|
 SCTP: An Overview Introduction SCTP Core Features: Multi-streaming SCTP Core Features: Multi-homing Other SCTP Features Features of the SCTP Initiation Procedure Cookie Mechanism INIT Collision Resolution SCTP DATA Exchange Features SCTP Shutdown Features SCTP Message Format Error Cases – Retransmission Error Cases – Path Failure Error Cases – Endpoint Failure API Security Extensions Introduction The Stream Control Transmission Protocol (SCTP) is a new IP transport protocol, existing at an equivalent level as UDP (User Datagram Protocol) and TCP (Transmission Control Protocol), which currently provide transport layer functions to all of the main Internet applications. SCTP has been approved by the IETF as a Proposed Standard, and is currently awaiting allocation of an RFC number. Like TCP, SCTP provides a reliable transport service, ensuring that data is transported across the network without error and in sequence. Like TCP, SCTP is a connection-oriented mechanism, meaning that an relationship is created between the endpoints of an SCTP session prior to data being transmitted, and this relationship is maintained until all data transmission has been successfully completed. Unlike TCP, SCTP provides a number of functions that are considered critical for signaling transport, and which at the same time can provide transport benefits to other applications requiring additional performance and reliability. The core features of SCTP are multi-streaming and multi-homing. SCTP Core Features: Multi-streaming The name Stream Control Transmission Protocol is derived from the multi-streaming function provided by SCTP. This feature allows data to be partitioned into multiple streams that have the property of being delivered independently, so that message loss in any of the streams will only affect delivery within that stream, and not in other streams. In contrast, TCP provides a single stream of data and ensures that delivery of that stream takes place with perfect sequence preservation. While this is desirable for delivery of a file or record, it causes additional delay when message loss or sequence error occurs within the network. When this happens, TCP must delay delivery of data until the correct sequencing is restored, either by receipt of an out-of-sequence message, or by retransmission of a lost message. For a number of applications, this characteristic of strict sequence preservation is not truly necessary. In signaling, for example, it is only necessary to maintain sequencing of messages that affect the same resource (e.g., the same call, or the same channel). Other messages are only loosely correlated and can be delivered without having to maintain overall sequence integrity. Another application having this property is delivery of web page objects, if done over a single session. It is generally not necessary to maintain sequence between the presentation of objects, and in some cases (esp. some types of graphic images) it may be possible to present parts of a single object out of sequence. Eventually the goal is to deliver all objects, however the ability to deliver objects out of sequence may result in at least better perceived performance, as parts of the web page can be displayed rather than waiting for all of the information to be received before displaying it. SCTP accomplishes multi-streaming by creating independence between data transmission and data delivery. In particular, each DATA “chunk” (or PDU) in the protocol uses two sets of sequence numbers, a Transmission Sequence Number that governs the transmission of messages and the detection of message loss, and the Stream ID/Stream Sequence Number pair, which is used to determine the sequence of delivery of received data. This independence of mechanisms allows the receiver to determine immediately when a gap in the transmission sequence occurs (e.g., due to message loss), and also whether or not messages received following the gap are within an affected stream or not. If a message is received within the affected stream, there will be a corresponding gap in the Stream Sequence Number, while messages from other streams will not show a gap. The receiver can therefore continue to deliver messages to the unaffected streams while buffering messages in the affected stream until retransmission occurs. SCTP Core Features: Multi-homing The other core feature of SCTP is multi-homing, or the ability for a single SCTP endpoint to support multiple IP addresses. The benefit of multi-homing is potentially greater survivability of the session in the presence of network failures. In a conventional single-homed session, the failure of a local LAN access can isolate the end system, while failures within the core network can cause temporary unavailability of transport until the IP routing protocols can reconverge around the point of failure. Using multi-homed SCTP, redundant LANs can be used to reinforce the local access, while various options are possible in the core network to reduce the dependency of failures for different addresses. Use of addresses with different prefixes can force routing to go through different carriers, for example, while route-pinning techniques or even redundant core networks can also be used if there is control over the network architecture and protocols. In its current form, SCTP does not do load-sharing, that is, multi-homing is used for redundancy purposes only. A single address is chosen as the “primary” address and is used as the destination for all DATA chunks for normal transmission. Retransmitted DATA chunks use the alternate address(es) to improve the probability of reaching the remote endpoint, while continued failure to send to the primary address ultimately results in the decision to transmit all DATA chunks to the alternate until heartbeats can reestablish the reachability of the primary. To support multi-homing, SCTP endpoints can exchange lists of addresses during initiation of the association. Each endpoint must be able to receive messages from any of the addresses associated with the remote endpoint; in practice, certain operating systems will utilize available source addresses in round robin fashion, in which case receipt of messages from different source addresses will be the normal case. A single port number is used across the entire address list at an endpoint for a specific session. In order to reduce the potential for security problems, it is required that some response messages be sent specifically to the source address in the message that caused the response. For example, when the server receives an INIT chunk from a client to initiate an SCTP association, the server always sends the response INIT ACK chunk to the source address that was in the IP header of the INIT. Other SCTP Features SCTP is a unicast protocol, and supports data exchange between exactly 2 endpoints, although these may be represented by multiple IP addresses. SCTP provides reliable transmission, detecting when data is discarded, reordered, duplicated or corrupted, and retransmitting damaged data as necessary. SCTP transmission is full duplex. SCTP is message oriented and supports framing of individual message boundaries. In comparison, TCP is stream oriented and does not preserve any implicit structure within a transmitted byte stream. SCTP is rate adaptive similar to TCP, and will scale back data transfer to the prevailing load conditions in the network. It is designed to behave cooperatively with TCP sessions attempting to use the same bandwidth. Features of the SCTP Initiation Procedure The SCTP Initiation Procedure relies on a 4 message sequence, where DATA can be included on the 3rd and 4th messages of the sequence, as these messages are sent when the association has already been validated. A “cookie” mechanism has been incorporated into the sequence to guard against some types of denial of service attacks.
The “cookie” mechanism guards specifically against a blind attacker generating false INIT chunks to try to overload the resources of an SCTP server by causing it to use up memory and resources handling new INIT requests. Rather than allocating memory for a Transmission Control Block (TCB), the server instead encrypts enough information to build its TCB an internal key and sends this as a Cookie parameter in the INIT ACK. Since the INIT ACK always goes back to the source address of the INIT, the blind attacker will not get the Cookie. A valid SCTP client will get the Cookie and return it in the COOKIE ECHO chunk, where the SCTP server can decrypt the Cookie and verify that it is valid – note that the SCTP client never has to understand the Cookie, so that no exchange of keys is required. Otherwise, the SCTP Initiation Procedure follows many TCP functions, so that the endpoints exchange receiver windows, initial sequence numbers, etc. In addition to this, the endpoints may exchange address lists as discussed above, and also mutually confirm the number of streams to be opened on each side. INIT Collision Resolution Multi-homing adds to the potential that messages will be received out of sequence or with different address pairs. This is a particular concern during initiation of the association, where without procedures for resolving collision of messages, you may easily end up with multiple parallel associations between the same endpoints. To avoid this, SCTP incorporates a number of procedures to resolve parallel initiation attempts into a single association. DATA chunk exchange in SCTP follows TCP’s Selective ACK procedure. Receipt of DATA chunks is acknowledged by sending SACK chunks, which indicate not only the cumulative Transmission Sequence Number (TSN) range received, but also any non-cumulative TSNs, implying gaps in the received TSN sequence. Following TCP procedures, SACKs are sent using the “delayed ack” method, normally one SACK per every other received packet, but with an upper limit on the delay between SACKs and an increase to once per received packet when there are gaps detected. Flow and Congestion Control follow TCP algorithms. The advertised receive window indicates buffer occupancy at the receiver, while a per-path congestion window is maintained to manage the packets in flight. Slow start, Congestion avoidance, Fast recovery and Fast retransmit are incorporated into the procedures as described in RFC2581, with the one change being that the endpoints must manage the conversion between bytes sent and received and TSNs sent and received, since TSN is per chunk rather than per byte. The application can specify a lifetime for data to be transmitted, so that if the lifetime has expired and the data has not yet been transmitted, it can be discarded (e.g., time-sensitive signaling messages). If the data has been transmitted, it must continue to be delivered to avoid creating a hole in the TSN sequence. SCTP Shutdown Features SCTP Shutdown uses a 3 message procedure to allow graceful shutdown, where each endpoint has confirmation of the DATA chunks received by the remote endpoint prior to completion of the shutdown. An Abort procedure is also provided for error cases when an immediate shutdown must take place. Note that SCTP does not support the function of a “half-open” connection as can occur in TCP, when one side indicates that it has no more data to send, but the other side can continue to send data indefinitely. SCTP assumes that once the shutdown procedure begins, both sides will stop sending new data across the association, and only need to clear up acknowledgements of previously sent data. SCTP Message Format The SCTP Message includes a common header plus one or more chunks, which can be control or data. The common header has source and destination port numbers to allow multiplexing of different SCTP associations at the same address, a 32-bit verification tag that guards against insertion of an out-of-date or false message into the SCTP association, and a 32-bit checksum (the Adler checksum, described in RFC1950) for end-to-end error detection. Each chunk includes chunk type, flag field, length and value. Control chunks incorporate different flags and parameters depending on the chunk type. DATA chunks in particular incorporate flags for control of segmentation and reassembly, and parameters for the TSN, Stream ID and Stream Sequence Number, and a Payload Protocol Identifier. The Payload Protocol ID has been included for future flexibility. Initially, values will reflect the adaptation layer protocol running over SCTP, and will match the registered port number assignments for adaptation layers. In future, it is envisioned that the functions of protocol identification and port number multiplexing will not be as closely linked so that the port number significance is not so much overloaded as it is now. The SCTP message format naturally allows support of bundling of multiple DATA and control chunks in a single message, to improve transport efficiency. Use of bundling is controllable by the application, so that bundling of initial transmission can be prohibited. Bundling will naturally occur on retransmission of DATA chunks, to further reduce any chance of congestion. Error Cases – Retransmission Retransmission of DATA chunks occurs from either (a) timeout of the retransmission timer; or (b) receipt of SACKs indicating the DATA chunk has not been received. To reduce the potential for congestion, the rate of retransmission of DATA chunks is limited. The retransmission timeout (RTO) is adjusted based on estimates of the round trip delay and backs off exponentially as message loss increases. In an active association with fairly constant DATA transmission, SACKs are more likely to cause retransmission than the timeout. To reduce the chance of an unnecessary retransmission, a 4 SACK rule is used, so that retransmission only occurs on receipt of the 4th SACK that indicates that the chunk is missing. This is intended to avoid retransmits due to normal occurrences such as packets received out of sequence. Error Cases – Path Failure A count is maintained of the number of retransmissions to a particular destination address without successful acknowledgement. When this count exceeds a configured maximum, the address is declared inactive, notification is given to the application, and the SCTP begins to use an alternate address for sending of DATA chunks. Also, Heartbeat chunks are sent periodically to all idle destinations (i.e., alternate addresses), and a counter is maintained on the number of Heartbeats sent to an inactive destination without receipt of a corresponding Heartbeat Ack. When this counter exceeds a configured maximum, that destination address is also declared inactive. Heartbeats continue to be sent to inactive destination addresses until an Ack is received, at which point the address can be made active again. The rate of sending of Heartbeats is tied to the RTO estimation plus an additional delay, to allow Heartbeat traffic to be tailored to the needs of the application. Error Cases – Endpoint Failure A count is maintained across all destination addresses on the number of retransmits or Heartbeats send to the remote endpoint without successful Ack. When this exceeds a configured maximum, the endpoint is delared unreachable, and the SCTP association is closed. API The specification includes a model of the primitives exchanged between the application and the SCTP layer, intended as informational material rather than a formal API statement. A follow-on document will provide a socket-like API to simplify any migration of TCP or UDP applications to the use of SCTP. Security In addition to the verification tag and cookie mechanisms, SCTP specifies the use of IPSec if strong security and integrity protection is required. The SCTP specification does not itself define any new security protocols or procedures. Extensions The SCTP format allows new chunk types, flags and parameter fields to be defined as extensions to the protocol. Any extensions must be based on standard agreements within IETF, as no vendor-specific extensions are supported in the protocol. Chunk Type values are organized into four ranges to allow extensions to be made with a pre-defined procedure for responding if the a new Chunk Type is not recognized at the remote endpoint. Responses include: whole packet discard; packet discard with reporting; ignoring the chunk; ignoring with reporting. Similar pre-defined responses are specified for unrecognized Parameter Type values. |