Add an extra reduction step to RSAZ mod_exp implementations

Inspired by BoringSSL fix by David Benjamin.

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18511)

diff --git a/crypto/bn/rsaz_exp.c b/crypto/bn/rsaz_exp.c
index 22455b8a63745965c68251211619b3b814ebb9c4..5c5cd4c2821523a05b63a251e120016ff82823c2 100644 (file)
--- a/crypto/bn/rsaz_exp.c
+++ b/crypto/bn/rsaz_exp.c
@@ -66,6 +66,7 @@ void RSAZ_1024_mod_exp_avx2(BN_ULONG result_norm[16],
     unsigned char *R2 = table_s; /* borrow */
     int index;
     int wvalue;
+    BN_ULONG tmp[16];
 
     if ((((size_t)p_str & 4095) + 320) >> 12) {
         result = p_str;
@@ -237,7 +238,10 @@ void RSAZ_1024_mod_exp_avx2(BN_ULONG result_norm[16],
 
     rsaz_1024_red2norm_avx2(result_norm, result);
 
+    bn_reduce_once_in_place(result_norm, /*carry=*/0, m_norm, tmp, 16);
+
     OPENSSL_cleanse(storage, sizeof(storage));
+    OPENSSL_cleanse(tmp, sizeof(tmp));
 }
 
 /*
@@ -266,6 +270,7 @@ void RSAZ_512_mod_exp(BN_ULONG result[8],
     unsigned char *p_str = (unsigned char *)exponent;
     int index;
     unsigned int wvalue;
+    BN_ULONG tmp[8];
 
     /* table[0] = 1_inv */
     temp[0] = 0 - m[0];
@@ -309,7 +314,10 @@ void RSAZ_512_mod_exp(BN_ULONG result[8],
     /* from Montgomery */
     rsaz_512_mul_by_one(result, temp, m, k0);
 
+    bn_reduce_once_in_place(result, /*carry=*/0, m, tmp, 8);
+
     OPENSSL_cleanse(storage, sizeof(storage));
+    OPENSSL_cleanse(tmp, sizeof(tmp));
 }
 
 #endif

diff --git a/crypto/bn/rsaz_exp.h b/crypto/bn/rsaz_exp.h
index 88f65a4bae46129128ec34ce4fcadf6895e20f9e..606496d45a06d29765afd037054254b6da0f4aaa 100644 (file)
--- a/crypto/bn/rsaz_exp.h
+++ b/crypto/bn/rsaz_exp.h
@@ -22,6 +22,8 @@
 #  define RSAZ_ENABLED
 
 #  include <openssl/bn.h>
+#  include "internal/constant_time.h"
+#  include "bn_local.h"
 
 void RSAZ_1024_mod_exp_avx2(BN_ULONG result[16],
                             const BN_ULONG base_norm[16],
@@ -35,6 +37,27 @@ void RSAZ_512_mod_exp(BN_ULONG result[8],
                       const BN_ULONG m_norm[8], BN_ULONG k0,
                       const BN_ULONG RR[8]);
 
+static ossl_inline void bn_select_words(BN_ULONG *r, BN_ULONG mask,
+                                        const BN_ULONG *a,
+                                        const BN_ULONG *b, size_t num)
+{
+    size_t i;
+
+    for (i = 0; i < num; i++) {
+        r[i] = constant_time_select_64(mask, a[i], b[i]);
+    }
+}
+
+static ossl_inline BN_ULONG bn_reduce_once_in_place(BN_ULONG *r,
+                                                    BN_ULONG carry,
+                                                    const BN_ULONG *m,
+                                                    BN_ULONG *tmp, size_t num)
+{
+    carry -= bn_sub_words(tmp, r, m, num);
+    bn_select_words(r, carry, r /* tmp < 0 */, tmp /* tmp >= 0 */, num);
+    return carry;
+}
+
 # endif
 
 #endif

diff --git a/test/recipes/10-test_bn_data/bnmod.txt b/test/recipes/10-test_bn_data/bnmod.txt
index 69f8af43d5a3f41ea47b9a0c992f8af3115d337c..edde03bd6294d677c40beffcc3b8699d350814e7 100644 (file)
--- a/test/recipes/10-test_bn_data/bnmod.txt
+++ b/test/recipes/10-test_bn_data/bnmod.txt
@@ -2493,12 +2493,10 @@ E = ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff
 M = 8f42c9e9e351ba9b32ab0cf69da43f4acf7028d19cff6e5059ea0e3fcc97c97f36a31470044737d4c0c933ac441ecb29e32c81401523afdac7de9c3fd8493c97
 
 # 1024-bit
-# TODO(davidben): This test breaks the RSAZ implementation. Fix it and enable
-# this test.
-# ModExp = 00
-# A = 800000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000002f
-# E = ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff
-# M = 9da8dc26fdf4d2e49833b240ee552beb7a6e251caa91bfb5d6cafaf8ed9461877fda8f6ac299036d35806bc1ae7872e54eaac1ec6bee6d02c6621a9cf8883b3abc33c49b3e601203e0e86ef8f0562412cc689ee2670704583909ca6d7774c9f9f9f4d77d37fedef9cb51d207cb629ec02fa03b526fd6594bfa8f2da71238a0b7
+ModExp = 00
+A = 800000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000002f
+E = ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff
+M = 9da8dc26fdf4d2e49833b240ee552beb7a6e251caa91bfb5d6cafaf8ed9461877fda8f6ac299036d35806bc1ae7872e54eaac1ec6bee6d02c6621a9cf8883b3abc33c49b3e601203e0e86ef8f0562412cc689ee2670704583909ca6d7774c9f9f9f4d77d37fedef9cb51d207cb629ec02fa03b526fd6594bfa8f2da71238a0b7
 
 # 1025-bit
 ModExp = 00