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/18510)

3 files changed, 34 insertions, 0 deletions

diff --git a/crypto/bn/rsaz_exp.c b/crypto/bn/rsaz_exp.c
index 2dbcb88ac3..e44eae43be 100644
--- 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 b4fd3cbbba..45dc9cc197 100644
--- 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],
@@ -52,6 +54,27 @@ int ossl_rsaz_mod_exp_avx512_x2(BN_ULONG *res1,
                                 BN_ULONG k0_2,
                                 int factor_size);
 
+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/crypto/bn/rsaz_exp_x2.c b/crypto/bn/rsaz_exp_x2.c
index 9969d45e40..6b04486e3f 100644
--- a/crypto/bn/rsaz_exp_x2.c
+++ b/crypto/bn/rsaz_exp_x2.c
@@ -257,6 +257,9 @@ int ossl_rsaz_mod_exp_avx512_x2(BN_ULONG *res1,
     from_words52(res1, factor_size, rr1_red);
     from_words52(res2, factor_size, rr2_red);
 
+    bn_reduce_once_in_place(res1, /*carry=*/0, m1, storage, factor_size);
+    bn_reduce_once_in_place(res2, /*carry=*/0, m2, storage, factor_size);
+
 err:
     if (storage != NULL) {
         OPENSSL_cleanse(storage, storage_len_bytes);