crypto/poly1305_generic.c

   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
  21 static inline u64 mlt(u64 a, u64 b)
  22 {
  23         return a * b;
  24 }
  25
  26 static inline u32 sr(u64 v, u_char n)
  27 {
  28         return v >> n;
  29 }
  30
  31 static inline u32 and(u32 v, u32 mask)
  32 {
  33         return v & mask;
  34 }
  35
  36 static inline u32 le32_to_cpuvp(const void *p)
  37 {
  38         return le32_to_cpup(p);
  39 }
  40
  41 int crypto_poly1305_init(struct shash_desc *desc)
  42 {
  43         struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
  44
  45         memset(dctx->h, 0, sizeof(dctx->h));
  46         dctx->buflen = 0;
  47         dctx->rset = false;
  48         dctx->sset = false;
  49
  50         return 0;
  51 }
  52 EXPORT_SYMBOL_GPL(crypto_poly1305_init);
  53
  54 static void poly1305_setrkey(struct poly1305_desc_ctx *dctx, const u8 *key)
  55 {
  56         /* r &= 0xffffffc0ffffffc0ffffffc0fffffff */
  57         dctx->r[0] = (le32_to_cpuvp(key +  0) >> 0) & 0x3ffffff;
  58         dctx->r[1] = (le32_to_cpuvp(key +  3) >> 2) & 0x3ffff03;
  59         dctx->r[2] = (le32_to_cpuvp(key +  6) >> 4) & 0x3ffc0ff;
  60         dctx->r[3] = (le32_to_cpuvp(key +  9) >> 6) & 0x3f03fff;
  61         dctx->r[4] = (le32_to_cpuvp(key + 12) >> 8) & 0x00fffff;
  62 }
  63
  64 static void poly1305_setskey(struct poly1305_desc_ctx *dctx, const u8 *key)
  65 {
  66         dctx->s[0] = le32_to_cpuvp(key +  0);
  67         dctx->s[1] = le32_to_cpuvp(key +  4);
  68         dctx->s[2] = le32_to_cpuvp(key +  8);
  69         dctx->s[3] = le32_to_cpuvp(key + 12);
  70 }
  71
  72 /*
  73  * Poly1305 requires a unique key for each tag, which implies that we can't set
  74  * it on the tfm that gets accessed by multiple users simultaneously. Instead we
  75  * expect the key as the first 32 bytes in the update() call.
  76  */
  77 unsigned int crypto_poly1305_setdesckey(struct poly1305_desc_ctx *dctx,
  78                                         const u8 *src, unsigned int srclen)
  79 {
  80         if (!dctx->sset) {
  81                 if (!dctx->rset && srclen >= POLY1305_BLOCK_SIZE) {
  82                         poly1305_setrkey(dctx, src);
  83                         src += POLY1305_BLOCK_SIZE;
  84                         srclen -= POLY1305_BLOCK_SIZE;
  85                         dctx->rset = true;
  86                 }
  87                 if (srclen >= POLY1305_BLOCK_SIZE) {
  88                         poly1305_setskey(dctx, src);
  89                         src += POLY1305_BLOCK_SIZE;
  90                         srclen -= POLY1305_BLOCK_SIZE;
  91                         dctx->sset = true;
  92                 }
  93         }
  94         return srclen;
  95 }
  96 EXPORT_SYMBOL_GPL(crypto_poly1305_setdesckey);
  97
  98 static unsigned int poly1305_blocks(struct poly1305_desc_ctx *dctx,
  99                                     const u8 *src, unsigned int srclen,
 100                                     u32 hibit)
 101 {
 102         u32 r0, r1, r2, r3, r4;
 103         u32 s1, s2, s3, s4;
 104         u32 h0, h1, h2, h3, h4;
 105         u64 d0, d1, d2, d3, d4;
 106         unsigned int datalen;
 107
 108         if (unlikely(!dctx->sset)) {
 109                 datalen = crypto_poly1305_setdesckey(dctx, src, srclen);
 110                 src += srclen - datalen;
 111                 srclen = datalen;
 112         }
 113
 114         r0 = dctx->r[0];
 115         r1 = dctx->r[1];
 116         r2 = dctx->r[2];
 117         r3 = dctx->r[3];
 118         r4 = dctx->r[4];
 119
 120         s1 = r1 * 5;
 121         s2 = r2 * 5;
 122         s3 = r3 * 5;
 123         s4 = r4 * 5;
 124
 125         h0 = dctx->h[0];
 126         h1 = dctx->h[1];
 127         h2 = dctx->h[2];
 128         h3 = dctx->h[3];
 129         h4 = dctx->h[4];
 130
 131         while (likely(srclen >= POLY1305_BLOCK_SIZE)) {
 132
 133                 /* h += m[i] */
 134                 h0 += (le32_to_cpuvp(src +  0) >> 0) & 0x3ffffff;
 135                 h1 += (le32_to_cpuvp(src +  3) >> 2) & 0x3ffffff;
 136                 h2 += (le32_to_cpuvp(src +  6) >> 4) & 0x3ffffff;
 137                 h3 += (le32_to_cpuvp(src +  9) >> 6) & 0x3ffffff;
 138                 h4 += (le32_to_cpuvp(src + 12) >> 8) | hibit;
 139
 140                 /* h *= r */
 141                 d0 = mlt(h0, r0) + mlt(h1, s4) + mlt(h2, s3) +
 142                      mlt(h3, s2) + mlt(h4, s1);
 143                 d1 = mlt(h0, r1) + mlt(h1, r0) + mlt(h2, s4) +
 144                      mlt(h3, s3) + mlt(h4, s2);
 145                 d2 = mlt(h0, r2) + mlt(h1, r1) + mlt(h2, r0) +
 146                      mlt(h3, s4) + mlt(h4, s3);
 147                 d3 = mlt(h0, r3) + mlt(h1, r2) + mlt(h2, r1) +
 148                      mlt(h3, r0) + mlt(h4, s4);
 149                 d4 = mlt(h0, r4) + mlt(h1, r3) + mlt(h2, r2) +
 150                      mlt(h3, r1) + mlt(h4, r0);
 151
 152                 /* (partial) h %= p */
 153                 d1 += sr(d0, 26);     h0 = and(d0, 0x3ffffff);
 154                 d2 += sr(d1, 26);     h1 = and(d1, 0x3ffffff);
 155                 d3 += sr(d2, 26);     h2 = and(d2, 0x3ffffff);
 156                 d4 += sr(d3, 26);     h3 = and(d3, 0x3ffffff);
 157                 h0 += sr(d4, 26) * 5; h4 = and(d4, 0x3ffffff);
 158                 h1 += h0 >> 26;       h0 = h0 & 0x3ffffff;
 159
 160                 src += POLY1305_BLOCK_SIZE;
 161                 srclen -= POLY1305_BLOCK_SIZE;
 162         }
 163
 164         dctx->h[0] = h0;
 165         dctx->h[1] = h1;
 166         dctx->h[2] = h2;
 167         dctx->h[3] = h3;
 168         dctx->h[4] = h4;
 169
 170         return srclen;
 171 }
 172
 173 int crypto_poly1305_update(struct shash_desc *desc,
 174                            const u8 *src, unsigned int srclen)
 175 {
 176         struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
 177         unsigned int bytes;
 178
 179         if (unlikely(dctx->buflen)) {
 180                 bytes = min(srclen, POLY1305_BLOCK_SIZE - dctx->buflen);
 181                 memcpy(dctx->buf + dctx->buflen, src, bytes);
 182                 src += bytes;
 183                 srclen -= bytes;
 184                 dctx->buflen += bytes;
 185
 186                 if (dctx->buflen == POLY1305_BLOCK_SIZE) {
 187                         poly1305_blocks(dctx, dctx->buf,
 188                                         POLY1305_BLOCK_SIZE, 1 << 24);
 189                         dctx->buflen = 0;
 190                 }
 191         }
 192
 193         if (likely(srclen >= POLY1305_BLOCK_SIZE)) {
 194                 bytes = poly1305_blocks(dctx, src, srclen, 1 << 24);
 195                 src += srclen - bytes;
 196                 srclen = bytes;
 197         }
 198
 199         if (unlikely(srclen)) {
 200                 dctx->buflen = srclen;
 201                 memcpy(dctx->buf, src, srclen);
 202         }
 203
 204         return 0;
 205 }
 206 EXPORT_SYMBOL_GPL(crypto_poly1305_update);
 207
 208 int crypto_poly1305_final(struct shash_desc *desc, u8 *dst)
 209 {
 210         struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
 211         __le32 *mac = (__le32 *)dst;
 212         u32 h0, h1, h2, h3, h4;
 213         u32 g0, g1, g2, g3, g4;
 214         u32 mask;
 215         u64 f = 0;
 216
 217         if (unlikely(!dctx->sset))
 218                 return -ENOKEY;
 219
 220         if (unlikely(dctx->buflen)) {
 221                 dctx->buf[dctx->buflen++] = 1;
 222                 memset(dctx->buf + dctx->buflen, 0,
 223                        POLY1305_BLOCK_SIZE - dctx->buflen);
 224                 poly1305_blocks(dctx, dctx->buf, POLY1305_BLOCK_SIZE, 0);
 225         }
 226
 227         /* fully carry h */
 228         h0 = dctx->h[0];
 229         h1 = dctx->h[1];
 230         h2 = dctx->h[2];
 231         h3 = dctx->h[3];
 232         h4 = dctx->h[4];
 233
 234         h2 += (h1 >> 26);     h1 = h1 & 0x3ffffff;
 235         h3 += (h2 >> 26);     h2 = h2 & 0x3ffffff;
 236         h4 += (h3 >> 26);     h3 = h3 & 0x3ffffff;
 237         h0 += (h4 >> 26) * 5; h4 = h4 & 0x3ffffff;
 238         h1 += (h0 >> 26);     h0 = h0 & 0x3ffffff;
 239
 240         /* compute h + -p */
 241         g0 = h0 + 5;
 242         g1 = h1 + (g0 >> 26);             g0 &= 0x3ffffff;
 243         g2 = h2 + (g1 >> 26);             g1 &= 0x3ffffff;
 244         g3 = h3 + (g2 >> 26);             g2 &= 0x3ffffff;
 245         g4 = h4 + (g3 >> 26) - (1 << 26); g3 &= 0x3ffffff;
 246
 247         /* select h if h < p, or h + -p if h >= p */
 248         mask = (g4 >> ((sizeof(u32) * 8) - 1)) - 1;
 249         g0 &= mask;
 250         g1 &= mask;
 251         g2 &= mask;
 252         g3 &= mask;
 253         g4 &= mask;
 254         mask = ~mask;
 255         h0 = (h0 & mask) | g0;
 256         h1 = (h1 & mask) | g1;
 257         h2 = (h2 & mask) | g2;
 258         h3 = (h3 & mask) | g3;
 259         h4 = (h4 & mask) | g4;
 260
 261         /* h = h % (2^128) */
 262         h0 = (h0 >>  0) | (h1 << 26);
 263         h1 = (h1 >>  6) | (h2 << 20);
 264         h2 = (h2 >> 12) | (h3 << 14);
 265         h3 = (h3 >> 18) | (h4 <<  8);
 266
 267         /* mac = (h + s) % (2^128) */
 268         f = (f >> 32) + h0 + dctx->s[0]; mac[0] = cpu_to_le32(f);
 269         f = (f >> 32) + h1 + dctx->s[1]; mac[1] = cpu_to_le32(f);
 270         f = (f >> 32) + h2 + dctx->s[2]; mac[2] = cpu_to_le32(f);
 271         f = (f >> 32) + h3 + dctx->s[3]; mac[3] = cpu_to_le32(f);
 272
 273         return 0;
 274 }
 275 EXPORT_SYMBOL_GPL(crypto_poly1305_final);
 276
 277 static struct shash_alg poly1305_alg = {
 278         .digestsize     = POLY1305_DIGEST_SIZE,
 279         .init           = crypto_poly1305_init,
 280         .update         = crypto_poly1305_update,
 281         .final          = crypto_poly1305_final,
 282         .descsize       = sizeof(struct poly1305_desc_ctx),
 283         .base           = {
 284                 .cra_name               = "poly1305",
 285                 .cra_driver_name        = "poly1305-generic",
 286                 .cra_priority           = 100,
 287                 .cra_flags              = CRYPTO_ALG_TYPE_SHASH,
 288                 .cra_alignmask          = sizeof(u32) - 1,
 289                 .cra_blocksize          = POLY1305_BLOCK_SIZE,
 290                 .cra_module             = THIS_MODULE,
 291         },
 292 };
 293
 294 static int __init poly1305_mod_init(void)
 295 {
 296         return crypto_register_shash(&poly1305_alg);
 297 }
 298
 299 static void __exit poly1305_mod_exit(void)
 300 {
 301         crypto_unregister_shash(&poly1305_alg);
 302 }
 303
 304 module_init(poly1305_mod_init);
 305 module_exit(poly1305_mod_exit);
 306
 307 MODULE_LICENSE("GPL");
 308 MODULE_AUTHOR("Martin Willi <martin@strongswan.org>");
 309 MODULE_DESCRIPTION("Poly1305 authenticator");
 310 MODULE_ALIAS_CRYPTO("poly1305");
 311 MODULE_ALIAS_CRYPTO("poly1305-generic");