This helps identify where the various functionalities are used, or rather, not
used. The `Ciphersuite` trait present in `patches/ciphersuite`, facilitating
the entire FCMP++ tree, only requires the markers _and_ canonical point
decoding. I've opened a PR to upstream such a trait into `group`
(https://github.com/zkcrypto/group/pull/68).
`WrappedGroup` is still justified for as long as `Group::generator` exists.
Moving `::generator()` to its own trait, on an independent structure (upstream)
would be massively appreciated. @tarcieri also wanted to update from
`fn generator()` to `const GENERATOR`, which would encourage further discussion
on https://github.com/zkcrypto/group/issues/32 and
https://github.com/zkcrypto/group/issues/45, which have been stagnant.
The `Id` trait is occasionally used yet really should be first off the chopping
block.
Finally, `WithPreferredHash` is only actually used around a third of the time,
which more than justifies it being a separate trait.
---
Updates `dalek_ff_group::Scalar` to directly re-export
`curve25519_dalek::Scalar`, as without issue. `dalek_ff_group::RistrettoPoint`
also could be replaced with an export of `curve25519_dalek::RistrettoPoint`,
yet the coordinator relies on how we implemented `Hash` on it for the hell of
it so it isn't worth it at this time. `dalek_ff_group::EdwardsPoint` can't be
replaced for an re-export of `curve25519_dalek::SubgroupPoint` as it doesn't
implement `zeroize`, `subtle` traits within a released, non-yanked version.
Relevance to https://github.com/serai-dex/serai/issues/201 and
https://github.com/dalek-cryptography/curve25519-dalek/issues/811#issuecomment-3247732746.
Also updates the `Ristretto` ciphersuite to prefer `Blake2b-512` over
`SHA2-512`. In order to maintain compliance with FROST's IETF standard,
`modular-frost` defines its own ciphersuite for Ristretto which still uses
`SHA2-512`.
The prior-present `Ciphersuite::hash_to_F` was a sin. Implementations took a
DST, yet were not require to securely handle it. It was also biased towards the
requirements of `modular-frost` as `ciphersuite` was originally written all
those years ago, when `modular-frost` had needs exceeding what `ff`, `group`
satisfied.
Now, the hash is bound to produce an output which can be converted to a scalar
with `ff::FromUniformBytes`. A new `hash_to_F`, which accepts a single argument
of the value to hash (removing the potential to insecurely handle the DST by
removing the DST entirely). Due to `digest` yielding a `GenericArray`, yet
`FromUniformBytes` taking a `const usize`, the `ciphersuite` crate now defines
a `FromUniformBytes` trait taking an array (then implemented for all satisfiers
of `ff::FromUniformBytes`). In order to get the array type from the
`GenericArray`, the output of the hash, `digest` is updated to the `0.11`
release candidate which moves to `flexible-array` which solves that problem.
The existing, specific `hash_to_F` functions have been moved to `modular-frost`
as necessary.
`flexible-array` itself is patched to a fork due to
https://github.com/RustCrypto/hybrid-array/issues/131.
Fixes `no-std` builds for packages which intended to be `no-std` (without
`alloc`).
Updates a variety of MSRVs to 1.73 due to `flexible-transcript` no longer using
`std-shims` to achieve 1.66 (as `std-shims` requires `alloc`). A future
improvement would be for `std-shims` to have an `alloc` feature and only
provide MSRV shims without it.
The prior workflow (now deleted) required manually specifying the packages to
check and only checked the package could compile under the stated MSRV. It
didn't verify it was actually the _minimum_ supported Rust version. The new
version finds the MSRV from scratch to check if the stated MSRV aligns.
Updates stated MSRVs accordingly.
Also removes many explicit dependencies from secq256k1 for their re-exports via
k256. Not directly relevant, just part of tidying up all the `toml`s.
monero-oxide relies on ciphersuite, which is in-tree, yet we've made breaking
changes since. This commit adds a patch so
monero-oxide -> patches/ciphersuite -> crypto/ciphersuite, with
patches/ciphersuite resolving the breaking changes.
This resolves the conflicts and gets the workspace `Cargo.toml`s to not be
invalid. It doesn't actually get clippy to pass again yet.
Does move `crypto/dkg/src/evrf` into a new `crypto/dkg/evrf` crate (which does
not yet compile).
flexible-transcript already had a shim to support <1.66. This was irrelevant
since flexible-transcript had a MSRV of 1.73. Due to how clunky it was, it has
been removed despite theoretically enabling an even lower MSRV.
Unused and unpublished. This was only added in the FCMP++ branch as a quick fix
for performance reasons. Finding a better API is still a tricky question, but
this API is _bad_.
promote and pedpop require dleq, which don't support no-std. All three should
be moved outside the Serai repository, per #597, as none are planned for use
and worth covering under our BBP.
* WIP constant-time implementation of the ec-divisors library
* Fix misc logic errors in poly.rs
* Remove accidentally committed test statements
* Fix ConstantTimeEq for CoefficientIndex
* Correct the iterations formula
x**3 / (0 y + x**1) would prior be considered indivisible with iterations = 0.
It is divisible however. The amount of iterations should be the amount of
coefficients within the numerator *excluding the coefficient for y**0 x**0*.
* Poly PartialEq, conditional_select_poly which checks poly structure equivalence
If the first passed argument is smaller than the latter, it's padded to the
necessary length.
Also adds code to trim the remainder as the remainder is the value modulo, so
it's very important it remains concise and workable.
* Fix the line function
It selected the case if both were identity before selecting the case if either
were identity, the latter overwriting the former.
* Final fixes re: ct_get
1) Our quotient structure does need to be of size equal to the numerator
entirely to prevent out-of-bounds reads on it
2) We need to get from yx_coefficients if of length >=, so if the length is 1
we can read y_pow=1 from it. If y_pow=0, and its length is 0 so it has no
inner Vecs, we need to fall back with the guard y_pow != 0.
* Add a trim algorithm to lib.rs to prevent Polys from becoming unbearably gigantic
Our Poly algorithm is incredibly leaky. While it presumably should be improved,
we can take advantage of our known structure while constructing divisors (and
the small modulus) to simply trim out the zero coefficients leaked. This
maintains Polys in a manageable size.
* Move constant-time scalar mul gadget divisor creation from dkg to ec-divisors
Anyone creating a divisor for the scalar mul gadget should use constant time
code, so this code should at least be in the EC gadgets crate It's of
non-trivial complexity to deal with otherwise.
* Remove unsafe, cache timing attacks from ec-divisors
* Upstream GBP, divisor, circuit abstraction, and EC gadgets from FCMP++
* Initial eVRF implementation
Not quite done yet. It needs to communicate the resulting points and proofs to
extract them from the Pedersen Commitments in order to return those, and then
be tested.
* Add the openings of the PCs to the eVRF as necessary
* Add implementation of secq256k1
* Make DKG Encryption a bit more flexible
No longer requires the use of an EncryptionKeyMessage, and allows pre-defined
keys for encryption.
* Make NUM_BITS an argument for the field macro
* Have the eVRF take a Zeroizing private key
* Initial eVRF-based DKG
* Add embedwards25519 curve
* Inline the eVRF into the DKG library
Due to how we're handling share encryption, we'd either need two circuits or to
dedicate this circuit to the DKG. The latter makes sense at this time.
* Add documentation to the eVRF-based DKG
* Add paragraph claiming robustness
* Update to the new eVRF proof
* Finish routing the eVRF functionality
Still needs errors and serialization, along with a few other TODOs.
* Add initial eVRF DKG test
* Improve eVRF DKG
Updates how we calculcate verification shares, improves performance when
extracting multiple sets of keys, and adds more to the test for it.
* Start using a proper error for the eVRF DKG
* Resolve various TODOs
Supports recovering multiple key shares from the eVRF DKG.
Inlines two loops to save 2**16 iterations.
Adds support for creating a constant time representation of scalars < NUM_BITS.
* Ban zero ECDH keys, document non-zero requirements
* Implement eVRF traits, all the way up to the DKG, for secp256k1/ed25519
* Add Ristretto eVRF trait impls
* Support participating multiple times in the eVRF DKG
* Only participate once per key, not once per key share
* Rewrite processor key-gen around the eVRF DKG
Still a WIP.
* Finish routing the new key gen in the processor
Doesn't touch the tests, coordinator, nor Substrate yet.
`cargo +nightly fmt && cargo +nightly-2024-07-01 clippy --all-features -p serai-processor`
does pass.
* Deduplicate and better document in processor key_gen
* Update serai-processor tests to the new key gen
* Correct amount of yx coefficients, get processor key gen test to pass
* Add embedded elliptic curve keys to Substrate
* Update processor key gen tests to the eVRF DKG
* Have set_keys take signature_participants, not removed_participants
Now no one is removed from the DKG. Only `t` people publish the key however.
Uses a BitVec for an efficient encoding of the participants.
* Update the coordinator binary for the new DKG
This does not yet update any tests.
* Add sensible Debug to key_gen::[Processor, Coordinator]Message
* Have the DKG explicitly declare how to interpolate its shares
Removes the hack for MuSig where we multiply keys by the inverse of their
lagrange interpolation factor.
* Replace Interpolation::None with Interpolation::Constant
Allows the MuSig DKG to keep the secret share as the original private key,
enabling deriving FROST nonces consistently regardless of the MuSig context.
* Get coordinator tests to pass
* Update spec to the new DKG
* Get clippy to pass across the repo
* cargo machete
* Add an extra sleep to ensure expected ordering of `Participation`s
* Update orchestration
* Remove bad panic in coordinator
It expected ConfirmationShare to be n-of-n, not t-of-n.
* Improve documentation on functions
* Update TX size limit
We now no longer have to support the ridiculous case of having 49 DKG
participations within a 101-of-150 DKG. It does remain quite high due to
needing to _sign_ so many times. It'd may be optimal for parties with multiple
key shares to independently send their preprocesses/shares (despite the
overhead that'll cause with signatures and the transaction structure).
* Correct error in the Processor spec document
* Update a few comments in the validator-sets pallet
* Send/Recv Participation one at a time
Sending all, then attempting to receive all in an expected order, wasn't working
even with notable delays between sending messages. This points to the mempool
not working as expected...
* Correct ThresholdKeys serialization in modular-frost test
* Updating existing TX size limit test for the new DKG parameters
* Increase time allowed for the DKG on the GH CI
* Correct construction of signature_participants in serai-client tests
Fault identified by akil.
* Further contextualize DkgConfirmer by ValidatorSet
Caught by a safety check we wouldn't reuse preprocesses across messages. That
raises the question of we were prior reusing preprocesses (reusing keys)?
Except that'd have caused a variety of signing failures (suggesting we had some
staggered timing avoiding it in practice but yes, this was possible in theory).
* Add necessary calls to set_embedded_elliptic_curve_key in coordinator set rotation tests
* Correct shimmed setting of a secq256k1 key
* cargo fmt
* Don't use `[0; 32]` for the embedded keys in the coordinator rotation test
The key_gen function expects the random values already decided.
* Big-endian secq256k1 scalars
Also restores the prior, safer, Encryption::register function.
* Move logic for evaluating if a cosign should occur to its own file
Cleans it up and makes it more robust.
* Have expected_next_batch return an error instead of retrying
While convenient to offer an error-free implementation, it potentially caused
very long lived lock acquisitions in handle_processor_message.
* Unify and clean DkgConfirmer and DkgRemoval
Does so via adding a new file for the common code, SigningProtocol.
Modifies from_cache to return the preprocess with the machine, as there's no
reason not to. Also removes an unused Result around the type.
Clarifies the security around deterministic nonces, removing them for
saved-to-disk cached preprocesses. The cached preprocesses are encrypted as the
DB is not a proper secret store.
Moves arguments always present in the protocol from function arguments into the
struct itself.
Removes the horribly ugly code in DkgRemoval, fixing multiple issues present
with it which would cause it to fail on use.
* Set SeraiBlockNumber in cosign.rs as it's used by the cosigning protocol
* Remove unnecessary Clone from lambdas in coordinator
* Remove the EventDb from Tributary scanner
We used per-Transaction DB TXNs so on error, we don't have to rescan the entire
block yet only the rest of it. We prevented scanning multiple transactions by
tracking which we already had.
This is over-engineered and not worth it.
* Implement borsh for HasEvents, removing the manual encoding
* Merge DkgConfirmer and DkgRemoval into signing_protocol.rs
Fixes a bug in DkgConfirmer which would cause it to improperly handle indexes
if any validator had multiple key shares.
* Strictly type DataSpecification's Label
* Correct threshold_i_map_to_keys_and_musig_i_map
It didn't include the participant's own index and accordingly was offset.
* Create TributaryBlockHandler
This struct contains all variables prior passed to handle_block and stops them
from being passed around again and again.
This also ensures fatal_slash is only called while handling a block, as needed
as it expects to operate under perfect consensus.
* Inline accumulate, store confirmation nonces with shares
Inlining accumulate makes sense due to the amount of data accumulate needed to
be passed.
Storing confirmation nonces with shares ensures that both are available or
neither. Prior, one could be yet the other may not have been (requiring an
assert in runtime to ensure we didn't bungle it somehow).
* Create helper functions for handling DkgRemoval/SubstrateSign/Sign Tributary TXs
* Move Label into SignData
All of our transactions which use SignData end up with the same common usage
pattern for Label, justifying this.
Removes 3 transactions, explicitly de-duplicating their handlers.
* Remove CurrentlyCompletingKeyPair for the non-contextual DkgKeyPair
* Remove the manual read/write for TributarySpec for borsh
This struct doesn't have any optimizations booned by the manual impl. Using
borsh reduces our scope.
* Use temporary variables to further minimize LoC in tributary handler
* Remove usage of tuples for non-trivial Tributary transactions
* Remove serde from dkg
serde could be used to deserialize intenrally inconsistent objects which could
lead to panics or faults.
The BorshDeserialize derives have been replaced with a manual implementation
which won't produce inconsistent objects.
* Abstract Future generics using new trait definitions in coordinator
* Move published_signed_transaction to tributary/mod.rs to reduce the size of main.rs
* Split coordinator/src/tributary/mod.rs into spec.rs and transaction.rs
* Add SignalsConfig to chain_spec
* Correct multiexp feature flagging for rand_core std
* Remove bincode for borsh
Replaces a non-canonical encoding with a canonical encoding which additionally
should be faster.
Also fixes an issue where we used bincode in transcripts where it cannot be
trusted.
This ended up fixing a myriad of other bugs observed, unfortunately.
Accordingly, it either has to be merged or the bug fixes from it must be ported
to a new PR.
* Make serde optional, minimize usage
* Make borsh an optional dependency of substrate/ crates
* Remove unused dependencies
* Use [u8; 64] where possible in the processor messages
* Correct borsh feature flagging
Resolves#391.
Given this code already wasn't modular/composable, this should be overall
equivalent regarding functionality and security. It's much less opinionated
though and has fewer dependencies.
