Inclusion proofs, error instead of panic

[vmx]

Description

Currently if we read in invalid data there is a panic:

rust-fil-proofs/filecoin-hashers/src/poseidon.rs

Line 345 in ccfea67

panic_any(format!("from_repr failure at {}", i));

. The problem is that it's hard to track down where the bad data actually is. Ideally you'd know at least which sector contains invalid data (see #1604 for an example of this issue).

If an error would be returned (instead of a panic), one could then catch that error here

rust-fil-proofs/storage-proofs-post/src/fallback/vanilla.rs

Line 638 in ccfea67

if inclusion_proof.root() != comm_r_last {

and return an error with the sector ID.

Acceptance criteria

The sector ID is somehow displayed if there is invalid data.

Risks + pitfalls

This also needs changes in the merkletree crate and I'm not sure how big the overall changes will be. So it might not be worth it.

Where to begin

Make

rust-fil-proofs/filecoin-hashers/src/poseidon.rs

Lines 335 to 354 in ccfea67

	fn multi_node(&mut self, parts: &[PoseidonDomain], _height: usize) -> PoseidonDomain {
	match parts.len() {
	1 | 2 | 4 | 8 | 16 => shared_hash_frs(
	&parts
	.iter()
	.enumerate()
	.map(|(i, x)| {
	if let Some(fr) = Fr::from_repr_vartime(x.0) {
	fr
	} else {
	panic_any(format!("from_repr failure at {}", i));
	}
	})
	.collect::<Vec<_>>(),
	)
	.into(),
	arity => panic_any(format!("unsupported arity {}", arity)),
	}
	}
	}

return Results instead of panicking and see which other changes are needed due to this.

[llifezou]

this problem may be related to special challange []uint64.

func GenerateSingleVanillaProof(
	replica PrivateSectorInfo,
	challange []uint64,
) ([]byte, error)

[storojs72]

The problem is that it's hard to track down where the bad data actually is

That's true.

Not sure if it does relate to merkletree crate... It looks like implementation of from_repr_vartime is located in blstrs crate, isn't it? And information about sector seems to be out of scope for it.

Can't it be resolved by adding information about sector into calling code where panic is invoked? I mean in such case we need to change caller that use Poseidon hasher by adding something like that:

...
trace!("sector is {}", sector_num);
multi_node( ... );
...

and of course, Rust logging should be enabled, as it looks like it is invoked from Go high-level code.

[vmx]

Not sure if it does relate to merkletree crate... I

If Fr::from_repr_vartime(x.0) returns None, we could error instead of panic. But then multi_node() would need to return a Result. And that would mean that the LightAlgorithm/Algorithm trait would need to be changed, which lives in merkletree.