rgb-cln/daemon/peer.c

5293 lines
148 KiB
C
Raw Normal View History

#include "bitcoind.h"
#include "chaintopology.h"
#include "close_tx.h"
#include "commit_tx.h"
#include "cryptopkt.h"
#include "db.h"
#include "dns.h"
#include "find_p2sh_out.h"
#include "invoice.h"
#include "jsonrpc.h"
#include "lightningd.h"
#include "log.h"
#include "names.h"
#include "netaddr.h"
#include "output_to_htlc.h"
#include "p2p_announce.h"
#include "packets.h"
#include "pay.h"
#include "peer.h"
#include "peer_internal.h"
#include "permute_tx.h"
#include "protobuf_convert.h"
#include "pseudorand.h"
#include "remove_dust.h"
#include "routing.h"
#include "secrets.h"
#include "sphinx.h"
#include "state.h"
#include "timeout.h"
#include "utils.h"
#include "wallet.h"
#include <bitcoin/base58.h>
#include <bitcoin/script.h>
#include <bitcoin/tx.h>
#include <ccan/array_size/array_size.h>
#include <ccan/cast/cast.h>
#include <ccan/io/io.h>
#include <ccan/list/list.h>
#include <ccan/mem/mem.h>
#include <ccan/noerr/noerr.h>
#include <ccan/ptrint/ptrint.h>
#include <ccan/str/hex/hex.h>
#include <ccan/structeq/structeq.h>
#include <ccan/tal/str/str.h>
#include <ccan/tal/tal.h>
#include <errno.h>
#include <inttypes.h>
#include <netdb.h>
#include <netinet/in.h>
#include <sodium/randombytes.h>
#include <stdlib.h>
#include <sys/socket.h>
#include <sys/types.h>
struct json_connecting {
/* This owns us, so we're freed after command_fail or command_success */
struct command *cmd;
const char *name, *port;
struct anchor_input *input;
};
static bool command_htlc_set_fail(struct peer *peer, struct htlc *htlc,
enum fail_error error_code, const char *why);
static bool command_htlc_fail(struct peer *peer, struct htlc *htlc);
static bool command_htlc_fulfill(struct peer *peer, struct htlc *htlc);
static void try_commit(struct peer *peer);
void peer_debug(struct peer *peer, const char *fmt, ...)
{
va_list ap;
va_start(ap, fmt);
logv(peer->log, LOG_DBG, fmt, ap);
va_end(ap);
}
void peer_add_their_commit(struct peer *peer,
const struct sha256_double *txid, u64 commit_num)
{
struct their_commit *tc = tal(peer, struct their_commit);
tc->txid = *txid;
tc->commit_num = commit_num;
list_add_tail(&peer->their_commits, &tc->list);
db_add_commit_map(peer, txid, commit_num);
}
/* Create a bitcoin close tx, using last signature they sent. */
static const struct bitcoin_tx *mk_bitcoin_close(const tal_t *ctx,
struct peer *peer)
{
struct bitcoin_tx *close_tx;
secp256k1_ecdsa_signature our_close_sig;
close_tx = peer_create_close_tx(ctx, peer, peer->closing.their_fee);
peer_sign_mutual_close(peer, close_tx, &our_close_sig);
close_tx->input[0].witness
= bitcoin_witness_2of2(close_tx->input,
peer->closing.their_sig,
&our_close_sig,
&peer->remote.commitkey,
&peer->local.commitkey);
return close_tx;
}
/* Create a bitcoin spend tx (to spend our commit's outputs) */
static const struct bitcoin_tx *bitcoin_spend_ours(struct peer *peer)
{
u8 *witnessscript;
const struct bitcoin_tx *commit = peer->local.commit->tx;
secp256k1_ecdsa_signature sig;
struct bitcoin_tx *tx;
unsigned int p2wsh_out;
uint64_t fee;
/* The redeemscript for a commit tx is fairly complex. */
witnessscript = bitcoin_redeem_secret_or_delay(peer,
&peer->local.finalkey,
&peer->remote.locktime,
&peer->remote.finalkey,
&peer->local.commit->revocation_hash);
/* Now, create transaction to spend it. */
tx = bitcoin_tx(peer, 1, 1);
tx->input[0].txid = peer->local.commit->txid;
p2wsh_out = find_p2wsh_out(commit, witnessscript);
tx->input[0].index = p2wsh_out;
tx->input[0].sequence_number = bitcoin_nsequence(&peer->remote.locktime);
tx->input[0].amount = tal_dup(tx->input, u64,
&commit->output[p2wsh_out].amount);
tx->output[0].script = scriptpubkey_p2sh(tx,
bitcoin_redeem_single(tx,
&peer->local.finalkey));
/* Witness length can vary, due to DER encoding of sigs, but we
* use 176 from an example run. */
assert(measure_tx_cost(tx) == 83 * 4);
fee = fee_by_feerate(83 + 176 / 4, get_feerate(peer->dstate->topology));
/* FIXME: Fail gracefully in these cases (not worth collecting) */
if (fee > commit->output[p2wsh_out].amount
|| is_dust(commit->output[p2wsh_out].amount - fee))
fatal("Amount of %"PRIu64" won't cover fee %"PRIu64,
commit->output[p2wsh_out].amount, fee);
tx->output[0].amount = commit->output[p2wsh_out].amount - fee;
peer_sign_spend(peer, tx, witnessscript, &sig);
tx->input[0].witness = bitcoin_witness_secret(tx,
NULL, 0, &sig,
witnessscript);
return tx;
}
/* Sign and local commit tx */
static void sign_commit_tx(struct peer *peer)
{
secp256k1_ecdsa_signature sig;
/* Can't be signed already, and can't have scriptsig! */
assert(!peer->local.commit->tx->input[0].script);
assert(!peer->local.commit->tx->input[0].witness);
peer_sign_ourcommit(peer, peer->local.commit->tx, &sig);
peer->local.commit->tx->input[0].witness
= bitcoin_witness_2of2(peer->local.commit->tx->input,
peer->local.commit->sig,
&sig,
&peer->remote.commitkey,
&peer->local.commitkey);
}
static u64 commit_tx_fee(const struct bitcoin_tx *commit, u64 anchor_satoshis)
{
uint64_t i, total = 0;
for (i = 0; i < tal_count(commit->output); i++)
total += commit->output[i].amount;
assert(anchor_satoshis >= total);
return anchor_satoshis - total;
}
struct peer *find_peer(struct lightningd_state *dstate, const struct pubkey *id)
{
struct peer *peer;
list_for_each(&dstate->peers, peer, list) {
if (peer->id && pubkey_eq(peer->id, id))
return peer;
}
return NULL;
}
struct peer *find_peer_by_pkhash(struct lightningd_state *dstate, const u8 *pkhash)
{
struct peer *peer;
struct ripemd160 addr;
list_for_each(&dstate->peers, peer, list) {
pubkey_to_hash160(peer->id, &addr);
if (memcmp(&addr, pkhash, sizeof(addr)) == 0)
return peer;
}
return NULL;
}
void debug_dump_peers(struct lightningd_state *dstate)
{
struct peer *peer;
list_for_each(&dstate->peers, peer, list) {
if (!peer->local.commit
|| !peer->remote.commit)
continue;
log_debug_struct(peer->log, "our cstate: %s",
struct channel_state,
peer->local.commit->cstate);
log_debug_struct(peer->log, "their cstate: %s",
struct channel_state,
peer->remote.commit->cstate);
}
}
static struct peer *find_peer_json(struct lightningd_state *dstate,
const char *buffer,
jsmntok_t *peeridtok)
{
struct pubkey peerid;
if (!pubkey_from_hexstr(buffer + peeridtok->start,
peeridtok->end - peeridtok->start, &peerid))
return NULL;
return find_peer(dstate, &peerid);
}
static bool peer_uncommitted_changes(const struct peer *peer)
{
struct htlc_map_iter it;
struct htlc *h;
enum feechange_state i;
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
if (htlc_has(h, HTLC_REMOTE_F_PENDING))
return true;
}
/* Pending feechange we sent, or pending ack of theirs. */
for (i = 0; i < ARRAY_SIZE(peer->feechanges); i++) {
if (!peer->feechanges[i])
continue;
if (feechange_state_flags(i) & HTLC_REMOTE_F_PENDING)
return true;
}
return false;
}
static void remote_changes_pending(struct peer *peer)
{
if (!peer->commit_timer) {
log_debug(peer->log, "remote_changes_pending: adding timer");
peer->commit_timer = new_reltimer(&peer->dstate->timers, peer,
peer->dstate->config.commit_time,
try_commit, peer);
} else
log_debug(peer->log, "remote_changes_pending: timer already exists");
}
static void peer_update_complete(struct peer *peer)
{
log_debug(peer->log, "peer_update_complete");
if (peer->commit_jsoncmd) {
command_success(peer->commit_jsoncmd,
null_response(peer->commit_jsoncmd));
peer->commit_jsoncmd = NULL;
}
/* Have we got more changes in the meantime? */
if (peer_uncommitted_changes(peer)) {
log_debug(peer->log, "peer_update_complete: more changes!");
remote_changes_pending(peer);
}
}
/* FIXME: Split success and fail functions, roll state changes etc into
* success case. */
static void peer_open_complete(struct peer *peer, const char *problem)
{
if (problem) {
log_unusual(peer->log, "peer open failed: %s", problem);
if (peer->open_jsoncmd) {
command_fail(peer->open_jsoncmd, "%s", problem);
peer->open_jsoncmd = NULL;
}
} else {
log_debug(peer->log, "peer open complete");
assert(!peer->nc);
/* We're connected, so record it. */
peer->nc = add_connection(peer->dstate->rstate,
&peer->dstate->id, peer->id,
peer->dstate->config.fee_base,
peer->dstate->config.fee_per_satoshi,
peer->dstate->config.min_htlc_expiry,
peer->dstate->config.min_htlc_expiry);
if (peer->open_jsoncmd) {
struct json_result *response;
response = new_json_result(peer->open_jsoncmd);
json_object_start(response, NULL);
json_add_pubkey(response, "id", peer->id);
json_object_end(response);
command_success(peer->open_jsoncmd, response);
peer->open_jsoncmd = NULL;
}
}
}
static void set_peer_state(struct peer *peer, enum state newstate,
const char *caller, bool db_commit)
{
log_debug(peer->log, "%s: %s => %s", caller,
state_name(peer->state), state_name(newstate));
peer->state = newstate;
/* We can only route in normal state. */
if (!state_is_normal(peer->state))
peer->nc = tal_free(peer->nc);
if (db_commit)
db_update_state(peer);
}
static void peer_breakdown(struct peer *peer)
{
if (peer->commit_jsoncmd) {
command_fail(peer->commit_jsoncmd, "peer breakdown");
peer->commit_jsoncmd = NULL;
}
/* FIXME: Reason. */
if (peer->open_jsoncmd) {
command_fail(peer->open_jsoncmd, "peer breakdown");
peer->open_jsoncmd = NULL;
}
/* If we have a closing tx, use it. */
if (peer->closing.their_sig) {
const struct bitcoin_tx *close = mk_bitcoin_close(peer, peer);
log_unusual(peer->log, "Peer breakdown: sending close tx");
broadcast_tx(peer->dstate->topology, peer, close, NULL);
tal_free(close);
/* If we have a signed commit tx (maybe not if we just offered
* anchor, or they supplied anchor, or no outputs to us). */
} else if (peer->local.commit && peer->local.commit->sig) {
log_unusual(peer->log, "Peer breakdown: sending commit tx");
sign_commit_tx(peer);
broadcast_tx(peer->dstate->topology, peer,
peer->local.commit->tx, NULL);
} else {
log_info(peer->log, "Peer breakdown: nothing to do");
/* We close immediately. */
set_peer_state(peer, STATE_CLOSED, __func__, false);
db_forget_peer(peer);
}
/* Always wake peer to close or flush packets. */
io_wake(peer);
}
/* All unrevoked commit txs must have no HTLCs in them. */
static bool committed_to_htlcs(const struct peer *peer)
{
struct htlc_map_iter it;
struct htlc *h;
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
if (htlc_is_dead(h))
continue;
return true;
}
return false;
}
static void peer_calculate_close_fee(struct peer *peer)
{
/* Use actual worst-case length of close tx: based on FIXME-OLD#02's
* commitment tx numbers, but only 1 byte for output count */
const uint64_t txsize = 41 + 221 + 10 + 32 + 32;
uint64_t maxfee;
peer->closing.our_fee
= fee_by_feerate(txsize, get_feerate(peer->dstate->topology));
/* FIXME-OLD #2:
* The sender MUST set `close_fee` lower than or equal to the
* fee of the final commitment transaction, and MUST set
* `close_fee` to an even number of satoshis.
*/
maxfee = commit_tx_fee(peer->local.commit->tx, peer->anchor.satoshis);
if (peer->closing.our_fee > maxfee) {
/* This could only happen if the fee rate dramatically */
log_unusual(peer->log,
"Closing fee %"PRIu64" exceeded commit fee %"PRIu64", reducing.",
peer->closing.our_fee, maxfee);
peer->closing.our_fee = maxfee;
/* This can happen if actual commit txfee is odd. */
if (peer->closing.our_fee & 1)
peer->closing.our_fee--;
}
assert(!(peer->closing.our_fee & 1));
}
static void start_closing_in_transaction(struct peer *peer)
{
assert(!committed_to_htlcs(peer));
set_peer_state(peer, STATE_MUTUAL_CLOSING, __func__, true);
peer_calculate_close_fee(peer);
peer->closing.closing_order = peer->order_counter++;
db_update_our_closing(peer);
queue_pkt_close_signature(peer);
}
static struct io_plan *peer_close(struct io_conn *conn, struct peer *peer)
{
/* Tell writer to wrap it up (may have to xmit first) */
io_wake(peer);
/* We do nothing more. */
return io_wait(conn, NULL, io_never, NULL);
}
void peer_fail(struct peer *peer, const char *caller)
{
/* Don't fail twice. */
if (state_is_error(peer->state) || state_is_onchain(peer->state))
return;
/* FIXME: Save state here? */
set_peer_state(peer, STATE_ERR_BREAKDOWN, caller, false);
peer_breakdown(peer);
}
/* Communication failed: send err (if non-NULL), then dump to chain and close. */
static bool peer_comms_err(struct peer *peer, Pkt *err)
{
if (err)
queue_pkt_err(peer, err);
peer_fail(peer, __func__);
return false;
}
static bool peer_database_err(struct peer *peer)
{
return peer_comms_err(peer, pkt_err(peer, "database error"));
}
/* Unexpected packet received: stop listening, send error, start
* breakdown procedure, return false. */
static bool peer_received_unexpected_pkt(struct peer *peer, const Pkt *pkt,
const char *where)
{
const char *p;
Pkt *err;
log_unusual(peer->log, "%s: received unexpected pkt %u (%s) in %s",
where, pkt->pkt_case, pkt_name(pkt->pkt_case),
state_name(peer->state));
if (pkt->pkt_case != PKT__PKT_ERROR) {
err = pkt_err_unexpected(peer, pkt);
goto out;
}
/* FIXME-OLD #2:
*
* A node MUST fail the connection if it receives an `err`
* message, and MUST NOT send an `err` message in this case.
* For other connection failures, a node SHOULD send an
* informative `err` message.
*/
err = NULL;
/* Check packet for weird chars. */
for (p = pkt->error->problem; *p; p++) {
if (cisprint(*p))
continue;
p = tal_hexstr(peer, pkt->error->problem,
strlen(pkt->error->problem));
log_unusual(peer->log, "Error pkt (hex) %s", p);
tal_free(p);
goto out;
}
log_unusual(peer->log, "Error pkt '%s'", pkt->error->problem);
out:
return peer_comms_err(peer, err);
}
/* Creation the bitcoin anchor tx, spending output user provided. */
static bool bitcoin_create_anchor(struct peer *peer)
{
struct bitcoin_tx *tx = bitcoin_tx(peer, 1, 1);
size_t i;
/* We must be offering anchor for us to try creating it */
assert(peer->local.offer_anchor);
tx->output[0].script = scriptpubkey_p2wsh(tx, peer->anchor.witnessscript);
tx->output[0].amount = peer->anchor.input->out_amount;
tx->input[0].txid = peer->anchor.input->txid;
tx->input[0].index = peer->anchor.input->index;
tx->input[0].amount = tal_dup(tx->input, u64,
&peer->anchor.input->in_amount);
if (!wallet_add_signed_input(peer->dstate,
&peer->anchor.input->walletkey,
tx, 0))
return false;
bitcoin_txid(tx, &peer->anchor.txid);
peer->anchor.tx = tx;
peer->anchor.index = 0;
/* We'll need this later, when we're told to broadcast it. */
peer->anchor.satoshis = tx->output[0].amount;
/* To avoid malleation, all inputs must be segwit! */
for (i = 0; i < tal_count(tx->input); i++)
assert(tx->input[i].witness);
return true;
}
static bool open_pkt_in(struct peer *peer, const Pkt *pkt)
{
Pkt *err;
struct commit_info *ci;
assert(peer->state == STATE_OPEN_WAIT_FOR_OPENPKT);
if (pkt->pkt_case != PKT__PKT_OPEN)
return peer_received_unexpected_pkt(peer, pkt, __func__);
db_start_transaction(peer);
ci = new_commit_info(peer, 0);
err = accept_pkt_open(peer, pkt, &ci->revocation_hash,
&peer->remote.next_revocation_hash);
if (err) {
db_abort_transaction(peer);
return peer_comms_err(peer, err);
}
db_set_visible_state(peer);
/* Set up their commit info now: rest gets done in setup_first_commit
* once anchor is established. */
peer->remote.commit = ci;
/* Witness script for anchor. */
peer->anchor.witnessscript
= bitcoin_redeem_2of2(peer,
&peer->local.commitkey,
&peer->remote.commitkey);
if (peer->local.offer_anchor) {
if (!bitcoin_create_anchor(peer)) {
db_abort_transaction(peer);
err = pkt_err(peer, "Own anchor unavailable");
return peer_comms_err(peer, err);
}
/* FIXME: Redundant with peer->local.offer_anchor? */
peer->anchor.ours = true;
/* This shouldn't happen! */
if (!setup_first_commit(peer)) {
db_abort_transaction(peer);
err = pkt_err(peer, "Own anchor has insufficient funds");
return peer_comms_err(peer, err);
}
set_peer_state(peer, STATE_OPEN_WAIT_FOR_COMMIT_SIGPKT,
__func__, true);
if (db_commit_transaction(peer) != NULL)
return peer_database_err(peer);
queue_pkt_anchor(peer);
return true;
} else {
set_peer_state(peer, STATE_OPEN_WAIT_FOR_ANCHORPKT,
__func__, true);
if (db_commit_transaction(peer) != NULL)
return peer_database_err(peer);
return true;
}
}
static void funding_tx_failed(struct peer *peer,
int exitstatus,
const char *err)
{
const char *str = tal_fmt(peer, "Broadcasting funding gave %i: %s",
exitstatus, err);
peer_open_complete(peer, str);
peer_breakdown(peer);
queue_pkt_err(peer, pkt_err(peer, "Funding failed"));
}
static bool open_ouranchor_pkt_in(struct peer *peer, const Pkt *pkt)
{
Pkt *err;
if (pkt->pkt_case != PKT__PKT_OPEN_COMMIT_SIG)
return peer_received_unexpected_pkt(peer, pkt, __func__);
peer->local.commit->sig = tal(peer->local.commit,
secp256k1_ecdsa_signature);
err = accept_pkt_open_commit_sig(peer, pkt,
peer->local.commit->sig);
if (!err &&
!check_tx_sig(peer->local.commit->tx, 0,
NULL,
peer->anchor.witnessscript,
&peer->remote.commitkey,
peer->local.commit->sig))
err = pkt_err(peer, "Bad signature");
if (err) {
peer->local.commit->sig = tal_free(peer->local.commit->sig);
return peer_comms_err(peer, err);
}
peer->their_commitsigs++;
db_start_transaction(peer);
db_set_anchor(peer);
db_new_commit_info(peer, LOCAL, NULL);
set_peer_state(peer,
STATE_OPEN_WAIT_ANCHORDEPTH_AND_THEIRCOMPLETE,
__func__, true);
if (db_commit_transaction(peer) != NULL)
return peer_database_err(peer);
broadcast_tx(peer->dstate->topology,
peer, peer->anchor.tx, funding_tx_failed);
peer_watch_anchor(peer, peer->local.mindepth);
return true;
}
static bool open_theiranchor_pkt_in(struct peer *peer, const Pkt *pkt)
{
Pkt *err;
const char *db_err;
if (pkt->pkt_case != PKT__PKT_OPEN_ANCHOR)
return peer_received_unexpected_pkt(peer, pkt, __func__);
err = accept_pkt_anchor(peer, pkt);
if (err) {
peer_open_complete(peer, err->error->problem);
return peer_comms_err(peer, err);
}
peer->anchor.ours = false;
if (!setup_first_commit(peer)) {
err = pkt_err(peer, "Insufficient funds for fee");
peer_open_complete(peer, err->error->problem);
return peer_comms_err(peer, err);
}
log_debug_struct(peer->log, "Creating sig for %s",
struct bitcoin_tx,
peer->remote.commit->tx);
log_add_struct(peer->log, " using key %s",
struct pubkey, &peer->local.commitkey);
peer->remote.commit->sig = tal(peer->remote.commit,
secp256k1_ecdsa_signature);
peer_sign_theircommit(peer, peer->remote.commit->tx,
peer->remote.commit->sig);
peer->remote.commit->order = peer->order_counter++;
db_start_transaction(peer);
db_set_anchor(peer);
db_new_commit_info(peer, REMOTE, NULL);
peer_add_their_commit(peer,
&peer->remote.commit->txid,
peer->remote.commit->commit_num);
set_peer_state(peer, STATE_OPEN_WAIT_ANCHORDEPTH_AND_THEIRCOMPLETE,
__func__, true);
db_err = db_commit_transaction(peer);
if (db_err) {
peer_open_complete(peer, db_err);
return peer_database_err(peer);
}
queue_pkt_open_commit_sig(peer);
peer_watch_anchor(peer, peer->local.mindepth);
return true;
}
/* Dump all known channels and nodes to the peer. Used when a new
* connection was established. */
static void sync_routing_table(struct lightningd_state *dstate, struct peer *peer)
{
struct node *n;
struct node_map_iter it;
int i;
struct node_connection *nc;
for (n = node_map_first(dstate->rstate->nodes, &it); n; n = node_map_next(dstate->rstate->nodes, &it)) {
size_t num_edges = tal_count(n->out);
for (i = 0; i < num_edges; i++) {
nc = n->out[i];
if (nc->channel_announcement)
queue_pkt_nested(peer, WIRE_CHANNEL_ANNOUNCEMENT, nc->channel_announcement);
if (nc->channel_update)
queue_pkt_nested(peer, WIRE_CHANNEL_UPDATE, nc->channel_update);
}
if (n->node_announcement && num_edges > 0)
queue_pkt_nested(peer, WIRE_NODE_ANNOUNCEMENT, n->node_announcement);
}
}
static bool open_wait_pkt_in(struct peer *peer, const Pkt *pkt)
{
Pkt *err;
const char *db_err;
/* If they want to shutdown during this, we do mutual close dance. */
if (pkt->pkt_case == PKT__PKT_CLOSE_SHUTDOWN) {
err = accept_pkt_close_shutdown(peer, pkt);
if (err)
return peer_comms_err(peer, err);
peer_open_complete(peer, "Shutdown request received");
db_start_transaction(peer);
db_set_their_closing_script(peer);
start_closing_in_transaction(peer);
if (db_commit_transaction(peer) != NULL)
return peer_database_err(peer);
return false;
}
switch (peer->state) {
case STATE_OPEN_WAIT_ANCHORDEPTH_AND_THEIRCOMPLETE:
case STATE_OPEN_WAIT_THEIRCOMPLETE:
if (pkt->pkt_case != PKT__PKT_OPEN_COMPLETE)
return peer_received_unexpected_pkt(peer, pkt, __func__);
err = accept_pkt_open_complete(peer, pkt);
if (err) {
peer_open_complete(peer, err->error->problem);
return peer_comms_err(peer, err);
}
db_start_transaction(peer);
if (peer->state == STATE_OPEN_WAIT_THEIRCOMPLETE) {
peer_open_complete(peer, NULL);
set_peer_state(peer, STATE_NORMAL, __func__, true);
announce_channel(peer->dstate, peer);
sync_routing_table(peer->dstate, peer);
} else {
set_peer_state(peer, STATE_OPEN_WAIT_ANCHORDEPTH,
__func__, true);
}
db_err = db_commit_transaction(peer);
if (db_err) {
peer_open_complete(peer, db_err);
return peer_database_err(peer);
}
return true;
case STATE_OPEN_WAIT_ANCHORDEPTH:
return peer_received_unexpected_pkt(peer, pkt, __func__);
default:
log_unusual(peer->log,
"%s: unexpected state %s",
__func__, state_name(peer->state));
peer_fail(peer, __func__);
return false;
}
}
static void set_htlc_rval(struct peer *peer,
struct htlc *htlc, const struct preimage *rval)
{
assert(!htlc->r);
assert(!htlc->fail);
htlc->r = tal_dup(htlc, struct preimage, rval);
db_htlc_fulfilled(peer, htlc);
}
static void set_htlc_fail(struct peer *peer,
struct htlc *htlc, const void *fail, size_t len)
{
assert(!htlc->r);
assert(!htlc->fail);
htlc->fail = tal_dup_arr(htlc, u8, fail, len, 0);
db_htlc_failed(peer, htlc);
}
static void route_htlc_onwards(struct peer *peer,
struct htlc *htlc,
u64 msatoshi,
const u8 *pb_id,
const u8 *rest_of_route,
const struct peer *only_dest)
{
struct peer *next;
struct htlc *newhtlc;
enum fail_error error_code;
const char *err;
if (!only_dest) {
log_debug_struct(peer->log, "Forwarding HTLC %s",
struct sha256, &htlc->rhash);
log_add(peer->log, " (id %"PRIu64")", htlc->id);
}
next = find_peer_by_pkhash(peer->dstate, pb_id);
if (!next || !next->nc) {
log_unusual(peer->log, "Can't route HTLC %"PRIu64": no %speer ",
htlc->id, next ? "ready " : "");
if (!peer->dstate->dev_never_routefail)
command_htlc_set_fail(peer, htlc, NOT_FOUND_404,
"Unknown peer");
return;
}
if (only_dest && next != only_dest)
return;
/* Offered fee must be sufficient. */
if ((s64)(htlc->msatoshi - msatoshi)
< connection_fee(next->nc, msatoshi)) {
log_unusual(peer->log,
"Insufficient fee for HTLC %"PRIu64
": %"PRIi64" on %"PRIu64,
htlc->id, htlc->msatoshi - msatoshi,
msatoshi);
command_htlc_set_fail(peer, htlc, PAYMENT_REQUIRED_402,
"Insufficent fee");
return;
}
log_debug_struct(peer->log, "HTLC forward to %s",
struct pubkey, next->id);
/* This checks the HTLC itself is possible. */
err = command_htlc_add(next, msatoshi,
abs_locktime_to_blocks(&htlc->expiry)
- next->nc->delay,
&htlc->rhash, htlc, rest_of_route,
&error_code, &newhtlc);
if (err)
command_htlc_set_fail(peer, htlc, error_code, err);
}
static void their_htlc_added(struct peer *peer, struct htlc *htlc,
struct peer *only_dest)
{
struct invoice *invoice;
struct onionpacket *packet;
struct route_step *step = NULL;
if (abs_locktime_is_seconds(&htlc->expiry)) {
log_unusual(peer->log, "HTLC %"PRIu64" is in seconds", htlc->id);
command_htlc_set_fail(peer, htlc, BAD_REQUEST_400,
"bad locktime");
return;
}
if (abs_locktime_to_blocks(&htlc->expiry) <=
get_block_height(peer->dstate->topology) + peer->dstate->config.min_htlc_expiry) {
log_unusual(peer->log, "HTLC %"PRIu64" expires too soon:"
" block %u",
htlc->id, abs_locktime_to_blocks(&htlc->expiry));
command_htlc_set_fail(peer, htlc, BAD_REQUEST_400,
"expiry too soon");
return;
}
if (abs_locktime_to_blocks(&htlc->expiry) >
get_block_height(peer->dstate->topology) + peer->dstate->config.max_htlc_expiry) {
log_unusual(peer->log, "HTLC %"PRIu64" expires too far:"
" block %u",
htlc->id, abs_locktime_to_blocks(&htlc->expiry));
command_htlc_set_fail(peer, htlc, BAD_REQUEST_400,
"expiry too far");
return;
}
packet = parse_onionpacket(peer,
htlc->routing, tal_count(htlc->routing));
if (packet) {
u8 shared_secret[32];
if (onion_shared_secret(shared_secret, packet,
peer->dstate->privkey))
step = process_onionpacket(packet, packet,
shared_secret,
htlc->rhash.u.u8,
sizeof(htlc->rhash));
}
if (!step) {
log_unusual(peer->log, "Bad onion, failing HTLC %"PRIu64,
htlc->id);
command_htlc_set_fail(peer, htlc, BAD_REQUEST_400,
"invalid onion");
goto free_packet;
}
switch (step->nextcase) {
case ONION_END:
if (only_dest)
goto free_packet;
invoice = find_unpaid(peer->dstate->invoices, &htlc->rhash);
if (!invoice) {
log_unusual(peer->log, "No invoice for HTLC %"PRIu64,
htlc->id);
log_add_struct(peer->log, " rhash=%s",
struct sha256, &htlc->rhash);
if (unlikely(!peer->dstate->dev_never_routefail))
command_htlc_set_fail(peer, htlc,
UNAUTHORIZED_401,
"unknown rhash");
goto free_packet;
}
if (htlc->msatoshi != invoice->msatoshi) {
log_unusual(peer->log, "Short payment for '%s' HTLC %"PRIu64
": %"PRIu64" not %"PRIu64 " satoshi!",
invoice->label,
htlc->id,
htlc->msatoshi,
invoice->msatoshi);
command_htlc_set_fail(peer, htlc,
UNAUTHORIZED_401,
"incorrect amount");
goto free_packet;
}
log_info(peer->log, "Immediately resolving '%s' HTLC %"PRIu64,
invoice->label, htlc->id);
resolve_invoice(peer->dstate, invoice);
set_htlc_rval(peer, htlc, &invoice->r);
command_htlc_fulfill(peer, htlc);
goto free_packet;
case ONION_FORWARD:
route_htlc_onwards(peer, htlc, step->hoppayload->amt_to_forward, step->next->nexthop,
serialize_onionpacket(step, step->next), only_dest);
goto free_packet;
default:
log_info(peer->log, "Unknown step type %u", step->nextcase);
command_htlc_set_fail(peer, htlc, VERSION_NOT_SUPPORTED_505,
"unknown step type");
goto free_packet;
}
free_packet:
tal_free(packet);
}
static void our_htlc_failed(struct peer *peer, struct htlc *htlc)
{
assert(htlc_owner(htlc) == LOCAL);
if (htlc->src) {
set_htlc_fail(htlc->src->peer, htlc->src,
htlc->fail, tal_count(htlc->fail));
command_htlc_fail(htlc->src->peer, htlc->src);
} else
complete_pay_command(peer->dstate, htlc);
}
static void our_htlc_fulfilled(struct peer *peer, struct htlc *htlc)
{
if (htlc->src) {
set_htlc_rval(htlc->src->peer, htlc->src, htlc->r);
command_htlc_fulfill(htlc->src->peer, htlc->src);
} else {
complete_pay_command(peer->dstate, htlc);
}
}
/* FIXME: Slow! */
static struct htlc *htlc_with_source(struct peer *peer, struct htlc *src)
{
struct htlc_map_iter it;
struct htlc *h;
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
if (h->src == src)
return h;
}
return NULL;
}
/* peer has come back online: re-send any we have to send to them. */
static void retry_all_routing(struct peer *restarted_peer)
{
struct peer *peer;
struct htlc_map_iter it;
struct htlc *h;
/* Look for added htlcs from other peers which need to go here. */
list_for_each(&restarted_peer->dstate->peers, peer, list) {
if (peer == restarted_peer)
continue;
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
if (h->state != RCVD_ADD_ACK_REVOCATION)
continue;
if (htlc_with_source(peer, h))
continue;
their_htlc_added(peer, h, restarted_peer);
}
}
/* Catch any HTLCs which are fulfilled, but the message got reset
* by reconnect. */
for (h = htlc_map_first(&restarted_peer->htlcs, &it);
h;
h = htlc_map_next(&restarted_peer->htlcs, &it)) {
if (h->state != RCVD_ADD_ACK_REVOCATION)
continue;
if (h->r)
command_htlc_fulfill(restarted_peer, h);
else if (h->fail)
command_htlc_fail(restarted_peer, h);
}
}
static bool adjust_cstate_side(struct channel_state *cstate,
struct htlc *h,
enum htlc_state old, enum htlc_state new,
enum side side)
{
int oldf = htlc_state_flags(old), newf = htlc_state_flags(new);
bool old_committed, new_committed;
/* We applied changes to staging_cstate when we first received
* add/remove packet, so we could make sure it was valid. Don't
* do that again. */
if (old == SENT_ADD_HTLC || old == RCVD_REMOVE_HTLC
|| old == RCVD_ADD_HTLC || old == SENT_REMOVE_HTLC)
return true;
old_committed = (oldf & HTLC_FLAG(side, HTLC_F_COMMITTED));
new_committed = (newf & HTLC_FLAG(side, HTLC_F_COMMITTED));
if (old_committed && !new_committed) {
if (h->r)
cstate_fulfill_htlc(cstate, h);
else
cstate_fail_htlc(cstate, h);
} else if (!old_committed && new_committed) {
if (!cstate_add_htlc(cstate, h, false)) {
log_broken_struct(h->peer->log,
"Cannot afford htlc %s",
struct htlc, h);
log_add_struct(h->peer->log, " channel state %s",
struct channel_state, cstate);
return false;
}
}
return true;
}
/* We apply changes to staging_cstate when we first PENDING, so we can
* make sure they're valid. So here we change the staging_cstate on
* the revocation receive (ie. when acked). */
static bool adjust_cstates(struct peer *peer, struct htlc *h,
enum htlc_state old, enum htlc_state new)
{
return adjust_cstate_side(peer->remote.staging_cstate, h, old, new,
REMOTE)
&& adjust_cstate_side(peer->local.staging_cstate, h, old, new,
LOCAL);
}
static void adjust_cstate_fee_side(struct channel_state *cstate,
const struct feechange *f,
enum feechange_state old,
enum feechange_state new,
enum side side)
{
/* We applied changes to staging_cstate when we first received
* feechange packet, so we could make sure it was valid. Don't
* do that again. */
if (old == SENT_FEECHANGE || old == RCVD_FEECHANGE)
return;
/* Feechanges only ever get applied to the side which created them:
* ours gets applied when they ack, theirs gets applied when we ack. */
if (side == LOCAL && new == RCVD_FEECHANGE_REVOCATION)
adjust_fee(cstate, f->fee_rate);
else if (side == REMOTE && new == SENT_FEECHANGE_REVOCATION)
adjust_fee(cstate, f->fee_rate);
}
static void adjust_cstates_fee(struct peer *peer, const struct feechange *f,
enum feechange_state old,
enum feechange_state new)
{
adjust_cstate_fee_side(peer->remote.staging_cstate, f, old, new, REMOTE);
adjust_cstate_fee_side(peer->local.staging_cstate, f, old, new, LOCAL);
}
static void check_both_committed(struct peer *peer, struct htlc *h)
{
if (!htlc_has(h, HTLC_ADDING) && !htlc_has(h, HTLC_REMOVING))
log_debug(peer->log,
"Both committed to %s of %s HTLC %"PRIu64 "(%s)",
h->state == SENT_ADD_ACK_REVOCATION
|| h->state == RCVD_ADD_ACK_REVOCATION ? "ADD"
: h->r ? "FULFILL" : "FAIL",
htlc_owner(h) == LOCAL ? "our" : "their",
h->id, htlc_state_name(h->state));
switch (h->state) {
case RCVD_REMOVE_ACK_REVOCATION:
/* If it was fulfilled, we handled it immediately. */
if (h->fail)
our_htlc_failed(peer, h);
break;
case RCVD_ADD_ACK_REVOCATION:
their_htlc_added(peer, h, NULL);
break;
default:
break;
}
}
struct htlcs_table {
enum htlc_state from, to;
};
struct feechanges_table {
enum feechange_state from, to;
};
static const char *changestates(struct peer *peer,
const struct htlcs_table *table,
size_t n,
const struct feechanges_table *ftable,
size_t n_ftable,
bool db_commit)
{
struct htlc_map_iter it;
struct htlc *h;
bool changed = false;
size_t i;
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
for (i = 0; i < n; i++) {
if (h->state == table[i].from) {
if (!adjust_cstates(peer, h,
table[i].from, table[i].to))
return "accounting error";
htlc_changestate(h, table[i].from,
table[i].to, db_commit);
check_both_committed(peer, h);
changed = true;
}
}
}
for (i = 0; i < n_ftable; i++) {
struct feechange *f = peer->feechanges[ftable[i].from];
if (!f)
continue;
adjust_cstates_fee(peer, f, ftable[i].from, ftable[i].to);
feechange_changestate(peer, f,
ftable[i].from, ftable[i].to,
db_commit);
changed = true;
}
/* FIXME-OLD #2:
*
* A node MUST NOT send an `update_commit` message which does
* not include any updates.
*/
if (!changed)
return "no changes made";
return NULL;
}
/* This is the io loop while we're negotiating closing tx. */
static bool closing_pkt_in(struct peer *peer, const Pkt *pkt)
{
const CloseSignature *c = pkt->close_signature;
struct bitcoin_tx *close_tx;
secp256k1_ecdsa_signature theirsig;
assert(peer->state == STATE_MUTUAL_CLOSING);
if (pkt->pkt_case != PKT__PKT_CLOSE_SIGNATURE)
return peer_received_unexpected_pkt(peer, pkt, __func__);
log_info(peer->log, "closing_pkt_in: they offered close fee %"PRIu64,
c->close_fee);
/* FIXME-OLD #2:
*
* The sender MUST set `close_fee` lower than or equal to the fee of the
* final commitment transaction, and MUST set `close_fee` to an even
* number of satoshis.
*/
if ((c->close_fee & 1)
|| c->close_fee > commit_tx_fee(peer->remote.commit->tx,
peer->anchor.satoshis)) {
return peer_comms_err(peer, pkt_err(peer, "Invalid close fee"));
}
/* FIXME: Don't accept tiny fee at all? */
/* FIXME-OLD #2:
... otherwise it SHOULD propose a
value strictly between the received `close_fee` and its
previously-sent `close_fee`.
*/
if (peer->closing.their_sig) {
/* We want more, they should give more. */
if (peer->closing.our_fee > peer->closing.their_fee) {
if (c->close_fee <= peer->closing.their_fee)
return peer_comms_err(peer,
pkt_err(peer, "Didn't increase close fee"));
} else {
if (c->close_fee >= peer->closing.their_fee)
return peer_comms_err(peer,
pkt_err(peer, "Didn't decrease close fee"));
}
}
/* FIXME-OLD #2:
*
* The receiver MUST check `sig` is valid for the close
* transaction with the given `close_fee`, and MUST fail the
* connection if it is not. */
if (!proto_to_signature(c->sig, &theirsig))
return peer_comms_err(peer,
pkt_err(peer, "Invalid signature format"));
close_tx = peer_create_close_tx(c, peer, c->close_fee);
if (!check_tx_sig(close_tx, 0,
NULL,
peer->anchor.witnessscript,
&peer->remote.commitkey, &theirsig))
return peer_comms_err(peer,
pkt_err(peer, "Invalid signature"));
tal_free(peer->closing.their_sig);
peer->closing.their_sig = tal_dup(peer,
secp256k1_ecdsa_signature, &theirsig);
peer->closing.their_fee = c->close_fee;
peer->closing.sigs_in++;
if (!db_update_their_closing(peer))
return peer_database_err(peer);
if (peer->closing.our_fee != peer->closing.their_fee) {
/* FIXME-OLD #2:
*
* If the receiver agrees with the fee, it SHOULD reply with a
* `close_signature` with the same `close_fee` value,
* otherwise it SHOULD propose a value strictly between the
* received `close_fee` and its previously-sent `close_fee`.
*/
/* Adjust our fee to close on their fee. */
u64 sum;
/* Beware overflow! */
sum = (u64)peer->closing.our_fee + peer->closing.their_fee;
peer->closing.our_fee = sum / 2;
if (peer->closing.our_fee & 1)
peer->closing.our_fee++;
log_info(peer->log, "accept_pkt_close_sig: we change to %"PRIu64,
peer->closing.our_fee);
peer->closing.closing_order = peer->order_counter++;
db_start_transaction(peer);
db_update_our_closing(peer);
if (db_commit_transaction(peer) != NULL)
return peer_database_err(peer);
queue_pkt_close_signature(peer);
}
/* Note corner case: we may *now* agree with them! */
if (peer->closing.our_fee == peer->closing.their_fee) {
const struct bitcoin_tx *close;
log_info(peer->log, "accept_pkt_close_sig: we agree");
/* FIXME-OLD #2:
*
* Once a node has sent or received a `close_signature` with
* matching `close_fee` it SHOULD close the connection and
* SHOULD sign and broadcast the final closing transaction.
*/
close = mk_bitcoin_close(peer, peer);
broadcast_tx(peer->dstate->topology, peer, close, NULL);
tal_free(close);
return false;
}
return true;
}
/* We can get update_commit in both normal and shutdown states. */
static Pkt *handle_pkt_commit(struct peer *peer, const Pkt *pkt)
{
Pkt *err;
const char *errmsg;
struct sha256 preimage;
struct commit_info *ci;
bool to_them_only;
/* FIXME: We can actually merge these two... */
static const struct htlcs_table commit_changes[] = {
{ RCVD_ADD_REVOCATION, RCVD_ADD_ACK_COMMIT },
{ RCVD_REMOVE_HTLC, RCVD_REMOVE_COMMIT },
{ RCVD_ADD_HTLC, RCVD_ADD_COMMIT },
{ RCVD_REMOVE_REVOCATION, RCVD_REMOVE_ACK_COMMIT }
};
static const struct feechanges_table commit_feechanges[] = {
{ RCVD_FEECHANGE_REVOCATION, RCVD_FEECHANGE_ACK_COMMIT },
{ RCVD_FEECHANGE, RCVD_FEECHANGE_COMMIT }
};
static const struct htlcs_table revocation_changes[] = {
{ RCVD_ADD_ACK_COMMIT, SENT_ADD_ACK_REVOCATION },
{ RCVD_REMOVE_COMMIT, SENT_REMOVE_REVOCATION },
{ RCVD_ADD_COMMIT, SENT_ADD_REVOCATION },
{ RCVD_REMOVE_ACK_COMMIT, SENT_REMOVE_ACK_REVOCATION }
};
static const struct feechanges_table revocation_feechanges[] = {
{ RCVD_FEECHANGE_ACK_COMMIT, SENT_FEECHANGE_ACK_REVOCATION },
{ RCVD_FEECHANGE_COMMIT, SENT_FEECHANGE_REVOCATION }
};
ci = new_commit_info(peer, peer->local.commit->commit_num + 1);
db_start_transaction(peer);
/* FIXME-OLD #2:
*
* A node MUST NOT send an `update_commit` message which does
* not include any updates.
*/
errmsg = changestates(peer,
commit_changes, ARRAY_SIZE(commit_changes),
commit_feechanges, ARRAY_SIZE(commit_feechanges),
true);
if (errmsg) {
db_abort_transaction(peer);
return pkt_err(peer, "%s", errmsg);
}
/* Create new commit info for this commit tx. */
ci->revocation_hash = peer->local.next_revocation_hash;
/* FIXME-OLD #2:
*
* A receiving node MUST apply all local acked and unacked
* changes except unacked fee changes to the local commitment
*/
/* (We already applied them to staging_cstate as we went) */
ci->cstate = copy_cstate(ci, peer->local.staging_cstate);
ci->tx = create_commit_tx(ci, peer, &ci->revocation_hash,
ci->cstate, LOCAL, &to_them_only);
bitcoin_txid(ci->tx, &ci->txid);
log_debug(peer->log, "Check tx %"PRIu64" sig", ci->commit_num);
log_add_struct(peer->log, " for %s", struct channel_state, ci->cstate);
log_add_struct(peer->log, " (txid %s)", struct sha256_double, &ci->txid);
/* FIXME-OLD #2:
*
* If the commitment transaction has only a single output which pays
* to the other node, `sig` MUST be unset. Otherwise, a sending node
* MUST apply all remote acked and unacked changes except unacked fee
* changes to the remote commitment before generating `sig`.
*/
if (!to_them_only)
ci->sig = tal(ci, secp256k1_ecdsa_signature);
err = accept_pkt_commit(peer, pkt, ci->sig);
if (err)
return err;
/* FIXME-OLD #2:
*
* A receiving node MUST apply all local acked and unacked changes
* except unacked fee changes to the local commitment, then it MUST
* check `sig` is valid for that transaction.
*/
if (ci->sig && !check_tx_sig(ci->tx, 0,
NULL,
peer->anchor.witnessscript,
&peer->remote.commitkey,
ci->sig)) {
db_abort_transaction(peer);
return pkt_err(peer, "Bad signature");
}
/* Switch to the new commitment. */
tal_free(peer->local.commit);
peer->local.commit = ci;
peer->local.commit->order = peer->order_counter++;
db_new_commit_info(peer, LOCAL, NULL);
peer_get_revocation_hash(peer, ci->commit_num + 1,
&peer->local.next_revocation_hash);
peer->their_commitsigs++;
/* Now, send the revocation. */
/* We have their signature on the current one, right? */
assert(to_them_only || peer->local.commit->sig);
assert(peer->local.commit->commit_num > 0);
errmsg = changestates(peer,
revocation_changes, ARRAY_SIZE(revocation_changes),
revocation_feechanges,
ARRAY_SIZE(revocation_feechanges),
true);
if (errmsg) {
log_broken(peer->log, "queue_pkt_revocation: %s", errmsg);
db_abort_transaction(peer);
return pkt_err(peer, "Database error");
}
peer_get_revocation_preimage(peer, peer->local.commit->commit_num - 1,
&preimage);
/* Fire off timer if this ack caused new changes */
if (peer_uncommitted_changes(peer))
remote_changes_pending(peer);
queue_pkt_revocation(peer, &preimage, &peer->local.next_revocation_hash);
/* If we're shutting down and no more HTLCs, begin closing */
if (peer->closing.their_script && !committed_to_htlcs(peer))
start_closing_in_transaction(peer);
if (db_commit_transaction(peer) != NULL)
return pkt_err(peer, "Database error");
return NULL;
}
static Pkt *handle_pkt_htlc_add(struct peer *peer, const Pkt *pkt)
{
struct htlc *htlc;
Pkt *err;
err = accept_pkt_htlc_add(peer, pkt, &htlc);
if (err)
return err;
assert(htlc->state == RCVD_ADD_HTLC);
/* FIXME-OLD #2:
*
* A node MUST NOT offer `amount_msat` it cannot pay for in
* the remote commitment transaction at the current `fee_rate` (see
* "Fee Calculation" ). A node SHOULD fail the connection if
* this occurs.
*/
if (!cstate_add_htlc(peer->local.staging_cstate, htlc, true)) {
u64 id = htlc->id;
log_broken_struct(peer->log, "They cannot afford htlc %s",
struct htlc, htlc);
log_add_struct(peer->log, " cstate %s",
struct channel_state,
peer->local.staging_cstate);
tal_free(htlc);
return pkt_err(peer, "Cannot afford htlc %"PRIu64, id);
}
return NULL;
}
static Pkt *handle_pkt_htlc_fail(struct peer *peer, const Pkt *pkt)
{
struct htlc *htlc;
u8 *fail;
Pkt *err;
err = accept_pkt_htlc_fail(peer, pkt, &htlc, &fail);
if (err)
return err;
/* This can happen with re-transmissions; simply note it. */
if (htlc->fail) {
log_debug(peer->log, "HTLC %"PRIu64" failed twice", htlc->id);
htlc->fail = tal_free(htlc->fail);
}
db_start_transaction(peer);
set_htlc_fail(peer, htlc, fail, tal_count(fail));
tal_free(fail);
if (db_commit_transaction(peer) != NULL)
return pkt_err(peer, "database error");
cstate_fail_htlc(peer->local.staging_cstate, htlc);
/* FIXME-OLD #2:
*
* ... and the receiving node MUST add the HTLC fulfill/fail
* to the unacked changeset for its local commitment.
*/
htlc_changestate(htlc, SENT_ADD_ACK_REVOCATION, RCVD_REMOVE_HTLC, false);
return NULL;
}
static Pkt *handle_pkt_htlc_fulfill(struct peer *peer, const Pkt *pkt)
{
struct htlc *htlc;
Pkt *err;
struct preimage r;
err = accept_pkt_htlc_fulfill(peer, pkt, &htlc, &r);
if (err)
return err;
/* Reconnect may mean HTLC was already fulfilled. That's OK. */
if (!htlc->r) {
db_start_transaction(peer);
set_htlc_rval(peer, htlc, &r);
/* We can relay this upstream immediately. */
our_htlc_fulfilled(peer, htlc);
if (db_commit_transaction(peer) != NULL)
return pkt_err(peer, "database error");
}
/* FIXME-OLD #2:
*
* ... and the receiving node MUST add the HTLC fulfill/fail
* to the unacked changeset for its local commitment.
*/
cstate_fulfill_htlc(peer->local.staging_cstate, htlc);
htlc_changestate(htlc, SENT_ADD_ACK_REVOCATION, RCVD_REMOVE_HTLC, false);
return NULL;
}
static void set_feechange(struct peer *peer, u64 fee_rate,
enum feechange_state state)
{
/* If we already have a feechange for this commit, simply update it. */
if (peer->feechanges[state]) {
log_debug(peer->log, "Feechange: fee %"PRIu64" to %"PRIu64,
peer->feechanges[state]->fee_rate,
fee_rate);
peer->feechanges[state]->fee_rate = fee_rate;
} else {
log_debug(peer->log, "Feechange: New fee %"PRIu64, fee_rate);
peer->feechanges[state] = new_feechange(peer, fee_rate, state);
}
}
static Pkt *handle_pkt_feechange(struct peer *peer, const Pkt *pkt)
{
u64 feerate;
Pkt *err;
err = accept_pkt_update_fee(peer, pkt, &feerate);
if (err)
return err;
/* FIXME-OLD #2:
*
* The sending node MUST NOT send a `fee_rate` which it could not
* afford (see "Fee Calculation), were it applied to the receiving
* node's commitment transaction. The receiving node SHOULD fail the
* connection if this occurs.
*/
if (!can_afford_feerate(peer->local.staging_cstate, feerate, REMOTE))
return pkt_err(peer, "Cannot afford feerate %"PRIu64,
feerate);
set_feechange(peer, feerate, RCVD_FEECHANGE);
return NULL;
}
static Pkt *handle_pkt_revocation(struct peer *peer, const Pkt *pkt,
enum state next_state)
{
Pkt *err;
const char *errmsg;
static const struct htlcs_table changes[] = {
{ SENT_ADD_COMMIT, RCVD_ADD_REVOCATION },
{ SENT_REMOVE_ACK_COMMIT, RCVD_REMOVE_ACK_REVOCATION },
{ SENT_ADD_ACK_COMMIT, RCVD_ADD_ACK_REVOCATION },
{ SENT_REMOVE_COMMIT, RCVD_REMOVE_REVOCATION }
};
static const struct feechanges_table feechanges[] = {
{ SENT_FEECHANGE_COMMIT, RCVD_FEECHANGE_REVOCATION },
{ SENT_FEECHANGE_ACK_COMMIT, RCVD_FEECHANGE_ACK_REVOCATION }
};
err = accept_pkt_revocation(peer, pkt);
if (err)
return err;
/* FIXME-OLD #2:
*
* The receiver of `update_revocation`... MUST add the remote
* unacked changes to the set of local acked changes.
*/
db_start_transaction(peer);
errmsg = changestates(peer, changes, ARRAY_SIZE(changes),
feechanges, ARRAY_SIZE(feechanges), true);
if (errmsg) {
log_broken(peer->log, "accept_pkt_revocation: %s", errmsg);
db_abort_transaction(peer);
return pkt_err(peer, "failure accepting update_revocation: %s",
errmsg);
}
db_save_shachain(peer);
db_update_next_revocation_hash(peer);
set_peer_state(peer, next_state, __func__, true);
db_remove_their_prev_revocation_hash(peer);
/* If we're shutting down and no more HTLCs, begin closing */
if (peer->closing.their_script && !committed_to_htlcs(peer))
start_closing_in_transaction(peer);
if (db_commit_transaction(peer) != NULL)
return pkt_err(peer, "database error");
return NULL;
}
/* This is the io loop while we're doing shutdown. */
static bool shutdown_pkt_in(struct peer *peer, const Pkt *pkt)
{
Pkt *err = NULL;
assert(peer->state == STATE_SHUTDOWN
|| peer->state == STATE_SHUTDOWN_COMMITTING);
switch (pkt->pkt_case) {
case PKT__PKT_UPDATE_REVOCATION:
if (peer->state == STATE_SHUTDOWN)
return peer_received_unexpected_pkt(peer, pkt, __func__);
else {
err = handle_pkt_revocation(peer, pkt, STATE_SHUTDOWN);
if (!err)
peer_update_complete(peer);
}
break;
case PKT__PKT_UPDATE_ADD_HTLC:
/* FIXME-OLD #2:
*
* A node MUST NOT send a `update_add_htlc` after a
* `close_shutdown` */
if (peer->closing.their_script)
err = pkt_err(peer, "Update during shutdown");
else
err = handle_pkt_htlc_add(peer, pkt);
break;
case PKT__PKT_CLOSE_SHUTDOWN:
/* FIXME-OLD #2:
*
* A node... MUST NOT send more than one `close_shutdown`. */
if (peer->closing.their_script)
return peer_received_unexpected_pkt(peer, pkt, __func__);
else {
err = accept_pkt_close_shutdown(peer, pkt);
if (!err) {
db_start_transaction(peer);
db_set_their_closing_script(peer);
/* If no more HTLCs, we're closing. */
if (!committed_to_htlcs(peer))
start_closing_in_transaction(peer);
if (db_commit_transaction(peer) != NULL)
err = pkt_err(peer, "database error");
}
}
break;
case PKT__PKT_UPDATE_FULFILL_HTLC:
err = handle_pkt_htlc_fulfill(peer, pkt);
break;
case PKT__PKT_UPDATE_FAIL_HTLC:
err = handle_pkt_htlc_fail(peer, pkt);
break;
case PKT__PKT_UPDATE_FEE:
err = handle_pkt_feechange(peer, pkt);
break;
case PKT__PKT_UPDATE_COMMIT:
err = handle_pkt_commit(peer, pkt);
break;
case PKT__PKT_ERROR:
return peer_received_unexpected_pkt(peer, pkt, __func__);
case PKT__PKT_AUTH:
case PKT__PKT_OPEN:
case PKT__PKT_OPEN_ANCHOR:
case PKT__PKT_OPEN_COMMIT_SIG:
case PKT__PKT_OPEN_COMPLETE:
case PKT__PKT_CLOSE_SIGNATURE:
default:
return peer_received_unexpected_pkt(peer, pkt, __func__);
}
if (err)
return peer_comms_err(peer, err);
return true;
}
static bool do_commit(struct peer *peer, struct command *jsoncmd)
{
struct commit_info *ci;
const char *errmsg;
static const struct htlcs_table changes[] = {
{ SENT_ADD_HTLC, SENT_ADD_COMMIT },
{ SENT_REMOVE_REVOCATION, SENT_REMOVE_ACK_COMMIT },
{ SENT_ADD_REVOCATION, SENT_ADD_ACK_COMMIT},
{ SENT_REMOVE_HTLC, SENT_REMOVE_COMMIT}
};
static const struct feechanges_table feechanges[] = {
{ SENT_FEECHANGE, SENT_FEECHANGE_COMMIT },
{ SENT_FEECHANGE_REVOCATION, SENT_FEECHANGE_ACK_COMMIT}
};
bool to_us_only;
/* We can have changes we suggested, or changes they suggested. */
if (!peer_uncommitted_changes(peer)) {
log_debug(peer->log, "do_commit: no changes to commit");
if (jsoncmd)
command_fail(jsoncmd, "no changes to commit");
return true;
}
log_debug(peer->log, "do_commit: sending commit command %"PRIu64,
peer->remote.commit->commit_num + 1);
assert(state_can_commit(peer->state));
assert(!peer->commit_jsoncmd);
peer->commit_jsoncmd = jsoncmd;
ci = new_commit_info(peer, peer->remote.commit->commit_num + 1);
assert(!peer->their_prev_revocation_hash);
peer->their_prev_revocation_hash
= tal_dup(peer, struct sha256,
&peer->remote.commit->revocation_hash);
db_start_transaction(peer);
errmsg = changestates(peer, changes, ARRAY_SIZE(changes),
feechanges, ARRAY_SIZE(feechanges), true);
if (errmsg) {
log_broken(peer->log, "queue_pkt_commit: %s", errmsg);
goto database_error;
}
/* Create new commit info for this commit tx. */
ci->revocation_hash = peer->remote.next_revocation_hash;
/* FIXME-OLD #2:
*
* ...a sending node MUST apply all remote acked and unacked
* changes except unacked fee changes to the remote commitment
* before generating `sig`. */
ci->cstate = copy_cstate(ci, peer->remote.staging_cstate);
ci->tx = create_commit_tx(ci, peer, &ci->revocation_hash,
ci->cstate, REMOTE, &to_us_only);
bitcoin_txid(ci->tx, &ci->txid);
if (!to_us_only) {
log_debug(peer->log, "Signing tx %"PRIu64, ci->commit_num);
log_add_struct(peer->log, " for %s",
struct channel_state, ci->cstate);
log_add_struct(peer->log, " (txid %s)",
struct sha256_double, &ci->txid);
ci->sig = tal(ci, secp256k1_ecdsa_signature);
peer_sign_theircommit(peer, ci->tx, ci->sig);
}
/* Switch to the new commitment. */
tal_free(peer->remote.commit);
peer->remote.commit = ci;
peer->remote.commit->order = peer->order_counter++;
db_new_commit_info(peer, REMOTE, peer->their_prev_revocation_hash);
/* We don't need to remember their commit if we don't give sig. */
if (ci->sig)
peer_add_their_commit(peer, &ci->txid, ci->commit_num);
if (peer->state == STATE_SHUTDOWN) {
set_peer_state(peer, STATE_SHUTDOWN_COMMITTING, __func__, true);
} else {
assert(peer->state == STATE_NORMAL);
set_peer_state(peer, STATE_NORMAL_COMMITTING, __func__, true);
}
if (db_commit_transaction(peer) != NULL)
goto database_error;
queue_pkt_commit(peer, ci->sig);
return true;
database_error:
db_abort_transaction(peer);
peer_fail(peer, __func__);
return false;
}
static bool peer_start_shutdown(struct peer *peer)
{
enum state newstate;
u8 *redeemscript;
/* We might have uncommited changes; if so, commit them now. */
if (!do_commit(peer, NULL))
return false;
db_start_transaction(peer);
db_begin_shutdown(peer);
/* If they started close, we might not have sent ours. */
assert(!peer->closing.our_script);
redeemscript = bitcoin_redeem_single(peer, &peer->local.finalkey);
peer->closing.our_script = scriptpubkey_p2sh(peer, redeemscript);
tal_free(redeemscript);
/* FIXME-OLD #2:
*
* A node SHOULD send a `close_shutdown` (if it has
* not already) after receiving `close_shutdown`.
*/
peer->closing.shutdown_order = peer->order_counter++;
db_set_our_closing_script(peer);
queue_pkt_close_shutdown(peer);
if (peer->state == STATE_NORMAL_COMMITTING) {
newstate = STATE_SHUTDOWN_COMMITTING;
} else {
newstate = STATE_SHUTDOWN;
}
set_peer_state(peer, newstate, __func__, true);
/* Catch case where we've exchanged and had no HTLCs anyway. */
if (peer->closing.their_script && !committed_to_htlcs(peer))
start_closing_in_transaction(peer);
return db_commit_transaction(peer) == NULL;
}
/* Shim to handle the new packet format until we complete the
* switch. Handing the protobuf in anyway to fall back on protobuf
* based error handling. */
static bool nested_pkt_in(struct peer *peer, const u32 type,
const u8 *innerpkt, size_t innerpktlen,
const Pkt *pkt)
{
switch (type) {
case WIRE_CHANNEL_ANNOUNCEMENT:
handle_channel_announcement(peer->dstate->rstate, innerpkt, innerpktlen);
break;
case WIRE_CHANNEL_UPDATE:
handle_channel_update(peer->dstate->rstate, innerpkt, innerpktlen);
break;
case WIRE_NODE_ANNOUNCEMENT:
handle_node_announcement(peer->dstate->rstate, innerpkt, innerpktlen);
break;
default:
/* BOLT01: Unknown even typed packets MUST kill the
connection, unknown odd-typed packets MAY be ignored. */
if (type % 2 == 0){
return peer_received_unexpected_pkt(peer, pkt, __func__);
} else {
log_debug(peer->log, "Ignoring odd typed (%d) unknown packet.", type);
return true;
}
}
return true;
}
/* This is the io loop while we're in normal mode. */
static bool normal_pkt_in(struct peer *peer, const Pkt *pkt)
{
Pkt *err = NULL;
assert(peer->state == STATE_NORMAL
|| peer->state == STATE_NORMAL_COMMITTING);
switch (pkt->pkt_case) {
case PKT__PKT_UPDATE_ADD_HTLC:
err = handle_pkt_htlc_add(peer, pkt);
break;
case PKT__PKT_UPDATE_FULFILL_HTLC:
err = handle_pkt_htlc_fulfill(peer, pkt);
break;
case PKT__PKT_UPDATE_FAIL_HTLC:
err = handle_pkt_htlc_fail(peer, pkt);
break;
case PKT__PKT_UPDATE_FEE:
err = handle_pkt_feechange(peer, pkt);
break;
case PKT__PKT_UPDATE_COMMIT:
err = handle_pkt_commit(peer, pkt);
break;
case PKT__PKT_CLOSE_SHUTDOWN:
err = accept_pkt_close_shutdown(peer, pkt);
if (err)
break;
if (!peer_start_shutdown(peer)) {
err = pkt_err(peer, "database error");
break;
}
return true;
case PKT__PKT_UPDATE_REVOCATION:
if (peer->state == STATE_NORMAL_COMMITTING) {
err = handle_pkt_revocation(peer, pkt, STATE_NORMAL);
if (!err)
peer_update_complete(peer);
break;
}
/* Fall thru. */
default:
return peer_received_unexpected_pkt(peer, pkt, __func__);
}
if (err) {
return peer_comms_err(peer, err);
}
return true;
}
/* Create a HTLC fulfill transaction for onchain.tx[out_num]. */
static const struct bitcoin_tx *htlc_fulfill_tx(const struct peer *peer,
unsigned int out_num)
{
struct bitcoin_tx *tx = bitcoin_tx(peer, 1, 1);
const struct htlc *htlc = peer->onchain.htlcs[out_num];
const u8 *wscript = peer->onchain.wscripts[out_num];
secp256k1_ecdsa_signature sig;
u64 fee, satoshis;
assert(htlc->r);
tx->input[0].index = out_num;
tx->input[0].txid = peer->onchain.txid;
satoshis = htlc->msatoshi / 1000;
tx->input[0].amount = tal_dup(tx->input, u64, &satoshis);
tx->input[0].sequence_number = bitcoin_nsequence(&peer->remote.locktime);
/* Using a new output address here would be useless: they can tell
* it's their HTLC, and that we collected it via rval. */
tx->output[0].script = scriptpubkey_p2sh(tx,
bitcoin_redeem_single(tx,
&peer->local.finalkey));
log_debug(peer->log, "Pre-witness txlen = %zu\n",
measure_tx_cost(tx) / 4);
assert(measure_tx_cost(tx) == 83 * 4);
/* Witness length can vary, due to DER encoding of sigs, but we
* use 539 from an example run. */
fee = fee_by_feerate(83 + 539 / 4, get_feerate(peer->dstate->topology));
/* FIXME: Fail gracefully in these cases (not worth collecting) */
if (fee > satoshis || is_dust(satoshis - fee))
fatal("HTLC fulfill amount of %"PRIu64" won't cover fee %"PRIu64,
satoshis, fee);
tx->output[0].amount = satoshis - fee;
peer_sign_htlc_fulfill(peer, tx, wscript, &sig);
tx->input[0].witness = bitcoin_witness_htlc(tx,
htlc->r, &sig, wscript);
log_debug(peer->log, "tx cost for htlc fulfill tx: %zu",
measure_tx_cost(tx));
return tx;
}
static bool command_htlc_set_fail(struct peer *peer, struct htlc *htlc,
enum fail_error error_code, const char *why)
{
const u8 *fail = failinfo_create(htlc,
&peer->dstate->id, error_code, why);
set_htlc_fail(peer, htlc, fail, tal_count(fail));
tal_free(fail);
return command_htlc_fail(peer, htlc);
}
static bool command_htlc_fail(struct peer *peer, struct htlc *htlc)
{
/* If onchain, nothing we can do. */
if (!state_can_remove_htlc(peer->state))
return false;
/* FIXME-OLD #2:
*
* The sending node MUST add the HTLC fulfill/fail to the
* unacked changeset for its remote commitment
*/
cstate_fail_htlc(peer->remote.staging_cstate, htlc);
htlc_changestate(htlc, RCVD_ADD_ACK_REVOCATION, SENT_REMOVE_HTLC, false);
remote_changes_pending(peer);
queue_pkt_htlc_fail(peer, htlc);
return true;
}
/* FIXME-OLD #onchain:
*
* If the node receives... a redemption preimage for an unresolved *commitment
* tx* output it was offered, it MUST *resolve* the output by spending it using
* the preimage.
*/
static bool fulfill_onchain(struct peer *peer, struct htlc *htlc)
{
size_t i;
for (i = 0; i < tal_count(peer->onchain.htlcs); i++) {
if (peer->onchain.htlcs[i] == htlc) {
/* Already irrevocably resolved? */
if (peer->onchain.resolved[i])
return false;
peer->onchain.resolved[i]
= htlc_fulfill_tx(peer, i);
broadcast_tx(peer->dstate->topology,
peer, peer->onchain.resolved[i], NULL);
return true;
}
}
fatal("Unknown HTLC to fulfill onchain");
}
static bool command_htlc_fulfill(struct peer *peer, struct htlc *htlc)
{
if (peer->state == STATE_CLOSE_ONCHAIN_THEIR_UNILATERAL
|| peer->state == STATE_CLOSE_ONCHAIN_OUR_UNILATERAL) {
return fulfill_onchain(peer, htlc);
}
if (!state_can_remove_htlc(peer->state))
return false;
/* FIXME-OLD #2:
*
* The sending node MUST add the HTLC fulfill/fail to the
* unacked changeset for its remote commitment
*/
cstate_fulfill_htlc(peer->remote.staging_cstate, htlc);
htlc_changestate(htlc, RCVD_ADD_ACK_REVOCATION, SENT_REMOVE_HTLC, false);
remote_changes_pending(peer);
queue_pkt_htlc_fulfill(peer, htlc);
return true;
}
const char *command_htlc_add(struct peer *peer, u64 msatoshi,
unsigned int expiry,
const struct sha256 *rhash,
struct htlc *src,
const u8 *route,
u32 *error_code,
struct htlc **htlc)
{
struct abs_locktime locktime;
if (!blocks_to_abs_locktime(expiry, &locktime)) {
log_unusual(peer->log, "add_htlc: fail: bad expiry %u", expiry);
*error_code = BAD_REQUEST_400;
return "bad expiry";
}
if (expiry < get_block_height(peer->dstate->topology) + peer->dstate->config.min_htlc_expiry) {
log_unusual(peer->log, "add_htlc: fail: expiry %u is too soon",
expiry);
*error_code = BAD_REQUEST_400;
return "expiry too soon";
}
if (expiry > get_block_height(peer->dstate->topology) + peer->dstate->config.max_htlc_expiry) {
log_unusual(peer->log, "add_htlc: fail: expiry %u is too far",
expiry);
*error_code = BAD_REQUEST_400;
return "expiry too far";
}
/* FIXME-OLD #2:
*
* A node MUST NOT add a HTLC if it would result in it
* offering more than 300 HTLCs in the remote commitment transaction.
*/
if (peer->remote.staging_cstate->side[LOCAL].num_htlcs == 300) {
log_unusual(peer->log, "add_htlc: fail: already at limit");
*error_code = SERVICE_UNAVAILABLE_503;
return "channel full";
}
if (!state_can_add_htlc(peer->state)) {
log_unusual(peer->log, "add_htlc: fail: peer state %s",
state_name(peer->state));
*error_code = NOT_FOUND_404;
return "peer not available";
}
*htlc = peer_new_htlc(peer, peer->htlc_id_counter,
msatoshi, rhash, expiry, route, tal_count(route),
src, SENT_ADD_HTLC);
/* FIXME-OLD #2:
*
* The sending node MUST add the HTLC addition to the unacked
* changeset for its remote commitment
*/
if (!cstate_add_htlc(peer->remote.staging_cstate, *htlc, true)) {
/* FIXME-OLD #2:
*
* A node MUST NOT offer `amount_msat` it cannot pay for in
* the remote commitment transaction at the current `fee_rate`
*/
log_unusual(peer->log, "add_htlc: fail: Cannot afford %"PRIu64
" milli-satoshis in their commit tx",
msatoshi);
log_add_struct(peer->log, " channel state %s",
struct channel_state,
peer->remote.staging_cstate);
*htlc = tal_free(*htlc);
*error_code = SERVICE_UNAVAILABLE_503;
return "cannot afford htlc";
}
remote_changes_pending(peer);
queue_pkt_htlc_add(peer, *htlc);
/* Make sure we never offer the same one twice. */
peer->htlc_id_counter++;
return NULL;
}
static struct io_plan *pkt_out(struct io_conn *conn, struct peer *peer)
{
Pkt *out;
size_t n = tal_count(peer->outpkt);
if (n == 0) {
/* We close the connection once we've sent everything. */
if (!state_can_io(peer->state)) {
log_debug(peer->log, "pkt_out: no IO possible, closing");
return io_close(conn);
}
return io_out_wait(conn, peer, pkt_out, peer);
}
if (peer->fake_close || !peer->output_enabled)
return io_out_wait(conn, peer, pkt_out, peer);
out = peer->outpkt[0];
memmove(peer->outpkt, peer->outpkt + 1, (sizeof(*peer->outpkt)*(n-1)));
tal_resize(&peer->outpkt, n-1);
log_debug(peer->log, "pkt_out: writing %s", pkt_name(out->pkt_case));
return peer_write_packet(conn, peer, out, pkt_out);
}
static void clear_output_queue(struct peer *peer)
{
size_t i, n = tal_count(peer->outpkt);
for (i = 0; i < n; i++)
tal_free(peer->outpkt[i]);
tal_resize(&peer->outpkt, 0);
}
static struct io_plan *pkt_in(struct io_conn *conn, struct peer *peer)
{
bool keep_going = true;
/* We ignore packets if they tell us to, or we're closing already */
if (peer->fake_close || !state_can_io(peer->state))
keep_going = true;
/* Sidestep the state machine for nested packets */
else if (peer->inpkt->pkt_case == PKT__PKT_NESTED)
keep_going = nested_pkt_in(peer, peer->inpkt->nested->type,
peer->inpkt->nested->inner_pkt.data,
peer->inpkt->nested->inner_pkt.len,
peer->inpkt);
else if (state_is_normal(peer->state))
keep_going = normal_pkt_in(peer, peer->inpkt);
else if (state_is_shutdown(peer->state))
keep_going = shutdown_pkt_in(peer, peer->inpkt);
else if (peer->state == STATE_MUTUAL_CLOSING)
keep_going = closing_pkt_in(peer, peer->inpkt);
else if (peer->state == STATE_OPEN_WAIT_FOR_OPENPKT)
keep_going = open_pkt_in(peer, peer->inpkt);
else if (peer->state == STATE_OPEN_WAIT_FOR_COMMIT_SIGPKT)
keep_going = open_ouranchor_pkt_in(peer, peer->inpkt);
else if (peer->state == STATE_OPEN_WAIT_FOR_ANCHORPKT)
keep_going = open_theiranchor_pkt_in(peer, peer->inpkt);
else if (state_is_openwait(peer->state))
keep_going = open_wait_pkt_in(peer, peer->inpkt);
else {
log_unusual(peer->log,
"Unexpected state %s", state_name(peer->state));
keep_going = false;
}
peer->inpkt = tal_free(peer->inpkt);
if (keep_going)
return peer_read_packet(conn, peer, pkt_in);
else
return peer_close(conn, peer);
}
/*
* This only works because we send one update at a time, and they can't
* ask for it again if they've already sent the `update_revocation` acking it.
*/
static void retransmit_updates(struct peer *peer)
{
struct htlc_map_iter it;
struct htlc *h;
/* FIXME-OLD #2:
*
* A node MAY simply retransmit messages which are identical to the
* previous transmission. */
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
switch (h->state) {
case SENT_ADD_COMMIT:
log_debug(peer->log, "Retransmitting add HTLC %"PRIu64,
h->id);
queue_pkt_htlc_add(peer, h);
break;
case SENT_REMOVE_COMMIT:
log_debug(peer->log, "Retransmitting %s HTLC %"PRIu64,
h->r ? "fulfill" : "fail", h->id);
if (h->r)
queue_pkt_htlc_fulfill(peer, h);
else
queue_pkt_htlc_fail(peer, h);
break;
default:
break;
}
}
assert(!peer->feechanges[SENT_FEECHANGE]);
}
/* FIXME-OLD #2:
*
* On disconnection, a node MUST reverse any uncommitted changes sent by the
* other side (ie. `update_add_htlc`, `update_fee`, `update_fail_htlc` and
* `update_fulfill_htlc` for which no `update_commit` has been received). A
* node SHOULD retain the `r` value from the `update_fulfill_htlc`, however.
*/
static void forget_uncommitted_changes(struct peer *peer)
{
struct htlc *h;
struct htlc_map_iter it;
bool retry;
if (!peer->remote.commit || !peer->remote.commit->cstate)
return;
log_debug(peer->log, "Forgetting uncommitted");
log_debug_struct(peer->log, "LOCAL: changing from %s",
struct channel_state, peer->local.staging_cstate);
log_add_struct(peer->log, " to %s",
struct channel_state, peer->local.commit->cstate);
log_debug_struct(peer->log, "REMOTE: changing from %s",
struct channel_state, peer->remote.staging_cstate);
log_add_struct(peer->log, " to %s",
struct channel_state, peer->remote.commit->cstate);
tal_free(peer->local.staging_cstate);
tal_free(peer->remote.staging_cstate);
peer->local.staging_cstate
= copy_cstate(peer, peer->local.commit->cstate);
peer->remote.staging_cstate
= copy_cstate(peer, peer->remote.commit->cstate);
/* We forget everything we're routing, and re-send. This
* works for the reload-from-database case as well as the
* normal reconnect. */
again:
retry = false;
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
switch (h->state) {
case SENT_ADD_HTLC:
/* Adjust counter to lowest HTLC removed */
if (peer->htlc_id_counter > h->id) {
log_debug(peer->log,
"Lowering htlc_id_counter to %"PRIu64,
h->id);
peer->htlc_id_counter = h->id;
}
/* Fall thru */
case RCVD_ADD_HTLC:
log_debug(peer->log, "Forgetting %s %"PRIu64,
htlc_state_name(h->state), h->id);
/* May miss some due to delete reorg. */
tal_free(h);
retry = true;
break;
case RCVD_REMOVE_HTLC:
log_debug(peer->log, "Undoing %s %"PRIu64,
htlc_state_name(h->state), h->id);
htlc_undostate(h, RCVD_REMOVE_HTLC,
SENT_ADD_ACK_REVOCATION);
break;
case SENT_REMOVE_HTLC:
log_debug(peer->log, "Undoing %s %"PRIu64,
htlc_state_name(h->state), h->id);
htlc_undostate(h, SENT_REMOVE_HTLC,
RCVD_ADD_ACK_REVOCATION);
break;
default:
break;
}
}
if (retry)
goto again;
/* Forget uncommitted feechanges */
peer->feechanges[SENT_FEECHANGE]
= tal_free(peer->feechanges[SENT_FEECHANGE]);
peer->feechanges[RCVD_FEECHANGE]
= tal_free(peer->feechanges[RCVD_FEECHANGE]);
/* Make sure our HTLC counter is correct. */
if (peer->htlc_id_counter != 0)
assert(htlc_get(&peer->htlcs, peer->htlc_id_counter-1, LOCAL));
assert(!htlc_get(&peer->htlcs, peer->htlc_id_counter, LOCAL));
}
static void retransmit_pkts(struct peer *peer, s64 ack)
{
log_debug(peer->log, "Our order counter is %"PRIi64", their ack %"PRIi64,
peer->order_counter, ack);
if (ack > peer->order_counter) {
log_unusual(peer->log, "reconnect ack %"PRIi64" > %"PRIi64,
ack, peer->order_counter);
peer_comms_err(peer, pkt_err(peer, "invalid ack"));
return;
}
log_debug(peer->log, "They acked %"PRIi64", remote=%"PRIi64" local=%"PRIi64,
ack, peer->remote.commit ? peer->remote.commit->order : -2,
peer->local.commit ? peer->local.commit->order : -2);
/* FIXME-OLD #2:
*
* A node MAY assume that only one of each type of message need be
* retransmitted. A node SHOULD retransmit the last of each message
* type which was not counted by the `ack` field.
*/
while (ack < peer->order_counter) {
if (ack == 0) {
queue_pkt_open(peer, peer->local.offer_anchor);
} else if (ack == 1) {
queue_pkt_open_commit_sig(peer);
} else if (peer->remote.commit
&& ack == peer->remote.commit->order) {
/* FIXME-OLD #2:
*
* Before retransmitting `update_commit`, the node
* MUST send appropriate `update_add_htlc`,
* `update_fee`, `update_fail_htlc` or
* `update_fulfill_htlc` messages (the other node will
* have forgotten them, as required above).
*/
retransmit_updates(peer);
queue_pkt_commit(peer, peer->remote.commit->sig);
} else if (peer->local.commit
&& ack == peer->local.commit->order) {
/* Re-transmit revocation. */
struct sha256 preimage, next;
u64 commit_num = peer->local.commit->commit_num - 1;
/* Make sure we don't revoke current commit! */
assert(commit_num < peer->local.commit->commit_num);
peer_get_revocation_preimage(peer, commit_num,&preimage);
peer_get_revocation_hash(peer, commit_num + 2, &next);
log_debug(peer->log, "Re-sending revocation hash %"PRIu64,
commit_num + 2);
log_add_struct(peer->log, "value %s", struct sha256,
&next);
log_add_struct(peer->log, "local.next=%s", struct sha256,
&peer->local.next_revocation_hash);
log_debug(peer->log, "Re-sending revocation %"PRIu64,
commit_num);
queue_pkt_revocation(peer, &preimage, &next);
} else if (ack == peer->closing.shutdown_order) {
log_debug(peer->log, "Re-sending shutdown");
queue_pkt_close_shutdown(peer);
} else if (ack == peer->closing.closing_order) {
log_debug(peer->log, "Re-sending closing order");
queue_pkt_close_signature(peer);
} else {
log_broken(peer->log, "Can't rexmit %"PRIu64
" when local commit %"PRIi64" and remote %"PRIi64,
ack,
peer->local.commit ? peer->local.commit->order : -2,
peer->remote.commit ? peer->remote.commit->order : -2);
peer_comms_err(peer, pkt_err(peer, "invalid ack"));
return;
}
ack++;
}
/* We might need to update HTLCs which were from other peers. */
retry_all_routing(peer);
}
static u64 desired_commit_feerate(struct lightningd_state *dstate)
{
return get_feerate(dstate->topology) * dstate->config.commitment_fee_percent / 100;
}
static bool want_feechange(const struct peer *peer)
{
if (!state_is_normal(peer->state) && !state_is_shutdown(peer->state))
return false;
log_debug(peer->log, "Current fee_rate: %"PRIu64" want %"PRIu64,
peer->local.staging_cstate->fee_rate,
desired_commit_feerate(peer->dstate));
/* FIXME: Send fee changes when we want it */
return false;
}
static void peer_has_connected(struct peer *peer)
{
assert(!peer->connected);
peer->connected = true;
/* Do we want to send something? */
if (peer_uncommitted_changes(peer) || want_feechange(peer)) {
log_debug(peer->log, "connected: changes pending");
remote_changes_pending(peer);
}
}
static struct io_plan *init_pkt_in(struct io_conn *conn, struct peer *peer)
{
if (peer->inpkt->pkt_case != PKT__PKT_INIT) {
peer_received_unexpected_pkt(peer, peer->inpkt, __func__);
goto fail;
}
/* They might have missed the error, tell them before hanging up */
if (state_is_error(peer->state)) {
queue_pkt_err(peer, pkt_err(peer, "In error state %s",
state_name(peer->state)));
goto fail;
}
daemon/peer: handle narrow reconnect race on close. Usually if we get a packet while closing (onchain event), we're going through pkt_in which discards it. However, if we're reconnecting, we simply process the init packet and get upset because they've forgotten us. Hard to reproduce, but here's the log (in this case, test-routing --reconnect and we have just done mutual close): We reconnect in STATE_MUTUAL_CLOSING, send INIT pkt: +19.397025114 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: Init with ack 1 opens + 9 sigs + 8 revokes + 1 shutdown + 1 closing While waiting for response, we see the mutual close... +19.398732602 lightningd(4637):DEBUG: reaped 6370: bitcoin-cli -regtest=1 -datadir=/tmp/bitcoin-lightning2 getblock 2a63b209e17aedc5b1bcc6c2f9e044f97c9c3ca136fc64a719f704d2f632df5f false +19.401834422 lightningd(4637):DEBUG: Adding block 5fdf32f6d204f719a764fc36a13c9c7cf944e0f9c2c6bcb1c5ed7ae109b2632a +19.405167334 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: Got UTXO spend for 8bb48a:0: 7f5e422f... +19.412543610 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: anchor_spent: STATE_MUTUAL_CLOSING => STATE_CLOSE_ONCHAIN_MUTUAL And we also see it buried "forever" (10 blocks in test mode), so we forget peer: +19.423045014 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: Anchor at depth 13 +19.426775063 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: check_for_resolution: STATE_CLOSE_ONCHAIN_MUTUAL => STATE_CLOSED +19.427613109 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: db_forget_peer(023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898) +19.428130685 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: db_start_transaction(023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898) +19.501027511 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: db_commit_transaction(023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898) Now, we get their reply, but they've forgotten us: +19.520208608 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: Decrypted header len 5 +19.520872035 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: Received packet LEN=5, type=PKT__PKT_INIT +19.520999082 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: Our order counter is 19, their ack 0 +19.521078913 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: They acked 0, remote=16 local=15 +19.521447174 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: Queued pkt PKT__PKT_OPEN (order=19) +19.522563794 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: Queued pkt PKT__PKT_OPEN_COMMIT_SIG (order=19) +19.523517319 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:BROKEN: Can't rexmit 2 when local commit 15 and remote 16 +19.524613177 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:UNUSUAL: Sending PKT_ERROR: invalid ack +19.526638447 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: Queued pkt PKT__PKT_ERROR (order=19) +19.527508022 023ec94fb93c669154ba7b08907276e8c8661b2e65d80fc2c089215d5395574898:DEBUG: peer_comms_err: STATE_CLOSED => STATE_ERR_BREAKDOWN We should never transition from STATE_CLOSED to STATE_ERR_BREAKDOWn, and that's what this check prevents. Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
2017-03-02 12:21:49 +00:00
/* We might have had an onchain event while handshaking! */
if (!state_can_io(peer->state))
goto fail;
if (peer->inpkt->init->has_features) {
size_t i;
/* FIXME-OLD #2:
*
* The receiving node SHOULD ignore any odd feature bits it
* does not support, and MUST fail the connection if any
* unsupported even `features` bit is set. */
for (i = 0; i < peer->inpkt->init->features.len*CHAR_BIT; i++) {
size_t byte = i / CHAR_BIT, bit = i % CHAR_BIT;
if (peer->inpkt->init->features.data[byte] & (1<<bit)) {
/* Can't handle even features. */
if (i % 2 != 0) {
log_debug(peer->log,
"They offered feature %zu", i);
continue;
}
queue_pkt_err(peer,
pkt_err(peer,
"Unsupported feature %zu",
i));
goto fail;
}
}
}
/* Send any packets they missed. */
retransmit_pkts(peer, peer->inpkt->init->ack);
/* We let the conversation go this far in case they missed the
* close packets. But now we can close if we're done. */
if (!state_can_io(peer->state)) {
log_debug(peer->log, "State %s, closing immediately",
state_name(peer->state));
goto fail;
}
/* Back into normal mode. */
peer->inpkt = tal_free(peer->inpkt);
peer_has_connected(peer);
if (state_is_normal(peer->state)){
announce_channel(peer->dstate, peer);
sync_routing_table(peer->dstate, peer);
}
return io_duplex(conn,
peer_read_packet(conn, peer, pkt_in),
pkt_out(conn, peer));
fail:
/* We always free inpkt; they may yet reconnect. */
peer->inpkt = tal_free(peer->inpkt);
return pkt_out(conn, peer);
}
static struct io_plan *read_init_pkt(struct io_conn *conn,
struct peer *peer)
{
return peer_read_packet(conn, peer, init_pkt_in);
}
static u64 peer_commitsigs_received(struct peer *peer)
{
return peer->their_commitsigs;
}
static u64 peer_revocations_received(struct peer *peer)
{
/* How many preimages we've received. */
return -peer->their_preimages.min_index;
}
static struct io_plan *peer_send_init(struct io_conn *conn, struct peer *peer)
{
u64 open, sigs, revokes, shutdown, closing;
open = (peer->state == STATE_OPEN_WAIT_FOR_OPENPKT ? 0 : 1);
sigs = peer_commitsigs_received(peer);
revokes = peer_revocations_received(peer);
shutdown = peer->closing.their_script ? 1 : 0;
closing = peer->closing.sigs_in;
log_debug(peer->log,
"Init with ack %"PRIu64" opens + %"PRIu64" sigs + %"
PRIu64" revokes + %"PRIu64" shutdown + %"PRIu64" closing",
open, sigs, revokes, shutdown, closing);
/* FIXME-OLD #2:
*
* A node MUST send an `init` message immediately immediately after
* it has validated the `authenticate` message. A node MUST set
* the `ack` field in the `init` message to the the sum of
* previously-processed messages of types `open`, `open_commit_sig`,
* `update_commit`, `update_revocation`, `close_shutdown` and
* `close_signature`. */
return peer_write_packet(conn, peer,
pkt_init(peer, open + sigs + revokes
+ shutdown + closing),
read_init_pkt);
}
/* Crypto is on, we are live. */
static struct io_plan *peer_crypto_on(struct io_conn *conn, struct peer *peer)
{
peer_secrets_init(peer);
peer_get_revocation_hash(peer, 0, &peer->local.next_revocation_hash);
assert(peer->state == STATE_INIT);
/* Counter is 1 for sending pkt_open: we'll do it in retransmit_pkts */
peer->order_counter++;
db_start_transaction(peer);
set_peer_state(peer, STATE_OPEN_WAIT_FOR_OPENPKT, __func__, true);
/* FIXME: Start timeout, and close peer if they don't progress! */
db_create_peer(peer);
if (db_commit_transaction(peer) != NULL) {
peer_database_err(peer);
return peer_close(conn, peer);
}
/* Set up out commit info now: rest gets done in setup_first_commit
* once anchor is established. */
peer->local.commit = new_commit_info(peer, 0);
peer->local.commit->revocation_hash = peer->local.next_revocation_hash;
peer_get_revocation_hash(peer, 1, &peer->local.next_revocation_hash);
return peer_send_init(conn,peer);
}
static void destroy_peer(struct peer *peer)
{
if (peer->conn)
io_close(peer->conn);
list_del_from(&peer->dstate->peers, &peer->list);
}
static void try_reconnect(struct peer *peer);
static void peer_disconnect(struct io_conn *conn, struct peer *peer)
{
log_info(peer->log, "Disconnected");
/* No longer connected. */
peer->conn = NULL;
peer->connected = false;
/* Not even set up yet? Simply free.*/
if (peer->state == STATE_INIT) {
/* This means we didn't get past crypto handshake or hit db */
tal_free(peer);
return;
}
/* Completely dead? Free it now. */
if (peer->state == STATE_CLOSED) {
io_break(peer);
return;
}
/* This is an unexpected close. */
if (state_can_io(peer->state)) {
forget_uncommitted_changes(peer);
try_reconnect(peer);
}
}
static void try_commit(struct peer *peer)
{
peer->commit_timer = NULL;
if (!peer->connected) {
log_debug(peer->log, "try_commit: state=%s, not connected",
state_name(peer->state));
return;
}
if (state_can_commit(peer->state))
do_commit(peer, NULL);
else {
/* FIXME: try again when we receive revocation /
* reconnect, rather than using timer! */
log_debug(peer->log, "try_commit: state=%s, re-queueing timer",
state_name(peer->state));
remote_changes_pending(peer);
}
}
struct commit_info *new_commit_info(const tal_t *ctx, u64 commit_num)
{
struct commit_info *ci = tal(ctx, struct commit_info);
ci->commit_num = commit_num;
ci->tx = NULL;
ci->cstate = NULL;
ci->sig = NULL;
ci->order = (s64)-1LL;
return ci;
}
static bool peer_reconnected(struct peer *peer,
struct io_conn *conn,
int addr_type, int addr_protocol,
struct io_data *iod,
const struct pubkey *id,
bool we_connected)
{
char *name;
struct netaddr addr;
assert(structeq(peer->id, id));
peer->io_data = tal_steal(peer, iod);
/* FIXME: Attach IO logging for this peer. */
if (!netaddr_from_fd(io_conn_fd(conn), addr_type, addr_protocol, &addr))
return false;
/* If we free peer, conn should be closed, but can't be freed
* immediately so don't make peer a parent. */
peer->conn = conn;
io_set_finish(conn, peer_disconnect, peer);
name = netaddr_name(peer, &addr);
log_info(peer->log, "Reconnected %s %s",
we_connected ? "out to" : "in from", name);
tal_free(name);
return true;
}
struct peer *new_peer(struct lightningd_state *dstate,
struct log *log,
enum state state,
bool offer_anchor)
{
struct peer *peer = tal(dstate, struct peer);
peer->state = state;
peer->connected = false;
peer->id = NULL;
peer->dstate = dstate;
peer->io_data = NULL;
peer->secrets = NULL;
list_head_init(&peer->watches);
peer->inpkt = NULL;
peer->outpkt = tal_arr(peer, Pkt *, 0);
peer->open_jsoncmd = NULL;
peer->commit_jsoncmd = NULL;
list_head_init(&peer->their_commits);
peer->anchor.ok_depth = -1;
peer->order_counter = 0;
peer->their_commitsigs = 0;
peer->cur_commit.watch = NULL;
peer->closing.their_sig = NULL;
peer->closing.our_script = NULL;
peer->closing.their_script = NULL;
peer->closing.shutdown_order = (s64)-1LL;
peer->closing.closing_order = (s64)-1LL;
peer->closing.sigs_in = 0;
peer->onchain.tx = NULL;
peer->onchain.resolved = NULL;
peer->onchain.htlcs = NULL;
peer->onchain.wscripts = NULL;
peer->commit_timer = NULL;
peer->nc = NULL;
peer->their_prev_revocation_hash = NULL;
peer->conn = NULL;
peer->fake_close = false;
peer->output_enabled = true;
peer->local.offer_anchor = offer_anchor;
peer->broadcast_index = 0;
if (!blocks_to_rel_locktime(dstate->config.locktime_blocks,
&peer->local.locktime))
fatal("Could not convert locktime_blocks");
peer->local.mindepth = dstate->config.anchor_confirms;
peer->local.commit = peer->remote.commit = NULL;
peer->local.staging_cstate = peer->remote.staging_cstate = NULL;
peer->log = tal_steal(peer, log);
log_debug(peer->log, "New peer %p", peer);
htlc_map_init(&peer->htlcs);
memset(peer->feechanges, 0, sizeof(peer->feechanges));
shachain_init(&peer->their_preimages);
list_add(&dstate->peers, &peer->list);
tal_add_destructor(peer, destroy_peer);
return peer;
}
static struct peer_address *find_address(struct lightningd_state *dstate,
const struct pubkey *id)
{
struct peer_address *i;
list_for_each(&dstate->addresses, i, list) {
if (structeq(&id->pubkey, &i->id.pubkey))
return i;
}
return NULL;
}
static bool add_peer_address(struct lightningd_state *dstate,
const struct pubkey *id,
const struct netaddr *addr)
{
struct peer_address *a = find_address(dstate, id);
if (a) {
a->addr = *addr;
} else {
a = tal(dstate, struct peer_address);
a->addr = *addr;
a->id = *id;
list_add_tail(&dstate->addresses, &a->list);
}
return db_add_peer_address(dstate, a);
}
static bool peer_first_connected(struct peer *peer,
struct io_conn *conn,
int addr_type, int addr_protocol,
struct io_data *iod,
const struct pubkey *id,
bool we_connected)
{
char *name, *idstr;
struct netaddr addr;
peer->io_data = tal_steal(peer, iod);
peer->id = tal_dup(peer, struct pubkey, id);
peer->local.commit_fee_rate = desired_commit_feerate(peer->dstate);
peer->htlc_id_counter = 0;
/* If we free peer, conn should be closed, but can't be freed
* immediately so don't make peer a parent. */
peer->conn = conn;
io_set_finish(conn, peer_disconnect, peer);
peer->anchor.min_depth = get_block_height(peer->dstate->topology);
/* FIXME: Attach IO logging for this peer. */
if (!netaddr_from_fd(io_conn_fd(conn), addr_type, addr_protocol, &addr))
return false;
/* Save/update address if we connected to them. */
if (we_connected && !add_peer_address(peer->dstate, peer->id, &addr))
return false;
name = netaddr_name(peer, &addr);
idstr = pubkey_to_hexstr(name, peer->id);
log_info(peer->log, "Connected %s %s id %s, changing prefix",
we_connected ? "out to" : "in from", name, idstr);
set_log_prefix(peer->log, tal_fmt(name, "%s:", idstr));
tal_free(name);
log_debug(peer->log, "Using fee rate %"PRIu64,
peer->local.commit_fee_rate);
return true;
}
static void htlc_destroy(struct htlc *htlc)
{
if (!htlc_map_del(&htlc->peer->htlcs, htlc))
fatal("Could not find htlc to destroy");
}
struct htlc *peer_new_htlc(struct peer *peer,
u64 id,
u64 msatoshi,
const struct sha256 *rhash,
u32 expiry,
const u8 *route,
size_t routelen,
struct htlc *src,
enum htlc_state state)
{
struct htlc *h = tal(peer, struct htlc);
h->peer = peer;
h->state = state;
h->id = id;
h->msatoshi = msatoshi;
h->rhash = *rhash;
h->r = NULL;
h->fail = NULL;
if (!blocks_to_abs_locktime(expiry, &h->expiry))
fatal("Invalid HTLC expiry %u", expiry);
h->routing = tal_dup_arr(h, u8, route, routelen, 0);
h->src = src;
if (htlc_owner(h) == LOCAL) {
if (src) {
h->deadline = abs_locktime_to_blocks(&src->expiry)
- peer->dstate->config.deadline_blocks;
} else
/* If we're paying, give it a little longer. */
h->deadline = expiry
+ peer->dstate->config.min_htlc_expiry;
} else {
assert(htlc_owner(h) == REMOTE);
}
htlc_map_add(&peer->htlcs, h);
tal_add_destructor(h, htlc_destroy);
return h;
}
static struct io_plan *crypto_on_reconnect(struct io_conn *conn,
struct lightningd_state *dstate,
struct io_data *iod,
const struct pubkey *id,
struct peer *peer,
bool we_connected)
{
/* Setup peer->conn and peer->io_data */
if (!peer_reconnected(peer, conn, SOCK_STREAM, IPPROTO_TCP,
iod, id, we_connected))
return io_close(conn);
/* We need to eliminate queue now. */
clear_output_queue(peer);
return peer_send_init(conn, peer);
}
static struct io_plan *crypto_on_reconnect_in(struct io_conn *conn,
struct lightningd_state *dstate,
struct io_data *iod,
struct log *log,
const struct pubkey *id,
struct peer *peer)
{
assert(log == peer->log);
return crypto_on_reconnect(conn, dstate, iod, id, peer, false);
}
static struct io_plan *crypto_on_reconnect_out(struct io_conn *conn,
struct lightningd_state *dstate,
struct io_data *iod,
struct log *log,
const struct pubkey *id,
struct peer *peer)
{
assert(log == peer->log);
return crypto_on_reconnect(conn, dstate, iod, id, peer, true);
}
static struct io_plan *crypto_on_out(struct io_conn *conn,
struct lightningd_state *dstate,
struct io_data *iod,
struct log *log,
const struct pubkey *id,
struct json_connecting *connect)
{
struct peer *peer;
if (find_peer(dstate, id)) {
command_fail(connect->cmd, "Already connected to peer %s",
pubkey_to_hexstr(connect->cmd, id));
return io_close(conn);
}
/* Initiator currently funds channel */
peer = new_peer(dstate, log, STATE_INIT, true);
if (!peer_first_connected(peer, conn, SOCK_STREAM, IPPROTO_TCP,
iod, id, true)) {
command_fail(connect->cmd, "Failed to make peer for %s:%s",
connect->name, connect->port);
return io_close(conn);
}
peer->anchor.input = tal_steal(peer, connect->input);
peer->open_jsoncmd = connect->cmd;
return peer_crypto_on(conn, peer);
}
static struct io_plan *peer_connected_out(struct io_conn *conn,
struct lightningd_state *dstate,
const struct netaddr *netaddr,
struct json_connecting *connect)
{
struct log *l;
l = new_log(conn, dstate->log_book, "OUT-%s:%s:",
connect->name, connect->port);
log_debug_struct(l, "Connected out to %s", struct netaddr, netaddr);
return peer_crypto_setup(conn, dstate, NULL, l, crypto_on_out, connect);
}
static struct io_plan *crypto_on_in(struct io_conn *conn,
struct lightningd_state *dstate,
struct io_data *iod,
struct log *log,
const struct pubkey *id,
void *unused)
{
struct peer *peer;
/* FIXME-OLD #2:
*
* A node MUST handle continuing a previous channel on a new encrypted
* transport. */
peer = find_peer(dstate, id);
if (peer) {
/* Close any existing connection, without side effects. */
if (peer->conn) {
log_debug(log, "This is reconnect for peer %p", peer);
log_debug(peer->log, "Reconnect: closing old conn %p for new conn %p",
peer->conn, conn);
io_set_finish(peer->conn, NULL, NULL);
io_close(peer->conn);
peer->conn = NULL;
peer->connected = false;
}
return crypto_on_reconnect_in(conn, dstate, iod, peer->log, id,
peer);
}
/* Initiator currently funds channel */
peer = new_peer(dstate, log, STATE_INIT, false);
if (!peer_first_connected(peer, conn, SOCK_STREAM, IPPROTO_TCP,
iod, id, false))
return io_close(conn);
return peer_crypto_on(conn, peer);
}
static struct io_plan *peer_connected_in(struct io_conn *conn,
struct lightningd_state *dstate)
{
struct netaddr addr;
struct log *l;
const char *name;
if (!netaddr_from_fd(io_conn_fd(conn), SOCK_STREAM, IPPROTO_TCP, &addr))
return false;
name = netaddr_name(conn, &addr);
l = new_log(conn, dstate->log_book, "IN-%s:", name);
log_debug(l, "Connected in");
return peer_crypto_setup(conn, dstate, NULL, l, crypto_on_in, NULL);
}
static int make_listen_fd(struct lightningd_state *dstate,
int domain, void *addr, socklen_t len)
{
int fd = socket(domain, SOCK_STREAM, 0);
if (fd < 0) {
log_debug(dstate->base_log, "Failed to create %u socket: %s",
domain, strerror(errno));
return -1;
}
if (addr) {
int on = 1;
/* Re-use, please.. */
if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on)))
log_unusual(dstate->base_log,
"Failed setting socket reuse: %s",
strerror(errno));
if (bind(fd, addr, len) != 0) {
log_unusual(dstate->base_log,
"Failed to bind on %u socket: %s",
domain, strerror(errno));
goto fail;
}
}
if (listen(fd, 5) != 0) {
log_unusual(dstate->base_log,
"Failed to listen on %u socket: %s",
domain, strerror(errno));
goto fail;
}
return fd;
fail:
close_noerr(fd);
return -1;
}
void setup_listeners(struct lightningd_state *dstate)
{
struct sockaddr_in addr;
struct sockaddr_in6 addr6;
socklen_t len;
int fd1, fd2;
if (!dstate->portnum)
return;
memset(&addr, 0, sizeof(addr));
addr.sin_family = AF_INET;
addr.sin_addr.s_addr = INADDR_ANY;
addr.sin_port = htons(dstate->portnum);
memset(&addr6, 0, sizeof(addr6));
addr6.sin6_family = AF_INET6;
addr6.sin6_addr = in6addr_any;
addr6.sin6_port = htons(dstate->portnum);
/* IPv6, since on Linux that (usually) binds to IPv4 too. */
fd1 = make_listen_fd(dstate, AF_INET6, &addr6, sizeof(addr6));
if (fd1 >= 0) {
struct sockaddr_in6 in6;
len = sizeof(in6);
if (getsockname(fd1, (void *)&in6, &len) != 0) {
log_unusual(dstate->base_log,
"Failed get IPv6 sockname: %s",
strerror(errno));
close_noerr(fd1);
} else {
addr.sin_port = in6.sin6_port;
assert(dstate->portnum == ntohs(addr.sin_port));
log_debug(dstate->base_log,
"Creating IPv6 listener on port %u",
dstate->portnum);
io_new_listener(dstate, fd1, peer_connected_in, dstate);
}
}
/* Just in case, aim for the same port... */
fd2 = make_listen_fd(dstate, AF_INET, &addr, sizeof(addr));
if (fd2 >= 0) {
len = sizeof(addr);
if (getsockname(fd2, (void *)&addr, &len) != 0) {
log_unusual(dstate->base_log,
"Failed get IPv4 sockname: %s",
strerror(errno));
close_noerr(fd2);
} else {
assert(dstate->portnum == ntohs(addr.sin_port));
log_debug(dstate->base_log,
"Creating IPv4 listener on port %u",
dstate->portnum);
io_new_listener(dstate, fd2, peer_connected_in, dstate);
}
}
if (fd1 < 0 && fd2 < 0)
fatal("Could not bind to a network address");
}
static void peer_failed(struct lightningd_state *dstate,
struct json_connecting *connect)
{
/* FIXME: Better diagnostics! */
command_fail(connect->cmd, "Failed to connect to peer %s:%s",
connect->name, connect->port);
}
static void json_connect(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct json_connecting *connect;
jsmntok_t *host, *port, *txtok;
struct bitcoin_tx *tx;
int output;
size_t txhexlen;
u64 fee;
const tal_t *tmpctx = tal_tmpctx(cmd);
if (!json_get_params(buffer, params,
"host", &host,
"port", &port,
"tx", &txtok,
NULL)) {
command_fail(cmd, "Need host, port and tx to a wallet address");
return;
}
connect = tal(cmd, struct json_connecting);
connect->cmd = cmd;
connect->name = tal_strndup(connect, buffer + host->start,
host->end - host->start);
connect->port = tal_strndup(connect, buffer + port->start,
port->end - port->start);
connect->input = tal(connect, struct anchor_input);
txhexlen = txtok->end - txtok->start;
tx = bitcoin_tx_from_hex(tmpctx, buffer + txtok->start, txhexlen);
if (!tx) {
command_fail(cmd, "'%.*s' is not a valid transaction",
txtok->end - txtok->start,
buffer + txtok->start);
return;
}
bitcoin_txid(tx, &connect->input->txid);
/* Find an output we know how to spend. */
for (output = 0; output < tal_count(tx->output); output++) {
if (wallet_can_spend(cmd->dstate, &tx->output[output],
&connect->input->walletkey))
break;
}
if (output == tal_count(tx->output)) {
command_fail(cmd, "Tx doesn't send to wallet address");
return;
}
connect->input->index = output;
connect->input->in_amount = tx->output[output].amount;
/* FIXME: This is normal case, not exact. */
fee = fee_by_feerate(94 + 1+73 + 1+33 + 1, get_feerate(cmd->dstate->topology));
if (fee >= connect->input->in_amount) {
command_fail(cmd, "Amount %"PRIu64" below fee %"PRIu64,
connect->input->in_amount, fee);
return;
}
connect->input->out_amount = connect->input->in_amount - fee;
if (anchor_too_large(connect->input->out_amount)) {
command_fail(cmd, "Amount %"PRIu64" is too large",
connect->input->out_amount);
return;
}
if (!dns_resolve_and_connect(cmd->dstate, connect->name, connect->port,
peer_connected_out, peer_failed, connect)) {
command_fail(cmd, "DNS failed");
return;
}
tal_free(tmpctx);
}
static const struct json_command connect_command = {
"connect",
json_connect,
"Connect to a {host} at {port} using hex-encoded {tx} to fund",
"Returns the {id} on success (once channel established)"
};
AUTODATA(json_command, &connect_command);
/* Have any of our HTLCs passed their deadline? */
static bool any_deadline_past(struct peer *peer)
{
u32 height = get_block_height(peer->dstate->topology);
struct htlc_map_iter it;
struct htlc *h;
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
if (htlc_is_dead(h))
continue;
if (htlc_owner(h) != LOCAL)
continue;
if (height >= h->deadline) {
log_unusual_struct(peer->log,
"HTLC %s deadline has passed",
struct htlc, h);
return true;
}
}
return false;
}
static void check_htlc_expiry(struct peer *peer)
{
u32 height = get_block_height(peer->dstate->topology);
struct htlc_map_iter it;
struct htlc *h;
/* Check their currently still-existing htlcs for expiry */
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
assert(!abs_locktime_is_seconds(&h->expiry));
/* Only their consider HTLCs which are completely locked in. */
if (h->state != RCVD_ADD_ACK_REVOCATION)
continue;
/* We give it an extra block, to avoid the worst of the
* inter-node timing issues. */
if (height <= abs_locktime_to_blocks(&h->expiry))
continue;
db_start_transaction(peer);
/* This can fail only if we're in an error state. */
command_htlc_set_fail(peer, h,
REQUEST_TIMEOUT_408, "timed out");
if (db_commit_transaction(peer) != NULL) {
peer_fail(peer, __func__);
return;
}
}
/* FIXME-OLD #2:
*
* A node MUST NOT offer a HTLC after this deadline, and MUST
* fail the connection if an HTLC which it offered is in
* either node's current commitment transaction past this
* deadline.
*/
/* To save logic elsewhere (ie. to avoid signing a new commit with a
* past-deadline HTLC) we also check staged HTLCs.
*/
if (!state_is_normal(peer->state))
return;
if (any_deadline_past(peer))
peer_fail(peer, __func__);
}
static void peer_depth_ok(struct peer *peer)
{
queue_pkt_open_complete(peer);
db_start_transaction(peer);
switch (peer->state) {
case STATE_OPEN_WAIT_ANCHORDEPTH_AND_THEIRCOMPLETE:
set_peer_state(peer, STATE_OPEN_WAIT_THEIRCOMPLETE,
__func__, true);
break;
case STATE_OPEN_WAIT_ANCHORDEPTH:
peer_open_complete(peer, NULL);
set_peer_state(peer, STATE_NORMAL, __func__, true);
announce_channel(peer->dstate, peer);
sync_routing_table(peer->dstate, peer);
break;
default:
log_broken(peer->log, "%s: state %s",
__func__, state_name(peer->state));
peer_fail(peer, __func__);
break;
}
if (db_commit_transaction(peer))
peer_database_err(peer);
}
static enum watch_result anchor_depthchange(struct peer *peer,
const struct bitcoin_tx *tx,
unsigned int depth,
void *unused)
{
log_debug(peer->log, "Anchor at depth %u", depth);
/* Still waiting for it to reach depth? */
if (state_is_waiting_for_anchor(peer->state)) {
log_debug(peer->log, "Waiting for depth %i",
peer->anchor.ok_depth);
/* We can see a run of blocks all at once, so may be > depth */
if ((int)depth >= peer->anchor.ok_depth) {
peer_depth_ok(peer);
peer->anchor.ok_depth = -1;
}
} else if (depth == 0)
/* FIXME: Report losses! */
fatal("Funding transaction was unspent!");
/* Since this gets called on every new block, check HTLCs here. */
check_htlc_expiry(peer);
/* If fee rate has changed, fire off update to change it. */
if (want_feechange(peer) && state_can_commit(peer->state)) {
log_debug(peer->log, "fee rate changed to %"PRIu64,
desired_commit_feerate(peer->dstate));
/* FIXME: If fee changes back before update, we screw
* up and send an empty commit. We need to generate a
* real packet here! */
remote_changes_pending(peer);
}
/* FIXME-OLD #2:
*
* A node MUST update bitcoin fees if it estimates that the
* current commitment transaction will not be processed in a
* timely manner (see "Risks With HTLC Timeouts").
*/
/* Note: we don't do this when we're told to ignore fees. */
/* FIXME: BOLT should say what to do if it can't! We drop conn. */
if (!state_is_onchain(peer->state) && !state_is_error(peer->state)
&& peer->dstate->config.commitment_fee_min_percent != 0
&& peer->local.commit->cstate->fee_rate < get_feerate(peer->dstate->topology)) {
log_broken(peer->log, "fee rate %"PRIu64" lower than %"PRIu64,
peer->local.commit->cstate->fee_rate,
get_feerate(peer->dstate->topology));
peer_fail(peer, __func__);
}
return KEEP_WATCHING;
}
void notify_new_block(struct chain_topology *topo, unsigned int height)
{
struct lightningd_state *dstate = tal_parent(topo);
/* This is where we check for anchor timeouts. */
struct peer *peer;
list_for_each(&dstate->peers, peer, list) {
if (!state_is_waiting_for_anchor(peer->state))
continue;
/* If we haven't seen anchor yet, we can timeout. */
if (height >= peer->anchor.min_depth
+ dstate->config.anchor_onchain_wait
+ dstate->config.anchor_confirms) {
queue_pkt_err(peer, pkt_err(peer, "Funding timeout"));
set_peer_state(peer, STATE_ERR_ANCHOR_TIMEOUT, __func__,
false);
peer_breakdown(peer);
}
}
}
static bool outputscript_eq(const struct bitcoin_tx_output *out,
size_t i, const u8 *script)
{
if (tal_count(out[i].script) != tal_count(script))
return false;
return memcmp(out[i].script, script, tal_count(script)) == 0;
}
/* This tx is their commitment;
* fill in onchain.htlcs[], wscripts[], to_us_idx and to_them_idx */
static bool map_onchain_outputs(struct peer *peer,
const struct sha256 *rhash,
const struct bitcoin_tx *tx,
enum side side,
unsigned int commit_num)
{
u8 *to_us, *to_them, *to_them_wscript, *to_us_wscript;
struct htlc_output_map *hmap;
size_t i;
peer->onchain.to_us_idx = peer->onchain.to_them_idx = -1;
peer->onchain.htlcs = tal_arr(tx, struct htlc *, tal_count(tx->output));
peer->onchain.wscripts = tal_arr(tx, const u8 *, tal_count(tx->output));
to_us = commit_output_to_us(tx, peer, rhash, side, &to_us_wscript);
to_them = commit_output_to_them(tx, peer, rhash, side,
&to_them_wscript);
/* Now generate the wscript hashes for every possible HTLC. */
hmap = get_htlc_output_map(tx, peer, rhash, side, commit_num);
for (i = 0; i < tal_count(tx->output); i++) {
log_debug(peer->log, "%s: output %zi", __func__, i);
if (peer->onchain.to_us_idx == -1
&& outputscript_eq(tx->output, i, to_us)) {
log_add(peer->log, " -> to us");
peer->onchain.htlcs[i] = NULL;
peer->onchain.wscripts[i] = to_us_wscript;
peer->onchain.to_us_idx = i;
continue;
}
if (peer->onchain.to_them_idx == -1
&& outputscript_eq(tx->output, i, to_them)) {
log_add(peer->log, " -> to them");
peer->onchain.htlcs[i] = NULL;
peer->onchain.wscripts[i] = to_them_wscript;
peer->onchain.to_them_idx = i;
continue;
}
/* Must be an HTLC output */
peer->onchain.htlcs[i] = txout_get_htlc(hmap,
tx->output[i].script,
peer->onchain.wscripts+i);
if (!peer->onchain.htlcs[i]) {
log_add(peer->log, "no HTLC found");
goto fail;
}
tal_steal(peer->onchain.htlcs, peer->onchain.htlcs[i]);
tal_steal(peer->onchain.wscripts, peer->onchain.wscripts[i]);
log_add(peer->log, "HTLC %"PRIu64, peer->onchain.htlcs[i]->id);
}
tal_free(hmap);
return true;
fail:
tal_free(hmap);
return false;
}
static bool is_mutual_close(const struct peer *peer,
const struct bitcoin_tx *tx)
{
const u8 *ours, *theirs;
ours = peer->closing.our_script;
theirs = peer->closing.their_script;
/* If we don't know the closing scripts, can't have signed them. */
if (!ours || !theirs)
return false;
if (tal_count(tx->output) != 2)
return false;
/* Without knowing fee amounts, can't determine order. Check both. */
if (scripteq(tx->output[0].script, ours)
&& scripteq(tx->output[1].script, theirs))
return true;
if (scripteq(tx->output[0].script, theirs)
&& scripteq(tx->output[1].script, ours))
return true;
return false;
}
/* Create a HTLC refund collection for onchain.tx output out_num. */
static const struct bitcoin_tx *htlc_timeout_tx(const struct peer *peer,
unsigned int out_num)
{
const struct htlc *htlc = peer->onchain.htlcs[out_num];
const u8 *wscript = peer->onchain.wscripts[out_num];
struct bitcoin_tx *tx = bitcoin_tx(peer, 1, 1);
secp256k1_ecdsa_signature sig;
u64 fee, satoshis;
/* We must set locktime so HTLC expiry can OP_CHECKLOCKTIMEVERIFY */
tx->lock_time = htlc->expiry.locktime;
tx->input[0].index = out_num;
tx->input[0].txid = peer->onchain.txid;
satoshis = htlc->msatoshi / 1000;
tx->input[0].amount = tal_dup(tx->input, u64, &satoshis);
tx->input[0].sequence_number = bitcoin_nsequence(&peer->remote.locktime);
/* Using a new output address here would be useless: they can tell
* it's our HTLC, and that we collected it via timeout. */
tx->output[0].script = scriptpubkey_p2sh(tx,
bitcoin_redeem_single(tx,
&peer->local.finalkey));
log_unusual(peer->log, "Pre-witness txlen = %zu\n",
measure_tx_cost(tx) / 4);
assert(measure_tx_cost(tx) == 83 * 4);
/* Witness length can vary, due to DER encoding of sigs, but we
* use 539 from an example run. */
fee = fee_by_feerate(83 + 539 / 4, get_feerate(peer->dstate->topology));
/* FIXME: Fail gracefully in these cases (not worth collecting) */
if (fee > satoshis || is_dust(satoshis - fee))
fatal("HTLC refund amount of %"PRIu64" won't cover fee %"PRIu64,
satoshis, fee);
tx->output[0].amount = satoshis - fee;
peer_sign_htlc_refund(peer, tx, wscript, &sig);
tx->input[0].witness = bitcoin_witness_htlc(tx,
NULL, &sig, wscript);
log_unusual(peer->log, "tx cost for htlc timeout tx: %zu",
measure_tx_cost(tx));
return tx;
}
static void reset_onchain_closing(struct peer *peer)
{
if (peer->onchain.tx) {
log_unusual_struct(peer->log,
"New anchor spend, forgetting old tx %s",
struct sha256_double, &peer->onchain.txid);
peer->onchain.tx = tal_free(peer->onchain.tx);
peer->onchain.resolved = NULL;
peer->onchain.htlcs = NULL;
peer->onchain.wscripts = NULL;
}
}
static const struct bitcoin_tx *irrevocably_resolved(struct peer *peer)
{
/* We can't all be irrevocably resolved until the commit tx is,
* so just mark that as resolving us. */
return peer->onchain.tx;
}
/* We usually don't fail HTLCs we offered, but if the peer breaks down
* before we've confirmed it, this is exactly what happens. */
static void fail_own_htlc(struct peer *peer, struct htlc *htlc)
{
/* We can't do anything about db failures; peer already closed. */
db_start_transaction(peer);
set_htlc_fail(peer, htlc, "peer closed", strlen("peer closed"));
our_htlc_failed(peer, htlc);
db_commit_transaction(peer);
}
/* We've spent an HTLC output to get our funds back. There's still a
* chance that they could also spend the HTLC output (using the preimage),
* so we need to wait for some confirms.
*
* However, we don't want to wait too long: our upstream will get upset if
* their HTLC has timed out and we don't close it. So we wait one less
* than the HTLC timeout difference.
*/
static enum watch_result our_htlc_timeout_depth(struct peer *peer,
const struct bitcoin_tx *tx,
unsigned int depth,
struct htlc *htlc)
{
if (depth == 0)
return KEEP_WATCHING;
if (depth + 1 < peer->dstate->config.min_htlc_expiry)
return KEEP_WATCHING;
fail_own_htlc(peer, htlc);
return DELETE_WATCH;
}
static enum watch_result our_htlc_depth(struct peer *peer,
unsigned int depth,
const struct sha256_double *txid,
enum side whose_commit,
unsigned int out_num)
{
struct htlc *h = peer->onchain.htlcs[out_num];
u32 height;
/* Must be in a block. */
if (depth == 0)
return KEEP_WATCHING;
height = get_block_height(peer->dstate->topology);
/* FIXME-OLD #onchain:
*
* If the *commitment tx* is the other node's, the output is
* considered *timed out* once the HTLC is expired. If the
* *commitment tx* is this node's, the output is considered *timed
* out* once the HTLC is expired, AND the output's
* `OP_CHECKSEQUENCEVERIFY` delay has passed.
*/
if (height < abs_locktime_to_blocks(&h->expiry))
return KEEP_WATCHING;
if (whose_commit == LOCAL) {
if (depth < rel_locktime_to_blocks(&peer->remote.locktime))
return KEEP_WATCHING;
}
/* FIXME-OLD #onchain:
*
* If the output has *timed out* and not been *resolved*, the node
* MUST *resolve* the output by spending it.
*/
/* FIXME: we should simply delete this watch if HTLC is fulfilled. */
if (!peer->onchain.resolved[out_num]) {
peer->onchain.resolved[out_num] = htlc_timeout_tx(peer, out_num);
watch_tx(peer->onchain.resolved[out_num],
peer->dstate->topology,
peer,
peer->onchain.resolved[out_num],
our_htlc_timeout_depth, h);
broadcast_tx(peer->dstate->topology,
peer, peer->onchain.resolved[out_num], NULL);
}
return DELETE_WATCH;
}
static enum watch_result our_htlc_depth_theircommit(struct peer *peer,
const struct bitcoin_tx *tx,
unsigned int depth,
ptrint_t *out_num)
{
struct sha256_double txid;
bitcoin_txid(tx, &txid);
return our_htlc_depth(peer, depth, &txid, REMOTE, ptr2int(out_num));
}
static enum watch_result our_htlc_depth_ourcommit(struct peer *peer,
const struct bitcoin_tx *tx,
unsigned int depth,
ptrint_t *out_num)
{
struct sha256_double txid;
bitcoin_txid(tx, &txid);
return our_htlc_depth(peer, depth, &txid, LOCAL, ptr2int(out_num));
}
static enum watch_result their_htlc_depth(struct peer *peer,
const struct bitcoin_tx *tx,
unsigned int depth,
ptrint_t *out_num)
{
u32 height;
const struct htlc *htlc = peer->onchain.htlcs[ptr2int(out_num)];
/* Must be in a block. */
if (depth == 0)
return KEEP_WATCHING;
height = get_block_height(peer->dstate->topology);
/* FIXME-OLD #onchain:
*
* Otherwise, if the output HTLC has expired, it is considered
* *irrevocably resolved*.
*/
if (height < abs_locktime_to_blocks(&htlc->expiry))
return KEEP_WATCHING;
peer->onchain.resolved[ptr2int(out_num)] = irrevocably_resolved(peer);
return DELETE_WATCH;
}
static enum watch_result our_main_output_depth(struct peer *peer,
const struct bitcoin_tx *tx,
unsigned int depth,
void *unused)
{
/* Not past CSV timeout? */
if (depth < rel_locktime_to_blocks(&peer->remote.locktime))
return KEEP_WATCHING;
assert(peer->onchain.to_us_idx != -1);
/* FIXME-OLD #onchain:
*
* 1. _A's main output_: A node SHOULD spend this output to a
* convenient address. This avoids having to remember the
* complicated witness script associated with that particular
* channel for later spending. ... If the output is spent (as
* recommended), the output is *resolved* by the spending
* transaction
*/
peer->onchain.resolved[peer->onchain.to_us_idx]
= bitcoin_spend_ours(peer);
broadcast_tx(peer->dstate->topology,
peer, peer->onchain.resolved[peer->onchain.to_us_idx],
NULL);
return DELETE_WATCH;
}
/* Any of our HTLCs we didn't have in our commitment tx, but they did,
* we can't fail until we're sure our commitment tx will win. */
static enum watch_result our_unilateral_depth(struct peer *peer,
const struct bitcoin_tx *tx,
unsigned int depth,
void *unused)
{
struct htlc_map_iter it;
struct htlc *h;
if (depth < peer->dstate->config.min_htlc_expiry)
return KEEP_WATCHING;
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
if (htlc_owner(h) == LOCAL
&& !htlc_has(h, HTLC_LOCAL_F_COMMITTED)
&& htlc_has(h, HTLC_REMOTE_F_COMMITTED)) {
log_debug(peer->log,
"%s:failing uncommitted htlc %"PRIu64,
__func__, h->id);
fail_own_htlc(peer, h);
}
}
return DELETE_WATCH;
}
static enum watch_result our_htlc_spent(struct peer *peer,
const struct bitcoin_tx *tx,
size_t input_num,
struct htlc *h)
{
struct sha256 sha;
struct preimage preimage;
/* FIXME-OLD #onchain:
*
* If a node sees a redemption transaction...the node MUST extract the
* preimage from the transaction input witness. This is either to
* prove payment (if this node originated the payment), or to redeem
* the corresponding incoming HTLC from another peer.
*/
/* This is the form of all HTLC spends. */
if (!tx->input[input_num].witness
|| tal_count(tx->input[input_num].witness) != 3
|| tal_count(tx->input[input_num].witness[1]) != sizeof(preimage))
fatal("Impossible HTLC spend for %"PRIu64, h->id);
/* Our timeout tx has all-zeroes, so we can distinguish it. */
if (memeqzero(tx->input[input_num].witness[1], sizeof(preimage)))
/* They might try to race us. */
return KEEP_WATCHING;
memcpy(&preimage, tx->input[input_num].witness[1], sizeof(preimage));
sha256(&sha, &preimage, sizeof(preimage));
/* FIXME: This could happen with a ripemd collision, since
* script.c only checks that ripemd matches... */
if (!structeq(&sha, &h->rhash))
fatal("HTLC redeemed with incorrect r value?");
log_unusual(peer->log, "Peer redeemed HTLC %"PRIu64" on-chain",
h->id);
log_add_struct(peer->log, " using rvalue %s", struct preimage, &preimage);
set_htlc_rval(peer, h, &preimage);
our_htlc_fulfilled(peer, h);
/* FIXME-OLD #onchain:
*
* If a node sees a redemption transaction, the output is considered
* *irrevocably resolved*... Note that we don't care about the fate of
* the redemption transaction itself once we've extracted the
* preimage; the knowledge is not revocable.
*/
peer->onchain.resolved[tx->input[input_num].index]
= irrevocably_resolved(peer);
return DELETE_WATCH;
}
static void resolve_our_htlc(struct peer *peer,
unsigned int out_num,
enum watch_result (*cb)(struct peer *peer,
const struct bitcoin_tx *tx,
unsigned int depth,
ptrint_t *out_num))
{
/* FIXME-OLD #onchain:
*
* A node MUST watch for spends of *commitment tx* outputs for HTLCs
* it offered; each one must be *resolved* by a timeout transaction
* (the node pays back to itself) or redemption transaction (the other
* node provides the redemption preimage).
*/
watch_txo(peer->onchain.tx,
peer->dstate->topology,
peer, &peer->onchain.txid, out_num,
our_htlc_spent, peer->onchain.htlcs[out_num]);
watch_txid(peer->onchain.tx,
peer->dstate->topology,
peer, &peer->onchain.txid, cb, int2ptr(out_num));
}
static void resolve_their_htlc(struct peer *peer, unsigned int out_num)
{
/* FIXME-OLD #onchain:
*
* If the node ... already knows... a redemption preimage for an
* unresolved *commitment tx* output it was offered, it MUST *resolve*
* the output by spending it using the preimage.
*/
if (peer->onchain.htlcs[out_num]->r) {
peer->onchain.resolved[out_num] = htlc_fulfill_tx(peer, out_num);
broadcast_tx(peer->dstate->topology,
peer, peer->onchain.resolved[out_num], NULL);
} else {
/* FIXME-OLD #onchain:
*
* Otherwise, if the output HTLC has expired, it is considered
* *irrevocably resolved*.
*/
watch_tx(peer->onchain.tx,
peer->dstate->topology,
peer, peer->onchain.tx,
their_htlc_depth, int2ptr(out_num));
}
}
/* FIXME-OLD #onchain:
*
* When node A sees its own *commitment tx*:
*/
static void resolve_our_unilateral(struct peer *peer)
{
unsigned int i;
struct chain_topology *topo = peer->dstate->topology;
const struct bitcoin_tx *tx = peer->onchain.tx;
/* This only works because we always watch for a long time before
* freeing peer, by which time this has resolved. We could create
* resolved[] entries for these uncommitted HTLCs, too. */
watch_tx(tx, topo, peer, tx, our_unilateral_depth, NULL);
for (i = 0; i < tal_count(tx->output); i++) {
/* FIXME-OLD #onchain:
*
* 1. _A's main output_: A node SHOULD spend this output to a
* convenient address. ... A node MUST wait until the
* `OP_CHECKSEQUENCEVERIFY` delay has passed (as specified
* by the other node's `open_channel` `delay` field) before
* spending the output.
*/
if (i == peer->onchain.to_us_idx)
watch_tx(tx, topo,
peer, tx, our_main_output_depth, NULL);
/* FIXME-OLD #onchain:
*
* 2. _B's main output_: No action required, this output is
* considered *resolved* by the *commitment tx*.
*/
else if (i == peer->onchain.to_them_idx)
peer->onchain.resolved[i] = tx;
/* FIXME-OLD #onchain:
*
* 3. _A's offered HTLCs_: See On-chain HTLC Handling: Our
* Offers below.
*/
else if (htlc_owner(peer->onchain.htlcs[i]) == LOCAL)
resolve_our_htlc(peer, i, our_htlc_depth_ourcommit);
/* FIXME-OLD #onchain:
*
* 4. _B's offered HTLCs_: See On-chain HTLC Handling: Their
* Offers below.
*/
else
resolve_their_htlc(peer, i);
}
}
/* FIXME-OLD #onchain:
*
* Similarly, when node A sees a *commitment tx* from B:
*/
static void resolve_their_unilateral(struct peer *peer)
{
unsigned int i;
const struct bitcoin_tx *tx = peer->onchain.tx;
for (i = 0; i < tal_count(tx->output); i++) {
/* FIXME-OLD #onchain:
*
* 1. _A's main output_: No action is required; this is a
* simple P2WPKH output. This output is considered
* *resolved* by the *commitment tx*.
*/
if (i == peer->onchain.to_us_idx)
peer->onchain.resolved[i] = tx;
/* FIXME-OLD #onchain:
*
* 2. _B's main output_: No action required, this output is
* considered *resolved* by the *commitment tx*.
*/
else if (i == peer->onchain.to_them_idx)
peer->onchain.resolved[i] = tx;
/* FIXME-OLD #onchain:
*
* 3. _A's offered HTLCs_: See On-chain HTLC Handling: Our
* Offers below.
*/
else if (htlc_owner(peer->onchain.htlcs[i]) == LOCAL)
resolve_our_htlc(peer, i, our_htlc_depth_theircommit);
/*
* 4. _B's offered HTLCs_: See On-chain HTLC Handling: Their
* Offers below.
*/
else
resolve_their_htlc(peer, i);
}
}
static void resolve_mutual_close(struct peer *peer)
{
unsigned int i;
/* FIXME-OLD #onchain:
*
* A node doesn't need to do anything else as it has already agreed to
* the output, which is sent to its specified scriptpubkey (see FIXME-OLD
* #2 "4.1: Closing initiation: close_shutdown").
*/
for (i = 0; i < tal_count(peer->onchain.tx->output); i++)
peer->onchain.resolved[i] = irrevocably_resolved(peer);
/* No HTLCs. */
peer->onchain.htlcs = tal_arrz(peer->onchain.tx,
struct htlc *,
tal_count(peer->onchain.tx->output));
}
/* Called every time the tx spending the funding tx changes depth. */
static enum watch_result check_for_resolution(struct peer *peer,
const struct bitcoin_tx *tx,
unsigned int depth,
void *unused)
{
size_t i, n = tal_count(peer->onchain.resolved);
size_t forever = peer->dstate->config.forever_confirms;
/* FIXME-OLD #onchain:
*
* A node MUST *resolve* all outputs as specified below, and MUST be
* prepared to resolve them multiple times in case of blockchain
* reorganizations.
*/
for (i = 0; i < n; i++)
if (!peer->onchain.resolved[i])
return KEEP_WATCHING;
/* FIXME-OLD #onchain:
*
* Outputs which are *resolved* by a transaction are considered
* *irrevocably resolved* once they are included in a block at least
* 100 deep on the most-work blockchain.
*/
if (depth < forever)
return KEEP_WATCHING;
for (i = 0; i < n; i++) {
struct sha256_double txid;
bitcoin_txid(peer->onchain.resolved[i], &txid);
if (get_tx_depth(peer->dstate->topology, &txid, NULL) < forever)
return KEEP_WATCHING;
}
/* FIXME-OLD #onchain:
*
* A node MUST monitor the blockchain for transactions which spend any
* output which is not *irrevocably resolved* until all outputs are
* *irrevocably resolved*.
*/
set_peer_state(peer, STATE_CLOSED, "check_for_resolution", false);
db_forget_peer(peer);
/* It's theoretically possible that peer is still writing output */
if (!peer->conn)
io_break(peer);
else
io_wake(peer);
return DELETE_WATCH;
}
static bool find_their_old_tx(struct peer *peer,
const struct sha256_double *txid,
u64 *idx)
{
/* FIXME: Don't keep these in memory, search db here. */
struct their_commit *tc;
log_debug_struct(peer->log, "Finding txid %s", struct sha256_double,
txid);
list_for_each(&peer->their_commits, tc, list) {
if (structeq(&tc->txid, txid)) {
*idx = tc->commit_num;
return true;
}
}
return false;
}
static void resolve_their_steal(struct peer *peer,
const struct sha256 *revocation_preimage)
{
int i, n;
const struct bitcoin_tx *tx = peer->onchain.tx;
struct bitcoin_tx *steal_tx;
size_t wsize = 0;
u64 input_total = 0, fee;
/* Create steal_tx: don't need to steal to_us output */
if (peer->onchain.to_us_idx == -1)
steal_tx = bitcoin_tx(tx, tal_count(tx->output), 1);
else
steal_tx = bitcoin_tx(tx, tal_count(tx->output) - 1, 1);
n = 0;
log_debug(peer->log, "Analyzing tx to steal:");
for (i = 0; i < tal_count(tx->output); i++) {
/* FIXME-OLD #onchain:
* 1. _A's main output_: No action is required; this is a
* simple P2WPKH output. This output is considered
* *resolved* by the *commitment tx*.
*/
if (i == peer->onchain.to_us_idx) {
log_debug(peer->log, "%i is to-us, ignoring", i);
peer->onchain.resolved[i] = tx;
continue;
}
/* FIXME-OLD #onchain:
*
* 2. _B's main output_: The node MUST *resolve* this by
* spending using the revocation preimage.
*
* 3. _A's offered HTLCs_: The node MUST *resolve* this by
* spending using the revocation preimage.
*
* 4. _B's offered HTLCs_: The node MUST *resolve* this by
* spending using the revocation preimage.
*/
peer->onchain.resolved[i] = steal_tx;
/* Connect it up. */
steal_tx->input[n].txid = peer->onchain.txid;
steal_tx->input[n].index = i;
steal_tx->input[n].amount = tal_dup(steal_tx, u64,
&tx->output[i].amount);
/* Track witness size, for fee. */
wsize += tal_count(peer->onchain.wscripts[i]);
input_total += tx->output[i].amount;
n++;
}
assert(n == tal_count(steal_tx->input));
fee = get_feerate(peer->dstate->topology)
* (measure_tx_cost(steal_tx) + wsize) / 1000;
if (fee > input_total || is_dust(input_total - fee)) {
log_unusual(peer->log, "Not worth stealing tiny amount %"PRIu64,
input_total);
/* Consider them all resolved by steal tx. */
for (i = 0; i < tal_count(peer->onchain.resolved); i++)
peer->onchain.resolved[i] = tx;
tal_free(steal_tx);
return;
}
steal_tx->output[0].amount = input_total - fee;
steal_tx->output[0].script = scriptpubkey_p2sh(steal_tx,
bitcoin_redeem_single(steal_tx,
&peer->local.finalkey));
/* Now, we can sign them all (they're all of same form). */
n = 0;
for (i = 0; i < tal_count(tx->output); i++) {
secp256k1_ecdsa_signature sig;
/* Don't bother stealing the output already to us. */
if (i == peer->onchain.to_us_idx)
continue;
peer_sign_steal_input(peer, steal_tx, n,
peer->onchain.wscripts[i],
&sig);
steal_tx->input[n].witness
= bitcoin_witness_secret(steal_tx,
revocation_preimage,
sizeof(*revocation_preimage),
&sig,
peer->onchain.wscripts[i]);
n++;
}
assert(n == tal_count(steal_tx->input));
broadcast_tx(peer->dstate->topology, peer, steal_tx, NULL);
}
static struct sha256 *get_rhash(struct peer *peer, u64 commit_num,
struct sha256 *rhash)
{
struct sha256 preimage;
/* Previous revoked tx? */
if (shachain_get_hash(&peer->their_preimages,
0xFFFFFFFFFFFFFFFFL - commit_num,
&preimage)) {
sha256(rhash, &preimage, sizeof(preimage));
return tal_dup(peer, struct sha256, &preimage);
}
/* Current tx? */
if (commit_num == peer->remote.commit->commit_num) {
*rhash = peer->remote.commit->revocation_hash;
return NULL;
}
/* Last tx, but we haven't got revoke for it yet? */
assert(commit_num == peer->remote.commit->commit_num-1);
*rhash = *peer->their_prev_revocation_hash;
return NULL;
}
/* We assume the tx is valid! Don't do a blockchain.info and feed this
* invalid transactions! */
static enum watch_result anchor_spent(struct peer *peer,
const struct bitcoin_tx *tx,
size_t input_num,
void *unused)
{
Pkt *err;
enum state newstate;
struct htlc_map_iter it;
struct htlc *h;
u64 commit_num;
assert(input_num < tal_count(tx->input));
/* We only ever sign single-input txs. */
if (input_num != 0) {
log_broken(peer->log, "Anchor spend by non-single input tx");
goto unknown_spend;
}
/* We may have been following a different spend. Forget it. */
reset_onchain_closing(peer);
peer->onchain.tx = tal_steal(peer, tx);
bitcoin_txid(tx, &peer->onchain.txid);
/* If we have any HTLCs we're not committed to yet, fail them now. */
for (h = htlc_map_first(&peer->htlcs, &it);
h;
h = htlc_map_next(&peer->htlcs, &it)) {
if (h->state == SENT_ADD_HTLC) {
fail_own_htlc(peer, h);
}
}
/* We need to resolve every output. */
peer->onchain.resolved
= tal_arrz(tx, const struct bitcoin_tx *,
tal_count(tx->output));
/* A mutual close tx. */
if (is_mutual_close(peer, tx)) {
newstate = STATE_CLOSE_ONCHAIN_MUTUAL;
err = NULL;
resolve_mutual_close(peer);
/* Our unilateral */
} else if (structeq(&peer->local.commit->txid,
&peer->onchain.txid)) {
newstate = STATE_CLOSE_ONCHAIN_OUR_UNILATERAL;
/* We're almost certainly closed to them by now. */
err = pkt_err(peer, "Our own unilateral close tx seen");
if (!map_onchain_outputs(peer,
&peer->local.commit->revocation_hash,
tx, LOCAL,
peer->local.commit->commit_num)) {
log_broken(peer->log,
"Can't resolve own anchor spend %"PRIu64"!",
peer->local.commit->commit_num);
goto unknown_spend;
}
resolve_our_unilateral(peer);
/* Must be their unilateral */
} else if (find_their_old_tx(peer, &peer->onchain.txid,
&commit_num)) {
struct sha256 *preimage, rhash;
preimage = get_rhash(peer, commit_num, &rhash);
if (!map_onchain_outputs(peer, &rhash, tx, REMOTE, commit_num)) {
/* Should not happen */
log_broken(peer->log,
"Can't resolve known anchor spend %"PRIu64"!",
commit_num);
goto unknown_spend;
}
if (preimage) {
newstate = STATE_CLOSE_ONCHAIN_CHEATED;
err = pkt_err(peer, "Revoked transaction seen");
resolve_their_steal(peer, preimage);
} else {
newstate = STATE_CLOSE_ONCHAIN_THEIR_UNILATERAL;
err = pkt_err(peer, "Unilateral close tx seen");
resolve_their_unilateral(peer);
}
} else {
/* FIXME: Log harder! */
log_broken(peer->log,
"Unknown anchor spend! Funds may be lost!");
goto unknown_spend;
}
/* FIXME-OLD #onchain:
*
* A node MAY send a descriptive error packet in this case.
*/
if (err && state_can_io(peer->state))
queue_pkt_err(peer, err);
/* Don't need to save to DB: it will be replayed if we crash. */
set_peer_state(peer, newstate, "anchor_spent", false);
/* If we've just closed connection, make output close it. */
io_wake(peer);
/* FIXME-OLD #onchain:
*
* A node SHOULD fail the connection if it is not already
* closed when it sees the funding transaction spent.
*/
assert(!state_can_io(peer->state));
assert(peer->onchain.resolved != NULL);
watch_tx(tx, peer->dstate->topology,
peer, tx, check_for_resolution, NULL);
return KEEP_WATCHING;
unknown_spend:
/* FIXME-OLD #onchain:
*
* A node SHOULD report an error to the operator if it
* sees a transaction spend the funding transaction
* output which does not fall into one of these
* categories (mutual close, unilateral close, or
* cheating attempt). Such a transaction implies its
* private key has leaked, and funds may be lost.
*/
/* FIXME: Save to db. */
set_peer_state(peer, STATE_ERR_INFORMATION_LEAK, "anchor_spent", false);
return DELETE_WATCH;
}
void peer_watch_anchor(struct peer *peer, int depth)
{
struct chain_topology *topo = peer->dstate->topology;
log_debug_struct(peer->log, "watching for anchor %s",
struct sha256_double, &peer->anchor.txid);
log_add(peer->log, " to hit depth %i", depth);
peer->anchor.ok_depth = depth;
watch_txid(peer, topo, peer,
&peer->anchor.txid, anchor_depthchange, NULL);
watch_txo(peer, topo, peer, &peer->anchor.txid, 0, anchor_spent, NULL);
}
struct bitcoin_tx *peer_create_close_tx(const tal_t *ctx,
struct peer *peer, u64 fee)
{
struct channel_state cstate;
/* We don't need a deep copy here, just fee levels. */
cstate = *peer->local.staging_cstate;
if (!force_fee(&cstate, fee)) {
log_unusual(peer->log,
"peer_create_close_tx: can't afford fee %"PRIu64,
fee);
return NULL;
}
log_debug(peer->log,
"creating close-tx with fee %"PRIu64" amounts %u/%u to ",
fee,
cstate.side[LOCAL].pay_msat / 1000,
cstate.side[REMOTE].pay_msat / 1000);
log_add_struct(peer->log, "%s", struct pubkey, &peer->local.finalkey);
log_add_struct(peer->log, "/%s", struct pubkey, &peer->remote.finalkey);
return create_close_tx(ctx,
peer->closing.our_script,
peer->closing.their_script,
&peer->anchor.txid,
peer->anchor.index,
peer->anchor.satoshis,
cstate.side[LOCAL].pay_msat / 1000,
cstate.side[REMOTE].pay_msat / 1000,
/* FIXME: Real dust limit! */
0);
}
/* Sets up the initial cstate and commit tx for both nodes: false if
* insufficient funds. */
bool setup_first_commit(struct peer *peer)
{
bool to_them_only, to_us_only;
assert(!peer->local.commit->tx);
assert(!peer->remote.commit->tx);
/* Revocation hashes already filled in, from pkt_open */
peer->local.commit->cstate = initial_cstate(peer->local.commit,
peer->anchor.satoshis,
peer->local.commit_fee_rate,
peer->local.offer_anchor ?
LOCAL : REMOTE);
if (!peer->local.commit->cstate)
return false;
peer->remote.commit->cstate = initial_cstate(peer->remote.commit,
peer->anchor.satoshis,
peer->remote.commit_fee_rate,
peer->local.offer_anchor ?
LOCAL : REMOTE);
if (!peer->remote.commit->cstate)
return false;
peer->local.commit->tx = create_commit_tx(peer->local.commit,
peer,
&peer->local.commit->revocation_hash,
peer->local.commit->cstate,
LOCAL, &to_them_only);
bitcoin_txid(peer->local.commit->tx, &peer->local.commit->txid);
peer->remote.commit->tx = create_commit_tx(peer->remote.commit,
peer,
&peer->remote.commit->revocation_hash,
peer->remote.commit->cstate,
REMOTE, &to_us_only);
assert(to_them_only != to_us_only);
/* If we offer anchor, their commit is to-us only. */
assert(to_us_only == peer->local.offer_anchor);
bitcoin_txid(peer->remote.commit->tx, &peer->remote.commit->txid);
peer->local.staging_cstate = copy_cstate(peer, peer->local.commit->cstate);
peer->remote.staging_cstate = copy_cstate(peer, peer->remote.commit->cstate);
return true;
}
static struct io_plan *peer_reconnect(struct io_conn *conn, struct peer *peer)
{
/* In case they reconnected to us already. */
if (peer->conn)
return io_close(conn);
log_debug(peer->log, "Reconnected, doing crypto...");
peer->conn = conn;
assert(!peer->connected);
assert(peer->id);
return peer_crypto_setup(conn, peer->dstate,
peer->id, peer->log,
crypto_on_reconnect_out, peer);
}
/* We can't only retry when we want to send: they may want to send us
* something but not be able to connect (NAT). So keep retrying.. */
static void reconnect_failed(struct io_conn *conn, struct peer *peer)
{
/* Already otherwise connected (ie. they connected in)? */
if (peer->conn) {
log_debug(peer->log, "reconnect_failed: already connected");
return;
}
log_debug(peer->log, "Setting timer to re-connect");
new_reltimer(&peer->dstate->timers, peer, peer->dstate->config.poll_time,
try_reconnect, peer);
}
static struct io_plan *init_conn(struct io_conn *conn, struct peer *peer)
{
struct addrinfo a;
struct peer_address *addr = find_address(peer->dstate, peer->id);
netaddr_to_addrinfo(&a, &addr->addr);
return io_connect(conn, &a, peer_reconnect, peer);
}
static void try_reconnect(struct peer *peer)
{
struct io_conn *conn;
struct peer_address *addr;
char *name;
int fd;
/* Already reconnected? */
if (peer->conn) {
log_debug(peer->log, "try_reconnect: already connected");
return;
}
addr = find_address(peer->dstate, peer->id);
if (!addr) {
log_debug(peer->log, "try_reconnect: no known address");
return;
}
fd = socket(addr->addr.saddr.s.sa_family, addr->addr.type,
addr->addr.protocol);
if (fd < 0) {
log_broken(peer->log, "do_reconnect: failed to create socket: %s",
strerror(errno));
peer_fail(peer, __func__);
return;
}
assert(!peer->conn);
conn = io_new_conn(peer->dstate, fd, init_conn, peer);
name = netaddr_name(peer, &addr->addr);
log_debug(peer->log, "Trying to reconnect to %s", name);
tal_free(name);
io_set_finish(conn, reconnect_failed, peer);
}
void reconnect_peers(struct lightningd_state *dstate)
{
struct peer *peer;
list_for_each(&dstate->peers, peer, list)
try_reconnect(peer);
}
/* Return earliest block we're interested in, or 0 for none. */
u32 get_peer_min_block(struct lightningd_state *dstate)
{
u32 min_block = 0;
struct peer *peer;
/* If loaded from database, go back to earliest possible peer anchor. */
list_for_each(&dstate->peers, peer, list) {
if (!peer->anchor.min_depth)
continue;
if (min_block == 0 || peer->anchor.min_depth < min_block)
min_block = peer->anchor.min_depth;
}
return min_block;
}
/* We may have gone down before broadcasting the anchor. Try again. */
void rebroadcast_anchors(struct lightningd_state *dstate)
{
struct peer *peer;
list_for_each(&dstate->peers, peer, list) {
if (!state_is_waiting_for_anchor(peer->state))
continue;
if (!peer->anchor.ours)
continue;
if (!bitcoin_create_anchor(peer))
peer_fail(peer, __func__);
else
broadcast_tx(peer->dstate->topology,
peer, peer->anchor.tx, NULL);
}
}
static void json_add_abstime(struct json_result *response,
const char *id,
const struct abs_locktime *t)
{
json_object_start(response, id);
if (abs_locktime_is_seconds(t))
json_add_num(response, "second", abs_locktime_to_seconds(t));
else
json_add_num(response, "block", abs_locktime_to_blocks(t));
json_object_end(response);
}
static void json_add_htlcs(struct json_result *response,
const char *id,
struct peer *peer,
enum side owner)
{
struct htlc_map_iter it;
struct htlc *h;
const struct htlc_map *htlcs = &peer->htlcs;
json_array_start(response, id);
for (h = htlc_map_first(htlcs, &it); h; h = htlc_map_next(htlcs, &it)) {
if (htlc_owner(h) != owner)
continue;
/* Ignore completed HTLCs. */
if (htlc_is_dead(h))
continue;
json_object_start(response, NULL);
json_add_u64(response, "msatoshi", h->msatoshi);
json_add_abstime(response, "expiry", &h->expiry);
json_add_hex(response, "rhash", &h->rhash, sizeof(h->rhash));
json_add_string(response, "state", htlc_state_name(h->state));
json_object_end(response);
}
json_array_end(response);
}
/* FIXME: add history command which shows all prior and current commit txs */
/* FIXME: Somehow we should show running DNS lookups! */
static void json_getpeers(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *p;
struct json_result *response = new_json_result(cmd);
json_object_start(response, NULL);
json_array_start(response, "peers");
list_for_each(&cmd->dstate->peers, p, list) {
const struct channel_state *last;
json_object_start(response, NULL);
json_add_string(response, "name", log_prefix(p->log));
json_add_string(response, "state", state_name(p->state));
if (p->id)
json_add_pubkey(response, "peerid", p->id);
json_add_bool(response, "connected", p->connected);
/* FIXME: Report anchor. */
if (!p->local.commit || !p->local.commit->cstate) {
json_object_end(response);
continue;
}
last = p->local.commit->cstate;
json_add_num(response, "our_amount", last->side[LOCAL].pay_msat);
json_add_num(response, "our_fee", last->side[LOCAL].fee_msat);
json_add_num(response, "their_amount", last->side[REMOTE].pay_msat);
json_add_num(response, "their_fee", last->side[REMOTE].fee_msat);
json_add_htlcs(response, "our_htlcs", p, LOCAL);
json_add_htlcs(response, "their_htlcs", p, REMOTE);
json_object_end(response);
}
json_array_end(response);
json_object_end(response);
command_success(cmd, response);
}
static const struct json_command getpeers_command = {
"getpeers",
json_getpeers,
"List the current peers",
"Returns a 'peers' array"
};
AUTODATA(json_command, &getpeers_command);
static void json_gethtlcs(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *peer;
jsmntok_t *peeridtok, *resolvedtok;
bool resolved = false;
struct json_result *response = new_json_result(cmd);
struct htlc *h;
struct htlc_map_iter it;
if (!json_get_params(buffer, params,
"peerid", &peeridtok,
"?resolved", &resolvedtok,
NULL)) {
command_fail(cmd, "Need peerid");
return;
}
peer = find_peer_json(cmd->dstate, buffer, peeridtok);
if (!peer) {
command_fail(cmd, "Could not find peer with that peerid");
return;
}
if (resolvedtok && !json_tok_bool(buffer, resolvedtok, &resolved)) {
command_fail(cmd, "resolved must be true or false");
return;
}
json_object_start(response, NULL);
json_array_start(response, "htlcs");
for (h = htlc_map_first(&peer->htlcs, &it);
h; h = htlc_map_next(&peer->htlcs, &it)) {
if (htlc_is_dead(h) && !resolved)
continue;
json_object_start(response, NULL);
json_add_u64(response, "id", h->id);
json_add_string(response, "state", htlc_state_name(h->state));
json_add_u64(response, "msatoshi", h->msatoshi);
json_add_abstime(response, "expiry", &h->expiry);
json_add_hex(response, "rhash", &h->rhash, sizeof(h->rhash));
if (h->r)
json_add_hex(response, "r", h->r, sizeof(*h->r));
if (htlc_owner(h) == LOCAL) {
json_add_num(response, "deadline", h->deadline);
if (h->src) {
json_object_start(response, "src");
json_add_pubkey(response,
"peerid", h->src->peer->id);
json_add_u64(response, "id", h->src->id);
json_object_end(response);
}
} else {
if (h->routing)
json_add_hex(response, "routing",
h->routing, tal_count(h->routing));
}
json_object_end(response);
}
json_array_end(response);
json_object_end(response);
command_success(cmd, response);
}
static const struct json_command gethtlcs_command = {
"gethtlcs",
json_gethtlcs,
"List HTLCs for {peer}; all if {resolved} is true.",
"Returns a 'htlcs' array"
};
AUTODATA(json_command, &gethtlcs_command);
/* To avoid freeing underneath ourselves, we free outside event loop. */
void cleanup_peers(struct lightningd_state *dstate)
{
struct peer *peer, *next;
list_for_each_safe(&dstate->peers, peer, next, list) {
/* Deletes itself from list. */
if (!peer->conn && peer->state == STATE_CLOSED)
tal_free(peer);
}
}
static void json_newhtlc(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *peer;
jsmntok_t *peeridtok, *msatoshitok, *expirytok, *rhashtok;
unsigned int expiry;
u64 msatoshi;
struct sha256 rhash;
struct json_result *response = new_json_result(cmd);
struct htlc *htlc;
const char *err;
enum fail_error error_code;
struct hoppayload *hoppayloads;
u8 sessionkey[32];
struct onionpacket *packet;
u8 *onion;
struct pubkey *path = tal_arrz(cmd, struct pubkey, 1);
if (!json_get_params(buffer, params,
"peerid", &peeridtok,
"msatoshi", &msatoshitok,
"expiry", &expirytok,
"rhash", &rhashtok,
NULL)) {
command_fail(cmd, "Need peerid, msatoshi, expiry and rhash");
return;
}
peer = find_peer_json(cmd->dstate, buffer, peeridtok);
if (!peer) {
command_fail(cmd, "Could not find peer with that peerid");
return;
}
if (!peer->remote.commit || !peer->remote.commit->cstate) {
command_fail(cmd, "peer not fully established");
return;
}
if (!peer->connected) {
command_fail(cmd, "peer not connected");
return;
}
if (!json_tok_u64(buffer, msatoshitok, &msatoshi)) {
command_fail(cmd, "'%.*s' is not a valid number",
(int)(msatoshitok->end - msatoshitok->start),
buffer + msatoshitok->start);
return;
}
if (!json_tok_number(buffer, expirytok, &expiry)) {
command_fail(cmd, "'%.*s' is not a valid number",
(int)(expirytok->end - expirytok->start),
buffer + expirytok->start);
return;
}
if (!hex_decode(buffer + rhashtok->start,
rhashtok->end - rhashtok->start,
&rhash, sizeof(rhash))) {
command_fail(cmd, "'%.*s' is not a valid sha256 hash",
(int)(rhashtok->end - rhashtok->start),
buffer + rhashtok->start);
return;
}
tal_arr(cmd, struct pubkey, 1);
hoppayloads = tal_arrz(cmd, struct hoppayload, 1);
memcpy(&path[0], peer->id, sizeof(struct pubkey));
randombytes_buf(&sessionkey, sizeof(sessionkey));
packet = create_onionpacket(cmd, path, hoppayloads, sessionkey,
rhash.u.u8, sizeof(rhash));
onion = serialize_onionpacket(cmd, packet);
log_debug(peer->log, "JSON command to add new HTLC");
err = command_htlc_add(peer, msatoshi, expiry, &rhash, NULL,
onion,
&error_code, &htlc);
if (err) {
command_fail(cmd, "could not add htlc: %u:%s", error_code, err);
return;
}
log_debug(peer->log, "JSON new HTLC is %"PRIu64, htlc->id);
json_object_start(response, NULL);
json_add_u64(response, "id", htlc->id);
json_object_end(response);
command_success(cmd, response);
}
static const struct json_command dev_newhtlc_command = {
"dev-newhtlc",
json_newhtlc,
"Offer {peerid} an HTLC worth {msatoshi} in {expiry} (block number) with {rhash}",
"Returns { id: u64 } result on success"
};
AUTODATA(json_command, &dev_newhtlc_command);
static void json_fulfillhtlc(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *peer;
jsmntok_t *peeridtok, *idtok, *rtok;
u64 id;
struct htlc *htlc;
struct sha256 rhash;
struct preimage r;
if (!json_get_params(buffer, params,
"peerid", &peeridtok,
"id", &idtok,
"r", &rtok,
NULL)) {
command_fail(cmd, "Need peerid, id and r");
return;
}
peer = find_peer_json(cmd->dstate, buffer, peeridtok);
if (!peer) {
command_fail(cmd, "Could not find peer with that peerid");
return;
}
if (!peer->remote.commit || !peer->remote.commit->cstate) {
command_fail(cmd, "peer not fully established");
return;
}
if (!peer->connected) {
command_fail(cmd, "peer not connected");
return;
}
if (!json_tok_u64(buffer, idtok, &id)) {
command_fail(cmd, "'%.*s' is not a valid id",
(int)(idtok->end - idtok->start),
buffer + idtok->start);
return;
}
if (!hex_decode(buffer + rtok->start,
rtok->end - rtok->start,
&r, sizeof(r))) {
command_fail(cmd, "'%.*s' is not a valid sha256 preimage",
(int)(rtok->end - rtok->start),
buffer + rtok->start);
return;
}
htlc = htlc_get(&peer->htlcs, id, REMOTE);
if (!htlc) {
command_fail(cmd, "preimage htlc not found");
return;
}
if (htlc->state != RCVD_ADD_ACK_REVOCATION) {
command_fail(cmd, "htlc in state %s",
htlc_state_name(htlc->state));
return;
}
sha256(&rhash, &r, sizeof(r));
if (!structeq(&htlc->rhash, &rhash)) {
command_fail(cmd, "preimage incorrect");
return;
}
/* This can happen if we're disconnected, and thus haven't sent
* fulfill yet; we stored r in database immediately. */
if (!htlc->r) {
const char *db_err;
db_start_transaction(peer);
set_htlc_rval(peer, htlc, &r);
/* We can relay this upstream immediately. */
our_htlc_fulfilled(peer, htlc);
db_err = db_commit_transaction(peer);
if (db_err) {
command_fail(cmd, "%s", db_err);
return;
}
}
if (command_htlc_fulfill(peer, htlc))
command_success(cmd, null_response(cmd));
else
command_fail(cmd,
"htlc_fulfill not possible in state %s",
state_name(peer->state));
}
static const struct json_command dev_fulfillhtlc_command = {
"dev-fulfillhtlc",
json_fulfillhtlc,
"Redeem htlc proposed by {peerid} of {id} using {r}",
"Returns an empty result on success"
};
AUTODATA(json_command, &dev_fulfillhtlc_command);
static void json_failhtlc(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *peer;
jsmntok_t *peeridtok, *idtok, *reasontok;
u64 id;
struct htlc *htlc;
const char *db_err;
if (!json_get_params(buffer, params,
"peerid", &peeridtok,
"id", &idtok,
"reason", &reasontok,
NULL)) {
command_fail(cmd, "Need peerid, id and reason");
return;
}
peer = find_peer_json(cmd->dstate, buffer, peeridtok);
if (!peer) {
command_fail(cmd, "Could not find peer with that peerid");
return;
}
if (!peer->remote.commit || !peer->remote.commit->cstate) {
command_fail(cmd, "peer not fully established");
return;
}
if (!peer->connected) {
command_fail(cmd, "peer not connected");
return;
}
if (!json_tok_u64(buffer, idtok, &id)) {
command_fail(cmd, "'%.*s' is not a valid id",
(int)(idtok->end - idtok->start),
buffer + idtok->start);
return;
}
htlc = htlc_get(&peer->htlcs, id, REMOTE);
if (!htlc) {
command_fail(cmd, "preimage htlc not found");
return;
}
if (htlc->state != RCVD_ADD_ACK_REVOCATION) {
command_fail(cmd, "htlc in state %s",
htlc_state_name(htlc->state));
return;
}
db_start_transaction(peer);
set_htlc_fail(peer, htlc, buffer + reasontok->start,
reasontok->end - reasontok->start);
db_err = db_commit_transaction(peer);
if (db_err) {
command_fail(cmd, "%s", db_err);
return;
}
if (command_htlc_fail(peer, htlc))
command_success(cmd, null_response(cmd));
else
command_fail(cmd,
"htlc_fail not possible in state %s",
state_name(peer->state));
}
static const struct json_command dev_failhtlc_command = {
"dev-failhtlc",
json_failhtlc,
"Fail htlc proposed by {peerid} which has {id}, using {reason}",
"Returns an empty result on success"
};
AUTODATA(json_command, &dev_failhtlc_command);
static void json_commit(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *peer;
jsmntok_t *peeridtok;
if (!json_get_params(buffer, params,
"peerid", &peeridtok,
NULL)) {
command_fail(cmd, "Need peerid");
return;
}
peer = find_peer_json(cmd->dstate, buffer, peeridtok);
if (!peer) {
command_fail(cmd, "Could not find peer with that peerid");
return;
}
if (!peer->remote.commit || !peer->remote.commit->cstate) {
command_fail(cmd, "peer not fully established");
return;
}
if (!peer->connected) {
command_fail(cmd, "peer not connected");
return;
}
if (!state_can_commit(peer->state)) {
command_fail(cmd, "peer in state %s", state_name(peer->state));
return;
}
do_commit(peer, cmd);
}
static const struct json_command dev_commit_command = {
"dev-commit",
json_commit,
"Commit all staged HTLC changes with {peerid}",
"Returns an empty result on success"
};
AUTODATA(json_command, &dev_commit_command);
static void json_close(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *peer;
jsmntok_t *peeridtok;
if (!json_get_params(buffer, params,
"peerid", &peeridtok,
NULL)) {
command_fail(cmd, "Need peerid");
return;
}
peer = find_peer_json(cmd->dstate, buffer, peeridtok);
if (!peer) {
command_fail(cmd, "Could not find peer with that peerid");
return;
}
if (!state_is_normal(peer->state) && !state_is_opening(peer->state)) {
command_fail(cmd, "Peer is already closing: state %s",
state_name(peer->state));
return;
}
if (!peer_start_shutdown(peer)) {
command_fail(cmd, "Database error");
return;
}
/* FIXME: Block until closed! */
command_success(cmd, null_response(cmd));
}
static const struct json_command close_command = {
"close",
json_close,
"Close the channel with peer {peerid}",
"Returns an empty result on success"
};
AUTODATA(json_command, &close_command);
static void json_feerate(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
jsmntok_t *feeratetok;
u64 feerate;
if (!json_get_params(buffer, params,
"feerate", &feeratetok,
NULL)) {
command_fail(cmd, "Need feerate");
return;
}
if (!json_tok_u64(buffer, feeratetok, &feerate)) {
command_fail(cmd, "Invalid feerate");
return;
}
log_debug(cmd->jcon->log, "Fee rate changed to %"PRIu64, feerate);
cmd->dstate->topology->default_fee_rate = feerate;
command_success(cmd, null_response(cmd));
}
static const struct json_command dev_feerate_command = {
"dev-feerate",
json_feerate,
"Change the (default) fee rate to {feerate}",
"Returns an empty result on success"
};
AUTODATA(json_command, &dev_feerate_command);
static void json_disconnect(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *peer;
jsmntok_t *peeridtok;
if (!json_get_params(buffer, params,
"peerid", &peeridtok,
NULL)) {
command_fail(cmd, "Need peerid");
return;
}
peer = find_peer_json(cmd->dstate, buffer, peeridtok);
if (!peer) {
command_fail(cmd, "Could not find peer with that peerid");
return;
}
if (!peer->conn) {
command_fail(cmd, "Peer is already disconnected");
return;
}
/* We don't actually close it, since for testing we want only
* one side to freak out. We just ensure we ignore it. */
log_debug(peer->log, "Pretending connection is closed");
peer->fake_close = true;
peer->connected = false;
peer_fail(peer, __func__);
command_success(cmd, null_response(cmd));
}
static void json_reconnect(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *peer;
jsmntok_t *peeridtok;
if (!json_get_params(buffer, params,
"peerid", &peeridtok,
NULL)) {
command_fail(cmd, "Need peerid");
return;
}
peer = find_peer_json(cmd->dstate, buffer, peeridtok);
if (!peer) {
command_fail(cmd, "Could not find peer with that peerid");
return;
}
if (!peer->conn) {
command_fail(cmd, "Peer is already disconnected");
return;
}
/* Should reconnect on its own. */
io_close(peer->conn);
command_success(cmd, null_response(cmd));
}
static void json_signcommit(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *peer;
jsmntok_t *peeridtok;
u8 *linear;
struct json_result *response = new_json_result(cmd);
if (!json_get_params(buffer, params,
"peerid", &peeridtok,
NULL)) {
command_fail(cmd, "Need peerid");
return;
}
peer = find_peer_json(cmd->dstate, buffer, peeridtok);
if (!peer) {
command_fail(cmd, "Could not find peer with that peerid");
return;
}
if (!peer->local.commit->sig) {
command_fail(cmd, "Peer has not given us a signature");
return;
}
sign_commit_tx(peer);
linear = linearize_tx(cmd, peer->local.commit->tx);
/* Clear witness for potential future uses. */
peer->local.commit->tx->input[0].witness
= tal_free(peer->local.commit->tx->input[0].witness);
json_object_start(response, NULL);
json_add_string(response, "tx", tal_hex(cmd, linear));
json_object_end(response);
command_success(cmd, response);
}
static void json_output(struct command *cmd,
const char *buffer, const jsmntok_t *params)
{
struct peer *peer;
jsmntok_t *peeridtok, *enabletok;
bool enable;
if (!json_get_params(buffer, params,
"peerid", &peeridtok,
"enable", &enabletok,
NULL)) {
command_fail(cmd, "Need peerid and enable");
return;
}
peer = find_peer_json(cmd->dstate, buffer, peeridtok);
if (!peer) {
command_fail(cmd, "Could not find peer with that peerid");
return;
}
if (!peer->conn) {
command_fail(cmd, "Peer is already disconnected");
return;
}
if (!json_tok_bool(buffer, enabletok, &enable)) {
command_fail(cmd, "enable must be true or false");
return;
}
log_debug(peer->log, "dev-output: output %s",
enable ? "enabled" : "disabled");
peer->output_enabled = enable;
/* Flush any outstanding output */
if (peer->output_enabled)
io_wake(peer);
command_success(cmd, null_response(cmd));
}
static const struct json_command dev_output_command = {
"dev-output",
json_output,
"Enable/disable any messages to peer {peerid} depending on {enable}",
"Returns an empty result on success"
};
AUTODATA(json_command, &dev_output_command);
static const struct json_command dev_disconnect_command = {
"dev-disconnect",
json_disconnect,
"Force a disconnect with peer {peerid}",
"Returns an empty result on success"
};
AUTODATA(json_command, &dev_disconnect_command);
static const struct json_command dev_reconnect_command = {
"dev-reconnect",
json_reconnect,
"Force a reconnect with peer {peerid}",
"Returns an empty result on success"
};
AUTODATA(json_command, &dev_reconnect_command);
static const struct json_command dev_signcommit_command = {
"dev-signcommit",
json_signcommit,
"Sign and return the current commit with peer {peerid}",
"Returns a hex string on success"
};
AUTODATA(json_command, &dev_signcommit_command);