lightningd: don't crash if peer manages to spend onchain HTLC after w…

…e've abandoned upstream.

Since we now abandon HTLCs to avoid breaking a channel (in limited circumstances), the
peer can take the funds if they beat us to spend the HTLC before we timeout.  This is extremely
unusual, but finally happend in CI.

First it fails our internal sanity checks, but then it would fail when we tried to fulfill
an already-failed HTLC:

```
**BROKEN** lightningd: fulfill_our_htlc_out:Output preimage, input failonion?
**BROKEN** lightningd: FATAL SIGNAL 6 (version f82fedb-modded)
**BROKEN** lightningd: backtrace: common/daemon.c:38 (send_backtrace) 0x55c09b04f0b5
**BROKEN** lightningd: backtrace: common/daemon.c:75 (crashdump) 0x55c09b04f247
**BROKEN** lightningd: backtrace: ./signal/../sysdeps/unix/sysv/linux/x86_64/libc_sigaction.c:0 ((null)) 0x7f607463551f
**BROKEN** lightningd: backtrace: ./nptl/pthread_kill.c:44 (__pthread_kill_implementation) 0x7f6074689a7c
**BROKEN** lightningd: backtrace: ./nptl/pthread_kill.c:78 (__pthread_kill_internal) 0x7f6074689a7c
**BROKEN** lightningd: backtrace: ./nptl/pthread_kill.c:89 (__GI___pthread_kill) 0x7f6074689a7c
**BROKEN** lightningd: backtrace: ../sysdeps/posix/raise.c:26 (__GI_raise) 0x7f6074635475
**BROKEN** lightningd: backtrace: ./stdlib/abort.c:79 (__GI_abort) 0x7f607461b7f2
**BROKEN** lightningd: backtrace: lightningd/log.c:1016 (fatal_vfmt) 0x55c09afdb7cb
**BROKEN** lightningd: backtrace: lightningd/log.c:1026 (fatal) 0x55c09afdb880
**BROKEN** lightningd: backtrace: lightningd/htlc_end.c:87 (corrupt) 0x55c09afc9472
**BROKEN** lightningd: backtrace: lightningd/htlc_end.c:207 (htlc_out_check) 0x55c09afc9c6b
**BROKEN** lightningd: backtrace: lightningd/peer_htlcs.c:1451 (fulfill_our_htlc_out) 0x55c09b004dd7
**BROKEN** lightningd: backtrace: lightningd/peer_htlcs.c:1526 (onchain_fulfilled_htlc) 0x55c09b0050fe
**BROKEN** lightningd: backtrace: lightningd/onchain_control.c:313 (handle_extracted_preimage) 0x55c09afdf9f8
**BROKEN** lightningd: backtrace: lightningd/onchain_control.c:1423 (onchain_msg) 0x55c09afe2da9
**BROKEN** lightningd: backtrace: lightningd/subd.c:557 (sd_msg_read) 0x55c09b019ac8
```

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

lightningd/htlc_end.c

@@ -203,9 +203,9 @@ struct htlc_out *htlc_out_check(const struct htlc_out *hout,
 				return corrupt(abortstr,
 					       "Output failmsg, input preimage");
 		} else if (hout->preimage) {
-			if (hout->in->failonion)
-				return corrupt(abortstr,
-					       "Output preimage, input failonion");
+			/* If we abandoned the HTLC to save the incoming channel,
+			 * (see consider_failing_incoming), hout->in->failonion
+			 * will be set! */
 			if (hout->in->badonion)
 				return corrupt(abortstr,
 					       "Output preimage, input badonion");

lightningd/peer_htlcs.c

@@ -1464,11 +1464,18 @@ static void fulfill_our_htlc_out(struct channel *channel, struct htlc_out *hout,
 	if (hout->am_origin)
 		payment_succeeded(ld, &hout->payment_hash, hout->partid, hout->groupid, preimage);
 	else if (hout->in) {
-		fulfill_htlc(hout->in, preimage);
-		wallet_forwarded_payment_add(ld->wallet, hout->in,
-					     FORWARD_STYLE_TLV,
-					     channel_scid_or_local_alias(hout->key.channel), hout,
-					     FORWARD_SETTLED, 0);
+		/* Did we abandon the incoming?  Oops! */
+		if (hout->in->failonion) {
+			/* FIXME: Accounting? */
+			log_unusual(channel->log, "FUNDS LOSS of %s: peer took funds onchain before we could time out the HTLC, but we abandoned incoming HTLC to save the incoming channel",
+				    fmt_amount_msat(tmpctx, hout->msat));
+		} else {
+			fulfill_htlc(hout->in, preimage);
+			wallet_forwarded_payment_add(ld->wallet, hout->in,
+						     FORWARD_STYLE_TLV,
+						     channel_scid_or_local_alias(hout->key.channel), hout,
+						     FORWARD_SETTLED, 0);
+		}
 	}
 }