htlcswitch: return inbound channel update #6967

joostjager · 2022-10-03T14:17:16Z

This PR shows the changes required to return the inbound channel update in a FeeInsufficient failure too. This would be an extension of #6703. The goal is to let a subsequent payment attempt succeed also in the case that the sender didn't have an up to date channel policy for the incoming channel and was using outdated inbound fees.

To remain backwards compatible, senders that are able to interpret long failure messages will signal this through onion tlv field 333. Its value is a single byte indicating the failure message version that the sender supports. To receive long failures, the sender should set the value to 1.

Builds on #6703.

joostjager · 2022-12-05T11:53:40Z

Updated PR to not always return 1024 byte (padded) failure messages, because old senders would still choke on those. Currently it pads to 256 bytes. Longer messages aren't padded at all.

t-bast · 2022-12-16T11:18:08Z

I'm trying to test this against ACINQ/eclair#2455 and I'm seeing issues.

First of all, I can't get lnd to send an error with a tlv stream: I do set tlv field 333 in the onion (the encoded tlv is: 0xfd014d0101), and send a payment Alice -> Bob -> Carol where the Bob -> Carol channel doesn't have enough liquidity, yet Bob doesn't add a tlv stream to the failure. Do I need to trigger a different, more specific error to get lnd to add tlvs?

Secondly, I tried it the other way around: eclair adds a tlv stream at the end of TemporaryChannelFailure. But this seems to break lnd:

2022-12-16 12:10:47.215 [ERR] CRTR: Unable to validate channel update: invalid signature for channel update (*lnwire.ChannelUpdate)(0xc00122fb80)({
 Signature: (lnwire.Sig) (len=64 cap=64) {
  00000000  c3 16 21 5b 71 35 90 bf  7e c2 54 a6 08 ea 6c 2c  |..![q5..~.T...l,|
  00000010  95 05 61 ed 1c 84 78 bf  ac 68 2d 03 2c cd 95 b5  |..a...x..h-.,...|
  00000020  41 fe aa 91 7d bb 25 e6  4d 24 bb 40 f0 be f9 38  |A...}.%.M$.@...8|
  00000030  f5 29 96 5c 71 8f 07 06  46 d1 ee 32 4f 21 1d d9  |.).\q...F..2O!..|
 },
 ChainHash: (chainhash.Hash) (len=32 cap=32) 0f9188f13cb7b2c71f2a335e3a4fc328bf5beb436012afca590b1a11466e2206,
 ShortChannelID: (lnwire.ShortChannelID) 1746:2:1,
 Timestamp: (uint32) 1671188835,
 MessageFlags: (lnwire.ChanUpdateMsgFlags) 00000001,
 ChannelFlags: (lnwire.ChanUpdateChanFlags) 00000001,
 TimeLockDelta: (uint16) 48,
 HtlcMinimumMsat: (lnwire.MilliSatoshi) 1 mSAT,
 BaseFee: (uint32) 1000,
 FeeRate: (uint32) 200,
 HtlcMaximumMsat: (lnwire.MilliSatoshi) 67500000 mSAT,
 ExtraOpaqueData: (lnwire.ExtraOpaqueData) (len=15 cap=512) {
  00000000  fd 02 31 04 de ad be ef  fd 02 33 03 2a 2a 2a     |..1.......3.***|
 }
})

I believe lnd incorrectly assigns the tlv stream to the channel update, whereas the channel update is length-prefixed so it should be able to figure out that this tlv stream is outside of the channel update and belongs to the failure message.

joostjager · 2022-12-16T12:05:23Z

The encoded TLV looks good. But for inbound fees, an additional channel update is only returned in the case of a fee_insufficient failure. LND does not attach anything extra for temp_channel_failure. If you keep using the 333 tlv record and under-pay Bob's routing fee, you should get the expected failure message.

The other issue I will take a look at, thanks for reporting!

joostjager · 2022-12-16T12:37:50Z

Fixed the invalid channel update issue, fix pushed.

t-bast · 2022-12-16T13:21:23Z

If you keep using the 333 tlv record and under-pay Bob's routing fee, you should get the expected failure message.

Cool thanks for the details, I've done that and eclair was correctly able to deserialize the tlv stream sent by lnd 👍

Fixed the invalid channel update issue, fix pushed.

It doesn't seem to be fixed as of 13a7e03
I'm still getting the following error from lnd when sending back a temporary_channel_failure containing a tlv stream:

2022-12-16 14:09:40.108 [WRN] CRTR: Attempt 1003 for payment fc5e53a0094807b1bc26e30a3baf5294b429e9cadacca04cd569a928fe8f906b failed: TemporaryChannelFailure(update=(*lnwire.ChannelUpdate)(0xc0007f40b0)({
 Signature: (lnwire.Sig) (len=64 cap=64) {
  00000000  60 5f 88 0f 54 45 b8 9c  38 6e 40 50 a6 8d 76 7d  |`_..TE..8n@P..v}|
  00000010  09 ac 5d 0c 5f c5 7b f7  b1 62 22 92 cd 25 a8 ef  |..]._.{..b"..%..|
  00000020  6e a4 bf d3 95 5b 3c a3  be 4a e4 4c 91 fe fe a8  |n....[<..J.L....|
  00000030  12 be 20 0d 0a a5 f8 03  2b e3 9a 7d 91 ef fe f7  |.. .....+..}....|
 },
 ChainHash: (chainhash.Hash) (len=32 cap=32) 0f9188f13cb7b2c71f2a335e3a4fc328bf5beb436012afca590b1a11466e2206,
 ShortChannelID: (lnwire.ShortChannelID) 1786:2:1,
 Timestamp: (uint32) 1671195821,
 MessageFlags: (lnwire.ChanUpdateMsgFlags) 00000001,
 ChannelFlags: (lnwire.ChanUpdateChanFlags) 00000000,
 TimeLockDelta: (uint16) 48,
 HtlcMinimumMsat: (lnwire.MilliSatoshi) 1 mSAT,
 BaseFee: (uint32) 1000,
 FeeRate: (uint32) 200,
 HtlcMaximumMsat: (lnwire.MilliSatoshi) 67500000 mSAT,
 ExtraOpaqueData: (lnwire.ExtraOpaqueData) (len=15 cap=512) {
  00000000  fd 02 31 04 de ad be ef  fd 02 33 03 2a 2a 2a     |..1.......3.***|
 }
})
)@1
2022-12-16 14:09:40.121 [ERR] CRTR: Unable to validate channel update: invalid signature for channel update (*lnwire.ChannelUpdate)(0xc0007f40b0)({
 Signature: (lnwire.Sig) (len=64 cap=64) {
  00000000  60 5f 88 0f 54 45 b8 9c  38 6e 40 50 a6 8d 76 7d  |`_..TE..8n@P..v}|
  00000010  09 ac 5d 0c 5f c5 7b f7  b1 62 22 92 cd 25 a8 ef  |..]._.{..b"..%..|
  00000020  6e a4 bf d3 95 5b 3c a3  be 4a e4 4c 91 fe fe a8  |n....[<..J.L....|
  00000030  12 be 20 0d 0a a5 f8 03  2b e3 9a 7d 91 ef fe f7  |.. .....+..}....|
 },
 ChainHash: (chainhash.Hash) (len=32 cap=32) 0f9188f13cb7b2c71f2a335e3a4fc328bf5beb436012afca590b1a11466e2206,
 ShortChannelID: (lnwire.ShortChannelID) 1786:2:1,
 Timestamp: (uint32) 1671195821,
 MessageFlags: (lnwire.ChanUpdateMsgFlags) 00000001,
 ChannelFlags: (lnwire.ChanUpdateChanFlags) 00000000,
 TimeLockDelta: (uint16) 48,
 HtlcMinimumMsat: (lnwire.MilliSatoshi) 1 mSAT,
 BaseFee: (uint32) 1000,
 FeeRate: (uint32) 200,
 HtlcMaximumMsat: (lnwire.MilliSatoshi) 67500000 mSAT,
 ExtraOpaqueData: (lnwire.ExtraOpaqueData) (len=15 cap=512) {
  00000000  fd 02 31 04 de ad be ef  fd 02 33 03 2a 2a 2a     |..1.......3.***|
 }
})

Here is the encoded failure message I'm sending:

6361222ca3f8091416c6bb7097bed80e5be5a88e5010eb461078d944eb978206 (mac)
009d (failure length)
1007008a0102605f880f5445b89c386e4050a68d767d09ac5d0c5fc57bf7b1622292cd25a8ef6ea4bfd3955b3ca3be4ae44c91fefea812be200d0aa5f8032be39a7d91effef706226e46111a0b59caaf126043eb5bbf28c34f3a5e332a1fc7b2b73cf188910f0006fa0000020001639c6cad010000300000000000000001000003e8000000c8000000000405f7e0fd023104deadbeeffd0233032a2a2a (failure)
0063 (padding length)
000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000 (padding)

Here is the detailed breakdown of the encoded failure:

1007 (temporary_channel_failure)
008a (channel_update length)
0102 (channel_update message type)
605f880f5445b89c386e4050a68d767d09ac5d0c5fc57bf7b1622292cd25a8ef6ea4bfd3955b3ca3be4ae44c91fefea812be200d0aa5f8032be39a7d91effef706226e46111a0b59caaf126043eb5bbf28c34f3a5e332a1fc7b2b73cf188910f0006fa0000020001639c6cad010000300000000000000001000003e8000000c8000000000405f7e0 (channel update)
fd023104deadbeef fd0233032a2a2a (two tlv fields in temporary_channel_failure's tlv stream)

I double-checked the lengths and they seem correct, so lnd should be able to separate the channel_update from the failure's tlv stream, but for some reason it doesn't seem to do it correctly.

joostjager · 2022-12-16T13:54:16Z

Got it, thanks for all those intermediate results. Turns out this bug exists for other messages that contain a channel update as well. Will fix.

joostjager · 2022-12-19T09:56:44Z

Channel update reader fixed in #7262

joostjager · 2023-01-03T09:56:41Z

@t-bast rebased on top of #7262 to fix the failure tlv parsing bug, as discussed in yesterday's spec meeting.

t-bast · 2023-01-03T14:12:05Z

Thanks, I've re-run my cross-compat tests and everything looks good!

In this commit, the tlv extension of a channel update message is parsed. If an inbound fee schedule is encountered, it is reported in the graph rpc calls.

Preparation for negative fees.

Ensure that negative fees are backwards compatible.

Crypt-iQ · 2023-10-06T19:33:09Z

lnwire/onion_error.go

+		return err
+	}
+
+	types, err := tlvStream.DecodeWithParsedTypes(r)


This should be using DecodeWithParsedTypesP2P to avoid a fake record with a maliciously long length

Crypt-iQ · 2023-10-06T19:33:41Z

lnwire/onion_error.go

+	}
+
+	var b bytes.Buffer
+	if err := f.IncomingUpdate.Encode(&b, pver); err != nil {


b isn't used again, so can remove these 4 lines?

Crypt-iQ · 2023-10-06T19:34:03Z

routing/router.go

 	case *lnwire.FailIncorrectCltvExpiry:
 		update = &onionErr.Update
 	case *lnwire.FailChannelDisabled:
 		update = &onionErr.Update
 	case *lnwire.FailTemporaryChannelFailure:
+		if onionErr.Update == nil {


Curious, why was this added?

IIRC, without this a nil update would be appended to the new updates list.

joostjager · 2023-10-18T07:28:16Z

The main open question for this PR is whether we still need a signal from the sender that they support long failure messages. Maybe in the mean time the proliferation of lnd 0.16 (which includes #6913) is sufficient to just assume that support?

joostjager · 2023-11-27T07:27:37Z

To get a better idea about whether the signaling is needed, advertized feature bits of public nodes could be useful. Long failure message incompatibility is purely an lnd issue, other implementations have supported it from the start afaik.

Feature 31 amp is lnd-specific and has been in lnd for a long time already. In the public graph, there are 7036 nodes that have feature 31. Probably very close to the total number of public lnd nodes.

Feature 27 shutdown-any-segwit has been added in lnd 0.15.1. The number of public nodes that advertize both 31 and 27 is 7003.

So it seems that nearly every lnd node on the network is 0.15.1 or above. Unfortunately no new feature bits have been added after that, so it is unclear what the proportion of nodes running 0.16+ is.

saubyk · 2023-11-27T15:33:59Z

Unfortunately no new feature bits have been added after that, so it is unclear what the proportion of nodes running 0.16+ is.

Given that mitigation for replacement recycling attack was made available with version 0.16.1 onwards, can we assume that most lnd nodes on the network would've upgraded?

Also, what would be the impact if a portion of the nodes are not supporting long failure message?

joostjager · 2023-11-27T16:12:22Z

Given that mitigation for replacement recycling attack was made available with version 0.16.1 onwards, can we assume that most lnd nodes on the network would've upgraded?

Sounds reasonable.

Also, what would be the impact if a portion of the nodes are not supporting long failure message?

A non-compatible sender isn't able to properly attribute the error to a (pair of) nodes and will penalize the full route, if I remember it correctly from #6913.

joostjager mentioned this pull request Oct 3, 2022

htlcswitch: add inbound routing fees receive support #6703

Merged

joostjager added the routing label Oct 3, 2022

joostjager mentioned this pull request Oct 6, 2022

TLV failure message and length relaxation lightning/bolts#1021

Merged

saubyk added the inbound fee Changes related to inbound routing fee label Nov 9, 2022

saubyk added this to the v0.17.0 milestone Nov 9, 2022

joostjager force-pushed the inbound-fees-signal branch 3 times, most recently from 7774d4d to 2f3b228 Compare December 5, 2022 11:51

joostjager force-pushed the inbound-fees-signal branch from 2f3b228 to 13a7e03 Compare December 16, 2022 12:37

joostjager force-pushed the inbound-fees-signal branch from 13a7e03 to e9bbfd9 Compare January 3, 2023 09:53

t-bast mentioned this pull request Jan 3, 2023

Add tlv stream to onion failures ACINQ/eclair#2455

Merged

joostjager added 9 commits January 5, 2023 13:43

lnwire+channeldb: parse inbound fees

23fb6eb

In this commit, the tlv extension of a channel update message is parsed. If an inbound fee schedule is encountered, it is reported in the graph rpc calls.

lnrpc: signed fees

30aa8f6

Preparation for negative fees.

htlcswitch: add receiver-side inbound fee support

70ea795

routing+lnrpc: add inbound fee policy update

c1b804a

itest: multi-hop payment test with negative inbound fee

998c37f

Ensure that negative fees are backwards compatible.

lnwire: allow encoding of onion failure messages longer than 256 bytes

785a9f9

htlcswitch+routing: signal failure message format version

e51609a

htlcswitch: include incoming channel update in failure message

0c1d909

routing: apply incoming channel update if present

d03cef8

joostjager force-pushed the inbound-fees-signal branch from e9bbfd9 to d03cef8 Compare January 5, 2023 12:47

saubyk added the P1 MUST be fixed or reviewed label Aug 8, 2023

saubyk removed this from the Low Priority milestone Aug 8, 2023

saubyk assigned joostjager Aug 10, 2023

saubyk added this to the v0.18.0 milestone Aug 10, 2023

Crypt-iQ reviewed Oct 6, 2023

View reviewed changes

saubyk removed this from the v0.18.0 milestone Jan 28, 2024

Roasbeef mentioned this pull request Mar 30, 2024

lnwire: hard reject positive inbound fees set over RPC #8605

Open

feelancer21 mentioned this pull request Sep 2, 2024

[bug]: Positive inbound fees mess with route finding #9058

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

htlcswitch: return inbound channel update #6967

htlcswitch: return inbound channel update #6967

joostjager commented Oct 3, 2022 •

edited

Loading

joostjager commented Dec 5, 2022

t-bast commented Dec 16, 2022

joostjager commented Dec 16, 2022

joostjager commented Dec 16, 2022

t-bast commented Dec 16, 2022

joostjager commented Dec 16, 2022

joostjager commented Dec 19, 2022

joostjager commented Jan 3, 2023

t-bast commented Jan 3, 2023

Crypt-iQ Oct 6, 2023

Crypt-iQ Oct 6, 2023

Crypt-iQ Oct 6, 2023

joostjager Oct 18, 2023

joostjager commented Oct 18, 2023

joostjager commented Nov 27, 2023

saubyk commented Nov 27, 2023

joostjager commented Nov 27, 2023

htlcswitch: return inbound channel update #6967

Are you sure you want to change the base?

htlcswitch: return inbound channel update #6967

Conversation

joostjager commented Oct 3, 2022 • edited Loading

joostjager commented Dec 5, 2022

t-bast commented Dec 16, 2022

joostjager commented Dec 16, 2022

joostjager commented Dec 16, 2022

t-bast commented Dec 16, 2022

joostjager commented Dec 16, 2022

joostjager commented Dec 19, 2022

joostjager commented Jan 3, 2023

t-bast commented Jan 3, 2023

Crypt-iQ Oct 6, 2023

Choose a reason for hiding this comment

Crypt-iQ Oct 6, 2023

Choose a reason for hiding this comment

Crypt-iQ Oct 6, 2023

Choose a reason for hiding this comment

joostjager Oct 18, 2023

Choose a reason for hiding this comment

joostjager commented Oct 18, 2023

joostjager commented Nov 27, 2023

saubyk commented Nov 27, 2023

joostjager commented Nov 27, 2023

joostjager commented Oct 3, 2022 •

edited

Loading