bgls

README

BGLS

Master: Develop:

Aggregate and Multi Signatures based on BGLS over Alt bn128 and BLS12-381

This library provides no security against side channel attacks. We provide no security guarantees of this implementation.

Design

The goal of this library is to create an efficient and secure ad hoc aggregate and multi signature scheme. It supports the curves bls12-381 and alt bn128. It implements hashing of arbitrary byte data to curve points, the standard BGLS scheme for aggregate signatures, and a custom multi signature scheme.

Multi Signature

The multi signature scheme is a modification of the BGLS scheme, where all signatures are on the same message. This allows verification with a constant number of pairing operations, at the cost of being insecure to rogue public key attacks. We have three separate solutions to the rogue public key attack implemented. (Proving knowlege of the secret key, Enforcing that messages are distinct, and performing aggregation with hashed exponents. These are described in Dan Boneh's recent paper)

Curves

Bls12-381

This is the set of curves which zcash is switching too. Its official documentation is located here.

Alt bn128

The group G_1 is a cyclic group of prime order on the curve Y^2 = X^3 + 3 defined over the field F_p with p = 21888242871839275222246405745257275088696311157297823662689037894645226208583.

The generator g_1 is (1,2)

Since this curve is of prime order, every non-identity point is a generator, therefore the cofactor is 1.

The group G_2 is a cyclic subgroup of the non-prime order elliptic curve Y^2 = X^3 + 3*((i + 9)^(-1)) over the field F_p^2 = F_p[X] / (X^2 + 1) (where p is the same as above). We can write our irreducible element as i. The cofactor of this group is 21888242871839275222246405745257275088844257914179612981679871602714643921549.

The generator g_2 is defined as: (11559732032986387107991004021392285783925812861821192530917403151452391805634*i + 10857046999023057135944570762232829481370756359578518086990519993285655852781, 4082367875863433681332203403145435568316851327593401208105741076214120093531*i + 8495653923123431417604973247489272438418190587263600148770280649306958101930)

The identity element for both groups (The point at infinity in affine space) is internally represented as (0,0)

Benchmarks

The following benchmarks are from a 3.80GHz i7-7700HQ CPU with 16GB ram. The aggregate verification is utilizing parallelization for the pairing operations. The multisignature has parellilization for the two involved pairing operations, and parallelization for the pairing checks at the end. Note, all of the benchmarks need to be updated.

For reference, the pairing operation on Altbn128 (the slowest operation involved) takes ~1.9 milliseconds.

BenchmarkPairing-8   	    1000	   1958898 ns/op

and for Bls12 its:

BenchmarkPairGT-8               	    1000	   1539918 ns/op

• Signing ~.22 milliseconds
• Signature verification ~3.1 milliseconds, using two pairings.
• Multi Signature verification ~2 milliseconds + ~1.1 microseconds per signer, two pairings + n point additions
• Aggregate Signature verification ~.36 milliseconds per signer/message pair, with n+1 pairings run in parallel. (4.45x speedup with 8 cores)

The following benchmarks are done with altbn128, before the product of pairings abstraction was included. These need to be updated.

$ go test github.com/Project-Arda/bgls/bgls/  -v -bench .
BenchmarkKeygen-8                  	    3000	    434484 ns/op
BenchmarkAltBnHashToCurve-8        	   20000	     91947 ns/op
BenchmarkSigning-8                 	   10000	    218670 ns/op
BenchmarkVerification-8            	     500	   3079415 ns/op
BenchmarkMultiVerification64-8     	    1000	   2056798 ns/op
BenchmarkMultiVerification128-8    	    1000	   2140613 ns/op
BenchmarkMultiVerification256-8    	     500	   2334271 ns/op
BenchmarkMultiVerification512-8    	     500	   2617277 ns/op
BenchmarkMultiVerification1024-8   	     500	   3243045 ns/op
BenchmarkMultiVerification2048-8   	     300	   4325183 ns/op
BenchmarkAggregateVerification-8   	    5000	    361270 ns/op
PASS
ok  	github.com/Project-Arda/bgls	31.043s

For comparison, the ed25519 implementation in go yields much faster key generation signing and single signature verification. However, at ~145 microseconds per verification, the multi signature verification is actually faster beyond ~26 signatures.

$ go test golang.org/x/crypto/ed25519 -bench .
BenchmarkKeyGeneration-8   	   30000	     51878 ns/op
BenchmarkSigning-8         	   30000	     54050 ns/op
BenchmarkVerification-8    	   10000	    145063 ns/op
PASS
ok  	golang.org/x/crypto/ed25519	5.750s

Hashing

Currently only hashing to G1 is supported. Hashing to G2 is planned. For altbn128, the hashing algorithm is currently try-and-increment, and we support SHA3, Kangaroo twelve, Keccak256, and Blake2b.

For bls12-381, we are using Fouque-Tibouchi hashing using blake2b. This is interoperable with ebfull's repository.

Future work

• Optimize bigint allocations.
• Add hashing to G2
• Integrations with bgls-on-evm.
• Add tests to show that none of the functions mutate data.
• More complete usage documentation.
• Add buffering for the channels used in parallelization.
• Make upstream libraries implement product of pairings algorithms

References

• Dan Boneh Methods to prevent the rogue public key attack
• Dan Boneh, Craig Gentry, Ben Lynn, and Hovav Shacham. Aggregate and verifiably encrypted signatures from bilinear maps
• Pierre-Alain Fouque and Mehdi Tibouchi. Indifferentiable Hashing to Barreto–Naehrig Curves
• Claus-Peter Schnorr. Efficient Signature Generation by Smart Cards