1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright 2025 Google LLC
4 */
5 #include <crypto/poly1305.h>
6 #include "poly1305-testvecs.h"
7
8 /*
9 * A fixed key used when presenting Poly1305 as an unkeyed hash function in
10 * order to reuse hash-test-template.h. At the beginning of the test suite,
11 * this is initialized to bytes generated from a fixed seed.
12 */
13 static u8 test_key[POLY1305_KEY_SIZE];
14
15 /* This probably should be in the actual API, but just define it here for now */
poly1305(const u8 key[POLY1305_KEY_SIZE],const u8 * data,size_t len,u8 out[POLY1305_DIGEST_SIZE])16 static void poly1305(const u8 key[POLY1305_KEY_SIZE], const u8 *data,
17 size_t len, u8 out[POLY1305_DIGEST_SIZE])
18 {
19 struct poly1305_desc_ctx ctx;
20
21 poly1305_init(&ctx, key);
22 poly1305_update(&ctx, data, len);
23 poly1305_final(&ctx, out);
24 }
25
poly1305_init_withtestkey(struct poly1305_desc_ctx * ctx)26 static void poly1305_init_withtestkey(struct poly1305_desc_ctx *ctx)
27 {
28 poly1305_init(ctx, test_key);
29 }
30
poly1305_withtestkey(const u8 * data,size_t len,u8 out[POLY1305_DIGEST_SIZE])31 static void poly1305_withtestkey(const u8 *data, size_t len,
32 u8 out[POLY1305_DIGEST_SIZE])
33 {
34 poly1305(test_key, data, len, out);
35 }
36
37 /* Generate the HASH_KUNIT_CASES using hash-test-template.h. */
38 #define HASH poly1305_withtestkey
39 #define HASH_CTX poly1305_desc_ctx
40 #define HASH_SIZE POLY1305_DIGEST_SIZE
41 #define HASH_INIT poly1305_init_withtestkey
42 #define HASH_UPDATE poly1305_update
43 #define HASH_FINAL poly1305_final
44 #include "hash-test-template.h"
45
poly1305_suite_init(struct kunit_suite * suite)46 static int poly1305_suite_init(struct kunit_suite *suite)
47 {
48 rand_bytes_seeded_from_len(test_key, POLY1305_KEY_SIZE);
49 return 0;
50 }
51
52 /*
53 * Poly1305 test case which uses a key and message consisting only of one bits:
54 *
55 * - Using an all-one-bits r_key tests the key clamping.
56 * - Using an all-one-bits s_key tests carries in implementations of the
57 * addition mod 2**128 during finalization.
58 * - Using all-one-bits message and r_key results in large values for the
59 * internal products of r_key by (accumulator + message block). This
60 * increases the chance of detecting bugs that occur only in rare cases where
61 * the sum of these internal products is very large, for example the bug fixed
62 * by commit 678cce4019d7 ("crypto: x86/poly1305 - fix overflow during partial
63 * reduction").
64 *
65 * Bugs with the handling of large internal product sums may be specific to
66 * particular update lengths (in blocks) and/or require that the accumulator
67 * also have a large value. Note that the accumulator starts at 0. Thus, a
68 * single all-one-bits test vector may be insufficient.
69 *
70 * Considering that, do the following test: continuously update a single
71 * Poly1305 context with all-one-bits data of varying lengths (0, 16, 32, ...,
72 * 4096 bytes). After each update, generate the MAC from the current context,
73 * and feed that MAC into a separate Poly1305 context. Repeat that entire
74 * sequence of updates 32 times without re-initializing either context,
75 * resulting in a total of 8224 MAC computations from a long-running, cumulative
76 * context. Finally, generate and verify the MAC of all the MACs.
77 */
test_poly1305_allones_keys_and_message(struct kunit * test)78 static void test_poly1305_allones_keys_and_message(struct kunit *test)
79 {
80 const size_t max_len = 4096;
81 u8 *data = alloc_buf(test, max_len);
82 struct poly1305_desc_ctx mac_ctx, macofmacs_ctx;
83 u8 mac[POLY1305_DIGEST_SIZE];
84
85 memset(data, 0xff, max_len);
86
87 poly1305_init(&mac_ctx, data);
88 poly1305_init(&macofmacs_ctx, data);
89 for (int i = 0; i < 32; i++) {
90 for (size_t len = 0; len <= max_len; len += 16) {
91 struct poly1305_desc_ctx tmp_ctx;
92
93 poly1305_update(&mac_ctx, data, len);
94 tmp_ctx = mac_ctx;
95 poly1305_final(&tmp_ctx, mac);
96 poly1305_update(&macofmacs_ctx, mac,
97 POLY1305_DIGEST_SIZE);
98 }
99 }
100 poly1305_final(&macofmacs_ctx, mac);
101 KUNIT_ASSERT_MEMEQ(test, mac, poly1305_allones_macofmacs,
102 POLY1305_DIGEST_SIZE);
103 }
104
105 /*
106 * Poly1305 test case which uses r_key=1, s_key=0, and a 48-byte message
107 * consisting of three blocks with integer values [2**128 - i, 0, 0]. In this
108 * case, the result of the polynomial evaluation is 2**130 - i. For small
109 * values of i, this is very close to the modulus 2**130 - 5, which helps catch
110 * edge case bugs in the modular reduction logic.
111 */
test_poly1305_reduction_edge_cases(struct kunit * test)112 static void test_poly1305_reduction_edge_cases(struct kunit *test)
113 {
114 static const u8 key[POLY1305_KEY_SIZE] = { 1 }; /* r_key=1, s_key=0 */
115 u8 data[3 * POLY1305_BLOCK_SIZE] = {};
116 u8 expected_mac[POLY1305_DIGEST_SIZE];
117 u8 actual_mac[POLY1305_DIGEST_SIZE];
118
119 for (int i = 1; i <= 10; i++) {
120 /* Set the first data block to 2**128 - i. */
121 data[0] = -i;
122 memset(&data[1], 0xff, POLY1305_BLOCK_SIZE - 1);
123
124 /*
125 * Assuming s_key=0, the expected MAC as an integer is
126 * (2**130 - i mod 2**130 - 5) + 0 mod 2**128. If 1 <= i <= 5,
127 * that's 5 - i. If 6 <= i <= 10, that's 2**128 - i.
128 */
129 if (i <= 5) {
130 expected_mac[0] = 5 - i;
131 memset(&expected_mac[1], 0, POLY1305_DIGEST_SIZE - 1);
132 } else {
133 expected_mac[0] = -i;
134 memset(&expected_mac[1], 0xff,
135 POLY1305_DIGEST_SIZE - 1);
136 }
137
138 /* Compute and verify the MAC. */
139 poly1305(key, data, sizeof(data), actual_mac);
140 KUNIT_ASSERT_MEMEQ(test, actual_mac, expected_mac,
141 POLY1305_DIGEST_SIZE);
142 }
143 }
144
145 /*
146 * Test that Poly1305 MACs are computed correctly when r_key=1, s_key=0, and the
147 * message consists of all-ones blocks.
148 *
149 * With r_key=1 and s_key=0, Poly1305 degrades to the sum of the padded blocks
150 * mod 2**130 - 5, then reduced mod 2**128. A padded all-ones block maps to
151 * 2**128 + (2**128 - 1) = 2**129 - 1. Two such blocks sum to 2**130 - 2, which
152 * is congruent to 3 mod 2**130 - 5. Thus, the expected MAC is just 3 *
153 * floor(nblocks / 2), minus 1 if nblocks is odd, then reduced mod 2**128. If
154 * nblocks == 1 this is 2**128 - 1, otherwise it's just a small integer.
155 *
156 * Inside the Poly1305 implementation in this case, the multiplication does
157 * nothing and each block just adds 2**129 - 1 to the accumulator. Without the
158 * multiplication to mix things up, this results in some interesting values
159 * being reached where the limbs are at or near their maximum values, even after
160 * (lazy) reduction. This can reproduce bugs that happen only in such cases.
161 *
162 * For example, this test case reproduces the bug fixed by commit 24f07e8c07a0
163 * ("lib/crypto: arm64: Fix lost Poly1305 carry when resuming NEON state").
164 *
165 * However, reaching that bug required two consecutive updates: one with >= 8
166 * blocks to cause NEON code to be used and leave the accumulator in base 2**26
167 * with its limbs in a certain pattern, then one with an odd number of blocks >=
168 * 9 to cause NEON to be used again, but first processing one block using scalar
169 * code as a special case, triggering a conversion to base 2**64 where the carry
170 * into the 2**128 bit was lost. To cover this and other similar edge cases
171 * that may exist in other Poly1305 implementations, just test all possible
172 * block-aligned splits between three poly1305_update() calls.
173 */
test_poly1305_split_update_carry(struct kunit * test)174 static void test_poly1305_split_update_carry(struct kunit *test)
175 {
176 static const u8 key[POLY1305_KEY_SIZE] = { 1 }; /* r_key=1, s_key=0 */
177 /*
178 * Use max_nblocks=66 so that it's a bit more than twice the AVX-512
179 * threshold of 32 blocks.
180 */
181 const int max_nblocks = 66;
182 u8 *data = alloc_buf(test, max_nblocks * POLY1305_BLOCK_SIZE);
183 u8 expected_mac[POLY1305_DIGEST_SIZE];
184 u8 actual_mac[POLY1305_DIGEST_SIZE];
185 struct poly1305_desc_ctx ctx;
186
187 KUNIT_ASSERT_LE(test, 3 * (max_nblocks / 2), U8_MAX);
188
189 memset(data, 0xff, max_nblocks * POLY1305_BLOCK_SIZE);
190
191 for (int nblocks = 0; nblocks <= max_nblocks; nblocks++) {
192 size_t len = nblocks * POLY1305_BLOCK_SIZE;
193
194 expected_mac[0] = 3 * (nblocks / 2) - (nblocks % 2);
195 memset(&expected_mac[1], nblocks == 1 ? 0xff : 0,
196 POLY1305_DIGEST_SIZE - 1);
197 for (size_t part1_len = 0; part1_len <= len;
198 part1_len += POLY1305_BLOCK_SIZE) {
199 for (size_t part2_len = 0; part2_len <= len - part1_len;
200 part2_len += POLY1305_BLOCK_SIZE) {
201 size_t part3_len = len - part1_len - part2_len;
202
203 poly1305_init(&ctx, key);
204 poly1305_update(&ctx, data, part1_len);
205 poly1305_update(&ctx, &data[part1_len],
206 part2_len);
207 poly1305_update(&ctx,
208 &data[part1_len + part2_len],
209 part3_len);
210 poly1305_final(&ctx, actual_mac);
211 KUNIT_ASSERT_MEMEQ_MSG(
212 test, actual_mac, expected_mac,
213 POLY1305_DIGEST_SIZE,
214 "Failed with nblocks=%d, part1_len=%zu, part2_len=%zu, part3_len=%zu",
215 nblocks, part1_len, part2_len,
216 part3_len);
217 }
218 }
219 }
220 }
221
222 static struct kunit_case poly1305_test_cases[] = {
223 HASH_KUNIT_CASES,
224 KUNIT_CASE(test_poly1305_allones_keys_and_message),
225 KUNIT_CASE(test_poly1305_reduction_edge_cases),
226 KUNIT_CASE(test_poly1305_split_update_carry),
227 KUNIT_CASE(benchmark_hash),
228 {},
229 };
230
231 static struct kunit_suite poly1305_test_suite = {
232 .name = "poly1305",
233 .test_cases = poly1305_test_cases,
234 .suite_init = poly1305_suite_init,
235 };
236 kunit_test_suite(poly1305_test_suite);
237
238 MODULE_DESCRIPTION("KUnit tests and benchmark for Poly1305");
239 MODULE_LICENSE("GPL");
240