FATAL SIGNAL on daemon.c in 24.02 and 24.05

[btweenthebars]

Issue and Steps to Reproduce

I keep seeing these errors on the log and console but it doesn't crash my node.

There are 2 variations:

lightning_channeld: FATAL SIGNAL 11 (version v24.05-modded)
0x5ed1c6727571 send_backtrace
common/daemon.c:33
0x5ed1c67275f6 crashdump
common/daemon.c:75
0x73fe53d16adf ???
???:0
0x5ed1c6719538 peer_reconnect
channeld/channeld.c:5205
0x5ed1c6719dab init_channel
channeld/channeld.c:5959
0x5ed1c6719f04 main
channeld/channeld.c:6005
0x73fe53cffc87 ???
???:0
0x73fe53cffd4b ???
???:0
0x5ed1c670e6f4 ???
???:0
0xffffffffffffffff ???
???:0

and

lightning_channeld: FATAL SIGNAL (version v24.05-modded)
0x5ed1c6727571 send_backtrace
common/daemon.c:33
0x5ed1c67322cc status_failed
common/status.c:221
0x5ed1c6732482 status_backtrace_exit
common/subdaemon.c:18
0x5ed1c67275fc crashdump
common/daemon.c:78
0x73fe53d16adf ???
???:0
0x5ed1c6719538 peer_reconnect
channeld/channeld.c:5205
0x5ed1c6719dab init_channel
channeld/channeld.c:5959
0x5ed1c6719f04 main
channeld/channeld.c:6005
0x73fe53cffc87 ???
???:0
0x73fe53cffd4b ???
???:0
0x5ed1c670e6f4 ???
???:0
0xffffffffffffffff ???
???:0

It's been going on since 24.02. I didn't have that in 23.11

getinfo output

"version": "v24.05-modded",
"blockheight": 853495,
"network": "bitcoin",
"fees_collected_msat": 111111,
"lightning-dir": "/cln/bitcoin",
"our_features": {
"init": "080000000000000000000000000008a0080a8a59a1",
"node": "080000000000000000000000000088a0080a8a59a1",
"channel": "",
"invoice": "02000002024100"
}

[cdecker]

Sounds like we are dereferencing the initr pointer here:

lightning/channeld/channeld.c

Lines 5205 to 5209 in 11586ab

	if (match_type(initr->desired_channel_type,
	ninitr->current_channel_type)
	|| match_type(initr->desired_channel_type,
	ninitr->upgradable_channel_type))
	type = initr->desired_channel_type;

How cound that be NULL? It looks as if either the send_tlvs or recv_tlvs is set to NULL. Git blame says @rustyrussell last edited the code here, so he might know how to address this best.

[rustyrussell]

Good catch!

send_tlvs can indeed be NULL, with older peers.