2 * Bignum routines for RSA and DH and stuff.
14 * * Do not call the DIVMOD_WORD macro with expressions such as array
15 * subscripts, as some implementations object to this (see below).
16 * * Note that none of the division methods below will cope if the
17 * quotient won't fit into BIGNUM_INT_BITS. Callers should be careful
19 * If this condition occurs, in the case of the x86 DIV instruction,
20 * an overflow exception will occur, which (according to a correspondent)
21 * will manifest on Windows as something like
22 * 0xC0000095: Integer overflow
23 * The C variant won't give the right answer, either.
26 #if defined __GNUC__ && defined __i386__
27 typedef unsigned long BignumInt;
28 typedef unsigned long long BignumDblInt;
29 #define BIGNUM_INT_MASK 0xFFFFFFFFUL
30 #define BIGNUM_TOP_BIT 0x80000000UL
31 #define BIGNUM_INT_BITS 32
32 #define MUL_WORD(w1, w2) ((BignumDblInt)w1 * w2)
33 #define DIVMOD_WORD(q, r, hi, lo, w) \
35 "=d" (r), "=a" (q) : \
36 "r" (w), "d" (hi), "a" (lo))
37 #elif defined _MSC_VER && defined _M_IX86
38 typedef unsigned __int32 BignumInt;
39 typedef unsigned __int64 BignumDblInt;
40 #define BIGNUM_INT_MASK 0xFFFFFFFFUL
41 #define BIGNUM_TOP_BIT 0x80000000UL
42 #define BIGNUM_INT_BITS 32
43 #define MUL_WORD(w1, w2) ((BignumDblInt)w1 * w2)
44 /* Note: MASM interprets array subscripts in the macro arguments as
45 * assembler syntax, which gives the wrong answer. Don't supply them.
46 * <http://msdn2.microsoft.com/en-us/library/bf1dw62z.aspx> */
47 #define DIVMOD_WORD(q, r, hi, lo, w) do { \
55 /* 64-bit architectures can do 32x32->64 chunks at a time */
56 typedef unsigned int BignumInt;
57 typedef unsigned long BignumDblInt;
58 #define BIGNUM_INT_MASK 0xFFFFFFFFU
59 #define BIGNUM_TOP_BIT 0x80000000U
60 #define BIGNUM_INT_BITS 32
61 #define MUL_WORD(w1, w2) ((BignumDblInt)w1 * w2)
62 #define DIVMOD_WORD(q, r, hi, lo, w) do { \
63 BignumDblInt n = (((BignumDblInt)hi) << BIGNUM_INT_BITS) | lo; \
68 /* 64-bit architectures in which unsigned long is 32 bits, not 64 */
69 typedef unsigned long BignumInt;
70 typedef unsigned long long BignumDblInt;
71 #define BIGNUM_INT_MASK 0xFFFFFFFFUL
72 #define BIGNUM_TOP_BIT 0x80000000UL
73 #define BIGNUM_INT_BITS 32
74 #define MUL_WORD(w1, w2) ((BignumDblInt)w1 * w2)
75 #define DIVMOD_WORD(q, r, hi, lo, w) do { \
76 BignumDblInt n = (((BignumDblInt)hi) << BIGNUM_INT_BITS) | lo; \
81 /* Fallback for all other cases */
82 typedef unsigned short BignumInt;
83 typedef unsigned long BignumDblInt;
84 #define BIGNUM_INT_MASK 0xFFFFU
85 #define BIGNUM_TOP_BIT 0x8000U
86 #define BIGNUM_INT_BITS 16
87 #define MUL_WORD(w1, w2) ((BignumDblInt)w1 * w2)
88 #define DIVMOD_WORD(q, r, hi, lo, w) do { \
89 BignumDblInt n = (((BignumDblInt)hi) << BIGNUM_INT_BITS) | lo; \
95 #define BIGNUM_INT_BYTES (BIGNUM_INT_BITS / 8)
97 #define BIGNUM_INTERNAL
98 typedef BignumInt *Bignum;
102 BignumInt bnZero[1] = { 0 };
103 BignumInt bnOne[2] = { 1, 1 };
106 * The Bignum format is an array of `BignumInt'. The first
107 * element of the array counts the remaining elements. The
108 * remaining elements express the actual number, base 2^BIGNUM_INT_BITS, _least_
109 * significant digit first. (So it's trivial to extract the bit
110 * with value 2^n for any n.)
112 * All Bignums in this module are positive. Negative numbers must
113 * be dealt with outside it.
115 * INVARIANT: the most significant word of any Bignum must be
119 Bignum Zero = bnZero, One = bnOne;
121 static Bignum newbn(int length)
123 Bignum b = snewn(length + 1, BignumInt);
126 memset(b, 0, (length + 1) * sizeof(*b));
131 void bn_restore_invariant(Bignum b)
133 while (b[0] > 1 && b[b[0]] == 0)
137 Bignum copybn(Bignum orig)
139 Bignum b = snewn(orig[0] + 1, BignumInt);
142 memcpy(b, orig, (orig[0] + 1) * sizeof(*b));
146 void freebn(Bignum b)
149 * Burn the evidence, just in case.
151 memset(b, 0, sizeof(b[0]) * (b[0] + 1));
155 Bignum bn_power_2(int n)
157 Bignum ret = newbn(n / BIGNUM_INT_BITS + 1);
158 bignum_set_bit(ret, n, 1);
163 * Internal addition. Sets c = a - b, where 'a', 'b' and 'c' are all
164 * big-endian arrays of 'len' BignumInts. Returns a BignumInt carried
167 static BignumInt internal_add(const BignumInt *a, const BignumInt *b,
168 BignumInt *c, int len)
171 BignumDblInt carry = 0;
173 for (i = len-1; i >= 0; i--) {
174 carry += (BignumDblInt)a[i] + b[i];
175 c[i] = (BignumInt)carry;
176 carry >>= BIGNUM_INT_BITS;
179 return (BignumInt)carry;
183 * Internal subtraction. Sets c = a - b, where 'a', 'b' and 'c' are
184 * all big-endian arrays of 'len' BignumInts. Any borrow from the top
187 static void internal_sub(const BignumInt *a, const BignumInt *b,
188 BignumInt *c, int len)
191 BignumDblInt carry = 1;
193 for (i = len-1; i >= 0; i--) {
194 carry += (BignumDblInt)a[i] + (b[i] ^ BIGNUM_INT_MASK);
195 c[i] = (BignumInt)carry;
196 carry >>= BIGNUM_INT_BITS;
202 * Input is in the first len words of a and b.
203 * Result is returned in the first 2*len words of c.
205 #define KARATSUBA_THRESHOLD 50
206 static void internal_mul(BignumInt *a, BignumInt *b,
207 BignumInt *c, int len)
212 if (len > KARATSUBA_THRESHOLD) {
215 * Karatsuba divide-and-conquer algorithm. Cut each input in
216 * half, so that it's expressed as two big 'digits' in a giant
222 * Then the product is of course
224 * ab = a_1 b_1 D^2 + (a_1 b_0 + a_0 b_1) D + a_0 b_0
226 * and we compute the three coefficients by recursively
227 * calling ourself to do half-length multiplications.
229 * The clever bit that makes this worth doing is that we only
230 * need _one_ half-length multiplication for the central
231 * coefficient rather than the two that it obviouly looks
232 * like, because we can use a single multiplication to compute
234 * (a_1 + a_0) (b_1 + b_0) = a_1 b_1 + a_1 b_0 + a_0 b_1 + a_0 b_0
236 * and then we subtract the other two coefficients (a_1 b_1
237 * and a_0 b_0) which we were computing anyway.
239 * Hence we get to multiply two numbers of length N in about
240 * three times as much work as it takes to multiply numbers of
241 * length N/2, which is obviously better than the four times
242 * as much work it would take if we just did a long
243 * conventional multiply.
246 int toplen = len/2, botlen = len - toplen; /* botlen is the bigger */
247 int midlen = botlen + 1;
252 * The coefficients a_1 b_1 and a_0 b_0 just avoid overlapping
253 * in the output array, so we can compute them immediately in
258 internal_mul(a, b, c, toplen);
261 internal_mul(a + toplen, b + toplen, c + 2*toplen, botlen);
264 * We must allocate scratch space for the central coefficient,
265 * and also for the two input values that we multiply when
266 * computing it. Since either or both may carry into the
267 * (botlen+1)th word, we must use a slightly longer length
270 scratch = snewn(4 * midlen, BignumInt);
272 /* Zero padding. midlen exceeds toplen by at most 2, so just
273 * zero the first two words of each input and the rest will be
275 scratch[0] = scratch[1] = scratch[midlen] = scratch[midlen+1] = 0;
277 for (j = 0; j < toplen; j++) {
278 scratch[midlen - toplen + j] = a[j]; /* a_1 */
279 scratch[2*midlen - toplen + j] = b[j]; /* b_1 */
282 /* compute a_1 + a_0 */
283 scratch[0] = internal_add(scratch+1, a+toplen, scratch+1, botlen);
284 /* compute b_1 + b_0 */
285 scratch[midlen] = internal_add(scratch+midlen+1, b+toplen,
286 scratch+midlen+1, botlen);
289 * Now we can do the third multiplication.
291 internal_mul(scratch, scratch + midlen, scratch + 2*midlen, midlen);
294 * Now we can reuse the first half of 'scratch' to compute the
295 * sum of the outer two coefficients, to subtract from that
296 * product to obtain the middle one.
298 scratch[0] = scratch[1] = scratch[2] = scratch[3] = 0;
299 for (j = 0; j < 2*toplen; j++)
300 scratch[2*midlen - 2*toplen + j] = c[j];
301 scratch[1] = internal_add(scratch+2, c + 2*toplen,
302 scratch+2, 2*botlen);
304 internal_sub(scratch + 2*midlen, scratch,
305 scratch + 2*midlen, 2*midlen);
308 * And now all we need to do is to add that middle coefficient
309 * back into the output. We may have to propagate a carry
310 * further up the output, but we can be sure it won't
311 * propagate right the way off the top.
313 carry = internal_add(c + 2*len - botlen - 2*midlen,
315 c + 2*len - botlen - 2*midlen, 2*midlen);
316 j = 2*len - botlen - 2*midlen - 1;
320 c[j] = (BignumInt)carry;
321 carry >>= BIGNUM_INT_BITS;
325 for (j = 0; j < 4 * midlen; j++)
332 * Multiply in the ordinary O(N^2) way.
335 for (j = 0; j < 2 * len; j++)
338 for (i = len - 1; i >= 0; i--) {
340 for (j = len - 1; j >= 0; j--) {
341 t += MUL_WORD(a[i], (BignumDblInt) b[j]);
342 t += (BignumDblInt) c[i + j + 1];
343 c[i + j + 1] = (BignumInt) t;
344 t = t >> BIGNUM_INT_BITS;
346 c[i] = (BignumInt) t;
351 static void internal_add_shifted(BignumInt *number,
352 unsigned n, int shift)
354 int word = 1 + (shift / BIGNUM_INT_BITS);
355 int bshift = shift % BIGNUM_INT_BITS;
358 addend = (BignumDblInt)n << bshift;
361 addend += number[word];
362 number[word] = (BignumInt) addend & BIGNUM_INT_MASK;
363 addend >>= BIGNUM_INT_BITS;
370 * Input in first alen words of a and first mlen words of m.
371 * Output in first alen words of a
372 * (of which first alen-mlen words will be zero).
373 * The MSW of m MUST have its high bit set.
374 * Quotient is accumulated in the `quotient' array, which is a Bignum
375 * rather than the internal bigendian format. Quotient parts are shifted
376 * left by `qshift' before adding into quot.
378 static void internal_mod(BignumInt *a, int alen,
379 BignumInt *m, int mlen,
380 BignumInt *quot, int qshift)
392 for (i = 0; i <= alen - mlen; i++) {
394 unsigned int q, r, c, ai1;
408 /* Find q = h:a[i] / m0 */
413 * To illustrate it, suppose a BignumInt is 8 bits, and
414 * we are dividing (say) A1:23:45:67 by A1:B2:C3. Then
415 * our initial division will be 0xA123 / 0xA1, which
416 * will give a quotient of 0x100 and a divide overflow.
417 * However, the invariants in this division algorithm
418 * are not violated, since the full number A1:23:... is
419 * _less_ than the quotient prefix A1:B2:... and so the
420 * following correction loop would have sorted it out.
422 * In this situation we set q to be the largest
423 * quotient we _can_ stomach (0xFF, of course).
427 /* Macro doesn't want an array subscript expression passed
428 * into it (see definition), so use a temporary. */
429 BignumInt tmplo = a[i];
430 DIVMOD_WORD(q, r, h, tmplo, m0);
432 /* Refine our estimate of q by looking at
433 h:a[i]:a[i+1] / m0:m1 */
435 if (t > ((BignumDblInt) r << BIGNUM_INT_BITS) + ai1) {
438 r = (r + m0) & BIGNUM_INT_MASK; /* overflow? */
439 if (r >= (BignumDblInt) m0 &&
440 t > ((BignumDblInt) r << BIGNUM_INT_BITS) + ai1) q--;
444 /* Subtract q * m from a[i...] */
446 for (k = mlen - 1; k >= 0; k--) {
447 t = MUL_WORD(q, m[k]);
449 c = (unsigned)(t >> BIGNUM_INT_BITS);
450 if ((BignumInt) t > a[i + k])
452 a[i + k] -= (BignumInt) t;
455 /* Add back m in case of borrow */
458 for (k = mlen - 1; k >= 0; k--) {
461 a[i + k] = (BignumInt) t;
462 t = t >> BIGNUM_INT_BITS;
467 internal_add_shifted(quot, q, qshift + BIGNUM_INT_BITS * (alen - mlen - i));
472 * Compute (base ^ exp) % mod.
474 Bignum modpow(Bignum base_in, Bignum exp, Bignum mod)
476 BignumInt *a, *b, *n, *m;
482 * The most significant word of mod needs to be non-zero. It
483 * should already be, but let's make sure.
485 assert(mod[mod[0]] != 0);
488 * Make sure the base is smaller than the modulus, by reducing
489 * it modulo the modulus if not.
491 base = bigmod(base_in, mod);
493 /* Allocate m of size mlen, copy mod to m */
494 /* We use big endian internally */
496 m = snewn(mlen, BignumInt);
497 for (j = 0; j < mlen; j++)
498 m[j] = mod[mod[0] - j];
500 /* Shift m left to make msb bit set */
501 for (mshift = 0; mshift < BIGNUM_INT_BITS-1; mshift++)
502 if ((m[0] << mshift) & BIGNUM_TOP_BIT)
505 for (i = 0; i < mlen - 1; i++)
506 m[i] = (m[i] << mshift) | (m[i + 1] >> (BIGNUM_INT_BITS - mshift));
507 m[mlen - 1] = m[mlen - 1] << mshift;
510 /* Allocate n of size mlen, copy base to n */
511 n = snewn(mlen, BignumInt);
513 for (j = 0; j < i; j++)
515 for (j = 0; j < (int)base[0]; j++)
516 n[i + j] = base[base[0] - j];
518 /* Allocate a and b of size 2*mlen. Set a = 1 */
519 a = snewn(2 * mlen, BignumInt);
520 b = snewn(2 * mlen, BignumInt);
521 for (i = 0; i < 2 * mlen; i++)
525 /* Skip leading zero bits of exp. */
527 j = BIGNUM_INT_BITS-1;
528 while (i < (int)exp[0] && (exp[exp[0] - i] & (1 << j)) == 0) {
532 j = BIGNUM_INT_BITS-1;
536 /* Main computation */
537 while (i < (int)exp[0]) {
539 internal_mul(a + mlen, a + mlen, b, mlen);
540 internal_mod(b, mlen * 2, m, mlen, NULL, 0);
541 if ((exp[exp[0] - i] & (1 << j)) != 0) {
542 internal_mul(b + mlen, n, a, mlen);
543 internal_mod(a, mlen * 2, m, mlen, NULL, 0);
553 j = BIGNUM_INT_BITS-1;
556 /* Fixup result in case the modulus was shifted */
558 for (i = mlen - 1; i < 2 * mlen - 1; i++)
559 a[i] = (a[i] << mshift) | (a[i + 1] >> (BIGNUM_INT_BITS - mshift));
560 a[2 * mlen - 1] = a[2 * mlen - 1] << mshift;
561 internal_mod(a, mlen * 2, m, mlen, NULL, 0);
562 for (i = 2 * mlen - 1; i >= mlen; i--)
563 a[i] = (a[i] >> mshift) | (a[i - 1] << (BIGNUM_INT_BITS - mshift));
566 /* Copy result to buffer */
567 result = newbn(mod[0]);
568 for (i = 0; i < mlen; i++)
569 result[result[0] - i] = a[i + mlen];
570 while (result[0] > 1 && result[result[0]] == 0)
573 /* Free temporary arrays */
574 for (i = 0; i < 2 * mlen; i++)
577 for (i = 0; i < 2 * mlen; i++)
580 for (i = 0; i < mlen; i++)
583 for (i = 0; i < mlen; i++)
593 * Compute (p * q) % mod.
594 * The most significant word of mod MUST be non-zero.
595 * We assume that the result array is the same size as the mod array.
597 Bignum modmul(Bignum p, Bignum q, Bignum mod)
599 BignumInt *a, *n, *m, *o;
601 int pqlen, mlen, rlen, i, j;
604 /* Allocate m of size mlen, copy mod to m */
605 /* We use big endian internally */
607 m = snewn(mlen, BignumInt);
608 for (j = 0; j < mlen; j++)
609 m[j] = mod[mod[0] - j];
611 /* Shift m left to make msb bit set */
612 for (mshift = 0; mshift < BIGNUM_INT_BITS-1; mshift++)
613 if ((m[0] << mshift) & BIGNUM_TOP_BIT)
616 for (i = 0; i < mlen - 1; i++)
617 m[i] = (m[i] << mshift) | (m[i + 1] >> (BIGNUM_INT_BITS - mshift));
618 m[mlen - 1] = m[mlen - 1] << mshift;
621 pqlen = (p[0] > q[0] ? p[0] : q[0]);
623 /* Allocate n of size pqlen, copy p to n */
624 n = snewn(pqlen, BignumInt);
626 for (j = 0; j < i; j++)
628 for (j = 0; j < (int)p[0]; j++)
629 n[i + j] = p[p[0] - j];
631 /* Allocate o of size pqlen, copy q to o */
632 o = snewn(pqlen, BignumInt);
634 for (j = 0; j < i; j++)
636 for (j = 0; j < (int)q[0]; j++)
637 o[i + j] = q[q[0] - j];
639 /* Allocate a of size 2*pqlen for result */
640 a = snewn(2 * pqlen, BignumInt);
642 /* Main computation */
643 internal_mul(n, o, a, pqlen);
644 internal_mod(a, pqlen * 2, m, mlen, NULL, 0);
646 /* Fixup result in case the modulus was shifted */
648 for (i = 2 * pqlen - mlen - 1; i < 2 * pqlen - 1; i++)
649 a[i] = (a[i] << mshift) | (a[i + 1] >> (BIGNUM_INT_BITS - mshift));
650 a[2 * pqlen - 1] = a[2 * pqlen - 1] << mshift;
651 internal_mod(a, pqlen * 2, m, mlen, NULL, 0);
652 for (i = 2 * pqlen - 1; i >= 2 * pqlen - mlen; i--)
653 a[i] = (a[i] >> mshift) | (a[i - 1] << (BIGNUM_INT_BITS - mshift));
656 /* Copy result to buffer */
657 rlen = (mlen < pqlen * 2 ? mlen : pqlen * 2);
658 result = newbn(rlen);
659 for (i = 0; i < rlen; i++)
660 result[result[0] - i] = a[i + 2 * pqlen - rlen];
661 while (result[0] > 1 && result[result[0]] == 0)
664 /* Free temporary arrays */
665 for (i = 0; i < 2 * pqlen; i++)
668 for (i = 0; i < mlen; i++)
671 for (i = 0; i < pqlen; i++)
674 for (i = 0; i < pqlen; i++)
683 * The most significant word of mod MUST be non-zero.
684 * We assume that the result array is the same size as the mod array.
685 * We optionally write out a quotient if `quotient' is non-NULL.
686 * We can avoid writing out the result if `result' is NULL.
688 static void bigdivmod(Bignum p, Bignum mod, Bignum result, Bignum quotient)
692 int plen, mlen, i, j;
694 /* Allocate m of size mlen, copy mod to m */
695 /* We use big endian internally */
697 m = snewn(mlen, BignumInt);
698 for (j = 0; j < mlen; j++)
699 m[j] = mod[mod[0] - j];
701 /* Shift m left to make msb bit set */
702 for (mshift = 0; mshift < BIGNUM_INT_BITS-1; mshift++)
703 if ((m[0] << mshift) & BIGNUM_TOP_BIT)
706 for (i = 0; i < mlen - 1; i++)
707 m[i] = (m[i] << mshift) | (m[i + 1] >> (BIGNUM_INT_BITS - mshift));
708 m[mlen - 1] = m[mlen - 1] << mshift;
712 /* Ensure plen > mlen */
716 /* Allocate n of size plen, copy p to n */
717 n = snewn(plen, BignumInt);
718 for (j = 0; j < plen; j++)
720 for (j = 1; j <= (int)p[0]; j++)
723 /* Main computation */
724 internal_mod(n, plen, m, mlen, quotient, mshift);
726 /* Fixup result in case the modulus was shifted */
728 for (i = plen - mlen - 1; i < plen - 1; i++)
729 n[i] = (n[i] << mshift) | (n[i + 1] >> (BIGNUM_INT_BITS - mshift));
730 n[plen - 1] = n[plen - 1] << mshift;
731 internal_mod(n, plen, m, mlen, quotient, 0);
732 for (i = plen - 1; i >= plen - mlen; i--)
733 n[i] = (n[i] >> mshift) | (n[i - 1] << (BIGNUM_INT_BITS - mshift));
736 /* Copy result to buffer */
738 for (i = 1; i <= (int)result[0]; i++) {
740 result[i] = j >= 0 ? n[j] : 0;
744 /* Free temporary arrays */
745 for (i = 0; i < mlen; i++)
748 for (i = 0; i < plen; i++)
754 * Decrement a number.
756 void decbn(Bignum bn)
759 while (i < (int)bn[0] && bn[i] == 0)
760 bn[i++] = BIGNUM_INT_MASK;
764 Bignum bignum_from_bytes(const unsigned char *data, int nbytes)
769 w = (nbytes + BIGNUM_INT_BYTES - 1) / BIGNUM_INT_BYTES; /* bytes->words */
772 for (i = 1; i <= w; i++)
774 for (i = nbytes; i--;) {
775 unsigned char byte = *data++;
776 result[1 + i / BIGNUM_INT_BYTES] |= byte << (8*i % BIGNUM_INT_BITS);
779 while (result[0] > 1 && result[result[0]] == 0)
785 * Read an SSH-1-format bignum from a data buffer. Return the number
786 * of bytes consumed, or -1 if there wasn't enough data.
788 int ssh1_read_bignum(const unsigned char *data, int len, Bignum * result)
790 const unsigned char *p = data;
798 for (i = 0; i < 2; i++)
800 b = (w + 7) / 8; /* bits -> bytes */
805 if (!result) /* just return length */
808 *result = bignum_from_bytes(p, b);
814 * Return the bit count of a bignum, for SSH-1 encoding.
816 int bignum_bitcount(Bignum bn)
818 int bitcount = bn[0] * BIGNUM_INT_BITS - 1;
820 && (bn[bitcount / BIGNUM_INT_BITS + 1] >> (bitcount % BIGNUM_INT_BITS)) == 0) bitcount--;
825 * Return the byte length of a bignum when SSH-1 encoded.
827 int ssh1_bignum_length(Bignum bn)
829 return 2 + (bignum_bitcount(bn) + 7) / 8;
833 * Return the byte length of a bignum when SSH-2 encoded.
835 int ssh2_bignum_length(Bignum bn)
837 return 4 + (bignum_bitcount(bn) + 8) / 8;
841 * Return a byte from a bignum; 0 is least significant, etc.
843 int bignum_byte(Bignum bn, int i)
845 if (i >= (int)(BIGNUM_INT_BYTES * bn[0]))
846 return 0; /* beyond the end */
848 return (bn[i / BIGNUM_INT_BYTES + 1] >>
849 ((i % BIGNUM_INT_BYTES)*8)) & 0xFF;
853 * Return a bit from a bignum; 0 is least significant, etc.
855 int bignum_bit(Bignum bn, int i)
857 if (i >= (int)(BIGNUM_INT_BITS * bn[0]))
858 return 0; /* beyond the end */
860 return (bn[i / BIGNUM_INT_BITS + 1] >> (i % BIGNUM_INT_BITS)) & 1;
864 * Set a bit in a bignum; 0 is least significant, etc.
866 void bignum_set_bit(Bignum bn, int bitnum, int value)
868 if (bitnum >= (int)(BIGNUM_INT_BITS * bn[0]))
869 abort(); /* beyond the end */
871 int v = bitnum / BIGNUM_INT_BITS + 1;
872 int mask = 1 << (bitnum % BIGNUM_INT_BITS);
881 * Write a SSH-1-format bignum into a buffer. It is assumed the
882 * buffer is big enough. Returns the number of bytes used.
884 int ssh1_write_bignum(void *data, Bignum bn)
886 unsigned char *p = data;
887 int len = ssh1_bignum_length(bn);
889 int bitc = bignum_bitcount(bn);
891 *p++ = (bitc >> 8) & 0xFF;
892 *p++ = (bitc) & 0xFF;
893 for (i = len - 2; i--;)
894 *p++ = bignum_byte(bn, i);
899 * Compare two bignums. Returns like strcmp.
901 int bignum_cmp(Bignum a, Bignum b)
903 int amax = a[0], bmax = b[0];
904 int i = (amax > bmax ? amax : bmax);
906 BignumInt aval = (i > amax ? 0 : a[i]);
907 BignumInt bval = (i > bmax ? 0 : b[i]);
918 * Right-shift one bignum to form another.
920 Bignum bignum_rshift(Bignum a, int shift)
923 int i, shiftw, shiftb, shiftbb, bits;
926 bits = bignum_bitcount(a) - shift;
927 ret = newbn((bits + BIGNUM_INT_BITS - 1) / BIGNUM_INT_BITS);
930 shiftw = shift / BIGNUM_INT_BITS;
931 shiftb = shift % BIGNUM_INT_BITS;
932 shiftbb = BIGNUM_INT_BITS - shiftb;
935 for (i = 1; i <= (int)ret[0]; i++) {
937 ai1 = (i + shiftw + 1 <= (int)a[0] ? a[i + shiftw + 1] : 0);
938 ret[i] = ((ai >> shiftb) | (ai1 << shiftbb)) & BIGNUM_INT_MASK;
946 * Non-modular multiplication and addition.
948 Bignum bigmuladd(Bignum a, Bignum b, Bignum addend)
950 int alen = a[0], blen = b[0];
951 int mlen = (alen > blen ? alen : blen);
952 int rlen, i, maxspot;
953 BignumInt *workspace;
956 /* mlen space for a, mlen space for b, 2*mlen for result */
957 workspace = snewn(mlen * 4, BignumInt);
958 for (i = 0; i < mlen; i++) {
959 workspace[0 * mlen + i] = (mlen - i <= (int)a[0] ? a[mlen - i] : 0);
960 workspace[1 * mlen + i] = (mlen - i <= (int)b[0] ? b[mlen - i] : 0);
963 internal_mul(workspace + 0 * mlen, workspace + 1 * mlen,
964 workspace + 2 * mlen, mlen);
966 /* now just copy the result back */
967 rlen = alen + blen + 1;
968 if (addend && rlen <= (int)addend[0])
969 rlen = addend[0] + 1;
972 for (i = 1; i <= (int)ret[0]; i++) {
973 ret[i] = (i <= 2 * mlen ? workspace[4 * mlen - i] : 0);
979 /* now add in the addend, if any */
981 BignumDblInt carry = 0;
982 for (i = 1; i <= rlen; i++) {
983 carry += (i <= (int)ret[0] ? ret[i] : 0);
984 carry += (i <= (int)addend[0] ? addend[i] : 0);
985 ret[i] = (BignumInt) carry & BIGNUM_INT_MASK;
986 carry >>= BIGNUM_INT_BITS;
987 if (ret[i] != 0 && i > maxspot)
998 * Non-modular multiplication.
1000 Bignum bigmul(Bignum a, Bignum b)
1002 return bigmuladd(a, b, NULL);
1006 * Create a bignum which is the bitmask covering another one. That
1007 * is, the smallest integer which is >= N and is also one less than
1010 Bignum bignum_bitmask(Bignum n)
1012 Bignum ret = copybn(n);
1017 while (n[i] == 0 && i > 0)
1020 return ret; /* input was zero */
1026 ret[i] = BIGNUM_INT_MASK;
1031 * Convert a (max 32-bit) long into a bignum.
1033 Bignum bignum_from_long(unsigned long nn)
1036 BignumDblInt n = nn;
1039 ret[1] = (BignumInt)(n & BIGNUM_INT_MASK);
1040 ret[2] = (BignumInt)((n >> BIGNUM_INT_BITS) & BIGNUM_INT_MASK);
1042 ret[0] = (ret[2] ? 2 : 1);
1047 * Add a long to a bignum.
1049 Bignum bignum_add_long(Bignum number, unsigned long addendx)
1051 Bignum ret = newbn(number[0] + 1);
1053 BignumDblInt carry = 0, addend = addendx;
1055 for (i = 1; i <= (int)ret[0]; i++) {
1056 carry += addend & BIGNUM_INT_MASK;
1057 carry += (i <= (int)number[0] ? number[i] : 0);
1058 addend >>= BIGNUM_INT_BITS;
1059 ret[i] = (BignumInt) carry & BIGNUM_INT_MASK;
1060 carry >>= BIGNUM_INT_BITS;
1069 * Compute the residue of a bignum, modulo a (max 16-bit) short.
1071 unsigned short bignum_mod_short(Bignum number, unsigned short modulus)
1073 BignumDblInt mod, r;
1078 for (i = number[0]; i > 0; i--)
1079 r = (r * (BIGNUM_TOP_BIT % mod) * 2 + number[i] % mod) % mod;
1080 return (unsigned short) r;
1084 void diagbn(char *prefix, Bignum md)
1086 int i, nibbles, morenibbles;
1087 static const char hex[] = "0123456789ABCDEF";
1089 debug(("%s0x", prefix ? prefix : ""));
1091 nibbles = (3 + bignum_bitcount(md)) / 4;
1094 morenibbles = 4 * md[0] - nibbles;
1095 for (i = 0; i < morenibbles; i++)
1097 for (i = nibbles; i--;)
1099 hex[(bignum_byte(md, i / 2) >> (4 * (i % 2))) & 0xF]));
1109 Bignum bigdiv(Bignum a, Bignum b)
1111 Bignum q = newbn(a[0]);
1112 bigdivmod(a, b, NULL, q);
1119 Bignum bigmod(Bignum a, Bignum b)
1121 Bignum r = newbn(b[0]);
1122 bigdivmod(a, b, r, NULL);
1127 * Greatest common divisor.
1129 Bignum biggcd(Bignum av, Bignum bv)
1131 Bignum a = copybn(av);
1132 Bignum b = copybn(bv);
1134 while (bignum_cmp(b, Zero) != 0) {
1135 Bignum t = newbn(b[0]);
1136 bigdivmod(a, b, t, NULL);
1137 while (t[0] > 1 && t[t[0]] == 0)
1149 * Modular inverse, using Euclid's extended algorithm.
1151 Bignum modinv(Bignum number, Bignum modulus)
1153 Bignum a = copybn(modulus);
1154 Bignum b = copybn(number);
1155 Bignum xp = copybn(Zero);
1156 Bignum x = copybn(One);
1159 while (bignum_cmp(b, One) != 0) {
1160 Bignum t = newbn(b[0]);
1161 Bignum q = newbn(a[0]);
1162 bigdivmod(a, b, t, q);
1163 while (t[0] > 1 && t[t[0]] == 0)
1170 x = bigmuladd(q, xp, t);
1180 /* now we know that sign * x == 1, and that x < modulus */
1182 /* set a new x to be modulus - x */
1183 Bignum newx = newbn(modulus[0]);
1184 BignumInt carry = 0;
1188 for (i = 1; i <= (int)newx[0]; i++) {
1189 BignumInt aword = (i <= (int)modulus[0] ? modulus[i] : 0);
1190 BignumInt bword = (i <= (int)x[0] ? x[i] : 0);
1191 newx[i] = aword - bword - carry;
1193 carry = carry ? (newx[i] >= bword) : (newx[i] > bword);
1207 * Render a bignum into decimal. Return a malloced string holding
1208 * the decimal representation.
1210 char *bignum_decimal(Bignum x)
1212 int ndigits, ndigit;
1216 BignumInt *workspace;
1219 * First, estimate the number of digits. Since log(10)/log(2)
1220 * is just greater than 93/28 (the joys of continued fraction
1221 * approximations...) we know that for every 93 bits, we need
1222 * at most 28 digits. This will tell us how much to malloc.
1224 * Formally: if x has i bits, that means x is strictly less
1225 * than 2^i. Since 2 is less than 10^(28/93), this is less than
1226 * 10^(28i/93). We need an integer power of ten, so we must
1227 * round up (rounding down might make it less than x again).
1228 * Therefore if we multiply the bit count by 28/93, rounding
1229 * up, we will have enough digits.
1231 * i=0 (i.e., x=0) is an irritating special case.
1233 i = bignum_bitcount(x);
1235 ndigits = 1; /* x = 0 */
1237 ndigits = (28 * i + 92) / 93; /* multiply by 28/93 and round up */
1238 ndigits++; /* allow for trailing \0 */
1239 ret = snewn(ndigits, char);
1242 * Now allocate some workspace to hold the binary form as we
1243 * repeatedly divide it by ten. Initialise this to the
1244 * big-endian form of the number.
1246 workspace = snewn(x[0], BignumInt);
1247 for (i = 0; i < (int)x[0]; i++)
1248 workspace[i] = x[x[0] - i];
1251 * Next, write the decimal number starting with the last digit.
1252 * We use ordinary short division, dividing 10 into the
1255 ndigit = ndigits - 1;
1260 for (i = 0; i < (int)x[0]; i++) {
1261 carry = (carry << BIGNUM_INT_BITS) + workspace[i];
1262 workspace[i] = (BignumInt) (carry / 10);
1267 ret[--ndigit] = (char) (carry + '0');
1271 * There's a chance we've fallen short of the start of the
1272 * string. Correct if so.
1275 memmove(ret, ret + ndigit, ndigits - ndigit);