... by calling it both before and after every chain event in
testing and fuzzing.
This requires fixing some blockchain inconsistencies in
`do_test_onchain_htlc_reorg`, `do_retry_with_no_persist`, and
`do_test_dup_htlc_onchain_fails_on_reload` where we'd connect
conflicting transactions in the same chain.
When handling the broadcast of a local commitment transactions
(with associated CSV delays prior to spendability), we incorrectly
handled the CSV delays on HTLC transactions. This caused us to miss
spendable outputs for HTLCs which were awaiting a CSV delay.
Further, because of this, we could hit an assertion as
`get_claimable_balances` asserted that HTLCs were resolved after
the funding spend was resolved, which was not true if the HTLC did
not have a CSV delay attached (due to the above bug or due to it
being an HTLC claim by our counterparty).
This fixes both bugs, also converting some assertions to
`debug_assert`s to avoid future issues as balance mis-calculation
is not currently an indication of potential funds loss.
Thanks to Cash App for reporting this bug.
Failing an HTLC with onion error channel_disabled requires encoding a 'flags' field into the failure
packet. However, we were encoding this 'flags' field for all failures packets that were failing on
update_add_htlc with an update (error 0x1000 UPDATE).
Discovered in the course of adding phantom payment failure tests, which also added testing for this bug
In any place where fail_htlc_backwards_internal was called for a phantom payment
failure, we weren't encoding the onion failure as if the phantom were the one
failing. Instead, we were encoding the failure as if it were coming from the
second-to-last hop. This caused our failures to not be parsed properly on the
payer's side.
Places we were encoding failures incorrectly include:
* on failure of a call to inbound_payment::verify
* on a user call to fail_htlc_backwards
Also drop some unnecessary panics when reading OnionHopData objects. This also
enables one of the phantom failure tests because we can construct OnionHopDatas
with invalid amounts.
Lastly, remove a bogus comment
This also fixes a bug where we were failing back phantom payments with the
wrong scid, causing them to never actually be failed backwards (L3022 in
channelmanager.rs)
This new field will be used in upcoming commit(s) to encrypt phantom payment failure
packets.
The take-self-return-Self idiom in Rust is substantially less
usable than it is in Java, where its more common. Because we have
to take self by move, it prevents using the update methods to
actually update features, something we occasionally want to do.
See, eg, the change in lightning-invoice where we previously had
to copy and re-create an entire vec of fields just to update the
features field, which is nuts.
There are a few places where this makes things a little less clean,
but the tradeoff to enable more effecient and broader uses of the
update methods seems worth it.
Prior to cryptographic payment secrets, when we process a received
payment in `process_pending_htlc_fowards` we'd remove its entry
from the `pending_inbound_payments` map and give the user a
`PaymentReceived` event.
Thereafter, if a second HTLC came in with the same payment hash, it
would find no entry in the `pending_inbound_payments` map and be
immediately failed in `process_pending_htlc_forwards`.
Thus, each HTLC will either result in a `PaymentReceived` event or
be failed, with no possibility for both.
As of 8464875555, we no longer
materially have a pending-inbound-payments map, and thus
more-than-happily accept a second payment with the same payment
hash even if we just failed a previous one for having mis-matched
payment data.
This can cause an issue if the two HTLCs are received back-to-back,
with the first being accepted as valid, generating a
`PaymentReceived` event. Then, when the second comes in we'll hit
the "total value {} ran over expected value" condition and fail
*all* pending HTLCs with the same payment hash. At this point,
we'll have a pending failure for both HTLCs, as well as a
`PaymentReceived` event for the user.
Thereafter, if the user attempts to fail the HTLC in response to
the `PaymentReceived`, they'll get a debug panic at channel.rs:1657
'Tried to fail an HTLC that was already failed'.
The solution is to avoid bulk-failing all pending HTLCs for a
payment. This feels like the right thing to do anyway - if a sender
accidentally sends an extra HTLC after a payment has ben fully
paid, we shouldn't fail the entire payment.
Found by the `chanmon_consistency` fuzz test.
Somehow, our channel type implementation doesn't echo back the
channel type as we believe it was negotiated, as we should. Though
the spec doesn't explicitly require this, some implementations may
require it and it appears to have been in the BOLTs from the start
of the channel type logic.
Add functional tests for manually responding to inbound channel requests.
Responding to inbound channel requests are required when the
`manually_accept_inbound_channels` config flag is set to true.
The tests cover the following cases:
* Accepting an inbound channel request
* Rejecting an inbound channel request
* FundingCreated message sent by the counterparty before accepting the
inbound channel request
* Attempting to accept an inbound channel request twice
* Attempting to accept an unkown inbound channel
Add a new config flag `UserConfig::manually_accept_inbound_channels`,
which when set to true allows the node operator to accept or reject new
channel requests.
When set to true, `Event::OpenChannelRequest` will be triggered once a
request to open a new inbound channel is received. When accepting the
request, `ChannelManager::accept_inbound_channel` should be called.
Rejecting the request is done through
`ChannelManager::force_close_channel`.
`cargo bench` sets `cfg(test)`, causing us to hit some test-only
code in the router when benchmarking, throwing off our benchmarks
substantially. Here we swap from the `unstable` feature to a more
clearly internal feature (`_bench_unstable`) and also checking for
it when enabling test-only code.
Given the balance is reported as "total balance if we went to chain
ignoring fees", it seems reasonable to include claimed HTLCs - if
we went to chain we'd get those funds, less on-chain fees. Further,
if we do not include them, its possible to have pending outbound
holding-cell HTLCs underflow the balance calculation, causing a
panic in debug mode, and bogus values in release.
This resolves a subtraction underflow bug found by the
`chanmon_consistency` fuzz target.
Apparently rustc doesn't (actually) provide any kind of
compilation-stability guarantees, despite their claims. Here we
work around rustc being unstable by making the trait call explicit.
See also https://github.com/rust-lang/rust/issues/93599
We currently allow users to provide an `override_config` in
`ChannelManager::create_channel` which it seems should apply to the
channel. However, because we don't store any of it, the only parts
which we apply to the channel are those which are set in the
`Channel` object immediately in `Channel::new_outbound` and used
from there.
This is great in most cases, however the
`UserConfig::peer_channel_config_limits` `ChannelHandshakeLimits`
object is used in `accept_channel` to bound what is acceptable in
our peer's `AcceptChannel` message. Thus, for outbound channels, we
are given a full `UserConfig` object to "override" the default
config, but we don't use any of the handshake limits specified in
it.
Here, we move to storing the `ChannelHandshakeLimits` explicitly
and applying it when we receive our peer's `AcceptChannel`. Note
that we don't need to store it anywhere because if we haven't
received an `AcceptChannel` from our peer when we reload from disk
we will forget the channel entirely anyway.
Channel::get_announcement_sigs is only used in contexts where we
have a logger already, and the error returned is always ignored, so
instead of returning an ignored error message we return an `Option`
directly and log when it won't be too verbose.
The spec actually requires we never send `announcement_signatures`
(and, thus, `channel_announcement`s) until after six confirmations.
However, we would happily have sent them prior to that as long as
we exchange `funding_locked` messages with our countarparty. Thanks
to re-broadcasting this issue is largely harmless, however it could
have some negative interactions with less-robust peers. Much more
importantly, this represents an important step towards supporting
0-conf channels, where `funding_locked` messages may be exchanged
before we even have an SCID to construct the messages with.
Because there is no ACK mechanism for `announcement_signatures` we
rely on existing channel updates to stop rebroadcasting them - if
we sent a `commitment_signed` after an `announcement_signatures`
and later receive a `revoke_and_ack`, we know our counterparty also
received our `announcement_signatures`. This may resolve some rare
edge-cases where we send a `funding_locked` which our counterparty
receives, but lose connection before the `announcement_signatures`
(usually the very next message) arrives.
Sadly, because the set of places where an `announcement_signatures`
may now be generated more closely mirrors where `funding_locked`
messages may be generated, but they are now separate, there is a
substantial amount of code motion providing relevant parameters
about current block information and ensuring we can return new
`announcement_signatures` messages.
If we have not yet sent `funding_locked` only because of a pending
channel monitor update, we shouldn't consider a channel
`is_usable`. This has a number of downstream effects, including
not attempting to route payments through the channel, not sending
private `channel_update` messages to our counterparty, or sending
channel_announcement messages if our couterparty has already signed
for it.
We further gate generation of `node_announcement`s on `is_usable`,
preventing generation of those or `announcement_signatures` until
we've sent our `funding_locked`.
Finally, `during_funding_monitor_fail` is updated to test a case
where we see the funding transaction lock in but have a pending
monitor update failure, then receive `funding_locked` from our
counterparty and ensure we don't generate the above messages until
after the monitor update completes.
While its generally harmless to do so (the messages will simply be
dropped in `PeerManager`) there is a potential race condition where
the FundingLocked message enters the outbound message queue, then
the peer reconnects, and then the FundingLocked message is
delivered prior to the normal ChannelReestablish flow.
We also take this opportunity to rewrite
`test_funding_peer_disconnect` to be explicit instead of using
`reconnect_peers`. This allows it to check each message being sent
carefully, whereas `reconnect_peers` is rather lazy and accepts
that sometimes signatures will be exchanged, and sometimes not.
Quite some time ago, `UnknownRequiredFeature` was only used when a
gossip message has a missing required feature. These days, its also
used for any required TLV which we do not understand in any
message. However, the handling of it was never updated in
`PeerManager`, leaving it printing a warning about gossip and
ignoring the message entirely.
Instead, we send a warning message and disconnect.
Closes#1236, as caught by @jkczyz.
This is harmless outside of debug builds - the feerate will
overflow causing it to either spuriously fail the first check, or
correctly pass it and fail the second check. In debug builds,
however, it panics due to integer overflow.
Found by the `full_stack_target` fuzz test in the
Chaincode-provided continuous fuzzing. Thanks Chaincode!
