lnd.xprv/lnwallet/channel.go

2509 lines
89 KiB
Go
Raw Normal View History

package lnwallet
import (
"bytes"
2016-07-06 03:01:55 +03:00
"container/list"
"crypto/sha256"
"fmt"
"sync"
"sync/atomic"
"github.com/davecgh/go-spew/spew"
"github.com/lightningnetwork/lnd/chainntnfs"
2016-01-16 21:45:54 +03:00
"github.com/lightningnetwork/lnd/channeldb"
2016-07-06 03:01:55 +03:00
"github.com/lightningnetwork/lnd/lnwire"
"github.com/roasbeef/btcd/blockchain"
"github.com/roasbeef/btcd/chaincfg/chainhash"
"github.com/roasbeef/btcd/btcec"
"github.com/roasbeef/btcd/txscript"
"github.com/roasbeef/btcd/wire"
"github.com/roasbeef/btcutil"
"github.com/roasbeef/btcutil/txsort"
)
var zeroHash chainhash.Hash
2016-07-06 03:01:55 +03:00
var (
// ErrChanClosing is returned when a caller attempts to close a channel
// that has already been closed or is in the process of being closed.
ErrChanClosing = fmt.Errorf("channel is being closed, operation disallowed")
// ErrNoWindow is returned when revocation window is exausted.
ErrNoWindow = fmt.Errorf("unable to sign new commitment, the current" +
" revocation window is exhausted")
// ErrMaxWeightCost is returned when the cost/weight (see segwit)
// exceeds the widely used maximum allowed policy weight limit. In this
// case the commitment transaction can't be propagated through the
// network.
ErrMaxWeightCost = fmt.Errorf("commitment transaction exceed max " +
"available cost")
// ErrMaxHTLCNumber is returned when a proposed HTLC would exceed the
// maximum number of allowed HTLC's if committed in a state transition
ErrMaxHTLCNumber = fmt.Errorf("commitment transaction exceed max " +
"htlc number")
)
const (
// InitialRevocationWindow is the number of revoked commitment
// transactions allowed within the commitment chain. This value allows
// a greater degree of de-synchronization by allowing either parties to
// extend the other's commitment chain non-interactively, and also
// serves as a flow control mechanism to a degree.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
InitialRevocationWindow = 1
)
// channelState is an enum like type which represents the current state of a
// particular channel.
2016-07-06 03:01:55 +03:00
// TODO(roasbeef): actually update state
type channelState uint8
const (
// channelPending indicates this channel is still going through the
// funding workflow, and isn't yet open.
channelPending channelState = iota
// channelOpen represents an open, active channel capable of
// sending/receiving HTLCs.
channelOpen
// channelClosing represents a channel which is in the process of being
// closed.
channelClosing
// channelClosed represents a channel which has been fully closed. Note
// that before a channel can be closed, ALL pending HTLCs must be
// settled/removed.
channelClosed
// channelDispute indicates that an un-cooperative closure has been
// detected within the channel.
channelDispute
// channelPendingPayment indicates that there a currently outstanding
// HTLCs within the channel.
channelPendingPayment
)
2016-07-06 03:01:55 +03:00
// PaymentHash represents the sha256 of a random value. This hash is used to
// uniquely track incoming/outgoing payments within this channel, as well as
// payments requested by the wallet/daemon.
type PaymentHash [32]byte
// updateType is the exact type of an entry within the shared HTLC log.
2016-07-06 03:01:55 +03:00
type updateType uint8
const (
// Add is an update type that adds a new HTLC entry into the log.
// Either side can add a new penidng HTLC by adding a new Add entry
// into their update log.
2016-07-06 03:01:55 +03:00
Add updateType = iota
// Fail is an update type which removes a prior HTLC entry from the
// log. Adding a Fail entry to ones log will modify the _remote_
// parties update log once a new commitment view has been evaluated
// which contains the Fail entry.
Fail
// Settle is an update type which settles a prior HTLC crediting the
// balance of the receiving node. Adding a Settle entry to a log will
// result in the settle entry being removed on the log as well as the
// original add entry from the remote party's log after the next state
// transition.
2016-07-06 03:01:55 +03:00
Settle
)
// PaymentDescriptor represents a commitment state update which either adds,
// settles, or removes an HTLC. PaymentDescriptors encapsulate all necessary
// metadata w.r.t to an HTLC, and additional data pairing a settle message to
2016-07-06 03:01:55 +03:00
// the original added HTLC.
// TODO(roasbeef): LogEntry interface??
// * need to separate attrs for cancel/add/settle
2016-07-06 03:01:55 +03:00
type PaymentDescriptor struct {
sync.RWMutex
// RHash is the payment hash for this HTLC. The HTLC can be settled iff
// the preimage to this hash is presented.
RHash PaymentHash
// RPreimage is the preimage that settles the HTLC pointed to wthin the
// log by the ParentIndex.
RPreimage PaymentHash
// Timeout is the absolute timeout in blocks, after which this HTLC
2016-07-06 03:01:55 +03:00
// expires.
Timeout uint32
// Amount is the HTLC amount in satoshis.
Amount btcutil.Amount
// Index is the log entry number that his HTLC update has within the
// log. Depending on if IsIncoming is true, this is either an entry the
// remote party added, or one that we added locally.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
Index uint64
2016-07-06 03:01:55 +03:00
// ParentIndex is the index of the log entry that this HTLC update
// settles or times out.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
ParentIndex uint64
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// Payload is an opaque blob which is used to complete multi-hop
// routing.
2016-07-06 03:01:55 +03:00
Payload []byte
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// EntryType denotes the exact type of the PaymentDescriptor. In the
// case of a Timeout, or Settle type, then the Parent field will point
// into the log to the HTLC being modified.
EntryType updateType
2016-07-06 03:01:55 +03:00
// addCommitHeight[Remote|Local] encodes the height of the commitment
// which included this HTLC on either the remote or local commitment
// chain. This value is used to determine when an HTLC is fully
// "locked-in".
addCommitHeightRemote uint64
addCommitHeightLocal uint64
// removeCommitHeight[Remote|Local] encodes the height of the
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// commitment which removed the parent pointer of this
// PaymentDescriptor either due to a timeout or a settle. Once both
// these heights are above the tail of both chains, the log entries can
// safely be removed.
2016-07-06 03:01:55 +03:00
removeCommitHeightRemote uint64
removeCommitHeightLocal uint64
// isForwarded denotes if an incoming HTLC has been forwarded to any
// possible upstream peers in the route.
isForwarded bool
lnwallet: fix bug that retrieves incorrect pkScript in toChannelDelta() 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.
2017-01-20 14:37:30 +03:00
// [our|their|theirPrev]PkScript are the raw public key scripts that
// encodes the redemption rules for this particular HTLC. These fields
// will only be populated iff the EntryType of this PaymentDescriptor
// is Add. ourPkScript is the ourPkScript from the context of our local
// commitment chain. [their|theirPrev]PkScript are the two latest
// pkScripts from the context of the remote commitment chain.
ourPkScript []byte
theirPkScript []byte
theirPrevPkScript []byte
2016-07-06 03:01:55 +03:00
}
// commitment represents a commitment to a new state within an active channel.
// New commitments can be initiated by either side. Commitments are ordered
// into a commitment chain, with one existing for both parties. Each side can
2016-10-15 16:18:38 +03:00
// independently extend the other side's commitment chain, up to a certain
// "revocation window", which once reached, disallows new commitments until
2016-07-06 03:01:55 +03:00
// the local nodes receives the revocation for the remote node's chain tail.
type commitment struct {
// height represents the commitment height of this commitment, or the
// update number of this commitment.
height uint64
// [our|their]MessageIndex are indexes into the HTLC log, up to which
// this commitment transaction includes. These indexes allow both sides
2016-10-15 16:18:38 +03:00
// to independently, and concurrent send create new commitments. Each
2016-07-06 03:01:55 +03:00
// new commitment sent to the remote party includes an index in the
// shared log which details which of their updates we're including in
// this new commitment.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
ourMessageIndex uint64
theirMessageIndex uint64
2016-07-06 03:01:55 +03:00
// txn is the commitment transaction generated by including any HTLC
// updates whose index are below the two indexes listed above. If this
// commitment is being added to the remote chain, then this txn is
// their version of the commitment transactions. If the local commit
// chain is being modified, the opposite is true.
txn *wire.MsgTx
// sig is a signature for the above commitment transaction.
sig []byte
// [our|their]Balance represents the settled balances at this point
// within the commitment chain. This balance is computed by properly
// evaluating all the add/remove/settle log entries before the listed
// indexes.
ourBalance btcutil.Amount
theirBalance btcutil.Amount
// htlcs is the set of HTLCs which remain unsettled within this
// commitment.
outgoingHTLCs []*PaymentDescriptor
incomingHTLCs []*PaymentDescriptor
}
// toChannelDelta converts the target commitment into a format suitable to be
// written to disk after an accepted state transition.
// TODO(roasbeef): properly fill in refund timeouts
lnwallet: fix bug that retrieves incorrect pkScript in toChannelDelta() 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.
2017-01-20 14:37:30 +03:00
func (c *commitment) toChannelDelta(ourCommit bool) (*channeldb.ChannelDelta, error) {
numHtlcs := len(c.outgoingHTLCs) + len(c.incomingHTLCs)
// Save output indexes for RHash values found, so we don't return the
// same output index more than once.
dups := make(map[PaymentHash][]uint16)
delta := &channeldb.ChannelDelta{
LocalBalance: c.ourBalance,
RemoteBalance: c.theirBalance,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
UpdateNum: c.height,
Htlcs: make([]*channeldb.HTLC, 0, numHtlcs),
}
// Check to see if element (e) exists in slice (s)
contains := func(s []uint16, e uint16) bool {
for _, a := range s {
if a == e {
return true
}
}
return false
}
// As we also store the output index of the HTLC for continence
// purposes, we create a small helper function to locate the output
// index of a particular HTLC within the current commitment
// transaction.
locateOutputIndex := func(p *PaymentDescriptor) (uint16, error) {
var idx uint16
var found bool
lnwallet: fix bug that retrieves incorrect pkScript in toChannelDelta() 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.
2017-01-20 14:37:30 +03:00
pkScript := p.theirPrevPkScript
if ourCommit {
pkScript = p.ourPkScript
}
for i, txOut := range c.txn.TxOut {
lnwallet: fix bug that retrieves incorrect pkScript in toChannelDelta() 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.
2017-01-20 14:37:30 +03:00
if bytes.Equal(txOut.PkScript, pkScript) {
if contains(dups[p.RHash], uint16(i)) {
continue
}
found = true
idx = uint16(i)
dups[p.RHash] = append(dups[p.RHash], idx)
break
}
}
if !found {
return 0, fmt.Errorf("could not find a matching output for the HTLC " +
"in the commitment transaction")
}
return idx, nil
}
for _, htlc := range c.outgoingHTLCs {
h := &channeldb.HTLC{
Incoming: false,
Amt: htlc.Amount,
RHash: htlc.RHash,
RefundTimeout: htlc.Timeout,
RevocationDelay: 0,
OutputIndex: index,
}
delta.Htlcs = append(delta.Htlcs, h)
}
for _, htlc := range c.incomingHTLCs {
h := &channeldb.HTLC{
Incoming: true,
Amt: htlc.Amount,
RHash: htlc.RHash,
RefundTimeout: htlc.Timeout,
RevocationDelay: 0,
OutputIndex: index,
}
delta.Htlcs = append(delta.Htlcs, h)
}
return delta, nil
2016-07-06 03:01:55 +03:00
}
// commitmentChain represents a chain of unrevoked commitments. The tail of the
// chain is the latest fully signed, yet unrevoked commitment. Two chains are
// tracked, one for the local node, and another for the remote node. New
2016-10-15 16:18:38 +03:00
// commitments we create locally extend the remote node's chain, and vice
2016-07-06 03:01:55 +03:00
// versa. Commitment chains are allowed to grow to a bounded length, after
// which the tail needs to be "dropped" before new commitments can be received.
// The tail is "dropped" when the owner of the chain sends a revocation for the
// previous tail.
type commitmentChain struct {
// commitments is a linked list of commitments to new states. New
// commitments are added to the end of the chain with increase height.
// Once a commitment transaction is revoked, the tail is incremented,
// freeing up the revocation window for new commitments.
commitments *list.List
// startingHeight is the starting height of this commitment chain on a
// session basis.
startingHeight uint64
}
// newCommitmentChain creates a new commitment chain from an initial height.
func newCommitmentChain(initialHeight uint64) *commitmentChain {
return &commitmentChain{
commitments: list.New(),
startingHeight: initialHeight,
}
}
// addCommitment extends the commitment chain by a single commitment. This
// added commitment represents a state update propsed by either party. Once the
// commitment prior to this commitment is revoked, the commitment becomes the
// new defacto state within the channel.
func (s *commitmentChain) addCommitment(c *commitment) {
s.commitments.PushBack(c)
}
// advanceTail reduces the length of the commitment chain by one. The tail of
// the chain should be advanced once a revocation for the lowest unrevoked
// commitment in the chain is received.
func (s *commitmentChain) advanceTail() {
s.commitments.Remove(s.commitments.Front())
}
// tip returns the latest commitment added to the chain.
func (s *commitmentChain) tip() *commitment {
return s.commitments.Back().Value.(*commitment)
}
// tail returns the lowest unrevoked commitment transaction in the chain.
func (s *commitmentChain) tail() *commitment {
return s.commitments.Front().Value.(*commitment)
}
// updateLog is an append-only log that stores updates to a node's commitment
// chain. This structure can be seen as the "mempool" within Lightning where
// changes are stored before they're committed to the chain. Once an entry has
// been committed in both the local and remote commitment chain, then it can be
// removed from this log.
//
// TODO(roasbeef): create lightning package, move commitment and update to
// package?
// * also move state machine, separate from lnwallet package
// * possible embed updateLog within commitmentChain.
type updateLog struct {
// logIndex is a monotonically increasing integer that tracks the total
// number of update entries ever applied to the log. When sending new
// commitment states, we include all updates up to this index.
logIndex uint64
// ackIndex is a special "pointer" index into the log that tracks the
// position which, up to, all changes have been ACK'd by the remote
// party. When receiving new commitment states, we include all of our
// updates up to this index to restore the commitment view.
ackedIndex uint64
// pendingACKIndex is another special "pointer" index into the log that
// tracks our logIndex value right before we extend the remote party's
// commitment chain. Once we receive an ACK for this changes, then we
// set ackedIndex=pendingAckIndex.
//
// TODO(roasbeef): eventually expand into list when we go back to a
// sliding window format
pendingAckIndex uint64
// List is the updatelog itself, we embed this value so updateLog has
// access to all the method of a list.List.
*list.List
// updateIndex is an index that maps a particular entries index to the
// list element within the list.List above.
updateIndex map[uint64]*list.Element
}
// newUpdateLog creates a new updateLog instance.
func newUpdateLog() *updateLog {
return &updateLog{
List: list.New(),
updateIndex: make(map[uint64]*list.Element),
}
}
// appendUpdate appends a new update to the tip of the updateLog. The entry is
// also added to index accordingly.
func (u *updateLog) appendUpdate(pd *PaymentDescriptor) {
u.updateIndex[u.logIndex] = u.PushBack(pd)
u.logIndex++
}
// lookup attempts to look up an update entry according to it's index value. In
// the case that the entry isn't found, a nil pointer is returned.
func (u *updateLog) lookup(i uint64) *PaymentDescriptor {
return u.updateIndex[i].Value.(*PaymentDescriptor)
}
// remove attempts to remove an entry from the update log. If the entry is
// found, then the entry will be removed from the update log and index.
func (u *updateLog) remove(i uint64) {
entry := u.updateIndex[i]
u.Remove(entry)
delete(u.updateIndex, i)
}
// initiateTransition marks that the caller has extended the commitment chain
// of the remote party with the contents of the updateLog. This function will
// mark the log index value at this point so it can later be marked as ACK'd.
func (u *updateLog) initiateTransition() {
u.pendingAckIndex = u.logIndex
}
// ackTransition updates the internal indexes of the updateLog to mark that the
// last pending state transition has been accepted by the remote party. To do
// so, we mark the prior pendingAckIndex as fully ACK'd.
func (u *updateLog) ackTransition() {
u.ackedIndex = u.pendingAckIndex
u.pendingAckIndex = 0
}
// compactLogs performs garbage collection within the log removing HTLCs which
// have been removed from the point-of-view of the tail of both chains. The
// entries which timeout/settle HTLCs are also removed.
func compactLogs(ourLog, theirLog *updateLog,
localChainTail, remoteChainTail uint64) {
compactLog := func(logA, logB *updateLog) {
var nextA *list.Element
for e := logA.Front(); e != nil; e = nextA {
nextA = e.Next()
htlc := e.Value.(*PaymentDescriptor)
if htlc.EntryType == Add {
continue
}
// If the HTLC hasn't yet been removed from either
// chain, the skip it.
if htlc.removeCommitHeightRemote == 0 ||
htlc.removeCommitHeightLocal == 0 {
continue
}
// Otherwise if the height of the tail of both chains
// is at least the height in which the HTLC was
// removed, then evict the settle/timeout entry along
// with the original add entry.
if remoteChainTail >= htlc.removeCommitHeightRemote &&
localChainTail >= htlc.removeCommitHeightLocal {
logA.remove(htlc.Index)
logB.remove(htlc.ParentIndex)
}
}
}
compactLog(ourLog, theirLog)
compactLog(theirLog, ourLog)
}
2016-07-06 03:01:55 +03:00
// LightningChannel implements the state machine which corresponds to the
// current commitment protocol wire spec. The state machine implemented allows
// for asynchronous fully desynchronized, batched+pipelined updates to
// commitment transactions allowing for a high degree of non-blocking
// bi-directional payment throughput.
//
// In order to allow updates to be fully non-blocking, either side is able to
// create multiple new commitment states up to a pre-determined window size.
// This window size is encoded within InitialRevocationWindow. Before the start
// of a session, both side should send out revocation messages with nil
// preimages in order to populate their revocation window for the remote party.
// Ths method .ExtendRevocationWindow() is used to extend the revocation window
// by a single revocation.
//
// The state machine has for main methods:
// * .SignNextCommitment()
// * Called one one wishes to sign the next commitment, either initiating a
// new state update, or responding to a received commitment.
// * .ReceiveNewCommitment()
2016-07-06 03:01:55 +03:00
// * Called upon receipt of a new commitment from the remote party. If the
// new commitment is valid, then a revocation should immediately be
// generated and sent.
// * .RevokeCurrentCommitment()
// * Revokes the current commitment. Should be called directly after
// receiving a new commitment.
// * .ReceiveRevocation()
// * Processes a revocation from the remote party. If successful creates a
// new defacto broadcastable state.
//
// See the individual comments within the above methods for further details.
type LightningChannel struct {
signer Signer
signDesc *SignDescriptor
channelEvents chainntnfs.ChainNotifier
sync.RWMutex
2016-07-06 03:01:55 +03:00
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
pendingACK bool
2016-07-06 03:01:55 +03:00
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
status channelState
// Capcity is the total capacity of this channel.
Capacity btcutil.Amount
2016-07-06 03:01:55 +03:00
// currentHeight is the current height of our local commitment chain.
// This is also the same as the number of updates to the channel we've
// accepted.
currentHeight uint64
// revocationWindowEdge is the edge of the current revocation window.
// New revocations for prior states created by this channel extend the
// edge of this revocation window. The existence of a revocation window
// allows the remote party to initiate new state updates independently
2016-07-06 03:01:55 +03:00
// until the window is exhausted.
revocationWindowEdge uint64
// usedRevocations is a slice of revocations given to us by the remote
// party that we've used. This slice is extended each time we create a
// new commitment. The front of the slice is popped off once we receive
// a revocation for a prior state. This head element then becomes the
// next set of keys/hashes we expect to be revoked.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
usedRevocations []*lnwire.RevokeAndAck
2016-07-06 03:01:55 +03:00
// revocationWindow is a window of revocations sent to use by the
// remote party, allowing us to create new commitment transactions
// until depleted. The revocations don't contain a valid preimage,
2016-07-06 03:01:55 +03:00
// only an additional key/hash allowing us to create a new commitment
// transaction for the remote node that they are able to revoke. If
// this slice is empty, then we cannot make any new updates to their
// commitment chain.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
revocationWindow []*lnwire.RevokeAndAck
2016-07-06 03:01:55 +03:00
// remoteCommitChain is the remote node's commitment chain. Any new
// commitments we initiate are added to the tip of this chain.
remoteCommitChain *commitmentChain
// localCommitChain is our local commitment chain. Any new commitments
// received are added to the tip of this chain. The tail (or lowest
// height) in this chain is our current accepted state, which we are
// able to broadcast safely.
localCommitChain *commitmentChain
// stateMtx protects concurrent access to the state struct.
stateMtx sync.RWMutex
channelState *channeldb.OpenChannel
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// [local|remote]Log is a (mostly) append-only log storing all the HTLC
2016-07-06 03:01:55 +03:00
// updates to this channel. The log is walked backwards as HTLC updates
// are applied in order to re-construct a commitment transaction from a
// commitment. The log is compacted once a revocation is received.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
localUpdateLog *updateLog
remoteUpdateLog *updateLog
// rHashMap is a map with PaymentHashes pointing to their respective
// PaymentDescriptors. We insert *PaymentDescriptors whenever we
// receive HTLCs. When a state transition happens (settling or
// canceling the HTLC), rHashMap will provide an efficient
// way to lookup the original PaymentDescriptor.
rHashMap map[PaymentHash][]*PaymentDescriptor
LocalDeliveryScript []byte
RemoteDeliveryScript []byte
2016-07-06 03:01:55 +03:00
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// FundingWitnessScript is the witness script for the 2-of-2 multi-sig
// that opened the channel.
FundingWitnessScript []byte
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
fundingTxIn *wire.TxIn
fundingP2WSH []byte
// ForceCloseSignal is a channel that is closed to indicate that a
// local system has initiated a force close by broadcasting the current
// commitment transaction directly on-chain.
ForceCloseSignal chan struct{}
// UnilateralCloseSignal is a channel that is closed to indicate that
// the remote party has performed a unilateral close by broadcasting
// their version of the commitment transaction on-chain.
UnilateralCloseSignal chan struct{}
// ContractBreach is a channel that is used to communicate the data
// necessary to fully resolve the channel in the case that a contract
// breach is detected. A contract breach occurs it is detected that the
// counterparty has broadcast a prior *revoked* state.
ContractBreach chan *BreachRetribution
// LocalFundingKey is the public key under control by the wallet that
// was used for the 2-of-2 funding output which created this channel.
LocalFundingKey *btcec.PublicKey
// RemoteFundingKey is the public key for the remote channel counter
// party which used for the 2-of-2 funding output which created this
// channel.
RemoteFundingKey *btcec.PublicKey
shutdown int32
quit chan struct{}
}
2016-07-06 03:01:55 +03:00
// NewLightningChannel creates a new, active payment channel given an
// implementation of the chain notifier, channel database, and the current
// settled channel state. Throughout state transitions, then channel will
// automatically persist pertinent state to the database in an efficient
// manner.
func NewLightningChannel(signer Signer, events chainntnfs.ChainNotifier,
state *channeldb.OpenChannel) (*LightningChannel, error) {
lc := &LightningChannel{
signer: signer,
channelEvents: events,
currentHeight: state.NumUpdates,
remoteCommitChain: newCommitmentChain(state.NumUpdates),
localCommitChain: newCommitmentChain(state.NumUpdates),
channelState: state,
revocationWindowEdge: state.NumUpdates,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
localUpdateLog: newUpdateLog(),
remoteUpdateLog: newUpdateLog(),
rHashMap: make(map[PaymentHash][]*PaymentDescriptor),
Capacity: state.Capacity,
LocalDeliveryScript: state.OurDeliveryScript,
RemoteDeliveryScript: state.TheirDeliveryScript,
FundingWitnessScript: state.FundingWitnessScript,
ForceCloseSignal: make(chan struct{}),
UnilateralCloseSignal: make(chan struct{}),
ContractBreach: make(chan *BreachRetribution, 1),
LocalFundingKey: state.OurMultiSigKey,
RemoteFundingKey: state.TheirMultiSigKey,
quit: make(chan struct{}),
2016-07-06 03:01:55 +03:00
}
// Initialize both of our chains the current un-revoked commitment for
// each side.
// TODO(roasbeef): add chnneldb.RevocationLogTail method, then init
// their commitment from that as we may be de-synced
2016-07-06 03:01:55 +03:00
initialCommitment := &commitment{
height: lc.currentHeight,
ourBalance: state.OurBalance,
ourMessageIndex: 0,
theirBalance: state.TheirBalance,
theirMessageIndex: 0,
}
2016-07-06 03:01:55 +03:00
lc.localCommitChain.addCommitment(initialCommitment)
lc.remoteCommitChain.addCommitment(initialCommitment)
// If we're restarting from a channel with history, then restore the
// update in-memory update logs to that of the prior state.
if lc.currentHeight != 0 {
lc.restoreStateLogs()
}
// Create the sign descriptor which we'll be using very frequently to
// request a signature for the 2-of-2 multi-sig from the signer in
// order to complete channel state transitions.
fundingPkScript, err := witnessScriptHash(state.FundingWitnessScript)
if err != nil {
return nil, err
}
lc.fundingTxIn = wire.NewTxIn(state.FundingOutpoint, nil, nil)
2016-07-06 03:01:55 +03:00
lc.fundingP2WSH = fundingPkScript
lc.signDesc = &SignDescriptor{
PubKey: lc.channelState.OurMultiSigKey,
WitnessScript: lc.channelState.FundingWitnessScript,
Output: &wire.TxOut{
PkScript: lc.fundingP2WSH,
Value: int64(lc.channelState.Capacity),
},
HashType: txscript.SigHashAll,
InputIndex: 0,
}
// We'll only launch a close observer if the ChainNotifier
// implementation is non-nil. Passing a nil value indicates that the
// channel shouldn't be actively watched for.
if lc.channelEvents != nil {
// Register for a notification to be dispatched if the funding
// outpoint has been spent. This indicates that either us or
// the remote party has broadcasted a commitment transaction
// on-chain.
fundingOut := &lc.fundingTxIn.PreviousOutPoint
channelCloseNtfn, err := lc.channelEvents.RegisterSpendNtfn(fundingOut)
if err != nil {
return nil, err
}
// Launch the close observer which will vigilantly watch the
// network for any broadcasts the current or prior commitment
// transactions, taking action accordingly.
go lc.closeObserver(channelCloseNtfn)
}
return lc, nil
}
// BreachRetribution contains all the data necessary to bring a channel
// counterparty to justice claiming ALL lingering funds within the channel in
// the scenario that they broadcast a revoked commitment transaction. A
// BreachRetribution is created by the closeObserver if it detects an
// uncooperative close of the channel which uses a revoked commitment
// transaction. The BreachRetribution is then sent over the ContractBreach
// channel in order to allow the subscriber of the channel to dispatch justice.
type BreachRetribution struct {
// BreachTransaction is the transaction which breached the channel
// contract by spending from the funding multi-sig with a revoked
// commitment transaction.
BreachTransaction *wire.MsgTx
// RevokedStateNum is the revoked state number which was broadcast.
RevokedStateNum uint64
// PendingHTLCs is a slice of the HTLCs which were pending at this
// point within the channel's history transcript.
PendingHTLCs []*channeldb.HTLC
// LocalOutputSignDesc is a SignDescriptor which is capable of
// generating the signature necessary to sweep the output within the
// BreachTransaction that pays directly us.
LocalOutputSignDesc *SignDescriptor
// LocalOutpoint is the outpoint of the output paying to us (the local
// party) within the breach transaction.
LocalOutpoint wire.OutPoint
// RemoteOutputSignDesc is a SignDescriptor which is capable of
// generating the signature required to claim the funds as described
// within the revocation clause of the remote party's commitment
// output.
RemoteOutputSignDesc *SignDescriptor
// RemoteOutpoint is the output of the output paying to the remote
// party within the breach transaction.
RemoteOutpoint wire.OutPoint
}
// newBreachRetribution creates a new fully populated BreachRetribution for the
// passed channel, at a particular revoked state number, and one which targets
// the passed commitment transaction.
func newBreachRetribution(chanState *channeldb.OpenChannel, stateNum uint64,
broadcastCommitment *wire.MsgTx) (*BreachRetribution, error) {
commitHash := broadcastCommitment.TxHash()
// Query the on-disk revocation log for the snapshot which was recorded
// at this particular state num.
revokedSnapshot, err := chanState.FindPreviousState(stateNum)
if err != nil {
return nil, err
}
// With the state number broadcast known, we can now derive/restore the
// proper revocation preimage necessary to sweep the remote party's
// output.
revocationPreimage, err := chanState.RevocationStore.LookUp(stateNum)
if err != nil {
return nil, err
}
// Once we derive the revocation leaf, we can then re-create the
// revocation public key used within this state. This is needed in
// order to create the proper script below.
localCommitKey := chanState.OurCommitKey
revocationKey := DeriveRevocationPubkey(localCommitKey, revocationPreimage[:])
remoteCommitkey := chanState.TheirCommitKey
remoteDelay := chanState.RemoteCsvDelay
// Next, reconstruct the scripts as they were present at this state
// number so we can have the proper witness script to sign and include
// within the final witness.
remotePkScript, err := commitScriptToSelf(remoteDelay,
remoteCommitkey, revocationKey)
if err != nil {
return nil, err
}
remoteWitnessHash, err := witnessScriptHash(remotePkScript)
if err != nil {
return nil, err
}
localPkScript, err := commitScriptUnencumbered(localCommitKey)
if err != nil {
return nil, err
}
// In order to fully populate the breach retribution struct, we'll need
// to find the exact index of the local+remote commitment outputs.
localOutpoint := wire.OutPoint{
Hash: commitHash,
}
remoteOutpoint := wire.OutPoint{
Hash: commitHash,
}
for i, txOut := range broadcastCommitment.TxOut {
switch {
case bytes.Equal(txOut.PkScript, localPkScript):
localOutpoint.Index = uint32(i)
case bytes.Equal(txOut.PkScript, remoteWitnessHash):
remoteOutpoint.Index = uint32(i)
}
}
// Finally, with all the necessary data constructed, we can create the
// BreachRetribution struct which houses all the data necessary to
// swiftly bring justice to the cheating remote party.
return &BreachRetribution{
BreachTransaction: broadcastCommitment,
RevokedStateNum: stateNum,
PendingHTLCs: revokedSnapshot.Htlcs,
LocalOutpoint: localOutpoint,
LocalOutputSignDesc: &SignDescriptor{
PubKey: localCommitKey,
Output: &wire.TxOut{
PkScript: localPkScript,
Value: int64(revokedSnapshot.LocalBalance),
},
HashType: txscript.SigHashAll,
},
RemoteOutpoint: remoteOutpoint,
RemoteOutputSignDesc: &SignDescriptor{
PubKey: localCommitKey,
PrivateTweak: revocationPreimage[:],
WitnessScript: remotePkScript,
Output: &wire.TxOut{
PkScript: remoteWitnessHash,
Value: int64(revokedSnapshot.RemoteBalance),
},
HashType: txscript.SigHashAll,
},
}, nil
}
// closeObserver is a goroutine which watches the network for any spends of the
// multi-sig funding output. A spend from the multi-sig output may occur under
// the following three scenarios: a cooperative close, a unilateral close, and
// a uncooperative contract breaching close. In the case of the last scenario a
// BreachRetribution struct is created and sent over the ContractBreach channel
// notifying subscribers that the counterparty has violated the condition of
// the channel by broadcasting a revoked prior state.
//
// NOTE: This MUST be run as a goroutine.
func (lc *LightningChannel) closeObserver(channelCloseNtfn *chainntnfs.SpendEvent) {
walletLog.Infof("Close observer for ChannelPoint(%v) active",
lc.channelState.ChanID)
var (
commitSpend *chainntnfs.SpendDetail
ok bool
)
select {
// If the daemon is shutting down, then this notification channel will
// be closed, so check the second read-value to avoid a false positive.
case commitSpend, ok = <-channelCloseNtfn.Spend:
if !ok {
return
}
// Otherwise, we've beeen signalled to bail out early by the
// caller/maintainer of this channel.
case <-lc.quit:
// As we're exiting before the spend notification has been
// triggered, we'll cancel the notification intent so the
// ChainNotiifer can free up the resources.
channelCloseNtfn.Cancel()
return
}
// If we've already initiated a local cooperative or unilateral close
// locally, then we have nothing more to do.
lc.RLock()
if lc.status == channelClosed || lc.status == channelDispute ||
lc.status == channelClosing {
lc.RUnlock()
return
}
lc.RUnlock()
lc.Lock()
defer lc.Unlock()
walletLog.Warnf("Unprompted commitment broadcast for ChannelPoint(%v) "+
"detected!", lc.channelState.ChanID)
// Otherwise, the remote party might have broadcast a prior revoked
// state...!!!
commitTxBroadcast := commitSpend.SpendingTx
// Decode the state hint encoded within the commitment transaction to
// determine if this is a revoked state or not.
obsfucator := lc.channelState.StateHintObsfucator
broadcastStateNum := GetStateNumHint(commitTxBroadcast, obsfucator)
currentStateNum := lc.currentHeight
switch {
// If state number spending transaction matches the current latest
// state, then they've initiated a unilateral close. So we'll trigger
// the unilateral close signal so subscribers can clean up the state as
// necessary.
//
// We'll also handle the case of the remote party broadcasting their
// commitment transaction which is one height above ours. This case an
// arise when we initiate a state transition, but the remote party has
// a fail crash _after_ accepting the new state, but _before_ sending
// their signature to us.
case broadcastStateNum >= currentStateNum:
walletLog.Infof("Unilateral close of ChannelPoint(%v) "+
"detected", lc.channelState.ChanID)
// As we've deleted that the channel has been closed,
// immediately delete the state from disk, creating a close
// summary for future usage by related sub-systems.
if err := lc.DeleteState(); err != nil {
walletLog.Errorf("unable to delete channel state: %v",
err)
}
// Notify any subscribers that we've detected a unilateral
// commitment transaction broadcast.
close(lc.UnilateralCloseSignal)
// If the state number broadcast is lower than the remote node's
// current un-revoked height, then THEY'RE ATTEMPTING TO VIOLATE THE
// CONTRACT LAID OUT WITHIN THE PAYMENT CHANNEL. Therefore we close
// the signal indicating a revoked broadcast to allow subscribers to
// swiftly dispatch justice!!!
case broadcastStateNum < currentStateNum:
walletLog.Warnf("Remote peer has breached the channel "+
"contract for ChannelPoint(%v). Revoked state #%v was "+
"broadcast!!!", lc.channelState.ChanID,
broadcastStateNum)
// Create a new reach retribution struct which contains all the
// data needed to swiftly bring the cheating peer to justice.
retribution, err := newBreachRetribution(lc.channelState,
broadcastStateNum, commitTxBroadcast)
if err != nil {
walletLog.Errorf("unable to create breach retribution: %v", err)
return
}
walletLog.Debugf("Punishment breach retribution created: %v",
spew.Sdump(retribution))
// Finally, send the retribution struct over the contract beach
// channel to allow the observer the use the breach retribution
// to sweep ALL funds.
lc.ContractBreach <- retribution
}
}
// Stop gracefully shuts down any active goroutines spawned by the
// LightningChannel during regular duties.
func (lc *LightningChannel) Stop() {
if !atomic.CompareAndSwapInt32(&lc.shutdown, 0, 1) {
return
}
close(lc.quit)
}
// restoreStateLogs runs through the current locked-in HTLCs from the point of
// view of the channel and insert corresponding log entries (both local and
// remote) for each HTLC read from disk. This method is required sync the
// in-memory state of the state machine with that read from persistent storage.
func (lc *LightningChannel) restoreStateLogs() error {
var pastHeight uint64
if lc.currentHeight > 0 {
pastHeight = lc.currentHeight - 1
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
var ourCounter, theirCounter uint64
for _, htlc := range lc.channelState.Htlcs {
// TODO(roasbeef): set isForwarded to false for all? need to
// persist state w.r.t to if forwarded or not, or can
// inadvertently trigger replays
pd := &PaymentDescriptor{
RHash: htlc.RHash,
Timeout: htlc.RefundTimeout,
Amount: htlc.Amt,
EntryType: Add,
addCommitHeightRemote: pastHeight,
addCommitHeightLocal: pastHeight,
}
if !htlc.Incoming {
pd.Index = ourCounter
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
lc.localUpdateLog.appendUpdate(pd)
ourCounter++
} else {
pd.Index = theirCounter
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
lc.remoteUpdateLog.appendUpdate(pd)
lc.rHashMap[pd.RHash] = append(lc.rHashMap[pd.RHash], pd)
theirCounter++
}
}
lc.localCommitChain.tail().ourMessageIndex = ourCounter
lc.localCommitChain.tail().theirMessageIndex = theirCounter
lc.remoteCommitChain.tail().ourMessageIndex = ourCounter
lc.remoteCommitChain.tail().theirMessageIndex = theirCounter
return nil
}
// htlcView represents the "active" HTLCs at a particular point within the
// history of the HTLC update log.
type htlcView struct {
ourUpdates []*PaymentDescriptor
theirUpdates []*PaymentDescriptor
}
2016-07-06 03:01:55 +03:00
// fetchHTLCView returns all the candidate HTLC updates which should be
// considered for inclusion within a commitment based on the passed HTLC log
// indexes.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) fetchHTLCView(theirLogIndex, ourLogIndex uint64) *htlcView {
var ourHTLCs []*PaymentDescriptor
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
for e := lc.localUpdateLog.Front(); e != nil; e = e.Next() {
htlc := e.Value.(*PaymentDescriptor)
2016-07-06 03:01:55 +03:00
// This HTLC is active from this point-of-view iff the log
// index of the state update is below the specified index in
// our update log.
if htlc.Index < ourLogIndex {
ourHTLCs = append(ourHTLCs, htlc)
}
}
var theirHTLCs []*PaymentDescriptor
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
for e := lc.remoteUpdateLog.Front(); e != nil; e = e.Next() {
2016-07-06 03:01:55 +03:00
htlc := e.Value.(*PaymentDescriptor)
// If this is an incoming HTLC, then it is only active from
// this point-of-view if the index of the HTLC addition in
// their log is below the specified view index.
if htlc.Index < theirLogIndex {
theirHTLCs = append(theirHTLCs, htlc)
2016-07-06 03:01:55 +03:00
}
}
return &htlcView{
ourUpdates: ourHTLCs,
theirUpdates: theirHTLCs,
}
2016-07-06 03:01:55 +03:00
}
2016-07-06 03:01:55 +03:00
// fetchCommitmentView returns a populated commitment which expresses the state
// of the channel from the point of view of a local or remote chain, evaluating
// the HTLC log up to the passed indexes. This function is used to construct
// both local and remote commitment transactions in order to sign or verify new
// commitment updates. A fully populated commitment is returned which reflects
// the proper balances for both sides at this point in the commitment chain.
func (lc *LightningChannel) fetchCommitmentView(remoteChain bool,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
ourLogIndex, theirLogIndex uint64, revocationKey *btcec.PublicKey,
2016-07-06 03:01:55 +03:00
revocationHash [32]byte) (*commitment, error) {
var commitChain *commitmentChain
if remoteChain {
commitChain = lc.remoteCommitChain
} else {
commitChain = lc.localCommitChain
}
2016-07-06 03:01:55 +03:00
// TODO(roasbeef): don't assume view is always fetched from tip?
var ourBalance, theirBalance btcutil.Amount
if commitChain.tip() == nil {
ourBalance = lc.channelState.OurBalance
theirBalance = lc.channelState.TheirBalance
} else {
ourBalance = commitChain.tip().ourBalance
theirBalance = commitChain.tip().theirBalance
}
nextHeight := commitChain.tip().height + 1
// Run through all the HTLCs that will be covered by this transaction
2016-07-06 03:01:55 +03:00
// in order to update their commitment addition height, and to adjust
// the balances on the commitment transaction accordingly.
// TODO(roasbeef): error if log empty?
htlcView := lc.fetchHTLCView(theirLogIndex, ourLogIndex)
filteredHTLCView := lc.evaluateHTLCView(htlcView, &ourBalance, &theirBalance,
nextHeight, remoteChain)
2016-07-06 03:01:55 +03:00
var selfKey *btcec.PublicKey
var remoteKey *btcec.PublicKey
var delay uint32
var delayBalance, p2wkhBalance, dustLimit btcutil.Amount
2016-07-06 03:01:55 +03:00
if remoteChain {
selfKey = lc.channelState.TheirCommitKey
remoteKey = lc.channelState.OurCommitKey
2016-07-06 03:01:55 +03:00
delay = lc.channelState.RemoteCsvDelay
delayBalance = theirBalance
p2wkhBalance = ourBalance
dustLimit = lc.channelState.TheirDustLimit
2016-07-06 03:01:55 +03:00
} else {
selfKey = lc.channelState.OurCommitKey
2016-07-06 03:01:55 +03:00
remoteKey = lc.channelState.TheirCommitKey
delay = lc.channelState.LocalCsvDelay
delayBalance = ourBalance
p2wkhBalance = theirBalance
dustLimit = lc.channelState.OurDustLimit
2016-07-06 03:01:55 +03:00
}
2016-07-06 03:01:55 +03:00
// Generate a new commitment transaction with all the latest
// unsettled/un-timed out HTLCs.
2016-07-06 03:01:55 +03:00
ourCommitTx := !remoteChain
commitTx, err := CreateCommitTx(lc.fundingTxIn, selfKey, remoteKey,
revocationKey, delay, delayBalance, p2wkhBalance, dustLimit)
if err != nil {
return nil, err
}
for _, htlc := range filteredHTLCView.ourUpdates {
if htlc.Amount < dustLimit {
continue
}
err := lc.addHTLC(commitTx, ourCommitTx, htlc,
revocationHash, delay, false)
if err != nil {
return nil, err
}
}
for _, htlc := range filteredHTLCView.theirUpdates {
if htlc.Amount < dustLimit {
continue
}
err := lc.addHTLC(commitTx, ourCommitTx, htlc,
revocationHash, delay, true)
if err != nil {
2016-07-06 03:01:55 +03:00
return nil, err
}
}
// Set the state hint of the commitment transaction to facilitate
// quickly recovering the necessary penalty state in the case of an
// uncooperative broadcast.
obsfucator := lc.channelState.StateHintObsfucator
stateNum := nextHeight
if err := SetStateNumHint(commitTx, stateNum, obsfucator); err != nil {
return nil, err
}
// Sort the transactions according to the agreed upon canonical
2016-07-06 03:01:55 +03:00
// ordering. This lets us skip sending the entire transaction over,
// instead we'll just send signatures.
txsort.InPlaceSort(commitTx)
return &commitment{
txn: commitTx,
height: nextHeight,
ourBalance: ourBalance,
ourMessageIndex: ourLogIndex,
theirMessageIndex: theirLogIndex,
theirBalance: theirBalance,
outgoingHTLCs: filteredHTLCView.ourUpdates,
incomingHTLCs: filteredHTLCView.theirUpdates,
2016-07-06 03:01:55 +03:00
}, nil
}
// evaluateHTLCView processes all update entries in both HTLC update logs,
// producing a final view which is the result of properly applying all adds,
// settles, and timeouts found in both logs. The resulting view returned
// reflects the current state of HTLCs within the remote or local commitment
// chain.
func (lc *LightningChannel) evaluateHTLCView(view *htlcView, ourBalance,
theirBalance *btcutil.Amount, nextHeight uint64, remoteChain bool) *htlcView {
newView := &htlcView{}
// We use two maps, one for the local log and one for the remote log to
// keep track of which entries we need to skip when creating the final
// htlc view. We skip an entry whenever we find a settle or a timeout
// modifying an entry.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
skipUs := make(map[uint64]struct{})
skipThem := make(map[uint64]struct{})
// First we run through non-add entries in both logs, populating the
// skip sets and mutating the current chain state (crediting balances, etc) to
// reflect the settle/timeout entry encountered.
for _, entry := range view.ourUpdates {
if entry.EntryType == Add {
continue
}
// If we're settling in inbound HTLC, and it hasn't been
// processed, yet, the increment our state tracking the total
// number of satoshis we've received within the channel.
if entry.EntryType == Settle && !remoteChain &&
entry.removeCommitHeightLocal == 0 {
lc.channelState.TotalSatoshisReceived += uint64(entry.Amount)
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
addEntry := lc.remoteUpdateLog.lookup(entry.ParentIndex)
skipThem[addEntry.Index] = struct{}{}
processRemoveEntry(entry, ourBalance, theirBalance,
nextHeight, remoteChain, true)
2016-07-06 03:01:55 +03:00
}
for _, entry := range view.theirUpdates {
if entry.EntryType == Add {
continue
}
// If the remote party is settling one of our outbound HTLC's,
// and it hasn't been processed, yet, the increment our state
// tracking the total number of satoshis we've sent within the
// channel.
if entry.EntryType == Settle && !remoteChain &&
entry.removeCommitHeightLocal == 0 {
lc.channelState.TotalSatoshisSent += uint64(entry.Amount)
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
addEntry := lc.localUpdateLog.lookup(entry.ParentIndex)
skipUs[addEntry.Index] = struct{}{}
processRemoveEntry(entry, ourBalance, theirBalance,
nextHeight, remoteChain, false)
}
// Next we take a second pass through all the log entries, skipping any
// settled HTLCs, and debiting the chain state balance due to any
// newly added HTLCs.
for _, entry := range view.ourUpdates {
isAdd := entry.EntryType == Add
if _, ok := skipUs[entry.Index]; !isAdd || ok {
continue
}
2016-07-06 03:01:55 +03:00
processAddEntry(entry, ourBalance, theirBalance, nextHeight,
remoteChain, false)
newView.ourUpdates = append(newView.ourUpdates, entry)
}
for _, entry := range view.theirUpdates {
isAdd := entry.EntryType == Add
if _, ok := skipThem[entry.Index]; !isAdd || ok {
continue
}
processAddEntry(entry, ourBalance, theirBalance, nextHeight,
remoteChain, true)
newView.theirUpdates = append(newView.theirUpdates, entry)
}
return newView
2016-07-06 03:01:55 +03:00
}
// processAddEntry evaluates the effect of an add entry within the HTLC log.
2016-07-06 03:01:55 +03:00
// If the HTLC hasn't yet been committed in either chain, then the height it
// was committed is updated. Keeping track of this inclusion height allows us to
2016-07-06 03:01:55 +03:00
// later compact the log once the change is fully committed in both chains.
func processAddEntry(htlc *PaymentDescriptor, ourBalance, theirBalance *btcutil.Amount,
nextHeight uint64, remoteChain bool, isIncoming bool) {
2016-07-06 03:01:55 +03:00
// If we're evaluating this entry for the remote chain (to create/view
// a new commitment), then we'll may be updating the height this entry
// was added to the chain. Otherwise, we may be updating the entry's
// height w.r.t the local chain.
var addHeight *uint64
if remoteChain {
addHeight = &htlc.addCommitHeightRemote
} else {
addHeight = &htlc.addCommitHeightLocal
}
2016-07-06 03:01:55 +03:00
if *addHeight != 0 {
return
}
if isIncoming {
2016-07-06 03:01:55 +03:00
// If this is a new incoming (un-committed) HTLC, then we need
// to update their balance accordingly by subtracting the
// amount of the HTLC that are funds pending.
*theirBalance -= htlc.Amount
} else {
// Similarly, we need to debit our balance if this is an out
2016-07-06 03:01:55 +03:00
// going HTLC to reflect the pending balance.
*ourBalance -= htlc.Amount
}
2016-07-06 03:01:55 +03:00
*addHeight = nextHeight
}
2016-07-06 03:01:55 +03:00
// processRemoveEntry processes a log entry which settles or timesout a
// previously added HTLC. If the removal entry has already been processed, it
// is skipped.
func processRemoveEntry(htlc *PaymentDescriptor, ourBalance,
theirBalance *btcutil.Amount, nextHeight uint64,
remoteChain bool, isIncoming bool) {
2016-07-06 03:01:55 +03:00
var removeHeight *uint64
if remoteChain {
removeHeight = &htlc.removeCommitHeightRemote
} else {
removeHeight = &htlc.removeCommitHeightLocal
}
2016-07-06 03:01:55 +03:00
// Ignore any removal entries which have already been processed.
if *removeHeight != 0 {
return
}
2016-07-06 03:01:55 +03:00
switch {
// If an incoming HTLC is being settled, then this means that we've
// received the preimage either from another subsystem, or the
2016-07-06 03:01:55 +03:00
// upstream peer in the route. Therefore, we increase our balance by
// the HTLC amount.
case isIncoming && htlc.EntryType == Settle:
2016-07-06 03:01:55 +03:00
*ourBalance += htlc.Amount
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// Otherwise, this HTLC is being failed out, therefore the value of the
2016-07-06 03:01:55 +03:00
// HTLC should return to the remote party.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
case isIncoming && htlc.EntryType == Fail:
2016-07-06 03:01:55 +03:00
*theirBalance += htlc.Amount
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
2016-07-06 03:01:55 +03:00
// If an outgoing HTLC is being settled, then this means that the
// downstream party resented the preimage or learned of it via a
// downstream peer. In either case, we credit their settled value with
// the value of the HTLC.
case !isIncoming && htlc.EntryType == Settle:
2016-07-06 03:01:55 +03:00
*theirBalance += htlc.Amount
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// Otherwise, one of our outgoing HTLC's has timed out, so the value of
// the HTLC should be returned to our settled balance.
case !isIncoming && htlc.EntryType == Fail:
2016-07-06 03:01:55 +03:00
*ourBalance += htlc.Amount
}
2016-07-06 03:01:55 +03:00
*removeHeight = nextHeight
}
2016-07-06 03:01:55 +03:00
// SignNextCommitment signs a new commitment which includes any previous
// unsettled HTLCs, any new HTLCs, and any modifications to prior HTLCs
// committed in previous commitment updates. Signing a new commitment
// decrements the available revocation window by 1. After a successful method
// call, the remote party's commitment chain is extended by a new commitment
// which includes all updates to the HTLC log prior to this method invocation.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) SignNextCommitment() ([]byte, error) {
lc.Lock()
defer lc.Unlock()
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// If we're awaiting an ACK to a commitment signature, then we're
// unable to create new states as we don't have any revocations we can
// use.
if lc.pendingACK {
return nil, ErrNoWindow
2016-11-23 11:36:55 +03:00
}
2016-07-06 03:01:55 +03:00
// Ensure that we have enough unused revocation hashes given to us by the
// remote party. If the set is empty, then we're unable to create a new
// state unless they first revoke a prior commitment transaction.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// TODO(roasbeef): remove now due to above?
2016-07-06 03:01:55 +03:00
if len(lc.revocationWindow) == 0 ||
len(lc.usedRevocations) == InitialRevocationWindow {
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
return nil, ErrNoWindow
}
// Before we extend this new commitment to the remote commitment chain,
// ensure that we aren't violating any of the constraints the remote
// party set up when we initially set up the channel. If we are, then
// we'll abort this state transition.
err := lc.validateCommitmentSanity(lc.remoteUpdateLog.ackedIndex,
lc.localUpdateLog.logIndex, false)
if err != nil {
return nil, err
2016-07-06 03:01:55 +03:00
}
2016-07-06 03:01:55 +03:00
// Grab the next revocation hash and key to use for this new commitment
// transaction, if no errors occur then this revocation tuple will be
// moved to the used set.
nextRevocation := lc.revocationWindow[0]
remoteRevocationKey := nextRevocation.NextRevocationKey
remoteRevocationHash := nextRevocation.NextRevocationHash
// Create a new commitment view which will calculate the evaluated
// state of the remote node's new commitment including our latest added
// HTLCs. The view includes the latest balances for both sides on the
2016-07-06 03:01:55 +03:00
// remote node's chain, and also update the addition height of any new
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// HTLC log entries. When we creating a new remote view, we include
// _all_ of our changes (pending or committed) but only the remote
// node's changes up to the last change we've ACK'd.
newCommitView, err := lc.fetchCommitmentView(true, lc.localUpdateLog.logIndex,
lc.remoteUpdateLog.ackedIndex, remoteRevocationKey, remoteRevocationHash)
if err != nil {
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
return nil, err
}
walletLog.Tracef("ChannelPoint(%v): extending remote chain to height %v",
lc.channelState.ChanID, newCommitView.height)
walletLog.Tracef("ChannelPoint(%v): remote chain: our_balance=%v, "+
"their_balance=%v, commit_tx: %v", lc.channelState.ChanID,
newCommitView.ourBalance, newCommitView.theirBalance,
newLogClosure(func() string {
return spew.Sdump(newCommitView.txn)
}))
2016-07-06 03:01:55 +03:00
// Sign their version of the new commitment transaction.
lc.signDesc.SigHashes = txscript.NewTxSigHashes(newCommitView.txn)
sig, err := lc.signer.SignOutputRaw(newCommitView.txn, lc.signDesc)
2016-07-06 03:01:55 +03:00
if err != nil {
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
return nil, err
2016-07-06 03:01:55 +03:00
}
2016-07-06 03:01:55 +03:00
// Extend the remote commitment chain by one with the addition of our
// latest commitment update.
lc.remoteCommitChain.addCommitment(newCommitView)
2016-07-06 03:01:55 +03:00
// Move the now used revocation hash from the unused set to the used set.
// We only do this at the end, as we know at this point the procedure will
// succeed without any errors.
lc.usedRevocations = append(lc.usedRevocations, nextRevocation)
lc.revocationWindow[0] = nil // Avoid a GC leak.
lc.revocationWindow = lc.revocationWindow[1:]
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// As we've just created a new update for the remote commitment chain,
// we set the bool indicating that we're waiting for an ACK to our new
// changes.
lc.pendingACK = true
// Additionally, we'll remember our log index at this point, so we can
// properly track which changes have been ACK'd.
lc.localUpdateLog.initiateTransition()
return sig, nil
2016-07-06 03:01:55 +03:00
}
2016-11-23 11:36:55 +03:00
// validateCommitmentSanity is used to validate that on current state the commitment
// transaction is valid in terms of propagating it over Bitcoin network, and
// also that all outputs are meet Bitcoin spec requirements and they are
// spendable.
func (lc *LightningChannel) validateCommitmentSanity(theirLogCounter,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
ourLogCounter uint64, prediction bool) error {
2016-11-23 11:36:55 +03:00
htlcCount := 0
if prediction {
htlcCount++
}
// Run through all the HTLCs that will be covered by this transaction
2016-11-23 11:36:55 +03:00
// in order to calculate theirs count.
htlcView := lc.fetchHTLCView(theirLogCounter, ourLogCounter)
for _, entry := range htlcView.ourUpdates {
if entry.EntryType == Add {
htlcCount++
} else {
htlcCount--
}
}
for _, entry := range htlcView.theirUpdates {
if entry.EntryType == Add {
htlcCount++
} else {
htlcCount--
}
}
if htlcCount > MaxHTLCNumber {
return ErrMaxHTLCNumber
}
return nil
}
// ReceiveNewCommitment process a signature for a new commitment state sent by
2016-07-06 03:01:55 +03:00
// the remote party. This method will should be called in response to the
// remote party initiating a new change, or when the remote party sends a
// signature fully accepting a new state we've initiated. If we are able to
// successfully validate the signature, then the generated commitment is added
2016-07-06 03:01:55 +03:00
// to our local commitment chain. Once we send a revocation for our prior
// state, then this newly added commitment becomes our current accepted channel
// state.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) ReceiveNewCommitment(rawSig []byte) error {
lc.Lock()
defer lc.Unlock()
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// Ensure that this new local update from the remote node respects all
// the constraints we specified during initial channel setup. If not,
// then we'll abort the channel as they've violated our constraints.
err := lc.validateCommitmentSanity(lc.remoteUpdateLog.logIndex,
lc.localUpdateLog.ackedIndex, false)
2016-11-23 11:36:55 +03:00
if err != nil {
return err
}
2016-07-06 03:01:55 +03:00
theirCommitKey := lc.channelState.TheirCommitKey
theirMultiSigKey := lc.channelState.TheirMultiSigKey
// We're receiving a new commitment which attempts to extend our local
2016-07-06 03:01:55 +03:00
// commitment chain height by one, so fetch the proper revocation to
// derive the key+hash needed to construct the new commitment view and
// state.
nextHeight := lc.currentHeight + 1
revocation, err := lc.channelState.RevocationProducer.AtIndex(nextHeight)
2016-07-06 03:01:55 +03:00
if err != nil {
return err
}
revocationKey := DeriveRevocationPubkey(theirCommitKey, revocation[:])
revocationHash := sha256.Sum256(revocation[:])
2016-07-06 03:01:55 +03:00
// With the revocation information calculated, construct the new
// commitment view which includes all the entries we know of in their
// HTLC log, and up to ourLogIndex in our HTLC log.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
localCommitmentView, err := lc.fetchCommitmentView(false,
lc.localUpdateLog.ackedIndex, lc.remoteUpdateLog.logIndex,
revocationKey, revocationHash)
2016-07-06 03:01:55 +03:00
if err != nil {
return err
}
walletLog.Tracef("ChannelPoint(%v): extending local chain to height %v",
lc.channelState.ChanID, localCommitmentView.height)
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
walletLog.Tracef("ChannelPoint(%v): local chain: our_balance=%v, "+
"their_balance=%v, commit_tx: %v", lc.channelState.ChanID,
localCommitmentView.ourBalance, localCommitmentView.theirBalance,
newLogClosure(func() string {
return spew.Sdump(localCommitmentView.txn)
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
}),
)
2016-07-06 03:01:55 +03:00
// Construct the sighash of the commitment transaction corresponding to
// this newly proposed state update.
localCommitTx := localCommitmentView.txn
multiSigScript := lc.channelState.FundingWitnessScript
2016-07-06 03:01:55 +03:00
hashCache := txscript.NewTxSigHashes(localCommitTx)
sigHash, err := txscript.CalcWitnessSigHash(multiSigScript, hashCache,
txscript.SigHashAll, localCommitTx, 0, int64(lc.channelState.Capacity))
if err != nil {
// TODO(roasbeef): fetchview has already mutated the HTLCs...
2016-07-06 03:01:55 +03:00
// * need to either roll-back, or make pure
return err
}
2016-07-06 03:01:55 +03:00
// Ensure that the newly constructed commitment state has a valid
// signature.
theirMultiSigKey.Curve = btcec.S256()
2016-07-06 03:01:55 +03:00
sig, err := btcec.ParseSignature(rawSig, btcec.S256())
if err != nil {
return err
} else if !sig.Verify(sigHash, theirMultiSigKey) {
return fmt.Errorf("invalid commitment signature")
}
2016-07-06 03:01:55 +03:00
// The signature checks out, so we can now add the new commitment to
// our local commitment chain.
localCommitmentView.sig = rawSig
lc.localCommitChain.addCommitment(localCommitmentView)
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// Finally we'll keep track of the current pending index for the remote
// party so we can ACK up to this value once we revoke our current
// commitment.
lc.remoteUpdateLog.initiateTransition()
return nil
}
// PendingUpdates returns a boolean value reflecting if there are any pending
// updates which need to be committed. The state machine has pending updates if
// the local log index on the local and remote chain tip aren't identical. This
// indicates that either we have pending updates they need to commit, or vice
// versa.
func (lc *LightningChannel) PendingUpdates() bool {
lc.RLock()
defer lc.RUnlock()
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// TODO(roasbeef): instead check our current counter?
fullySynced := (lc.localCommitChain.tip().ourMessageIndex ==
lc.remoteCommitChain.tip().ourMessageIndex)
return !fullySynced
}
2016-07-06 03:01:55 +03:00
// RevokeCurrentCommitment revokes the next lowest unrevoked commitment
// transaction in the local commitment chain. As a result the edge of our
// revocation window is extended by one, and the tail of our local commitment
// chain is advanced by a single commitment. This now lowest unrevoked
// commitment becomes our currently accepted state within the channel.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) RevokeCurrentCommitment() (*lnwire.RevokeAndAck, error) {
lc.Lock()
defer lc.Unlock()
2016-07-06 03:01:55 +03:00
theirCommitKey := lc.channelState.TheirCommitKey
// Now that we've accept a new state transition, we send the remote
// party the revocation for our current commitment state.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
revocationMsg := &lnwire.RevokeAndAck{}
currentRevocation, err := lc.channelState.RevocationProducer.AtIndex(lc.currentHeight)
if err != nil {
return nil, err
}
2016-07-06 03:01:55 +03:00
copy(revocationMsg.Revocation[:], currentRevocation[:])
2016-07-06 03:01:55 +03:00
// Along with this revocation, we'll also send an additional extension
// to our revocation window to the remote party.
lc.revocationWindowEdge++
revocationEdge, err := lc.channelState.RevocationProducer.AtIndex(lc.revocationWindowEdge)
2016-07-06 03:01:55 +03:00
if err != nil {
return nil, err
}
revocationMsg.NextRevocationKey = DeriveRevocationPubkey(theirCommitKey,
2016-07-06 03:01:55 +03:00
revocationEdge[:])
revocationMsg.NextRevocationHash = sha256.Sum256(revocationEdge[:])
walletLog.Tracef("ChannelPoint(%v): revoking height=%v, now at height=%v, window_edge=%v",
lc.channelState.ChanID, lc.localCommitChain.tail().height,
lc.currentHeight+1, lc.revocationWindowEdge)
2016-07-06 03:01:55 +03:00
// Advance our tail, as we've revoked our previous state.
lc.localCommitChain.advanceTail()
lc.currentHeight++
// Additionally, generate a channel delta for this state transition for
// persistent storage.
2016-07-06 03:01:55 +03:00
tail := lc.localCommitChain.tail()
lnwallet: fix bug that retrieves incorrect pkScript in toChannelDelta() 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.
2017-01-20 14:37:30 +03:00
delta, err := tail.toChannelDelta(true)
if err != nil {
return nil, err
}
err = lc.channelState.UpdateCommitment(tail.txn, tail.sig, delta)
if err != nil {
return nil, err
}
2016-07-06 03:01:55 +03:00
walletLog.Tracef("ChannelPoint(%v): state transition accepted: "+
"our_balance=%v, their_balance=%v", lc.channelState.ChanID,
tail.ourBalance, tail.theirBalance)
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// In the process of revoking our current commitment, we've also
// implicitly ACK'd their set of pending changes that arrived before
// the signature the triggered this revocation. So we'll move up their
// ACK'd index within the log to right at this set of pending changes.
lc.remoteUpdateLog.ackTransition()
revocationMsg.ChannelPoint = *lc.channelState.ChanID
2016-07-06 03:01:55 +03:00
return revocationMsg, nil
}
// ReceiveRevocation processes a revocation sent by the remote party for the
// lowest unrevoked commitment within their commitment chain. We receive a
// revocation either during the initial session negotiation wherein revocation
// windows are extended, or in response to a state update that we initiate. If
// successful, then the remote commitment chain is advanced by a single
// commitment, and a log compaction is attempted. In addition, a slice of
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// HTLC's which can be forwarded upstream are returned.
func (lc *LightningChannel) ReceiveRevocation(revMsg *lnwire.RevokeAndAck) ([]*PaymentDescriptor, error) {
lc.Lock()
defer lc.Unlock()
// The revocation has a nil (zero) preimage, then this should simply be
2016-07-06 03:01:55 +03:00
// added to the end of the revocation window for the remote node.
if bytes.Equal(zeroHash[:], revMsg.Revocation[:]) {
lc.revocationWindow = append(lc.revocationWindow, revMsg)
return nil, nil
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// Now that we've received a new revocation from the remote party,
// we'll toggle our pendingACk bool to indicate that we can create a
// new commitment state after we finish processing this revocation.
lc.pendingACK = false
2016-07-06 03:01:55 +03:00
ourCommitKey := lc.channelState.OurCommitKey
currentRevocationKey := lc.channelState.TheirCurrentRevocation
pendingRevocation := chainhash.Hash(revMsg.Revocation)
2016-07-06 03:01:55 +03:00
// Ensure that the new pre-image can be placed in preimage store.
2016-07-06 03:01:55 +03:00
// TODO(rosbeef): abstract into func
store := lc.channelState.RevocationStore
if err := store.AddNextEntry(&pendingRevocation); err != nil {
return nil, err
}
2016-07-06 03:01:55 +03:00
// Verify that the revocation public key we can derive using this
// preimage and our private key is identical to the revocation key we
2016-07-06 03:01:55 +03:00
// were given for their current (prior) commitment transaction.
revocationPub := DeriveRevocationPubkey(ourCommitKey, pendingRevocation[:])
if !revocationPub.IsEqual(currentRevocationKey) {
2016-07-06 03:01:55 +03:00
return nil, fmt.Errorf("revocation key mismatch")
}
// Additionally, we need to ensure we were given the proper preimage
// to the revocation hash used within any current HTLCs.
2016-07-06 03:01:55 +03:00
if !bytes.Equal(lc.channelState.TheirCurrentRevocationHash[:], zeroHash[:]) {
revokeHash := sha256.Sum256(pendingRevocation[:])
2016-07-06 03:01:55 +03:00
// TODO(roasbeef): rename to drop the "Their"
if !bytes.Equal(lc.channelState.TheirCurrentRevocationHash[:], revokeHash[:]) {
return nil, fmt.Errorf("revocation hash mismatch")
}
}
2016-07-06 03:01:55 +03:00
// Advance the head of the revocation queue now that this revocation has
// been verified. Additionally, extend the end of our unused revocation
// queue with the newly extended revocation window update.
nextRevocation := lc.usedRevocations[0]
lc.channelState.TheirCurrentRevocation = nextRevocation.NextRevocationKey
lc.channelState.TheirCurrentRevocationHash = nextRevocation.NextRevocationHash
lc.usedRevocations[0] = nil // Prevent GC leak.
lc.usedRevocations = lc.usedRevocations[1:]
lc.revocationWindow = append(lc.revocationWindow, revMsg)
walletLog.Tracef("ChannelPoint(%v): remote party accepted state transition, "+
"revoked height %v, now at %v", lc.channelState.ChanID,
lc.remoteCommitChain.tail().height,
lc.remoteCommitChain.tail().height+1)
// At this point, the revocation has been accepted, and we've rotated
// the current revocation key+hash for the remote party. Therefore we
// sync now to ensure the revocation producer state is consistent with
// the current commitment height.
tail := lc.remoteCommitChain.tail()
lnwallet: fix bug that retrieves incorrect pkScript in toChannelDelta() 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.
2017-01-20 14:37:30 +03:00
delta, err := tail.toChannelDelta(false)
if err != nil {
return nil, err
}
if err := lc.channelState.AppendToRevocationLog(delta); err != nil {
return nil, err
}
2016-07-06 03:01:55 +03:00
// Since they revoked the current lowest height in their commitment
// chain, we can advance their chain by a single commitment.
lc.remoteCommitChain.advanceTail()
remoteChainTail := lc.remoteCommitChain.tail().height
localChainTail := lc.localCommitChain.tail().height
// Now that we've verified the revocation update the state of the HTLC
// log as we may be able to prune portions of it now, and update their
// balance.
var htlcsToForward []*PaymentDescriptor
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
for e := lc.remoteUpdateLog.Front(); e != nil; e = e.Next() {
2016-07-06 03:01:55 +03:00
htlc := e.Value.(*PaymentDescriptor)
if htlc.isForwarded {
continue
}
// TODO(roasbeef): re-visit after adding persistence to HTLCs
// * either record add height, or set to N - 1
uncomitted := (htlc.addCommitHeightRemote == 0 ||
htlc.addCommitHeightLocal == 0)
if htlc.EntryType == Add && uncomitted {
continue
}
if htlc.EntryType == Add &&
remoteChainTail >= htlc.addCommitHeightRemote &&
localChainTail >= htlc.addCommitHeightLocal {
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
htlc.isForwarded = true
htlcsToForward = append(htlcsToForward, htlc)
} else if htlc.EntryType != Add &&
remoteChainTail >= htlc.removeCommitHeightRemote &&
localChainTail >= htlc.removeCommitHeightLocal {
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
htlc.isForwarded = true
htlcsToForward = append(htlcsToForward, htlc)
}
}
// As we've just completed a new state transition, attempt to see if we
// can remove any entries from the update log which have been removed
// from the PoV of both commitment chains.
compactLogs(lc.localUpdateLog, lc.remoteUpdateLog,
localChainTail, remoteChainTail)
// As a final step, now that we've received an ACK for our last batch
// of pending changes, we'll update our local ACK'd index to the now
// commitment index, and reset our pendingACKIndex.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
lc.localUpdateLog.ackTransition()
return htlcsToForward, nil
}
2016-07-06 03:01:55 +03:00
// ExtendRevocationWindow extends our revocation window by a single revocation,
// increasing the number of new commitment updates the remote party can
// initiate without our cooperation.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) ExtendRevocationWindow() (*lnwire.RevokeAndAck, error) {
lc.Lock()
defer lc.Unlock()
2016-07-06 03:01:55 +03:00
/// TODO(roasbeef): error if window edge differs from tail by more than
// InitialRevocationWindow
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
revMsg := &lnwire.RevokeAndAck{}
revMsg.ChannelPoint = *lc.channelState.ChanID
2016-07-06 03:01:55 +03:00
nextHeight := lc.revocationWindowEdge + 1
revocation, err := lc.channelState.RevocationProducer.AtIndex(nextHeight)
if err != nil {
2016-07-06 03:01:55 +03:00
return nil, err
}
2016-07-06 03:01:55 +03:00
theirCommitKey := lc.channelState.TheirCommitKey
revMsg.NextRevocationKey = DeriveRevocationPubkey(theirCommitKey,
2016-07-06 03:01:55 +03:00
revocation[:])
revMsg.NextRevocationHash = sha256.Sum256(revocation[:])
2016-07-06 03:01:55 +03:00
lc.revocationWindowEdge++
return revMsg, nil
}
// NextRevocationkey returns the revocation key for the _next_ commitment
// height. The pubkey returned by this function is required by the remote party
// to extend our commitment chain with a new commitment.
//
// TODO(roasbeef): after commitment tx re-write add methdod to ingest
// revocation key
func (lc *LightningChannel) NextRevocationkey() (*btcec.PublicKey, error) {
lc.RLock()
defer lc.RUnlock()
nextHeight := lc.currentHeight + 1
revocation, err := lc.channelState.RevocationProducer.AtIndex(nextHeight)
if err != nil {
return nil, err
}
theirCommitKey := lc.channelState.TheirCommitKey
return DeriveRevocationPubkey(theirCommitKey,
revocation[:]), nil
}
// AddHTLC adds an HTLC to the state machine's local update log. This method
// should be called when preparing to send an outgoing HTLC.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
//
// TODO(roasbeef): check for duplicates below? edge case during restart w/ HTLC
// persistence
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) AddHTLC(htlc *lnwire.UpdateAddHTLC) (uint64, error) {
lc.Lock()
defer lc.Unlock()
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
if err := lc.validateCommitmentSanity(lc.remoteUpdateLog.logIndex,
lc.localUpdateLog.logIndex, true); err != nil {
2016-11-23 11:36:55 +03:00
return 0, err
}
2016-07-06 03:01:55 +03:00
pd := &PaymentDescriptor{
EntryType: Add,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
RHash: PaymentHash(htlc.PaymentHash),
Timeout: htlc.Expiry,
Amount: htlc.Amount,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
Index: lc.localUpdateLog.logIndex,
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
lc.localUpdateLog.appendUpdate(pd)
2016-11-23 11:36:55 +03:00
return pd.Index, nil
}
// ReceiveHTLC adds an HTLC to the state machine's remote update log. This
// method should be called in response to receiving a new HTLC from the remote
// party.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) ReceiveHTLC(htlc *lnwire.UpdateAddHTLC) (uint64, error) {
lc.Lock()
defer lc.Unlock()
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
if err := lc.validateCommitmentSanity(lc.remoteUpdateLog.logIndex,
lc.localUpdateLog.logIndex, true); err != nil {
2016-11-23 11:36:55 +03:00
return 0, err
}
pd := &PaymentDescriptor{
EntryType: Add,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
RHash: PaymentHash(htlc.PaymentHash),
Timeout: htlc.Expiry,
Amount: htlc.Amount,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
Index: lc.remoteUpdateLog.logIndex,
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
lc.remoteUpdateLog.appendUpdate(pd)
2016-07-06 03:01:55 +03:00
lc.rHashMap[pd.RHash] = append(lc.rHashMap[pd.RHash], pd)
2016-11-23 11:36:55 +03:00
return pd.Index, nil
}
2016-10-15 16:18:38 +03:00
// SettleHTLC attempts to settle an existing outstanding received HTLC. The
// remote log index of the HTLC settled is returned in order to facilitate
// creating the corresponding wire message. In the case the supplied preimage
// is invalid, an error is returned.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) SettleHTLC(preimage [32]byte) (uint64, error) {
lc.Lock()
defer lc.Unlock()
paymentHash := sha256.Sum256(preimage[:])
targetHTLCs, ok := lc.rHashMap[paymentHash]
if !ok {
return 0, fmt.Errorf("invalid payment hash")
}
targetHTLC := targetHTLCs[0]
pd := &PaymentDescriptor{
Amount: targetHTLC.Amount,
RPreimage: preimage,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
Index: lc.localUpdateLog.logIndex,
ParentIndex: targetHTLC.Index,
EntryType: Settle,
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
lc.localUpdateLog.appendUpdate(pd)
lc.rHashMap[paymentHash][0] = nil
lc.rHashMap[paymentHash] = lc.rHashMap[paymentHash][1:]
if len(lc.rHashMap[paymentHash]) == 0 {
delete(lc.rHashMap, paymentHash)
}
return targetHTLC.Index, nil
2016-07-06 03:01:55 +03:00
}
// ReceiveHTLCSettle attempts to settle an existing outgoing HTLC indexed by an
// index into the local log. If the specified index doesn't exist within the
// log, and error is returned. Similarly if the preimage is invalid w.r.t to
// the referenced of then a distinct error is returned.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) ReceiveHTLCSettle(preimage [32]byte, logIndex uint64) error {
lc.Lock()
defer lc.Unlock()
paymentHash := sha256.Sum256(preimage[:])
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
htlc := lc.localUpdateLog.lookup(logIndex)
if htlc == nil {
return fmt.Errorf("non existant log entry")
}
if !bytes.Equal(htlc.RHash[:], paymentHash[:]) {
return fmt.Errorf("invalid payment hash")
}
pd := &PaymentDescriptor{
Amount: htlc.Amount,
RPreimage: preimage,
ParentIndex: htlc.Index,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
Index: lc.remoteUpdateLog.logIndex,
EntryType: Settle,
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
lc.remoteUpdateLog.appendUpdate(pd)
return nil
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// FailHTLC attempts to fail a targeted HTLC by its payment hash, inserting an
// entry which will remove the target log entry within the next commitment
// update. This method is intended to be called in order to cancel in
// _incoming_ HTLC.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) FailHTLC(rHash [32]byte) (uint64, error) {
lc.Lock()
defer lc.Unlock()
addEntries, ok := lc.rHashMap[rHash]
if !ok {
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
return 0, fmt.Errorf("unable to find HTLC to fail")
}
addEntry := addEntries[0]
pd := &PaymentDescriptor{
Amount: addEntry.Amount,
RHash: addEntry.RHash,
ParentIndex: addEntry.Index,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
Index: lc.localUpdateLog.logIndex,
EntryType: Fail,
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
lc.localUpdateLog.appendUpdate(pd)
lc.rHashMap[rHash][0] = nil
lc.rHashMap[rHash] = lc.rHashMap[rHash][1:]
if len(lc.rHashMap[rHash]) == 0 {
delete(lc.rHashMap, rHash)
}
return addEntry.Index, nil
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
// ReceiveFailHTLC attempts to cancel a targeted HTLC by its log index,
// inserting an entry which will remove the target log entry within the next
// commitment update. This method should be called in response to the upstream
// party cancelling an outgoing HTLC.
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
func (lc *LightningChannel) ReceiveFailHTLC(logIndex uint64) error {
lc.Lock()
defer lc.Unlock()
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
htlc := lc.localUpdateLog.lookup(logIndex)
if htlc == nil {
return fmt.Errorf("unable to find HTLC to fail")
}
pd := &PaymentDescriptor{
Amount: htlc.Amount,
RHash: htlc.RHash,
ParentIndex: htlc.Index,
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
Index: lc.remoteUpdateLog.logIndex,
EntryType: Fail,
}
lnwallet: update state machine to the version within the spec 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.
2017-02-21 04:55:33 +03:00
lc.remoteUpdateLog.appendUpdate(pd)
2016-07-06 03:01:55 +03:00
return nil
}
// ChannelPoint returns the outpoint of the original funding transaction which
// created this active channel. This outpoint is used throughout various
// subsystems to uniquely identify an open channel.
2016-07-06 03:01:55 +03:00
func (lc *LightningChannel) ChannelPoint() *wire.OutPoint {
return lc.channelState.ChanID
}
2016-07-06 03:01:55 +03:00
// addHTLC adds a new HTLC to the passed commitment transaction. One of four
// full scripts will be generated for the HTLC output depending on if the HTLC
// is incoming and if it's being applied to our commitment transaction or that
// of the remote node's. Additionally, in order to be able to efficiently
// locate the added HTLC on the commitment transaction from the
// PaymentDescriptor that generated it, the generated script is stored within
// the descriptor itself.
2016-07-06 03:01:55 +03:00
func (lc *LightningChannel) addHTLC(commitTx *wire.MsgTx, ourCommit bool,
paymentDesc *PaymentDescriptor, revocation [32]byte, delay uint32,
isIncoming bool) error {
localKey := lc.channelState.OurCommitKey
2016-07-06 03:01:55 +03:00
remoteKey := lc.channelState.TheirCommitKey
timeout := paymentDesc.Timeout
rHash := paymentDesc.RHash
2016-07-06 03:01:55 +03:00
// Generate the proper redeem scripts for the HTLC output modified by
// two-bits denoting if this is an incoming HTLC, and if the HTLC is
// being applied to their commitment transaction or ours.
var pkScript []byte
var err error
switch {
// The HTLC is paying to us, and being applied to our commitment
// transaction. So we need to use the receiver's version of HTLC the
// script.
case isIncoming && ourCommit:
2016-07-06 03:01:55 +03:00
pkScript, err = receiverHTLCScript(timeout, delay, remoteKey,
localKey, revocation[:], rHash[:])
// We're being paid via an HTLC by the remote party, and the HTLC is
// being added to their commitment transaction, so we use the sender's
// version of the HTLC script.
case isIncoming && !ourCommit:
2016-07-06 03:01:55 +03:00
pkScript, err = senderHTLCScript(timeout, delay, remoteKey,
localKey, revocation[:], rHash[:])
// We're sending an HTLC which is being added to our commitment
// transaction. Therefore, we need to use the sender's version of the
// HTLC script.
case !isIncoming && ourCommit:
2016-07-06 03:01:55 +03:00
pkScript, err = senderHTLCScript(timeout, delay, localKey,
remoteKey, revocation[:], rHash[:])
// Finally, we're paying the remote party via an HTLC, which is being
// added to their commitment transaction. Therefore, we use the
// receiver's version of the HTLC script.
case !isIncoming && !ourCommit:
2016-07-06 03:01:55 +03:00
pkScript, err = receiverHTLCScript(timeout, delay, localKey,
remoteKey, revocation[:], rHash[:])
}
if err != nil {
return err
}
2016-07-06 03:01:55 +03:00
// Now that we have the redeem scripts, create the P2WSH public key
// script for the output itself.
htlcP2WSH, err := witnessScriptHash(pkScript)
if err != nil {
return err
}
// Add the new HTLC outputs to the respective commitment transactions.
amountPending := int64(paymentDesc.Amount)
commitTx.AddTxOut(wire.NewTxOut(amountPending, htlcP2WSH))
// Store the pkScript of this particular PaymentDescriptor so we can
// quickly locate it within the commitment transaction later.
lnwallet: fix bug that retrieves incorrect pkScript in toChannelDelta() 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.
2017-01-20 14:37:30 +03:00
if ourCommit {
paymentDesc.ourPkScript = htlcP2WSH
} else {
paymentDesc.theirPrevPkScript = paymentDesc.theirPkScript
paymentDesc.theirPkScript = htlcP2WSH
}
return nil
}
// ForceCloseSummary describes the final commitment state before the channel is
// locked-down to initiate a force closure by broadcasting the latest state
// on-chain. The summary includes all the information required to claim all
// rightfully owned outputs.
// TODO(roasbeef): generalize, add HTLC info, etc.
type ForceCloseSummary struct {
// CloseTx is the transaction which closed the channel on-chain. If we
// initiate the force close, then this'll be our latest commitment
// state. Otherwise, this'll be the state that the remote peer
// broadcasted on-chain.
CloseTx *wire.MsgTx
// SelfOutpoint is the output created by the above close tx which is
// spendable by us after a relative time delay.
SelfOutpoint wire.OutPoint
// SelfOutputMaturity is the relative maturity period before the above
// output can be claimed.
SelfOutputMaturity uint32
// SelfOutputSignDesc is a fully populated sign descriptor capable of
// generating a valid signature to sweep the self output.
SelfOutputSignDesc *SignDescriptor
}
// getSignedCommitTx function take the latest commitment transaction and populate
// it with witness data.
func (lc *LightningChannel) getSignedCommitTx() (*wire.MsgTx, error) {
// Fetch the current commitment transaction, along with their signature
// for the transaction.
commitTx := lc.channelState.OurCommitTx
theirSig := append(lc.channelState.OurCommitSig, byte(txscript.SigHashAll))
// With this, we then generate the full witness so the caller can
// broadcast a fully signed transaction.
lc.signDesc.SigHashes = txscript.NewTxSigHashes(commitTx)
ourSigRaw, err := lc.signer.SignOutputRaw(commitTx, lc.signDesc)
if err != nil {
return nil, err
}
ourSig := append(ourSigRaw, byte(txscript.SigHashAll))
// With the final signature generated, create the witness stack
// required to spend from the multi-sig output.
ourKey := lc.channelState.OurMultiSigKey.SerializeCompressed()
theirKey := lc.channelState.TheirMultiSigKey.SerializeCompressed()
commitTx.TxIn[0].Witness = SpendMultiSig(lc.FundingWitnessScript, ourKey,
ourSig, theirKey, theirSig)
return commitTx, nil
}
// ForceClose executes a unilateral closure of the transaction at the current
// lowest commitment height of the channel. Following a force closure, all
// state transitions, or modifications to the state update logs will be
// rejected. Additionally, this function also returns a ForceCloseSummary which
// includes the necessary details required to sweep all the time-locked within
// the commitment transaction.
//
// TODO(roasbeef): all methods need to abort if in dispute state
// TODO(roasbeef): method to generate CloseSummaries for when the remote peer
// does a unilateral close
func (lc *LightningChannel) ForceClose() (*ForceCloseSummary, error) {
lc.Lock()
defer lc.Unlock()
// Set the channel state to indicate that the channel is now in a
// contested state.
lc.status = channelDispute
commitTx, err := lc.getSignedCommitTx()
if err != nil {
return nil, err
}
// Re-derive the original pkScript for to-self output within the
// commitment transaction. We'll need this to find the corresponding
// output in the commitment transaction and potentially for creating
// the sign descriptor.
csvTimeout := lc.channelState.LocalCsvDelay
selfKey := lc.channelState.OurCommitKey
producer := lc.channelState.RevocationProducer
unusedRevocation, err := producer.AtIndex(lc.currentHeight)
if err != nil {
return nil, err
}
revokeKey := DeriveRevocationPubkey(lc.channelState.TheirCommitKey,
unusedRevocation[:])
selfScript, err := commitScriptToSelf(csvTimeout, selfKey, revokeKey)
if err != nil {
return nil, err
}
payToUsScriptHash, err := witnessScriptHash(selfScript)
if err != nil {
return nil, err
}
// Locate the output index of the delayed commitment output back to us.
// We'll return the details of this output to the caller so they can
// sweep it once it's mature.
// TODO(roasbeef): also return HTLC info, assumes only p2wsh is commit
// tx
var delayIndex uint32
var delayScript []byte
var selfSignDesc *SignDescriptor
for i, txOut := range commitTx.TxOut {
if !bytes.Equal(payToUsScriptHash, txOut.PkScript) {
continue
}
delayIndex = uint32(i)
delayScript = txOut.PkScript
break
}
// With the necessary information gathered above, create a new sign
// descriptor which is capable of generating the signature the caller
// needs to sweep this output. The hash cache, and input index are not
// set as the caller will decide these values once sweeping the output.
// If the output is non-existant (dust), have the sign descriptor be nil.
if len(delayScript) != 0 {
selfSignDesc = &SignDescriptor{
PubKey: selfKey,
WitnessScript: selfScript,
Output: &wire.TxOut{
PkScript: delayScript,
Value: int64(lc.channelState.OurBalance),
},
HashType: txscript.SigHashAll,
}
}
// Finally, close the channel force close signal which notifies any
// subscribers that the channel has now been forcibly closed. This
// allows callers to begin to carry out any post channel closure
// activities.
close(lc.ForceCloseSignal)
return &ForceCloseSummary{
CloseTx: commitTx,
SelfOutpoint: wire.OutPoint{
Hash: commitTx.TxHash(),
Index: delayIndex,
},
SelfOutputMaturity: csvTimeout,
SelfOutputSignDesc: selfSignDesc,
}, nil
}
// InitCooperativeClose initiates a cooperative closure of an active lightning
// channel. This method should only be executed once all pending HTLCs (if any)
// on the channel have been cleared/removed. Upon completion, the source
// channel will shift into the "closing" state, which indicates that all
// incoming/outgoing HTLC requests should be rejected. A signature for the
// closing transaction, and the txid of the closing transaction are returned.
// The initiator of the channel closure should then watch the blockchain for a
// confirmation of the closing transaction before considering the channel
// terminated. In the case of an unresponsive remote party, the initiator can
// either choose to execute a force closure, or backoff for a period of time,
// and retry the cooperative closure.
//
// TODO(roasbeef): caller should initiate signal to reject all incoming HTLCs,
// settle any inflight.
func (lc *LightningChannel) InitCooperativeClose() ([]byte, *chainhash.Hash, error) {
lc.Lock()
defer lc.Unlock()
// If we're already closing the channel, then ignore this request.
if lc.status == channelClosing || lc.status == channelClosed {
// TODO(roasbeef): check to ensure no pending payments
return nil, nil, ErrChanClosing
}
closeTx := CreateCooperativeCloseTx(lc.fundingTxIn,
lc.channelState.OurBalance, lc.channelState.TheirBalance,
lc.channelState.OurDeliveryScript, lc.channelState.TheirDeliveryScript,
lc.channelState.IsInitiator)
2017-03-15 08:53:15 +03:00
// Ensure that the transaction doesn't explicitly violate any
// consensus rules such as being too big, or having any value with a
// negative output.
tx := btcutil.NewTx(closeTx)
if err := blockchain.CheckTransactionSanity(tx); err != nil {
return nil, nil, err
}
// Finally, sign the completed cooperative closure transaction. As the
// initiator we'll simply send our signature over to the remote party,
// using the generated txid to be notified once the closure transaction
// has been confirmed.
lc.signDesc.SigHashes = txscript.NewTxSigHashes(closeTx)
closeSig, err := lc.signer.SignOutputRaw(closeTx, lc.signDesc)
if err != nil {
return nil, nil, err
}
// As everything checks out, indicate in the channel status that a
// channel closure has been initiated.
lc.status = channelClosing
closeTxSha := closeTx.TxHash()
return closeSig, &closeTxSha, nil
}
// CompleteCooperativeClose completes the cooperative closure of the target
// active lightning channel. This method should be called in response to the
// remote node initiating a cooperative channel closure. A fully signed closure
// transaction is returned. It is the duty of the responding node to broadcast
// a signed+valid closure transaction to the network.
//
// NOTE: The passed remote sig is expected to be a fully complete signature
// including the proper sighash byte.
func (lc *LightningChannel) CompleteCooperativeClose(remoteSig []byte) (*wire.MsgTx, error) {
lc.Lock()
defer lc.Unlock()
// If we're already closing the channel, then ignore this request.
if lc.status == channelClosing || lc.status == channelClosed {
// TODO(roasbeef): check to ensure no pending payments
return nil, ErrChanClosing
}
// Create the transaction used to return the current settled balance
// on this active channel back to both parties. In this current model,
// the initiator pays full fees for the cooperative close transaction.
closeTx := CreateCooperativeCloseTx(lc.fundingTxIn,
lc.channelState.OurBalance, lc.channelState.TheirBalance,
lc.channelState.OurDeliveryScript, lc.channelState.TheirDeliveryScript,
lc.channelState.IsInitiator)
// Ensure that the transaction doesn't explicitly validate any
// consensus rules such as being too big, or having any value with a
// negative output.
tx := btcutil.NewTx(closeTx)
if err := blockchain.CheckTransactionSanity(tx); err != nil {
return nil, err
}
// With the transaction created, we can finally generate our half of
// the 2-of-2 multi-sig needed to redeem the funding output.
hashCache := txscript.NewTxSigHashes(closeTx)
lc.signDesc.SigHashes = hashCache
closeSig, err := lc.signer.SignOutputRaw(closeTx, lc.signDesc)
if err != nil {
return nil, err
}
// Finally, construct the witness stack minding the order of the
// pubkeys+sigs on the stack.
ourKey := lc.channelState.OurMultiSigKey.SerializeCompressed()
theirKey := lc.channelState.TheirMultiSigKey.SerializeCompressed()
ourSig := append(closeSig, byte(txscript.SigHashAll))
witness := SpendMultiSig(lc.signDesc.WitnessScript, ourKey, ourSig,
theirKey, remoteSig)
closeTx.TxIn[0].Witness = witness
// Validate the finalized transaction to ensure the output script is
// properly met, and that the remote peer supplied a valid signature.
vm, err := txscript.NewEngine(lc.fundingP2WSH, closeTx, 0,
txscript.StandardVerifyFlags, nil, hashCache,
int64(lc.channelState.Capacity))
if err != nil {
return nil, err
}
if err := vm.Execute(); err != nil {
return nil, err
}
// As the transaction is sane, and the scripts are valid we'll mark the
// channel now as closed as the closure transaction should get into the
// chain in a timely manner and possibly be re-broadcast by the wallet.
lc.status = channelClosed
return closeTx, nil
}
// DeleteState deletes all state concerning the channel from the underlying
// database, only leaving a small summary describing metadata of the
// channel's lifetime.
func (lc *LightningChannel) DeleteState() error {
return lc.channelState.CloseChannel()
}
// StateSnapshot returns a snapshot of the current fully committed state within
// the channel.
func (lc *LightningChannel) StateSnapshot() *channeldb.ChannelSnapshot {
lc.stateMtx.RLock()
defer lc.stateMtx.RUnlock()
return lc.channelState.Snapshot()
}
// CreateCommitTx creates a commitment transaction, spending from specified
// funding output. The commitment transaction contains two outputs: one paying
// to the "owner" of the commitment transaction which can be spent after a
// relative block delay or revocation event, and the other paying the the
// counterparty within the channel, which can be spent immediately.
func CreateCommitTx(fundingOutput *wire.TxIn, selfKey, theirKey *btcec.PublicKey,
revokeKey *btcec.PublicKey, csvTimeout uint32, amountToSelf,
amountToThem, dustLimit btcutil.Amount) (*wire.MsgTx, error) {
// First, we create the script for the delayed "pay-to-self" output.
// This output has 2 main redemption clauses: either we can redeem the
// output after a relative block delay, or the remote node can claim
// the funds with the revocation key if we broadcast a revoked
// commitment transaction.
ourRedeemScript, err := commitScriptToSelf(csvTimeout, selfKey,
revokeKey)
if err != nil {
return nil, err
}
payToUsScriptHash, err := witnessScriptHash(ourRedeemScript)
if err != nil {
return nil, err
}
// Next, we create the script paying to them. This is just a regular
// P2WPKH output, without any added CSV delay.
theirWitnessKeyHash, err := commitScriptUnencumbered(theirKey)
if err != nil {
return nil, err
}
// Now that both output scripts have been created, we can finally create
// the transaction itself. We use a transaction version of 2 since CSV
// will fail unless the tx version is >= 2.
commitTx := wire.NewMsgTx(2)
commitTx.AddTxIn(fundingOutput)
// Avoid creating dust outputs within the commitment transaction.
if amountToSelf >= dustLimit {
commitTx.AddTxOut(wire.NewTxOut(int64(amountToSelf), payToUsScriptHash))
}
if amountToThem >= dustLimit {
commitTx.AddTxOut(wire.NewTxOut(int64(amountToThem), theirWitnessKeyHash))
}
return commitTx, nil
}
// CreateCooperativeCloseTx creates a transaction which if signed by both
// parties, then broadcast cooperatively closes an active channel. The creation
// of the closure transaction is modified by a boolean indicating if the party
// constructing the channel is the initiator of the closure. Currently it is
// expected that the initiator pays the transaction fees for the closing
// transaction in full.
func CreateCooperativeCloseTx(fundingTxIn *wire.TxIn,
ourBalance, theirBalance btcutil.Amount,
ourDeliveryScript, theirDeliveryScript []byte,
initiator bool) *wire.MsgTx {
// Construct the transaction to perform a cooperative closure of the
// channel. In the event that one side doesn't have any settled funds
// within the channel then a refund output for that particular side can
// be omitted.
closeTx := wire.NewMsgTx(2)
closeTx.AddTxIn(fundingTxIn)
// The initiator of a cooperative closure pays the fee in entirety.
// Determine if we're the initiator so we can compute fees properly.
if initiator {
// TODO(roasbeef): take sat/byte here instead of properly calc
ourBalance -= 5000
} else {
theirBalance -= 5000
}
// TODO(roasbeef): dust check...
// * although upper layers should prevent
if ourBalance != 0 {
closeTx.AddTxOut(&wire.TxOut{
PkScript: ourDeliveryScript,
Value: int64(ourBalance),
})
}
if theirBalance != 0 {
closeTx.AddTxOut(&wire.TxOut{
PkScript: theirDeliveryScript,
Value: int64(theirBalance),
})
}
txsort.InPlaceSort(closeTx)
return closeTx
}