Internet Engineering Task Force (IETF) A. Begen Request for Comments: 6284 D. Wing Category: Standards Track Cisco ISSN: 2070-1721 T. Van Caenegem Alcatel-Lucent June 2011
Port Mapping between Unicast and Multicast RTP Sessions
This document presents a port mapping solution that allows RTP receivers to choose their own ports for an auxiliary unicast session in RTP applications using both unicast and multicast services. The solution provides protection against denial-of-service or packet amplification attacks that could be used to cause one or more RTP packets to be sent to a victim client.
Status of This Memo
This is an Internet Standards Track document.
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Further information on Internet Standards is available in Section 2 of RFC 5741.
Copyright (c) 2011 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
In (any-source or source-specific) multicast RTP applications, destination ports (i.e., the ports on which the multicast receivers receive the RTP and RTP Control Protocol (RTCP) packets) are defined declaratively. In other words, the receivers cannot choose their receive ports, and the sender(s) use the predefined ports.
In unicast RTP applications, the receiving end needs to choose its ports for RTP and RTCP since these ports are local resources and only the receiving end can determine which ports are available to use. In addition, Network Address Port Translation (NAPT, hereafter simply called NAT) devices are commonly deployed in networks; thus, static port assignments cannot be used. The receiving end may convey its request to the sending end through different ways, one of which is the Offer/Answer Model [RFC3264] for the Session Description Protocol (SDP) [RFC4566]. However, the Offer/Answer Model requires offer/ answer exchange(s) between the endpoints, and the resulting delay may not be desirable in delay-sensitive real-time applications. Furthermore, the Offer/Answer Model may be burdensome for the endpoints that are concurrently running a large number of unicast sessions with other endpoints.
In this specification, we consider an RTP application that uses one or more unicast and multicast RTP sessions together. While the declaration and selection of the ports are well defined and work well for multicast and unicast RTP applications, respectively, the usage of the ports introduces complications when a receiving end mixes unicast and multicast RTP sessions within the same RTP application.
An example scenario is where the RTP packets are distributed through source-specific multicast (SSM) [RFC4607] and a receiver sends unicast RTCP NACK feedback [RFC4585] to a local repair server (also functioning as a unicast RTCP feedback target) [RFC5760] asking for a retransmission of the packets it is missing, and the local repair server sends the retransmission packets over a unicast RTP session [RETRANSMISSION-FOR-SSM].
Another scenario is where a receiver wants to rapidly acquire a new primary multicast RTP session and receives one or more RTP burst packets over a unicast session before joining the SSM session; see [RFC6285] regarding Rapid Acquisition of Multicast RTP Sessions (RAMS). Similar scenarios exist in applications where some part of the content is distributed through multicast while the receivers get additional and/or auxiliary content through one or more unicast connections, as illustrated in Figure 1.
Begen, et al. Standards Track [Page 3]
RFC 6284 Port Mapping June 2011
In this document, we discuss this problem and introduce a solution that we refer to as port mapping. This solution allows receivers to choose their desired UDP ports for RTP and RTCP in every unicast session when they are running RTP applications using both unicast and multicast services and offer/answer exchange is not available. The solution includes a Token-based protection mechanism against denial- of-service (DoS) or packet amplification attacks that could be used to cause one or more RTP packets to be sent to a victim client. This solution is not applicable in cases where TCP is used as the transport protocol in the unicast sessions. For such scenarios, refer to [RFC4145].
Figure 1: RTP Applications Simultaneously Using Both Unicast and Multicast Services
In the remainder of this document, we refer to the RTP endpoints that serve other RTP endpoints over a unicast session as the Servers. The receiving RTP endpoints are referred to as Clients. This terminology also reflects the fact that when port mapping is used, the RTP packets can only flow in one direction (from the server to the client) in the unicast sessions.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
Token-based port mapping consists of the server providing the client a Token that can be used to establish a unicast session without the possibility of an attacker redirecting traffic to an unsuspecting third party to create a DoS attack. The Token is essentially an opaque encapsulation that is based on the client's IP address (as seen by the server), a time-to-live value, and a random nonce provided by the client.
Token-based port mapping consists of two steps: (i) Token request and retrieval, and (ii) unicast session establishment.
When a Token request is received, the server creates a Token for this particular client and sends it back to the client.
Once a Token is retrieved from a particular server, it can be used for all the unicast sessions the client will be running with this particular server until the Token expires. By default, Tokens are server specific. However, the client can use the same Token to communicate with different servers if these servers are provided with the same secret key and algorithm used to generate the Token and are at least loosely clock-synchronized.
The Token becomes invalid if the client's IP address (as seen by the server) changes (note that the client cannot necessarily detect this in a timely manner) or if the server expires the Token. In these cases, the client has to request a new Token.
As the second step, when the client wants to establish a unicast session, the client includes the Token with its RTCP feedback message. The server validates the Token, making sure that the IP address information matches. This is effective against DoS attacks, e.g., an attacker cannot simply spoof another client's IP address and start a unicast transmission towards random clients. If the validation passes, the unicast session gets established. Otherwise, the server notifies the client that the validation has failed, and in this case, the unicast session will not be established.
Upon successful validation and once the unicast session is established, all the RTP and RTCP rules specified in [RFC3550] and other relevant specifications also apply in this session until it is terminated. During the lifetime of a unicast session, a client might need to send RTCP messages that require authorization. Since such messages require a valid Token for authorization, the client needs to include the Token along with such RTCP messages as explained in detail in later sections of this document.
Begen, et al. Standards Track [Page 5]
RFC 6284 Port Mapping June 2011
Below, we first present a motivating scenario for port mapping and then describe the normative behavior and requirements.
Consider an SSM distribution network where a distribution source multicasts RTP packets to a large number of clients, and one or more retransmission servers function as feedback targets to collect unicast RTCP feedback from these clients [RFC5760]. The retransmission servers also join the multicast session to receive the multicast packets and cache them for a certain time period. When a client detects missing packets in the multicast session, it requests a retransmission from one of the retransmission servers by using an RTCP NACK message [RFC4585]. The retransmission server pulls the requested packet(s) out of the cache and retransmits them to the requesting client [RETRANSMISSION-FOR-SSM].
The RTP and RTCP flows pertaining to the scenario described above are illustrated in Figure 2. Between the client and server, we assume there exists at least one NAT device [RFC4787]. (If there are no NAT devices between the server and client, the method still works in the same fashion.) The multicast and unicast sessions are clearly identified with their individual RTP and RTCP flows and port numbers.
Figure 2: Example Scenario Showing an SSM Distribution with Support for Retransmissions from a Server
In Figure 2, we have the following multicast and unicast ports:
o Ports P1 and P2 denote the destination RTP and RTCP ports in the multicast session, respectively. The clients listen to these ports to receive the multicast RTP and RTCP packets. Ports P1 and P2 are defined declaratively.
Begen, et al. Standards Track [Page 7]
RFC 6284 Port Mapping June 2011
o Port P3 denotes the RTCP port on the feedback target running on the retransmission server to collect any RTCP packet sent by the clients, including feedback messages and RTCP receiver and extended reports. This is also the port that the retransmission server uses to send the RTP packets and RTCP sender reports in the unicast session. Port P3 is defined declaratively.
o Port P4 denotes the RTCP port on the retransmission server used to collect the RTCP receiver and extended reports for the unicast session. Port P4 is defined declaratively.
o Ports *c0, *c1, and *c2 are chosen by the client. (Note: "*" indicates that the port can be chosen randomly; once chosen, the "*" is no longer used.) *c0 denotes the port on the client used to send the RTCP reports for the multicast session. *c1 denotes the port on the client used to send the unicast RTCP feedback messages in the multicast session and to receive the RTP packets and RTCP sender reports in the unicast session. *c2 denotes the port on the client used to send the RTCP receiver and extended reports in the unicast session. Ports c0, c1, and c2 could be the same port or different ports. There are two advantages of using the same port for both c0 and c1:
1. Some NATs only keep bindings active when a packet goes from the inside to the outside of the NAT (see REQ-6 of Section 4.3 of [RFC4787]). When the gap between the packets sent from the client to the server is long, this can exceed the timeout limit. If c0=c1, the occasional (periodic) RTCP receiver reports sent from port c0 (for the multicast session's RTCP port P3) will ensure the NAT does not time out the public port associated with the incoming unicast traffic to port c1.
2. Having c0=c1 conserves NAT port bindings.
o Ports PT and *cT denote the ports through which the Token request and retrieval occur at the server and client sides, respectively. Port PT is declared on a per-unicast-session basis, although the same port could be used for two or more unicast sessions sourced by the server. A Token once requested and retrieved by a client from port PT remains valid until its expiration time.
We assume that the information declaratively defined is available as part of the session description information and is provided to the clients. The Session Description Protocol (SDP) [RFC4566] and other session description methods can be used for this purpose.
In this section, we describe the normative behavior and requirements. To simplify the presentation, we refer to the port numbers described in the example presented in Figure 2. However, the behavior and requirements described here are not specific to that particular example and can be applied to any scenario where analogous ports can be identified.
First of all, a client compliant with this specification MUST be able to include a Token with any type of RTCP message (as described below) when it is needed.
Second, the solution provided in this specification is not applicable in cases where there is RTP traffic flowing from the client to the server in the unicast session. In other words, the direction of RTP traffic MUST be only from the server to the client in the unicast session. If the client wants to send RTP traffic back to the server, the regular session establishment methods such as [RFC3264] need to be used.
The following steps summarize the Token-based solution:
1. The client ascertains server address and port numbers (P3, P4 and PT) from the session description. Port P4 MUST be different from port P3. Port PT MAY be equal to port P3.
2. The client selects its local port numbers (*c0, *c1, *c2 and *cT). It is strongly RECOMMENDED that the client uses the same port for c0 and c1. Port cT MAY be equal to ports c0 and c1.
3. If the client does not have a Token (or the existing Token has expired):
A. The client first sends a Port Mapping Request message (Section 4.1) to port PT. This message is sent from port cT on the client side. The server learns the client's IP address from the received message. The client can send this message anytime it wants (e.g., during initialization) and does not normally ever need to resend this message (see Section 6).
B. The server generates an opaque encapsulation (i.e., the Token) based on certain information, including the client's IP address.
Begen, et al. Standards Track [Page 9]
RFC 6284 Port Mapping June 2011
C. The server sends the Token back to the client using a Port Mapping Response message (Section 4.2). This message MUST be sent from port PT towards port cT.
4. The client needs to provide the Token to the server using a Token Verification Request message (Section 4.3) whenever the client sends an RTCP feedback message for triggering or controlling a unicast session (see Section 4.3.1). If the Token is invalid or missing, the server sends a Token Verification Failure message (Section 4.4) to the client.
Note that the unicast session is only established after the server has received a feedback message (along with a valid Token) from the client for which it needs to react by sending unicast data. Until a unicast session is established, neither the server nor the client needs to send RTCP reports for the unicast session.
5. Normal flows ensue as shown in Figure 2. It is strongly RECOMMENDED that the client uses the same port for both c0 and c1, as this causes the periodic RTCP reports to keep the NAT mapping alive. However, if the client uses different ports for c0 and c1, the client MUST keep its own NAT mapping alive for the P3->c1 session (see [RFC6263] for additional information).
In the unicast session, traffic from the server to the client (i.e., both the RTP and RTCP packets sent from port P3 towards port c1) MUST be multiplexed on the same port [RFC5761].
The client sends the RTCP receiver and extended reports in the unicast session from port c2 towards port P4. The server correlates these reports with the reports received in the multicast session based on the client's RTCP Canonical Name (CNAME). Thus, the client MUST use the same RTCP CNAME in both sessions, and its RTCP CNAME MUST be unique [RFC6222].
A unicast session on a particular receive port c1 can last as long as the associated multicast session lasts. However, a client cannot keep using the same receive port c1 for different subsequent unicast sessions since there could be packet leakage when switching from one unicast session to another unless each received unicast stream has its own distinct Synchronization Source (SSRC) identifier to allow the client to filter out the undesired packets. Unless this is guaranteed (which is not often easy), a client SHOULD use separate receive ports for subsequent unicast sessions. After a sufficient time (two minutes is RECOMMENDED, similar to one TCP Maximum Segment Lifetime specified in [RFC0793]), a previously used receive port can be used again.
Begen, et al. Standards Track [Page 10]
RFC 6284 Port Mapping June 2011
The established unicast session can be explicitly terminated by the procedures specified by an application or extension using the port mapping approach described in this document. In addition, the unicast session can also be terminated by the procedure defined below, which is based on timing all participants out following the timeout rules of [RFC3550]. Both the server and client periodically check the liveness of the other peer, and if there is no RTCP traffic from the other peer for a certain amount of time (Section 6.3.5 of [RFC3550] suggests five RTCP reporting intervals), the unicast session SHOULD be considered terminated, and no further RTP and/or RTCP packets SHOULD be sent in that session. The client can attempt to establish a new unicast session if needed. If no explicit procedure for session termination exists, the client MAY stop sending RTCP to the server to accomplish session termination. However, the server SHALL NOT stop sending RTCP until the unicast session is terminated. If Token-based authentication is also signaled to be allowed in the unicast session, i.e., in the RTCP messages sent from port c2 towards port P4, the client SHOULD terminate the unicast session by sending an RTCP BYE message for each SSRC it has used in that unicast session.
This section defines the formats of the RTCP messages that are exchanged between a server and a client for the purpose of port mapping. A new RTCP control packet type is introduced, and four port mapping messages using this control packet are defined:
1. Port Mapping Request
2. Port Mapping Response
3. Token Verification Request
4. Token Verification Failure
Each message has a fixed-length field for version (V), padding (P), sub-message type (SMT), packet type (PT), length, and SSRC of packet sender. Messages have other fields as defined below. In all messages defined in this section, the PT field is set to TOKEN (210). Individual messages are identified by the SMT field. The length field indicates the message size in 32-bit words minus one, including the header and any padding. This definition is in line with the definition of the Length field used in RTCP sender and receiver reports. In all messages, any Reserved field SHALL be set to zero and ignored.
Begen, et al. Standards Track [Page 11]
RFC 6284 Port Mapping June 2011
Following the rules specified in [RFC3550], all integer fields in the messages defined below are carried in network-byte order, that is, most significant byte (octet) first, also known as big-endian. Unless otherwise stated, numeric constants are in decimal (base 10).
Note that RTCP is not a timely or reliable protocol. The RTCP packets might get lost or reordered in the network, and it is not easy to detect these events. When sending a new Port Mapping Request message, the scheduling rules that apply to sending initial RTCP messages [RFC4585] apply. When a client sends a Port Mapping Request or Token Verification Request message but it does not receive a response back from the server (either a Port Mapping Response or Token Verification Failure message), it MAY resend its request by following the timer rules defined for RTCP feedback messages in Section 3.5 of [RFC4585] as a good practice. However, implementations are advised to avoid sending spurious RTCP messages just because the timer rules (based on some RTCP configuration parameters) allow. Reasonably safe practices are to be used to detect RTCP message loss. When sending an RTCP (feedback) message bundled with a Token Verification Request message, the timer rules of [RFC4585] apply as usual.
The Port Mapping Request message is identified by SMT=1. This message is transmitted by the client to a dedicated server port (and possibly a dedicated address) to request a Token. In the Port Mapping Request message, the packet sender's SSRC is set to the client's SSRC, which is chosen randomly by the client. The packet format has the structure depicted in Figure 3.
Figure 3: Packet Format for the Port Mapping Request Message
o Random Nonce (64 bits): Field that contains a random value generated by the client following the procedures of [RFC4086]. This nonce is taken into account by the server when generating a Token for the client to enable better security for clients that
Begen, et al. Standards Track [Page 12]
RFC 6284 Port Mapping June 2011
share the same IP address (such clients need to produce a random value extremely unlikely to collide with other clients sharing the same IP address). If the same Port Mapping Request message is transmitted multiple times for redundancy reasons, the random nonce value MUST remain the same in these duplicated messages. However, the client MUST generate a new random nonce for every new Port Mapping Request message.
The Port Mapping Response message is identified by SMT=2. This message is sent by the server and delivers the Token to the client as a response to the Port Mapping Request message. In the Port Mapping Response message, the packet sender's SSRC is set to the server's SSRC. The packet format has the structure depicted in Figure 4.
Figure 4: Packet Format for the Port Mapping Response Message
o SSRC of Requesting Client (32 bits): Field that contains the SSRC of the client who sent the request.
o Associated Nonce (64 bits): Field that contains the nonce received in the Port Mapping Request message and used in Token construction.
Begen, et al. Standards Track [Page 13]
RFC 6284 Port Mapping June 2011
o Token Element (variable size): Element that is used to carry the Token generated by the server. This element is a 32-bit aligned Length-Value element. The Length field, which is 16 bits, indicates the length (in octets) of the Value field that follows the Length field. While a 16-bit length allows for Tokens with a size of up to 65535 bytes, using Tokens of sizes that make the RTCP compound packet larger than the MTU might have a negative impact on functionality because of IP fragmentation. Some NATs or other middleboxes do not pass IP fragments; thus, a large Token can cause the whole mechanism to fail. In addition, fragmentation increases the risk for packet loss.
The length does not include any padding that is required for alignment. The Value field carries the Token (or more accurately, the output of the encoding process on the server). If the Token element does not fall on a 32-bit boundary, the last word MUST be padded to the boundary using further bits set to zero.
o Absolute Expiration Time (64 bits): Field that contains the absolute expiration time of the Token. The absolute expiration time is expressed as a Network Time Protocol (NTP) timestamp value in seconds since the year 1900 [RFC5905]. The client does not need to use this element directly and thus does not need to synchronize its clock with the server. However, the client needs to send this element back to the server along with the associated nonce in the Token Verification Request message and thus needs to keep it associated with the Token.
o Relative Expiration Time (32 bits): Field that contains the relative expiration time of the Token. The relative expiration time is expressed in seconds from the time the Token was generated. Whenever a server decides to not grant a Token to a requesting client, the relative expiration time will be set to zero (and hence, the accompanying Token will be invalid).
The server conveys the relative expiration time in the clear to the client to allow the client to request a new Token well before the expiration time.
o Packet Types Element (variable size): Element that is used to signal which RTCP packet types require Token-based authentication. This element is a 32-bit aligned Length-Value element. The Length field, which is 8 bits, indicates the length (in octets) of the Value field that follows the Length field. This length does not include any padding that is required for alignment. The Value field carries zero or more 8-bit sub-fields, each carrying an RTCP packet type. If the Packet Types element does not fall on a
Begen, et al. Standards Track [Page 14]
RFC 6284 Port Mapping June 2011
32-bit boundary, the last word MUST be padded to the boundary using further bits set to zero. An example Packet Types element is shown in Figure 5.
A server MAY change its policy on which RTCP packet types would require Token-based authentication based on observations, configuration, or other policies. However, upon such a change, the server SHALL NOT send a new Port Mapping Response message to the clients who requested a Token earlier. A client learns about this change when and if it gets a Token Verification Failure message.
The Token Verification Request message is identified by SMT=3. This message contains the Token and accompanies any RTCP message that would trigger a new unicast session or control an existing unicast session. For a list of such messages, see Section 4.3.1.
In the Token Verification Request message, the packet sender's SSRC is set to the client's SSRC. The client MUST NOT send a Token Verification Request message with a Token that has expired. The packet format has the structure depicted in Figure 6.
This section provides guidelines about which RTCP packet types would need to be accompanied by a Token Verification Request message. However, since a server might determine in real time that other RTCP messages also need to be authenticated by a Token, a client MUST act according to the up-to-date list provided to the client in the Port Mapping Response message (in the Packet Types element). Clients need to support the use of Token-based authentication with any necessary RTCP message (see Section 3.2).
As a general rule, when the Token capability is declared in the session description, the RTCP messages that trigger transmission of RTP packets in a port mapped unicast session are REQUIRED to be authenticated by using a Token. Such messages include but are not limited to:
Additionally, some RTCP messages might directly or indirectly control an existing unicast session associated with a multicast session. Unless another authentication method as described in their respective specifications is used, implementations MUST support authenticating such RTCP messages by using a Token.
Note that even if a packet type is listed to require Token-based authentication, it does not need to be authenticated when it does not control the unicast session. For example, if BYE (203) is listed in the Port Mapping Response message as one of the packet types that requires authentication, the client does not need to bundle the RTCP BYE message with a Token when it is sending it for the multicast session.
The Token Verification Request message might also be bundled with packets carrying RTCP receiver and/or extended reports. While such packets do not have a strong security impact, a specific application might desire to have a more controlled reporting scheme from the clients. In this case, the server lists the packet types for the receiver (201) and/or extended reports (207) in the Port Mapping Response message.
The Token Verification Failure message is identified by SMT=4. This message is sent by the server and notifies the client that the Token was invalid or that the client did not include a Token Verification Request message in the RTCP packet although it was supposed to (the message is sent from port P3 towards port c1). The packet format has the structure depicted in Figure 7.
Figure 7: Packet Format for the Token Verification Failure Message
o SSRC of Packet Sender: This is the server's SSRC, which equals the SSRC of the respective multicast stream. Note that this SSRC value is from a different SSRC space than the one used in the unicast session.
o SSRC of Requesting Client (32 bits): Field that contains the SSRC of the client.
o Failed PT (8 bits): Field that indicates the type of the RTCP packet that caused this failure message.
o FMT (5 bits): Field that indicates the feedback message type (FMT) value of the RTCP packet that caused this failure. Together with the field above, the client can infer which RTCP message it had previously sent caused this failure message to be sent by the server. For example, if the client did not include a valid Token with an RTCP NACK message, the Failed PT field will indicate 205 (RTPFB) and the FMT field will indicate 1 (Generic NACK). If the RTCP message did not have an associated FMT value (such as an RTCP BYE message), the FMT field SHALL be set to zero.
o Associated Nonce (64 bits): Field that contains the nonce received in the Token Verification Request message. If there was no Token Verification Request message included by the client, this field is set to zero.
The Token encoding is known to the server but opaque to the client. Implementations MUST encode the following information into the Token as a minimum, in order to provide adequate security:
Begen, et al. Standards Track [Page 18]
RFC 6284 Port Mapping June 2011
o Client's IP address as seen by the server (32/128 bits for IPv4/ IPv6 addresses)
o The nonce generated and inserted in the Port Mapping Request message by the client (64 bits)
o The absolute expiration time chosen by the server indicated as an NTP timestamp value in seconds since the year 1900 [RFC5905] (64 bits, to protect against replay attacks)
The RECOMMENDED way for constructing Tokens is to perform HMAC-SHA1 [RFC2104] on the concatenated values of the information listed above (implementations might adopt different approaches). If HMAC-SHA1 is used, the Hashed Message Authentication Code (HMAC) key MUST be at least 160 bits long and generated using a cryptographically secure random source [RFC4086].
In addition to the information listed above, implementations are encouraged to encode whatever additional information is deemed necessary or useful. For example, key rollover is simplified by encoding a key-id into the Token. As another example, a cluster of anycast servers could find advantage by encoding a server identifier into the Token. As another example, while HMAC-SHA1 provides a level of security that is widely regarded as being more than sufficient for providing message authentication and it is secure against all known cryptanalytic attacks that use computational resources that are currently economically feasible, a replacement HMAC algorithm (e.g., HMAC-SHA256) could be used instead if HMAC-SHA1 has been compromised.
To protect from offline attacks, the server SHOULD occasionally choose a new HMAC key. To ease implementation, a key-id can be assigned to each HMAC key. This can be encoded as simply as one bit (where the new key is X (e.g., 1) and the old key is the inverted value of X (e.g., 0)), or if several keys are supported at once, the key-id could be encoded into several bits. As the encoding of the Token is entirely private to the server and opaque to the clients, any encoding can be used. By encoding the key-id into the Token element, the server can reject an old key without bothering to do HMAC validation (saving CPU cycles). The key-id can be encoded into the Value field of the Token element by simply concatenating the (plaintext) key-id with the hashed information (i.e., the Token itself).
For example, the Value field in the Token element can be computed as:
During Token construction, the expiration time has to be chosen carefully based on the intended service duration. Tokens that are valid for an unnecessarily long period of time (e.g., several hours) might impose security risks. Depending on the application and use cases, a reasonable value needs to be chosen by the server. Note that using shorter lifetimes requires the clients to acquire Tokens more frequently. However, since a client can acquire a new Token well before it will need to use it, the client will not necessarily be penalized for the acquisition delay.
Finally, be aware that NTP timestamps will wrap around in the year 2036. Refer to Section 6 of [RFC5905] for further details.
The server MUST validate the Token upon receipt of an RTCP feedback message along with the Token Verification Request message that contains a Token, nonce, and absolute expiration time.
The server first applies its own procedure for constructing the Tokens by using the client's IP address from the received Token Verification Request message and the nonce and absolute expiration time values reported in the received Token Verification Request message. The server then compares the resulting output with the Token sent by the client in the Token Verification Request message. If they match and the absolute expiration time has not passed yet, the server declares that the Token is valid.
Note that if the client's IP address changes, the Token will not validate. Similarly, if the client inserts an incorrect nonce or absolute expiration time value in the Token Verification Request message, validation will fail. It is also possible that the server wants to expire the Token prematurely. In these cases, the server MUST reply back to the client with a Token Verification Failure message (that goes from port P3 on the server towards port c1 on the client).
In addition to the Token Verification Failure message, it is RECOMMENDED that applications define an application-specific error response to be sent by the server when the server detects that the Token is invalid. For applications using [RFC6285], this document defines a new 4xx-level response code in the RAMS Response Code Space Registry. A client that receives a Token Verification Failure message can request a new Token from the server.
If a client receives a Port Mapping Response message with an invalid Token (i.e., the relative expiration time is set to zero) two or more times for a particular Port Mapping Request message or the client
Begen, et al. Standards Track [Page 20]
RFC 6284 Port Mapping June 2011
receives a Token Verification Failure message two or more times for the same Token Verification Request message, the client SHOULD do the following:
1. Check whether or not the session description has been updated. If it was updated, act according to the new session description.
2. Exponentially back off for the third and subsequent attempts. Exponential back-off does not apply when the client sends a Port Mapping Request or Token Verification Request message to a new address and/or port.
This attribute is used declaratively in any media block that describes an RTP session that uses Token-based authentication for one or more RTCP messages relating to that session. It indicates the port and optionally the address for obtaining a Token.
The presence of the 'portmapping-req' attribute indicates that (i) a Token MUST be included in certain RTCP messages sent to the server triggering or controlling a unicast session (see Section 4.3.1) and (ii) the client MUST receive the unicast session's RTP and RTCP packets from the server on the port from which it sent the RTCP message triggering the unicast session.
Note: This does not imply that Token Verification Request messages always need to be sent in the unicast session. Token Verification Request messages accompany RTCP messages that trigger or control this unicast session and are sent either in the multicast session or the unicast session, depending on the RTCP message (see Section 4.3.1).
Here, 'port', 'nettype', 'addrtype', and 'connection-address' are defined as specified in Section 9 of [RFC4566].
The 'portmapping-req' attribute SHALL only be used as a media-level attribute.
Begen, et al. Standards Track [Page 21]
RFC 6284 Port Mapping June 2011
In the optional address value, only unicast addresses SHOULD be used unless one wants to use a multicast address after evaluating the additional security risks such as non-legit servers generating fake Tokens. If the address is not specified, the (source) address in the "c" line applicable to the media description SHALL be used.
When using the 'portmapping-req' attribute in SDP offer/answer exchanges [RFC3264], the following considerations apply. When an offerer sends an answerer an offer of an SDP description making use of the Token approach described in this specification, the 'portmapping-req' attribute is included declaratively. There will not be offer/answer exchanges between the answerer and the actual server providing the unicast service(s).
When the answerer supports the Token approach, it MUST echo in its answer back to the offerer the 'portmapping-req' attribute from the offer including the same port number and address (if any). If the answerer does not implement this specification, it follows normal SDP parsing of unknown attributes (they are ignored and are not sent in the answer). This means that the answerer can still join the multicast session but will not be able to use the unicast service(s) that require the use of Tokens.
Figure 8: SDP Describing an SSM Distribution with Support for Retransmissions from a Local Server
In this description, we highlight the following notes:
Note 1: The source stream is multicast from a distribution source with a source IP address of 198.51.100.1 to the multicast destination address of 184.108.40.206 and port 41000 (P1). The associated RTCP packets are multicast in the same group to port 41500 (P2).
Note 2: A retransmission server including feedback target functionality with an IP address of 192.0.2.1 and port of 42000 (P3) is specified with the 'rtcp' attribute. The feedback functionality is enabled for the RTP stream with payload type 98 through the 'rtcp-fb' attribute [RFC4585].
Begen, et al. Standards Track [Page 23]
RFC 6284 Port Mapping June 2011
Note 3: The "a=portmapping-req" line indicates that one or more RTCP messages relating to the RTP session described in this media block uses Token-based authentication, and a Token needs to be retrieved first from the designated port (PT) before the unicast session can be established. In the first appearance, an explicit address is provided. In the second appearance, there is no address indicated in this line and the client needs to send the Token request to the address specified in the "c" line in the unicast media block.
Note 4: The port specified in the second "m" line (for the unicast stream) does not mean anything in this scenario as the client does not send any RTP traffic back to the server.
Note 5: The server multiplexes RTP and RTCP packets sent towards c1 on the same port.
Note 6: The server uses port 42500 (P4) for the unicast session.
Large-scale NAT devices have a pool of public IPv4 addresses and map internal hosts to one of those public IPv4 addresses. As long as an internal host maintains an active mapping in the NAT, the same IPv4 address is assigned to new connections. However, once all of the host's mappings have been deleted (e.g., because of timeout), it is possible that a new connection from that same host will be assigned a different IPv4 address from the pool. When that occurs, the Token will be considered invalid by the server, causing an additional round trip for the client to acquire a fresh Token.
Any traffic from the host that traverses the NAT will prevent this problem. As the host is sending RTCP receiver reports at least every 5 seconds (Section 6.2 of [RFC3550]) for the multicast session it is receiving, those RTCP messages will be sufficient to prevent this problem.
The Token, which is generated based on a client's IP address and expiration date, provides protection against off-path denial-of- service (DoS) attacks. An attacker using a certain IP address cannot cause one or more RTP packets to be sent to a victim client who has a different IP address. However, if the attacker acquires a valid Token for a victim and can spoof the victim's source address, this
Begen, et al. Standards Track [Page 24]
RFC 6284 Port Mapping June 2011
approach becomes vulnerable to replay attacks. This is especially easy if the attacker and victim are behind a large-scale NAT and share the same IP address.
Multicast is deployed on managed networks, not the Internet. These managed networks will choose whether or not to enable network ingress filtering [RFC2827]. If ingress filtering is enabled on a network, an attacker cannot spoof a victim's IP address to use a Token to initiate an attack against a victim. However, if ingress filtering is not enabled on a network, an attacker could obtain a Token and spoof the victim's address, causing traffic to flood the victim. On such a network, the server can reduce the time period for such an attack by expiring a Token in a short period of time. In the extreme case, the server can expire the Token in such a short period of time that the client will have to acquire a new Token immediately before using it in a Token Verification Request message. One should, however, note that such a behavior might have an adverse effect on the delay in establishing or controlling a unicast session.
RTCP messages could be subject to on-path or man-in-the-middle attacks. For example, an attacker can modify a value in one or more fields in the Port Mapping Response or the Token Verification Request message that are used in Token construction. This will result in Token validation failure. Consequently, the client ends up asking the server to generate a new Token. The resulting delay and extra processing on the server is undesirable.
Alternatively, the attacker can modify a value in a field that is not used in Token construction. For example, the attacker can reduce the value in the Relative Expiration Time field in the Port Mapping Response message from two hours to two minutes. While the Token will still validate, this attack will result in more frequent requests to the server for a new Token. Oppositely, the attacker can increase the value in the Relative Expiration Time field and make the client think the Token will be valid for a longer time. This attack can be only detected by monitoring the activity on the server. Note that using the relative expiration time in Token construction does not necessarily make this attack easier to detect since the attacker might revert the modified value back to its original value in the Token Verification Request message. This allows the Token to still validate on the server. In this case, the attack is still only detectable by monitoring the server activity.
If there is a risk or concern for on-path or man-in-the-middle attacks, RTCP messages SHOULD be protected by Secure RTCP (SRTCP) [RFC3711].
Begen, et al. Standards Track [Page 25]
RFC 6284 Port Mapping June 2011
To minimize the risk of cross-protocol attacks, a server MUST NOT use the same secret key it used for Token construction for other purposes.
The 'portmapping-req' attribute is not believed to introduce any significant security risk to multimedia applications. A malevolent third party could use this attribute to redirect the Port Mapping Request messages by altering the port number or cause the unicast session establishment to fail by removing it from the SDP description. However, this requires intercepting and rewriting the packets carrying the SDP description, and if an interceptor can do that, many more attacks are possible, including a wholesale change of the addresses and port numbers at which the media will be sent.
In order to avoid attacks of this sort, the SDP description needs to be integrity protected and provided with source authentication. This can, for example, be achieved on an end-to-end basis using Secure/ Multipurpose Internet Mail Extensions (S/MIME) [RFC5652] [RFC5751] when SDP is used in a signaling packet using MIME types (application/ sdp). Alternatively, HTTPS [RFC2818] or the authentication method in the Session Announcement Protocol (SAP) [RFC2974] could be used as well.
This document registers one new attribute name in SDP.
SDP Attribute ("att-field"): Attribute name: portmapping-req Long form: Port and address for requesting Token Type of name: att-field Type of attribute: Media level Subject to charset: No Purpose: See this document Reference: [RFC6284] Values: See this document
This document creates a new sub-registry for the sub-message type (SMT) values to be used with the TOKEN packet type. The registry is called the SMT Values for TOKEN Packet Type Registry. This registry is managed by the IANA according to the IETF Review policy of [RFC5226].
The length of the SMT field is five bits, allowing 32 values. The registry is initialized with the following entries:
Value Name Reference ----- -------------------------------------------------- ------------ 0 Reserved [RFC6284] 1 Port Mapping Request [RFC6284] 2 Port Mapping Response [RFC6284] 3 Token Verification Request [RFC6284] 4 Token Verification Failure [RFC6284] 5-30 Unassigned IETF Review 31 Reserved [RFC6284]
The SMT values 0 and 31 are reserved for future use.
The approach presented in this document came out after discussions with various individuals in the AVT and MMUSIC WGs and the breakout session held at the Anaheim meeting. We thank each of these individuals, particularly Magnus Westerlund and Colin Perkins.