`ChannelManager::fail_htlc_backwards`' bool return value is quite
confusing - just because it returns false doesn't mean the payment
wasn't (already) failed. Worse, in some race cases around shutdown
where a payment was claimed before an unclean shutdown and then
retried on startup, `fail_htlc_backwards` could return true even
though (a duplicate copy of the same payment) was claimed, but the
claim event has not been seen by the user yet.
While its possible to use it correctly, its somewhat confusing to
have a return value at all, and definitely lends itself to misuse.
Instead, we should push users towards a model where they don't care
if `fail_htlc_backwards` succeeds - either they've locally marked
the payment as failed (prior to seeing any `PaymentReceived`
events) and will fail any attempts to pay it, or they have not and
the payment is still receivable until its timeout time is reached.
We can revisit this decision based on user feedback, but will need
to very carefully document the potential failure modes here if we
do.
As additional sanity checks, before claiming a payment, we check
that we have the full amount available in `claimable_htlcs` that
the payment should be for. Concretely, this prevents one
somewhat-absurd edge case where a user may receive an MPP payment,
wait many *blocks* before claiming it, allowing us to fail the
pending HTLCs and the sender to retry some subset of the payment
before we go to claim. More generally, this is just good
belt-and-suspenders against any edge cases we may have missed.
If we crashed during a payment claim and then detected a partial
claim on restart, we should ensure the user is aware that the
payment has been claimed. We do so here by using the new
partial-claim detection logic to create a `PaymentClaimed` event.
While the HTLC-claim process happens across all MPP parts under one
lock, this doesn't imply that they are claimed fully atomically on
disk. Ultimately, an application can crash after persisting one
`ChannelMonitorUpdate` out of multiple monitor updates needed for
the full claim.
Previously, this would leave us in a very bad state - because of
the all-channels-available check in `claim_funds` we'd refuse to
claim the payment again on restart (even though the
`PaymentReceived` event will be passed to the user again), and we'd
end up having partially claimed the payment!
The fix for the consistency part of this issue is pretty
straightforward - just check for this condition on startup and
complete the claim across all channels/`ChannelMonitor`s if we
detect it.
This still leaves us in a confused state from the perspective of
the user, however - we've actually claimed a payment but when they
call `claim_funds` we return `false` indicating it could not be
claimed.
In fc77c57c3c we stopped using the
`FInalOnionHopData` in `OnionPayload::Invoice` directly and intend
to remove it eventually. However, in the next few commits we need
access to the payment secret when claimaing a payment, as we create
a new `PaymentPurpose` during the claim process for a new event.
In order to get access to a `PaymentPurpose` without having access
to the `FinalOnionHopData` we here change the storage of
`claimable_htlcs` to store a single `PaymentPurpose` explicitly
with each set of claimable HTLCs.
In fc77c57c3c we stopped using the
`FinalOnionHopData` in `OnionPayload::Invoice` directly and renamed
it `_legacy_hop_data` with the intent of removing it in a few
versions. However, we continue to check that it was included in the
serialized data, meaning we would not be able to remove it without
breaking ability to serialize full `ChannelManager`s.
This fixes that by making the `_legacy_hop_data` an `Option` which
we will happily handle just fine if its `None`.
This update also includes a minor refactor. The return type of
`pending_monitor_events` has been changed to a `Vec` tuple with the
`OutPoint` type. This associates a `Vec` of `MonitorEvent`s with a
funding outpoint.
We've also renamed `source/sink_channel_id` to `prev/next_channel_id` in
the favour of clarity.
As the `counterparty_node_id` is now required to be passed back to the
`ChannelManager` to accept or reject an inbound channel request, the
documentation is updated to reflect that.
Instead of including a `Secp256k1` context per
`PeerChannelEncryptor`, which is relatively expensive memory-wise
and nontrivial CPU-wise to construct, we should keep one for all
peers and simply reuse it.
This is relatively trivial so we do so in this commit.
Since its trivial to do so, we also take this opportunity to
randomize the new PeerManager context.
Because we handle messages (which can take some time, persisting
things to disk or validating cryptographic signatures) with the
top-level read lock, but require the top-level write lock to
connect new peers or handle disconnection, we are particularly
sensitive to writer starvation issues.
Rust's libstd RwLock does not provide any fairness guarantees,
using whatever the OS provides as-is. On Linux, pthreads defaults
to starving writers, which Rust's RwLock exposes to us (without
any configurability).
Here we work around that issue by blocking readers if there are
pending writers, optimizing for readable code over
perfectly-optimized blocking.
This avoids any extra calls to `read_event` after a write fails to
flush the write buffer fully, as is required by the PeerManager
API (though it isn't critical).
Only one instance of PeerManager::process_events can run at a time,
and each run always finishes all available work before returning.
Thus, having several threads blocked on the process_events lock
doesn't accomplish anything but blocking more threads.
Here we limit the number of blocked calls on process_events to two
- one processing events and one blocked at the top which will
process all available events after the first completes.
Because the peers write lock "blocks the world", and happens after
each read event, always taking the write lock has pretty severe
impacts on parallelism. Instead, here, we only take the global
write lock if we have to disconnect a peer.
Unlike very ancient versions of lightning-net-tokio, this does not
rely on a single global process_events future, but instead has one
per connection. This could still cause significant contention, so
we'll ensure only two process_events calls can exist at once in
the next few commits.
Users are required to only ever call `read_event` serially
per-peer, thus we actually don't need any locks while we're
processing messages - we can only be processing messages in one
thread per-peer.
That said, we do need to ensure that another thread doesn't
disconnect the peer we're processing messages for, as that could
result in a peer_disconencted call while we're processing a
message for the same peer - somewhat nonsensical.
This significantly improves parallelism especially during gossip
processing as it avoids waiting on the entire set of individual
peer locks to forward a gossip message while several other threads
are validating gossip messages with their individual peer locks
held.
This adds the required locking to process messages from different
peers simultaneously in `PeerManager`. Note that channel messages
are still processed under a global lock in `ChannelManager`, and
most work is still processed under a global lock in gossip message
handling, but parallelizing message deserialization and message
decryption is somewhat helpful.
