lightning/doc/PLUGINS.md

Plugins

Plugins are a simple yet powerful way to extend the functionality provided by c-lightning. They are subprocesses that are started by the main lightningd daemon and can interact with lightningd in a variety of ways:

• Command line option passthrough allows plugins to register their own command line options that are exposed through lightningd so that only the main process needs to be configured.
• JSON-RPC command passthrough adds a way for plugins to add their own commands to the JSON-RPC interface.
• Event stream subscriptions provide plugins with a push-based notification mechanism about events from the lightningd.
• Hooks are a primitive that allows plugins to be notified about internal events in lightningd and alter its behavior or inject custom behaviors.

A plugin may be written in any language, and communicates with lightningd through the plugin's stdin and stdout. JSON-RPCv2 is used as protocol on top of the two streams, with the plugin acting as server and lightningd acting as client. The plugin file needs to be executable (e.g. use chmod a+x plugin_name)

A day in the life of a plugin

During startup of lightningd you can use the --plugin= option to register one or more plugins that should be started. In case you wish to start several plugins you have to use the --plugin= argument once for each plugin. An example call might look like:

lightningd --plugin=/path/to/plugin1 --plugin=path/to/plugin2

lightningd will write JSON-RPC requests to the plugin's stdin and will read replies from its stdout. To initialize the plugin two RPC methods are required:

• getmanifest asks the plugin for command line options and JSON-RPC commands that should be passed through
• init is called after the command line options have been parsed and passes them through with the real values (if specified). This is also the signal that lightningd's JSON-RPC over Unix Socket is now up and ready to receive incoming requests from the plugin.

Once those two methods were called lightningd will start passing through incoming JSON-RPC commands that were registered and the plugin may interact with lightningd using the JSON-RPC over Unix-Socket interface.

The getmanifest method

The getmanifest method is required for all plugins and will be called on startup without any params. It MUST return a JSON object similar to this example:

{
	"options": [
		{
			"name": "greeting",
			"type": "string",
			"default": "World",
			"description": "What name should I call you?"
		}
	],
	"rpcmethods": [
		{
			"name": "hello",
			"usage": "[name]",
			"description": "Returns a personalized greeting for {greeting} (set via options)."
		},
		{
			"name": "gettime",
			"usage": "",
			"description": "Returns the current time in {timezone}",
			"long_description": "Returns the current time in the timezone that is given as the only parameter.\nThis description may be quite long and is allowed to span multiple lines."
		}
	],
	"subscriptions": [
		"connect",
		"disconnect"
	],
	"hooks": [
		"openchannel",
		"htlc_accepted"
	],
	"dynamic": true
}

The options will be added to the list of command line options that lightningd accepts. The above will add a --greeting option with a default value of World and the specified description. Notice that currently string, (unsigned) integers, and bool options are supported.

The rpcmethods are methods that will be exposed via lightningd's JSON-RPC over Unix-Socket interface, just like the builtin commands. Any parameters given to the JSON-RPC calls will be passed through verbatim. Notice that the name, description and usage fields are mandatory, while the long_description can be omitted (it'll be set to description if it was not provided). usage should surround optional parameter names in [].

The dynamic indicates if the plugin can be managed after lightningd has been started. Critical plugins that should not be stop should set it to false.

Plugins are free to register any name for their rpcmethod as long as the name was not previously registered. This includes both built-in methods, such as help and getinfo, as well as methods registered by other plugins. If there is a conflict then lightningd will report an error and exit.

The init method

The init method is required so that lightningd can pass back the filled command line options and notify the plugin that lightningd is now ready to receive JSON-RPC commands. The params of the call are a simple JSON object containing the options:

{
	"options": {
		"greeting": "World"
	},
	"configuration": {
		 "lightning-dir": "/home/user/.lightning",
		 "rpc-file": "lightning-rpc",
		 "startup": true
	}
}

The plugin must respond to init calls, however the response can be arbitrary and will currently be discarded by lightningd. JSON-RPC commands were chosen over notifications in order not to force plugins to implement notifications which are not that well supported.

The startup field allows a plugin to detect if it was started at lightningd startup (true), or at runtime (false).

JSON-RPC passthrough

Plugins may register their own JSON-RPC methods that are exposed through the JSON-RPC provided by lightningd. This provides users with a single interface to interact with, while allowing the addition of custom methods without having to modify the daemon itself.

JSON-RPC methods are registered as part of the getmanifest result. Each registered method must provide a name and a description. An optional long_description may also be provided. This information is then added to the internal dispatch table, and used to return the help text when using lightning-cli help, and the methods can be called using the name.

For example the above getmanifest result will register two methods, called hello and gettime:

    ...
	"rpcmethods": [
		{
			"name": "hello",
			"usage": "[name]",
			"description": "Returns a personalized greeting for {greeting} (set via options)."
		},
		{
			"name": "gettime",
			"description": "Returns the current time in {timezone}",
			"usage": "",
			"long_description": "Returns the current time in the timezone that is given as the only parameter.\nThis description may be quite long and is allowed to span multiple lines."
		}
	],
	...

The RPC call will be passed through unmodified, with the exception of the JSON-RPC call id, which is internally remapped to a unique integer instead, in order to avoid collisions. When passing the result back the id field is restored to its original value.

Note that if your result for an RPC call includes "format-hint": "simple", then lightning-cli will default to printing your output in "human-readable" flat form.

Event notifications

Event notifications allow a plugin to subscribe to events in lightningd. lightningd will then send a push notification if an event matching the subscription occurred. A notification is defined in the JSON-RPC specification as an RPC call that does not include an id parameter:

A Notification is a Request object without an "id" member. A Request object that is a Notification signifies the Client's lack of interest in the corresponding Response object, and as such no Response object needs to be returned to the client. The Server MUST NOT reply to a Notification, including those that are within a batch request.

Notifications are not confirmable by definition, since they do not have a Response object to be returned. As such, the Client would not be aware of any errors (like e.g. "Invalid params","Internal error").

Plugins subscribe by returning an array of subscriptions as part of the getmanifest response. The result for the getmanifest call above for example subscribes to the two topics connect and disconnect. The topics that are currently defined and the corresponding payloads are listed below.

Notification Types

channel_opened

A notification for topic channel_opened is sent if a peer successfully funded a channel with us. It contains the peer id, the funding amount (in millisatoshis), the funding transaction id, and a boolean indicating if the funding transaction has been included into a block.

{
	"channel_opened": {
		"id": "03864ef025fde8fb587d989186ce6a4a186895ee44a926bfc370e2c366597a3f8f",
		"funding_satoshis": "100000000msat",
		"funding_txid": "4a5e1e4baab89f3a32518a88c31bc87f618f76673e2cc77ab2127b7afdeda33b",
		"funding_locked": false
	}
}

connect

A notification for topic connect is sent every time a new connection to a peer is established.

{
	"id": "02f6725f9c1c40333b67faea92fd211c183050f28df32cac3f9d69685fe9665432",
	"address": "1.2.3.4"
}

disconnect

A notification for topic disconnect is sent every time a connection to a peer was lost.

{
	"id": "02f6725f9c1c40333b67faea92fd211c183050f28df32cac3f9d69685fe9665432"
}

invoice_payment

A notification for topic invoice_payment is sent every time an invoie is paid.

{
	"invoice_payment": {
		"label": "unique-label-for-invoice",
		"preimage": "0000000000000000000000000000000000000000000000000000000000000000",
		"msat": "10000msat"
	}
}

warning

A notification for topic warning is sent every time a new BROKEN /UNUSUAL level(in plugins, we use error/warn) log generated, which means an unusual/borken thing happens, such as channel failed, message resolving failed...

{
	"warning": {
	"level": "warn",
	"time": "1559743608.565342521",
	"source": "lightningd(17652): 0821f80652fb840239df8dc99205792bba2e559a05469915804c08420230e23c7c chan #7854:",
	"log": "Peer permanent failure in CHANNELD_NORMAL: lightning_channeld: sent ERROR bad reestablish dataloss msg"
  }
}

1. level is warn or error: warn means something seems bad happened and it's under control, but we'd better check it; error means something extremely bad is out of control, and it may lead to crash;
2. time is the second since epoch;
3. source means where the event happened, it may have the following forms:
   <node_id> chan #<db_id_of_channel>:,lightningd(<lightningd_pid>):, plugin-<plugin_name>:, <daemon_name>(<daemon_pid>):, jsonrpc:, jcon fd <error_fd_to_jsonrpc>:, plugin-manager;
4. log is the context of the original log entry.

Hooks

Hooks allow a plugin to define custom behavior for lightningd without having to modify the c-lightning source code itself. A plugin declares that it'd like to consulted on what to do next for certain events in the daemon. A hook can then decide how lightningd should react to the given event.

Hooks and notifications sounds very similar, however there are a few key differences:

• Notifications are asynchronous, i.e., lightningd will send the notifications but not wait for the plugin to process them. Hooks on the other hand are synchronous, lightningd cannot finish processing the event until the plugin has returned.
• Any number of plugins can subscribe to a notification topic, however only one plugin may register for any hook topic at any point in time (we cannot disambiguate between multiple plugins returning contradictory results from a hook callback).

Hooks are considered to be an advanced feature due to the fact that lightningd relies on the plugin to tell it what to do next. Use them carefully, and make sure your plugins always return a valid response to any hook invocation.

Hook Types

peer_connected

This hook is called whenever a peer has connected and successfully completed the cryptographic handshake. The parameters have the following structure if there is a channel with the peer:

{
  "peer": {
	"id": "03864ef025fde8fb587d989186ce6a4a186895ee44a926bfc370e2c366597a3f8f",
	"addr": "34.239.230.56:9735",
	"globalfeatures": "",
	"localfeatures": ""
  }
}

The hook is sparse on purpose, since the plugin can use the JSON-RPC listpeers command to get additional details should they be required. The addr field shows the address that we are connected to ourselves, not the gossiped list of known addresses. In particular this means that the port for incoming connections is an ephemeral port, that may not be available for reconnections.

The returned result must contain a result member which is either the string disconnect or continue. If disconnect and there's a member error_message, that member is sent to the peer before disconnection.

db_write

This hook is called whenever a change is about to be committed to the database. It is currently extremely restricted:

1. a plugin registering for this hook should not perform anything that may cause a db operation in response (pretty much, anything but logging).
2. a plugin registering for this hook should not register for other hooks or commands, as these may become intermingled and break rule #1.
3. the hook will be called before your plugin is initialized!

{
  "writes": [ "PRAGMA foreign_keys = ON" ]
}

Any response but "true" will cause lightningd to error without committing to the database!

invoice_payment

This hook is called whenever a valid payment for an unpaid invoice has arrived.

{
  "payment": {
	"label": "unique-label-for-invoice",
	"preimage": "0000000000000000000000000000000000000000000000000000000000000000",
	"msat": "10000msat"
  }
}

The hook is sparse on purpose, since the plugin can use the JSON-RPC listinvoices command to get additional details about this invoice. It can return a non-zero failure_code field as defined for final nodes in BOLT 4, or otherwise an empty object to accept the payment.

openchannel

This hook is called whenever a remote peer tries to fund a channel to us, and it has passed basic sanity checks:

{
  "openchannel": {
	"id": "03864ef025fde8fb587d989186ce6a4a186895ee44a926bfc370e2c366597a3f8f",
	"funding_satoshis": "100000000msat",
	"push_msat": "0msat",
	"dust_limit_satoshis": "546000msat",
	"max_htlc_value_in_flight_msat": "18446744073709551615msat",
	"channel_reserve_satoshis": "1000000msat",
	"htlc_minimum_msat": "0msat",
	"feerate_per_kw": 7500,
	"to_self_delay": 5,
	"max_accepted_htlcs": 483,
	"channel_flags": 1
  }
}

There may be additional fields, including shutdown_scriptpubkey and a hex-string. You can see the definitions of these fields in BOLT 2's description of the open_channel message.

The returned result must contain a result member which is either the string reject or continue. If reject and there's a member error_message, that member is sent to the peer before disconnection.

htlc_accepted

The htlc_accepted hook is called whenever an incoming HTLC is accepted, and its result determines how lightningd should treat that HTLC.

The payload of the hook call has the following format:

{
  "onion": {
    "payload": "",
    "per_hop_v0": {
      "realm": "00",
      "short_channel_id": "1x2x3",
      "forward_amount": "42msat",
      "outgoing_cltv_value": 500014
    }
  },
  "next_onion": "[1365bytes of serialized onion]",
  "shared_secret": "0000000000000000000000000000000000000000000000000000000000000000",
  "htlc": {
    "amount": "43msat",
    "cltv_expiry": 500028,
    "cltv_expiry_relative": 10,
    "payment_hash": "0000000000000000000000000000000000000000000000000000000000000000"
  }
}

The per_hop_v0 will only be present if the per hop payload has format 0x00 as defined by the specification. If not present an object representing the type-length-vale (TLV) payload will be added (pending specification). For detailed information about each field please refer to BOLT 04 of the specification, the following is just a brief summary:

• onion.payload contains the unparsed payload that was sent to us from the sender of the payment.
• onion.per_hop_v0:
  • realm will always be 00 since that value determines that we are using the per_hop_v0 format.
  • short_channel_id determines the channel that the sender is hinting should be used next (set to 0x0x0 if we are the recipient of the payment).
  • forward_amount is the amount we should be forwarding to the next hop, and should match the incoming funds in case we are the recipient.
  • outgoing_cltv_value determines what the CLTV value for the HTLC that we forward to the next hop should be.
• next_onion is the fully processed onion that we should be sending to the next hop as part of the outgoing HTLC. Processed in this case means that we took the incoming onion, decrypted it, extracted the payload destined for us, and serialized the resulting onion again.
• shared_secret is the shared secret we used to decrypt the incoming onion. It is shared with the sender that constructed the onion.
• htlc:
  • amount is the amount that we received with the HTLC. This amount minus the forward_amount is the fee that will stay with us.
  • cltv_expiry determines when the HTLC reverts back to the sender. cltv_expiry minus outgoing_cltv_expiry should be equal or larger than our cltv_delta setting.
  • cltv_expiry_relative hints how much time we still have to claim the HTLC. It is the cltv_expiry minus the current blockheight and is passed along mainly to avoid the plugin having to look up the current blockheight.
  • payment_hash is the hash whose payment_preimage will unlock the funds and allow us to claim the HTLC.

The hook response must have one of the following formats:

{
  "result": "continue"
}

This means that the plugin does not want to do anything special and lightningd should continue processing it normally, i.e., resolve the payment if we're the recipient, or attempt to forward it otherwise. Notice that the usual checks such as sufficient fees and CLTV deltas are still enforced.

{
  "result": "fail",
  "failure_code": 4301
}

fail will tell lightningd to fail the HTLC with a given numeric failure_code (please refer to the spec for details).

{
  "result": "resolve",
  "payment_key": "0000000000000000000000000000000000000000000000000000000000000000"
}

resolve instructs lightningd to claim the HTLC by providing the preimage matching the payment_hash presented in the call. Notice that the plugin must ensure that the payment_key really matches the payment_hash since lightningd will not check and the wrong value could result in the channel being closed.

Warning: lightningd will replay the HTLCs for which it doesn't have a final verdict during startup. This means that, if the plugin response wasn't processed before the HTLC was forwarded, failed, or resolved, then the plugin may see the same HTLC again during startup. It is therefore paramount that the plugin is idempotent if it talks to an external system.