1 /* Copyright (c) 2015, Google Inc.
2  *
3  * Permission to use, copy, modify, and/or distribute this software for any
4  * purpose with or without fee is hereby granted, provided that the above
5  * copyright notice and this permission notice appear in all copies.
6  *
7  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
8  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
9  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
10  * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
11  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
12  * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
13  * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
14 
15 #include <openssl/base.h>
16 
17 
18 #if defined(OPENSSL_64_BIT) && !defined(OPENSSL_WINDOWS)
19 
20 #include <openssl/ec.h>
21 
22 #include "internal.h"
23 
24 /* Convert an array of points into affine coordinates. (If the point at
25  * infinity is found (Z = 0), it remains unchanged.) This function is
26  * essentially an equivalent to EC_POINTs_make_affine(), but works with the
27  * internal representation of points as used by ecp_nistp###.c rather than
28  * with (BIGNUM-based) EC_POINT data structures. point_array is the
29  * input/output buffer ('num' points in projective form, i.e. three
30  * coordinates each), based on an internal representation of field elements
31  * of size 'felem_size'. tmp_felems needs to point to a temporary array of
32  * 'num'+1 field elements for storage of intermediate values. */
ec_GFp_nistp_points_make_affine_internal(size_t num,void * point_array,size_t felem_size,void * tmp_felems,void (* felem_one)(void * out),int (* felem_is_zero)(const void * in),void (* felem_assign)(void * out,const void * in),void (* felem_square)(void * out,const void * in),void (* felem_mul)(void * out,const void * in1,const void * in2),void (* felem_inv)(void * out,const void * in),void (* felem_contract)(void * out,const void * in))33 void ec_GFp_nistp_points_make_affine_internal(
34     size_t num, void *point_array, size_t felem_size, void *tmp_felems,
35     void (*felem_one)(void *out), int (*felem_is_zero)(const void *in),
36     void (*felem_assign)(void *out, const void *in),
37     void (*felem_square)(void *out, const void *in),
38     void (*felem_mul)(void *out, const void *in1, const void *in2),
39     void (*felem_inv)(void *out, const void *in),
40     void (*felem_contract)(void *out, const void *in)) {
41   int i = 0;
42 
43 #define tmp_felem(I) (&((char *)tmp_felems)[(I)*felem_size])
44 #define X(I) (&((char *)point_array)[3 * (I)*felem_size])
45 #define Y(I) (&((char *)point_array)[(3 * (I) + 1) * felem_size])
46 #define Z(I) (&((char *)point_array)[(3 * (I) + 2) * felem_size])
47 
48   if (!felem_is_zero(Z(0))) {
49     felem_assign(tmp_felem(0), Z(0));
50   } else {
51     felem_one(tmp_felem(0));
52   }
53 
54   for (i = 1; i < (int)num; i++) {
55     if (!felem_is_zero(Z(i))) {
56       felem_mul(tmp_felem(i), tmp_felem(i - 1), Z(i));
57     } else {
58       felem_assign(tmp_felem(i), tmp_felem(i - 1));
59     }
60   }
61   /* Now each tmp_felem(i) is the product of Z(0) .. Z(i), skipping any
62    * zero-valued factors: if Z(i) = 0, we essentially pretend that Z(i) = 1. */
63 
64   felem_inv(tmp_felem(num - 1), tmp_felem(num - 1));
65   for (i = num - 1; i >= 0; i--) {
66     if (i > 0) {
67       /* tmp_felem(i-1) is the product of Z(0) .. Z(i-1), tmp_felem(i)
68        * is the inverse of the product of Z(0) .. Z(i). */
69       /* 1/Z(i) */
70       felem_mul(tmp_felem(num), tmp_felem(i - 1), tmp_felem(i));
71     } else {
72       felem_assign(tmp_felem(num), tmp_felem(0)); /* 1/Z(0) */
73     }
74 
75     if (!felem_is_zero(Z(i))) {
76       if (i > 0) {
77         /* For next iteration, replace tmp_felem(i-1) by its inverse. */
78         felem_mul(tmp_felem(i - 1), tmp_felem(i), Z(i));
79       }
80 
81       /* Convert point (X, Y, Z) into affine form (X/(Z^2), Y/(Z^3), 1). */
82       felem_square(Z(i), tmp_felem(num));    /* 1/(Z^2) */
83       felem_mul(X(i), X(i), Z(i));           /* X/(Z^2) */
84       felem_mul(Z(i), Z(i), tmp_felem(num)); /* 1/(Z^3) */
85       felem_mul(Y(i), Y(i), Z(i));           /* Y/(Z^3) */
86       felem_contract(X(i), X(i));
87       felem_contract(Y(i), Y(i));
88       felem_one(Z(i));
89     } else {
90       if (i > 0) {
91         /* For next iteration, replace tmp_felem(i-1) by its inverse. */
92         felem_assign(tmp_felem(i - 1), tmp_felem(i));
93       }
94     }
95   }
96 }
97 
98 /* This function looks at 5+1 scalar bits (5 current, 1 adjacent less
99  * significant bit), and recodes them into a signed digit for use in fast point
100  * multiplication: the use of signed rather than unsigned digits means that
101  * fewer points need to be precomputed, given that point inversion is easy (a
102  * precomputed point dP makes -dP available as well).
103  *
104  * BACKGROUND:
105  *
106  * Signed digits for multiplication were introduced by Booth ("A signed binary
107  * multiplication technique", Quart. Journ. Mech. and Applied Math., vol. IV,
108  * pt. 2 (1951), pp. 236-240), in that case for multiplication of integers.
109  * Booth's original encoding did not generally improve the density of nonzero
110  * digits over the binary representation, and was merely meant to simplify the
111  * handling of signed factors given in two's complement; but it has since been
112  * shown to be the basis of various signed-digit representations that do have
113  * further advantages, including the wNAF, using the following general
114  * approach:
115  *
116  * (1) Given a binary representation
117  *
118  *       b_k  ...  b_2  b_1  b_0,
119  *
120  *     of a nonnegative integer (b_k in {0, 1}), rewrite it in digits 0, 1, -1
121  *     by using bit-wise subtraction as follows:
122  *
123  *        b_k b_(k-1)  ...  b_2  b_1  b_0
124  *      -     b_k      ...  b_3  b_2  b_1  b_0
125  *       -------------------------------------
126  *        s_k b_(k-1)  ...  s_3  s_2  s_1  s_0
127  *
128  *     A left-shift followed by subtraction of the original value yields a new
129  *     representation of the same value, using signed bits s_i = b_(i+1) - b_i.
130  *     This representation from Booth's paper has since appeared in the
131  *     literature under a variety of different names including "reversed binary
132  *     form", "alternating greedy expansion", "mutual opposite form", and
133  *     "sign-alternating {+-1}-representation".
134  *
135  *     An interesting property is that among the nonzero bits, values 1 and -1
136  *     strictly alternate.
137  *
138  * (2) Various window schemes can be applied to the Booth representation of
139  *     integers: for example, right-to-left sliding windows yield the wNAF
140  *     (a signed-digit encoding independently discovered by various researchers
141  *     in the 1990s), and left-to-right sliding windows yield a left-to-right
142  *     equivalent of the wNAF (independently discovered by various researchers
143  *     around 2004).
144  *
145  * To prevent leaking information through side channels in point multiplication,
146  * we need to recode the given integer into a regular pattern: sliding windows
147  * as in wNAFs won't do, we need their fixed-window equivalent -- which is a few
148  * decades older: we'll be using the so-called "modified Booth encoding" due to
149  * MacSorley ("High-speed arithmetic in binary computers", Proc. IRE, vol. 49
150  * (1961), pp. 67-91), in a radix-2^5 setting.  That is, we always combine five
151  * signed bits into a signed digit:
152  *
153  *       s_(4j + 4) s_(4j + 3) s_(4j + 2) s_(4j + 1) s_(4j)
154  *
155  * The sign-alternating property implies that the resulting digit values are
156  * integers from -16 to 16.
157  *
158  * Of course, we don't actually need to compute the signed digits s_i as an
159  * intermediate step (that's just a nice way to see how this scheme relates
160  * to the wNAF): a direct computation obtains the recoded digit from the
161  * six bits b_(4j + 4) ... b_(4j - 1).
162  *
163  * This function takes those five bits as an integer (0 .. 63), writing the
164  * recoded digit to *sign (0 for positive, 1 for negative) and *digit (absolute
165  * value, in the range 0 .. 8).  Note that this integer essentially provides the
166  * input bits "shifted to the left" by one position: for example, the input to
167  * compute the least significant recoded digit, given that there's no bit b_-1,
168  * has to be b_4 b_3 b_2 b_1 b_0 0. */
ec_GFp_nistp_recode_scalar_bits(uint8_t * sign,uint8_t * digit,uint8_t in)169 void ec_GFp_nistp_recode_scalar_bits(uint8_t *sign, uint8_t *digit,
170                                      uint8_t in) {
171   uint8_t s, d;
172 
173   s = ~((in >> 5) - 1); /* sets all bits to MSB(in), 'in' seen as
174                           * 6-bit value */
175   d = (1 << 6) - in - 1;
176   d = (d & s) | (in & ~s);
177   d = (d >> 1) + (d & 1);
178 
179   *sign = s & 1;
180   *digit = d;
181 }
182 
183 #endif  /* 64_BIT && !WINDOWS */
184