1 // SPDX-License-Identifier: GPL-2.0 OR MIT
2 /*
3 * Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
4 */
5
6 #include <crypto/internal/blake2s.h>
7 #include <crypto/internal/simd.h>
8 #include <crypto/internal/hash.h>
9
10 #include <linux/types.h>
11 #include <linux/jump_label.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/sizes.h>
15
16 #include <asm/cpufeature.h>
17 #include <asm/fpu/api.h>
18 #include <asm/processor.h>
19 #include <asm/simd.h>
20
21 asmlinkage void blake2s_compress_ssse3(struct blake2s_state *state,
22 const u8 *block, const size_t nblocks,
23 const u32 inc);
24 asmlinkage void blake2s_compress_avx512(struct blake2s_state *state,
25 const u8 *block, const size_t nblocks,
26 const u32 inc);
27
28 static __ro_after_init DEFINE_STATIC_KEY_FALSE(blake2s_use_ssse3);
29 static __ro_after_init DEFINE_STATIC_KEY_FALSE(blake2s_use_avx512);
30
blake2s_compress_arch(struct blake2s_state * state,const u8 * block,size_t nblocks,const u32 inc)31 void blake2s_compress_arch(struct blake2s_state *state,
32 const u8 *block, size_t nblocks,
33 const u32 inc)
34 {
35 /* SIMD disables preemption, so relax after processing each page. */
36 BUILD_BUG_ON(SZ_4K / BLAKE2S_BLOCK_SIZE < 8);
37
38 if (!static_branch_likely(&blake2s_use_ssse3) || !crypto_simd_usable()) {
39 blake2s_compress_generic(state, block, nblocks, inc);
40 return;
41 }
42
43 do {
44 const size_t blocks = min_t(size_t, nblocks,
45 SZ_4K / BLAKE2S_BLOCK_SIZE);
46
47 kernel_fpu_begin();
48 if (IS_ENABLED(CONFIG_AS_AVX512) &&
49 static_branch_likely(&blake2s_use_avx512))
50 blake2s_compress_avx512(state, block, blocks, inc);
51 else
52 blake2s_compress_ssse3(state, block, blocks, inc);
53 kernel_fpu_end();
54
55 nblocks -= blocks;
56 block += blocks * BLAKE2S_BLOCK_SIZE;
57 } while (nblocks);
58 }
59 EXPORT_SYMBOL(blake2s_compress_arch);
60
crypto_blake2s_setkey(struct crypto_shash * tfm,const u8 * key,unsigned int keylen)61 static int crypto_blake2s_setkey(struct crypto_shash *tfm, const u8 *key,
62 unsigned int keylen)
63 {
64 struct blake2s_tfm_ctx *tctx = crypto_shash_ctx(tfm);
65
66 if (keylen == 0 || keylen > BLAKE2S_KEY_SIZE)
67 return -EINVAL;
68
69 memcpy(tctx->key, key, keylen);
70 tctx->keylen = keylen;
71
72 return 0;
73 }
74
crypto_blake2s_init(struct shash_desc * desc)75 static int crypto_blake2s_init(struct shash_desc *desc)
76 {
77 struct blake2s_tfm_ctx *tctx = crypto_shash_ctx(desc->tfm);
78 struct blake2s_state *state = shash_desc_ctx(desc);
79 const int outlen = crypto_shash_digestsize(desc->tfm);
80
81 if (tctx->keylen)
82 blake2s_init_key(state, outlen, tctx->key, tctx->keylen);
83 else
84 blake2s_init(state, outlen);
85
86 return 0;
87 }
88
crypto_blake2s_update(struct shash_desc * desc,const u8 * in,unsigned int inlen)89 static int crypto_blake2s_update(struct shash_desc *desc, const u8 *in,
90 unsigned int inlen)
91 {
92 struct blake2s_state *state = shash_desc_ctx(desc);
93 const size_t fill = BLAKE2S_BLOCK_SIZE - state->buflen;
94
95 if (unlikely(!inlen))
96 return 0;
97 if (inlen > fill) {
98 memcpy(state->buf + state->buflen, in, fill);
99 blake2s_compress_arch(state, state->buf, 1, BLAKE2S_BLOCK_SIZE);
100 state->buflen = 0;
101 in += fill;
102 inlen -= fill;
103 }
104 if (inlen > BLAKE2S_BLOCK_SIZE) {
105 const size_t nblocks = DIV_ROUND_UP(inlen, BLAKE2S_BLOCK_SIZE);
106 /* Hash one less (full) block than strictly possible */
107 blake2s_compress_arch(state, in, nblocks - 1, BLAKE2S_BLOCK_SIZE);
108 in += BLAKE2S_BLOCK_SIZE * (nblocks - 1);
109 inlen -= BLAKE2S_BLOCK_SIZE * (nblocks - 1);
110 }
111 memcpy(state->buf + state->buflen, in, inlen);
112 state->buflen += inlen;
113
114 return 0;
115 }
116
crypto_blake2s_final(struct shash_desc * desc,u8 * out)117 static int crypto_blake2s_final(struct shash_desc *desc, u8 *out)
118 {
119 struct blake2s_state *state = shash_desc_ctx(desc);
120
121 blake2s_set_lastblock(state);
122 memset(state->buf + state->buflen, 0,
123 BLAKE2S_BLOCK_SIZE - state->buflen); /* Padding */
124 blake2s_compress_arch(state, state->buf, 1, state->buflen);
125 cpu_to_le32_array(state->h, ARRAY_SIZE(state->h));
126 memcpy(out, state->h, state->outlen);
127 memzero_explicit(state, sizeof(*state));
128
129 return 0;
130 }
131
132 static struct shash_alg blake2s_algs[] = {{
133 .base.cra_name = "blake2s-128",
134 .base.cra_driver_name = "blake2s-128-x86",
135 .base.cra_flags = CRYPTO_ALG_OPTIONAL_KEY,
136 .base.cra_ctxsize = sizeof(struct blake2s_tfm_ctx),
137 .base.cra_priority = 200,
138 .base.cra_blocksize = BLAKE2S_BLOCK_SIZE,
139 .base.cra_module = THIS_MODULE,
140
141 .digestsize = BLAKE2S_128_HASH_SIZE,
142 .setkey = crypto_blake2s_setkey,
143 .init = crypto_blake2s_init,
144 .update = crypto_blake2s_update,
145 .final = crypto_blake2s_final,
146 .descsize = sizeof(struct blake2s_state),
147 }, {
148 .base.cra_name = "blake2s-160",
149 .base.cra_driver_name = "blake2s-160-x86",
150 .base.cra_flags = CRYPTO_ALG_OPTIONAL_KEY,
151 .base.cra_ctxsize = sizeof(struct blake2s_tfm_ctx),
152 .base.cra_priority = 200,
153 .base.cra_blocksize = BLAKE2S_BLOCK_SIZE,
154 .base.cra_module = THIS_MODULE,
155
156 .digestsize = BLAKE2S_160_HASH_SIZE,
157 .setkey = crypto_blake2s_setkey,
158 .init = crypto_blake2s_init,
159 .update = crypto_blake2s_update,
160 .final = crypto_blake2s_final,
161 .descsize = sizeof(struct blake2s_state),
162 }, {
163 .base.cra_name = "blake2s-224",
164 .base.cra_driver_name = "blake2s-224-x86",
165 .base.cra_flags = CRYPTO_ALG_OPTIONAL_KEY,
166 .base.cra_ctxsize = sizeof(struct blake2s_tfm_ctx),
167 .base.cra_priority = 200,
168 .base.cra_blocksize = BLAKE2S_BLOCK_SIZE,
169 .base.cra_module = THIS_MODULE,
170
171 .digestsize = BLAKE2S_224_HASH_SIZE,
172 .setkey = crypto_blake2s_setkey,
173 .init = crypto_blake2s_init,
174 .update = crypto_blake2s_update,
175 .final = crypto_blake2s_final,
176 .descsize = sizeof(struct blake2s_state),
177 }, {
178 .base.cra_name = "blake2s-256",
179 .base.cra_driver_name = "blake2s-256-x86",
180 .base.cra_flags = CRYPTO_ALG_OPTIONAL_KEY,
181 .base.cra_ctxsize = sizeof(struct blake2s_tfm_ctx),
182 .base.cra_priority = 200,
183 .base.cra_blocksize = BLAKE2S_BLOCK_SIZE,
184 .base.cra_module = THIS_MODULE,
185
186 .digestsize = BLAKE2S_256_HASH_SIZE,
187 .setkey = crypto_blake2s_setkey,
188 .init = crypto_blake2s_init,
189 .update = crypto_blake2s_update,
190 .final = crypto_blake2s_final,
191 .descsize = sizeof(struct blake2s_state),
192 }};
193
blake2s_mod_init(void)194 static int __init blake2s_mod_init(void)
195 {
196 if (!boot_cpu_has(X86_FEATURE_SSSE3))
197 return 0;
198
199 static_branch_enable(&blake2s_use_ssse3);
200
201 if (IS_ENABLED(CONFIG_AS_AVX512) &&
202 boot_cpu_has(X86_FEATURE_AVX) &&
203 boot_cpu_has(X86_FEATURE_AVX2) &&
204 boot_cpu_has(X86_FEATURE_AVX512F) &&
205 boot_cpu_has(X86_FEATURE_AVX512VL) &&
206 cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM |
207 XFEATURE_MASK_AVX512, NULL))
208 static_branch_enable(&blake2s_use_avx512);
209
210 return IS_REACHABLE(CONFIG_CRYPTO_HASH) ?
211 crypto_register_shashes(blake2s_algs,
212 ARRAY_SIZE(blake2s_algs)) : 0;
213 }
214
blake2s_mod_exit(void)215 static void __exit blake2s_mod_exit(void)
216 {
217 if (IS_REACHABLE(CONFIG_CRYPTO_HASH) && boot_cpu_has(X86_FEATURE_SSSE3))
218 crypto_unregister_shashes(blake2s_algs, ARRAY_SIZE(blake2s_algs));
219 }
220
221 module_init(blake2s_mod_init);
222 module_exit(blake2s_mod_exit);
223
224 MODULE_ALIAS_CRYPTO("blake2s-128");
225 MODULE_ALIAS_CRYPTO("blake2s-128-x86");
226 MODULE_ALIAS_CRYPTO("blake2s-160");
227 MODULE_ALIAS_CRYPTO("blake2s-160-x86");
228 MODULE_ALIAS_CRYPTO("blake2s-224");
229 MODULE_ALIAS_CRYPTO("blake2s-224-x86");
230 MODULE_ALIAS_CRYPTO("blake2s-256");
231 MODULE_ALIAS_CRYPTO("blake2s-256-x86");
232 MODULE_LICENSE("GPL v2");
233