Added intermediate casts to void* to prevent spurious warnings (with Clang and -Wcast...

[BearSSL] / src / ec / ec_c25519_i15.c

diff --git a/src/ec/ec_c25519_i15.c b/src/ec/ec_c25519_i15.c
index 79560ae..361e75f 100644 (file)
--- a/src/ec/ec_c25519_i15.c
+++ b/src/ec/ec_c25519_i15.c
@@ -46,6 +46,27 @@ static const uint16_t C255_R2[] = {
        0x0000
 };
 
+/* obsolete
+#include <stdio.h>
+#include <stdlib.h>
+static void
+print_int_mont(const char *name, const uint16_t *x)
+{
+       uint16_t y[18];
+       unsigned char tmp[32];
+       size_t u;
+
+       printf("%s = ", name);
+       memcpy(y, x, sizeof y);
+       br_i15_from_monty(y, C255_P, P0I);
+       br_i15_encode(tmp, sizeof tmp, y);
+       for (u = 0; u < sizeof tmp; u ++) {
+               printf("%02X", tmp[u]);
+       }
+       printf("\n");
+}
+*/
+
 static const uint16_t C255_A24[] = {
        0x0110,
        0x45D3, 0x0046, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
@@ -61,10 +82,10 @@ static const unsigned char GEN[] = {
 };
 
 static const unsigned char ORDER[] = {
-       0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-       0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-       0x14, 0xDE, 0xF9, 0xDE, 0xA2, 0xF7, 0x9C, 0xD6,
-       0x58, 0x12, 0x63, 0x1A, 0x5C, 0xF5, 0xD3, 0xED
+       0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+       0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+       0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+       0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF
 };
 
 static const unsigned char *
@@ -83,6 +104,14 @@ api_order(int curve, size_t *len)
        return ORDER;
 }
 
+static size_t
+api_xoff(int curve, size_t *len)
+{
+       (void)curve;
+       *len = 32;
+       return 0;
+}
+
 static void
 cswap(uint16_t *a, uint16_t *b, uint32_t ctl)
 {
@@ -150,8 +179,14 @@ static uint32_t
 api_mul(unsigned char *G, size_t Glen,
        const unsigned char *kb, size_t kblen, int curve)
 {
+#define ILEN   (18 * sizeof(uint16_t))
+
+       /*
+        * The a[] and b[] arrays have an extra word to allow for
+        * decoding without using br_i15_decode_reduce().
+        */
        uint16_t x1[18], x2[18], x3[18], z2[18], z3[18];
-       uint16_t a[18], aa[18], b[18], bb[18];
+       uint16_t a[19], aa[18], b[19], bb[18];
        uint16_t c[18], d[18], e[18], da[18], cb[18];
        unsigned char k[32];
        uint32_t swap;
@@ -176,17 +211,33 @@ api_mul(unsigned char *G, size_t Glen,
         */
        byteswap(G);
 
+       /*
+        * Decode the point ('u' coordinate). This should be reduced
+        * modulo p, but we prefer to avoid the dependency on
+        * br_i15_decode_reduce(). Instead, we use br_i15_decode_mod()
+        * with a synthetic modulus of value 2^255 (this must work
+        * since G was truncated to 255 bits), then use a conditional
+        * subtraction. We use br_i15_decode_mod() and not
+        * br_i15_decode(), because the ec_prime_i15 implementation uses
+        * the former but not the latter.
+        *    br_i15_decode_reduce(a, G, 32, C255_P);
+        */
+       br_i15_zero(b, 0x111);
+       b[18] = 1;
+       br_i15_decode_mod(a, G, 32, b);
+       a[0] = 0x110;
+       br_i15_sub(a, C255_P, NOT(br_i15_sub(a, C255_P, 0)));
+
        /*
         * Initialise variables x1, x2, z2, x3 and z3. We set all of them
         * into Montgomery representation.
         */
-       br_i15_decode_reduce(a, G, 32, C255_P);
        br_i15_montymul(x1, a, C255_R2, C255_P, P0I);
-       memcpy(x3, x1, sizeof x1);
+       memcpy(x3, x1, ILEN);
        br_i15_zero(z2, C255_P[0]);
-       memcpy(x2, z2, sizeof z2);
+       memcpy(x2, z2, ILEN);
        x2[1] = 19;
-       memcpy(z3, x2, sizeof x2);
+       memcpy(z3, x2, ILEN);
 
        memcpy(k, kb, kblen);
        memset(k + kblen, 0, (sizeof k) - kblen);
@@ -194,6 +245,10 @@ api_mul(unsigned char *G, size_t Glen,
        k[31] &= 0x7F;
        k[31] |= 0x40;
 
+       /* obsolete
+       print_int_mont("x1", x1);
+       */
+
        swap = 0;
        for (i = 254; i >= 0; i --) {
                uint32_t kt;
@@ -204,6 +259,13 @@ api_mul(unsigned char *G, size_t Glen,
                cswap(z2, z3, swap);
                swap = kt;
 
+               /* obsolete
+               print_int_mont("x2", x2);
+               print_int_mont("z2", z2);
+               print_int_mont("x3", x3);
+               print_int_mont("z3", z3);
+               */
+
                c255_add(a, x2, z2);
                c255_mul(aa, a, a);
                c255_sub(b, x2, z2);
@@ -213,6 +275,19 @@ api_mul(unsigned char *G, size_t Glen,
                c255_sub(d, x3, z3);
                c255_mul(da, d, a);
                c255_mul(cb, c, b);
+
+               /* obsolete
+               print_int_mont("a ", a);
+               print_int_mont("aa", aa);
+               print_int_mont("b ", b);
+               print_int_mont("bb", bb);
+               print_int_mont("e ", e);
+               print_int_mont("c ", c);
+               print_int_mont("d ", d);
+               print_int_mont("da", da);
+               print_int_mont("cb", cb);
+               */
+
                c255_add(x3, da, cb);
                c255_mul(x3, x3, x3);
                c255_sub(z3, da, cb);
@@ -222,6 +297,13 @@ api_mul(unsigned char *G, size_t Glen,
                c255_mul(z2, C255_A24, e);
                c255_add(z2, z2, aa);
                c255_mul(z2, e, z2);
+
+               /* obsolete
+               print_int_mont("x2", x2);
+               print_int_mont("z2", z2);
+               print_int_mont("x3", x3);
+               print_int_mont("z3", z3);
+               */
        }
        cswap(x2, x3, swap);
        cswap(z2, z3, swap);
@@ -231,12 +313,12 @@ api_mul(unsigned char *G, size_t Glen,
         * square-and-multiply algorithm; we mutualise most non-squarings
         * since the exponent contains almost only ones.
         */
-       memcpy(a, z2, sizeof z2);
+       memcpy(a, z2, ILEN);
        for (i = 0; i < 15; i ++) {
                c255_mul(a, a, a);
                c255_mul(a, a, z2);
        }
-       memcpy(b, a, sizeof a);
+       memcpy(b, a, ILEN);
        for (i = 0; i < 14; i ++) {
                int j;
 
@@ -251,11 +333,23 @@ api_mul(unsigned char *G, size_t Glen,
                        c255_mul(b, z2, b);
                }
        }
-       c255_mul(x2, x2, b);
-       br_i15_from_monty(x2, C255_P, P0I);
+       c255_mul(b, x2, b);
+
+       /*
+        * To avoid a dependency on br_i15_from_monty(), we use a
+        * Montgomery multiplication with 1.
+        *    memcpy(x2, b, ILEN);
+        *    br_i15_from_monty(x2, C255_P, P0I);
+        */
+       br_i15_zero(a, C255_P[0]);
+       a[1] = 1;
+       br_i15_montymul(x2, a, b, C255_P, P0I);
+
        br_i15_encode(G, 32, x2);
        byteswap(G);
        return 1;
+
+#undef ILEN
 }
 
 static size_t
@@ -297,6 +391,7 @@ const br_ec_impl br_ec_c25519_i15 = {
        (uint32_t)0x20000000,
        &api_generator,
        &api_order,
+       &api_xoff,
        &api_mul,
        &api_mulgen,
        &api_muladd