From 58d4d9bca3ae5896eeea4b46324df27b8ecb0ce1 Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Mon, 2 Oct 2017 20:19:26 +1030 Subject: [PATCH] BOLT 2: Details of HTLC Timeouts, ie. `cltv_expiry_delta`. Complete rewrite, including a routing example and the new min_final_cltv expirt. I hope this makes it clear. (Thanks to everyone who reviewed and gave feedback; you rock!) Signed-off-by: Rusty Russell --- 02-peer-protocol.md | 134 ++++++++++++++++++++++++++++++------------- 04-onion-routing.md | 4 +- 07-routing-gossip.md | 90 +++++++++++++++++++++++++++++ 3 files changed, 185 insertions(+), 43 deletions(-) diff --git a/02-peer-protocol.md b/02-peer-protocol.md index cb6343f..049dd5d 100644 --- a/02-peer-protocol.md +++ b/02-peer-protocol.md @@ -16,7 +16,7 @@ operation, and closing. * [Closing negotiation: `closing_signed`](#closing-negotiation-closing_signed) * [Normal Operation](#normal-operation) * [Forwarding HTLCs](#forwarding-htlcs) - * [Risks With HTLC Timeouts](#risks-with-htlc-timeouts) + * [`cltv_expiry_delta` Selection](#cltv_expiry_delta-selection) * [Adding an HTLC: `update_add_htlc`](#adding-an-htlc-update_add_htlc) * [Removing an HTLC: `update_fulfill_htlc`, `update_fail_htlc` and `update_fail_malformed_htlc`](#removing-an-htlc-update_fulfill_htlc-update_fail_htlc-and-update_fail_malformed_htlc) * [Committing Updates So Far: `commitment_signed`](#committing-updates-so-far-commitment_signed) @@ -508,7 +508,7 @@ the incoming HTLC has been irrevocably committed. A node MUST NOT fail an incoming HTLC (`update_fail_htlc`) for which it has committed to an outgoing HTLC, until the removal of the outgoing HTLC is irrevocably committed, or the outgoing on-chain HTLC output has been spent via the HTLC-timeout transaction with sufficient depth. -A node MUST fail an incoming HTLC (`update_fail_htlc`) once its `cltv_expiry` has been reached. +A node MUST fail an incoming HTLC (`update_fail_htlc`) once its `cltv_expiry` has been reached, or if `cltv_expiry` - `current_height` < `cltv_expiry_delta` for the outgoing channel. A node MUST fulfill an incoming HTLC for which it has committed to an outgoing HTLC, as soon as it receives `update_fulfill_htlc` for the outgoing HTLC, or has discovered the `payment_preimage` from an on-chain HTLC spend. @@ -520,61 +520,113 @@ Fulfilling an HTLC is different: knowledge of the preimage is by definition irre so we should fulfill the incoming HTLC as soon as we can to reduce latency. -### Risks With HTLC Timeouts +### `cltv_expiry_delta` Selection -Once an HTLC has timed out where it could either be fulfilled or timed-out; +Once an HTLC has timed out it could either be fulfilled or timed-out; care must be taken around this transition both for offered and received HTLCs. -As a result of forwarding an HTLC from node A to node C, B will end up having an incoming -HTLC from A and an outgoing HTLC to C. B will make sure that the incoming HTLC has a greater -timeout than the outgoing HTLC, so that B can get refunded from C sooner than it has to refund - A if the payment does not complete. +Consider the following scenario, where A sends an HTLC to B, who +forwards to C, who delivers the goods as soon as the payment is +received. -For example, node A might offer node B an HTLC with a timeout of 3 days, and node B might -offer node C the same HTLC with a timeout of 2 days: +1. C needs to be sure that the HTLC from B cannot time out, even if B becomes + unresponsive; i.e. C can fulfill the incoming HTLC on-chain before B can + time it out on-chain. -``` - 3 days timeout 2 days timeout -A ------------------> B ------------------> C -``` +2. B needs to be sure that if C fulfills the HTLC from B, it can fulfill the + incoming HTLC from A. i.e. B can get the preimage from C and fulfill incoming the + HTLC on-chain before A can time it out on-chain. -The difference in timeouts is called `cltv_expiry_delta` in -[BOLT #7](07-routing-gossip.md). +The critical settings here are the `cltv_expiry_delta` in +[BOLT #7](07-routing-gossip.md#the-channel_update-message), and the +related +[`min_final_cltv_expiry` in BOLT #11](11-payment-encoding.md#tagged-fields). +`cltv_expiry_delta` is the minimum difference in HTLC CLTV timeouts in +the forwarding case (B) and `min_final_ctlv_expiry` is the minimum difference +between HTLC CLTV timeout and the current block height for the +terminal case (C). -This difference is important: after 2 days B can try to -remove the offer to C even if C is unresponsive, by broadcasting the -commitment transaction it has with C and spending the HTLC output. -Even though C might race to try to use its payment preimage at that point to -also spend the HTLC, it should be resolved well before the 3 day -deadline so B can either redeem the HTLC off A or close it. +Note that if this value is too low for a channel, the risk is only to +the node *accepting* the HTLC, not the node offering it. For this +reason, the `cltv_expiry_delta` for the *outgoing* channel is used as +the delta across a node. +We can derive the worst-case number of blocks between outgoing and +incoming HTLC resolution, given a few assumptions: -If the timing is too close, there is a risk of "one-sided redemption", -where the payment preimage received from an offered HTLC is too late -to be used for an incoming HTLC, leaving the node with unexpected -liability. +* A worst-case reorganization depth `R` blocks +* A grace-period `G` blocks after HTLC timeout before we give up on + an unresponsive peer and drop to chain. +* A number of blocks `S` between transaction broadcast and the + transaction being included in a block. +The worst case is for a forwarding node (B) which takes the longest +possible time to spot the outgoing HTLC fulfillment, and then takes +the longest possible time to redeem it on-chain: -Thus the effective timeout of the HTLC is the `cltv_expiry`, plus some -additional delay for the transaction which redeems the HTLC output to -be irreversibly committed to the blockchain. +1. The B->C HTLC times out at block `N`, and B waits `G` blocks until + it gives up waiting for C. B or C commits to the blockchain, + and B spends HTLC, which takes `S` blocks to be included. +2. Bad case: C wins the race (just) and fulfills the HTLC, B only sees + that transaction when it sees block `N+G+S+1`. +3. Worst case: There's reorganization `R` deep in which C wins and + fulfills. B only sees transaction at `N+G+S+R`. +4. B now needs to fulfill the incoming A->B HTLC, but A is unresponsive: B waits `G` more + blocks before giving up waiting for A. A or B commits to the blockchain. +5. Bad case: B sees A's commitment transaction in block `N+G+S+R+G+1`, and has + to spend the HTLC output, which takes `S` blocks to be mined. +6. Worst case: There's another reorganization `R` deep which A uses to + spend the commitment transaction, so B sees A's commitment + transaction in block `N+G+S+R+G+R`, and has to spend the HTLC output, which + takes `S` blocks to be mined. +7. B's HTLC spend needs to be at least `R` deep before it times out, + otherwise another reorganization could allow A to timeout the + transaction. + +Thus the worst case is `3R+2G+2S` assuming `R` is at least 1. Note that the +chances of three reorganizations in which the other node wins all of them is +low for `R` of 2 or more. Since we use high fees (and HTLC spends can use +almost arbitrary fees) `S` should be small, though given that block times are +irregular and empty blocks still occur, `S = 2` should be considered a +minimum. Similarly, the grace period `G` can be low (1 or 2), as nodes are +required to timeout or fulfill as soon as possible; but too low increases the +risk of unnecessary channel closure due to networking delays. + +There are four values we need to derive: + +1. The `cltv_expiry_delta` for channels. `3R+2G+2S`; if in doubt, a + `cltv_expiry_delta` of 12 is reasonable (R=2, G=1, S=2). + +2. For HTLCs we offer: the timeout deadline when we have to fail the channel + and time it out on-chain. This is `G` blocks after the HTLC + `cltv_expiry`; 1 block is reasonable. + +3. For HTLCs we accept and have a preimage: the fulfillment deadline when we + have to fail the channel and fulfill the HTLC onchain before its + `cltv_expiry`. This is steps 4-7 above, which means a deadline of `2R+G+S` + blocks before `cltv_expiry`; 7 blocks is reasonable. + +4. The minimum `cltv_expiry` we will accept for terminal payments: the + worst case for the terminal node C lower at `2R+G+S` blocks (steps + 1-3 above don't apply). The default in + [BOLT 11](11-payment-encoding.md) is 9, which is slightly more + conservative than the 6 this calculation suggests. -The fulfillment risk is similar: if a node C fulfills an HTLC after -its timeout, B might broadcast the commitment transaction and -immediately broadcast the HTLC timeout transaction. In this scenario, -B would gain knowledge of the preimage without paying C. #### Requirements -A node MUST estimate the deadline for successful redemption for each -HTLC. A node MUST NOT offer a HTLC after this deadline, and -MUST fail the channel if an HTLC which it offered is in either node's -current commitment transaction past this deadline. +A node MUST estimate a timeout deadline for each HTLC it offers. A node MUST +NOT offer an HTLC with a timeout deadline before its `cltv_expiry`, and MUST +fail the channel if an HTLC which it offered is in either node's current +commitment transaction past this timeout deadline. + +A node MUST estimate a fulfillment deadline for each HTLC it is attempting to +fulfill. A node MUST fail (and not forward) an HTLC whose fulfillment +deadline is already past, and MUST fail the connection if a HTLC it has +fulfilled is in either node's current commitment transaction past this +fulfillment deadline. -A node MUST NOT fulfill an HTLC after this deadline, and MUST fail the -connection if a HTLC it has fulfilled is in either node's current -commitment transaction past this deadline. ### Adding an HTLC: `update_add_htlc` @@ -743,7 +795,7 @@ using the `failure_code` given and setting the data to #### Rationale A node which doesn't time out HTLCs risks channel failure (see -"Risks With HTLC Timeouts"). +[`cltv_expiry_delta` Selection](#cltv_expiry_delta-selection). A node which sends `update_fulfill_htlc` before the sender is also committed to the HTLC risks losing funds. diff --git a/04-onion-routing.md b/04-onion-routing.md index b7f143a..dd35c1c 100644 --- a/04-onion-routing.md +++ b/04-onion-routing.md @@ -146,7 +146,7 @@ Field Description: * `outgoing_cltv_value` - The CLTV value that the _outgoing_ HTLC carrying the packet should have. - cltv_expiry - cltv_expiry_delta = outgoing_cltv_value + cltv_expiry - cltv_expiry_delta >= outgoing_cltv_value Inclusion of this field allows a node to both authenticate the information specified by the original sender and the parameters of the HTLC forwarded, @@ -557,7 +557,7 @@ outgoing channel: * [`2`:`len`] * [`len`:`channel_update`] -If the cltv-expiry is too near, we tell them the the current channel +If the `cltv_expiry` is too near, we tell them the the current channel setting for the outgoing channel: 1. type: UPDATE|14 (`expiry_too_soon`) diff --git a/07-routing-gossip.md b/07-routing-gossip.md index 51dca7d..c58790e 100644 --- a/07-routing-gossip.md +++ b/07-routing-gossip.md @@ -448,6 +448,96 @@ Other more advanced considerations involve diversity of routes to avoid single points of failure and detection, and channel balance of local channels. +### Routing Example + +Consider four nodes: + + +``` + B + / \ + / \ +A C + \ / + \ / + D +``` + +Each advertises the following `cltv_expiry_delta` on its end of every +channel: + +1. A: 10 blocks +2. B: 20 blocks +3. C: 30 blocks +4. D: 40 blocks + +C also uses a`min_final_cltv_expiry` of 9 (the default) when requesting +payments. + +Also, each node has the same fee scheme which it uses for each of its +channels: + +1. A: 100 base + 1000 millionths +1. B: 200 base + 2000 millionths +1. C: 300 base + 3000 millionths +1. D: 400 base + 4000 millionths + +The network will see eight `channel_update` messages: + +1. A->B: `cltv_expiry_delta` = 10, `fee_base_msat` = 100, `fee_proportional_millionths` = 1000 +1. A->D: `cltv_expiry_delta` = 10, `fee_base_msat` = 100, `fee_proportional_millionths` = 1000 +1. B->A: `cltv_expiry_delta` = 20, `fee_base_msat` = 200, `fee_proportional_millionths` = 2000 +1. D->A: `cltv_expiry_delta` = 40, `fee_base_msat` = 400, `fee_proportional_millionths` = 4000 +1. B->C: `cltv_expiry_delta` = 20, `fee_base_msat` = 200, `fee_proportional_millionths` = 2000 +1. D->C: `cltv_expiry_delta` = 40, `fee_base_msat` = 400, `fee_proportional_millionths` = 4000 +1. C->B: `cltv_expiry_delta` = 30, `fee_base_msat` = 300, `fee_proportional_millionths` = 3000 +1. C->D: `cltv_expiry_delta` = 30, `fee_base_msat` = 300, `fee_proportional_millionths` = 3000 + +If B were to send 4,999,999 millisatoshi directly to C, it wouldn't +charge itself a fee nor add its own `cltv_expiry_delta`, so it would +use C's requested `min_final_cltv_expiry` of 9. We also assume it adds a +"shadow route" to give an extra CLTV of 42. It could also add extra +cltv deltas at other hops, as these values are a minimum, but we don't +here for simplicity: + + * `amount_msat`: 4999999 + * `cltv_expiry`: current-block-height + 9 + 42 + * `onion_routing_packet`: + * `amt_to_forward` = 4999999 + * `outgoing_cltv_value` = current-block-height + 9 + 42 + +If A were to send an 4,999,999 millisatoshi to C via B, it needs to +pay B the fee it specified in the B->C `channel_update`, calculated as +per [HTLC Fees](#htlc_fees): + + 200 + 4999999 * 2000 / 1000000 = 10199 + +Similarly, it would need to add the `cltv_expiry` from B->C's +`channel_update` (20), plus C's requested `min_final_cltv_expiry` (9), plus 42 for the +"shadow route". Thus the `update_add_htlc` message from A to B would +be: + + * `amount_msat`: 5010198 + * `cltv_expiry`: current-block-height + 20 + 9 + 42 + * `onion_routing_packet`: + * `amt_to_forward` = 4999999 + * `outgoing_cltv_value` = current-block-height + 9 + 42 + +The `update_add_htlc` from B to C would be the same as the B->C direct +payment above. + +Finally, if for some reason A chose the more expensive route via D, it +would send the following `update_add_htlc` to D: + + * `amount_msat`: 5020398 + * `cltv_expiry`: current-block-height + 40 + 9 + 42 + * `onion_routing_packet`: + * `amt_to_forward` = 4999999 + * `outgoing_cltv_value` = current-block-height + 9 + 42 + +And the `update_add_htlc` from D to C would be the same as the B->C +direct payment again. + ## References ![Creative Commons License](https://i.creativecommons.org/l/by/4.0/88x31.png "License CC-BY")