boringssl/ssl/internal.h

3400 lines
135 KiB
C
Raw Normal View History

/* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
* All rights reserved.
*
* This package is an SSL implementation written
* by Eric Young (eay@cryptsoft.com).
* The implementation was written so as to conform with Netscapes SSL.
*
* This library is free for commercial and non-commercial use as long as
* the following conditions are aheared to. The following conditions
* apply to all code found in this distribution, be it the RC4, RSA,
* lhash, DES, etc., code; not just the SSL code. The SSL documentation
* included with this distribution is covered by the same copyright terms
* except that the holder is Tim Hudson (tjh@cryptsoft.com).
*
* Copyright remains Eric Young's, and as such any Copyright notices in
* the code are not to be removed.
* If this package is used in a product, Eric Young should be given attribution
* as the author of the parts of the library used.
* This can be in the form of a textual message at program startup or
* in documentation (online or textual) provided with the package.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* "This product includes cryptographic software written by
* Eric Young (eay@cryptsoft.com)"
* The word 'cryptographic' can be left out if the rouines from the library
* being used are not cryptographic related :-).
* 4. If you include any Windows specific code (or a derivative thereof) from
* the apps directory (application code) you must include an acknowledgement:
* "This product includes software written by Tim Hudson (tjh@cryptsoft.com)"
*
* THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* The licence and distribution terms for any publically available version or
* derivative of this code cannot be changed. i.e. this code cannot simply be
* copied and put under another distribution licence
* [including the GNU Public Licence.]
*/
/* ====================================================================
* Copyright (c) 1998-2007 The OpenSSL Project. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
*
* 3. All advertising materials mentioning features or use of this
* software must display the following acknowledgment:
* "This product includes software developed by the OpenSSL Project
* for use in the OpenSSL Toolkit. (http://www.openssl.org/)"
*
* 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
* endorse or promote products derived from this software without
* prior written permission. For written permission, please contact
* openssl-core@openssl.org.
*
* 5. Products derived from this software may not be called "OpenSSL"
* nor may "OpenSSL" appear in their names without prior written
* permission of the OpenSSL Project.
*
* 6. Redistributions of any form whatsoever must retain the following
* acknowledgment:
* "This product includes software developed by the OpenSSL Project
* for use in the OpenSSL Toolkit (http://www.openssl.org/)"
*
* THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
* EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR
* ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
* STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
* OF THE POSSIBILITY OF SUCH DAMAGE.
* ====================================================================
*
* This product includes cryptographic software written by Eric Young
* (eay@cryptsoft.com). This product includes software written by Tim
* Hudson (tjh@cryptsoft.com).
*
*/
/* ====================================================================
* Copyright 2002 Sun Microsystems, Inc. ALL RIGHTS RESERVED.
* ECC cipher suite support in OpenSSL originally developed by
* SUN MICROSYSTEMS, INC., and contributed to the OpenSSL project.
*/
/* ====================================================================
* Copyright 2005 Nokia. All rights reserved.
*
* The portions of the attached software ("Contribution") is developed by
* Nokia Corporation and is licensed pursuant to the OpenSSL open source
* license.
*
* The Contribution, originally written by Mika Kousa and Pasi Eronen of
* Nokia Corporation, consists of the "PSK" (Pre-Shared Key) ciphersuites
* support (see RFC 4279) to OpenSSL.
*
* No patent licenses or other rights except those expressly stated in
* the OpenSSL open source license shall be deemed granted or received
* expressly, by implication, estoppel, or otherwise.
*
* No assurances are provided by Nokia that the Contribution does not
* infringe the patent or other intellectual property rights of any third
* party or that the license provides you with all the necessary rights
* to make use of the Contribution.
*
* THE SOFTWARE IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY KIND. IN
* ADDITION TO THE DISCLAIMERS INCLUDED IN THE LICENSE, NOKIA
* SPECIFICALLY DISCLAIMS ANY LIABILITY FOR CLAIMS BROUGHT BY YOU OR ANY
* OTHER ENTITY BASED ON INFRINGEMENT OF INTELLECTUAL PROPERTY RIGHTS OR
* OTHERWISE.
*/
#ifndef OPENSSL_HEADER_SSL_INTERNAL_H
#define OPENSSL_HEADER_SSL_INTERNAL_H
#include <openssl/base.h>
#include <stdlib.h>
#include <limits>
#include <new>
#include <type_traits>
#include <utility>
#include <openssl/aead.h>
#include <openssl/err.h>
#include <openssl/lhash.h>
#include <openssl/mem.h>
#include <openssl/span.h>
#include <openssl/ssl.h>
#include <openssl/stack.h>
#include "../crypto/err/internal.h"
#include "../crypto/internal.h"
#if defined(OPENSSL_WINDOWS)
// Windows defines struct timeval in winsock2.h.
OPENSSL_MSVC_PRAGMA(warning(push, 3))
#include <winsock2.h>
OPENSSL_MSVC_PRAGMA(warning(pop))
#else
#include <sys/time.h>
#endif
BSSL_NAMESPACE_BEGIN
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
struct SSL_CONFIG;
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
struct SSL_HANDSHAKE;
struct SSL_PROTOCOL_METHOD;
struct SSL_X509_METHOD;
Revise version negotiation logic on the C side. This is in preparation for upcoming experiments which will require supporting multiple experimental versions of TLS 1.3 with, on the server, the ability to enable multiple variants at once. This means the version <-> wire bijection no longer exists, even when limiting to a single SSL*. Thus version_to_wire is removed and instead we treat the wire version as the canonical version value. There is a mapping from valid wire versions to protocol versions which describe the high-level handshake protocol in use. This mapping is not injective, so uses of version_from_wire are rewritten differently. All the version-munging logic is moved to ssl_versions.c with a master preference list of all TLS and DTLS versions. The legacy version negotiation is converted to the new scheme. The version lists and negotiation are driven by the preference lists and a ssl_supports_version API. To simplify the mess around SSL_SESSION and versions, version_from_wire is now DTLS/TLS-agnostic, with any filtering being done by ssl_supports_version. This is screwy but allows parsing SSL_SESSIONs to sanity-check it and reject all bogus versions in SSL_SESSION. This reduces a mess of error cases. As part of this, the weird logic where ssl->version is set early when sending the ClientHello is removed. The one place where we were relying on this behavior is tweaked to query hs->max_version instead. Change-Id: Ic91b348481ceba94d9ae06d6781187c11adc15b0 Reviewed-on: https://boringssl-review.googlesource.com/17524 Reviewed-by: David Benjamin <davidben@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2017-06-20 15:55:02 +01:00
// C++ utilities.
// New behaves like |new| but uses |OPENSSL_malloc| for memory allocation. It
// returns nullptr on allocation error. It only implements single-object
// allocation and not new T[n].
//
// Note: unlike |new|, this does not support non-public constructors.
template <typename T, typename... Args>
T *New(Args &&... args) {
void *t = OPENSSL_malloc(sizeof(T));
if (t == nullptr) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return nullptr;
}
return new (t) T(std::forward<Args>(args)...);
}
// Delete behaves like |delete| but uses |OPENSSL_free| to release memory.
//
// Note: unlike |delete| this does not support non-public destructors.
template <typename T>
void Delete(T *t) {
if (t != nullptr) {
t->~T();
OPENSSL_free(t);
}
}
// All types with kAllowUniquePtr set may be used with UniquePtr. Other types
// may be C structs which require a |BORINGSSL_MAKE_DELETER| registration.
namespace internal {
template <typename T>
struct DeleterImpl<T, typename std::enable_if<T::kAllowUniquePtr>::type> {
static void Free(T *t) { Delete(t); }
};
} // namespace internal
// MakeUnique behaves like |std::make_unique| but returns nullptr on allocation
// error.
template <typename T, typename... Args>
UniquePtr<T> MakeUnique(Args &&... args) {
return UniquePtr<T>(New<T>(std::forward<Args>(args)...));
}
#if defined(BORINGSSL_ALLOW_CXX_RUNTIME)
#define HAS_VIRTUAL_DESTRUCTOR
#define PURE_VIRTUAL = 0
#else
// HAS_VIRTUAL_DESTRUCTOR should be declared in any base class which defines a
// virtual destructor. This avoids a dependency on |_ZdlPv| and prevents the
// class from being used with |delete|.
#define HAS_VIRTUAL_DESTRUCTOR \
void operator delete(void *) { abort(); }
// PURE_VIRTUAL should be used instead of = 0 when defining pure-virtual
// functions. This avoids a dependency on |__cxa_pure_virtual| but loses
// compile-time checking.
#define PURE_VIRTUAL \
{ abort(); }
#endif
// CONSTEXPR_ARRAY works around a VS 2015 bug where ranged for loops don't work
// on constexpr arrays.
#if defined(_MSC_VER) && !defined(__clang__) && _MSC_VER < 1910
#define CONSTEXPR_ARRAY const
#else
#define CONSTEXPR_ARRAY constexpr
#endif
// Array<T> is an owning array of elements of |T|.
template <typename T>
class Array {
public:
// Array's default constructor creates an empty array.
Array() {}
Array(const Array &) = delete;
Array(Array &&other) { *this = std::move(other); }
~Array() { Reset(); }
Array &operator=(const Array &) = delete;
Array &operator=(Array &&other) {
Reset();
other.Release(&data_, &size_);
return *this;
}
const T *data() const { return data_; }
T *data() { return data_; }
size_t size() const { return size_; }
bool empty() const { return size_ == 0; }
const T &operator[](size_t i) const { return data_[i]; }
T &operator[](size_t i) { return data_[i]; }
T *begin() { return data_; }
const T *cbegin() const { return data_; }
T *end() { return data_ + size_; }
const T *cend() const { return data_ + size_; }
void Reset() { Reset(nullptr, 0); }
// Reset releases the current contents of the array and takes ownership of the
// raw pointer supplied by the caller.
void Reset(T *new_data, size_t new_size) {
for (size_t i = 0; i < size_; i++) {
data_[i].~T();
}
OPENSSL_free(data_);
data_ = new_data;
size_ = new_size;
}
// Release releases ownership of the array to a raw pointer supplied by the
// caller.
void Release(T **out, size_t *out_size) {
*out = data_;
*out_size = size_;
data_ = nullptr;
size_ = 0;
}
// Init replaces the array with a newly-allocated array of |new_size|
// default-constructed copies of |T|. It returns true on success and false on
// error.
//
// Note that if |T| is a primitive type like |uint8_t|, it is uninitialized.
bool Init(size_t new_size) {
Reset();
if (new_size == 0) {
return true;
}
if (new_size > std::numeric_limits<size_t>::max() / sizeof(T)) {
OPENSSL_PUT_ERROR(SSL, ERR_R_OVERFLOW);
return false;
}
data_ = reinterpret_cast<T *>(OPENSSL_malloc(new_size * sizeof(T)));
if (data_ == nullptr) {
OPENSSL_PUT_ERROR(SSL, ERR_R_MALLOC_FAILURE);
return false;
}
size_ = new_size;
for (size_t i = 0; i < size_; i++) {
new (&data_[i]) T;
}
return true;
}
// CopyFrom replaces the array with a newly-allocated copy of |in|. It returns
// true on success and false on error.
bool CopyFrom(Span<const T> in) {
if (!Init(in.size())) {
return false;
}
OPENSSL_memcpy(data_, in.data(), sizeof(T) * in.size());
return true;
}
// Shrink shrinks the stored size of the array to |new_size|. It crashes if
// the new size is larger. Note this does not shrink the allocation itself.
void Shrink(size_t new_size) {
if (new_size > size_) {
abort();
}
size_ = new_size;
}
private:
T *data_ = nullptr;
size_t size_ = 0;
};
// CBBFinishArray behaves like |CBB_finish| but stores the result in an Array.
OPENSSL_EXPORT bool CBBFinishArray(CBB *cbb, Array<uint8_t> *out);
// Protocol versions.
//
// Due to DTLS's historical wire version differences, we maintain two notions of
// version.
//
// The "version" or "wire version" is the actual 16-bit value that appears on
// the wire. It uniquely identifies a version and is also used at API
// boundaries. The set of supported versions differs between TLS and DTLS. Wire
// versions are opaque values and may not be compared numerically.
//
// The "protocol version" identifies the high-level handshake variant being
// used. DTLS versions map to the corresponding TLS versions. Protocol versions
// are sequential and may be compared numerically.
// ssl_protocol_version_from_wire sets |*out| to the protocol version
// corresponding to wire version |version| and returns true. If |version| is not
// a valid TLS or DTLS version, it returns false.
//
// Note this simultaneously handles both DTLS and TLS. Use one of the
// higher-level functions below for most operations.
bool ssl_protocol_version_from_wire(uint16_t *out, uint16_t version);
Revise version negotiation logic on the C side. This is in preparation for upcoming experiments which will require supporting multiple experimental versions of TLS 1.3 with, on the server, the ability to enable multiple variants at once. This means the version <-> wire bijection no longer exists, even when limiting to a single SSL*. Thus version_to_wire is removed and instead we treat the wire version as the canonical version value. There is a mapping from valid wire versions to protocol versions which describe the high-level handshake protocol in use. This mapping is not injective, so uses of version_from_wire are rewritten differently. All the version-munging logic is moved to ssl_versions.c with a master preference list of all TLS and DTLS versions. The legacy version negotiation is converted to the new scheme. The version lists and negotiation are driven by the preference lists and a ssl_supports_version API. To simplify the mess around SSL_SESSION and versions, version_from_wire is now DTLS/TLS-agnostic, with any filtering being done by ssl_supports_version. This is screwy but allows parsing SSL_SESSIONs to sanity-check it and reject all bogus versions in SSL_SESSION. This reduces a mess of error cases. As part of this, the weird logic where ssl->version is set early when sending the ClientHello is removed. The one place where we were relying on this behavior is tweaked to query hs->max_version instead. Change-Id: Ic91b348481ceba94d9ae06d6781187c11adc15b0 Reviewed-on: https://boringssl-review.googlesource.com/17524 Reviewed-by: David Benjamin <davidben@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2017-06-20 15:55:02 +01:00
// ssl_get_version_range sets |*out_min_version| and |*out_max_version| to the
// minimum and maximum enabled protocol versions, respectively.
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
bool ssl_get_version_range(const SSL_HANDSHAKE *hs, uint16_t *out_min_version,
uint16_t *out_max_version);
Revise version negotiation logic on the C side. This is in preparation for upcoming experiments which will require supporting multiple experimental versions of TLS 1.3 with, on the server, the ability to enable multiple variants at once. This means the version <-> wire bijection no longer exists, even when limiting to a single SSL*. Thus version_to_wire is removed and instead we treat the wire version as the canonical version value. There is a mapping from valid wire versions to protocol versions which describe the high-level handshake protocol in use. This mapping is not injective, so uses of version_from_wire are rewritten differently. All the version-munging logic is moved to ssl_versions.c with a master preference list of all TLS and DTLS versions. The legacy version negotiation is converted to the new scheme. The version lists and negotiation are driven by the preference lists and a ssl_supports_version API. To simplify the mess around SSL_SESSION and versions, version_from_wire is now DTLS/TLS-agnostic, with any filtering being done by ssl_supports_version. This is screwy but allows parsing SSL_SESSIONs to sanity-check it and reject all bogus versions in SSL_SESSION. This reduces a mess of error cases. As part of this, the weird logic where ssl->version is set early when sending the ClientHello is removed. The one place where we were relying on this behavior is tweaked to query hs->max_version instead. Change-Id: Ic91b348481ceba94d9ae06d6781187c11adc15b0 Reviewed-on: https://boringssl-review.googlesource.com/17524 Reviewed-by: David Benjamin <davidben@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2017-06-20 15:55:02 +01:00
// ssl_supports_version returns whether |hs| supports |version|.
bool ssl_supports_version(SSL_HANDSHAKE *hs, uint16_t version);
Revise version negotiation logic on the C side. This is in preparation for upcoming experiments which will require supporting multiple experimental versions of TLS 1.3 with, on the server, the ability to enable multiple variants at once. This means the version <-> wire bijection no longer exists, even when limiting to a single SSL*. Thus version_to_wire is removed and instead we treat the wire version as the canonical version value. There is a mapping from valid wire versions to protocol versions which describe the high-level handshake protocol in use. This mapping is not injective, so uses of version_from_wire are rewritten differently. All the version-munging logic is moved to ssl_versions.c with a master preference list of all TLS and DTLS versions. The legacy version negotiation is converted to the new scheme. The version lists and negotiation are driven by the preference lists and a ssl_supports_version API. To simplify the mess around SSL_SESSION and versions, version_from_wire is now DTLS/TLS-agnostic, with any filtering being done by ssl_supports_version. This is screwy but allows parsing SSL_SESSIONs to sanity-check it and reject all bogus versions in SSL_SESSION. This reduces a mess of error cases. As part of this, the weird logic where ssl->version is set early when sending the ClientHello is removed. The one place where we were relying on this behavior is tweaked to query hs->max_version instead. Change-Id: Ic91b348481ceba94d9ae06d6781187c11adc15b0 Reviewed-on: https://boringssl-review.googlesource.com/17524 Reviewed-by: David Benjamin <davidben@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2017-06-20 15:55:02 +01:00
// ssl_method_supports_version returns whether |method| supports |version|.
bool ssl_method_supports_version(const SSL_PROTOCOL_METHOD *method,
uint16_t version);
// ssl_add_supported_versions writes the supported versions of |hs| to |cbb|, in
// decreasing preference order.
bool ssl_add_supported_versions(SSL_HANDSHAKE *hs, CBB *cbb);
Revise version negotiation logic on the C side. This is in preparation for upcoming experiments which will require supporting multiple experimental versions of TLS 1.3 with, on the server, the ability to enable multiple variants at once. This means the version <-> wire bijection no longer exists, even when limiting to a single SSL*. Thus version_to_wire is removed and instead we treat the wire version as the canonical version value. There is a mapping from valid wire versions to protocol versions which describe the high-level handshake protocol in use. This mapping is not injective, so uses of version_from_wire are rewritten differently. All the version-munging logic is moved to ssl_versions.c with a master preference list of all TLS and DTLS versions. The legacy version negotiation is converted to the new scheme. The version lists and negotiation are driven by the preference lists and a ssl_supports_version API. To simplify the mess around SSL_SESSION and versions, version_from_wire is now DTLS/TLS-agnostic, with any filtering being done by ssl_supports_version. This is screwy but allows parsing SSL_SESSIONs to sanity-check it and reject all bogus versions in SSL_SESSION. This reduces a mess of error cases. As part of this, the weird logic where ssl->version is set early when sending the ClientHello is removed. The one place where we were relying on this behavior is tweaked to query hs->max_version instead. Change-Id: Ic91b348481ceba94d9ae06d6781187c11adc15b0 Reviewed-on: https://boringssl-review.googlesource.com/17524 Reviewed-by: David Benjamin <davidben@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2017-06-20 15:55:02 +01:00
// ssl_negotiate_version negotiates a common version based on |hs|'s preferences
// and the peer preference list in |peer_versions|. On success, it returns true
// and sets |*out_version| to the selected version. Otherwise, it returns false
// and sets |*out_alert| to an alert to send.
bool ssl_negotiate_version(SSL_HANDSHAKE *hs, uint8_t *out_alert,
uint16_t *out_version, const CBS *peer_versions);
Revise version negotiation logic on the C side. This is in preparation for upcoming experiments which will require supporting multiple experimental versions of TLS 1.3 with, on the server, the ability to enable multiple variants at once. This means the version <-> wire bijection no longer exists, even when limiting to a single SSL*. Thus version_to_wire is removed and instead we treat the wire version as the canonical version value. There is a mapping from valid wire versions to protocol versions which describe the high-level handshake protocol in use. This mapping is not injective, so uses of version_from_wire are rewritten differently. All the version-munging logic is moved to ssl_versions.c with a master preference list of all TLS and DTLS versions. The legacy version negotiation is converted to the new scheme. The version lists and negotiation are driven by the preference lists and a ssl_supports_version API. To simplify the mess around SSL_SESSION and versions, version_from_wire is now DTLS/TLS-agnostic, with any filtering being done by ssl_supports_version. This is screwy but allows parsing SSL_SESSIONs to sanity-check it and reject all bogus versions in SSL_SESSION. This reduces a mess of error cases. As part of this, the weird logic where ssl->version is set early when sending the ClientHello is removed. The one place where we were relying on this behavior is tweaked to query hs->max_version instead. Change-Id: Ic91b348481ceba94d9ae06d6781187c11adc15b0 Reviewed-on: https://boringssl-review.googlesource.com/17524 Reviewed-by: David Benjamin <davidben@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2017-06-20 15:55:02 +01:00
// ssl_protocol_version returns |ssl|'s protocol version. It is an error to
// call this function before the version is determined.
uint16_t ssl_protocol_version(const SSL *ssl);
Revise version negotiation logic on the C side. This is in preparation for upcoming experiments which will require supporting multiple experimental versions of TLS 1.3 with, on the server, the ability to enable multiple variants at once. This means the version <-> wire bijection no longer exists, even when limiting to a single SSL*. Thus version_to_wire is removed and instead we treat the wire version as the canonical version value. There is a mapping from valid wire versions to protocol versions which describe the high-level handshake protocol in use. This mapping is not injective, so uses of version_from_wire are rewritten differently. All the version-munging logic is moved to ssl_versions.c with a master preference list of all TLS and DTLS versions. The legacy version negotiation is converted to the new scheme. The version lists and negotiation are driven by the preference lists and a ssl_supports_version API. To simplify the mess around SSL_SESSION and versions, version_from_wire is now DTLS/TLS-agnostic, with any filtering being done by ssl_supports_version. This is screwy but allows parsing SSL_SESSIONs to sanity-check it and reject all bogus versions in SSL_SESSION. This reduces a mess of error cases. As part of this, the weird logic where ssl->version is set early when sending the ClientHello is removed. The one place where we were relying on this behavior is tweaked to query hs->max_version instead. Change-Id: Ic91b348481ceba94d9ae06d6781187c11adc15b0 Reviewed-on: https://boringssl-review.googlesource.com/17524 Reviewed-by: David Benjamin <davidben@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2017-06-20 15:55:02 +01:00
// Cipher suites.
BSSL_NAMESPACE_END
struct ssl_cipher_st {
// name is the OpenSSL name for the cipher.
const char *name;
// standard_name is the IETF name for the cipher.
const char *standard_name;
// id is the cipher suite value bitwise OR-d with 0x03000000.
uint32_t id;
// algorithm_* determine the cipher suite. See constants below for the values.
uint32_t algorithm_mkey;
uint32_t algorithm_auth;
uint32_t algorithm_enc;
uint32_t algorithm_mac;
uint32_t algorithm_prf;
};
BSSL_NAMESPACE_BEGIN
// Bits for |algorithm_mkey| (key exchange algorithm).
#define SSL_kRSA 0x00000001u
#define SSL_kECDHE 0x00000002u
// SSL_kPSK is only set for plain PSK, not ECDHE_PSK.
#define SSL_kPSK 0x00000004u
#define SSL_kGENERIC 0x00000008u
// Bits for |algorithm_auth| (server authentication).
#define SSL_aRSA 0x00000001u
#define SSL_aECDSA 0x00000002u
// SSL_aPSK is set for both PSK and ECDHE_PSK.
#define SSL_aPSK 0x00000004u
#define SSL_aGENERIC 0x00000008u
#define SSL_aCERT (SSL_aRSA | SSL_aECDSA)
// Bits for |algorithm_enc| (symmetric encryption).
#define SSL_3DES 0x00000001u
#define SSL_AES128 0x00000002u
#define SSL_AES256 0x00000004u
#define SSL_AES128GCM 0x00000008u
#define SSL_AES256GCM 0x00000010u
#define SSL_eNULL 0x00000020u
#define SSL_CHACHA20POLY1305 0x00000040u
#define SSL_AES (SSL_AES128 | SSL_AES256 | SSL_AES128GCM | SSL_AES256GCM)
// Bits for |algorithm_mac| (symmetric authentication).
#define SSL_SHA1 0x00000001u
// SSL_AEAD is set for all AEADs.
#define SSL_AEAD 0x00000002u
// Bits for |algorithm_prf| (handshake digest).
#define SSL_HANDSHAKE_MAC_DEFAULT 0x1
#define SSL_HANDSHAKE_MAC_SHA256 0x2
#define SSL_HANDSHAKE_MAC_SHA384 0x4
// An SSLCipherPreferenceList contains a list of SSL_CIPHERs with equal-
// preference groups. For TLS clients, the groups are moot because the server
// picks the cipher and groups cannot be expressed on the wire. However, for
// servers, the equal-preference groups allow the client's preferences to be
// partially respected. (This only has an effect with
// SSL_OP_CIPHER_SERVER_PREFERENCE).
//
// The equal-preference groups are expressed by grouping SSL_CIPHERs together.
// All elements of a group have the same priority: no ordering is expressed
// within a group.
//
// The values in |ciphers| are in one-to-one correspondence with
// |in_group_flags|. (That is, sk_SSL_CIPHER_num(ciphers) is the number of
// bytes in |in_group_flags|.) The bytes in |in_group_flags| are either 1, to
// indicate that the corresponding SSL_CIPHER is not the last element of a
// group, or 0 to indicate that it is.
//
// For example, if |in_group_flags| contains all zeros then that indicates a
// traditional, fully-ordered preference. Every SSL_CIPHER is the last element
// of the group (i.e. they are all in a one-element group).
//
// For a more complex example, consider:
// ciphers: A B C D E F
// in_group_flags: 1 1 0 0 1 0
//
// That would express the following, order:
//
// A E
// B -> D -> F
// C
struct SSLCipherPreferenceList {
static constexpr bool kAllowUniquePtr = true;
SSLCipherPreferenceList() = default;
~SSLCipherPreferenceList();
bool Init(UniquePtr<STACK_OF(SSL_CIPHER)> ciphers,
Span<const bool> in_group_flags);
bool Init(const SSLCipherPreferenceList &);
void Remove(const SSL_CIPHER *cipher);
UniquePtr<STACK_OF(SSL_CIPHER)> ciphers;
bool *in_group_flags = nullptr;
};
// AllCiphers returns an array of all supported ciphers, sorted by id.
Span<const SSL_CIPHER> AllCiphers();
// ssl_cipher_get_evp_aead sets |*out_aead| to point to the correct EVP_AEAD
// object for |cipher| protocol version |version|. It sets |*out_mac_secret_len|
// and |*out_fixed_iv_len| to the MAC key length and fixed IV length,
// respectively. The MAC key length is zero except for legacy block and stream
// ciphers. It returns true on success and false on error.
bool ssl_cipher_get_evp_aead(const EVP_AEAD **out_aead,
size_t *out_mac_secret_len,
size_t *out_fixed_iv_len, const SSL_CIPHER *cipher,
uint16_t version, bool is_dtls);
// ssl_get_handshake_digest returns the |EVP_MD| corresponding to |version| and
// |cipher|.
const EVP_MD *ssl_get_handshake_digest(uint16_t version,
const SSL_CIPHER *cipher);
// ssl_create_cipher_list evaluates |rule_str|. It sets |*out_cipher_list| to a
// newly-allocated |SSLCipherPreferenceList| containing the result. It returns
// true on success and false on failure. If |strict| is true, nonsense will be
// rejected. If false, nonsense will be silently ignored. An empty result is
// considered an error regardless of |strict|.
bool ssl_create_cipher_list(UniquePtr<SSLCipherPreferenceList> *out_cipher_list,
const char *rule_str, bool strict);
// ssl_cipher_get_value returns the cipher suite id of |cipher|.
uint16_t ssl_cipher_get_value(const SSL_CIPHER *cipher);
// ssl_cipher_auth_mask_for_key returns the mask of cipher |algorithm_auth|
// values suitable for use with |key| in TLS 1.2 and below.
uint32_t ssl_cipher_auth_mask_for_key(const EVP_PKEY *key);
// ssl_cipher_uses_certificate_auth returns whether |cipher| authenticates the
// server and, optionally, the client with a certificate.
bool ssl_cipher_uses_certificate_auth(const SSL_CIPHER *cipher);
// ssl_cipher_requires_server_key_exchange returns whether |cipher| requires a
// ServerKeyExchange message.
//
// This function may return false while still allowing |cipher| an optional
// ServerKeyExchange. This is the case for plain PSK ciphers.
bool ssl_cipher_requires_server_key_exchange(const SSL_CIPHER *cipher);
// ssl_cipher_get_record_split_len, for TLS 1.0 CBC mode ciphers, returns the
// length of an encrypted 1-byte record, for use in record-splitting. Otherwise
// it returns zero.
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
size_t ssl_cipher_get_record_split_len(const SSL_CIPHER *cipher);
// Transcript layer.
// SSLTranscript maintains the handshake transcript as a combination of a
// buffer and running hash.
class SSLTranscript {
public:
SSLTranscript();
~SSLTranscript();
// Init initializes the handshake transcript. If called on an existing
// transcript, it resets the transcript and hash. It returns true on success
// and false on failure.
bool Init();
// InitHash initializes the handshake hash based on the PRF and contents of
// the handshake transcript. Subsequent calls to |Update| will update the
// rolling hash. It returns one on success and zero on failure. It is an error
// to call this function after the handshake buffer is released.
bool InitHash(uint16_t version, const SSL_CIPHER *cipher);
// UpdateForHelloRetryRequest resets the rolling hash with the
// HelloRetryRequest construction. It returns true on success and false on
// failure. It is an error to call this function before the handshake buffer
// is released.
bool UpdateForHelloRetryRequest();
// CopyHashContext copies the hash context into |ctx| and returns true on
// success.
bool CopyHashContext(EVP_MD_CTX *ctx);
Span<const uint8_t> buffer() {
return MakeConstSpan(reinterpret_cast<const uint8_t *>(buffer_->data),
buffer_->length);
}
// FreeBuffer releases the handshake buffer. Subsequent calls to
// |Update| will not update the handshake buffer.
void FreeBuffer();
// DigestLen returns the length of the PRF hash.
size_t DigestLen() const;
// Digest returns the PRF hash. For TLS 1.1 and below, this is
// |EVP_md5_sha1|.
const EVP_MD *Digest() const;
// Update adds |in| to the handshake buffer and handshake hash, whichever is
// enabled. It returns true on success and false on failure.
bool Update(Span<const uint8_t> in);
// GetHash writes the handshake hash to |out| which must have room for at
// least |DigestLen| bytes. On success, it returns true and sets |*out_len| to
// the number of bytes written. Otherwise, it returns false.
bool GetHash(uint8_t *out, size_t *out_len);
// GetFinishedMAC computes the MAC for the Finished message into the bytes
// pointed by |out| and writes the number of bytes to |*out_len|. |out| must
// have room for |EVP_MAX_MD_SIZE| bytes. It returns true on success and false
// on failure.
bool GetFinishedMAC(uint8_t *out, size_t *out_len, const SSL_SESSION *session,
bool from_server);
private:
// buffer_, if non-null, contains the handshake transcript.
UniquePtr<BUF_MEM> buffer_;
// hash, if initialized with an |EVP_MD|, maintains the handshake hash.
ScopedEVP_MD_CTX hash_;
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
};
// tls1_prf computes the PRF function for |ssl|. It fills |out|, using |secret|
// as the secret and |label| as the label. |seed1| and |seed2| are concatenated
// to form the seed parameter. It returns true on success and false on failure.
bool tls1_prf(const EVP_MD *digest, Span<uint8_t> out,
Span<const uint8_t> secret, Span<const char> label,
Span<const uint8_t> seed1, Span<const uint8_t> seed2);
// Encryption layer.
// SSLAEADContext contains information about an AEAD that is being used to
// encrypt an SSL connection.
class SSLAEADContext {
public:
SSLAEADContext(uint16_t version, bool is_dtls, const SSL_CIPHER *cipher);
~SSLAEADContext();
static constexpr bool kAllowUniquePtr = true;
SSLAEADContext(const SSLAEADContext &&) = delete;
SSLAEADContext &operator=(const SSLAEADContext &&) = delete;
// CreateNullCipher creates an |SSLAEADContext| for the null cipher.
static UniquePtr<SSLAEADContext> CreateNullCipher(bool is_dtls);
// Create creates an |SSLAEADContext| using the supplied key material. It
// returns nullptr on error. Only one of |Open| or |Seal| may be used with the
// resulting object, depending on |direction|. |version| is the normalized
// protocol version, so DTLS 1.0 is represented as 0x0301, not 0xffef.
static UniquePtr<SSLAEADContext> Create(enum evp_aead_direction_t direction,
uint16_t version, bool is_dtls,
const SSL_CIPHER *cipher,
Span<const uint8_t> enc_key,
Span<const uint8_t> mac_key,
Span<const uint8_t> fixed_iv);
// CreatePlaceholderForQUIC creates a placeholder |SSLAEADContext| for the
// given cipher and version. The resulting object can be queried for various
// properties but cannot encrypt or decrypt data.
static UniquePtr<SSLAEADContext> CreatePlaceholderForQUIC(
uint16_t version, const SSL_CIPHER *cipher);
// SetVersionIfNullCipher sets the version the SSLAEADContext for the null
// cipher, to make version-specific determinations in the record layer prior
// to a cipher being selected.
void SetVersionIfNullCipher(uint16_t version);
// ProtocolVersion returns the protocol version associated with this
// SSLAEADContext. It can only be called once |version_| has been set to a
// valid value.
uint16_t ProtocolVersion() const;
// RecordVersion returns the record version that should be used with this
// SSLAEADContext for record construction and crypto.
uint16_t RecordVersion() const;
const SSL_CIPHER *cipher() const { return cipher_; }
// is_null_cipher returns true if this is the null cipher.
bool is_null_cipher() const { return !cipher_; }
// ExplicitNonceLen returns the length of the explicit nonce.
size_t ExplicitNonceLen() const;
// MaxOverhead returns the maximum overhead of calling |Seal|.
size_t MaxOverhead() const;
// SuffixLen calculates the suffix length written by |SealScatter| and writes
// it to |*out_suffix_len|. It returns true on success and false on error.
// |in_len| and |extra_in_len| should equal the argument of the same names
// passed to |SealScatter|.
bool SuffixLen(size_t *out_suffix_len, size_t in_len,
size_t extra_in_len) const;
// CiphertextLen calculates the total ciphertext length written by
// |SealScatter| and writes it to |*out_len|. It returns true on success and
// false on error. |in_len| and |extra_in_len| should equal the argument of
// the same names passed to |SealScatter|.
bool CiphertextLen(size_t *out_len, size_t in_len, size_t extra_in_len) const;
// Open authenticates and decrypts |in| in-place. On success, it sets |*out|
// to the plaintext in |in| and returns true. Otherwise, it returns
// false. The output will always be |ExplicitNonceLen| bytes ahead of |in|.
bool Open(Span<uint8_t> *out, uint8_t type, uint16_t record_version,
const uint8_t seqnum[8], Span<const uint8_t> header,
Span<uint8_t> in);
// Seal encrypts and authenticates |in_len| bytes from |in| and writes the
// result to |out|. It returns true on success and false on error.
//
// If |in| and |out| alias then |out| + |ExplicitNonceLen| must be == |in|.
bool Seal(uint8_t *out, size_t *out_len, size_t max_out, uint8_t type,
uint16_t record_version, const uint8_t seqnum[8],
Span<const uint8_t> header, const uint8_t *in, size_t in_len);
// SealScatter encrypts and authenticates |in_len| bytes from |in| and splits
// the result between |out_prefix|, |out| and |out_suffix|. It returns one on
// success and zero on error.
//
// On successful return, exactly |ExplicitNonceLen| bytes are written to
// |out_prefix|, |in_len| bytes to |out|, and |SuffixLen| bytes to
// |out_suffix|.
//
// |extra_in| may point to an additional plaintext buffer. If present,
// |extra_in_len| additional bytes are encrypted and authenticated, and the
// ciphertext is written to the beginning of |out_suffix|. |SuffixLen| should
// be used to size |out_suffix| accordingly.
//
// If |in| and |out| alias then |out| must be == |in|. Other arguments may not
// alias anything.
bool SealScatter(uint8_t *out_prefix, uint8_t *out, uint8_t *out_suffix,
uint8_t type, uint16_t record_version,
const uint8_t seqnum[8], Span<const uint8_t> header,
const uint8_t *in, size_t in_len, const uint8_t *extra_in,
size_t extra_in_len);
bool GetIV(const uint8_t **out_iv, size_t *out_iv_len) const;
private:
// GetAdditionalData returns the additional data, writing into |storage| if
// necessary.
Span<const uint8_t> GetAdditionalData(uint8_t storage[13], uint8_t type,
uint16_t record_version,
const uint8_t seqnum[8],
size_t plaintext_len,
Span<const uint8_t> header);
const SSL_CIPHER *cipher_;
ScopedEVP_AEAD_CTX ctx_;
// fixed_nonce_ contains any bytes of the nonce that are fixed for all
// records.
uint8_t fixed_nonce_[12];
uint8_t fixed_nonce_len_ = 0, variable_nonce_len_ = 0;
// version_ is the wire version that should be used with this AEAD.
uint16_t version_;
// is_dtls_ is whether DTLS is being used with this AEAD.
bool is_dtls_;
// variable_nonce_included_in_record_ is true if the variable nonce
// for a record is included as a prefix before the ciphertext.
bool variable_nonce_included_in_record_ : 1;
// random_variable_nonce_ is true if the variable nonce is
// randomly generated, rather than derived from the sequence
// number.
bool random_variable_nonce_ : 1;
// xor_fixed_nonce_ is true if the fixed nonce should be XOR'd into the
// variable nonce rather than prepended.
bool xor_fixed_nonce_ : 1;
// omit_length_in_ad_ is true if the length should be omitted in the
// AEAD's ad parameter.
bool omit_length_in_ad_ : 1;
// ad_is_header_ is true if the AEAD's ad parameter is the record header.
bool ad_is_header_ : 1;
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
};
// DTLS replay bitmap.
// DTLS1_BITMAP maintains a sliding window of 64 sequence numbers to detect
// replayed packets. It should be initialized by zeroing every field.
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
struct DTLS1_BITMAP {
// map is a bit mask of the last 64 sequence numbers. Bit
// |1<<i| corresponds to |max_seq_num - i|.
uint64_t map = 0;
// max_seq_num is the largest sequence number seen so far as a 64-bit
// integer.
uint64_t max_seq_num = 0;
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
};
// Record layer.
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
// ssl_record_sequence_update increments the sequence number in |seq|. It
// returns true on success and false on wraparound.
bool ssl_record_sequence_update(uint8_t *seq, size_t seq_len);
// ssl_record_prefix_len returns the length of the prefix before the ciphertext
// of a record for |ssl|.
//
// TODO(davidben): Expose this as part of public API once the high-level
// buffer-free APIs are available.
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
size_t ssl_record_prefix_len(const SSL *ssl);
enum ssl_open_record_t {
ssl_open_record_success,
ssl_open_record_discard,
ssl_open_record_partial,
ssl_open_record_close_notify,
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
ssl_open_record_error,
};
// tls_open_record decrypts a record from |in| in-place.
//
// If the input did not contain a complete record, it returns
// |ssl_open_record_partial|. It sets |*out_consumed| to the total number of
// bytes necessary. It is guaranteed that a successful call to |tls_open_record|
// will consume at least that many bytes.
//
// Otherwise, it sets |*out_consumed| to the number of bytes of input
// consumed. Note that input may be consumed on all return codes if a record was
// decrypted.
//
// On success, it returns |ssl_open_record_success|. It sets |*out_type| to the
// record type and |*out| to the record body in |in|. Note that |*out| may be
// empty.
//
// If a record was successfully processed but should be discarded, it returns
// |ssl_open_record_discard|.
//
// If a record was successfully processed but is a close_notify, it returns
// |ssl_open_record_close_notify|.
//
// On failure or fatal alert, it returns |ssl_open_record_error| and sets
// |*out_alert| to an alert to emit, or zero if no alert should be emitted.
enum ssl_open_record_t tls_open_record(SSL *ssl, uint8_t *out_type,
Span<uint8_t> *out, size_t *out_consumed,
uint8_t *out_alert, Span<uint8_t> in);
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
// dtls_open_record implements |tls_open_record| for DTLS. It only returns
// |ssl_open_record_partial| if |in| was empty and sets |*out_consumed| to
// zero. The caller should read one packet and try again.
enum ssl_open_record_t dtls_open_record(SSL *ssl, uint8_t *out_type,
Span<uint8_t> *out,
size_t *out_consumed,
uint8_t *out_alert, Span<uint8_t> in);
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
// ssl_seal_align_prefix_len returns the length of the prefix before the start
// of the bulk of the ciphertext when sealing a record with |ssl|. Callers may
// use this to align buffers.
//
// Note when TLS 1.0 CBC record-splitting is enabled, this includes the one byte
// record and is the offset into second record's ciphertext. Thus sealing a
// small record may result in a smaller output than this value.
//
// TODO(davidben): Is this alignment valuable? Record-splitting makes this a
// mess.
size_t ssl_seal_align_prefix_len(const SSL *ssl);
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
// tls_seal_record seals a new record of type |type| and body |in| and writes it
// to |out|. At most |max_out| bytes will be written. It returns true on success
// and false on error. If enabled, |tls_seal_record| implements TLS 1.0 CBC
// 1/n-1 record splitting and may write two records concatenated.
//
// For a large record, the bulk of the ciphertext will begin
// |ssl_seal_align_prefix_len| bytes into out. Aligning |out| appropriately may
// improve performance. It writes at most |in_len| + |SSL_max_seal_overhead|
// bytes to |out|.
//
// |in| and |out| may not alias.
bool tls_seal_record(SSL *ssl, uint8_t *out, size_t *out_len, size_t max_out,
uint8_t type, const uint8_t *in, size_t in_len);
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
enum dtls1_use_epoch_t {
dtls1_use_previous_epoch,
dtls1_use_current_epoch,
};
// dtls_max_seal_overhead returns the maximum overhead, in bytes, of sealing a
// record.
size_t dtls_max_seal_overhead(const SSL *ssl, enum dtls1_use_epoch_t use_epoch);
// dtls_seal_prefix_len returns the number of bytes of prefix to reserve in
// front of the plaintext when sealing a record in-place.
size_t dtls_seal_prefix_len(const SSL *ssl, enum dtls1_use_epoch_t use_epoch);
// dtls_seal_record implements |tls_seal_record| for DTLS. |use_epoch| selects
// which epoch's cipher state to use. Unlike |tls_seal_record|, |in| and |out|
// may alias but, if they do, |in| must be exactly |dtls_seal_prefix_len| bytes
// ahead of |out|.
bool dtls_seal_record(SSL *ssl, uint8_t *out, size_t *out_len, size_t max_out,
uint8_t type, const uint8_t *in, size_t in_len,
enum dtls1_use_epoch_t use_epoch);
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
// ssl_process_alert processes |in| as an alert and updates |ssl|'s shutdown
// state. It returns one of |ssl_open_record_discard|, |ssl_open_record_error|,
// |ssl_open_record_close_notify|, or |ssl_open_record_fatal_alert| as
// appropriate.
enum ssl_open_record_t ssl_process_alert(SSL *ssl, uint8_t *out_alert,
Span<const uint8_t> in);
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
// Private key operations.
// ssl_has_private_key returns whether |hs| has a private key configured.
bool ssl_has_private_key(const SSL_HANDSHAKE *hs);
// ssl_private_key_* perform the corresponding operation on
// |SSL_PRIVATE_KEY_METHOD|. If there is a custom private key configured, they
// call the corresponding function or |complete| depending on whether there is a
// pending operation. Otherwise, they implement the operation with
// |EVP_PKEY|.
enum ssl_private_key_result_t ssl_private_key_sign(
Simplify ssl_private_key_* state machine points. The original motivation behind the sign/complete split was to avoid needlessly hashing the input on each pass through the state machine, but we're payload-based now and, in all cases, the payload is either cheap to compute or readily available. (Even the hashing worry was probably unnecessary.) Tweak ssl_private_key_{sign,decrypt} to automatically call ssl_private_key_complete as needed and take advantage of this in the handshake state machines: - TLS 1.3 signing now computes the payload each pass. The payload is small and we're already allocating a comparable-sized buffer each iteration to hold the signature. This shouldn't be a big deal. - TLS 1.2 decryption code still needs two states due to reading the message (fixed in new state machine style), but otherwise it just performs cheap idempotent tasks again. The PSK code is reshuffled to guarantee the callback is not called twice (though this was impossible anyway because we don't support RSA_PSK). - TLS 1.2 CertificateVerify signing is easy as the transcript is readily available. The buffer is released very slightly later, but it shouldn't matter. - TLS 1.2 ServerKeyExchange signing required some reshuffling. Assembling the ServerKeyExchange parameters is moved to the previous state. The signing payload has some randoms prepended. This is cheap enough, but a nuisance in C. Pre-prepend the randoms in hs->server_params. With this change, we are *nearly* rid of the A/B => same function pattern. BUG=128 Change-Id: Iec4fe0be7cfc88a6de027ba2760fae70794ea810 Reviewed-on: https://boringssl-review.googlesource.com/17265 Commit-Queue: David Benjamin <davidben@google.com> Commit-Queue: Steven Valdez <svaldez@google.com> Reviewed-by: Steven Valdez <svaldez@google.com>
2017-06-17 18:20:59 +01:00
SSL_HANDSHAKE *hs, uint8_t *out, size_t *out_len, size_t max_out,
uint16_t sigalg, Span<const uint8_t> in);
enum ssl_private_key_result_t ssl_private_key_decrypt(SSL_HANDSHAKE *hs,
uint8_t *out,
size_t *out_len,
size_t max_out,
Span<const uint8_t> in);
// ssl_private_key_supports_signature_algorithm returns whether |hs|'s private
// key supports |sigalg|.
bool ssl_private_key_supports_signature_algorithm(SSL_HANDSHAKE *hs,
uint16_t sigalg);
// ssl_public_key_verify verifies that the |signature| is valid for the public
// key |pkey| and input |in|, using the signature algorithm |sigalg|.
bool ssl_public_key_verify(SSL *ssl, Span<const uint8_t> signature,
uint16_t sigalg, EVP_PKEY *pkey,
Span<const uint8_t> in);
// Key shares.
Implement draft-ietf-tls-curve25519-01 in C. The new curve is not enabled by default. As EC_GROUP/EC_POINT is a bit too complex for X25519, this introduces an SSL_ECDH_METHOD abstraction which wraps just the raw ECDH operation. It also tidies up some of the curve code which kept converting back and force between NIDs and curve IDs. Now everything transits as curve IDs except for API entry points (SSL_set1_curves) which take NIDs. Those convert immediately and act on curve IDs from then on. Note that, like the Go implementation, this slightly tweaks the order of operations. The client sees the server public key before sending its own. To keep the abstraction simple, SSL_ECDH_METHOD expects to generate a keypair before consuming the peer's public key. Instead, the client handshake stashes the serialized peer public value and defers parsing it until it comes time to send ClientKeyExchange. (This is analogous to what it was doing before where it stashed the parsed peer public value instead.) It still uses TLS 1.2 terminology everywhere, but this abstraction should also be compatible with TLS 1.3 which unifies (EC)DH-style key exchanges. (Accordingly, this abstraction intentionally does not handle parsing the ClientKeyExchange/ServerKeyExchange framing or attempt to handle asynchronous plain RSA or the authentication bits.) BUG=571231 Change-Id: Iba09dddee5bcdfeb2b70185308e8ab0632717932 Reviewed-on: https://boringssl-review.googlesource.com/6780 Reviewed-by: Adam Langley <agl@google.com>
2015-12-19 05:18:25 +00:00
// SSLKeyShare abstracts over Diffie-Hellman-like key exchanges.
class SSLKeyShare {
public:
virtual ~SSLKeyShare() {}
static constexpr bool kAllowUniquePtr = true;
HAS_VIRTUAL_DESTRUCTOR
// Create returns a SSLKeyShare instance for use with group |group_id| or
// nullptr on error.
static UniquePtr<SSLKeyShare> Create(uint16_t group_id);
Implement draft-ietf-tls-curve25519-01 in C. The new curve is not enabled by default. As EC_GROUP/EC_POINT is a bit too complex for X25519, this introduces an SSL_ECDH_METHOD abstraction which wraps just the raw ECDH operation. It also tidies up some of the curve code which kept converting back and force between NIDs and curve IDs. Now everything transits as curve IDs except for API entry points (SSL_set1_curves) which take NIDs. Those convert immediately and act on curve IDs from then on. Note that, like the Go implementation, this slightly tweaks the order of operations. The client sees the server public key before sending its own. To keep the abstraction simple, SSL_ECDH_METHOD expects to generate a keypair before consuming the peer's public key. Instead, the client handshake stashes the serialized peer public value and defers parsing it until it comes time to send ClientKeyExchange. (This is analogous to what it was doing before where it stashed the parsed peer public value instead.) It still uses TLS 1.2 terminology everywhere, but this abstraction should also be compatible with TLS 1.3 which unifies (EC)DH-style key exchanges. (Accordingly, this abstraction intentionally does not handle parsing the ClientKeyExchange/ServerKeyExchange framing or attempt to handle asynchronous plain RSA or the authentication bits.) BUG=571231 Change-Id: Iba09dddee5bcdfeb2b70185308e8ab0632717932 Reviewed-on: https://boringssl-review.googlesource.com/6780 Reviewed-by: Adam Langley <agl@google.com>
2015-12-19 05:18:25 +00:00
// Create deserializes an SSLKeyShare instance previously serialized by
// |Serialize|.
static UniquePtr<SSLKeyShare> Create(CBS *in);
// GroupID returns the group ID.
virtual uint16_t GroupID() const PURE_VIRTUAL;
Implement draft-ietf-tls-curve25519-01 in C. The new curve is not enabled by default. As EC_GROUP/EC_POINT is a bit too complex for X25519, this introduces an SSL_ECDH_METHOD abstraction which wraps just the raw ECDH operation. It also tidies up some of the curve code which kept converting back and force between NIDs and curve IDs. Now everything transits as curve IDs except for API entry points (SSL_set1_curves) which take NIDs. Those convert immediately and act on curve IDs from then on. Note that, like the Go implementation, this slightly tweaks the order of operations. The client sees the server public key before sending its own. To keep the abstraction simple, SSL_ECDH_METHOD expects to generate a keypair before consuming the peer's public key. Instead, the client handshake stashes the serialized peer public value and defers parsing it until it comes time to send ClientKeyExchange. (This is analogous to what it was doing before where it stashed the parsed peer public value instead.) It still uses TLS 1.2 terminology everywhere, but this abstraction should also be compatible with TLS 1.3 which unifies (EC)DH-style key exchanges. (Accordingly, this abstraction intentionally does not handle parsing the ClientKeyExchange/ServerKeyExchange framing or attempt to handle asynchronous plain RSA or the authentication bits.) BUG=571231 Change-Id: Iba09dddee5bcdfeb2b70185308e8ab0632717932 Reviewed-on: https://boringssl-review.googlesource.com/6780 Reviewed-by: Adam Langley <agl@google.com>
2015-12-19 05:18:25 +00:00
// Offer generates a keypair and writes the public value to
// |out_public_key|. It returns true on success and false on error.
virtual bool Offer(CBB *out_public_key) PURE_VIRTUAL;
// Accept performs a key exchange against the |peer_key| generated by |Offer|.
// On success, it returns true, writes the public value to |out_public_key|,
// and sets |*out_secret| to the shared secret. On failure, it returns false
// and sets |*out_alert| to an alert to send to the peer.
//
// The default implementation calls |Offer| and then |Finish|, assuming a key
// exchange protocol where the peers are symmetric.
virtual bool Accept(CBB *out_public_key, Array<uint8_t> *out_secret,
uint8_t *out_alert, Span<const uint8_t> peer_key);
// Finish performs a key exchange against the |peer_key| generated by
// |Accept|. On success, it returns true and sets |*out_secret| to the shared
// secret. On failure, it returns false and sets |*out_alert| to an alert to
// send to the peer.
virtual bool Finish(Array<uint8_t> *out_secret, uint8_t *out_alert,
Span<const uint8_t> peer_key) PURE_VIRTUAL;
// Serialize writes the state of the key exchange to |out|, returning true if
// successful and false otherwise.
virtual bool Serialize(CBB *out) { return false; }
// Deserialize initializes the state of the key exchange from |in|, returning
// true if successful and false otherwise. It is called by |Create|.
virtual bool Deserialize(CBS *in) { return false; }
};
Implement draft-ietf-tls-curve25519-01 in C. The new curve is not enabled by default. As EC_GROUP/EC_POINT is a bit too complex for X25519, this introduces an SSL_ECDH_METHOD abstraction which wraps just the raw ECDH operation. It also tidies up some of the curve code which kept converting back and force between NIDs and curve IDs. Now everything transits as curve IDs except for API entry points (SSL_set1_curves) which take NIDs. Those convert immediately and act on curve IDs from then on. Note that, like the Go implementation, this slightly tweaks the order of operations. The client sees the server public key before sending its own. To keep the abstraction simple, SSL_ECDH_METHOD expects to generate a keypair before consuming the peer's public key. Instead, the client handshake stashes the serialized peer public value and defers parsing it until it comes time to send ClientKeyExchange. (This is analogous to what it was doing before where it stashed the parsed peer public value instead.) It still uses TLS 1.2 terminology everywhere, but this abstraction should also be compatible with TLS 1.3 which unifies (EC)DH-style key exchanges. (Accordingly, this abstraction intentionally does not handle parsing the ClientKeyExchange/ServerKeyExchange framing or attempt to handle asynchronous plain RSA or the authentication bits.) BUG=571231 Change-Id: Iba09dddee5bcdfeb2b70185308e8ab0632717932 Reviewed-on: https://boringssl-review.googlesource.com/6780 Reviewed-by: Adam Langley <agl@google.com>
2015-12-19 05:18:25 +00:00
struct NamedGroup {
int nid;
uint16_t group_id;
const char name[8], alias[11];
};
// NamedGroups returns all supported groups.
Span<const NamedGroup> NamedGroups();
// ssl_nid_to_group_id looks up the group corresponding to |nid|. On success, it
// sets |*out_group_id| to the group ID and returns true. Otherwise, it returns
// false.
bool ssl_nid_to_group_id(uint16_t *out_group_id, int nid);
Implement draft-ietf-tls-curve25519-01 in C. The new curve is not enabled by default. As EC_GROUP/EC_POINT is a bit too complex for X25519, this introduces an SSL_ECDH_METHOD abstraction which wraps just the raw ECDH operation. It also tidies up some of the curve code which kept converting back and force between NIDs and curve IDs. Now everything transits as curve IDs except for API entry points (SSL_set1_curves) which take NIDs. Those convert immediately and act on curve IDs from then on. Note that, like the Go implementation, this slightly tweaks the order of operations. The client sees the server public key before sending its own. To keep the abstraction simple, SSL_ECDH_METHOD expects to generate a keypair before consuming the peer's public key. Instead, the client handshake stashes the serialized peer public value and defers parsing it until it comes time to send ClientKeyExchange. (This is analogous to what it was doing before where it stashed the parsed peer public value instead.) It still uses TLS 1.2 terminology everywhere, but this abstraction should also be compatible with TLS 1.3 which unifies (EC)DH-style key exchanges. (Accordingly, this abstraction intentionally does not handle parsing the ClientKeyExchange/ServerKeyExchange framing or attempt to handle asynchronous plain RSA or the authentication bits.) BUG=571231 Change-Id: Iba09dddee5bcdfeb2b70185308e8ab0632717932 Reviewed-on: https://boringssl-review.googlesource.com/6780 Reviewed-by: Adam Langley <agl@google.com>
2015-12-19 05:18:25 +00:00
// ssl_name_to_group_id looks up the group corresponding to the |name| string of
// length |len|. On success, it sets |*out_group_id| to the group ID and returns
// true. Otherwise, it returns false.
bool ssl_name_to_group_id(uint16_t *out_group_id, const char *name, size_t len);
Implement draft-ietf-tls-curve25519-01 in C. The new curve is not enabled by default. As EC_GROUP/EC_POINT is a bit too complex for X25519, this introduces an SSL_ECDH_METHOD abstraction which wraps just the raw ECDH operation. It also tidies up some of the curve code which kept converting back and force between NIDs and curve IDs. Now everything transits as curve IDs except for API entry points (SSL_set1_curves) which take NIDs. Those convert immediately and act on curve IDs from then on. Note that, like the Go implementation, this slightly tweaks the order of operations. The client sees the server public key before sending its own. To keep the abstraction simple, SSL_ECDH_METHOD expects to generate a keypair before consuming the peer's public key. Instead, the client handshake stashes the serialized peer public value and defers parsing it until it comes time to send ClientKeyExchange. (This is analogous to what it was doing before where it stashed the parsed peer public value instead.) It still uses TLS 1.2 terminology everywhere, but this abstraction should also be compatible with TLS 1.3 which unifies (EC)DH-style key exchanges. (Accordingly, this abstraction intentionally does not handle parsing the ClientKeyExchange/ServerKeyExchange framing or attempt to handle asynchronous plain RSA or the authentication bits.) BUG=571231 Change-Id: Iba09dddee5bcdfeb2b70185308e8ab0632717932 Reviewed-on: https://boringssl-review.googlesource.com/6780 Reviewed-by: Adam Langley <agl@google.com>
2015-12-19 05:18:25 +00:00
// Handshake messages.
Simplify handshake message size limits. A handshake message can go up to 2^24 bytes = 16MB which is a little large for the peer to force us to buffer. Accordingly, we bound the size of a handshake message. Rather than have a global limit, the existing logic uses a different limit at each state in the handshake state machine and, for certificates, allows configuring the maximum certificate size. This is nice in that we engage larger limits iff the relevant state is reachable from the handshake. Servers without client auth get a tighter limit "for free". However, this doesn't work for DTLS due to out-of-order messages and we use a simpler scheme for DTLS. This scheme also is tricky on optional messages and makes the handshake <-> message layer communication complex. Apart from an ignored 20,000 byte limit on ServerHello, the largest non-certificate limit is the common 16k limit on ClientHello. So this complexity wasn't buying us anything. Unify everything on the DTLS scheme except, so as not to regress bounds on client-auth-less servers, also correctly check for whether client auth is configured. The value of 16k was chosen based on this value. (The 20,000 byte ServerHello limit makes no sense. We can easily bound the ServerHello because servers may not send extensions we don't implement. But it gets overshadowed by the certificate anyway.) Change-Id: I00309b16d809a3c2a1543f99fd29c4163e3add81 Reviewed-on: https://boringssl-review.googlesource.com/7941 Reviewed-by: David Benjamin <davidben@google.com>
2016-05-12 05:43:05 +01:00
struct SSLMessage {
bool is_v2_hello;
uint8_t type;
CBS body;
// raw is the entire serialized handshake message, including the TLS or DTLS
// message header.
CBS raw;
};
// SSL_MAX_HANDSHAKE_FLIGHT is the number of messages, including
// ChangeCipherSpec, in the longest handshake flight. Currently this is the
// client's second leg in a full handshake when client certificates, NPN, and
// Channel ID, are all enabled.
#define SSL_MAX_HANDSHAKE_FLIGHT 7
extern const uint8_t kHelloRetryRequest[SSL3_RANDOM_SIZE];
extern const uint8_t kTLS12DowngradeRandom[8];
extern const uint8_t kTLS13DowngradeRandom[8];
extern const uint8_t kJDK11DowngradeRandom[8];
// ssl_max_handshake_message_len returns the maximum number of bytes permitted
// in a handshake message for |ssl|.
Simplify handshake message size limits. A handshake message can go up to 2^24 bytes = 16MB which is a little large for the peer to force us to buffer. Accordingly, we bound the size of a handshake message. Rather than have a global limit, the existing logic uses a different limit at each state in the handshake state machine and, for certificates, allows configuring the maximum certificate size. This is nice in that we engage larger limits iff the relevant state is reachable from the handshake. Servers without client auth get a tighter limit "for free". However, this doesn't work for DTLS due to out-of-order messages and we use a simpler scheme for DTLS. This scheme also is tricky on optional messages and makes the handshake <-> message layer communication complex. Apart from an ignored 20,000 byte limit on ServerHello, the largest non-certificate limit is the common 16k limit on ClientHello. So this complexity wasn't buying us anything. Unify everything on the DTLS scheme except, so as not to regress bounds on client-auth-less servers, also correctly check for whether client auth is configured. The value of 16k was chosen based on this value. (The 20,000 byte ServerHello limit makes no sense. We can easily bound the ServerHello because servers may not send extensions we don't implement. But it gets overshadowed by the certificate anyway.) Change-Id: I00309b16d809a3c2a1543f99fd29c4163e3add81 Reviewed-on: https://boringssl-review.googlesource.com/7941 Reviewed-by: David Benjamin <davidben@google.com>
2016-05-12 05:43:05 +01:00
size_t ssl_max_handshake_message_len(const SSL *ssl);
// tls_can_accept_handshake_data returns whether |ssl| is able to accept more
// data into handshake buffer.
bool tls_can_accept_handshake_data(const SSL *ssl, uint8_t *out_alert);
// tls_has_unprocessed_handshake_data returns whether there is buffered
// handshake data that has not been consumed by |get_message|.
bool tls_has_unprocessed_handshake_data(const SSL *ssl);
// tls_append_handshake_data appends |data| to the handshake buffer. It returns
// true on success and false on allocation failure.
bool tls_append_handshake_data(SSL *ssl, Span<const uint8_t> data);
// dtls_has_unprocessed_handshake_data behaves like
// |tls_has_unprocessed_handshake_data| for DTLS.
bool dtls_has_unprocessed_handshake_data(const SSL *ssl);
// tls_flush_pending_hs_data flushes any handshake plaintext data.
bool tls_flush_pending_hs_data(SSL *ssl);
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
struct DTLS_OUTGOING_MESSAGE {
DTLS_OUTGOING_MESSAGE() {}
DTLS_OUTGOING_MESSAGE(const DTLS_OUTGOING_MESSAGE &) = delete;
DTLS_OUTGOING_MESSAGE &operator=(const DTLS_OUTGOING_MESSAGE &) = delete;
~DTLS_OUTGOING_MESSAGE() { Clear(); }
void Clear();
uint8_t *data = nullptr;
uint32_t len = 0;
uint16_t epoch = 0;
bool is_ccs = false;
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
};
// dtls_clear_outgoing_messages releases all buffered outgoing messages.
void dtls_clear_outgoing_messages(SSL *ssl);
Simplify handshake message size limits. A handshake message can go up to 2^24 bytes = 16MB which is a little large for the peer to force us to buffer. Accordingly, we bound the size of a handshake message. Rather than have a global limit, the existing logic uses a different limit at each state in the handshake state machine and, for certificates, allows configuring the maximum certificate size. This is nice in that we engage larger limits iff the relevant state is reachable from the handshake. Servers without client auth get a tighter limit "for free". However, this doesn't work for DTLS due to out-of-order messages and we use a simpler scheme for DTLS. This scheme also is tricky on optional messages and makes the handshake <-> message layer communication complex. Apart from an ignored 20,000 byte limit on ServerHello, the largest non-certificate limit is the common 16k limit on ClientHello. So this complexity wasn't buying us anything. Unify everything on the DTLS scheme except, so as not to regress bounds on client-auth-less servers, also correctly check for whether client auth is configured. The value of 16k was chosen based on this value. (The 20,000 byte ServerHello limit makes no sense. We can easily bound the ServerHello because servers may not send extensions we don't implement. But it gets overshadowed by the certificate anyway.) Change-Id: I00309b16d809a3c2a1543f99fd29c4163e3add81 Reviewed-on: https://boringssl-review.googlesource.com/7941 Reviewed-by: David Benjamin <davidben@google.com>
2016-05-12 05:43:05 +01:00
// Callbacks.
// ssl_do_info_callback calls |ssl|'s info callback, if set.
void ssl_do_info_callback(const SSL *ssl, int type, int value);
// ssl_do_msg_callback calls |ssl|'s message callback, if set.
void ssl_do_msg_callback(SSL *ssl, int is_write, int content_type,
Span<const uint8_t> in);
// Transport buffers.
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
class SSLBuffer {
public:
SSLBuffer() {}
~SSLBuffer() { Clear(); }
SSLBuffer(const SSLBuffer &) = delete;
SSLBuffer &operator=(const SSLBuffer &) = delete;
uint8_t *data() { return buf_ + offset_; }
size_t size() const { return size_; }
bool empty() const { return size_ == 0; }
size_t cap() const { return cap_; }
Span<uint8_t> span() { return MakeSpan(data(), size()); }
Span<uint8_t> remaining() {
return MakeSpan(data() + size(), cap() - size());
}
// Clear releases the buffer.
void Clear();
// EnsureCap ensures the buffer has capacity at least |new_cap|, aligned such
// that data written after |header_len| is aligned to a
// |SSL3_ALIGN_PAYLOAD|-byte boundary. It returns true on success and false
// on error.
bool EnsureCap(size_t header_len, size_t new_cap);
// DidWrite extends the buffer by |len|. The caller must have filled in to
// this point.
void DidWrite(size_t len);
// Consume consumes |len| bytes from the front of the buffer. The memory
// consumed will remain valid until the next call to |DiscardConsumed| or
// |Clear|.
void Consume(size_t len);
// DiscardConsumed discards the consumed bytes from the buffer. If the buffer
// is now empty, it releases memory used by it.
void DiscardConsumed();
private:
// buf_ is the memory allocated for this buffer.
uint8_t *buf_ = nullptr;
// offset_ is the offset into |buf_| which the buffer contents start at.
uint16_t offset_ = 0;
// size_ is the size of the buffer contents from |buf_| + |offset_|.
uint16_t size_ = 0;
// cap_ is how much memory beyond |buf_| + |offset_| is available.
uint16_t cap_ = 0;
};
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
// ssl_read_buffer_extend_to extends the read buffer to the desired length. For
// TLS, it reads to the end of the buffer until the buffer is |len| bytes
// long. For DTLS, it reads a new packet and ignores |len|. It returns one on
// success, zero on EOF, and a negative number on error.
//
// It is an error to call |ssl_read_buffer_extend_to| in DTLS when the buffer is
// non-empty.
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
int ssl_read_buffer_extend_to(SSL *ssl, size_t len);
// ssl_handle_open_record handles the result of passing |ssl->s3->read_buffer|
// to a record-processing function. If |ret| is a success or if the caller
// should retry, it returns one and sets |*out_retry|. Otherwise, it returns <=
// 0.
int ssl_handle_open_record(SSL *ssl, bool *out_retry, ssl_open_record_t ret,
size_t consumed, uint8_t alert);
// ssl_write_buffer_flush flushes the write buffer to the transport. It returns
// one on success and <= 0 on error. For DTLS, whether or not the write
// succeeds, the write buffer will be cleared.
Factor out the buffering and low-level record code. This begins decoupling the transport from the SSL state machine. The buffering logic is hidden behind an opaque API. Fields like ssl->packet and ssl->packet_length are gone. ssl3_get_record and dtls1_get_record now call low-level tls_open_record and dtls_open_record functions that unpack a single record independent of who owns the buffer. Both may be called in-place. This removes ssl->rstate which was redundant with the buffer length. Future work will push the buffer up the stack until it is above the handshake. Then we can expose SSL_open and SSL_seal APIs which act like *_open_record but return a slightly larger enum due to other events being possible. Likewise the handshake state machine will be detached from its buffer. The existing SSL_read, SSL_write, etc., APIs will be implemented on top of SSL_open, etc., combined with ssl_read_buffer_* and ssl_write_buffer_*. (Which is why ssl_read_buffer_extend still tries to abstract between TLS's and DTLS's fairly different needs.) The new buffering logic does not support read-ahead (removed previously) since it lacks a memmove on ssl_read_buffer_discard for TLS, but this could be added if desired. The old buffering logic wasn't quite right anyway; it tried to avoid the memmove in some cases and could get stuck too far into the buffer and not accept records. (The only time the memmove is optional is in DTLS or if enough of the record header is available to know that the entire next record would fit in the buffer.) The new logic also now actually decrypts the ciphertext in-place again, rather than almost in-place when there's an explicit nonce/IV. (That accidentally switched in https://boringssl-review.googlesource.com/#/c/4792/; see 3d59e04bce96474099ba76786a2337e99ae14505.) BUG=468889 Change-Id: I403c1626253c46897f47c7ae93aeab1064b767b2 Reviewed-on: https://boringssl-review.googlesource.com/5715 Reviewed-by: Adam Langley <agl@google.com>
2015-07-29 02:34:45 +01:00
int ssl_write_buffer_flush(SSL *ssl);
// Certificate functions.
// ssl_has_certificate returns whether a certificate and private key are
// configured.
bool ssl_has_certificate(const SSL_HANDSHAKE *hs);
// ssl_parse_cert_chain parses a certificate list from |cbs| in the format used
// by a TLS Certificate message. On success, it advances |cbs| and returns
// true. Otherwise, it returns false and sets |*out_alert| to an alert to send
// to the peer.
//
// If the list is non-empty then |*out_chain| and |*out_pubkey| will be set to
// the certificate chain and the leaf certificate's public key
// respectively. Otherwise, both will be set to nullptr.
//
// If the list is non-empty and |out_leaf_sha256| is non-NULL, it writes the
// SHA-256 hash of the leaf to |out_leaf_sha256|.
bool ssl_parse_cert_chain(uint8_t *out_alert,
UniquePtr<STACK_OF(CRYPTO_BUFFER)> *out_chain,
UniquePtr<EVP_PKEY> *out_pubkey,
uint8_t *out_leaf_sha256, CBS *cbs,
CRYPTO_BUFFER_POOL *pool);
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// ssl_add_cert_chain adds |hs->ssl|'s certificate chain to |cbb| in the format
// used by a TLS Certificate message. If there is no certificate chain, it emits
// an empty certificate list. It returns true on success and false on error.
bool ssl_add_cert_chain(SSL_HANDSHAKE *hs, CBB *cbb);
enum ssl_key_usage_t {
key_usage_digital_signature = 0,
key_usage_encipherment = 2,
};
// ssl_cert_check_key_usage parses the DER-encoded, X.509 certificate in |in|
// and returns true if doesn't specify a key usage or, if it does, if it
// includes |bit|. Otherwise it pushes to the error queue and returns false.
bool ssl_cert_check_key_usage(const CBS *in, enum ssl_key_usage_t bit);
// ssl_cert_parse_pubkey extracts the public key from the DER-encoded, X.509
// certificate in |in|. It returns an allocated |EVP_PKEY| or else returns
// nullptr and pushes to the error queue.
UniquePtr<EVP_PKEY> ssl_cert_parse_pubkey(const CBS *in);
// ssl_parse_client_CA_list parses a CA list from |cbs| in the format used by a
// TLS CertificateRequest message. On success, it returns a newly-allocated
// |CRYPTO_BUFFER| list and advances |cbs|. Otherwise, it returns nullptr and
// sets |*out_alert| to an alert to send to the peer.
UniquePtr<STACK_OF(CRYPTO_BUFFER)> ssl_parse_client_CA_list(SSL *ssl,
uint8_t *out_alert,
CBS *cbs);
// ssl_has_client_CAs returns there are configured CAs.
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
bool ssl_has_client_CAs(const SSL_CONFIG *cfg);
// ssl_add_client_CA_list adds the configured CA list to |cbb| in the format
// used by a TLS CertificateRequest message. It returns true on success and
// false on error.
bool ssl_add_client_CA_list(SSL_HANDSHAKE *hs, CBB *cbb);
// ssl_check_leaf_certificate returns one if |pkey| and |leaf| are suitable as
// a server's leaf certificate for |hs|. Otherwise, it returns zero and pushes
// an error on the error queue.
bool ssl_check_leaf_certificate(SSL_HANDSHAKE *hs, EVP_PKEY *pkey,
const CRYPTO_BUFFER *leaf);
// ssl_on_certificate_selected is called once the certificate has been selected.
// It finalizes the certificate and initializes |hs->local_pubkey|. It returns
// true on success and false on error.
bool ssl_on_certificate_selected(SSL_HANDSHAKE *hs);
// TLS 1.3 key derivation.
// tls13_init_key_schedule initializes the handshake hash and key derivation
// state, and incorporates the PSK. The cipher suite and PRF hash must have been
// selected at this point. It returns true on success and false on error.
bool tls13_init_key_schedule(SSL_HANDSHAKE *hs, const uint8_t *psk,
size_t psk_len);
// tls13_init_early_key_schedule initializes the handshake hash and key
// derivation state from the resumption secret and incorporates the PSK to
// derive the early secrets. It returns one on success and zero on error.
bool tls13_init_early_key_schedule(SSL_HANDSHAKE *hs, const uint8_t *psk,
size_t psk_len);
// tls13_advance_key_schedule incorporates |in| into the key schedule with
// HKDF-Extract. It returns true on success and false on error.
bool tls13_advance_key_schedule(SSL_HANDSHAKE *hs, const uint8_t *in,
size_t len);
// tls13_set_traffic_key sets the read or write traffic keys to
// |traffic_secret|. It returns true on success and false on error.
bool tls13_set_traffic_key(SSL *ssl, enum ssl_encryption_level_t level,
enum evp_aead_direction_t direction,
const uint8_t *traffic_secret,
size_t traffic_secret_len);
// tls13_derive_early_secrets derives the early traffic secret. It returns true
// on success and false on error.
bool tls13_derive_early_secrets(SSL_HANDSHAKE *hs);
// tls13_derive_handshake_secrets derives the handshake traffic secret. It
// returns true on success and false on error.
bool tls13_derive_handshake_secrets(SSL_HANDSHAKE *hs);
// tls13_rotate_traffic_key derives the next read or write traffic secret. It
// returns true on success and false on error.
bool tls13_rotate_traffic_key(SSL *ssl, enum evp_aead_direction_t direction);
// tls13_derive_application_secrets derives the initial application data traffic
// and exporter secrets based on the handshake transcripts and |master_secret|.
// It returns true on success and false on error.
bool tls13_derive_application_secrets(SSL_HANDSHAKE *hs);
// tls13_derive_resumption_secret derives the |resumption_secret|.
bool tls13_derive_resumption_secret(SSL_HANDSHAKE *hs);
// tls13_export_keying_material provides an exporter interface to use the
// |exporter_secret|.
bool tls13_export_keying_material(SSL *ssl, Span<uint8_t> out,
Span<const uint8_t> secret,
Span<const char> label,
Span<const uint8_t> context);
// tls13_finished_mac calculates the MAC of the handshake transcript to verify
// the integrity of the Finished message, and stores the result in |out| and
// length in |out_len|. |is_server| is true if this is for the Server Finished
// and false for the Client Finished.
bool tls13_finished_mac(SSL_HANDSHAKE *hs, uint8_t *out, size_t *out_len,
bool is_server);
// tls13_derive_session_psk calculates the PSK for this session based on the
// resumption master secret and |nonce|. It returns true on success, and false
// on failure.
bool tls13_derive_session_psk(SSL_SESSION *session, Span<const uint8_t> nonce);
// tls13_write_psk_binder calculates the PSK binder value and replaces the last
// bytes of |msg| with the resulting value. It returns true on success, and
// false on failure.
bool tls13_write_psk_binder(SSL_HANDSHAKE *hs, uint8_t *msg, size_t len);
// tls13_verify_psk_binder verifies that the handshake transcript, truncated up
// to the binders has a valid signature using the value of |session|'s
// resumption secret. It returns true on success, and false on failure.
bool tls13_verify_psk_binder(SSL_HANDSHAKE *hs, SSL_SESSION *session,
const SSLMessage &msg, CBS *binders);
// Handshake functions.
enum ssl_hs_wait_t {
ssl_hs_error,
ssl_hs_ok,
ssl_hs_read_server_hello,
ssl_hs_read_message,
ssl_hs_flush,
ssl_hs_certificate_selection_pending,
ssl_hs_handoff,
ssl_hs_handback,
ssl_hs_x509_lookup,
ssl_hs_channel_id_lookup,
ssl_hs_private_key_operation,
ssl_hs_pending_session,
ssl_hs_pending_ticket,
ssl_hs_early_return,
ssl_hs_early_data_rejected,
ssl_hs_read_end_of_early_data,
ssl_hs_read_change_cipher_spec,
ssl_hs_certificate_verify,
};
enum ssl_grease_index_t {
ssl_grease_cipher = 0,
ssl_grease_group,
ssl_grease_extension1,
ssl_grease_extension2,
ssl_grease_version,
ssl_grease_ticket_extension,
ssl_grease_last_index = ssl_grease_ticket_extension,
};
enum tls12_server_hs_state_t {
state12_start_accept = 0,
state12_read_client_hello,
state12_select_certificate,
state12_tls13,
state12_select_parameters,
state12_send_server_hello,
state12_send_server_certificate,
state12_send_server_key_exchange,
state12_send_server_hello_done,
state12_read_client_certificate,
state12_verify_client_certificate,
state12_read_client_key_exchange,
state12_read_client_certificate_verify,
state12_read_change_cipher_spec,
state12_process_change_cipher_spec,
state12_read_next_proto,
state12_read_channel_id,
state12_read_client_finished,
state12_send_server_finished,
state12_finish_server_handshake,
state12_done,
};
// handback_t lists the points in the state machine where a handback can occur.
// These are the different points at which key material is no longer needed.
enum handback_t {
handback_after_session_resumption,
handback_after_ecdhe,
handback_after_handshake,
};
// Delegated credentials.
// This structure stores a delegated credential (DC) as defined by
// draft-ietf-tls-subcerts-03.
struct DC {
static constexpr bool kAllowUniquePtr = true;
~DC();
// Dup returns a copy of this DC and takes references to |raw| and |pkey|.
UniquePtr<DC> Dup();
// Parse parses the delegated credential stored in |in|. If successful it
// returns the parsed structure, otherwise it returns |nullptr| and sets
// |*out_alert|.
static UniquePtr<DC> Parse(CRYPTO_BUFFER *in, uint8_t *out_alert);
// raw is the delegated credential encoded as specified in draft-ietf-tls-
// subcerts-03.
UniquePtr<CRYPTO_BUFFER> raw;
// expected_cert_verify_algorithm is the signature scheme of the DC public
// key.
uint16_t expected_cert_verify_algorithm = 0;
// pkey is the public key parsed from |public_key|.
UniquePtr<EVP_PKEY> pkey;
private:
friend DC* New<DC>();
DC();
};
// ssl_signing_with_dc returns true if the peer has indicated support for
// delegated credentials and this host has sent a delegated credential in
// response. If this is true then we've committed to using the DC in the
// handshake.
bool ssl_signing_with_dc(const SSL_HANDSHAKE *hs);
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
struct SSL_HANDSHAKE {
explicit SSL_HANDSHAKE(SSL *ssl);
~SSL_HANDSHAKE();
static constexpr bool kAllowUniquePtr = true;
// ssl is a non-owning pointer to the parent |SSL| object.
SSL *ssl;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// config is a non-owning pointer to the handshake configuration.
SSL_CONFIG *config;
// wait contains the operation the handshake is currently blocking on or
// |ssl_hs_ok| if none.
enum ssl_hs_wait_t wait = ssl_hs_ok;
// state is the internal state for the TLS 1.2 and below handshake. Its
// values depend on |do_handshake| but the starting state is always zero.
int state = 0;
// tls13_state is the internal state for the TLS 1.3 handshake. Its values
// depend on |do_handshake| but the starting state is always zero.
int tls13_state = 0;
// min_version is the minimum accepted protocol version, taking account both
// |SSL_OP_NO_*| and |SSL_CTX_set_min_proto_version| APIs.
uint16_t min_version = 0;
// max_version is the maximum accepted protocol version, taking account both
// |SSL_OP_NO_*| and |SSL_CTX_set_max_proto_version| APIs.
uint16_t max_version = 0;
size_t hash_len = 0;
uint8_t secret[EVP_MAX_MD_SIZE] = {0};
uint8_t early_traffic_secret[EVP_MAX_MD_SIZE] = {0};
uint8_t client_handshake_secret[EVP_MAX_MD_SIZE] = {0};
uint8_t server_handshake_secret[EVP_MAX_MD_SIZE] = {0};
uint8_t client_traffic_secret_0[EVP_MAX_MD_SIZE] = {0};
uint8_t server_traffic_secret_0[EVP_MAX_MD_SIZE] = {0};
uint8_t expected_client_finished[EVP_MAX_MD_SIZE] = {0};
union {
// sent is a bitset where the bits correspond to elements of kExtensions
// in t1_lib.c. Each bit is set if that extension was sent in a
// ClientHello. It's not used by servers.
uint32_t sent = 0;
// received is a bitset, like |sent|, but is used by servers to record
// which extensions were received from a client.
uint32_t received;
} extensions;
// retry_group is the group ID selected by the server in HelloRetryRequest in
// TLS 1.3.
uint16_t retry_group = 0;
// error, if |wait| is |ssl_hs_error|, is the error the handshake failed on.
UniquePtr<ERR_SAVE_STATE> error;
// key_shares are the current key exchange instances. The second is only used
// as a client if we believe that we should offer two key shares in a
// ClientHello.
UniquePtr<SSLKeyShare> key_shares[2];
// transcript is the current handshake transcript.
SSLTranscript transcript;
// cookie is the value of the cookie received from the server, if any.
Array<uint8_t> cookie;
// key_share_bytes is the value of the previously sent KeyShare extension by
// the client in TLS 1.3.
Array<uint8_t> key_share_bytes;
// ecdh_public_key, for servers, is the key share to be sent to the client in
// TLS 1.3.
Array<uint8_t> ecdh_public_key;
// peer_sigalgs are the signature algorithms that the peer supports. These are
// taken from the contents of the signature algorithms extension for a server
// or from the CertificateRequest for a client.
Array<uint16_t> peer_sigalgs;
// peer_supported_group_list contains the supported group IDs advertised by
// the peer. This is only set on the server's end. The server does not
// advertise this extension to the client.
Array<uint16_t> peer_supported_group_list;
// peer_key is the peer's ECDH key for a TLS 1.2 client.
Array<uint8_t> peer_key;
// negotiated_token_binding_version is used by a server to store the
// on-the-wire encoding of the Token Binding protocol version to advertise in
// the ServerHello/EncryptedExtensions if the Token Binding extension is to be
// sent.
uint16_t negotiated_token_binding_version;
// cert_compression_alg_id, for a server, contains the negotiated certificate
// compression algorithm for this client. It is only valid if
// |cert_compression_negotiated| is true.
uint16_t cert_compression_alg_id;
// server_params, in a TLS 1.2 server, stores the ServerKeyExchange
// parameters. It has client and server randoms prepended for signing
// convenience.
Array<uint8_t> server_params;
// peer_psk_identity_hint, on the client, is the psk_identity_hint sent by the
// server when using a TLS 1.2 PSK key exchange.
UniquePtr<char> peer_psk_identity_hint;
// ca_names, on the client, contains the list of CAs received in a
// CertificateRequest message.
UniquePtr<STACK_OF(CRYPTO_BUFFER)> ca_names;
// cached_x509_ca_names contains a cache of parsed versions of the elements of
// |ca_names|. This pointer is left non-owning so only
// |ssl_crypto_x509_method| needs to link against crypto/x509.
STACK_OF(X509_NAME) *cached_x509_ca_names = nullptr;
// certificate_types, on the client, contains the set of certificate types
// received in a CertificateRequest message.
Array<uint8_t> certificate_types;
// local_pubkey is the public key we are authenticating as.
UniquePtr<EVP_PKEY> local_pubkey;
// peer_pubkey is the public key parsed from the peer's leaf certificate.
UniquePtr<EVP_PKEY> peer_pubkey;
// new_session is the new mutable session being established by the current
// handshake. It should not be cached.
UniquePtr<SSL_SESSION> new_session;
// early_session is the session corresponding to the current 0-RTT state on
// the client if |in_early_data| is true.
UniquePtr<SSL_SESSION> early_session;
// new_cipher is the cipher being negotiated in this handshake.
const SSL_CIPHER *new_cipher = nullptr;
// key_block is the record-layer key block for TLS 1.2 and earlier.
Array<uint8_t> key_block;
// scts_requested is true if the SCT extension is in the ClientHello.
bool scts_requested : 1;
// needs_psk_binder is true if the ClientHello has a placeholder PSK binder to
// be filled in.
bool needs_psk_binder : 1;
bool received_hello_retry_request : 1;
bool sent_hello_retry_request : 1;
// handshake_finalized is true once the handshake has completed, at which
// point accessors should use the established state.
bool handshake_finalized : 1;
// accept_psk_mode stores whether the client's PSK mode is compatible with our
// preferences.
bool accept_psk_mode : 1;
// cert_request is true if a client certificate was requested.
bool cert_request : 1;
// certificate_status_expected is true if OCSP stapling was negotiated and the
// server is expected to send a CertificateStatus message. (This is used on
// both the client and server sides.)
bool certificate_status_expected : 1;
// ocsp_stapling_requested is true if a client requested OCSP stapling.
bool ocsp_stapling_requested : 1;
// delegated_credential_requested is true if the peer indicated support for
// the delegated credential extension.
bool delegated_credential_requested : 1;
// should_ack_sni is used by a server and indicates that the SNI extension
// should be echoed in the ServerHello.
bool should_ack_sni : 1;
// in_false_start is true if there is a pending client handshake in False
// Start. The client may write data at this point.
bool in_false_start : 1;
// in_early_data is true if there is a pending handshake that has progressed
// enough to send and receive early data.
bool in_early_data : 1;
// early_data_offered is true if the client sent the early_data extension.
bool early_data_offered : 1;
// can_early_read is true if application data may be read at this point in the
// handshake.
bool can_early_read : 1;
// can_early_write is true if application data may be written at this point in
// the handshake.
bool can_early_write : 1;
// next_proto_neg_seen is one of NPN was negotiated.
bool next_proto_neg_seen : 1;
// ticket_expected is true if a TLS 1.2 NewSessionTicket message is to be sent
// or received.
bool ticket_expected : 1;
// extended_master_secret is true if the extended master secret extension is
// negotiated in this handshake.
bool extended_master_secret : 1;
// pending_private_key_op is true if there is a pending private key operation
// in progress.
bool pending_private_key_op : 1;
Simplify ssl_private_key_* state machine points. The original motivation behind the sign/complete split was to avoid needlessly hashing the input on each pass through the state machine, but we're payload-based now and, in all cases, the payload is either cheap to compute or readily available. (Even the hashing worry was probably unnecessary.) Tweak ssl_private_key_{sign,decrypt} to automatically call ssl_private_key_complete as needed and take advantage of this in the handshake state machines: - TLS 1.3 signing now computes the payload each pass. The payload is small and we're already allocating a comparable-sized buffer each iteration to hold the signature. This shouldn't be a big deal. - TLS 1.2 decryption code still needs two states due to reading the message (fixed in new state machine style), but otherwise it just performs cheap idempotent tasks again. The PSK code is reshuffled to guarantee the callback is not called twice (though this was impossible anyway because we don't support RSA_PSK). - TLS 1.2 CertificateVerify signing is easy as the transcript is readily available. The buffer is released very slightly later, but it shouldn't matter. - TLS 1.2 ServerKeyExchange signing required some reshuffling. Assembling the ServerKeyExchange parameters is moved to the previous state. The signing payload has some randoms prepended. This is cheap enough, but a nuisance in C. Pre-prepend the randoms in hs->server_params. With this change, we are *nearly* rid of the A/B => same function pattern. BUG=128 Change-Id: Iec4fe0be7cfc88a6de027ba2760fae70794ea810 Reviewed-on: https://boringssl-review.googlesource.com/17265 Commit-Queue: David Benjamin <davidben@google.com> Commit-Queue: Steven Valdez <svaldez@google.com> Reviewed-by: Steven Valdez <svaldez@google.com>
2017-06-17 18:20:59 +01:00
// grease_seeded is true if |grease_seed| has been initialized.
bool grease_seeded : 1;
// handback indicates that a server should pause the handshake after
// finishing operations that require private key material, in such a way that
// |SSL_get_error| returns |SSL_HANDBACK|. It is set by |SSL_apply_handoff|.
bool handback : 1;
// cert_compression_negotiated is true iff |cert_compression_alg_id| is valid.
bool cert_compression_negotiated : 1;
// apply_jdk11_workaround is true if the peer is probably a JDK 11 client
// which implemented TLS 1.3 incorrectly.
bool apply_jdk11_workaround : 1;
// client_version is the value sent or received in the ClientHello version.
uint16_t client_version = 0;
// early_data_read is the amount of early data that has been read by the
// record layer.
uint16_t early_data_read = 0;
// early_data_written is the amount of early data that has been written by the
// record layer.
uint16_t early_data_written = 0;
// session_id is the session ID in the ClientHello.
uint8_t session_id[SSL_MAX_SSL_SESSION_ID_LENGTH] = {0};
uint8_t session_id_len = 0;
// grease_seed is the entropy for GREASE values. It is valid if
// |grease_seeded| is true.
uint8_t grease_seed[ssl_grease_last_index + 1] = {0};
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
};
UniquePtr<SSL_HANDSHAKE> ssl_handshake_new(SSL *ssl);
// ssl_check_message_type checks if |msg| has type |type|. If so it returns
// one. Otherwise, it sends an alert and returns zero.
bool ssl_check_message_type(SSL *ssl, const SSLMessage &msg, int type);
// ssl_run_handshake runs the TLS handshake. It returns one on success and <= 0
// on error. It sets |out_early_return| to one if we've completed the handshake
// early.
int ssl_run_handshake(SSL_HANDSHAKE *hs, bool *out_early_return);
// The following are implementations of |do_handshake| for the client and
// server.
enum ssl_hs_wait_t ssl_client_handshake(SSL_HANDSHAKE *hs);
enum ssl_hs_wait_t ssl_server_handshake(SSL_HANDSHAKE *hs);
enum ssl_hs_wait_t tls13_client_handshake(SSL_HANDSHAKE *hs);
enum ssl_hs_wait_t tls13_server_handshake(SSL_HANDSHAKE *hs);
// The following functions return human-readable representations of the TLS
// handshake states for debugging.
const char *ssl_client_handshake_state(SSL_HANDSHAKE *hs);
const char *ssl_server_handshake_state(SSL_HANDSHAKE *hs);
const char *tls13_client_handshake_state(SSL_HANDSHAKE *hs);
const char *tls13_server_handshake_state(SSL_HANDSHAKE *hs);
// tls13_add_key_update queues a KeyUpdate message on |ssl|. The
// |update_requested| argument must be one of |SSL_KEY_UPDATE_REQUESTED| or
// |SSL_KEY_UPDATE_NOT_REQUESTED|.
bool tls13_add_key_update(SSL *ssl, int update_requested);
// tls13_post_handshake processes a post-handshake message. It returns true on
// success and false on failure.
bool tls13_post_handshake(SSL *ssl, const SSLMessage &msg);
bool tls13_process_certificate(SSL_HANDSHAKE *hs, const SSLMessage &msg,
bool allow_anonymous);
bool tls13_process_certificate_verify(SSL_HANDSHAKE *hs, const SSLMessage &msg);
// tls13_process_finished processes |msg| as a Finished message from the
// peer. If |use_saved_value| is true, the verify_data is compared against
// |hs->expected_client_finished| rather than computed fresh.
bool tls13_process_finished(SSL_HANDSHAKE *hs, const SSLMessage &msg,
bool use_saved_value);
bool tls13_add_certificate(SSL_HANDSHAKE *hs);
Simplify ssl_private_key_* state machine points. The original motivation behind the sign/complete split was to avoid needlessly hashing the input on each pass through the state machine, but we're payload-based now and, in all cases, the payload is either cheap to compute or readily available. (Even the hashing worry was probably unnecessary.) Tweak ssl_private_key_{sign,decrypt} to automatically call ssl_private_key_complete as needed and take advantage of this in the handshake state machines: - TLS 1.3 signing now computes the payload each pass. The payload is small and we're already allocating a comparable-sized buffer each iteration to hold the signature. This shouldn't be a big deal. - TLS 1.2 decryption code still needs two states due to reading the message (fixed in new state machine style), but otherwise it just performs cheap idempotent tasks again. The PSK code is reshuffled to guarantee the callback is not called twice (though this was impossible anyway because we don't support RSA_PSK). - TLS 1.2 CertificateVerify signing is easy as the transcript is readily available. The buffer is released very slightly later, but it shouldn't matter. - TLS 1.2 ServerKeyExchange signing required some reshuffling. Assembling the ServerKeyExchange parameters is moved to the previous state. The signing payload has some randoms prepended. This is cheap enough, but a nuisance in C. Pre-prepend the randoms in hs->server_params. With this change, we are *nearly* rid of the A/B => same function pattern. BUG=128 Change-Id: Iec4fe0be7cfc88a6de027ba2760fae70794ea810 Reviewed-on: https://boringssl-review.googlesource.com/17265 Commit-Queue: David Benjamin <davidben@google.com> Commit-Queue: Steven Valdez <svaldez@google.com> Reviewed-by: Steven Valdez <svaldez@google.com>
2017-06-17 18:20:59 +01:00
// tls13_add_certificate_verify adds a TLS 1.3 CertificateVerify message to the
// handshake. If it returns |ssl_private_key_retry|, it should be called again
// to retry when the signing operation is completed.
Simplify ssl_private_key_* state machine points. The original motivation behind the sign/complete split was to avoid needlessly hashing the input on each pass through the state machine, but we're payload-based now and, in all cases, the payload is either cheap to compute or readily available. (Even the hashing worry was probably unnecessary.) Tweak ssl_private_key_{sign,decrypt} to automatically call ssl_private_key_complete as needed and take advantage of this in the handshake state machines: - TLS 1.3 signing now computes the payload each pass. The payload is small and we're already allocating a comparable-sized buffer each iteration to hold the signature. This shouldn't be a big deal. - TLS 1.2 decryption code still needs two states due to reading the message (fixed in new state machine style), but otherwise it just performs cheap idempotent tasks again. The PSK code is reshuffled to guarantee the callback is not called twice (though this was impossible anyway because we don't support RSA_PSK). - TLS 1.2 CertificateVerify signing is easy as the transcript is readily available. The buffer is released very slightly later, but it shouldn't matter. - TLS 1.2 ServerKeyExchange signing required some reshuffling. Assembling the ServerKeyExchange parameters is moved to the previous state. The signing payload has some randoms prepended. This is cheap enough, but a nuisance in C. Pre-prepend the randoms in hs->server_params. With this change, we are *nearly* rid of the A/B => same function pattern. BUG=128 Change-Id: Iec4fe0be7cfc88a6de027ba2760fae70794ea810 Reviewed-on: https://boringssl-review.googlesource.com/17265 Commit-Queue: David Benjamin <davidben@google.com> Commit-Queue: Steven Valdez <svaldez@google.com> Reviewed-by: Steven Valdez <svaldez@google.com>
2017-06-17 18:20:59 +01:00
enum ssl_private_key_result_t tls13_add_certificate_verify(SSL_HANDSHAKE *hs);
bool tls13_add_finished(SSL_HANDSHAKE *hs);
bool tls13_process_new_session_ticket(SSL *ssl, const SSLMessage &msg);
bool ssl_ext_key_share_parse_serverhello(SSL_HANDSHAKE *hs,
Array<uint8_t> *out_secret,
uint8_t *out_alert, CBS *contents);
bool ssl_ext_key_share_parse_clienthello(SSL_HANDSHAKE *hs, bool *out_found,
Array<uint8_t> *out_secret,
uint8_t *out_alert, CBS *contents);
bool ssl_ext_key_share_add_serverhello(SSL_HANDSHAKE *hs, CBB *out);
bool ssl_ext_pre_shared_key_parse_serverhello(SSL_HANDSHAKE *hs,
uint8_t *out_alert,
CBS *contents);
bool ssl_ext_pre_shared_key_parse_clienthello(
SSL_HANDSHAKE *hs, CBS *out_ticket, CBS *out_binders,
uint32_t *out_obfuscated_ticket_age, uint8_t *out_alert, CBS *contents);
bool ssl_ext_pre_shared_key_add_serverhello(SSL_HANDSHAKE *hs, CBB *out);
// ssl_is_sct_list_valid does a shallow parse of the SCT list in |contents| and
// returns whether it's valid.
bool ssl_is_sct_list_valid(const CBS *contents);
bool ssl_write_client_hello(SSL_HANDSHAKE *hs);
enum ssl_cert_verify_context_t {
ssl_cert_verify_server,
ssl_cert_verify_client,
ssl_cert_verify_channel_id,
};
// tls13_get_cert_verify_signature_input generates the message to be signed for
// TLS 1.3's CertificateVerify message. |cert_verify_context| determines the
// type of signature. It sets |*out| to a newly allocated buffer containing the
// result. This function returns true on success and false on failure.
bool tls13_get_cert_verify_signature_input(
SSL_HANDSHAKE *hs, Array<uint8_t> *out,
enum ssl_cert_verify_context_t cert_verify_context);
// ssl_is_alpn_protocol_allowed returns whether |protocol| is a valid server
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// selection for |hs->ssl|'s client preferences.
bool ssl_is_alpn_protocol_allowed(const SSL_HANDSHAKE *hs,
Span<const uint8_t> protocol);
// ssl_negotiate_alpn negotiates the ALPN extension, if applicable. It returns
// true on successful negotiation or if nothing was negotiated. It returns false
// and sets |*out_alert| to an alert on error.
bool ssl_negotiate_alpn(SSL_HANDSHAKE *hs, uint8_t *out_alert,
const SSL_CLIENT_HELLO *client_hello);
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
struct SSL_EXTENSION_TYPE {
uint16_t type;
bool *out_present;
CBS *out_data;
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
};
// ssl_parse_extensions parses a TLS extensions block out of |cbs| and advances
// it. It writes the parsed extensions to pointers denoted by |ext_types|. On
// success, it fills in the |out_present| and |out_data| fields and returns one.
// Otherwise, it sets |*out_alert| to an alert to send and returns zero. Unknown
// extensions are rejected unless |ignore_unknown| is 1.
int ssl_parse_extensions(const CBS *cbs, uint8_t *out_alert,
const SSL_EXTENSION_TYPE *ext_types,
size_t num_ext_types, int ignore_unknown);
// ssl_verify_peer_cert verifies the peer certificate for |hs|.
enum ssl_verify_result_t ssl_verify_peer_cert(SSL_HANDSHAKE *hs);
// ssl_reverify_peer_cert verifies the peer certificate for |hs| when resuming a
// session.
enum ssl_verify_result_t ssl_reverify_peer_cert(SSL_HANDSHAKE *hs);
enum ssl_hs_wait_t ssl_get_finished(SSL_HANDSHAKE *hs);
bool ssl_send_finished(SSL_HANDSHAKE *hs);
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
bool ssl_output_cert_chain(SSL_HANDSHAKE *hs);
// SSLKEYLOGFILE functions.
2016-07-19 06:26:49 +01:00
// ssl_log_secret logs |secret| with label |label|, if logging is enabled for
// |ssl|. It returns one on success and zero on failure.
2016-07-19 06:26:49 +01:00
int ssl_log_secret(const SSL *ssl, const char *label, const uint8_t *secret,
size_t secret_len);
// ClientHello functions.
bool ssl_client_hello_init(SSL *ssl, SSL_CLIENT_HELLO *out,
const SSLMessage &msg);
bool ssl_client_hello_get_extension(const SSL_CLIENT_HELLO *client_hello,
CBS *out, uint16_t extension_type);
bool ssl_client_cipher_list_contains_cipher(
const SSL_CLIENT_HELLO *client_hello, uint16_t id);
// GREASE.
// ssl_get_grease_value returns a GREASE value for |hs|. For a given
// connection, the values for each index will be deterministic. This allows the
// same ClientHello be sent twice for a HelloRetryRequest or the same group be
// advertised in both supported_groups and key_shares.
uint16_t ssl_get_grease_value(SSL_HANDSHAKE *hs, enum ssl_grease_index_t index);
// Signature algorithms.
// tls1_parse_peer_sigalgs parses |sigalgs| as the list of peer signature
// algorithms and saves them on |hs|. It returns true on success and false on
// error.
bool tls1_parse_peer_sigalgs(SSL_HANDSHAKE *hs, const CBS *sigalgs);
// tls1_get_legacy_signature_algorithm sets |*out| to the signature algorithm
// that should be used with |pkey| in TLS 1.1 and earlier. It returns true on
// success and false if |pkey| may not be used at those versions.
bool tls1_get_legacy_signature_algorithm(uint16_t *out, const EVP_PKEY *pkey);
// tls1_choose_signature_algorithm sets |*out| to a signature algorithm for use
// with |hs|'s private key based on the peer's preferences and the algorithms
// supported. It returns true on success and false on error.
bool tls1_choose_signature_algorithm(SSL_HANDSHAKE *hs, uint16_t *out);
// tls1_get_peer_verify_algorithms returns the signature schemes for which the
// peer indicated support.
//
// NOTE: The related function |SSL_get0_peer_verify_algorithms| only has
// well-defined behavior during the callbacks set by |SSL_CTX_set_cert_cb| and
// |SSL_CTX_set_client_cert_cb|, or when the handshake is paused because of
// them.
Span<const uint16_t> tls1_get_peer_verify_algorithms(const SSL_HANDSHAKE *hs);
// tls12_add_verify_sigalgs adds the signature algorithms acceptable for the
// peer signature to |out|. It returns true on success and false on error. If
// |for_certs| is true, the potentially more restrictive list of algorithms for
// certificates is used. Otherwise, the online signature one is used.
bool tls12_add_verify_sigalgs(const SSL *ssl, CBB *out, bool for_certs);
// tls12_check_peer_sigalg checks if |sigalg| is acceptable for the peer
// signature. It returns true on success and false on error, setting
// |*out_alert| to an alert to send.
bool tls12_check_peer_sigalg(const SSL *ssl, uint8_t *out_alert,
uint16_t sigalg);
// tls12_has_different_verify_sigalgs_for_certs returns whether |ssl| has a
// different, more restrictive, list of signature algorithms acceptable for the
// certificate than the online signature.
bool tls12_has_different_verify_sigalgs_for_certs(const SSL *ssl);
// Underdocumented functions.
//
// Functions below here haven't been touched up and may be underdocumented.
#define TLSEXT_CHANNEL_ID_SIZE 128
// From RFC4492, used in encoding the curve type in ECParameters
#define NAMED_CURVE_TYPE 3
struct CERT {
static constexpr bool kAllowUniquePtr = true;
explicit CERT(const SSL_X509_METHOD *x509_method);
~CERT();
UniquePtr<EVP_PKEY> privatekey;
// chain contains the certificate chain, with the leaf at the beginning. The
// first element of |chain| may be NULL to indicate that the leaf certificate
// has not yet been set.
// If |chain| != NULL -> len(chain) >= 1
// If |chain[0]| == NULL -> len(chain) >= 2.
// |chain[1..]| != NULL
UniquePtr<STACK_OF(CRYPTO_BUFFER)> chain;
// x509_chain may contain a parsed copy of |chain[1..]|. This is only used as
// a cache in order to implement “get0” functions that return a non-owning
// pointer to the certificate chain.
STACK_OF(X509) *x509_chain = nullptr;
// x509_leaf may contain a parsed copy of the first element of |chain|. This
// is only used as a cache in order to implement “get0” functions that return
// a non-owning pointer to the certificate chain.
X509 *x509_leaf = nullptr;
// x509_stash contains the last |X509| object append to the chain. This is a
// workaround for some third-party code that continue to use an |X509| object
// even after passing ownership with an “add0” function.
X509 *x509_stash = nullptr;
// key_method, if non-NULL, is a set of callbacks to call for private key
// operations.
const SSL_PRIVATE_KEY_METHOD *key_method = nullptr;
// x509_method contains pointers to functions that might deal with |X509|
// compatibility, or might be a no-op, depending on the application.
const SSL_X509_METHOD *x509_method = nullptr;
// sigalgs, if non-empty, is the set of signature algorithms supported by
// |privatekey| in decreasing order of preference.
Array<uint16_t> sigalgs;
// Certificate setup callback: if set is called whenever a
// certificate may be required (client or server). the callback
// can then examine any appropriate parameters and setup any
// certificates required. This allows advanced applications
// to select certificates on the fly: for example based on
// supported signature algorithms or curves.
int (*cert_cb)(SSL *ssl, void *arg) = nullptr;
void *cert_cb_arg = nullptr;
// Optional X509_STORE for certificate validation. If NULL the parent SSL_CTX
// store is used instead.
X509_STORE *verify_store = nullptr;
// Signed certificate timestamp list to be sent to the client, if requested
UniquePtr<CRYPTO_BUFFER> signed_cert_timestamp_list;
// OCSP response to be sent to the client, if requested.
UniquePtr<CRYPTO_BUFFER> ocsp_response;
// sid_ctx partitions the session space within a shared session cache or
// ticket key. Only sessions with a matching value will be accepted.
uint8_t sid_ctx_length = 0;
uint8_t sid_ctx[SSL_MAX_SID_CTX_LENGTH] = {0};
// Delegated credentials.
// dc is the delegated credential to send to the peer (if requested).
UniquePtr<DC> dc = nullptr;
// dc_privatekey is used instead of |privatekey| or |key_method| to
// authenticate the host if a delegated credential is used in the handshake.
UniquePtr<EVP_PKEY> dc_privatekey = nullptr;
// dc_key_method, if not NULL, is used instead of |dc_privatekey| to
// authenticate the host.
const SSL_PRIVATE_KEY_METHOD *dc_key_method = nullptr;
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
};
// |SSL_PROTOCOL_METHOD| abstracts between TLS and DTLS.
struct SSL_PROTOCOL_METHOD {
bool is_dtls;
bool (*ssl_new)(SSL *ssl);
void (*ssl_free)(SSL *ssl);
// get_message sets |*out| to the current handshake message and returns true
// if one has been received. It returns false if more input is needed.
bool (*get_message)(SSL *ssl, SSLMessage *out);
// next_message is called to release the current handshake message.
void (*next_message)(SSL *ssl);
// Use the |ssl_open_handshake| wrapper.
ssl_open_record_t (*open_handshake)(SSL *ssl, size_t *out_consumed,
uint8_t *out_alert, Span<uint8_t> in);
// Use the |ssl_open_change_cipher_spec| wrapper.
ssl_open_record_t (*open_change_cipher_spec)(SSL *ssl, size_t *out_consumed,
uint8_t *out_alert,
Span<uint8_t> in);
// Use the |ssl_open_app_data| wrapper.
ssl_open_record_t (*open_app_data)(SSL *ssl, Span<uint8_t> *out,
size_t *out_consumed, uint8_t *out_alert,
Span<uint8_t> in);
int (*write_app_data)(SSL *ssl, bool *out_needs_handshake, const uint8_t *buf,
int len);
int (*dispatch_alert)(SSL *ssl);
// init_message begins a new handshake message of type |type|. |cbb| is the
// root CBB to be passed into |finish_message|. |*body| is set to a child CBB
// the caller should write to. It returns true on success and false on error.
bool (*init_message)(SSL *ssl, CBB *cbb, CBB *body, uint8_t type);
// finish_message finishes a handshake message. It sets |*out_msg| to the
// serialized message. It returns true on success and false on error.
bool (*finish_message)(SSL *ssl, CBB *cbb, bssl::Array<uint8_t> *out_msg);
// add_message adds a handshake message to the pending flight. It returns
// true on success and false on error.
bool (*add_message)(SSL *ssl, bssl::Array<uint8_t> msg);
// add_change_cipher_spec adds a ChangeCipherSpec record to the pending
// flight. It returns true on success and false on error.
bool (*add_change_cipher_spec)(SSL *ssl);
// flush_flight flushes the pending flight to the transport. It returns one on
// success and <= 0 on error.
int (*flush_flight)(SSL *ssl);
// on_handshake_complete is called when the handshake is complete.
void (*on_handshake_complete)(SSL *ssl);
// set_read_state sets |ssl|'s read cipher state to |aead_ctx|. It returns
// true on success and false if changing the read state is forbidden at this
// point.
bool (*set_read_state)(SSL *ssl, UniquePtr<SSLAEADContext> aead_ctx);
// set_write_state sets |ssl|'s write cipher state to |aead_ctx|. It returns
// true on success and false if changing the write state is forbidden at this
// point.
bool (*set_write_state)(SSL *ssl, UniquePtr<SSLAEADContext> aead_ctx);
};
// The following wrappers call |open_*| but handle |read_shutdown| correctly.
// ssl_open_handshake processes a record from |in| for reading a handshake
// message.
ssl_open_record_t ssl_open_handshake(SSL *ssl, size_t *out_consumed,
uint8_t *out_alert, Span<uint8_t> in);
// ssl_open_change_cipher_spec processes a record from |in| for reading a
// ChangeCipherSpec.
ssl_open_record_t ssl_open_change_cipher_spec(SSL *ssl, size_t *out_consumed,
uint8_t *out_alert,
Span<uint8_t> in);
// ssl_open_app_data processes a record from |in| for reading application data.
// On success, it returns |ssl_open_record_success| and sets |*out| to the
// input. If it encounters a post-handshake message, it returns
// |ssl_open_record_discard|. The caller should then retry, after processing any
// messages received with |get_message|.
ssl_open_record_t ssl_open_app_data(SSL *ssl, Span<uint8_t> *out,
size_t *out_consumed, uint8_t *out_alert,
Span<uint8_t> in);
struct SSL_X509_METHOD {
// check_client_CA_list returns one if |names| is a good list of X.509
// distinguished names and zero otherwise. This is used to ensure that we can
// reject unparsable values at handshake time when using crypto/x509.
int (*check_client_CA_list)(STACK_OF(CRYPTO_BUFFER) *names);
// cert_clear frees and NULLs all X509 certificate-related state.
void (*cert_clear)(CERT *cert);
// cert_free frees all X509-related state.
void (*cert_free)(CERT *cert);
// cert_flush_cached_chain drops any cached |X509|-based certificate chain
// from |cert|.
// cert_dup duplicates any needed fields from |cert| to |new_cert|.
void (*cert_dup)(CERT *new_cert, const CERT *cert);
void (*cert_flush_cached_chain)(CERT *cert);
// cert_flush_cached_chain drops any cached |X509|-based leaf certificate
// from |cert|.
void (*cert_flush_cached_leaf)(CERT *cert);
// session_cache_objects fills out |sess->x509_peer| and |sess->x509_chain|
// from |sess->certs| and erases |sess->x509_chain_without_leaf|. It returns
// one on success or zero on error.
int (*session_cache_objects)(SSL_SESSION *session);
// session_dup duplicates any needed fields from |session| to |new_session|.
// It returns one on success or zero on error.
int (*session_dup)(SSL_SESSION *new_session, const SSL_SESSION *session);
// session_clear frees any X509-related state from |session|.
void (*session_clear)(SSL_SESSION *session);
// session_verify_cert_chain verifies the certificate chain in |session|,
// sets |session->verify_result| and returns one on success or zero on
// error.
int (*session_verify_cert_chain)(SSL_SESSION *session, SSL_HANDSHAKE *ssl,
uint8_t *out_alert);
// hs_flush_cached_ca_names drops any cached |X509_NAME|s from |hs|.
void (*hs_flush_cached_ca_names)(SSL_HANDSHAKE *hs);
// ssl_new does any neccessary initialisation of |hs|. It returns one on
// success or zero on error.
int (*ssl_new)(SSL_HANDSHAKE *hs);
// ssl_free frees anything created by |ssl_new|.
void (*ssl_config_free)(SSL_CONFIG *cfg);
// ssl_flush_cached_client_CA drops any cached |X509_NAME|s from |ssl|.
void (*ssl_flush_cached_client_CA)(SSL_CONFIG *cfg);
// ssl_auto_chain_if_needed runs the deprecated auto-chaining logic if
// necessary. On success, it updates |ssl|'s certificate configuration as
// needed and returns one. Otherwise, it returns zero.
int (*ssl_auto_chain_if_needed)(SSL_HANDSHAKE *hs);
// ssl_ctx_new does any neccessary initialisation of |ctx|. It returns one on
// success or zero on error.
int (*ssl_ctx_new)(SSL_CTX *ctx);
// ssl_ctx_free frees anything created by |ssl_ctx_new|.
void (*ssl_ctx_free)(SSL_CTX *ctx);
// ssl_ctx_flush_cached_client_CA drops any cached |X509_NAME|s from |ctx|.
void (*ssl_ctx_flush_cached_client_CA)(SSL_CTX *ssl);
};
// ssl_crypto_x509_method provides the |SSL_X509_METHOD| functions using
// crypto/x509.
Move libssl's internals into the bssl namespace. This is horrible, but everything else I tried was worse. The goal with this CL is to take the extern "C" out of ssl/internal.h and move most symbols to namespace bssl, so we can start using C++ helpers and destructors without worry. Complications: - Public API functions must be extern "C" and match their declaration in ssl.h, which is unnamespaced. C++ really does not want you to interleave namespaced and unnamespaced things. One can actually write a namespaced extern "C" function, but this means, from C++'s perspective, the function is namespaced. Trying to namespace the public header would worked but ended up too deep a rabbithole. - Our STACK_OF macros do not work right in namespaces. - The typedefs for our exposed but opaque types are visible in the header files and copied into consuming projects as forward declarations. We ultimately want to give SSL a destructor, but clobbering an unnamespaced ssl_st::~ssl_st seems bad manners. - MSVC complains about ambiguous names if one typedefs SSL to bssl::SSL. This CL opts for: - ssl/*.cc must begin with #define BORINGSSL_INTERNAL_CXX_TYPES. This informs the public headers to create forward declarations which are compatible with our namespaces. - For now, C++-defined type FOO ends up at bssl::FOO with a typedef outside. Later I imagine we'll rename many of them. - Internal functions get namespace bssl, so we stop worrying about stomping the tls1_prf symbol. Exported C functions are stuck as they are. Rather than try anything weird, bite the bullet and reorder files which have a mix of public and private functions. I expect that over time, the public functions will become fairly small as we move logic to more idiomatic C++. Files without any public C functions can just be written normally. - To avoid MSVC troubles, some bssl types are renamed to CPlusPlusStyle in advance of them being made idiomatic C++. Bug: 132 Change-Id: Ic931895e117c38b14ff8d6e5a273e868796c7581 Reviewed-on: https://boringssl-review.googlesource.com/18124 Reviewed-by: David Benjamin <davidben@google.com>
2017-07-18 21:34:25 +01:00
extern const SSL_X509_METHOD ssl_crypto_x509_method;
// ssl_noop_x509_method provides the |SSL_X509_METHOD| functions that avoid
// crypto/x509.
extern const SSL_X509_METHOD ssl_noop_x509_method;
struct TicketKey {
static constexpr bool kAllowUniquePtr = true;
uint8_t name[SSL_TICKET_KEY_NAME_LEN] = {0};
uint8_t hmac_key[16] = {0};
uint8_t aes_key[16] = {0};
// next_rotation_tv_sec is the time (in seconds from the epoch) when the
// current key should be superseded by a new key, or the time when a previous
// key should be dropped. If zero, then the key should not be automatically
// rotated.
uint64_t next_rotation_tv_sec = 0;
};
struct CertCompressionAlg {
static constexpr bool kAllowUniquePtr = true;
ssl_cert_compression_func_t compress = nullptr;
ssl_cert_decompression_func_t decompress = nullptr;
uint16_t alg_id = 0;
};
BSSL_NAMESPACE_END
DEFINE_LHASH_OF(SSL_SESSION)
DEFINE_NAMED_STACK_OF(CertCompressionAlg, bssl::CertCompressionAlg)
BSSL_NAMESPACE_BEGIN
// An ssl_shutdown_t describes the shutdown state of one end of the connection,
// whether it is alive or has been shutdown via close_notify or fatal alert.
enum ssl_shutdown_t {
ssl_shutdown_none = 0,
ssl_shutdown_close_notify = 1,
ssl_shutdown_error = 2,
};
struct SSL3_STATE {
static constexpr bool kAllowUniquePtr = true;
SSL3_STATE();
~SSL3_STATE();
uint8_t read_sequence[8] = {0};
uint8_t write_sequence[8] = {0};
uint8_t server_random[SSL3_RANDOM_SIZE] = {0};
uint8_t client_random[SSL3_RANDOM_SIZE] = {0};
// read_buffer holds data from the transport to be processed.
SSLBuffer read_buffer;
// write_buffer holds data to be written to the transport.
SSLBuffer write_buffer;
// pending_app_data is the unconsumed application data. It points into
// |read_buffer|.
Span<uint8_t> pending_app_data;
// partial write - check the numbers match
unsigned int wnum = 0; // number of bytes sent so far
int wpend_tot = 0; // number bytes written
int wpend_type = 0;
int wpend_ret = 0; // number of bytes submitted
const uint8_t *wpend_buf = nullptr;
// read_shutdown is the shutdown state for the read half of the connection.
enum ssl_shutdown_t read_shutdown = ssl_shutdown_none;
// write_shutdown is the shutdown state for the write half of the connection.
enum ssl_shutdown_t write_shutdown = ssl_shutdown_none;
// read_error, if |read_shutdown| is |ssl_shutdown_error|, is the error for
// the receive half of the connection.
UniquePtr<ERR_SAVE_STATE> read_error;
int alert_dispatch = 0;
int total_renegotiations = 0;
// This holds a variable that indicates what we were doing when a 0 or -1 is
// returned. This is needed for non-blocking IO so we know what request
// needs re-doing when in SSL_accept or SSL_connect
int rwstate = SSL_NOTHING;
enum ssl_encryption_level_t read_level = ssl_encryption_initial;
enum ssl_encryption_level_t write_level = ssl_encryption_initial;
// early_data_skipped is the amount of early data that has been skipped by the
// record layer.
uint16_t early_data_skipped = 0;
// empty_record_count is the number of consecutive empty records received.
uint8_t empty_record_count = 0;
// warning_alert_count is the number of consecutive warning alerts
// received.
uint8_t warning_alert_count = 0;
// key_update_count is the number of consecutive KeyUpdates received.
uint8_t key_update_count = 0;
// The negotiated Token Binding key parameter. Only valid if
// |token_binding_negotiated| is set.
uint8_t negotiated_token_binding_param = 0;
// skip_early_data instructs the record layer to skip unexpected early data
// messages when 0RTT is rejected.
bool skip_early_data : 1;
// have_version is true if the connection's final version is known. Otherwise
// the version has not been negotiated yet.
bool have_version : 1;
// v2_hello_done is true if the peer's V2ClientHello, if any, has been handled
// and future messages should use the record layer.
bool v2_hello_done : 1;
// is_v2_hello is true if the current handshake message was derived from a
// V2ClientHello rather than received from the peer directly.
bool is_v2_hello : 1;
// has_message is true if the current handshake message has been returned
// at least once by |get_message| and false otherwise.
bool has_message : 1;
// initial_handshake_complete is true if the initial handshake has
// completed.
bool initial_handshake_complete : 1;
// session_reused indicates whether a session was resumed.
bool session_reused : 1;
bool send_connection_binding : 1;
// In a client, this means that the server supported Channel ID and that a
// Channel ID was sent. In a server it means that we echoed support for
// Channel IDs and that |channel_id| will be valid after the handshake.
bool channel_id_valid : 1;
// key_update_pending is true if we have a KeyUpdate acknowledgment
// outstanding.
bool key_update_pending : 1;
// wpend_pending is true if we have a pending write outstanding.
bool wpend_pending : 1;
// early_data_accepted is true if early data was accepted by the server.
bool early_data_accepted : 1;
// tls13_downgrade is whether the TLS 1.3 anti-downgrade logic fired.
bool tls13_downgrade : 1;
// token_binding_negotiated is set if Token Binding was negotiated.
bool token_binding_negotiated : 1;
// hs_buf is the buffer of handshake data to process.
UniquePtr<BUF_MEM> hs_buf;
// pending_hs_data contains the pending handshake data that has not yet
// been encrypted to |pending_flight|. This allows packing the handshake into
// fewer records.
UniquePtr<BUF_MEM> pending_hs_data;
// pending_flight is the pending outgoing flight. This is used to flush each
// handshake flight in a single write. |write_buffer| must be written out
// before this data.
UniquePtr<BUF_MEM> pending_flight;
// pending_flight_offset is the number of bytes of |pending_flight| which have
// been successfully written.
uint32_t pending_flight_offset = 0;
// ticket_age_skew is the difference, in seconds, between the client-sent
// ticket age and the server-computed value in TLS 1.3 server connections
// which resumed a session.
int32_t ticket_age_skew = 0;
// aead_read_ctx is the current read cipher state.
UniquePtr<SSLAEADContext> aead_read_ctx;
// aead_write_ctx is the current write cipher state.
UniquePtr<SSLAEADContext> aead_write_ctx;
// hs is the handshake state for the current handshake or NULL if there isn't
// one.
UniquePtr<SSL_HANDSHAKE> hs;
uint8_t write_traffic_secret[EVP_MAX_MD_SIZE] = {0};
uint8_t read_traffic_secret[EVP_MAX_MD_SIZE] = {0};
uint8_t exporter_secret[EVP_MAX_MD_SIZE] = {0};
uint8_t early_exporter_secret[EVP_MAX_MD_SIZE] = {0};
uint8_t write_traffic_secret_len = 0;
uint8_t read_traffic_secret_len = 0;
uint8_t exporter_secret_len = 0;
uint8_t early_exporter_secret_len = 0;
// Connection binding to prevent renegotiation attacks
uint8_t previous_client_finished[12] = {0};
uint8_t previous_client_finished_len = 0;
uint8_t previous_server_finished_len = 0;
uint8_t previous_server_finished[12] = {0};
uint8_t send_alert[2] = {0};
// established_session is the session established by the connection. This
// session is only filled upon the completion of the handshake and is
// immutable.
UniquePtr<SSL_SESSION> established_session;
// Next protocol negotiation. For the client, this is the protocol that we
// sent in NextProtocol and is set when handling ServerHello extensions.
//
// For a server, this is the client's selected_protocol from NextProtocol and
// is set when handling the NextProtocol message, before the Finished
// message.
Array<uint8_t> next_proto_negotiated;
// ALPN information
// (we are in the process of transitioning from NPN to ALPN.)
// In a server these point to the selected ALPN protocol after the
// ClientHello has been processed. In a client these contain the protocol
// that the server selected once the ServerHello has been processed.
Array<uint8_t> alpn_selected;
// hostname, on the server, is the value of the SNI extension.
UniquePtr<char> hostname;
// For a server:
// If |channel_id_valid| is true, then this contains the
// verified Channel ID from the client: a P256 point, (x,y), where
// each are big-endian values.
uint8_t channel_id[64] = {0};
// Contains the QUIC transport params received by the peer.
Array<uint8_t> peer_quic_transport_params;
// srtp_profile is the selected SRTP protection profile for
// DTLS-SRTP.
const SRTP_PROTECTION_PROFILE *srtp_profile = nullptr;
};
// lengths of messages
#define DTLS1_COOKIE_LENGTH 256
#define DTLS1_RT_HEADER_LENGTH 13
#define DTLS1_HM_HEADER_LENGTH 12
#define DTLS1_CCS_HEADER_LENGTH 1
#define DTLS1_AL_HEADER_LENGTH 2
struct hm_header_st {
uint8_t type;
uint32_t msg_len;
uint16_t seq;
uint32_t frag_off;
uint32_t frag_len;
};
// An hm_fragment is an incoming DTLS message, possibly not yet assembled.
struct hm_fragment {
static constexpr bool kAllowUniquePtr = true;
hm_fragment() {}
hm_fragment(const hm_fragment &) = delete;
hm_fragment &operator=(const hm_fragment &) = delete;
~hm_fragment();
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// type is the type of the message.
uint8_t type = 0;
// seq is the sequence number of this message.
uint16_t seq = 0;
// msg_len is the length of the message body.
uint32_t msg_len = 0;
// data is a pointer to the message, including message header. It has length
// |DTLS1_HM_HEADER_LENGTH| + |msg_len|.
uint8_t *data = nullptr;
// reassembly is a bitmask of |msg_len| bits corresponding to which parts of
// the message have been received. It is NULL if the message is complete.
uint8_t *reassembly = nullptr;
};
struct OPENSSL_timeval {
uint64_t tv_sec;
uint32_t tv_usec;
};
struct DTLS1_STATE {
static constexpr bool kAllowUniquePtr = true;
DTLS1_STATE();
~DTLS1_STATE();
// has_change_cipher_spec is true if we have received a ChangeCipherSpec from
// the peer in this epoch.
bool has_change_cipher_spec : 1;
// outgoing_messages_complete is true if |outgoing_messages| has been
// completed by an attempt to flush it. Future calls to |add_message| and
// |add_change_cipher_spec| will start a new flight.
bool outgoing_messages_complete : 1;
// flight_has_reply is true if the current outgoing flight is complete and has
// processed at least one message. This is used to detect whether we or the
// peer sent the final flight.
bool flight_has_reply : 1;
uint8_t cookie[DTLS1_COOKIE_LENGTH] = {0};
size_t cookie_len = 0;
// The current data and handshake epoch. This is initially undefined, and
// starts at zero once the initial handshake is completed.
uint16_t r_epoch = 0;
uint16_t w_epoch = 0;
// records being received in the current epoch
DTLS1_BITMAP bitmap;
uint16_t handshake_write_seq = 0;
uint16_t handshake_read_seq = 0;
// save last sequence number for retransmissions
uint8_t last_write_sequence[8] = {0};
UniquePtr<SSLAEADContext> last_aead_write_ctx;
// incoming_messages is a ring buffer of incoming handshake messages that have
// yet to be processed. The front of the ring buffer is message number
// |handshake_read_seq|, at position |handshake_read_seq| %
// |SSL_MAX_HANDSHAKE_FLIGHT|.
UniquePtr<hm_fragment> incoming_messages[SSL_MAX_HANDSHAKE_FLIGHT];
// outgoing_messages is the queue of outgoing messages from the last handshake
// flight.
DTLS_OUTGOING_MESSAGE outgoing_messages[SSL_MAX_HANDSHAKE_FLIGHT];
uint8_t outgoing_messages_len = 0;
// outgoing_written is the number of outgoing messages that have been
// written.
uint8_t outgoing_written = 0;
// outgoing_offset is the number of bytes of the next outgoing message have
// been written.
uint32_t outgoing_offset = 0;
unsigned mtu = 0; // max DTLS packet size
// num_timeouts is the number of times the retransmit timer has fired since
// the last time it was reset.
unsigned num_timeouts = 0;
// Indicates when the last handshake msg or heartbeat sent will
// timeout.
struct OPENSSL_timeval next_timeout = {0, 0};
// timeout_duration_ms is the timeout duration in milliseconds.
unsigned timeout_duration_ms = 0;
};
// SSL_CONFIG contains configuration bits that can be shed after the handshake
// completes. Objects of this type are not shared; they are unique to a
// particular |SSL|.
//
// See SSL_shed_handshake_config() for more about the conditions under which
// configuration can be shed.
struct SSL_CONFIG {
static constexpr bool kAllowUniquePtr = true;
explicit SSL_CONFIG(SSL *ssl_arg);
~SSL_CONFIG();
// ssl is a non-owning pointer to the parent |SSL| object.
SSL *const ssl = nullptr;
// conf_max_version is the maximum acceptable version configured by
// |SSL_set_max_proto_version|. Note this version is not normalized in DTLS
// and is further constrained by |SSL_OP_NO_*|.
uint16_t conf_max_version = 0;
// conf_min_version is the minimum acceptable version configured by
// |SSL_set_min_proto_version|. Note this version is not normalized in DTLS
// and is further constrained by |SSL_OP_NO_*|.
uint16_t conf_min_version = 0;
X509_VERIFY_PARAM *param = nullptr;
// crypto
UniquePtr<SSLCipherPreferenceList> cipher_list;
// This is used to hold the local certificate used (i.e. the server
// certificate for a server or the client certificate for a client).
UniquePtr<CERT> cert;
int (*verify_callback)(int ok,
X509_STORE_CTX *ctx) =
nullptr; // fail if callback returns 0
enum ssl_verify_result_t (*custom_verify_callback)(
SSL *ssl, uint8_t *out_alert) = nullptr;
// Server-only: psk_identity_hint is the identity hint to send in
// PSK-based key exchanges.
UniquePtr<char> psk_identity_hint;
unsigned (*psk_client_callback)(SSL *ssl, const char *hint, char *identity,
unsigned max_identity_len, uint8_t *psk,
unsigned max_psk_len) = nullptr;
unsigned (*psk_server_callback)(SSL *ssl, const char *identity, uint8_t *psk,
unsigned max_psk_len) = nullptr;
// for server side, keep the list of CA_dn we can use
UniquePtr<STACK_OF(CRYPTO_BUFFER)> client_CA;
// cached_x509_client_CA is a cache of parsed versions of the elements of
// |client_CA|.
STACK_OF(X509_NAME) *cached_x509_client_CA = nullptr;
Array<uint16_t> supported_group_list; // our list
// The client's Channel ID private key.
UniquePtr<EVP_PKEY> channel_id_private;
// For a client, this contains the list of supported protocols in wire
// format.
Array<uint8_t> alpn_client_proto_list;
// Contains a list of supported Token Binding key parameters.
Array<uint8_t> token_binding_params;
// Contains the QUIC transport params that this endpoint will send.
Array<uint8_t> quic_transport_params;
Don't use the buffer BIO in TLS. On the TLS side, we introduce a running buffer of ciphertext. Queuing up pending data consists of encrypting the record into the buffer. This effectively reimplements what the buffer BIO was doing previously, but this resizes to fit the whole flight. As part of this, rename all the functions to add to the pending flight to be more uniform. This CL proposes "add_foo" to add to the pending flight and "flush_flight" to drain it. We add an add_alert hook for alerts but, for now, only the SSL 3.0 warning alert (sent mid-handshake) uses this mechanism. Later work will push this down to the rest of the write path so closure alerts use it too, as in DTLS. The intended end state is that all the ssl_buffer.c and wpend_ret logic will only be used for application data and eventually optionally replaced by the in-place API, while all "incidental" data will be handled internally. For now, the two buffers are mutually exclusive. Moving closure alerts to "incidentals" will change this, but flushing application data early is tricky due to wpend_ret. (If we call ssl_write_buffer_flush, do_ssl3_write doesn't realize it still has a wpend_ret to replay.) That too is all left alone in this change. To keep the diff down, write_message is retained for now and will be removed from the state machines in a follow-up change. BUG=72 Change-Id: Ibce882f5f7196880648f25d5005322ca4055c71d Reviewed-on: https://boringssl-review.googlesource.com/13224 Reviewed-by: Adam Langley <agl@google.com>
2017-01-03 23:37:41 +00:00
// verify_sigalgs, if not empty, is the set of signature algorithms
// accepted from the peer in decreasing order of preference.
Array<uint16_t> verify_sigalgs;
// srtp_profiles is the list of configured SRTP protection profiles for
// DTLS-SRTP.
UniquePtr<STACK_OF(SRTP_PROTECTION_PROFILE)> srtp_profiles;
// verify_mode is a bitmask of |SSL_VERIFY_*| values.
uint8_t verify_mode = SSL_VERIFY_NONE;
// Enable signed certificate time stamps. Currently client only.
bool signed_cert_timestamps_enabled : 1;
// ocsp_stapling_enabled is only used by client connections and indicates
// whether OCSP stapling will be requested.
bool ocsp_stapling_enabled : 1;
// channel_id_enabled is copied from the |SSL_CTX|. For a server, means that
// we'll accept Channel IDs from clients. For a client, means that we'll
// advertise support.
bool channel_id_enabled : 1;
// If enforce_rsa_key_usage is true, the handshake will fail if the
// keyUsage extension is present and incompatible with the TLS usage.
// This field is not read until after certificate verification.
bool enforce_rsa_key_usage : 1;
// retain_only_sha256_of_client_certs is true if we should compute the SHA256
// hash of the peer's certificate and then discard it to save memory and
// session space. Only effective on the server side.
bool retain_only_sha256_of_client_certs : 1;
// handoff indicates that a server should stop after receiving the
// ClientHello and pause the handshake in such a way that |SSL_get_error|
// returns |SSL_HANDOFF|. This is copied in |SSL_new| from the |SSL_CTX|
// element of the same name and may be cleared if the handoff is declined.
bool handoff : 1;
// shed_handshake_config indicates that the handshake config (this object!)
// should be freed after the handshake completes.
bool shed_handshake_config : 1;
// ignore_tls13_downgrade is whether the connection should continue when the
// server random signals a downgrade.
bool ignore_tls13_downgrade : 1;
// jdk11_workaround is whether to disable TLS 1.3 for JDK 11 clients, as a
// workaround for https://bugs.openjdk.java.net/browse/JDK-8211806.
bool jdk11_workaround : 1;
};
// From RFC 8446, used in determining PSK modes.
#define SSL_PSK_DHE_KE 0x1
// kMaxEarlyDataAccepted is the advertised number of plaintext bytes of early
// data that will be accepted. This value should be slightly below
// kMaxEarlyDataSkipped in tls_record.c, which is measured in ciphertext.
static const size_t kMaxEarlyDataAccepted = 14336;
UniquePtr<CERT> ssl_cert_dup(CERT *cert);
void ssl_cert_clear_certs(CERT *cert);
bool ssl_set_cert(CERT *cert, UniquePtr<CRYPTO_BUFFER> buffer);
bool ssl_is_key_type_supported(int key_type);
// ssl_compare_public_and_private_key returns true if |pubkey| is the public
// counterpart to |privkey|. Otherwise it returns false and pushes a helpful
// message on the error queue.
bool ssl_compare_public_and_private_key(const EVP_PKEY *pubkey,
const EVP_PKEY *privkey);
bool ssl_cert_check_private_key(const CERT *cert, const EVP_PKEY *privkey);
int ssl_get_new_session(SSL_HANDSHAKE *hs, int is_server);
int ssl_encrypt_ticket(SSL_HANDSHAKE *hs, CBB *out, const SSL_SESSION *session);
int ssl_ctx_rotate_ticket_encryption_key(SSL_CTX *ctx);
// ssl_session_new returns a newly-allocated blank |SSL_SESSION| or nullptr on
// error.
UniquePtr<SSL_SESSION> ssl_session_new(const SSL_X509_METHOD *x509_method);
// ssl_hash_session_id returns a hash of |session_id|, suitable for a hash table
// keyed on session IDs.
uint32_t ssl_hash_session_id(Span<const uint8_t> session_id);
// SSL_SESSION_parse parses an |SSL_SESSION| from |cbs| and advances |cbs| over
// the parsed data.
OPENSSL_EXPORT UniquePtr<SSL_SESSION> SSL_SESSION_parse(
CBS *cbs, const SSL_X509_METHOD *x509_method, CRYPTO_BUFFER_POOL *pool);
// ssl_session_serialize writes |in| to |cbb| as if it were serialising a
// session for Session-ID resumption. It returns one on success and zero on
// error.
OPENSSL_EXPORT int ssl_session_serialize(const SSL_SESSION *in, CBB *cbb);
// ssl_session_is_context_valid returns one if |session|'s session ID context
// matches the one set on |hs| and zero otherwise.
int ssl_session_is_context_valid(const SSL_HANDSHAKE *hs,
const SSL_SESSION *session);
// ssl_session_is_time_valid returns one if |session| is still valid and zero if
// it has expired.
int ssl_session_is_time_valid(const SSL *ssl, const SSL_SESSION *session);
// ssl_session_is_resumable returns one if |session| is resumable for |hs| and
// zero otherwise.
int ssl_session_is_resumable(const SSL_HANDSHAKE *hs,
const SSL_SESSION *session);
// ssl_session_protocol_version returns the protocol version associated with
// |session|. Note that despite the name, this is not the same as
// |SSL_SESSION_get_protocol_version|. The latter is based on upstream's name.
uint16_t ssl_session_protocol_version(const SSL_SESSION *session);
// ssl_session_get_digest returns the digest used in |session|.
const EVP_MD *ssl_session_get_digest(const SSL_SESSION *session);
void ssl_set_session(SSL *ssl, SSL_SESSION *session);
// ssl_get_prev_session looks up the previous session based on |client_hello|.
// On success, it sets |*out_session| to the session or nullptr if none was
// found. If the session could not be looked up synchronously, it returns
// |ssl_hs_pending_session| and should be called again. If a ticket could not be
// decrypted immediately it returns |ssl_hs_pending_ticket| and should also
// be called again. Otherwise, it returns |ssl_hs_error|.
enum ssl_hs_wait_t ssl_get_prev_session(SSL_HANDSHAKE *hs,
UniquePtr<SSL_SESSION> *out_session,
bool *out_tickets_supported,
bool *out_renew_ticket,
const SSL_CLIENT_HELLO *client_hello);
// The following flags determine which parts of the session are duplicated.
#define SSL_SESSION_DUP_AUTH_ONLY 0x0
#define SSL_SESSION_INCLUDE_TICKET 0x1
#define SSL_SESSION_INCLUDE_NONAUTH 0x2
#define SSL_SESSION_DUP_ALL \
(SSL_SESSION_INCLUDE_TICKET | SSL_SESSION_INCLUDE_NONAUTH)
// SSL_SESSION_dup returns a newly-allocated |SSL_SESSION| with a copy of the
// fields in |session| or nullptr on error. The new session is non-resumable and
// must be explicitly marked resumable once it has been filled in.
OPENSSL_EXPORT UniquePtr<SSL_SESSION> SSL_SESSION_dup(SSL_SESSION *session,
int dup_flags);
// ssl_session_rebase_time updates |session|'s start time to the current time,
// adjusting the timeout so the expiration time is unchanged.
void ssl_session_rebase_time(SSL *ssl, SSL_SESSION *session);
// ssl_session_renew_timeout calls |ssl_session_rebase_time| and renews
// |session|'s timeout to |timeout| (measured from the current time). The
// renewal is clamped to the session's auth_timeout.
void ssl_session_renew_timeout(SSL *ssl, SSL_SESSION *session,
uint32_t timeout);
void ssl_update_cache(SSL_HANDSHAKE *hs, int mode);
int ssl_send_alert(SSL *ssl, int level, int desc);
bool ssl3_get_message(SSL *ssl, SSLMessage *out);
ssl_open_record_t ssl3_open_handshake(SSL *ssl, size_t *out_consumed,
uint8_t *out_alert, Span<uint8_t> in);
void ssl3_next_message(SSL *ssl);
int ssl3_dispatch_alert(SSL *ssl);
ssl_open_record_t ssl3_open_app_data(SSL *ssl, Span<uint8_t> *out,
size_t *out_consumed, uint8_t *out_alert,
Span<uint8_t> in);
ssl_open_record_t ssl3_open_change_cipher_spec(SSL *ssl, size_t *out_consumed,
uint8_t *out_alert,
Span<uint8_t> in);
int ssl3_write_app_data(SSL *ssl, bool *out_needs_handshake, const uint8_t *buf,
int len);
bool ssl3_new(SSL *ssl);
void ssl3_free(SSL *ssl);
bool ssl3_init_message(SSL *ssl, CBB *cbb, CBB *body, uint8_t type);
bool ssl3_finish_message(SSL *ssl, CBB *cbb, Array<uint8_t> *out_msg);
bool ssl3_add_message(SSL *ssl, Array<uint8_t> msg);
bool ssl3_add_change_cipher_spec(SSL *ssl);
int ssl3_flush_flight(SSL *ssl);
bool dtls1_init_message(SSL *ssl, CBB *cbb, CBB *body, uint8_t type);
bool dtls1_finish_message(SSL *ssl, CBB *cbb, Array<uint8_t> *out_msg);
bool dtls1_add_message(SSL *ssl, Array<uint8_t> msg);
bool dtls1_add_change_cipher_spec(SSL *ssl);
int dtls1_flush_flight(SSL *ssl);
// ssl_add_message_cbb finishes the handshake message in |cbb| and adds it to
// the pending flight. It returns true on success and false on error.
bool ssl_add_message_cbb(SSL *ssl, CBB *cbb);
// ssl_hash_message incorporates |msg| into the handshake hash. It returns true
// on success and false on allocation failure.
bool ssl_hash_message(SSL_HANDSHAKE *hs, const SSLMessage &msg);
ssl_open_record_t dtls1_open_app_data(SSL *ssl, Span<uint8_t> *out,
size_t *out_consumed, uint8_t *out_alert,
Span<uint8_t> in);
ssl_open_record_t dtls1_open_change_cipher_spec(SSL *ssl, size_t *out_consumed,
uint8_t *out_alert,
Span<uint8_t> in);
int dtls1_write_app_data(SSL *ssl, bool *out_needs_handshake,
const uint8_t *buf, int len);
// dtls1_write_record sends a record. It returns one on success and <= 0 on
// error.
int dtls1_write_record(SSL *ssl, int type, const uint8_t *buf, size_t len,
enum dtls1_use_epoch_t use_epoch);
int dtls1_retransmit_outgoing_messages(SSL *ssl);
bool dtls1_parse_fragment(CBS *cbs, struct hm_header_st *out_hdr,
CBS *out_body);
bool dtls1_check_timeout_num(SSL *ssl);
void dtls1_start_timer(SSL *ssl);
void dtls1_stop_timer(SSL *ssl);
bool dtls1_is_timer_expired(SSL *ssl);
unsigned int dtls1_min_mtu(void);
bool dtls1_new(SSL *ssl);
void dtls1_free(SSL *ssl);
bool dtls1_get_message(SSL *ssl, SSLMessage *out);
ssl_open_record_t dtls1_open_handshake(SSL *ssl, size_t *out_consumed,
uint8_t *out_alert, Span<uint8_t> in);
void dtls1_next_message(SSL *ssl);
int dtls1_dispatch_alert(SSL *ssl);
// tls1_configure_aead configures either the read or write direction AEAD (as
// determined by |direction|) using the keys generated by the TLS KDF. The
// |key_block_cache| argument is used to store the generated key block, if
// empty. Otherwise it's assumed that the key block is already contained within
// it. Returns one on success or zero on error.
int tls1_configure_aead(SSL *ssl, evp_aead_direction_t direction,
Array<uint8_t> *key_block_cache,
const SSL_CIPHER *cipher,
Span<const uint8_t> iv_override);
int tls1_change_cipher_state(SSL_HANDSHAKE *hs, evp_aead_direction_t direction);
int tls1_generate_master_secret(SSL_HANDSHAKE *hs, uint8_t *out,
Span<const uint8_t> premaster);
// tls1_get_grouplist returns the locally-configured group preference list.
Span<const uint16_t> tls1_get_grouplist(const SSL_HANDSHAKE *ssl);
// tls1_check_group_id returns whether |group_id| is consistent with locally-
// configured group preferences.
bool tls1_check_group_id(const SSL_HANDSHAKE *ssl, uint16_t group_id);
// tls1_get_shared_group sets |*out_group_id| to the first preferred shared
// group between client and server preferences and returns true. If none may be
// found, it returns false.
bool tls1_get_shared_group(SSL_HANDSHAKE *hs, uint16_t *out_group_id);
// tls1_set_curves converts the array of NIDs in |curves| into a newly allocated
// array of TLS group IDs. On success, the function returns true and writes the
// array to |*out_group_ids|. Otherwise, it returns false.
bool tls1_set_curves(Array<uint16_t> *out_group_ids, Span<const int> curves);
// tls1_set_curves_list converts the string of curves pointed to by |curves|
// into a newly allocated array of TLS group IDs. On success, the function
// returns true and writes the array to |*out_group_ids|. Otherwise, it returns
// false.
bool tls1_set_curves_list(Array<uint16_t> *out_group_ids, const char *curves);
// ssl_add_clienthello_tlsext writes ClientHello extensions to |out|. It returns
// true on success and false on failure. The |header_len| argument is the length
// of the ClientHello written so far and is used to compute the padding length.
// (It does not include the record header.)
bool ssl_add_clienthello_tlsext(SSL_HANDSHAKE *hs, CBB *out, size_t header_len);
bool ssl_add_serverhello_tlsext(SSL_HANDSHAKE *hs, CBB *out);
bool ssl_parse_clienthello_tlsext(SSL_HANDSHAKE *hs,
const SSL_CLIENT_HELLO *client_hello);
bool ssl_parse_serverhello_tlsext(SSL_HANDSHAKE *hs, CBS *cbs);
#define tlsext_tick_md EVP_sha256
// ssl_process_ticket processes a session ticket from the client. It returns
// one of:
// |ssl_ticket_aead_success|: |*out_session| is set to the parsed session and
// |*out_renew_ticket| is set to whether the ticket should be renewed.
// |ssl_ticket_aead_ignore_ticket|: |*out_renew_ticket| is set to whether a
// fresh ticket should be sent, but the given ticket cannot be used.
// |ssl_ticket_aead_retry|: the ticket could not be immediately decrypted.
// Retry later.
// |ssl_ticket_aead_error|: an error occured that is fatal to the connection.
enum ssl_ticket_aead_result_t ssl_process_ticket(
SSL_HANDSHAKE *hs, UniquePtr<SSL_SESSION> *out_session,
bool *out_renew_ticket, Span<const uint8_t> ticket,
Span<const uint8_t> session_id);
// tls1_verify_channel_id processes |msg| as a Channel ID message, and verifies
// the signature. If the key is valid, it saves the Channel ID and returns true.
// Otherwise, it returns false.
bool tls1_verify_channel_id(SSL_HANDSHAKE *hs, const SSLMessage &msg);
// tls1_write_channel_id generates a Channel ID message and puts the output in
// |cbb|. |ssl->channel_id_private| must already be set before calling. This
// function returns true on success and false on error.
bool tls1_write_channel_id(SSL_HANDSHAKE *hs, CBB *cbb);
// tls1_channel_id_hash computes the hash to be signed by Channel ID and writes
// it to |out|, which must contain at least |EVP_MAX_MD_SIZE| bytes. It returns
// true on success and false on failure.
bool tls1_channel_id_hash(SSL_HANDSHAKE *hs, uint8_t *out, size_t *out_len);
// tls1_record_handshake_hashes_for_channel_id records the current handshake
// hashes in |hs->new_session| so that Channel ID resumptions can sign that
// data.
bool tls1_record_handshake_hashes_for_channel_id(SSL_HANDSHAKE *hs);
// ssl_do_channel_id_callback checks runs |hs->ssl->ctx->channel_id_cb| if
// necessary. It returns true on success and false on fatal error. Note that, on
// success, |hs->ssl->channel_id_private| may be unset, in which case the
// operation should be retried later.
bool ssl_do_channel_id_callback(SSL_HANDSHAKE *hs);
// ssl_can_write returns whether |ssl| is allowed to write.
bool ssl_can_write(const SSL *ssl);
// ssl_can_read returns wheter |ssl| is allowed to read.
bool ssl_can_read(const SSL *ssl);
void ssl_get_current_time(const SSL *ssl, struct OPENSSL_timeval *out_clock);
void ssl_ctx_get_current_time(const SSL_CTX *ctx,
struct OPENSSL_timeval *out_clock);
// ssl_reset_error_state resets state for |SSL_get_error|.
void ssl_reset_error_state(SSL *ssl);
// ssl_set_read_error sets |ssl|'s read half into an error state, saving the
// current state of the error queue.
void ssl_set_read_error(SSL *ssl);
BSSL_NAMESPACE_END
// Opaque C types.
//
// The following types are exported to C code as public typedefs, so they must
// be defined outside of the namespace.
// ssl_method_st backs the public |SSL_METHOD| type. It is a compatibility
// structure to support the legacy version-locked methods.
struct ssl_method_st {
// version, if non-zero, is the only protocol version acceptable to an
// SSL_CTX initialized from this method.
uint16_t version;
// method is the underlying SSL_PROTOCOL_METHOD that initializes the
// SSL_CTX.
const bssl::SSL_PROTOCOL_METHOD *method;
// x509_method contains pointers to functions that might deal with |X509|
// compatibility, or might be a no-op, depending on the application.
const bssl::SSL_X509_METHOD *x509_method;
};
struct ssl_ctx_st {
explicit ssl_ctx_st(const SSL_METHOD *ssl_method);
ssl_ctx_st(const ssl_ctx_st &) = delete;
ssl_ctx_st &operator=(const ssl_ctx_st &) = delete;
const bssl::SSL_PROTOCOL_METHOD *method = nullptr;
const bssl::SSL_X509_METHOD *x509_method = nullptr;
// lock is used to protect various operations on this object.
CRYPTO_MUTEX lock;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// conf_max_version is the maximum acceptable protocol version configured by
// |SSL_CTX_set_max_proto_version|. Note this version is normalized in DTLS
// and is further constrainted by |SSL_OP_NO_*|.
uint16_t conf_max_version = 0;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// conf_min_version is the minimum acceptable protocol version configured by
// |SSL_CTX_set_min_proto_version|. Note this version is normalized in DTLS
// and is further constrainted by |SSL_OP_NO_*|.
uint16_t conf_min_version = 0;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// quic_method is the method table corresponding to the QUIC hooks.
const SSL_QUIC_METHOD *quic_method = nullptr;
bssl::UniquePtr<bssl::SSLCipherPreferenceList> cipher_list;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
X509_STORE *cert_store = nullptr;
LHASH_OF(SSL_SESSION) *sessions = nullptr;
// Most session-ids that will be cached, default is
// SSL_SESSION_CACHE_MAX_SIZE_DEFAULT. 0 is unlimited.
unsigned long session_cache_size = SSL_SESSION_CACHE_MAX_SIZE_DEFAULT;
SSL_SESSION *session_cache_head = nullptr;
SSL_SESSION *session_cache_tail = nullptr;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// handshakes_since_cache_flush is the number of successful handshakes since
// the last cache flush.
int handshakes_since_cache_flush = 0;
// This can have one of 2 values, ored together,
// SSL_SESS_CACHE_CLIENT,
// SSL_SESS_CACHE_SERVER,
// Default is SSL_SESSION_CACHE_SERVER, which means only
// SSL_accept which cache SSL_SESSIONS.
int session_cache_mode = SSL_SESS_CACHE_SERVER;
// session_timeout is the default lifetime for new sessions in TLS 1.2 and
// earlier, in seconds.
uint32_t session_timeout = SSL_DEFAULT_SESSION_TIMEOUT;
// session_psk_dhe_timeout is the default lifetime for new sessions in TLS
// 1.3, in seconds.
uint32_t session_psk_dhe_timeout = SSL_DEFAULT_SESSION_PSK_DHE_TIMEOUT;
// If this callback is not null, it will be called each time a session id is
// added to the cache. If this function returns 1, it means that the
// callback will do a SSL_SESSION_free() when it has finished using it.
// Otherwise, on 0, it means the callback has finished with it. If
// remove_session_cb is not null, it will be called when a session-id is
// removed from the cache. After the call, OpenSSL will SSL_SESSION_free()
// it.
int (*new_session_cb)(SSL *ssl, SSL_SESSION *sess) = nullptr;
void (*remove_session_cb)(SSL_CTX *ctx, SSL_SESSION *sess) = nullptr;
SSL_SESSION *(*get_session_cb)(SSL *ssl, const uint8_t *data, int len,
int *copy) = nullptr;
CRYPTO_refcount_t references = 1;
// if defined, these override the X509_verify_cert() calls
int (*app_verify_callback)(X509_STORE_CTX *store_ctx, void *arg) = nullptr;
void *app_verify_arg = nullptr;
ssl_verify_result_t (*custom_verify_callback)(SSL *ssl,
uint8_t *out_alert) = nullptr;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// Default password callback.
pem_password_cb *default_passwd_callback = nullptr;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// Default password callback user data.
void *default_passwd_callback_userdata = nullptr;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// get client cert callback
int (*client_cert_cb)(SSL *ssl, X509 **out_x509,
EVP_PKEY **out_pkey) = nullptr;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// get channel id callback
void (*channel_id_cb)(SSL *ssl, EVP_PKEY **out_pkey) = nullptr;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
CRYPTO_EX_DATA ex_data;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// Default values used when no per-SSL value is defined follow
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
void (*info_callback)(const SSL *ssl, int type, int value) = nullptr;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// what we put in client cert requests
bssl::UniquePtr<STACK_OF(CRYPTO_BUFFER)> client_CA;
// cached_x509_client_CA is a cache of parsed versions of the elements of
// |client_CA|.
STACK_OF(X509_NAME) *cached_x509_client_CA = nullptr;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
// Default values to use in SSL structures follow (these are copied by
// SSL_new)
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
uint32_t options = 0;
// Disable the auto-chaining feature by default. wpa_supplicant relies on this
// feature, but require callers opt into it.
uint32_t mode = SSL_MODE_NO_AUTO_CHAIN;
uint32_t max_cert_list = SSL_MAX_CERT_LIST_DEFAULT;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
bssl::UniquePtr<bssl::CERT> cert;
// callback that allows applications to peek at protocol messages
void (*msg_callback)(int write_p, int version, int content_type,
const void *buf, size_t len, SSL *ssl,
void *arg) = nullptr;
void *msg_callback_arg = nullptr;
SSL_CONFIG: new struct for sheddable handshake configuration. |SSL_CONFIG| is a container for bits of configuration that are unneeded after the handshake completes. By default it is retained for the life of the |SSL|, but it may be shed at the caller's option by calling SSL_set_shed_handshake_config(). This is incompatible with renegotiation, and with SSL_clear(). |SSL_CONFIG| is reachable by |ssl->config| and by |hs->config|. The latter is always non-NULL. To avoid null checks, I've changed the signature of a number of functions from |SSL*| arguments to |SSL_HANDSHAKE*| arguments. When configuration has been shed, setters that touch |SSL_CONFIG| return an error value if that is possible. Setters that return |void| do nothing. Getters that request |SSL_CONFIG| values will fail with an |assert| if the configuration has been shed. When asserts are compiled out, they will return an error value. The aim of this commit is to simplify analysis of split-handshakes by making it obvious that some bits of state have no effects beyond the handshake. It also cuts down on memory usage. Of note: |SSL_CTX| is still reachable after the configuration has been shed, and a couple things need to be retained only for the sake of post-handshake hooks. Perhaps these can be fixed in time. Change-Id: Idf09642e0518945b81a1e9fcd7331cc9cf7cc2d6 Bug: 123 Reviewed-on: https://boringssl-review.googlesource.com/27644 Commit-Queue: David Benjamin <davidben@google.com> CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org> Reviewed-by: David Benjamin <davidben@google.com>
2018-04-13 23:51:30 +01:00
int verify_mode = SSL_VERIFY_NONE;
int (*default_verify_callback)(int ok, X509_STORE_CTX *ctx) =
nullptr; // called 'verify_callback' in the SSL
X509_VERIFY_PARAM *param = nullptr;
// select_certificate_cb is called before most ClientHello processing and
// before the decision whether to resume a session is made. See
// |ssl_select_cert_result_t| for details of the return values.
ssl_select_cert_result_t (*select_certificate_cb)(const SSL_CLIENT_HELLO *) =
nullptr;
// dos_protection_cb is called once the resumption decision for a ClientHello
// has been made. It returns one to continue the handshake or zero to
// abort.
int (*dos_protection_cb)(const SSL_CLIENT_HELLO *) = nullptr;
// Controls whether to verify certificates when resuming connections. They
// were already verified when the connection was first made, so the default is
// false. For now, this is only respected on clients, not servers.
bool reverify_on_resume = false;
// Maximum amount of data to send in one fragment. actual record size can be
// more than this due to padding and MAC overheads.
uint16_t max_send_fragment = SSL3_RT_MAX_PLAIN_LENGTH;
// TLS extensions servername callback
int (*servername_callback)(SSL *, int *, void *) = nullptr;
void *servername_arg = nullptr;
// RFC 4507 session ticket keys. |ticket_key_current| may be NULL before the
// first handshake and |ticket_key_prev| may be NULL at any time.
// Automatically generated ticket keys are rotated as needed at handshake
// time. Hence, all access must be synchronized through |lock|.
bssl::UniquePtr<bssl::TicketKey> ticket_key_current;
bssl::UniquePtr<bssl::TicketKey> ticket_key_prev;
// Callback to support customisation of ticket key setting
int (*ticket_key_cb)(SSL *ssl, uint8_t *name, uint8_t *iv,
EVP_CIPHER_CTX *ectx, HMAC_CTX *hctx, int enc) = nullptr;
// Server-only: psk_identity_hint is the default identity hint to send in
// PSK-based key exchanges.
bssl::UniquePtr<char> psk_identity_hint;
unsigned (*psk_client_callback)(SSL *ssl, const char *hint, char *identity,
unsigned max_identity_len, uint8_t *psk,
unsigned max_psk_len) = nullptr;
unsigned (*psk_server_callback)(SSL *ssl, const char *identity, uint8_t *psk,
unsigned max_psk_len) = nullptr;
// Next protocol negotiation information
// (for experimental NPN extension).
// For a server, this contains a callback function by which the set of
// advertised protocols can be provided.
int (*next_protos_advertised_cb)(SSL *ssl, const uint8_t **out,
unsigned *out_len, void *arg) = nullptr;
void *next_protos_advertised_cb_arg = nullptr;
// For a client, this contains a callback function that selects the
// next protocol from the list provided by the server.
int (*next_proto_select_cb)(SSL *ssl, uint8_t **out, uint8_t *out_len,
const uint8_t *in, unsigned in_len,
void *arg) = nullptr;
void *next_proto_select_cb_arg = nullptr;
Revise version negotiation logic on the C side. This is in preparation for upcoming experiments which will require supporting multiple experimental versions of TLS 1.3 with, on the server, the ability to enable multiple variants at once. This means the version <-> wire bijection no longer exists, even when limiting to a single SSL*. Thus version_to_wire is removed and instead we treat the wire version as the canonical version value. There is a mapping from valid wire versions to protocol versions which describe the high-level handshake protocol in use. This mapping is not injective, so uses of version_from_wire are rewritten differently. All the version-munging logic is moved to ssl_versions.c with a master preference list of all TLS and DTLS versions. The legacy version negotiation is converted to the new scheme. The version lists and negotiation are driven by the preference lists and a ssl_supports_version API. To simplify the mess around SSL_SESSION and versions, version_from_wire is now DTLS/TLS-agnostic, with any filtering being done by ssl_supports_version. This is screwy but allows parsing SSL_SESSIONs to sanity-check it and reject all bogus versions in SSL_SESSION. This reduces a mess of error cases. As part of this, the weird logic where ssl->version is set early when sending the ClientHello is removed. The one place where we were relying on this behavior is tweaked to query hs->max_version instead. Change-Id: Ic91b348481ceba94d9ae06d6781187c11adc15b0 Reviewed-on: https://boringssl-review.googlesource.com/17524 Reviewed-by: David Benjamin <davidben@google.com> Commit-Queue: David Benjamin <davidben@google.com>
2017-06-20 15:55:02 +01:00
// ALPN information
// (we are in the process of transitioning from NPN to ALPN.)
// For a server, this contains a callback function that allows the
// server to select the protocol for the connection.
// out: on successful return, this must point to the raw protocol
// name (without the length prefix).
// outlen: on successful return, this contains the length of |*out|.
// in: points to the client's list of supported protocols in
// wire-format.
// inlen: the length of |in|.
int (*alpn_select_cb)(SSL *ssl, const uint8_t **out, uint8_t *out_len,
const uint8_t *in, unsigned in_len,
void *arg) = nullptr;
void *alpn_select_cb_arg = nullptr;
// For a client, this contains the list of supported protocols in wire
// format.
bssl::Array<uint8_t> alpn_client_proto_list;
// SRTP profiles we are willing to do from RFC 5764
bssl::UniquePtr<STACK_OF(SRTP_PROTECTION_PROFILE)> srtp_profiles;
// Defined compression algorithms for certificates.
bssl::UniquePtr<STACK_OF(CertCompressionAlg)> cert_compression_algs;
// Supported group values inherited by SSL structure
bssl::Array<uint16_t> supported_group_list;
2017-01-28 19:00:32 +00:00
// The client's Channel ID private key.
bssl::UniquePtr<EVP_PKEY> channel_id_private;
// keylog_callback, if not NULL, is the key logging callback. See
// |SSL_CTX_set_keylog_callback|.
void (*keylog_callback)(const SSL *ssl, const char *line) = nullptr;
// current_time_cb, if not NULL, is the function to use to get the current
// time. It sets |*out_clock| to the current time. The |ssl| argument is
// always NULL. See |SSL_CTX_set_current_time_cb|.
void (*current_time_cb)(const SSL *ssl, struct timeval *out_clock) = nullptr;
// pool is used for all |CRYPTO_BUFFER|s in case we wish to share certificate
// memory.
CRYPTO_BUFFER_POOL *pool = nullptr;
// ticket_aead_method contains function pointers for opening and sealing
// session tickets.
const SSL_TICKET_AEAD_METHOD *ticket_aead_method = nullptr;
// legacy_ocsp_callback implements an OCSP-related callback for OpenSSL
// compatibility.
int (*legacy_ocsp_callback)(SSL *ssl, void *arg) = nullptr;
void *legacy_ocsp_callback_arg = nullptr;
// verify_sigalgs, if not empty, is the set of signature algorithms
// accepted from the peer in decreasing order of preference.
bssl::Array<uint16_t> verify_sigalgs;
// retain_only_sha256_of_client_certs is true if we should compute the SHA256
// hash of the peer's certificate and then discard it to save memory and
// session space. Only effective on the server side.
bool retain_only_sha256_of_client_certs : 1;
// quiet_shutdown is true if the connection should not send a close_notify on
// shutdown.
bool quiet_shutdown : 1;
// ocsp_stapling_enabled is only used by client connections and indicates
// whether OCSP stapling will be requested.
bool ocsp_stapling_enabled : 1;
// If true, a client will request certificate timestamps.
bool signed_cert_timestamps_enabled : 1;
// channel_id_enabled is whether Channel ID is enabled. For a server, means
// that we'll accept Channel IDs from clients. For a client, means that we'll
// advertise support.
bool channel_id_enabled : 1;
// grease_enabled is whether draft-davidben-tls-grease-01 is enabled.
bool grease_enabled : 1;
// allow_unknown_alpn_protos is whether the client allows unsolicited ALPN
// protocols from the peer.
bool allow_unknown_alpn_protos : 1;
// ed25519_enabled is whether Ed25519 is advertised in the handshake.
bool ed25519_enabled : 1;
// rsa_pss_rsae_certs_enabled is whether rsa_pss_rsae_* are supported by the
// certificate verifier.
bool rsa_pss_rsae_certs_enabled : 1;
// false_start_allowed_without_alpn is whether False Start (if
// |SSL_MODE_ENABLE_FALSE_START| is enabled) is allowed without ALPN.
bool false_start_allowed_without_alpn : 1;
// ignore_tls13_downgrade is whether a connection should continue when the
// server random signals a downgrade.
bool ignore_tls13_downgrade:1;
// handoff indicates that a server should stop after receiving the
// ClientHello and pause the handshake in such a way that |SSL_get_error|
// returns |SSL_HANDOFF|.
bool handoff : 1;
// If enable_early_data is true, early data can be sent and accepted.
bool enable_early_data : 1;
private:
~ssl_ctx_st();
friend void SSL_CTX_free(SSL_CTX *);
};
struct ssl_st {
explicit ssl_st(SSL_CTX *ctx_arg);
ssl_st(const ssl_st &) = delete;
ssl_st &operator=(const ssl_st &) = delete;
~ssl_st();
// method is the method table corresponding to the current protocol (DTLS or
// TLS).
const bssl::SSL_PROTOCOL_METHOD *method = nullptr;
Implement draft-ietf-tls-curve25519-01 in C. The new curve is not enabled by default. As EC_GROUP/EC_POINT is a bit too complex for X25519, this introduces an SSL_ECDH_METHOD abstraction which wraps just the raw ECDH operation. It also tidies up some of the curve code which kept converting back and force between NIDs and curve IDs. Now everything transits as curve IDs except for API entry points (SSL_set1_curves) which take NIDs. Those convert immediately and act on curve IDs from then on. Note that, like the Go implementation, this slightly tweaks the order of operations. The client sees the server public key before sending its own. To keep the abstraction simple, SSL_ECDH_METHOD expects to generate a keypair before consuming the peer's public key. Instead, the client handshake stashes the serialized peer public value and defers parsing it until it comes time to send ClientKeyExchange. (This is analogous to what it was doing before where it stashed the parsed peer public value instead.) It still uses TLS 1.2 terminology everywhere, but this abstraction should also be compatible with TLS 1.3 which unifies (EC)DH-style key exchanges. (Accordingly, this abstraction intentionally does not handle parsing the ClientKeyExchange/ServerKeyExchange framing or attempt to handle asynchronous plain RSA or the authentication bits.) BUG=571231 Change-Id: Iba09dddee5bcdfeb2b70185308e8ab0632717932 Reviewed-on: https://boringssl-review.googlesource.com/6780 Reviewed-by: Adam Langley <agl@google.com>
2015-12-19 05:18:25 +00:00
// config is a container for handshake configuration. Accesses to this field
// should check for nullptr, since configuration may be shed after the
// handshake completes. (If you have the |SSL_HANDSHAKE| object at hand, use
// that instead, and skip the null check.)
bssl::UniquePtr<bssl::SSL_CONFIG> config;
// version is the protocol version.
uint16_t version = 0;
uint16_t max_send_fragment = 0;
// There are 2 BIO's even though they are normally both the same. This is so
// data can be read and written to different handlers
bssl::UniquePtr<BIO> rbio; // used by SSL_read
bssl::UniquePtr<BIO> wbio; // used by SSL_write
// do_handshake runs the handshake. On completion, it returns |ssl_hs_ok|.
// Otherwise, it returns a value corresponding to what operation is needed to
// progress.
bssl::ssl_hs_wait_t (*do_handshake)(bssl::SSL_HANDSHAKE *hs) = nullptr;
bssl::SSL3_STATE *s3 = nullptr; // TLS variables
bssl::DTLS1_STATE *d1 = nullptr; // DTLS variables
// callback that allows applications to peek at protocol messages
void (*msg_callback)(int write_p, int version, int content_type,
const void *buf, size_t len, SSL *ssl,
void *arg) = nullptr;
void *msg_callback_arg = nullptr;
// session info
// initial_timeout_duration_ms is the default DTLS timeout duration in
// milliseconds. It's used to initialize the timer any time it's restarted.
//
// RFC 6347 states that implementations SHOULD use an initial timer value of 1
// second.
unsigned initial_timeout_duration_ms = 1000;
// session is the configured session to be offered by the client. This session
// is immutable.
bssl::UniquePtr<SSL_SESSION> session;
void (*info_callback)(const SSL *ssl, int type, int value) = nullptr;
bssl::UniquePtr<SSL_CTX> ctx;
// session_ctx is the |SSL_CTX| used for the session cache and related
// settings.
bssl::UniquePtr<SSL_CTX> session_ctx;
// extra application data
CRYPTO_EX_DATA ex_data;
uint32_t options = 0; // protocol behaviour
uint32_t mode = 0; // API behaviour
uint32_t max_cert_list = 0;
bssl::UniquePtr<char> hostname;
// quic_method is the method table corresponding to the QUIC hooks.
const SSL_QUIC_METHOD *quic_method = nullptr;
// renegotiate_mode controls how peer renegotiation attempts are handled.
ssl_renegotiate_mode_t renegotiate_mode = ssl_renegotiate_never;
// server is true iff the this SSL* is the server half. Note: before the SSL*
// is initialized by either SSL_set_accept_state or SSL_set_connect_state,
// the side is not determined. In this state, server is always false.
bool server : 1;
// quiet_shutdown is true if the connection should not send a close_notify on
// shutdown.
bool quiet_shutdown : 1;
// If enable_early_data is true, early data can be sent and accepted.
bool enable_early_data : 1;
};
struct ssl_session_st {
explicit ssl_session_st(const bssl::SSL_X509_METHOD *method);
ssl_session_st(const ssl_session_st &) = delete;
ssl_session_st &operator=(const ssl_session_st &) = delete;
CRYPTO_refcount_t references = 1;
// ssl_version is the (D)TLS version that established the session.
uint16_t ssl_version = 0;
// group_id is the ID of the ECDH group used to establish this session or zero
// if not applicable or unknown.
uint16_t group_id = 0;
// peer_signature_algorithm is the signature algorithm used to authenticate
// the peer, or zero if not applicable or unknown.
uint16_t peer_signature_algorithm = 0;
// master_key, in TLS 1.2 and below, is the master secret associated with the
// session. In TLS 1.3 and up, it is the resumption secret.
int master_key_length = 0;
uint8_t master_key[SSL_MAX_MASTER_KEY_LENGTH] = {0};
// session_id - valid?
unsigned session_id_length = 0;
uint8_t session_id[SSL_MAX_SSL_SESSION_ID_LENGTH] = {0};
// this is used to determine whether the session is being reused in
// the appropriate context. It is up to the application to set this,
// via SSL_new
uint8_t sid_ctx_length = 0;
uint8_t sid_ctx[SSL_MAX_SID_CTX_LENGTH] = {0};
bssl::UniquePtr<char> psk_identity;
// certs contains the certificate chain from the peer, starting with the leaf
// certificate.
bssl::UniquePtr<STACK_OF(CRYPTO_BUFFER)> certs;
const bssl::SSL_X509_METHOD *x509_method = nullptr;
// x509_peer is the peer's certificate.
X509 *x509_peer = nullptr;
// x509_chain is the certificate chain sent by the peer. NOTE: for historical
// reasons, when a client (so the peer is a server), the chain includes
// |peer|, but when a server it does not.
STACK_OF(X509) *x509_chain = nullptr;
// x509_chain_without_leaf is a lazily constructed copy of |x509_chain| that
// omits the leaf certificate. This exists because OpenSSL, historically,
// didn't include the leaf certificate in the chain for a server, but did for
// a client. The |x509_chain| always includes it and, if an API call requires
// a chain without, it is stored here.
STACK_OF(X509) *x509_chain_without_leaf = nullptr;
// verify_result is the result of certificate verification in the case of
// non-fatal certificate errors.
long verify_result = X509_V_ERR_INVALID_CALL;
// timeout is the lifetime of the session in seconds, measured from |time|.
// This is renewable up to |auth_timeout|.
uint32_t timeout = SSL_DEFAULT_SESSION_TIMEOUT;
// auth_timeout is the non-renewable lifetime of the session in seconds,
// measured from |time|.
uint32_t auth_timeout = SSL_DEFAULT_SESSION_TIMEOUT;
// time is the time the session was issued, measured in seconds from the UNIX
// epoch.
uint64_t time = 0;
const SSL_CIPHER *cipher = nullptr;
CRYPTO_EX_DATA ex_data; // application specific data
// These are used to make removal of session-ids more efficient and to
// implement a maximum cache size.
SSL_SESSION *prev = nullptr, *next = nullptr;
bssl::Array<uint8_t> ticket;
bssl::UniquePtr<CRYPTO_BUFFER> signed_cert_timestamp_list;
// The OCSP response that came with the session.
bssl::UniquePtr<CRYPTO_BUFFER> ocsp_response;
// peer_sha256 contains the SHA-256 hash of the peer's certificate if
// |peer_sha256_valid| is true.
uint8_t peer_sha256[SHA256_DIGEST_LENGTH] = {0};
// original_handshake_hash contains the handshake hash (either SHA-1+MD5 or
// SHA-2, depending on TLS version) for the original, full handshake that
// created a session. This is used by Channel IDs during resumption.
uint8_t original_handshake_hash[EVP_MAX_MD_SIZE] = {0};
uint8_t original_handshake_hash_len = 0;
uint32_t ticket_lifetime_hint = 0; // Session lifetime hint in seconds
uint32_t ticket_age_add = 0;
// ticket_max_early_data is the maximum amount of data allowed to be sent as
// early data. If zero, 0-RTT is disallowed.
uint32_t ticket_max_early_data = 0;
// early_alpn is the ALPN protocol from the initial handshake. This is only
// stored for TLS 1.3 and above in order to enforce ALPN matching for 0-RTT
// resumptions.
bssl::Array<uint8_t> early_alpn;
// extended_master_secret is whether the master secret in this session was
// generated using EMS and thus isn't vulnerable to the Triple Handshake
// attack.
bool extended_master_secret : 1;
// peer_sha256_valid is whether |peer_sha256| is valid.
bool peer_sha256_valid : 1; // Non-zero if peer_sha256 is valid
// not_resumable is used to indicate that session resumption is disallowed.
bool not_resumable : 1;
// ticket_age_add_valid is whether |ticket_age_add| is valid.
bool ticket_age_add_valid : 1;
// is_server is whether this session was created by a server.
bool is_server : 1;
private:
~ssl_session_st();
friend void SSL_SESSION_free(SSL_SESSION *);
};
#endif // OPENSSL_HEADER_SSL_INTERNAL_H