Fixes prover behavior with singleton proof trees (#859)

* add logs and test

* add Merkle proof checks

* factor out Circom input normalization, fix proof input serialization

* add test and update existing ones

* update circuit assets

* add back trace message

* switch contracts to fix branch

* update codex-contracts-eth to latest

* do not expose prove with prenormalized inputs
This commit is contained in:
Giuliano Mega 2024-07-18 10:25:06 -03:00 committed by GitHub
parent 8f740b42e6
commit fbce240e3d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
11 changed files with 111 additions and 87 deletions

View File

@ -9,17 +9,14 @@
{.push raises: [].}
import std/sequtils
import std/sugar
import pkg/chronos
import pkg/questionable/results
import pkg/circomcompat
import pkg/poseidon2/io
import ../../types
import ../../../stores
import ../../../merkletree
import ../../../codextypes
import ../../../contracts
import ./converters
@ -39,6 +36,41 @@ type
backendCfg : ptr CircomBn254Cfg
vkp* : ptr CircomKey
NormalizedProofInputs*[H] {.borrow: `.`.} = distinct ProofInputs[H]
func normalizeInput*[H](self: CircomCompat, input: ProofInputs[H]):
NormalizedProofInputs[H] =
## Parameters in CIRCOM circuits are statically sized and must be properly
## padded before they can be passed onto the circuit. This function takes
## variable length parameters and performs that padding.
##
## The output from this function can be JSON-serialized and used as direct
## inputs to the CIRCOM circuit for testing and debugging when one wishes
## to bypass the Rust FFI.
let normSamples = collect:
for sample in input.samples:
var merklePaths = sample.merklePaths
merklePaths.setLen(self.slotDepth)
Sample[H](
cellData: sample.cellData,
merklePaths: merklePaths
)
var normSlotProof = input.slotProof
normSlotProof.setLen(self.datasetDepth)
NormalizedProofInputs[H] ProofInputs[H](
entropy: input.entropy,
datasetRoot: input.datasetRoot,
slotIndex: input.slotIndex,
slotRoot: input.slotRoot,
nCellsPerSlot: input.nCellsPerSlot,
nSlotsPerDataSet: input.nSlotsPerDataSet,
slotProof: normSlotProof,
samples: normSamples
)
proc release*(self: CircomCompat) =
## Release the ctx
##
@ -49,27 +81,20 @@ proc release*(self: CircomCompat) =
if not isNil(self.vkp):
self.vkp.unsafeAddr.release_key()
proc prove*[H](
proc prove[H](
self: CircomCompat,
input: ProofInputs[H]): ?!CircomProof =
## Encode buffers using a ctx
##
input: NormalizedProofInputs[H]): ?!CircomProof =
# NOTE: All inputs are statically sized per circuit
# and adjusted accordingly right before being passed
# to the circom ffi - `setLen` is used to adjust the
# sequence length to the correct size which also 0 pads
# to the correct length
doAssert input.samples.len == self.numSamples,
"Number of samples does not match"
doAssert input.slotProof.len <= self.datasetDepth,
"Number of slot proofs does not match"
"Slot proof is too deep - dataset has more slots than what we can handle?"
doAssert input.samples.allIt(
block:
(it.merklePaths.len <= self.slotDepth + self.blkDepth and
it.cellData.len <= self.cellElms * 32)), "Merkle paths length does not match"
it.cellData.len == self.cellElms)), "Merkle paths too deep or cells too big for circuit"
# TODO: All parameters should match circom's static parametter
var
@ -116,8 +141,7 @@ proc prove*[H](
var
slotProof = input.slotProof.mapIt( it.toBytes ).concat
slotProof.setLen(self.datasetDepth) # zero pad inputs to correct size
doAssert(slotProof.len == self.datasetDepth)
# arrays are always flattened
if ctx.pushInputU256Array(
"slotProof".cstring,
@ -128,16 +152,14 @@ proc prove*[H](
for s in input.samples:
var
merklePaths = s.merklePaths.mapIt( it.toBytes )
data = s.cellData
data = s.cellData.mapIt( @(it.toBytes) ).concat
merklePaths.setLen(self.slotDepth) # zero pad inputs to correct size
if ctx.pushInputU256Array(
"merklePaths".cstring,
merklePaths[0].addr,
uint (merklePaths[0].len * merklePaths.len)) != ERR_OK:
return failure("Failed to push merkle paths")
data.setLen(self.cellElms * 32) # zero pad inputs to correct size
if ctx.pushInputU256Array(
"cellData".cstring,
data[0].addr,
@ -162,6 +184,12 @@ proc prove*[H](
success proof
proc prove*[H](
self: CircomCompat,
input: ProofInputs[H]): ?!CircomProof =
self.prove(self.normalizeInput(input))
proc verify*[H](
self: CircomCompat,
proof: CircomProof,

View File

@ -38,7 +38,7 @@ type
func getCell*[T, H](
self: DataSampler[T, H],
blkBytes: seq[byte],
blkCellIdx: Natural): seq[byte] =
blkCellIdx: Natural): seq[H] =
let
cellSize = self.builder.cellSize.uint64
@ -47,7 +47,7 @@ func getCell*[T, H](
doAssert (dataEnd - dataStart) == cellSize, "Invalid cell size"
toInputData[H](blkBytes[dataStart ..< dataEnd])
blkBytes[dataStart ..< dataEnd].elements(H).toSeq()
proc getSample*[T, H](
self: DataSampler[T, H],

View File

@ -7,23 +7,13 @@
## This file may not be copied, modified, or distributed except according to
## those terms.
import std/sugar
import std/bitops
import std/sequtils
import pkg/questionable/results
import pkg/poseidon2
import pkg/poseidon2/io
import pkg/constantine/math/arithmetic
import pkg/constantine/math/io/io_fields
import ../../merkletree
func toInputData*[H](data: seq[byte]): seq[byte] =
return toSeq(data.elements(H)).mapIt( @(it.toBytes) ).concat
func extractLowBits*[n: static int](elm: BigInt[n], k: int): uint64 =
doAssert( k > 0 and k <= 64 )
var r = 0'u64
@ -39,6 +29,7 @@ func extractLowBits(fld: Poseidon2Hash, k: int): uint64 =
return extractLowBits(elm, k);
func floorLog2*(x : int) : int =
doAssert ( x > 0 )
var k = -1
var y = x
while (y > 0):
@ -47,9 +38,7 @@ func floorLog2*(x : int) : int =
return k
func ceilingLog2*(x : int) : int =
if (x == 0):
return -1
else:
doAssert ( x > 0 )
return (floorLog2(x - 1) + 1)
func toBlkInSlot*(cell: Natural, numCells: Natural): Natural =
@ -80,7 +69,7 @@ func cellIndices*(
numCells: Natural, nSamples: Natural): seq[Natural] =
var indices: seq[Natural]
while (indices.len < nSamples):
let idx = cellIndex(entropy, slotRoot, numCells, indices.len + 1)
indices.add(idx.Natural)
for i in 1..nSamples:
indices.add(cellIndex(entropy, slotRoot, numCells, i))
indices

View File

@ -9,7 +9,7 @@
type
Sample*[H] = object
cellData*: seq[byte]
cellData*: seq[H]
merklePaths*: seq[H]
PublicInputs*[H] = object
@ -24,5 +24,5 @@ type
slotRoot*: H
nCellsPerSlot*: Natural
nSlotsPerDataSet*: Natural
slotProof*: seq[H]
samples*: seq[Sample[H]]
slotProof*: seq[H] # inclusion proof that shows that the slot root (leaf) is part of the dataset (root)
samples*: seq[Sample[H]] # inclusion proofs which show that the selected cells (leafs) are part of the slot (roots)

View File

@ -17,21 +17,6 @@ import pkg/codex/utils/json
export types
func fromCircomData*(_: type Poseidon2Hash, cellData: seq[byte]): seq[Poseidon2Hash] =
var
pos = 0
cellElms: seq[Bn254Fr]
while pos < cellData.len:
var
step = 32
offset = min(pos + step, cellData.len)
data = cellData[pos..<offset]
let ff = Bn254Fr.fromBytes(data.toArray32).get
cellElms.add(ff)
pos += data.len
cellElms
func toJsonDecimal*(big: BigInt[254]): string =
let s = big.toDecimal.strip( leading = true, trailing = false, chars = {'0'} )
if s.len == 0: "0" else: s
@ -78,13 +63,16 @@ func toJson*(input: ProofInputs[Poseidon2Hash]): JsonNode =
"slotRoot": input.slotRoot.toDecimal,
"slotProof": input.slotProof.mapIt( it.toBig.toJsonDecimal ),
"cellData": input.samples.mapIt(
toSeq( it.cellData.elements(Poseidon2Hash) ).mapIt( it.toBig.toJsonDecimal )
it.cellData.mapIt( it.toBig.toJsonDecimal )
),
"merklePaths": input.samples.mapIt(
it.merklePaths.mapIt( it.toBig.toJsonDecimal )
)
}
func toJson*(input: NormalizedProofInputs[Poseidon2Hash]): JsonNode =
toJson(ProofInputs[Poseidon2Hash](input))
func jsonToProofInput*(_: type Poseidon2Hash, inputJson: JsonNode): ProofInputs[Poseidon2Hash] =
let
cellData =
@ -93,10 +81,12 @@ func jsonToProofInput*(_: type Poseidon2Hash, inputJson: JsonNode): ProofInputs[
block:
var
big: BigInt[256]
data = newSeq[byte](big.bits div 8)
hash: Poseidon2Hash
data: array[32, byte]
assert bool(big.fromDecimal( it.str ))
data.marshal(big, littleEndian)
data
assert data.marshal(big, littleEndian)
Poseidon2Hash.fromBytes(data).get
).concat # flatten out elements
)

View File

@ -58,7 +58,7 @@ suite "Test Sampler - control samples":
proofInput.nCellsPerSlot,
sample.merklePaths[5..<9]).tryGet
cellData = Poseidon2Hash.fromCircomData(sample.cellData)
cellData = sample.cellData
cellLeaf = Poseidon2Hash.spongeDigest(cellData, rate = 2).tryGet
slotLeaf = cellProof.reconstructRoot(cellLeaf).tryGet
@ -158,7 +158,7 @@ suite "Test Sampler":
nSlotCells,
sample.merklePaths[5..<sample.merklePaths.len]).tryGet
cellData = Poseidon2Hash.fromCircomData(sample.cellData)
cellData = sample.cellData
cellLeaf = Poseidon2Hash.spongeDigest(cellData, rate = 2).tryGet
slotLeaf = cellProof.reconstructRoot(cellLeaf).tryGet

View File

@ -24,23 +24,19 @@ import ./backends/helpers
suite "Test Prover":
let
slotId = 1
samples = 5
ecK = 3
ecM = 2
numDatasetBlocks = 8
blockSize = DefaultBlockSize
cellSize = DefaultCellSize
repoTmp = TempLevelDb.new()
metaTmp = TempLevelDb.new()
r1cs = "tests/circuits/fixtures/proof_main.r1cs"
wasm = "tests/circuits/fixtures/proof_main.wasm"
circomBackend = CircomCompat.init(r1cs, wasm)
challenge = 1234567.toF.toBytes.toArray32
var
datasetBlocks: seq[bt.Block]
store: BlockStore
manifest: Manifest
protected: Manifest
verifiable: Manifest
sampler: Poseidon2Sampler
prover: Prover
setup:
let
@ -48,14 +44,7 @@ suite "Test Prover":
metaDs = metaTmp.newDb()
store = RepoStore.new(repoDs, metaDs)
(manifest, protected, verifiable) =
await createVerifiableManifest(
store,
numDatasetBlocks,
ecK, ecM,
blockSize,
cellSize)
prover = Prover.new(store, circomBackend, samples)
teardown:
await repoTmp.destroyDb()
@ -63,13 +52,41 @@ suite "Test Prover":
test "Should sample and prove a slot":
let
r1cs = "tests/circuits/fixtures/proof_main.r1cs"
wasm = "tests/circuits/fixtures/proof_main.wasm"
(_, _, verifiable) =
await createVerifiableManifest(
store,
8, # number of blocks in the original dataset (before EC)
5, # ecK
3, # ecM
blockSize,
cellSize)
circomBackend = CircomCompat.init(r1cs, wasm)
prover = Prover.new(store, circomBackend, samples)
challenge = 1234567.toF.toBytes.toArray32
(inputs, proof) = (await prover.prove(1, verifiable, challenge)).tryGet
let
(inputs, proof) = (
await prover.prove(1, verifiable, challenge)).tryGet
check:
(await prover.verify(proof, inputs)).tryGet == true
test "Should generate valid proofs when slots consist of single blocks":
# To get single-block slots, we just need to set the number of blocks in
# the original dataset to be the same as ecK. The total number of blocks
# after generating random data for parity will be ecK + ecM, which will
# match the number of slots.
let
(_, _, verifiable) =
await createVerifiableManifest(
store,
2, # number of blocks in the original dataset (before EC)
2, # ecK
1, # ecM
blockSize,
cellSize)
let
(inputs, proof) = (
await prover.prove(1, verifiable, challenge)).tryGet
check:
(await prover.verify(proof, inputs)).tryGet == true

@ -1 +1 @@
Subproject commit 57e8cd5013325f05e16833a5320b575d32a403f3
Subproject commit 7ad26688a3b75b914d626e2623174a36f4425f51