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.

1172 lines
42 KiB

/* Copyright (c) 2021, Google Inc.
*
* Permission to use, copy, modify, and/or distribute this software for any
* purpose with or without fee is hereby granted, provided that the above
* copyright notice and this permission notice appear in all copies.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
* SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
* OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
* CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
#include <openssl/ssl.h>
#include <assert.h>
#include <string.h>
#include <algorithm>
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
#include <utility>
#include <openssl/aead.h>
#include <openssl/bytestring.h>
#include <openssl/curve25519.h>
#include <openssl/err.h>
#include <openssl/hkdf.h>
#include <openssl/hpke.h>
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
#include <openssl/rand.h>
#include "internal.h"
#if defined(OPENSSL_MSAN)
#define NO_SANITIZE_MEMORY __attribute__((no_sanitize("memory")))
#else
#define NO_SANITIZE_MEMORY
#endif
BSSL_NAMESPACE_BEGIN
// ECH reuses the extension code point for the version number.
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>
4 years ago
static constexpr uint16_t kECHConfigVersion =
TLSEXT_TYPE_encrypted_client_hello;
static const decltype(&EVP_hpke_aes_128_gcm) kSupportedAEADs[] = {
&EVP_hpke_aes_128_gcm,
&EVP_hpke_aes_256_gcm,
&EVP_hpke_chacha20_poly1305,
};
static const EVP_HPKE_AEAD *get_ech_aead(uint16_t aead_id) {
for (const auto aead_func : kSupportedAEADs) {
const EVP_HPKE_AEAD *aead = aead_func();
if (aead_id == EVP_HPKE_AEAD_id(aead)) {
return aead;
}
}
return nullptr;
}
// ssl_client_hello_write_without_extensions serializes |client_hello| into
// |out|, omitting the length-prefixed extensions. It serializes individual
// fields, starting with |client_hello->version|, and ignores the
// |client_hello->client_hello| field. It returns true on success and false on
// failure.
static bool ssl_client_hello_write_without_extensions(
const SSL_CLIENT_HELLO *client_hello, CBB *out) {
CBB cbb;
if (!CBB_add_u16(out, client_hello->version) ||
!CBB_add_bytes(out, client_hello->random, client_hello->random_len) ||
!CBB_add_u8_length_prefixed(out, &cbb) ||
!CBB_add_bytes(&cbb, client_hello->session_id,
client_hello->session_id_len) ||
!CBB_add_u16_length_prefixed(out, &cbb) ||
!CBB_add_bytes(&cbb, client_hello->cipher_suites,
client_hello->cipher_suites_len) ||
!CBB_add_u8_length_prefixed(out, &cbb) ||
!CBB_add_bytes(&cbb, client_hello->compression_methods,
client_hello->compression_methods_len) ||
!CBB_flush(out)) {
return false;
}
return true;
}
bool ssl_decode_client_hello_inner(
SSL *ssl, uint8_t *out_alert, Array<uint8_t> *out_client_hello_inner,
Span<const uint8_t> encoded_client_hello_inner,
const SSL_CLIENT_HELLO *client_hello_outer) {
SSL_CLIENT_HELLO client_hello_inner;
if (!ssl_client_hello_init(ssl, &client_hello_inner,
encoded_client_hello_inner)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
// TLS 1.3 ClientHellos must have extensions, and EncodedClientHelloInners use
// ClientHelloOuter's session_id.
if (client_hello_inner.extensions_len == 0 ||
client_hello_inner.session_id_len != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
client_hello_inner.session_id = client_hello_outer->session_id;
client_hello_inner.session_id_len = client_hello_outer->session_id_len;
// Begin serializing a message containing the ClientHelloInner in |cbb|.
ScopedCBB cbb;
CBB body, extensions;
if (!ssl->method->init_message(ssl, cbb.get(), &body, SSL3_MT_CLIENT_HELLO) ||
!ssl_client_hello_write_without_extensions(&client_hello_inner, &body) ||
!CBB_add_u16_length_prefixed(&body, &extensions)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
// Sort the extensions in ClientHelloOuter, so ech_outer_extensions may be
// processed in O(n*log(n)) time, rather than O(n^2).
struct Extension {
uint16_t extension = 0;
Span<const uint8_t> body;
bool copied = false;
};
// MSan's libc interceptors do not handle |bsearch|. See b/182583130.
auto compare_extension = [](const void *a, const void *b)
NO_SANITIZE_MEMORY -> int {
const Extension *extension_a = reinterpret_cast<const Extension *>(a);
const Extension *extension_b = reinterpret_cast<const Extension *>(b);
if (extension_a->extension < extension_b->extension) {
return -1;
} else if (extension_a->extension > extension_b->extension) {
return 1;
}
return 0;
};
GrowableArray<Extension> sorted_extensions;
CBS unsorted_extensions(MakeConstSpan(client_hello_outer->extensions,
client_hello_outer->extensions_len));
while (CBS_len(&unsorted_extensions) > 0) {
Extension extension;
CBS extension_body;
if (!CBS_get_u16(&unsorted_extensions, &extension.extension) ||
!CBS_get_u16_length_prefixed(&unsorted_extensions, &extension_body)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
extension.body = extension_body;
if (!sorted_extensions.Push(extension)) {
return false;
}
}
qsort(sorted_extensions.data(), sorted_extensions.size(), sizeof(Extension),
compare_extension);
// Copy extensions from |client_hello_inner|, expanding ech_outer_extensions.
CBS inner_extensions(MakeConstSpan(client_hello_inner.extensions,
client_hello_inner.extensions_len));
while (CBS_len(&inner_extensions) > 0) {
uint16_t extension_id;
CBS extension_body;
if (!CBS_get_u16(&inner_extensions, &extension_id) ||
!CBS_get_u16_length_prefixed(&inner_extensions, &extension_body)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
if (extension_id != TLSEXT_TYPE_ech_outer_extensions) {
if (!CBB_add_u16(&extensions, extension_id) ||
!CBB_add_u16(&extensions, CBS_len(&extension_body)) ||
!CBB_add_bytes(&extensions, CBS_data(&extension_body),
CBS_len(&extension_body))) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
continue;
}
// Replace ech_outer_extensions with the corresponding outer extensions.
CBS outer_extensions;
if (!CBS_get_u8_length_prefixed(&extension_body, &outer_extensions) ||
CBS_len(&extension_body) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
while (CBS_len(&outer_extensions) > 0) {
uint16_t extension_needed;
if (!CBS_get_u16(&outer_extensions, &extension_needed)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
if (extension_needed == TLSEXT_TYPE_encrypted_client_hello) {
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
// Find the referenced extension.
Extension key;
key.extension = extension_needed;
Extension *result = reinterpret_cast<Extension *>(
bsearch(&key, sorted_extensions.data(), sorted_extensions.size(),
sizeof(Extension), compare_extension));
if (result == nullptr) {
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
// Extensions may be referenced at most once, to bound the result size.
if (result->copied) {
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
OPENSSL_PUT_ERROR(SSL, SSL_R_DUPLICATE_EXTENSION);
return false;
}
result->copied = true;
if (!CBB_add_u16(&extensions, extension_needed) ||
!CBB_add_u16(&extensions, result->body.size()) ||
!CBB_add_bytes(&extensions, result->body.data(),
result->body.size())) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return false;
}
}
}
if (!CBB_flush(&body)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
// See https://github.com/tlswg/draft-ietf-tls-esni/pull/411
CBS extension;
if (!ssl_client_hello_init(ssl, &client_hello_inner,
MakeConstSpan(CBB_data(&body), CBB_len(&body))) ||
!ssl_client_hello_get_extension(&client_hello_inner, &extension,
TLSEXT_TYPE_ech_is_inner) ||
CBS_len(&extension) != 0 ||
ssl_client_hello_get_extension(&client_hello_inner, &extension,
TLSEXT_TYPE_encrypted_client_hello) ||
!ssl_client_hello_get_extension(&client_hello_inner, &extension,
TLSEXT_TYPE_supported_versions)) {
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
OPENSSL_PUT_ERROR(SSL, SSL_R_INVALID_CLIENT_HELLO_INNER);
return false;
}
// Parse supported_versions and reject TLS versions prior to TLS 1.3. Older
// versions are incompatible with ECH.
CBS versions;
if (!CBS_get_u8_length_prefixed(&extension, &versions) ||
CBS_len(&extension) != 0 || //
CBS_len(&versions) == 0) {
*out_alert = SSL_AD_DECODE_ERROR;
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
while (CBS_len(&versions) != 0) {
uint16_t version;
if (!CBS_get_u16(&versions, &version)) {
*out_alert = SSL_AD_DECODE_ERROR;
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
if (version == SSL3_VERSION || version == TLS1_VERSION ||
version == TLS1_1_VERSION || version == TLS1_2_VERSION ||
version == DTLS1_VERSION || version == DTLS1_2_VERSION) {
*out_alert = SSL_AD_ILLEGAL_PARAMETER;
OPENSSL_PUT_ERROR(SSL, SSL_R_INVALID_CLIENT_HELLO_INNER);
return false;
}
}
if (!ssl->method->finish_message(ssl, cbb.get(), out_client_hello_inner)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
return true;
}
bool ssl_client_hello_decrypt(
EVP_HPKE_CTX *hpke_ctx, Array<uint8_t> *out_encoded_client_hello_inner,
bool *out_is_decrypt_error, const SSL_CLIENT_HELLO *client_hello_outer,
uint16_t kdf_id, uint16_t aead_id, const uint8_t config_id,
Span<const uint8_t> enc, Span<const uint8_t> payload) {
*out_is_decrypt_error = false;
// Compute the ClientHello portion of the ClientHelloOuterAAD value. See
// draft-ietf-tls-esni-10, section 5.2.
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
ScopedCBB aad;
CBB enc_cbb, outer_hello_cbb, extensions_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_init(aad.get(), 256) ||
!CBB_add_u16(aad.get(), kdf_id) ||
!CBB_add_u16(aad.get(), aead_id) ||
!CBB_add_u8(aad.get(), config_id) ||
!CBB_add_u16_length_prefixed(aad.get(), &enc_cbb) ||
!CBB_add_bytes(&enc_cbb, enc.data(), enc.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_add_u24_length_prefixed(aad.get(), &outer_hello_cbb) ||
!ssl_client_hello_write_without_extensions(client_hello_outer,
&outer_hello_cbb) ||
!CBB_add_u16_length_prefixed(&outer_hello_cbb, &extensions_cbb)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return false;
}
CBS extensions(MakeConstSpan(client_hello_outer->extensions,
client_hello_outer->extensions_len));
while (CBS_len(&extensions) > 0) {
uint16_t extension_id;
CBS extension_body;
if (!CBS_get_u16(&extensions, &extension_id) ||
!CBS_get_u16_length_prefixed(&extensions, &extension_body)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
if (extension_id == TLSEXT_TYPE_encrypted_client_hello) {
continue;
}
if (!CBB_add_u16(&extensions_cbb, extension_id) ||
!CBB_add_u16(&extensions_cbb, CBS_len(&extension_body)) ||
!CBB_add_bytes(&extensions_cbb, CBS_data(&extension_body),
CBS_len(&extension_body))) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
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
if (!CBB_flush(aad.get())) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return false;
}
#if defined(BORINGSSL_UNSAFE_FUZZER_MODE)
// In fuzzer mode, disable encryption to improve coverage. We reserve a short
// input to signal decryption failure, so the fuzzer can explore fallback to
// ClientHelloOuter.
const uint8_t kBadPayload[] = {0xff};
if (payload == kBadPayload) {
*out_is_decrypt_error = true;
OPENSSL_PUT_ERROR(SSL, SSL_R_DECRYPTION_FAILED);
return false;
}
if (!out_encoded_client_hello_inner->CopyFrom(payload)) {
return false;
}
#else
// Attempt to decrypt into |out_encoded_client_hello_inner|.
if (!out_encoded_client_hello_inner->Init(payload.size())) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return false;
}
size_t encoded_client_hello_inner_len;
if (!EVP_HPKE_CTX_open(hpke_ctx, out_encoded_client_hello_inner->data(),
&encoded_client_hello_inner_len,
out_encoded_client_hello_inner->size(), payload.data(),
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
payload.size(), CBB_data(aad.get()),
CBB_len(aad.get()))) {
*out_is_decrypt_error = true;
OPENSSL_PUT_ERROR(SSL, SSL_R_DECRYPTION_FAILED);
return false;
}
out_encoded_client_hello_inner->Shrink(encoded_client_hello_inner_len);
#endif
return true;
}
static bool parse_ipv4_number(Span<const uint8_t> in, uint32_t *out) {
// See https://url.spec.whatwg.org/#ipv4-number-parser.
uint32_t base = 10;
if (in.size() >= 2 && in[0] == '0' && (in[1] == 'x' || in[1] == 'X')) {
in = in.subspan(2);
base = 16;
} else if (in.size() >= 1 && in[0] == '0') {
in = in.subspan(1);
base = 8;
}
*out = 0;
for (uint8_t c : in) {
uint32_t d;
if ('0' <= c && c <= '9') {
d = c - '0';
} else if ('a' <= c && c <= 'f') {
d = c - 'a' + 10;
} else if ('A' <= c && c <= 'F') {
d = c - 'A' + 10;
} else {
return false;
}
if (d >= base ||
*out > UINT32_MAX / base) {
return false;
}
*out *= base;
if (*out > UINT32_MAX - d) {
return false;
}
*out += d;
}
return true;
}
static bool is_ipv4_address(Span<const uint8_t> in) {
// See https://url.spec.whatwg.org/#concept-ipv4-parser
uint32_t numbers[4];
size_t num_numbers = 0;
while (!in.empty()) {
if (num_numbers == 4) {
// Too many components.
return false;
}
// Find the next dot-separated component.
auto dot = std::find(in.begin(), in.end(), '.');
if (dot == in.begin()) {
// Empty components are not allowed.
return false;
}
Span<const uint8_t> component;
if (dot == in.end()) {
component = in;
in = Span<const uint8_t>();
} else {
component = in.subspan(0, dot - in.begin());
in = in.subspan(dot - in.begin() + 1); // Skip the dot.
}
if (!parse_ipv4_number(component, &numbers[num_numbers])) {
return false;
}
num_numbers++;
}
if (num_numbers == 0) {
return false;
}
for (size_t i = 0; i < num_numbers - 1; i++) {
if (numbers[i] > 255) {
return false;
}
}
return num_numbers == 1 ||
numbers[num_numbers - 1] < 1u << (8 * (5 - num_numbers));
}
bool ssl_is_valid_ech_public_name(Span<const uint8_t> public_name) {
// See draft-ietf-tls-esni-11, Section 4 and RFC 5890, Section 2.3.1. The
// public name must be a dot-separated sequence of LDH labels and not begin or
// end with a dot.
auto copy = public_name;
if (copy.empty()) {
return false;
}
while (!copy.empty()) {
// Find the next dot-separated component.
auto dot = std::find(copy.begin(), copy.end(), '.');
Span<const uint8_t> component;
if (dot == copy.end()) {
component = copy;
copy = Span<const uint8_t>();
} else {
component = copy.subspan(0, dot - copy.begin());
copy = copy.subspan(dot - copy.begin() + 1); // Skip the dot.
if (copy.empty()) {
// Trailing dots are not allowed.
return false;
}
}
// |component| must be a valid LDH label. Checking for empty components also
// rejects leading dots.
if (component.empty() || component.size() > 63 ||
component.front() == '-' || component.back() == '-') {
return false;
}
for (uint8_t c : component) {
if (!('a' <= c && c <= 'z') && !('A' <= c && c <= 'Z') &&
!('0' <= c && c <= '9') && c != '-') {
return false;
}
}
}
return !is_ipv4_address(public_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
static bool parse_ech_config(CBS *cbs, ECHConfig *out, bool *out_supported,
bool all_extensions_mandatory) {
uint16_t version;
CBS orig = *cbs;
CBS contents;
if (!CBS_get_u16(cbs, &version) ||
!CBS_get_u16_length_prefixed(cbs, &contents)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
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
if (version != kECHConfigVersion) {
*out_supported = 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
// Make a copy of the ECHConfig and parse from it, so the results alias into
// the saved copy.
if (!out->raw.CopyFrom(
MakeConstSpan(CBS_data(&orig), CBS_len(&orig) - CBS_len(cbs)))) {
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
CBS ech_config(out->raw);
CBS public_name, public_key, cipher_suites, extensions;
if (!CBS_skip(&ech_config, 2) || // version
!CBS_get_u16_length_prefixed(&ech_config, &contents) ||
!CBS_get_u8(&contents, &out->config_id) ||
!CBS_get_u16(&contents, &out->kem_id) ||
!CBS_get_u16_length_prefixed(&contents, &public_key) ||
CBS_len(&public_key) == 0 ||
!CBS_get_u16_length_prefixed(&contents, &cipher_suites) ||
CBS_len(&cipher_suites) == 0 || CBS_len(&cipher_suites) % 4 != 0 ||
!CBS_get_u16(&contents, &out->maximum_name_length) ||
!CBS_get_u16_length_prefixed(&contents, &public_name) ||
CBS_len(&public_name) == 0 ||
!CBS_get_u16_length_prefixed(&contents, &extensions) ||
CBS_len(&contents) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
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
if (!ssl_is_valid_ech_public_name(public_name)) {
// TODO(https://crbug.com/boringssl/275): The draft says ECHConfigs with
// invalid public names should be ignored, but LDH syntax failures are
// unambiguously invalid.
*out_supported = 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
out->public_key = public_key;
out->public_name = public_name;
// This function does not ensure |out->kem_id| and |out->cipher_suites| use
// supported algorithms. The caller must do this.
out->cipher_suites = cipher_suites;
bool has_unknown_mandatory_extension = false;
while (CBS_len(&extensions) != 0) {
uint16_t type;
CBS body;
if (!CBS_get_u16(&extensions, &type) ||
!CBS_get_u16_length_prefixed(&extensions, &body)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
// We currently do not support any extensions.
if (type & 0x8000 || all_extensions_mandatory) {
// Extension numbers with the high bit set are mandatory. Continue parsing
// to enforce syntax, but we will ultimately ignore this ECHConfig as a
// client and reject it as a server.
has_unknown_mandatory_extension = true;
}
}
*out_supported = !has_unknown_mandatory_extension;
return true;
}
bool ECHServerConfig::Init(Span<const uint8_t> ech_config,
const EVP_HPKE_KEY *key, bool is_retry_config) {
is_retry_config_ = is_retry_config;
// Parse the ECHConfig, rejecting all unsupported parameters and extensions.
// Unlike most server options, ECH's server configuration is serialized and
// configured in both the server and DNS. If the caller configures an
// unsupported parameter, this is a deployment error. To catch these errors,
// we fail early.
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
CBS cbs = ech_config;
bool supported;
if (!parse_ech_config(&cbs, &ech_config_, &supported,
/*all_extensions_mandatory=*/true)) {
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
if (CBS_len(&cbs) != 0) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
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
if (!supported) {
OPENSSL_PUT_ERROR(SSL, SSL_R_UNSUPPORTED_ECH_SERVER_CONFIG);
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
CBS cipher_suites = ech_config_.cipher_suites;
while (CBS_len(&cipher_suites) > 0) {
uint16_t kdf_id, aead_id;
if (!CBS_get_u16(&cipher_suites, &kdf_id) ||
!CBS_get_u16(&cipher_suites, &aead_id)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return false;
}
// The server promises to support every option in the ECHConfig, so reject
// any unsupported cipher suites.
if (kdf_id != EVP_HPKE_HKDF_SHA256 || get_ech_aead(aead_id) == nullptr) {
OPENSSL_PUT_ERROR(SSL, SSL_R_UNSUPPORTED_ECH_SERVER_CONFIG);
return false;
}
}
// Check the public key in the ECHConfig matches |key|.
uint8_t expected_public_key[EVP_HPKE_MAX_PUBLIC_KEY_LENGTH];
size_t expected_public_key_len;
if (!EVP_HPKE_KEY_public_key(key, expected_public_key,
&expected_public_key_len,
sizeof(expected_public_key))) {
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
if (ech_config_.kem_id != EVP_HPKE_KEM_id(EVP_HPKE_KEY_kem(key)) ||
MakeConstSpan(expected_public_key, expected_public_key_len) !=
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
ech_config_.public_key) {
OPENSSL_PUT_ERROR(SSL, SSL_R_ECH_SERVER_CONFIG_AND_PRIVATE_KEY_MISMATCH);
return false;
}
if (!EVP_HPKE_KEY_copy(key_.get(), key)) {
return false;
}
return true;
}
bool ECHServerConfig::SetupContext(EVP_HPKE_CTX *ctx, uint16_t kdf_id,
uint16_t aead_id,
Span<const uint8_t> enc) const {
// Check the cipher suite is supported by this ECHServerConfig.
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
CBS cbs(ech_config_.cipher_suites);
bool cipher_ok = false;
while (CBS_len(&cbs) != 0) {
uint16_t supported_kdf_id, supported_aead_id;
if (!CBS_get_u16(&cbs, &supported_kdf_id) ||
!CBS_get_u16(&cbs, &supported_aead_id)) {
return false;
}
if (kdf_id == supported_kdf_id && aead_id == supported_aead_id) {
cipher_ok = true;
break;
}
}
if (!cipher_ok) {
return false;
}
static const uint8_t kInfoLabel[] = "tls ech";
ScopedCBB info_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_init(info_cbb.get(), sizeof(kInfoLabel) + ech_config_.raw.size()) ||
!CBB_add_bytes(info_cbb.get(), kInfoLabel,
sizeof(kInfoLabel) /* includes trailing NUL */) ||
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(info_cbb.get(), ech_config_.raw.data(),
ech_config_.raw.size())) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return false;
}
assert(kdf_id == EVP_HPKE_HKDF_SHA256);
assert(get_ech_aead(aead_id) != NULL);
return EVP_HPKE_CTX_setup_recipient(
ctx, key_.get(), EVP_hpke_hkdf_sha256(), get_ech_aead(aead_id), enc.data(),
enc.size(), CBB_data(info_cbb.get()), CBB_len(info_cbb.get()));
}
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 ssl_is_valid_ech_config_list(Span<const uint8_t> ech_config_list) {
CBS cbs = ech_config_list, child;
if (!CBS_get_u16_length_prefixed(&cbs, &child) || //
CBS_len(&child) == 0 || //
CBS_len(&cbs) > 0) {
return false;
}
while (CBS_len(&child) > 0) {
ECHConfig ech_config;
bool supported;
if (!parse_ech_config(&child, &ech_config, &supported,
/*all_extensions_mandatory=*/false)) {
return false;
}
}
return true;
}
static bool select_ech_cipher_suite(const EVP_HPKE_KDF **out_kdf,
const EVP_HPKE_AEAD **out_aead,
Span<const uint8_t> cipher_suites) {
const bool has_aes_hardware = EVP_has_aes_hardware();
const EVP_HPKE_AEAD *aead = nullptr;
CBS cbs = cipher_suites;
while (CBS_len(&cbs) != 0) {
uint16_t kdf_id, aead_id;
if (!CBS_get_u16(&cbs, &kdf_id) || //
!CBS_get_u16(&cbs, &aead_id)) {
return false;
}
// Pick the first common cipher suite, but prefer ChaCha20-Poly1305 if we
// don't have AES hardware.
const EVP_HPKE_AEAD *candidate = get_ech_aead(aead_id);
if (kdf_id != EVP_HPKE_HKDF_SHA256 || candidate == nullptr) {
continue;
}
if (aead == nullptr ||
(!has_aes_hardware && aead_id == EVP_HPKE_CHACHA20_POLY1305)) {
aead = candidate;
}
}
if (aead == nullptr) {
return false;
}
*out_kdf = EVP_hpke_hkdf_sha256();
*out_aead = aead;
return true;
}
bool ssl_select_ech_config(SSL_HANDSHAKE *hs, Span<uint8_t> out_enc,
size_t *out_enc_len) {
*out_enc_len = 0;
if (hs->max_version < TLS1_3_VERSION) {
// ECH requires TLS 1.3.
return true;
}
if (!hs->config->client_ech_config_list.empty()) {
CBS cbs = MakeConstSpan(hs->config->client_ech_config_list);
CBS child;
if (!CBS_get_u16_length_prefixed(&cbs, &child) || //
CBS_len(&child) == 0 || //
CBS_len(&cbs) > 0) {
return false;
}
// Look for the first ECHConfig with supported parameters.
while (CBS_len(&child) > 0) {
ECHConfig ech_config;
bool supported;
if (!parse_ech_config(&child, &ech_config, &supported,
/*all_extensions_mandatory=*/false)) {
return false;
}
const EVP_HPKE_KEM *kem = EVP_hpke_x25519_hkdf_sha256();
const EVP_HPKE_KDF *kdf;
const EVP_HPKE_AEAD *aead;
if (supported && //
ech_config.kem_id == EVP_HPKE_DHKEM_X25519_HKDF_SHA256 &&
select_ech_cipher_suite(&kdf, &aead, ech_config.cipher_suites)) {
ScopedCBB info;
static const uint8_t kInfoLabel[] = "tls ech"; // includes trailing NUL
if (!CBB_init(info.get(), sizeof(kInfoLabel) + ech_config.raw.size()) ||
!CBB_add_bytes(info.get(), kInfoLabel, sizeof(kInfoLabel)) ||
!CBB_add_bytes(info.get(), ech_config.raw.data(),
ech_config.raw.size())) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return false;
}
if (!EVP_HPKE_CTX_setup_sender(
hs->ech_hpke_ctx.get(), out_enc.data(), out_enc_len,
out_enc.size(), kem, kdf, aead, ech_config.public_key.data(),
ech_config.public_key.size(), CBB_data(info.get()),
CBB_len(info.get())) ||
!hs->inner_transcript.Init()) {
return false;
}
hs->selected_ech_config = MakeUnique<ECHConfig>(std::move(ech_config));
return hs->selected_ech_config != nullptr;
}
}
}
return true;
}
static size_t aead_overhead(const EVP_HPKE_AEAD *aead) {
#if defined(BORINGSSL_UNSAFE_FUZZER_MODE)
// TODO(https://crbug.com/boringssl/275): Having to adjust the overhead
// everywhere is tedious. Change fuzzer mode to append a fake tag but still
// otherwise be cleartext, refresh corpora, and then inline this function.
return 0;
#else
return EVP_AEAD_max_overhead(EVP_HPKE_AEAD_aead(aead));
#endif
}
static size_t compute_extension_length(const EVP_HPKE_AEAD *aead,
size_t enc_len, size_t in_len) {
size_t ret = 4; // HpkeSymmetricCipherSuite cipher_suite
ret++; // uint8 config_id
ret += 2 + enc_len; // opaque enc<1..2^16-1>
ret += 2 + in_len + aead_overhead(aead); // opaque payload<1..2^16-1>
return ret;
}
// random_size returns a random value between |min| and |max|, inclusive.
static size_t random_size(size_t min, size_t max) {
assert(min < max);
size_t value;
RAND_bytes(reinterpret_cast<uint8_t *>(&value), sizeof(value));
return value % (max - min + 1) + min;
}
static bool setup_ech_grease(SSL_HANDSHAKE *hs) {
assert(!hs->selected_ech_config);
if (hs->max_version < TLS1_3_VERSION || !hs->config->ech_grease_enabled) {
return true;
}
const uint16_t kdf_id = EVP_HPKE_HKDF_SHA256;
const EVP_HPKE_AEAD *aead = EVP_has_aes_hardware()
? EVP_hpke_aes_128_gcm()
: EVP_hpke_chacha20_poly1305();
static_assert(ssl_grease_ech_config_id < sizeof(hs->grease_seed),
"hs->grease_seed is too small");
uint8_t config_id = hs->grease_seed[ssl_grease_ech_config_id];
uint8_t enc[X25519_PUBLIC_VALUE_LEN];
uint8_t private_key_unused[X25519_PRIVATE_KEY_LEN];
X25519_keypair(enc, private_key_unused);
// To determine a plausible length for the payload, we estimate the size of a
// typical EncodedClientHelloInner without resumption:
//
// 2+32+1+2 version, random, legacy_session_id, legacy_compression_methods
// 2+4*2 cipher_suites (three TLS 1.3 ciphers, GREASE)
// 2 extensions prefix
// 4 ech_is_inner
// 4+1+2*2 supported_versions (TLS 1.3, GREASE)
// 4+1+10*2 outer_extensions (key_share, sigalgs, sct, alpn,
// supported_groups, status_request, psk_key_exchange_modes,
// compress_certificate, GREASE x2)
//
// The server_name extension has an overhead of 9 bytes. For now, arbitrarily
// estimate maximum_name_length to be between 32 and 100 bytes.
//
// TODO(https://crbug.com/boringssl/275): If the padding scheme changes to
// also round the entire payload, adjust this to match. See
// https://github.com/tlswg/draft-ietf-tls-esni/issues/433
const size_t overhead = aead_overhead(aead);
const size_t in_len = random_size(128, 196);
const size_t extension_len =
compute_extension_length(aead, sizeof(enc), in_len);
bssl::ScopedCBB cbb;
CBB enc_cbb, payload_cbb;
uint8_t *payload;
if (!CBB_init(cbb.get(), extension_len) ||
!CBB_add_u16(cbb.get(), kdf_id) ||
!CBB_add_u16(cbb.get(), EVP_HPKE_AEAD_id(aead)) ||
!CBB_add_u8(cbb.get(), config_id) ||
!CBB_add_u16_length_prefixed(cbb.get(), &enc_cbb) ||
!CBB_add_bytes(&enc_cbb, enc, sizeof(enc)) ||
!CBB_add_u16_length_prefixed(cbb.get(), &payload_cbb) ||
!CBB_add_space(&payload_cbb, &payload, in_len + overhead) ||
!RAND_bytes(payload, in_len + overhead) ||
!CBBFinishArray(cbb.get(), &hs->ech_client_bytes)) {
return false;
}
assert(hs->ech_client_bytes.size() == extension_len);
return true;
}
bool ssl_encrypt_client_hello(SSL_HANDSHAKE *hs, Span<const uint8_t> enc) {
SSL *const ssl = hs->ssl;
if (!hs->selected_ech_config) {
return setup_ech_grease(hs);
}
// Construct ClientHelloInner and EncodedClientHelloInner. See
// draft-ietf-tls-esni-10, sections 5.1 and 6.1.
ScopedCBB cbb, encoded_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 body;
bool needs_psk_binder;
Array<uint8_t> hello_inner, encoded;
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->method->init_message(ssl, cbb.get(), &body, SSL3_MT_CLIENT_HELLO) ||
!CBB_init(encoded_cbb.get(), 256) ||
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_write_client_hello_without_extensions(hs, &body,
ssl_client_hello_inner,
/*empty_session_id=*/false) ||
!ssl_write_client_hello_without_extensions(hs, encoded_cbb.get(),
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_client_hello_inner,
/*empty_session_id=*/true) ||
!ssl_add_clienthello_tlsext(hs, &body, encoded_cbb.get(),
&needs_psk_binder, ssl_client_hello_inner,
CBB_len(&body),
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
/*omit_ech_len=*/0) ||
!ssl->method->finish_message(ssl, cbb.get(), &hello_inner) ||
!CBBFinishArray(encoded_cbb.get(), &encoded)) {
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
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
if (needs_psk_binder) {
size_t binder_len;
if (!tls13_write_psk_binder(hs, hs->inner_transcript, MakeSpan(hello_inner),
&binder_len)) {
return false;
}
// Also update the EncodedClientHelloInner.
auto encoded_binder = MakeSpan(encoded).last(binder_len);
auto hello_inner_binder = MakeConstSpan(hello_inner).last(binder_len);
OPENSSL_memcpy(encoded_binder.data(), hello_inner_binder.data(),
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
binder_len);
}
if (!hs->inner_transcript.Update(hello_inner)) {
return false;
}
// Construct ClientHelloOuterAAD. See draft-ietf-tls-esni-10, section 5.2.
// TODO(https://crbug.com/boringssl/275): This ends up constructing the
// ClientHelloOuter twice. Revisit this in the next draft, which uses a more
// forgiving construction.
const EVP_HPKE_KDF *kdf = EVP_HPKE_CTX_kdf(hs->ech_hpke_ctx.get());
const EVP_HPKE_AEAD *aead = EVP_HPKE_CTX_aead(hs->ech_hpke_ctx.get());
const size_t extension_len =
compute_extension_length(aead, enc.size(), encoded.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
bssl::ScopedCBB aad;
CBB outer_hello;
CBB enc_cbb;
if (!CBB_init(aad.get(), 256) ||
!CBB_add_u16(aad.get(), EVP_HPKE_KDF_id(kdf)) ||
!CBB_add_u16(aad.get(), EVP_HPKE_AEAD_id(aead)) ||
!CBB_add_u8(aad.get(), hs->selected_ech_config->config_id) ||
!CBB_add_u16_length_prefixed(aad.get(), &enc_cbb) ||
!CBB_add_bytes(&enc_cbb, enc.data(), enc.size()) ||
!CBB_add_u24_length_prefixed(aad.get(), &outer_hello) ||
!ssl_write_client_hello_without_extensions(hs, &outer_hello,
ssl_client_hello_outer,
/*empty_session_id=*/false) ||
!ssl_add_clienthello_tlsext(hs, &outer_hello, /*out_encoded=*/nullptr,
&needs_psk_binder, ssl_client_hello_outer,
CBB_len(&outer_hello),
/*omit_ech_len=*/4 + extension_len) ||
!CBB_flush(aad.get())) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return false;
}
// ClientHelloOuter may not require a PSK binder. Otherwise, we have a
// circular dependency.
assert(!needs_psk_binder);
CBB payload_cbb;
if (!CBB_init(cbb.get(), extension_len) ||
!CBB_add_u16(cbb.get(), EVP_HPKE_KDF_id(kdf)) ||
!CBB_add_u16(cbb.get(), EVP_HPKE_AEAD_id(aead)) ||
!CBB_add_u8(cbb.get(), hs->selected_ech_config->config_id) ||
!CBB_add_u16_length_prefixed(cbb.get(), &enc_cbb) ||
!CBB_add_bytes(&enc_cbb, enc.data(), enc.size()) ||
!CBB_add_u16_length_prefixed(cbb.get(), &payload_cbb)) {
return false;
}
#if defined(BORINGSSL_UNSAFE_FUZZER_MODE)
// In fuzzer mode, the server expects a cleartext payload.
if (!CBB_add_bytes(&payload_cbb, encoded.data(), encoded.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
return false;
}
#else
uint8_t *payload;
size_t payload_len =
encoded.size() + EVP_AEAD_max_overhead(EVP_HPKE_AEAD_aead(aead));
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_reserve(&payload_cbb, &payload, payload_len) ||
!EVP_HPKE_CTX_seal(hs->ech_hpke_ctx.get(), payload, &payload_len,
payload_len, encoded.data(), encoded.size(),
CBB_data(aad.get()), CBB_len(aad.get())) ||
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_did_write(&payload_cbb, payload_len)) {
return false;
}
#endif // BORINGSSL_UNSAFE_FUZZER_MODE
if (!CBBFinishArray(cbb.get(), &hs->ech_client_bytes)) {
return false;
}
// The |aad| calculation relies on |extension_length| being correct.
assert(hs->ech_client_bytes.size() == extension_len);
return true;
}
BSSL_NAMESPACE_END
using namespace bssl;
void SSL_set_enable_ech_grease(SSL *ssl, int enable) {
if (!ssl->config) {
return;
}
ssl->config->ech_grease_enabled = !!enable;
}
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
int SSL_set1_ech_config_list(SSL *ssl, const uint8_t *ech_config_list,
size_t ech_config_list_len) {
if (!ssl->config) {
return 0;
}
auto span = MakeConstSpan(ech_config_list, ech_config_list_len);
if (!ssl_is_valid_ech_config_list(span)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_INVALID_ECH_CONFIG_LIST);
return 0;
}
return ssl->config->client_ech_config_list.CopyFrom(span);
}
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>
4 years ago
void SSL_get0_ech_name_override(const SSL *ssl, const char **out_name,
size_t *out_name_len) {
// When ECH is rejected, we use the public name. Note that, if
// |SSL_CTX_set_reverify_on_resume| is enabled, we reverify the certificate
// before the 0-RTT point. If also offering ECH, we verify as if
// ClientHelloInner was accepted and do not override. This works because, at
// this point, |ech_status| will be |ssl_ech_none|. See the
// ECH-Client-Reject-EarlyDataReject-OverrideNameOnRetry tests in runner.go.
const SSL_HANDSHAKE *hs = ssl->s3->hs.get();
if (!ssl->server && hs && 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>
4 years ago
*out_name = reinterpret_cast<const char *>(
hs->selected_ech_config->public_name.data());
*out_name_len = hs->selected_ech_config->public_name.size();
} else {
*out_name = nullptr;
*out_name_len = 0;
}
}
void SSL_get0_ech_retry_configs(
const SSL *ssl, const uint8_t **out_retry_configs,
size_t *out_retry_configs_len) {
const SSL_HANDSHAKE *hs = ssl->s3->hs.get();
if (!hs || !hs->ech_authenticated_reject) {
// It is an error to call this function except in response to
// |SSL_R_ECH_REJECTED|. Returning an empty string risks the caller
// mistakenly believing the server has disabled ECH. Instead, return a
// non-empty ECHConfigList with a syntax error, so the subsequent
// |SSL_set1_ech_config_list| call will fail.
assert(0);
static const uint8_t kPlaceholder[] = {
kECHConfigVersion >> 8, kECHConfigVersion & 0xff, 0xff, 0xff, 0xff};
*out_retry_configs = kPlaceholder;
*out_retry_configs_len = sizeof(kPlaceholder);
return;
}
*out_retry_configs = hs->ech_retry_configs.data();
*out_retry_configs_len = hs->ech_retry_configs.size();
}
int SSL_marshal_ech_config(uint8_t **out, size_t *out_len, uint8_t config_id,
const EVP_HPKE_KEY *key, const char *public_name,
size_t max_name_len) {
Span<const uint8_t> public_name_u8 = MakeConstSpan(
reinterpret_cast<const uint8_t *>(public_name), strlen(public_name));
if (!ssl_is_valid_ech_public_name(public_name_u8)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_INVALID_ECH_PUBLIC_NAME);
return 0;
}
// See draft-ietf-tls-esni-10, section 4.
ScopedCBB cbb;
CBB contents, child;
uint8_t *public_key;
size_t public_key_len;
if (!CBB_init(cbb.get(), 128) || //
!CBB_add_u16(cbb.get(), kECHConfigVersion) ||
!CBB_add_u16_length_prefixed(cbb.get(), &contents) ||
!CBB_add_u8(&contents, config_id) ||
!CBB_add_u16(&contents, EVP_HPKE_KEM_id(EVP_HPKE_KEY_kem(key))) ||
!CBB_add_u16_length_prefixed(&contents, &child) ||
!CBB_reserve(&child, &public_key, EVP_HPKE_MAX_PUBLIC_KEY_LENGTH) ||
!EVP_HPKE_KEY_public_key(key, public_key, &public_key_len,
EVP_HPKE_MAX_PUBLIC_KEY_LENGTH) ||
!CBB_did_write(&child, public_key_len) ||
!CBB_add_u16_length_prefixed(&contents, &child) ||
// Write a default cipher suite configuration.
!CBB_add_u16(&child, EVP_HPKE_HKDF_SHA256) ||
!CBB_add_u16(&child, EVP_HPKE_AES_128_GCM) ||
!CBB_add_u16(&child, EVP_HPKE_HKDF_SHA256) ||
!CBB_add_u16(&child, EVP_HPKE_CHACHA20_POLY1305) ||
!CBB_add_u16(&contents, max_name_len) ||
!CBB_add_u16_length_prefixed(&contents, &child) ||
!CBB_add_bytes(&child, public_name_u8.data(), public_name_u8.size()) ||
// TODO(https://crbug.com/boringssl/275): Reserve some GREASE extensions
// and include some.
!CBB_add_u16(&contents, 0 /* no extensions */) ||
!CBB_finish(cbb.get(), out, out_len)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_INTERNAL_ERROR);
return 0;
}
return 1;
}
SSL_ECH_KEYS *SSL_ECH_KEYS_new() { return New<SSL_ECH_KEYS>(); }
void SSL_ECH_KEYS_up_ref(SSL_ECH_KEYS *keys) {
CRYPTO_refcount_inc(&keys->references);
}
void SSL_ECH_KEYS_free(SSL_ECH_KEYS *keys) {
if (keys == nullptr ||
!CRYPTO_refcount_dec_and_test_zero(&keys->references)) {
return;
}
keys->~ssl_ech_keys_st();
OPENSSL_free(keys);
}
int SSL_ECH_KEYS_add(SSL_ECH_KEYS *configs, int is_retry_config,
const uint8_t *ech_config, size_t ech_config_len,
const EVP_HPKE_KEY *key) {
UniquePtr<ECHServerConfig> parsed_config = MakeUnique<ECHServerConfig>();
if (!parsed_config) {
return 0;
}
if (!parsed_config->Init(MakeConstSpan(ech_config, ech_config_len), key,
!!is_retry_config)) {
OPENSSL_PUT_ERROR(SSL, SSL_R_DECODE_ERROR);
return 0;
}
if (!configs->configs.Push(std::move(parsed_config))) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return 0;
}
return 1;
}
int SSL_ECH_KEYS_has_duplicate_config_id(const SSL_ECH_KEYS *keys) {
bool seen[256] = {false};
for (const auto &config : keys->configs) {
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 (seen[config->ech_config().config_id]) {
return 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
seen[config->ech_config().config_id] = true;
}
return 0;
}
int SSL_ECH_KEYS_marshal_retry_configs(const SSL_ECH_KEYS *keys, uint8_t **out,
size_t *out_len) {
ScopedCBB cbb;
CBB child;
if (!CBB_init(cbb.get(), 128) ||
!CBB_add_u16_length_prefixed(cbb.get(), &child)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return false;
}
for (const auto &config : keys->configs) {
if (config->is_retry_config() &&
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(&child, config->ech_config().raw.data(),
config->ech_config().raw.size())) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return false;
}
}
return CBB_finish(cbb.get(), out, out_len);
}
int SSL_CTX_set1_ech_keys(SSL_CTX *ctx, SSL_ECH_KEYS *keys) {
bool has_retry_config = false;
for (const auto &config : keys->configs) {
if (config->is_retry_config()) {
has_retry_config = true;
break;
}
}
if (!has_retry_config) {
OPENSSL_PUT_ERROR(SSL, SSL_R_ECH_SERVER_WOULD_HAVE_NO_RETRY_CONFIGS);
return 0;
}
UniquePtr<SSL_ECH_KEYS> owned_keys = UpRef(keys);
MutexWriteLock lock(&ctx->lock);
ctx->ech_keys.swap(owned_keys);
return 1;
}
int SSL_ech_accepted(const SSL *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_in_early_data(ssl) && !ssl->server) {
// In the client early data state, we report properties as if the server
// accepted early data. The server can only accept early data with
// ClientHelloInner.
return ssl->s3->hs->selected_ech_config != nullptr;
}
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>
4 years ago
return ssl->s3->ech_status == ssl_ech_accepted;
}