previous offer or answer from the agent. If the agent wants to change them, this is accomplished by changing the candidate ID as well. That will have the effect of removing the old candidate and adding a new one with the updated information. removed: A candidate is removed if its candidate ID appeared in a previous offer or answer, and that candidate ID is not present in the new offer. Rosenberg Expires September 7, 2006 [Page 40] Internet-Draft ICE March 2006 added: A candidate is added if its candidate ID appeared in the new offer, but was not present in a previous offer or answer from that agent. The following rules are used to determine the disposition of the each of the current native candidates in the new offer: o If a candidate is invalid, and all peer reflexive candidates generated from it are invalid as well, it SHOULD be removed. o If the candidate in the m/c-line is valid, all other candidates SHOULD be removed. This has the effect of stopping connectivity checks of other candidates. This SHOULD would not be followed if an agent wanted to keep a candidate ready for usage should, for some reason, the active candidate later become invalid. o If the candidate in the m/c-line is valid, and it is not peer reflexive, that candidate MUST be retained. If the candidate in the m/c-line is peer reflexive, its generating candidate MUST be retained, even if it is itself invalid. o If the candidate in the m/c-line has not been validated, all other candidates that are not invalid, or candidates for whom their derived candidates are not invalid, SHOULD be retained. o Peer reflexive candidates MUST NOT be added; they continue to be used as long as their generating candidate was retained. Peer derived candidates are learned exclusively through the STUN connectivity checks. A new candidate MAY be added. This can happen when the candidate is a new one, learned since the previous offer/answer exchange, and it has a higher priority than the currently active candidate. It can also occur when an agent wishes to restart checks for a transport address it had tried previously. Effectively, changing the candidate ID value in an updated offer will "restart" connectivity checks for that candidate. If a candidate is removed, the agent takes the following steps once the offer is sent: 1. The agent eliminates any candidate pairs whose native candidate equalled the candidate that was removed. Equality is based on comparison of candidate IDs. 2. The agent eliminates any candidate pairs that had a native candidate that is a peer reflexive candidate generated from the candidate that was removed. Rosenberg Expires September 7, 2006 [Page 41] Internet-Draft ICE March 2006 3. The candidate pairs that are eliminated are removed from the candidate pair priority ordered list and candidate pair check ordered list. As a consequence of this, if connectivity checks had not yet begun for the candidate pair, they won't. 4. If connectivity checks were already in progress for transport addresses in a candidate pair that was removed, the agent SHOULD immediately terminate them. No further retransmissions take place, and no further transactions from that candidate will be made. 5. If the removed candidate was a relayed candidate, the agent SHOULD de-allocate its transport addresses from the STUN relay if it is not using those resources elswhere. If a local candidate was removed, and all of its derived candidates were also removed (including any peer reflexive candidates), local operating system resources for each of the transport addresses in the local candidate SHOULD be de-allocated, as long as it is not using those resources elsewhere. The resources may be in use elsewhere if they were included in an initial offer which generated multiple answers (as can happen with SIP forking). In such a case, a subsequent offer which removes the candidate will not imply its removal with the other branches; each becomes a separate offer/answer relationship. Subsequent offers MUST contain the a=ice-pwd attribute. This SHOULD have the same value as in previous offers. However, an agent MAY change it if, for some reason, the agent believes that the password may have been compromised. Since the same password is applied across all transport addresses in all candidates for all media streams, a change in the password impacts all of them. An agent MUST be prepared to receive connectivity checks that use either the new or old password until Tpw seconds after it receives the answer. Tpw SHOULD be configurable, and SHOULD default to 2 seconds. 7.11.2 Receiving the Offer and Sending an Answer To generate the answer, the answerer has to decide which transport addresses to include in the m/c line, and which to include in candidate attributes. The first step in the process is to look for the a=remote-candidate attribute in the offer. The a=remote-candidate exists to eliminate a race condition between the updated offer and the response to the STUN Binding Request that moved a candidate into the Valid state. This race condition is shown in Figure 6. On receipt of message 5, agent A can move its transport address pair state machine into the Valid state. It sends a STUN response to the request (message 6), but this Rosenberg Expires September 7, 2006 [Page 42] Internet-Draft ICE March 2006 is lost. Agent A proceeds with an updated offer (message 7), which is received at agent B. As far as agent B is concerned, the transport address pair is still in the Send-Valid state. It will move into the Valid state only on receipt of the STUN response in message 10. Thus, upon receipt of the offer, agent B cannot determine which candidate to include in its answer. To eliminate this condition, the identity of the validated candidate is included in the offer itself. Note, however, that the answerer will not send media until it has received this STUN response. Agent A Network Agent B |(1) Offer | | |------------------------------------------>| |(2) Answer | | |<------------------------------------------| |(3) STUN Req. | | |------------------------------------------>| |(4) STUN Res. | | |<------------------------------------------| |(5) STUN Req. | | |<------------------------------------------| |(6) STUN Res. | | |-------------------->| | | |Lost | |(7) Offer | | |------------------------------------------>| |(8) Answer | | |<------------------------------------------| |(9) STUN Req. | | |<------------------------------------------| |(10) STUN Res. | | |------------------------------------------>| Figure 6 If the a=remote-candidate attribute is present, the agent examines the transport addresses in the m/c-line of the offer. It compares these with the transport addresses in the remote candidates of all candidate pairs. If there is at least one match, the agent compares the native candidate ID of each matching pair with the value of the a=remote-candidate attribute. If there is a match, that candidate pair is selected. For each transport address pair in that candidate pair, if the state of the transport address pair is Send-Valid, the agent considers the state to be Valid just for the purpose of selecting the m/c-line as discussed in the paragraph below. The actual state MUST remain Send-Valid. This is necessary to prevent Rosenberg Expires September 7, 2006 [Page 43] Internet-Draft ICE March 2006 against DoS attacks. Rules for choosing transport addresses for the m/c-line are as follows. The agent examines the transport addresses in the m/c-line of the offer. It compares these with the transport addresses in the remote candidates of candidate pairs whose states are Valid. If there is a matching candidate pair in that state, the pair with the highest priority MUST be chosen, and the native candidate from that pair used as the active candidate. If there were no matching candidate pairs in the Valid state, the candidate that is most likely to work with this peer, as described in Section 7.2, SHOULD be used. Like the offerer, the answerer can decide, for each of its candidates, whether they are retained or removed. The same rules defined in Section 7.11.1 for determining their disposition apply to the answerer. Similarly, if a candidate is removed, the same rules in Section 7.11.1 regarding removal of canididate pairs and freeing of resources apply. Once the answer is sent, the answerer will have the set of native and remote candidates before this offer/answer exchange, and the set of native and remote candidates afterwards. A peer derived candidate continues to be used as long as its generating parent continues to be used. The agent then pairs up the native and remote candidates which were added or retained. This leads to a set of current candidate pairs. If a candidate pair existed previously, but as a consequence of the offer/answer exchange, it no longer exists, the agent takes the following steps: 1. The candidate pair is removed from the candidate pair priority ordered list and candidate pair check ordered list. As a consequence of this, if connectivity checks had not yet begun for the candidate pair, they won't. 2. If connectivity checks were already in progress for that candidate pair, the agent SHOULD immediately terminate any STUN transactions in progress from that candidate. No further retransmissions take place, and no further transactions from that candidate will be made. 3. If the agent receives a STUN Binding Request for that candidate pair, the agent SHOULD generate a 430 response. If a candidate pair existed previously, and continues to exist, no changes are made; any STUN transactions in progress for that candidate pair continue, and it remains on the candidate pair Rosenberg Expires September 7, 2006 [Page 44] Internet-Draft ICE March 2006 priority ordered list and candidate pair check ordered list. If a candidate pair is new (because either its native candidate is new, or its remote candidate is new, or both), the agent takes the role of answerer for this candidate pair. The new candidate pair is inserted into the candidate pair priority ordered list and candidate pair check ordered list. STUN connectivity checks will start for them based on the logic described in Section 7.6. 7.11.3 Receiving the Answer Once the answer is received, the answerer will have the set of native and remote candidates before this offer/answer exchange, and the set of native and remote candidates afterwards. It then follows the same logic described in Section 7.11.2, pairing up the candidate pairs, removing ones that are no longer in use, and beginning of processing for ones that are new. 7.12 Binding Keepalives Once a candidate is promoted to active, and media begins flowing, it is still necessary to keep the bindings alive at intermediate NATs for the duration of the session. Normally, the media stream packets themselves (e.g., RTP) meet this objective. However, several cases merit further discussion. Firstly, in some RTP usages, such as SIP, the media streams can be "put on hold". This is accomplished by using the SDP "sendonly" or "inactive" attributes, as defined in RFC 3264 [4]. RFC 3264 directs implementations to cease transmission of media in these cases. However, doing so may cause NAT bindings to timeout, and media won't be able to come off hold. Secondly, some RTP payload formats, such as the payload format for text conversation [36], may send packets so infrequently that the interval exceeds the NAT binding timeouts. Thirdly, if silence suppression is in use, long periods of silence may cause media transmission to cease sufficiently long for NAT bindings to time out. To prevent these problems, ICE implementations MUST continue to list their active candidate in a=candidate lines for UDP-based media streams. As a consequence of this, STUN packets will be transmitted periodically independently of the transmission (or lack thereof) of media packets. This provides a media independent, RTP independent, and codec independent solution for keeping the NAT bindings alive. If an ICE implementation is communciating with one that does not support ICE, keepalives MUST still be sent. Indeed, these keepalives Rosenberg Expires September 7, 2006 [Page 45] Internet-Draft ICE March 2006 are essential even if neither endpoint implements ICE. As such, this specification defines keepalive behavior generally, for endpoints that support ICE, and those that do not. All endpoints MUST send keepalives for each media session. These keepalives MUST be sent regardless of whether the media stream is currently inactive, sendonly, recvonly or sendrecv. The keepalive SHOULD be sent using a format which is supported by its peer. ICE endpoints allow for STUN-based keepalives for UDP streams, and as such, STUN keepalives MUST be used when an agent is communicating with a peer that supports ICE. An agent can determine that its peer supports ICE by the presence of the a=candidate attributes for each media session. If the peer does not support ICE, the choice of a packet format for keepalives is a matter of local implementation. A format which allows packets to easily be sent in the absence of actual media content is RECOMMENDED. Examples of formats which readily meet this goal are RTP No-Op [31] and RTP comfort noise [26]. STUN-based keepalives will be sent periodically every Tr seconds as a consequence of the rules in in Section 7.7. If STUN keepalives are not in use (because the peer does not support ICE), an agent SHOULD ensure that a media packet is sent every Tr seconds. If one is not sent as a consequence of normal media communications, a keepalive packet using one of the formats discussed above SHOULD be sent. 7.13 Sending Media When an agent receives an offer and sends an answer, or when it receives an answer to an offer it sent, it begins connectivity checks. These checks will include validation of the active candidate pair, if there was one. An agent SHOULD NOT send media on the active candidate pair until that candidate pair has reached the Valid or Recv-Valid state. This is to help prevent a denial-of-service attack, described in Section 13. Once the active candidate pair reaches the Valid or Recv-Valid state, an agent MAY start sending media to that candidate pair. However, offer/answer exchanges are used with protocols, like SIP, which require media to be sent "early", from the answerer to the offer, prior to completion of the initial offer/answer exchange. It is highly desirable (and sometimes necessary) for this early media to use the candidate pair ultimately selected by ICE connectivity checks. For this reason, ICE provides an early media mechanism that allows for a candidate pair to be used in one direction prior to its promotion to active in a subsequent offer/answer exchange. Note that, with ICE, early media pertains to media sent to a candidate pair until its promotion to active in a subsequent offer/answer exchange. This is a broader definition than is used in [29], which Rosenberg Expires September 7, 2006 [Page 46] Internet-Draft ICE March 2006 defines early media as media sent prior to acceptance of a call. As a consequence of the connectivity checks, an agent will change the states for each transport address pair, and consequently, for the candidate pairs. When a candidate pair becomes Valid or Recv-Valid, and the candidate pair is not equal to the active candidate pair, and the agent is in the role of answerer for that candidate pair, the agent checks the position of that pair in the candidate pair priority ordered list. If it is the first, the agent selects this candidate pair for early media. If this candidate