Internet Engineering Task Force A. Ford Internet-Draft Roke Manor Research Intended status: Experimental C. Raiciu Expires: April 28, 2011 M. Handley University College London October 25, 2010 TCP Extensions for Multipath Operation with Multiple Addresses draft-ietf-mptcp-multiaddressed-02 Abstract TCP/IP communication is currently restricted to a single path per connection, yet multiple paths often exist between peers. The simultaneous use of these multiple paths for a TCP/IP session would improve resource usage within the network, and thus improve user experience through higher throughput and improved resilience to network failure. Multipath TCP provides the ability to simultaneously use multiple paths between peers. This document presents a set of extensions to traditional TCP to support multipath operation. The protocol offers the same type of service to applications as TCP - reliable bytestream - and provides the components necessary to establish and use multiple TCP flows across potentially disjoint paths. Status of this Memo This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at http://datatracker.ietf.org/drafts/current/. 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." This Internet-Draft will expire on April 28, 2011. Copyright Notice Copyright (c) 2010 IETF Trust and the persons identified as the document authors. All rights reserved. Ford, et al. Expires April 28, 2011 [Page 1] Internet-Draft Multipath TCP October 2010 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. Ford, et al. Expires April 28, 2011 [Page 2] Internet-Draft Multipath TCP October 2010 Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.1. Design Assumptions . . . . . . . . . . . . . . . . . . . . 4 1.2. Multipath TCP in the Networking Stack . . . . . . . . . . 5 1.3. Operation Summary . . . . . . . . . . . . . . . . . . . . 6 1.4. Requirements Language . . . . . . . . . . . . . . . . . . 7 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 7 3. MPTCP Protocol . . . . . . . . . . . . . . . . . . . . . . . . 8 3.1. Connection Initiation . . . . . . . . . . . . . . . . . . 8 3.2. Starting a New Subflow . . . . . . . . . . . . . . . . . . 11 3.3. General MPTCP Operation . . . . . . . . . . . . . . . . . 15 3.3.1. Data Sequence Numbering . . . . . . . . . . . . . . . 15 3.3.2. Data Acknowledgements . . . . . . . . . . . . . . . . 17 3.3.3. Receiver Considerations . . . . . . . . . . . . . . . 18 3.3.4. Sender Considerations . . . . . . . . . . . . . . . . 19 3.3.5. Congestion Control Considerations . . . . . . . . . . 21 3.3.6. Subflow Policy . . . . . . . . . . . . . . . . . . . . 21 3.4. Closing a Connection . . . . . . . . . . . . . . . . . . . 22 3.5. Address Knowledge Exchange (Path Management) . . . . . . . 24 3.5.1. Address Advertisement . . . . . . . . . . . . . . . . 25 3.5.2. Remove Address . . . . . . . . . . . . . . . . . . . . 27 3.6. Fallback . . . . . . . . . . . . . . . . . . . . . . . . . 28 3.7. Error Handling . . . . . . . . . . . . . . . . . . . . . . 31 3.8. Heuristics . . . . . . . . . . . . . . . . . . . . . . . . 32 3.8.1. Port Usage . . . . . . . . . . . . . . . . . . . . . . 32 4. Semantic Issues . . . . . . . . . . . . . . . . . . . . . . . 32 5. Security Considerations . . . . . . . . . . . . . . . . . . . 34 6. Interactions with Middleboxes . . . . . . . . . . . . . . . . 34 7. Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . 38 8. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 38 9. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 38 10. References . . . . . . . . . . . . . . . . . . . . . . . . . . 39 10.1. Normative References . . . . . . . . . . . . . . . . . . . 39 10.2. Informative References . . . . . . . . . . . . . . . . . . 39 Appendix A. Notes on use of TCP Options . . . . . . . . . . . . . 40 Appendix B. Resync Packet . . . . . . . . . . . . . . . . . . . . 42 Appendix C. Changelog . . . . . . . . . . . . . . . . . . . . . . 42 C.1. Changes since draft-ietf-mptcp-multiaddressed-01 . . . . . 42 C.2. Changes since draft-ietf-mptcp-multiaddressed-00 . . . . . 43 C.3. Changes since draft-ford-mptcp-multiaddressed-03 . . . . . 43 C.4. Changes since draft-ford-mptcp-multiaddressed-02 . . . . . 43 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 43 Ford, et al. Expires April 28, 2011 [Page 3] Internet-Draft Multipath TCP October 2010 1. Introduction Multipath TCP (henceforth referred to as MPTCP) is a set of extensions to regular TCP [2] to allow a transport connection to operate across multiple paths simultaneously. This document presents the protocol changes required to add multipath capability to TCP; specifically, those for signalling and setting up multiple paths ("subflows"), managing these subflows, reassembly of data, and termination of sessions. This is not the only information required to create a Multipath TCP implementation, however. This document is complemented by several others: o Architecture [3], which explains the motivations behind Multipath TCP, contains a discussion of high-level design decisions on which this design is based, and an explanation of a functional separation through which an extensible MPTCP implementation can be developed. o Congestion Control [4], presenting a safe congestion control algorithm for coupling the behaviour of the multiple paths in order to "do no harm" to other network users. o Application Considerations [5], discussing what impact MPTCP will have on applications, what applications will want to do with MPTCP, and as a consequence of these factors, what API extensions an MPTCP implementation should present. 1.1. Design Assumptions In order to limit the potentially huge design space, the authors imposed two key constraints on the multipath TCP design presented in this document: o It must be backwards-compatible with current, regular TCP, to increase its chances of deployment o It can be assumed that one or both endpoints are multihomed and multiaddressed To simplify the design we assume that the presence of multiple addresses at an endpoint is sufficient to indicate the existence of multiple paths. These paths need not be entirely disjoint: they may share one or many routers between them. Even in such a situation making use of multiple paths is beneficial, improving resource utilisation and resilience to a subset of node failures. The congestion control algorithms as discussed in [4] ensure this does not act detrimentally. Ford, et al. Expires April 28, 2011 [Page 4] Internet-Draft Multipath TCP October 2010 There are three aspects to the backwards-compatibility listed above (discussed in more detail in [3]): External Constraints: The protocol must function through the vast majority of existing middleboxes such as NATs, firewalls and proxies, and as such must resemble existing TCP as far as possible on the wire. Furthermore, the protocol must not assume the segments it sends on the wire arrive unmodified at the destination: they may be split or coalesced; options may be removed or duplicated. Application Constraints: The protocol must be usable with no change to existing applications that use the standard TCP API (although it is reasonable that not all features would be available to such legacy applications). Furthermore, the protocol must provide the same service model as regular TCP to the application. Fall-back: The protocol should be able to fall back to standard TCP with no interference from the user, to be able to communicate with legacy hosts. Areas for further study: o In theory, since this is purely a TCP extension, it should be possible to use MPTCP with both IPv4 and IPv6 subflows for the same connection on dual-stack hosts, thus having the additional possible benefit of aiding transition. o The design presented should work with network provided multipath, for instance ECMP routing; subflows could be opened with different source/destination ports between the same addreses to allow ECMP to place the subflows on different paths. 1.2. Multipath TCP in the Networking Stack MPTCP operates at the transport layer and aims to be transparent to both higher and lower layers. It is a set of additional features on top of standard TCP; Figure 1 illustrates this layering. MPTCP is designed to be usable by legacy applications with no changes; detailed discussion of its interactions with applications is given in [5]. Ford, et al. Expires April 28, 2011 [Page 5] Internet-Draft Multipath TCP October 2010 +-------------------------------+ | Application | +---------------+ +-------------------------------+ | Application | | MPTCP | +---------------+ + - - - - - - - + - - - - - - - + | TCP | | Subflow (TCP) | Subflow (TCP) | +---------------+ +-------------------------------+ | IP | | IP | IP | +---------------+ +-------------------------------+ Figure 1: Comparison of Standard TCP and MPTCP Protocol Stacks Detailed discussion of an architecture for developing a multipath TCP implementation, especially regarding the functional separation by which different components should be developed, is given in [3]. 1.3. Operation Summary This section provides a high-level summary of normal operation of MPTCP, and is illustrated by the scenario shown in Figure 2. A detailed description of operation is given in Section 3. o To a non-MPTCP-aware application, MPTCP will behave the same as normal TCP. Extended APIs could provide additional control to MPTCP-aware applications [5]. An application begins by opening a TCP socket in the normal way. MPTCP signaling and operation is handled by the MPTCP implementation. o An MPTCP connection begins similarly to a regular TCP connection. This is illustrated in Figure 2 where a TCP connection is established between addresses A1 and B1 on Hosts A and B respectively. o If extra paths are available, additional TCP sessions (termed "subflows") are created on these paths, and are combined with the existing session, which continues to appear as a single connection to the applications at both ends. The creation of the additional TCP session is illustrated between Address A2 on Host A and Address B1 on Host B. o MPTCP identifies multiple paths by the presence of multiple addresses at endpoints. Combinations of these multiple addresses equate to the additional paths. In the example, other potential paths that could be set up are A1<->B2 and A2<->B2. Although this additional session is shown as being initiated from A2, it could equally have been initiated from B1. Ford, et al. Expires April 28, 2011 [Page 6] Internet-Draft Multipath TCP October 2010 o The discovery and setup of additional subflows will be achieved through a path management method; this document describes a mechanism by which an endpoint can initiate new subflows by using its own additional addresses, or by signalling its available addresses to the other endpoint. o MPTCP adds connection-level sequence numbers to allow the reassembly of the in-order data stream from multiple subflows which may deliver packets out-of-order due to differing network delays. o Subflows are terminated as regular TCP connections, with a four way FIN handshake. The MPTCP connection is terminated by a connection-level FIN packet, sent together with the FIN on the last subflow of the connection. Host A Host B ------------------------ ------------------------ Address A1 Address A2 Address B1 Address B2 ---------- ---------- ---------- ---------- | | | | | (initial connection setup) | | |----------------------------------->| | |<-----------------------------------| | | | | | | (additional subflow setup) | | |--------------------->| | | |<---------------------| | | | | | | | | | Figure 2: Example MPTCP Usage Scenario 1.4. Requirements Language The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [1]. 2. Terminology Path: A sequence of links between a sender and a receiver, defined in this context by a source and destination address pair. Ford, et al. Expires April 28, 2011 [Page 7] Internet-Draft Multipath TCP October 2010 Subflow: A stream of TCP packets sent over a path, started and terminated similarly to a regular TCP connection. (MPTCP) Connection: A collection of one or more subflows, over which an application can communicate between two endpoints. There is a one-to-one mapping between a connection and an application socket. Data-level: The payload data is nominally transfered over a connection, which in turn is transported over subflows. Thus the term "data-level" is synonymous with "connection level", in contrast to "subflow-level" which refers to properties of an individual subflow. Token: A locally unique identifier given to a multipath connection by an endpoint. May also be referred to as a "Connection ID". Endpoint: A host operating an MPTCP implementation, and either initiating or accepting an MPTCP connection. 3. MPTCP Protocol This section describes the operation of the MPTCP protocol, and is subdivided into sections for each key part of the protocol operation. All MPTCP operations are signalled using optional TCP header fields. These TCP Options will have option numbers allocated by IANA, as listed in Section 9, and are defined throughout the following subsections. 3.1. Connection Initiation Connection Initiation begins with a SYN, SYN/ACK, ACK exchange on a single path. Each packet contains the Multipath Capable (MP_CAPABLE) TCP option (Figure 3). This option declares its sender is capable of performing multipath TCP and wishes to do so on this particular connection. This option contains a 64-bit key that is used to authenticate the addition of future subflows. This is the only time the key will be sent in clear on the wire; all future subflows will identify the connection using a 32-bit "token". This token is a cryptographically secure hash of this key. This will be a truncated (most significant 32 bits) SHA-1 hash [6]. A different, 64-bit truncation (the least significant 64 bits) of the hash of the key will be used as the Initial Data Sequence Number. This key is generated by the sender and has local meaning only, and Ford, et al. Expires April 28, 2011 [Page 8] Internet-Draft Multipath TCP October 2010 its method of generation is implementation-specific. The key SHOULD be hard to guess, and it MUST be unique for the sending host at any one time. Connections will be indexed at each host by the token (the truncated SHA-1 hash of the key), but an implementation will require a mapping from the token to the key for each connection. The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK packets that start the first subflow of an MPTCP connection. The data carried by each packet is as follows, where A = initiator and B = listener. o SYN (A->B): A's Key. o SYN/ACK (B->A): B's Key. o ACK (A->B): Both A's Key and B's Key. The contents of the option is determined by the SYN and ACK flags of the packet, verified by the option's length field. For the diagram shown in Figure 3, "sender" and "receiver" refer to the sender or receiver of the TCP packet. The keys are echoed in the ACK in order to allow the listener to act statelessly until the TCP connection reaches the ESTABLISHED state. If the listener acts in this way, however, it MUST generate its key in a verifiable fashion, allowing it to verify that it generated the key when it is echoed in the ACK. If this ACK does not carry data, it MUST still be ACKed by the receiver in order for the sender to ensure the ACK with MP_JOIN option has been received. The first octet of this option specifies the MPTCP version in use (for this specification, this is 0). The second octet is reserved for flags, and currently MUST be set to all zeros. The meaning of such flags will be determined in future revisions of MPTCP, however some possible uses may be to enable or disable certain MPTCP features, and to provide a mechanism for crypto agility. The MP_CAPABLE option is only used in the first subflow of a connection, in order to identify the connection; all following subflows will use the "Join" option (see Section 3.2) to join the existing connection. Ford, et al. Expires April 28, 2011 [Page 9] Internet-Draft Multipath TCP October 2010 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +---------------+---------------+---------------+---------------+ |Kind=MP_CAPABLE| Length | Version | (reserved) | +---------------+---------------+---------------+---------------+ | Sender Key | | (64 bits) | | | +---------------------------------------------------------------+ | Receiver Key (64 bits) | | (if Length==20) | | | +---------------------------------------------------------------+ Figure 3: Multipath Capable (MP_CAPABLE) option If a SYN contains an MP_CAPABLE option but the SYN/ACK does not, it is assumed that the passive opener is not multipath capable and thus the MPTCP session will operate as regular, single-path TCP. If a SYN does not contain a MP_CAPABLE option, the SYN/ACK MUST NOT contain one in response. If the SYN packets are unacknowledged, it is up to local policy to decide how to respond. It is expected that a sender will eventually fall back to single-path TCP (i.e. without the MP_CAPABLE Option) in order to work around middleboxes that may drop packets with unknown options; however, the number of multipath-capable attempts that are made first will be up to local policy. Once the active opener has sent a SYN without the MP_CAPABLE option, it MUST fall back to regular TCP behavior, even if it subsequently receives a SYN/ACK that contains an MP_CAPABLE option. This might happen if the MP_CAPABLE SYN and subsequent non-MP-capable SYN are reordered. This is to ensure that the two endpoints end up in an interoperable state, no matter what order the SYNs arrive at the passive opener. This final state is inferred from the presence or absence of the MP_CAPABLE option in the third packet of the TCP handshake. If this option is not present, the connection should fall back to regular TCP, as documented in Section 3.6. The initial Data Sequence Number (IDSN) is generated as a hash from the Key, in the same way as the token, i.e. IDSN-A = Hash(Key-A) and IDSN-B = Hash(Key-B). The Hash mechanism here provides the least significant 64 bits of the SHA-1 hash of the key. The SYN with MP_CAPABLE occupies the first octet of Data Sequence Space. Ford, et al. Expires April 28, 2011 [Page 10] Internet-Draft Multipath TCP October 2010 3.2. Starting a New Subflow Once a MPTCP connection has begun with the MP_CAPABLE exchange, further subflows can be added to the connection. Endpoints have knowledge of their own address(es), and can become aware of the other endpoint's addresses through signalling exchanges as described in Section 3.5. Using this knowledge, an endpoint can initiate a new subflow over a currently unused pair of addresses. The protocol permits either endpoint of a connection to initiate the creation of a new subflow (but see Section 3.8 for heuristics). A new subflow is started as a normal TCP SYN/ACK exchange. The Join Connection (MP_JOIN) TCP option (Figure 4) is used to identify the connection to be joined by the new subflow. The tokens used to identify the MPTCP connection are cryptographically secure hashes of the keys exchanged in the initial MP_CAPABLE handshake. The tokens presented in this option are generated by the SHA-1 [6] algorithm, truncated to the most significant 32 bits. The token included in the MP_JOIN option is the token that the receiver of the packet uses to identify this connection, i.e. Host A will send Token-B (which is generated from Key-B), and vice versa. The MP_JOIN SYN/SYN-ACK handshake not only exchanges the tokens (which are static for a connection) but also Random Numbers (nonces) that are used to prevent replay attacks on the authentication method. Whilst these data are transferred in the SYN exchange, the actual cryptographic authentication is undertaken in the first two payload segments of the connection. Once the peers have successfully authenticated themselves, the subflow is handed over to the scheduler to be used for data (the presense of a DSN_MAP option Section 3.3 indicates this). The MP_JOIN option also contains an "Address ID" to identify the source address of this packet if it has changed in transit; the behaviour of this ID is explained later in this section. 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +---------------+---------------+-------------------------------+ | Kind=MP_JOIN | Length = 8 | Address ID | (reserved) |B| +---------------+---------------+----------------+--------------+ | Receiver Token (32 bits) | +---------------------------------------------------------------+ | Sender Random Number (32 bits) | +---------------------------------------------------------------+ Figure 4: Join Connection (MP_JOIN) option (only valid on SYN packets) Ford, et al. Expires April 28, 2011 [Page 11] Internet-Draft Multipath TCP October 2010 On the third and fourth packets of the handshake, the following data is sent in the TCP payload: 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +---------------+---------------+-------------------------------+ | Kind=MP_AUTH | Length | (reserved) | +---------------+---------------+-------------------------------+ | | | | | HMAC (256 bits for SHA-256) | | | | | +---------------------------------------------------------------+ Figure 5: Authentication Data For consistancy, this follows the same format as a TCP Option, although it is sent in the TCP payload. The HMAC algorithm is as defined in [6], using the SHA-256 hash algorithm (thus generating a 256-bit / 32 octet HMAC), however in the future some of the reserved bits could be used to enable alternative algorithms. The key for the HMAC algorithm, in the case of the message transmitted by Host A, will be Key-A followed by Key-B, and in the case of Host B, Key-B followed by Key-A. The message in each case is the concatenations of Random Number for each host (denoted by R): for Host A, R-A followed by R-B; and for Host B, R-B followed by R-A. When receiving a SYN with a MP_JOIN option that contains a valid token for an existing MPTCP connection, the recipient SHOULD respond with a SYN/ACK also containing an MP_JOIN option containing the initiator's token. This will then lead on to the authentication HMAC exchange described above. This behaviour is illustrated in Figure 6. Ford, et al. Expires April 28, 2011 [Page 12] Internet-Draft Multipath TCP October 2010 Host A Host B ------------------------ ------------------------ Address A1 Address A2 Address B1 Address B2 ---------- ---------- ---------- ---------- | | | | | SYN + MP_CAPABLE(Key-A) | | |----------------------------------------->| | |<-----------------------------------------| | | SYN/ACK + MP_CAPABLE(Key-B) | | | | | | | ACK + MP_CAPABLE(Key-A, Key-B) | | |----------------------------------------->| | | | | | | | SYN + MP_JOIN(Token-B, R-A) | | |----------------------------------------->| | |<-----------------------------------------| | | SYN/ACK + MP_JOIN(Token-A, R-B) | | | | | | | HMAC(Key=(Key-A+Key-B), Msg=(R-A+R-B)) | | |----------------------------------------->| | |<-----------------------------------------| | | HMAC(Key=(Key-B+Key-A), Msg=(R-B+R-A)) | | | | | Figure 6: Example use of MPTCP Authentication If the token received at Host B is unknown or local policy prohibits the acceptance of the new subflow, the recipient MUST respond with a TCP RST. If the token is accepted at Host B, but the token returned to Host A is not the one expected, Host A MUST close the subflow with a TCP RST. If either host receives an incorrect HMAC (i.e. it does not match what the host believes it should be), it MUST close the subflow with a TCP RST. The echoing of the token serves two purposes: it ensures both endpoints agree on the connection being referred to (this is particularly relevant when both addresses being used are new to the connection); and it ensures there are no middleboxes on this new path that will drop MPTCP options on the return path. If the SYN/ACK as received at Host A does not have an MP_JOIN option, Host A MUST close the subflow with a RST. If MP_JOIN is stripped from the SYN on the path from A to B, and Host Ford, et al. Expires April 28, 2011 [Page 13] Internet-Draft Multipath TCP October 2010 B does not have a passive opener on the relevant port, it will respond with an RST in the normal way. If in response to a SYN with an MP_JOIN option, a SYN/ACK is received without the MP_JOIN option (either since it was stripped on the return path, or it was stripped on the outgoing path but the passive opener on Host B responded as if it were a new regular TCP session), then the subflow is unusable and Host A MUST close it with a RST. It should be noted that additional subflows can be created between any pair of ports (but see Section 3.8 for heuristics); no explicit application-level accept calls or bind calls are required to open additional subflows. To associate a new subflow with an existing connection, the token supplied in the subflow's SYN exchange is used for demultiplexing. This then binds the 5-tuple of the TCP subflow to the local token of the connection. A consequence is that it is possible to allow any port pairs to be used for a connection. Deumultiplexing subflow SYNs MUST be done using the token; this is unlike traditional TCP, where the destination port is used for demultiplexing SYN packets. Once a subflow is setup, demultiplexing packets is done using the five-tuple, as in traditional TCP. The five-tuples will be mapped to the local connection ID. The MP_JOIN option includes an "Address ID". This is an identifier that only has significance within a single connection, where it identifies the source address of this packet. The key purpose of this identifier is to allow address removal without needing to know what the source address at the receiver is, thus allowing the use of NATs. The sender can signal this to the receiver via the REMOVE_ADDR option (Section 3.5.2). It also allows correlation between new subflow setup attempts and address signalling (Section 3.5.1), to prevent setting up duplicate subflows on the same path. The Address IDs of the subflow used in the initial SYN exchange of the first subflow in the connection are implicit, and have the value zero. The Address ID must be stored by the receiver in a data structure that gathers all the Address ID to address mappings for a connection identified by a token pair. In this way there is a stored mapping between Address ID, observed source address and token pair for future processing of control information for a connection. The MP_JOIN option also includes 8 bits of flags, 7 of which are currently reserved. The final bit, labelled 'B', indicates whether the initiator wishes this subflow to be used purely as a backup path (B=1) in the event of failure of other paths, or whether it wants it to be used as part of the connection immediately. Subflow policy is Ford, et al. Expires April 28, 2011 [Page 14] Internet-Draft Multipath TCP October 2010 discussed in more detail in Section 3.3.6. 3.3. General MPTCP Operation This section discusses operation of MPTCP for data transfer. At a high level, an MPTCP implementation will take one input data stream from an application, and split it into one or more subflows, with sufficient control information to allow it to be reassembled and delivered reliably and in-order to the recipient application. The following subsections define this behaviour in detail. 3.3.1. Data Sequence Numbering The data stream as a whole can be reassembled through the use of the Data Sequence Mapping (DSN_MAP, Figure 7) option, which defines the mapping from the data sequence number to the subflow sequence number. This is used by the receiver to ensure in-order delivery to the application layer. Meanwhile, the subflow-level sequence numbers (i.e. the regular sequence numbers in the TCP header) have subflow- only relevance. It is expected (but not mandated) that SACK [7] is used at the subflow level to improve efficiency. 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +---------------+---------------+------------------------------+ | Kind=DSN_MAP | Length | Data Sequence Number ... : +---------------+---------------+------------------------------+ : ... ( (length-10) octets ) | Data-level Length (2 octets) | +-------------------------------+------------------------------+ | Subflow Sequence Number (4 octets) | +-------------------------------+------------------------------+ | Checksum (2 octets) | +-------------------------------+ Figure 7: Data Sequence Mapping (DSN_MAP) option This option specifies a full mapping from data sequence number to subflow sequence number, informing the receiver that there is a one- to-one correspondence between the two sequence spaces for the specified length (number of bytes of data). The purpose of the explicit mapping is to assist with compatibility with situations where TCP/IP segmentation or coalescing is undertaken separately from the stack that is generating the data flow (e.g. through the use of TCP segmentation offloading on network interface cards, or by middleboxes such as performance enhancing proxies). It also allows a single mapping to cover many packets, which may be useful in bulk transfer situations. Ford, et al. Expires April 28, 2011 [Page 15] Internet-Draft Multipath TCP October 2010 The data sequence number specified in this option is absolute, whereas the subflow sequence numbering is relative (the SYN at the start of the subflow has relative subflow sequence number 1). This is allow middleboxes to change the Initial Sequence Number of a subflow, since the data stream itself will not be affected (some firewalls do ISN randomization). The final two octets of this option contain a checksum of the data that this mapping covers. This is used to detect if the payload has been adjusted in any way by a non-MPTCP-aware middlebox. If this checksum fails, it will trigger a failure of the subflow, or a fallback to regular TCP, as documented in Section 3.6. The checksum algorithm used is the standard TCP checksum [2], operating only over the data covered by this DSN_MAP (i.e. there is no pseudo-header). This algorithm has been chosen since it will be calculated anyway for the TCP subflow, and if calculated first over the data before adding the pseudo-header, it only needs to be calculated once. Furthermore, since the TCP checksum is additive, the checksum for a DSN_MAP can be constructed by simply adding together the checksums for the data of each constituent TCP segment. This relies on the TCP subflow containing contiguous data, however, and thus a TCP subflow MUST NOT use the Urgent Pointer (i.e. the URG flag MUST be zero). A mapping is unique, in that the subflow sequence number is bound to the data sequence number after the mapping has been processed. It is not possible to change this mapping afterwards (although the length of a mapping can extend); however, the same data sequence number can be mapped on different subflows for retransmission purposes (see Section 3.3.4). To avoid possible deadlock scenarios, subflow-level processing should be undertaken separately from that at connection-level. Therefore, even if a mapping does not exist from the subflow space to the data- level space, the data should still be ACKed at the subflow. This data cannot, however, be acknowledged at the data level (Section 3.3.2) because its data sequence numbers are unknown. Implementations MAY hold onto such unmapped data for a short while in the expectation than a mapping will arrive shortly. Such unmapped data cannot be counted as being within the receive window because this is relative to the data sequence numbers, so if the receiver runs out of memory to hold this data, it will have to be discarded. If a mapping for that subflow-level sequence space does not arrive within a receive window of data, that subflow should be treated as broken, closed with an RST, and an unmapped data silently discarded. Data sequence numbers are always 64-bit quantities, and MUST be maintained as such in implementations. If a connection is progressing at a slow rate, so protection against wrapped sequence Ford, et al. Expires April 28, 2011 [Page 16] Internet-Draft Multipath TCP October 2010 numbers is not required, and if security requirements against blind insertion attacks are not stringent, then it is permissible to include just the lower 32 bits of the sequence number in the DSN_MAP option as an optimization. Implementations MUST accept this and implicitly promote it to a 64-bit quantity by incrementing the upper 32 bits of sequence number each time the lower 32 bits wrap. By defauly, the full 64 bit DSN_MAP should be sent. Security implications are discussed in Section 5. As with the standard TCP sequence number, the data sequence number should not start at zero, but at a random value to make blind session hijacking harder. This is done by setting the initial data sequence number (IDSN) of each host to the least significant 64 bits of the SHA-1 hash of the host's key (as declared in the MP_CAPABLE option in the initial connection SYN, which itself occupies the first octet of data sequence space). This handshake is described in more detail in Section 3.1. The DSN_MAP option does not need to be included in every MPTCP packet, as long as the subflow sequence space in that packet is covered by a mapping known at the receiver. This can be used to reduce overhead in cases where the mapping is known in advance; one such case is when there is a single subflow between the endpoints, another is when segments of data are scheduled in larger than packet- sized chunks. An "infinite" mapping can be used to fallback to regular TCP by mapping the subflow-level data to the connection-level data for the remainder of the connection (see Section 3.6). This is achieved by setting the data-level length field to the reserved value of 0. 3.3.2. Data Acknowledgements To provide full end-to-end resilience, MPTCP provides a connection- level acknowledgement, the DATA_ACK, illustrated in Figure 8, to act as a cumulative ACK for the connection as a whole. This is analogous to the behaviour of the standard TCP cumulative ACK in TCP SACK - indicating how much data has been successfully received (with no holes). The rationale for the inclusion of the DATA_ACK includes the existence of certain middleboxes that pro-actively ACK packets, and thus might cause deadlock conditions if data were acked at the subflow level but then fails to reach the receiver. This sort of bad interaction might be expecially prevalent when the receiver is mobile. The DATA_ACK ensures the data has been delieverd to the receiver. An MPTCP sender MUST only free data from the send buffer when it has Ford, et al. Expires April 28, 2011 [Page 17] Internet-Draft Multipath TCP October 2010 been acknowledged by both a DATA_ACK received on any subflow and at the subflow level by any subflows the data was sent on. The former condition ensures liveness of the connection and the latter condition ensures liveness and self-consistence of a subflow when data needs to be restransmited. The DATA_ACK option MAY be included in all segments, analogous to a standard TCP ACK. However, optimisations SHOULD be considered in more advanced implementations, where the DATA_ACK option is present in segments (data or pure ACKs) only when the DATA_ACK advances, and this behaviour MUST be treated as valid. This behaviour ensures the sender buffer is freed, while reducing overhead when the data transfer is unidirectional. 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +---------------+---------------+------------------------------+ | Kind=DATA_ACK | Length | Data Sequence Number ... : +---------------+---------------+------------------------------+ : ... ( (length-2) octets ) | +-------------------------------+ Figure 8: Connection-level Acknowledgement (DATA_ACK) 3.3.3. Receiver Considerations Regular TCP advertises a receive window in each packet, telling the sender how much data the receiver is willing to accept past the cumulative ack. The receive window is used to implement flow control, throttling down fast senders when receivers cannot keep up. MPTCP also uses a unique receive window, shared between the subflows. The idea is to allow any subflow to send data as long as the receiver is willing to accept it; the alternative, maintaining per subflow receive windows, could end-up stalling some subflows while others would not use up their window. The receive window is relative to the DATA_ACK. As in TCP, a receiver MUST NOT shrink the right edge of the receive window (e.g. DATA_ACK + receive window). The receiver will use the Data Sequence Number to tell if a packet should be accepted at connection level. When deciding to accept packets at subflow level, normal TCP uses the sequence number in the packet and checks it against the allowed receive window. With multipath, such a check is done using only the connection level window. A sanity check could be performed at subflow level to ensure that: SSN - SUBFLOW_ACK <= DSN - DATA_ACK. Ford, et al. Expires April 28, 2011 [Page 18] Internet-Draft Multipath TCP October 2010 When should segments be processed at connection level? An implementation might wait until they arrive in order at subflow level, and only then do connection level processing. However, if many segments of data are restransmitted on more than one subflow, then because some data is duplicated then the sum total of unacknowledged data on all subflows might exceed the receive window that was advertised, which indicates buffering available for data sequence space. This such a strategy is probably undesirable. An alternative implementation might process segments at the connection level segments that have not yet been acked at subflow level; the only requirement for this is to have a valid data sequence mapping for the segment. This removes such duplicate data from the receive buffer, so avoids running out of buffer space. Such implementations SHOULD keep track of which subflow sequence numbers have already been accepted in this way, so they can be ACKed appropriately when the hole in the subflow sequence space in subsequently filled. An implementation that does store such metadata would still progress (the rules for freeing data at the sender ensure this), but unnecessary retransmissions will result. It is important for implementers to understand how large a receiver buffer is appropriate. The lower bound for full network utilization is the maximum bandwidth-delay product of any of the paths. However this might be insufficient when a packet is lost on a slower subflow and needs to be retransmitted (see Section 3.3.4). A tight upper bound would be the maximum RTT of any path multiplied by the total bandwidth available across all paths. This permits all subflows to continue at full speed while a packet is fast-retransmitted on the maximum RTT path. Even this might be insufficient to maintain full performance in the event of a retransmit timeout on the maximum RTT path. It is for future study to determine the relationship between retransmission strategies and receive buffer sizing. 3.3.4. Sender Considerations The sender remembers receiver window advertisements from the receiver. It should only update its local receive window values when the largest sequence number allowed (i.e. DATA_ACK + receive window) increases. This is important to allow using paths with different RTTs, and thus different feedback loops. Some classes of middleboxes may alter the TCP-level receive window. Typically these will shrink the offered window, although for short periods of time it may be possible for the window to be larger (however note that this would not continue for long periods since ultimately the middlebox must keep up with delivering data to the receiver). Therefore, if receive window sizes differ on multiple Ford, et al. Expires April 28, 2011 [Page 19] Internet-Draft Multipath TCP October 2010 subflows, when sending data MPTCP SHOULD take the largest of the most recent window sizes as the one to use in calculations. (this rule is implicit in the requirement not to move back the right edge of the window). The sender also remembers the receive windows advertised by each subflow. The allowed window for subflow i is (ack_i, ack_i + rcv_wnd_i), where ack_i is the subflow-level cumulative ack of subflow i. This ensures data will not be sent to a middlebox unless there is enough buffering for the data. Putting the two rules together, we get the following: a sender is allowed to send data segments with data-level sequence numbers between (DATA_ACK, DATA_ACK + receive_window). Each of these segments will be mapped onto subflows, as long as subflow sequence numbers are in the the allowed windows for those subflows. Note that subflow sequence numbers do not generally affect flow control if the same receive window is advertised across all subflows. They will perform flow control for those subflows with a smaller advertised receive window. The data sequence mapping allows senders to re-send data with the same data sequence number on a different subflow. When doing this, an endpoint must still retransmit the original data on the original subflow, in order to preserve the subflow integrity (middleboxes could replay old data, and/or could reject holes in subflows), and a receiver will ignore these retransmissions. While this is clearly suboptimal, for compatibility reasons this is the best behaviour. Optimisations could be negotiated in future versions of this protocol. This protocol specification does not mandate any mechanisms for handling retransmissions, and much will be dependent upon local policy (as discussed in Section 3.3.6). One can imagine aggressive connection level retransmissions policies where every packet lost at subflow level is retransmitted on a different subflow (hence wasting bandwidth but possibly reducing application-to-application delays), or conservative retransmission policies where connection-level retransmits are only used after a few subflow level retransmission timeouts occur. It is envisaged that a standard connection-level retransmission mechanism would be implemented around a connection-level data queue: all segments that haven't been DATA_ACKed are stored. A timer (based on the subflow timer values) is set when the head of the connection- level is ACKed at subflow level but its corresponding data is not acked at data level. Ford, et al. Expires April 28, 2011 [Page 20] Internet-Draft Multipath TCP October 2010 The sender MUST keep data in its send buffer as long as the data has not been acked at connection level and on all subflows it has been sent on. In this way, the sender can always retransmit the data if needed, on the same subflow or on a different one. A special case is when a subflow fails: the sender will typically resend the data on other working subflows, and will keep trying to retransmit the data on the failed subflow too. The sender will declare the subflow failed after a predefined upper bound on retransmissions is reached, and only then delete the outstanding data segments. A sender will maintain connection level timers for unacknowledged segments. These timers will be based on the subflow timers, and will guard against pro-active acking by middleboxes. The send buffer must be, at the minimum, as big as the receive buffer, to enable the sender to reach maximum throughput. 3.3.5. Congestion Control Considerations Different subflows in an MPTCP connection have different congestion windows. To achieve fairness at bottlenecks and resource pooling, it is necessary to couple the congestion windows in use on each subflow, in order to push most traffic to uncongested links. One algorithm for achieving this is presented in [4]; the algorithm does not achieve perfect resource pooling but is "safe" in that it is readily deployable in the current Internet. It is foreseeable that different congestion controllers will be implemented for MPTCP, each aiming to achieve different properties in the resource pooling/fairness/stability design space. Much research is expected in this area in the near future. Regardless of the algorithm used, the design of the MPTCP protocol aims to provide the congestion control implementations sufficient information to take the right decisions; this information includes, for each subflow, which packets where lost and when. 3.3.6. Subflow Policy Within a local MPTCP implementation, a host may use any local policy it wishes to decide how to share the traffic to be sent over the available paths. In the typical use case, where the goal is to maximise throughput, all available paths will be used simultaneously for data transfer, using coupled congestion control as described in [4]. It is expected, however, that other use cases will appear. Ford, et al. Expires April 28, 2011 [Page 21] Internet-Draft Multipath TCP October 2010 For instance, a possibility is an 'all-or-nothing' approach, i.e. have a second path ready for use in the event of failure of the first path, but alternatives could include entirely saturating one path before using an additional path (the 'overflow' case). Such choices would be most likely based on the monetary cost of links, but may also be based on properties such as the delay or jitter of links, where stability is more important than throughput. Application requirements such as these are discussed in detail in [5]. The ability to make effective choices at the sender requires full knowledge of the path "cost", which is unlikely to be the case. It would be desirable for a receiver to be able to signal their own preferences for paths, since they will often be the multihomed party, and may have to pay for metered incoming bandwidth. Whilst fine-grained control may be the most powerful solution, that would require some mechanism such as overloading the ECN signal [8], which is undesirable, and it is felt that there would not be sufficient benefit to justify an entirely new signal. Therefore the MP_JOIN Section 3.2 and ADD_ADDR Section 3.5 options contain the 'B' bit, which allows a host to indicate to its peer that this path should be treated as a backup path to use only in the event of failure of other working subflows (i.e. a subflow where the receiver has indicated B=1 SHOULD NOT be used to send data unless there are no usable subflows where B=0). In the event that the available set of paths changes, a host may wish to signal a change in priority of subflows to the peer. Therefore, the MP_PRIO option, shown in Figure 9, can be used to change the 'B' flag of the subflow on which it is sent. 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +---------------+---------------+-------------+-+ | Kind=MP_PRIO | Length=3 | (reserved) |B| +---------------+---------------+-------------+-+ Figure 9: MP_PRIO option It should be noted that the backup flag is a request from the receiver to the sender only, and the sender SHOULD adhere to these requests. The reciever, however, may continue using the subflow to send data even if it has signalled B=1 to the other host. 3.4. Closing a Connection In regular TCP a FIN announces the receiver that the sender has no more data to send. In order to allow subflows to operate Ford, et al. Expires April 28, 2011 [Page 22] Internet-Draft Multipath TCP October 2010 independently and to keep the appearance of TCP over the wire, a FIN in MPTCP only affects the subflow on which it is sent. This allows nodes to exercise considerable freedom over which paths are in use at any one time. The semantics of a FIN remain as for regular TCP, i.e. it is not until both sides have ACKed each other's FINs that the subflow is fully closed. When an application calls close() on a socket, this indicates that it has no more data to send, and for regular TCP this would result in a FIN on the connection. For MPTCP, an equivalent mechanism is needed, and this is the DATA_FIN. This option, shown in Figure 10, is attached to a regular FIN option on a subflow. A DATA_FIN is an indication that the sender has no more data to send, and as such can be used as a rapid indication of the end of data from a sender. A DATA_FIN, as with the FIN on a regular TCP connection, is a unidirectional signal. A DATA_FIN occupies one octet (the final octet) of Data Sequence Number space. This number is included in the option, and will be ACKed at data level to ensure reliable delivery. The DATA_FIN is an optimisation to rapidly indicate the end of a data stream and clean up state associated with a MPTCP connection, especially when some subflows may have failed. Specifically, when a DATA_FIN has been received, IF all data has been successfully received, timeouts on all subflows MAY be reduced. Similarly, when sending a DATA_FIN, once all data (including the DATA_FIN, since it occupies one octet of data sequence space) has been acknowledged, FINs must be sent on every subflow. This applies to both endpoints, and is required in order to clean up state in middleboxes. The interactions between a DATA_FIN and subflow properties are as follows: o A DATA_FIN MUST only be sent on a packet which also has the FIN flag set. o When DATA_FIN is sent, it should be sent on all active subflows. o There is a one-to-one mapping between the DATA_FIN and the subflow's FIN flag (and its associated sequence space and thus its acknowlegement). o The data sequence number included in the DATA_FIN is used to verify that all data has been successfully received. It should be noted that an endpoint may also send a FIN on an Ford, et al. Expires April 28, 2011 [Page 23] Internet-Draft Multipath TCP October 2010 individual subflow to shut it down, but this impact is limited to the subflow in question. If all subflows have been closed with a FIN, that is equivalent to having closed the connection with a DATA_FIN. The full eight-byte data sequence number is always included in a DATA_FIN. 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +---------------+---------------+------------------------------+ | Kind=DATA_FIN | Length=10 | Data Sequence Number (8B) : +---------------+---------------+------------------------------+ : Data Sequence Number (contd.) : +-------------------------------+------------------------------+ : Data Sequence Number (contd.)| +-------------------------------+ Figure 10: DATA_FIN option 3.5. Address Knowledge Exchange (Path Management) We use the term "path management" to refer to the exchange of information about additional paths between endpoints, which in this design is managed by multiple addresses at endpoints. For more detail of the architectural thinking behind this design, see the separate architecture document [3]. This design makes use of two methods of sharing such information, used simultaneously. The first is the direct setup of new subflows, already described in Section 3.2, where the initiator has an additional address. The second method, described in the following subsections, signals addresses explicitly to the other endpoint to allow it to initiate new subflows. The two mechanisms are complementary: the first is implicit and simple, while the explicit is more complex but is more robust. Together, the mechanisms allow addresses to change in flight (and thus support operation through NATs, since the source address need not be known), and also allow the signalling of previously unknown addresses, and of addresses belonging to other address families (e.g. IPv4 and IPv6). Here is an example of typical operation of the protocol: o A1 of host A and address/port B1 of host B. If host A is multihomed and multi-addressed, it can start an additional subflow from its address A2 to B1, by sending a SYN with a Join option from A2 to B1, using B's previously declared token for this connection. Alternatively, if B is multhomed, it can try to set Ford, et al. Expires April 28, 2011 [Page 24] Internet-Draft Multipath TCP October 2010 up a new subflow from B2 to A1, using A's previously declared token. In either case, the SYN will be sent to the port already in use for the original subflow on the receiving host. o Simultaneously (or after a timeout), an ADD_ADDR option (Section 3.5.1) is sent on an existing subflow, informing the receiver of the sender's alternative address(es). The recipient can use this information to open a new subflow to the sender's additional address. In our example, A will send ADD_ADDR option informing B of address A2. The mix of using the SYN-based option and the ADD_ADDR option, including timeouts, is implementation- specific and can be tailored to agree with local policy. o If subflow A2-B1 is succesfully setup, host B1 can use the Address ID in the Join option to correlate this with the ADD_ADDR option that will also arrive on an existing subflow; now B knows not to open A2-B1, ignoring the ADD_ADDR. Otherwise, if B has not received the A2-B1 SYN join but received the ADD_ADDR, it will try to initiate a new subflow from one or more of its addresses to address A2. This permits new sessions to be opened if one endpoint is behind a NAT. A slight security improvement can be gained if a host ensures there is a correlated ADD_ADDR option before responding to the SYN. Other ways of using the two signaling mechanisms are possible; for instance, signaling addresses in other address families can only be done explicitly using the Add Address option. 3.5.1. Address Advertisement The Add Address (ADD_ADDR) TCP Option announces additional addresses on which an endpoint can be reached (Figure 11). It can be used to announce several (ID, address) pairs to be announced to the other endpoint. Multiple addresses can be added in a single message if there is sufficient TCP option space, otherwise multiple TCP messages containing this option will be sent. This option can be used at any time during a connection, depending on when the sender wishes to enable multiple paths and/or when paths become available. Every address has an ID which can be used for address removal, and therefore endpoints must cache the mapping between ID and address. This is also used to identify Join Connection options (Section 3.2) relating to the same address, even when address translators are in use. The ID must uniquely identify the address to the sender (within the connection), but its mechanism for allocating such IDs is implementation-specific. This option is shown for IPv4. For IPv6, the IPVer field will read Ford, et al. Expires April 28, 2011 [Page 25] Internet-Draft Multipath TCP October 2010 6, and the length of the address will be 16 octets (instead of 4), and the length of the option will be 2 + (18 * number_of_entries). If there is sufficient TCP option space, multiple addresses can be included, with an ID following on immediately from the previous address. The number of addresses can be deduced from the option length and version fields. The 'P' bit is used to indicate the presence of an additional two octets specifying the port number to use. Although it is expected that the majority of use cases will use the same port pairs as used for the initial subflow (e.g. port 80 remains port 80 on all subflows, as does the ephemeral port at the client, there may be cases (such as port-based load balancing) where the explicit specification of a different port is required. If the P bit is not specified, MPTCP MUST attempt to connect to the specified address on same port as is already in use by the signalling subflow. The 'B' bit is used to indicate that this specified address (and port, if applicable) should be treated as a backup subflow to use only in the event of failure of other working subflows. A receiver of this option SHOULD set up a TCP subflow to the specified address and port, but SHOULD NOT send data on it until the other paths have failed. 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +---------------+---------------+---------------+-------+---+-+-+ | Kind=ADD_ADDR | Length | Address ID | IPVer | |B|P| +---------------+---------------+---------------+-------+---+-+-+ | Address (IPv4 - 4 octets / IPv6 - 16 octets) | +-------------------------------+-------------------------------+ | Port (2 octets if P=1) | ... +-------------------------------+ ( ... further ID/Version/Address/Port fields as required ... ) Figure 11: Add Address (ADD_ADDR) option (shown for IPv4) Due to the proliferation of NATs, it is reasonably likely that one endpoint may attempt to advertise private addresses [9]. We do not wish to blanket prohibit this, since there may be cases where both endpoints have additional interfaces on the same private network. We must ensure, however, that such advertisements do not cause harm. The standard mechanism to create a new subflow (Section 3.2) contains a 32-bit token that uniquely identifies the connection to the receiving endpoint . If the token is unknown, the endpoint will return with a RST. If the token is known, subflow setup will continue, but the sender's token will be sent back. In order for a new subflow to be setup, both tokens must match what each endpoint Ford, et al. Expires April 28, 2011 [Page 26] Internet-Draft Multipath TCP October 2010 expects. This will be further followed by the HMAC exchange for authentication. This will provide sufficient protection against two unconnected endpoints accidentally setting up a new subflow upon the signal of a private address. Ideally, we'd like to ensure the ADD_ADDR (and REMOVE_ADDR) option is sent reliably and in order to the other end. This is to ensure that we don't close the connection when remove/add addresses are processed in reverse order, and to ensure that all possible paths are used. We note, however, that losing reliability and ordering it will not break the multipath connections; they will just reduce the opportunity to open multipath paths and to survive different patterns of path failures. Subflow level ACKs do not cover options, so if we want explicit guarantees we need to build in other mechanisms. Solutions include echoing the options and sending one option per RTT, or adding a sequence number to the option which is explicitly acked in another option. However, we feel these mechanisms' added complexity is not worth the benefits they bring. There are two basic failure modes for options: a) every new option gets stripped or b) some options get stripped, randomly. The second option looks more like a middlebox implementation error, so we believe it is not worth optimizing for. In the first case, resending the option on a different subflow is the thing to do. To achieve similar reliability without explicit ACKs, we propose sending all ADD_ADDR/REMOVE_ADDR options on all existing subflows. If ordering is needed, we should only send one ADD_ADDR/ REMOVE_ADDR option per RTT (modulo lost packets at subflow level). When receiving an ADD_ADDR message with an address ID already in use for that connection, the receiver SHOULD silently ignore the ADD_ADDR. During normal MPTCP operation, it is unlikely that there will be sufficient TCP option space for ADD_ADDR to be included along with those for data sequence numbering (Section 3.3.1). Therefore, it is expected that an MPTCP implementation will send the ADD_ADDR option on separate (either duplicate, or normal but lacking any payload) ACKs. As with all TCP Options, the ADD_ADDR option does not have reliable delivery. Therefore, a sender should send a duplicate ACK with this option on all available subflows. 3.5.2. Remove Address If, during the lifetime of a MPTCP connection, a previously-announced address becomes invalid (e.g. if the interface disappears), the Ford, et al. Expires April 28, 2011 [Page 27] Internet-Draft Multipath TCP October 2010 affected endpoint should announce this so that the other endpoint can remove subflows related to this address. This is achieved through the Remove Address (REMOVE_ADDR) option (Figure 12), which will remove a previously-added address (or list of addresses) from a connection and terminate any subflows currently using that address. For security purposes, if a host receives a REMOVE_ADDR option, it must ensure the affected path(s) are no longer in use before it instigates closure. The receipt of REMOVE_ADDR should first trigger the sending of a TCP Keepalive [10] on the path, and if a response is received the path is not removed. Typical TCP validity tests on the subflow (e.g. ensuring sequence and ack numbers are correct) MUST also be undertaken. The sending and receipt (if no keepalive response was received) of this message SHOULD trigger the sending of RSTs by both endpoints on the affected subflow(s) (if possible), as a courtesy to cleaning up middlebox state, but endpoints may clean up their internal state without a long timeout. Address removal is undertaken by ID, so as to permit the use of NATs and other middleboxes. If there is no address at the requested ID, the receiver will silently ignore the request. The standard way to close a subflow (so long as it is still functioning) is to use a FIN exchange as in regular TCP - for more information, see Section 3.4. 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +---------------+---------------+---------------+ |Kind=REMOVEADDR| Length = 2+n | Address ID | ... +---------------+---------------+---------------+ Figure 12: Remove Address (REMOVE_ADDR) option 3.6. Fallback At the start of a MPTCP connection (i.e. the first subflow), it is important to ensure that the path is fully MPTCP-capable and the necessary TCP options can reach each endpoint. The handshake as described in Section 3.1 will fall back to regular TCP if either of the SYN messages do not have the MPTCP options: this is the same, and desired, behaviour in the case where an endpoint is not MPTCP capable, or the path does not support he MPTCP options. When attempting to join an existing MPTCP connection (Section 3.2), if a Ford, et al. Expires April 28, 2011 [Page 28] Internet-Draft Multipath TCP October 2010 path is not MPTCP capable, the TCP options will not get through on the SYNs and the subflow will be closed. There is, however, another corner case which should be addressed. That is one of MPTCP options getting through on the SYN, but not on regular packets. This can be resolved if the subflow is the first subflow, and thus all data in flight is contiguous. This resolution mechanism is as follows: o The first window's worth of data MUST be DATA_ACKed on every packet o If the first data packet does not have a Data Sequence Mapping option, drop out of MPTCP mode back to regular TCP (and thus send a regular, subflow-level ACK, without a DATA_ACK) o If an ACK is received without a DATA_ACK within the first window, drop out of MPTCP mode back to regular TCP (and thus stop sending data with a Data Sequence Mapping) These rules should cover all cases where such a failure could happen: whether it's on the forward or reverse path, and whether the server or the client first sends data. If lost options on data packets occur on any other subflow apart from the start of the initial subflow, it should be treated as a standard path failure. The data would not be DATA_ACKed (since there is no mapping for the data), and the subflow can be closed with an RST. The case described above is a specialised case of fallback. More generally, fallback to regular TCP can become necessary at any point during a connection if a non-MPTCP-aware middlebox changes the data stream. As described in Section 3.3, each portion of data for which there is a mapping is protected by a checksum. This mechanism is used to detect if middleboxes have made any adjustments to the payload (added, removed, or changed data). A checksum will fail if the data has been changed in any way. This will also detect if the length of data on the subflow is increased or decreased, and this means the Data Sequence Mapping is no longer valid. The sender no longer knows what subflow-level sequence number the receiver is genuinely operating at (the middlebox will be faking ACKs in return), and cannot signal any further mappings. Furthermore, in addition to the possibility of payload modifications that are valid at the application layer, there is the possibility that false-positives could be hit across segment boundaries, corrupting the data. Therefore, all data from the start of the segment that failed the checksum onwards is not trustworthy. Ford, et al. Expires April 28, 2011 [Page 29] Internet-Draft Multipath TCP October 2010 When multiple subflows are in use, the data in-flight on a subflow will likely involve data that is not contiguously part of the connection-level stream, since segments will be spread across the multiple subflows. Due to the problems identified above, it is not possible to determine what the adjustment has done to the data (notably, any changes to the subflow sequence numbering). Therefore, it is not possible to recover the subflow, and the affected subflow must be immediately closed with an RST, featuring a "checksum failed" option, which defines the Data Sequence Number at the start of the segment (defined by the Data Sequence Mapping) which had the checksum failure (see Figure 13). 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +---------------+---------------+---------------+--------------+ | Kind=MP_FAIL | Length=10 | Data Sequence Number (8B) : +---------------+---------------+------------------------------+ : Data Sequence Number (contd.) : +-------------------------------+------------------------------+ : Data Sequence Number (contd.)| +-------------------------------+ Figure 13: Fallback (MP_FAIL) option TBD: In this case, is there any point in signalling Checksum Failed, or could we just RST the subflow? The signal would allow the sender to know there is something wrong with the path and not try to re- establish the subflow (if that was otherwise the policy). Failed data will not be DATA_ACKed and so will be re-transmitted on other subflows (Section 3.3.4). A special case is when there is a single subflow and it fails with a checksum error. Here, MPTCP should be able to recover and continue sending data. There are two possible mechanisms to support this. The first and simplest is to nevertheless close the subflow with a RST, and immediately establish a new one as part of the same MPTCP connection. Since it is known that the path may be compromised, it is not desirable to use MPTCP's segmentation on this path any longer. The new subflow will begin and will signal an infinite mapping (indicated by length=0 in the Data Sequence Mapping option, Section 3.3) from the data sequence number of the segment that failed the checksum. This connection will then continue to appear as a regular TCP session, and a middlebox may change the payload without causing unintentional harm. An optimisation is possible, however. If it is known that all Ford, et al. Expires April 28, 2011 [Page 30] Internet-Draft Multipath TCP October 2010 unacknowledged data in flight is contiguous, an infinite mapping could be applied to the subflow without the need to close it first, and essentially turn off all further MPTCP signalling. In this case, if a receiver identifies a checksum failure when there is only one path, it will send back an MP_FAIL option on the subflow-level ACK. The sender will receive this, and if all unacknowledged data in flight is contiguous, will signal an infinite mapping (if the data is not contiguous, the sender MUST send an RST). This infinite mapping will be a Data Sequence Mapping option on the first new packet, but it acts retroactively, referring to the start of the subflow sequence number of the last segment that was known to be delivered intact. From that point onwards data can be altered by a middlebox without affecting MPTCP, as the data stream is equivalent to a regular, legacy TCP session. After a sender signals an infinite mapping it MUST only use subflow ACKs to clear its send buffer. This is because Data ACKs may become misaligned with the subflow ACKs when middleboxes insert or delete data. The receive SHOULD stop generating Data ACKs after it receives an infinite mapping. When a connection is in fallback mode, only one subflow can send data at a time. Otherwise, the receiver would not know how to reorder the data. However, subflows can be opened and closed as necessary, as long as a single one is active at any point. It should be emphasised that we are not attempting to prevent the use of middleboxes that want to adjust the payload. An MPTCP-aware middlebox to provide such functionality could be designed that would re-write checksums if needed, and additionally would be able to parse the data sequence mappings, and thus not hit false positives though not knowing where data boundaries lie. 3.7. Error Handling In addition to the fallback mechanism as described above, the standard classes of TCP errors may need to be handled in an MPTCP- specific way. Note that changing semantics - such as the relevance of an RST - has already been covered in Section 4. Where possible, we do not want to deviate from regular TCP behaviour. The following list covers possible errors and the appropriate MPTCP behaviour: o Unknown token in MP_JOIN (or token mismatch in MP_JOIN ACK, or missing MP_JOIN in SYN/ACK response): send RST (analogous to TCP's behaviour on an unknown port) Ford, et al. Expires April 28, 2011 [Page 31] Internet-Draft Multipath TCP October 2010 o DSN out of Window (during normal operation): just ignore, however if at the beginning of a new subflow we might want to RST it as a security mechanism o Remove request for unknown address ID: silently ignore o DATA_ACK for data not yet sent: abort connection by RST on every subflow. 3.8. Heuristics There are a number of heuristics that are needed for performance or deployment but which are not required for protocol correctness. In this section we detail such heuristics 3.8.1. Port Usage Under typical operation an MPTCP implementation SHOULD use the same ports as already in use. In other words, the destination port of a SYN containing a MP_JOIN option SHOULD be the same as the remote port of the first subflow in the connection. The local port for such SYNs SHOULD also be the same as for the first subflow (and as such, an implementation SHOULD reserve ephemeral ports across all local IP addresses), although there may be cases where this is infeasible. This strategy is intended to maximize the probability of the SYN being permitted by a firewall or NAT at the recipient and to avoid confusing any network monitoring software. There may also be cases, however, where the passive opener wishes to signal to the other endpoint that a specific port should be used, and this facility is provided in the Add Address option as documented in Section 3.5.1. It is therefore feasible to allow multiple subflows between the same two addresses but using different port pairs, and such a facility could be such a facility could be used to allow load balancing within the network based on 5-tuples (e.g. ECMP). 4. Semantic Issues In order to support multipath operation, the semantics of some TCP components have changed. To aid clarity, this section collects these semantic changes as a reference. Sequence Number: The (in-header) TCP sequence number is specific to the subflow. To allow the receiver to reorder application data, an additional data-level sequence space is used. In this data- level sequence space, the initial SYN and the final DATA_FIN occupy one octet of sequence space. There is an explicit mapping Ford, et al. Expires April 28, 2011 [Page 32] Internet-Draft Multipath TCP October 2010 of data sequence space to subflow sequence space, which is signalled through TCP options in data packets. ACK: The ACK field in the TCP header acknowledges only the subflow sequence number, not the data-level sequence space. Implementations SHOULD NOT attempt to infer a data-level acknowledgement from the subflow ACKs. Instead an explicit data- level DATA_ACK is used. This avoids possible deadlock scenarios when a non-TCP-aware middlebox pro-actively ACKs at the subflow level. Receive Window: The receive window in the TCP header indicates the amount of free buffer space for the whole data-level connection (as opposed to for this subflow) that is available at the receiver. This is the same semantics as regular TCP, but to maintain these semantics the receive window must be interpreted at the sender as relative to the sequence number given in the DATA_ACK rather than the subflow ACK in the TCP header. In this way the original flow control role is preserved. FIN: The FIN flag in the TCP header applies only to the subflow it is sent on, not to the whole connection. For connection-level FIN semantics, the DATA_FIN option is used. RST: The RST flag in the TCP header applies only to the subflow it is sent on, not to the whole connection. A connection is considered reset if a RST is received on every subflow. Address List: Address list management (i.e. knowledge of the local and remote hosts' lists of available IP addresses) is handled on a per-connection basis (as opposed to per-subflow, per host, or per pair of communicating hosts). This permits the application of per-connection local policy. Adding an address to one connection (either explicitly through an Add Address message, or implicitly through a Join) has no implication for other connections between the same pair of hosts. 5-tuple: The 5-tuple (protocol, local address, local port, remote address, remote port) presented by kernel APIs to the application layer in a non-multipath-aware application is that of the first subflow, even if the subflow has since been closed and removed from the connection. This decision, and other related API issues, are discussed in more detail in [5]. Ford, et al. Expires April 28, 2011 [Page 33] Internet-Draft Multipath TCP October 2010 5. Security Considerations As identified in [11], the addition of multipath capability to TCP will bring with it a number of new classes of threat. In order to prevent these, [3] presents a set of requirements for a security solution for MPTCP. The fundamental goal is for the security of MPTCP to be "no worse" than regular TCP today, and the key security requirements are: o Provide a mechanism to confirm that the parties in a subflow handshake are the same as in the original connection setup. o Provide verification that the peer can receive traffic at a new address before using it as part of a connection. o Provide replay protection, i.e. ensure that a request to add/ remove a subflow is 'fresh'. In order to achieve these goals, MPTCP includes a hash-based handshake algorithm documented in Section 3.1 and Section 3.2. The security of the MPTCP connection hangs on the use of keys that are shared once at the start of the first subflow, and never again in the clear. To ease demultiplexing whilst not giving away any cryptographic material, future subflows use a truncated SHA-1 hash of this key as the connection identification "token". The keys are used as keys in a HMAC, and this should verify that the parties in the handshake are the same as in the original connection setup. It also provides verification that the peer can receive traffic at this new address. Replay attacks would still be possible in this scenario, and therefore the handshakes use single-use random numbers (nonces) at both ends - this ensures the HMAC will never be the same on two handshakes. The security mechanism presented in this draft should therefore protect against all forms of flooding and hijacking attacks suggested in [11]. 6. Interactions with Middleboxes Multipath TCP was designed to be deployable in the present world. Its design takes into account "reasonable" existing middlebox behaviour. In this section we outline a few representative middlebox-related failure scenarios and show how multipath TCP handles them. Next, we list the design decisions multipath has made to accomodate the different middleboxes. A primary concern is our use of new TCP options. Most middleboxes should just forward packets with new options unchanged, yet there are Ford, et al. Expires April 28, 2011 [Page 34] Internet-Draft Multipath TCP October 2010 some that don't. These we expect will either strip options and pass the data, drop packets with new options, copy the same option into multiple segments (e.g. when doing segmentation) or drop options during segment coalescing. MPTCP SYN packets contain the MP_CAPABLE option to indicate the use of MPTCP. When the middlebox drops the packet containing the MP_CAPABLE option either on the outgoing or the return path, the connection will fail. Host A SHOULD fall back to TCP in such cases (studies suggest that few middleboxes drop packets with unknown options). The same applies for subflow setup. The second case is when the middleboxes strip options. Let's first discuss behaviour for initial connection SYNs (see Figure 14). If the option is stripped from the packet on the outgoing path, the connection falls back to regular TCP. If the option is stripped on the return path, host B will wait for a DATA_ACK of its connection SYN, retransmitting the SYN/ACK until it declares the connection failed. Host A thinks it is talking to a regular host, and may send data segments, but these will not be acked by host B as they do not have the proper mapping. Hence the connection fails. Host A SHOULD fall back to regular TCP after the connection times out. Subflow SYNs contain the MP_JOIN option. If this option is stripped on the outgoing path the SYN will appear to be a regular SYN to host B. Depending on whether there is a listening socket on the target port, host B will reply either with SYN/ACK or RST (subflow connection fails). When host A receives the SYN/ACK it sends a RST because the SYN/ACK does not contain the MP_JOIN option and its token. Either way, the connection fails. Ford, et al. Expires April 28, 2011 [Page 35] Internet-Draft Multipath TCP October 2010 Host A Host B | Middlebox M | | | | | SYN(MP_CAPABLE) | SYN | |-------------------|---------------->| | SYN/ACK | |<------------------------------------| a) MP_CAPABLE option stripped on outgoing path Host A Host B | SYN(MP_CAPABLE) | |------------------------------------>| | Middlebox M | | | | | SYN/ACK |SYN/ACK(MP_CAPABLE)| |<----------------|-------------------| b) MP_CAPABLE option stripped on return path Figure 14: Connection Setup with Middleboxes that Strip Options from Packets We now examine data flow with MPTCP, assuming the flow is correctly setup which implies the options in the SYN packets were allowed through by the relevant middleboxes. If options are allowed through and there is no resegmentation or coalescing to TCP segments, multipath TCP flows can proceed without problems. The case when options get stripped on data packets has been discussed in the Fallback section. We can further analyze what happens when a fraction of options is stripped. The multipath subflow should survive losing a fraction of DATA_ACKs and data sequence mappings, but performance will degrade as the fraction of stripped options increases. We do not expect such cases to appear in practice, though: most middleboxes will either strip all options or let them all through. We end this section with a list of middlebox classes, their behaviour and the elements in the MPTCP design that allow operation through such middleboxes. Issues surrounding dropping packets with options or stripping options were discussed above, and are not included here: o NAT [12]: changes the source address and port of packets. This means that a host will not know its public-facing address for signalling in MPTCP. Therefore, MPTCP permits implicit address addition via the MP_JOIN option, and has heuristics to ensure that connection attempts to private addresses [9] do not cause problems. Address removal is undertaken by an ID number to allow no knowledge of the source address. Ford, et al. Expires April 28, 2011 [Page 36] Internet-Draft Multipath TCP October 2010 o Performance Enhancing Proxies (PEPs) [13]: might pro-actively ACK data to increase performance. Problems will occur if a PEP ACKs data and then fails before sending it on to the receiver, of it the receiver is mobile and moves away before proactively ACKed data is forwarded on. If subflow ACKs were used to control send buffering, the data could be lost and never be retransmitted, thus causing the subflow to permanently stall. MPTCP therefore uses the DATA_ACK to make progress when one of its subflows fails in this way. This is why MPTCP does not use subflow ACKs to infer connection level ACKs. o Traffic Normalizers [14]: do not allow holes in sequence numbers, cache packets and retransmit the same data. MPTCP looks like standard TCP on the wire, and will not retransmit different data on the same subflow sequence number. o Firewalls [15]: might perform sequence number randomization on TCP connections. MPTCP uses relative sequence numbers in data sequence mapping to cope with this. Like NATs, firewalls will not permit many incoming connections, so MPTCP supports address signalling (ADD_ADDR) so that a multi-addressed endpoint can invite its peer behind the firewall/NAT to connect out to its additional interface. o Intrusion Detection Systems: look out for traffic patterns and content that could threaten a network. Multipath will mean that such data is potentially spread, so it is more difficult for an IDS to analyse the whole traffic, and potentially increasint the risk of false positives. However, for an MPTCP-aware IDS, connection IDs can be easily read by such systems to correlate multiple subflows and re-assemble for analysis. o Application level NATs: may alter the payload within a subflow. Multipath TCP will detect these using the checksum and close the affected subflow(s), if there are other subflows that can be used. If all subflows are affected multipath will fallback to TCP, allowing middleboxes to change the payload. o Middleboxes that alter the receive window: MPTCP will use the maximum window at data-level, but will also obey subflow specific windows. In addition, all classes of middleboxes may affect TCP traffic in the following ways: o TCP Options: may be removed, or packets with unknown options dropped, by many classes of middleboxes. It is intended that the initial SYN exchange, with a TCP Option, will be sufficient to Ford, et al. Expires April 28, 2011 [Page 37] Internet-Draft Multipath TCP October 2010 identify the path capabilities. If such a packet does not get through, MPTCP will end up falling back to regular TCP. o Segmentation/Coalescing (e.g. tcp segmentation offloading, etc): might copy options between packets and might strip some options. MPTCP's data sequence mapping includes the subflow sequence number instead of using the sequence number in the segment. In this way, the mapping is independent of the packets that carry it. 7. Interfaces TBD Interface with applications, interface with TCP, interface with lower layers... Discussion of interaction with applications (both in terms of how MPTCP will affect an application's assumptions of the transport layer, and what API extensions an application may wish to use with MPTCP) are discussed in [5]. 8. Acknowledgements The authors are supported by Trilogy (http://www.trilogy-project.org), a research project (ICT-216372) partially funded by the European Community under its Seventh Framework Program. The views expressed here are those of the author(s) only. The European Commission is not liable for any use that may be made of the information in this document. The authors gratefully acknowledge significant input into this document from Olivier Bonaventure and Andrew McDonald. The authors also wish to acknowledge reviews and contributions from Iljitsch van Beijnum, Lars Eggert, Marcelo Bagnulo, Robert Hancock, Pasi Sarolahti, Toby Moncaster, Philip Eardley, Sergio Lembo, and Lawrence Conroy. 9. IANA Considerations This document will make a request to IANA to allocate new values for TCP Option identifiers, as follows: Ford, et al. Expires April 28, 2011 [Page 38] Internet-Draft Multipath TCP October 2010 +-------------+-----------------------------+---------------+-------+ | Symbol | Name | Ref | Value | +-------------+-----------------------------+---------------+-------+ | MP_CAPABLE | Multipath Capable | Section 3.1 | (tbc) | | MP_JOIN | Join Connection | Section 3.2 | (tbc) | | ADD_ADDR | Add Address | Section 3.5.1 | (tbc) | | REMOVE_ADDR | Remove Address | Section 3.5.2 | (tbc) | | DSN_MAP | Data Sequence Number | Section 3.3 | (tbc) | | | Mapping | | | | DATA_ACK | Data-level Acknowledgment | Section 3.3 | (tbc) | | DATA_FIN | Data-level FIN | Section 3.4 | (tbc) | | MP_PRIO | Change Subflow Priority | Section 3.3.6 | (tbc) | | MP_FAIL | Fallback | Section 3.6 | (tbc) | +-------------+-----------------------------+---------------+-------+ Table 1: TCP Options for MPTCP 10. References 10.1. Normative References [1] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. 10.2. Informative References [2] Postel, J., "Transmission Control Protocol", STD 7, RFC 793, September 1981. [3] Ford, A., Raiciu, C., Handley, M., and J. Iyengar, "Architectural Guidelines for Multipath TCP Development", draft-ietf-mptcp-architecture-02 (work in progress), October 2010. [4] Raiciu, C., Handley, M., and D. Wischik, "Coupled Multipath- Aware Congestion Control", draft-ietf-mptcp-congestion-00 (work in progress), July 2010. [5] Scharf, M. and A. Ford, "MPTCP Application Interface Considerations", draft-scharf-mptcp-api-02 (work in progress), July 2010. [6] Eastlake, D. and T. Hansen, "US Secure Hash Algorithms (SHA and HMAC-SHA)", RFC 4634, July 2006. [7] Mathis, M., Mahdavi, J., Floyd, S., and A. Romanow, "TCP Selective Acknowledgment Options", RFC 2018, October 1996. Ford, et al. Expires April 28, 2011 [Page 39] Internet-Draft Multipath TCP October 2010 [8] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition of Explicit Congestion Notification (ECN) to IP", RFC 3168, September 2001. [9] Rekhter, Y., Moskowitz, R., Karrenberg, D., Groot, G., and E. Lear, "Address Allocation for Private Internets", BCP 5, RFC 1918, February 1996. [10] Braden, R., "Requirements for Internet Hosts - Communication Layers", STD 3, RFC 1122, October 1989. [11] Bagnulo, M., "Threat Analysis for Multi-addressed/Multi-path TCP", draft-ietf-mptcp-threat-03 (work in progress), October 2010. [12] Srisuresh, P. and K. Egevang, "Traditional IP Network Address Translator (Traditional NAT)", RFC 3022, January 2001. [13] Border, J., Kojo, M., Griner, J., Montenegro, G., and Z. Shelby, "Performance Enhancing Proxies Intended to Mitigate Link-Related Degradations", RFC 3135, June 2001. [14] Handley, M., Paxson, V., and C. Kreibich, "Network Intrusion Detection: Evasion, Traffic Normalization, and End-to-End Protocol Semantics", Usenix Security 2001, 2001, . [15] Freed, N., "Behavior of and Requirements for Internet Firewalls", RFC 2979, October 2000. Appendix A. Notes on use of TCP Options The TCP option space is limited due to the length of the Data Offset field in the TCP header (4 bits), which defines the TCP header length in 32-bit words. With the standard TCP header being 20 bytes, this leaves a maximum of 40 bytes for options, and many of these may already be used by options such as timestamp and SACK. We have performed a brief study on the commonly used TCP options in SYN, data, and pure ACK packets, and found that there is enough room to fit all the options we propose using in this draft. SYN packets typically include MSS (4 bytes), window scale (3 bytes), SACK permitted (2 bytes) and timestamp (10 bytes) options. Together these sum to 19 bytes. Some operating systems appear to pad each option up to a word boundary, thus using 24 bytes (a brief survey suggests Windows XP and Mac OS X do this, whereas Linux does not). Ford, et al. Expires April 28, 2011 [Page 40] Internet-Draft Multipath TCP October 2010 Optimistically, therefore, we have 21 bytes spare, or 16 if it has to be word-aligned. In either case, however, the Multipath Capable (12 bytes) and Join (12 bytes) options will fit in this remaining space. TCP data packets typically carry timestamp options in every packet, taking 10 bytes (or 12 with padding). That leaves 30 bytes (or 28, if word-aligned), which are enough to encode the data sequence mapping (14 or 18 bytes, depending on the length of the sequence number in use) and the DATA_ACK if the flow is bidirectional (6 or 10 bytes). Such options will just fit in the available option space, although 8 byte data-level sequence numbers in both will only fit if word-alignment is not required. If this proves to be a problem, it is not necessary to include the Data Sequence Mapping and DATA_ACK in each packet, and in many cases it may be possible to alternate their presence (so long as the mapping covers the data being sent in the following packet). Other options include: wrapping the DATA_ACK into the Data Sequence Mapping option; alternating between 4 and 8 byte sequence numbers in each option; and sending the DATA_ACK on a duplicate subflow-level ACK. Pure ACKs in TCP typically contain only timestamps (10B). Here, multipath TCP typically needs to encode the DATA_ACK (max 10B). Occasionally ACKs will contain SACK information. Depending on the number of lost packets, SACK may utilize the entire option space. If a DATA_ACK had to be included, then it is probably necessary to reduce the number of SACK blocks by one to accomodate the DATA_ACK. However, the presence of the DATA_ACK is unlikely to be necessary in a case where SACK is in use, however, since until at least some of the SACK blocks have been retransmitted, the cumulative data-level ACK will not be moving forward (or if it does, due to retransmissions on antoher path, then that path can also be used to transmit the new DATA_ACK). The ADD_ADDR option can be between 8 and 22 bytes, depending on whether IPv4 or IPv6 is used, and whether the Port number is present or not. It is unlikely that such signalling would fit in a data packet (although if there is space, it is fine to include it). It is recommended to use duplicate ACKs with no other payload or options in order to transmit these rare signals. Finally, there are issues with options reliability. As options can also be sent on pure ACKs, these are not reliably sent. This is not an issue for DATA_ACK due to their cumulative nature, but may be an issue for ADD_ADDR/REMOVE_ADDR options. Here we favour redundant transmissions at the sender (whether on multiple paths, or on the same path on a number of ACKs). The cases where options are stripped by middleboxes are discussed in Section 6. Ford, et al. Expires April 28, 2011 [Page 41] Internet-Draft Multipath TCP October 2010 Appendix B. Resync Packet In earlier versions of this draft, we proposed the use of a "re-sync" option that would be used in certain circumstances when a sender needs to instruct the receiver to skip over certain subflow sequence numbers (i.e. to treat the specified sequence space as having been received and acknowledged). The typical use of this option will be when packets are retransmitted on different subflows, after failing to be acknowledged on the original subflow. In such a case, it becomes necessary to move forward the original subflow's sequence numbering so as not to later transmit different data with a previously used sequence number (i.e. when more data comes to be transmitted on the original subflow, it would be different data, and so must not be sent with previously-used (but unacknowledged) sequence numbering). The rationale for needing to do this is two-fold: firstly, when ACKs are received they are for the subflow only, and the sender infers from this the data that was sent - if the same sequence space could be occupied by different data, the sender won't know whether the intended data was received. Secondly, certain classes of middleboxes may cache data and not send the new data on a previously-seen sequence number. This option was dropped, however, since some middleboxes may get confused when they meet a hole in the sequence space, and do not understand the resync option. It is therefore felt that the same data must continue to be retransmitted on a subflow even if it is already received after being retransmitted on another. There should not be a significant performance hit from this since the amount of data involved and needing to be retransmitted multiple times will be relatively small. Appendix C. Changelog This section maintains logs of significant changes made to this document between versions. C.1. Changes since draft-ietf-mptcp-multiaddressed-01 o Added proposal for hash-based security mechanism. o Added receiver subflow policy control (backup path flags and MP_PRIO option). Ford, et al. Expires April 28, 2011 [Page 42] Internet-Draft Multipath TCP October 2010 o Changed DSN_MAP checksum to use the TCP checksum algorithm. C.2. Changes since draft-ietf-mptcp-multiaddressed-00 o Various clarifications and minor re-structuring in response to comments. C.3. Changes since draft-ford-mptcp-multiaddressed-03 o Clarified handshake mechanism, especially with regard to error cases (Section 3.2). o Added optional port to ADD_ADDR and clarified situation with private addresses (Section 3.5.1). o Added path liveness check to REMOVE_ADDR (Section 3.5.2). o Added chunk checksumming to DSN_MAP (Section 3.3.1) to detect payload-altering middleboxes, and defined fallback mechanism (Section 3.6). o Major clarifications to receive window discussion (Section 3.3.4). o Various textual clarifications, especially in examples. C.4. Changes since draft-ford-mptcp-multiaddressed-02 o Remove Version and Address ID in MP_CAPABLE in Section 3.1, and make ISN be 6 bytes. o Data sequence numbers are now always 8 bytes. But in some cases where it is unambiguous it is permissible to only send the lower 4 bytes if space is at a premium. o Clarified behaviour of MP_JOIN in Section 3.2. o Added DATA_ACK to Section 3.3. o Clarified fallback to non-multipath once a non-MP-capable SYN is sent. Ford, et al. Expires April 28, 2011 [Page 43] Internet-Draft Multipath TCP October 2010 Authors' Addresses Alan Ford Roke Manor Research Old Salisbury Lane Romsey, Hampshire SO51 0ZN UK Phone: +44 1794 833 465 Email: alan.ford@roke.co.uk Costin Raiciu University College London Gower Street London WC1E 6BT UK Email: c.raiciu@cs.ucl.ac.uk Mark Handley University College London Gower Street London WC1E 6BT UK Email: m.handley@cs.ucl.ac.uk Ford, et al. Expires April 28, 2011 [Page 44]