Added HKDF tests
Mark Qvist
parent
929aba0282
commit
cfddb9abe6
|
|
@ -0,0 +1,188 @@
|
|||
import unittest
|
||||
|
||||
import os
|
||||
import time
|
||||
import random
|
||||
|
||||
import RNS
|
||||
from RNS.Cryptography import hkdf
|
||||
from RNS.Cryptography.HKDF import hkdf_legacy
|
||||
|
||||
HKDF_TEST_SEED = 0x5EE5FAC
|
||||
|
||||
# RFC 5869 test vectors HKDF SHA-256 vectors
|
||||
RFC5869_CASES = [
|
||||
# A.1: basic test case
|
||||
(
|
||||
bytes([0x0b]) * 22,
|
||||
bytes(range(0x00, 0x0d)),
|
||||
bytes(range(0xf0, 0xfa)),
|
||||
42,
|
||||
"3cb25f25faacd57a90434f64d0362f2a2d2d0a90cf1a5a4c5db02d56ecc4c5bf34007208d5b887185865",
|
||||
),
|
||||
# A.2: longer inputs/outputs
|
||||
(
|
||||
bytes(range(0x00, 0x50)),
|
||||
bytes(range(0x60, 0xb0)),
|
||||
bytes(range(0xb0, 0x100)),
|
||||
82,
|
||||
"b11e398dc80327a1c8e7f78c596a49344f012eda2d4efad8a050cc4c19afa97c"
|
||||
"59045a99cac7827271cb41c65e590e09da3275600c2f09b8367793a9aca3db71"
|
||||
"cc30c58179ec3e87c14c01d5c1f3434f1d87",
|
||||
),
|
||||
# A.3: zero-length salt/info
|
||||
(
|
||||
bytes([0x0b]) * 22,
|
||||
b"",
|
||||
b"",
|
||||
42,
|
||||
"8da4e775a563c18f715f802a063c5a31b8a11f5c5ee1879ec3454e5f3c738d2d9d201395faa4b61a96c8",
|
||||
),
|
||||
]
|
||||
|
||||
|
||||
def rand_bytes(rng, length):
|
||||
return bytes(rng.randrange(256) for _ in range(length))
|
||||
|
||||
|
||||
class TestHKDF(unittest.TestCase):
|
||||
|
||||
def test_01_rfc5869_vectors(self):
|
||||
print("")
|
||||
|
||||
for index, (ikm, salt, info, length, expected_hex) in enumerate(RFC5869_CASES):
|
||||
expected = bytes.fromhex(expected_hex)
|
||||
|
||||
new_out = hkdf(length=length, derive_from=ikm, salt=salt, context=info)
|
||||
leg_out = hkdf_legacy(length=length, derive_from=ikm, salt=salt, context=info)
|
||||
|
||||
self.assertEqual(new_out, expected)
|
||||
self.assertEqual(leg_out, expected)
|
||||
|
||||
print(f"RFC 5869 test case {index+1}: OK (optimized = legacy = reference, {len(expected)} bytes)")
|
||||
|
||||
def test_02_parity_matrix(self):
|
||||
print("")
|
||||
|
||||
rng = random.Random(HKDF_TEST_SEED)
|
||||
|
||||
# Core lengths: block boundaries, block+1, counter wrap at
|
||||
# 256 blocks (8192 bytes), +1/-1 around it, and medium sizes.
|
||||
lengths_core = [ 1, 2, 31, 32, 33, 63, 64, 65, 96, 127, 128, 129,
|
||||
255, 256, 257, 511, 512, 513, 1024, 4096, 8191, 8192, 8193 ]
|
||||
# Large lengths: exercised with a reduced combination set.
|
||||
lengths_large = [16384, 65536, 262144]
|
||||
|
||||
derive_lens = [1, 7, 16, 32, 64]
|
||||
salt_lens = [None, 0, 1, 16, 32, 64, 100]
|
||||
context_lens = [None, 0, 1, 16, 64, 128]
|
||||
derive_lens_large = [16, 64]
|
||||
salt_lens_large = [None, 32, 100]
|
||||
context_lens_large = [None, 16, 128]
|
||||
|
||||
verified = 0
|
||||
for length in lengths_core:
|
||||
for dlen in derive_lens:
|
||||
for slen in salt_lens:
|
||||
for clen in context_lens:
|
||||
derive_from = rand_bytes(rng, dlen)
|
||||
salt = None if slen is None else (b"" if slen == 0 else rand_bytes(rng, slen))
|
||||
context = None if clen is None else (b"" if clen == 0 else rand_bytes(rng, clen))
|
||||
|
||||
new_out = hkdf(length=length, derive_from=derive_from, salt=salt, context=context)
|
||||
leg_out = hkdf_legacy(length=length, derive_from=derive_from, salt=salt, context=context)
|
||||
|
||||
self.assertEqual(new_out, leg_out)
|
||||
verified += 1
|
||||
|
||||
for length in lengths_large:
|
||||
for dlen in derive_lens_large:
|
||||
for slen in salt_lens_large:
|
||||
for clen in context_lens_large:
|
||||
derive_from = rand_bytes(rng, dlen)
|
||||
salt = None if slen is None else (b"" if slen == 0 else rand_bytes(rng, slen))
|
||||
context = None if clen is None else (b"" if clen == 0 else rand_bytes(rng, clen))
|
||||
|
||||
new_out = hkdf(length=length, derive_from=derive_from, salt=salt, context=context)
|
||||
leg_out = hkdf_legacy(length=length, derive_from=derive_from, salt=salt, context=context)
|
||||
|
||||
self.assertEqual(new_out, leg_out)
|
||||
verified += 1
|
||||
|
||||
print(f"HKDF parity matrix: {verified} deterministic cases verified (optimized == legacy)")
|
||||
|
||||
def test_03_error_parity(self):
|
||||
print("")
|
||||
|
||||
invalid_cases = [
|
||||
dict(length=0, derive_from=b"x"),
|
||||
dict(length=None, derive_from=b"x"),
|
||||
dict(length=-1, derive_from=b"x"),
|
||||
dict(length="x", derive_from=b"x"),
|
||||
dict(length=32, derive_from=""),
|
||||
dict(length=32, derive_from=None),
|
||||
]
|
||||
|
||||
def raised_exception(fn, kwargs):
|
||||
try:
|
||||
fn(**kwargs)
|
||||
return None
|
||||
except Exception as e: return (type(e).__name__, str(e))
|
||||
|
||||
for kwargs in invalid_cases:
|
||||
new_exc = raised_exception(hkdf, kwargs)
|
||||
leg_exc = raised_exception(hkdf_legacy, kwargs)
|
||||
|
||||
# Both implementations must raise the exact same exception
|
||||
# type with the exact same message on invalid arguments
|
||||
self.assertIsNotNone(new_exc)
|
||||
self.assertEqual(new_exc, leg_exc)
|
||||
|
||||
print(f"HKDF error parity: {len(invalid_cases)} invalid-argument cases raise identical exception types and messages")
|
||||
|
||||
@unittest.skipIf(os.getenv("RNS_SKIP_PERF_TESTS") != None, "Skipping performance tests")
|
||||
def test_04_benchmark(self):
|
||||
print("")
|
||||
|
||||
rng = random.Random(HKDF_TEST_SEED ^ 0xA11CE)
|
||||
salt = RNS.Reticulum.IFAC_SALT
|
||||
|
||||
sizes = [1500, 16384, 262144]
|
||||
runs = {1500: 40, 16384: 20, 262144: 8}
|
||||
|
||||
print("HKDF implementation benchmark (optimized vs legacy):")
|
||||
print(f"{'length':>9} {'legacy avg':>12} {'opt avg':>9} {'speedup':>10}")
|
||||
print("-" * 48)
|
||||
|
||||
leg_total = new_total = 0.0
|
||||
for size in sizes:
|
||||
derive_from = rand_bytes(rng, 16)
|
||||
context = rand_bytes(rng, 7)
|
||||
n = runs[size]
|
||||
|
||||
hkdf_legacy(length=size, derive_from=derive_from, salt=salt, context=context)
|
||||
hkdf(length=size, derive_from=derive_from, salt=salt, context=context)
|
||||
|
||||
t0 = time.perf_counter()
|
||||
for _ in range(n):
|
||||
hkdf_legacy(length=size, derive_from=derive_from, salt=salt, context=context)
|
||||
leg_avg = (time.perf_counter() - t0) / n
|
||||
|
||||
t0 = time.perf_counter()
|
||||
for _ in range(n):
|
||||
hkdf(length=size, derive_from=derive_from, salt=salt, context=context)
|
||||
new_avg = (time.perf_counter() - t0) / n
|
||||
|
||||
leg_total += leg_avg
|
||||
new_total += new_avg
|
||||
|
||||
print(f"{size:>9} {leg_avg*1000:9.3f}m {new_avg*1000:9.3f}m {leg_avg/new_avg:8.1f}x")
|
||||
|
||||
print(f"\nThroughput at {sizes[-1]} bytes: legacy {sizes[-1]/leg_avg/1e6*8:.2f} Mbps, "
|
||||
f"new {sizes[-1]/new_avg/1e6*8:.2f} Mbps")
|
||||
print(f"Overall speedup across benchmarks: {leg_total/new_total:.1f}x")
|
||||
print()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main(verbosity=2)
|
||||
Loading…
Reference in New Issue