X-Git-Url: https://www.bearssl.org/gitweb//home/git/?p=BearSSL;a=blobdiff_plain;f=src%2Frsa%2Frsa_i15_priv.c;h=177cc3a6643ab4359d2385db95cf86b21c557464;hp=8dc839ae81472a0bbb368d20e76f95e6c27f2081;hb=b2a08e908d6a1802c9d50aca699193b38b31a0b2;hpb=556e525d62cd5559e74fe4d2777a59d33590a033

diff --git a/src/rsa/rsa_i15_priv.c b/src/rsa/rsa_i15_priv.c
index 8dc839a..177cc3a 100644
--- a/src/rsa/rsa_i15_priv.c
+++ b/src/rsa/rsa_i15_priv.c
@@ -24,27 +24,9 @@
 
 #include "inner.h"
 
-#define U      (1 + ((BR_MAX_RSA_FACTOR + 14) / 15))
+#define U      (2 + ((BR_MAX_RSA_FACTOR + 14) / 15))
 #define TLEN   (8 * U)
 
-/* obsolete
-static void
-print_int(const char *name, const uint16_t *x)
-{
-	extern int printf(const char *fmt, ...);
-	unsigned char tmp[1000];
-	size_t u, len;
-
-	len = (x[0] - (x[0] >> 4) + 7) >> 3;
-	br_i15_encode(tmp, len, x);
-	printf("%s = ", name);
-	for (u = 0; u < len; u ++) {
-		printf("%02X", tmp[u]);
-	}
-	printf("\n");
-}
-*/
-
 /* see bearssl_rsa.h */
 uint32_t
 br_rsa_i15_private(unsigned char *x, const br_rsa_private_key *sk)
@@ -53,8 +35,8 @@ br_rsa_i15_private(unsigned char *x, const br_rsa_private_key *sk)
 	size_t plen, qlen;
 	size_t fwlen;
 	uint16_t p0i, q0i;
-	size_t xlen;
-	uint16_t tmp[TLEN];
+	size_t xlen, u;
+	uint16_t tmp[1 + TLEN];
 	long z;
 	uint16_t *mp, *mq, *s1, *s2, *t1, *t2, *t3;
 	uint32_t r;
@@ -86,6 +68,10 @@ br_rsa_i15_private(unsigned char *x, const br_rsa_private_key *sk)
 		z -= 15;
 		fwlen ++;
 	}
+	/*
+	 * Round up the word length to an even number.
+	 */
+	fwlen += (fwlen & 1);
 
 	/*
 	 * We need to fit at least 6 values in the stack buffer.
@@ -100,34 +86,78 @@ br_rsa_i15_private(unsigned char *x, const br_rsa_private_key *sk)
 	xlen = (sk->n_bitlen + 7) >> 3;
 
 	/*
-	 * Decode q.
+	 * Ensure 32-bit alignment for value words.
 	 */
 	mq = tmp;
-	br_i15_decode(mq, q, qlen);
+	if (((uintptr_t)mq & 2) == 0) {
+		mq ++;
+	}
 
 	/*
-	 * Compute s2 = x^dq mod q.
+	 * Decode q.
 	 */
-	q0i = br_i15_ninv15(mq[1]);
-	s2 = tmp + fwlen;
-	br_i15_decode_reduce(s2, x, xlen, mq);
-	r = br_i15_modpow_opt(s2, sk->dq, sk->dqlen, mq, q0i,
-		tmp + 2 * fwlen, TLEN - 2 * fwlen);
+	br_i15_decode(mq, q, qlen);
 
 	/*
 	 * Decode p.
 	 */
-	mp = tmp + 2 * fwlen;
-	br_i15_decode(mp, p, plen);
+	t1 = mq + fwlen;
+	br_i15_decode(t1, p, plen);
+
+	/*
+	 * Compute the modulus (product of the two factors), to compare
+	 * it with the source value. We use br_i15_mulacc(), since it's
+	 * already used later on.
+	 */
+	t2 = mq + 2 * fwlen;
+	br_i15_zero(t2, mq[0]);
+	br_i15_mulacc(t2, mq, t1);
+
+	/*
+	 * We encode the modulus into bytes, to perform the comparison
+	 * with bytes. We know that the product length, in bytes, is
+	 * exactly xlen.
+	 * The comparison actually computes the carry when subtracting
+	 * the modulus from the source value; that carry must be 1 for
+	 * a value in the correct range. We keep it in r, which is our
+	 * accumulator for the error code.
+	 */
+	t3 = mq + 4 * fwlen;
+	br_i15_encode(t3, xlen, t2);
+	u = xlen;
+	r = 0;
+	while (u > 0) {
+		uint32_t wn, wx;
+
+		u --;
+		wn = ((unsigned char *)t3)[u];
+		wx = x[u];
+		r = ((wx - (wn + r)) >> 8) & 1;
+	}
+
+	/*
+	 * Move the decoded p to another temporary buffer.
+	 */
+	mp = mq + 2 * fwlen;
+	memmove(mp, t1, fwlen * sizeof *t1);
+
+	/*
+	 * Compute s2 = x^dq mod q.
+	 */
+	q0i = br_i15_ninv15(mq[1]);
+	s2 = mq + fwlen;
+	br_i15_decode_reduce(s2, x, xlen, mq);
+	r &= br_i15_modpow_opt(s2, sk->dq, sk->dqlen, mq, q0i,
+		mq + 3 * fwlen, TLEN - 3 * fwlen);
 
 	/*
 	 * Compute s1 = x^dq mod q.
 	 */
 	p0i = br_i15_ninv15(mp[1]);
-	s1 = tmp + 3 * fwlen;
+	s1 = mq + 3 * fwlen;
 	br_i15_decode_reduce(s1, x, xlen, mp);
 	r &= br_i15_modpow_opt(s1, sk->dp, sk->dplen, mp, p0i,
-		tmp + 4 * fwlen, TLEN - 4 * fwlen);
+		mq + 4 * fwlen, TLEN - 4 * fwlen);
 
 	/*
 	 * Compute:
@@ -142,8 +172,8 @@ br_rsa_i15_private(unsigned char *x, const br_rsa_private_key *sk)
 	 * inverse of q modulo p), we also tolerate improperly large
 	 * values for this parameter.
 	 */
-	t1 = tmp + 4 * fwlen;
-	t2 = tmp + 5 * fwlen;
+	t1 = mq + 4 * fwlen;
+	t2 = mq + 5 * fwlen;
 	br_i15_reduce(t2, s2, mp);
 	br_i15_add(s1, mp, br_i15_sub(s1, t2, 1));
 	br_i15_to_monty(s1, mp);