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 int crypto_poly1305_setkey(struct crypto_shash *tfm,
  55                            const u8 *key, unsigned int keylen)
  56 {
  57         /* Poly1305 requires a unique key for each tag, which implies that
  58          * we can't set it on the tfm that gets accessed by multiple users
  59          * simultaneously. Instead we expect the key as the first 32 bytes in
  60          * the update() call. */
  61         return -ENOTSUPP;
  62 }
  63 EXPORT_SYMBOL_GPL(crypto_poly1305_setkey);
  64
  65 static void poly1305_setrkey(struct poly1305_desc_ctx *dctx, const u8 *key)
  66 {
  67         /* r &= 0xffffffc0ffffffc0ffffffc0fffffff */
  68         dctx->r[0] = (le32_to_cpuvp(key +  0) >> 0) & 0x3ffffff;
  69         dctx->r[1] = (le32_to_cpuvp(key +  3) >> 2) & 0x3ffff03;
  70         dctx->r[2] = (le32_to_cpuvp(key +  6) >> 4) & 0x3ffc0ff;
  71         dctx->r[3] = (le32_to_cpuvp(key +  9) >> 6) & 0x3f03fff;
  72         dctx->r[4] = (le32_to_cpuvp(key + 12) >> 8) & 0x00fffff;
  73 }
  74
  75 static void poly1305_setskey(struct poly1305_desc_ctx *dctx, const u8 *key)
  76 {
  77         dctx->s[0] = le32_to_cpuvp(key +  0);
  78         dctx->s[1] = le32_to_cpuvp(key +  4);
  79         dctx->s[2] = le32_to_cpuvp(key +  8);
  80         dctx->s[3] = le32_to_cpuvp(key + 12);
  81 }
  82
  83 unsigned int crypto_poly1305_setdesckey(struct poly1305_desc_ctx *dctx,
  84                                         const u8 *src, unsigned int srclen)
  85 {
  86         if (!dctx->sset) {
  87                 if (!dctx->rset && srclen >= POLY1305_BLOCK_SIZE) {
  88                         poly1305_setrkey(dctx, src);
  89                         src += POLY1305_BLOCK_SIZE;
  90                         srclen -= POLY1305_BLOCK_SIZE;
  91                         dctx->rset = true;
  92                 }
  93                 if (srclen >= POLY1305_BLOCK_SIZE) {
  94                         poly1305_setskey(dctx, src);
  95                         src += POLY1305_BLOCK_SIZE;
  96                         srclen -= POLY1305_BLOCK_SIZE;
  97                         dctx->sset = true;
  98                 }
  99         }
 100         return srclen;
 101 }
 102 EXPORT_SYMBOL_GPL(crypto_poly1305_setdesckey);
 103
 104 static unsigned int poly1305_blocks(struct poly1305_desc_ctx *dctx,
 105                                     const u8 *src, unsigned int srclen,
 106                                     u32 hibit)
 107 {
 108         u32 r0, r1, r2, r3, r4;
 109         u32 s1, s2, s3, s4;
 110         u32 h0, h1, h2, h3, h4;
 111         u64 d0, d1, d2, d3, d4;
 112         unsigned int datalen;
 113
 114         if (unlikely(!dctx->sset)) {
 115                 datalen = crypto_poly1305_setdesckey(dctx, src, srclen);
 116                 src += srclen - datalen;
 117                 srclen = datalen;
 118         }
 119
 120         r0 = dctx->r[0];
 121         r1 = dctx->r[1];
 122         r2 = dctx->r[2];
 123         r3 = dctx->r[3];
 124         r4 = dctx->r[4];
 125
 126         s1 = r1 * 5;
 127         s2 = r2 * 5;
 128         s3 = r3 * 5;
 129         s4 = r4 * 5;
 130
 131         h0 = dctx->h[0];
 132         h1 = dctx->h[1];
 133         h2 = dctx->h[2];
 134         h3 = dctx->h[3];
 135         h4 = dctx->h[4];
 136
 137         while (likely(srclen >= POLY1305_BLOCK_SIZE)) {
 138
 139                 /* h += m[i] */
 140                 h0 += (le32_to_cpuvp(src +  0) >> 0) & 0x3ffffff;
 141                 h1 += (le32_to_cpuvp(src +  3) >> 2) & 0x3ffffff;
 142                 h2 += (le32_to_cpuvp(src +  6) >> 4) & 0x3ffffff;
 143                 h3 += (le32_to_cpuvp(src +  9) >> 6) & 0x3ffffff;
 144                 h4 += (le32_to_cpuvp(src + 12) >> 8) | hibit;
 145
 146                 /* h *= r */
 147                 d0 = mlt(h0, r0) + mlt(h1, s4) + mlt(h2, s3) +
 148                      mlt(h3, s2) + mlt(h4, s1);
 149                 d1 = mlt(h0, r1) + mlt(h1, r0) + mlt(h2, s4) +
 150                      mlt(h3, s3) + mlt(h4, s2);
 151                 d2 = mlt(h0, r2) + mlt(h1, r1) + mlt(h2, r0) +
 152                      mlt(h3, s4) + mlt(h4, s3);
 153                 d3 = mlt(h0, r3) + mlt(h1, r2) + mlt(h2, r1) +
 154                      mlt(h3, r0) + mlt(h4, s4);
 155                 d4 = mlt(h0, r4) + mlt(h1, r3) + mlt(h2, r2) +
 156                      mlt(h3, r1) + mlt(h4, r0);
 157
 158                 /* (partial) h %= p */
 159                 d1 += sr(d0, 26);     h0 = and(d0, 0x3ffffff);
 160                 d2 += sr(d1, 26);     h1 = and(d1, 0x3ffffff);
 161                 d3 += sr(d2, 26);     h2 = and(d2, 0x3ffffff);
 162                 d4 += sr(d3, 26);     h3 = and(d3, 0x3ffffff);
 163                 h0 += sr(d4, 26) * 5; h4 = and(d4, 0x3ffffff);
 164                 h1 += h0 >> 26;       h0 = h0 & 0x3ffffff;
 165
 166                 src += POLY1305_BLOCK_SIZE;
 167                 srclen -= POLY1305_BLOCK_SIZE;
 168         }
 169
 170         dctx->h[0] = h0;
 171         dctx->h[1] = h1;
 172         dctx->h[2] = h2;
 173         dctx->h[3] = h3;
 174         dctx->h[4] = h4;
 175
 176         return srclen;
 177 }
 178
 179 int crypto_poly1305_update(struct shash_desc *desc,
 180                            const u8 *src, unsigned int srclen)
 181 {
 182         struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
 183         unsigned int bytes;
 184
 185         if (unlikely(dctx->buflen)) {
 186                 bytes = min(srclen, POLY1305_BLOCK_SIZE - dctx->buflen);
 187                 memcpy(dctx->buf + dctx->buflen, src, bytes);
 188                 src += bytes;
 189                 srclen -= bytes;
 190                 dctx->buflen += bytes;
 191
 192                 if (dctx->buflen == POLY1305_BLOCK_SIZE) {
 193                         poly1305_blocks(dctx, dctx->buf,
 194                                         POLY1305_BLOCK_SIZE, 1 << 24);
 195                         dctx->buflen = 0;
 196                 }
 197         }
 198
 199         if (likely(srclen >= POLY1305_BLOCK_SIZE)) {
 200                 bytes = poly1305_blocks(dctx, src, srclen, 1 << 24);
 201                 src += srclen - bytes;
 202                 srclen = bytes;
 203         }
 204
 205         if (unlikely(srclen)) {
 206                 dctx->buflen = srclen;
 207                 memcpy(dctx->buf, src, srclen);
 208         }
 209
 210         return 0;
 211 }
 212 EXPORT_SYMBOL_GPL(crypto_poly1305_update);
 213
 214 int crypto_poly1305_final(struct shash_desc *desc, u8 *dst)
 215 {
 216         struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
 217         __le32 *mac = (__le32 *)dst;
 218         u32 h0, h1, h2, h3, h4;
 219         u32 g0, g1, g2, g3, g4;
 220         u32 mask;
 221         u64 f = 0;
 222
 223         if (unlikely(!dctx->sset))
 224                 return -ENOKEY;
 225
 226         if (unlikely(dctx->buflen)) {
 227                 dctx->buf[dctx->buflen++] = 1;
 228                 memset(dctx->buf + dctx->buflen, 0,
 229                        POLY1305_BLOCK_SIZE - dctx->buflen);
 230                 poly1305_blocks(dctx, dctx->buf, POLY1305_BLOCK_SIZE, 0);
 231         }
 232
 233         /* fully carry h */
 234         h0 = dctx->h[0];
 235         h1 = dctx->h[1];
 236         h2 = dctx->h[2];
 237         h3 = dctx->h[3];
 238         h4 = dctx->h[4];
 239
 240         h2 += (h1 >> 26);     h1 = h1 & 0x3ffffff;
 241         h3 += (h2 >> 26);     h2 = h2 & 0x3ffffff;
 242         h4 += (h3 >> 26);     h3 = h3 & 0x3ffffff;
 243         h0 += (h4 >> 26) * 5; h4 = h4 & 0x3ffffff;
 244         h1 += (h0 >> 26);     h0 = h0 & 0x3ffffff;
 245
 246         /* compute h + -p */
 247         g0 = h0 + 5;
 248         g1 = h1 + (g0 >> 26);             g0 &= 0x3ffffff;
 249         g2 = h2 + (g1 >> 26);             g1 &= 0x3ffffff;
 250         g3 = h3 + (g2 >> 26);             g2 &= 0x3ffffff;
 251         g4 = h4 + (g3 >> 26) - (1 << 26); g3 &= 0x3ffffff;
 252
 253         /* select h if h < p, or h + -p if h >= p */
 254         mask = (g4 >> ((sizeof(u32) * 8) - 1)) - 1;
 255         g0 &= mask;
 256         g1 &= mask;
 257         g2 &= mask;
 258         g3 &= mask;
 259         g4 &= mask;
 260         mask = ~mask;
 261         h0 = (h0 & mask) | g0;
 262         h1 = (h1 & mask) | g1;
 263         h2 = (h2 & mask) | g2;
 264         h3 = (h3 & mask) | g3;
 265         h4 = (h4 & mask) | g4;
 266
 267         /* h = h % (2^128) */
 268         h0 = (h0 >>  0) | (h1 << 26);
 269         h1 = (h1 >>  6) | (h2 << 20);
 270         h2 = (h2 >> 12) | (h3 << 14);
 271         h3 = (h3 >> 18) | (h4 <<  8);
 272
 273         /* mac = (h + s) % (2^128) */
 274         f = (f >> 32) + h0 + dctx->s[0]; mac[0] = cpu_to_le32(f);
 275         f = (f >> 32) + h1 + dctx->s[1]; mac[1] = cpu_to_le32(f);
 276         f = (f >> 32) + h2 + dctx->s[2]; mac[2] = cpu_to_le32(f);
 277         f = (f >> 32) + h3 + dctx->s[3]; mac[3] = cpu_to_le32(f);
 278
 279         return 0;
 280 }
 281 EXPORT_SYMBOL_GPL(crypto_poly1305_final);
 282
 283 static struct shash_alg poly1305_alg = {
 284         .digestsize     = POLY1305_DIGEST_SIZE,
 285         .init           = crypto_poly1305_init,
 286         .update         = crypto_poly1305_update,
 287         .final          = crypto_poly1305_final,
 288         .setkey         = crypto_poly1305_setkey,
 289         .descsize       = sizeof(struct poly1305_desc_ctx),
 290         .base           = {
 291                 .cra_name               = "poly1305",
 292                 .cra_driver_name        = "poly1305-generic",
 293                 .cra_priority           = 100,
 294                 .cra_flags              = CRYPTO_ALG_TYPE_SHASH,
 295                 .cra_alignmask          = sizeof(u32) - 1,
 296                 .cra_blocksize          = POLY1305_BLOCK_SIZE,
 297                 .cra_module             = THIS_MODULE,
 298         },
 299 };
 300
 301 static int __init poly1305_mod_init(void)
 302 {
 303         return crypto_register_shash(&poly1305_alg);
 304 }
 305
 306 static void __exit poly1305_mod_exit(void)
 307 {
 308         crypto_unregister_shash(&poly1305_alg);
 309 }
 310
 311 module_init(poly1305_mod_init);
 312 module_exit(poly1305_mod_exit);
 313
 314 MODULE_LICENSE("GPL");
 315 MODULE_AUTHOR("Martin Willi <martin@strongswan.org>");
 316 MODULE_DESCRIPTION("Poly1305 authenticator");
 317 MODULE_ALIAS_CRYPTO("poly1305");
 318 MODULE_ALIAS_CRYPTO("poly1305-generic");