Providing LDK-specific data in a BlindedPath allows for the data to be
received with the corresponding payment. This is useful as it can then
be surfaced in PaymentClaimable events where the user may correlated
with an Offer, for instance. Define a PaymentContext enum for including
various context (e.g., offer, refund, static invoice).
When sending an invoice for a refund, information from the invoice may
be useful for caller. For instance, the payment_hash can be used to
track whether the refund was paid. Return the invoice to facilitate this
use case.
Previously, we would hold the `per_peer_state` `read` lock during the
serialization of channel objects. While this might have reduced lock
congestion minimally, we also calculated the number of serializable
peers in this phase, only to use it later after dropping and reacquiring
the `per_peer_state` `write` lock.
This could potentially result in inconsistiencies if the
`serializable_peer_count` would change after we dropped the `read` lock
but before we acquired the `write` lock.
To mitigate the issue we just acquire the `write` lock ealier and hold
it for the remainder of `ChannelManager::write`.
Bolt12Invoice will be added to a new Event::InvoiceGenerated variant.
These traits along with PartialEq are required to be implemented for any
type used in an Event.
In order to provide an InvoiceGenerated event, it would be cleaner to
have one location where a Bolt12Invoice is successfully created.
Refactor the handling code to this end and clean-up line length by
making some of the type conversions more streamlined.
In order to generate and InvoiceSent event, it would be cleaner to have
one location where a Bolt12Invoice is successfully generated. Refactor
the handling code to this end and clean-up line length by making some of
the type conversions more streamlined.
Add a version of the create_onion_message utility function that attempts
to resolved the introduction node of the destination's BlindedPath using
a ReadOnlyNetworkGraph`. Otherwise, if given a path using the compact
representation, using create_onion_message would fail. Keep the current
version for use internally and for external uses where the blinded path
is known to be resolved.
When forwarding onion messages, the next node may be represented by a
short channel id instead of a node id. Parameterize OnionMessenger with
a NodeIdLookUp trait to find which node is the next hop. Implement the
trait for ChannelManager for forwarding to channel counterparties.
Also use this trait when advancing a blinded path one hop when the
sender is the introduction node.
When OnionMessenger creates an OnionMessage to a Destination, the latter
may contain an IntroductionNode::DirectedShortChannelId inside a
BlindedPath. Resolve these in DefaultMessageRouter and handle unresolved
ones in OnionMessenger.
Allow using either a node id or a directed short channel id in blinded
paths. This allows for a more compact representation of blinded paths,
which is advantageous for reducing offer QR code size.
Follow-up commits will implement handling the directed short channel id
case in OnionMessenger as it requires resolving the introduction node in
MessageRouter.
Blinded paths currently contain a node id for the introduction node.
However, a compact representation will allow a directed short channel id
instead. Update BlindedPathCandidate and OneHopBlindedPathCandidate to
store a NodeId reference from either the NetworkGraph or from the user-
provided first hops.
This approach avoids looking up the introduction node id on demand,
which may not be resolvable. Thus, that would require returning an
Option from CandidateRouteHop::source and handle None accordingly.
The current `cmp::max` doesnt align with the function comment, ie its
comparing 2530 and `feerate_plus_quarter` instead of `feerate_per_kw
+ 2530` and `feerate_plus_quarter` which is fixed in this commit
The initial noise handshake on connection establishment must complete
within a single timer tick. This timeout is enforced via
`awaiting_pong_timer_tick_intervals` whenever a timer tick fires while
our handshake has yet to complete. Currently, on an inbound connection,
if a timer tick fires after we've sent act two of the noise handshake
along with our init message and before receiving the counterparty's init
message, we begin enforcing such timeout. Even if we immediately
continue to process the counterparty's init message to complete to
handshake, the timeout enforcement is not cleared. With the handshake
complete, `awaiting_pong_timer_tick_intervals` is now tracked to enforce
a pong timeout, except a ping was never actually sent. If a single timer
tick fires again without having received a message from the peer, or
enough timer ticks fire to trigger the
`MAX_BUFFER_DRAIN_TICK_INTERVALS_PER_PEER` logic, we'll end up
disconnecting the peer due to a timeout for a pong we'll never receive.
We fix this by always resetting `awaiting_pong_timer_tick_intervals`
upon processing our counterparty's init message.
If we are missing a `ChannelMonitor` for which the `ChannelManager`
has pending updates, it may be useful to print what the updates we
have actually are, at least useful in identifying the bug(s).
Blinded paths use a pubkey to identify the introduction node, but it
will soon allow using a directed short channel id instead. Add an
introduction_node_id method to BlindedPath to facilitate lookup in the
latter case.
Allow either using a node id or a short channel id when forwarding an
onion message. This allows for a more compact representation of blinded
paths, which is advantageous for reducing offer QR code size.
Follow-up commits will implement handling the short channel id case as
it requires looking up the destination node id.
cc78b77c71 fixed an important
downgrade bug, but neglected to add test coverage. Here we recitfy
that by adding a few simple tests of common cases.
Tests heavily tweaked by Matt Corallo <git@bluematt.me>.
If we are reading an object that is `MaybeReadable` in a TLV stream
using `upgradable_required`, it may return early with `Ok(None)`.
In this case, it will not read any further TLVs from the TLV
stream. This is fine, except that we generally expect
`MaybeReadable` always consume the correct number of bytes for the
full object, even if it doesn't understand it.
This could pose a problem, for example, in cases where we're
reading a TLV-stream `MaybeReadable` object inside another
TLV-stream object. In that case, the `MaybeReadable` object may
return `Ok(None)` and not consume all the available bytes, causing
the outer TLV read to fail as the TLV length does not match.
