Currently we always generate a
`MessageSendEvent::BroadcastChannelUpdate` when a channel is closed
even if the channel is private. Our immediate peers should ignore
such messages as they haven't seen a corresponding
`channel_announcement`, but we are still giving up some privacy by
informing our immediate peers of which channels were ours.
Here we split `ChannelManager::get_channel_update` into a
`get_channel_update_for_broadcast` and
`get_channel_update_for_unicast`. The first is used when we are
broadcasting a `channel_update`, allowing us to refuse to do so
for private channels. The second is used when failing a payment (in
which case the recipient has already shown that they are aware of
the channel so no such privacy concerns exist).
We had a user who pointed out that we weren't creating
`SpendableOutputs` events when we should have been after they
called `ChannelMonitor::best_block_updated` with a block well
after a CSV locktime and then called
`ChannelMonitor::transactions_confirmed` with the transaction which
we should have been spending (with a block height/hash a ways in
the past).
This was due to `ChannelMonitor::transactions_confirmed` only
calling `ChannelMonitor::block_confirmed` with the height at which
the transactions were confirmed, resulting in all checks being done
against that, not the current height.
Further, in the same scenario, we also would not fail-back and HTLC
where the HTLC-Timeout transaction was confirmed more than
ANTI_REORG_DELAY blocks ago.
To address this, we use the best block height for confirmation
threshold checks in `ChannelMonitor::block_confirmed` and pass both
the confirmation and current heights through to
`OnchainTx::update_claims_view`, using each as appropriate.
Fixes#962.
No matter the context, if we're told about a block which is
guaranteed by our API semantics to be on the best chain, and it has
a higher height than our current understanding of the best chain,
we should update our understanding. This avoids complexity
in `block_confirmed` by never having a height set which is *higher*
than our current best chain, potentially avoiding some bugs in the
rather-complicated code.
It also requires a minor test tweak as we in some cases now no
longer broadcast a conflicting transaction after the original has
reached the ANTI_REORG_DELAY.
There are no visible effects of this, but it seems like good code
hygiene to not call a disconnect function in a different file if no
disconnect happened.
We use `Channel::is_live()` to gate inclusion of a channel in
`ChannelManager::list_usable_channels()` and when sending an
HTLC to select whether a channel is available for
forwarding through/sending to.
In both of these cases, we should consider a channel `is_live()` when
they are pending a monitor update. Some clients may update monitors
asynchronously, thus we may simply be waiting a short duration for a
monitor update to complete, and shouldn't fail all forwarding HTLCs
during that time.
After #851, we always ensure any holding cells are free'd when
sending P2P messages, making this change much more trivially
correct - instead of having to ensure that we always free the holding
cell when a channel becomes live again after adding something to the
holding cell, we can simply rely on the fact that it always happens.
Fixes#661.
If we receive a `channel_update` message for a channel unrelated to
our own, we shouldn't trigger a persistence of our
`ChannelManager`. This avoids significant persistence traffic during
initial node startup.
We changed the sort order of log levels to be more natural, but this
comparison wasn't updated accordingly. Likely the reason it was
left strange for so long is it also had the comparison argument
ordering flipped.
ChannelMonitor and related log entries can generally lean towards
being higher log levels than they necessarily need to be, as they
should be exceedingly rare, if only because they require
confirmation of an on-chain transaction.
This updates a number of log sites in channel and channelmanager to
* Be a bit more verbose at the TRACE level,
* Move some error/useful messages to the ERROR/WARN/INFO level,
* Add new logs to always log once at the DEBUG level when we
send/receive a commitment_signed (with some extra data),
* Include the channel id being operated on in more log messages.
We very often receive duplicate gossip messages, which now causes us
to log at the DEBUG level, which is almost certainly not what a
user wants. Instead, we add a new form of ErrorAction which causes
us to only log at the TRACE level.
For log entries which may have a variable level, we can't call an
arbitrary macro and need to be able to pass an explicit level. This
does so without breaking the compile-time disabling of certain log
levels.
Further, we "fix" the comparison order of log levels to make more
significant levels sort "higher", which implicitly makes more sense
than sorting "lower".
Finally, we remove the "Off" log level as no log entry should ever
be logged at the "Off" level - that would be nonsensical.
This much more consistently logs information about messages
sent/received, including logging the full messages being
sent/received at the TRACE log level. Many other log messages which
are more often of interest were moved to the DEBUG log level.
We had a client application which provided inconsistent monitor
state when deserializing a ChannelManager, resulting in opaque and
generic "InvalidData" deserialization failures. Instead, we log
some informative (and appropriately scary) warning messages in
such cases.
`Event::PaymentSent` serialization has a bug where we
double-serialize the payment_preimage field but do not attempt to
read it twice. This results in a failure to read `ChannelManager`s
from disk if we have a pending `Event::PaymentSent` pending
awaiting handling when we serialize.
Instead of attempting to read both versions, we opt to simply fix
the serialization, assuming it is exceedingly rare for such a
scenario to appear (and if it does, we can assist in manual
recovery).
The release notes have been updated to note this inconsistency.
Found by the chanmon_consistency fuzz target.
lnd has a long-standing bug where, upon reconnection, if the
channel is not yet confirmed they will not send a
channel_reestablish until the channel locks in. Then, they will
send a funding_locked *before* sending the channel_reestablish
(which is clearly a violation of the BOLT specs). We copy
c-lightning's workaround here and simply store the funding_locked
message until we receive a channel_reestablish.
See-also https://github.com/lightningnetwork/lnd/issues/4006Fixes#963
Previous to this PR, TLV serialization involved iterating from 0 to the highest
given TLV type. This worked until we decided to implement keysend, which has a
TLV type of ~5.48 billion.
So instead, we now specify the type of whatever is being (de)serialized (which
can be an Option, a Vec type, or a non-Option (specified in the serialization macros as "required").
When a peer sends us the routing graph, it may include gossip
messages for our channels, despite it not being a party to them.
This is completely fine, but we currently print a somewhat-scary
looking log messages in these cases, eg:
```
ERROR [lightning::ln::channelmanager:4104] Got a message for a channel from the wrong node!
TRACE [lightning::ln::peer_handler:1267] Handling SendErrorMessage HandleError event in peer_handler for node ... with message Got a message for a channel from the wrong node!
```
Instead, we should simply not consider this an "error" condition
and stay silent.
The only practical way to meet this requirement is to block
disconnect_socket until any pending events are fully processed,
leading to this trivial deadlock:
* Thread 1: select() woken up due to a read event
* Thread 2: Event processing causes a disconnect_socket call to
fire while the PeerManager lock is held.
* Thread 2: disconnect_socket blocks until the read event in
thread 1 completes.
* Thread 1: bytes are read from the socket and
PeerManager::read_event is called, waiting on the lock
still held by thread 2.
There isn't a trivial way to address this deadlock without simply
making the final read_event call return immediately, which we do
here. This also implies that users can freely call event methods
after disconnect_socket, but only so far as the socket descriptor
is different from any later socket descriptor (ie until the file
descriptor is re-used).
`Julian Knutsen <julianknutsen@users.noreply.github.com>` pointed
out in a previous discussion that `read_event` can reenter user
code despite the documentation stating explicitly that it will not.
This was addressed in #456 by simply codifying the reentrancy, but
its somewhat simpler to just drop the `do_attempt_write_data` call.
Ideally we could land most of Julian's work, but its still in need
of substantial git history cleanup to get it in a reviewable state
and this solves the immediate issue.
