]>
Commit | Line | Data |
---|---|---|
d02b48c6 | 1 | /* crypto/bn/bn_div.c */ |
58964a49 | 2 | /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com) |
d02b48c6 RE |
3 | * All rights reserved. |
4 | * | |
5 | * This package is an SSL implementation written | |
6 | * by Eric Young (eay@cryptsoft.com). | |
7 | * The implementation was written so as to conform with Netscapes SSL. | |
8 | * | |
9 | * This library is free for commercial and non-commercial use as long as | |
10 | * the following conditions are aheared to. The following conditions | |
11 | * apply to all code found in this distribution, be it the RC4, RSA, | |
12 | * lhash, DES, etc., code; not just the SSL code. The SSL documentation | |
13 | * included with this distribution is covered by the same copyright terms | |
14 | * except that the holder is Tim Hudson (tjh@cryptsoft.com). | |
15 | * | |
16 | * Copyright remains Eric Young's, and as such any Copyright notices in | |
17 | * the code are not to be removed. | |
18 | * If this package is used in a product, Eric Young should be given attribution | |
19 | * as the author of the parts of the library used. | |
20 | * This can be in the form of a textual message at program startup or | |
21 | * in documentation (online or textual) provided with the package. | |
22 | * | |
23 | * Redistribution and use in source and binary forms, with or without | |
24 | * modification, are permitted provided that the following conditions | |
25 | * are met: | |
26 | * 1. Redistributions of source code must retain the copyright | |
27 | * notice, this list of conditions and the following disclaimer. | |
28 | * 2. Redistributions in binary form must reproduce the above copyright | |
29 | * notice, this list of conditions and the following disclaimer in the | |
30 | * documentation and/or other materials provided with the distribution. | |
31 | * 3. All advertising materials mentioning features or use of this software | |
32 | * must display the following acknowledgement: | |
33 | * "This product includes cryptographic software written by | |
34 | * Eric Young (eay@cryptsoft.com)" | |
35 | * The word 'cryptographic' can be left out if the rouines from the library | |
36 | * being used are not cryptographic related :-). | |
37 | * 4. If you include any Windows specific code (or a derivative thereof) from | |
38 | * the apps directory (application code) you must include an acknowledgement: | |
39 | * "This product includes software written by Tim Hudson (tjh@cryptsoft.com)" | |
40 | * | |
41 | * THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND | |
42 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
43 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
44 | * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE | |
45 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | |
46 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS | |
47 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
48 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT | |
49 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY | |
50 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF | |
51 | * SUCH DAMAGE. | |
52 | * | |
53 | * The licence and distribution terms for any publically available version or | |
54 | * derivative of this code cannot be changed. i.e. this code cannot simply be | |
55 | * copied and put under another distribution licence | |
56 | * [including the GNU Public Licence.] | |
57 | */ | |
58 | ||
59 | #include <stdio.h> | |
84c15db5 | 60 | #include <openssl/bn.h> |
d02b48c6 RE |
61 | #include "cryptlib.h" |
62 | #include "bn_lcl.h" | |
63 | ||
78a0c1f1 | 64 | |
d02b48c6 | 65 | /* The old slow way */ |
4a6222d7 | 66 | #if 0 |
0bde1089 UM |
67 | int BN_div(BIGNUM *dv, BIGNUM *rem, const BIGNUM *m, const BIGNUM *d, |
68 | BN_CTX *ctx) | |
d02b48c6 RE |
69 | { |
70 | int i,nm,nd; | |
9b141126 | 71 | int ret = 0; |
d02b48c6 RE |
72 | BIGNUM *D; |
73 | ||
dfeab068 RE |
74 | bn_check_top(m); |
75 | bn_check_top(d); | |
d02b48c6 RE |
76 | if (BN_is_zero(d)) |
77 | { | |
78 | BNerr(BN_F_BN_DIV,BN_R_DIV_BY_ZERO); | |
79 | return(0); | |
80 | } | |
81 | ||
82 | if (BN_ucmp(m,d) < 0) | |
83 | { | |
84 | if (rem != NULL) | |
85 | { if (BN_copy(rem,m) == NULL) return(0); } | |
86 | if (dv != NULL) BN_zero(dv); | |
87 | return(1); | |
88 | } | |
89 | ||
9b141126 UM |
90 | BN_CTX_start(ctx); |
91 | D = BN_CTX_get(ctx); | |
92 | if (dv == NULL) dv = BN_CTX_get(ctx); | |
93 | if (rem == NULL) rem = BN_CTX_get(ctx); | |
94 | if (D == NULL || dv == NULL || rem == NULL) | |
95 | goto end; | |
d02b48c6 RE |
96 | |
97 | nd=BN_num_bits(d); | |
98 | nm=BN_num_bits(m); | |
9b141126 UM |
99 | if (BN_copy(D,d) == NULL) goto end; |
100 | if (BN_copy(rem,m) == NULL) goto end; | |
d02b48c6 RE |
101 | |
102 | /* The next 2 are needed so we can do a dv->d[0]|=1 later | |
103 | * since BN_lshift1 will only work once there is a value :-) */ | |
104 | BN_zero(dv); | |
dfeab068 | 105 | bn_wexpand(dv,1); |
d02b48c6 RE |
106 | dv->top=1; |
107 | ||
9b141126 | 108 | if (!BN_lshift(D,D,nm-nd)) goto end; |
d02b48c6 RE |
109 | for (i=nm-nd; i>=0; i--) |
110 | { | |
9b141126 | 111 | if (!BN_lshift1(dv,dv)) goto end; |
d02b48c6 RE |
112 | if (BN_ucmp(rem,D) >= 0) |
113 | { | |
114 | dv->d[0]|=1; | |
9b141126 | 115 | if (!BN_usub(rem,rem,D)) goto end; |
d02b48c6 RE |
116 | } |
117 | /* CAN IMPROVE (and have now :=) */ | |
9b141126 | 118 | if (!BN_rshift1(D,D)) goto end; |
d02b48c6 RE |
119 | } |
120 | rem->neg=BN_is_zero(rem)?0:m->neg; | |
121 | dv->neg=m->neg^d->neg; | |
9b141126 UM |
122 | ret = 1; |
123 | end: | |
124 | BN_CTX_end(ctx); | |
125 | return(ret); | |
d02b48c6 RE |
126 | } |
127 | ||
128 | #else | |
129 | ||
cf1b7d96 RL |
130 | #if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) \ |
131 | && !defined(PEDANTIC) && !defined(BN_DIV3W) | |
4a6222d7 | 132 | # if defined(__GNUC__) && __GNUC__>=2 |
5dd955dc | 133 | # if defined(__i386) || defined (__i386__) |
4a6222d7 UM |
134 | /* |
135 | * There were two reasons for implementing this template: | |
136 | * - GNU C generates a call to a function (__udivdi3 to be exact) | |
137 | * in reply to ((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0 (I fail to | |
138 | * understand why...); | |
139 | * - divl doesn't only calculate quotient, but also leaves | |
140 | * remainder in %edx which we can definitely use here:-) | |
141 | * | |
142 | * <appro@fy.chalmers.se> | |
143 | */ | |
144 | # define bn_div_words(n0,n1,d0) \ | |
145 | ({ asm volatile ( \ | |
146 | "divl %4" \ | |
147 | : "=a"(q), "=d"(rem) \ | |
148 | : "a"(n1), "d"(n0), "g"(d0) \ | |
149 | : "cc"); \ | |
150 | q; \ | |
151 | }) | |
152 | # define REMAINDER_IS_ALREADY_CALCULATED | |
2f98abbc AP |
153 | # elif defined(__x86_64) && defined(SIXTY_FOUR_BIT_LONG) |
154 | /* | |
155 | * Same story here, but it's 128-bit by 64-bit division. Wow! | |
156 | * <appro@fy.chalmers.se> | |
157 | */ | |
158 | # define bn_div_words(n0,n1,d0) \ | |
159 | ({ asm volatile ( \ | |
160 | "divq %4" \ | |
161 | : "=a"(q), "=d"(rem) \ | |
162 | : "a"(n1), "d"(n0), "g"(d0) \ | |
163 | : "cc"); \ | |
164 | q; \ | |
165 | }) | |
166 | # define REMAINDER_IS_ALREADY_CALCULATED | |
4a6222d7 UM |
167 | # endif /* __<cpu> */ |
168 | # endif /* __GNUC__ */ | |
cf1b7d96 | 169 | #endif /* OPENSSL_NO_ASM */ |
4a6222d7 | 170 | |
78a0c1f1 | 171 | |
55525742 AP |
172 | /* BN_div[_no_branch] computes dv := num / divisor, rounding towards |
173 | * zero, and sets up rm such that dv*divisor + rm = num holds. | |
78a0c1f1 BM |
174 | * Thus: |
175 | * dv->neg == num->neg ^ divisor->neg (unless the result is zero) | |
176 | * rm->neg == num->neg (unless the remainder is zero) | |
177 | * If 'dv' or 'rm' is NULL, the respective value is not returned. | |
178 | */ | |
55525742 AP |
179 | static int BN_div_no_branch(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, |
180 | const BIGNUM *divisor, BN_CTX *ctx); | |
84c15db5 BL |
181 | int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor, |
182 | BN_CTX *ctx) | |
d02b48c6 | 183 | { |
4d6e1e4f BL |
184 | int norm_shift,i; |
185 | size_t loop; | |
d02b48c6 RE |
186 | BIGNUM *tmp,wnum,*snum,*sdiv,*res; |
187 | BN_ULONG *resp,*wnump; | |
188 | BN_ULONG d0,d1; | |
4d6e1e4f | 189 | size_t num_n,div_n; |
d02b48c6 | 190 | |
f8d6be3f BM |
191 | /* Invalid zero-padding would have particularly bad consequences |
192 | * in the case of 'num', so don't just rely on bn_check_top() for this one | |
193 | * (bn_check_top() works only for BN_DEBUG builds) */ | |
194 | if (num->top > 0 && num->d[num->top - 1] == 0) | |
195 | { | |
196 | BNerr(BN_F_BN_DIV,BN_R_NOT_INITIALIZED); | |
197 | return 0; | |
198 | } | |
199 | ||
200 | bn_check_top(num); | |
201 | ||
b002265e | 202 | if ((BN_get_flags(num, BN_FLG_CONSTTIME) != 0) || (BN_get_flags(divisor, BN_FLG_CONSTTIME) != 0)) |
bd31fb21 BM |
203 | { |
204 | return BN_div_no_branch(dv, rm, num, divisor, ctx); | |
205 | } | |
206 | ||
8215e7a9 NL |
207 | bn_check_top(dv); |
208 | bn_check_top(rm); | |
f8d6be3f | 209 | /* bn_check_top(num); */ /* 'num' has been checked already */ |
dfeab068 RE |
210 | bn_check_top(divisor); |
211 | ||
58964a49 | 212 | if (BN_is_zero(divisor)) |
d02b48c6 RE |
213 | { |
214 | BNerr(BN_F_BN_DIV,BN_R_DIV_BY_ZERO); | |
215 | return(0); | |
216 | } | |
217 | ||
218 | if (BN_ucmp(num,divisor) < 0) | |
219 | { | |
220 | if (rm != NULL) | |
221 | { if (BN_copy(rm,num) == NULL) return(0); } | |
222 | if (dv != NULL) BN_zero(dv); | |
223 | return(1); | |
224 | } | |
225 | ||
9b141126 UM |
226 | BN_CTX_start(ctx); |
227 | tmp=BN_CTX_get(ctx); | |
9b141126 UM |
228 | snum=BN_CTX_get(ctx); |
229 | sdiv=BN_CTX_get(ctx); | |
d02b48c6 | 230 | if (dv == NULL) |
9b141126 | 231 | res=BN_CTX_get(ctx); |
d02b48c6 | 232 | else res=dv; |
7f7b8d68 | 233 | if (sdiv == NULL || res == NULL) goto err; |
d02b48c6 RE |
234 | |
235 | /* First we normalise the numbers */ | |
236 | norm_shift=BN_BITS2-((BN_num_bits(divisor))%BN_BITS2); | |
9cdf87f1 | 237 | if (!(BN_lshift(sdiv,divisor,norm_shift))) goto err; |
d02b48c6 RE |
238 | sdiv->neg=0; |
239 | norm_shift+=BN_BITS2; | |
9cdf87f1 | 240 | if (!(BN_lshift(snum,num,norm_shift))) goto err; |
d02b48c6 RE |
241 | snum->neg=0; |
242 | div_n=sdiv->top; | |
243 | num_n=snum->top; | |
244 | loop=num_n-div_n; | |
d02b48c6 RE |
245 | /* Lets setup a 'window' into snum |
246 | * This is the part that corresponds to the current | |
247 | * 'area' being divided */ | |
9e989810 GT |
248 | wnum.neg = 0; |
249 | wnum.d = &(snum->d[loop]); | |
250 | wnum.top = div_n; | |
9e989810 GT |
251 | /* only needed when BN_ucmp messes up the values between top and max */ |
252 | wnum.dmax = snum->dmax - loop; /* so we don't step out of bounds */ | |
d02b48c6 RE |
253 | |
254 | /* Get the top 2 words of sdiv */ | |
5c0c2280 | 255 | /* div_n=sdiv->top; */ |
d02b48c6 RE |
256 | d0=sdiv->d[div_n-1]; |
257 | d1=(div_n == 1)?0:sdiv->d[div_n-2]; | |
258 | ||
259 | /* pointer to the 'top' of snum */ | |
260 | wnump= &(snum->d[num_n-1]); | |
261 | ||
262 | /* Setup to 'res' */ | |
263 | res->neg= (num->neg^divisor->neg); | |
58964a49 | 264 | if (!bn_wexpand(res,(loop+1))) goto err; |
dfeab068 | 265 | res->top=loop; |
d02b48c6 RE |
266 | resp= &(res->d[loop-1]); |
267 | ||
268 | /* space for temp */ | |
4d6e1e4f | 269 | if (!bn_wexpand(tmp, div_n+1)) goto err; |
d02b48c6 RE |
270 | |
271 | if (BN_ucmp(&wnum,sdiv) >= 0) | |
272 | { | |
9e989810 GT |
273 | /* If BN_DEBUG_RAND is defined BN_ucmp changes (via |
274 | * bn_pollute) the const bignum arguments => | |
275 | * clean the values between top and max again */ | |
276 | bn_clear_top2max(&wnum); | |
9e989810 | 277 | bn_sub_words(wnum.d, wnum.d, sdiv->d, div_n); |
d02b48c6 | 278 | *resp=1; |
d02b48c6 RE |
279 | } |
280 | else | |
281 | res->top--; | |
9e989810 GT |
282 | /* if res->top == 0 then clear the neg value otherwise decrease |
283 | * the resp pointer */ | |
80d89e6a BM |
284 | if (res->top == 0) |
285 | res->neg = 0; | |
bd31fb21 BM |
286 | else |
287 | resp--; | |
288 | ||
289 | for (i=0; i<loop-1; i++, wnump--, resp--) | |
290 | { | |
291 | BN_ULONG q,l0; | |
292 | /* the first part of the loop uses the top two words of | |
293 | * snum and sdiv to calculate a BN_ULONG q such that | |
294 | * | wnum - sdiv * q | < sdiv */ | |
295 | #if defined(BN_DIV3W) && !defined(OPENSSL_NO_ASM) | |
296 | BN_ULONG bn_div_3_words(BN_ULONG*,BN_ULONG,BN_ULONG); | |
297 | q=bn_div_3_words(wnump,d1,d0); | |
298 | #else | |
299 | BN_ULONG n0,n1,rem=0; | |
300 | ||
301 | n0=wnump[0]; | |
302 | n1=wnump[-1]; | |
303 | if (n0 == d0) | |
304 | q=BN_MASK2; | |
305 | else /* n0 < d0 */ | |
306 | { | |
307 | #ifdef BN_LLONG | |
308 | BN_ULLONG t2; | |
309 | ||
310 | #if defined(BN_LLONG) && defined(BN_DIV2W) && !defined(bn_div_words) | |
311 | q=(BN_ULONG)(((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0); | |
312 | #else | |
313 | q=bn_div_words(n0,n1,d0); | |
314 | #ifdef BN_DEBUG_LEVITTE | |
315 | fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\ | |
316 | X) -> 0x%08X\n", | |
317 | n0, n1, d0, q); | |
318 | #endif | |
319 | #endif | |
320 | ||
321 | #ifndef REMAINDER_IS_ALREADY_CALCULATED | |
322 | /* | |
323 | * rem doesn't have to be BN_ULLONG. The least we | |
324 | * know it's less that d0, isn't it? | |
325 | */ | |
326 | rem=(n1-q*d0)&BN_MASK2; | |
327 | #endif | |
328 | t2=(BN_ULLONG)d1*q; | |
329 | ||
330 | for (;;) | |
331 | { | |
332 | if (t2 <= ((((BN_ULLONG)rem)<<BN_BITS2)|wnump[-2])) | |
333 | break; | |
334 | q--; | |
335 | rem += d0; | |
336 | if (rem < d0) break; /* don't let rem overflow */ | |
337 | t2 -= d1; | |
338 | } | |
339 | #else /* !BN_LLONG */ | |
56c7754c | 340 | BN_ULONG t2l,t2h; |
bd31fb21 BM |
341 | |
342 | q=bn_div_words(n0,n1,d0); | |
343 | #ifdef BN_DEBUG_LEVITTE | |
344 | fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\ | |
345 | X) -> 0x%08X\n", | |
346 | n0, n1, d0, q); | |
347 | #endif | |
348 | #ifndef REMAINDER_IS_ALREADY_CALCULATED | |
349 | rem=(n1-q*d0)&BN_MASK2; | |
350 | #endif | |
351 | ||
352 | #if defined(BN_UMULT_LOHI) | |
353 | BN_UMULT_LOHI(t2l,t2h,d1,q); | |
354 | #elif defined(BN_UMULT_HIGH) | |
355 | t2l = d1 * q; | |
356 | t2h = BN_UMULT_HIGH(d1,q); | |
357 | #else | |
56c7754c DSH |
358 | { |
359 | BN_ULONG ql, qh; | |
bd31fb21 BM |
360 | t2l=LBITS(d1); t2h=HBITS(d1); |
361 | ql =LBITS(q); qh =HBITS(q); | |
362 | mul64(t2l,t2h,ql,qh); /* t2=(BN_ULLONG)d1*q; */ | |
56c7754c | 363 | } |
bd31fb21 BM |
364 | #endif |
365 | ||
366 | for (;;) | |
367 | { | |
368 | if ((t2h < rem) || | |
369 | ((t2h == rem) && (t2l <= wnump[-2]))) | |
370 | break; | |
371 | q--; | |
372 | rem += d0; | |
373 | if (rem < d0) break; /* don't let rem overflow */ | |
374 | if (t2l < d1) t2h--; t2l -= d1; | |
375 | } | |
376 | #endif /* !BN_LLONG */ | |
377 | } | |
378 | #endif /* !BN_DIV3W */ | |
379 | ||
380 | l0=bn_mul_words(tmp->d,sdiv->d,div_n,q); | |
381 | tmp->d[div_n]=l0; | |
382 | wnum.d--; | |
383 | /* ingore top values of the bignums just sub the two | |
384 | * BN_ULONG arrays with bn_sub_words */ | |
385 | if (bn_sub_words(wnum.d, wnum.d, tmp->d, div_n+1)) | |
386 | { | |
387 | /* Note: As we have considered only the leading | |
388 | * two BN_ULONGs in the calculation of q, sdiv * q | |
389 | * might be greater than wnum (but then (q-1) * sdiv | |
390 | * is less or equal than wnum) | |
391 | */ | |
392 | q--; | |
393 | if (bn_add_words(wnum.d, wnum.d, sdiv->d, div_n)) | |
394 | /* we can't have an overflow here (assuming | |
395 | * that q != 0, but if q == 0 then tmp is | |
396 | * zero anyway) */ | |
397 | (*wnump)++; | |
398 | } | |
399 | /* store part of the result */ | |
400 | *resp = q; | |
401 | } | |
402 | bn_correct_top(snum); | |
403 | if (rm != NULL) | |
404 | { | |
405 | /* Keep a copy of the neg flag in num because if rm==num | |
406 | * BN_rshift() will overwrite it. | |
407 | */ | |
408 | int neg = num->neg; | |
409 | BN_rshift(rm,snum,norm_shift); | |
410 | if (!BN_is_zero(rm)) | |
411 | rm->neg = neg; | |
412 | bn_check_top(rm); | |
413 | } | |
414 | BN_CTX_end(ctx); | |
415 | return(1); | |
416 | err: | |
417 | bn_check_top(rm); | |
418 | BN_CTX_end(ctx); | |
419 | return(0); | |
420 | } | |
421 | ||
422 | ||
423 | /* BN_div_no_branch is a special version of BN_div. It does not contain | |
424 | * branches that may leak sensitive information. | |
425 | */ | |
55525742 | 426 | static int BN_div_no_branch(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, |
bd31fb21 BM |
427 | const BIGNUM *divisor, BN_CTX *ctx) |
428 | { | |
429 | int norm_shift,i,loop; | |
430 | BIGNUM *tmp,wnum,*snum,*sdiv,*res; | |
431 | BN_ULONG *resp,*wnump; | |
432 | BN_ULONG d0,d1; | |
4d6e1e4f | 433 | size_t num_n,div_n; |
bd31fb21 BM |
434 | |
435 | bn_check_top(dv); | |
436 | bn_check_top(rm); | |
f8d6be3f | 437 | /* bn_check_top(num); */ /* 'num' has been checked in BN_div() */ |
bd31fb21 BM |
438 | bn_check_top(divisor); |
439 | ||
440 | if (BN_is_zero(divisor)) | |
441 | { | |
24a8c25a | 442 | BNerr(BN_F_BN_DIV_NO_BRANCH,BN_R_DIV_BY_ZERO); |
bd31fb21 BM |
443 | return(0); |
444 | } | |
445 | ||
446 | BN_CTX_start(ctx); | |
447 | tmp=BN_CTX_get(ctx); | |
448 | snum=BN_CTX_get(ctx); | |
449 | sdiv=BN_CTX_get(ctx); | |
450 | if (dv == NULL) | |
451 | res=BN_CTX_get(ctx); | |
452 | else res=dv; | |
453 | if (sdiv == NULL || res == NULL) goto err; | |
454 | ||
455 | /* First we normalise the numbers */ | |
456 | norm_shift=BN_BITS2-((BN_num_bits(divisor))%BN_BITS2); | |
457 | if (!(BN_lshift(sdiv,divisor,norm_shift))) goto err; | |
458 | sdiv->neg=0; | |
459 | norm_shift+=BN_BITS2; | |
460 | if (!(BN_lshift(snum,num,norm_shift))) goto err; | |
461 | snum->neg=0; | |
462 | ||
463 | /* Since we don't know whether snum is larger than sdiv, | |
464 | * we pad snum with enough zeroes without changing its | |
465 | * value. | |
466 | */ | |
467 | if (snum->top <= sdiv->top+1) | |
468 | { | |
469 | if (bn_wexpand(snum, sdiv->top + 2) == NULL) goto err; | |
470 | for (i = snum->top; i < sdiv->top + 2; i++) snum->d[i] = 0; | |
471 | snum->top = sdiv->top + 2; | |
472 | } | |
473 | else | |
474 | { | |
475 | if (bn_wexpand(snum, snum->top + 1) == NULL) goto err; | |
476 | snum->d[snum->top] = 0; | |
477 | snum->top ++; | |
478 | } | |
479 | ||
480 | div_n=sdiv->top; | |
481 | num_n=snum->top; | |
482 | loop=num_n-div_n; | |
483 | /* Lets setup a 'window' into snum | |
484 | * This is the part that corresponds to the current | |
485 | * 'area' being divided */ | |
486 | wnum.neg = 0; | |
487 | wnum.d = &(snum->d[loop]); | |
488 | wnum.top = div_n; | |
489 | /* only needed when BN_ucmp messes up the values between top and max */ | |
490 | wnum.dmax = snum->dmax - loop; /* so we don't step out of bounds */ | |
491 | ||
492 | /* Get the top 2 words of sdiv */ | |
493 | /* div_n=sdiv->top; */ | |
494 | d0=sdiv->d[div_n-1]; | |
495 | d1=(div_n == 1)?0:sdiv->d[div_n-2]; | |
496 | ||
497 | /* pointer to the 'top' of snum */ | |
498 | wnump= &(snum->d[num_n-1]); | |
499 | ||
500 | /* Setup to 'res' */ | |
501 | res->neg= (num->neg^divisor->neg); | |
4d6e1e4f | 502 | if (!bn_wexpand(res,loop+1U)) goto err; |
bd31fb21 BM |
503 | res->top=loop-1; |
504 | resp= &(res->d[loop-1]); | |
505 | ||
506 | /* space for temp */ | |
4d6e1e4f | 507 | if (!bn_wexpand(tmp,div_n+1U)) goto err; |
bd31fb21 BM |
508 | |
509 | /* if res->top == 0 then clear the neg value otherwise decrease | |
510 | * the resp pointer */ | |
511 | if (res->top == 0) | |
512 | res->neg = 0; | |
9e989810 GT |
513 | else |
514 | resp--; | |
d02b48c6 | 515 | |
9e989810 | 516 | for (i=0; i<loop-1; i++, wnump--, resp--) |
d02b48c6 | 517 | { |
0dd25e36 | 518 | BN_ULONG q,l0; |
9e989810 GT |
519 | /* the first part of the loop uses the top two words of |
520 | * snum and sdiv to calculate a BN_ULONG q such that | |
521 | * | wnum - sdiv * q | < sdiv */ | |
cf1b7d96 | 522 | #if defined(BN_DIV3W) && !defined(OPENSSL_NO_ASM) |
500230ee | 523 | BN_ULONG bn_div_3_words(BN_ULONG*,BN_ULONG,BN_ULONG); |
0bbd0352 | 524 | q=bn_div_3_words(wnump,d1,d0); |
0dd25e36 | 525 | #else |
4c22909e | 526 | BN_ULONG n0,n1,rem=0; |
d02b48c6 | 527 | |
d02b48c6 RE |
528 | n0=wnump[0]; |
529 | n1=wnump[-1]; | |
530 | if (n0 == d0) | |
531 | q=BN_MASK2; | |
4a6222d7 UM |
532 | else /* n0 < d0 */ |
533 | { | |
534 | #ifdef BN_LLONG | |
535 | BN_ULLONG t2; | |
536 | ||
4c22909e | 537 | #if defined(BN_LLONG) && defined(BN_DIV2W) && !defined(bn_div_words) |
8e1589ec | 538 | q=(BN_ULONG)(((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0); |
0dd25e36 | 539 | #else |
dfeab068 | 540 | q=bn_div_words(n0,n1,d0); |
3c801fa4 RL |
541 | #ifdef BN_DEBUG_LEVITTE |
542 | fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\ | |
543 | X) -> 0x%08X\n", | |
544 | n0, n1, d0, q); | |
545 | #endif | |
0dd25e36 | 546 | #endif |
0dd25e36 | 547 | |
fe7cd164 | 548 | #ifndef REMAINDER_IS_ALREADY_CALCULATED |
4a6222d7 UM |
549 | /* |
550 | * rem doesn't have to be BN_ULLONG. The least we | |
551 | * know it's less that d0, isn't it? | |
552 | */ | |
553 | rem=(n1-q*d0)&BN_MASK2; | |
4c22909e | 554 | #endif |
4a6222d7 UM |
555 | t2=(BN_ULLONG)d1*q; |
556 | ||
557 | for (;;) | |
558 | { | |
559 | if (t2 <= ((((BN_ULLONG)rem)<<BN_BITS2)|wnump[-2])) | |
560 | break; | |
561 | q--; | |
562 | rem += d0; | |
563 | if (rem < d0) break; /* don't let rem overflow */ | |
564 | t2 -= d1; | |
565 | } | |
566 | #else /* !BN_LLONG */ | |
56c7754c | 567 | BN_ULONG t2l,t2h; |
0dd25e36 | 568 | |
4a6222d7 | 569 | q=bn_div_words(n0,n1,d0); |
3c801fa4 RL |
570 | #ifdef BN_DEBUG_LEVITTE |
571 | fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\ | |
572 | X) -> 0x%08X\n", | |
573 | n0, n1, d0, q); | |
574 | #endif | |
fe7cd164 | 575 | #ifndef REMAINDER_IS_ALREADY_CALCULATED |
4a6222d7 | 576 | rem=(n1-q*d0)&BN_MASK2; |
4c22909e | 577 | #endif |
4a6222d7 | 578 | |
2f98abbc AP |
579 | #if defined(BN_UMULT_LOHI) |
580 | BN_UMULT_LOHI(t2l,t2h,d1,q); | |
581 | #elif defined(BN_UMULT_HIGH) | |
4a6222d7 UM |
582 | t2l = d1 * q; |
583 | t2h = BN_UMULT_HIGH(d1,q); | |
fb81ac5e | 584 | #else |
56c7754c DSH |
585 | { |
586 | BN_ULONG ql, qh; | |
4a6222d7 UM |
587 | t2l=LBITS(d1); t2h=HBITS(d1); |
588 | ql =LBITS(q); qh =HBITS(q); | |
589 | mul64(t2l,t2h,ql,qh); /* t2=(BN_ULLONG)d1*q; */ | |
56c7754c | 590 | } |
fb81ac5e | 591 | #endif |
0dd25e36 | 592 | |
4a6222d7 UM |
593 | for (;;) |
594 | { | |
595 | if ((t2h < rem) || | |
596 | ((t2h == rem) && (t2l <= wnump[-2]))) | |
597 | break; | |
598 | q--; | |
599 | rem += d0; | |
600 | if (rem < d0) break; /* don't let rem overflow */ | |
601 | if (t2l < d1) t2h--; t2l -= d1; | |
602 | } | |
603 | #endif /* !BN_LLONG */ | |
d02b48c6 | 604 | } |
4c22909e | 605 | #endif /* !BN_DIV3W */ |
4a6222d7 | 606 | |
58964a49 | 607 | l0=bn_mul_words(tmp->d,sdiv->d,div_n,q); |
d02b48c6 | 608 | tmp->d[div_n]=l0; |
9e989810 GT |
609 | wnum.d--; |
610 | /* ingore top values of the bignums just sub the two | |
611 | * BN_ULONG arrays with bn_sub_words */ | |
612 | if (bn_sub_words(wnum.d, wnum.d, tmp->d, div_n+1)) | |
d02b48c6 | 613 | { |
9e989810 GT |
614 | /* Note: As we have considered only the leading |
615 | * two BN_ULONGs in the calculation of q, sdiv * q | |
616 | * might be greater than wnum (but then (q-1) * sdiv | |
617 | * is less or equal than wnum) | |
618 | */ | |
d02b48c6 | 619 | q--; |
9e989810 GT |
620 | if (bn_add_words(wnum.d, wnum.d, sdiv->d, div_n)) |
621 | /* we can't have an overflow here (assuming | |
622 | * that q != 0, but if q == 0 then tmp is | |
623 | * zero anyway) */ | |
624 | (*wnump)++; | |
d02b48c6 | 625 | } |
9e989810 GT |
626 | /* store part of the result */ |
627 | *resp = q; | |
d02b48c6 | 628 | } |
a8aa764d | 629 | bn_correct_top(snum); |
d02b48c6 RE |
630 | if (rm != NULL) |
631 | { | |
3d2e469c DSH |
632 | /* Keep a copy of the neg flag in num because if rm==num |
633 | * BN_rshift() will overwrite it. | |
634 | */ | |
635 | int neg = num->neg; | |
d02b48c6 | 636 | BN_rshift(rm,snum,norm_shift); |
78a0c1f1 | 637 | if (!BN_is_zero(rm)) |
3d2e469c | 638 | rm->neg = neg; |
d870740c | 639 | bn_check_top(rm); |
d02b48c6 | 640 | } |
d1e7d1d9 | 641 | bn_correct_top(res); |
9b141126 | 642 | BN_CTX_end(ctx); |
d02b48c6 RE |
643 | return(1); |
644 | err: | |
8215e7a9 | 645 | bn_check_top(rm); |
4a6222d7 | 646 | BN_CTX_end(ctx); |
d02b48c6 RE |
647 | return(0); |
648 | } | |
649 | ||
650 | #endif |