1 /* mpihelp-mul.c  -  MPI helper functions
2  * Copyright (C) 1994, 1996, 1998, 1999,
3  *               2000 Free Software Foundation, Inc.
4  *
5  * This file is part of GnuPG.
6  *
7  * GnuPG is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License as published by
9  * the Free Software Foundation; either version 2 of the License, or
10  * (at your option) any later version.
11  *
12  * GnuPG is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with this program; if not, write to the Free Software
19  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA
20  *
21  * Note: This code is heavily based on the GNU MP Library.
22  *	 Actually it's the same code with only minor changes in the
23  *	 way the data is stored; this is to support the abstraction
24  *	 of an optional secure memory allocation which may be used
25  *	 to avoid revealing of sensitive data due to paging etc.
26  *	 The GNU MP Library itself is published under the LGPL;
27  *	 however I decided to publish this code under the plain GPL.
28  */
29 
30 #include <linux/string.h>
31 #include "mpi-internal.h"
32 #include "longlong.h"
33 
34 #define MPN_MUL_N_RECURSE(prodp, up, vp, size, tspace)		\
35 	do {							\
36 		if ((size) < KARATSUBA_THRESHOLD)		\
37 			mul_n_basecase(prodp, up, vp, size);	\
38 		else						\
39 			mul_n(prodp, up, vp, size, tspace);	\
40 	} while (0);
41 
42 #define MPN_SQR_N_RECURSE(prodp, up, size, tspace)		\
43 	do {							\
44 		if ((size) < KARATSUBA_THRESHOLD)		\
45 			mpih_sqr_n_basecase(prodp, up, size);	\
46 		else						\
47 			mpih_sqr_n(prodp, up, size, tspace);	\
48 	} while (0);
49 
50 /* Multiply the natural numbers u (pointed to by UP) and v (pointed to by VP),
51  * both with SIZE limbs, and store the result at PRODP.  2 * SIZE limbs are
52  * always stored.  Return the most significant limb.
53  *
54  * Argument constraints:
55  * 1. PRODP != UP and PRODP != VP, i.e. the destination
56  *    must be distinct from the multiplier and the multiplicand.
57  *
58  *
59  * Handle simple cases with traditional multiplication.
60  *
61  * This is the most critical code of multiplication.  All multiplies rely
62  * on this, both small and huge.  Small ones arrive here immediately.  Huge
63  * ones arrive here as this is the base case for Karatsuba's recursive
64  * algorithm below.
65  */
66 
67 static mpi_limb_t
mul_n_basecase(mpi_ptr_t prodp,mpi_ptr_t up,mpi_ptr_t vp,mpi_size_t size)68 mul_n_basecase(mpi_ptr_t prodp, mpi_ptr_t up, mpi_ptr_t vp, mpi_size_t size)
69 {
70 	mpi_size_t i;
71 	mpi_limb_t cy;
72 	mpi_limb_t v_limb;
73 
74 	/* Multiply by the first limb in V separately, as the result can be
75 	 * stored (not added) to PROD.  We also avoid a loop for zeroing.  */
76 	v_limb = vp[0];
77 	if (v_limb <= 1) {
78 		if (v_limb == 1)
79 			MPN_COPY(prodp, up, size);
80 		else
81 			MPN_ZERO(prodp, size);
82 		cy = 0;
83 	} else
84 		cy = mpihelp_mul_1(prodp, up, size, v_limb);
85 
86 	prodp[size] = cy;
87 	prodp++;
88 
89 	/* For each iteration in the outer loop, multiply one limb from
90 	 * U with one limb from V, and add it to PROD.  */
91 	for (i = 1; i < size; i++) {
92 		v_limb = vp[i];
93 		if (v_limb <= 1) {
94 			cy = 0;
95 			if (v_limb == 1)
96 				cy = mpihelp_add_n(prodp, prodp, up, size);
97 		} else
98 			cy = mpihelp_addmul_1(prodp, up, size, v_limb);
99 
100 		prodp[size] = cy;
101 		prodp++;
102 	}
103 
104 	return cy;
105 }
106 
107 static void
mul_n(mpi_ptr_t prodp,mpi_ptr_t up,mpi_ptr_t vp,mpi_size_t size,mpi_ptr_t tspace)108 mul_n(mpi_ptr_t prodp, mpi_ptr_t up, mpi_ptr_t vp,
109 		mpi_size_t size, mpi_ptr_t tspace)
110 {
111 	if (size & 1) {
112 		/* The size is odd, and the code below doesn't handle that.
113 		 * Multiply the least significant (size - 1) limbs with a recursive
114 		 * call, and handle the most significant limb of S1 and S2
115 		 * separately.
116 		 * A slightly faster way to do this would be to make the Karatsuba
117 		 * code below behave as if the size were even, and let it check for
118 		 * odd size in the end.  I.e., in essence move this code to the end.
119 		 * Doing so would save us a recursive call, and potentially make the
120 		 * stack grow a lot less.
121 		 */
122 		mpi_size_t esize = size - 1;	/* even size */
123 		mpi_limb_t cy_limb;
124 
125 		MPN_MUL_N_RECURSE(prodp, up, vp, esize, tspace);
126 		cy_limb = mpihelp_addmul_1(prodp + esize, up, esize, vp[esize]);
127 		prodp[esize + esize] = cy_limb;
128 		cy_limb = mpihelp_addmul_1(prodp + esize, vp, size, up[esize]);
129 		prodp[esize + size] = cy_limb;
130 	} else {
131 		/* Anatolij Alekseevich Karatsuba's divide-and-conquer algorithm.
132 		 *
133 		 * Split U in two pieces, U1 and U0, such that
134 		 * U = U0 + U1*(B**n),
135 		 * and V in V1 and V0, such that
136 		 * V = V0 + V1*(B**n).
137 		 *
138 		 * UV is then computed recursively using the identity
139 		 *
140 		 *        2n   n          n                     n
141 		 * UV = (B  + B )U V  +  B (U -U )(V -V )  +  (B + 1)U V
142 		 *                1 1        1  0   0  1              0 0
143 		 *
144 		 * Where B = 2**BITS_PER_MP_LIMB.
145 		 */
146 		mpi_size_t hsize = size >> 1;
147 		mpi_limb_t cy;
148 		int negflg;
149 
150 		/* Product H.      ________________  ________________
151 		 *                |_____U1 x V1____||____U0 x V0_____|
152 		 * Put result in upper part of PROD and pass low part of TSPACE
153 		 * as new TSPACE.
154 		 */
155 		MPN_MUL_N_RECURSE(prodp + size, up + hsize, vp + hsize, hsize,
156 				  tspace);
157 
158 		/* Product M.      ________________
159 		 *                |_(U1-U0)(V0-V1)_|
160 		 */
161 		if (mpihelp_cmp(up + hsize, up, hsize) >= 0) {
162 			mpihelp_sub_n(prodp, up + hsize, up, hsize);
163 			negflg = 0;
164 		} else {
165 			mpihelp_sub_n(prodp, up, up + hsize, hsize);
166 			negflg = 1;
167 		}
168 		if (mpihelp_cmp(vp + hsize, vp, hsize) >= 0) {
169 			mpihelp_sub_n(prodp + hsize, vp + hsize, vp, hsize);
170 			negflg ^= 1;
171 		} else {
172 			mpihelp_sub_n(prodp + hsize, vp, vp + hsize, hsize);
173 			/* No change of NEGFLG.  */
174 		}
175 		/* Read temporary operands from low part of PROD.
176 		 * Put result in low part of TSPACE using upper part of TSPACE
177 		 * as new TSPACE.
178 		 */
179 		MPN_MUL_N_RECURSE(tspace, prodp, prodp + hsize, hsize,
180 				  tspace + size);
181 
182 		/* Add/copy product H. */
183 		MPN_COPY(prodp + hsize, prodp + size, hsize);
184 		cy = mpihelp_add_n(prodp + size, prodp + size,
185 				   prodp + size + hsize, hsize);
186 
187 		/* Add product M (if NEGFLG M is a negative number) */
188 		if (negflg)
189 			cy -=
190 			    mpihelp_sub_n(prodp + hsize, prodp + hsize, tspace,
191 					  size);
192 		else
193 			cy +=
194 			    mpihelp_add_n(prodp + hsize, prodp + hsize, tspace,
195 					  size);
196 
197 		/* Product L.      ________________  ________________
198 		 *                |________________||____U0 x V0_____|
199 		 * Read temporary operands from low part of PROD.
200 		 * Put result in low part of TSPACE using upper part of TSPACE
201 		 * as new TSPACE.
202 		 */
203 		MPN_MUL_N_RECURSE(tspace, up, vp, hsize, tspace + size);
204 
205 		/* Add/copy Product L (twice) */
206 
207 		cy += mpihelp_add_n(prodp + hsize, prodp + hsize, tspace, size);
208 		if (cy)
209 			mpihelp_add_1(prodp + hsize + size,
210 				      prodp + hsize + size, hsize, cy);
211 
212 		MPN_COPY(prodp, tspace, hsize);
213 		cy = mpihelp_add_n(prodp + hsize, prodp + hsize, tspace + hsize,
214 				   hsize);
215 		if (cy)
216 			mpihelp_add_1(prodp + size, prodp + size, size, 1);
217 	}
218 }
219 
mpih_sqr_n_basecase(mpi_ptr_t prodp,mpi_ptr_t up,mpi_size_t size)220 void mpih_sqr_n_basecase(mpi_ptr_t prodp, mpi_ptr_t up, mpi_size_t size)
221 {
222 	mpi_size_t i;
223 	mpi_limb_t cy_limb;
224 	mpi_limb_t v_limb;
225 
226 	/* Multiply by the first limb in V separately, as the result can be
227 	 * stored (not added) to PROD.  We also avoid a loop for zeroing.  */
228 	v_limb = up[0];
229 	if (v_limb <= 1) {
230 		if (v_limb == 1)
231 			MPN_COPY(prodp, up, size);
232 		else
233 			MPN_ZERO(prodp, size);
234 		cy_limb = 0;
235 	} else
236 		cy_limb = mpihelp_mul_1(prodp, up, size, v_limb);
237 
238 	prodp[size] = cy_limb;
239 	prodp++;
240 
241 	/* For each iteration in the outer loop, multiply one limb from
242 	 * U with one limb from V, and add it to PROD.  */
243 	for (i = 1; i < size; i++) {
244 		v_limb = up[i];
245 		if (v_limb <= 1) {
246 			cy_limb = 0;
247 			if (v_limb == 1)
248 				cy_limb = mpihelp_add_n(prodp, prodp, up, size);
249 		} else
250 			cy_limb = mpihelp_addmul_1(prodp, up, size, v_limb);
251 
252 		prodp[size] = cy_limb;
253 		prodp++;
254 	}
255 }
256 
257 void
mpih_sqr_n(mpi_ptr_t prodp,mpi_ptr_t up,mpi_size_t size,mpi_ptr_t tspace)258 mpih_sqr_n(mpi_ptr_t prodp, mpi_ptr_t up, mpi_size_t size, mpi_ptr_t tspace)
259 {
260 	if (size & 1) {
261 		/* The size is odd, and the code below doesn't handle that.
262 		 * Multiply the least significant (size - 1) limbs with a recursive
263 		 * call, and handle the most significant limb of S1 and S2
264 		 * separately.
265 		 * A slightly faster way to do this would be to make the Karatsuba
266 		 * code below behave as if the size were even, and let it check for
267 		 * odd size in the end.  I.e., in essence move this code to the end.
268 		 * Doing so would save us a recursive call, and potentially make the
269 		 * stack grow a lot less.
270 		 */
271 		mpi_size_t esize = size - 1;	/* even size */
272 		mpi_limb_t cy_limb;
273 
274 		MPN_SQR_N_RECURSE(prodp, up, esize, tspace);
275 		cy_limb = mpihelp_addmul_1(prodp + esize, up, esize, up[esize]);
276 		prodp[esize + esize] = cy_limb;
277 		cy_limb = mpihelp_addmul_1(prodp + esize, up, size, up[esize]);
278 
279 		prodp[esize + size] = cy_limb;
280 	} else {
281 		mpi_size_t hsize = size >> 1;
282 		mpi_limb_t cy;
283 
284 		/* Product H.      ________________  ________________
285 		 *                |_____U1 x U1____||____U0 x U0_____|
286 		 * Put result in upper part of PROD and pass low part of TSPACE
287 		 * as new TSPACE.
288 		 */
289 		MPN_SQR_N_RECURSE(prodp + size, up + hsize, hsize, tspace);
290 
291 		/* Product M.      ________________
292 		 *                |_(U1-U0)(U0-U1)_|
293 		 */
294 		if (mpihelp_cmp(up + hsize, up, hsize) >= 0)
295 			mpihelp_sub_n(prodp, up + hsize, up, hsize);
296 		else
297 			mpihelp_sub_n(prodp, up, up + hsize, hsize);
298 
299 		/* Read temporary operands from low part of PROD.
300 		 * Put result in low part of TSPACE using upper part of TSPACE
301 		 * as new TSPACE.  */
302 		MPN_SQR_N_RECURSE(tspace, prodp, hsize, tspace + size);
303 
304 		/* Add/copy product H  */
305 		MPN_COPY(prodp + hsize, prodp + size, hsize);
306 		cy = mpihelp_add_n(prodp + size, prodp + size,
307 				   prodp + size + hsize, hsize);
308 
309 		/* Add product M (if NEGFLG M is a negative number).  */
310 		cy -= mpihelp_sub_n(prodp + hsize, prodp + hsize, tspace, size);
311 
312 		/* Product L.      ________________  ________________
313 		 *                |________________||____U0 x U0_____|
314 		 * Read temporary operands from low part of PROD.
315 		 * Put result in low part of TSPACE using upper part of TSPACE
316 		 * as new TSPACE.  */
317 		MPN_SQR_N_RECURSE(tspace, up, hsize, tspace + size);
318 
319 		/* Add/copy Product L (twice).  */
320 		cy += mpihelp_add_n(prodp + hsize, prodp + hsize, tspace, size);
321 		if (cy)
322 			mpihelp_add_1(prodp + hsize + size,
323 				      prodp + hsize + size, hsize, cy);
324 
325 		MPN_COPY(prodp, tspace, hsize);
326 		cy = mpihelp_add_n(prodp + hsize, prodp + hsize, tspace + hsize,
327 				   hsize);
328 		if (cy)
329 			mpihelp_add_1(prodp + size, prodp + size, size, 1);
330 	}
331 }
332 
333 /* This should be made into an inline function in gmp.h.  */
mpihelp_mul_n(mpi_ptr_t prodp,mpi_ptr_t up,mpi_ptr_t vp,mpi_size_t size)334 int mpihelp_mul_n(mpi_ptr_t prodp, mpi_ptr_t up, mpi_ptr_t vp, mpi_size_t size)
335 {
336 	if (up == vp) {
337 		if (size < KARATSUBA_THRESHOLD)
338 			mpih_sqr_n_basecase(prodp, up, size);
339 		else {
340 			mpi_ptr_t tspace;
341 			tspace = mpi_alloc_limb_space(2 * size);
342 			if (!tspace)
343 				return -ENOMEM;
344 			mpih_sqr_n(prodp, up, size, tspace);
345 			mpi_free_limb_space(tspace);
346 		}
347 	} else {
348 		if (size < KARATSUBA_THRESHOLD)
349 			mul_n_basecase(prodp, up, vp, size);
350 		else {
351 			mpi_ptr_t tspace;
352 			tspace = mpi_alloc_limb_space(2 * size);
353 			if (!tspace)
354 				return -ENOMEM;
355 			mul_n(prodp, up, vp, size, tspace);
356 			mpi_free_limb_space(tspace);
357 		}
358 	}
359 
360 	return 0;
361 }
362 
363 int
mpihelp_mul_karatsuba_case(mpi_ptr_t prodp,mpi_ptr_t up,mpi_size_t usize,mpi_ptr_t vp,mpi_size_t vsize,struct karatsuba_ctx * ctx)364 mpihelp_mul_karatsuba_case(mpi_ptr_t prodp,
365 			   mpi_ptr_t up, mpi_size_t usize,
366 			   mpi_ptr_t vp, mpi_size_t vsize,
367 			   struct karatsuba_ctx *ctx)
368 {
369 	mpi_limb_t cy;
370 
371 	if (!ctx->tspace || ctx->tspace_size < vsize) {
372 		if (ctx->tspace)
373 			mpi_free_limb_space(ctx->tspace);
374 		ctx->tspace = mpi_alloc_limb_space(2 * vsize);
375 		if (!ctx->tspace)
376 			return -ENOMEM;
377 		ctx->tspace_size = vsize;
378 	}
379 
380 	MPN_MUL_N_RECURSE(prodp, up, vp, vsize, ctx->tspace);
381 
382 	prodp += vsize;
383 	up += vsize;
384 	usize -= vsize;
385 	if (usize >= vsize) {
386 		if (!ctx->tp || ctx->tp_size < vsize) {
387 			if (ctx->tp)
388 				mpi_free_limb_space(ctx->tp);
389 			ctx->tp = mpi_alloc_limb_space(2 * vsize);
390 			if (!ctx->tp) {
391 				if (ctx->tspace)
392 					mpi_free_limb_space(ctx->tspace);
393 				ctx->tspace = NULL;
394 				return -ENOMEM;
395 			}
396 			ctx->tp_size = vsize;
397 		}
398 
399 		do {
400 			MPN_MUL_N_RECURSE(ctx->tp, up, vp, vsize, ctx->tspace);
401 			cy = mpihelp_add_n(prodp, prodp, ctx->tp, vsize);
402 			mpihelp_add_1(prodp + vsize, ctx->tp + vsize, vsize,
403 				      cy);
404 			prodp += vsize;
405 			up += vsize;
406 			usize -= vsize;
407 		} while (usize >= vsize);
408 	}
409 
410 	if (usize) {
411 		if (usize < KARATSUBA_THRESHOLD) {
412 			mpi_limb_t tmp;
413 			if (mpihelp_mul(ctx->tspace, vp, vsize, up, usize, &tmp)
414 			    < 0)
415 				return -ENOMEM;
416 		} else {
417 			if (!ctx->next) {
418 				ctx->next = kzalloc(sizeof *ctx, GFP_KERNEL);
419 				if (!ctx->next)
420 					return -ENOMEM;
421 			}
422 			if (mpihelp_mul_karatsuba_case(ctx->tspace,
423 						       vp, vsize,
424 						       up, usize,
425 						       ctx->next) < 0)
426 				return -ENOMEM;
427 		}
428 
429 		cy = mpihelp_add_n(prodp, prodp, ctx->tspace, vsize);
430 		mpihelp_add_1(prodp + vsize, ctx->tspace + vsize, usize, cy);
431 	}
432 
433 	return 0;
434 }
435 
mpihelp_release_karatsuba_ctx(struct karatsuba_ctx * ctx)436 void mpihelp_release_karatsuba_ctx(struct karatsuba_ctx *ctx)
437 {
438 	struct karatsuba_ctx *ctx2;
439 
440 	if (ctx->tp)
441 		mpi_free_limb_space(ctx->tp);
442 	if (ctx->tspace)
443 		mpi_free_limb_space(ctx->tspace);
444 	for (ctx = ctx->next; ctx; ctx = ctx2) {
445 		ctx2 = ctx->next;
446 		if (ctx->tp)
447 			mpi_free_limb_space(ctx->tp);
448 		if (ctx->tspace)
449 			mpi_free_limb_space(ctx->tspace);
450 		kfree(ctx);
451 	}
452 }
453 
454 /* Multiply the natural numbers u (pointed to by UP, with USIZE limbs)
455  * and v (pointed to by VP, with VSIZE limbs), and store the result at
456  * PRODP.  USIZE + VSIZE limbs are always stored, but if the input
457  * operands are normalized.  Return the most significant limb of the
458  * result.
459  *
460  * NOTE: The space pointed to by PRODP is overwritten before finished
461  * with U and V, so overlap is an error.
462  *
463  * Argument constraints:
464  * 1. USIZE >= VSIZE.
465  * 2. PRODP != UP and PRODP != VP, i.e. the destination
466  *    must be distinct from the multiplier and the multiplicand.
467  */
468 
469 int
mpihelp_mul(mpi_ptr_t prodp,mpi_ptr_t up,mpi_size_t usize,mpi_ptr_t vp,mpi_size_t vsize,mpi_limb_t * _result)470 mpihelp_mul(mpi_ptr_t prodp, mpi_ptr_t up, mpi_size_t usize,
471 	    mpi_ptr_t vp, mpi_size_t vsize, mpi_limb_t *_result)
472 {
473 	mpi_ptr_t prod_endp = prodp + usize + vsize - 1;
474 	mpi_limb_t cy;
475 	struct karatsuba_ctx ctx;
476 
477 	if (vsize < KARATSUBA_THRESHOLD) {
478 		mpi_size_t i;
479 		mpi_limb_t v_limb;
480 
481 		if (!vsize) {
482 			*_result = 0;
483 			return 0;
484 		}
485 
486 		/* Multiply by the first limb in V separately, as the result can be
487 		 * stored (not added) to PROD.  We also avoid a loop for zeroing.  */
488 		v_limb = vp[0];
489 		if (v_limb <= 1) {
490 			if (v_limb == 1)
491 				MPN_COPY(prodp, up, usize);
492 			else
493 				MPN_ZERO(prodp, usize);
494 			cy = 0;
495 		} else
496 			cy = mpihelp_mul_1(prodp, up, usize, v_limb);
497 
498 		prodp[usize] = cy;
499 		prodp++;
500 
501 		/* For each iteration in the outer loop, multiply one limb from
502 		 * U with one limb from V, and add it to PROD.  */
503 		for (i = 1; i < vsize; i++) {
504 			v_limb = vp[i];
505 			if (v_limb <= 1) {
506 				cy = 0;
507 				if (v_limb == 1)
508 					cy = mpihelp_add_n(prodp, prodp, up,
509 							   usize);
510 			} else
511 				cy = mpihelp_addmul_1(prodp, up, usize, v_limb);
512 
513 			prodp[usize] = cy;
514 			prodp++;
515 		}
516 
517 		*_result = cy;
518 		return 0;
519 	}
520 
521 	memset(&ctx, 0, sizeof ctx);
522 	if (mpihelp_mul_karatsuba_case(prodp, up, usize, vp, vsize, &ctx) < 0)
523 		return -ENOMEM;
524 	mpihelp_release_karatsuba_ctx(&ctx);
525 	*_result = *prod_endp;
526 	return 0;
527 }
528