Artifact Content
Not logged in

Artifact 0831d91f3c8b4e2ecef71aebdbe5aae7f15a4fea:

title: Bilateral Transfer Protocol 2.0 (BTP/2.0)

draft: 6

Bilateral Transfer Protocol 2.0 (BTP/2.0)


This document describes version 2.0 of the Bilateral Transfer Protocol (BTP), a request/response protocol for bilateral WebSocket links between Interledger connectors.



When two Interledger connectors send ILPv4 packets over HTTP POST, they each need to act as an HTTP server at times. If one of the connectors runs behind a firewall, this may be impossible. Therefore, BTP uses WebSockets instead of HTTP. With WebSockets, only one of the connectors needs to be publicly addressable.

However, WebSockets don't provide a mechanism for relating responses to requests. BTP adds this missing request/response layer, between WebSockets and the request/response pairs exchanged by the connectors.


This document describes the flow and data format that BTP uses, but not sub-protocols. Sub-protocols include optional functionality like ledger metadata, balance, automated settlement, and dispute resolution. Some protocols are documented on the wiki page. They are carried in the protocol data of BTP packets.

The BTP packet format is described exactly in the BTP ASN.1 spec.



BTP is broken up into one different RPC requests, which can get two different responses. Every BTP packet follows a common structure:

+---------------+ 1 | Type (1) | +---------------+ 1 | Request ID | 2 | (2) | 3 | | 4 | | +---------------+ 1 | Length Prefix | 2 | (3) | +---------------+ | Packet- | | Specific | | Data | . (4) | . | . | | | +---------------+ 1 | Sub-Protocol | 2 | Count (5) | +---------------+ | Sub-Protocol | | Data | . (6) | . | . | | | +---------------+

  1. Type: A 1-byte value describing what type of BTP packet this is. The values are described below, in BTP Type IDs.

  2. Request ID: A random 4-byte value used to correlate requests and responses. This value MAY be sequential instead of random, but care must be taken so that duplicate IDs are never in-flight at the same time.

  3. Length Prefix: A 1-byte (if under 128) or 2-byte value, containing the combined length of the packet-specific data and protocol data sections.

  4. Packet-Specific Data: Fields specific to the type of BTP packet. Variable length.

  5. Sub-Protocol Count: Variable-length integer containing the number of sub-protocols carried by this packet.

  6. Sub-Protocol Data: A list of protocols, containing a string (the protocol's name), a 1-byte flag (containing the MIME type), and a length-prefixed octet string (containing the protocol's data). Exact description is below in Sub-Protocol Data Format.

BTP Type IDs

ID Type Request/Response
1 Response Response
2 Error Response
3 (not used)
4 (not used)
5 (not used)
6 Message Request
7 Transfer Request

Sub-Protocol Data Format

```asn1 ContentType ::= INTEGER { applicationOctetString (0), textPlainUtf8 (1), applicationJson (2) } (0..255)

ProtocolData ::= SEQUENCE OF SEQUENCE { protocolName IA5String, contentType ContentType, data OCTET STRING } ```


Before anything else, when a client connects to a server, it sends a special Message request. Its primary protocolData entry MUST have name 'auth', content type MIME_APPLICATION_OCTET_STREAM, and empty data, and among the secondary entries, there MUST be a UTF-8 <<<<<<< HEAD 'auth_token' entry, and there MAY be a UTF-8 'auth_username' entry. The further secondary protocol data entries of this Message request MAY also be used to send additional information to the server. In situations where no authentication is needed, the 'auth_token' data can be set to the empty string, but it cannot be omitted. In situations where the 'auth_token' acts as a bearer token, and the value of 'auth_username' can be derived from the value of 'auth_token' deterministically, the 'auth_username' entry can either

be omitted, or set to the empty string.

'auth_token' entry. The further secondary protocol data entries of this Message request MAY also be used to send additional information to the server. In situations where no authentication is needed, the 'auth_token' data can be set to the empty string, but it cannot be omitted. >>>>>>> 35e6dd7... docs: BTP/2.0

If the client sends any BTP call that is not a Message, or sends a Message call whose primary sub-protocol is not auth, the server should respond with an Error and close the connection.

The server responds with a Response or Error as appropriate. Again, the protocolData field there MAY be used to send additional information to the client. To be clear, the server responds with an Error if:

If the server sent an Error, it subsequently closes the connection. If the server sent a Response, the BTP connection is open, until either one of the parties closes it. At the BTP level, the client and server play identical roles.

If the client does not send an Auth packet within a reasonable time, the server optionally sends a Message informing the client that the authentication timed out, and then closes the connection. If the client did send an Auth packet, but got neither a Response nor an Error back from the server, the client closes the connection.

If the connection is ever dropped and reconnected then it must be re-authenticated.


In order to understand the different BTP calls, it is necessary to distinguish between the first ("primary") and subsequent ("secondary") sub-protocol entries. The primary sub-protocol entry defines what type of action or information is requested of the recipient of the message. The secondary sub-protocols should not request additional actions or information. If multiple actions or pieces of information are required, multiple separate Messages should be sent. The secondary sub-protocols should only modify the request made in the primary sub-protocol, or provide additional contextual data which can be consumed in a readonly way (without affecting the result).

For example, the primary sub-protocol entry of a Message might represent a quote request, while one additional secondary sub-protocol entry may be present, indicating this request was forwarded by a proxy.

Likewise, only the primary sub-protocol data in a Response indicates whether result of the request from the Message being responded to actually succeeded or not.

In Error calls, the distinction between primary and secondary sub-protocol entries is less strict.


BTP uses a simple RPC flow. A request-type BTP packet is sent, and a response-type BTP packet is sent in response with the same request ID. The request types are Message and Transfer, and the response types are Response and Error.

Because it would be too slow to atomically save all requestIds that are processed, they are not idempotent. It is the responsibility of the requestor to make sure they don't duplicate requestIds. The implementation should ensure that no two in-flight requests are sent out with the same requestId. The responder should always send back a response to a request with the same requestId.

There are also a couple of tricky cases to handle:

These behaviors are important for preventing accidental feedback loops. If an unexpected packet triggered an error, that error may be unexpected to the sender. The sender would reply with another unexpected error, causing an infinite loop. Unreadable packets must be ignored too. If an application got onto a BTP connection and spoke the wrong protocol, it would trigger an error from BTP. This might trigger an error from the application, and it would devolve into another infinite loop.


asn1 Message ::= SEQUENCE { protocolData ProtocolData }

Message is used for sending information to the peer. It contains no packet-specific data, only protocol data. ILP packets are attached under the protocol name ilp with content-type application/octet-stream.


Error ::= SEQUENCE { -- Standardized error code code IA5String (SIZE (3)), -- Corresponding error code name IA5String, -- Time of emission triggeredAt GeneralizedTime, -- Additional data data OCTET STRING (SIZE (0..8192)), -- protocolData ProtocolData }

Error is a response-type message, returned when an error occurs on the BTP level. It has packet-specific data which resembles the ILP Error format, but irrelevant fields have been taken off and new error codes have been written:

Error Codes

Errors marked with a T are temporary, and can be retried after a short (1-60s) wait. If a retry fails again with a temporary error, a BTP client SHOULD wait longer before trying again. Errors marked with F are final, and the same request MUST NOT be retried.

Code Name Description
T00 UnreachableError Temporary error, indicating that the connector cannot process this request at the moment. Try again later.
F00 NotAcceptedError Data were symantically invalid.
F01 InvalidFieldsError At least one field contained structurally invalid data, e.g. timestamp full of garbage characters.
F03 TransferNotFoundError The transferId included in the packet does not reference an existing transfer.
F04 InvalidFulfillmentError The fulfillment included in the packet does not match the transfer's condition.
F05 DuplicateIdError The transferId and method match a previous request, but other data do not.
F06 AlreadyRolledBackError The transfer cannot be fulfilled because it has already been rejected or expired.
F07 AlreadyFulfilledError The transfer cannot be rejected because it has already been fulfilled.
F08 InsufficientBalanceError The transfer cannot be prepared because there is not enough available liquidity.


asn1 Transfer ::= SEQUENCE { amount UInt64, -- protocolData ProtocolData }

Transfer is used to send proof of payment, payment channel claims, or other settlement information to the other connector. The amount should indicate the additional value of this settlement state (compared to the previous settlement state), in a unit that was agreed out-of-band.