1/*
2 * Poly1305 authenticator algorithm, RFC7539
3 *
4 * Copyright (C) 2015 Martin Willi
5 *
6 * Based on public domain code by Andrew Moon and Daniel J. Bernstein.
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 */
13
14#include <crypto/algapi.h>
15#include <crypto/internal/hash.h>
16#include <crypto/poly1305.h>
17#include <linux/crypto.h>
18#include <linux/kernel.h>
19#include <linux/module.h>
20#include <asm/unaligned.h>
21
22static inline u64 mlt(u64 a, u64 b)
23{
24 return a * b;
25}
26
27static inline u32 sr(u64 v, u_char n)
28{
29 return v >> n;
30}
31
32static inline u32 and(u32 v, u32 mask)
33{
34 return v & mask;
35}
36
37int crypto_poly1305_init(struct shash_desc *desc)
38{
39 struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
40
41 poly1305_core_init(&dctx->h);
42 dctx->buflen = 0;
43 dctx->rset = false;
44 dctx->sset = false;
45
46 return 0;
47}
48EXPORT_SYMBOL_GPL(crypto_poly1305_init);
49
50void poly1305_core_setkey(struct poly1305_key *key, const u8 *raw_key)
51{
52 /* r &= 0xffffffc0ffffffc0ffffffc0fffffff */
53 key->r[0] = (get_unaligned_le32(raw_key + 0) >> 0) & 0x3ffffff;
54 key->r[1] = (get_unaligned_le32(raw_key + 3) >> 2) & 0x3ffff03;
55 key->r[2] = (get_unaligned_le32(raw_key + 6) >> 4) & 0x3ffc0ff;
56 key->r[3] = (get_unaligned_le32(raw_key + 9) >> 6) & 0x3f03fff;
57 key->r[4] = (get_unaligned_le32(raw_key + 12) >> 8) & 0x00fffff;
58}
59EXPORT_SYMBOL_GPL(poly1305_core_setkey);
60
61/*
62 * Poly1305 requires a unique key for each tag, which implies that we can't set
63 * it on the tfm that gets accessed by multiple users simultaneously. Instead we
64 * expect the key as the first 32 bytes in the update() call.
65 */
66unsigned int crypto_poly1305_setdesckey(struct poly1305_desc_ctx *dctx,
67 const u8 *src, unsigned int srclen)
68{
69 if (!dctx->sset) {
70 if (!dctx->rset && srclen >= POLY1305_BLOCK_SIZE) {
71 poly1305_core_setkey(&dctx->r, src);
72 src += POLY1305_BLOCK_SIZE;
73 srclen -= POLY1305_BLOCK_SIZE;
74 dctx->rset = true;
75 }
76 if (srclen >= POLY1305_BLOCK_SIZE) {
77 dctx->s[0] = get_unaligned_le32(src + 0);
78 dctx->s[1] = get_unaligned_le32(src + 4);
79 dctx->s[2] = get_unaligned_le32(src + 8);
80 dctx->s[3] = get_unaligned_le32(src + 12);
81 src += POLY1305_BLOCK_SIZE;
82 srclen -= POLY1305_BLOCK_SIZE;
83 dctx->sset = true;
84 }
85 }
86 return srclen;
87}
88EXPORT_SYMBOL_GPL(crypto_poly1305_setdesckey);
89
90static void poly1305_blocks_internal(struct poly1305_state *state,
91 const struct poly1305_key *key,
92 const void *src, unsigned int nblocks,
93 u32 hibit)
94{
95 u32 r0, r1, r2, r3, r4;
96 u32 s1, s2, s3, s4;
97 u32 h0, h1, h2, h3, h4;
98 u64 d0, d1, d2, d3, d4;
99
100 if (!nblocks)
101 return;
102
103 r0 = key->r[0];
104 r1 = key->r[1];
105 r2 = key->r[2];
106 r3 = key->r[3];
107 r4 = key->r[4];
108
109 s1 = r1 * 5;
110 s2 = r2 * 5;
111 s3 = r3 * 5;
112 s4 = r4 * 5;
113
114 h0 = state->h[0];
115 h1 = state->h[1];
116 h2 = state->h[2];
117 h3 = state->h[3];
118 h4 = state->h[4];
119
120 do {
121 /* h += m[i] */
122 h0 += (get_unaligned_le32(src + 0) >> 0) & 0x3ffffff;
123 h1 += (get_unaligned_le32(src + 3) >> 2) & 0x3ffffff;
124 h2 += (get_unaligned_le32(src + 6) >> 4) & 0x3ffffff;
125 h3 += (get_unaligned_le32(src + 9) >> 6) & 0x3ffffff;
126 h4 += (get_unaligned_le32(src + 12) >> 8) | hibit;
127
128 /* h *= r */
129 d0 = mlt(h0, r0) + mlt(h1, s4) + mlt(h2, s3) +
130 mlt(h3, s2) + mlt(h4, s1);
131 d1 = mlt(h0, r1) + mlt(h1, r0) + mlt(h2, s4) +
132 mlt(h3, s3) + mlt(h4, s2);
133 d2 = mlt(h0, r2) + mlt(h1, r1) + mlt(h2, r0) +
134 mlt(h3, s4) + mlt(h4, s3);
135 d3 = mlt(h0, r3) + mlt(h1, r2) + mlt(h2, r1) +
136 mlt(h3, r0) + mlt(h4, s4);
137 d4 = mlt(h0, r4) + mlt(h1, r3) + mlt(h2, r2) +
138 mlt(h3, r1) + mlt(h4, r0);
139
140 /* (partial) h %= p */
141 d1 += sr(d0, 26); h0 = and(d0, 0x3ffffff);
142 d2 += sr(d1, 26); h1 = and(d1, 0x3ffffff);
143 d3 += sr(d2, 26); h2 = and(d2, 0x3ffffff);
144 d4 += sr(d3, 26); h3 = and(d3, 0x3ffffff);
145 h0 += sr(d4, 26) * 5; h4 = and(d4, 0x3ffffff);
146 h1 += h0 >> 26; h0 = h0 & 0x3ffffff;
147
148 src += POLY1305_BLOCK_SIZE;
149 } while (--nblocks);
150
151 state->h[0] = h0;
152 state->h[1] = h1;
153 state->h[2] = h2;
154 state->h[3] = h3;
155 state->h[4] = h4;
156}
157
158void poly1305_core_blocks(struct poly1305_state *state,
159 const struct poly1305_key *key,
160 const void *src, unsigned int nblocks)
161{
162 poly1305_blocks_internal(state, key, src, nblocks, 1 << 24);
163}
164EXPORT_SYMBOL_GPL(poly1305_core_blocks);
165
166static void poly1305_blocks(struct poly1305_desc_ctx *dctx,
167 const u8 *src, unsigned int srclen, u32 hibit)
168{
169 unsigned int datalen;
170
171 if (unlikely(!dctx->sset)) {
172 datalen = crypto_poly1305_setdesckey(dctx, src, srclen);
173 src += srclen - datalen;
174 srclen = datalen;
175 }
176
177 poly1305_blocks_internal(&dctx->h, &dctx->r,
178 src, srclen / POLY1305_BLOCK_SIZE, hibit);
179}
180
181int crypto_poly1305_update(struct shash_desc *desc,
182 const u8 *src, unsigned int srclen)
183{
184 struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
185 unsigned int bytes;
186
187 if (unlikely(dctx->buflen)) {
188 bytes = min(srclen, POLY1305_BLOCK_SIZE - dctx->buflen);
189 memcpy(dctx->buf + dctx->buflen, src, bytes);
190 src += bytes;
191 srclen -= bytes;
192 dctx->buflen += bytes;
193
194 if (dctx->buflen == POLY1305_BLOCK_SIZE) {
195 poly1305_blocks(dctx, dctx->buf,
196 POLY1305_BLOCK_SIZE, 1 << 24);
197 dctx->buflen = 0;
198 }
199 }
200
201 if (likely(srclen >= POLY1305_BLOCK_SIZE)) {
202 poly1305_blocks(dctx, src, srclen, 1 << 24);
203 src += srclen - (srclen % POLY1305_BLOCK_SIZE);
204 srclen %= POLY1305_BLOCK_SIZE;
205 }
206
207 if (unlikely(srclen)) {
208 dctx->buflen = srclen;
209 memcpy(dctx->buf, src, srclen);
210 }
211
212 return 0;
213}
214EXPORT_SYMBOL_GPL(crypto_poly1305_update);
215
216void poly1305_core_emit(const struct poly1305_state *state, void *dst)
217{
218 u32 h0, h1, h2, h3, h4;
219 u32 g0, g1, g2, g3, g4;
220 u32 mask;
221
222 /* fully carry h */
223 h0 = state->h[0];
224 h1 = state->h[1];
225 h2 = state->h[2];
226 h3 = state->h[3];
227 h4 = state->h[4];
228
229 h2 += (h1 >> 26); h1 = h1 & 0x3ffffff;
230 h3 += (h2 >> 26); h2 = h2 & 0x3ffffff;
231 h4 += (h3 >> 26); h3 = h3 & 0x3ffffff;
232 h0 += (h4 >> 26) * 5; h4 = h4 & 0x3ffffff;
233 h1 += (h0 >> 26); h0 = h0 & 0x3ffffff;
234
235 /* compute h + -p */
236 g0 = h0 + 5;
237 g1 = h1 + (g0 >> 26); g0 &= 0x3ffffff;
238 g2 = h2 + (g1 >> 26); g1 &= 0x3ffffff;
239 g3 = h3 + (g2 >> 26); g2 &= 0x3ffffff;
240 g4 = h4 + (g3 >> 26) - (1 << 26); g3 &= 0x3ffffff;
241
242 /* select h if h < p, or h + -p if h >= p */
243 mask = (g4 >> ((sizeof(u32) * 8) - 1)) - 1;
244 g0 &= mask;
245 g1 &= mask;
246 g2 &= mask;
247 g3 &= mask;
248 g4 &= mask;
249 mask = ~mask;
250 h0 = (h0 & mask) | g0;
251 h1 = (h1 & mask) | g1;
252 h2 = (h2 & mask) | g2;
253 h3 = (h3 & mask) | g3;
254 h4 = (h4 & mask) | g4;
255
256 /* h = h % (2^128) */
257 put_unaligned_le32((h0 >> 0) | (h1 << 26), dst + 0);
258 put_unaligned_le32((h1 >> 6) | (h2 << 20), dst + 4);
259 put_unaligned_le32((h2 >> 12) | (h3 << 14), dst + 8);
260 put_unaligned_le32((h3 >> 18) | (h4 << 8), dst + 12);
261}
262EXPORT_SYMBOL_GPL(poly1305_core_emit);
263
264int crypto_poly1305_final(struct shash_desc *desc, u8 *dst)
265{
266 struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
267 __le32 digest[4];
268 u64 f = 0;
269
270 if (unlikely(!dctx->sset))
271 return -ENOKEY;
272
273 if (unlikely(dctx->buflen)) {
274 dctx->buf[dctx->buflen++] = 1;
275 memset(dctx->buf + dctx->buflen, 0,
276 POLY1305_BLOCK_SIZE - dctx->buflen);
277 poly1305_blocks(dctx, dctx->buf, POLY1305_BLOCK_SIZE, 0);
278 }
279
280 poly1305_core_emit(&dctx->h, digest);
281
282 /* mac = (h + s) % (2^128) */
283 f = (f >> 32) + le32_to_cpu(digest[0]) + dctx->s[0];
284 put_unaligned_le32(f, dst + 0);
285 f = (f >> 32) + le32_to_cpu(digest[1]) + dctx->s[1];
286 put_unaligned_le32(f, dst + 4);
287 f = (f >> 32) + le32_to_cpu(digest[2]) + dctx->s[2];
288 put_unaligned_le32(f, dst + 8);
289 f = (f >> 32) + le32_to_cpu(digest[3]) + dctx->s[3];
290 put_unaligned_le32(f, dst + 12);
291
292 return 0;
293}
294EXPORT_SYMBOL_GPL(crypto_poly1305_final);
295
296static struct shash_alg poly1305_alg = {
297 .digestsize = POLY1305_DIGEST_SIZE,
298 .init = crypto_poly1305_init,
299 .update = crypto_poly1305_update,
300 .final = crypto_poly1305_final,
301 .descsize = sizeof(struct poly1305_desc_ctx),
302 .base = {
303 .cra_name = "poly1305",
304 .cra_driver_name = "poly1305-generic",
305 .cra_priority = 100,
306 .cra_blocksize = POLY1305_BLOCK_SIZE,
307 .cra_module = THIS_MODULE,
308 },
309};
310
311static int __init poly1305_mod_init(void)
312{
313 return crypto_register_shash(&poly1305_alg);
314}
315
316static void __exit poly1305_mod_exit(void)
317{
318 crypto_unregister_shash(&poly1305_alg);
319}
320
321module_init(poly1305_mod_init);
322module_exit(poly1305_mod_exit);
323
324MODULE_LICENSE("GPL");
325MODULE_AUTHOR("Martin Willi <martin@strongswan.org>");
326MODULE_DESCRIPTION("Poly1305 authenticator");
327MODULE_ALIAS_CRYPTO("poly1305");
328MODULE_ALIAS_CRYPTO("poly1305-generic");
329