diff options
| -rw-r--r-- | analysis/countermeasures/countermeasures.ipynb | 1045 | ||||
| -rw-r--r-- | analysis/countermeasures/simulation.ipynb | 1910 |
2 files changed, 1910 insertions, 1045 deletions
diff --git a/analysis/countermeasures/countermeasures.ipynb b/analysis/countermeasures/countermeasures.ipynb deleted file mode 100644 index 58c654d..0000000 --- a/analysis/countermeasures/countermeasures.ipynb +++ /dev/null @@ -1,1045 +0,0 @@ -{ - "cells": [ - { - "cell_type": "markdown", - "id": "bafc2f4e-05a3-4120-bcd6-5d1f5fb91cd9", - "metadata": {}, - "source": [ - "# Distinguishing countermeasures by output" - ] - }, - { - "cell_type": "code", - "execution_count": 69, - "id": "33ee6084-2ac3-4f95-9610-0fbc06026538", - "metadata": {}, - "outputs": [], - "source": [ - "import io\n", - "import random\n", - "import itertools\n", - "import warnings\n", - "\n", - "import cypari2\n", - "from cysignals.alarm import alarm, AlarmInterrupt\n", - "\n", - "from matplotlib import pyplot as plt\n", - "from collections import Counter\n", - "from tqdm.auto import tqdm, trange\n", - "\n", - "from pyecsca.misc.utils import TaskExecutor\n", - "from pyecsca.ec.mod import mod, RandomModAction\n", - "from pyecsca.ec.point import Point\n", - "from pyecsca.ec.model import ShortWeierstrassModel\n", - "from pyecsca.ec.params import load_params_ectester\n", - "from pyecsca.ec.mult import LTRMultiplier\n", - "from pyecsca.ec.context import local, DefaultContext\n", - "from pyecsca.ec.countermeasures import GroupScalarRandomization, AdditiveSplitting, MultiplicativeSplitting, EuclideanSplitting, BrumleyTuveri\n", - "\n", - "%matplotlib ipympl" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "b1b9596c-1eba-4ace-af84-8cb279d84cc2", - "metadata": {}, - "outputs": [], - "source": [ - "model = ShortWeierstrassModel()\n", - "coords = model.coordinates[\"projective\"]" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "b0afb195-8390-44c5-931e-75a70ccd4e9e", - "metadata": {}, - "outputs": [], - "source": [ - "add = coords.formulas[\"add-2015-rcb\"]\n", - "dbl = coords.formulas[\"dbl-2015-rcb\"]\n", - "mult = LTRMultiplier(add, dbl, complete=False)" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "52c877e1-5021-4ec2-9daa-dd20bec6bcb2", - "metadata": {}, - "outputs": [], - "source": [ - "gsr = GroupScalarRandomization(mult)\n", - "asplit = AdditiveSplitting(mult)\n", - "msplit = MultiplicativeSplitting(mult)\n", - "esplit = EuclideanSplitting(mult)\n", - "bt = BrumleyTuveri(mult)" - ] - }, - { - "cell_type": "markdown", - "id": "27626337-dcbc-497c-a54e-02d50e2b8f34", - "metadata": {}, - "source": [ - "## 3n test" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "c3088419-161b-4193-a1b6-6f623f217fcd", - "metadata": {}, - "outputs": [], - "source": [ - "key3n = 0x20959f2b437de1e522baf6d814911938157390d3ea5118660b852ab0d5387006\n", - "params3n = load_params_ectester(io.BytesIO(b\"0xc381bb0394f34b5ed061c9107b66974f4d0a8ec89b9fe73b98b6d1368c7d974d,0x5ca6c5ee0a10097af291a8f125303fb1a3e35e8100411902245d691e0e5cb497,0x385a5a8bb8af94721f6fd10b562606d9b9df931f7fd966e96859bb9bd7c05836,0x4616af1898b92cac0f902a9daee24bbae63571cead270467c6a7886ced421f5e,0x34e896bdb1337e0ae5960fa3389fb59c2c8d6c7dbfd9aac33a844f8f98e433ef,0x412b3e5686fbc3ca4575edb0292232702ae721a7d4a230cc170a5561aa70e00f,0x01\"), \"projective\")\n", - "bits3n = params3n.full_order.bit_length()\n", - "point3n = Point(X=mod(0x4a48addb2e471767b7cd0f6f1d4c27fe46f4a828fc20f950bd1f72c939b36a84, params3n.curve.prime),\n", - " Y=mod(0x13384d38c353f862832c0f067e46a3e510bb6803c20745dfb31929f4a18d890d, params3n.curve.prime),\n", - " Z=mod(1, params3n.curve.prime), model=coords)" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "a8dde7e6-cd48-4f99-9677-23a19e4c2e5b", - "metadata": {}, - "outputs": [], - "source": [ - "print(f\"prime:\\t0x{params3n.curve.prime:x}\")\n", - "print(f\"a:\\t0x{params3n.curve.parameters['a']:x}\")\n", - "print(f\"b:\\t0x{params3n.curve.parameters['b']:x}\")\n", - "print(f\"G:\\t[0x{params3n.generator.X:x},\\n\\t 0x{params3n.generator.Y:x}]\")\n", - "print(f\"n:\\t0x{params3n.order:x}\")\n", - "print(f\"3n:\\t0x{3 * params3n.order:x}\")\n", - "print(f\"\\nP:\\t[0x{point3n.X:x},\\n\\t 0x{point3n.Y:x}]\")" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "cd6f8500-7509-45b0-8b23-471ee5014f42", - "metadata": {}, - "outputs": [], - "source": [ - "def generate_scalars_mod3(rem, samples):\n", - " scalars = []\n", - " while True:\n", - " scalar = random.randint(0, params3n.full_order)\n", - " if scalar % 3 == rem:\n", - " scalars.append(scalar)\n", - " if len(scalars) == samples:\n", - " break\n", - " return scalars\n", - "\n", - "def test_3n(countermeasure, scalars):\n", - " ctr = Counter()\n", - " for k in tqdm(scalars, leave=False):\n", - " mult.init(params3n, point3n)\n", - " kP = mult.multiply(k).to_affine()\n", - " mult.init(params3n, point3n)\n", - " knP = mult.multiply(k + params3n.full_order).to_affine()\n", - " mult.init(params3n, point3n)\n", - " k2nP = mult.multiply(k + 2 * params3n.full_order).to_affine()\n", - "\n", - " countermeasure.init(params3n, point3n)\n", - " res = countermeasure.multiply(k)\n", - " aff = res.to_affine()\n", - " if aff.equals(kP):\n", - " ctr[\"k\"] += 1\n", - " elif aff.equals(knP):\n", - " ctr[\"k + 1n\"] += 1\n", - " elif aff.equals(k2nP):\n", - " ctr[\"k + 2n\"] += 1\n", - " else:\n", - " ctr[aff] += 1\n", - " for name, count in sorted(ctr.items()):\n", - " print(f\"{name}:\\t{count}\")\n", - "\n", - "def test_3n_fixed_scalar(countermeasure, samples):\n", - " test_3n(countermeasure, [key3n for _ in range(samples)])\n", - "\n", - "def test_3n_random_scalar(countermeasure, samples):\n", - " test_3n(countermeasure, [random.randint(0, params3n.full_order) for _ in range(samples)])\n", - "\n", - "def test_3n_random_scalar_projected(countermeasure, samples):\n", - " print(\"k = 0 mod 3\")\n", - " test_3n(countermeasure, generate_scalars_mod3(0, samples))\n", - " print()\n", - " print(\"k = 1 mod 3\")\n", - " test_3n(countermeasure, generate_scalars_mod3(1, samples))\n", - " print()\n", - " print(\"k = 2 mod 3\")\n", - " test_3n(countermeasure, generate_scalars_mod3(2, samples))" - ] - }, - { - "cell_type": "markdown", - "id": "46b8f74a-433d-48c9-b5b9-6bb7d2731246", - "metadata": {}, - "source": [ - "### Fixed scalar experiments" - ] - }, - { - "cell_type": "markdown", - "id": "fc82d4b9-91cd-423c-83aa-89721efa1ae9", - "metadata": {}, - "source": [ - "#### Group scalar randomization" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "86532d50-2db7-4370-b449-c545b330a852", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_fixed_scalar(gsr, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "aba3e713-6246-435d-93af-2e8b42ee9582", - "metadata": {}, - "source": [ - "#### Additive splitting" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "ad421630-606f-4666-9bbf-1a446eec1b59", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_fixed_scalar(asplit, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "a1284a11-4ace-437d-9826-2035cce36756", - "metadata": {}, - "source": [ - "#### Multiplicative splitting" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "3ed5d7f3-0ba1-4b62-9635-aeb492499175", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_fixed_scalar(msplit, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "5e40f8e9-d26f-41e7-adbf-a0fbdb680677", - "metadata": {}, - "source": [ - "#### Euclidean splitting" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "314447c6-a1fb-4d3a-8988-b34c8912dd5e", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_fixed_scalar(esplit, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "ff7185a2-3cd9-44d7-b1a9-982eaed561dc", - "metadata": {}, - "source": [ - "#### Brumley and Tuveri bit-length fixing" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "f41dfc1d-1017-4aa0-bcd4-6569c53bf81e", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_fixed_scalar(bt, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "28915553-b5e6-4108-bc23-e5844b6a63b8", - "metadata": {}, - "source": [ - "### Random scalar experiments" - ] - }, - { - "cell_type": "markdown", - "id": "566ddd10-2d0e-4b32-9b27-60770ab68155", - "metadata": {}, - "source": [ - "#### Group scalar randomization" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "7255321a-6ad6-4938-8ec9-dd8d977686db", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_random_scalar(gsr, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "376c0da8-b92d-4151-ac30-839ab5c0ceae", - "metadata": {}, - "source": [ - "#### Additive splitting" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "b0146a9a-0803-43c4-ab29-8ba6e15934b5", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_random_scalar(asplit, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "96823352-cd63-4cf8-8ab7-cf6e025837b9", - "metadata": {}, - "source": [ - "#### Multiplicative splitting" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "5645ae6f-f5f4-419d-ba47-248532dc2114", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_random_scalar(msplit, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "1784d763-932a-4f66-a1da-d7ef51cbc88a", - "metadata": {}, - "source": [ - "#### Euclidean splitting" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "c9fc4f35-1c25-4cac-bb63-8bd70263db47", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_random_scalar(esplit, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "8917dfd5-1548-414b-846a-685857bfb427", - "metadata": {}, - "source": [ - "#### Brumley and Tuveri bit-length fixing" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "4fd6b288-08a9-4dbe-9145-e96401805315", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_random_scalar(bt, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "ee9e23f6-edd9-4fc2-9e7c-b5c176991071", - "metadata": {}, - "source": [ - "### Random scalar experiments projected to scalar divisor classes mod 3" - ] - }, - { - "cell_type": "markdown", - "id": "e17fcb12-1e02-4bcf-a2b7-785c16c03028", - "metadata": {}, - "source": [ - "#### Group scalar randomization" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "6c46fdbb-2ffb-4169-8e00-6d93b8407ee5", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_random_scalar_projected(gsr, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "b9d635c8-f788-4700-876b-ac48e89557a7", - "metadata": {}, - "source": [ - "#### Additive splitting" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "344a4f90-3470-40e9-a75f-b925a88c2480", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_random_scalar_projected(asplit, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "78aaf83f-a689-40ad-ae88-58ab20e5e6f9", - "metadata": {}, - "source": [ - "#### Multiplicative splitting" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "616a7726-01e6-4e9c-b7f2-fe8f14b60071", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_random_scalar_projected(msplit, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "018c8f31-d7c3-497b-bdaf-580c6465753f", - "metadata": {}, - "source": [ - "#### Euclidean splitting" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "adced4e4-37a7-43ed-97b5-01cb5d274d6b", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_random_scalar_projected(esplit, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "8d7b0a42-7be7-464d-932d-8386ad912034", - "metadata": {}, - "source": [ - "#### Brumley and Tuveri bit-length fixing" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "fe8d8295-3e69-4b60-b8c3-5710deaeb0b3", - "metadata": {}, - "outputs": [], - "source": [ - "test_3n_random_scalar_projected(bt, 1000)" - ] - }, - { - "cell_type": "markdown", - "id": "43b309af-5683-4384-9623-d7633723177c", - "metadata": {}, - "source": [ - "## Mask recovery\n", - "Using a composite order curve we can recover the size and the actual mask values (in a known key scenario) in both GSR and multiplicative splitting. However, real-world targets do not like composite order curves and may either check the order or otherwise fail to compute on such curves. Thus, we lie to them and set the order to the next-prime of the true order, in this case $n + 92$." - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "20a26f27-620d-4d7f-92bd-b949482b5c9a", - "metadata": {}, - "outputs": [], - "source": [ - "pari = cypari2.Pari(256_000_000, 2_000_000_000)" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "144340bd-5372-4beb-a46e-fd60c596b254", - "metadata": {}, - "outputs": [], - "source": [ - "real_n = 0xa9fa3419aca88bade2cba14e317816c6828910c6ce04fcd2a2e857d25df50775\n", - "# = 2898786277 * 2916913393 * 3067509271 * 3248233993 * 3894099889 * 4099407227 * 4101666977 * 13936975277\n", - "real_n_facts = pari.factor(real_n)\n", - "params92pn = load_params_ectester(io.BytesIO(b\"0xa9fa3419aca88bade2cba14e317816c79d52481d463dc9bcb12c37f45aa3b4e1,0x2ea3bfe6659f8e035735349b91fbfa2baf0cf8e640315f0fe03c1136813dec99,0x2b07c518e04b02158651e3dbbef7720015dd496bf15af02f8439f8e1503b8370,0x90fb04b1af19e8e20396ac052f260a9fb5f736b97e3cd4af08fe81a1e75dac6d,0x2302bcf700d3d5899f04d0c7441f5017c9758bfafd6ce15dbe36fb4eea76baec,0xa9fa3419aca88bade2cba14e317816c6828910c6ce04fcd2a2e857d25df507d1,0x01\"), \"projective\")\n", - "e = pari.ellinit([int(params92pn.curve.parameters[\"a\"]), int(params92pn.curve.parameters[\"b\"])], int(params92pn.curve.prime))\n", - "e[15][0] = real_n" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "f103129c-17d3-4217-999b-94ecb4ec523d", - "metadata": {}, - "outputs": [], - "source": [ - "print(f\"prime:\\t0x{params92pn.curve.prime:x}\")\n", - "print(f\"a:\\t0x{params92pn.curve.parameters['a']:x}\")\n", - "print(f\"b:\\t0x{params92pn.curve.parameters['b']:x}\")\n", - "print(f\"G:\\t[0x{params92pn.generator.X:x},\\n\\t 0x{params92pn.generator.Y:x}]\")\n", - "print(f\"n+92:\\t0x{params92pn.order:x} (fake order, given to the target, prime)\")\n", - "print(f\"n:\\t0x{real_n:x} (real order, composite)\")" - ] - }, - { - "cell_type": "markdown", - "id": "322d2e68-5259-4ea6-9748-2b0aa21b557f", - "metadata": {}, - "source": [ - "### Group scalar randomization\n", - "In GSR getting the mask out this way is quite simple. The target believes it is operating on a curve of order $n+92$ so it will use that value multiplied with the mask to randomize the scalar. Thus as a result we get:\n", - "$$ P = [k + r(n + 92)]G $$\n", - "\n", - "However, the curve is truly of order $n$, thus arithmetic on its group will make this equal to: \n", - "$$ P = [k + r 92]G $$\n", - "\n", - "Since this is a composite order curve, we can solve the dlog and obtain $k + r 92$ and since we assume we know $k$ we can easily compute both the mask size and mask value $r$." - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "08d99bd5-2b87-4a04-995d-7a87f9b67102", - "metadata": {}, - "outputs": [], - "source": [ - "key = 0x20959f2b437de1e522baf6d814911938157390d3ea5118660b852ab0d5387006 # any key works ofc\n", - "gsr.init(params92pn, params92pn.generator)\n", - "res = gsr.multiply(key)" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "2a869bed-8e21-46af-8f70-065f4afd6a82", - "metadata": {}, - "outputs": [], - "source": [ - "affine_gen = params92pn.generator.to_affine()\n", - "affine_res = res.to_affine()" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "e440399a-bc01-488b-8822-08cc0bf1672d", - "metadata": {}, - "outputs": [], - "source": [ - "dlog = pari.elllog(e,\n", - " [int(affine_res.x), int(affine_res.y)],\n", - " [int(affine_gen.x), int(affine_gen.y)],\n", - " real_n)" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "7ea6d6ae-a6f5-4b53-8c40-787d79970cb6", - "metadata": {}, - "outputs": [], - "source": [ - "mask = int((dlog - key) / 92)\n", - "mask_len = mask.bit_length()\n", - "print(mask)\n", - "print(mask_len)" - ] - }, - { - "cell_type": "markdown", - "id": "d40ec035-0656-4eda-8ef4-c14f9d53f49f", - "metadata": {}, - "source": [ - "### Multiplicative splitting\n", - "In multiplicative splitting the situation is a bit more complicated. Doing the same computation, where the target thinks the curve order is $n+92$ leads to:\n", - "$$ P = [k r^{-1}\\pmod{n+92}][r \\mod n]G $$\n", - "\n", - "Since the curve is composite order we can easily compute the dlog $d$ of P to G, we get:\n", - "$$ d = (k r^{-1})\\pmod{n+92}\\: r = k + t (n + 92) $$\n", - "\n", - "However, the dlog is computed $\\mod n$ so we really get: $ d = k + t 92$. We extract the $t$ out of this.\n", - "Note that $t$ will have roughly the same size as the mask $r$, since at the left side we have $(k r^{-1}) {\\mod (n+92)}$\n", - "of size $n$ and $r$ of size of the mask and on the right size we have $t n$ that dominates.\n", - "Thus at this point we have recovered the mask size.\n", - "However, $t$ is always smaller than $r$, sometimes also in bitsize.\n", - "\n", - "Now that we have $s$ we can go back to the original equation and get:\n", - "$$ (k r^{-1})\\pmod{n+92}\\: r = k + t (n + 92) $$\n", - "\n", - "We can then factor this value, lets call it $full$, and look for divisors that are larger than $t$ but smaller than the mask length\n", - "that we recovered before. There may be multiple candidates here and we don't know how to distinguish between\n", - "them. It holds for all of the candidates $c$ that the rest of the value $full$ is equal to the inverse of $c \\mod (n+92)$.\n", - "However, sometimes there is only one candidate, which is equal to the true mask value $r$." - ] - }, - { - "cell_type": "code", - "execution_count": 80, - "id": "b5f398fc-90d7-455e-97bd-62b682d55961", - "metadata": {}, - "outputs": [], - "source": [ - "def divisors(primes, powers):\n", - " for comb in itertools.product(*[range(power+1) for power in powers]):\n", - " value = 1\n", - " for prime, power in zip(primes, comb):\n", - " value *= prime**power\n", - " yield value\n", - "\n", - "def pari_factor(number, flag=1+8, duration=5*60):\n", - " # Use the 1 + 8 option to factorint to skip some costly factorization methods, we don't need the huge factors.\n", - " # This means that some factors may remain, which in turn means that sometimes we may not find the true mask.\n", - " pari = cypari2.Pari(256_000_000, 2_000_000_000)\n", - " with warnings.catch_warnings(category=RuntimeWarning):\n", - " warnings.simplefilter(\"ignore\")\n", - " try:\n", - " alarm(duration)\n", - " factors = pari.factorint(number, flag)\n", - " except AlarmInterrupt:\n", - " return None\n", - " primes = list(map(int, factors[0]))\n", - " powers = list(map(int, factors[1]))\n", - " return primes, powers\n", - "\n", - "def pari_dlog(params, P, G, real_n, facts_str):\n", - " pari = cypari2.Pari(256_000_000, 2_000_000_000)\n", - " e = pari.ellinit([int(params.curve.parameters[\"a\"]), int(params.curve.parameters[\"b\"])], int(params.curve.prime))\n", - " e[15][0] = real_n\n", - " facts = pari(facts_str)\n", - " dlog = pari.elllog(e, P, G, facts)\n", - " return int(dlog)" - ] - }, - { - "cell_type": "code", - "execution_count": 81, - "id": "5f03e586-33df-4525-a722-f5f63d6ca28d", - "metadata": {}, - "outputs": [ - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "c3b98a7bfa2e4e568f83ee3d035266bb", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Collecting scalarmults: 0%| | 0/1000 [00:00<?, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "3768e6da82324ae9891666bcf6d952b7", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Computing dlogs: 0%| | 0/1000 [00:00<?, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "32 True\n" - ] - } - ], - "source": [ - "key = 0x20959f2b437de1e522baf6d814911938157390d3ea5118660b852ab0d5387006 # any key works\n", - "msplit = MultiplicativeSplitting(mult, rand_bits=32) # change the mask size here to your liking\n", - "tries = 1000\n", - "num_workers = 30\n", - "\n", - "blens = [None for _ in range(tries)]\n", - "dlogs = [None for _ in range(tries)]\n", - "ts = [None for _ in range(tries)]\n", - "\n", - "results = [None for _ in range(tries)]\n", - "rs = [None for _ in range(tries)]\n", - "\n", - "with TaskExecutor(max_workers=num_workers) as pool:\n", - " for i in trange(tries, desc=\"Collecting scalarmults\"):\n", - " msplit.init(params92pn, params92pn.generator)\n", - " with local(DefaultContext()) as ctx:\n", - " res = msplit.multiply(key)\n", - " \n", - " affine_res = res.to_affine()\n", - " affine_gen = params92pn.generator.to_affine()\n", - " results[i] = affine_res\n", - " \n", - " rval = []\n", - " ctx.actions[0].walk(lambda action: rval.append(int(action.result)) if isinstance(action, RandomModAction) else None)\n", - " rs[i] = rval[0]\n", - " \n", - " pool.submit_task(i,\n", - " pari_dlog,\n", - " params92pn,\n", - " [int(affine_res.x), int(affine_res.y)],\n", - " [int(affine_gen.x), int(affine_gen.y)],\n", - " real_n,\n", - " repr(real_n_facts))\n", - " \n", - " for i, future in tqdm(pool.as_completed(), desc=\"Computing dlogs\", total=len(pool.tasks)):\n", - " dlog = future.result()\n", - " val = dlog - key\n", - " if val % 92 != 0:\n", - " print(f\"Bad val! {i}\")\n", - " t = val // 92\n", - " ts[i] = t\n", - " dlogs[i] = dlog\n", - " blens[i] = t.bit_length()\n", - "\n", - "mask_len = max(blens)\n", - "print(mask_len, mask_len == msplit.rand_bits)" - ] - }, - { - "cell_type": "code", - "execution_count": 85, - "id": "5fbf8a38-983d-49a6-9cac-5350f960dc3e", - "metadata": {}, - "outputs": [ - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "8237fd8893084365a474553995944794", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Factoring: 0%| | 0/1000 [00:00<?, ?it/s]" - ] - }, - "metadata": {}, - "output_type": "display_data" - } - ], - "source": [ - "with TaskExecutor(max_workers=num_workers) as pool:\n", - " fulls = []\n", - " for t in ts:\n", - " full = t * (real_n + 92) + key\n", - " fulls.append(full)\n", - " pool.submit_task(t,\n", - " pari_factor,\n", - " full, flag=0)\n", - " facts = [None for _ in ts]\n", - " for t, future in tqdm(pool.as_completed(), desc=\"Factoring\", total=len(ts)):\n", - " result = future.result()\n", - " if result is None:\n", - " print(\"Timed out.\")\n", - " facts[ts.index(t)] = result" - ] - }, - { - "cell_type": "code", - "execution_count": 86, - "id": "0973fe4b-cdf5-4e91-850b-25375eeabb7e", - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Only one candidate, we got the mask: 2842006421 True \n", - "Only one candidate, we got the mask: 1367232818 True \n", - "Only one candidate, we got the mask: 2240904829 True \n", - "Only one candidate, we got the mask: 1039351451 True \n", - "Only one candidate, we got the mask: 108996955 True \n", - "Only one candidate, we got the mask: 2881524907 True \n", - "Only one candidate, we got the mask: 2189838553 True \n", - "Only one candidate, we got the mask: 3969096149 True \n", - "Only one candidate, we got the mask: 2287426132 True \n", - "Only one candidate, we got the mask: 4074964099 True \n", - "Only one candidate, we got the mask: 2669990606 True \n", - "Only one candidate, we got the mask: 1456246563 True \n", - "Only one candidate, we got the mask: 2710951939 True \n", - "Only one candidate, we got the mask: 2825913889 True \n", - "Only one candidate, we got the mask: 1638243357 True \n", - "Only one candidate, we got the mask: 3811276388 True \n", - "Only one candidate, we got the mask: 3123140239 True \n", - "Only one candidate, we got the mask: 3595726430 True \n", - "Only one candidate, we got the mask: 4044959749 True \n", - "Only one candidate, we got the mask: 1018157261 True \n", - "Only one candidate, we got the mask: 3129245929 True \n", - "Only one candidate, we got the mask: 2809475199 True \n", - "Only one candidate, we got the mask: 3952225418 True \n", - "Only one candidate, we got the mask: 832268494 True \n", - "Only one candidate, we got the mask: 1264726273 True \n", - "Only one candidate, we got the mask: 616082151 True \n", - "Only one candidate, we got the mask: 2208359603 True \n", - "Only one candidate, we got the mask: 4071547089 True \n", - "Only one candidate, we got the mask: 12516967 True \n", - "Only one candidate, we got the mask: 3559760625 True \n", - "Only one candidate, we got the mask: 2760746593 True \n", - "Only one candidate, we got the mask: 3819292718 True \n", - "Only one candidate, we got the mask: 2244540383 True \n", - "Only one candidate, we got the mask: 3057851337 True \n", - "Only one candidate, we got the mask: 2604694145 True \n", - "Only one candidate, we got the mask: 2715382871 True \n", - "Only one candidate, we got the mask: 4149507669 True \n", - "Only one candidate, we got the mask: 2502506233 True \n", - "Only one candidate, we got the mask: 4060064857 True \n", - "Only one candidate, we got the mask: 3733544567 True \n", - "Only one candidate, we got the mask: 4274624339 True \n", - "Only one candidate, we got the mask: 2590849447 True \n", - "Only one candidate, we got the mask: 3426697997 True \n", - "Only one candidate, we got the mask: 3882456323 True \n", - "Only one candidate, we got the mask: 3269089457 True \n", - "Only one candidate, we got the mask: 3761164241 True \n", - "Only one candidate, we got the mask: 2802249770 True \n", - "Only one candidate, we got the mask: 2199311258 True \n", - "Only one candidate, we got the mask: 4044571507 True \n", - "Only one candidate, we got the mask: 2531874111 True \n", - "Only one candidate, we got the mask: 3110772661 True \n", - "Only one candidate, we got the mask: 1450255847 True \n", - "Only one candidate, we got the mask: 3506577773 True \n", - "Only one candidate, we got the mask: 864589123 True \n", - "Only one candidate, we got the mask: 2460981459 True \n", - "Only one candidate, we got the mask: 1852927003 True \n", - "Only one candidate, we got the mask: 4049366217 True \n", - "Only one candidate, we got the mask: 4139493739 True \n", - "Only one candidate, we got the mask: 1970200502 True \n", - "Only one candidate, we got the mask: 2982933973 True \n", - "Only one candidate, we got the mask: 1185834757 True \n", - "Only one candidate, we got the mask: 4057506562 True \n", - "Only one candidate, we got the mask: 3553777159 True \n", - "Only one candidate, we got the mask: 3228784961 True \n", - "Only one candidate, we got the mask: 3361181844 True \n", - "Only one candidate, we got the mask: 1946286323 True \n", - "Only one candidate, we got the mask: 3295077587 True \n", - "Only one candidate, we got the mask: 5406775 True \n", - "Only one candidate, we got the mask: 2712315191 True \n", - "Only one candidate, we got the mask: 97620847 True \n", - "Only one candidate, we got the mask: 3352788414 True \n", - "Only one candidate, we got the mask: 2067199152 True \n", - "Only one candidate, we got the mask: 2780511079 True \n", - "Only one candidate, we got the mask: 2422242923 True \n", - "Only one candidate, we got the mask: 1410524515 True \n", - "Only one candidate, we got the mask: 1313538691 True \n", - "Only one candidate, we got the mask: 2212262809 True \n", - "Only one candidate, we got the mask: 1889821971 True \n", - "Only one candidate, we got the mask: 4205618623 True \n", - "Only one candidate, we got the mask: 368456507 True \n", - "Only one candidate, we got the mask: 3659588587 True \n", - "Only one candidate, we got the mask: 3907266774 True \n", - "Only one candidate, we got the mask: 1421268181 True \n", - "Only one candidate, we got the mask: 3594406679 True \n", - "Only one candidate, we got the mask: 2835610927 True \n", - "Only one candidate, we got the mask: 1750708313 True \n", - "Only one candidate, we got the mask: 3994812597 True \n", - "Only one candidate, we got the mask: 3416127039 True \n", - "Only one candidate, we got the mask: 2918932879 True \n", - "Only one candidate, we got the mask: 2903903777 True \n", - "Only one candidate, we got the mask: 3751353247 True \n", - "Only one candidate, we got the mask: 1914133021 True \n", - "Only one candidate, we got the mask: 1041740814 True \n", - "Only one candidate, we got the mask: 2717146511 True \n", - "Only one candidate, we got the mask: 3787563800 True \n", - "Only one candidate, we got the mask: 2532401805 True \n", - "Only one candidate, we got the mask: 1580928500 True \n", - "Only one candidate, we got the mask: 3553660109 True \n", - "Only one candidate, we got the mask: 3548089540 True \n", - "Only one candidate, we got the mask: 3102746041 True \n", - "Only one candidate, we got the mask: 4065155265 True \n", - "Only one candidate, we got the mask: 3111552963 True \n", - "Only one candidate, we got the mask: 922382643 True \n", - "Only one candidate, we got the mask: 1896551582 True \n", - "Only one candidate, we got the mask: 4036077281 True \n", - "Only one candidate, we got the mask: 1202517347 True \n", - "Only one candidate, we got the mask: 4166638824 True \n", - "Only one candidate, we got the mask: 4196515943 True \n", - "Only one candidate, we got the mask: 2409770161 True \n", - "Only one candidate, we got the mask: 2156195057 True \n", - "Only one candidate, we got the mask: 3303218539 True \n", - "Only one candidate, we got the mask: 4019401523 True \n", - "Only one candidate, we got the mask: 3383631109 True \n", - "Only one candidate, we got the mask: 4045144939 True \n", - "Total recovered masks: 114 out of 1000\n" - ] - } - ], - "source": [ - "candidate_amounts = []\n", - "i = 0\n", - "for dlog, t, blen, r, fact, full, result in zip(dlogs, ts, blens, rs, facts, fulls, results):\n", - " if fact is None:\n", - " continue\n", - " \n", - " (primes, powers) = fact\n", - " complete = True\n", - " for prime in primes:\n", - " if not pari.isprime(prime):\n", - " complete = False\n", - " break\n", - "\n", - " i += 1\n", - " candidates = set()\n", - " for divisor in divisors(primes, powers):\n", - " if divisor.bit_length() <= mask_len and divisor >= t:\n", - " candidates.add(divisor)\n", - " candidate_amounts.append(len(candidates))\n", - " if len(candidates) == 1:\n", - " candidate = candidates.pop()\n", - " print(\"Only one candidate, we got the mask:\", candidate, candidate == r, \"incomplete factoring this may not be the r\" if not complete else \"\")\n", - "print(f\"Total recovered masks: {len(list(filter(lambda a: a == 1, candidate_amounts)))} out of {i}\")" - ] - }, - { - "cell_type": "code", - "execution_count": 89, - "id": "6274ff91-325f-4c6b-a4d7-d66b994d730f", - "metadata": {}, - "outputs": [ - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "3a03974d3d6c4f159824b17e4bfb0467", - "version_major": 2, - "version_minor": 0 - }, - "image/png": "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", - "text/html": [ - "\n", - " <div style=\"display: inline-block;\">\n", - " <div class=\"jupyter-widgets widget-label\" style=\"text-align: center;\">\n", - " Figure\n", - " </div>\n", - " <img src='data:image/png;base64,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' width=640.0/>\n", - " </div>\n", - " " - ], - "text/plain": [ - "Canvas(toolbar=Toolbar(toolitems=[('Home', 'Reset original view', 'home', 'home'), ('Back', 'Back to previous …" - ] - }, - "metadata": {}, - "output_type": "display_data" - } - ], - "source": [ - "max_amount = max(candidate_amounts)\n", - "fig = plt.subplots()\n", - "plt.hist(candidate_amounts, range=(1, max_amount), align=\"left\", density=True, bins=range(1, max_amount))#, bins=list(range(20)) + list(range(20, 100, 5)) + list(range(100, max(candidate_amounts), 10)))\n", - "plt.xlabel(\"number of candidate masks\")\n", - "plt.ylabel(\"proportion\")\n", - "plt.xticks(range(max_amount))\n", - "plt.xlim(0, 20);\n", - "plt.show()" - ] - }, - { - "cell_type": "code", - "execution_count": 90, - "id": "9f22ca9d-bdc2-4ea5-b2bc-249a256bb8ad", - "metadata": {}, - "outputs": [ - { - "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "2710c2726d1e452d8abbcb45d8cff8e0", - "version_major": 2, - "version_minor": 0 - }, - "image/png": "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", - "text/html": [ - "\n", - " <div style=\"display: inline-block;\">\n", - " <div class=\"jupyter-widgets widget-label\" style=\"text-align: center;\">\n", - " Figure\n", - " </div>\n", - " <img src='data:image/png;base64,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' width=640.0/>\n", - " </div>\n", - " " - ], - "text/plain": [ - "Canvas(toolbar=Toolbar(toolitems=[('Home', 'Reset original view', 'home', 'home'), ('Back', 'Back to previous …" - ] - }, - "metadata": {}, - "output_type": "display_data" - } - ], - "source": [ - "max_amount = max(candidate_amounts)\n", - "fig = plt.subplots()\n", - "plt.hist(candidate_amounts, range=(1, max_amount), align=\"left\", density=True, bins=range(1, max_amount))#, bins=list(range(20)) + list(range(20, 100, 5)) + list(range(100, max(candidate_amounts), 10)))\n", - "plt.xlabel(\"number of candidate masks\")\n", - "plt.ylabel(\"proportion\")\n", - "plt.show()" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "id": "b323aec4-e6d6-4722-b677-1ed1b04a5b02", - "metadata": {}, - "outputs": [], - "source": [] - } - ], - "metadata": { - "kernelspec": { - "display_name": "Python 3 (ipykernel)", - "language": "python", - "name": "python3" - }, - "language_info": { - "codemirror_mode": { - "name": "ipython", - "version": 3 - }, - "file_extension": ".py", - "mimetype": "text/x-python", - "name": "python", - "nbconvert_exporter": "python", - "pygments_lexer": "ipython3", - "version": "3.12.3" - } - }, - "nbformat": 4, - "nbformat_minor": 5 -} diff --git a/analysis/countermeasures/simulation.ipynb b/analysis/countermeasures/simulation.ipynb new file mode 100644 index 0000000..1d1a8d5 --- /dev/null +++ b/analysis/countermeasures/simulation.ipynb @@ -0,0 +1,1910 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "bafc2f4e-05a3-4120-bcd6-5d1f5fb91cd9", + "metadata": {}, + "source": [ + "# Distinguishing countermeasures by output" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "id": "33ee6084-2ac3-4f95-9610-0fbc06026538", + "metadata": {}, + "outputs": [], + "source": [ + "import io\n", + "import math\n", + "import random\n", + "import itertools\n", + "import warnings\n", + "\n", + "import cypari2\n", + "from cysignals.alarm import alarm, AlarmInterrupt\n", + "\n", + "from matplotlib import pyplot as plt\n", + "from collections import Counter\n", + "from tqdm.auto import tqdm, trange\n", + "\n", + "from pyecsca.misc.utils import TaskExecutor\n", + "from pyecsca.ec.mod import mod, RandomModAction\n", + "from pyecsca.ec.point import Point\n", + "from pyecsca.ec.model import ShortWeierstrassModel\n", + "from pyecsca.ec.params import load_params_ectester, get_params\n", + "from pyecsca.ec.mult import LTRMultiplier, RTLMultiplier, ScalarMultiplicationAction\n", + "from pyecsca.ec.context import local, DefaultContext\n", + "from pyecsca.ec.countermeasures import GroupScalarRandomization, AdditiveSplitting, MultiplicativeSplitting, EuclideanSplitting, BrumleyTuveri\n", + "\n", + "%matplotlib ipympl" + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "id": "b1b9596c-1eba-4ace-af84-8cb279d84cc2", + "metadata": {}, + "outputs": [], + "source": [ + "model = ShortWeierstrassModel()\n", + "coords = model.coordinates[\"projective\"]" + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "id": "b0afb195-8390-44c5-931e-75a70ccd4e9e", + "metadata": {}, + "outputs": [], + "source": [ + "add = coords.formulas[\"add-2007-bl\"]\n", + "dbl = coords.formulas[\"dbl-2007-bl\"]\n", + "ltr = LTRMultiplier(add, dbl, complete=False)\n", + "rtl = RTLMultiplier(add, dbl, complete=False)\n", + "mult = ltr" + ] + }, + { + "cell_type": "code", + "execution_count": 4, + "id": "52c877e1-5021-4ec2-9daa-dd20bec6bcb2", + "metadata": {}, + "outputs": [], + "source": [ + "gsr = GroupScalarRandomization(mult)\n", + "asplit = AdditiveSplitting(mult)\n", + "msplit = MultiplicativeSplitting(mult)\n", + "esplit = EuclideanSplitting(mult)\n", + "bt = BrumleyTuveri(mult)" + ] + }, + { + "cell_type": "markdown", + "id": "27626337-dcbc-497c-a54e-02d50e2b8f34", + "metadata": {}, + "source": [ + "## 3n test" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "id": "c3088419-161b-4193-a1b6-6f623f217fcd", + "metadata": {}, + "outputs": [], + "source": [ + "key3n = 0x20959f2b437de1e522baf6d814911938157390d3ea5118660b852ab0d5387006\n", + "params3n = load_params_ectester(io.BytesIO(b\"0xc381bb0394f34b5ed061c9107b66974f4d0a8ec89b9fe73b98b6d1368c7d974d,0x5ca6c5ee0a10097af291a8f125303fb1a3e35e8100411902245d691e0e5cb497,0x385a5a8bb8af94721f6fd10b562606d9b9df931f7fd966e96859bb9bd7c05836,0x4616af1898b92cac0f902a9daee24bbae63571cead270467c6a7886ced421f5e,0x34e896bdb1337e0ae5960fa3389fb59c2c8d6c7dbfd9aac33a844f8f98e433ef,0x412b3e5686fbc3ca4575edb0292232702ae721a7d4a230cc170a5561aa70e00f,0x01\"), \"projective\")\n", + "bits3n = params3n.full_order.bit_length()\n", + "point3n = Point(X=mod(0x4a48addb2e471767b7cd0f6f1d4c27fe46f4a828fc20f950bd1f72c939b36a84, params3n.curve.prime),\n", + " Y=mod(0x13384d38c353f862832c0f067e46a3e510bb6803c20745dfb31929f4a18d890d, params3n.curve.prime),\n", + " Z=mod(1, params3n.curve.prime), model=coords)" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "id": "a8dde7e6-cd48-4f99-9677-23a19e4c2e5b", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "prime:\t0xc381bb0394f34b5ed061c9107b66974f4d0a8ec89b9fe73b98b6d1368c7d974d\n", + "a:\t0x5ca6c5ee0a10097af291a8f125303fb1a3e35e8100411902245d691e0e5cb497\n", + "b:\t0x385a5a8bb8af94721f6fd10b562606d9b9df931f7fd966e96859bb9bd7c05836\n", + "G:\t[0x4616af1898b92cac0f902a9daee24bbae63571cead270467c6a7886ced421f5e,\n", + "\t 0x34e896bdb1337e0ae5960fa3389fb59c2c8d6c7dbfd9aac33a844f8f98e433ef]\n", + "n:\t0x412b3e5686fbc3ca4575edb0292232702ae721a7d4a230cc170a5561aa70e00f\n", + "3n:\t0xc381bb0394f34b5ed061c9107b66975080b564f77de69264451f0024ff52a02d\n", + "\n", + "P:\t[0x4a48addb2e471767b7cd0f6f1d4c27fe46f4a828fc20f950bd1f72c939b36a84,\n", + "\t 0x13384d38c353f862832c0f067e46a3e510bb6803c20745dfb31929f4a18d890d]\n" + ] + } + ], + "source": [ + "print(f\"prime:\\t0x{params3n.curve.prime:x}\")\n", + "print(f\"a:\\t0x{params3n.curve.parameters['a']:x}\")\n", + "print(f\"b:\\t0x{params3n.curve.parameters['b']:x}\")\n", + "print(f\"G:\\t[0x{params3n.generator.X:x},\\n\\t 0x{params3n.generator.Y:x}]\")\n", + "print(f\"n:\\t0x{params3n.order:x}\")\n", + "print(f\"3n:\\t0x{3 * params3n.order:x}\")\n", + "print(f\"\\nP:\\t[0x{point3n.X:x},\\n\\t 0x{point3n.Y:x}]\")" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "id": "cd6f8500-7509-45b0-8b23-471ee5014f42", + "metadata": {}, + "outputs": [], + "source": [ + "def generate_scalars_mod3(rem, samples):\n", + " scalars = []\n", + " while True:\n", + " scalar = random.randint(0, params3n.full_order)\n", + " if scalar % 3 == rem:\n", + " scalars.append(scalar)\n", + " if len(scalars) == samples:\n", + " break\n", + " return scalars\n", + "\n", + "def test_3n(countermeasure, scalars):\n", + " ctr = Counter()\n", + " for k in tqdm(scalars, leave=False):\n", + " mult.init(params3n, point3n)\n", + " kP = mult.multiply(k).to_affine()\n", + " mult.init(params3n, point3n)\n", + " knP = mult.multiply(k + params3n.full_order).to_affine()\n", + " mult.init(params3n, point3n)\n", + " k2nP = mult.multiply(k + 2 * params3n.full_order).to_affine()\n", + "\n", + " countermeasure.init(params3n, point3n)\n", + " res = countermeasure.multiply(k)\n", + " aff = res.to_affine()\n", + " if aff.equals(kP):\n", + " ctr[\"k\"] += 1\n", + " elif aff.equals(knP):\n", + " ctr[\"k + 1n\"] += 1\n", + " elif aff.equals(k2nP):\n", + " ctr[\"k + 2n\"] += 1\n", + " else:\n", + " ctr[aff] += 1\n", + " for name, count in sorted(ctr.items()):\n", + " print(f\"{name}:\\t{count}\")\n", + "\n", + "def test_3n_fixed_scalar(countermeasure, samples):\n", + " test_3n(countermeasure, [key3n for _ in range(samples)])\n", + "\n", + "def test_3n_random_scalar(countermeasure, samples):\n", + " test_3n(countermeasure, [random.randint(0, params3n.full_order) for _ in range(samples)])\n", + "\n", + "def test_3n_random_scalar_projected(countermeasure, samples):\n", + " print(\"k = 0 mod 3\")\n", + " test_3n(countermeasure, generate_scalars_mod3(0, samples))\n", + " print()\n", + " print(\"k = 1 mod 3\")\n", + " test_3n(countermeasure, generate_scalars_mod3(1, samples))\n", + " print()\n", + " print(\"k = 2 mod 3\")\n", + " test_3n(countermeasure, generate_scalars_mod3(2, samples))" + ] + }, + { + "cell_type": "markdown", + "id": "46b8f74a-433d-48c9-b5b9-6bb7d2731246", + "metadata": {}, + "source": [ + "### Fixed scalar experiments" + ] + }, + { + "cell_type": "markdown", + "id": "fc82d4b9-91cd-423c-83aa-89721efa1ae9", + "metadata": {}, + "source": [ + "#### Group scalar randomization" + ] + }, + { + "cell_type": "code", + "execution_count": 8, + "id": "86532d50-2db7-4370-b449-c545b330a852", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "7c11b83ca4fa40eaac982481494bd7ce", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t354\n", + "k + 1n:\t318\n", + "k + 2n:\t328\n" + ] + } + ], + "source": [ + "test_3n_fixed_scalar(gsr, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "aba3e713-6246-435d-93af-2e8b42ee9582", + "metadata": {}, + "source": [ + "#### Additive splitting" + ] + }, + { + "cell_type": "code", + "execution_count": 9, + "id": "ad421630-606f-4666-9bbf-1a446eec1b59", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "4ac4f4baa0864f4ab34d422e13507754", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t508\n", + "k + 1n:\t492\n" + ] + } + ], + "source": [ + "test_3n_fixed_scalar(asplit, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "a1284a11-4ace-437d-9826-2035cce36756", + "metadata": {}, + "source": [ + "#### Multiplicative splitting" + ] + }, + { + "cell_type": "code", + "execution_count": 10, + "id": "3ed5d7f3-0ba1-4b62-9635-aeb492499175", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "4193d9eadc644cad957e15d9c9ca98a7", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t192\n", + "k + 1n:\t575\n", + "k + 2n:\t233\n" + ] + } + ], + "source": [ + "test_3n_fixed_scalar(msplit, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "5e40f8e9-d26f-41e7-adbf-a0fbdb680677", + "metadata": {}, + "source": [ + "#### Euclidean splitting" + ] + }, + { + "cell_type": "code", + "execution_count": 11, + "id": "314447c6-a1fb-4d3a-8988-b34c8912dd5e", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "8fce34bb47b44e17a2eaf9b20041e5db", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t1000\n" + ] + } + ], + "source": [ + "test_3n_fixed_scalar(esplit, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "ff7185a2-3cd9-44d7-b1a9-982eaed561dc", + "metadata": {}, + "source": [ + "#### Brumley and Tuveri bit-length fixing" + ] + }, + { + "cell_type": "code", + "execution_count": 12, + "id": "f41dfc1d-1017-4aa0-bcd4-6569c53bf81e", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "f6bb27fe24f7463e85b1b37848f5444d", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k + 2n:\t1000\n" + ] + } + ], + "source": [ + "test_3n_fixed_scalar(bt, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "28915553-b5e6-4108-bc23-e5844b6a63b8", + "metadata": {}, + "source": [ + "### Random scalar experiments" + ] + }, + { + "cell_type": "markdown", + "id": "566ddd10-2d0e-4b32-9b27-60770ab68155", + "metadata": {}, + "source": [ + "#### Group scalar randomization" + ] + }, + { + "cell_type": "code", + "execution_count": 13, + "id": "7255321a-6ad6-4938-8ec9-dd8d977686db", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "0c52e094bfd84c70874bd1444cec7c18", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t331\n", + "k + 1n:\t339\n", + "k + 2n:\t330\n" + ] + } + ], + "source": [ + "test_3n_random_scalar(gsr, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "376c0da8-b92d-4151-ac30-839ab5c0ceae", + "metadata": {}, + "source": [ + "#### Additive splitting" + ] + }, + { + "cell_type": "code", + "execution_count": 14, + "id": "b0146a9a-0803-43c4-ab29-8ba6e15934b5", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "c98aabf325d14c85844ce03540166769", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t499\n", + "k + 1n:\t501\n" + ] + } + ], + "source": [ + "test_3n_random_scalar(asplit, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "96823352-cd63-4cf8-8ab7-cf6e025837b9", + "metadata": {}, + "source": [ + "#### Multiplicative splitting" + ] + }, + { + "cell_type": "code", + "execution_count": 15, + "id": "5645ae6f-f5f4-419d-ba47-248532dc2114", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "ad009d473f5345f89f72f5d892886672", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t313\n", + "k + 1n:\t352\n", + "k + 2n:\t335\n" + ] + } + ], + "source": [ + "test_3n_random_scalar(msplit, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "1784d763-932a-4f66-a1da-d7ef51cbc88a", + "metadata": {}, + "source": [ + "#### Euclidean splitting" + ] + }, + { + "cell_type": "code", + "execution_count": 16, + "id": "c9fc4f35-1c25-4cac-bb63-8bd70263db47", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "59c26f78ce4d48e6b0f48c56cb2a1957", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t1000\n" + ] + } + ], + "source": [ + "test_3n_random_scalar(esplit, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "8917dfd5-1548-414b-846a-685857bfb427", + "metadata": {}, + "source": [ + "#### Brumley and Tuveri bit-length fixing" + ] + }, + { + "cell_type": "code", + "execution_count": 17, + "id": "4fd6b288-08a9-4dbe-9145-e96401805315", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "0a9b4d4abf5544658185577fb6a6d7c8", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k + 1n:\t28\n", + "k + 2n:\t972\n" + ] + } + ], + "source": [ + "test_3n_random_scalar(bt, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "ee9e23f6-edd9-4fc2-9e7c-b5c176991071", + "metadata": {}, + "source": [ + "### Random scalar experiments projected to scalar divisor classes mod 3" + ] + }, + { + "cell_type": "markdown", + "id": "e17fcb12-1e02-4bcf-a2b7-785c16c03028", + "metadata": {}, + "source": [ + "#### Group scalar randomization" + ] + }, + { + "cell_type": "code", + "execution_count": 18, + "id": "6c46fdbb-2ffb-4169-8e00-6d93b8407ee5", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k = 0 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "d99eea3415564fac926c237689f65b3c", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t312\n", + "k + 1n:\t357\n", + "k + 2n:\t331\n", + "\n", + "k = 1 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "9d75e1bf8df240bf9c19d3885d253186", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t291\n", + "k + 1n:\t358\n", + "k + 2n:\t351\n", + "\n", + "k = 2 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "d43f5b16d2d444cda559bf18976a2e2a", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t342\n", + "k + 1n:\t338\n", + "k + 2n:\t320\n" + ] + } + ], + "source": [ + "test_3n_random_scalar_projected(gsr, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "b9d635c8-f788-4700-876b-ac48e89557a7", + "metadata": {}, + "source": [ + "#### Additive splitting" + ] + }, + { + "cell_type": "code", + "execution_count": 19, + "id": "344a4f90-3470-40e9-a75f-b925a88c2480", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k = 0 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "e60f8fd6e60f4edc8c841a194e874a70", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t500\n", + "k + 1n:\t500\n", + "\n", + "k = 1 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "f08cebf9ab274dad8164b57f18417627", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t528\n", + "k + 1n:\t472\n", + "\n", + "k = 2 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "772c89d8a3a048f5934cb451f0b12be3", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t531\n", + "k + 1n:\t469\n" + ] + } + ], + "source": [ + "test_3n_random_scalar_projected(asplit, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "78aaf83f-a689-40ad-ae88-58ab20e5e6f9", + "metadata": {}, + "source": [ + "#### Multiplicative splitting" + ] + }, + { + "cell_type": "code", + "execution_count": 20, + "id": "616a7726-01e6-4e9c-b7f2-fe8f14b60071", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k = 0 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "0dfabf99286f48c9b2015742783669b9", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t571\n", + "k + 1n:\t190\n", + "k + 2n:\t239\n", + "\n", + "k = 1 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "d310342f047747a097f635e63540c42a", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t217\n", + "k + 1n:\t232\n", + "k + 2n:\t551\n", + "\n", + "k = 2 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "5581e145df664b6c87957f1791aea1b3", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t209\n", + "k + 1n:\t546\n", + "k + 2n:\t245\n" + ] + } + ], + "source": [ + "test_3n_random_scalar_projected(msplit, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "018c8f31-d7c3-497b-bdaf-580c6465753f", + "metadata": {}, + "source": [ + "#### Euclidean splitting" + ] + }, + { + "cell_type": "code", + "execution_count": 21, + "id": "adced4e4-37a7-43ed-97b5-01cb5d274d6b", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k = 0 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "29f009a782ee4dd4bff9938d655db831", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t1000\n", + "\n", + "k = 1 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "f2d438308994479b8e1086621b24466e", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t1000\n", + "\n", + "k = 2 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "00f137d1135741ada0978bd00c9ff194", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k:\t1000\n" + ] + } + ], + "source": [ + "test_3n_random_scalar_projected(esplit, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "8d7b0a42-7be7-464d-932d-8386ad912034", + "metadata": {}, + "source": [ + "#### Brumley and Tuveri bit-length fixing" + ] + }, + { + "cell_type": "code", + "execution_count": 22, + "id": "fe8d8295-3e69-4b60-b8c3-5710deaeb0b3", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k = 0 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "4448b9a556cc4acd8d31e60277e09488", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k + 1n:\t35\n", + "k + 2n:\t965\n", + "\n", + "k = 1 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "1dbe7fc3fbbf4f92a889ca37b66a27b3", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k + 1n:\t36\n", + "k + 2n:\t964\n", + "\n", + "k = 2 mod 3\n" + ] + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "56b3afd278b8491e80dabc3b1d676631", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + " 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "k + 1n:\t37\n", + "k + 2n:\t963\n" + ] + } + ], + "source": [ + "test_3n_random_scalar_projected(bt, 1000)" + ] + }, + { + "cell_type": "markdown", + "id": "43b309af-5683-4384-9623-d7633723177c", + "metadata": {}, + "source": [ + "## Mask recovery ($n + \\epsilon$ test)\n", + "Using a composite order curve we can recover the size and the actual mask values (in a known key scenario) in both GSR and multiplicative splitting. However, real-world targets do not like composite order curves and may either check the order or otherwise fail to compute on such curves. Thus, we lie to them and set the order to the next-prime of the true order, in this case $n + 92$." + ] + }, + { + "cell_type": "code", + "execution_count": 23, + "id": "20a26f27-620d-4d7f-92bd-b949482b5c9a", + "metadata": {}, + "outputs": [], + "source": [ + "pari = cypari2.Pari(256_000_000, 2_000_000_000)" + ] + }, + { + "cell_type": "code", + "execution_count": 24, + "id": "144340bd-5372-4beb-a46e-fd60c596b254", + "metadata": {}, + "outputs": [], + "source": [ + "real_n = 0xa9fa3419aca88bade2cba14e317816c6828910c6ce04fcd2a2e857d25df50775\n", + "# = 2898786277 * 2916913393 * 3067509271 * 3248233993 * 3894099889 * 4099407227 * 4101666977 * 13936975277\n", + "real_n_facts = pari.factor(real_n)\n", + "params92pn = load_params_ectester(io.BytesIO(b\"0xa9fa3419aca88bade2cba14e317816c79d52481d463dc9bcb12c37f45aa3b4e1,0x2ea3bfe6659f8e035735349b91fbfa2baf0cf8e640315f0fe03c1136813dec99,0x2b07c518e04b02158651e3dbbef7720015dd496bf15af02f8439f8e1503b8370,0x90fb04b1af19e8e20396ac052f260a9fb5f736b97e3cd4af08fe81a1e75dac6d,0x2302bcf700d3d5899f04d0c7441f5017c9758bfafd6ce15dbe36fb4eea76baec,0xa9fa3419aca88bade2cba14e317816c6828910c6ce04fcd2a2e857d25df507d1,0x01\"), \"projective\")\n", + "e = pari.ellinit([int(params92pn.curve.parameters[\"a\"]), int(params92pn.curve.parameters[\"b\"])], int(params92pn.curve.prime))\n", + "e[15][0] = real_n" + ] + }, + { + "cell_type": "code", + "execution_count": 25, + "id": "f103129c-17d3-4217-999b-94ecb4ec523d", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "prime:\t0xa9fa3419aca88bade2cba14e317816c79d52481d463dc9bcb12c37f45aa3b4e1\n", + "a:\t0x2ea3bfe6659f8e035735349b91fbfa2baf0cf8e640315f0fe03c1136813dec99\n", + "b:\t0x2b07c518e04b02158651e3dbbef7720015dd496bf15af02f8439f8e1503b8370\n", + "G:\t[0x90fb04b1af19e8e20396ac052f260a9fb5f736b97e3cd4af08fe81a1e75dac6d,\n", + "\t 0x2302bcf700d3d5899f04d0c7441f5017c9758bfafd6ce15dbe36fb4eea76baec]\n", + "n+92:\t0xa9fa3419aca88bade2cba14e317816c6828910c6ce04fcd2a2e857d25df507d1 (fake order, given to the target, prime)\n", + "n:\t0xa9fa3419aca88bade2cba14e317816c6828910c6ce04fcd2a2e857d25df50775 (real order, composite)\n" + ] + } + ], + "source": [ + "print(f\"prime:\\t0x{params92pn.curve.prime:x}\")\n", + "print(f\"a:\\t0x{params92pn.curve.parameters['a']:x}\")\n", + "print(f\"b:\\t0x{params92pn.curve.parameters['b']:x}\")\n", + "print(f\"G:\\t[0x{params92pn.generator.X:x},\\n\\t 0x{params92pn.generator.Y:x}]\")\n", + "print(f\"n+92:\\t0x{params92pn.order:x} (fake order, given to the target, prime)\")\n", + "print(f\"n:\\t0x{real_n:x} (real order, composite)\")" + ] + }, + { + "cell_type": "markdown", + "id": "322d2e68-5259-4ea6-9748-2b0aa21b557f", + "metadata": {}, + "source": [ + "### Group scalar randomization\n", + "In GSR getting the mask out this way is quite simple. The target believes it is operating on a curve of order $n+92$ so it will use that value multiplied with the mask to randomize the scalar. Thus as a result we get:\n", + "$$ P = [k + r(n + 92)]G $$\n", + "\n", + "However, the curve is truly of order $n$, thus arithmetic on its group will make this equal to: \n", + "$$ P = [k + r 92]G $$\n", + "\n", + "Since this is a composite order curve, we can solve the dlog and obtain $k + r 92$ and since we assume we know $k$ we can easily compute both the mask size and mask value $r$." + ] + }, + { + "cell_type": "code", + "execution_count": 26, + "id": "08d99bd5-2b87-4a04-995d-7a87f9b67102", + "metadata": {}, + "outputs": [], + "source": [ + "key = 0x20959f2b437de1e522baf6d814911938157390d3ea5118660b852ab0d5387006 # any key works ofc\n", + "gsr.init(params92pn, params92pn.generator)\n", + "res = gsr.multiply(key)" + ] + }, + { + "cell_type": "code", + "execution_count": 27, + "id": "2a869bed-8e21-46af-8f70-065f4afd6a82", + "metadata": {}, + "outputs": [], + "source": [ + "affine_gen = params92pn.generator.to_affine()\n", + "affine_res = res.to_affine()" + ] + }, + { + "cell_type": "code", + "execution_count": 28, + "id": "e440399a-bc01-488b-8822-08cc0bf1672d", + "metadata": {}, + "outputs": [], + "source": [ + "dlog = pari.elllog(e,\n", + " [int(affine_res.x), int(affine_res.y)],\n", + " [int(affine_gen.x), int(affine_gen.y)],\n", + " real_n)" + ] + }, + { + "cell_type": "code", + "execution_count": 29, + "id": "7ea6d6ae-a6f5-4b53-8c40-787d79970cb6", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "2799400670\n", + "32\n" + ] + } + ], + "source": [ + "mask = int((dlog - key) / 92)\n", + "mask_len = mask.bit_length()\n", + "print(mask)\n", + "print(mask_len)" + ] + }, + { + "cell_type": "markdown", + "id": "d40ec035-0656-4eda-8ef4-c14f9d53f49f", + "metadata": {}, + "source": [ + "### Multiplicative splitting\n", + "In multiplicative splitting the situation is a bit more complicated. Doing the same computation, where the target thinks the curve order is $n+92$ leads to:\n", + "$$ P = [k r^{-1}\\pmod{n+92}][r \\mod n]G $$\n", + "\n", + "Since the curve is composite order we can easily compute the dlog $d$ of P to G, we get:\n", + "$$ d = (k r^{-1})\\pmod{n+92}\\: r = k + t (n + 92) $$\n", + "\n", + "However, the dlog is computed $\\mod n$ so we really get: $ d = k + t 92$. We extract the $t$ out of this.\n", + "Note that $t$ will have roughly the same size as the mask $r$, since at the left side we have $(k r^{-1}) {\\mod (n+92)}$\n", + "of size $n$ and $r$ of size of the mask and on the right size we have $t n$ that dominates.\n", + "Thus at this point we have recovered the mask size.\n", + "However, $t$ is always smaller than $r$, sometimes also in bitsize.\n", + "\n", + "Now that we have $s$ we can go back to the original equation and get:\n", + "$$ (k r^{-1})\\pmod{n+92}\\: r = k + t (n + 92) $$\n", + "\n", + "We can then factor this value, lets call it $full$, and look for divisors that are larger than $t$ but smaller than the mask length\n", + "that we recovered before. There may be multiple candidates here and we don't know how to distinguish between\n", + "them. It holds for all of the candidates $c$ that the rest of the value $full$ is equal to the inverse of $c \\mod (n+92)$.\n", + "However, sometimes there is only one candidate, which is equal to the true mask value $r$." + ] + }, + { + "cell_type": "code", + "execution_count": 30, + "id": "b5f398fc-90d7-455e-97bd-62b682d55961", + "metadata": {}, + "outputs": [], + "source": [ + "def divisors(primes, powers):\n", + " for comb in itertools.product(*[range(power+1) for power in powers]):\n", + " value = 1\n", + " for prime, power in zip(primes, comb):\n", + " value *= prime**power\n", + " yield value\n", + "\n", + "def pari_factor(number, flag=1+8, duration=5*60):\n", + " # Use the 1 + 8 option to factorint to skip some costly factorization methods, we don't need the huge factors.\n", + " # This means that some factors may remain, which in turn means that sometimes we may not find the true mask.\n", + " pari = cypari2.Pari(256_000_000, 2_000_000_000)\n", + " with warnings.catch_warnings(category=RuntimeWarning):\n", + " warnings.simplefilter(\"ignore\")\n", + " try:\n", + " alarm(duration)\n", + " factors = pari.factorint(number, flag)\n", + " except AlarmInterrupt:\n", + " return None\n", + " primes = list(map(int, factors[0]))\n", + " powers = list(map(int, factors[1]))\n", + " return primes, powers\n", + "\n", + "def pari_dlog(params, P, G, real_n, facts_str):\n", + " pari = cypari2.Pari(256_000_000, 2_000_000_000)\n", + " e = pari.ellinit([int(params.curve.parameters[\"a\"]), int(params.curve.parameters[\"b\"])], int(params.curve.prime))\n", + " e[15][0] = real_n\n", + " facts = pari(facts_str)\n", + " dlog = pari.elllog(e, P, G, facts)\n", + " return int(dlog)" + ] + }, + { + "cell_type": "code", + "execution_count": 31, + "id": "5f03e586-33df-4525-a722-f5f63d6ca28d", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "28ab3630b5084b6aa80ae732f85fd8c9", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + "Collecting scalarmults: 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "b79a4b8ae44c494e81087a5c0e413ec9", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + "Computing dlogs: 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "32 True\n" + ] + } + ], + "source": [ + "key = 0x20959f2b437de1e522baf6d814911938157390d3ea5118660b852ab0d5387006 # any key works\n", + "msplit = MultiplicativeSplitting(mult, rand_bits=32) # change the mask size here to your liking\n", + "tries = 1000\n", + "num_workers = 30\n", + "\n", + "blens = [None for _ in range(tries)]\n", + "dlogs = [None for _ in range(tries)]\n", + "ts = [None for _ in range(tries)]\n", + "\n", + "results = [None for _ in range(tries)]\n", + "rs = [None for _ in range(tries)]\n", + "\n", + "with TaskExecutor(max_workers=num_workers) as pool:\n", + " for i in trange(tries, desc=\"Collecting scalarmults\"):\n", + " msplit.init(params92pn, params92pn.generator)\n", + " with local(DefaultContext()) as ctx:\n", + " res = msplit.multiply(key)\n", + " \n", + " affine_res = res.to_affine()\n", + " affine_gen = params92pn.generator.to_affine()\n", + " results[i] = affine_res\n", + " \n", + " rval = []\n", + " ctx.actions[0].walk(lambda action: rval.append(int(action.result)) if isinstance(action, RandomModAction) else None)\n", + " rs[i] = rval[0]\n", + " \n", + " pool.submit_task(i,\n", + " pari_dlog,\n", + " params92pn,\n", + " [int(affine_res.x), int(affine_res.y)],\n", + " [int(affine_gen.x), int(affine_gen.y)],\n", + " real_n,\n", + " repr(real_n_facts))\n", + " \n", + " for i, future in tqdm(pool.as_completed(), desc=\"Computing dlogs\", total=len(pool.tasks)):\n", + " dlog = future.result()\n", + " val = dlog - key\n", + " if val % 92 != 0:\n", + " print(f\"Bad val! {i}\")\n", + " t = val // 92\n", + " ts[i] = t\n", + " dlogs[i] = dlog\n", + " blens[i] = t.bit_length()\n", + "\n", + "mask_len = max(blens)\n", + "print(mask_len, mask_len == msplit.rand_bits)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "5fbf8a38-983d-49a6-9cac-5350f960dc3e", + "metadata": {}, + "outputs": [ + { + "data": { + "application/vnd.jupyter.widget-view+json": { + "model_id": "239ea218388d4c988e39e3ec54765325", + "version_major": 2, + "version_minor": 0 + }, + "text/plain": [ + "Factoring: 0%| | 0/1000 [00:00<?, ?it/s]" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "with TaskExecutor(max_workers=num_workers) as pool:\n", + " fulls = []\n", + " for t in ts:\n", + " full = t * (real_n + 92) + key\n", + " fulls.append(full)\n", + " pool.submit_task(t,\n", + " pari_factor,\n", + " full, flag=0)\n", + " facts = [None for _ in ts]\n", + " for t, future in tqdm(pool.as_completed(), desc=\"Factoring\", total=len(ts)):\n", + " result = future.result()\n", + " if result is None:\n", + " print(\"Timed out.\")\n", + " facts[ts.index(t)] = result" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "0973fe4b-cdf5-4e91-850b-25375eeabb7e", + "metadata": {}, + "outputs": [], + "source": [ + "candidate_amounts = []\n", + "i = 0\n", + "for dlog, t, blen, r, fact, full, result in zip(dlogs, ts, blens, rs, facts, fulls, results):\n", + " if fact is None:\n", + " continue\n", + " \n", + " (primes, powers) = fact\n", + " complete = True\n", + " for prime in primes:\n", + " if not pari.isprime(prime):\n", + " complete = False\n", + " break\n", + "\n", + " i += 1\n", + " candidates = set()\n", + " for divisor in divisors(primes, powers):\n", + " if divisor.bit_length() <= mask_len and divisor >= t:\n", + " candidates.add(divisor)\n", + " candidate_amounts.append(len(candidates))\n", + " if len(candidates) == 1:\n", + " candidate = candidates.pop()\n", + " print(\"Only one candidate, we got the mask:\", candidate, candidate == r, \"incomplete factoring this may not be the r\" if not complete else \"\")\n", + "print(f\"Total recovered masks: {len(list(filter(lambda a: a == 1, candidate_amounts)))} out of {i}\")" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "6274ff91-325f-4c6b-a4d7-d66b994d730f", + "metadata": {}, + "outputs": [], + "source": [ + "max_amount = max(candidate_amounts)\n", + "fig = plt.subplots()\n", + "plt.hist(candidate_amounts, range=(1, max_amount), align=\"left\", density=True, bins=range(1, max_amount))#, bins=list(range(20)) + list(range(20, 100, 5)) + list(range(100, max(candidate_amounts), 10)))\n", + "plt.xlabel(\"number of candidate masks\")\n", + "plt.ylabel(\"proportion\")\n", + "plt.xticks(range(max_amount))\n", + "plt.xlim(0, 20);\n", + "plt.show()" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "9f22ca9d-bdc2-4ea5-b2bc-249a256bb8ad", + "metadata": {}, + "outputs": [], + "source": [ + "max_amount = max(candidate_amounts)\n", + "fig = plt.subplots()\n", + "plt.hist(candidate_amounts, range=(1, max_amount), align=\"left\", density=True, bins=range(1, max_amount))#, bins=list(range(20)) + list(range(20, 100, 5)) + list(range(100, max(candidate_amounts), 10)))\n", + "plt.xlabel(\"number of candidate masks\")\n", + "plt.ylabel(\"proportion\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "91bbda3c-5670-49c9-85f0-99fb36a8acf9", + "metadata": {}, + "source": [ + "## $k = 10$ test" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "92781e99-15fc-4499-a24e-5ccc20ed3707", + "metadata": {}, + "outputs": [], + "source": [ + "paramsk10 = get_params(\"secg\", \"secp256r1\", \"projective\")" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "b31b2023-c0f7-47b2-8e99-b1f7f6e734b6", + "metadata": {}, + "outputs": [], + "source": [ + "def test_k10(countermeasure, samples, k_range = None, zero_fails = True, plot=True):\n", + " G = paramsk10.generator\n", + " Gaff = G.to_affine()\n", + " if k_range is None:\n", + " ks = list(range(2, 21))\n", + " else:\n", + " ks = list(k_range)\n", + " fails = []\n", + " for k in ks:\n", + " correct = 0\n", + " failed = 0\n", + " expected = paramsk10.curve.affine_multiply(Gaff, k).to_model(paramsk10.curve.coordinate_model, paramsk10.curve)\n", + " for _ in range(samples):\n", + " with local(DefaultContext()) as ctx:\n", + " countermeasure.init(paramsk10, paramsk10.generator)\n", + " res = countermeasure.multiply(k)\n", + " smults = set()\n", + " ctx.actions[0].walk(lambda action: smults.add(action.scalar) if isinstance(action, ScalarMultiplicationAction) else None)\n", + " if 0 in smults and zero_fails:\n", + " failed += 1\n", + " continue\n", + " try:\n", + " if res.equals_scaled(expected):\n", + " correct += 1\n", + " else:\n", + " failed += 1\n", + " except:\n", + " failed += 1\n", + " print(f\"k = {k}: failed in {failed} out of {samples}.\")\n", + " fails.append(failed / samples)\n", + " if plot:\n", + " fig, ax = plt.subplots()\n", + " xs = list(range(len(ks)))\n", + " ax.plot(xs, fails, label=\"Error rate\")\n", + " if any(k > 100 for k in ks):\n", + " ax.set_xticks(xs, (f\"2^{int(math.log2(k))}\" for k in ks))\n", + " else:\n", + " ax.set_xticks(xs, ks)\n", + " ax.set_ylim(-0.05, 1.05)\n", + " ax.set_xlabel(\"k\")\n", + " ax.set_ylabel(\"Error rate\")\n", + " ax.legend()\n", + " plt.show()\n", + " return fails" + ] + }, + { + "cell_type": "markdown", + "id": "37e1ce04-5487-484c-9bf1-d6a8f7076390", + "metadata": {}, + "source": [ + "### Group scalar randomization\n", + "\n", + "In GSR, the LTR simple double-and-add errors quite a lot for small scalars." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "d4f5298c-6277-40f4-af5d-590df6d61d6e", + "metadata": {}, + "outputs": [], + "source": [ + "test_k10(gsr, 100);" + ] + }, + { + "cell_type": "markdown", + "id": "fa4e3ee7-2900-4b30-9b85-066d4fa5db96", + "metadata": {}, + "source": [ + "However, the RTL double-and-add does not error at all." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "b4271143-5365-43b1-9767-d14feb278939", + "metadata": {}, + "outputs": [], + "source": [ + "test_k10(GroupScalarRandomization(rtl), 100);" + ] + }, + { + "cell_type": "markdown", + "id": "8eda1c29-f522-4c61-b445-c1c48278e88f", + "metadata": {}, + "source": [ + "### Multiplicative splitting\n", + "Multiplicative splitting has no reason to error out." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "7cd5bbf5-46e4-4d27-a803-79e35585c250", + "metadata": {}, + "outputs": [], + "source": [ + "test_k10(msplit, 100);" + ] + }, + { + "cell_type": "markdown", + "id": "ca68b8f2-c256-40f9-9135-961877342736", + "metadata": {}, + "source": [ + "### Euclidean splitting\n", + "In Euclidean splitting, it matters whether we consider the computation $[0]G$ to error or not, i.e., whether the implementation is setup to handle a zero scalar gracefully. If it is not, we get 100% of errors for small scalars." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "f205287f-1174-4815-bc0a-2a56ff806f11", + "metadata": {}, + "outputs": [], + "source": [ + "test_k10(esplit, 100);" + ] + }, + { + "cell_type": "markdown", + "id": "cab4594e-8489-4512-98f5-dc56bef3472f", + "metadata": {}, + "source": [ + "However, if it **is** setup to handle a zero scalar, we get no errors on small scalars." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "a6b36c65-3016-4d83-a352-3cb409319e31", + "metadata": {}, + "outputs": [], + "source": [ + "test_k10(esplit, 100, zero_fails=False);" + ] + }, + { + "cell_type": "markdown", + "id": "b1f68d3e-0b55-450e-a831-fa0a85fd06a1", + "metadata": {}, + "source": [ + "When we look back at the case where a zero scalar is mishandled, we see the errors drop-off when we reach the size of tha random mask." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "54716a04-ed6b-4ad3-b92f-2d49ddbcdd73", + "metadata": {}, + "outputs": [], + "source": [ + "test_k10(esplit, 100, (2**i for i in range(120, 134)));" + ] + }, + { + "cell_type": "markdown", + "id": "a0402510-6565-4a63-a72c-fef5fc0cd7f4", + "metadata": {}, + "source": [ + "In the \"no-error-on-zero\" case, we see no errors even up to and past the random mask size." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "187be558-c52e-4ae5-8b1e-a492e913cfb9", + "metadata": {}, + "outputs": [], + "source": [ + "test_k10(esplit, 100, (2**i for i in range(2, 34)), zero_fails=False);" + ] + }, + { + "cell_type": "markdown", + "id": "98ec9216-b704-4d63-a83a-322d4b1eced0", + "metadata": {}, + "source": [ + "### Additive splitting\n", + "Additive splitting has no reason to error out." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "ca9068dc-7752-482e-b56d-de69c0849a07", + "metadata": {}, + "outputs": [], + "source": [ + "test_k10(asplit, 100);" + ] + }, + { + "cell_type": "markdown", + "id": "867fedf2-0d95-4012-b57c-f2e3dcf0826c", + "metadata": {}, + "source": [ + "## Composite test" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "7fdd24cc-7c52-4222-b473-79b09c0df810", + "metadata": {}, + "outputs": [], + "source": [ + "composite = load_params_ectester(io.BytesIO(b\"0x5033ba46674a215a5d62c329a1d017a7909408d59823423b9b273a129828ccc3,0x0b0600851039db449b034a92fd9a7120ab7ee4186c805da4ae42d1755c5f199d,0x2f86acaeaf18553eb283bc126d07f84e9f388d36f5745789b64042bd5ee0277c,0x20a7d7eb71da5e6cbf7dfb64c3e7f6183ffa85f47b58fe0c934fcfb234067664,0x071727449f4309638e74b6a0dad7daa707657427cef3ef8441386b393e4ebf90,0x1abbe8c2226e0b1e1f20ebb88b455d383402bf0803931d1b48eff141bacf7801,0x01\"), \"projective\")\n", + "factors = [2524621529, 2536125749, 2667393247, 3038161487, 3561529459, 3938093561, 4007208383, 4146424241]" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "9db677c5-34e3-4b5e-93dc-11b9b7e2cf3a", + "metadata": {}, + "outputs": [], + "source": [ + "print(f\"prime:\\t0x{composite.curve.prime:x}\")\n", + "print(f\"a:\\t0x{composite.curve.parameters['a']:x}\")\n", + "print(f\"b:\\t0x{composite.curve.parameters['b']:x}\")\n", + "print(f\"G:\\t[0x{composite.generator.X:x},\\n\\t 0x{composite.generator.Y:x}]\")\n", + "print(f\"n:\\t0x{composite.order:x}\")" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "db1e4115-d0cf-4558-93e5-d60781407548", + "metadata": {}, + "outputs": [], + "source": [ + "def test_composite(countermeasure, samples):\n", + " G = composite.generator\n", + " Gaff = G.to_affine()\n", + " correct = 0\n", + " failed = 0\n", + " for _ in range(samples):\n", + " k = random.randrange(0, composite.full_order)\n", + " countermeasure.init(composite, G)\n", + " res = countermeasure.multiply(k)\n", + " res_aff = composite.curve.affine_multiply(Gaff, k)\n", + " try:\n", + " if res.equals_scaled(res_aff.to_model(composite.curve.coordinate_model, composite.curve)):\n", + " correct += 1\n", + " else:\n", + " failed += 1\n", + " except:\n", + " failed += 1\n", + " print(f\"{failed} errors\")" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "9946acdf-41db-4aa2-864a-1e1e771bc6cc", + "metadata": {}, + "outputs": [], + "source": [ + "test_composite(ltr, 100)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "48eff976-9c37-471f-94ab-d09fd6cd53ae", + "metadata": {}, + "outputs": [], + "source": [ + "test_composite(rtl, 100)" + ] + }, + { + "cell_type": "markdown", + "id": "54c9ef0f-bc9e-47b9-a7e4-3821c2a2f93a", + "metadata": {}, + "source": [ + "### Group scalar randomization" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "3420068e-70f3-4fd7-a986-b61e6e21cb54", + "metadata": {}, + "outputs": [], + "source": [ + "test_composite(gsr, 100)" + ] + }, + { + "cell_type": "markdown", + "id": "fea32f4e-dca2-4a5c-bd93-6580183e2d02", + "metadata": {}, + "source": [ + "### Multiplicative splitting" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "1f540d89-af11-47ff-9477-32db8ccee045", + "metadata": {}, + "outputs": [], + "source": [ + "test_composite(msplit, 100)" + ] + }, + { + "cell_type": "markdown", + "id": "189d4f2a-1f0c-473f-b9fe-988bf4fbe9f7", + "metadata": {}, + "source": [ + "### Additive splitting" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "c5fb0d90-0f65-44a5-abdc-10d12345a132", + "metadata": {}, + "outputs": [], + "source": [ + "test_composite(asplit, 100)" + ] + }, + { + "cell_type": "markdown", + "id": "287b7945-c538-4b40-a65b-1081f68107ab", + "metadata": {}, + "source": [ + "### Euclidean splitting" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "61936c3a-6350-45dc-9870-20bac5b8c3e3", + "metadata": {}, + "outputs": [], + "source": [ + "test_composite(esplit, 100)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "f1532a7a-b089-4c27-bc52-c515d71e1a7a", + "metadata": {}, + "outputs": [], + "source": [] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3 (ipykernel)", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.12.3" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} |