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 #include <asm/unaligned.h>
  21
  22 static inline u64 mlt(u64 a, u64 b)
  23 {
  24         return a * b;
  25 }
  26
  27 static inline u32 sr(u64 v, u_char n)
  28 {
  29         return v >> n;
  30 }
  31
  32 static inline u32 and(u32 v, u32 mask)
  33 {
  34         return v & mask;
  35 }
  36
  37 int crypto_poly1305_init(struct shash_desc *desc)
  38 {
  39         struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
  40
  41         memset(dctx->h, 0, sizeof(dctx->h));
  42         dctx->buflen = 0;
  43         dctx->rset = false;
  44         dctx->sset = false;
  45
  46         return 0;
  47 }
  48 EXPORT_SYMBOL_GPL(crypto_poly1305_init);
  49
  50 int crypto_poly1305_setkey(struct crypto_shash *tfm,
  51                            const u8 *key, unsigned int keylen)
  52 {
  53         /* Poly1305 requires a unique key for each tag, which implies that
  54          * we can't set it on the tfm that gets accessed by multiple users
  55          * simultaneously. Instead we expect the key as the first 32 bytes in
  56          * the update() call. */
  57         return -ENOTSUPP;
  58 }
  59 EXPORT_SYMBOL_GPL(crypto_poly1305_setkey);
  60
  61 static void poly1305_setrkey(struct poly1305_desc_ctx *dctx, const u8 *key)
  62 {
  63         /* r &= 0xffffffc0ffffffc0ffffffc0fffffff */
  64         dctx->r[0] = (get_unaligned_le32(key +  0) >> 0) & 0x3ffffff;
  65         dctx->r[1] = (get_unaligned_le32(key +  3) >> 2) & 0x3ffff03;
  66         dctx->r[2] = (get_unaligned_le32(key +  6) >> 4) & 0x3ffc0ff;
  67         dctx->r[3] = (get_unaligned_le32(key +  9) >> 6) & 0x3f03fff;
  68         dctx->r[4] = (get_unaligned_le32(key + 12) >> 8) & 0x00fffff;
  69 }
  70
  71 static void poly1305_setskey(struct poly1305_desc_ctx *dctx, const u8 *key)
  72 {
  73         dctx->s[0] = get_unaligned_le32(key +  0);
  74         dctx->s[1] = get_unaligned_le32(key +  4);
  75         dctx->s[2] = get_unaligned_le32(key +  8);
  76         dctx->s[3] = get_unaligned_le32(key + 12);
  77 }
  78
  79 unsigned int crypto_poly1305_setdesckey(struct poly1305_desc_ctx *dctx,
  80                                         const u8 *src, unsigned int srclen)
  81 {
  82         if (!dctx->sset) {
  83                 if (!dctx->rset && srclen >= POLY1305_BLOCK_SIZE) {
  84                         poly1305_setrkey(dctx, src);
  85                         src += POLY1305_BLOCK_SIZE;
  86                         srclen -= POLY1305_BLOCK_SIZE;
  87                         dctx->rset = true;
  88                 }
  89                 if (srclen >= POLY1305_BLOCK_SIZE) {
  90                         poly1305_setskey(dctx, src);
  91                         src += POLY1305_BLOCK_SIZE;
  92                         srclen -= POLY1305_BLOCK_SIZE;
  93                         dctx->sset = true;
  94                 }
  95         }
  96         return srclen;
  97 }
  98 EXPORT_SYMBOL_GPL(crypto_poly1305_setdesckey);
  99
 100 static unsigned int poly1305_blocks(struct poly1305_desc_ctx *dctx,
 101                                     const u8 *src, unsigned int srclen,
 102                                     u32 hibit)
 103 {
 104         u32 r0, r1, r2, r3, r4;
 105         u32 s1, s2, s3, s4;
 106         u32 h0, h1, h2, h3, h4;
 107         u64 d0, d1, d2, d3, d4;
 108         unsigned int datalen;
 109
 110         if (unlikely(!dctx->sset)) {
 111                 datalen = crypto_poly1305_setdesckey(dctx, src, srclen);
 112                 src += srclen - datalen;
 113                 srclen = datalen;
 114         }
 115
 116         r0 = dctx->r[0];
 117         r1 = dctx->r[1];
 118         r2 = dctx->r[2];
 119         r3 = dctx->r[3];
 120         r4 = dctx->r[4];
 121
 122         s1 = r1 * 5;
 123         s2 = r2 * 5;
 124         s3 = r3 * 5;
 125         s4 = r4 * 5;
 126
 127         h0 = dctx->h[0];
 128         h1 = dctx->h[1];
 129         h2 = dctx->h[2];
 130         h3 = dctx->h[3];
 131         h4 = dctx->h[4];
 132
 133         while (likely(srclen >= POLY1305_BLOCK_SIZE)) {
 134
 135                 /* h += m[i] */
 136                 h0 += (get_unaligned_le32(src +  0) >> 0) & 0x3ffffff;
 137                 h1 += (get_unaligned_le32(src +  3) >> 2) & 0x3ffffff;
 138                 h2 += (get_unaligned_le32(src +  6) >> 4) & 0x3ffffff;
 139                 h3 += (get_unaligned_le32(src +  9) >> 6) & 0x3ffffff;
 140                 h4 += (get_unaligned_le32(src + 12) >> 8) | hibit;
 141
 142                 /* h *= r */
 143                 d0 = mlt(h0, r0) + mlt(h1, s4) + mlt(h2, s3) +
 144                      mlt(h3, s2) + mlt(h4, s1);
 145                 d1 = mlt(h0, r1) + mlt(h1, r0) + mlt(h2, s4) +
 146                      mlt(h3, s3) + mlt(h4, s2);
 147                 d2 = mlt(h0, r2) + mlt(h1, r1) + mlt(h2, r0) +
 148                      mlt(h3, s4) + mlt(h4, s3);
 149                 d3 = mlt(h0, r3) + mlt(h1, r2) + mlt(h2, r1) +
 150                      mlt(h3, r0) + mlt(h4, s4);
 151                 d4 = mlt(h0, r4) + mlt(h1, r3) + mlt(h2, r2) +
 152                      mlt(h3, r1) + mlt(h4, r0);
 153
 154                 /* (partial) h %= p */
 155                 d1 += sr(d0, 26);     h0 = and(d0, 0x3ffffff);
 156                 d2 += sr(d1, 26);     h1 = and(d1, 0x3ffffff);
 157                 d3 += sr(d2, 26);     h2 = and(d2, 0x3ffffff);
 158                 d4 += sr(d3, 26);     h3 = and(d3, 0x3ffffff);
 159                 h0 += sr(d4, 26) * 5; h4 = and(d4, 0x3ffffff);
 160                 h1 += h0 >> 26;       h0 = h0 & 0x3ffffff;
 161
 162                 src += POLY1305_BLOCK_SIZE;
 163                 srclen -= POLY1305_BLOCK_SIZE;
 164         }
 165
 166         dctx->h[0] = h0;
 167         dctx->h[1] = h1;
 168         dctx->h[2] = h2;
 169         dctx->h[3] = h3;
 170         dctx->h[4] = h4;
 171
 172         return srclen;
 173 }
 174
 175 int crypto_poly1305_update(struct shash_desc *desc,
 176                            const u8 *src, unsigned int srclen)
 177 {
 178         struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
 179         unsigned int bytes;
 180
 181         if (unlikely(dctx->buflen)) {
 182                 bytes = min(srclen, POLY1305_BLOCK_SIZE - dctx->buflen);
 183                 memcpy(dctx->buf + dctx->buflen, src, bytes);
 184                 src += bytes;
 185                 srclen -= bytes;
 186                 dctx->buflen += bytes;
 187
 188                 if (dctx->buflen == POLY1305_BLOCK_SIZE) {
 189                         poly1305_blocks(dctx, dctx->buf,
 190                                         POLY1305_BLOCK_SIZE, 1 << 24);
 191                         dctx->buflen = 0;
 192                 }
 193         }
 194
 195         if (likely(srclen >= POLY1305_BLOCK_SIZE)) {
 196                 bytes = poly1305_blocks(dctx, src, srclen, 1 << 24);
 197                 src += srclen - bytes;
 198                 srclen = bytes;
 199         }
 200
 201         if (unlikely(srclen)) {
 202                 dctx->buflen = srclen;
 203                 memcpy(dctx->buf, src, srclen);
 204         }
 205
 206         return 0;
 207 }
 208 EXPORT_SYMBOL_GPL(crypto_poly1305_update);
 209
 210 int crypto_poly1305_final(struct shash_desc *desc, u8 *dst)
 211 {
 212         struct poly1305_desc_ctx *dctx = shash_desc_ctx(desc);
 213         __le32 *mac = (__le32 *)dst;
 214         u32 h0, h1, h2, h3, h4;
 215         u32 g0, g1, g2, g3, g4;
 216         u32 mask;
 217         u64 f = 0;
 218
 219         if (unlikely(!dctx->sset))
 220                 return -ENOKEY;
 221
 222         if (unlikely(dctx->buflen)) {
 223                 dctx->buf[dctx->buflen++] = 1;
 224                 memset(dctx->buf + dctx->buflen, 0,
 225                        POLY1305_BLOCK_SIZE - dctx->buflen);
 226                 poly1305_blocks(dctx, dctx->buf, POLY1305_BLOCK_SIZE, 0);
 227         }
 228
 229         /* fully carry h */
 230         h0 = dctx->h[0];
 231         h1 = dctx->h[1];
 232         h2 = dctx->h[2];
 233         h3 = dctx->h[3];
 234         h4 = dctx->h[4];
 235
 236         h2 += (h1 >> 26);     h1 = h1 & 0x3ffffff;
 237         h3 += (h2 >> 26);     h2 = h2 & 0x3ffffff;
 238         h4 += (h3 >> 26);     h3 = h3 & 0x3ffffff;
 239         h0 += (h4 >> 26) * 5; h4 = h4 & 0x3ffffff;
 240         h1 += (h0 >> 26);     h0 = h0 & 0x3ffffff;
 241
 242         /* compute h + -p */
 243         g0 = h0 + 5;
 244         g1 = h1 + (g0 >> 26);             g0 &= 0x3ffffff;
 245         g2 = h2 + (g1 >> 26);             g1 &= 0x3ffffff;
 246         g3 = h3 + (g2 >> 26);             g2 &= 0x3ffffff;
 247         g4 = h4 + (g3 >> 26) - (1 << 26); g3 &= 0x3ffffff;
 248
 249         /* select h if h < p, or h + -p if h >= p */
 250         mask = (g4 >> ((sizeof(u32) * 8) - 1)) - 1;
 251         g0 &= mask;
 252         g1 &= mask;
 253         g2 &= mask;
 254         g3 &= mask;
 255         g4 &= mask;
 256         mask = ~mask;
 257         h0 = (h0 & mask) | g0;
 258         h1 = (h1 & mask) | g1;
 259         h2 = (h2 & mask) | g2;
 260         h3 = (h3 & mask) | g3;
 261         h4 = (h4 & mask) | g4;
 262
 263         /* h = h % (2^128) */
 264         h0 = (h0 >>  0) | (h1 << 26);
 265         h1 = (h1 >>  6) | (h2 << 20);
 266         h2 = (h2 >> 12) | (h3 << 14);
 267         h3 = (h3 >> 18) | (h4 <<  8);
 268
 269         /* mac = (h + s) % (2^128) */
 270         f = (f >> 32) + h0 + dctx->s[0]; mac[0] = cpu_to_le32(f);
 271         f = (f >> 32) + h1 + dctx->s[1]; mac[1] = cpu_to_le32(f);
 272         f = (f >> 32) + h2 + dctx->s[2]; mac[2] = cpu_to_le32(f);
 273         f = (f >> 32) + h3 + dctx->s[3]; mac[3] = cpu_to_le32(f);
 274
 275         return 0;
 276 }
 277 EXPORT_SYMBOL_GPL(crypto_poly1305_final);
 278
 279 static struct shash_alg poly1305_alg = {
 280         .digestsize     = POLY1305_DIGEST_SIZE,
 281         .init           = crypto_poly1305_init,
 282         .update         = crypto_poly1305_update,
 283         .final          = crypto_poly1305_final,
 284         .setkey         = crypto_poly1305_setkey,
 285         .descsize       = sizeof(struct poly1305_desc_ctx),
 286         .base           = {
 287                 .cra_name               = "poly1305",
 288                 .cra_driver_name        = "poly1305-generic",
 289                 .cra_priority           = 100,
 290                 .cra_flags              = CRYPTO_ALG_TYPE_SHASH,
 291                 .cra_alignmask          = sizeof(u32) - 1,
 292                 .cra_blocksize          = POLY1305_BLOCK_SIZE,
 293                 .cra_module             = THIS_MODULE,
 294         },
 295 };
 296
 297 static int __init poly1305_mod_init(void)
 298 {
 299         return crypto_register_shash(&poly1305_alg);
 300 }
 301
 302 static void __exit poly1305_mod_exit(void)
 303 {
 304         crypto_unregister_shash(&poly1305_alg);
 305 }
 306
 307 module_init(poly1305_mod_init);
 308 module_exit(poly1305_mod_exit);
 309
 310 MODULE_LICENSE("GPL");
 311 MODULE_AUTHOR("Martin Willi <martin@strongswan.org>");
 312 MODULE_DESCRIPTION("Poly1305 authenticator");
 313 MODULE_ALIAS_CRYPTO("poly1305");
 314 MODULE_ALIAS_CRYPTO("poly1305-generic");