grpc 1.61.0 → 1.62.0

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/bytes.c

@@ -63,26 +63,31 @@
 
 void bn_big_endian_to_words(BN_ULONG *out, size_t out_len, const uint8_t *in,
                             size_t in_len) {
-  for (size_t i = 0; i < out_len; i++) {
-    if (in_len < sizeof(BN_ULONG)) {
-      // Load the last partial word.
-      BN_ULONG word = 0;
-      for (size_t j = 0; j < in_len; j++) {
-        word = (word << 8) | in[j];
-      }
-      in_len = 0;
-      out[i] = word;
-      // Fill the remainder with zeros.
-      OPENSSL_memset(out + i + 1, 0, (out_len - i - 1) * sizeof(BN_ULONG));
-      break;
-    }
+  // The caller should have sized |out| to fit |in| without truncating. This
+  // condition ensures we do not overflow |out|, so use a runtime check.
+  BSSL_CHECK(in_len <= out_len * sizeof(BN_ULONG));
 
+  // Load whole words.
+  while (in_len >= sizeof(BN_ULONG)) {
     in_len -= sizeof(BN_ULONG);
-    out[i] = CRYPTO_load_word_be(in + in_len);
+    out[0] = CRYPTO_load_word_be(in + in_len);
+    out++;
+    out_len--;
+  }
+
+  // Load the last partial word.
+  if (in_len != 0) {
+    BN_ULONG word = 0;
+    for (size_t i = 0; i < in_len; i++) {
+      word = (word << 8) | in[i];
+    }
+    out[0] = word;
+    out++;
+    out_len--;
   }
 
-  // The caller should have sized the output to avoid truncation.
-  assert(in_len == 0);
+  // Fill the remainder with zeros.
+  OPENSSL_memset(out, 0, out_len * sizeof(BN_ULONG));
 }
 
 BIGNUM *BN_bin2bn(const uint8_t *in, size_t len, BIGNUM *ret) {
@@ -116,7 +121,7 @@ BIGNUM *BN_bin2bn(const uint8_t *in, size_t len, BIGNUM *ret) {
   return ret;
 }
 
-BIGNUM *BN_le2bn(const uint8_t *in, size_t len, BIGNUM *ret) {
+BIGNUM *BN_lebin2bn(const uint8_t *in, size_t len, BIGNUM *ret) {
   BIGNUM *bn = NULL;
   if (ret == NULL) {
     bn = BN_new();
@@ -149,6 +154,10 @@ BIGNUM *BN_le2bn(const uint8_t *in, size_t len, BIGNUM *ret) {
   return ret;
 }
 
+BIGNUM *BN_le2bn(const uint8_t *in, size_t len, BIGNUM *ret) {
+  return BN_lebin2bn(in, len, ret);
+}
+
 // fits_in_bytes returns one if the |num_words| words in |words| can be
 // represented in |num_bytes| bytes.
 static int fits_in_bytes(const BN_ULONG *words, size_t num_words,

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/ctx.c

@@ -210,7 +210,7 @@ static int BN_STACK_push(BN_STACK *st, size_t idx) {
     // This function intentionally does not push to the error queue on error.
     // Error-reporting is deferred to |BN_CTX_get|.
     size_t new_size = st->size != 0 ? st->size * 3 / 2 : BN_CTX_START_FRAMES;
-    if (new_size <= st->size || new_size > ((size_t)-1) / sizeof(size_t)) {
+    if (new_size <= st->size || new_size > SIZE_MAX / sizeof(size_t)) {
       return 0;
     }
     size_t *new_indexes =

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/exponentiation.c

@@ -724,7 +724,7 @@ void bn_mod_exp_mont_small(BN_ULONG *r, const BN_ULONG *a, size_t num,
                            const BN_ULONG *p, size_t num_p,
                            const BN_MONT_CTX *mont) {
   if (num != (size_t)mont->N.width || num > BN_SMALL_MAX_WORDS ||
-      num_p > ((size_t)-1) / BN_BITS2) {
+      num_p > SIZE_MAX / BN_BITS2) {
     abort();
   }
   assert(BN_is_odd(&mont->N));
@@ -898,7 +898,9 @@ int BN_mod_exp_mont_consttime(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p,
     OPENSSL_PUT_ERROR(BN, BN_R_NEGATIVE_NUMBER);
     return 0;
   }
-  if (a->neg || BN_ucmp(a, m) >= 0) {
+  // |a| is secret, but it is required to be in range, so these comparisons may
+  // be leaked.
+  if (a->neg || constant_time_declassify_int(BN_ucmp(a, m) >= 0)) {
     OPENSSL_PUT_ERROR(BN, BN_R_INPUT_NOT_REDUCED);
     return 0;
   }

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/gcd.c

@@ -327,7 +327,10 @@ int BN_mod_inverse_blinded(BIGNUM *out, int *out_no_inverse, const BIGNUM *a,
                            const BN_MONT_CTX *mont, BN_CTX *ctx) {
   *out_no_inverse = 0;
 
-  if (BN_is_negative(a) || BN_cmp(a, &mont->N) >= 0) {
+  // |a| is secret, but it is required to be in range, so these comparisons may
+  // be leaked.
+  if (BN_is_negative(a) ||
+      constant_time_declassify_int(BN_cmp(a, &mont->N) >= 0)) {
     OPENSSL_PUT_ERROR(BN, BN_R_INPUT_NOT_REDUCED);
     return 0;
   }
@@ -336,11 +339,29 @@ int BN_mod_inverse_blinded(BIGNUM *out, int *out_no_inverse, const BIGNUM *a,
   BIGNUM blinding_factor;
   BN_init(&blinding_factor);
 
-  if (!BN_rand_range_ex(&blinding_factor, 1, &mont->N) ||
-      !BN_mod_mul_montgomery(out, &blinding_factor, a, mont, ctx) ||
-      !BN_mod_inverse_odd(out, out_no_inverse, out, &mont->N, ctx) ||
+  // |BN_mod_inverse_odd| is leaky, so generate a secret blinding factor and
+  // blind |a|. This works because (ar)^-1 * r = a^-1, supposing r is
+  // invertible. If r is not invertible, this function will fail. However, we
+  // only use this in RSA, where stumbling on an uninvertible element means
+  // stumbling on the key's factorization. That is, if this function fails, the
+  // RSA key was not actually a product of two large primes.
+  //
+  // TODO(crbug.com/boringssl/677): When the PRNG output is marked secret by
+  // default, the explicit |bn_secret| call can be removed.
+  if (!BN_rand_range_ex(&blinding_factor, 1, &mont->N)) {
+    goto err;
+  }
+  bn_secret(&blinding_factor);
+  if (!BN_mod_mul_montgomery(out, &blinding_factor, a, mont, ctx)) {
+    goto err;
+  }
+
+  // Once blinded, |out| is no longer secret, so it may be passed to a leaky
+  // mod inverse function. Note |blinding_factor| is secret, so |out| will be
+  // secret again after multiplying.
+  bn_declassify(out);
+  if (!BN_mod_inverse_odd(out, out_no_inverse, out, &mont->N, ctx) ||
       !BN_mod_mul_montgomery(out, &blinding_factor, out, mont, ctx)) {
-    OPENSSL_PUT_ERROR(BN, ERR_R_BN_LIB);
     goto err;
   }
 

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/generic.c

@@ -567,37 +567,6 @@ void bn_sqr_comba4(BN_ULONG r[8], const BN_ULONG a[4]) {
 
 #if !defined(BN_ADD_ASM)
 
-// bn_add_with_carry returns |x + y + carry|, and sets |*out_carry| to the
-// carry bit. |carry| must be zero or one.
-static inline BN_ULONG bn_add_with_carry(BN_ULONG x, BN_ULONG y, BN_ULONG carry,
-                                         BN_ULONG *out_carry) {
-  assert(carry == 0 || carry == 1);
-#if defined(BN_ULLONG)
-  BN_ULLONG ret = carry;
-  ret += (BN_ULLONG)x + y;
-  *out_carry = (BN_ULONG)(ret >> BN_BITS2);
-  return (BN_ULONG)ret;
-#else
-  x += carry;
-  carry = x < carry;
-  BN_ULONG ret = x + y;
-  carry += ret < x;
-  *out_carry = carry;
-  return ret;
-#endif
-}
-
-// bn_sub_with_borrow returns |x - y - borrow|, and sets |*out_borrow| to the
-// borrow bit. |borrow| must be zero or one.
-static inline BN_ULONG bn_sub_with_borrow(BN_ULONG x, BN_ULONG y,
-                                          BN_ULONG borrow,
-                                          BN_ULONG *out_borrow) {
-  assert(borrow == 0 || borrow == 1);
-  BN_ULONG ret = x - y - borrow;
-  *out_borrow = (x < y) | ((x == y) & borrow);
-  return ret;
-}
-
 BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
                       size_t n) {
   if (n == 0) {
@@ -606,17 +575,17 @@ BN_ULONG bn_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
 
   BN_ULONG carry = 0;
   while (n & ~3) {
-    r[0] = bn_add_with_carry(a[0], b[0], carry, &carry);
-    r[1] = bn_add_with_carry(a[1], b[1], carry, &carry);
-    r[2] = bn_add_with_carry(a[2], b[2], carry, &carry);
-    r[3] = bn_add_with_carry(a[3], b[3], carry, &carry);
+    r[0] = CRYPTO_addc_w(a[0], b[0], carry, &carry);
+    r[1] = CRYPTO_addc_w(a[1], b[1], carry, &carry);
+    r[2] = CRYPTO_addc_w(a[2], b[2], carry, &carry);
+    r[3] = CRYPTO_addc_w(a[3], b[3], carry, &carry);
     a += 4;
     b += 4;
     r += 4;
     n -= 4;
   }
   while (n) {
-    r[0] = bn_add_with_carry(a[0], b[0], carry, &carry);
+    r[0] = CRYPTO_addc_w(a[0], b[0], carry, &carry);
     a++;
     b++;
     r++;
@@ -633,17 +602,17 @@ BN_ULONG bn_sub_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
 
   BN_ULONG borrow = 0;
   while (n & ~3) {
-    r[0] = bn_sub_with_borrow(a[0], b[0], borrow, &borrow);
-    r[1] = bn_sub_with_borrow(a[1], b[1], borrow, &borrow);
-    r[2] = bn_sub_with_borrow(a[2], b[2], borrow, &borrow);
-    r[3] = bn_sub_with_borrow(a[3], b[3], borrow, &borrow);
+    r[0] = CRYPTO_subc_w(a[0], b[0], borrow, &borrow);
+    r[1] = CRYPTO_subc_w(a[1], b[1], borrow, &borrow);
+    r[2] = CRYPTO_subc_w(a[2], b[2], borrow, &borrow);
+    r[3] = CRYPTO_subc_w(a[3], b[3], borrow, &borrow);
     a += 4;
     b += 4;
     r += 4;
     n -= 4;
   }
   while (n) {
-    r[0] = bn_sub_with_borrow(a[0], b[0], borrow, &borrow);
+    r[0] = CRYPTO_subc_w(a[0], b[0], borrow, &borrow);
     a++;
     b++;
     r++;

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/internal.h

@@ -149,6 +149,7 @@ extern "C" {
 #endif
 
 #define BN_BITS2 64
+#define BN_BITS2_LG 6
 #define BN_BYTES 8
 #define BN_BITS4 32
 #define BN_MASK2 (0xffffffffffffffffUL)
@@ -165,6 +166,7 @@ extern "C" {
 #define BN_ULLONG uint64_t
 #define BN_CAN_DIVIDE_ULLONG
 #define BN_BITS2 32
+#define BN_BITS2_LG 5
 #define BN_BYTES 4
 #define BN_BITS4 16
 #define BN_MASK2 (0xffffffffUL)
@@ -269,6 +271,18 @@ int bn_copy_words(BN_ULONG *out, size_t num, const BIGNUM *bn);
 // validation.
 void bn_assert_fits_in_bytes(const BIGNUM *bn, size_t num);
 
+// bn_secret marks |bn|'s contents, but not its width or sign, as secret. See
+// |CONSTTIME_SECRET| for details.
+OPENSSL_INLINE void bn_secret(BIGNUM *bn) {
+  CONSTTIME_SECRET(bn->d, bn->width * sizeof(BN_ULONG));
+}
+
+// bn_declassify marks |bn|'s value as public. See |CONSTTIME_DECLASSIFY| for
+// details.
+OPENSSL_INLINE void bn_declassify(BIGNUM *bn) {
+  CONSTTIME_DECLASSIFY(bn->d, bn->width * sizeof(BN_ULONG));
+}
+
 // bn_mul_add_words multiples |ap| by |w|, adds the result to |rp|, and places
 // the result in |rp|. |ap| and |rp| must both be |num| words long. It returns
 // the carry word of the operation. |ap| and |rp| may be equal but otherwise may
@@ -386,8 +400,41 @@ int bn_rand_secret_range(BIGNUM *r, int *out_is_uniform, BN_ULONG min_inclusive,
 // inputs.
 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
                 const BN_ULONG *np, const BN_ULONG *n0, size_t num);
+
+#if defined(OPENSSL_X86_64)
+OPENSSL_INLINE int bn_mulx_adx_capable(void) {
+  // MULX is in BMI2.
+  return CRYPTO_is_BMI2_capable() && CRYPTO_is_ADX_capable();
+}
+int bn_mul_mont_nohw(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
+                     const BN_ULONG *np, const BN_ULONG *n0, size_t num);
+OPENSSL_INLINE int bn_mul4x_mont_capable(size_t num) {
+  return num >= 8 && (num & 3) == 0;
+}
+int bn_mul4x_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
+                  const BN_ULONG *np, const BN_ULONG *n0, size_t num);
+OPENSSL_INLINE int bn_mulx4x_mont_capable(size_t num) {
+  return bn_mul4x_mont_capable(num) && bn_mulx_adx_capable();
+}
+int bn_mulx4x_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
+                   const BN_ULONG *np, const BN_ULONG *n0, size_t num);
+OPENSSL_INLINE int bn_sqr8x_mont_capable(size_t num) {
+  return num >= 8 && (num & 7) == 0;
+}
+int bn_sqr8x_mont(BN_ULONG *rp, const BN_ULONG *ap, BN_ULONG mulx_adx_capable,
+                  const BN_ULONG *np, const BN_ULONG *n0, size_t num);
+#elif defined(OPENSSL_ARM)
+OPENSSL_INLINE int bn_mul8x_mont_neon_capable(size_t num) {
+  return (num & 7) == 0 && CRYPTO_is_NEON_capable();
+}
+int bn_mul8x_mont_neon(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
+                       const BN_ULONG *np, const BN_ULONG *n0, size_t num);
+int bn_mul_mont_nohw(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
+                     const BN_ULONG *np, const BN_ULONG *n0, size_t num);
 #endif
 
+#endif  // OPENSSL_BN_ASM_MONT
+
 #if !defined(OPENSSL_NO_ASM) && defined(OPENSSL_X86_64)
 #define OPENSSL_BN_ASM_MONT5
 
@@ -754,8 +801,8 @@ void bn_mod_inverse0_prime_mont_small(BN_ULONG *r, const BN_ULONG *a,
 
 // bn_big_endian_to_words interprets |in_len| bytes from |in| as a big-endian,
 // unsigned integer and writes the result to |out_len| words in |out|. |out_len|
-// must be large enough to represent any |in_len|-byte value. That is, |out_len|
-// must be at least |BN_BYTES * in_len|.
+// must be large enough to represent any |in_len|-byte value. That is, |in_len|
+// must be at most |BN_BYTES * out_len|.
 void bn_big_endian_to_words(BN_ULONG *out, size_t out_len, const uint8_t *in,
                             size_t in_len);
 

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/montgomery.c

@@ -504,3 +504,29 @@ void bn_mod_mul_montgomery_small(BN_ULONG *r, const BN_ULONG *a,
   }
   OPENSSL_cleanse(tmp, 2 * num * sizeof(BN_ULONG));
 }
+
+#if defined(OPENSSL_BN_ASM_MONT) && defined(OPENSSL_X86_64)
+int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
+                const BN_ULONG *np, const BN_ULONG *n0, size_t num) {
+  if (ap == bp && bn_sqr8x_mont_capable(num)) {
+    return bn_sqr8x_mont(rp, ap, bn_mulx_adx_capable(), np, n0, num);
+  }
+  if (bn_mulx4x_mont_capable(num)) {
+    return bn_mulx4x_mont(rp, ap, bp, np, n0, num);
+  }
+  if (bn_mul4x_mont_capable(num)) {
+    return bn_mul4x_mont(rp, ap, bp, np, n0, num);
+  }
+  return bn_mul_mont_nohw(rp, ap, bp, np, n0, num);
+}
+#endif
+
+#if defined(OPENSSL_BN_ASM_MONT) && defined(OPENSSL_ARM)
+int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
+                const BN_ULONG *np, const BN_ULONG *n0, size_t num) {
+  if (bn_mul8x_mont_neon_capable(num)) {
+    return bn_mul8x_mont_neon(rp, ap, bp, np, n0, num);
+  }
+  return bn_mul_mont_nohw(rp, ap, bp, np, n0, num);
+}
+#endif

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/montgomery_inv.c

@@ -179,42 +179,43 @@ int bn_mont_ctx_set_RR_consttime(BN_MONT_CTX *mont, BN_CTX *ctx) {
   // Montgomery domain, 2R or 2^(lgBigR+1), and then use Montgomery
   // square-and-multiply to exponentiate.
   //
-  // The multiply steps take 2^n R to 2^(n+1) R. It is faster to double
-  // the value instead. The square steps take 2^n R to 2^(2n) R. This is
-  // equivalent to doubling n times. When n is below some threshold, doubling is
-  // faster. When above, squaring is faster.
+  // The square steps take 2^n R to (2^n)*(2^n) R = 2^2n R. This is the same as
+  // doubling 2^n R, n times (doubling any x, n times, computes 2^n * x). When n
+  // is below some threshold, doubling is faster; when above, squaring is
+  // faster. From benchmarking various 32-bit and 64-bit architectures, the word
+  // count seems to work well as a threshold. (Doubling scales linearly and
+  // Montgomery reduction scales quadratically, so the threshold should scale
+  // roughly linearly.)
   //
-  // We double to this threshold, then switch to Montgomery squaring. From
-  // benchmarking various 32-bit and 64-bit architectures, the word count seems
-  // to work well as a threshold. (Doubling scales linearly and Montgomery
-  // reduction scales quadratically, so the threshold should scale roughly
-  // linearly.)
-  unsigned threshold = mont->N.width;
-  unsigned iters;
-  for (iters = 0; iters < sizeof(lgBigR) * 8; iters++) {
-    if ((lgBigR >> iters) <= threshold) {
-      break;
-    }
-  }
-
-  // Compute 2^(lgBigR >> iters) R, or 2^((lgBigR >> iters) + lgBigR), by
-  // doubling. The first n_bits - 1 doubles can be skipped because we don't need
-  // to reduce.
+  // The multiply steps take 2^n R to 2*2^n R = 2^(n+1) R. It is faster to
+  // double the value instead, so the square-and-multiply exponentiation would
+  // become square-and-double. However, when using the word count as the
+  // threshold, it turns out that no multiply/double steps will be needed at
+  // all, because squaring any x, i times, computes x^(2^i):
+  //
+  //   (2^threshold)^(2^BN_BITS2_LG) R
+  //   (2^mont->N.width)^BN_BITS2 R
+  // = 2^(mont->N.width*BN_BITS2) R
+  // = 2^lgBigR R
+  // = RR
+  int threshold = mont->N.width;
+
+  // Calculate 2^threshold R = 2^(threshold + lgBigR) by doubling. The
+  // first n_bits - 1 doubles can be skipped because we don't need to reduce.
   if (!BN_set_bit(&mont->RR, n_bits - 1) ||
       !bn_mod_lshift_consttime(&mont->RR, &mont->RR,
-                               (lgBigR >> iters) + lgBigR - (n_bits - 1),
+                               threshold + (lgBigR - (n_bits - 1)),
                                &mont->N, ctx)) {
     return 0;
   }
 
-  for (unsigned i = iters - 1; i < iters; i--) {
+  // The above steps are the same regardless of the threshold. The steps below
+  // need to be modified if the threshold changes.
+  assert(threshold == mont->N.width);
+  for (unsigned i = 0; i < BN_BITS2_LG; i++) {
     if (!BN_mod_mul_montgomery(&mont->RR, &mont->RR, &mont->RR, mont, ctx)) {
       return 0;
     }
-    if ((lgBigR & (1u << i)) != 0 &&
-        !bn_mod_lshift1_consttime(&mont->RR, &mont->RR, &mont->N, ctx)) {
-      return 0;
-    }
   }
 
   return bn_resize_words(&mont->RR, mont->N.width);

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/mul.c

@@ -143,17 +143,13 @@ static BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a,
     // in |a| were zeros.
     dl = -dl;
     for (int i = 0; i < dl; i++) {
-      r[i] = 0u - b[i] - borrow;
-      borrow |= r[i] != 0;
+      r[i] = CRYPTO_subc_w(0, b[i], borrow, &borrow);
     }
   } else {
     // |b| is shorter than |a|. Complete the subtraction as if the excess words
     // in |b| were zeros.
     for (int i = 0; i < dl; i++) {
-      // |r| and |a| may alias, so use a temporary.
-      BN_ULONG tmp = a[i];
-      r[i] = a[i] - borrow;
-      borrow = tmp < r[i];
+      r[i] = CRYPTO_subc_w(a[i], 0, borrow, &borrow);
     }
   }
 

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/prime.c

@@ -359,14 +359,7 @@ static int probable_prime_dh(BIGNUM *rnd, int bits, const BIGNUM *add,
 static int probable_prime_dh_safe(BIGNUM *rnd, int bits, const BIGNUM *add,
                                   const BIGNUM *rem, BN_CTX *ctx);
 
-BN_GENCB *BN_GENCB_new(void) {
-  BN_GENCB *callback = OPENSSL_malloc(sizeof(BN_GENCB));
-  if (callback == NULL) {
-    return NULL;
-  }
-  OPENSSL_memset(callback, 0, sizeof(BN_GENCB));
-  return callback;
-}
+BN_GENCB *BN_GENCB_new(void) { return OPENSSL_zalloc(sizeof(BN_GENCB)); }
 
 void BN_GENCB_free(BN_GENCB *callback) { OPENSSL_free(callback); }
 

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/bn/random.c

@@ -281,8 +281,14 @@ int bn_rand_range_words(BN_ULONG *out, BN_ULONG min_inclusive,
     out[words - 1] &= mask;
 
     // If out >= max_exclusive or out < min_inclusive, retry. This implements
-    // the equivalent of steps 6 and 7 without leaking the value of |out|.
-  } while (!bn_in_range_words(out, min_inclusive, max_exclusive, words));
+    // the equivalent of steps 6 and 7 without leaking the value of |out|. The
+    // result of this comparison may be treated as public. It only reveals how
+    // many attempts were needed before we found a value in range. This is
+    // independent of the final secret output, and has a distribution that
+    // depends only on |min_inclusive| and |max_exclusive|, both of which are
+    // public.
+  } while (!constant_time_declassify_int(
+      bn_in_range_words(out, min_inclusive, max_exclusive, words)));
   return 1;
 }
 

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/cipher/cipher.c

@@ -113,12 +113,11 @@ int EVP_CIPHER_CTX_copy(EVP_CIPHER_CTX *out, const EVP_CIPHER_CTX *in) {
   OPENSSL_memcpy(out, in, sizeof(EVP_CIPHER_CTX));
 
   if (in->cipher_data && in->cipher->ctx_size) {
-    out->cipher_data = OPENSSL_malloc(in->cipher->ctx_size);
+    out->cipher_data = OPENSSL_memdup(in->cipher_data, in->cipher->ctx_size);
     if (!out->cipher_data) {
       out->cipher = NULL;
       return 0;
     }
-    OPENSSL_memcpy(out->cipher_data, in->cipher_data, in->cipher->ctx_size);
   }
 
   if (in->cipher->flags & EVP_CIPH_CUSTOM_COPY) {
@@ -586,6 +585,16 @@ unsigned EVP_CIPHER_CTX_key_length(const EVP_CIPHER_CTX *ctx) {
 }
 
 unsigned EVP_CIPHER_CTX_iv_length(const EVP_CIPHER_CTX *ctx) {
+  if (EVP_CIPHER_mode(ctx->cipher) == EVP_CIPH_GCM_MODE) {
+    int length;
+    int res = EVP_CIPHER_CTX_ctrl((EVP_CIPHER_CTX *)ctx, EVP_CTRL_GET_IVLEN, 0,
+                                  &length);
+    // EVP_CIPHER_CTX_ctrl returning an error should be impossible under this
+    // circumstance. If it somehow did, fallback to the static cipher iv_len.
+    if (res == 1) {
+      return length;
+    }
+  }
   return ctx->cipher->iv_len;
 }
 

data/third_party/boringssl-with-bazel/src/crypto/fipsmodule/cipher/e_aes.c

@@ -408,22 +408,6 @@ static void aes_gcm_cleanup(EVP_CIPHER_CTX *c) {
   }
 }
 
-// increment counter (64-bit int) by 1
-static void ctr64_inc(uint8_t *counter) {
-  int n = 8;
-  uint8_t c;
-
-  do {
-    --n;
-    c = counter[n];
-    ++c;
-    counter[n] = c;
-    if (c) {
-      return;
-    }
-  } while (n);
-}
-
 static int aes_gcm_ctrl(EVP_CIPHER_CTX *c, int type, int arg, void *ptr) {
   EVP_AES_GCM_CTX *gctx = aes_gcm_from_cipher_ctx(c);
   switch (type) {
@@ -454,6 +438,10 @@ static int aes_gcm_ctrl(EVP_CIPHER_CTX *c, int type, int arg, void *ptr) {
       gctx->ivlen = arg;
       return 1;
 
+    case EVP_CTRL_GET_IVLEN:
+      *(int *)ptr = gctx->ivlen;
+      return 1;
+
     case EVP_CTRL_AEAD_SET_TAG:
       if (arg <= 0 || arg > 16 || c->encrypt) {
         return 0;
@@ -481,9 +469,7 @@ static int aes_gcm_ctrl(EVP_CIPHER_CTX *c, int type, int arg, void *ptr) {
       if (arg < 4 || (gctx->ivlen - arg) < 8) {
         return 0;
       }
-      if (arg) {
-        OPENSSL_memcpy(gctx->iv, ptr, arg);
-      }
+      OPENSSL_memcpy(gctx->iv, ptr, arg);
       if (c->encrypt) {
         // |RAND_bytes| calls within the fipsmodule should be wrapped with state
         // lock functions to avoid updating the service indicator with the DRBG
@@ -495,7 +481,7 @@ static int aes_gcm_ctrl(EVP_CIPHER_CTX *c, int type, int arg, void *ptr) {
       gctx->iv_gen = 1;
       return 1;
 
-    case EVP_CTRL_GCM_IV_GEN:
+    case EVP_CTRL_GCM_IV_GEN: {
       if (gctx->iv_gen == 0 || gctx->key_set == 0) {
         return 0;
       }
@@ -504,12 +490,13 @@ static int aes_gcm_ctrl(EVP_CIPHER_CTX *c, int type, int arg, void *ptr) {
         arg = gctx->ivlen;
       }
       OPENSSL_memcpy(ptr, gctx->iv + gctx->ivlen - arg, arg);
-      // Invocation field will be at least 8 bytes in size and
-      // so no need to check wrap around or increment more than
-      // last 8 bytes.
-      ctr64_inc(gctx->iv + gctx->ivlen - 8);
+      // Invocation field will be at least 8 bytes in size, so no need to check
+      // wrap around or increment more than last 8 bytes.
+      uint8_t *ctr = gctx->iv + gctx->ivlen - 8;
+      CRYPTO_store_u64_be(ctr, CRYPTO_load_u64_be(ctr) + 1);
       gctx->iv_set = 1;
       return 1;
+    }
 
     case EVP_CTRL_GCM_SET_IV_INV:
       if (gctx->iv_gen == 0 || gctx->key_set == 0 || c->encrypt) {