RFC 6824
TCP Extensions for Multipath Operation with Multiple Addresses
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 signaled using optional TCP header fields. A single TCP option number ("Kind") has been assigned by IANA for MPTCP (see Section 8), and then individual messages will be determined by a "subtype", the values of which are also stored in an IANA registry (and are also listed in Section 8).
Throughout this document, when reference is made to an MPTCP option by symbolic name, such as "MP_CAPABLE", this refers to a TCP option with the single MPTCP option type, and with the subtype value of the symbolic name as defined in Section 8. This subtype is a 4-bit field -- the first 4 bits of the option payload, as shown in Figure 3. The MPTCP messages are defined in the following sections. 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 | Length |Subtype| | +---------------+---------------+-------+ | | Subtype-specific data | | (variable length) | +---------------------------------------------------------------+ Figure 3: MPTCP Option Format Those MPTCP options associated with subflow initiation are used on packets with the SYN flag set. Additionally, there is one MPTCP option for signaling metadata to ensure segmented data can be recombined for delivery to the application. The remaining options, however, are signals that do not need to be on a specific packet, such as those for signaling additional addresses. Whilst an implementation may desire to send MPTCP options as soon as possible, it may not be possible to combine all desired options (both those for MPTCP and for regular TCP, such as SACK (selective acknowledgment) [11]) on a single packet. Therefore, an implementation may choose to send duplicate ACKs containing the additional signaling information. This changes the semantics of a duplicate ACK; these are usually only sent as a signal of a lost segment [12] in regular TCP. Therefore, an MPTCP implementation receiving a duplicate ACK that contains an MPTCP option MUST NOT treat it as a signal of congestion. Additionally, an MPTCP implementation SHOULD NOT send more than two duplicate ACKs in a row for the purposes of sending MPTCP options alone, in order to ensure no middleboxes misinterpret this as a sign of congestion. Furthermore, standard TCP validity checks (such as ensuring the sequence number and acknowledgment number are within window) MUST be undertaken before processing any MPTCP signals, as described in [13].
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 4). This option declares its sender is capable of performing Multipath TCP and wishes to do so on this particular connection. This option is used to declare the 64-bit key that the sender has generated for this MPTCP connection. This key is used to authenticate the addition of future subflows to this connection. This is the only time the key will be sent in clear on the wire (unless "fast close", Section 3.5, is used); all future subflows will identify the connection using a 32-bit "token". This token is a cryptographic hash of this key. The algorithm for this process is dependent on the authentication algorithm selected; the method of selection is defined later in this section. This key is generated by its sender, and its method of generation is implementation specific. The key MUST be hard to guess, and it MUST be unique for the sending host at any one time. Recommendations for generating random numbers for use in keys are given in [14]. Connections will be indexed at each host by the token (a one-way hash of the key). Therefore, an implementation will require a mapping from each token to the corresponding connection, and in turn to the keys for the connection. There is a risk that two different keys will hash to the same token. The risk of hash collisions is usually small, unless the host is handling many tens of thousands of connections. Therefore, an implementation SHOULD check its list of connection tokens to ensure there is not a collision before sending its key in the SYN/ACK. This would, however, be costly for a server with thousands of connections. The subflow handshake mechanism (Section 3.2) will ensure that new subflows only join the correct connection, however, through the cryptographic handshake, as well as checking the connection tokens in both directions, and ensuring sequence numbers are in-window. So in the worst case if there was a token collision, the new subflow would not succeed, but the MPTCP connection would continue to provide a regular TCP service. 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 for this connection. o SYN/ACK (B->A): B's Key for this connection. o ACK (A->B): A's Key followed by 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 4, "sender" and "receiver" refer to the sender or receiver of the TCP packet (which can be either host). If the SYN flag is set, a single key is included; if only an ACK flag is set, both keys are present. B's Key is echoed in the ACK in order to allow the listener (Host B) 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 way that would allow it to verify that it generated the key when it is echoed in the ACK. This exchange allows the safe passage of MPTCP options on SYN packets to be determined. If any of these options are dropped, MPTCP will gracefully fall back to regular single-path TCP, as documented in Section 3.6. Note that new subflows MUST NOT be established (using the process documented in Section 3.2) until a Digital Signature Standard (DSS) option has been successfully received across the path (as documented in Section 3.3). 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 | Length |Subtype|Version|A|B|C|D|E|F|G|H| +---------------+---------------+-------+-------+---------------+ | Option Sender's Key (64 bits) | | | | | +---------------------------------------------------------------+ | Option Receiver's Key (64 bits) | | (if option Length == 20) | | | +---------------------------------------------------------------+ Figure 4: Multipath Capable (MP_CAPABLE) Option The first 4 bits of the first octet in the MP_CAPABLE option (Figure 4) define the MPTCP option subtype (see Section 8; for MP_CAPABLE, this is 0), and the remaining 4 bits of this octet specify the MPTCP version in use (for this specification, this is 0).
The second octet is reserved for flags, allocated as follows:
A: The leftmost bit, labeled "A", SHOULD be set to 1 to indicate
"Checksum Required", unless the system administrator has decided
that checksums are not required (for example, if the environment
is controlled and no middleboxes exist that might adjust the
payload).
B: The second bit, labeled "B", is an extensibility flag, and MUST be
set to 0 for current implementations. This will be used for an
extensibility mechanism in a future specification, and the impact
of this flag will be defined at a later date. If receiving a
message with the 'B' flag set to 1, and this is not understood,
then this SYN MUST be silently ignored; the sender is expected to
retry with a format compatible with this legacy specification.
Note that the length of the MP_CAPABLE option, and the meanings of
bits "C" through "H", may be altered by setting B=1.
C through H: The remaining bits, labeled "C" through "H", are used
for crypto algorithm negotiation. Currently only the rightmost
bit, labeled "H", is assigned. Bit "H" indicates the use of HMAC-
SHA1 (as defined in Section 3.2). An implementation that only
supports this method MUST set bit "H" to 1, and bits "C" through
"G" to 0.
A crypto algorithm MUST be specified. If flag bits C through H are
all 0, the MP_CAPABLE option MUST be treated as invalid and ignored
(that is, it must be treated as a regular TCP handshake).
The selection of the authentication algorithm also impacts the
algorithm used to generate the token and the initial data sequence
number (IDSN). In this specification, with only the SHA-1 algorithm
(bit "H") specified and selected, the token MUST be a truncated (most
significant 32 bits) SHA-1 hash ([4], [15]) of the key. A different,
64-bit truncation (the least significant 64 bits) of the SHA-1 hash
of the key MUST be used as the initial data sequence number. Note
that the key MUST be hashed in network byte order. Also note that
the "least significant" bits MUST be the rightmost bits of the SHA-1
digest, as per [4]. Future specifications of the use of the crypto
bits may choose to specify different algorithms for token and IDSN
generation.
Both the crypto and checksum bits negotiate capabilities in similar
ways. For the Checksum Required bit (labeled "A"), if either host
requires the use of checksums, checksums MUST be used. In other
words, the only way for checksums not to be used is if both hosts in
their SYNs set A=0. This decision is confirmed by the setting of the
"A" bit in the third packet (the ACK) of the handshake. For example,
if the initiator sets A=0 in the SYN, but the responder sets A=1 in the SYN/ACK, checksums MUST be used in both directions, and the initiator will set A=1 in the ACK. The decision whether to use checksums will be stored by an implementation in a per-connection binary state variable. For crypto negotiation, the responder has the choice. The initiator creates a proposal setting a bit for each algorithm it supports to 1 (in this version of the specification, there is only one proposal, so bit "H" will be always set to 1). The responder responds with only 1 bit set -- this is the chosen algorithm. The rationale for this behavior is that the responder will typically be a server with potentially many thousands of connections, so it may wish to choose an algorithm with minimal computational complexity, depending on the load. If a responder does not support (or does not want to support) any of the initiator's proposals, it can respond without an MP_CAPABLE option, thus forcing a fallback to regular TCP. 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. 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; thus, the MPTCP session MUST operate as a 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 third packet (the ACK) does not contain the MP_CAPABLE option, then the session MUST fall back to operating as a regular, single-path TCP. This is to maintain compatibility with middleboxes on the path that drop some or all TCP options. Note that an implementation MAY choose to attempt sending MPTCP options more than one time before making this decision to operate as regular TCP (see Section 3.8). 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. It is possible that MPTCP and non-MPTCP SYNs could get reordered in the network. Therefore, the 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 on an MPTCP connection is generated from the key. The algorithm for IDSN generation is also determined from the negotiated authentication algorithm. In this specification, with only the SHA-1 algorithm specified and selected, the IDSN of a host MUST be the least significant 64 bits of the SHA-1 hash of its key, i.e., IDSN-A = Hash(Key-A) and IDSN-B = Hash(Key-B). This deterministic generation of the IDSN allows a receiver to ensure that there are no gaps in sequence space at the start of the connection. The SYN with MP_CAPABLE occupies the first octet of data sequence space, although this does not need to be acknowledged at the connection level until the first data is sent (see Section 3.3).3.2. Starting a New Subflow
Once an MPTCP connection has begun with the MP_CAPABLE exchange, further subflows can be added to the connection. Hosts have knowledge of their own address(es), and can become aware of the other host's addresses through signaling exchanges as described in Section 3.4. Using this knowledge, a host can initiate a new subflow over a currently unused pair of addresses. It is permitted for either host in a connection to initiate the creation of a new subflow, but it is expected that this will normally be the original connection initiator (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 is used to identify the connection to be joined by the new subflow. It uses keying material that was exchanged in the initial MP_CAPABLE handshake (Section 3.1), and that handshake also negotiates the crypto algorithm in use for the MP_JOIN handshake. This section specifies the behavior of MP_JOIN using the HMAC-SHA1 algorithm. An MP_JOIN option is present in the SYN, SYN/ACK, and ACK of the three-way handshake, although in each case with a different format. In the first MP_JOIN on the SYN packet, illustrated in Figure 5, the initiator sends a token, random number, and address ID. The token is used to identify the MPTCP connection and is a cryptographic hash of the receiver's key, as exchanged in the initial MP_CAPABLE handshake (Section 3.1). In this specification, the tokens presented in this option are generated by the SHA-1 ([4], [15]) 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). Note that the hash generation algorithm can be overridden by the choice of cryptographic handshake algorithm, as defined in Section 3.1. The MP_JOIN SYN sends not only the token (which is static for a connection) but also random numbers (nonces) that are used to prevent replay attacks on the authentication method. Recommendations for the generation of random numbers for this purpose are given in [14]. 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, even if the IP header has been changed in transit by a middlebox. The Address ID allows address removal (Section 3.4.2) without needing to know what the source address at the receiver is, thus allowing address removal through NATs. The Address ID also allows correlation between new subflow setup attempts and address signaling (Section 3.4.1), to prevent setting up duplicate subflows on the same path, if an MP_JOIN and ADD_ADDR are sent at the same time. 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. A host MUST store the mappings between Address IDs and addresses both for itself and the remote host. An implementation will also need to know which local and remote Address IDs are associated with which established subflows, for when addresses are removed from a local or remote host. The MP_JOIN option on packets with the SYN flag set also includes 4 bits of flags, 3 of which are currently reserved and MUST be set to zero by the sender. The final bit, labeled "B", indicates whether the sender of this option wishes this subflow to be used 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. By setting B=1, the sender of the option is requesting the other host to only send data on this subflow if there are no available subflows where B=0. Subflow policy is discussed in more detail in Section 3.3.8.
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 | Length = 12 |Subtype| |B| Address ID | +---------------+---------------+-------+-----+-+---------------+ | Receiver's Token (32 bits) | +---------------------------------------------------------------+ | Sender's Random Number (32 bits) | +---------------------------------------------------------------+ Figure 5: Join Connection (MP_JOIN) Option (for Initial SYN) When receiving a SYN with an 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 a random number and a truncated (leftmost 64 bits) Hash-based Message Authentication Code (HMAC). This version of the option is shown in Figure 6. If the token is unknown, or the host wants to refuse subflow establishment (for example, due to a limit on the number of subflows it will permit), the receiver will send back a reset (RST) signal, analogous to an unknown port in TCP. Although calculating an HMAC requires cryptographic operations, it is believed that the 32- bit token in the MP_JOIN SYN gives sufficient protection against blind state exhaustion attacks; therefore, there is no need to provide mechanisms to allow a responder to operate statelessly at the MP_JOIN stage. An HMAC is sent by both hosts -- by the initiator (Host A) in the third packet (the ACK) and by the responder (Host B) in the second packet (the SYN/ACK). Doing the HMAC exchange at this stage allows both hosts to have first exchanged random data (in the first two SYN packets) that is used as the "message". This specification defines that HMAC as defined in [10] is used, along with the SHA-1 hash algorithm [4] (potentially implemented as in [15]), thus generating a 160-bit / 20-octet HMAC. Due to option space limitations, the HMAC included in the SYN/ACK is truncated to the leftmost 64 bits, but this is acceptable since random numbers are used; thus, an attacker only has one chance to guess the HMAC correctly (if the HMAC is incorrect, the TCP connection is closed, so a new MP_JOIN negotiation with a new random number is required). The initiator's authentication information is sent in its first ACK (the third packet of the handshake), as shown in Figure 7. This data needs to be sent reliably, since it is the only time this HMAC is sent; therefore, receipt of this packet MUST trigger a regular TCP ACK in response, and the packet MUST be retransmitted if this ACK is not received. In other words, sending the ACK/MP_JOIN packet places the subflow in the PRE_ESTABLISHED state, and it moves to the
ESTABLISHED state only on receipt of an ACK from the receiver. It is not permitted to send data while in the PRE_ESTABLISHED state. The reserved bits in this option MUST be set to zero by the sender. 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. These are the keys that were exchanged in the original MP_CAPABLE handshake. The "message" for the HMAC algorithm 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. 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 | Length = 16 |Subtype| |B| Address ID | +---------------+---------------+-------+-----+-+---------------+ | | | Sender's Truncated HMAC (64 bits) | | | +---------------------------------------------------------------+ | Sender's Random Number (32 bits) | +---------------------------------------------------------------+ Figure 6: Join Connection (MP_JOIN) Option (for Responding SYN/ACK) 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 | Length = 24 |Subtype| (reserved) | +---------------+---------------+-------+-----------------------+ | | | | | Sender's HMAC (160 bits) | | | | | +---------------------------------------------------------------+ Figure 7: Join Connection (MP_JOIN) Option (for Third ACK) These various TCP options fit together to enable authenticated subflow setup as illustrated in Figure 8.
Host A Host B ------------------------ ---------- Address A1 Address A2 Address B1 ---------- ---------- ---------- | | | | 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(HMAC-B, R-B) | | | | | | ACK + MP_JOIN(HMAC-A) | | |------------------------------->| | |<-------------------------------| | | ACK | HMAC-A = HMAC(Key=(Key-A+Key-B), Msg=(R-A+R-B)) HMAC-B = HMAC(Key=(Key-B+Key-A), Msg=(R-B+R-A)) Figure 8: 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 for the subflow. If the token is accepted at Host B, but the HMAC returned to Host A does not match the one expected, Host A MUST close the subflow with a TCP RST. If Host B does not receive the expected HMAC, or the MP_JOIN option is missing from the ACK, it MUST close the subflow with a TCP RST. If the HMACs are verified as correct, then both hosts have authenticated each other as being the same peers as existed at the start of the connection, and they have agreed of which connection this subflow will become a part. 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.
This covers all cases of the loss of an MP_JOIN. In more detail, if MP_JOIN is stripped from the SYN on the path from A to B, and Host B does not have a passive opener on the relevant port, it will respond with a 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. Note 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. Demultiplexing 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 set up, demultiplexing packets is done using the 5-tuple, as in traditional TCP. The 5-tuples will be mapped to the local connection identifier (token). Note that Host A will know its local token for the subflow even though it is not sent on the wire -- only the responder's token is sent.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 behavior in detail. The data sequence mapping and the Data ACK are signaled in the Data Sequence Signal (DSS) option (Figure 9). Either or both can be signaled in one DSS, dependent on the flags set. The data sequence mapping defines how the sequence space on the subflow maps to the connection level, and the Data ACK acknowledges receipt of data at the connection level. These functions are described in more detail in the following two subsections. Either or both the data sequence mapping and the Data ACK can be signaled in the DSS option, dependent on the flags set.
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 | Length |Subtype| (reserved) |F|m|M|a|A| +---------------+---------------+-------+----------------------+ | Data ACK (4 or 8 octets, depending on flags) | +--------------------------------------------------------------+ | Data sequence number (4 or 8 octets, depending on flags) | +--------------------------------------------------------------+ | Subflow Sequence Number (4 octets) | +-------------------------------+------------------------------+ | Data-Level Length (2 octets) | Checksum (2 octets) | +-------------------------------+------------------------------+ Figure 9: Data Sequence Signal (DSS) Option The flags, when set, define the contents of this option, as follows: o A = Data ACK present o a = Data ACK is 8 octets (if not set, Data ACK is 4 octets) o M = Data Sequence Number (DSN), Subflow Sequence Number (SSN), Data-Level Length, and Checksum present o m = Data sequence number is 8 octets (if not set, DSN is 4 octets) The flags 'a' and 'm' only have meaning if the corresponding 'A' or 'M' flags are set; otherwise, they will be ignored. The maximum length of this option, with all flags set, is 28 octets. The 'F' flag indicates "DATA_FIN". If present, this means that this mapping covers the final data from the sender. This is the connection-level equivalent to the FIN flag in single-path TCP. A connection is not closed unless there has been a DATA_FIN exchange or a timeout. The purpose of the DATA_FIN and the interactions between this flag, the subflow-level FIN flag, and the data sequence mapping are described in Section 3.3.3. The remaining reserved bits MUST be set to zero by an implementation of this specification. Note that the checksum is only present in this option if the use of MPTCP checksumming has been negotiated at the MP_CAPABLE handshake (see Section 3.1). The presence of the checksum can be inferred from the length of the option. If a checksum is present, but its use had not been negotiated in the MP_CAPABLE handshake, the checksum field MUST be ignored. If a checksum is not present when its use has been negotiated, the receiver MUST close the subflow with a RST as it is considered broken.
3.3.1. Data Sequence Mapping
The data stream as a whole can be reassembled through the use of the data sequence mapping components of the DSS option (Figure 9), which define the mapping from the subflow sequence number to the data 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 [11] is used at the subflow level to improve efficiency. The data sequence mapping specifies a mapping from subflow sequence space to data sequence space. This is expressed in terms of starting sequence numbers for the subflow and the data level, and a length of bytes for which this mapping is valid. This explicit mapping for a range of data was chosen rather than per-packet signaling 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. A mapping is fixed, in that the subflow sequence number is bound to the data sequence number after the mapping has been processed. A sender MUST NOT change this mapping after it has been declared; however, the same data sequence number can be mapped to by different subflows for retransmission purposes (see Section 3.3.6). This would also permit the same data to be sent simultaneously on multiple subflows for resilience or efficiency purposes, especially in the case of lossy links. Although the detailed specification of such operation is outside the scope of this document, an implementation SHOULD treat the first data that is received at a subflow for the data sequence space as that which should be delivered to the application, and any later data for that sequence space ignored. The data sequence number is specified as an absolute value, whereas the subflow sequence numbering is relative (the SYN at the start of the subflow has relative subflow sequence number 0). This is to allow middleboxes to change the initial sequence number of a subflow, such as firewalls that undertake ISN randomization. The data sequence mapping also contains a checksum of the data that this mapping covers, if use of checksums has been negotiated at the MP_CAPABLE exchange. Checksums are 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, since MPTCP can no longer reliably know the subflow sequence space at the receiver to build data sequence mappings. The checksum algorithm used is the standard TCP checksum [1], operating over the data covered by this mapping, along with a pseudo- header as shown in Figure 10. 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 +--------------------------------------------------------------+ | | | Data Sequence Number (8 octets) | | | +--------------------------------------------------------------+ | Subflow Sequence Number (4 octets) | +-------------------------------+------------------------------+ | Data-Level Length (2 octets) | Zeros (2 octets) | +-------------------------------+------------------------------+ Figure 10: Pseudo-Header for DSS Checksum Note that the data sequence number used in the pseudo-header is always the 64-bit value, irrespective of what length is used in the DSS option itself. The standard TCP checksum 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-headers, 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, and adding the checksum for the DSS pseudo- header. Note that checksumming relies on the TCP subflow containing contiguous data; therefore, a TCP subflow MUST NOT use the Urgent Pointer to interrupt an existing mapping. Further note, however, that if Urgent data is received on a subflow, it SHOULD be mapped to the data sequence space and delivered to the application analogous to Urgent data in regular TCP. 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 (if it is in-window). 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 that a mapping will arrive shortly. Such unmapped
data cannot be counted as being within the connection level 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 a RST, and any 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 numbers is not required, then it is permissible to include just the lower 32 bits of the data sequence number in the data sequence mapping and/or Data ACK as an optimization, and an implementation can make this choice independently for each packet. An implementation MUST send the full 64-bit data sequence number if it is transmitting at a sufficiently high rate that the 32-bit value could wrap within the Maximum Segment Lifetime (MSL) [16]. The lengths of the DSNs used in these values (which may be different) are declared with flags in the DSS option. Implementations MUST accept a 32-bit DSN 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. A sanity check MUST be implemented to ensure that a wrap occurs at an expected time (e.g., the sequence number jumps from a very high number to a very low number) and is not triggered by out- of-order packets. 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 specification requires 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 described in Section 3.1. A data sequence mapping 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 hosts, another is when segments of data are scheduled in larger than packet- sized chunks. An "infinite" mapping can be used to fall back 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 of the DSS option to the reserved value of 0. The checksum, in such a case, will also be set to zero.
3.3.2. Data Acknowledgments
To provide full end-to-end resilience, MPTCP provides a connection- level acknowledgment, to act as a cumulative ACK for the connection as a whole. This is the "Data ACK" field of the DSS option (Figure 9). The Data ACK is analogous to the behavior of the standard TCP cumulative ACK -- indicating how much data has been successfully received (with no holes). This is in comparison to the subflow-level ACK, which acts analogous to TCP SACK, given that there may still be holes in the data stream at the connection level. The Data ACK specifies the next data sequence number it expects to receive. The Data ACK, as for the DSN, can be sent as the full 64-bit value, or as the lower 32 bits. If data is received with a 64-bit DSN, it MUST be acknowledged with a 64-bit Data ACK. If the DSN received is 32 bits, it is valid for the implementation to choose whether to send a 32-bit or 64-bit Data ACK. The Data ACK proves that the data, and all required MPTCP signaling, has been received and accepted by the remote end. One key use of the Data ACK signal is that it is used to indicate the left edge of the advertised receive window. As explained in Section 3.3.4, the receive window is shared by all subflows and is relative to the Data ACK. Because of this, an implementation MUST NOT use the RCV.WND field of a TCP segment at the connection level if it does not also carry a DSS option with a Data ACK field. Furthermore, separating the connection-level acknowledgments from the subflow level allows processing to be done separately, and a receiver has the freedom to drop segments after acknowledgment at the subflow level, for example, due to memory constraints when many segments arrive out of order. An MPTCP sender MUST NOT free data from the send buffer until it has been acknowledged by both a Data ACK received on any subflow and at the subflow level by all subflows on which the data was sent. 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 retransmitted. Note, however, that if some data needs to be retransmitted multiple times over a subflow, there is a risk of blocking the sending window. In this case, the MPTCP sender can decide to terminate the subflow that is behaving badly by sending a RST. The Data ACK MAY be included in all segments; however, optimizations SHOULD be considered in more advanced implementations, where the Data ACK is present in segments only when the Data ACK value advances, and
this behavior MUST be treated as valid. This behavior ensures the sender buffer is freed, while reducing overhead when the data transfer is unidirectional.3.3.3. 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 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; for regular TCP, this would result in a FIN on the connection. For MPTCP, an equivalent mechanism is needed, and this is referred to as the DATA_FIN. A DATA_FIN is an indication that the sender has no more data to send, and as such can be used to verify that all data has been successfully received. A DATA_FIN, as with the FIN on a regular TCP connection, is a unidirectional signal. The DATA_FIN is signaled by setting the 'F' flag in the Data Sequence Signal option (Figure 9) to 1. A DATA_FIN occupies 1 octet (the final octet) of the connection-level sequence space. Note that the DATA_FIN is included in the Data-Level Length, but not at the subflow level: for example, a segment with DSN 80, and Data-Level Length 11, with DATA_FIN set, would map 10 octets from the subflow into data sequence space 80-89, the DATA_FIN is DSN 90; therefore, this segment including DATA_FIN would be acknowledged with a DATA_ACK of 91. Note that when the DATA_FIN is not attached to a TCP segment containing data, the Data Sequence Signal MUST have a subflow sequence number of 0, a Data-Level Length of 1, and the data sequence number that corresponds with the DATA_FIN itself. The checksum in this case will only cover the pseudo-header. A DATA_FIN has the semantics and behavior as a regular TCP FIN, but at the connection level. Notably, it is only DATA_ACKed once all data has been successfully received at the connection level. Note, therefore, that a DATA_FIN is decoupled from a subflow FIN. It is only permissible to combine these signals on one subflow if there is no data outstanding on other subflows. Otherwise, it may be necessary to retransmit data on different subflows. Essentially, a host MUST NOT close all functioning subflows unless it is safe to do
so, i.e., until all outstanding data has been DATA_ACKed, or until the segment with the DATA_FIN flag set is the only outstanding segment. Once a DATA_FIN has been acknowledged, all remaining subflows MUST be closed with standard FIN exchanges. Both hosts SHOULD send FINs on all subflows, as a courtesy to allow middleboxes to clean up state even if an individual subflow has failed. It is also encouraged to reduce the timeouts (Maximum Segment Life) on subflows at end hosts. In particular, any subflows where there is still outstanding data queued (which has been retransmitted on other subflows in order to get the DATA_FIN acknowledged) MAY be closed with a RST. A connection is considered closed once both hosts' DATA_FINs have been acknowledged by DATA_ACKs. As specified above, a standard TCP FIN on an individual subflow only shuts down the subflow on which it was sent. If all subflows have been closed with a FIN exchange, but no DATA_FIN has been received and acknowledged, the MPTCP connection is treated as closed only after a timeout. This implies that an implementation will have TIME_WAIT states at both the subflow and connection levels (see Appendix C). This permits "break-before-make" scenarios where connectivity is lost on all subflows before a new one can be re- established.3.3.4. 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 (i.e., DATA_ACK + receive window). The receiver will use the data sequence number to tell if a packet should be accepted at the connection level. When deciding to accept packets at subflow level, regular TCP checks the sequence number in the packet against the allowed receive window. With multipath, such a check is done using only the connection-level
window. A sanity check SHOULD be performed at subflow level to ensure that the subflow and mapped sequence numbers meet the following test: SSN - SUBFLOW_ACK <= DSN - DATA_ACK, where SSN is the subflow sequence number of the received packet and SUBFLOW_ACK is the RCV.NXT (next expected sequence number) of the subflow (with the equivalent connection-level definitions for DSN and DATA_ACK). In regular TCP, once a segment is deemed in-window, it is put either in the in-order receive queue or in the out-of-order queue. In Multipath TCP, the same happens but at the connection level: a segment is placed in the connection level in-order or out-of-order queue if it is in-window at both connection and subflow levels. The stack still has to remember, for each subflow, which segments were received successfully so that it can ACK them at subflow level appropriately. Typically, this will be implemented by keeping per subflow out-of-order queues (containing only message headers, not the payloads) and remembering the value of the cumulative ACK. 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 one of the paths. However, this might be insufficient when a packet is lost on a slower subflow and needs to be retransmitted (see Section 3.3.6). A tight upper bound would be the maximum round-trip time (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.3.3.5. 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, on the receipt of a DATA_ACK. This is important to allow using paths with different RTTs, and thus different feedback loops. MPTCP uses a single receive window across all subflows, and if the receive window was guaranteed to be unchanged end-to-end, a host could always read the most recent receive window value. However, 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 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 reduce the right edge of the window. The sender MUST also remember 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 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 send buffer MUST, at a minimum, be as big as the receive buffer, to enable the sender to reach maximum throughput.3.3.6. Reliability and Retransmissions
The data sequence mapping allows senders to resend data with the same data sequence number on a different subflow. When doing this, a host 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 sensible behavior. Optimizations 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 3.3.8). 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 is set when the head of the connection-level is ACKed at subflow level but its corresponding data is not ACKed at data level. This timer will guard against failures in retransmission by middleboxes that proactively ACK data. The sender MUST keep data in its send buffer as long as the data has not been acknowledged at both connection level and on all subflows on which it has been sent. 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 after a timeout, 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 (which MAY be lower than the usual TCP limits of the Maximum Segment Life), or on the receipt of an ICMP error, and only then delete the outstanding data segments. Multiple retransmissions are triggers that will indicate that a subflow performs badly and could lead to a host resetting the subflow with a RST. However, additional research is required to understand the heuristics of how and when to reset underperforming subflows. For example, a highly asymmetric path may be misdiagnosed as underperforming.3.3.7. 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 [5]; the algorithm does not achieve perfect resource pooling but is "safe" in that it is readily deployable in the current Internet. By this, we mean that it does not take up more capacity on any one path than if it was a single path flow using only that route, so this ensures fair coexistence with single-path TCP at shared bottlenecks. 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, as well as those for achieving different properties in quality of service, reliability, and resilience.
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 were lost and when.3.3.8. 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 maximize throughput, all available paths will be used simultaneously for data transfer, using coupled congestion control as described in [5]. It is expected, however, that other use cases will appear. 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 (of delay or bandwidth) is more important than throughput. Application requirements such as these are discussed in detail in [6]. 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 Explicit Congestion Notification (ECN) signal [17], which is undesirable, and it is felt that there would not be sufficient benefit to justify an entirely new signal. Therefore, the MP_JOIN option (see Section 3.2) contains 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 (e.g., a subflow that was previously set as backup should now take priority
over all remaining subflows). Therefore, the MP_PRIO option, shown in Figure 11, can be used to change the 'B' flag of the subflow on which it is sent. 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 | Length |Subtype| |B| AddrID (opt) | +---------------+---------------+-------+-----+-+--------------+ Figure 11: Change Subflow Priority (MP_PRIO) Option It should be noted that the backup flag is a request from a data receiver to a data sender only, and the data sender SHOULD adhere to these requests. A host cannot assume that the data sender will do so, however, since local policies -- or technical difficulties -- may override MP_PRIO requests. Note also that this signal applies to a single direction, and so the sender of this option could choose to continue using the subflow to send data even if it has signaled B=1 to the other host. This option can also be applied to other subflows than the one on which it is sent, by setting the optional Address ID field. This applies the given setting of B to all subflows in this connection that use the address identified by the given Address ID. The presence of this field is determined by the option length; if Length==4 then it is present. If Length==3, then it applies to the current subflow only. The use case of this is that a host can signal to its peer that an address is temporarily unavailable (for example, if it has radio coverage issues) and the peer should therefore drop to backup state on all subflows using that Address ID.
(page 35 continued on part 3)
