Mirror of BoringSSL (grpc依赖) https://boringssl.googlesource.com/boringssl
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

4372 lines
138 KiB

/* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
* All rights reserved.
*
* This package is an SSL implementation written
* by Eric Young (eay@cryptsoft.com).
* The implementation was written so as to conform with Netscapes SSL.
*
* This library is free for commercial and non-commercial use as long as
* the following conditions are aheared to. The following conditions
* apply to all code found in this distribution, be it the RC4, RSA,
* lhash, DES, etc., code; not just the SSL code. The SSL documentation
* included with this distribution is covered by the same copyright terms
* except that the holder is Tim Hudson (tjh@cryptsoft.com).
*
* Copyright remains Eric Young's, and as such any Copyright notices in
* the code are not to be removed.
* If this package is used in a product, Eric Young should be given attribution
* as the author of the parts of the library used.
* This can be in the form of a textual message at program startup or
* in documentation (online or textual) provided with the package.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the 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 acknowledgement:
* "This product includes cryptographic software written by
* Eric Young (eay@cryptsoft.com)"
* The word 'cryptographic' can be left out if the rouines from the library
* being used are not cryptographic related :-).
* 4. If you include any Windows specific code (or a derivative thereof) from
* the apps directory (application code) you must include an acknowledgement:
* "This product includes software written by Tim Hudson (tjh@cryptsoft.com)"
*
* THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND
* ANY EXPRESS 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 AUTHOR OR 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.
*
* The licence and distribution terms for any publically available version or
* derivative of this code cannot be changed. i.e. this code cannot simply be
* copied and put under another distribution licence
* [including the GNU Public Licence.]
*/
/* ====================================================================
* Copyright (c) 1998-2007 The OpenSSL Project. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 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.
* ====================================================================
*
* This product includes cryptographic software written by Eric Young
* (eay@cryptsoft.com). This product includes software written by Tim
* Hudson (tjh@cryptsoft.com). */
#include <openssl/ssl.h>
#include <assert.h>
#include <limits.h>
#include <stdlib.h>
#include <string.h>
#include <algorithm>
#include <utility>
#include <openssl/aead.h>
#include <openssl/bytestring.h>
#include <openssl/chacha.h>
#include <openssl/curve25519.h>
#include <openssl/digest.h>
#include <openssl/err.h>
#include <openssl/evp.h>
#include <openssl/hmac.h>
#include <openssl/hpke.h>
#include <openssl/mem.h>
#include <openssl/nid.h>
#include <openssl/rand.h>
#include "../crypto/internal.h"
#include "internal.h"
BSSL_NAMESPACE_BEGIN
static bool ssl_check_clienthello_tlsext(SSL_HANDSHAKE *hs);
static bool ssl_check_serverhello_tlsext(SSL_HANDSHAKE *hs);
static int compare_uint16_t(const void *p1, const void *p2) {
uint16_t u1 = *((const uint16_t *)p1);
uint16_t u2 = *((const uint16_t *)p2);
if (u1 < u2) {
return -1;
} else if (u1 > u2) {
return 1;
} else {
return 0;
}
}
// Per http://tools.ietf.org/html/rfc5246#section-7.4.1.4, there may not be
// more than one extension of the same type in a ClientHello or ServerHello.
// This function does an initial scan over the extensions block to filter those
// out.
static bool tls1_check_duplicate_extensions(const CBS *cbs) {
// First pass: count the extensions.
size_t num_extensions = 0;
CBS extensions = *cbs;
while (CBS_len(&extensions) > 0) {
uint16_t type;
CBS extension;
if (!CBS_get_u16(&extensions, &type) ||
!CBS_get_u16_length_prefixed(&extensions, &extension)) {
return false;
}
num_extensions++;
}
if (num_extensions == 0) {
return true;
}
Array<uint16_t> extension_types;
if (!extension_types.Init(num_extensions)) {
return false;
}
// Second pass: gather the extension types.
extensions = *cbs;
for (size_t i = 0; i < extension_types.size(); i++) {
CBS extension;
if (!CBS_get_u16(&extensions, &extension_types[i]) ||
!CBS_get_u16_length_prefixed(&extensions, &extension)) {
// This should not happen.
return false;
}
}
assert(CBS_len(&extensions) == 0);
// Sort the extensions and make sure there are no duplicates.
qsort(extension_types.data(), extension_types.size(), sizeof(uint16_t),
compare_uint16_t);
for (size_t i = 1; i < num_extensions; i++) {
if (extension_types[i - 1] == extension_types[i]) {
return false;
}
}
return true;
}
static bool is_post_quantum_group(uint16_t id) {
switch (id) {
case SSL_GROUP_X25519_KYBER768_DRAFT00:
return true;
default:
return false;
}
}
bool ssl_client_hello_init(const SSL *ssl, SSL_CLIENT_HELLO *out,
Span<const uint8_t> body) {
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
CBS cbs = body;
if (!ssl_parse_client_hello_with_trailing_data(ssl, &cbs, out) ||
CBS_len(&cbs) != 0) {
return false;
}
return true;
}
bool ssl_parse_client_hello_with_trailing_data(const SSL *ssl, CBS *cbs,
SSL_CLIENT_HELLO *out) {
OPENSSL_memset(out, 0, sizeof(*out));
out->ssl = const_cast<SSL *>(ssl);
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
CBS copy = *cbs;
CBS random, session_id;
if (!CBS_get_u16(cbs, &out->version) ||
!CBS_get_bytes(cbs, &random, SSL3_RANDOM_SIZE) ||
!CBS_get_u8_length_prefixed(cbs, &session_id) ||
CBS_len(&session_id) > SSL_MAX_SSL_SESSION_ID_LENGTH) {
return false;
}
out->random = CBS_data(&random);
out->random_len = CBS_len(&random);
out->session_id = CBS_data(&session_id);
out->session_id_len = CBS_len(&session_id);
// Skip past DTLS cookie
if (SSL_is_dtls(out->ssl)) {
CBS cookie;
if (!CBS_get_u8_length_prefixed(cbs, &cookie)) {
return false;
}
}
CBS cipher_suites, compression_methods;
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
if (!CBS_get_u16_length_prefixed(cbs, &cipher_suites) ||
CBS_len(&cipher_suites) < 2 || (CBS_len(&cipher_suites) & 1) != 0 ||
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
!CBS_get_u8_length_prefixed(cbs, &compression_methods) ||
CBS_len(&compression_methods) < 1) {
return false;
}
out->cipher_suites = CBS_data(&cipher_suites);
out->cipher_suites_len = CBS_len(&cipher_suites);
out->compression_methods = CBS_data(&compression_methods);
out->compression_methods_len = CBS_len(&compression_methods);
// If the ClientHello ends here then it's valid, but doesn't have any
// extensions.
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
if (CBS_len(cbs) == 0) {
out->extensions = nullptr;
out->extensions_len = 0;
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
} else {
// Extract extensions and check it is valid.
CBS extensions;
if (!CBS_get_u16_length_prefixed(cbs, &extensions) ||
!tls1_check_duplicate_extensions(&extensions)) {
return false;
}
out->extensions = CBS_data(&extensions);
out->extensions_len = CBS_len(&extensions);
}
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
out->client_hello = CBS_data(&copy);
out->client_hello_len = CBS_len(&copy) - CBS_len(cbs);
return true;
}
bool ssl_client_hello_get_extension(const SSL_CLIENT_HELLO *client_hello,
CBS *out, uint16_t extension_type) {
CBS extensions;
CBS_init(&extensions, client_hello->extensions, client_hello->extensions_len);
while (CBS_len(&extensions) != 0) {
// Decode the next extension.
uint16_t type;
CBS extension;
if (!CBS_get_u16(&extensions, &type) ||
!CBS_get_u16_length_prefixed(&extensions, &extension)) {
return false;
}
if (type == extension_type) {
*out = extension;
return true;
}
}
return false;
}
static const uint16_t kDefaultGroups[] = {
SSL_GROUP_X25519,
SSL_GROUP_SECP256R1,
SSL_GROUP_SECP384R1,
};
Span<const uint16_t> tls1_get_grouplist(const SSL_HANDSHAKE *hs) {
if (!hs->config->supported_group_list.empty()) {
return hs->config->supported_group_list;
}
return Span<const uint16_t>(kDefaultGroups);
}
bool tls1_get_shared_group(SSL_HANDSHAKE *hs, uint16_t *out_group_id) {
SSL *const ssl = hs->ssl;
assert(ssl->server);
// Clients are not required to send a supported_groups extension. In this
// case, the server is free to pick any group it likes. See RFC 4492,
// section 4, paragraph 3.
//
// However, in the interests of compatibility, we will skip ECDH if the
// client didn't send an extension because we can't be sure that they'll
// support our favoured group. Thus we do not special-case an emtpy
// |peer_supported_group_list|.
Span<const uint16_t> groups = tls1_get_grouplist(hs);
Span<const uint16_t> pref, supp;
if (ssl->options & SSL_OP_CIPHER_SERVER_PREFERENCE) {
pref = groups;
supp = hs->peer_supported_group_list;
} else {
pref = hs->peer_supported_group_list;
supp = groups;
}
for (uint16_t pref_group : pref) {
for (uint16_t supp_group : supp) {
if (pref_group == supp_group &&
// Post-quantum key agreements don't fit in the u8-length-prefixed
// ECPoint field in TLS 1.2 and below.
(ssl_protocol_version(ssl) >= TLS1_3_VERSION ||
!is_post_quantum_group(pref_group))) {
*out_group_id = pref_group;
return true;
}
}
}
return false;
}
bool tls1_check_group_id(const SSL_HANDSHAKE *hs, uint16_t group_id) {
if (is_post_quantum_group(group_id) &&
ssl_protocol_version(hs->ssl) < TLS1_3_VERSION) {
// Post-quantum "groups" require TLS 1.3.
return false;
}
// We internally assume zero is never allocated as a group ID.
if (group_id == 0) {
return false;
}
for (uint16_t supported : tls1_get_grouplist(hs)) {
if (supported == group_id) {
return true;
}
}
return false;
}
// kVerifySignatureAlgorithms is the default list of accepted signature
// algorithms for verifying.
static const uint16_t kVerifySignatureAlgorithms[] = {
// List our preferred algorithms first.
SSL_SIGN_ECDSA_SECP256R1_SHA256,
SSL_SIGN_RSA_PSS_RSAE_SHA256,
SSL_SIGN_RSA_PKCS1_SHA256,
// Larger hashes are acceptable.
SSL_SIGN_ECDSA_SECP384R1_SHA384,
SSL_SIGN_RSA_PSS_RSAE_SHA384,
SSL_SIGN_RSA_PKCS1_SHA384,
SSL_SIGN_RSA_PSS_RSAE_SHA512,
SSL_SIGN_RSA_PKCS1_SHA512,
// For now, SHA-1 is still accepted but least preferable.
SSL_SIGN_RSA_PKCS1_SHA1,
};
// kSignSignatureAlgorithms is the default list of supported signature
// algorithms for signing.
static const uint16_t kSignSignatureAlgorithms[] = {
// List our preferred algorithms first.
SSL_SIGN_ED25519,
SSL_SIGN_ECDSA_SECP256R1_SHA256,
SSL_SIGN_RSA_PSS_RSAE_SHA256,
SSL_SIGN_RSA_PKCS1_SHA256,
// If needed, sign larger hashes.
//
// TODO(davidben): Determine which of these may be pruned.
SSL_SIGN_ECDSA_SECP384R1_SHA384,
SSL_SIGN_RSA_PSS_RSAE_SHA384,
SSL_SIGN_RSA_PKCS1_SHA384,
SSL_SIGN_ECDSA_SECP521R1_SHA512,
SSL_SIGN_RSA_PSS_RSAE_SHA512,
SSL_SIGN_RSA_PKCS1_SHA512,
// If the peer supports nothing else, sign with SHA-1.
SSL_SIGN_ECDSA_SHA1,
SSL_SIGN_RSA_PKCS1_SHA1,
};
static Span<const uint16_t> tls12_get_verify_sigalgs(const SSL_HANDSHAKE *hs) {
if (hs->config->verify_sigalgs.empty()) {
return Span<const uint16_t>(kVerifySignatureAlgorithms);
}
return hs->config->verify_sigalgs;
}
bool tls12_add_verify_sigalgs(const SSL_HANDSHAKE *hs, CBB *out) {
for (uint16_t sigalg : tls12_get_verify_sigalgs(hs)) {
if (!CBB_add_u16(out, sigalg)) {
return false;
}
}
return true;
}
bool tls12_check_peer_sigalg(const SSL_HANDSHAKE *hs, uint8_t *out_alert,
uint16_t sigalg) {
for (uint16_t verify_sigalg : tls12_get_verify_sigalgs(hs)) {
if (verify_sigalg == sigalg) {
return true;
}
}
OPENSSL_PUT_ERROR(SSL, SSL_R_WRONG_SIGNATURE_TYPE);
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
// tls_extension represents a TLS extension that is handled internally.
//
// The parse callbacks receive a |CBS| that contains the contents of the
// extension (i.e. not including the type and length bytes). If an extension is
// not received then the parse callbacks will be called with a NULL CBS so that
// they can do any processing needed to handle the absence of an extension.
//
// The add callbacks receive a |CBB| to which the extension can be appended but
// the function is responsible for appending the type and length bytes too.
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
//
// |add_clienthello| may be called multiple times and must not mutate |hs|. It
// is additionally passed two output |CBB|s. If the extension is the same
// independent of the value of |type|, the callback may write to
// |out_compressible| instead of |out|. When serializing the ClientHelloInner,
// all compressible extensions will be made continguous and replaced with
// ech_outer_extensions when encrypted. When serializing the ClientHelloOuter
// or not offering ECH, |out| will be equal to |out_compressible|, so writing to
// |out_compressible| still works.
//
// Note the |parse_serverhello| and |add_serverhello| callbacks refer to the
// TLS 1.2 ServerHello. In TLS 1.3, these callbacks act on EncryptedExtensions,
// with ServerHello extensions handled elsewhere in the handshake.
//
// All callbacks return true for success and false for error. If a parse
// function returns zero then a fatal alert with value |*out_alert| will be
// sent. If |*out_alert| isn't set, then a |decode_error| alert will be sent.
struct tls_extension {
uint16_t value;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
bool (*add_clienthello)(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible, ssl_client_hello_type_t type);
bool (*parse_serverhello)(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents);
bool (*parse_clienthello)(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents);
bool (*add_serverhello)(SSL_HANDSHAKE *hs, CBB *out);
};
static bool forbid_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
if (contents != NULL) {
// Servers MUST NOT send this extension.
*out_alert = SSL_AD_UNSUPPORTED_EXTENSION;
OPENSSL_PUT_ERROR(SSL, SSL_R_UNEXPECTED_EXTENSION);
return false;
}
return true;
}
static bool ignore_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
// This extension from the client is handled elsewhere.
return true;
}
static bool dont_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
return true;
}
// Server name indication (SNI).
//
// https://tools.ietf.org/html/rfc6066#section-3.
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_sni_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
// If offering ECH, send the public name instead of the configured name.
Span<const uint8_t> hostname;
if (type == ssl_client_hello_outer) {
hostname = hs->selected_ech_config->public_name;
} else {
if (ssl->hostname == nullptr) {
return true;
}
hostname =
MakeConstSpan(reinterpret_cast<const uint8_t *>(ssl->hostname.get()),
strlen(ssl->hostname.get()));
}
CBB contents, server_name_list, name;
if (!CBB_add_u16(out, TLSEXT_TYPE_server_name) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &server_name_list) ||
!CBB_add_u8(&server_name_list, TLSEXT_NAMETYPE_host_name) ||
!CBB_add_u16_length_prefixed(&server_name_list, &name) ||
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
!CBB_add_bytes(&name, hostname.data(), hostname.size()) ||
!CBB_flush(out)) {
return false;
}
return true;
}
static bool ext_sni_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
// The server may acknowledge SNI with an empty extension. We check the syntax
// but otherwise ignore this signal.
return contents == NULL || CBS_len(contents) == 0;
}
static bool ext_sni_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
// SNI has already been parsed earlier in the handshake. See |extract_sni|.
return true;
}
static bool ext_sni_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
if (hs->ssl->s3->session_reused ||
!hs->should_ack_sni) {
return true;
}
if (!CBB_add_u16(out, TLSEXT_TYPE_server_name) ||
!CBB_add_u16(out, 0 /* length */)) {
return false;
}
return true;
}
// Encrypted ClientHello (ECH)
//
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
// https://tools.ietf.org/html/draft-ietf-tls-esni-13
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_ech_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
if (type == ssl_client_hello_inner) {
if (!CBB_add_u16(out, TLSEXT_TYPE_encrypted_client_hello) ||
!CBB_add_u16(out, /* length */ 1) ||
!CBB_add_u8(out, ECH_CLIENT_INNER)) {
return false;
}
return true;
}
if (hs->ech_client_outer.empty()) {
return true;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
CBB ech_body;
if (!CBB_add_u16(out, TLSEXT_TYPE_encrypted_client_hello) ||
!CBB_add_u16_length_prefixed(out, &ech_body) ||
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
!CBB_add_u8(&ech_body, ECH_CLIENT_OUTER) ||
!CBB_add_bytes(&ech_body, hs->ech_client_outer.data(),
hs->ech_client_outer.size()) ||
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
!CBB_flush(out)) {
return false;
}
return true;
}
static bool ext_ech_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents == NULL) {
return true;
}
// The ECH extension may not be sent in TLS 1.2 ServerHello, only TLS 1.3
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
// EncryptedExtensions. It also may not be sent in response to an inner ECH
// extension.
if (ssl_protocol_version(ssl) < TLS1_3_VERSION ||
ssl->s3->ech_status == ssl_ech_accepted) {
*out_alert = SSL_AD_UNSUPPORTED_EXTENSION;
OPENSSL_PUT_ERROR(SSL, SSL_R_UNEXPECTED_EXTENSION);
return false;
}
Implement ClientHelloOuter handshakes. If a client offers ECH, but the server rejects it, the client completes the handshake with ClientHelloOuter in order to authenticate retry keys. Implement this flow. This is largely allowing the existing handshake to proceed, but with some changes: - Certificate verification uses the other name. This CL routes this up to the built-in verifier and adds SSL_get0_ech_name_override for the callback. - We need to disable False Start to pick up server Finished in TLS 1.2. - Client certificates, notably in TLS 1.3 where they're encrypted, should only be revealed to the true server. Fortunately, not sending client certs is always an option, so do that. Channel ID has a similar issue. I've just omitted the extension in ClientHelloOuter because it's deprecated and is unlikely to be used with ECH at this point. ALPS may be worth some pondering but, the way it's currently used, is not sensitive. (Possibly we should change the draft to terminate the handshake before even sending that flight...) - The session is never offered in ClientHelloOuter, but our internal book-keeping doesn't quite notice. I had to replace ech_accept with a tri-state ech_status to correctly handle an edge case in SSL_get0_ech_name_override: when ECH + 0-RTT + reverify_on_resume are all enabled, the first certificate verification is for the 0-RTT session and should be against the true name, yet we have selected_ech_config && !ech_accept. A tri-state tracks when ECH is actually rejected. I've maintained this on the server as well, though the server never actually cares. Bug: 275 Change-Id: Ie55966ca3dc4ffcc8c381479f0fe9bcacd34d0f8 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48135 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
if (!ssl_is_valid_ech_config_list(*contents)) {
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
if (ssl->s3->ech_status == ssl_ech_rejected &&
Implement ClientHelloOuter handshakes. If a client offers ECH, but the server rejects it, the client completes the handshake with ClientHelloOuter in order to authenticate retry keys. Implement this flow. This is largely allowing the existing handshake to proceed, but with some changes: - Certificate verification uses the other name. This CL routes this up to the built-in verifier and adds SSL_get0_ech_name_override for the callback. - We need to disable False Start to pick up server Finished in TLS 1.2. - Client certificates, notably in TLS 1.3 where they're encrypted, should only be revealed to the true server. Fortunately, not sending client certs is always an option, so do that. Channel ID has a similar issue. I've just omitted the extension in ClientHelloOuter because it's deprecated and is unlikely to be used with ECH at this point. ALPS may be worth some pondering but, the way it's currently used, is not sensitive. (Possibly we should change the draft to terminate the handshake before even sending that flight...) - The session is never offered in ClientHelloOuter, but our internal book-keeping doesn't quite notice. I had to replace ech_accept with a tri-state ech_status to correctly handle an edge case in SSL_get0_ech_name_override: when ECH + 0-RTT + reverify_on_resume are all enabled, the first certificate verification is for the 0-RTT session and should be against the true name, yet we have selected_ech_config && !ech_accept. A tri-state tracks when ECH is actually rejected. I've maintained this on the server as well, though the server never actually cares. Bug: 275 Change-Id: Ie55966ca3dc4ffcc8c381479f0fe9bcacd34d0f8 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48135 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
!hs->ech_retry_configs.CopyFrom(*contents)) {
*out_alert = SSL_AD_INTERNAL_ERROR;
return false;
}
Implement ClientHelloOuter handshakes. If a client offers ECH, but the server rejects it, the client completes the handshake with ClientHelloOuter in order to authenticate retry keys. Implement this flow. This is largely allowing the existing handshake to proceed, but with some changes: - Certificate verification uses the other name. This CL routes this up to the built-in verifier and adds SSL_get0_ech_name_override for the callback. - We need to disable False Start to pick up server Finished in TLS 1.2. - Client certificates, notably in TLS 1.3 where they're encrypted, should only be revealed to the true server. Fortunately, not sending client certs is always an option, so do that. Channel ID has a similar issue. I've just omitted the extension in ClientHelloOuter because it's deprecated and is unlikely to be used with ECH at this point. ALPS may be worth some pondering but, the way it's currently used, is not sensitive. (Possibly we should change the draft to terminate the handshake before even sending that flight...) - The session is never offered in ClientHelloOuter, but our internal book-keeping doesn't quite notice. I had to replace ech_accept with a tri-state ech_status to correctly handle an edge case in SSL_get0_ech_name_override: when ECH + 0-RTT + reverify_on_resume are all enabled, the first certificate verification is for the 0-RTT session and should be against the true name, yet we have selected_ech_config && !ech_accept. A tri-state tracks when ECH is actually rejected. I've maintained this on the server as well, though the server never actually cares. Bug: 275 Change-Id: Ie55966ca3dc4ffcc8c381479f0fe9bcacd34d0f8 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48135 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
return true;
}
static bool ext_ech_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
if (contents == nullptr) {
return true;
}
uint8_t type;
if (!CBS_get_u8(contents, &type)) {
return false;
}
if (type == ECH_CLIENT_OUTER) {
// Outer ECH extensions are handled outside the callback.
return true;
}
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
if (type != ECH_CLIENT_INNER || CBS_len(contents) != 0) {
return false;
}
hs->ech_is_inner = true;
return true;
}
static bool ext_ech_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
SSL *const ssl = hs->ssl;
Implement ClientHelloOuter handshakes. If a client offers ECH, but the server rejects it, the client completes the handshake with ClientHelloOuter in order to authenticate retry keys. Implement this flow. This is largely allowing the existing handshake to proceed, but with some changes: - Certificate verification uses the other name. This CL routes this up to the built-in verifier and adds SSL_get0_ech_name_override for the callback. - We need to disable False Start to pick up server Finished in TLS 1.2. - Client certificates, notably in TLS 1.3 where they're encrypted, should only be revealed to the true server. Fortunately, not sending client certs is always an option, so do that. Channel ID has a similar issue. I've just omitted the extension in ClientHelloOuter because it's deprecated and is unlikely to be used with ECH at this point. ALPS may be worth some pondering but, the way it's currently used, is not sensitive. (Possibly we should change the draft to terminate the handshake before even sending that flight...) - The session is never offered in ClientHelloOuter, but our internal book-keeping doesn't quite notice. I had to replace ech_accept with a tri-state ech_status to correctly handle an edge case in SSL_get0_ech_name_override: when ECH + 0-RTT + reverify_on_resume are all enabled, the first certificate verification is for the 0-RTT session and should be against the true name, yet we have selected_ech_config && !ech_accept. A tri-state tracks when ECH is actually rejected. I've maintained this on the server as well, though the server never actually cares. Bug: 275 Change-Id: Ie55966ca3dc4ffcc8c381479f0fe9bcacd34d0f8 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48135 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
if (ssl_protocol_version(ssl) < TLS1_3_VERSION ||
ssl->s3->ech_status == ssl_ech_accepted || //
hs->ech_keys == nullptr) {
return true;
}
// Write the list of retry configs to |out|. Note |SSL_CTX_set1_ech_keys|
// ensures |ech_keys| contains at least one retry config.
CBB body, retry_configs;
if (!CBB_add_u16(out, TLSEXT_TYPE_encrypted_client_hello) ||
!CBB_add_u16_length_prefixed(out, &body) ||
!CBB_add_u16_length_prefixed(&body, &retry_configs)) {
return false;
}
for (const auto &config : hs->ech_keys->configs) {
if (!config->is_retry_config()) {
continue;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_bytes(&retry_configs, config->ech_config().raw.data(),
config->ech_config().raw.size())) {
return false;
}
}
return CBB_flush(out);
}
// Renegotiation indication.
//
// https://tools.ietf.org/html/rfc5746
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_ri_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
// Renegotiation indication is not necessary in TLS 1.3.
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (hs->min_version >= TLS1_3_VERSION ||
type == ssl_client_hello_inner) {
return true;
}
assert(ssl->s3->initial_handshake_complete ==
(ssl->s3->previous_client_finished_len != 0));
CBB contents, prev_finished;
if (!CBB_add_u16(out, TLSEXT_TYPE_renegotiate) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_u8_length_prefixed(&contents, &prev_finished) ||
!CBB_add_bytes(&prev_finished, ssl->s3->previous_client_finished,
ssl->s3->previous_client_finished_len) ||
!CBB_flush(out)) {
return false;
}
return true;
}
static bool ext_ri_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents != NULL && ssl_protocol_version(ssl) >= TLS1_3_VERSION) {
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
// Servers may not switch between omitting the extension and supporting it.
// See RFC 5746, sections 3.5 and 4.2.
if (ssl->s3->initial_handshake_complete &&
(contents != NULL) != ssl->s3->send_connection_binding) {
*out_alert = SSL_AD_HANDSHAKE_FAILURE;
OPENSSL_PUT_ERROR(SSL, SSL_R_RENEGOTIATION_MISMATCH);
return false;
}
if (contents == NULL) {
// Strictly speaking, if we want to avoid an attack we should *always* see
// RI even on initial ServerHello because the client doesn't see any
// renegotiation during an attack. However this would mean we could not
// connect to any server which doesn't support RI.
//
// OpenSSL has |SSL_OP_LEGACY_SERVER_CONNECT| to control this, but in
// practical terms every client sets it so it's just assumed here.
return true;
}
const size_t expected_len = ssl->s3->previous_client_finished_len +
ssl->s3->previous_server_finished_len;
// Check for logic errors
assert(!expected_len || ssl->s3->previous_client_finished_len);
assert(!expected_len || ssl->s3->previous_server_finished_len);
assert(ssl->s3->initial_handshake_complete ==
(ssl->s3->previous_client_finished_len != 0));
assert(ssl->s3->initial_handshake_complete ==
(ssl->s3->previous_server_finished_len != 0));
// Parse out the extension contents.
CBS renegotiated_connection;
if (!CBS_get_u8_length_prefixed(contents, &renegotiated_connection) ||
CBS_len(contents) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_RENEGOTIATION_ENCODING_ERR);
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
// Check that the extension matches.
if (CBS_len(&renegotiated_connection) != expected_len) {
OPENSSL_PUT_ERROR(SSL, SSL_R_RENEGOTIATION_MISMATCH);
*out_alert = SSL_AD_HANDSHAKE_FAILURE;
return false;
}
const uint8_t *d = CBS_data(&renegotiated_connection);
bool ok = CRYPTO_memcmp(d, ssl->s3->previous_client_finished,
ssl->s3->previous_client_finished_len) == 0;
#if defined(BORINGSSL_UNSAFE_FUZZER_MODE)
ok = true;
#endif
if (!ok) {
OPENSSL_PUT_ERROR(SSL, SSL_R_RENEGOTIATION_MISMATCH);
*out_alert = SSL_AD_HANDSHAKE_FAILURE;
return false;
}
d += ssl->s3->previous_client_finished_len;
ok = CRYPTO_memcmp(d, ssl->s3->previous_server_finished,
ssl->s3->previous_server_finished_len) == 0;
#if defined(BORINGSSL_UNSAFE_FUZZER_MODE)
ok = true;
#endif
if (!ok) {
OPENSSL_PUT_ERROR(SSL, SSL_R_RENEGOTIATION_MISMATCH);
*out_alert = SSL_AD_HANDSHAKE_FAILURE;
return false;
}
ssl->s3->send_connection_binding = true;
return true;
}
static bool ext_ri_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
// Renegotiation isn't supported as a server so this function should never be
// called after the initial handshake.
assert(!ssl->s3->initial_handshake_complete);
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION) {
return true;
}
if (contents == NULL) {
return true;
}
CBS renegotiated_connection;
if (!CBS_get_u8_length_prefixed(contents, &renegotiated_connection) ||
CBS_len(contents) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_RENEGOTIATION_ENCODING_ERR);
return false;
}
// Check that the extension matches. We do not support renegotiation as a
// server, so this must be empty.
if (CBS_len(&renegotiated_connection) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_RENEGOTIATION_MISMATCH);
*out_alert = SSL_AD_HANDSHAKE_FAILURE;
return false;
}
ssl->s3->send_connection_binding = true;
return true;
}
static bool ext_ri_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
SSL *const ssl = hs->ssl;
// Renegotiation isn't supported as a server so this function should never be
// called after the initial handshake.
assert(!ssl->s3->initial_handshake_complete);
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION) {
return true;
}
if (!CBB_add_u16(out, TLSEXT_TYPE_renegotiate) ||
!CBB_add_u16(out, 1 /* length */) ||
!CBB_add_u8(out, 0 /* empty renegotiation info */)) {
return false;
}
return true;
}
// Extended Master Secret.
//
// https://tools.ietf.org/html/rfc7627
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_ems_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
// Extended master secret is not necessary in TLS 1.3.
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (hs->min_version >= TLS1_3_VERSION || type == ssl_client_hello_inner) {
return true;
}
if (!CBB_add_u16(out, TLSEXT_TYPE_extended_master_secret) ||
!CBB_add_u16(out, 0 /* length */)) {
return false;
}
return true;
}
static bool ext_ems_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents != NULL) {
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION ||
CBS_len(contents) != 0) {
return false;
}
hs->extended_master_secret = true;
}
// Whether EMS is negotiated may not change on renegotiation.
if (ssl->s3->established_session != nullptr &&
hs->extended_master_secret !=
!!ssl->s3->established_session->extended_master_secret) {
OPENSSL_PUT_ERROR(SSL, SSL_R_RENEGOTIATION_EMS_MISMATCH);
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
return true;
}
static bool ext_ems_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
if (ssl_protocol_version(hs->ssl) >= TLS1_3_VERSION) {
return true;
}
if (contents == NULL) {
return true;
}
if (CBS_len(contents) != 0) {
return false;
}
hs->extended_master_secret = true;
return true;
}
static bool ext_ems_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
if (!hs->extended_master_secret) {
return true;
}
if (!CBB_add_u16(out, TLSEXT_TYPE_extended_master_secret) ||
!CBB_add_u16(out, 0 /* length */)) {
return false;
}
return true;
}
// Session tickets.
//
// https://tools.ietf.org/html/rfc5077
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_ticket_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
// TLS 1.3 uses a different ticket extension.
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (hs->min_version >= TLS1_3_VERSION || type == ssl_client_hello_inner ||
SSL_get_options(ssl) & SSL_OP_NO_TICKET) {
return true;
}
Span<const uint8_t> ticket;
// Renegotiation does not participate in session resumption. However, still
// advertise the extension to avoid potentially breaking servers which carry
// over the state from the previous handshake, such as OpenSSL servers
// without upstream's 3c3f0259238594d77264a78944d409f2127642c4.
if (!ssl->s3->initial_handshake_complete &&
ssl->session != nullptr &&
!ssl->session->ticket.empty() &&
// Don't send TLS 1.3 session tickets in the ticket extension.
ssl_session_protocol_version(ssl->session.get()) < TLS1_3_VERSION) {
ticket = ssl->session->ticket;
}
CBB ticket_cbb;
if (!CBB_add_u16(out, TLSEXT_TYPE_session_ticket) ||
!CBB_add_u16_length_prefixed(out, &ticket_cbb) ||
!CBB_add_bytes(&ticket_cbb, ticket.data(), ticket.size()) ||
!CBB_flush(out)) {
return false;
}
return true;
}
static bool ext_ticket_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents == NULL) {
return true;
}
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION) {
return false;
}
// If |SSL_OP_NO_TICKET| is set then no extension will have been sent and
// this function should never be called, even if the server tries to send the
// extension.
assert((SSL_get_options(ssl) & SSL_OP_NO_TICKET) == 0);
if (CBS_len(contents) != 0) {
return false;
}
hs->ticket_expected = true;
return true;
}
static bool ext_ticket_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
if (!hs->ticket_expected) {
return true;
}
// If |SSL_OP_NO_TICKET| is set, |ticket_expected| should never be true.
assert((SSL_get_options(hs->ssl) & SSL_OP_NO_TICKET) == 0);
if (!CBB_add_u16(out, TLSEXT_TYPE_session_ticket) ||
!CBB_add_u16(out, 0 /* length */)) {
return false;
}
return true;
}
// Signature Algorithms.
//
// https://tools.ietf.org/html/rfc5246#section-7.4.1.4.1
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_sigalgs_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
if (hs->max_version < TLS1_2_VERSION) {
return true;
}
CBB contents, sigalgs_cbb;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_u16(out_compressible, TLSEXT_TYPE_signature_algorithms) ||
!CBB_add_u16_length_prefixed(out_compressible, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &sigalgs_cbb) ||
!tls12_add_verify_sigalgs(hs, &sigalgs_cbb) ||
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
!CBB_flush(out_compressible)) {
return false;
}
return true;
}
static bool ext_sigalgs_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
hs->peer_sigalgs.Reset();
if (contents == NULL) {
return true;
}
CBS supported_signature_algorithms;
if (!CBS_get_u16_length_prefixed(contents, &supported_signature_algorithms) ||
CBS_len(contents) != 0 ||
!tls1_parse_peer_sigalgs(hs, &supported_signature_algorithms)) {
return false;
}
return true;
}
// OCSP Stapling.
//
// https://tools.ietf.org/html/rfc6066#section-8
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_ocsp_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
if (!hs->config->ocsp_stapling_enabled) {
return true;
}
CBB contents;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_u16(out_compressible, TLSEXT_TYPE_status_request) ||
!CBB_add_u16_length_prefixed(out_compressible, &contents) ||
!CBB_add_u8(&contents, TLSEXT_STATUSTYPE_ocsp) ||
!CBB_add_u16(&contents, 0 /* empty responder ID list */) ||
!CBB_add_u16(&contents, 0 /* empty request extensions */) ||
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
!CBB_flush(out_compressible)) {
return false;
}
return true;
}
static bool ext_ocsp_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents == NULL) {
return true;
}
// TLS 1.3 OCSP responses are included in the Certificate extensions.
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION) {
return false;
}
// OCSP stapling is forbidden on non-certificate ciphers.
if (CBS_len(contents) != 0 ||
!ssl_cipher_uses_certificate_auth(hs->new_cipher)) {
return false;
}
// Note this does not check for resumption in TLS 1.2. Sending
// status_request here does not make sense, but OpenSSL does so and the
// specification does not say anything. Tolerate it but ignore it.
hs->certificate_status_expected = true;
return true;
}
static bool ext_ocsp_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
if (contents == NULL) {
return true;
}
uint8_t status_type;
if (!CBS_get_u8(contents, &status_type)) {
return false;
}
// We cannot decide whether OCSP stapling will occur yet because the correct
// SSL_CTX might not have been selected.
hs->ocsp_stapling_requested = status_type == TLSEXT_STATUSTYPE_ocsp;
return true;
}
static bool ext_ocsp_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
SSL *const ssl = hs->ssl;
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION ||
!hs->ocsp_stapling_requested || hs->config->cert->ocsp_response == NULL ||
ssl->s3->session_reused ||
!ssl_cipher_uses_certificate_auth(hs->new_cipher)) {
return true;
}
hs->certificate_status_expected = true;
return CBB_add_u16(out, TLSEXT_TYPE_status_request) &&
CBB_add_u16(out, 0 /* length */);
}
// Next protocol negotiation.
//
// https://htmlpreview.github.io/?https://github.com/agl/technotes/blob/master/nextprotoneg.html
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_npn_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (ssl->ctx->next_proto_select_cb == NULL ||
// Do not allow NPN to change on renegotiation.
ssl->s3->initial_handshake_complete ||
// NPN is not defined in DTLS or TLS 1.3.
SSL_is_dtls(ssl) || hs->min_version >= TLS1_3_VERSION ||
type == ssl_client_hello_inner) {
return true;
}
if (!CBB_add_u16(out, TLSEXT_TYPE_next_proto_neg) ||
!CBB_add_u16(out, 0 /* length */)) {
return false;
}
return true;
}
static bool ext_npn_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents == NULL) {
return true;
}
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION) {
return false;
}
// If any of these are false then we should never have sent the NPN
// extension in the ClientHello and thus this function should never have been
// called.
assert(!ssl->s3->initial_handshake_complete);
assert(!SSL_is_dtls(ssl));
assert(ssl->ctx->next_proto_select_cb != NULL);
if (!ssl->s3->alpn_selected.empty()) {
// NPN and ALPN may not be negotiated in the same connection.
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
OPENSSL_PUT_ERROR(SSL, SSL_R_NEGOTIATED_BOTH_NPN_AND_ALPN);
return false;
}
const uint8_t *const orig_contents = CBS_data(contents);
const size_t orig_len = CBS_len(contents);
while (CBS_len(contents) != 0) {
CBS proto;
if (!CBS_get_u8_length_prefixed(contents, &proto) ||
CBS_len(&proto) == 0) {
return false;
}
}
// |orig_len| fits in |unsigned| because TLS extensions use 16-bit lengths.
uint8_t *selected;
uint8_t selected_len;
if (ssl->ctx->next_proto_select_cb(
ssl, &selected, &selected_len, orig_contents,
static_cast<unsigned>(orig_len),
ssl->ctx->next_proto_select_cb_arg) != SSL_TLSEXT_ERR_OK ||
!ssl->s3->next_proto_negotiated.CopyFrom(
MakeConstSpan(selected, selected_len))) {
*out_alert = SSL_AD_INTERNAL_ERROR;
return false;
}
hs->next_proto_neg_seen = true;
return true;
}
static bool ext_npn_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION) {
return true;
}
if (contents != NULL && CBS_len(contents) != 0) {
return false;
}
if (contents == NULL ||
ssl->s3->initial_handshake_complete ||
ssl->ctx->next_protos_advertised_cb == NULL ||
SSL_is_dtls(ssl)) {
return true;
}
hs->next_proto_neg_seen = true;
return true;
}
static bool ext_npn_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
SSL *const ssl = hs->ssl;
// |next_proto_neg_seen| might have been cleared when an ALPN extension was
// parsed.
if (!hs->next_proto_neg_seen) {
return true;
}
const uint8_t *npa;
unsigned npa_len;
if (ssl->ctx->next_protos_advertised_cb(
ssl, &npa, &npa_len, ssl->ctx->next_protos_advertised_cb_arg) !=
SSL_TLSEXT_ERR_OK) {
hs->next_proto_neg_seen = false;
return true;
}
CBB contents;
if (!CBB_add_u16(out, TLSEXT_TYPE_next_proto_neg) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_bytes(&contents, npa, npa_len) ||
!CBB_flush(out)) {
return false;
}
return true;
}
// Signed certificate timestamps.
//
// https://tools.ietf.org/html/rfc6962#section-3.3.1
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_sct_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
if (!hs->config->signed_cert_timestamps_enabled) {
return true;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_u16(out_compressible, TLSEXT_TYPE_certificate_timestamp) ||
!CBB_add_u16(out_compressible, 0 /* length */)) {
return false;
}
return true;
}
static bool ext_sct_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents == NULL) {
return true;
}
// TLS 1.3 SCTs are included in the Certificate extensions.
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION) {
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
// If this is false then we should never have sent the SCT extension in the
// ClientHello and thus this function should never have been called.
assert(hs->config->signed_cert_timestamps_enabled);
if (!ssl_is_sct_list_valid(contents)) {
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
// Session resumption uses the original session information. The extension
// should not be sent on resumption, but RFC 6962 did not make it a
// requirement, so tolerate this.
//
// TODO(davidben): Enforce this anyway.
if (!ssl->s3->session_reused) {
hs->new_session->signed_cert_timestamp_list.reset(
CRYPTO_BUFFER_new_from_CBS(contents, ssl->ctx->pool));
if (hs->new_session->signed_cert_timestamp_list == nullptr) {
*out_alert = SSL_AD_INTERNAL_ERROR;
return false;
}
}
return true;
}
static bool ext_sct_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
if (contents == NULL) {
return true;
}
if (CBS_len(contents) != 0) {
return false;
}
hs->scts_requested = true;
return true;
}
static bool ext_sct_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
SSL *const ssl = hs->ssl;
// The extension shouldn't be sent when resuming sessions.
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION || ssl->s3->session_reused ||
hs->config->cert->signed_cert_timestamp_list == NULL) {
return true;
}
CBB contents;
return CBB_add_u16(out, TLSEXT_TYPE_certificate_timestamp) &&
CBB_add_u16_length_prefixed(out, &contents) &&
CBB_add_bytes(
&contents,
CRYPTO_BUFFER_data(
hs->config->cert->signed_cert_timestamp_list.get()),
CRYPTO_BUFFER_len(
hs->config->cert->signed_cert_timestamp_list.get())) &&
CBB_flush(out);
}
// Application-level Protocol Negotiation.
//
// https://tools.ietf.org/html/rfc7301
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_alpn_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
if (hs->config->alpn_client_proto_list.empty() && ssl->quic_method) {
// ALPN MUST be used with QUIC.
OPENSSL_PUT_ERROR(SSL, SSL_R_NO_APPLICATION_PROTOCOL);
return false;
}
if (hs->config->alpn_client_proto_list.empty() ||
ssl->s3->initial_handshake_complete) {
return true;
}
CBB contents, proto_list;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_u16(out_compressible,
TLSEXT_TYPE_application_layer_protocol_negotiation) ||
!CBB_add_u16_length_prefixed(out_compressible, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &proto_list) ||
!CBB_add_bytes(&proto_list, hs->config->alpn_client_proto_list.data(),
hs->config->alpn_client_proto_list.size()) ||
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
!CBB_flush(out_compressible)) {
return false;
}
return true;
}
static bool ext_alpn_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents == NULL) {
if (ssl->quic_method) {
// ALPN is required when QUIC is used.
OPENSSL_PUT_ERROR(SSL, SSL_R_NO_APPLICATION_PROTOCOL);
*out_alert = SSL_AD_NO_APPLICATION_PROTOCOL;
return false;
}
return true;
}
assert(!ssl->s3->initial_handshake_complete);
assert(!hs->config->alpn_client_proto_list.empty());
if (hs->next_proto_neg_seen) {
// NPN and ALPN may not be negotiated in the same connection.
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
OPENSSL_PUT_ERROR(SSL, SSL_R_NEGOTIATED_BOTH_NPN_AND_ALPN);
return false;
}
// The extension data consists of a ProtocolNameList which must have
// exactly one ProtocolName. Each of these is length-prefixed.
CBS protocol_name_list, protocol_name;
if (!CBS_get_u16_length_prefixed(contents, &protocol_name_list) ||
CBS_len(contents) != 0 ||
!CBS_get_u8_length_prefixed(&protocol_name_list, &protocol_name) ||
// Empty protocol names are forbidden.
CBS_len(&protocol_name) == 0 ||
CBS_len(&protocol_name_list) != 0) {
return false;
}
if (!ssl_is_alpn_protocol_allowed(hs, protocol_name)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_INVALID_ALPN_PROTOCOL);
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
if (!ssl->s3->alpn_selected.CopyFrom(protocol_name)) {
*out_alert = SSL_AD_INTERNAL_ERROR;
return false;
}
return true;
}
bool ssl_is_valid_alpn_list(Span<const uint8_t> in) {
CBS protocol_name_list = in;
if (CBS_len(&protocol_name_list) == 0) {
return false;
}
while (CBS_len(&protocol_name_list) > 0) {
CBS protocol_name;
if (!CBS_get_u8_length_prefixed(&protocol_name_list, &protocol_name) ||
// Empty protocol names are forbidden.
CBS_len(&protocol_name) == 0) {
return false;
}
}
return true;
}
bool ssl_is_alpn_protocol_allowed(const SSL_HANDSHAKE *hs,
Span<const uint8_t> protocol) {
if (hs->config->alpn_client_proto_list.empty()) {
return false;
}
if (hs->ssl->ctx->allow_unknown_alpn_protos) {
return true;
}
// Check that the protocol name is one of the ones we advertised.
CBS client_protocol_name_list =
MakeConstSpan(hs->config->alpn_client_proto_list),
client_protocol_name;
while (CBS_len(&client_protocol_name_list) > 0) {
if (!CBS_get_u8_length_prefixed(&client_protocol_name_list,
&client_protocol_name)) {
return false;
}
if (client_protocol_name == protocol) {
return true;
}
}
return false;
}
bool ssl_negotiate_alpn(SSL_HANDSHAKE *hs, uint8_t *out_alert,
const SSL_CLIENT_HELLO *client_hello) {
SSL *const ssl = hs->ssl;
CBS contents;
if (ssl->ctx->alpn_select_cb == NULL ||
!ssl_client_hello_get_extension(
client_hello, &contents,
TLSEXT_TYPE_application_layer_protocol_negotiation)) {
if (ssl->quic_method) {
// ALPN is required when QUIC is used.
OPENSSL_PUT_ERROR(SSL, SSL_R_NO_APPLICATION_PROTOCOL);
*out_alert = SSL_AD_NO_APPLICATION_PROTOCOL;
return false;
}
// Ignore ALPN if not configured or no extension was supplied.
return true;
}
// ALPN takes precedence over NPN.
hs->next_proto_neg_seen = false;
CBS protocol_name_list;
if (!CBS_get_u16_length_prefixed(&contents, &protocol_name_list) ||
CBS_len(&contents) != 0 ||
!ssl_is_valid_alpn_list(protocol_name_list)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_PARSE_TLSEXT);
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
// |protocol_name_list| fits in |unsigned| because TLS extensions use 16-bit
// lengths.
const uint8_t *selected;
uint8_t selected_len;
int ret = ssl->ctx->alpn_select_cb(
ssl, &selected, &selected_len, CBS_data(&protocol_name_list),
static_cast<unsigned>(CBS_len(&protocol_name_list)),
ssl->ctx->alpn_select_cb_arg);
// ALPN is required when QUIC is used.
if (ssl->quic_method &&
(ret == SSL_TLSEXT_ERR_NOACK || ret == SSL_TLSEXT_ERR_ALERT_WARNING)) {
ret = SSL_TLSEXT_ERR_ALERT_FATAL;
}
switch (ret) {
case SSL_TLSEXT_ERR_OK:
if (selected_len == 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_INVALID_ALPN_PROTOCOL);
*out_alert = SSL_AD_INTERNAL_ERROR;
return false;
}
if (!ssl->s3->alpn_selected.CopyFrom(
MakeConstSpan(selected, selected_len))) {
*out_alert = SSL_AD_INTERNAL_ERROR;
return false;
}
break;
case SSL_TLSEXT_ERR_NOACK:
case SSL_TLSEXT_ERR_ALERT_WARNING:
break;
case SSL_TLSEXT_ERR_ALERT_FATAL:
*out_alert = SSL_AD_NO_APPLICATION_PROTOCOL;
OPENSSL_PUT_ERROR(SSL, SSL_R_NO_APPLICATION_PROTOCOL);
return false;
default:
// Invalid return value.
*out_alert = SSL_AD_INTERNAL_ERROR;
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
return true;
}
static bool ext_alpn_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
SSL *const ssl = hs->ssl;
if (ssl->s3->alpn_selected.empty()) {
return true;
}
CBB contents, proto_list, proto;
if (!CBB_add_u16(out, TLSEXT_TYPE_application_layer_protocol_negotiation) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &proto_list) ||
!CBB_add_u8_length_prefixed(&proto_list, &proto) ||
!CBB_add_bytes(&proto, ssl->s3->alpn_selected.data(),
ssl->s3->alpn_selected.size()) ||
!CBB_flush(out)) {
return false;
}
return true;
}
// Channel ID.
//
// https://tools.ietf.org/html/draft-balfanz-tls-channelid-01
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_channel_id_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
Implement ClientHelloOuter handshakes. If a client offers ECH, but the server rejects it, the client completes the handshake with ClientHelloOuter in order to authenticate retry keys. Implement this flow. This is largely allowing the existing handshake to proceed, but with some changes: - Certificate verification uses the other name. This CL routes this up to the built-in verifier and adds SSL_get0_ech_name_override for the callback. - We need to disable False Start to pick up server Finished in TLS 1.2. - Client certificates, notably in TLS 1.3 where they're encrypted, should only be revealed to the true server. Fortunately, not sending client certs is always an option, so do that. Channel ID has a similar issue. I've just omitted the extension in ClientHelloOuter because it's deprecated and is unlikely to be used with ECH at this point. ALPS may be worth some pondering but, the way it's currently used, is not sensitive. (Possibly we should change the draft to terminate the handshake before even sending that flight...) - The session is never offered in ClientHelloOuter, but our internal book-keeping doesn't quite notice. I had to replace ech_accept with a tri-state ech_status to correctly handle an edge case in SSL_get0_ech_name_override: when ECH + 0-RTT + reverify_on_resume are all enabled, the first certificate verification is for the 0-RTT session and should be against the true name, yet we have selected_ech_config && !ech_accept. A tri-state tracks when ECH is actually rejected. I've maintained this on the server as well, though the server never actually cares. Bug: 275 Change-Id: Ie55966ca3dc4ffcc8c381479f0fe9bcacd34d0f8 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48135 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
if (!hs->config->channel_id_private || SSL_is_dtls(ssl) ||
// Don't offer Channel ID in ClientHelloOuter. ClientHelloOuter handshakes
// are not authenticated for the name that can learn the Channel ID.
//
// We could alternatively offer the extension but sign with a random key.
// For other extensions, we try to align |ssl_client_hello_outer| and
// |ssl_client_hello_unencrypted|, to improve the effectiveness of ECH
// GREASE. However, Channel ID is deprecated and unlikely to be used with
// ECH, so do the simplest thing.
type == ssl_client_hello_outer) {
return true;
}
Implement ClientHelloOuter handshakes. If a client offers ECH, but the server rejects it, the client completes the handshake with ClientHelloOuter in order to authenticate retry keys. Implement this flow. This is largely allowing the existing handshake to proceed, but with some changes: - Certificate verification uses the other name. This CL routes this up to the built-in verifier and adds SSL_get0_ech_name_override for the callback. - We need to disable False Start to pick up server Finished in TLS 1.2. - Client certificates, notably in TLS 1.3 where they're encrypted, should only be revealed to the true server. Fortunately, not sending client certs is always an option, so do that. Channel ID has a similar issue. I've just omitted the extension in ClientHelloOuter because it's deprecated and is unlikely to be used with ECH at this point. ALPS may be worth some pondering but, the way it's currently used, is not sensitive. (Possibly we should change the draft to terminate the handshake before even sending that flight...) - The session is never offered in ClientHelloOuter, but our internal book-keeping doesn't quite notice. I had to replace ech_accept with a tri-state ech_status to correctly handle an edge case in SSL_get0_ech_name_override: when ECH + 0-RTT + reverify_on_resume are all enabled, the first certificate verification is for the 0-RTT session and should be against the true name, yet we have selected_ech_config && !ech_accept. A tri-state tracks when ECH is actually rejected. I've maintained this on the server as well, though the server never actually cares. Bug: 275 Change-Id: Ie55966ca3dc4ffcc8c381479f0fe9bcacd34d0f8 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48135 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
if (!CBB_add_u16(out, TLSEXT_TYPE_channel_id) ||
!CBB_add_u16(out, 0 /* length */)) {
return false;
}
return true;
}
static bool ext_channel_id_parse_serverhello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
if (contents == NULL) {
return true;
}
assert(!SSL_is_dtls(hs->ssl));
assert(hs->config->channel_id_private);
if (CBS_len(contents) != 0) {
return false;
}
hs->channel_id_negotiated = true;
return true;
}
static bool ext_channel_id_parse_clienthello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents == NULL || !hs->config->channel_id_enabled || SSL_is_dtls(ssl)) {
return true;
}
if (CBS_len(contents) != 0) {
return false;
}
hs->channel_id_negotiated = true;
return true;
}
static bool ext_channel_id_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
if (!hs->channel_id_negotiated) {
return true;
}
if (!CBB_add_u16(out, TLSEXT_TYPE_channel_id) ||
!CBB_add_u16(out, 0 /* length */)) {
return false;
}
return true;
}
// Secure Real-time Transport Protocol (SRTP) extension.
//
// https://tools.ietf.org/html/rfc5764
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_srtp_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
const STACK_OF(SRTP_PROTECTION_PROFILE) *profiles =
SSL_get_srtp_profiles(ssl);
if (profiles == NULL ||
sk_SRTP_PROTECTION_PROFILE_num(profiles) == 0 ||
!SSL_is_dtls(ssl)) {
return true;
}
CBB contents, profile_ids;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_u16(out_compressible, TLSEXT_TYPE_srtp) ||
!CBB_add_u16_length_prefixed(out_compressible, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &profile_ids)) {
return false;
}
for (const SRTP_PROTECTION_PROFILE *profile : profiles) {
if (!CBB_add_u16(&profile_ids, profile->id)) {
return false;
}
}
if (!CBB_add_u8(&contents, 0 /* empty use_mki value */) ||
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
!CBB_flush(out_compressible)) {
return false;
}
return true;
}
static bool ext_srtp_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents == NULL) {
return true;
}
// The extension consists of a u16-prefixed profile ID list containing a
// single uint16_t profile ID, then followed by a u8-prefixed srtp_mki field.
//
// See https://tools.ietf.org/html/rfc5764#section-4.1.1
assert(SSL_is_dtls(ssl));
CBS profile_ids, srtp_mki;
uint16_t profile_id;
if (!CBS_get_u16_length_prefixed(contents, &profile_ids) ||
!CBS_get_u16(&profile_ids, &profile_id) ||
CBS_len(&profile_ids) != 0 ||
!CBS_get_u8_length_prefixed(contents, &srtp_mki) ||
CBS_len(contents) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_BAD_SRTP_PROTECTION_PROFILE_LIST);
return false;
}
if (CBS_len(&srtp_mki) != 0) {
// Must be no MKI, since we never offer one.
OPENSSL_PUT_ERROR(SSL, SSL_R_BAD_SRTP_MKI_VALUE);
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
// Check to see if the server gave us something we support and offered.
for (const SRTP_PROTECTION_PROFILE *profile : SSL_get_srtp_profiles(ssl)) {
if (profile->id == profile_id) {
ssl->s3->srtp_profile = profile;
return true;
}
}
OPENSSL_PUT_ERROR(SSL, SSL_R_BAD_SRTP_PROTECTION_PROFILE_LIST);
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
static bool ext_srtp_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
// DTLS-SRTP is only defined for DTLS.
if (contents == NULL || !SSL_is_dtls(ssl)) {
return true;
}
CBS profile_ids, srtp_mki;
if (!CBS_get_u16_length_prefixed(contents, &profile_ids) ||
CBS_len(&profile_ids) < 2 ||
!CBS_get_u8_length_prefixed(contents, &srtp_mki) ||
CBS_len(contents) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_BAD_SRTP_PROTECTION_PROFILE_LIST);
return false;
}
// Discard the MKI value for now.
const STACK_OF(SRTP_PROTECTION_PROFILE) *server_profiles =
SSL_get_srtp_profiles(ssl);
// Pick the server's most preferred profile.
for (const SRTP_PROTECTION_PROFILE *server_profile : server_profiles) {
CBS profile_ids_tmp;
CBS_init(&profile_ids_tmp, CBS_data(&profile_ids), CBS_len(&profile_ids));
while (CBS_len(&profile_ids_tmp) > 0) {
uint16_t profile_id;
if (!CBS_get_u16(&profile_ids_tmp, &profile_id)) {
return false;
}
if (server_profile->id == profile_id) {
ssl->s3->srtp_profile = server_profile;
return true;
}
}
}
return true;
}
static bool ext_srtp_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
SSL *const ssl = hs->ssl;
if (ssl->s3->srtp_profile == NULL) {
return true;
}
assert(SSL_is_dtls(ssl));
CBB contents, profile_ids;
if (!CBB_add_u16(out, TLSEXT_TYPE_srtp) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &profile_ids) ||
!CBB_add_u16(&profile_ids, ssl->s3->srtp_profile->id) ||
!CBB_add_u8(&contents, 0 /* empty MKI */) ||
!CBB_flush(out)) {
return false;
}
return true;
}
// EC point formats.
//
// https://tools.ietf.org/html/rfc4492#section-5.1.2
static bool ext_ec_point_add_extension(const SSL_HANDSHAKE *hs, CBB *out) {
CBB contents, formats;
if (!CBB_add_u16(out, TLSEXT_TYPE_ec_point_formats) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_u8_length_prefixed(&contents, &formats) ||
!CBB_add_u8(&formats, TLSEXT_ECPOINTFORMAT_uncompressed) ||
!CBB_flush(out)) {
return false;
}
return true;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_ec_point_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
// The point format extension is unnecessary in TLS 1.3.
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (hs->min_version >= TLS1_3_VERSION || type == ssl_client_hello_inner) {
return true;
}
return ext_ec_point_add_extension(hs, out);
}
static bool ext_ec_point_parse_serverhello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
if (contents == NULL) {
return true;
}
if (ssl_protocol_version(hs->ssl) >= TLS1_3_VERSION) {
return false;
}
CBS ec_point_format_list;
if (!CBS_get_u8_length_prefixed(contents, &ec_point_format_list) ||
CBS_len(contents) != 0) {
return false;
}
// Per RFC 4492, section 5.1.2, implementations MUST support the uncompressed
// point format.
if (OPENSSL_memchr(CBS_data(&ec_point_format_list),
TLSEXT_ECPOINTFORMAT_uncompressed,
CBS_len(&ec_point_format_list)) == NULL) {
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
return true;
}
static bool ext_ec_point_parse_clienthello(SSL_HANDSHAKE *hs, uint8_t *out_alert,
CBS *contents) {
if (ssl_protocol_version(hs->ssl) >= TLS1_3_VERSION) {
return true;
}
return ext_ec_point_parse_serverhello(hs, out_alert, contents);
}
static bool ext_ec_point_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
SSL *const ssl = hs->ssl;
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION) {
return true;
}
const uint32_t alg_k = hs->new_cipher->algorithm_mkey;
const uint32_t alg_a = hs->new_cipher->algorithm_auth;
const bool using_ecc = (alg_k & SSL_kECDHE) || (alg_a & SSL_aECDSA);
if (!using_ecc) {
return true;
}
return ext_ec_point_add_extension(hs, out);
}
// Pre Shared Key
//
// https://tools.ietf.org/html/rfc8446#section-4.2.11
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool should_offer_psk(const SSL_HANDSHAKE *hs,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
if (hs->max_version < TLS1_3_VERSION || ssl->session == nullptr ||
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
ssl_session_protocol_version(ssl->session.get()) < TLS1_3_VERSION ||
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
// TODO(https://crbug.com/boringssl/275): Should we synthesize a
// placeholder PSK, at least when we offer early data? Otherwise
// ClientHelloOuter will contain an early_data extension without a
// pre_shared_key extension and potentially break the recovery flow.
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
type == ssl_client_hello_outer) {
return false;
}
// Per RFC 8446 section 4.1.4, skip offering the session if the selected
// cipher in HelloRetryRequest does not match. This avoids performing the
// transcript hash transformation for multiple hashes.
if (ssl->s3->used_hello_retry_request &&
ssl->session->cipher->algorithm_prf != hs->new_cipher->algorithm_prf) {
return false;
}
return true;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static size_t ext_pre_shared_key_clienthello_length(
const SSL_HANDSHAKE *hs, ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!should_offer_psk(hs, type)) {
return 0;
}
size_t binder_len = EVP_MD_size(ssl_session_get_digest(ssl->session.get()));
return 15 + ssl->session->ticket.size() + binder_len;
}
static bool ext_pre_shared_key_add_clienthello(const SSL_HANDSHAKE *hs,
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
CBB *out, bool *out_needs_binder,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
*out_needs_binder = false;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!should_offer_psk(hs, type)) {
return true;
}
struct OPENSSL_timeval now;
ssl_get_current_time(ssl, &now);
uint32_t ticket_age = 1000 * (now.tv_sec - ssl->session->time);
uint32_t obfuscated_ticket_age = ticket_age + ssl->session->ticket_age_add;
// Fill in a placeholder zero binder of the appropriate length. It will be
// computed and filled in later after length prefixes are computed.
size_t binder_len = EVP_MD_size(ssl_session_get_digest(ssl->session.get()));
CBB contents, identity, ticket, binders, binder;
if (!CBB_add_u16(out, TLSEXT_TYPE_pre_shared_key) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &identity) ||
!CBB_add_u16_length_prefixed(&identity, &ticket) ||
!CBB_add_bytes(&ticket, ssl->session->ticket.data(),
ssl->session->ticket.size()) ||
!CBB_add_u32(&identity, obfuscated_ticket_age) ||
!CBB_add_u16_length_prefixed(&contents, &binders) ||
!CBB_add_u8_length_prefixed(&binders, &binder) ||
!CBB_add_zeros(&binder, binder_len)) {
return false;
}
*out_needs_binder = true;
return CBB_flush(out);
}
bool ssl_ext_pre_shared_key_parse_serverhello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
uint16_t psk_id;
if (!CBS_get_u16(contents, &psk_id) ||
CBS_len(contents) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
// We only advertise one PSK identity, so the only legal index is zero.
if (psk_id != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_PSK_IDENTITY_NOT_FOUND);
*out_alert = SSL_AD_UNKNOWN_PSK_IDENTITY;
return false;
}
return true;
}
bool ssl_ext_pre_shared_key_parse_clienthello(
SSL_HANDSHAKE *hs, CBS *out_ticket, CBS *out_binders,
uint32_t *out_obfuscated_ticket_age, uint8_t *out_alert,
const SSL_CLIENT_HELLO *client_hello, CBS *contents) {
// Verify that the pre_shared_key extension is the last extension in
// ClientHello.
if (CBS_data(contents) + CBS_len(contents) !=
client_hello->extensions + client_hello->extensions_len) {
OPENSSL_PUT_ERROR(SSL, SSL_R_PRE_SHARED_KEY_MUST_BE_LAST);
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
// We only process the first PSK identity since we don't support pure PSK.
CBS identities, binders;
if (!CBS_get_u16_length_prefixed(contents, &identities) ||
!CBS_get_u16_length_prefixed(&identities, out_ticket) ||
!CBS_get_u32(&identities, out_obfuscated_ticket_age) ||
!CBS_get_u16_length_prefixed(contents, &binders) ||
CBS_len(&binders) == 0 ||
CBS_len(contents) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
*out_binders = binders;
// Check the syntax of the remaining identities, but do not process them.
size_t num_identities = 1;
while (CBS_len(&identities) != 0) {
CBS unused_ticket;
uint32_t unused_obfuscated_ticket_age;
if (!CBS_get_u16_length_prefixed(&identities, &unused_ticket) ||
!CBS_get_u32(&identities, &unused_obfuscated_ticket_age)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
num_identities++;
}
// Check the syntax of the binders. The value will be checked later if
// resuming.
size_t num_binders = 0;
while (CBS_len(&binders) != 0) {
CBS binder;
if (!CBS_get_u8_length_prefixed(&binders, &binder)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
num_binders++;
}
if (num_identities != num_binders) {
OPENSSL_PUT_ERROR(SSL, SSL_R_PSK_IDENTITY_BINDER_COUNT_MISMATCH);
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
return true;
}
bool ssl_ext_pre_shared_key_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
if (!hs->ssl->s3->session_reused) {
return true;
}
CBB contents;
if (!CBB_add_u16(out, TLSEXT_TYPE_pre_shared_key) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
// We only consider the first identity for resumption
!CBB_add_u16(&contents, 0) ||
!CBB_flush(out)) {
return false;
}
return true;
}
// Pre-Shared Key Exchange Modes
//
// https://tools.ietf.org/html/rfc8446#section-4.2.9
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_psk_key_exchange_modes_add_clienthello(
const SSL_HANDSHAKE *hs, CBB *out, CBB *out_compressible,
ssl_client_hello_type_t type) {
if (hs->max_version < TLS1_3_VERSION) {
return true;
}
CBB contents, ke_modes;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_u16(out_compressible, TLSEXT_TYPE_psk_key_exchange_modes) ||
!CBB_add_u16_length_prefixed(out_compressible, &contents) ||
!CBB_add_u8_length_prefixed(&contents, &ke_modes) ||
!CBB_add_u8(&ke_modes, SSL_PSK_DHE_KE)) {
return false;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
return CBB_flush(out_compressible);
}
static bool ext_psk_key_exchange_modes_parse_clienthello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
if (contents == NULL) {
return true;
}
CBS ke_modes;
if (!CBS_get_u8_length_prefixed(contents, &ke_modes) ||
CBS_len(&ke_modes) == 0 ||
CBS_len(contents) != 0) {
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
// We only support tickets with PSK_DHE_KE.
hs->accept_psk_mode = OPENSSL_memchr(CBS_data(&ke_modes), SSL_PSK_DHE_KE,
CBS_len(&ke_modes)) != NULL;
return true;
}
// Early Data Indication
//
// https://tools.ietf.org/html/rfc8446#section-4.2.10
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_early_data_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
// The second ClientHello never offers early data, and we must have already
// filled in |early_data_reason| by this point.
if (ssl->s3->used_hello_retry_request) {
assert(ssl->s3->early_data_reason != ssl_early_data_unknown);
return true;
}
if (!hs->early_data_offered) {
return true;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
// If offering ECH, the extension only applies to ClientHelloInner, but we
// send the extension in both ClientHellos. This ensures that, if the server
// handshakes with ClientHelloOuter, it can skip past early data. See
// draft-ietf-tls-esni-13, section 6.1.
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_u16(out_compressible, TLSEXT_TYPE_early_data) ||
!CBB_add_u16(out_compressible, 0) ||
!CBB_flush(out_compressible)) {
return false;
}
return true;
}
static bool ext_early_data_parse_serverhello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents == NULL) {
if (hs->early_data_offered && !ssl->s3->used_hello_retry_request) {
ssl->s3->early_data_reason = ssl->s3->session_reused
? ssl_early_data_peer_declined
: ssl_early_data_session_not_resumed;
} else {
// We already filled in |early_data_reason| when declining to offer 0-RTT
// or handling the implicit HelloRetryRequest reject.
assert(ssl->s3->early_data_reason != ssl_early_data_unknown);
}
return true;
}
// If we received an HRR, the second ClientHello never offers early data, so
// the extensions logic will automatically reject early data extensions as
// unsolicited. This covered by the ServerAcceptsEarlyDataOnHRR test.
assert(!ssl->s3->used_hello_retry_request);
if (CBS_len(contents) != 0) {
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
if (!ssl->s3->session_reused) {
*out_alert = SSL_AD_UNSUPPORTED_EXTENSION;
OPENSSL_PUT_ERROR(SSL, SSL_R_UNEXPECTED_EXTENSION);
return false;
}
ssl->s3->early_data_reason = ssl_early_data_accepted;
ssl->s3->early_data_accepted = true;
return true;
}
static bool ext_early_data_parse_clienthello(SSL_HANDSHAKE *hs,
uint8_t *out_alert, CBS *contents) {
SSL *const ssl = hs->ssl;
if (contents == NULL ||
ssl_protocol_version(ssl) < TLS1_3_VERSION) {
return true;
}
if (CBS_len(contents) != 0) {
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
hs->early_data_offered = true;
return true;
}
static bool ext_early_data_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
if (!hs->ssl->s3->early_data_accepted) {
return true;
}
if (!CBB_add_u16(out, TLSEXT_TYPE_early_data) ||
!CBB_add_u16(out, 0) ||
!CBB_flush(out)) {
return false;
}
return true;
}
// Key Share
//
// https://tools.ietf.org/html/rfc8446#section-4.2.8
bool ssl_setup_key_shares(SSL_HANDSHAKE *hs, uint16_t override_group_id) {
SSL *const ssl = hs->ssl;
hs->key_shares[0].reset();
hs->key_shares[1].reset();
hs->key_share_bytes.Reset();
if (hs->max_version < TLS1_3_VERSION) {
return true;
}
bssl::ScopedCBB cbb;
if (!CBB_init(cbb.get(), 64)) {
return false;
}
if (override_group_id == 0 && ssl->ctx->grease_enabled) {
// Add a fake group. See RFC 8701.
if (!CBB_add_u16(cbb.get(), ssl_get_grease_value(hs, ssl_grease_group)) ||
!CBB_add_u16(cbb.get(), 1 /* length */) ||
!CBB_add_u8(cbb.get(), 0 /* one byte key share */)) {
return false;
}
}
uint16_t group_id = override_group_id;
uint16_t second_group_id = 0;
if (override_group_id == 0) {
// Predict the most preferred group.
Span<const uint16_t> groups = tls1_get_grouplist(hs);
if (groups.empty()) {
OPENSSL_PUT_ERROR(SSL, SSL_R_NO_GROUPS_SPECIFIED);
return false;
}
group_id = groups[0];
// We'll try to include one post-quantum and one classical initial key
// share.
for (size_t i = 1; i < groups.size() && second_group_id == 0; i++) {
if (is_post_quantum_group(group_id) != is_post_quantum_group(groups[i])) {
second_group_id = groups[i];
assert(second_group_id != group_id);
}
}
}
CBB key_exchange;
hs->key_shares[0] = SSLKeyShare::Create(group_id);
if (!hs->key_shares[0] || //
!CBB_add_u16(cbb.get(), group_id) ||
!CBB_add_u16_length_prefixed(cbb.get(), &key_exchange) ||
Use KEM terminology in TLS ECDHE and key_share abstractions TLS 1.2 ECDHE and TLS 1.3 key shares were originally designed around Diffie-Hellman-like primitives and use language based on that. Post-quantum replacements do not look like Diffie-Hellman, where each part exchanges a public key, but schemes that work differently can still slot in without protocol changes. We previously came up with our own Offer/Accept/Finish abstraction for early post-quantum experiments, but the NIST constructions are all expressed as KEMs: First, the recipient generates a keypair and sends the public key. Then the sender encapsulates a symmetric secret and sends the ciphertext. Finally, the recipient decapsulates the ciphertext to get the secret. Align our C++ and Go abstractions to this terminology. The functions are now called Generate/Encap/Decap, and the output of Encap is called "ciphertext", which seems to align with what most folks use. (RFC 9180 uses "enc" for "encapsulated key", but they staple a KEM to an AEAD, so "ciphertext" would be ambiguous.) Where variable names refer to parts of the protocol, rather than the the underlying KEM-like construction, I've kept variable names matching the protocol mechanism, so we still talk about "curves" and "key shares", but, when using the post-quantum replacements, the terminology is no longer quite accurate. I've also not yet renamed SSLKeyShare yet, though the name is now inaccurate. Also ideally we'd touch that up so the stateful object is just a KEM private key, for SSLKEMKey. Though at that point, we maybe should just add EVP_KEM and EVP_KEM_KEY APIs to libcrypto. Change-Id: Icbcc1840c5d2dfad210ef4caad2a7c4bf8146553 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/57726 Reviewed-by: Adam Langley <agl@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2 years ago
!hs->key_shares[0]->Generate(&key_exchange)) {
return false;
}
if (second_group_id != 0) {
hs->key_shares[1] = SSLKeyShare::Create(second_group_id);
if (!hs->key_shares[1] || //
!CBB_add_u16(cbb.get(), second_group_id) ||
!CBB_add_u16_length_prefixed(cbb.get(), &key_exchange) ||
Use KEM terminology in TLS ECDHE and key_share abstractions TLS 1.2 ECDHE and TLS 1.3 key shares were originally designed around Diffie-Hellman-like primitives and use language based on that. Post-quantum replacements do not look like Diffie-Hellman, where each part exchanges a public key, but schemes that work differently can still slot in without protocol changes. We previously came up with our own Offer/Accept/Finish abstraction for early post-quantum experiments, but the NIST constructions are all expressed as KEMs: First, the recipient generates a keypair and sends the public key. Then the sender encapsulates a symmetric secret and sends the ciphertext. Finally, the recipient decapsulates the ciphertext to get the secret. Align our C++ and Go abstractions to this terminology. The functions are now called Generate/Encap/Decap, and the output of Encap is called "ciphertext", which seems to align with what most folks use. (RFC 9180 uses "enc" for "encapsulated key", but they staple a KEM to an AEAD, so "ciphertext" would be ambiguous.) Where variable names refer to parts of the protocol, rather than the the underlying KEM-like construction, I've kept variable names matching the protocol mechanism, so we still talk about "curves" and "key shares", but, when using the post-quantum replacements, the terminology is no longer quite accurate. I've also not yet renamed SSLKeyShare yet, though the name is now inaccurate. Also ideally we'd touch that up so the stateful object is just a KEM private key, for SSLKEMKey. Though at that point, we maybe should just add EVP_KEM and EVP_KEM_KEY APIs to libcrypto. Change-Id: Icbcc1840c5d2dfad210ef4caad2a7c4bf8146553 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/57726 Reviewed-by: Adam Langley <agl@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2 years ago
!hs->key_shares[1]->Generate(&key_exchange)) {
return false;
}
}
return CBBFinishArray(cbb.get(), &hs->key_share_bytes);
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_key_share_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
if (hs->max_version < TLS1_3_VERSION) {
return true;
}
assert(!hs->key_share_bytes.empty());
CBB contents, kse_bytes;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_u16(out_compressible, TLSEXT_TYPE_key_share) ||
!CBB_add_u16_length_prefixed(out_compressible, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &kse_bytes) ||
!CBB_add_bytes(&kse_bytes, hs->key_share_bytes.data(),
hs->key_share_bytes.size()) ||
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
!CBB_flush(out_compressible)) {
return false;
}
return true;
}
bool ssl_ext_key_share_parse_serverhello(SSL_HANDSHAKE *hs,
Array<uint8_t> *out_secret,
uint8_t *out_alert, CBS *contents) {
Use KEM terminology in TLS ECDHE and key_share abstractions TLS 1.2 ECDHE and TLS 1.3 key shares were originally designed around Diffie-Hellman-like primitives and use language based on that. Post-quantum replacements do not look like Diffie-Hellman, where each part exchanges a public key, but schemes that work differently can still slot in without protocol changes. We previously came up with our own Offer/Accept/Finish abstraction for early post-quantum experiments, but the NIST constructions are all expressed as KEMs: First, the recipient generates a keypair and sends the public key. Then the sender encapsulates a symmetric secret and sends the ciphertext. Finally, the recipient decapsulates the ciphertext to get the secret. Align our C++ and Go abstractions to this terminology. The functions are now called Generate/Encap/Decap, and the output of Encap is called "ciphertext", which seems to align with what most folks use. (RFC 9180 uses "enc" for "encapsulated key", but they staple a KEM to an AEAD, so "ciphertext" would be ambiguous.) Where variable names refer to parts of the protocol, rather than the the underlying KEM-like construction, I've kept variable names matching the protocol mechanism, so we still talk about "curves" and "key shares", but, when using the post-quantum replacements, the terminology is no longer quite accurate. I've also not yet renamed SSLKeyShare yet, though the name is now inaccurate. Also ideally we'd touch that up so the stateful object is just a KEM private key, for SSLKEMKey. Though at that point, we maybe should just add EVP_KEM and EVP_KEM_KEY APIs to libcrypto. Change-Id: Icbcc1840c5d2dfad210ef4caad2a7c4bf8146553 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/57726 Reviewed-by: Adam Langley <agl@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2 years ago
CBS ciphertext;
uint16_t group_id;
if (!CBS_get_u16(contents, &group_id) ||
Use KEM terminology in TLS ECDHE and key_share abstractions TLS 1.2 ECDHE and TLS 1.3 key shares were originally designed around Diffie-Hellman-like primitives and use language based on that. Post-quantum replacements do not look like Diffie-Hellman, where each part exchanges a public key, but schemes that work differently can still slot in without protocol changes. We previously came up with our own Offer/Accept/Finish abstraction for early post-quantum experiments, but the NIST constructions are all expressed as KEMs: First, the recipient generates a keypair and sends the public key. Then the sender encapsulates a symmetric secret and sends the ciphertext. Finally, the recipient decapsulates the ciphertext to get the secret. Align our C++ and Go abstractions to this terminology. The functions are now called Generate/Encap/Decap, and the output of Encap is called "ciphertext", which seems to align with what most folks use. (RFC 9180 uses "enc" for "encapsulated key", but they staple a KEM to an AEAD, so "ciphertext" would be ambiguous.) Where variable names refer to parts of the protocol, rather than the the underlying KEM-like construction, I've kept variable names matching the protocol mechanism, so we still talk about "curves" and "key shares", but, when using the post-quantum replacements, the terminology is no longer quite accurate. I've also not yet renamed SSLKeyShare yet, though the name is now inaccurate. Also ideally we'd touch that up so the stateful object is just a KEM private key, for SSLKEMKey. Though at that point, we maybe should just add EVP_KEM and EVP_KEM_KEY APIs to libcrypto. Change-Id: Icbcc1840c5d2dfad210ef4caad2a7c4bf8146553 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/57726 Reviewed-by: Adam Langley <agl@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2 years ago
!CBS_get_u16_length_prefixed(contents, &ciphertext) ||
CBS_len(contents) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
SSLKeyShare *key_share = hs->key_shares[0].get();
if (key_share->GroupID() != group_id) {
if (!hs->key_shares[1] || hs->key_shares[1]->GroupID() != group_id) {
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
OPENSSL_PUT_ERROR(SSL, SSL_R_WRONG_CURVE);
return false;
}
key_share = hs->key_shares[1].get();
}
Use KEM terminology in TLS ECDHE and key_share abstractions TLS 1.2 ECDHE and TLS 1.3 key shares were originally designed around Diffie-Hellman-like primitives and use language based on that. Post-quantum replacements do not look like Diffie-Hellman, where each part exchanges a public key, but schemes that work differently can still slot in without protocol changes. We previously came up with our own Offer/Accept/Finish abstraction for early post-quantum experiments, but the NIST constructions are all expressed as KEMs: First, the recipient generates a keypair and sends the public key. Then the sender encapsulates a symmetric secret and sends the ciphertext. Finally, the recipient decapsulates the ciphertext to get the secret. Align our C++ and Go abstractions to this terminology. The functions are now called Generate/Encap/Decap, and the output of Encap is called "ciphertext", which seems to align with what most folks use. (RFC 9180 uses "enc" for "encapsulated key", but they staple a KEM to an AEAD, so "ciphertext" would be ambiguous.) Where variable names refer to parts of the protocol, rather than the the underlying KEM-like construction, I've kept variable names matching the protocol mechanism, so we still talk about "curves" and "key shares", but, when using the post-quantum replacements, the terminology is no longer quite accurate. I've also not yet renamed SSLKeyShare yet, though the name is now inaccurate. Also ideally we'd touch that up so the stateful object is just a KEM private key, for SSLKEMKey. Though at that point, we maybe should just add EVP_KEM and EVP_KEM_KEY APIs to libcrypto. Change-Id: Icbcc1840c5d2dfad210ef4caad2a7c4bf8146553 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/57726 Reviewed-by: Adam Langley <agl@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2 years ago
if (!key_share->Decap(out_secret, out_alert, ciphertext)) {
*out_alert = SSL_AD_INTERNAL_ERROR;
return false;
}
hs->new_session->group_id = group_id;
hs->key_shares[0].reset();
hs->key_shares[1].reset();
return true;
}
bool ssl_ext_key_share_parse_clienthello(SSL_HANDSHAKE *hs, bool *out_found,
Span<const uint8_t> *out_peer_key,
uint8_t *out_alert,
const SSL_CLIENT_HELLO *client_hello) {
// We only support connections that include an ECDHE key exchange.
CBS contents;
if (!ssl_client_hello_get_extension(client_hello, &contents,
TLSEXT_TYPE_key_share)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_MISSING_KEY_SHARE);
*out_alert = SSL_AD_MISSING_EXTENSION;
return false;
}
CBS key_shares;
if (!CBS_get_u16_length_prefixed(&contents, &key_shares) ||
CBS_len(&contents) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
// Find the corresponding key share.
const uint16_t group_id = hs->new_session->group_id;
CBS peer_key;
CBS_init(&peer_key, nullptr, 0);
while (CBS_len(&key_shares) > 0) {
uint16_t id;
CBS peer_key_tmp;
if (!CBS_get_u16(&key_shares, &id) ||
!CBS_get_u16_length_prefixed(&key_shares, &peer_key_tmp) ||
CBS_len(&peer_key_tmp) == 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
if (id == group_id) {
if (CBS_len(&peer_key) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DUPLICATE_KEY_SHARE);
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
return false;
}
peer_key = peer_key_tmp;
// Continue parsing the structure to keep peers honest.
}
}
if (out_peer_key != nullptr) {
*out_peer_key = peer_key;
}
*out_found = CBS_len(&peer_key) != 0;
return true;
}
bool ssl_ext_key_share_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
Use KEM terminology in TLS ECDHE and key_share abstractions TLS 1.2 ECDHE and TLS 1.3 key shares were originally designed around Diffie-Hellman-like primitives and use language based on that. Post-quantum replacements do not look like Diffie-Hellman, where each part exchanges a public key, but schemes that work differently can still slot in without protocol changes. We previously came up with our own Offer/Accept/Finish abstraction for early post-quantum experiments, but the NIST constructions are all expressed as KEMs: First, the recipient generates a keypair and sends the public key. Then the sender encapsulates a symmetric secret and sends the ciphertext. Finally, the recipient decapsulates the ciphertext to get the secret. Align our C++ and Go abstractions to this terminology. The functions are now called Generate/Encap/Decap, and the output of Encap is called "ciphertext", which seems to align with what most folks use. (RFC 9180 uses "enc" for "encapsulated key", but they staple a KEM to an AEAD, so "ciphertext" would be ambiguous.) Where variable names refer to parts of the protocol, rather than the the underlying KEM-like construction, I've kept variable names matching the protocol mechanism, so we still talk about "curves" and "key shares", but, when using the post-quantum replacements, the terminology is no longer quite accurate. I've also not yet renamed SSLKeyShare yet, though the name is now inaccurate. Also ideally we'd touch that up so the stateful object is just a KEM private key, for SSLKEMKey. Though at that point, we maybe should just add EVP_KEM and EVP_KEM_KEY APIs to libcrypto. Change-Id: Icbcc1840c5d2dfad210ef4caad2a7c4bf8146553 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/57726 Reviewed-by: Adam Langley <agl@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2 years ago
CBB entry, ciphertext;
if (!CBB_add_u16(out, TLSEXT_TYPE_key_share) ||
Use KEM terminology in TLS ECDHE and key_share abstractions TLS 1.2 ECDHE and TLS 1.3 key shares were originally designed around Diffie-Hellman-like primitives and use language based on that. Post-quantum replacements do not look like Diffie-Hellman, where each part exchanges a public key, but schemes that work differently can still slot in without protocol changes. We previously came up with our own Offer/Accept/Finish abstraction for early post-quantum experiments, but the NIST constructions are all expressed as KEMs: First, the recipient generates a keypair and sends the public key. Then the sender encapsulates a symmetric secret and sends the ciphertext. Finally, the recipient decapsulates the ciphertext to get the secret. Align our C++ and Go abstractions to this terminology. The functions are now called Generate/Encap/Decap, and the output of Encap is called "ciphertext", which seems to align with what most folks use. (RFC 9180 uses "enc" for "encapsulated key", but they staple a KEM to an AEAD, so "ciphertext" would be ambiguous.) Where variable names refer to parts of the protocol, rather than the the underlying KEM-like construction, I've kept variable names matching the protocol mechanism, so we still talk about "curves" and "key shares", but, when using the post-quantum replacements, the terminology is no longer quite accurate. I've also not yet renamed SSLKeyShare yet, though the name is now inaccurate. Also ideally we'd touch that up so the stateful object is just a KEM private key, for SSLKEMKey. Though at that point, we maybe should just add EVP_KEM and EVP_KEM_KEY APIs to libcrypto. Change-Id: Icbcc1840c5d2dfad210ef4caad2a7c4bf8146553 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/57726 Reviewed-by: Adam Langley <agl@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2 years ago
!CBB_add_u16_length_prefixed(out, &entry) ||
!CBB_add_u16(&entry, hs->new_session->group_id) ||
!CBB_add_u16_length_prefixed(&entry, &ciphertext) ||
!CBB_add_bytes(&ciphertext, hs->key_share_ciphertext.data(),
hs->key_share_ciphertext.size()) ||
!CBB_flush(out)) {
return false;
}
return true;
}
// Supported Versions
//
// https://tools.ietf.org/html/rfc8446#section-4.2.1
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_supported_versions_add_clienthello(
const SSL_HANDSHAKE *hs, CBB *out, CBB *out_compressible,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
if (hs->max_version <= TLS1_2_VERSION) {
return true;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
// supported_versions is compressible in ECH if ClientHelloOuter already
// requires TLS 1.3. Otherwise the extensions differ in the older versions.
if (hs->min_version >= TLS1_3_VERSION) {
out = out_compressible;
}
CBB contents, versions;
if (!CBB_add_u16(out, TLSEXT_TYPE_supported_versions) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_u8_length_prefixed(&contents, &versions)) {
return false;
}
// Add a fake version. See RFC 8701.
if (ssl->ctx->grease_enabled &&
!CBB_add_u16(&versions, ssl_get_grease_value(hs, ssl_grease_version))) {
return false;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
// Encrypted ClientHellos requires TLS 1.3 or later.
uint16_t extra_min_version =
type == ssl_client_hello_inner ? TLS1_3_VERSION : 0;
if (!ssl_add_supported_versions(hs, &versions, extra_min_version) ||
!CBB_flush(out)) {
return false;
}
return true;
}
// Cookie
//
// https://tools.ietf.org/html/rfc8446#section-4.2.2
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_cookie_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
if (hs->cookie.empty()) {
return true;
}
CBB contents, cookie;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_u16(out_compressible, TLSEXT_TYPE_cookie) ||
!CBB_add_u16_length_prefixed(out_compressible, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &cookie) ||
!CBB_add_bytes(&cookie, hs->cookie.data(), hs->cookie.size()) ||
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
!CBB_flush(out_compressible)) {
return false;
}
return true;
}
// Supported Groups
//
// https://tools.ietf.org/html/rfc4492#section-5.1.1
// https://tools.ietf.org/html/rfc8446#section-4.2.7
static bool ext_supported_groups_add_clienthello(const SSL_HANDSHAKE *hs,
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
const SSL *const ssl = hs->ssl;
CBB contents, groups_bytes;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!CBB_add_u16(out_compressible, TLSEXT_TYPE_supported_groups) ||
!CBB_add_u16_length_prefixed(out_compressible, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &groups_bytes)) {
return false;
}
// Add a fake group. See RFC 8701.
if (ssl->ctx->grease_enabled &&
!CBB_add_u16(&groups_bytes,
ssl_get_grease_value(hs, ssl_grease_group))) {
return false;
}
for (uint16_t group : tls1_get_grouplist(hs)) {
if (is_post_quantum_group(group) &&
hs->max_version < TLS1_3_VERSION) {
continue;
}
if (!CBB_add_u16(&groups_bytes, group)) {
return false;
}
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
return CBB_flush(out_compressible);
}
static bool ext_supported_groups_parse_serverhello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
// This extension is not expected to be echoed by servers in TLS 1.2, but some
// BigIP servers send it nonetheless, so do not enforce this.
return true;
}
static bool parse_u16_array(const CBS *cbs, Array<uint16_t> *out) {
CBS copy = *cbs;
if ((CBS_len(&copy) & 1) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
Array<uint16_t> ret;
if (!ret.Init(CBS_len(&copy) / 2)) {
return false;
}
for (size_t i = 0; i < ret.size(); i++) {
if (!CBS_get_u16(&copy, &ret[i])) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
}
assert(CBS_len(&copy) == 0);
*out = std::move(ret);
return true;
}
static bool ext_supported_groups_parse_clienthello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
if (contents == NULL) {
return true;
}
CBS supported_group_list;
if (!CBS_get_u16_length_prefixed(contents, &supported_group_list) ||
CBS_len(&supported_group_list) == 0 ||
CBS_len(contents) != 0 ||
!parse_u16_array(&supported_group_list, &hs->peer_supported_group_list)) {
return false;
}
return true;
}
// QUIC Transport Parameters
static bool ext_quic_transport_params_add_clienthello_impl(
const SSL_HANDSHAKE *hs, CBB *out, bool use_legacy_codepoint) {
if (hs->config->quic_transport_params.empty() && !hs->ssl->quic_method) {
return true;
}
if (hs->config->quic_transport_params.empty() || !hs->ssl->quic_method) {
// QUIC Transport Parameters must be sent over QUIC, and they must not be
// sent over non-QUIC transports. If transport params are set, then
// SSL(_CTX)_set_quic_method must also be called.
OPENSSL_PUT_ERROR(SSL, SSL_R_QUIC_TRANSPORT_PARAMETERS_MISCONFIGURED);
return false;
}
assert(hs->min_version > TLS1_2_VERSION);
if (use_legacy_codepoint != hs->config->quic_use_legacy_codepoint) {
// Do nothing, we'll send the other codepoint.
return true;
}
uint16_t extension_type = TLSEXT_TYPE_quic_transport_parameters;
if (hs->config->quic_use_legacy_codepoint) {
extension_type = TLSEXT_TYPE_quic_transport_parameters_legacy;
}
CBB contents;
if (!CBB_add_u16(out, extension_type) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_bytes(&contents, hs->config->quic_transport_params.data(),
hs->config->quic_transport_params.size()) ||
!CBB_flush(out)) {
return false;
}
return true;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_quic_transport_params_add_clienthello(
const SSL_HANDSHAKE *hs, CBB *out, CBB *out_compressible,
ssl_client_hello_type_t type) {
return ext_quic_transport_params_add_clienthello_impl(
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
hs, out_compressible, /*use_legacy_codepoint=*/false);
}
static bool ext_quic_transport_params_add_clienthello_legacy(
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
const SSL_HANDSHAKE *hs, CBB *out, CBB *out_compressible,
ssl_client_hello_type_t type) {
return ext_quic_transport_params_add_clienthello_impl(
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
hs, out_compressible, /*use_legacy_codepoint=*/true);
}
static bool ext_quic_transport_params_parse_serverhello_impl(
SSL_HANDSHAKE *hs, uint8_t *out_alert, CBS *contents,
bool used_legacy_codepoint) {
SSL *const ssl = hs->ssl;
if (contents == nullptr) {
if (used_legacy_codepoint != hs->config->quic_use_legacy_codepoint) {
// Silently ignore because we expect the other QUIC codepoint.
return true;
}
if (!ssl->quic_method) {
return true;
}
*out_alert = SSL_AD_MISSING_EXTENSION;
return false;
}
// The extensions parser will check for unsolicited extensions before
// calling the callback.
assert(ssl->quic_method != nullptr);
assert(ssl_protocol_version(ssl) == TLS1_3_VERSION);
assert(used_legacy_codepoint == hs->config->quic_use_legacy_codepoint);
return ssl->s3->peer_quic_transport_params.CopyFrom(*contents);
}
static bool ext_quic_transport_params_parse_serverhello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
return ext_quic_transport_params_parse_serverhello_impl(
hs, out_alert, contents, /*used_legacy_codepoint=*/false);
}
static bool ext_quic_transport_params_parse_serverhello_legacy(
SSL_HANDSHAKE *hs, uint8_t *out_alert, CBS *contents) {
return ext_quic_transport_params_parse_serverhello_impl(
hs, out_alert, contents, /*used_legacy_codepoint=*/true);
}
static bool ext_quic_transport_params_parse_clienthello_impl(
SSL_HANDSHAKE *hs, uint8_t *out_alert, CBS *contents,
bool used_legacy_codepoint) {
SSL *const ssl = hs->ssl;
if (!contents) {
if (!ssl->quic_method) {
if (hs->config->quic_transport_params.empty()) {
return true;
}
// QUIC transport parameters must not be set if |ssl| is not configured
// for QUIC.
OPENSSL_PUT_ERROR(SSL, SSL_R_QUIC_TRANSPORT_PARAMETERS_MISCONFIGURED);
*out_alert = SSL_AD_INTERNAL_ERROR;
return false;
}
if (used_legacy_codepoint != hs->config->quic_use_legacy_codepoint) {
// Silently ignore because we expect the other QUIC codepoint.
return true;
}
*out_alert = SSL_AD_MISSING_EXTENSION;
return false;
}
if (!ssl->quic_method) {
if (used_legacy_codepoint) {
// Ignore the legacy private-use codepoint because that could be sent
// to mean something else than QUIC transport parameters.
return true;
}
// Fail if we received the codepoint registered with IANA for QUIC
// because that is not allowed outside of QUIC.
*out_alert = SSL_AD_UNSUPPORTED_EXTENSION;
return false;
}
assert(ssl_protocol_version(ssl) == TLS1_3_VERSION);
if (used_legacy_codepoint != hs->config->quic_use_legacy_codepoint) {
// Silently ignore because we expect the other QUIC codepoint.
return true;
}
return ssl->s3->peer_quic_transport_params.CopyFrom(*contents);
}
static bool ext_quic_transport_params_parse_clienthello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
return ext_quic_transport_params_parse_clienthello_impl(
hs, out_alert, contents, /*used_legacy_codepoint=*/false);
}
static bool ext_quic_transport_params_parse_clienthello_legacy(
SSL_HANDSHAKE *hs, uint8_t *out_alert, CBS *contents) {
return ext_quic_transport_params_parse_clienthello_impl(
hs, out_alert, contents, /*used_legacy_codepoint=*/true);
}
static bool ext_quic_transport_params_add_serverhello_impl(
SSL_HANDSHAKE *hs, CBB *out, bool use_legacy_codepoint) {
if (hs->ssl->quic_method == nullptr && use_legacy_codepoint) {
// Ignore the legacy private-use codepoint because that could be sent
// to mean something else than QUIC transport parameters.
return true;
}
assert(hs->ssl->quic_method != nullptr);
if (hs->config->quic_transport_params.empty()) {
// Transport parameters must be set when using QUIC.
OPENSSL_PUT_ERROR(SSL, SSL_R_QUIC_TRANSPORT_PARAMETERS_MISCONFIGURED);
return false;
}
if (use_legacy_codepoint != hs->config->quic_use_legacy_codepoint) {
// Do nothing, we'll send the other codepoint.
return true;
}
uint16_t extension_type = TLSEXT_TYPE_quic_transport_parameters;
if (hs->config->quic_use_legacy_codepoint) {
extension_type = TLSEXT_TYPE_quic_transport_parameters_legacy;
}
CBB contents;
if (!CBB_add_u16(out, extension_type) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_bytes(&contents, hs->config->quic_transport_params.data(),
hs->config->quic_transport_params.size()) ||
!CBB_flush(out)) {
return false;
}
return true;
}
static bool ext_quic_transport_params_add_serverhello(SSL_HANDSHAKE *hs,
CBB *out) {
return ext_quic_transport_params_add_serverhello_impl(
hs, out, /*use_legacy_codepoint=*/false);
}
static bool ext_quic_transport_params_add_serverhello_legacy(SSL_HANDSHAKE *hs,
CBB *out) {
return ext_quic_transport_params_add_serverhello_impl(
hs, out, /*use_legacy_codepoint=*/true);
}
// Delegated credentials.
//
// https://tools.ietf.org/html/draft-ietf-tls-subcerts
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ext_delegated_credential_add_clienthello(
const SSL_HANDSHAKE *hs, CBB *out, CBB *out_compressible,
ssl_client_hello_type_t type) {
return true;
}
static bool ext_delegated_credential_parse_clienthello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
if (contents == nullptr || ssl_protocol_version(hs->ssl) < TLS1_3_VERSION) {
// Don't use delegated credentials unless we're negotiating TLS 1.3 or
// higher.
return true;
}
// The contents of the extension are the signature algorithms the client will
// accept for a delegated credential.
CBS sigalg_list;
if (!CBS_get_u16_length_prefixed(contents, &sigalg_list) ||
CBS_len(&sigalg_list) == 0 ||
CBS_len(contents) != 0 ||
!parse_u16_array(&sigalg_list, &hs->peer_delegated_credential_sigalgs)) {
return false;
}
hs->delegated_credential_requested = true;
return true;
}
// Certificate compression
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool cert_compression_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
bool first = true;
CBB contents, algs;
for (const auto &alg : hs->ssl->ctx->cert_compression_algs) {
if (alg.decompress == nullptr) {
continue;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (first &&
(!CBB_add_u16(out_compressible, TLSEXT_TYPE_cert_compression) ||
!CBB_add_u16_length_prefixed(out_compressible, &contents) ||
!CBB_add_u8_length_prefixed(&contents, &algs))) {
return false;
}
first = false;
if (!CBB_add_u16(&algs, alg.alg_id)) {
return false;
}
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
return first || CBB_flush(out_compressible);
}
static bool cert_compression_parse_serverhello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
if (contents == nullptr) {
return true;
}
// The server may not echo this extension. Any server to client negotiation is
// advertised in the CertificateRequest message.
return false;
}
static bool cert_compression_parse_clienthello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
if (contents == nullptr) {
return true;
}
const SSL_CTX *ctx = hs->ssl->ctx.get();
const size_t num_algs = ctx->cert_compression_algs.size();
CBS alg_ids;
if (!CBS_get_u8_length_prefixed(contents, &alg_ids) ||
CBS_len(contents) != 0 ||
CBS_len(&alg_ids) == 0 ||
CBS_len(&alg_ids) % 2 == 1) {
return false;
}
const size_t num_given_alg_ids = CBS_len(&alg_ids) / 2;
Array<uint16_t> given_alg_ids;
if (!given_alg_ids.Init(num_given_alg_ids)) {
return false;
}
size_t best_index = num_algs;
size_t given_alg_idx = 0;
while (CBS_len(&alg_ids) > 0) {
uint16_t alg_id;
if (!CBS_get_u16(&alg_ids, &alg_id)) {
return false;
}
given_alg_ids[given_alg_idx++] = alg_id;
for (size_t i = 0; i < num_algs; i++) {
const auto &alg = ctx->cert_compression_algs[i];
if (alg.alg_id == alg_id && alg.compress != nullptr) {
if (i < best_index) {
best_index = i;
}
break;
}
}
}
qsort(given_alg_ids.data(), given_alg_ids.size(), sizeof(uint16_t),
compare_uint16_t);
for (size_t i = 1; i < num_given_alg_ids; i++) {
if (given_alg_ids[i - 1] == given_alg_ids[i]) {
return false;
}
}
if (best_index < num_algs &&
ssl_protocol_version(hs->ssl) >= TLS1_3_VERSION) {
hs->cert_compression_negotiated = true;
hs->cert_compression_alg_id = ctx->cert_compression_algs[best_index].alg_id;
}
return true;
}
static bool cert_compression_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
return true;
}
// Application-level Protocol Settings
//
// https://tools.ietf.org/html/draft-vvv-tls-alps-01
bool ssl_get_local_application_settings(const SSL_HANDSHAKE *hs,
Span<const uint8_t> *out_settings,
Span<const uint8_t> protocol) {
for (const ALPSConfig &config : hs->config->alps_configs) {
if (protocol == config.protocol) {
*out_settings = config.settings;
return true;
}
}
return false;
}
static bool ext_alps_add_clienthello_impl(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type,
bool use_new_codepoint) {
const SSL *const ssl = hs->ssl;
if (// ALPS requires TLS 1.3.
hs->max_version < TLS1_3_VERSION ||
// Do not offer ALPS without ALPN.
hs->config->alpn_client_proto_list.empty() ||
// Do not offer ALPS if not configured.
hs->config->alps_configs.empty() ||
// Do not offer ALPS on renegotiation handshakes.
ssl->s3->initial_handshake_complete) {
return true;
}
if (use_new_codepoint != hs->config->alps_use_new_codepoint) {
// Do nothing, we'll send the other codepoint.
return true;
}
uint16_t extension_type = TLSEXT_TYPE_application_settings_old;
if (hs->config->alps_use_new_codepoint) {
extension_type = TLSEXT_TYPE_application_settings;
}
CBB contents, proto_list, proto;
if (!CBB_add_u16(out_compressible, extension_type) ||
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
!CBB_add_u16_length_prefixed(out_compressible, &contents) ||
!CBB_add_u16_length_prefixed(&contents, &proto_list)) {
return false;
}
for (const ALPSConfig &config : hs->config->alps_configs) {
if (!CBB_add_u8_length_prefixed(&proto_list, &proto) ||
!CBB_add_bytes(&proto, config.protocol.data(),
config.protocol.size())) {
return false;
}
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
return CBB_flush(out_compressible);
}
static bool ext_alps_add_clienthello(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
return ext_alps_add_clienthello_impl(hs, out, out_compressible, type,
/*use_new_codepoint=*/true);
}
static bool ext_alps_add_clienthello_old(const SSL_HANDSHAKE *hs, CBB *out,
CBB *out_compressible,
ssl_client_hello_type_t type) {
return ext_alps_add_clienthello_impl(hs, out, out_compressible, type,
/*use_new_codepoint=*/false);
}
static bool ext_alps_parse_serverhello_impl(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents,
bool use_new_codepoint) {
SSL *const ssl = hs->ssl;
if (contents == nullptr) {
return true;
}
assert(!ssl->s3->initial_handshake_complete);
assert(!hs->config->alpn_client_proto_list.empty());
assert(!hs->config->alps_configs.empty());
assert(use_new_codepoint == hs->config->alps_use_new_codepoint);
// ALPS requires TLS 1.3.
if (ssl_protocol_version(ssl) < TLS1_3_VERSION) {
*out_alert = SSL_AD_UNSUPPORTED_EXTENSION;
OPENSSL_PUT_ERROR(SSL, SSL_R_UNEXPECTED_EXTENSION);
return false;
}
// Note extension callbacks may run in any order, so we defer checking
// consistency with ALPN to |ssl_check_serverhello_tlsext|.
if (!hs->new_session->peer_application_settings.CopyFrom(*contents)) {
*out_alert = SSL_AD_INTERNAL_ERROR;
return false;
}
hs->new_session->has_application_settings = true;
return true;
}
static bool ext_alps_parse_serverhello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
return ext_alps_parse_serverhello_impl(hs, out_alert, contents,
/*use_new_codepoint=*/true);
}
static bool ext_alps_parse_serverhello_old(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents) {
return ext_alps_parse_serverhello_impl(hs, out_alert, contents,
/*use_new_codepoint=*/false);
}
static bool ext_alps_add_serverhello_impl(SSL_HANDSHAKE *hs, CBB *out,
bool use_new_codepoint) {
SSL *const ssl = hs->ssl;
// If early data is accepted, we omit the ALPS extension. It is implicitly
// carried over from the previous connection.
if (hs->new_session == nullptr ||
!hs->new_session->has_application_settings ||
ssl->s3->early_data_accepted) {
return true;
}
if (use_new_codepoint != hs->config->alps_use_new_codepoint) {
// Do nothing, we'll send the other codepoint.
return true;
}
uint16_t extension_type = TLSEXT_TYPE_application_settings_old;
if (hs->config->alps_use_new_codepoint) {
extension_type = TLSEXT_TYPE_application_settings;
}
CBB contents;
if (!CBB_add_u16(out, extension_type) ||
!CBB_add_u16_length_prefixed(out, &contents) ||
!CBB_add_bytes(&contents,
hs->new_session->local_application_settings.data(),
hs->new_session->local_application_settings.size()) ||
!CBB_flush(out)) {
return false;
}
return true;
}
static bool ext_alps_add_serverhello(SSL_HANDSHAKE *hs, CBB *out) {
return ext_alps_add_serverhello_impl(hs, out, /*use_new_codepoint=*/true);
}
static bool ext_alps_add_serverhello_old(SSL_HANDSHAKE *hs, CBB *out) {
return ext_alps_add_serverhello_impl(hs, out, /*use_new_codepoint=*/false);
}
bool ssl_negotiate_alps(SSL_HANDSHAKE *hs, uint8_t *out_alert,
const SSL_CLIENT_HELLO *client_hello) {
SSL *const ssl = hs->ssl;
if (ssl->s3->alpn_selected.empty()) {
return true;
}
// If we negotiate ALPN over TLS 1.3, try to negotiate ALPS.
CBS alps_contents;
Span<const uint8_t> settings;
uint16_t extension_type = TLSEXT_TYPE_application_settings_old;
if (hs->config->alps_use_new_codepoint) {
extension_type = TLSEXT_TYPE_application_settings;
}
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION &&
ssl_get_local_application_settings(hs, &settings,
ssl->s3->alpn_selected) &&
ssl_client_hello_get_extension(client_hello, &alps_contents,
extension_type)) {
// Check if the client supports ALPS with the selected ALPN.
bool found = false;
CBS alps_list;
if (!CBS_get_u16_length_prefixed(&alps_contents, &alps_list) ||
CBS_len(&alps_contents) != 0 ||
CBS_len(&alps_list) == 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
while (CBS_len(&alps_list) > 0) {
CBS protocol_name;
if (!CBS_get_u8_length_prefixed(&alps_list, &protocol_name) ||
// Empty protocol names are forbidden.
CBS_len(&protocol_name) == 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
if (protocol_name == MakeConstSpan(ssl->s3->alpn_selected)) {
found = true;
}
}
// Negotiate ALPS if both client also supports ALPS for this protocol.
if (found) {
hs->new_session->has_application_settings = true;
if (!hs->new_session->local_application_settings.CopyFrom(settings)) {
*out_alert = SSL_AD_INTERNAL_ERROR;
return false;
}
}
}
return true;
}
// kExtensions contains all the supported extensions.
static const struct tls_extension kExtensions[] = {
{
TLSEXT_TYPE_server_name,
ext_sni_add_clienthello,
ext_sni_parse_serverhello,
ext_sni_parse_clienthello,
ext_sni_add_serverhello,
},
{
TLSEXT_TYPE_encrypted_client_hello,
ext_ech_add_clienthello,
ext_ech_parse_serverhello,
ext_ech_parse_clienthello,
ext_ech_add_serverhello,
},
{
TLSEXT_TYPE_extended_master_secret,
ext_ems_add_clienthello,
ext_ems_parse_serverhello,
ext_ems_parse_clienthello,
ext_ems_add_serverhello,
},
{
TLSEXT_TYPE_renegotiate,
ext_ri_add_clienthello,
ext_ri_parse_serverhello,
ext_ri_parse_clienthello,
ext_ri_add_serverhello,
},
{
TLSEXT_TYPE_supported_groups,
ext_supported_groups_add_clienthello,
ext_supported_groups_parse_serverhello,
ext_supported_groups_parse_clienthello,
dont_add_serverhello,
},
{
TLSEXT_TYPE_ec_point_formats,
ext_ec_point_add_clienthello,
ext_ec_point_parse_serverhello,
ext_ec_point_parse_clienthello,
ext_ec_point_add_serverhello,
},
{
TLSEXT_TYPE_session_ticket,
ext_ticket_add_clienthello,
ext_ticket_parse_serverhello,
// Ticket extension client parsing is handled in ssl_session.c
ignore_parse_clienthello,
ext_ticket_add_serverhello,
},
{
TLSEXT_TYPE_application_layer_protocol_negotiation,
ext_alpn_add_clienthello,
ext_alpn_parse_serverhello,
// ALPN is negotiated late in |ssl_negotiate_alpn|.
ignore_parse_clienthello,
ext_alpn_add_serverhello,
},
{
TLSEXT_TYPE_status_request,
ext_ocsp_add_clienthello,
ext_ocsp_parse_serverhello,
ext_ocsp_parse_clienthello,
ext_ocsp_add_serverhello,
},
{
TLSEXT_TYPE_signature_algorithms,
ext_sigalgs_add_clienthello,
forbid_parse_serverhello,
ext_sigalgs_parse_clienthello,
dont_add_serverhello,
},
{
TLSEXT_TYPE_next_proto_neg,
ext_npn_add_clienthello,
ext_npn_parse_serverhello,
ext_npn_parse_clienthello,
ext_npn_add_serverhello,
},
{
TLSEXT_TYPE_certificate_timestamp,
ext_sct_add_clienthello,
ext_sct_parse_serverhello,
ext_sct_parse_clienthello,
ext_sct_add_serverhello,
},
{
TLSEXT_TYPE_channel_id,
ext_channel_id_add_clienthello,
ext_channel_id_parse_serverhello,
ext_channel_id_parse_clienthello,
ext_channel_id_add_serverhello,
},
{
TLSEXT_TYPE_srtp,
ext_srtp_add_clienthello,
ext_srtp_parse_serverhello,
ext_srtp_parse_clienthello,
ext_srtp_add_serverhello,
},
{
TLSEXT_TYPE_key_share,
ext_key_share_add_clienthello,
forbid_parse_serverhello,
ignore_parse_clienthello,
dont_add_serverhello,
},
{
TLSEXT_TYPE_psk_key_exchange_modes,
ext_psk_key_exchange_modes_add_clienthello,
forbid_parse_serverhello,
ext_psk_key_exchange_modes_parse_clienthello,
dont_add_serverhello,
},
{
TLSEXT_TYPE_early_data,
ext_early_data_add_clienthello,
ext_early_data_parse_serverhello,
ext_early_data_parse_clienthello,
ext_early_data_add_serverhello,
},
{
TLSEXT_TYPE_supported_versions,
ext_supported_versions_add_clienthello,
forbid_parse_serverhello,
ignore_parse_clienthello,
dont_add_serverhello,
},
{
TLSEXT_TYPE_cookie,
ext_cookie_add_clienthello,
forbid_parse_serverhello,
ignore_parse_clienthello,
dont_add_serverhello,
},
{
TLSEXT_TYPE_quic_transport_parameters,
ext_quic_transport_params_add_clienthello,
ext_quic_transport_params_parse_serverhello,
ext_quic_transport_params_parse_clienthello,
ext_quic_transport_params_add_serverhello,
},
{
TLSEXT_TYPE_quic_transport_parameters_legacy,
ext_quic_transport_params_add_clienthello_legacy,
ext_quic_transport_params_parse_serverhello_legacy,
ext_quic_transport_params_parse_clienthello_legacy,
ext_quic_transport_params_add_serverhello_legacy,
},
{
TLSEXT_TYPE_cert_compression,
cert_compression_add_clienthello,
cert_compression_parse_serverhello,
cert_compression_parse_clienthello,
cert_compression_add_serverhello,
},
{
TLSEXT_TYPE_delegated_credential,
ext_delegated_credential_add_clienthello,
forbid_parse_serverhello,
ext_delegated_credential_parse_clienthello,
dont_add_serverhello,
},
{
TLSEXT_TYPE_application_settings,
ext_alps_add_clienthello,
ext_alps_parse_serverhello,
// ALPS is negotiated late in |ssl_negotiate_alpn|.
ignore_parse_clienthello,
ext_alps_add_serverhello,
},
{
TLSEXT_TYPE_application_settings_old,
ext_alps_add_clienthello_old,
ext_alps_parse_serverhello_old,
// ALPS is negotiated late in |ssl_negotiate_alpn|.
ignore_parse_clienthello,
ext_alps_add_serverhello_old,
},
};
#define kNumExtensions (sizeof(kExtensions) / sizeof(struct tls_extension))
static_assert(kNumExtensions <=
sizeof(((SSL_HANDSHAKE *)NULL)->extensions.sent) * 8,
"too many extensions for sent bitset");
static_assert(kNumExtensions <=
sizeof(((SSL_HANDSHAKE *)NULL)->extensions.received) * 8,
"too many extensions for received bitset");
bool ssl_setup_extension_permutation(SSL_HANDSHAKE *hs) {
if (!hs->config->permute_extensions) {
return true;
}
static_assert(kNumExtensions <= UINT8_MAX,
"extensions_permutation type is too small");
uint32_t seeds[kNumExtensions - 1];
Array<uint8_t> permutation;
if (!RAND_bytes(reinterpret_cast<uint8_t *>(seeds), sizeof(seeds)) ||
!permutation.Init(kNumExtensions)) {
return false;
}
for (size_t i = 0; i < kNumExtensions; i++) {
permutation[i] = i;
}
for (size_t i = kNumExtensions - 1; i > 0; i--) {
// Set element |i| to a randomly-selected element 0 <= j <= i.
std::swap(permutation[i], permutation[seeds[i - 1] % (i + 1)]);
}
hs->extension_permutation = std::move(permutation);
return true;
}
static const struct tls_extension *tls_extension_find(uint32_t *out_index,
uint16_t value) {
unsigned i;
for (i = 0; i < kNumExtensions; i++) {
if (kExtensions[i].value == value) {
*out_index = i;
return &kExtensions[i];
}
}
return NULL;
}
static bool add_padding_extension(CBB *cbb, uint16_t ext, size_t len) {
CBB child;
if (!CBB_add_u16(cbb, ext) || //
!CBB_add_u16_length_prefixed(cbb, &child) ||
!CBB_add_zeros(&child, len)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
return CBB_flush(cbb);
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
static bool ssl_add_clienthello_tlsext_inner(SSL_HANDSHAKE *hs, CBB *out,
CBB *out_encoded,
bool *out_needs_psk_binder) {
// When writing ClientHelloInner, we construct the real and encoded
// ClientHellos concurrently, to handle compression. Uncompressed extensions
// are written to |extensions| and copied to |extensions_encoded|. Compressed
// extensions are buffered in |compressed| and written to the end. (ECH can
// only compress continguous extensions.)
SSL *const ssl = hs->ssl;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
bssl::ScopedCBB compressed, outer_extensions;
CBB extensions, extensions_encoded;
if (!CBB_add_u16_length_prefixed(out, &extensions) ||
!CBB_add_u16_length_prefixed(out_encoded, &extensions_encoded) ||
!CBB_init(compressed.get(), 64) ||
!CBB_init(outer_extensions.get(), 64)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
hs->inner_extensions_sent = 0;
if (ssl->ctx->grease_enabled) {
// Add a fake empty extension. See RFC 8701. This always matches
// |ssl_add_clienthello_tlsext|, so compress it.
uint16_t grease_ext = ssl_get_grease_value(hs, ssl_grease_extension1);
if (!add_padding_extension(compressed.get(), grease_ext, 0) ||
!CBB_add_u16(outer_extensions.get(), grease_ext)) {
return false;
}
}
for (size_t unpermuted = 0; unpermuted < kNumExtensions; unpermuted++) {
size_t i = hs->extension_permutation.empty()
? unpermuted
: hs->extension_permutation[unpermuted];
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
const size_t len_before = CBB_len(&extensions);
const size_t len_compressed_before = CBB_len(compressed.get());
if (!kExtensions[i].add_clienthello(hs, &extensions, compressed.get(),
ssl_client_hello_inner)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_ERROR_ADDING_EXTENSION);
ERR_add_error_dataf("extension %u", (unsigned)kExtensions[i].value);
return false;
}
const size_t bytes_written = CBB_len(&extensions) - len_before;
const size_t bytes_written_compressed =
CBB_len(compressed.get()) - len_compressed_before;
// The callback may write to at most one output.
assert(bytes_written == 0 || bytes_written_compressed == 0);
if (bytes_written != 0 || bytes_written_compressed != 0) {
hs->inner_extensions_sent |= (1u << i);
}
// If compressed, update the running ech_outer_extensions extension.
if (bytes_written_compressed != 0 &&
!CBB_add_u16(outer_extensions.get(), kExtensions[i].value)) {
return false;
}
}
if (ssl->ctx->grease_enabled) {
// Add a fake non-empty extension. See RFC 8701. This always matches
// |ssl_add_clienthello_tlsext|, so compress it.
uint16_t grease_ext = ssl_get_grease_value(hs, ssl_grease_extension2);
if (!add_padding_extension(compressed.get(), grease_ext, 1) ||
!CBB_add_u16(outer_extensions.get(), grease_ext)) {
return false;
}
}
// Uncompressed extensions are encoded as-is.
if (!CBB_add_bytes(&extensions_encoded, CBB_data(&extensions),
CBB_len(&extensions))) {
return false;
}
// Flush all the compressed extensions.
if (CBB_len(compressed.get()) != 0) {
CBB extension, child;
// Copy them as-is in the real ClientHelloInner.
if (!CBB_add_bytes(&extensions, CBB_data(compressed.get()),
CBB_len(compressed.get())) ||
// Replace with ech_outer_extensions in the encoded form.
!CBB_add_u16(&extensions_encoded, TLSEXT_TYPE_ech_outer_extensions) ||
!CBB_add_u16_length_prefixed(&extensions_encoded, &extension) ||
!CBB_add_u8_length_prefixed(&extension, &child) ||
!CBB_add_bytes(&child, CBB_data(outer_extensions.get()),
CBB_len(outer_extensions.get())) ||
!CBB_flush(&extensions_encoded)) {
return false;
}
}
// The PSK extension must be last. It is never compressed. Note, if there is a
// binder, the caller will need to update both ClientHelloInner and
// EncodedClientHelloInner after computing it.
const size_t len_before = CBB_len(&extensions);
if (!ext_pre_shared_key_add_clienthello(hs, &extensions, out_needs_psk_binder,
ssl_client_hello_inner) ||
!CBB_add_bytes(&extensions_encoded, CBB_data(&extensions) + len_before,
CBB_len(&extensions) - len_before) ||
!CBB_flush(out) || //
!CBB_flush(out_encoded)) {
return false;
}
return true;
}
bool ssl_add_clienthello_tlsext(SSL_HANDSHAKE *hs, CBB *out, CBB *out_encoded,
bool *out_needs_psk_binder,
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
ssl_client_hello_type_t type,
size_t header_len) {
*out_needs_psk_binder = false;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (type == ssl_client_hello_inner) {
return ssl_add_clienthello_tlsext_inner(hs, out, out_encoded,
out_needs_psk_binder);
}
assert(out_encoded == nullptr); // Only ClientHelloInner needs two outputs.
SSL *const ssl = hs->ssl;
CBB extensions;
if (!CBB_add_u16_length_prefixed(out, &extensions)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
// Note we may send multiple ClientHellos for DTLS HelloVerifyRequest and TLS
// 1.3 HelloRetryRequest. For the latter, the extensions may change, so it is
// important to reset this value.
hs->extensions.sent = 0;
// Add a fake empty extension. See RFC 8701.
if (ssl->ctx->grease_enabled &&
!add_padding_extension(
&extensions, ssl_get_grease_value(hs, ssl_grease_extension1), 0)) {
return false;
}
bool last_was_empty = false;
for (size_t unpermuted = 0; unpermuted < kNumExtensions; unpermuted++) {
size_t i = hs->extension_permutation.empty()
? unpermuted
: hs->extension_permutation[unpermuted];
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
const size_t len_before = CBB_len(&extensions);
if (!kExtensions[i].add_clienthello(hs, &extensions, &extensions, type)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_ERROR_ADDING_EXTENSION);
ERR_add_error_dataf("extension %u", (unsigned)kExtensions[i].value);
return false;
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
}
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
const size_t bytes_written = CBB_len(&extensions) - len_before;
if (bytes_written != 0) {
hs->extensions.sent |= (1u << i);
}
// If the difference in lengths is only four bytes then the extension had
// an empty body.
last_was_empty = (bytes_written == 4);
}
if (ssl->ctx->grease_enabled) {
// Add a fake non-empty extension. See RFC 8701.
if (!add_padding_extension(
&extensions, ssl_get_grease_value(hs, ssl_grease_extension2), 1)) {
return false;
}
last_was_empty = false;
}
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
// In cleartext ClientHellos, we add the padding extension to work around
// bugs. We also apply this padding to ClientHelloOuter, to keep the wire
// images aligned.
size_t psk_extension_len = ext_pre_shared_key_clienthello_length(hs, type);
if (!SSL_is_dtls(ssl) && !ssl->quic_method &&
!ssl->s3->used_hello_retry_request) {
Update to draft-ietf-tls-esni-13. Later CLs will clean up the ClientHello construction a bit (draft-12 avoids computing ClientHelloOuter twice). I suspect the transcript handling on the client can also be simpler, but I'll see what's convenient after I've changed how ClientHelloOuter is constructed. Changes of note between draft-10 and draft-13: - There is now an ECH confirmation signal in both HRR and SH. We don't actually make much use of this in our client right now, but it resolves a bunch of weird issues around HRR, including edge cases if HRR applies to one ClientHello but not the other. - The confirmation signal no longer depends on key_share and PSK, so we don't have to work around a weird ordering issue. - ech_is_inner is now folded into the main encrypted_client_hello code point. This works better with some stuff around HRR. - Padding is moved from the padding extension, computed with ClientHelloInner, to something we fill in afterwards. This makes it easier to pad up the whole thing to a multiple of 32. I've accordingly updated to the latest recommended padding construction, and updated the GREASE logic to match. - ech_outer_extensions is much easier to process because the order is required to be consistent. We were doing that anyway, and now a simple linear scan works. - ClientHelloOuterAAD now uses an all zero placeholder payload of the same length. This lets us simplify the server code, but, for now, I've kept the client code the same. I'll follow this up with a CL to avoid computing ClientHelloOuter twice. - ClientHelloOuterAAD is allowed to contain a placeholder PSK. I haven't filled that in and will do it in a follow-up CL. Bug: 275 Change-Id: I7464345125c53968b2fe692f9268e392120fc2eb Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48912 Commit-Queue: David Benjamin <davidben@google.com> Reviewed-by: Adam Langley <agl@google.com>
3 years ago
header_len +=
SSL3_HM_HEADER_LENGTH + 2 + CBB_len(&extensions) + psk_extension_len;
size_t padding_len = 0;
// The final extension must be non-empty. WebSphere Application
// Server 7.0 is intolerant to the last extension being zero-length. See
// https://crbug.com/363583.
if (last_was_empty && psk_extension_len == 0) {
padding_len = 1;
// The addition of the padding extension may push us into the F5 bug.
header_len += 4 + padding_len;
}
// Add padding to workaround bugs in F5 terminators. See RFC 7685.
//
// NB: because this code works out the length of all existing extensions
// it MUST always appear last (save for any PSK extension).
if (header_len > 0xff && header_len < 0x200) {
// If our calculations already included a padding extension, remove that
// factor because we're about to change its length.
if (padding_len != 0) {
header_len -= 4 + padding_len;
}
padding_len = 0x200 - header_len;
// Extensions take at least four bytes to encode. Always include at least
// one byte of data if including the extension. WebSphere Application
// Server 7.0 is intolerant to the last extension being zero-length. See
// https://crbug.com/363583.
if (padding_len >= 4 + 1) {
padding_len -= 4;
} else {
padding_len = 1;
}
}
if (padding_len != 0 &&
!add_padding_extension(&extensions, TLSEXT_TYPE_padding, padding_len)) {
return false;
}
}
// The PSK extension must be last, including after the padding.
const size_t len_before = CBB_len(&extensions);
Add most of an ECH client implementation. Based on an initial implementation by Dan McArdle at https://boringssl-review.googlesource.com/c/boringssl/+/46784 This CL contains most of a client implementation for draft-ietf-tls-esni-10. The pieces missing so far, which will be done in follow-up CLs are: 1. While the ClientHelloInner is padded, the server Certificate message is not. I'll add that once we resolve the spec discussions on how to do that. (We were originally going to use TLS record-level padding, but that doesn't work well with QUIC.) 2. The client should check the public name is a valid DNS name before copying it into ClientHelloOuter.server_name. 3. The ClientHelloOuter handshake flow is not yet implemented. This CL can detect when the server selects ClientHelloOuter, but for now the handshake immediately fails. A follow-up CL will remove that logic and instead add the APIs and extra checks needed. Otherwise, this should be complete, including padding and compression. The main interesting point design-wise is that we run through ClientHello construction multiple times. We need to construct ClientHelloInner and ClientHelloOuter. Then each of those has slight variants: EncodedClientHelloInner is the compressed form, and ClientHelloOuterAAD just has the ECH extension erased to avoid a circular dependency. I've computed ClientHelloInner and EncodedClientHelloInner concurrently because the compression scheme requires shifting the extensions around to be contiguous. However, I've computed ClientHelloOuterAAD and ClientHelloOuter by running through the logic twice. This probably can be done better, but the next draft revises the construction anyway, so I'm thinking I'll rework it then. (In the next draft, we use a placeholder payload of the same length, so we can construct the ClientHello once and fill in the payload.) Additionally, now that we have a client available in ssl_test, this adds a threading test to confirm that SSL_CTX_set1_ech_keys is properly synchronized. (Confirmed that, if I drop the lock in SSL_CTX_set1_ech_keys, TSan notices.) Change-Id: Icaff68b595035bdcc73c468ff638e67c84239ef4 Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48004 Reviewed-by: Adam Langley <agl@google.com>
4 years ago
if (!ext_pre_shared_key_add_clienthello(hs, &extensions, out_needs_psk_binder,
type)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
assert(psk_extension_len == CBB_len(&extensions) - len_before);
(void)len_before; // |assert| is omitted in release builds.
// Discard empty extensions blocks.
if (CBB_len(&extensions) == 0) {
CBB_discard_child(out);
}
return CBB_flush(out);
}
bool ssl_add_serverhello_tlsext(SSL_HANDSHAKE *hs, CBB *out) {
SSL *const ssl = hs->ssl;
CBB extensions;
if (!CBB_add_u16_length_prefixed(out, &extensions)) {
goto err;
}
for (unsigned i = 0; i < kNumExtensions; i++) {
if (!(hs->extensions.received & (1u << i))) {
// Don't send extensions that were not received.
continue;
}
if (!kExtensions[i].add_serverhello(hs, &extensions)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_ERROR_ADDING_EXTENSION);
ERR_add_error_dataf("extension %u", (unsigned)kExtensions[i].value);
goto err;
}
}
// Discard empty extensions blocks before TLS 1.3.
if (ssl_protocol_version(ssl) < TLS1_3_VERSION &&
CBB_len(&extensions) == 0) {
CBB_discard_child(out);
}
return CBB_flush(out);
err:
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
static bool ssl_scan_clienthello_tlsext(SSL_HANDSHAKE *hs,
const SSL_CLIENT_HELLO *client_hello,
int *out_alert) {
hs->extensions.received = 0;
CBS extensions;
CBS_init(&extensions, client_hello->extensions, client_hello->extensions_len);
while (CBS_len(&extensions) != 0) {
uint16_t type;
CBS extension;
// Decode the next extension.
if (!CBS_get_u16(&extensions, &type) ||
!CBS_get_u16_length_prefixed(&extensions, &extension)) {
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
unsigned ext_index;
const struct tls_extension *const ext =
tls_extension_find(&ext_index, type);
if (ext == NULL) {
continue;
}
hs->extensions.received |= (1u << ext_index);
uint8_t alert = SSL_AD_DECODE_ERROR;
if (!ext->parse_clienthello(hs, &alert, &extension)) {
*out_alert = alert;
OPENSSL_PUT_ERROR(SSL, SSL_R_ERROR_PARSING_EXTENSION);
ERR_add_error_dataf("extension %u", (unsigned)type);
return false;
}
}
for (size_t i = 0; i < kNumExtensions; i++) {
if (hs->extensions.received & (1u << i)) {
continue;
}
CBS *contents = NULL, fake_contents;
static const uint8_t kFakeRenegotiateExtension[] = {0};
if (kExtensions[i].value == TLSEXT_TYPE_renegotiate &&
ssl_client_cipher_list_contains_cipher(client_hello,
SSL3_CK_SCSV & 0xffff)) {
// The renegotiation SCSV was received so pretend that we received a
// renegotiation extension.
CBS_init(&fake_contents, kFakeRenegotiateExtension,
sizeof(kFakeRenegotiateExtension));
contents = &fake_contents;
hs->extensions.received |= (1u << i);
}
// Extension wasn't observed so call the callback with a NULL
// parameter.
uint8_t alert = SSL_AD_DECODE_ERROR;
if (!kExtensions[i].parse_clienthello(hs, &alert, contents)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_MISSING_EXTENSION);
ERR_add_error_dataf("extension %u", (unsigned)kExtensions[i].value);
*out_alert = alert;
return false;
}
}
return true;
}
bool ssl_parse_clienthello_tlsext(SSL_HANDSHAKE *hs,
const SSL_CLIENT_HELLO *client_hello) {
SSL *const ssl = hs->ssl;
int alert = SSL_AD_DECODE_ERROR;
if (!ssl_scan_clienthello_tlsext(hs, client_hello, &alert)) {
ssl_send_alert(ssl, SSL3_AL_FATAL, alert);
return false;
}
if (!ssl_check_clienthello_tlsext(hs)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_CLIENTHELLO_TLSEXT);
return false;
}
return true;
}
static bool ssl_scan_serverhello_tlsext(SSL_HANDSHAKE *hs, const CBS *cbs,
int *out_alert) {
CBS extensions = *cbs;
if (!tls1_check_duplicate_extensions(&extensions)) {
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
uint32_t received = 0;
while (CBS_len(&extensions) != 0) {
uint16_t type;
CBS extension;
// Decode the next extension.
if (!CBS_get_u16(&extensions, &type) ||
!CBS_get_u16_length_prefixed(&extensions, &extension)) {
*out_alert = SSL_AD_DECODE_ERROR;
return false;
}
unsigned ext_index;
const struct tls_extension *const ext =
tls_extension_find(&ext_index, type);
if (ext == NULL) {
OPENSSL_PUT_ERROR(SSL, SSL_R_UNEXPECTED_EXTENSION);
ERR_add_error_dataf("extension %u", (unsigned)type);
*out_alert = SSL_AD_UNSUPPORTED_EXTENSION;
return false;
}
static_assert(kNumExtensions <= sizeof(hs->extensions.sent) * 8,
"too many bits");
if (!(hs->extensions.sent & (1u << ext_index))) {
// If the extension was never sent then it is illegal.
OPENSSL_PUT_ERROR(SSL, SSL_R_UNEXPECTED_EXTENSION);
ERR_add_error_dataf("extension :%u", (unsigned)type);
*out_alert = SSL_AD_UNSUPPORTED_EXTENSION;
return false;
}
received |= (1u << ext_index);
uint8_t alert = SSL_AD_DECODE_ERROR;
if (!ext->parse_serverhello(hs, &alert, &extension)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_ERROR_PARSING_EXTENSION);
ERR_add_error_dataf("extension %u", (unsigned)type);
*out_alert = alert;
return false;
}
}
for (size_t i = 0; i < kNumExtensions; i++) {
if (!(received & (1u << i))) {
// Extension wasn't observed so call the callback with a NULL
// parameter.
uint8_t alert = SSL_AD_DECODE_ERROR;
if (!kExtensions[i].parse_serverhello(hs, &alert, NULL)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_MISSING_EXTENSION);
ERR_add_error_dataf("extension %u", (unsigned)kExtensions[i].value);
*out_alert = alert;
return false;
}
}
}
return true;
}
static bool ssl_check_clienthello_tlsext(SSL_HANDSHAKE *hs) {
SSL *const ssl = hs->ssl;
int ret = SSL_TLSEXT_ERR_NOACK;
int al = SSL_AD_UNRECOGNIZED_NAME;
if (ssl->ctx->servername_callback != 0) {
ret = ssl->ctx->servername_callback(ssl, &al, ssl->ctx->servername_arg);
} else if (ssl->session_ctx->servername_callback != 0) {
ret = ssl->session_ctx->servername_callback(
ssl, &al, ssl->session_ctx->servername_arg);
}
switch (ret) {
case SSL_TLSEXT_ERR_ALERT_FATAL:
ssl_send_alert(ssl, SSL3_AL_FATAL, al);
return false;
case SSL_TLSEXT_ERR_NOACK:
hs->should_ack_sni = false;
return true;
default:
return true;
}
}
static bool ssl_check_serverhello_tlsext(SSL_HANDSHAKE *hs) {
SSL *const ssl = hs->ssl;
// ALPS and ALPN have a dependency between each other, so we defer checking
// consistency to after the callbacks run.
if (hs->new_session != nullptr && hs->new_session->has_application_settings) {
// ALPN must be negotiated.
if (ssl->s3->alpn_selected.empty()) {
OPENSSL_PUT_ERROR(SSL, SSL_R_NEGOTIATED_ALPS_WITHOUT_ALPN);
ssl_send_alert(ssl, SSL3_AL_FATAL, SSL_AD_ILLEGAL_PARAMETER);
return false;
}
// The negotiated protocol must be one of the ones we advertised for ALPS.
Span<const uint8_t> settings;
if (!ssl_get_local_application_settings(hs, &settings,
ssl->s3->alpn_selected)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_INVALID_ALPN_PROTOCOL);
ssl_send_alert(ssl, SSL3_AL_FATAL, SSL_AD_ILLEGAL_PARAMETER);
return false;
}
if (!hs->new_session->local_application_settings.CopyFrom(settings)) {
ssl_send_alert(ssl, SSL3_AL_FATAL, SSL_AD_INTERNAL_ERROR);
return false;
}
}
return true;
}
bool ssl_parse_serverhello_tlsext(SSL_HANDSHAKE *hs, const CBS *cbs) {
SSL *const ssl = hs->ssl;
int alert = SSL_AD_DECODE_ERROR;
if (!ssl_scan_serverhello_tlsext(hs, cbs, &alert)) {
ssl_send_alert(ssl, SSL3_AL_FATAL, alert);
return false;
}
if (!ssl_check_serverhello_tlsext(hs)) {
return false;
}
return true;
}
static enum ssl_ticket_aead_result_t decrypt_ticket_with_cipher_ctx(
Array<uint8_t> *out, EVP_CIPHER_CTX *cipher_ctx, HMAC_CTX *hmac_ctx,
Span<const uint8_t> ticket) {
size_t iv_len = EVP_CIPHER_CTX_iv_length(cipher_ctx);
// Check the MAC at the end of the ticket.
uint8_t mac[EVP_MAX_MD_SIZE];
size_t mac_len = HMAC_size(hmac_ctx);
if (ticket.size() < SSL_TICKET_KEY_NAME_LEN + iv_len + 1 + mac_len) {
// The ticket must be large enough for key name, IV, data, and MAC.
return ssl_ticket_aead_ignore_ticket;
}
// Split the ticket into the ticket and the MAC.
auto ticket_mac = ticket.last(mac_len);
ticket = ticket.first(ticket.size() - mac_len);
HMAC_Update(hmac_ctx, ticket.data(), ticket.size());
HMAC_Final(hmac_ctx, mac, NULL);
assert(mac_len == ticket_mac.size());
bool mac_ok = CRYPTO_memcmp(mac, ticket_mac.data(), mac_len) == 0;
#if defined(BORINGSSL_UNSAFE_FUZZER_MODE)
mac_ok = true;
#endif
if (!mac_ok) {
return ssl_ticket_aead_ignore_ticket;
}
// Decrypt the session data.
auto ciphertext = ticket.subspan(SSL_TICKET_KEY_NAME_LEN + iv_len);
Array<uint8_t> plaintext;
#if defined(BORINGSSL_UNSAFE_FUZZER_MODE)
if (!plaintext.CopyFrom(ciphertext)) {
return ssl_ticket_aead_error;
}
#else
if (ciphertext.size() >= INT_MAX) {
return ssl_ticket_aead_ignore_ticket;
}
if (!plaintext.Init(ciphertext.size())) {
return ssl_ticket_aead_error;
}
int len1, len2;
if (!EVP_DecryptUpdate(cipher_ctx, plaintext.data(), &len1, ciphertext.data(),
(int)ciphertext.size()) ||
!EVP_DecryptFinal_ex(cipher_ctx, plaintext.data() + len1, &len2)) {
ERR_clear_error();
return ssl_ticket_aead_ignore_ticket;
}
plaintext.Shrink(static_cast<size_t>(len1) + len2);
#endif
*out = std::move(plaintext);
return ssl_ticket_aead_success;
}
static enum ssl_ticket_aead_result_t ssl_decrypt_ticket_with_cb(
SSL_HANDSHAKE *hs, Array<uint8_t> *out, bool *out_renew_ticket,
Span<const uint8_t> ticket) {
assert(ticket.size() >= SSL_TICKET_KEY_NAME_LEN + EVP_MAX_IV_LENGTH);
ScopedEVP_CIPHER_CTX cipher_ctx;
ScopedHMAC_CTX hmac_ctx;
auto name = ticket.subspan(0, SSL_TICKET_KEY_NAME_LEN);
// The actual IV is shorter, but the length is determined by the callback's
// chosen cipher. Instead we pass in |EVP_MAX_IV_LENGTH| worth of IV to ensure
// the callback has enough.
auto iv = ticket.subspan(SSL_TICKET_KEY_NAME_LEN, EVP_MAX_IV_LENGTH);
int cb_ret = hs->ssl->session_ctx->ticket_key_cb(
hs->ssl, const_cast<uint8_t *>(name.data()),
const_cast<uint8_t *>(iv.data()), cipher_ctx.get(), hmac_ctx.get(),
0 /* decrypt */);
if (cb_ret < 0) {
return ssl_ticket_aead_error;
} else if (cb_ret == 0) {
return ssl_ticket_aead_ignore_ticket;
} else if (cb_ret == 2) {
*out_renew_ticket = true;
} else {
assert(cb_ret == 1);
}
return decrypt_ticket_with_cipher_ctx(out, cipher_ctx.get(), hmac_ctx.get(),
ticket);
}
static enum ssl_ticket_aead_result_t ssl_decrypt_ticket_with_ticket_keys(
SSL_HANDSHAKE *hs, Array<uint8_t> *out, Span<const uint8_t> ticket) {
assert(ticket.size() >= SSL_TICKET_KEY_NAME_LEN + EVP_MAX_IV_LENGTH);
SSL_CTX *ctx = hs->ssl->session_ctx.get();
// Rotate the ticket key if necessary.
if (!ssl_ctx_rotate_ticket_encryption_key(ctx)) {
return ssl_ticket_aead_error;
}
const EVP_CIPHER *cipher = EVP_aes_128_cbc();
auto name = ticket.subspan(0, SSL_TICKET_KEY_NAME_LEN);
auto iv =
ticket.subspan(SSL_TICKET_KEY_NAME_LEN, EVP_CIPHER_iv_length(cipher));
// Pick the matching ticket key and decrypt.
ScopedEVP_CIPHER_CTX cipher_ctx;
ScopedHMAC_CTX hmac_ctx;
{
MutexReadLock lock(&ctx->lock);
const TicketKey *key;
if (ctx->ticket_key_current && name == ctx->ticket_key_current->name) {
key = ctx->ticket_key_current.get();
} else if (ctx->ticket_key_prev && name == ctx->ticket_key_prev->name) {
key = ctx->ticket_key_prev.get();
} else {
return ssl_ticket_aead_ignore_ticket;
}
if (!HMAC_Init_ex(hmac_ctx.get(), key->hmac_key, sizeof(key->hmac_key),
tlsext_tick_md(), NULL) ||
!EVP_DecryptInit_ex(cipher_ctx.get(), cipher, NULL,
key->aes_key, iv.data())) {
return ssl_ticket_aead_error;
}
}
return decrypt_ticket_with_cipher_ctx(out, cipher_ctx.get(), hmac_ctx.get(),
ticket);
}
static enum ssl_ticket_aead_result_t ssl_decrypt_ticket_with_method(
SSL_HANDSHAKE *hs, Array<uint8_t> *out, bool *out_renew_ticket,
Span<const uint8_t> ticket) {
Array<uint8_t> plaintext;
if (!plaintext.Init(ticket.size())) {
return ssl_ticket_aead_error;
}
size_t plaintext_len;
const enum ssl_ticket_aead_result_t result =
hs->ssl->session_ctx->ticket_aead_method->open(
hs->ssl, plaintext.data(), &plaintext_len, ticket.size(),
ticket.data(), ticket.size());
if (result != ssl_ticket_aead_success) {
return result;
}
plaintext.Shrink(plaintext_len);
*out = std::move(plaintext);
return ssl_ticket_aead_success;
}
enum ssl_ticket_aead_result_t ssl_process_ticket(
SSL_HANDSHAKE *hs, UniquePtr<SSL_SESSION> *out_session,
bool *out_renew_ticket, Span<const uint8_t> ticket,
Span<const uint8_t> session_id) {
SSL *const ssl = hs->ssl;
*out_renew_ticket = false;
out_session->reset();
if ((SSL_get_options(hs->ssl) & SSL_OP_NO_TICKET) ||
session_id.size() > SSL_MAX_SSL_SESSION_ID_LENGTH) {
return ssl_ticket_aead_ignore_ticket;
}
// Tickets in TLS 1.3 are tied into pre-shared keys (PSKs), unlike in TLS 1.2
// where that concept doesn't exist. The |decrypted_psk| and |ignore_psk|
// hints only apply to PSKs. We check the version to determine which this is.
const bool is_psk = ssl_protocol_version(ssl) >= TLS1_3_VERSION;
Array<uint8_t> plaintext;
enum ssl_ticket_aead_result_t result;
SSL_HANDSHAKE_HINTS *const hints = hs->hints.get();
if (is_psk && hints && !hs->hints_requested &&
!hints->decrypted_psk.empty()) {
result = plaintext.CopyFrom(hints->decrypted_psk) ? ssl_ticket_aead_success
: ssl_ticket_aead_error;
} else if (is_psk && hints && !hs->hints_requested && hints->ignore_psk) {
result = ssl_ticket_aead_ignore_ticket;
} else if (!is_psk && hints && !hs->hints_requested &&
!hints->decrypted_ticket.empty()) {
if (plaintext.CopyFrom(hints->decrypted_ticket)) {
result = ssl_ticket_aead_success;
*out_renew_ticket = hints->renew_ticket;
} else {
result = ssl_ticket_aead_error;
}
} else if (!is_psk && hints && !hs->hints_requested && hints->ignore_ticket) {
result = ssl_ticket_aead_ignore_ticket;
} else if (ssl->session_ctx->ticket_aead_method != NULL) {
result = ssl_decrypt_ticket_with_method(hs, &plaintext, out_renew_ticket,
ticket);
} else {
// Ensure there is room for the key name and the largest IV |ticket_key_cb|
// may try to consume. The real limit may be lower, but the maximum IV
// length should be well under the minimum size for the session material and
// HMAC.
if (ticket.size() < SSL_TICKET_KEY_NAME_LEN + EVP_MAX_IV_LENGTH) {
result = ssl_ticket_aead_ignore_ticket;
} else if (ssl->session_ctx->ticket_key_cb != NULL) {
result =
ssl_decrypt_ticket_with_cb(hs, &plaintext, out_renew_ticket, ticket);
} else {
result = ssl_decrypt_ticket_with_ticket_keys(hs, &plaintext, ticket);
}
}
if (hints && hs->hints_requested) {
if (result == ssl_ticket_aead_ignore_ticket) {
if (is_psk) {
hints->ignore_psk = true;
} else {
hints->ignore_ticket = true;
}
} else if (result == ssl_ticket_aead_success) {
if (is_psk) {
if (!hints->decrypted_psk.CopyFrom(plaintext)) {
return ssl_ticket_aead_error;
}
} else {
if (!hints->decrypted_ticket.CopyFrom(plaintext)) {
return ssl_ticket_aead_error;
}
hints->renew_ticket = *out_renew_ticket;
}
}
}
if (result != ssl_ticket_aead_success) {
return result;
}
// Decode the session.
UniquePtr<SSL_SESSION> session(SSL_SESSION_from_bytes(
plaintext.data(), plaintext.size(), ssl->ctx.get()));
if (!session) {
ERR_clear_error(); // Don't leave an error on the queue.
return ssl_ticket_aead_ignore_ticket;
}
// Envoy's tests expect the session to have a session ID that matches the
// placeholder used by the client. It's unclear whether this is a good idea,
// but we maintain it for now.
SHA256(ticket.data(), ticket.size(), session->session_id);
// Other consumers may expect a non-empty session ID to indicate resumption.
session->session_id_length = SHA256_DIGEST_LENGTH;
*out_session = std::move(session);
return ssl_ticket_aead_success;
}
bool tls1_parse_peer_sigalgs(SSL_HANDSHAKE *hs, const CBS *in_sigalgs) {
// Extension ignored for inappropriate versions
if (ssl_protocol_version(hs->ssl) < TLS1_2_VERSION) {
return true;
}
// In all contexts, the signature algorithms list may not be empty. (It may be
// omitted by clients in TLS 1.2, but then the entire extension is omitted.)
return CBS_len(in_sigalgs) != 0 &&
parse_u16_array(in_sigalgs, &hs->peer_sigalgs);
}
bool tls1_get_legacy_signature_algorithm(uint16_t *out, const EVP_PKEY *pkey) {
switch (EVP_PKEY_id(pkey)) {
case EVP_PKEY_RSA:
*out = SSL_SIGN_RSA_PKCS1_MD5_SHA1;
return true;
case EVP_PKEY_EC:
*out = SSL_SIGN_ECDSA_SHA1;
return true;
default:
return false;
}
}
bool tls1_choose_signature_algorithm(SSL_HANDSHAKE *hs, uint16_t *out) {
SSL *const ssl = hs->ssl;
CERT *cert = hs->config->cert.get();
DC *dc = cert->dc.get();
// Before TLS 1.2, the signature algorithm isn't negotiated as part of the
// handshake.
if (ssl_protocol_version(ssl) < TLS1_2_VERSION) {
if (!tls1_get_legacy_signature_algorithm(out, hs->local_pubkey.get())) {
OPENSSL_PUT_ERROR(SSL, SSL_R_NO_COMMON_SIGNATURE_ALGORITHMS);
return false;
}
return true;
}
Span<const uint16_t> sigalgs = kSignSignatureAlgorithms;
if (ssl_signing_with_dc(hs)) {
sigalgs = MakeConstSpan(&dc->expected_cert_verify_algorithm, 1);
} else if (!cert->sigalgs.empty()) {
sigalgs = cert->sigalgs;
}
Span<const uint16_t> peer_sigalgs = tls1_get_peer_verify_algorithms(hs);
for (uint16_t sigalg : sigalgs) {
if (!ssl_private_key_supports_signature_algorithm(hs, sigalg)) {
continue;
}
for (uint16_t peer_sigalg : peer_sigalgs) {
if (sigalg == peer_sigalg) {
*out = sigalg;
return true;
}
}
}
OPENSSL_PUT_ERROR(SSL, SSL_R_NO_COMMON_SIGNATURE_ALGORITHMS);
return false;
}
Span<const uint16_t> tls1_get_peer_verify_algorithms(const SSL_HANDSHAKE *hs) {
Span<const uint16_t> peer_sigalgs = hs->peer_sigalgs;
if (peer_sigalgs.empty() && ssl_protocol_version(hs->ssl) < TLS1_3_VERSION) {
// If the client didn't specify any signature_algorithms extension then
// we can assume that it supports SHA1. See
// http://tools.ietf.org/html/rfc5246#section-7.4.1.4.1
static const uint16_t kDefaultPeerAlgorithms[] = {SSL_SIGN_RSA_PKCS1_SHA1,
SSL_SIGN_ECDSA_SHA1};
peer_sigalgs = kDefaultPeerAlgorithms;
}
return peer_sigalgs;
}
bool tls1_verify_channel_id(SSL_HANDSHAKE *hs, const SSLMessage &msg) {
SSL *const ssl = hs->ssl;
// A Channel ID handshake message is structured to contain multiple
// extensions, but the only one that can be present is Channel ID.
uint16_t extension_type;
CBS channel_id = msg.body, extension;
if (!CBS_get_u16(&channel_id, &extension_type) ||
!CBS_get_u16_length_prefixed(&channel_id, &extension) ||
CBS_len(&channel_id) != 0 ||
extension_type != TLSEXT_TYPE_channel_id ||
CBS_len(&extension) != TLSEXT_CHANNEL_ID_SIZE) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
ssl_send_alert(ssl, SSL3_AL_FATAL, SSL_AD_DECODE_ERROR);
return false;
}
const EC_GROUP *p256 = EC_group_p256();
UniquePtr<ECDSA_SIG> sig(ECDSA_SIG_new());
UniquePtr<BIGNUM> x(BN_new()), y(BN_new());
if (!sig || !x || !y) {
return false;
}
const uint8_t *p = CBS_data(&extension);
if (BN_bin2bn(p + 0, 32, x.get()) == NULL ||
BN_bin2bn(p + 32, 32, y.get()) == NULL ||
BN_bin2bn(p + 64, 32, sig->r) == NULL ||
BN_bin2bn(p + 96, 32, sig->s) == NULL) {
return false;
}
UniquePtr<EC_KEY> key(EC_KEY_new());
UniquePtr<EC_POINT> point(EC_POINT_new(p256));
if (!key || !point ||
!EC_POINT_set_affine_coordinates_GFp(p256, point.get(), x.get(), y.get(),
nullptr) ||
!EC_KEY_set_group(key.get(), p256) ||
!EC_KEY_set_public_key(key.get(), point.get())) {
return false;
}
uint8_t digest[EVP_MAX_MD_SIZE];
size_t digest_len;
if (!tls1_channel_id_hash(hs, digest, &digest_len)) {
return false;
}
bool sig_ok = ECDSA_do_verify(digest, digest_len, sig.get(), key.get());
#if defined(BORINGSSL_UNSAFE_FUZZER_MODE)
sig_ok = true;
ERR_clear_error();
#endif
if (!sig_ok) {
OPENSSL_PUT_ERROR(SSL, SSL_R_CHANNEL_ID_SIGNATURE_INVALID);
ssl_send_alert(ssl, SSL3_AL_FATAL, SSL_AD_DECRYPT_ERROR);
return false;
}
OPENSSL_memcpy(ssl->s3->channel_id, p, 64);
ssl->s3->channel_id_valid = true;
return true;
}
bool tls1_write_channel_id(SSL_HANDSHAKE *hs, CBB *cbb) {
uint8_t digest[EVP_MAX_MD_SIZE];
size_t digest_len;
if (!tls1_channel_id_hash(hs, digest, &digest_len)) {
return false;
}
EC_KEY *ec_key = EVP_PKEY_get0_EC_KEY(hs->config->channel_id_private.get());
if (ec_key == nullptr) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
UniquePtr<BIGNUM> x(BN_new()), y(BN_new());
if (!x || !y ||
!EC_POINT_get_affine_coordinates_GFp(EC_KEY_get0_group(ec_key),
EC_KEY_get0_public_key(ec_key),
x.get(), y.get(), nullptr)) {
return false;
}
UniquePtr<ECDSA_SIG> sig(ECDSA_do_sign(digest, digest_len, ec_key));
if (!sig) {
return false;
}
CBB child;
if (!CBB_add_u16(cbb, TLSEXT_TYPE_channel_id) ||
!CBB_add_u16_length_prefixed(cbb, &child) ||
!BN_bn2cbb_padded(&child, 32, x.get()) ||
!BN_bn2cbb_padded(&child, 32, y.get()) ||
!BN_bn2cbb_padded(&child, 32, sig->r) ||
!BN_bn2cbb_padded(&child, 32, sig->s) ||
!CBB_flush(cbb)) {
return false;
}
return true;
}
bool tls1_channel_id_hash(SSL_HANDSHAKE *hs, uint8_t *out, size_t *out_len) {
SSL *const ssl = hs->ssl;
if (ssl_protocol_version(ssl) >= TLS1_3_VERSION) {
Array<uint8_t> msg;
if (!tls13_get_cert_verify_signature_input(hs, &msg,
ssl_cert_verify_channel_id)) {
return false;
}
SHA256(msg.data(), msg.size(), out);
*out_len = SHA256_DIGEST_LENGTH;
return true;
}
SHA256_CTX ctx;
SHA256_Init(&ctx);
static const char kClientIDMagic[] = "TLS Channel ID signature";
SHA256_Update(&ctx, kClientIDMagic, sizeof(kClientIDMagic));
if (ssl->session != NULL) {
static const char kResumptionMagic[] = "Resumption";
SHA256_Update(&ctx, kResumptionMagic, sizeof(kResumptionMagic));
if (ssl->session->original_handshake_hash_len == 0) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
SHA256_Update(&ctx, ssl->session->original_handshake_hash,
ssl->session->original_handshake_hash_len);
}
uint8_t hs_hash[EVP_MAX_MD_SIZE];
size_t hs_hash_len;
if (!hs->transcript.GetHash(hs_hash, &hs_hash_len)) {
return false;
}
SHA256_Update(&ctx, hs_hash, (size_t)hs_hash_len);
SHA256_Final(out, &ctx);
*out_len = SHA256_DIGEST_LENGTH;
return true;
}
bool tls1_record_handshake_hashes_for_channel_id(SSL_HANDSHAKE *hs) {
SSL *const ssl = hs->ssl;
// This function should never be called for a resumed session because the
// handshake hashes that we wish to record are for the original, full
// handshake.
if (ssl->session != NULL) {
return false;
}
static_assert(
sizeof(hs->new_session->original_handshake_hash) == EVP_MAX_MD_SIZE,
"original_handshake_hash is too small");
size_t digest_len;
if (!hs->transcript.GetHash(hs->new_session->original_handshake_hash,
&digest_len)) {
return false;
}
static_assert(EVP_MAX_MD_SIZE <= 0xff,
"EVP_MAX_MD_SIZE does not fit in uint8_t");
hs->new_session->original_handshake_hash_len = (uint8_t)digest_len;
return true;
}
bool ssl_is_sct_list_valid(const CBS *contents) {
// Shallow parse the SCT list for sanity. By the RFC
// (https://tools.ietf.org/html/rfc6962#section-3.3) neither the list nor any
// of the SCTs may be empty.
CBS copy = *contents;
CBS sct_list;
if (!CBS_get_u16_length_prefixed(&copy, &sct_list) ||
CBS_len(&copy) != 0 ||
CBS_len(&sct_list) == 0) {
return false;
}
while (CBS_len(&sct_list) > 0) {
CBS sct;
if (!CBS_get_u16_length_prefixed(&sct_list, &sct) ||
CBS_len(&sct) == 0) {
return false;
}
}
return true;
}
BSSL_NAMESPACE_END
using namespace bssl;
int SSL_early_callback_ctx_extension_get(const SSL_CLIENT_HELLO *client_hello,
uint16_t extension_type,
const uint8_t **out_data,
size_t *out_len) {
CBS cbs;
if (!ssl_client_hello_get_extension(client_hello, &cbs, extension_type)) {
return 0;
}
*out_data = CBS_data(&cbs);
*out_len = CBS_len(&cbs);
return 1;
}