This commit modifies the actions of the closeObserver goroutine to
utilize a _new_ channel to send channel close details over. The
original close signal channel is still used to notify observers that a
channel _has_ been closed, but this new channel will provide a single
observer with details w.r.t _how_ a channel was closed.
This commit adds an additional field to the ForceCloseSummary that
allows observers of the channel that sends this struct to track _which_
channel the force close came from.
This commit improves the channel state machine by converting the
objective PendingUpdates method to a subjective FullySynced method
which is to be used in place of PendingUpdates. The new FullySynced
method is fully encompassing and replaces any upper state required by
the state machine which wraps this one.
The new FullySynced method is identical to PendingUpdates aside from
the fact that: it now also factors in the log message index of the
remote commitment chain, and also introduces a concept of an “owed
commitment”. A commitment chain owes a commitment if the height of the
local commitment chain is higher than that of the remote chain.
This commit removes the theirPrevPkScript field from the
LightningChannel struct all together. It’s no longer needed as the more
fundamental mutation bug has been fixed within the channel state
machine.
This commit fixes a class of bug that can arise in the channel state
machine when a very high throughput workflow is attempted. Since the
PaymentDescriptor’s within a commitment pointed directly into the log,
any changes to a payment descriptor would also be reflected in all
other ones. Due to this mutation possibility, at times, the
locateOutputIndex method would fail since the HTLC’s pkScript was
modified, causing the channel to fail.
We fix this class of bug by simply ensure that once an HTLC has been
associated with a particular commitment, then it becomes immutable.
This commit fixes a slight oversight in the current state machine which
assumes that both commitment chains are always at the same height. In a
future where we move back to allowing nodes to pipeline commitment
updates, this will not always be the case.
This commit fixes a lingering TODO within the wallet portion of the
codebase by properly adhering to the set dust limits when closing a
channel. With this new commit if a party’s current settled balance is
below their current dust-limit, then it will be omitted from the
commitment transaction.
The prior test that asserted negative outputs are rejected has been
removed as they’ll now be avoided by ensuring we omit dust outputs from
the commitment transaction.
This commit does some minor shuffling around and also adds some
additional comments to the restoreStateLogs method within the channel
state machine. After the latest merge in this area, the code has
diverged slightly from what’s considered typical within the rest of the
codebase.
It is possible that that there are multiple HTLCs with different values,
but the same public key script. As such, a check against the value should
be performed when looking for HTLC outputs in a commitment transaction.
Create a new helper method called genHtlcScript which will
generate the public key scripts for a supplied HTLC. This functionality
from addHTLC is removed, and addHTLC will instead call this new
method.
In restoreStateLogs we will regenerate the public key scripts for the
HTLCs with genHtlcScript and restore the proper values.
When an HTLC is either cancelled or settled we must properly set the
pkScript for the HTLC on the remote commitment, such that we can
generate a valid ChannelDelta.
Description of bug:
When calling ReceiveNewCommitment() we will progress through methods
fetchCommitmentView and addHTLC which will add HTLC outputs to the
commitment transaction in the local commitment chain and save the
pkScript to the relevant PaymentDescriptor which resides in the
corresponding updateLog. Finally the local commitment will be added
to the local commitment chain.
When the same user next calls SignNextCommitment we will again
progress through fetchCommitmentView and addHTLC. In addHTLC we will
now overwrite the pkScripts in the PaymentDescriptors with the
pkScript from the context of the remote commitment. When we later
call RevokeCurrentCommitment and proceed into toChannelDelta, we
will not be able to find the correct pkScript in the PaymentDescriptor
to match it against the outputs in the commitment transaction.
This will lead to the nested function locateOutputIndex returning
incorrect values.
Fixing the bug:
We introduce three new fields in PaymentDescriptor:
* ourPkScript
* theirPkScript
* theirPrevPkScript
ourPkScript will include the pkScript for the HTLC from the context
of the local commitment.
theirPkScript will take the value of the latest pkScript for the HTLC
from the context of the remote commitment.
theirPrevPkScript will take the second-latest pkScript for the HTLC
from the context of the remote commitment. This is the value we use
in toChannelDelta when we save a revoked commitment from our peer.
The appropriate value of these fields are set in the addHTLC method.
Additionally we pass a boolean value to toChannelDelta so we know
whether we are operating on a local or remote commitment and grab
the correct pkScript in locateUpdateIndex.
This commit removes all instances of the fastsha256 library and
replaces it with the sha256 library in the standard library. This
change should see a number of performance improvements as the standard
library has highly optimized assembly instructions with use vectorized
instructions as the platform supports.
If the value of the to-local output is below the dust limit, the
ForceCloseSummary should not include a sign descriptor for this output.
We also find the proper to-self output by looking for the expected public
key script and not assume that no HTLC outputs exist.
Currently non-HTLC outputs will be accepted in the commitment
transaction as long as it is non-zero. We change this by not allowing
outputs with a value lower than the dust limit. The value of such
an output will go towards transaction fees.
This commit fixes a class of bug that currently exists within the
cooperative closure methods for the channel state machine. As an
example, due to the current hard coded fees, if one of the outputs
generated within the generated closure transaction has a negative
output, then the initiating node would gladly forward this to the
remote node. The remote node would then reject the closure as the
transaction is invalid. However, the act of completing the closure
would cause the remote node’s state machine to shift into a “closed”
state. As a result, any further closure attempts by the first node
(force or regular) would go unnoticed by the remote node.
We fix this issue by ensuring the transaction is “sane” before
initiating of completing a cooperative channel closure.
At test case has been added exercising the particular erroneous case
reported by “moli” on IRC.
This commit avoids a class of bug wherein the state of the channel
would be marked as closing enough though an error occurred somewhere in
the function. The bug was due to the fact that the channel `status` was
shifted before any actual logic within the function(s) were executed.
We fix this bug by _only_ shifting the channel status once the function
has completed without any error.
This commit adds a new method to the channel’s state machine:
NextRevocationKey. This method is being added in preparation for the
upcoming change to switch to the commitment transaction format outlined
in the spec. When this comes to pass, the ExtendRevocationWindow method
will be removed, as it will no longer be needed.
The NextRevocationKey method will be needed as to conform to the spec,
we’ll need to send the next revocation key within the `fundingLocked`
message.
In this commit the initial implementation of revocation hash
generation 'elkrem' was replaced with 'shachain' Rusty Russel
implementation which currently enshrined in the spec. This alghoritm has
the same asymptotic characteristics but has more complex scheme
to determine wish hash we can drop and what needs to be stored
in order to be able to achive full compression.
Fix SetStateNumHint and GetStateNumHint to properly
set and get the stateNumHints using the lower 24 bits
of the locktime of the commitment transaction as the
lower 24 bits of the obfuscated state number and the
lower 24 bits of the sequence field as the higher 24
bits.
This commit adds an additional case of the closeObserver that will
properly handle the case of a channel being closed by a de-sync’d
commitment transaction from the PoV of the local node. In the case of a
minor 1-state divergence, the commitment transaction broadcast by the
remote node will be 1 state ahead of the commitment transaction we have
locally. This should be seen as a regular unilateral close as they
remote peer didn’t violate the channel contract in any way.
We address this case by changing the `==` to a `>=`.
This commit updates the internal channel state machine to the one as
described within the spec and currently implemented within the rest of
the other Lightning implementations.
At a high level the following modifications have been made:
* When signing we no loner include the index of the remote party’s
log
that our signature covers. Instead we include ALL of our current
updates, but only the updates of the remote party that we’ve
ACK’d.
* A pending change is considered ACK’d once a revocation message
has been received, locking in the changes in the remote party’s
commitment transaction.
* When sending a new commitment, we remember the index of our
log at that point so we can mark that portion of the log as ACK’d
once we receive a revocation message from the remote party.
* When receiving a new commitment signature, we include ALL of
the remote party’s changes that we’ve received but only our set
of changes that’ve been ACK’d by the remote party.
* Implicitly a revocation message now also implicitly serves to ACK
all the changes that were included in the CommitSig message
received before it.
The resulting change is a rather minor diff. However, with this state
machine it’s important to note that the order to sig/revoke messages
has been swapped. A proper exchange now looks like the following:
* Alice -> Add, Add, Add
* Alice -> Sig
* Revoke <- Bob
* Sig <- Bob
* Alice -> Revoke
One other thing that’s worth noting is that with this state machine,
since what’s included in an update is implicit, both side may need to
at times send a new commitment update in the case of a concurrent state
transition initiated by both sides.
Finally, all counters/indexes have been made 64-bit integers in order
to properly match the spec.
This commit adds a new struct to the channel state machine: updateLog.
updateLog encapsulates the update log linked list itself, a series of
new counters we’ll need in order to switch to the spec’s state machine
and also the index into the log itself. This new struct serves to
simplify much of the logic surrounding the update log and also
elminates a bit of code duplication within the current state machine.
This commit only adds the new struct. The rest of the state machine
will be updated in a later commit to use the new log and its new
counters.
This commit fixes a bug in the LightingChannel commitment state machine
which could occasionally result in the total number of satoshis sent or
received being counted twice if a redundant state transition were
initiated.
To fix this bug, we now only increment the number of satoshi
sent/recv’d iff it’s the first time the HTLC has been processed.
This commit ensures that when a channel’s closeObserver is signaled to
exit before a channel closure has been detected, then the resources
dedicated to the pending spend notification can be freed up.
rHashMap is used to store the PaymentDescriptor belonging to a received
HTLC's revocation hash. This improves the efficiency of looking up
PaymentDescriptors from their RHash whenever we want to settle or cancel
that HTLC.
This commit adds the ability to cancel a channel’s internal
closeObserver goroutine by adding a new public facing Stop method.
Additionally, we now make passing a ChainNotifier interface completely
optional. If the ChainNotifier isn’t passed in as a constructor, then
the closeObserver goroutine will never be launched. This new feature
lets the caller ensure that only a single closeObsever for any given
channels exists.
This commit removes the BlockChainIO interface as a dependency to the
LightningChannel struct as the interface is no longer used within the
operation of the LightningChannel.
This commit fixes a bug introduced by the past attempt to Make Logging
Great Again. Since we unset the curve parameters when reading/writing
the messages, if we have a lingering reference that’s active elsewhere
in the daemon, then we’ll modify that reference. To fix this, we now
explicitly set the Curve parameters in two areas.
A similar commit has been pushed to lightning-onion.
This commit modifies the closeObserver goroutine to ensure that a close
summary has been inserted into the database before signalling any
observers that a unilateral channel closure was detected. This fixes a
slight bug where a peer would force close a channel, but we wouldn’t
properly detect that and clean up the channel state if had a failed
cooperative channel closure.
This commit modifies the logic of the closeObserver slightly to not
incorrectly mark the broadcast of the commitment transaction triggered
by a cooperative channel closure as an unprompted broadcast.
This commit modifies the logic around adding cancel entries to the
update log for the commitment state machine slightly by also including
the r-hash of the HTLC that’s been cancelled in the entry for the
cancellation. With this change, we can accurately track which HTLC is
being cancelled within outer sub-systems.
This commit alters the new HTLC cancellation logic to instead allow the
canceller of an HTLC to cancel the HTLC by the payment hash rather than
the index of the HTLC.
