BEP:10
Title:Extension Protocol
Version: 023256c7581a4bed356e47caf8632be2834211bd
Last-Modified:Thu Jan 12 12:29:12 2017 -0800
Author: Arvid Norberg <arvid@bittorrent.com>, Ludvig Strigeus <bittorrent@strigeus.com>, Greg Hazel <greg@bittorrent.com>
Status: Accepted
Type:Standards Track
Created:31-Jan-2008
Post-History:

The intention of this protocol is to provide a simple and thin transport for extensions to the bittorrent protocol. Supporting this protocol makes it easy to add new extensions without interfering with the standard bittorrent protocol or clients that don't support this extension or the one you want to add.

To advertise to other clients that you support, one bit from the reserved bytes is used.

The bit selected for the extension protocol is bit 20 from the right (counting starts at 0). So (reserved_byte[5] & 0x10) is the expression to use for checking if the client supports extended messaging.

Once support for the protocol is established, the client is supposed to support 1 new message:

name id
extended 20

This message is sent as any other bittorrent message, with a 4 byte length prefix and a single byte identifying the message (the single byte being 20 in this case). At the start of the payload of the message, is a single byte message identifier. This identifier can refer to different extension messages and only one ID is specified, 0. If the ID is 0, the message is a handshake message which is described below. The layout of a general extended message follows (including the message headers used by the bittorrent protocol):

size description
uint32_t length prefix. Specifies the number of bytes for the entire message. (Big endian)
uint8_t bittorrent message ID, = 20
uint8_t extended message ID. 0 = handshake, >0 = extended message as specified by the handshake.

handshake message

The payload of the handshake message is a bencoded dictionary. All items in the dictionary are optional. Any unknown names should be ignored by the client. All parts of the dictionary are case sensitive. This is the defined item in the dictionary:

name description
m

Dictionary of supported extension messages which maps names of extensions to an extended message ID for each extension message. The only requirement on these IDs is that no extension message share the same one. Setting an extension number to zero means that the extension is not supported/disabled. The client should ignore any extension names it doesn't recognize.

The extension message IDs are the IDs used to send the extension messages to the peer sending this handshake. i.e. The IDs are local to this particular peer.

Here are some other items that an implementation may choose to support:

name description
p Local TCP listen port. Allows each side to learn about the TCP port number of the other side. Note that there is no need for the receiving side of the connection to send this extension message, since its port number is already known.
v Client name and version (as a utf-8 string). This is a much more reliable way of identifying the client than relying on the peer id encoding.
yourip A string containing the compact representation of the ip address this peer sees you as. i.e. this is the receiver's external ip address (no port is included). This may be either an IPv4 (4 bytes) or an IPv6 (16 bytes) address.
ipv6 If this peer has an IPv6 interface, this is the compact representation of that address (16 bytes). The client may prefer to connect back via the IPv6 address.
ipv4 If this peer has an IPv4 interface, this is the compact representation of that address (4 bytes). The client may prefer to connect back via this interface.
reqq An integer, the number of outstanding request messages this client supports without dropping any. The default in in libtorrent is 250.

The handshake dictionary could also include extended handshake information, such as support for encrypted headers or anything imaginable.

An example of what the payload of a handshake message could look like:

Dictionary
m
Dictionary
LT_metadata 1
ut_pex 2
p 6881
v "µTorrent 1.2"

and in the encoded form:

d1:md11:LT_metadatai1e6:µT_PEXi2ee1:pi6881e1:v13:\xc2\xb5Torrent 1.2e

To make sure the extension names do not collide by mistake, they should be prefixed with the two (or one) character code that is used to identify the client that introduced the extension. This applies for both the names of extension messages, and for any additional information put inside the top-level dictionary. All one and two byte identifiers are invalid to use unless defined by this specification.

This message should be sent immediately after the standard bittorrent handshake to any peer that supports this extension protocol. It is valid to send the handshake message more than once during the lifetime of a connection, the sending client should not be disconnected. An implementation may choose to ignore the subsequent handshake messages (or parts of them).

Subsequent handshake messages can be used to enable/disable extensions without restarting the connection. If a peer supports changing extensions at run time, it should note that the m dictionary is additive. It's enough that it contains the actual CHANGES to the extension list. To disable the support for LT_metadata at run-time, without affecting any other extensions, this message should be sent: d11:LT_metadatai0ee. As specified above, the value 0 is used to turn off an extension.

The extension IDs must be stored for every peer, becuase every peer may have different IDs for the same extension.

This specification, deliberately, does not specify any extensions such as peer-exchange or metadata exchange. This protocol is merely a transport for the actual extensions to the bittorrent protocol and the extensions named in the example above (such as p) are just examples of possible extensions.

rationale

The reason why the extension messages' IDs would be defined in the handshake is to avoid having a global registry of message IDs. Instead the names of the extension messages requires unique names, which is much easier to do without a global registry. The convention is to use a two letter prefix on the extension message names, the prefix would identify the client first implementing the extension message. e.g. LT_metadata is implemented by libtorrent, and hence it has the LT prefix.

If the client supporting the extensions can decide which numbers the messages it receives will have, it means they are constants within that client. i.e. they can be used in switch statements. It's easy for the other end to store an array with the ID's we expect for each message and use that for lookups each time it sends an extension message.

The reason for having a dictionary instead of having an array (using implicitly assigned index numbers to the extensions) is that if a client want to disable some extensions, the ID numbers would change, and it wouldn't be able to use constants (and hence, not use them in a switch). If the messages IDs would map directly to bittorrent message IDs, It would also make it possible to map extensions in the handshake to existing extensions with fixed message IDs.

The reasoning behind having a single byte as extended message identifier is to follow the the bittorrent spec. with its single byte message identifiers. It is also considered to be enough. It won't limit the total number of extensions, only the number of extensions used simultaneously.

The reason for using single byte identifiers for the standardized handshake identifiers is 1) The mainline DHT uses single byte identifiers. 2) Saves bandwidth. The only advantage of longer messages is that it makes the protocol more readable for a human, but the BT protocol wasn't designed to be a human readable protocol, so why bother.