grpc 1.24.0 → 1.25.0.pre1

data/third_party/boringssl/crypto/cipher_extra/e_aesccm.c

@@ -1,21 +1,56 @@
-/* Copyright (c) 2018, Google Inc.
+/* ====================================================================
+ * Copyright (c) 2008 The OpenSSL Project.  All rights reserved.
  *
- * Permission to use, copy, modify, and/or distribute this software for any
- * purpose with or without fee is hereby granted, provided that the above
- * copyright notice and this permission notice appear in all copies.
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
  *
- * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
- * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
- * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
- * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
- * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
- * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * 3. All advertising materials mentioning features or use of this
+ *    software must display the following acknowledgment:
+ *    "This product includes software developed by the OpenSSL Project
+ *    for use in the OpenSSL Toolkit. (http://www.openssl.org/)"
+ *
+ * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
+ *    endorse or promote products derived from this software without
+ *    prior written permission. For written permission, please contact
+ *    openssl-core@openssl.org.
+ *
+ * 5. Products derived from this software may not be called "OpenSSL"
+ *    nor may "OpenSSL" appear in their names without prior written
+ *    permission of the OpenSSL Project.
+ *
+ * 6. Redistributions of any form whatsoever must retain the following
+ *    acknowledgment:
+ *    "This product includes software developed by the OpenSSL Project
+ *    for use in the OpenSSL Toolkit (http://www.openssl.org/)"
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
+ * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE OpenSSL PROJECT OR
+ * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ * ==================================================================== */
 
 #include <openssl/aead.h>
 
 #include <assert.h>
 
+#include <openssl/cpu.h>
 #include <openssl/cipher.h>
 #include <openssl/err.h>
 #include <openssl/mem.h>
@@ -23,6 +58,213 @@
 #include "../fipsmodule/cipher/internal.h"
 
 
+struct ccm128_context {
+  block128_f block;
+  ctr128_f ctr;
+  unsigned M, L;
+};
+
+struct ccm128_state {
+  union {
+    uint64_t u[2];
+    uint8_t c[16];
+  } nonce, cmac;
+};
+
+static int CRYPTO_ccm128_init(struct ccm128_context *ctx, const AES_KEY *key,
+                              block128_f block, ctr128_f ctr, unsigned M,
+                              unsigned L) {
+  if (M < 4 || M > 16 || (M & 1) != 0 || L < 2 || L > 8) {
+    return 0;
+  }
+  ctx->block = block;
+  ctx->ctr = ctr;
+  ctx->M = M;
+  ctx->L = L;
+  return 1;
+}
+
+static size_t CRYPTO_ccm128_max_input(const struct ccm128_context *ctx) {
+  return ctx->L >= sizeof(size_t) ? (size_t)-1
+                                  : (((size_t)1) << (ctx->L * 8)) - 1;
+}
+
+static int ccm128_init_state(const struct ccm128_context *ctx,
+                             struct ccm128_state *state, const AES_KEY *key,
+                             const uint8_t *nonce, size_t nonce_len,
+                             const uint8_t *aad, size_t aad_len,
+                             size_t plaintext_len) {
+  const block128_f block = ctx->block;
+  const unsigned M = ctx->M;
+  const unsigned L = ctx->L;
+
+  // |L| determines the expected |nonce_len| and the limit for |plaintext_len|.
+  if (plaintext_len > CRYPTO_ccm128_max_input(ctx) ||
+      nonce_len != 15 - L) {
+    return 0;
+  }
+
+  // Assemble the first block for computing the MAC.
+  OPENSSL_memset(state, 0, sizeof(*state));
+  state->nonce.c[0] = (uint8_t)((L - 1) | ((M - 2) / 2) << 3);
+  if (aad_len != 0) {
+    state->nonce.c[0] |= 0x40;  // Set AAD Flag
+  }
+  OPENSSL_memcpy(&state->nonce.c[1], nonce, nonce_len);
+  for (unsigned i = 0; i < L; i++) {
+    state->nonce.c[15 - i] = (uint8_t)(plaintext_len >> (8 * i));
+  }
+
+  (*block)(state->nonce.c, state->cmac.c, key);
+  size_t blocks = 1;
+
+  if (aad_len != 0) {
+    unsigned i;
+    // Cast to u64 to avoid the compiler complaining about invalid shifts.
+    uint64_t aad_len_u64 = aad_len;
+    if (aad_len_u64 < 0x10000 - 0x100) {
+      state->cmac.c[0] ^= (uint8_t)(aad_len_u64 >> 8);
+      state->cmac.c[1] ^= (uint8_t)aad_len_u64;
+      i = 2;
+    } else if (aad_len_u64 <= 0xffffffff) {
+      state->cmac.c[0] ^= 0xff;
+      state->cmac.c[1] ^= 0xfe;
+      state->cmac.c[2] ^= (uint8_t)(aad_len_u64 >> 24);
+      state->cmac.c[3] ^= (uint8_t)(aad_len_u64 >> 16);
+      state->cmac.c[4] ^= (uint8_t)(aad_len_u64 >> 8);
+      state->cmac.c[5] ^= (uint8_t)aad_len_u64;
+      i = 6;
+    } else {
+      state->cmac.c[0] ^= 0xff;
+      state->cmac.c[1] ^= 0xff;
+      state->cmac.c[2] ^= (uint8_t)(aad_len_u64 >> 56);
+      state->cmac.c[3] ^= (uint8_t)(aad_len_u64 >> 48);
+      state->cmac.c[4] ^= (uint8_t)(aad_len_u64 >> 40);
+      state->cmac.c[5] ^= (uint8_t)(aad_len_u64 >> 32);
+      state->cmac.c[6] ^= (uint8_t)(aad_len_u64 >> 24);
+      state->cmac.c[7] ^= (uint8_t)(aad_len_u64 >> 16);
+      state->cmac.c[8] ^= (uint8_t)(aad_len_u64 >> 8);
+      state->cmac.c[9] ^= (uint8_t)aad_len_u64;
+      i = 10;
+    }
+
+    do {
+      for (; i < 16 && aad_len != 0; i++) {
+        state->cmac.c[i] ^= *aad;
+        aad++;
+        aad_len--;
+      }
+      (*block)(state->cmac.c, state->cmac.c, key);
+      blocks++;
+      i = 0;
+    } while (aad_len != 0);
+  }
+
+  // Per RFC 3610, section 2.6, the total number of block cipher operations done
+  // must not exceed 2^61. There are two block cipher operations remaining per
+  // message block, plus one block at the end to encrypt the MAC.
+  size_t remaining_blocks = 2 * ((plaintext_len + 15) / 16) + 1;
+  if (plaintext_len + 15 < plaintext_len ||
+      remaining_blocks + blocks < blocks ||
+      (uint64_t) remaining_blocks + blocks > UINT64_C(1) << 61) {
+    return 0;
+  }
+
+  // Assemble the first block for encrypting and decrypting. The bottom |L|
+  // bytes are replaced with a counter and all bit the encoding of |L| is
+  // cleared in the first byte.
+  state->nonce.c[0] &= 7;
+  return 1;
+}
+
+static int ccm128_encrypt(const struct ccm128_context *ctx,
+                          struct ccm128_state *state, const AES_KEY *key,
+                          uint8_t *out, const uint8_t *in, size_t len) {
+  // The counter for encryption begins at one.
+  for (unsigned i = 0; i < ctx->L; i++) {
+    state->nonce.c[15 - i] = 0;
+  }
+  state->nonce.c[15] = 1;
+
+  uint8_t partial_buf[16];
+  unsigned num = 0;
+  if (ctx->ctr != NULL) {
+    CRYPTO_ctr128_encrypt_ctr32(in, out, len, key, state->nonce.c, partial_buf,
+                                &num, ctx->ctr);
+  } else {
+    CRYPTO_ctr128_encrypt(in, out, len, key, state->nonce.c, partial_buf, &num,
+                          ctx->block);
+  }
+  return 1;
+}
+
+static int ccm128_compute_mac(const struct ccm128_context *ctx,
+                              struct ccm128_state *state, const AES_KEY *key,
+                              uint8_t *out_tag, size_t tag_len,
+                              const uint8_t *in, size_t len) {
+  block128_f block = ctx->block;
+  if (tag_len != ctx->M) {
+    return 0;
+  }
+
+  // Incorporate |in| into the MAC.
+  union {
+    uint64_t u[2];
+    uint8_t c[16];
+  } tmp;
+  while (len >= 16) {
+    OPENSSL_memcpy(tmp.c, in, 16);
+    state->cmac.u[0] ^= tmp.u[0];
+    state->cmac.u[1] ^= tmp.u[1];
+    (*block)(state->cmac.c, state->cmac.c, key);
+    in += 16;
+    len -= 16;
+  }
+  if (len > 0) {
+    for (size_t i = 0; i < len; i++) {
+      state->cmac.c[i] ^= in[i];
+    }
+    (*block)(state->cmac.c, state->cmac.c, key);
+  }
+
+  // Encrypt the MAC with counter zero.
+  for (unsigned i = 0; i < ctx->L; i++) {
+    state->nonce.c[15 - i] = 0;
+  }
+  (*block)(state->nonce.c, tmp.c, key);
+  state->cmac.u[0] ^= tmp.u[0];
+  state->cmac.u[1] ^= tmp.u[1];
+
+  OPENSSL_memcpy(out_tag, state->cmac.c, tag_len);
+  return 1;
+}
+
+static int CRYPTO_ccm128_encrypt(const struct ccm128_context *ctx,
+                                 const AES_KEY *key, uint8_t *out,
+                                 uint8_t *out_tag, size_t tag_len,
+                                 const uint8_t *nonce, size_t nonce_len,
+                                 const uint8_t *in, size_t len,
+                                 const uint8_t *aad, size_t aad_len) {
+  struct ccm128_state state;
+  return ccm128_init_state(ctx, &state, key, nonce, nonce_len, aad, aad_len,
+                           len) &&
+         ccm128_compute_mac(ctx, &state, key, out_tag, tag_len, in, len) &&
+         ccm128_encrypt(ctx, &state, key, out, in, len);
+}
+
+static int CRYPTO_ccm128_decrypt(const struct ccm128_context *ctx,
+                                 const AES_KEY *key, uint8_t *out,
+                                 uint8_t *out_tag, size_t tag_len,
+                                 const uint8_t *nonce, size_t nonce_len,
+                                 const uint8_t *in, size_t len,
+                                 const uint8_t *aad, size_t aad_len) {
+  struct ccm128_state state;
+  return ccm128_init_state(ctx, &state, key, nonce, nonce_len, aad, aad_len,
+                           len) &&
+         ccm128_encrypt(ctx, &state, key, out, in, len) &&
+         ccm128_compute_mac(ctx, &state, key, out_tag, tag_len, out, len);
+}
+
 #define EVP_AEAD_AES_CCM_MAX_TAG_LEN 16
 
 struct aead_aes_ccm_ctx {
@@ -30,9 +272,18 @@ struct aead_aes_ccm_ctx {
     double align;
     AES_KEY ks;
   } ks;
-  CCM128_CONTEXT ccm;
+  struct ccm128_context ccm;
 };
 
+OPENSSL_STATIC_ASSERT(sizeof(((EVP_AEAD_CTX *)NULL)->state) >=
+                          sizeof(struct aead_aes_ccm_ctx),
+                      "AEAD state is too small");
+#if defined(__GNUC__) || defined(__clang__)
+OPENSSL_STATIC_ASSERT(alignof(union evp_aead_ctx_st_state) >=
+                          alignof(struct aead_aes_ccm_ctx),
+                      "AEAD state has insufficient alignment");
+#endif
+
 static int aead_aes_ccm_init(EVP_AEAD_CTX *ctx, const uint8_t *key,
                              size_t key_len, size_t tag_len, unsigned M,
                              unsigned L) {
@@ -54,36 +305,28 @@ static int aead_aes_ccm_init(EVP_AEAD_CTX *ctx, const uint8_t *key,
     return 0;
   }
 
-  struct aead_aes_ccm_ctx *ccm_ctx =
-      OPENSSL_malloc(sizeof(struct aead_aes_ccm_ctx));
-  if (ccm_ctx == NULL) {
-    OPENSSL_PUT_ERROR(CIPHER, ERR_R_MALLOC_FAILURE);
-    return 0;
-  }
+  struct aead_aes_ccm_ctx *ccm_ctx = (struct aead_aes_ccm_ctx *)&ctx->state;
 
   block128_f block;
   ctr128_f ctr = aes_ctr_set_key(&ccm_ctx->ks.ks, NULL, &block, key, key_len);
   ctx->tag_len = tag_len;
   if (!CRYPTO_ccm128_init(&ccm_ctx->ccm, &ccm_ctx->ks.ks, block, ctr, M, L)) {
     OPENSSL_PUT_ERROR(CIPHER, ERR_R_INTERNAL_ERROR);
-    OPENSSL_free(ccm_ctx);
     return 0;
   }
 
-  ctx->aead_state = ccm_ctx;
   return 1;
 }
 
-static void aead_aes_ccm_cleanup(EVP_AEAD_CTX *ctx) {
-  OPENSSL_free(ctx->aead_state);
-}
+static void aead_aes_ccm_cleanup(EVP_AEAD_CTX *ctx) {}
 
 static int aead_aes_ccm_seal_scatter(
     const EVP_AEAD_CTX *ctx, uint8_t *out, uint8_t *out_tag,
     size_t *out_tag_len, size_t max_out_tag_len, const uint8_t *nonce,
     size_t nonce_len, const uint8_t *in, size_t in_len, const uint8_t *extra_in,
     size_t extra_in_len, const uint8_t *ad, size_t ad_len) {
-  const struct aead_aes_ccm_ctx *ccm_ctx = ctx->aead_state;
+  const struct aead_aes_ccm_ctx *ccm_ctx =
+      (struct aead_aes_ccm_ctx *)&ctx->state;
 
   if (in_len > CRYPTO_ccm128_max_input(&ccm_ctx->ccm)) {
     OPENSSL_PUT_ERROR(CIPHER, CIPHER_R_TOO_LARGE);
@@ -116,7 +359,8 @@ static int aead_aes_ccm_open_gather(const EVP_AEAD_CTX *ctx, uint8_t *out,
                                     const uint8_t *in, size_t in_len,
                                     const uint8_t *in_tag, size_t in_tag_len,
                                     const uint8_t *ad, size_t ad_len) {
-  const struct aead_aes_ccm_ctx *ccm_ctx = ctx->aead_state;
+  const struct aead_aes_ccm_ctx *ccm_ctx =
+      (struct aead_aes_ccm_ctx *)&ctx->state;
 
   if (in_len > CRYPTO_ccm128_max_input(&ccm_ctx->ccm)) {
     OPENSSL_PUT_ERROR(CIPHER, CIPHER_R_TOO_LARGE);

data/third_party/boringssl/crypto/cipher_extra/e_aesctrhmac.c

@@ -35,6 +35,15 @@ struct aead_aes_ctr_hmac_sha256_ctx {
   SHA256_CTX outer_init_state;
 };
 
+OPENSSL_STATIC_ASSERT(sizeof(((EVP_AEAD_CTX *)NULL)->state) >=
+                          sizeof(struct aead_aes_ctr_hmac_sha256_ctx),
+                      "AEAD state is too small");
+#if defined(__GNUC__) || defined(__clang__)
+OPENSSL_STATIC_ASSERT(alignof(union evp_aead_ctx_st_state) >=
+                          alignof(struct aead_aes_ctr_hmac_sha256_ctx),
+                      "AEAD state has insufficient alignment");
+#endif
+
 static void hmac_init(SHA256_CTX *out_inner, SHA256_CTX *out_outer,
                       const uint8_t hmac_key[32]) {
   static const size_t hmac_key_len = 32;
@@ -61,7 +70,8 @@ static void hmac_init(SHA256_CTX *out_inner, SHA256_CTX *out_outer,
 
 static int aead_aes_ctr_hmac_sha256_init(EVP_AEAD_CTX *ctx, const uint8_t *key,
                                          size_t key_len, size_t tag_len) {
-  struct aead_aes_ctr_hmac_sha256_ctx *aes_ctx;
+  struct aead_aes_ctr_hmac_sha256_ctx *aes_ctx =
+      (struct aead_aes_ctr_hmac_sha256_ctx *)&ctx->state;
   static const size_t hmac_key_len = 32;
 
   if (key_len < hmac_key_len) {
@@ -84,26 +94,16 @@ static int aead_aes_ctr_hmac_sha256_init(EVP_AEAD_CTX *ctx, const uint8_t *key,
     return 0;
   }
 
-  aes_ctx = OPENSSL_malloc(sizeof(struct aead_aes_ctr_hmac_sha256_ctx));
-  if (aes_ctx == NULL) {
-    OPENSSL_PUT_ERROR(CIPHER, ERR_R_MALLOC_FAILURE);
-    return 0;
-  }
-
   aes_ctx->ctr =
       aes_ctr_set_key(&aes_ctx->ks.ks, NULL, &aes_ctx->block, key, aes_key_len);
   ctx->tag_len = tag_len;
   hmac_init(&aes_ctx->inner_init_state, &aes_ctx->outer_init_state,
             key + aes_key_len);
 
-  ctx->aead_state = aes_ctx;
-
   return 1;
 }
 
-static void aead_aes_ctr_hmac_sha256_cleanup(EVP_AEAD_CTX *ctx) {
-  OPENSSL_free(ctx->aead_state);
-}
+static void aead_aes_ctr_hmac_sha256_cleanup(EVP_AEAD_CTX *ctx) {}
 
 static void hmac_update_uint64(SHA256_CTX *sha256, uint64_t value) {
   unsigned i;
@@ -178,7 +178,8 @@ static int aead_aes_ctr_hmac_sha256_seal_scatter(
     size_t *out_tag_len, size_t max_out_tag_len, const uint8_t *nonce,
     size_t nonce_len, const uint8_t *in, size_t in_len, const uint8_t *extra_in,
     size_t extra_in_len, const uint8_t *ad, size_t ad_len) {
-  const struct aead_aes_ctr_hmac_sha256_ctx *aes_ctx = ctx->aead_state;
+  const struct aead_aes_ctr_hmac_sha256_ctx *aes_ctx =
+      (struct aead_aes_ctr_hmac_sha256_ctx *) &ctx->state;
   const uint64_t in_len_64 = in_len;
 
   if (in_len_64 >= (UINT64_C(1) << 32) * AES_BLOCK_SIZE) {
@@ -212,7 +213,8 @@ static int aead_aes_ctr_hmac_sha256_open_gather(
     const EVP_AEAD_CTX *ctx, uint8_t *out, const uint8_t *nonce,
     size_t nonce_len, const uint8_t *in, size_t in_len, const uint8_t *in_tag,
     size_t in_tag_len, const uint8_t *ad, size_t ad_len) {
-  const struct aead_aes_ctr_hmac_sha256_ctx *aes_ctx = ctx->aead_state;
+  const struct aead_aes_ctr_hmac_sha256_ctx *aes_ctx =
+      (struct aead_aes_ctr_hmac_sha256_ctx *) &ctx->state;
 
   if (in_tag_len != ctx->tag_len) {
     OPENSSL_PUT_ERROR(CIPHER, CIPHER_R_BAD_DECRYPT);

data/third_party/boringssl/crypto/cipher_extra/e_aesgcmsiv.c

@@ -27,28 +27,47 @@
 #define EVP_AEAD_AES_GCM_SIV_NONCE_LEN 12
 #define EVP_AEAD_AES_GCM_SIV_TAG_LEN 16
 
-#if defined(OPENSSL_X86_64) && !defined(OPENSSL_NO_ASM)
+// TODO(davidben): AES-GCM-SIV assembly is not correct for Windows. It must save
+// and restore xmm6 through xmm15.
+#if defined(OPENSSL_X86_64) && !defined(OPENSSL_NO_ASM) && \
+    !defined(OPENSSL_WINDOWS)
+#define AES_GCM_SIV_ASM
 
 // Optimised AES-GCM-SIV
 
 struct aead_aes_gcm_siv_asm_ctx {
   alignas(16) uint8_t key[16*15];
   int is_128_bit;
-  // ptr contains the original pointer from |OPENSSL_malloc|, which may only be
-  // 8-byte aligned. When freeing this structure, actually call |OPENSSL_free|
-  // on this pointer.
-  void *ptr;
 };
 
+// The assembly code assumes 8-byte alignment of the EVP_AEAD_CTX's state, and
+// aligns to 16 bytes itself.
+OPENSSL_STATIC_ASSERT(sizeof(((EVP_AEAD_CTX *)NULL)->state) + 8 >=
+                          sizeof(struct aead_aes_gcm_siv_asm_ctx),
+                      "AEAD state is too small");
+#if defined(__GNUC__) || defined(__clang__)
+OPENSSL_STATIC_ASSERT(alignof(union evp_aead_ctx_st_state) >= 8,
+                      "AEAD state has insufficient alignment");
+#endif
+
+// asm_ctx_from_ctx returns a 16-byte aligned context pointer from |ctx|.
+static struct aead_aes_gcm_siv_asm_ctx *asm_ctx_from_ctx(
+    const EVP_AEAD_CTX *ctx) {
+  // ctx->state must already be 8-byte aligned. Thus, at most, we may need to
+  // add eight to align it to 16 bytes.
+  const uintptr_t offset = ((uintptr_t)&ctx->state) & 8;
+  return (struct aead_aes_gcm_siv_asm_ctx *)(&ctx->state.opaque[offset]);
+}
+
 // aes128gcmsiv_aes_ks writes an AES-128 key schedule for |key| to
 // |out_expanded_key|.
 extern void aes128gcmsiv_aes_ks(
     const uint8_t key[16], uint8_t out_expanded_key[16*15]);
 
-// aes128gcmsiv_aes_ks writes an AES-128 key schedule for |key| to
+// aes256gcmsiv_aes_ks writes an AES-256 key schedule for |key| to
 // |out_expanded_key|.
 extern void aes256gcmsiv_aes_ks(
-    const uint8_t key[16], uint8_t out_expanded_key[16*15]);
+    const uint8_t key[32], uint8_t out_expanded_key[16*15]);
 
 static int aead_aes_gcm_siv_asm_init(EVP_AEAD_CTX *ctx, const uint8_t *key,
                                      size_t key_len, size_t tag_len) {
@@ -68,18 +87,8 @@ static int aead_aes_gcm_siv_asm_init(EVP_AEAD_CTX *ctx, const uint8_t *key,
     return 0;
   }
 
-  char *ptr = OPENSSL_malloc(sizeof(struct aead_aes_gcm_siv_asm_ctx) + 8);
-  if (ptr == NULL) {
-    return 0;
-  }
-  assert((((uintptr_t)ptr) & 7) == 0);
-
-  // gcm_siv_ctx needs to be 16-byte aligned in a cross-platform way.
-  struct aead_aes_gcm_siv_asm_ctx *gcm_siv_ctx =
-      (struct aead_aes_gcm_siv_asm_ctx *)(ptr + (((uintptr_t)ptr) & 8));
-
+  struct aead_aes_gcm_siv_asm_ctx *gcm_siv_ctx = asm_ctx_from_ctx(ctx);
   assert((((uintptr_t)gcm_siv_ctx) & 15) == 0);
-  gcm_siv_ctx->ptr = ptr;
 
   if (key_bits == 128) {
     aes128gcmsiv_aes_ks(key, &gcm_siv_ctx->key[0]);
@@ -88,16 +97,13 @@ static int aead_aes_gcm_siv_asm_init(EVP_AEAD_CTX *ctx, const uint8_t *key,
     aes256gcmsiv_aes_ks(key, &gcm_siv_ctx->key[0]);
     gcm_siv_ctx->is_128_bit = 0;
   }
-  ctx->aead_state = gcm_siv_ctx;
+
   ctx->tag_len = tag_len;
 
   return 1;
 }
 
-static void aead_aes_gcm_siv_asm_cleanup(EVP_AEAD_CTX *ctx) {
-  const struct aead_aes_gcm_siv_asm_ctx *gcm_siv_ctx = ctx->aead_state;
-  OPENSSL_free(gcm_siv_ctx->ptr);
-}
+static void aead_aes_gcm_siv_asm_cleanup(EVP_AEAD_CTX *ctx) {}
 
 // aesgcmsiv_polyval_horner updates the POLYVAL value in |in_out_poly| to
 // include a number (|in_blocks|) of 16-byte blocks of data from |in|, given
@@ -337,7 +343,7 @@ static int aead_aes_gcm_siv_asm_seal_scatter(
     size_t *out_tag_len, size_t max_out_tag_len, const uint8_t *nonce,
     size_t nonce_len, const uint8_t *in, size_t in_len, const uint8_t *extra_in,
     size_t extra_in_len, const uint8_t *ad, size_t ad_len) {
-  const struct aead_aes_gcm_siv_asm_ctx *gcm_siv_ctx = ctx->aead_state;
+  const struct aead_aes_gcm_siv_asm_ctx *gcm_siv_ctx = asm_ctx_from_ctx(ctx);
   const uint64_t in_len_64 = in_len;
   const uint64_t ad_len_64 = ad_len;
 
@@ -420,7 +426,12 @@ static int aead_aes_gcm_siv_asm_open(const EVP_AEAD_CTX *ctx, uint8_t *out,
     return 0;
   }
 
-  const struct aead_aes_gcm_siv_asm_ctx *gcm_siv_ctx = ctx->aead_state;
+  if (nonce_len != EVP_AEAD_AES_GCM_SIV_NONCE_LEN) {
+    OPENSSL_PUT_ERROR(CIPHER, CIPHER_R_UNSUPPORTED_NONCE_SIZE);
+    return 0;
+  }
+
+  const struct aead_aes_gcm_siv_asm_ctx *gcm_siv_ctx = asm_ctx_from_ctx(ctx);
   const size_t plaintext_len = in_len - EVP_AEAD_AES_GCM_SIV_TAG_LEN;
   const uint8_t *const given_tag = in + plaintext_len;
 
@@ -547,7 +558,7 @@ static const EVP_AEAD aead_aes_256_gcm_siv_asm = {
     NULL /* tag_len */,
 };
 
-#endif  // X86_64 && !NO_ASM
+#endif  // X86_64 && !NO_ASM && !WINDOWS
 
 struct aead_aes_gcm_siv_ctx {
   union {
@@ -558,6 +569,15 @@ struct aead_aes_gcm_siv_ctx {
   unsigned is_256:1;
 };
 
+OPENSSL_STATIC_ASSERT(sizeof(((EVP_AEAD_CTX *)NULL)->state) >=
+                          sizeof(struct aead_aes_gcm_siv_ctx),
+                      "AEAD state is too small");
+#if defined(__GNUC__) || defined(__clang__)
+OPENSSL_STATIC_ASSERT(alignof(union evp_aead_ctx_st_state) >=
+                          alignof(struct aead_aes_gcm_siv_ctx),
+                      "AEAD state has insufficient alignment");
+#endif
+
 static int aead_aes_gcm_siv_init(EVP_AEAD_CTX *ctx, const uint8_t *key,
                                  size_t key_len, size_t tag_len) {
   const size_t key_bits = key_len * 8;
@@ -576,24 +596,18 @@ static int aead_aes_gcm_siv_init(EVP_AEAD_CTX *ctx, const uint8_t *key,
   }
 
   struct aead_aes_gcm_siv_ctx *gcm_siv_ctx =
-      OPENSSL_malloc(sizeof(struct aead_aes_gcm_siv_ctx));
-  if (gcm_siv_ctx == NULL) {
-    return 0;
-  }
+      (struct aead_aes_gcm_siv_ctx *)&ctx->state;
   OPENSSL_memset(gcm_siv_ctx, 0, sizeof(struct aead_aes_gcm_siv_ctx));
 
   aes_ctr_set_key(&gcm_siv_ctx->ks.ks, NULL, &gcm_siv_ctx->kgk_block, key,
                   key_len);
   gcm_siv_ctx->is_256 = (key_len == 32);
-  ctx->aead_state = gcm_siv_ctx;
   ctx->tag_len = tag_len;
 
   return 1;
 }
 
-static void aead_aes_gcm_siv_cleanup(EVP_AEAD_CTX *ctx) {
-  OPENSSL_free(ctx->aead_state);
-}
+static void aead_aes_gcm_siv_cleanup(EVP_AEAD_CTX *ctx) {}
 
 // gcm_siv_crypt encrypts (or decrypts—it's the same thing) |in_len| bytes from
 // |in| to |out|, using the block function |enc_block| with |key| in counter
@@ -718,7 +732,8 @@ static int aead_aes_gcm_siv_seal_scatter(
     size_t *out_tag_len, size_t max_out_tag_len, const uint8_t *nonce,
     size_t nonce_len, const uint8_t *in, size_t in_len, const uint8_t *extra_in,
     size_t extra_in_len, const uint8_t *ad, size_t ad_len) {
-  const struct aead_aes_gcm_siv_ctx *gcm_siv_ctx = ctx->aead_state;
+  const struct aead_aes_gcm_siv_ctx *gcm_siv_ctx =
+      (struct aead_aes_gcm_siv_ctx *)&ctx->state;
   const uint64_t in_len_64 = in_len;
   const uint64_t ad_len_64 = ad_len;
 
@@ -778,7 +793,8 @@ static int aead_aes_gcm_siv_open_gather(const EVP_AEAD_CTX *ctx, uint8_t *out,
     return 0;
   }
 
-  const struct aead_aes_gcm_siv_ctx *gcm_siv_ctx = ctx->aead_state;
+  const struct aead_aes_gcm_siv_ctx *gcm_siv_ctx =
+      (struct aead_aes_gcm_siv_ctx *)&ctx->state;
 
   struct gcm_siv_record_keys keys;
   gcm_siv_keys(gcm_siv_ctx, &keys, nonce);
@@ -831,7 +847,7 @@ static const EVP_AEAD aead_aes_256_gcm_siv = {
     NULL /* tag_len */,
 };
 
-#if defined(OPENSSL_X86_64) && !defined(OPENSSL_NO_ASM)
+#if defined(AES_GCM_SIV_ASM)
 
 static char avx_aesni_capable(void) {
   const uint32_t ecx = OPENSSL_ia32cap_P[1];
@@ -864,4 +880,4 @@ const EVP_AEAD *EVP_aead_aes_256_gcm_siv(void) {
   return &aead_aes_256_gcm_siv;
 }
 
-#endif  // X86_64 && !NO_ASM
+#endif  // AES_GCM_SIV_ASM