packet-forward-middleware

packet-forward-middleware

Middleware for forwarding IBC packets.

Asynchronous acknowledgements are utilized for atomic multi-hop packet flows. The acknowledgement will only be written on the chain where the user initiated the packet flow after the forward/multi-hop sequence has completed (success or failure). This means that a user (i.e. an IBC application) only needs to monitor the chain where the initial transfer was sent for the response of the entire process.

About

The packet-forward-middleware is an IBC middleware module built for Cosmos blockchains utilizing the IBC protocol. A chain which incorporates the packet-forward-middleware is able to route incoming IBC packets from a source chain to a destination chain. As the Cosmos SDK/IBC become commonplace in the blockchain space more and more zones will come online, these new zones joining are noticing a problem: they need to maintain a large amount of infrastructure (archive nodes and relayers for each counterparty chain) to connect with all the chains in the ecosystem, a number that is continuing to increase quickly. Luckily this problem has been anticipated and IBC has been architected to accommodate multi-hop transactions. However, a packet forwarding/routing feature was not in the initial IBC release.

Sequence diagrams

Let's stipulate the following connections between chains A, B, C, and D

flowchart LR
    A((Chain A))
    B((Chain B))
    C((Chain C))
    D((Chain D))

    A <--"ch-0 ch-1 (IBC)"--> B
    B <--"ch-2 ch-3 (IBC)"--> C
    C <--"ch-4 ch-5 (IBC)"--> D

Loading

SCENARIO: Via PFM, Chain A wants to pass a message to Chain D (to which it's not directly connected).

sequenceDiagram
    autonumber
    Chain A ->> Chain B: PFM transfer
    Chain B --> Chain B: recv_packet
    Chain B ->> Chain C: forward
    Chain C --> Chain C: recv_packet
    Chain C ->> Chain D: forward
    Chain D --> Chain D: recv_packet
    Chain D ->> Chain C: ack
    Chain C ->> Chain B: ack
    Chain B ->> Chain A: ack

Loading

SCENARIO: Multi-hop A->B->C->D, C->D recv_packet error, refund back to A

sequenceDiagram
    autonumber
    Chain A ->> Chain B: PFM transfer
    Chain B --> Chain B: recv_packet
    Chain B ->> Chain C: forward
    Chain C --> Chain C: recv_packet
    Chain C ->> Chain D: forward
    Chain D --> Chain D: ☠️ recv_packet ERR ☠️
    Chain D ->> Chain C: ☠️ ack ERR ☠️
    Chain C ->> Chain B: ☠️ ack ERR ☠️
    Chain B ->> Chain A: ☠️ ack ERR ☠️

Loading

SCENARIO: Forward A->B->C with 1 retry, max timeouts occurs, refund back to A

sequenceDiagram
    autonumber
    Chain A ->> Chain B: PFM transfer
    Chain B --> Chain B: recv_packet
    Chain B ->> Chain C: forward
    Chain C --x Chain B: timeout
    Chain B ->> Chain C: forward retry
    Chain C --x Chain B: timeout
    Chain B ->> Chain A: ☠️ ack ERR ☠️

Loading

Examples

Utilizing the packet memo field, instructions can be encoded as JSON for multi-hop sequences.

Minimal Example - Chain forward A->B->C

• The packet-forward-middleware integrated on Chain B.
• The packet data receiver for the MsgTransfer on Chain A is set to "pfm" or some other invalid bech32 string.*
• The packet memo is included in MsgTransfer by user on Chain A.

memo:

{
  "forward": {
    "receiver": "chain-c-bech32-address",
    "port": "transfer",
    "channel": "channel-123"
  }
}

Full Example - Chain forward A->B->C->D with retry on timeout

• The packet-forward-middleware integrated on Chain B and Chain C.
• The packet data receiver for the MsgTransfer on Chain A is set to "pfm" or some other invalid bech32 string.*
• The forward metadata receiver for the hop from Chain B to Chain C is set to "pfm" or some other invalid bech32 string.*
• The packet memo is included in MsgTransfer by user on Chain A.
• A packet timeout of 10 minutes and 2 retries is set for both forwards.

In the case of a timeout after 10 minutes for either forward, the packet would be retried up to 2 times, at which case an error ack would be written to issue a refund on the prior chain.

next is the memo to pass for the next transfer hop. Per memo intended usage of a JSON string, it should be either JSON which will be Marshaled retaining key order, or an escaped JSON string which will be passed directly.

next as JSON

{
  "forward": {
    "receiver": "pfm", // purposely using invalid bech32 here*
    "port": "transfer",
    "channel": "channel-123",
    "timeout": "10m",
    "retries": 2,
    "next": {
      "forward": {
        "receiver": "chain-d-bech32-address",
        "port": "transfer",
        "channel":"channel-234",
        "timeout":"10m",
        "retries": 2
      }
    }
  }
}

next as escaped JSON string

{
  "forward": {
    "receiver": "pfm", // purposely using invalid bech32 here*
    "port": "transfer",
    "channel": "channel-123",
    "timeout": "10m",
    "retries": 2,
    "next": "{\"forward\":{\"receiver\":\"chain-d-bech32-address\",\"port\":\"transfer\",\"channel\":\"channel-234\",\"timeout\":\"10m\",\"retries\":2}}"
  }
}

Intermediate Receivers*

PFM does not need the packet data receiver address to be valid, as it will create a hash of the sender and channel to derive a receiver address on the intermediate chains. This is done for security purposes to ensure that users cannot move funds through arbitrary accounts on intermediate chains.

To prevent accidentally sending funds to a chain which does not have PFM, it is recommended to use an invalid bech32 string (such as "pfm") for the receiver on intermediate chains. By using an invalid bech32 string, a transfer that is accidentally sent to a chain that does not have PFM would fail to be received, and properly refunded to the user on the source chain, rather than having funds get stuck on the intermediate chain.

The examples above show the intended usage of the receiver field for one or multiple intermediate PFM chains.

Implementation details

Flow sequence mainly encoded in middleware and in keeper.

Describes A sending to C via B in several scenarios with operational opened channels, enabled denom composition, fees and available to refund, but no retries.

Generally without memo to handle, all handling by this module is delegated to ICS-020. ICS-020 ACK are written and parsed in any case (ACK are backwarded).

A -> B -> C full success

1. A This sends packet over underlying ICS-004 wrapper with memo as is.
2. B This receives packet and parses it into ICS-020 packet.
3. B Validates forward packet on this step, return ACK error if fails.
4. B If other middleware not yet called ICS-020, call it and ACK error on fail. Tokens minted or unescrowed here.
5. B Handle denom. If denom prefix is from B, remove it. If denom prefix is other chain - add B prefix.
6. B Take fee, create new ICS-004 packet with timeout from forward for next step, and remaining inner memo.
7. B Send transfer to C with parameters obtained from memo. Tokens burnt or escrowed here.
8. B Store tracking in flight packet under next (channel, port, ICS-20 transfer sequence), do not ACK packet yet.
9. C Handle ICS-020 packet as usual.
10. B On ICS-020 ACK from C find in flight packet, delete it and write ACK for original packet from A.
11. A Handle ICS-020 ACK as usual

Example of USDC transfer from Osmosis -> Noble -> Sei

A -> B -> C with C error ACK

10. B On ICS-020 ACK from C find in flight packet, delete it
11. B Burns or escrows tokens.
12. B And write error ACK for original packet from A.
13. A Handle ICS-020 timeout as usual
14. C writes success ACK for packet from B

Same behavior in case of timeout on C

A packet timeouts on B before C timeouts packet from B

10. A Cannot timeout because in flight packet has proof on B of packet inclusion.
11. B waits for ACK or timeout from C.
12. B timeout from C becomes fail ACK on B for A
13. A receives success or fail ACK, but not timeout

In this case A assets hang until final hop timeouts or ACK.

References

• https://www.mintscan.io/cosmos/proposals/56
• PFM was originally implemented in cosmos/ibc-go#373