Delay-Tolerant Networking TCP Convergence Layer Protocol Version 4

Delay-Tolerant Networking TCP Convergence Layer Protocol Version 4 RKF Engineering Solutions, LLC

1229 19th Street NW Wasington DC 20036 US BSipos@rkf-eng.com

University of California, Berkeley

Computer Science Division 445 Soda Hall Berkeley CA 94720-1776 US demmer@cs.berkeley.edu

Aalto University

Department of Communications and Networking PO Box 13000 Aalto 02015 Finland jo@netlab.tkk.fi

Quebec QC Canada simon@per.reau.lt

Transport Delay Tolerant Networking This document describes a revised protocol for the TCP-based convergence layer for Delay-Tolerant Networking (DTN). The protocol revision is based on implementation issues in the original and updates to the Bundle Protocol contents, encodings, and convergence layer requirements in . The majority of this specification is unchanged from TCPCL version 3.

This document describes the TCP-based convergence-layer protocol for Delay-Tolerant Networking. Delay-Tolerant Networking is an end-to- end architecture providing communications in and/or through highly stressed environments, including those with intermittent connectivity, long and/or variable delays, and high bit error rates. More detailed descriptions of the rationale and capabilities of these networks can be found in "Delay-Tolerant Network Architecture" . An important goal of the DTN architecture is to accommodate a wide range of networking technologies and environments. The protocol used for DTN communications is the revsided Bundle Protocol (BP) , an application-layer protocol that is used to construct a store-and- forward overlay network. As described in the Bundle Protocol specification , it requires the services of a "convergence- layer adapter" (CLA) to send and receive bundles using the service of some "native" link, network, or Internet protocol. This document describes one such convergence-layer adapter that uses the well-known Transmission Control Protocol (TCP). This convergence layer is referred to as TCPCL. The locations of the TCPCL and the BP in the Internet model protocol stack are shown in Figure 1. In particular, when BP is using TCP as its bearer with TCPCL as its convergence layer, both BP and TCPCL reside at the application layer of the Internet model.

+-------------------------+ | DTN Application | -\ +-------------------------| | | Bundle Protocol (BP) | -> Application Layer +-------------------------+ | | TCP Conv. Layer (TCPCL) | -/ +-------------------------+ | TLS (optional) | ---> Presentation Layer +-------------------------+ | TCP | ---> Transport Layer +-------------------------+ | IP | ---> Network Layer +-------------------------+ | Link-Layer Protocol | ---> Link Layer +-------------------------+ | Physical Medium | ---> Physical Layer +-------------------------+ This document describes the format of the protocol data units passed between entities participating in TCPCL communications. This document does not address: The format of protocol data units of the Bundle Protocol, as those are defined elsewhere in and . This includes the concept of bundle fragmentation or bundle encapsulation. The TCPCL transfers bundles as opaque data blocks. Mechanisms for locating or identifying other bundle nodes within an internet.

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 .

This section contains definitions that are interpreted to be specific to the operation of the TCPCL protocol, as described below. A TCP connection refers to a transport connection using TCP as the transport protocol. A TCPCL session (as opposed to a TCP connection) is a TCPCL communication relationship between two bundle nodes. The lifetime of a TCPCL session is bound to the lifetime of an underlying TCP connection. Therefore, a TCPCL session is initiated when a bundle node initiates a TCP connection to be established for the purposes of bundle communication. A TCPCL session is terminated when the TCP connection ends, due either to one or both nodes actively terminating the TCP connection or due to network errors causing a failure of the TCP connection. For the remainder of this document, the term "session" without the prefix "TCPCL" refer to a TCPCL session. The session parameters are a set of values used to affect the operation of the TCPCL for a given session. The manner in which these parameters are conveyed to the bundle node and thereby to the TCPCL is implementation dependent. However, the mechanism by which two bundle nodes exchange and negotiate the values to be used for a given session is described in . Transmission refers to the procedures and mechanisms (described below) for conveyance of a bundle from one node to another.

The service of this protocol is the transmission of DTN bundles over TCP. This document specifies the encapsulation of bundles, procedures for TCP setup and teardown, and a set of messages and node requirements. The general operation of the protocol is as follows. First, one node establishes a TCPCL session to the other by initiating a TCP connection. After setup of the TCP connection is complete, an initial contact header is exchanged in both directions to set parameters of the TCPCL session and exchange a singleton endpoint identifier for each node (not the singleton Endpoint Identifier (EID) of any application running on the node) to denote the bundle-layer identity of each DTN node. This is used to assist in routing and forwarding messages, e.g., to prevent loops. Once the TCPCL session is established and configured in this way, bundles can be transmitted in either direction. Each bundle is transmitted in one or more logical segments of formatted bundle data. Each logical data segment consists of a DATA_SEGMENT message header, a count of the length of the segment, and finally the octet range of the bundle data. The choice of the length to use for segments is an implementation matter. The first segment for a bundle MUST set the 'start' flag, and the last one MUST set the 'end' flag in the DATA_SEGMENT message header. If multiple bundles are transmitted on a single TCPCL connection, they MUST be transmitted consecutively. Interleaving data segments from different bundles is not allowed. Bundle interleaving can be accomplished by fragmentation at the BP layer or by establishing multiple TCPCL sessions. A feature of this protocol is for the receiving node to send acknowledgments as bundle data segments arrive (ACK_SEGMENT). The rationale behind these acknowledgments is to enable the sender node to determine how much of the bundle has been received, so that in case the session is interrupted, it can perform reactive fragmentation to avoid re-sending the already transmitted part of the bundle. For each data segment that is received, the receiving node sends an ACK_SEGMENT code followed by an count containing the cumulative length of the bundle that has been received. The sending node MAY transmit multiple DATA_SEGMENT messages without necessarily waiting for the corresponding ACK_SEGMENT responses. This enables pipelining of messages on a channel. In addition, there is no explicit flow control on the TCPCL layer. Another feature is that a receiver MAY interrupt the transmission of a bundle at any point in time by replying with a REFUSE_BUNDLE message, which causes the sender to stop transmission of the current bundle, after completing transmission of a partially sent data segment. Note: This enables a cross-layer optimization in that it allows a receiver that detects that it already has received a certain bundle to interrupt transmission as early as possible and thus save transmission capacity for other bundles. For sessions that are idle, a KEEPALIVE message is sent at a negotiated interval. This is used to convey liveness information. Finally, before sessions close, a SHUTDOWN message is sent to the session peer. After sending a SHUTDOWN message, the sender of this message MAY send further acknowledgments (ACK_SEGMENT or REFUSE_BUNDLE) but no further data messages (DATA_SEGMENT). A SHUTDOWN message MAY also be used to refuse a session setup by a peer.

There are specific messages for sending and receiving operations (in addition to session setup/teardown). TCPCL is symmetric, i.e., both sides can start sending data segments in a session, and one side's bundle transfer does not have to complete before the other side can start sending data segments on its own. Hence, the protocol allows for a bi-directional mode of communication. Note that in the case of concurrent bidirectional transmission, acknowledgment segments MAY be interleaved with data segments.

The following figure visually depicts the protocol exchange for a simple session, showing the session establishment and the transmission of a single bundle split into three data segments (of lengths L1, L2, and L3) from Node A to Node B. Note that the sending node MAY transmit multiple DATA_SEGMENT messages without necessarily waiting for the corresponding ACK_SEGMENT responses. This enables pipelining of messages on a channel. Although this example only demonstrates a single bundle transmission, it is also possible to pipeline multiple DATA_SEGMENT messages for different bundles without necessarily waiting for ACK_SEGMENT messages to be returned for each one. However, interleaving data segments from different bundles is not allowed. No errors or rejections are shown in this example.

Node A Node B ====== ====== +-------------------------+ +-------------------------+ | Contact Header | -> <- | Contact Header | +-------------------------+ +-------------------------+ +-------------------------+ | DATA_SEGMENT (start) | -> | Transfer ID [I1] | -> | Length [L1] | -> | Bundle Data 0..(L1-1) | -> +-------------------------+ +-------------------------+ +-------------------------+ | DATA_SEGMENT | -> <- | ACK_SEGMENT | | Transfer ID [I1] | -> <- | Transfer ID [I1] | | Length [L2] | -> <- | Length [L1] | |Bundle Data L1..(L1+L2-1)| -> +-------------------------+ +-------------------------+ +-------------------------+ +-------------------------+ | DATA_SEGMENT (end) | -> <- | ACK_SEGMENT | | Transfer ID [I1] | -> <- | Transfer ID [I1] | | Length [L3] | -> <- | Length [L1+L2] | |Bundle Data | -> +-------------------------+ | (L1+L2)..(L1+L2+L3-1)| +-------------------------+ +-------------------------+ <- | ACK_SEGMENT | <- | Transfer ID [I1] | <- | Length [L1+L2+L3] | +-------------------------+ +-------------------------+ +-------------------------+ | SHUTDOWN | -> <- | SHUTDOWN | +-------------------------+ +-------------------------+

For bundle transmissions to occur using the TCPCL, a TCPCL session MUST first be established between communicating nodes. It is up to the implementation to decide how and when session setup is triggered. For example, some sessions MAY be opened proactively and maintained for as long as is possible given the network conditions, while other sessions MAY be opened only when there is a bundle that is queued for transmission and the routing algorithm selects a certain next-hop node. To establish a TCPCL session, a node MUST first establish a TCP connection with the intended peer node, typically by using the services provided by the operating system. Port number 4556 has been assigned by IANA as the well-known port number for the TCP convergence layer. Other port numbers MAY be used per local configuration. Determining a peer's port number (if different from the well-known TCPCL port) is up to the implementation. If the node is unable to establish a TCP connection for any reason, then it is an implementation matter to determine how to handle the connection failure. A node MAY decide to re-attempt to establish the connection. If it does so, it MUST NOT overwhelm its target with repeated connection attempts. Therefore, the node MUST retry the connection setup only after some delay (a 1-second minimum is RECOMMENDED), and it SHOULD use a (binary) exponential backoff mechanism to increase this delay in case of repeated failures. In case a SHUTDOWN message specifying a reconnection delay is received, that delay is used as the initial delay. The default initial delay SHOULD be at least 1 second but SHOULD be configurable since it will be application and network type dependent. The node MAY declare failure after one or more connection attempts and MAY attempt to find an alternate route for bundle data. Such decisions are up to the higher layer (i.e., the BP). Once a TCP connection is established, each node MUST immediately transmit a contact header over the TCP connection. The format of the contact header is described in . Upon receipt of the contact header, both nodes perform the validation and negotiation procedures defined in After receiving the contact header from the other node, either node MAY also refuse the session by sending a SHUTDOWN message. If session setup is refused, a reason MUST be included in the SHUTDOWN message.

Once a TCP connection is established, both parties exchange a contact header. This section describes the format of the contact header and the meaning of its fields. The format for the Contact Header is as follows:

1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 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 +---------------+---------------+---------------+---------------+ | magic='dtn!' | +---------------+---------------+---------------+---------------+ | Version | Flags | Keepalive Interval | +---------------+---------------+---------------+---------------+ | Segment MRU... | +---------------+---------------+---------------+---------------+ | contd. | +---------------+---------------+---------------+---------------+ | Transfer MRU... | +---------------+---------------+---------------+---------------+ | contd. | +---------------+---------------+---------------+---------------+ | EID Length | EID Data... | +---------------+---------------+---------------+---------------+ | contd. | +---------------+---------------+---------------+---------------+ The fields of the contact header are: A four-octet field that always contains the octet sequence 0x64 0x74 0x6e 0x21, i.e., the text string "dtn!" in US-ASCII (and UTF-8). A one-octet field value containing the value 4 (current version of the protocol). A one-octet field of single-bit flags, interpreted according to the descriptions in . A 16-bit unsigned integer indicating the longest allowable interval in seconds between KEEPALIVE messages received in this session. A 64-bit unsigned integer indicating the largest allowable single-segment data payload size to be received in this session. Any DATA_SEGMENT sent to this peer SHALL have a data payload no longer than the peer's Segment MRU. The two endpoints of a single session MAY have different Segment MRUs, and no relation between the two is required. A 64-bit unsigned integer indicating the largest allowable total-bundle data size to be received in this session. Any bundle transfer sent to this peer SHALL have a Total bundle data payload no longer than the peer's Transfer MRU. This value can be used to perform proactive bundle fragmentation. The two endpoints of a single session MAY have different Transfer MRUs, and no relation between the two is required. Together these fields represent a variable-length text string. The EID Length is a 16-bit unsigned integer indicating the number of octets of EID Data to follow. A zero EID Length is a special case which indicates the lack of EID rather than a truly empty EID. A non-zero-length EID Data contains the UTF-8 encoded EID of some singleton endpoint in which the sending node is a member, in the canonical format of <scheme name>:<scheme-specific part>. TypeCodeDescription CAN_TLS0x01If bit is set, indicates that the sending peer is capable of TLS security.

Upon reception of the contact header, each node follows the following procedures to ensure the validity of the TCPCL session and to negotiate values for the session parameters. If the magic string is not present or is not valid, the connection MUST be terminated. The intent of the magic string is to provide some protection against an inadvertent TCP connection by a different protocol than the one described in this document. To prevent a flood of repeated connections from a misconfigured application, a node MAY elect to hold an invalid connection open and idle for some time before closing it. If a node receives a contact header containing a version that is greater than the current version of the protocol that the node implements, then the node SHALL shutdown the session with a reason code of "Version mismatch". If a node receives a contact header with a version that is lower than the version of the protocol that the node implements, the node MAY either terminate the session (with a reason code of "Version mismatch"). Otherwise, the node MAY adapt its operation to conform to the older version of the protocol. This decision is an implementation matter. A node calculates the parameters for a TCPCL session by negotiating the values from its own preferences (conveyed by the contact header it sent to the peer) with the preferences of the peer node (expressed in the contact header that it received from the peer). The negotatiated parameters defined by this specification are described in the following paragraphs. Negotiation of the Session Keepalive parameter is performed by taking the minimum of this two contact headers' Keepalive Interval. If the negotiated Session Keepalve is zero (i.e. one or both contact headers contains a zero Keepalive Interval), then the keepalive feature (described in ) is disabled. Negotiation of the Enable TLS parameter is performed by taking the logical AND of the two contact headers' CAN_TLS flags. If the negotiated Enable TLS value is true then TLS negotiation feature (described in ) begins immediately following the contact header exchange. Once this process of parameter negotiation is completed, the protocol defines no additional mechanism to change the parameters of an established session; to effect such a change, the session MUST be terminated and a new session established.

This section describes the protocol operation for the duration of an established session, including the mechanisms for transmitting bundles over the session.

After the initial exchange of a contact header, all messages transmitted over the session are identified by a one-octet header with the following structure:

0 1 2 3 4 5 6 7 +-+-+-+-+-+-+-+-+ | type | flags | +-+-+-+-+-+-+-+-+ type: Indicates the type of the message as per below. flags: Optional flags defined based on message type. The types and values for the message type code are as follows. TypeCodeDescription DATA_SEGMENT0x1Indicates the transmission of a segment of bundle data, as described in . ACK_SEGMENT0x2Acknowledges reception of a data segment, as described in . REFUSE_BUNDLE0x3Indicates that the transmission of the current bundle SHALL be stopped, as described in . KEEPALIVE0x4KEEPALIVE message for the session, as described in . SHUTDOWN0x5Indicates that one of the nodes participating in the session wishes to cleanly terminate the session, as described in . LENGTH0x6Contains the length (in octets) of the next bundle, as described in .

The protocol includes a provision for transmission of KEEPALIVE messages over the TCPCL session to help determine if the underlying TCP connection has been disrupted. As described in , one of the parameters in the contact header is the keepalive_interval. Both sides populate this field with their requested intervals (in seconds) between KEEPALIVE messages. The format of a KEEPALIVE message is a one-octet message type code of KEEPALIVE (as described in Table 2) with no additional data. Both sides SHOULD send a KEEPALIVE message whenever the negotiated interval has elapsed with no transmission of any message (KEEPALIVE or other). If no message (KEEPALIVE or other) has been received for at least twice the keepalive_interval, then either party MAY terminate the session by transmitting a one-octet SHUTDOWN message (as described in Table 2) and by closing the session. Note: The keepalive_interval SHOULD not be chosen too short as TCP retransmissions MAY occur in case of packet loss. Those will have to be triggered by a timeout (TCP retransmission timeout (RTO)), which is dependent on the measured RTT for the TCP connection so that KEEPALIVE messages MAY experience noticeable latency.

If a TCPCL endpoint receives a message which is uknown to it (possibly due to an unhandled protocol mismatch) or is inappropriate for the current session state (e.g. a KEEPALIVE or LENGTH message received after feature negotation has disabled those features), there is a protocol-level message to signal this condition in the form of a REJECT reply. The format of a REJECT message follows:

+-----------------------------+ | Message Header | +-----------------------------+ | Reason Code (U8) | +-----------------------------+ | Rejected Message Header | +-----------------------------+ The Rejected Message Header is a copy of the Message Header to which the REJECT message is sent as a response. The REJECT Reason Code is an 8-bit unsigned integer and indicates why the REJECT itself was sent. The specified values of the reason code are: NameCodeDescription Message Type Unknown0x01A message was received with a Message Type code unknown to the TCPCL endpoint. Message Unsupported0x02A message was received but the TCPCL endpoint cannot comply with the message contents. Message Unexpected0x03A message was received while the session is in a state in which the message is not expected.

This version of the TCPCL supports establishing a session-level Transport Layer Security (TLS) session within an existing TCPCL session. When TLS is used within the TCPCL it affects the entire session. By convention, this protocol uses the endpoint which initiated the underlying TCP connection as the "client" role of the TLS handshake request. Once a TLS session is established within TCPCL, there is no mechanism provided to end the TLS session and downgrade the session. If a non-TLS session is desired after a TLS session is started then the entire TCPCL session MUST be shutdown first. After negotiating an Enable TLS parameter of true, and before any other TCPCL messages are sent within the session, the session endpoints SHALL begin a TLS handshake in accordance with . The parameters within each TLS negotation are implementation dependent but any TCPCL endpoint SHOULD follow all recommended best practices of .

If a TLS handshake cannot negotiate a TLS session, both endpoints of the TCPCL session SHALL cause a TCPCL shutdown with reason "TLS negotiation failed". Unless the TLS parameters change between two sequential handshakes, the subsequent handshake is likely to fail just as the earlier one. After a TLS session is successfuly established, both TCPCL endpoints SHALL re-exchange TCPCL Contact Header messages. Any information cached from the prior Contact Header exchange SHALL be discarded. This re-exchange avoids man-in-the-middle attack in identical fashon to .

A summary of a typical CAN_TLS usage is shown in the sequence below where the client/requester role is represented by the prefix "C" and the server/responder role is represented by the prefix "S". Unordered or "simultaneous" actions are shown as "C/S".

All of the message in this section are directly associated with tranfering a bundle between TCPCL endpoints.

Each of the bundle transfer messages contains a Transfer ID number which is used to correlate messages originating from sender and receiver of a bundle. The Transfer ID provides a similar behaivior to a datagram sequence number. A Transfer ID does not attempt to address uniqueness of the bundle data itself and has no relation to concepts such as bundle fragmentation. Transmitting the same bundle repeatedly, or fragments of the same bundle, or any other combination will result in a unique Transfer ID for each transmission sequence. Transfer IDs from each endpoint SHALL be unique within a single TCPCL session. The initial Transfer ID from each endpoint SHALL have value zero. Subsequent Transfer ID values SHALL be incremented from the prior Transfer ID value by one. Upon exhaustion of the entire 64-bit Transfer ID space, the sending endpoint SHALL terminate the session with SHUTDOWN reason code "Resource Exhaustion". For bidirectional bundle transfers, a TCPCL endpoint SHOULD NOT rely on any relation between Transfer IDs originating from each side of the TCPCL session.

The LENGTH message contains the total length, in octets, of the next bundle, formatted as a 64-bit unsigned integer. Its purpose is to allow nodes to preemptively refuse bundles that would exceed their resources or to prepare storage on the receiving node for the upcoming bundle data. The Total Bundle Length field within a LENGTH message SHALL be used as informative data by the receiver. If, for whatever reason, the actual total legnth of bundle data received differs from the value indicated by the LENGTH message, the receiver SHOULD accept the full set of bundle data as valid. The format of the LENGTH message is as follows:

+-----------------------------+ | Message Header | +-----------------------------+ | Transfer ID (U64) | +-----------------------------+ | Total bundle length (U64) | +-----------------------------+ LENGTH messages SHALL be sent immediately before transmission of any DATA_SEGMENT messages. LENGTH messages MUST NOT be sent unless the next DATA_SEGMENT message has the 'S' bit set to "1" (i.e., just before the start of a new bundle). A receiver MAY send a BUNDLE_REFUSE message as soon as it receives a LENGTH message without waiting for the next DATA_SEGMENT message. The sender MUST be prepared for this and MUST associate the refusal with the correct bundle via the Transfer ID fields. Upon reception of a LENGTH message not immediately before the start of a starting DATA_SEGMENT the reciever SHALL send a REJECT message with a Reason Code of "Message Unexpected".

Each bundle is transmitted in one or more data segments. The format of a DATA_SEGMENT message follows in and its use of header flags is shown in .

+------------------------------+ | Message Header | +------------------------------+ | Transfer ID (U64) | +------------------------------+ | Data length (U64) | +------------------------------+ | Data contents (octet string) | +------------------------------+

4 5 6 7 +-+-+-+-+ |0|0|S|E| +-+-+-+-+ The type portion of the message header contains the value 0x1. The flags portion of the message header octet contains two optional values in the two low-order bits, denoted 'S' and 'E' above. The 'S' bit MUST be set to one if it precedes the transmission of the first segment of a new bundle. The 'E' bit MUST be set to one when transmitting the last segment of a bundle. In the case where an entire transfer is accomplished in a single segment, both the 'S' and 'E' bits MUST be set to one. Following the message header, the length field is a 64-bit unsigned integer containing the number of octets of bundle data that are transmitted in this segment. Following this length is the actual data contents. Once a transmission of a bundle has commenced, the node MUST only send segments containing sequential portions of that bundle until it sends a segment with the 'E' bit set.

Although the TCP transport provides reliable transfer of data between transport peers, the typical BSD sockets interface provides no means to inform a sending application of when the receiving application has processed some amount of transmitted data. Thus, after transmitting some data, a Bundle Protocol agent needs an additional mechanism to determine whether the receiving agent has successfully received the segment. To this end, the TCPCL protocol provides feedback messaging whereby a receiving node transmits acknowledgments of reception of data segments. The format of an ACK_SEGMENT message follows in and its use of header flags is the same as for DATA_SEGMENT (shown in ). The flags of an ACK_SEGMENT message SHALL be identical to the flags of the DATA_SEGMENT message for which it is a reply.

+-----------------------------+ | Message Header | +-----------------------------+ | Transfer ID (U64) | +-----------------------------+ | Acknowledged length (U64) | +-----------------------------+ To transmit an acknowledgment, a node first transmits a message header with the ACK_SEGMENT type code and all flags set to zero, then transmits a 64-bit unsigned integer containing the cumulative length in octets of the received segment(s) of the current bundle. The length MUST fall on a segment boundary. That is, only full segments can be acknowledged. For example, suppose the sending node transmits four segments of bundle data with lengths 100, 200, 500, and 1000, respectively. After receiving the first segment, the node sends an acknowledgment of length 100. After the second segment is received, the node sends an acknowledgment of length 300. The third and fourth acknowledgments are of length 800 and 1800, respectively.

As bundles can be large, the TCPCL supports an optional mechanisms by which a receiving node MAY indicate to the sender that it does not want to receive the corresponding bundle. To do so, upon receiving a LENGTH or DATA_SEGMENT message, the node MAY transmit a REFUSE_BUNDLE message. As data segments and acknowledgments MAY cross on the wire, the bundle that is being refused SHALL be identified by the Transfer ID of the refusal. The format of the message is as follows:

+-----------------------------+ | Message Header | +-----------------------------+ | Transfer ID (U64) | +-----------------------------+

4 5 6 7 +-+-+-+-+ | RCode | +-+-+-+-+ The RCode field, which stands for "reason code", contains a value indicating why the bundle was refused. The following table contains semantics for some values. Other values MAY be registered with IANA, as defined in . NameRCodeSemantics Unknown0x0Reason for refusal is unknown or not specified. Completed0x1The receiver now has the complete bundle. The sender MAY now consider the bundle as completely received. No Resources0x2The receiver's resources are exhausted. The sender SHOULD apply reactive bundle fragmentation before retrying. Retransmit0x3The receiver has encountered a problem that requires the bundle to be retransmitted in its entirety. The receiver MUST, for each bundle preceding the one to be refused, have either acknowledged all DATA_SEGMENTs or refused the bundle. This allows the sender to identify the bundles accepted and refused by means of a simple FIFO list of segments and acknowledgments. The bundle refusal MAY be sent before the entire data segment is received. If a sender receives a REFUSE_BUNDLE message, the sender MUST complete the transmission of any partially sent DATA_SEGMENT message (so that the receiver stays in sync). The sender MUST NOT commence transmission of any further segments of the refused bundle subsequently. Note, however, that this requirement does not ensure that a node will not receive another DATA_SEGMENT for the same bundle after transmitting a REFUSE_BUNDLE message since messages MAY cross on the wire; if this happens, subsequent segments of the bundle SHOULD also be refused with a REFUSE_BUNDLE message. Note: If a bundle transmission is aborted in this way, the receiver MAY not receive a segment with the 'E' flag set to '1' for the aborted bundle. The beginning of the next bundle is identified by the 'S' bit set to '1', indicating the start of a new bundle.

This section describes the procedures for ending a TCPCL session.

To cleanly shut down a session, a SHUTDOWN message MUST be transmitted by either node at any point following complete transmission of any other message. A node SHOULD acknowledge all received data segments before sending a SHUTDOWN message to end the session. The format of the SHUTDOWN message is as follows:

+-----------------------------------+ | Message Header | +-----------------------------------+ | Reason Code (optional U8) | +-----------------------------------+ | Reconnection Delay (optional U16) | +-----------------------------------+

4 5 6 7 +-+-+-+-+ |0|0|R|D| +-+-+-+-+ It is possible for a node to convey additional information regarding the reason for session termination. To do so, the node MUST set the 'R' bit in the message header flags and transmit a one-octet reason code immediately following the message header. The specified values of the reason code are: NameCodeDescription Idle timeout0x00The session is being closed due to idleness. Version mismatch0x01The node cannot conform to the specified TCPCL protocol version. Busy0x02The node is too busy to handle the current session. Contact Failure0x03The node cannot interpret or negotiate contact header option. TLS failure0x04The node failed to negotiate TLS session and cannot continue the session. Resource Exhaustion0x05The node has run into some resoure limit and cannot continue the session. It is also possible to convey a requested reconnection delay to indicate how long the other node MUST wait before attempting session re-establishment. To do so, the node sets the 'D' bit in the message header flags and then transmits an 16-bit unsigned integer specifying the requested delay, in seconds, following the message header (and optionally, the SHUTDOWN reason code). The value 0 SHALL be interpreted as an infinite delay, i.e., that the connecting node MUST NOT re-establish the session. In contrast, if the node does not wish to request a delay, it SHOULD omit the reconnection delay field (and set the 'D' bit to zero). A session shutdown MAY occur immediately after TCP connection establishment or reception of a contact header (and prior to any further data exchange). This MAY, for example, be used to notify that the node is currently not able or willing to communicate. However, a node MUST always send the contact header to its peer before sending a SHUTDOWN message. If either node terminates a session prematurely in this manner, it SHOULD send a SHUTDOWN message and MUST indicate a reason code unless the incoming connection did not include the magic string. If a node does not want its peer to reopen a connection immediately, it SHOULD set the 'D' bit in the flags and include a reconnection delay to indicate when the peer is allowed to attempt another session setup. If a session is to be terminated before another protocol message has completed, then the node MUST NOT transmit the SHUTDOWN message but still SHOULD close the TCP connection. In particular, if the session is to be closed (for whatever reason) while a node is in the process of transmitting a bundle data segment, the receiving node is still expecting segment data and might erroneously interpret the SHUTDOWN message to be part of the data segment.

The protocol includes a provision for clean shutdown of idle sessions. Determining the length of time to wait before closing idle sessions, if they are to be closed at all, is an implementation and configuration matter. If there is a configured time to close idle links and if no bundle data (other than KEEPALIVE messages) has been received for at least that amount of time, then either node MAY terminate the session by transmitting a SHUTDOWN message indicating the reason code of 'Idle timeout' (as described in Table 4). After receiving a SHUTDOWN message in response, both sides MAY close the TCP connection.

One security consideration for this protocol relates to the fact that nodes present their endpoint identifier as part of the session header exchange. It would be possible for a node to fake this value and present the identity of a singleton endpoint in which the node is not a member, essentially masquerading as another DTN node. If this identifier is used outside of a TLS-secured session or without further verification as a means to determine which bundles are transmitted over the session, then the node that has falsified its identity MAY be able to obtain bundles that it SHOULD not have. Therefore, a node SHALL NOT use the endpoint identifier conveyed in the TCPCL session message to derive a peer node's identity unless it can corroborate it via other means. These concerns MAY be mitigated through the use of the Bundle Security Protocol . In particular, the Bundle Authentication Block defines mechanism for secure exchange of bundles between DTN nodes. Thus, an implementation could delay trusting the presented endpoint identifier until the node can securely validate that its peer is in fact the only member of the given singleton endpoint. TCPCL can be used to provide point-to-point transport security, but does not provide security of data-at-rest and does not guarantee end-to-end bundle security. The mechanisms defined in and are to be used instead. Even when using TLS to secure the TCPCL session, the actual ciphersuite negotiated between the TLS peers MAY be insecure. TLS can be used to perform authentication without data confidentiality, for example. It is up to security policies within each TCPCL node to ensure that the negotiated TLS ciphersuite meets transport security requirements. This is identical behavior to STARTTLS use in . Another consideration for this protocol relates to denial-of-service attacks. A node MAY send a large amount of data over a TCPCL session, requiring the receiving node to handle the data, attempt to stop the flood of data by sending a REFUSE_BUNDLE message, or forcibly terminate the session. This burden could cause denial of service on other, well-behaving sessions. There is also nothing to prevent a malicious node from continually establishing sessions and repeatedly trying to send copious amounts of bundle data. A listening node MAY take countermeasures such as ignoring TCP SYN messages, closing TCP connections as soon as they are established, waiting before sending the contact header, sending a SHUTDOWN message quickly or with a delay, etc.

In this section, registration procedures are as defined in

Port number 4556 has been previously assigned as the default port for the TCP convergence layer in . This assignment is unchanged by protocol version 4. ParameterValue Service Name:dtn-bundle Transport Protocol(s):TCP Assignee:Simon Perreault <simon@per.reau.lt> Contact:Simon Perreault <simon@per.reau.lt> Description:DTN Bundle TCP CL Protocol Reference: Port Number:4556

IANA has created, under the "Bundle Protocol" registry, a sub- registry titled "Bundle Protocol TCP Convergence-Layer Version Numbers" and initialized it with the following table. The registration procedure is RFC Required. ValueDescriptionReference 0Reserved 1Reserved 2Reserved 3TCPCL 4TCPCLbisThis specification. 5-255Unassigned

IANA has created, under the "Bundle Protocol" registry, a sub- registry titled "Bundle Protocol TCP Convergence-Layer Message Types" and initialized it with the contents below. The registration procedure is RFC Required. CodeMessage Type 0x0Reserved 0x1DATA_SEGMENT 0x2ACK_SEGMENT 0x3REFUSE_BUNDLE 0x4KEEPALIVE 0x5SHUTDOWN 0x6LENGTH 0x7REJECT 0x8STARTTLS 0x9--0xfUnassigned

IANA has created, under the "Bundle Protocol" registry, a sub- registry titled "Bundle Protocol TCP Convergence-Layer REFUSE_BUNDLE Reason Codes" and initialized it with the contents of Table 3. The registration procedure is RFC Required. CodeRefusal Reason 0x0Unknown 0x1Completed 0x2No Resources 0x3Retransmit 0x4--0x7Unassigned 0x8--0xfReserved for future usage

IANA has created, under the "Bundle Protocol" registry, a sub- registry titled "Bundle Protocol TCP Convergence-Layer SHUTDOWN Reason Codes" and initialized it with the contents of Table 4. The registration procedure is RFC Required. CodeShutdown Reason 0x00Idle timeout 0x01Version mismatch 0x02Busy 0x03Contact Failure 0x04TLS failure 0x05--0xFFUnassigned

EDITOR NOTE: sub-registry to-be-created upon publication of this specification. IANA will create, under the "Bundle Protocol" registry, a sub- registry titled "Bundle Protocol TCP Convergence-Layer REJECT Reason Codes" and initialized it with the contents of Table 4. The registration procedure is RFC Required. CodeRejection Reason 0x00reserved 0x01Message Type Unknown 0x02Message Unsupported 0x03Message Unexpected 0x04-0xFFUnassigned

This memo is based on comments on implementation of provided from Scott Burleigh.

Bundle Protocol registry IANA

The areas in which changes from have been made to existing messages are: Changed contact header content to limit number of negotated options. Added contact option to negotiate maximum segment size (per each direction). Added a bundle transfer identification number to all bundle-related messages (LENGTH, DATA_SEGMENT, ACK_SEGMENT, REFUSE_BUNDLE). Use flags in ACK_SEGMENT to mirror flags from DATA_SEGMENT. Removed all uses of SDNV fields and replaced with fixed-bit-length fields. The areas in which extensions from have been made as new messages and codes are: Added REJECT message to indicate an unknown or unhandled message was received. Added TLS session security mechanism. Added TLS failure SHUTDOWN reason code.