1 /*-
2  * Copyright 2009 Colin Percival
3  * Copyright 2012,2013 Alexander Peslyak
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * This file was originally written by Colin Percival as part of the Tarsnap
28  * online backup system.
29  */
30 
31 #include <errno.h>
32 #include <limits.h>
33 #include <stdint.h>
34 #include <stdlib.h>
35 #include <string.h>
36 
37 #include "private/common.h"
38 #include "private/sse2_64_32.h"
39 
40 #ifdef HAVE_EMMINTRIN_H
41 
42 # ifdef __GNUC__
43 #  pragma GCC target("sse2")
44 # endif
45 # include <emmintrin.h>
46 # if defined(__XOP__) && defined(DISABLED)
47 #  include <x86intrin.h>
48 # endif
49 
50 # include "../crypto_scrypt.h"
51 # include "../pbkdf2-sha256.h"
52 
53 # if defined(__XOP__) && defined(DISABLED)
54 #  define ARX(out, in1, in2, s) \
55     out = _mm_xor_si128(out, _mm_roti_epi32(_mm_add_epi32(in1, in2), s));
56 # else
57 #  define ARX(out, in1, in2, s)                                    \
58     {                                                              \
59         __m128i T = _mm_add_epi32(in1, in2);                       \
60         out       = _mm_xor_si128(out, _mm_slli_epi32(T, s));      \
61         out       = _mm_xor_si128(out, _mm_srli_epi32(T, 32 - s)); \
62     }
63 # endif
64 
65 # define SALSA20_2ROUNDS              \
66     /* Operate on "columns". */       \
67     ARX(X1, X0, X3, 7)                \
68     ARX(X2, X1, X0, 9)                \
69     ARX(X3, X2, X1, 13)               \
70     ARX(X0, X3, X2, 18)               \
71                                       \
72     /* Rearrange data. */             \
73     X1 = _mm_shuffle_epi32(X1, 0x93); \
74     X2 = _mm_shuffle_epi32(X2, 0x4E); \
75     X3 = _mm_shuffle_epi32(X3, 0x39); \
76                                       \
77     /* Operate on "rows". */          \
78     ARX(X3, X0, X1, 7)                \
79     ARX(X2, X3, X0, 9)                \
80     ARX(X1, X2, X3, 13)               \
81     ARX(X0, X1, X2, 18)               \
82                                       \
83     /* Rearrange data. */             \
84     X1 = _mm_shuffle_epi32(X1, 0x39); \
85     X2 = _mm_shuffle_epi32(X2, 0x4E); \
86     X3 = _mm_shuffle_epi32(X3, 0x93);
87 
88 /**
89  * Apply the salsa20/8 core to the block provided in (X0 ... X3) ^ (Z0 ... Z3).
90  */
91 # define SALSA20_8_XOR(in, out)                               \
92     {                                                         \
93         __m128i Y0 = X0 = _mm_xor_si128(X0, (in)[0]);         \
94         __m128i Y1 = X1 = _mm_xor_si128(X1, (in)[1]);         \
95         __m128i Y2 = X2 = _mm_xor_si128(X2, (in)[2]);         \
96         __m128i Y3 = X3 = _mm_xor_si128(X3, (in)[3]);         \
97         SALSA20_2ROUNDS                                       \
98         SALSA20_2ROUNDS                                       \
99         SALSA20_2ROUNDS                                       \
100         SALSA20_2ROUNDS(out)[0] = X0 = _mm_add_epi32(X0, Y0); \
101         (out)[1] = X1 = _mm_add_epi32(X1, Y1);                \
102         (out)[2] = X2 = _mm_add_epi32(X2, Y2);                \
103         (out)[3] = X3 = _mm_add_epi32(X3, Y3);                \
104     }
105 
106 /**
107  * blockmix_salsa8(Bin, Bout, r):
108  * Compute Bout = BlockMix_{salsa20/8, r}(Bin).  The input Bin must be 128r
109  * bytes in length; the output Bout must also be the same size.
110  */
111 static inline void
blockmix_salsa8(const __m128i * Bin,__m128i * Bout,size_t r)112 blockmix_salsa8(const __m128i *Bin, __m128i *Bout, size_t r)
113 {
114     __m128i X0, X1, X2, X3;
115     size_t  i;
116 
117     /* 1: X <-- B_{2r - 1} */
118     X0 = Bin[8 * r - 4];
119     X1 = Bin[8 * r - 3];
120     X2 = Bin[8 * r - 2];
121     X3 = Bin[8 * r - 1];
122 
123     /* 3: X <-- H(X \xor B_i) */
124     /* 4: Y_i <-- X */
125     /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
126     SALSA20_8_XOR(Bin, Bout)
127 
128     /* 2: for i = 0 to 2r - 1 do */
129     r--;
130     for (i = 0; i < r;) {
131         /* 3: X <-- H(X \xor B_i) */
132         /* 4: Y_i <-- X */
133         /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
134         SALSA20_8_XOR(&Bin[i * 8 + 4], &Bout[(r + i) * 4 + 4])
135 
136         i++;
137 
138         /* 3: X <-- H(X \xor B_i) */
139         /* 4: Y_i <-- X */
140         /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
141         SALSA20_8_XOR(&Bin[i * 8], &Bout[i * 4])
142     }
143 
144     /* 3: X <-- H(X \xor B_i) */
145     /* 4: Y_i <-- X */
146     /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
147     SALSA20_8_XOR(&Bin[i * 8 + 4], &Bout[(r + i) * 4 + 4])
148 }
149 
150 # define XOR4(in)                    \
151     X0 = _mm_xor_si128(X0, (in)[0]); \
152     X1 = _mm_xor_si128(X1, (in)[1]); \
153     X2 = _mm_xor_si128(X2, (in)[2]); \
154     X3 = _mm_xor_si128(X3, (in)[3]);
155 
156 # define XOR4_2(in1, in2)                   \
157     X0 = _mm_xor_si128((in1)[0], (in2)[0]); \
158     X1 = _mm_xor_si128((in1)[1], (in2)[1]); \
159     X2 = _mm_xor_si128((in1)[2], (in2)[2]); \
160     X3 = _mm_xor_si128((in1)[3], (in2)[3]);
161 
162 static inline uint32_t
blockmix_salsa8_xor(const __m128i * Bin1,const __m128i * Bin2,__m128i * Bout,size_t r)163 blockmix_salsa8_xor(const __m128i *Bin1, const __m128i *Bin2, __m128i *Bout,
164                     size_t r)
165 {
166     __m128i X0, X1, X2, X3;
167     size_t  i;
168 
169     /* 1: X <-- B_{2r - 1} */
170     XOR4_2(&Bin1[8 * r - 4], &Bin2[8 * r - 4])
171 
172     /* 3: X <-- H(X \xor B_i) */
173     /* 4: Y_i <-- X */
174     /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
175     XOR4(Bin1)
176     SALSA20_8_XOR(Bin2, Bout)
177 
178     /* 2: for i = 0 to 2r - 1 do */
179     r--;
180     for (i = 0; i < r;) {
181         /* 3: X <-- H(X \xor B_i) */
182         /* 4: Y_i <-- X */
183         /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
184         XOR4(&Bin1[i * 8 + 4])
185         SALSA20_8_XOR(&Bin2[i * 8 + 4], &Bout[(r + i) * 4 + 4])
186 
187         i++;
188 
189         /* 3: X <-- H(X \xor B_i) */
190         /* 4: Y_i <-- X */
191         /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
192         XOR4(&Bin1[i * 8])
193         SALSA20_8_XOR(&Bin2[i * 8], &Bout[i * 4])
194     }
195 
196     /* 3: X <-- H(X \xor B_i) */
197     /* 4: Y_i <-- X */
198     /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
199     XOR4(&Bin1[i * 8 + 4])
200     SALSA20_8_XOR(&Bin2[i * 8 + 4], &Bout[(r + i) * 4 + 4])
201 
202     return _mm_cvtsi128_si32(X0);
203 }
204 
205 # undef ARX
206 # undef SALSA20_2ROUNDS
207 # undef SALSA20_8_XOR
208 # undef XOR4
209 # undef XOR4_2
210 
211 /**
212  * integerify(B, r):
213  * Return the result of parsing B_{2r-1} as a little-endian integer.
214  * Note that B's layout is permuted compared to the generic implementation.
215  */
216 static inline uint32_t
integerify(const void * B,size_t r)217 integerify(const void *B, size_t r)
218 {
219     return *(const uint32_t *) ((uintptr_t)(B) + (2 * r - 1) * 64);
220 }
221 
222 /**
223  * smix(B, r, N, V, XY):
224  * Compute B = SMix_r(B, N).  The input B must be 128r bytes in length;
225  * the temporary storage V must be 128rN bytes in length; the temporary
226  * storage XY must be 256r + 64 bytes in length.  The value N must be a
227  * power of 2 greater than 1.  The arrays B, V, and XY must be aligned to a
228  * multiple of 64 bytes.
229  */
230 static void
smix(uint8_t * B,size_t r,uint32_t N,void * V,void * XY)231 smix(uint8_t *B, size_t r, uint32_t N, void *V, void *XY)
232 {
233     size_t    s   = 128 * r;
234     __m128i * X   = (__m128i *) V, *Y;
235     uint32_t *X32 = (uint32_t *) V;
236     uint32_t  i, j;
237     size_t    k;
238 
239     /* 1: X <-- B */
240     /* 3: V_i <-- X */
241     for (k = 0; k < 2 * r; k++) {
242         for (i = 0; i < 16; i++) {
243             X32[k * 16 + i] = LOAD32_LE(&B[(k * 16 + (i * 5 % 16)) * 4]);
244         }
245     }
246 
247     /* 2: for i = 0 to N - 1 do */
248     for (i = 1; i < N - 1; i += 2) {
249         /* 4: X <-- H(X) */
250         /* 3: V_i <-- X */
251         Y = (__m128i *) ((uintptr_t)(V) + i * s);
252         blockmix_salsa8(X, Y, r);
253 
254         /* 4: X <-- H(X) */
255         /* 3: V_i <-- X */
256         X = (__m128i *) ((uintptr_t)(V) + (i + 1) * s);
257         blockmix_salsa8(Y, X, r);
258     }
259 
260     /* 4: X <-- H(X) */
261     /* 3: V_i <-- X */
262     Y = (__m128i *) ((uintptr_t)(V) + i * s);
263     blockmix_salsa8(X, Y, r);
264 
265     /* 4: X <-- H(X) */
266     /* 3: V_i <-- X */
267     X = (__m128i *) XY;
268     blockmix_salsa8(Y, X, r);
269 
270     X32 = (uint32_t *) XY;
271     Y   = (__m128i *) ((uintptr_t)(XY) + s);
272 
273     /* 7: j <-- Integerify(X) mod N */
274     j = integerify(X, r) & (N - 1);
275 
276     /* 6: for i = 0 to N - 1 do */
277     for (i = 0; i < N; i += 2) {
278         __m128i *V_j = (__m128i *) ((uintptr_t)(V) + j * s);
279 
280         /* 8: X <-- H(X \xor V_j) */
281         /* 7: j <-- Integerify(X) mod N */
282         j   = blockmix_salsa8_xor(X, V_j, Y, r) & (N - 1);
283         V_j = (__m128i *) ((uintptr_t)(V) + j * s);
284 
285         /* 8: X <-- H(X \xor V_j) */
286         /* 7: j <-- Integerify(X) mod N */
287         j = blockmix_salsa8_xor(Y, V_j, X, r) & (N - 1);
288     }
289 
290     /* 10: B' <-- X */
291     for (k = 0; k < 2 * r; k++) {
292         for (i = 0; i < 16; i++) {
293             STORE32_LE(&B[(k * 16 + (i * 5 % 16)) * 4], X32[k * 16 + i]);
294         }
295     }
296 }
297 
298 /**
299  * escrypt_kdf(local, passwd, passwdlen, salt, saltlen,
300  *     N, r, p, buf, buflen):
301  * Compute scrypt(passwd[0 .. passwdlen - 1], salt[0 .. saltlen - 1], N, r,
302  * p, buflen) and write the result into buf.  The parameters r, p, and buflen
303  * must satisfy r * p < 2^30 and buflen <= (2^32 - 1) * 32.  The parameter N
304  * must be a power of 2 greater than 1.
305  *
306  * Return 0 on success; or -1 on error.
307  */
308 int
escrypt_kdf_sse(escrypt_local_t * local,const uint8_t * passwd,size_t passwdlen,const uint8_t * salt,size_t saltlen,uint64_t N,uint32_t _r,uint32_t _p,uint8_t * buf,size_t buflen)309 escrypt_kdf_sse(escrypt_local_t *local, const uint8_t *passwd, size_t passwdlen,
310                 const uint8_t *salt, size_t saltlen, uint64_t N, uint32_t _r,
311                 uint32_t _p, uint8_t *buf, size_t buflen)
312 {
313     size_t    B_size, V_size, XY_size, need;
314     uint8_t * B;
315     uint32_t *V, *XY;
316     size_t    r = _r, p = _p;
317     uint32_t  i;
318 
319 /* Sanity-check parameters. */
320 # if SIZE_MAX > UINT32_MAX
321 /* LCOV_EXCL_START */
322     if (buflen > (((uint64_t)(1) << 32) - 1) * 32) {
323         errno = EFBIG;
324         return -1;
325     }
326 /* LCOV_EXCL_END */
327 # endif
328     if ((uint64_t)(r) * (uint64_t)(p) >= ((uint64_t) 1 << 30)) {
329         errno = EFBIG;
330         return -1;
331     }
332     if (N > UINT32_MAX) {
333         errno = EFBIG;
334         return -1;
335     }
336     if (((N & (N - 1)) != 0) || (N < 2)) {
337         errno = EINVAL;
338         return -1;
339     }
340     if (r == 0 || p == 0) {
341         errno = EINVAL;
342         return -1;
343     }
344 /* LCOV_EXCL_START */
345     if ((r > SIZE_MAX / 128 / p) ||
346 # if SIZE_MAX / 256 <= UINT32_MAX
347         (r > SIZE_MAX / 256) ||
348 # endif
349         (N > SIZE_MAX / 128 / r)) {
350         errno = ENOMEM;
351         return -1;
352     }
353 /* LCOV_EXCL_END */
354 
355     /* Allocate memory. */
356     B_size = (size_t) 128 * r * p;
357     V_size = (size_t) 128 * r * N;
358     need   = B_size + V_size;
359 /* LCOV_EXCL_START */
360     if (need < V_size) {
361         errno = ENOMEM;
362         return -1;
363     }
364 /* LCOV_EXCL_END */
365     XY_size = (size_t) 256 * r + 64;
366     need += XY_size;
367 /* LCOV_EXCL_START */
368     if (need < XY_size) {
369         errno = ENOMEM;
370         return -1;
371     }
372 /* LCOV_EXCL_END */
373     if (local->size < need) {
374         if (free_region(local)) {
375             return -1; /* LCOV_EXCL_LINE */
376         }
377         if (!alloc_region(local, need)) {
378             return -1; /* LCOV_EXCL_LINE */
379         }
380     }
381     B  = (uint8_t *) local->aligned;
382     V  = (uint32_t *) ((uint8_t *) B + B_size);
383     XY = (uint32_t *) ((uint8_t *) V + V_size);
384 
385     /* 1: (B_0 ... B_{p-1}) <-- PBKDF2(P, S, 1, p * MFLen) */
386     PBKDF2_SHA256(passwd, passwdlen, salt, saltlen, 1, B, B_size);
387 
388     /* 2: for i = 0 to p - 1 do */
389     for (i = 0; i < p; i++) {
390         /* 3: B_i <-- MF(B_i, N) */
391         smix(&B[(size_t) 128 * i * r], r, (uint32_t) N, V, XY);
392     }
393 
394     /* 5: DK <-- PBKDF2(P, B, 1, dkLen) */
395     PBKDF2_SHA256(passwd, passwdlen, B, B_size, 1, buf, buflen);
396 
397     /* Success! */
398     return 0;
399 }
400 #endif
401