Files
secp256k1-zkp/tools/test_vectors_frost_enrollment_generate.py

364 lines
13 KiB
Python
Raw Normal View History

frost_enrollment: add the test suite and the regression vectors Fourth of six commits. Twelve tests replacing the Phase 2 smoke test, plus a vector generator and the frozen vectors it produces. The regression vectors are the one part of this worth being precise about, because they are easy to over-claim. FROST enrollment has no BIP and no published vectors, and the reference proof of concept draws its randomness from secrets.randbits, which is not seedable -- so there is nothing to cross-validate against. tools/test_vectors_frost_enrollment_generate.py therefore re-implements the math independently in stdlib-only Python, including the group arithmetic written from the secp256k1 parameters rather than borrowed, and freezes the output. What that buys: the two tag strings, the params hash serialization, the share-splitting derivation and the identifier conventions are now pinned, and changing any of them is a loud vector-breaking change. What it does not buy is evidence of protocol correctness. The generator header comment and the generated file both say so, as does frost_enrollment.md. The vectors passed on the first run against the C code, which is worth recording: two independent implementations agree byte for byte on the params hash, every delta, every sigma, the derived public share and the final share, across four cases (2-of-3 minimal, 2-of-3 oversized at u = 3 > t = 2, a 3-of-5 repair with a deliberately UNSORTED helper set, and a 4-of-6 enrollment), covering both threshold-key Y parities. The algebraic invariants are what actually carry correctness: - Reconstruction (PoC test_generate_frost_share): after a 2-of-3 group enrolls id 3, every pair {i, 3} reconstructs the original threshold secret, and so does the untouched pair {0, 1}. - Signing (PoC test_sign): a real BIP340 signature from {2, 3} verifying against the unchanged threshold public key, with every partial signature individually verified, plus the n -> n+1 bookkeeping -- secp256k1_frost_threshold_info_validate must accept the public share table extended with pubshare_derive's output at n+1. - Repair: byte-for-byte equality with the lost share, and the repaired participant keeps its old public share. - Oversized helper set: u = 3 and u = 2 over the same key material produce the same share and the same derived public share. - Randomized: COUNT iterations over 2 <= t <= u <= n <= 7, half enrollment and half repair, with EVERY HELPER GIVEN THE IDENTIFIER SET IN ITS OWN SHUFFLED ORDER. The params hash must come out identical while the delta buffers stay aligned per helper -- which is the whole point of canonicalizing ids inside the hash and nowhere else. Each iteration then checks every t-subset containing the new participant. The negative tests are organized around what each gate is actually for: - Fault injection flips a bit in one sigma. secshare_gen fails and wipes its output; the same call with expected_pubshare = NULL SUCCEEDS and returns a wrong share. That second assertion is the point -- it is the evidence that the parameter is load-bearing rather than decorative. Tampered public shares are caught earlier, by secp256k1_frost_threshold_info_validate, so the test exercises the recommended flow and not just the module. - Parameter mismatch, four angles: (a) one helper runs round 1.1 for a different target and every other helper's share_agg aborts naming it by identifier; (b) a caller that IGNORES that abort and finishes round 1.2 anyway still cannot produce a usable share, because the public-share check catches the inconsistent sum -- defence in depth, not a test of the test's own control flow; (c) the helpers agree with each other on new_id = 3 while the target expects 4, which round 1.2 cannot see and round 2's own recomputation does; (d) two groups with identical (t, n, ids, new_id) get different hashes, and a hash from one fails share_agg in the other. - Own-slot semantics: filling the caller's own slot of received_params_hashes32 with garbage changes nothing, because it is never read -- but the same garbage in a slot that IS read still aborts. That pair is what makes "recomputation, not string comparison" testable rather than merely asserted. - Invalid parameters, including both deliberate divergences: t = 1 refused, enrollment refused at n = 128 while repair at n = 128 is accepted, n_ids > 128 returning 0 with the output zeroed in a production build. Three bugs found while writing these, all in the tests, all worth naming: - pubshare_derive takes public shares ALIGNED WITH ids, and the test helper was handing it the participant-indexed table. Those coincide exactly when the helper set is 0..u-1, which every test until the repair case used, so the first non-contiguous helper set {0, 2} was what exposed it. There is now one helper that does the gather, with a comment saying which confusion it exists to prevent. - The fault-injection test compared against r.new_secshare without ever running round 2, and the mismatch test compared against r.params_hashes[0] one line before round 1.1 filled it. Both were reads of uninitialized memory that happened to pass; valgrind found both. The randomized test loops COUNT times so -i scales it, following the iceberg module (tests_impl.h:1322) rather than prefractal's run-once convention -- a fuzzing loop that ignores the iteration count is not much of one. Verification: all twelve tests pass at the default iteration count, at -i=200 and at -i=2000; ./tests, ./noverify_tests and ./exhaustive_tests exit 0 with all five FROST-stack modules enabled; the module runs clean under valgrind (0 errors from 0 contexts); ctime_tests is clean under valgrind; regenerating vectors.h reproduces it byte for byte. One note for anyone running these locally: ctime_tests must not be run against a CPPFLAGS='-DVERIFY' build. secp256k1_scalar_verify branches on scalar values, which ctime_tests deliberately marks secret, so every scalar operation in the library reports a finding -- 75997 of them, none in this module. The CI matrix already pairs -DVERIFY with CTIMETESTS: 'no' (.github/workflows/ci.yml:119, :596) for this reason. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-04 04:24:21 +02:00
#!/usr/bin/env python3
"""Generates src/modules/frost_enrollment/vectors.h.
These are REGRESSION vectors, not cross-validation vectors. FROST enrollment
has no BIP and therefore no published test vectors, and the reference proof of
concept (https://github.com/siv2r/frost-enrollment) draws its randomness from
secrets.randbits, which is not seedable -- so there is nothing to check the C
implementation against. This script therefore re-implements the same math
independently, in plain Python, and freezes the result.
What that buys is real but bounded: it pins the two tag strings, the exact
parameters hash serialization, the share-splitting derivation and the
identifier conventions, so that any change to them is a loud, deliberate,
vector-breaking change rather than a silent one. It is NOT evidence that the
protocol is implemented correctly -- the algebraic invariants in
src/modules/frost_enrollment/tests_impl.h are what carry that.
The one thing this file does establish independently is the group arithmetic:
the elliptic curve operations below are written from the secp256k1 parameters
rather than borrowed from the library, so a vector mismatch in the derived
public share or the threshold key really is a disagreement between two
implementations.
Usage: %s > src/modules/frost_enrollment/vectors.h
"""
import hashlib
import sys
import textwrap
# secp256k1 domain parameters.
P = 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFEFFFFFC2F
ORDER = 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFEBAAEDCE6AF48A03BBFD25E8CD0364141
GX = 0x79BE667EF9DCBBAC55A06295CE870B07029BFCDB2DCE28D959F2815B16F81798
GY = 0x483ADA7726A3C4655DA4FBFC0E1108A8FD17B448A68554199C47D08FFB10D4B8
G = (GX, GY)
MAX_PARTICIPANTS = 128
# --- group arithmetic (points are (x, y) or None for infinity) ---
def point_add(a, b):
if a is None:
return b
if b is None:
return a
if a[0] == b[0] and (a[1] + b[1]) % P == 0:
return None
if a == b:
lam = 3 * a[0] * a[0] * pow(2 * a[1], P - 2, P) % P
else:
lam = (b[1] - a[1]) * pow(b[0] - a[0], P - 2, P) % P
x = (lam * lam - a[0] - b[0]) % P
return (x, (lam * (a[0] - x) - a[1]) % P)
def point_mul(point, scalar):
result = None
scalar %= ORDER
while scalar:
if scalar & 1:
result = point_add(result, point)
point = point_add(point, point)
scalar >>= 1
return result
def cbytes(point):
"""33-byte compressed serialization."""
assert point is not None
return bytes([2 + (point[1] & 1)]) + point[0].to_bytes(32, "big")
# --- hashing ---
def tagged_hash(tag, msg):
tag_hash = hashlib.sha256(tag.encode()).digest()
return hashlib.sha256(tag_hash + tag_hash + msg).digest()
def ser32(x):
return x.to_bytes(4, "big")
# --- the protocol, mirroring src/modules/frost_enrollment/enrollment_impl.h ---
def params_hash(thresh_pk, ids, new_id, n_participants, threshold):
msg = cbytes(thresh_pk)
msg += ser32(n_participants) + ser32(threshold) + ser32(new_id)
msg += ser32(len(ids))
for i in sorted(ids):
msg += ser32(i)
return tagged_hash("FROST enrollment/params_hash", msg)
def lagrange_at(ids, my_id, new_id):
"""The Lagrange basis polynomial of my_id over ids, at the target.
Identifier space: the x-coordinate of identifier id is id + 1, so an
x-coordinate difference is an identifier difference and the +1 cancels."""
num, deno = 1, 1
for other in ids:
if other == my_id:
continue
num = num * (new_id - other) % ORDER
deno = deno * (my_id - other) % ORDER
return num * pow(deno, ORDER - 2, ORDER) % ORDER
def derive_mask(rand32, ph32, my_id, recipient_id):
msg = rand32 + ph32 + ser32(my_id) + ser32(recipient_id)
# from_bytes_wrapping: reduce mod the group order rather than reject.
return int.from_bytes(tagged_hash("FROST enrollment/share_split", msg), "big") % ORDER
def shares_gen(secshare, thresh_pk, ids, my_id, new_id, n_participants, threshold, secrand32):
ph32 = params_hash(thresh_pk, ids, new_id, n_participants, threshold)
v = lagrange_at(ids, my_id, new_id) * secshare % ORDER
rand32 = bytes(
a ^ b
for a, b in zip(
tagged_hash("FROST enrollment/share_split", secrand32),
secshare.to_bytes(32, "big"),
)
)
out = [0] * len(ids)
my_pos = ids.index(my_id)
for j, recipient in enumerate(ids):
if j == my_pos:
continue
out[j] = derive_mask(rand32, ph32, my_id, recipient)
v = (v - out[j]) % ORDER
out[my_pos] = v
return out, ph32
def trusted_dealer_keygen(thresh_sk, n_participants, threshold):
"""The frost module's trusted dealer (src/modules/frost/keygen_impl.h)."""
coeffs = []
for i in range(1, threshold):
h = tagged_hash("BIP0445/trusted/keygen", thresh_sk.to_bytes(32, "big") + ser32(i))
c = int.from_bytes(h, "big")
assert 0 < c < ORDER
coeffs.append(c)
secshares = []
for i in range(n_participants):
x = i + 1
share = 0
for c in coeffs:
share = (share * x + c) % ORDER
share = (share * x + thresh_sk) % ORDER
assert share != 0
secshares.append(share)
return secshares, point_mul(G, thresh_sk), [point_mul(G, s) for s in secshares]
def run_case(thresh_sk, n_participants, threshold, ids, new_id, seeds):
secshares, thresh_pk, pubshares = trusted_dealer_keygen(thresh_sk, n_participants, threshold)
ph32 = params_hash(thresh_pk, ids, new_id, n_participants, threshold)
shares = []
for k, my_id in enumerate(ids):
out, ph = shares_gen(
secshares[my_id], thresh_pk, ids, my_id, new_id, n_participants, threshold, seeds[k]
)
assert ph == ph32
shares.append(out)
# Round 1.2: helper j sums entry j of every helper's output.
sigmas = [sum(shares[i][j] for i in range(len(ids))) % ORDER for j in range(len(ids))]
# Round 2, and the independent check that the result really is f(x_new).
new_secshare = sum(sigmas) % ORDER
expected = sum(lagrange_at(ids, i, new_id) * secshares[i] for i in ids) % ORDER
assert new_secshare == expected
new_pubshare = None
for i in ids:
new_pubshare = point_add(new_pubshare, point_mul(pubshares[i], lagrange_at(ids, i, new_id)))
assert new_pubshare == point_mul(G, new_secshare)
return {
"n_participants": n_participants,
"threshold": threshold,
"ids": ids,
"new_id": new_id,
"thresh_pk": thresh_pk,
"pubshares": [pubshares[i] for i in ids],
"secshares": [secshares[i] for i in ids],
"seeds": seeds,
"params_hash": ph32,
"shares": shares,
"sigmas": sigmas,
"new_secshare": new_secshare,
"new_pubshare": new_pubshare,
}
# --- C emission ---
def byte_array(b):
return "{ %s }" % ", ".join("0x%02X" % x for x in b)
def scalar_array(x):
return byte_array(x.to_bytes(32, "big"))
def indent(s, level=1):
return textwrap.indent(s, 4 * level * " ")
def emit_case(c):
n_ids = len(c["ids"])
lines = []
lines.append("%d, %d, %d, %d," % (c["n_participants"], c["threshold"], n_ids, c["new_id"]))
lines.append("{ %s }," % ", ".join(str(i) for i in c["ids"]))
lines.append("%s," % byte_array(cbytes(c["thresh_pk"])))
lines.append("{ %s }," % ", ".join(byte_array(cbytes(p)) for p in c["pubshares"]))
lines.append("{ %s }," % ", ".join(scalar_array(s) for s in c["secshares"]))
lines.append("{ %s }," % ", ".join(byte_array(s) for s in c["seeds"]))
lines.append("%s," % byte_array(c["params_hash"]))
lines.append(
"{ %s },"
% ", ".join(
"{ %s }" % ", ".join("0x%02X" % b for s in row for b in s.to_bytes(32, "big"))
for row in c["shares"]
)
)
lines.append(
"{ %s },"
% ", ".join("0x%02X" % b for s in c["sigmas"] for b in s.to_bytes(32, "big"))
)
lines.append("%s," % scalar_array(c["new_secshare"]))
lines.append("%s" % byte_array(cbytes(c["new_pubshare"])))
return "{\n" + indent("\n".join(lines)) + "\n},"
# Fixed inputs. Nothing here is random at run time: the whole point is that
# regenerating this file without an intentional change reproduces it byte for
# byte.
CASES = [
# A 2-of-3 group enrolling a fourth participant with the minimum helper
# set. The base case, and the one the module documentation walks through.
frost_enrollment: fix the example's stale n and failure-path hygiene The last of the review findings, plus the comment and structure fixes it listed. The example's repair run used the pre-enrollment participant count. Two blocks earlier the example teaches that every participant must update its record of n from 3 to 4 after an enrollment, and the signing session duly uses N_PARTICIPANTS_AFTER. Then enroll() -- which hard-coded N_PARTICIPANTS -- ran the repair at n = 3. It worked only because the Lagrange math never involves n and every party in this single-process demo passed the same stale value. In a real post-enrollment repair it would not. n is bound into the parameters hash, so helpers feeding their updated n = 4 into shares_gen while the requester feeds n = 3 abort round 1.2 with no visible cause. enroll() now takes n_participants as a parameter, the repair passes N_PARTICIPANTS_AFTER, and both the function's contract comment and the repair call site say why. The repaired share is still byte-identical to the original, which is the point: n changes the hash, not the arithmetic. The example leaked secrets on its failure paths. enroll() erased the delta and sigma buffers only on success; four early returns left them live. sign_and_verify() returned from three places without erasing already-generated secnonces. Both now route every exit through a cleanup block. This example is otherwise more careful about erasure than its siblings, so the asymmetry was exactly what a reader copying it would carry into production -- on the fault paths where hygiene matters most. The double-wipe of session_secrand is gone with it: shares_gen and nonce_gen both wipe the seed on every path, and doing it again read as uncertainty about the contract. The comment now states the contract instead. The fill_random failure path does erase, since nothing else has touched the buffer there. The example's mismatch message asserted a cause it cannot know. It printed "Helper %u disagrees about the enrollment parameters" for what may equally be a corrupted share, per the previous commit's finding. It now says the helper "contributed a share this helper cannot use", with a comment noting that share_agg does not distinguish the two causes so neither can the message. Comment and structure fixes, all noted in the review: - The vector generator claimed case 4 was "the only case whose DERIVED public share has odd Y". It is not -- cases 1, 2 and 4 are odd and case 3 is even. The comment existed to justify a coverage choice and misinformed; both parity comments now describe the set accurately and say they document it rather than constrain it. Regenerating vectors.h still reproduces it byte for byte. - The secp256k1_frost_sort_ids declaration in frost/session.h no longer duplicates the definition's doc comment, which was two copies to keep in sync. It says what the function is for and points at the definition for the contract. - The t >= 2 rationale was stated in full in three places. The impl now states the conclusion and names frost_enrollment.md as the single place to edit if the policy moves. - The ctime_tests comments read ambiguously ("the parameters hash is public, the delta values are not" against a header calling deltas secret), and computing direct_hash without asserting anything invited a "forgotten assertion" reading. Both are now explicit. - The example moves next to frost_example in Makefile.am rather than after iceberg, matching the FROST-stack grouping used in configure.ac, ci.sh, ci.yml and README. - frost_enrollment.md now distinguishes what is unstable (the C API) from what is frozen (the wire-visible encodings), which the two statements together previously left easy to conflate. Not fixed, deliberately, and now recorded where the tree can see it: the plan called for a CHANGELOG.md entry. That file states in its first two lines that it is upstream libsecp256k1's changelog and not this fork's, and none of frost, chilldkg, iceberg or prefractal has an entry. Adding the first one is a decision about all five modules, not this one. The README link is the fork's actual convention for announcing a module and is in place. Verification: autotools builds warning-free and `make check` is 12/12 including the example; ctime_tests is clean under valgrind; `make dist` carries all nine frost_enrollment files; CMake with examples builds warning-free and ctest is 542/542; the example source is clean under gcc -std=c89 -pedantic -Wall -Wextra; regenerating vectors.h reproduces it byte for byte. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-04 10:27:17 +02:00
# This key and the next have EVEN Y, the last two odd Y; among the derived
# public shares, case 3 is the even one and the rest are odd. Nothing in
# enrollment depends on either parity -- unlike frost signing, it never
# takes an x-only view of a key -- so this is coverage rather than a
# distinction the code makes. run_case() in this file prints nothing about
# parity; the values above were read off the generated vectors, so treat
# this comment as documentation of the current set rather than a
# constraint on it.
frost_enrollment: add the test suite and the regression vectors Fourth of six commits. Twelve tests replacing the Phase 2 smoke test, plus a vector generator and the frozen vectors it produces. The regression vectors are the one part of this worth being precise about, because they are easy to over-claim. FROST enrollment has no BIP and no published vectors, and the reference proof of concept draws its randomness from secrets.randbits, which is not seedable -- so there is nothing to cross-validate against. tools/test_vectors_frost_enrollment_generate.py therefore re-implements the math independently in stdlib-only Python, including the group arithmetic written from the secp256k1 parameters rather than borrowed, and freezes the output. What that buys: the two tag strings, the params hash serialization, the share-splitting derivation and the identifier conventions are now pinned, and changing any of them is a loud vector-breaking change. What it does not buy is evidence of protocol correctness. The generator header comment and the generated file both say so, as does frost_enrollment.md. The vectors passed on the first run against the C code, which is worth recording: two independent implementations agree byte for byte on the params hash, every delta, every sigma, the derived public share and the final share, across four cases (2-of-3 minimal, 2-of-3 oversized at u = 3 > t = 2, a 3-of-5 repair with a deliberately UNSORTED helper set, and a 4-of-6 enrollment), covering both threshold-key Y parities. The algebraic invariants are what actually carry correctness: - Reconstruction (PoC test_generate_frost_share): after a 2-of-3 group enrolls id 3, every pair {i, 3} reconstructs the original threshold secret, and so does the untouched pair {0, 1}. - Signing (PoC test_sign): a real BIP340 signature from {2, 3} verifying against the unchanged threshold public key, with every partial signature individually verified, plus the n -> n+1 bookkeeping -- secp256k1_frost_threshold_info_validate must accept the public share table extended with pubshare_derive's output at n+1. - Repair: byte-for-byte equality with the lost share, and the repaired participant keeps its old public share. - Oversized helper set: u = 3 and u = 2 over the same key material produce the same share and the same derived public share. - Randomized: COUNT iterations over 2 <= t <= u <= n <= 7, half enrollment and half repair, with EVERY HELPER GIVEN THE IDENTIFIER SET IN ITS OWN SHUFFLED ORDER. The params hash must come out identical while the delta buffers stay aligned per helper -- which is the whole point of canonicalizing ids inside the hash and nowhere else. Each iteration then checks every t-subset containing the new participant. The negative tests are organized around what each gate is actually for: - Fault injection flips a bit in one sigma. secshare_gen fails and wipes its output; the same call with expected_pubshare = NULL SUCCEEDS and returns a wrong share. That second assertion is the point -- it is the evidence that the parameter is load-bearing rather than decorative. Tampered public shares are caught earlier, by secp256k1_frost_threshold_info_validate, so the test exercises the recommended flow and not just the module. - Parameter mismatch, four angles: (a) one helper runs round 1.1 for a different target and every other helper's share_agg aborts naming it by identifier; (b) a caller that IGNORES that abort and finishes round 1.2 anyway still cannot produce a usable share, because the public-share check catches the inconsistent sum -- defence in depth, not a test of the test's own control flow; (c) the helpers agree with each other on new_id = 3 while the target expects 4, which round 1.2 cannot see and round 2's own recomputation does; (d) two groups with identical (t, n, ids, new_id) get different hashes, and a hash from one fails share_agg in the other. - Own-slot semantics: filling the caller's own slot of received_params_hashes32 with garbage changes nothing, because it is never read -- but the same garbage in a slot that IS read still aborts. That pair is what makes "recomputation, not string comparison" testable rather than merely asserted. - Invalid parameters, including both deliberate divergences: t = 1 refused, enrollment refused at n = 128 while repair at n = 128 is accepted, n_ids > 128 returning 0 with the output zeroed in a production build. Three bugs found while writing these, all in the tests, all worth naming: - pubshare_derive takes public shares ALIGNED WITH ids, and the test helper was handing it the participant-indexed table. Those coincide exactly when the helper set is 0..u-1, which every test until the repair case used, so the first non-contiguous helper set {0, 2} was what exposed it. There is now one helper that does the gather, with a comment saying which confusion it exists to prevent. - The fault-injection test compared against r.new_secshare without ever running round 2, and the mismatch test compared against r.params_hashes[0] one line before round 1.1 filled it. Both were reads of uninitialized memory that happened to pass; valgrind found both. The randomized test loops COUNT times so -i scales it, following the iceberg module (tests_impl.h:1322) rather than prefractal's run-once convention -- a fuzzing loop that ignores the iteration count is not much of one. Verification: all twelve tests pass at the default iteration count, at -i=200 and at -i=2000; ./tests, ./noverify_tests and ./exhaustive_tests exit 0 with all five FROST-stack modules enabled; the module runs clean under valgrind (0 errors from 0 contexts); ctime_tests is clean under valgrind; regenerating vectors.h reproduces it byte for byte. One note for anyone running these locally: ctime_tests must not be run against a CPPFLAGS='-DVERIFY' build. secp256k1_scalar_verify branches on scalar values, which ctime_tests deliberately marks secret, so every scalar operation in the library reports a finding -- 75997 of them, none in this module. The CI matrix already pairs -DVERIFY with CTIMETESTS: 'no' (.github/workflows/ci.yml:119, :596) for this reason. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-04 04:24:21 +02:00
dict(
thresh_sk=0x0202020202020202020202020202020202020202020202020202020202020202,
n_participants=3,
threshold=2,
ids=[0, 1],
new_id=3,
seeds=[bytes([0x10 + i] * 32) for i in range(2)],
),
# The same group with an oversized helper set: u = 3 > t = 2. The resulting
# share must be the one the u = 2 case produces, which the C test checks
# separately; here it is simply frozen.
dict(
thresh_sk=0x0202020202020202020202020202020202020202020202020202020202020202,
n_participants=3,
threshold=2,
ids=[0, 1, 2],
new_id=3,
seeds=[bytes([0x20 + i] * 32) for i in range(3)],
),
# Repair: a 3-of-5 group reproducing participant 2's lost share. The helper
# set is deliberately unsorted, to pin that the parameters hash
# canonicalizes identifiers while the share buffers follow the caller's
# order.
dict(
thresh_sk=0x02030405060708090A0B0C0D0E0F101112131415161718191A1B1C1D1E1F2021,
n_participants=5,
threshold=3,
ids=[4, 0, 3],
new_id=2,
seeds=[bytes([0x30 + i] * 32) for i in range(3)],
),
frost_enrollment: fix the example's stale n and failure-path hygiene The last of the review findings, plus the comment and structure fixes it listed. The example's repair run used the pre-enrollment participant count. Two blocks earlier the example teaches that every participant must update its record of n from 3 to 4 after an enrollment, and the signing session duly uses N_PARTICIPANTS_AFTER. Then enroll() -- which hard-coded N_PARTICIPANTS -- ran the repair at n = 3. It worked only because the Lagrange math never involves n and every party in this single-process demo passed the same stale value. In a real post-enrollment repair it would not. n is bound into the parameters hash, so helpers feeding their updated n = 4 into shares_gen while the requester feeds n = 3 abort round 1.2 with no visible cause. enroll() now takes n_participants as a parameter, the repair passes N_PARTICIPANTS_AFTER, and both the function's contract comment and the repair call site say why. The repaired share is still byte-identical to the original, which is the point: n changes the hash, not the arithmetic. The example leaked secrets on its failure paths. enroll() erased the delta and sigma buffers only on success; four early returns left them live. sign_and_verify() returned from three places without erasing already-generated secnonces. Both now route every exit through a cleanup block. This example is otherwise more careful about erasure than its siblings, so the asymmetry was exactly what a reader copying it would carry into production -- on the fault paths where hygiene matters most. The double-wipe of session_secrand is gone with it: shares_gen and nonce_gen both wipe the seed on every path, and doing it again read as uncertainty about the contract. The comment now states the contract instead. The fill_random failure path does erase, since nothing else has touched the buffer there. The example's mismatch message asserted a cause it cannot know. It printed "Helper %u disagrees about the enrollment parameters" for what may equally be a corrupted share, per the previous commit's finding. It now says the helper "contributed a share this helper cannot use", with a comment noting that share_agg does not distinguish the two causes so neither can the message. Comment and structure fixes, all noted in the review: - The vector generator claimed case 4 was "the only case whose DERIVED public share has odd Y". It is not -- cases 1, 2 and 4 are odd and case 3 is even. The comment existed to justify a coverage choice and misinformed; both parity comments now describe the set accurately and say they document it rather than constrain it. Regenerating vectors.h still reproduces it byte for byte. - The secp256k1_frost_sort_ids declaration in frost/session.h no longer duplicates the definition's doc comment, which was two copies to keep in sync. It says what the function is for and points at the definition for the contract. - The t >= 2 rationale was stated in full in three places. The impl now states the conclusion and names frost_enrollment.md as the single place to edit if the policy moves. - The ctime_tests comments read ambiguously ("the parameters hash is public, the delta values are not" against a header calling deltas secret), and computing direct_hash without asserting anything invited a "forgotten assertion" reading. Both are now explicit. - The example moves next to frost_example in Makefile.am rather than after iceberg, matching the FROST-stack grouping used in configure.ac, ci.sh, ci.yml and README. - frost_enrollment.md now distinguishes what is unstable (the C API) from what is frozen (the wire-visible encodings), which the two statements together previously left easy to conflate. Not fixed, deliberately, and now recorded where the tree can see it: the plan called for a CHANGELOG.md entry. That file states in its first two lines that it is upstream libsecp256k1's changelog and not this fork's, and none of frost, chilldkg, iceberg or prefractal has an entry. Adding the first one is a decision about all five modules, not this one. The README link is the fork's actual convention for announcing a module and is in place. Verification: autotools builds warning-free and `make check` is 12/12 including the example; ctime_tests is clean under valgrind; `make dist` carries all nine frost_enrollment files; CMake with examples builds warning-free and ctest is 542/542; the example source is clean under gcc -std=c89 -pedantic -Wall -Wextra; regenerating vectors.h reproduces it byte for byte. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-04 10:27:17 +02:00
# A larger enrollment, 4-of-6 to 4-of-7, at the largest threshold and
# helper count in this set.
frost_enrollment: add the test suite and the regression vectors Fourth of six commits. Twelve tests replacing the Phase 2 smoke test, plus a vector generator and the frozen vectors it produces. The regression vectors are the one part of this worth being precise about, because they are easy to over-claim. FROST enrollment has no BIP and no published vectors, and the reference proof of concept draws its randomness from secrets.randbits, which is not seedable -- so there is nothing to cross-validate against. tools/test_vectors_frost_enrollment_generate.py therefore re-implements the math independently in stdlib-only Python, including the group arithmetic written from the secp256k1 parameters rather than borrowed, and freezes the output. What that buys: the two tag strings, the params hash serialization, the share-splitting derivation and the identifier conventions are now pinned, and changing any of them is a loud vector-breaking change. What it does not buy is evidence of protocol correctness. The generator header comment and the generated file both say so, as does frost_enrollment.md. The vectors passed on the first run against the C code, which is worth recording: two independent implementations agree byte for byte on the params hash, every delta, every sigma, the derived public share and the final share, across four cases (2-of-3 minimal, 2-of-3 oversized at u = 3 > t = 2, a 3-of-5 repair with a deliberately UNSORTED helper set, and a 4-of-6 enrollment), covering both threshold-key Y parities. The algebraic invariants are what actually carry correctness: - Reconstruction (PoC test_generate_frost_share): after a 2-of-3 group enrolls id 3, every pair {i, 3} reconstructs the original threshold secret, and so does the untouched pair {0, 1}. - Signing (PoC test_sign): a real BIP340 signature from {2, 3} verifying against the unchanged threshold public key, with every partial signature individually verified, plus the n -> n+1 bookkeeping -- secp256k1_frost_threshold_info_validate must accept the public share table extended with pubshare_derive's output at n+1. - Repair: byte-for-byte equality with the lost share, and the repaired participant keeps its old public share. - Oversized helper set: u = 3 and u = 2 over the same key material produce the same share and the same derived public share. - Randomized: COUNT iterations over 2 <= t <= u <= n <= 7, half enrollment and half repair, with EVERY HELPER GIVEN THE IDENTIFIER SET IN ITS OWN SHUFFLED ORDER. The params hash must come out identical while the delta buffers stay aligned per helper -- which is the whole point of canonicalizing ids inside the hash and nowhere else. Each iteration then checks every t-subset containing the new participant. The negative tests are organized around what each gate is actually for: - Fault injection flips a bit in one sigma. secshare_gen fails and wipes its output; the same call with expected_pubshare = NULL SUCCEEDS and returns a wrong share. That second assertion is the point -- it is the evidence that the parameter is load-bearing rather than decorative. Tampered public shares are caught earlier, by secp256k1_frost_threshold_info_validate, so the test exercises the recommended flow and not just the module. - Parameter mismatch, four angles: (a) one helper runs round 1.1 for a different target and every other helper's share_agg aborts naming it by identifier; (b) a caller that IGNORES that abort and finishes round 1.2 anyway still cannot produce a usable share, because the public-share check catches the inconsistent sum -- defence in depth, not a test of the test's own control flow; (c) the helpers agree with each other on new_id = 3 while the target expects 4, which round 1.2 cannot see and round 2's own recomputation does; (d) two groups with identical (t, n, ids, new_id) get different hashes, and a hash from one fails share_agg in the other. - Own-slot semantics: filling the caller's own slot of received_params_hashes32 with garbage changes nothing, because it is never read -- but the same garbage in a slot that IS read still aborts. That pair is what makes "recomputation, not string comparison" testable rather than merely asserted. - Invalid parameters, including both deliberate divergences: t = 1 refused, enrollment refused at n = 128 while repair at n = 128 is accepted, n_ids > 128 returning 0 with the output zeroed in a production build. Three bugs found while writing these, all in the tests, all worth naming: - pubshare_derive takes public shares ALIGNED WITH ids, and the test helper was handing it the participant-indexed table. Those coincide exactly when the helper set is 0..u-1, which every test until the repair case used, so the first non-contiguous helper set {0, 2} was what exposed it. There is now one helper that does the gather, with a comment saying which confusion it exists to prevent. - The fault-injection test compared against r.new_secshare without ever running round 2, and the mismatch test compared against r.params_hashes[0] one line before round 1.1 filled it. Both were reads of uninitialized memory that happened to pass; valgrind found both. The randomized test loops COUNT times so -i scales it, following the iceberg module (tests_impl.h:1322) rather than prefractal's run-once convention -- a fuzzing loop that ignores the iteration count is not much of one. Verification: all twelve tests pass at the default iteration count, at -i=200 and at -i=2000; ./tests, ./noverify_tests and ./exhaustive_tests exit 0 with all five FROST-stack modules enabled; the module runs clean under valgrind (0 errors from 0 contexts); ctime_tests is clean under valgrind; regenerating vectors.h reproduces it byte for byte. One note for anyone running these locally: ctime_tests must not be run against a CPPFLAGS='-DVERIFY' build. secp256k1_scalar_verify branches on scalar values, which ctime_tests deliberately marks secret, so every scalar operation in the library reports a finding -- 75997 of them, none in this module. The CI matrix already pairs -DVERIFY with CTIMETESTS: 'no' (.github/workflows/ci.yml:119, :596) for this reason. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-04 04:24:21 +02:00
dict(
thresh_sk=0x1122334455667788990011223344556677889900112233445566778899001122,
n_participants=6,
threshold=4,
ids=[0, 2, 3, 5],
new_id=6,
seeds=[bytes([0x40 + i] * 32) for i in range(4)],
),
]
def main():
cases = [run_case(**c) for c in CASES]
max_ids = max(len(c["ids"]) for c in cases)
out = """/**
* Automatically generated by tools/test_vectors_frost_enrollment_generate.py.
*
* REGRESSION vectors, not cross-validation vectors. FROST enrollment has no
* BIP and no published test vectors, and the reference proof of concept
* (https://github.com/siv2r/frost-enrollment) draws its randomness from
* secrets.randbits, which is not seedable, so there is nothing to check
* against. The generator re-implements the math independently in Python and
* freezes the result.
*
* What these pin: the two tag strings ("FROST enrollment/params_hash" and
* "FROST enrollment/share_split"), the parameters hash serialization, the
* share-splitting derivation, and the identifier conventions. Changing any of
* them is a vector-breaking change. What they do NOT establish is protocol
* correctness -- the algebraic invariants in tests_impl.h carry that.
*
* Used by the tests in src/modules/frost_enrollment/tests_impl.h. */
#ifndef SECP256K1_MODULE_FROST_ENROLLMENT_VECTORS_H
#define SECP256K1_MODULE_FROST_ENROLLMENT_VECTORS_H
#define FROST_ENROLLMENT_VEC_MAX_IDS %d
struct frost_enrollment_vec_case {
/* Parameters. */
size_t n_participants;
uint32_t threshold;
size_t n_ids;
uint32_t new_id;
uint32_t ids[FROST_ENROLLMENT_VEC_MAX_IDS];
/* Group key material, aligned with ids. */
unsigned char thresh_pk33[33];
unsigned char pubshares33[FROST_ENROLLMENT_VEC_MAX_IDS][33];
unsigned char secshares32[FROST_ENROLLMENT_VEC_MAX_IDS][32];
/* Round 1.1 inputs and outputs. shares32[i] is helper ids[i]'s output
* buffer, aligned with ids. */
unsigned char session_secrand32[FROST_ENROLLMENT_VEC_MAX_IDS][32];
unsigned char params_hash32[32];
unsigned char shares32[FROST_ENROLLMENT_VEC_MAX_IDS][FROST_ENROLLMENT_VEC_MAX_IDS * 32];
/* Round 1.2 and round 2 outputs. */
unsigned char sigmas32[FROST_ENROLLMENT_VEC_MAX_IDS * 32];
unsigned char new_secshare32[32];
unsigned char new_pubshare33[33];
};
static const struct frost_enrollment_vec_case frost_enrollment_vec_cases[%d] = {
""" % (
max_ids,
len(cases),
)
for c in cases:
out += indent(emit_case(c)) + "\n"
out += "};\n\n#endif /* SECP256K1_MODULE_FROST_ENROLLMENT_VECTORS_H */\n"
sys.stdout.write(out)
if __name__ == "__main__":
main()