openssl/ssl/statem/statem_srvr.c

4276 lines
140 KiB
C
Raw Normal View History

/*
* Copyright 1995-2019 The OpenSSL Project Authors. All Rights Reserved.
* Copyright (c) 2002, Oracle and/or its affiliates. All rights reserved
* Copyright 2005 Nokia. All rights reserved.
*
* Licensed under the OpenSSL license (the "License"). You may not use
* this file except in compliance with the License. You can obtain a copy
* in the file LICENSE in the source distribution or at
* https://www.openssl.org/source/license.html
*/
#include <stdio.h>
#include "../ssl_local.h"
#include "statem_local.h"
#include "internal/constant_time.h"
#include "internal/cryptlib.h"
#include <openssl/buffer.h>
#include <openssl/rand.h>
#include <openssl/objects.h>
#include <openssl/evp.h>
#include <openssl/hmac.h>
#include <openssl/x509.h>
#include <openssl/dh.h>
#include <openssl/bn.h>
#include <openssl/md5.h>
#include <openssl/asn1t.h>
#define TICKET_NONCE_SIZE 8
typedef struct {
ASN1_TYPE *kxBlob;
ASN1_TYPE *opaqueBlob;
} GOST_KX_MESSAGE;
DECLARE_ASN1_FUNCTIONS(GOST_KX_MESSAGE)
ASN1_SEQUENCE(GOST_KX_MESSAGE) = {
ASN1_SIMPLE(GOST_KX_MESSAGE, kxBlob, ASN1_ANY),
ASN1_OPT(GOST_KX_MESSAGE, opaqueBlob, ASN1_ANY),
} ASN1_SEQUENCE_END(GOST_KX_MESSAGE)
IMPLEMENT_ASN1_FUNCTIONS(GOST_KX_MESSAGE)
static int tls_construct_encrypted_extensions(SSL *s, WPACKET *pkt);
/*
* ossl_statem_server13_read_transition() encapsulates the logic for the allowed
* handshake state transitions when a TLSv1.3 server is reading messages from
* the client. The message type that the client has sent is provided in |mt|.
* The current state is in |s->statem.hand_state|.
*
* Return values are 1 for success (transition allowed) and 0 on error
* (transition not allowed)
*/
static int ossl_statem_server13_read_transition(SSL *s, int mt)
{
OSSL_STATEM *st = &s->statem;
/*
* Note: There is no case for TLS_ST_BEFORE because at that stage we have
* not negotiated TLSv1.3 yet, so that case is handled by
* ossl_statem_server_read_transition()
*/
switch (st->hand_state) {
default:
break;
case TLS_ST_EARLY_DATA:
if (s->hello_retry_request == SSL_HRR_PENDING) {
if (mt == SSL3_MT_CLIENT_HELLO) {
st->hand_state = TLS_ST_SR_CLNT_HELLO;
return 1;
}
break;
} else if (s->ext.early_data == SSL_EARLY_DATA_ACCEPTED) {
if (mt == SSL3_MT_END_OF_EARLY_DATA) {
st->hand_state = TLS_ST_SR_END_OF_EARLY_DATA;
return 1;
}
break;
}
/* Fall through */
case TLS_ST_SR_END_OF_EARLY_DATA:
case TLS_ST_SW_FINISHED:
if (s->s3->tmp.cert_request) {
if (mt == SSL3_MT_CERTIFICATE) {
st->hand_state = TLS_ST_SR_CERT;
return 1;
}
} else {
if (mt == SSL3_MT_FINISHED) {
st->hand_state = TLS_ST_SR_FINISHED;
return 1;
}
}
break;
case TLS_ST_SR_CERT:
if (s->session->peer == NULL) {
if (mt == SSL3_MT_FINISHED) {
st->hand_state = TLS_ST_SR_FINISHED;
return 1;
}
} else {
if (mt == SSL3_MT_CERTIFICATE_VERIFY) {
st->hand_state = TLS_ST_SR_CERT_VRFY;
return 1;
}
}
break;
case TLS_ST_SR_CERT_VRFY:
if (mt == SSL3_MT_FINISHED) {
st->hand_state = TLS_ST_SR_FINISHED;
return 1;
}
break;
case TLS_ST_OK:
/*
* Its never ok to start processing handshake messages in the middle of
* early data (i.e. before we've received the end of early data alert)
*/
if (s->early_data_state == SSL_EARLY_DATA_READING)
break;
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
if (mt == SSL3_MT_CERTIFICATE
&& s->post_handshake_auth == SSL_PHA_REQUESTED) {
st->hand_state = TLS_ST_SR_CERT;
return 1;
}
if (mt == SSL3_MT_KEY_UPDATE) {
st->hand_state = TLS_ST_SR_KEY_UPDATE;
return 1;
}
break;
}
/* No valid transition found */
return 0;
}
/*
* ossl_statem_server_read_transition() encapsulates the logic for the allowed
* handshake state transitions when the server is reading messages from the
* client. The message type that the client has sent is provided in |mt|. The
* current state is in |s->statem.hand_state|.
*
* Return values are 1 for success (transition allowed) and 0 on error
* (transition not allowed)
*/
int ossl_statem_server_read_transition(SSL *s, int mt)
{
OSSL_STATEM *st = &s->statem;
if (SSL_IS_TLS13(s)) {
if (!ossl_statem_server13_read_transition(s, mt))
goto err;
return 1;
}
switch (st->hand_state) {
default:
break;
case TLS_ST_BEFORE:
case TLS_ST_OK:
case DTLS_ST_SW_HELLO_VERIFY_REQUEST:
if (mt == SSL3_MT_CLIENT_HELLO) {
st->hand_state = TLS_ST_SR_CLNT_HELLO;
return 1;
}
break;
case TLS_ST_SW_SRVR_DONE:
/*
* If we get a CKE message after a ServerDone then either
* 1) We didn't request a Certificate
* OR
* 2) If we did request one then
* a) We allow no Certificate to be returned
* AND
* b) We are running SSL3 (in TLS1.0+ the client must return a 0
* list if we requested a certificate)
*/
if (mt == SSL3_MT_CLIENT_KEY_EXCHANGE) {
if (s->s3->tmp.cert_request) {
if (s->version == SSL3_VERSION) {
if ((s->verify_mode & SSL_VERIFY_PEER)
&& (s->verify_mode & SSL_VERIFY_FAIL_IF_NO_PEER_CERT)) {
/*
* This isn't an unexpected message as such - we're just
* not going to accept it because we require a client
* cert.
*/
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_OSSL_STATEM_SERVER_READ_TRANSITION,
SSL_R_PEER_DID_NOT_RETURN_A_CERTIFICATE);
return 0;
}
st->hand_state = TLS_ST_SR_KEY_EXCH;
return 1;
}
} else {
st->hand_state = TLS_ST_SR_KEY_EXCH;
return 1;
}
} else if (s->s3->tmp.cert_request) {
if (mt == SSL3_MT_CERTIFICATE) {
st->hand_state = TLS_ST_SR_CERT;
return 1;
}
}
break;
case TLS_ST_SR_CERT:
if (mt == SSL3_MT_CLIENT_KEY_EXCHANGE) {
st->hand_state = TLS_ST_SR_KEY_EXCH;
return 1;
}
break;
case TLS_ST_SR_KEY_EXCH:
/*
* We should only process a CertificateVerify message if we have
* received a Certificate from the client. If so then |s->session->peer|
* will be non NULL. In some instances a CertificateVerify message is
* not required even if the peer has sent a Certificate (e.g. such as in
* the case of static DH). In that case |st->no_cert_verify| should be
* set.
*/
if (s->session->peer == NULL || st->no_cert_verify) {
if (mt == SSL3_MT_CHANGE_CIPHER_SPEC) {
/*
* For the ECDH ciphersuites when the client sends its ECDH
* pub key in a certificate, the CertificateVerify message is
* not sent. Also for GOST ciphersuites when the client uses
* its key from the certificate for key exchange.
*/
st->hand_state = TLS_ST_SR_CHANGE;
return 1;
}
} else {
if (mt == SSL3_MT_CERTIFICATE_VERIFY) {
st->hand_state = TLS_ST_SR_CERT_VRFY;
return 1;
}
}
break;
case TLS_ST_SR_CERT_VRFY:
if (mt == SSL3_MT_CHANGE_CIPHER_SPEC) {
st->hand_state = TLS_ST_SR_CHANGE;
return 1;
}
break;
case TLS_ST_SR_CHANGE:
#ifndef OPENSSL_NO_NEXTPROTONEG
if (s->s3->npn_seen) {
if (mt == SSL3_MT_NEXT_PROTO) {
st->hand_state = TLS_ST_SR_NEXT_PROTO;
return 1;
}
} else {
#endif
if (mt == SSL3_MT_FINISHED) {
st->hand_state = TLS_ST_SR_FINISHED;
return 1;
}
#ifndef OPENSSL_NO_NEXTPROTONEG
}
#endif
break;
#ifndef OPENSSL_NO_NEXTPROTONEG
case TLS_ST_SR_NEXT_PROTO:
if (mt == SSL3_MT_FINISHED) {
st->hand_state = TLS_ST_SR_FINISHED;
return 1;
}
break;
#endif
case TLS_ST_SW_FINISHED:
if (mt == SSL3_MT_CHANGE_CIPHER_SPEC) {
st->hand_state = TLS_ST_SR_CHANGE;
return 1;
}
break;
}
err:
/* No valid transition found */
if (SSL_IS_DTLS(s) && mt == SSL3_MT_CHANGE_CIPHER_SPEC) {
BIO *rbio;
/*
* CCS messages don't have a message sequence number so this is probably
* because of an out-of-order CCS. We'll just drop it.
*/
s->init_num = 0;
s->rwstate = SSL_READING;
rbio = SSL_get_rbio(s);
BIO_clear_retry_flags(rbio);
BIO_set_retry_read(rbio);
return 0;
}
SSLfatal(s, SSL3_AD_UNEXPECTED_MESSAGE,
SSL_F_OSSL_STATEM_SERVER_READ_TRANSITION,
SSL_R_UNEXPECTED_MESSAGE);
return 0;
}
/*
* Should we send a ServerKeyExchange message?
*
* Valid return values are:
* 1: Yes
* 0: No
*/
static int send_server_key_exchange(SSL *s)
{
unsigned long alg_k = s->s3->tmp.new_cipher->algorithm_mkey;
/*
* only send a ServerKeyExchange if DH or fortezza but we have a
* sign only certificate PSK: may send PSK identity hints For
* ECC ciphersuites, we send a serverKeyExchange message only if
* the cipher suite is either ECDH-anon or ECDHE. In other cases,
* the server certificate contains the server's public key for
* key exchange.
*/
if (alg_k & (SSL_kDHE | SSL_kECDHE)
/*
* PSK: send ServerKeyExchange if PSK identity hint if
* provided
*/
#ifndef OPENSSL_NO_PSK
/* Only send SKE if we have identity hint for plain PSK */
|| ((alg_k & (SSL_kPSK | SSL_kRSAPSK))
&& s->cert->psk_identity_hint)
/* For other PSK always send SKE */
|| (alg_k & (SSL_PSK & (SSL_kDHEPSK | SSL_kECDHEPSK)))
#endif
#ifndef OPENSSL_NO_SRP
/* SRP: send ServerKeyExchange */
|| (alg_k & SSL_kSRP)
#endif
) {
return 1;
}
return 0;
}
/*
* Should we send a CertificateRequest message?
*
* Valid return values are:
* 1: Yes
* 0: No
*/
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
int send_certificate_request(SSL *s)
{
if (
/* don't request cert unless asked for it: */
s->verify_mode & SSL_VERIFY_PEER
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
/*
* don't request if post-handshake-only unless doing
* post-handshake in TLSv1.3:
*/
&& (!SSL_IS_TLS13(s) || !(s->verify_mode & SSL_VERIFY_POST_HANDSHAKE)
|| s->post_handshake_auth == SSL_PHA_REQUEST_PENDING)
/*
* if SSL_VERIFY_CLIENT_ONCE is set, don't request cert
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
* a second time:
*/
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
&& (s->certreqs_sent < 1 ||
!(s->verify_mode & SSL_VERIFY_CLIENT_ONCE))
/*
* never request cert in anonymous ciphersuites (see
* section "Certificate request" in SSL 3 drafts and in
* RFC 2246):
*/
&& (!(s->s3->tmp.new_cipher->algorithm_auth & SSL_aNULL)
/*
* ... except when the application insists on
* verification (against the specs, but statem_clnt.c accepts
* this for SSL 3)
*/
|| (s->verify_mode & SSL_VERIFY_FAIL_IF_NO_PEER_CERT))
/* don't request certificate for SRP auth */
&& !(s->s3->tmp.new_cipher->algorithm_auth & SSL_aSRP)
/*
* With normal PSK Certificates and Certificate Requests
* are omitted
*/
&& !(s->s3->tmp.new_cipher->algorithm_auth & SSL_aPSK)) {
return 1;
}
return 0;
}
/*
* ossl_statem_server13_write_transition() works out what handshake state to
* move to next when a TLSv1.3 server is writing messages to be sent to the
* client.
*/
static WRITE_TRAN ossl_statem_server13_write_transition(SSL *s)
{
OSSL_STATEM *st = &s->statem;
/*
* No case for TLS_ST_BEFORE, because at that stage we have not negotiated
* TLSv1.3 yet, so that is handled by ossl_statem_server_write_transition()
*/
switch (st->hand_state) {
default:
/* Shouldn't happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_OSSL_STATEM_SERVER13_WRITE_TRANSITION,
ERR_R_INTERNAL_ERROR);
return WRITE_TRAN_ERROR;
case TLS_ST_OK:
if (s->key_update != SSL_KEY_UPDATE_NONE) {
st->hand_state = TLS_ST_SW_KEY_UPDATE;
return WRITE_TRAN_CONTINUE;
}
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
if (s->post_handshake_auth == SSL_PHA_REQUEST_PENDING) {
st->hand_state = TLS_ST_SW_CERT_REQ;
return WRITE_TRAN_CONTINUE;
}
/* Try to read from the client instead */
return WRITE_TRAN_FINISHED;
case TLS_ST_SR_CLNT_HELLO:
st->hand_state = TLS_ST_SW_SRVR_HELLO;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_SRVR_HELLO:
if ((s->options & SSL_OP_ENABLE_MIDDLEBOX_COMPAT) != 0
&& s->hello_retry_request != SSL_HRR_COMPLETE)
st->hand_state = TLS_ST_SW_CHANGE;
else if (s->hello_retry_request == SSL_HRR_PENDING)
st->hand_state = TLS_ST_EARLY_DATA;
else
st->hand_state = TLS_ST_SW_ENCRYPTED_EXTENSIONS;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_CHANGE:
if (s->hello_retry_request == SSL_HRR_PENDING)
st->hand_state = TLS_ST_EARLY_DATA;
else
st->hand_state = TLS_ST_SW_ENCRYPTED_EXTENSIONS;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_ENCRYPTED_EXTENSIONS:
if (s->hit)
st->hand_state = TLS_ST_SW_FINISHED;
else if (send_certificate_request(s))
st->hand_state = TLS_ST_SW_CERT_REQ;
else
st->hand_state = TLS_ST_SW_CERT;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_CERT_REQ:
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
if (s->post_handshake_auth == SSL_PHA_REQUEST_PENDING) {
s->post_handshake_auth = SSL_PHA_REQUESTED;
st->hand_state = TLS_ST_OK;
} else {
st->hand_state = TLS_ST_SW_CERT;
}
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_CERT:
st->hand_state = TLS_ST_SW_CERT_VRFY;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_CERT_VRFY:
st->hand_state = TLS_ST_SW_FINISHED;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_FINISHED:
st->hand_state = TLS_ST_EARLY_DATA;
return WRITE_TRAN_CONTINUE;
case TLS_ST_EARLY_DATA:
return WRITE_TRAN_FINISHED;
case TLS_ST_SR_FINISHED:
/*
* Technically we have finished the handshake at this point, but we're
* going to remain "in_init" for now and write out any session tickets
* immediately.
*/
if (s->post_handshake_auth == SSL_PHA_REQUESTED) {
s->post_handshake_auth = SSL_PHA_EXT_RECEIVED;
} else if (!s->ext.ticket_expected) {
/*
* If we're not going to renew the ticket then we just finish the
* handshake at this point.
*/
st->hand_state = TLS_ST_OK;
return WRITE_TRAN_CONTINUE;
}
if (s->num_tickets > s->sent_tickets)
st->hand_state = TLS_ST_SW_SESSION_TICKET;
else
st->hand_state = TLS_ST_OK;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SR_KEY_UPDATE:
case TLS_ST_SW_KEY_UPDATE:
st->hand_state = TLS_ST_OK;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_SESSION_TICKET:
/* In a resumption we only ever send a maximum of one new ticket.
* Following an initial handshake we send the number of tickets we have
* been configured for.
*/
if (s->hit || s->num_tickets <= s->sent_tickets) {
/* We've written enough tickets out. */
st->hand_state = TLS_ST_OK;
}
return WRITE_TRAN_CONTINUE;
}
}
/*
* ossl_statem_server_write_transition() works out what handshake state to move
* to next when the server is writing messages to be sent to the client.
*/
WRITE_TRAN ossl_statem_server_write_transition(SSL *s)
{
OSSL_STATEM *st = &s->statem;
/*
* Note that before the ClientHello we don't know what version we are going
* to negotiate yet, so we don't take this branch until later
*/
if (SSL_IS_TLS13(s))
return ossl_statem_server13_write_transition(s);
switch (st->hand_state) {
default:
/* Shouldn't happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_OSSL_STATEM_SERVER_WRITE_TRANSITION,
ERR_R_INTERNAL_ERROR);
return WRITE_TRAN_ERROR;
case TLS_ST_OK:
if (st->request_state == TLS_ST_SW_HELLO_REQ) {
/* We must be trying to renegotiate */
st->hand_state = TLS_ST_SW_HELLO_REQ;
st->request_state = TLS_ST_BEFORE;
return WRITE_TRAN_CONTINUE;
}
/* Must be an incoming ClientHello */
if (!tls_setup_handshake(s)) {
/* SSLfatal() already called */
return WRITE_TRAN_ERROR;
}
/* Fall through */
case TLS_ST_BEFORE:
/* Just go straight to trying to read from the client */
return WRITE_TRAN_FINISHED;
case TLS_ST_SW_HELLO_REQ:
st->hand_state = TLS_ST_OK;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SR_CLNT_HELLO:
if (SSL_IS_DTLS(s) && !s->d1->cookie_verified
&& (SSL_get_options(s) & SSL_OP_COOKIE_EXCHANGE)) {
st->hand_state = DTLS_ST_SW_HELLO_VERIFY_REQUEST;
} else if (s->renegotiate == 0 && !SSL_IS_FIRST_HANDSHAKE(s)) {
/* We must have rejected the renegotiation */
st->hand_state = TLS_ST_OK;
return WRITE_TRAN_CONTINUE;
} else {
st->hand_state = TLS_ST_SW_SRVR_HELLO;
}
return WRITE_TRAN_CONTINUE;
case DTLS_ST_SW_HELLO_VERIFY_REQUEST:
return WRITE_TRAN_FINISHED;
case TLS_ST_SW_SRVR_HELLO:
if (s->hit) {
if (s->ext.ticket_expected)
st->hand_state = TLS_ST_SW_SESSION_TICKET;
else
st->hand_state = TLS_ST_SW_CHANGE;
} else {
/* Check if it is anon DH or anon ECDH, */
/* normal PSK or SRP */
if (!(s->s3->tmp.new_cipher->algorithm_auth &
(SSL_aNULL | SSL_aSRP | SSL_aPSK))) {
st->hand_state = TLS_ST_SW_CERT;
} else if (send_server_key_exchange(s)) {
st->hand_state = TLS_ST_SW_KEY_EXCH;
} else if (send_certificate_request(s)) {
st->hand_state = TLS_ST_SW_CERT_REQ;
} else {
st->hand_state = TLS_ST_SW_SRVR_DONE;
}
}
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_CERT:
if (s->ext.status_expected) {
st->hand_state = TLS_ST_SW_CERT_STATUS;
return WRITE_TRAN_CONTINUE;
}
/* Fall through */
case TLS_ST_SW_CERT_STATUS:
if (send_server_key_exchange(s)) {
st->hand_state = TLS_ST_SW_KEY_EXCH;
return WRITE_TRAN_CONTINUE;
}
/* Fall through */
case TLS_ST_SW_KEY_EXCH:
if (send_certificate_request(s)) {
st->hand_state = TLS_ST_SW_CERT_REQ;
return WRITE_TRAN_CONTINUE;
}
/* Fall through */
case TLS_ST_SW_CERT_REQ:
st->hand_state = TLS_ST_SW_SRVR_DONE;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_SRVR_DONE:
return WRITE_TRAN_FINISHED;
case TLS_ST_SR_FINISHED:
if (s->hit) {
st->hand_state = TLS_ST_OK;
return WRITE_TRAN_CONTINUE;
} else if (s->ext.ticket_expected) {
st->hand_state = TLS_ST_SW_SESSION_TICKET;
} else {
st->hand_state = TLS_ST_SW_CHANGE;
}
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_SESSION_TICKET:
st->hand_state = TLS_ST_SW_CHANGE;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_CHANGE:
st->hand_state = TLS_ST_SW_FINISHED;
return WRITE_TRAN_CONTINUE;
case TLS_ST_SW_FINISHED:
if (s->hit) {
return WRITE_TRAN_FINISHED;
}
st->hand_state = TLS_ST_OK;
return WRITE_TRAN_CONTINUE;
}
}
/*
* Perform any pre work that needs to be done prior to sending a message from
* the server to the client.
*/
WORK_STATE ossl_statem_server_pre_work(SSL *s, WORK_STATE wst)
{
OSSL_STATEM *st = &s->statem;
switch (st->hand_state) {
default:
/* No pre work to be done */
break;
case TLS_ST_SW_HELLO_REQ:
s->shutdown = 0;
if (SSL_IS_DTLS(s))
Fix DTLS buffered message DoS attack DTLS can handle out of order record delivery. Additionally since handshake messages can be bigger than will fit into a single packet, the messages can be fragmented across multiple records (as with normal TLS). That means that the messages can arrive mixed up, and we have to reassemble them. We keep a queue of buffered messages that are "from the future", i.e. messages we're not ready to deal with yet but have arrived early. The messages held there may not be full yet - they could be one or more fragments that are still in the process of being reassembled. The code assumes that we will eventually complete the reassembly and when that occurs the complete message is removed from the queue at the point that we need to use it. However, DTLS is also tolerant of packet loss. To get around that DTLS messages can be retransmitted. If we receive a full (non-fragmented) message from the peer after previously having received a fragment of that message, then we ignore the message in the queue and just use the non-fragmented version. At that point the queued message will never get removed. Additionally the peer could send "future" messages that we never get to in order to complete the handshake. Each message has a sequence number (starting from 0). We will accept a message fragment for the current message sequence number, or for any sequence up to 10 into the future. However if the Finished message has a sequence number of 2, anything greater than that in the queue is just left there. So, in those two ways we can end up with "orphaned" data in the queue that will never get removed - except when the connection is closed. At that point all the queues are flushed. An attacker could seek to exploit this by filling up the queues with lots of large messages that are never going to be used in order to attempt a DoS by memory exhaustion. I will assume that we are only concerned with servers here. It does not seem reasonable to be concerned about a memory exhaustion attack on a client. They are unlikely to process enough connections for this to be an issue. A "long" handshake with many messages might be 5 messages long (in the incoming direction), e.g. ClientHello, Certificate, ClientKeyExchange, CertificateVerify, Finished. So this would be message sequence numbers 0 to 4. Additionally we can buffer up to 10 messages in the future. Therefore the maximum number of messages that an attacker could send that could get orphaned would typically be 15. The maximum size that a DTLS message is allowed to be is defined by max_cert_list, which by default is 100k. Therefore the maximum amount of "orphaned" memory per connection is 1500k. Message sequence numbers get reset after the Finished message, so renegotiation will not extend the maximum number of messages that can be orphaned per connection. As noted above, the queues do get cleared when the connection is closed. Therefore in order to mount an effective attack, an attacker would have to open many simultaneous connections. Issue reported by Quan Luo. CVE-2016-2179 Reviewed-by: Richard Levitte <levitte@openssl.org>
2016-06-30 12:17:08 +00:00
dtls1_clear_sent_buffer(s);
break;
case DTLS_ST_SW_HELLO_VERIFY_REQUEST:
s->shutdown = 0;
if (SSL_IS_DTLS(s)) {
Fix DTLS buffered message DoS attack DTLS can handle out of order record delivery. Additionally since handshake messages can be bigger than will fit into a single packet, the messages can be fragmented across multiple records (as with normal TLS). That means that the messages can arrive mixed up, and we have to reassemble them. We keep a queue of buffered messages that are "from the future", i.e. messages we're not ready to deal with yet but have arrived early. The messages held there may not be full yet - they could be one or more fragments that are still in the process of being reassembled. The code assumes that we will eventually complete the reassembly and when that occurs the complete message is removed from the queue at the point that we need to use it. However, DTLS is also tolerant of packet loss. To get around that DTLS messages can be retransmitted. If we receive a full (non-fragmented) message from the peer after previously having received a fragment of that message, then we ignore the message in the queue and just use the non-fragmented version. At that point the queued message will never get removed. Additionally the peer could send "future" messages that we never get to in order to complete the handshake. Each message has a sequence number (starting from 0). We will accept a message fragment for the current message sequence number, or for any sequence up to 10 into the future. However if the Finished message has a sequence number of 2, anything greater than that in the queue is just left there. So, in those two ways we can end up with "orphaned" data in the queue that will never get removed - except when the connection is closed. At that point all the queues are flushed. An attacker could seek to exploit this by filling up the queues with lots of large messages that are never going to be used in order to attempt a DoS by memory exhaustion. I will assume that we are only concerned with servers here. It does not seem reasonable to be concerned about a memory exhaustion attack on a client. They are unlikely to process enough connections for this to be an issue. A "long" handshake with many messages might be 5 messages long (in the incoming direction), e.g. ClientHello, Certificate, ClientKeyExchange, CertificateVerify, Finished. So this would be message sequence numbers 0 to 4. Additionally we can buffer up to 10 messages in the future. Therefore the maximum number of messages that an attacker could send that could get orphaned would typically be 15. The maximum size that a DTLS message is allowed to be is defined by max_cert_list, which by default is 100k. Therefore the maximum amount of "orphaned" memory per connection is 1500k. Message sequence numbers get reset after the Finished message, so renegotiation will not extend the maximum number of messages that can be orphaned per connection. As noted above, the queues do get cleared when the connection is closed. Therefore in order to mount an effective attack, an attacker would have to open many simultaneous connections. Issue reported by Quan Luo. CVE-2016-2179 Reviewed-by: Richard Levitte <levitte@openssl.org>
2016-06-30 12:17:08 +00:00
dtls1_clear_sent_buffer(s);
/* We don't buffer this message so don't use the timer */
st->use_timer = 0;
}
break;
case TLS_ST_SW_SRVR_HELLO:
if (SSL_IS_DTLS(s)) {
/*
* Messages we write from now on should be buffered and
* retransmitted if necessary, so we need to use the timer now
*/
st->use_timer = 1;
}
break;
case TLS_ST_SW_SRVR_DONE:
#ifndef OPENSSL_NO_SCTP
if (SSL_IS_DTLS(s) && BIO_dgram_is_sctp(SSL_get_wbio(s))) {
/* Calls SSLfatal() as required */
return dtls_wait_for_dry(s);
}
#endif
return WORK_FINISHED_CONTINUE;
case TLS_ST_SW_SESSION_TICKET:
if (SSL_IS_TLS13(s) && s->sent_tickets == 0) {
/*
* Actually this is the end of the handshake, but we're going
* straight into writing the session ticket out. So we finish off
* the handshake, but keep the various buffers active.
*
* Calls SSLfatal as required.
*/
return tls_finish_handshake(s, wst, 0, 0);
} if (SSL_IS_DTLS(s)) {
/*
* We're into the last flight. We don't retransmit the last flight
* unless we need to, so we don't use the timer
*/
st->use_timer = 0;
}
break;
case TLS_ST_SW_CHANGE:
if (SSL_IS_TLS13(s))
break;
s->session->cipher = s->s3->tmp.new_cipher;
if (!s->method->ssl3_enc->setup_key_block(s)) {
/* SSLfatal() already called */
return WORK_ERROR;
}
if (SSL_IS_DTLS(s)) {
/*
* We're into the last flight. We don't retransmit the last flight
* unless we need to, so we don't use the timer. This might have
* already been set to 0 if we sent a NewSessionTicket message,
* but we'll set it again here in case we didn't.
*/
st->use_timer = 0;
}
return WORK_FINISHED_CONTINUE;
case TLS_ST_EARLY_DATA:
if (s->early_data_state != SSL_EARLY_DATA_ACCEPTING
&& (s->s3->flags & TLS1_FLAGS_STATELESS) == 0)
return WORK_FINISHED_CONTINUE;
/* Fall through */
case TLS_ST_OK:
/* Calls SSLfatal() as required */
return tls_finish_handshake(s, wst, 1, 1);
}
return WORK_FINISHED_CONTINUE;
}
static ossl_inline int conn_is_closed(void)
{
switch (get_last_sys_error()) {
#if defined(EPIPE)
case EPIPE:
return 1;
#endif
#if defined(ECONNRESET)
case ECONNRESET:
return 1;
#endif
#if defined(WSAECONNRESET)
case WSAECONNRESET:
return 1;
#endif
default:
return 0;
}
}
/*
* Perform any work that needs to be done after sending a message from the
* server to the client.
*/
WORK_STATE ossl_statem_server_post_work(SSL *s, WORK_STATE wst)
{
OSSL_STATEM *st = &s->statem;
s->init_num = 0;
switch (st->hand_state) {
default:
/* No post work to be done */
break;
case TLS_ST_SW_HELLO_REQ:
if (statem_flush(s) != 1)
return WORK_MORE_A;
if (!ssl3_init_finished_mac(s)) {
/* SSLfatal() already called */
return WORK_ERROR;
}
break;
case DTLS_ST_SW_HELLO_VERIFY_REQUEST:
if (statem_flush(s) != 1)
return WORK_MORE_A;
/* HelloVerifyRequest resets Finished MAC */
if (s->version != DTLS1_BAD_VER && !ssl3_init_finished_mac(s)) {
/* SSLfatal() already called */
return WORK_ERROR;
}
/*
* The next message should be another ClientHello which we need to
* treat like it was the first packet
*/
s->first_packet = 1;
break;
case TLS_ST_SW_SRVR_HELLO:
if (SSL_IS_TLS13(s) && s->hello_retry_request == SSL_HRR_PENDING) {
if ((s->options & SSL_OP_ENABLE_MIDDLEBOX_COMPAT) == 0
&& statem_flush(s) != 1)
return WORK_MORE_A;
break;
}
#ifndef OPENSSL_NO_SCTP
if (SSL_IS_DTLS(s) && s->hit) {
unsigned char sctpauthkey[64];
char labelbuffer[sizeof(DTLS1_SCTP_AUTH_LABEL)];
size_t labellen;
/*
* Add new shared key for SCTP-Auth, will be ignored if no
* SCTP used.
*/
memcpy(labelbuffer, DTLS1_SCTP_AUTH_LABEL,
sizeof(DTLS1_SCTP_AUTH_LABEL));
/* Don't include the terminating zero. */
labellen = sizeof(labelbuffer) - 1;
if (s->mode & SSL_MODE_DTLS_SCTP_LABEL_LENGTH_BUG)
labellen += 1;
if (SSL_export_keying_material(s, sctpauthkey,
sizeof(sctpauthkey), labelbuffer,
labellen, NULL, 0,
0) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_OSSL_STATEM_SERVER_POST_WORK,
ERR_R_INTERNAL_ERROR);
return WORK_ERROR;
}
BIO_ctrl(SSL_get_wbio(s), BIO_CTRL_DGRAM_SCTP_ADD_AUTH_KEY,
sizeof(sctpauthkey), sctpauthkey);
}
#endif
if (!SSL_IS_TLS13(s)
|| ((s->options & SSL_OP_ENABLE_MIDDLEBOX_COMPAT) != 0
&& s->hello_retry_request != SSL_HRR_COMPLETE))
break;
/* Fall through */
case TLS_ST_SW_CHANGE:
if (s->hello_retry_request == SSL_HRR_PENDING) {
if (!statem_flush(s))
return WORK_MORE_A;
break;
}
if (SSL_IS_TLS13(s)) {
if (!s->method->ssl3_enc->setup_key_block(s)
|| !s->method->ssl3_enc->change_cipher_state(s,
SSL3_CC_HANDSHAKE | SSL3_CHANGE_CIPHER_SERVER_WRITE)) {
/* SSLfatal() already called */
return WORK_ERROR;
}
if (s->ext.early_data != SSL_EARLY_DATA_ACCEPTED
&& !s->method->ssl3_enc->change_cipher_state(s,
SSL3_CC_HANDSHAKE |SSL3_CHANGE_CIPHER_SERVER_READ)) {
/* SSLfatal() already called */
return WORK_ERROR;
}
/*
* We don't yet know whether the next record we are going to receive
* is an unencrypted alert, an encrypted alert, or an encrypted
* handshake message. We temporarily tolerate unencrypted alerts.
*/
s->statem.enc_read_state = ENC_READ_STATE_ALLOW_PLAIN_ALERTS;
break;
}
#ifndef OPENSSL_NO_SCTP
if (SSL_IS_DTLS(s) && !s->hit) {
/*
* Change to new shared key of SCTP-Auth, will be ignored if
* no SCTP used.
*/
BIO_ctrl(SSL_get_wbio(s), BIO_CTRL_DGRAM_SCTP_NEXT_AUTH_KEY,
0, NULL);
}
#endif
if (!s->method->ssl3_enc->change_cipher_state(s,
SSL3_CHANGE_CIPHER_SERVER_WRITE))
{
/* SSLfatal() already called */
return WORK_ERROR;
}
if (SSL_IS_DTLS(s))
dtls1_reset_seq_numbers(s, SSL3_CC_WRITE);
break;
case TLS_ST_SW_SRVR_DONE:
if (statem_flush(s) != 1)
return WORK_MORE_A;
break;
case TLS_ST_SW_FINISHED:
if (statem_flush(s) != 1)
return WORK_MORE_A;
#ifndef OPENSSL_NO_SCTP
if (SSL_IS_DTLS(s) && s->hit) {
/*
* Change to new shared key of SCTP-Auth, will be ignored if
* no SCTP used.
*/
BIO_ctrl(SSL_get_wbio(s), BIO_CTRL_DGRAM_SCTP_NEXT_AUTH_KEY,
0, NULL);
}
#endif
if (SSL_IS_TLS13(s)) {
if (!s->method->ssl3_enc->generate_master_secret(s,
s->master_secret, s->handshake_secret, 0,
&s->session->master_key_length)
|| !s->method->ssl3_enc->change_cipher_state(s,
SSL3_CC_APPLICATION | SSL3_CHANGE_CIPHER_SERVER_WRITE))
/* SSLfatal() already called */
return WORK_ERROR;
}
break;
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
case TLS_ST_SW_CERT_REQ:
if (s->post_handshake_auth == SSL_PHA_REQUEST_PENDING) {
if (statem_flush(s) != 1)
return WORK_MORE_A;
}
break;
case TLS_ST_SW_KEY_UPDATE:
if (statem_flush(s) != 1)
return WORK_MORE_A;
if (!tls13_update_key(s, 1)) {
/* SSLfatal() already called */
return WORK_ERROR;
}
break;
case TLS_ST_SW_SESSION_TICKET:
clear_sys_error();
if (SSL_IS_TLS13(s) && statem_flush(s) != 1) {
if (SSL_get_error(s, 0) == SSL_ERROR_SYSCALL
&& conn_is_closed()) {
/*
* We ignore connection closed errors in TLSv1.3 when sending a
* NewSessionTicket and behave as if we were successful. This is
* so that we are still able to read data sent to us by a client
* that closes soon after the end of the handshake without
* waiting to read our post-handshake NewSessionTickets.
*/
s->rwstate = SSL_NOTHING;
break;
}
return WORK_MORE_A;
}
break;
}
return WORK_FINISHED_CONTINUE;
}
/*
* Get the message construction function and message type for sending from the
* server
*
* Valid return values are:
* 1: Success
* 0: Error
*/
int ossl_statem_server_construct_message(SSL *s, WPACKET *pkt,
confunc_f *confunc, int *mt)
{
OSSL_STATEM *st = &s->statem;
switch (st->hand_state) {
default:
/* Shouldn't happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_OSSL_STATEM_SERVER_CONSTRUCT_MESSAGE,
SSL_R_BAD_HANDSHAKE_STATE);
return 0;
case TLS_ST_SW_CHANGE:
if (SSL_IS_DTLS(s))
*confunc = dtls_construct_change_cipher_spec;
else
*confunc = tls_construct_change_cipher_spec;
*mt = SSL3_MT_CHANGE_CIPHER_SPEC;
break;
case DTLS_ST_SW_HELLO_VERIFY_REQUEST:
*confunc = dtls_construct_hello_verify_request;
*mt = DTLS1_MT_HELLO_VERIFY_REQUEST;
break;
case TLS_ST_SW_HELLO_REQ:
/* No construction function needed */
*confunc = NULL;
*mt = SSL3_MT_HELLO_REQUEST;
break;
case TLS_ST_SW_SRVR_HELLO:
*confunc = tls_construct_server_hello;
*mt = SSL3_MT_SERVER_HELLO;
break;
case TLS_ST_SW_CERT:
*confunc = tls_construct_server_certificate;
*mt = SSL3_MT_CERTIFICATE;
break;
case TLS_ST_SW_CERT_VRFY:
*confunc = tls_construct_cert_verify;
*mt = SSL3_MT_CERTIFICATE_VERIFY;
break;
case TLS_ST_SW_KEY_EXCH:
*confunc = tls_construct_server_key_exchange;
*mt = SSL3_MT_SERVER_KEY_EXCHANGE;
break;
case TLS_ST_SW_CERT_REQ:
*confunc = tls_construct_certificate_request;
*mt = SSL3_MT_CERTIFICATE_REQUEST;
break;
case TLS_ST_SW_SRVR_DONE:
*confunc = tls_construct_server_done;
*mt = SSL3_MT_SERVER_DONE;
break;
case TLS_ST_SW_SESSION_TICKET:
*confunc = tls_construct_new_session_ticket;
*mt = SSL3_MT_NEWSESSION_TICKET;
break;
case TLS_ST_SW_CERT_STATUS:
*confunc = tls_construct_cert_status;
*mt = SSL3_MT_CERTIFICATE_STATUS;
break;
case TLS_ST_SW_FINISHED:
*confunc = tls_construct_finished;
*mt = SSL3_MT_FINISHED;
break;
case TLS_ST_EARLY_DATA:
*confunc = NULL;
*mt = SSL3_MT_DUMMY;
break;
case TLS_ST_SW_ENCRYPTED_EXTENSIONS:
*confunc = tls_construct_encrypted_extensions;
*mt = SSL3_MT_ENCRYPTED_EXTENSIONS;
break;
case TLS_ST_SW_KEY_UPDATE:
*confunc = tls_construct_key_update;
*mt = SSL3_MT_KEY_UPDATE;
break;
}
return 1;
}
/*
* Maximum size (excluding the Handshake header) of a ClientHello message,
* calculated as follows:
*
* 2 + # client_version
* 32 + # only valid length for random
* 1 + # length of session_id
* 32 + # maximum size for session_id
* 2 + # length of cipher suites
* 2^16-2 + # maximum length of cipher suites array
* 1 + # length of compression_methods
* 2^8-1 + # maximum length of compression methods
* 2 + # length of extensions
* 2^16-1 # maximum length of extensions
*/
#define CLIENT_HELLO_MAX_LENGTH 131396
#define CLIENT_KEY_EXCH_MAX_LENGTH 2048
#define NEXT_PROTO_MAX_LENGTH 514
/*
* Returns the maximum allowed length for the current message that we are
* reading. Excludes the message header.
*/
size_t ossl_statem_server_max_message_size(SSL *s)
{
OSSL_STATEM *st = &s->statem;
switch (st->hand_state) {
default:
/* Shouldn't happen */
return 0;
case TLS_ST_SR_CLNT_HELLO:
return CLIENT_HELLO_MAX_LENGTH;
case TLS_ST_SR_END_OF_EARLY_DATA:
return END_OF_EARLY_DATA_MAX_LENGTH;
case TLS_ST_SR_CERT:
return s->max_cert_list;
case TLS_ST_SR_KEY_EXCH:
return CLIENT_KEY_EXCH_MAX_LENGTH;
case TLS_ST_SR_CERT_VRFY:
return SSL3_RT_MAX_PLAIN_LENGTH;
#ifndef OPENSSL_NO_NEXTPROTONEG
case TLS_ST_SR_NEXT_PROTO:
return NEXT_PROTO_MAX_LENGTH;
#endif
case TLS_ST_SR_CHANGE:
return CCS_MAX_LENGTH;
case TLS_ST_SR_FINISHED:
return FINISHED_MAX_LENGTH;
case TLS_ST_SR_KEY_UPDATE:
return KEY_UPDATE_MAX_LENGTH;
}
}
/*
* Process a message that the server has received from the client.
*/
MSG_PROCESS_RETURN ossl_statem_server_process_message(SSL *s, PACKET *pkt)
{
OSSL_STATEM *st = &s->statem;
switch (st->hand_state) {
default:
/* Shouldn't happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_OSSL_STATEM_SERVER_PROCESS_MESSAGE,
ERR_R_INTERNAL_ERROR);
return MSG_PROCESS_ERROR;
case TLS_ST_SR_CLNT_HELLO:
return tls_process_client_hello(s, pkt);
case TLS_ST_SR_END_OF_EARLY_DATA:
return tls_process_end_of_early_data(s, pkt);
case TLS_ST_SR_CERT:
return tls_process_client_certificate(s, pkt);
case TLS_ST_SR_KEY_EXCH:
return tls_process_client_key_exchange(s, pkt);
case TLS_ST_SR_CERT_VRFY:
return tls_process_cert_verify(s, pkt);
#ifndef OPENSSL_NO_NEXTPROTONEG
case TLS_ST_SR_NEXT_PROTO:
return tls_process_next_proto(s, pkt);
#endif
case TLS_ST_SR_CHANGE:
return tls_process_change_cipher_spec(s, pkt);
case TLS_ST_SR_FINISHED:
return tls_process_finished(s, pkt);
case TLS_ST_SR_KEY_UPDATE:
return tls_process_key_update(s, pkt);
}
}
/*
* Perform any further processing required following the receipt of a message
* from the client
*/
WORK_STATE ossl_statem_server_post_process_message(SSL *s, WORK_STATE wst)
{
OSSL_STATEM *st = &s->statem;
switch (st->hand_state) {
default:
/* Shouldn't happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_OSSL_STATEM_SERVER_POST_PROCESS_MESSAGE,
ERR_R_INTERNAL_ERROR);
return WORK_ERROR;
case TLS_ST_SR_CLNT_HELLO:
return tls_post_process_client_hello(s, wst);
case TLS_ST_SR_KEY_EXCH:
return tls_post_process_client_key_exchange(s, wst);
}
}
2011-03-12 17:01:19 +00:00
#ifndef OPENSSL_NO_SRP
/* Returns 1 on success, 0 for retryable error, -1 for fatal error */
static int ssl_check_srp_ext_ClientHello(SSL *s)
{
int ret;
int al = SSL_AD_UNRECOGNIZED_NAME;
if ((s->s3->tmp.new_cipher->algorithm_mkey & SSL_kSRP) &&
(s->srp_ctx.TLS_ext_srp_username_callback != NULL)) {
if (s->srp_ctx.login == NULL) {
/*
* RFC 5054 says SHOULD reject, we do so if There is no srp
* login name
*/
SSLfatal(s, SSL_AD_UNKNOWN_PSK_IDENTITY,
SSL_F_SSL_CHECK_SRP_EXT_CLIENTHELLO,
SSL_R_PSK_IDENTITY_NOT_FOUND);
return -1;
} else {
ret = SSL_srp_server_param_with_username(s, &al);
if (ret < 0)
return 0;
if (ret == SSL3_AL_FATAL) {
SSLfatal(s, al, SSL_F_SSL_CHECK_SRP_EXT_CLIENTHELLO,
al == SSL_AD_UNKNOWN_PSK_IDENTITY
? SSL_R_PSK_IDENTITY_NOT_FOUND
: SSL_R_CLIENTHELLO_TLSEXT);
return -1;
}
}
}
return 1;
}
2011-03-12 17:01:19 +00:00
#endif
int dtls_raw_hello_verify_request(WPACKET *pkt, unsigned char *cookie,
size_t cookie_len)
{
/* Always use DTLS 1.0 version: see RFC 6347 */
if (!WPACKET_put_bytes_u16(pkt, DTLS1_VERSION)
|| !WPACKET_sub_memcpy_u8(pkt, cookie, cookie_len))
return 0;
return 1;
}
int dtls_construct_hello_verify_request(SSL *s, WPACKET *pkt)
{
unsigned int cookie_leni;
if (s->ctx->app_gen_cookie_cb == NULL ||
s->ctx->app_gen_cookie_cb(s, s->d1->cookie,
&cookie_leni) == 0 ||
cookie_leni > 255) {
SSLfatal(s, SSL_AD_NO_ALERT, SSL_F_DTLS_CONSTRUCT_HELLO_VERIFY_REQUEST,
SSL_R_COOKIE_GEN_CALLBACK_FAILURE);
return 0;
}
s->d1->cookie_len = cookie_leni;
if (!dtls_raw_hello_verify_request(pkt, s->d1->cookie,
s->d1->cookie_len)) {
SSLfatal(s, SSL_AD_NO_ALERT, SSL_F_DTLS_CONSTRUCT_HELLO_VERIFY_REQUEST,
ERR_R_INTERNAL_ERROR);
return 0;
}
return 1;
}
#ifndef OPENSSL_NO_EC
/*-
* ssl_check_for_safari attempts to fingerprint Safari using OS X
* SecureTransport using the TLS extension block in |hello|.
* Safari, since 10.6, sends exactly these extensions, in this order:
* SNI,
* elliptic_curves
* ec_point_formats
* signature_algorithms (for TLSv1.2 only)
*
* We wish to fingerprint Safari because they broke ECDHE-ECDSA support in 10.8,
* but they advertise support. So enabling ECDHE-ECDSA ciphers breaks them.
* Sadly we cannot differentiate 10.6, 10.7 and 10.8.4 (which work), from
* 10.8..10.8.3 (which don't work).
*/
static void ssl_check_for_safari(SSL *s, const CLIENTHELLO_MSG *hello)
{
static const unsigned char kSafariExtensionsBlock[] = {
0x00, 0x0a, /* elliptic_curves extension */
0x00, 0x08, /* 8 bytes */
0x00, 0x06, /* 6 bytes of curve ids */
0x00, 0x17, /* P-256 */
0x00, 0x18, /* P-384 */
0x00, 0x19, /* P-521 */
0x00, 0x0b, /* ec_point_formats */
0x00, 0x02, /* 2 bytes */
0x01, /* 1 point format */
0x00, /* uncompressed */
/* The following is only present in TLS 1.2 */
0x00, 0x0d, /* signature_algorithms */
0x00, 0x0c, /* 12 bytes */
0x00, 0x0a, /* 10 bytes */
0x05, 0x01, /* SHA-384/RSA */
0x04, 0x01, /* SHA-256/RSA */
0x02, 0x01, /* SHA-1/RSA */
0x04, 0x03, /* SHA-256/ECDSA */
0x02, 0x03, /* SHA-1/ECDSA */
};
/* Length of the common prefix (first two extensions). */
static const size_t kSafariCommonExtensionsLength = 18;
unsigned int type;
PACKET sni, tmppkt;
size_t ext_len;
tmppkt = hello->extensions;
if (!PACKET_forward(&tmppkt, 2)
|| !PACKET_get_net_2(&tmppkt, &type)
|| !PACKET_get_length_prefixed_2(&tmppkt, &sni)) {
return;
}
if (type != TLSEXT_TYPE_server_name)
return;
ext_len = TLS1_get_client_version(s) >= TLS1_2_VERSION ?
sizeof(kSafariExtensionsBlock) : kSafariCommonExtensionsLength;
s->s3->is_probably_safari = PACKET_equal(&tmppkt, kSafariExtensionsBlock,
ext_len);
}
#endif /* !OPENSSL_NO_EC */
MSG_PROCESS_RETURN tls_process_client_hello(SSL *s, PACKET *pkt)
{
/* |cookie| will only be initialized for DTLS. */
PACKET session_id, compression, extensions, cookie;
static const unsigned char null_compression = 0;
CLIENTHELLO_MSG *clienthello = NULL;
/* Check if this is actually an unexpected renegotiation ClientHello */
if (s->renegotiate == 0 && !SSL_IS_FIRST_HANDSHAKE(s)) {
if (!ossl_assert(!SSL_IS_TLS13(s))) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
ERR_R_INTERNAL_ERROR);
goto err;
}
if ((s->options & SSL_OP_NO_RENEGOTIATION) != 0
|| (!s->s3->send_connection_binding
&& (s->options
& SSL_OP_ALLOW_UNSAFE_LEGACY_RENEGOTIATION) == 0)) {
ssl3_send_alert(s, SSL3_AL_WARNING, SSL_AD_NO_RENEGOTIATION);
return MSG_PROCESS_FINISHED_READING;
}
s->renegotiate = 1;
s->new_session = 1;
}
clienthello = OPENSSL_zalloc(sizeof(*clienthello));
if (clienthello == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
ERR_R_INTERNAL_ERROR);
goto err;
}
/*
* First, parse the raw ClientHello data into the CLIENTHELLO_MSG structure.
*/
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
clienthello->isv2 = RECORD_LAYER_is_sslv2_record(&s->rlayer);
PACKET_null_init(&cookie);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (clienthello->isv2) {
unsigned int mt;
if (!SSL_IS_FIRST_HANDSHAKE(s)
|| s->hello_retry_request != SSL_HRR_NONE) {
SSLfatal(s, SSL_AD_UNEXPECTED_MESSAGE,
SSL_F_TLS_PROCESS_CLIENT_HELLO, SSL_R_UNEXPECTED_MESSAGE);
goto err;
}
/*-
* An SSLv3/TLSv1 backwards-compatible CLIENT-HELLO in an SSLv2
* header is sent directly on the wire, not wrapped as a TLS
* record. Our record layer just processes the message length and passes
* the rest right through. Its format is:
* Byte Content
* 0-1 msg_length - decoded by the record layer
* 2 msg_type - s->init_msg points here
* 3-4 version
* 5-6 cipher_spec_length
* 7-8 session_id_length
* 9-10 challenge_length
* ... ...
*/
if (!PACKET_get_1(pkt, &mt)
|| mt != SSL2_MT_CLIENT_HELLO) {
/*
* Should never happen. We should have tested this in the record
* layer in order to have determined that this is a SSLv2 record
* in the first place
*/
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
ERR_R_INTERNAL_ERROR);
goto err;
}
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (!PACKET_get_net_2(pkt, &clienthello->legacy_version)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
SSL_R_LENGTH_TOO_SHORT);
goto err;
}
/* Parse the message and load client random. */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (clienthello->isv2) {
/*
* Handle an SSLv2 backwards compatible ClientHello
* Note, this is only for SSLv3+ using the backward compatible format.
* Real SSLv2 is not supported, and is rejected below.
*/
unsigned int ciphersuite_len, session_id_len, challenge_len;
PACKET challenge;
if (!PACKET_get_net_2(pkt, &ciphersuite_len)
|| !PACKET_get_net_2(pkt, &session_id_len)
|| !PACKET_get_net_2(pkt, &challenge_len)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
SSL_R_RECORD_LENGTH_MISMATCH);
goto err;
}
if (session_id_len > SSL_MAX_SSL_SESSION_ID_LENGTH) {
SSLfatal(s, SSL_AD_ILLEGAL_PARAMETER,
SSL_F_TLS_PROCESS_CLIENT_HELLO, SSL_R_LENGTH_MISMATCH);
goto err;
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (!PACKET_get_sub_packet(pkt, &clienthello->ciphersuites,
ciphersuite_len)
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
|| !PACKET_copy_bytes(pkt, clienthello->session_id, session_id_len)
|| !PACKET_get_sub_packet(pkt, &challenge, challenge_len)
/* No extensions. */
|| PACKET_remaining(pkt) != 0) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
SSL_R_RECORD_LENGTH_MISMATCH);
goto err;
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
clienthello->session_id_len = session_id_len;
/* Load the client random and compression list. We use SSL3_RANDOM_SIZE
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
* here rather than sizeof(clienthello->random) because that is the limit
* for SSLv3 and it is fixed. It won't change even if
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
* sizeof(clienthello->random) does.
*/
challenge_len = challenge_len > SSL3_RANDOM_SIZE
? SSL3_RANDOM_SIZE : challenge_len;
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
memset(clienthello->random, 0, SSL3_RANDOM_SIZE);
if (!PACKET_copy_bytes(&challenge,
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
clienthello->random + SSL3_RANDOM_SIZE -
challenge_len, challenge_len)
/* Advertise only null compression. */
|| !PACKET_buf_init(&compression, &null_compression, 1)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
ERR_R_INTERNAL_ERROR);
goto err;
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
PACKET_null_init(&clienthello->extensions);
} else {
/* Regular ClientHello. */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (!PACKET_copy_bytes(pkt, clienthello->random, SSL3_RANDOM_SIZE)
|| !PACKET_get_length_prefixed_1(pkt, &session_id)
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
|| !PACKET_copy_all(&session_id, clienthello->session_id,
SSL_MAX_SSL_SESSION_ID_LENGTH,
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
&clienthello->session_id_len)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
SSL_R_LENGTH_MISMATCH);
goto err;
}
if (SSL_IS_DTLS(s)) {
if (!PACKET_get_length_prefixed_1(pkt, &cookie)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
SSL_R_LENGTH_MISMATCH);
goto err;
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (!PACKET_copy_all(&cookie, clienthello->dtls_cookie,
DTLS1_COOKIE_LENGTH,
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
&clienthello->dtls_cookie_len)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_PROCESS_CLIENT_HELLO, ERR_R_INTERNAL_ERROR);
goto err;
}
/*
* If we require cookies and this ClientHello doesn't contain one,
* just return since we do not want to allocate any memory yet.
* So check cookie length...
*/
if (SSL_get_options(s) & SSL_OP_COOKIE_EXCHANGE) {
if (clienthello->dtls_cookie_len == 0) {
OPENSSL_free(clienthello);
return MSG_PROCESS_FINISHED_READING;
}
}
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (!PACKET_get_length_prefixed_2(pkt, &clienthello->ciphersuites)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
SSL_R_LENGTH_MISMATCH);
goto err;
}
if (!PACKET_get_length_prefixed_1(pkt, &compression)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
SSL_R_LENGTH_MISMATCH);
goto err;
}
/* Could be empty. */
if (PACKET_remaining(pkt) == 0) {
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
PACKET_null_init(&clienthello->extensions);
} else {
if (!PACKET_get_length_prefixed_2(pkt, &clienthello->extensions)
|| PACKET_remaining(pkt) != 0) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
SSL_R_LENGTH_MISMATCH);
goto err;
}
}
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (!PACKET_copy_all(&compression, clienthello->compressions,
MAX_COMPRESSIONS_SIZE,
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
&clienthello->compressions_len)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CLIENT_HELLO,
ERR_R_INTERNAL_ERROR);
goto err;
}
/* Preserve the raw extensions PACKET for later use */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
extensions = clienthello->extensions;
if (!tls_collect_extensions(s, &extensions, SSL_EXT_CLIENT_HELLO,
&clienthello->pre_proc_exts,
&clienthello->pre_proc_exts_len, 1)) {
/* SSLfatal already been called */
goto err;
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
s->clienthello = clienthello;
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
return MSG_PROCESS_CONTINUE_PROCESSING;
err:
if (clienthello != NULL)
OPENSSL_free(clienthello->pre_proc_exts);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
OPENSSL_free(clienthello);
return MSG_PROCESS_ERROR;
}
static int tls_early_post_process_client_hello(SSL *s)
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
{
unsigned int j;
int i, al = SSL_AD_INTERNAL_ERROR;
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
int protverr;
size_t loop;
unsigned long id;
#ifndef OPENSSL_NO_COMP
SSL_COMP *comp = NULL;
#endif
const SSL_CIPHER *c;
STACK_OF(SSL_CIPHER) *ciphers = NULL;
STACK_OF(SSL_CIPHER) *scsvs = NULL;
CLIENTHELLO_MSG *clienthello = s->clienthello;
DOWNGRADE dgrd = DOWNGRADE_NONE;
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
/* Finished parsing the ClientHello, now we can start processing it */
/* Give the ClientHello callback a crack at things */
if (s->ctx->client_hello_cb != NULL) {
/* A failure in the ClientHello callback terminates the connection. */
switch (s->ctx->client_hello_cb(s, &al, s->ctx->client_hello_cb_arg)) {
case SSL_CLIENT_HELLO_SUCCESS:
break;
case SSL_CLIENT_HELLO_RETRY:
s->rwstate = SSL_CLIENT_HELLO_CB;
return -1;
case SSL_CLIENT_HELLO_ERROR:
default:
SSLfatal(s, al,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_CALLBACK_FAILED);
goto err;
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
}
}
/* Set up the client_random */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
memcpy(s->s3->client_random, clienthello->random, SSL3_RANDOM_SIZE);
/* Choose the version */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (clienthello->isv2) {
if (clienthello->legacy_version == SSL2_VERSION
|| (clienthello->legacy_version & 0xff00)
!= (SSL3_VERSION_MAJOR << 8)) {
/*
* This is real SSLv2 or something completely unknown. We don't
* support it.
*/
SSLfatal(s, SSL_AD_PROTOCOL_VERSION,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_UNKNOWN_PROTOCOL);
goto err;
}
/* SSLv3/TLS */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
s->client_version = clienthello->legacy_version;
}
/*
* Do SSL/TLS version negotiation if applicable. For DTLS we just check
* versions are potentially compatible. Version negotiation comes later.
*/
if (!SSL_IS_DTLS(s)) {
protverr = ssl_choose_server_version(s, clienthello, &dgrd);
} else if (s->method->version != DTLS_ANY_VERSION &&
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
DTLS_VERSION_LT((int)clienthello->legacy_version, s->version)) {
protverr = SSL_R_VERSION_TOO_LOW;
} else {
protverr = 0;
}
if (protverr) {
if (SSL_IS_FIRST_HANDSHAKE(s)) {
/* like ssl3_get_record, send alert using remote version number */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
s->version = s->client_version = clienthello->legacy_version;
}
SSLfatal(s, SSL_AD_PROTOCOL_VERSION,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO, protverr);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
/* TLSv1.3 specifies that a ClientHello must end on a record boundary */
if (SSL_IS_TLS13(s) && RECORD_LAYER_processed_read_pending(&s->rlayer)) {
SSLfatal(s, SSL_AD_UNEXPECTED_MESSAGE,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_NOT_ON_RECORD_BOUNDARY);
goto err;
}
if (SSL_IS_DTLS(s)) {
/* Empty cookie was already handled above by returning early. */
if (SSL_get_options(s) & SSL_OP_COOKIE_EXCHANGE) {
if (s->ctx->app_verify_cookie_cb != NULL) {
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (s->ctx->app_verify_cookie_cb(s, clienthello->dtls_cookie,
clienthello->dtls_cookie_len) == 0) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_COOKIE_MISMATCH);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
/* else cookie verification succeeded */
}
/* default verification */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
} else if (s->d1->cookie_len != clienthello->dtls_cookie_len
|| memcmp(clienthello->dtls_cookie, s->d1->cookie,
s->d1->cookie_len) != 0) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_COOKIE_MISMATCH);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
s->d1->cookie_verified = 1;
}
if (s->method->version == DTLS_ANY_VERSION) {
protverr = ssl_choose_server_version(s, clienthello, &dgrd);
if (protverr != 0) {
s->version = s->client_version;
SSLfatal(s, SSL_AD_PROTOCOL_VERSION,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO, protverr);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
}
}
s->hit = 0;
if (!ssl_cache_cipherlist(s, &clienthello->ciphersuites,
clienthello->isv2) ||
!bytes_to_cipher_list(s, &clienthello->ciphersuites, &ciphers, &scsvs,
clienthello->isv2, 1)) {
/* SSLfatal() already called */
goto err;
}
s->s3->send_connection_binding = 0;
/* Check what signalling cipher-suite values were received. */
if (scsvs != NULL) {
for(i = 0; i < sk_SSL_CIPHER_num(scsvs); i++) {
c = sk_SSL_CIPHER_value(scsvs, i);
if (SSL_CIPHER_get_id(c) == SSL3_CK_SCSV) {
if (s->renegotiate) {
/* SCSV is fatal if renegotiating */
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_SCSV_RECEIVED_WHEN_RENEGOTIATING);
goto err;
}
s->s3->send_connection_binding = 1;
} else if (SSL_CIPHER_get_id(c) == SSL3_CK_FALLBACK_SCSV &&
!ssl_check_version_downgrade(s)) {
/*
* This SCSV indicates that the client previously tried
* a higher version. We should fail if the current version
* is an unexpected downgrade, as that indicates that the first
* connection may have been tampered with in order to trigger
* an insecure downgrade.
*/
SSLfatal(s, SSL_AD_INAPPROPRIATE_FALLBACK,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_INAPPROPRIATE_FALLBACK);
goto err;
}
}
}
/* For TLSv1.3 we must select the ciphersuite *before* session resumption */
if (SSL_IS_TLS13(s)) {
const SSL_CIPHER *cipher =
ssl3_choose_cipher(s, ciphers, SSL_get_ciphers(s));
if (cipher == NULL) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_NO_SHARED_CIPHER);
goto err;
}
if (s->hello_retry_request == SSL_HRR_PENDING
&& (s->s3->tmp.new_cipher == NULL
|| s->s3->tmp.new_cipher->id != cipher->id)) {
/*
* A previous HRR picked a different ciphersuite to the one we
* just selected. Something must have changed.
*/
SSLfatal(s, SSL_AD_ILLEGAL_PARAMETER,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_BAD_CIPHER);
goto err;
}
s->s3->tmp.new_cipher = cipher;
}
/* We need to do this before getting the session */
if (!tls_parse_extension(s, TLSEXT_IDX_extended_master_secret,
SSL_EXT_CLIENT_HELLO,
clienthello->pre_proc_exts, NULL, 0)) {
/* SSLfatal() already called */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
/*
* We don't allow resumption in a backwards compatible ClientHello.
* TODO(openssl-team): in TLS1.1+, session_id MUST be empty.
*
* Versions before 0.9.7 always allow clients to resume sessions in
* renegotiation. 0.9.7 and later allow this by default, but optionally
* ignore resumption requests with flag
* SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION (it's a new flag rather
* than a change to default behavior so that applications relying on
* this for security won't even compile against older library versions).
* 1.0.1 and later also have a function SSL_renegotiate_abbreviated() to
* request renegotiation but not a new session (s->new_session remains
* unset): for servers, this essentially just means that the
* SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION setting will be
* ignored.
*/
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (clienthello->isv2 ||
(s->new_session &&
(s->options & SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION))) {
if (!ssl_get_new_session(s, 1)) {
/* SSLfatal() already called */
goto err;
}
} else {
i = ssl_get_prev_session(s, clienthello);
if (i == 1) {
/* previous session */
s->hit = 1;
} else if (i == -1) {
/* SSLfatal() already called */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
} else {
/* i == 0 */
if (!ssl_get_new_session(s, 1)) {
/* SSLfatal() already called */
goto err;
}
}
}
if (SSL_IS_TLS13(s)) {
memcpy(s->tmp_session_id, s->clienthello->session_id,
s->clienthello->session_id_len);
s->tmp_session_id_len = s->clienthello->session_id_len;
}
/*
* If it is a hit, check that the cipher is in the list. In TLSv1.3 we check
* ciphersuite compatibility with the session as part of resumption.
*/
if (!SSL_IS_TLS13(s) && s->hit) {
j = 0;
id = s->session->cipher->id;
#ifdef CIPHER_DEBUG
fprintf(stderr, "client sent %d ciphers\n", sk_SSL_CIPHER_num(ciphers));
#endif
for (i = 0; i < sk_SSL_CIPHER_num(ciphers); i++) {
c = sk_SSL_CIPHER_value(ciphers, i);
#ifdef CIPHER_DEBUG
fprintf(stderr, "client [%2d of %2d]:%s\n",
i, sk_SSL_CIPHER_num(ciphers), SSL_CIPHER_get_name(c));
#endif
if (c->id == id) {
j = 1;
break;
}
}
if (j == 0) {
/*
* we need to have the cipher in the cipher list if we are asked
* to reuse it
*/
SSLfatal(s, SSL_AD_ILLEGAL_PARAMETER,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_REQUIRED_CIPHER_MISSING);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
for (loop = 0; loop < clienthello->compressions_len; loop++) {
if (clienthello->compressions[loop] == 0)
break;
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (loop >= clienthello->compressions_len) {
/* no compress */
SSLfatal(s, SSL_AD_DECODE_ERROR,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_NO_COMPRESSION_SPECIFIED);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
#ifndef OPENSSL_NO_EC
if (s->options & SSL_OP_SAFARI_ECDHE_ECDSA_BUG)
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
ssl_check_for_safari(s, clienthello);
#endif /* !OPENSSL_NO_EC */
/* TLS extensions */
if (!tls_parse_all_extensions(s, SSL_EXT_CLIENT_HELLO,
clienthello->pre_proc_exts, NULL, 0, 1)) {
/* SSLfatal() already called */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
/*
* Check if we want to use external pre-shared secret for this handshake
* for not reused session only. We need to generate server_random before
* calling tls_session_secret_cb in order to allow SessionTicket
* processing to use it in key derivation.
*/
{
unsigned char *pos;
pos = s->s3->server_random;
if (ssl_fill_hello_random(s, 1, pos, SSL3_RANDOM_SIZE, dgrd) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
ERR_R_INTERNAL_ERROR);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
}
if (!s->hit
&& s->version >= TLS1_VERSION
&& !SSL_IS_TLS13(s)
&& !SSL_IS_DTLS(s)
&& s->ext.session_secret_cb) {
const SSL_CIPHER *pref_cipher = NULL;
/*
* s->session->master_key_length is a size_t, but this is an int for
* backwards compat reasons
*/
int master_key_length;
master_key_length = sizeof(s->session->master_key);
if (s->ext.session_secret_cb(s, s->session->master_key,
&master_key_length, ciphers,
&pref_cipher,
s->ext.session_secret_cb_arg)
&& master_key_length > 0) {
s->session->master_key_length = master_key_length;
s->hit = 1;
s->peer_ciphers = ciphers;
s->session->verify_result = X509_V_OK;
ciphers = NULL;
/* check if some cipher was preferred by call back */
if (pref_cipher == NULL)
pref_cipher = ssl3_choose_cipher(s, s->peer_ciphers,
SSL_get_ciphers(s));
if (pref_cipher == NULL) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_NO_SHARED_CIPHER);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
s->session->cipher = pref_cipher;
sk_SSL_CIPHER_free(s->cipher_list);
s->cipher_list = sk_SSL_CIPHER_dup(s->peer_ciphers);
sk_SSL_CIPHER_free(s->cipher_list_by_id);
s->cipher_list_by_id = sk_SSL_CIPHER_dup(s->peer_ciphers);
}
}
/*
* Worst case, we will use the NULL compression, but if we have other
* options, we will now look for them. We have complen-1 compression
* algorithms from the client, starting at q.
*/
s->s3->tmp.new_compression = NULL;
if (SSL_IS_TLS13(s)) {
/*
* We already checked above that the NULL compression method appears in
* the list. Now we check there aren't any others (which is illegal in
* a TLSv1.3 ClientHello.
*/
if (clienthello->compressions_len != 1) {
SSLfatal(s, SSL_AD_ILLEGAL_PARAMETER,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_INVALID_COMPRESSION_ALGORITHM);
goto err;
}
}
2005-09-30 23:35:33 +00:00
#ifndef OPENSSL_NO_COMP
/* This only happens if we have a cache hit */
else if (s->session->compress_meth != 0) {
int m, comp_id = s->session->compress_meth;
unsigned int k;
/* Perform sanity checks on resumed compression algorithm */
/* Can't disable compression */
if (!ssl_allow_compression(s)) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_INCONSISTENT_COMPRESSION);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
/* Look for resumed compression method */
for (m = 0; m < sk_SSL_COMP_num(s->ctx->comp_methods); m++) {
comp = sk_SSL_COMP_value(s->ctx->comp_methods, m);
if (comp_id == comp->id) {
s->s3->tmp.new_compression = comp;
break;
}
}
if (s->s3->tmp.new_compression == NULL) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_INVALID_COMPRESSION_ALGORITHM);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
/* Look for resumed method in compression list */
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
for (k = 0; k < clienthello->compressions_len; k++) {
if (clienthello->compressions[k] == comp_id)
break;
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (k >= clienthello->compressions_len) {
SSLfatal(s, SSL_AD_ILLEGAL_PARAMETER,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_REQUIRED_COMPRESSION_ALGORITHM_MISSING);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
} else if (s->hit) {
comp = NULL;
} else if (ssl_allow_compression(s) && s->ctx->comp_methods) {
/* See if we have a match */
int m, nn, v, done = 0;
unsigned int o;
nn = sk_SSL_COMP_num(s->ctx->comp_methods);
for (m = 0; m < nn; m++) {
comp = sk_SSL_COMP_value(s->ctx->comp_methods, m);
v = comp->id;
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
for (o = 0; o < clienthello->compressions_len; o++) {
if (v == clienthello->compressions[o]) {
done = 1;
break;
}
}
if (done)
break;
}
if (done)
s->s3->tmp.new_compression = comp;
else
comp = NULL;
}
#else
/*
* If compression is disabled we'd better not try to resume a session
* using compression.
*/
if (s->session->compress_meth != 0) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
SSL_R_INCONSISTENT_COMPRESSION);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
2005-09-30 23:35:33 +00:00
#endif
/*
* Given s->peer_ciphers and SSL_get_ciphers, we must pick a cipher
*/
if (!s->hit || SSL_IS_TLS13(s)) {
sk_SSL_CIPHER_free(s->peer_ciphers);
s->peer_ciphers = ciphers;
if (ciphers == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO,
ERR_R_INTERNAL_ERROR);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
goto err;
}
ciphers = NULL;
}
if (!s->hit) {
#ifdef OPENSSL_NO_COMP
s->session->compress_meth = 0;
#else
s->session->compress_meth = (comp == NULL) ? 0 : comp->id;
#endif
if (!tls1_set_server_sigalgs(s)) {
/* SSLfatal() already called */
goto err;
}
}
sk_SSL_CIPHER_free(ciphers);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
sk_SSL_CIPHER_free(scsvs);
OPENSSL_free(clienthello->pre_proc_exts);
OPENSSL_free(s->clienthello);
s->clienthello = NULL;
return 1;
err:
sk_SSL_CIPHER_free(ciphers);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
sk_SSL_CIPHER_free(scsvs);
OPENSSL_free(clienthello->pre_proc_exts);
OPENSSL_free(s->clienthello);
s->clienthello = NULL;
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
return 0;
}
/*
* Call the status request callback if needed. Upon success, returns 1.
* Upon failure, returns 0.
*/
static int tls_handle_status_request(SSL *s)
{
s->ext.status_expected = 0;
/*
* If status request then ask callback what to do. Note: this must be
* called after servername callbacks in case the certificate has changed,
* and must be called after the cipher has been chosen because this may
* influence which certificate is sent
*/
if (s->ext.status_type != TLSEXT_STATUSTYPE_nothing && s->ctx != NULL
&& s->ctx->ext.status_cb != NULL) {
int ret;
/* If no certificate can't return certificate status */
if (s->s3->tmp.cert != NULL) {
/*
* Set current certificate to one we will use so SSL_get_certificate
* et al can pick it up.
*/
s->cert->key = s->s3->tmp.cert;
ret = s->ctx->ext.status_cb(s, s->ctx->ext.status_arg);
switch (ret) {
/* We don't want to send a status request response */
case SSL_TLSEXT_ERR_NOACK:
s->ext.status_expected = 0;
break;
/* status request response should be sent */
case SSL_TLSEXT_ERR_OK:
if (s->ext.ocsp.resp)
s->ext.status_expected = 1;
break;
/* something bad happened */
case SSL_TLSEXT_ERR_ALERT_FATAL:
default:
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_HANDLE_STATUS_REQUEST,
SSL_R_CLIENTHELLO_TLSEXT);
return 0;
}
}
}
return 1;
}
/*
* Call the alpn_select callback if needed. Upon success, returns 1.
* Upon failure, returns 0.
*/
int tls_handle_alpn(SSL *s)
{
const unsigned char *selected = NULL;
unsigned char selected_len = 0;
if (s->ctx->ext.alpn_select_cb != NULL && s->s3->alpn_proposed != NULL) {
int r = s->ctx->ext.alpn_select_cb(s, &selected, &selected_len,
s->s3->alpn_proposed,
(unsigned int)s->s3->alpn_proposed_len,
s->ctx->ext.alpn_select_cb_arg);
if (r == SSL_TLSEXT_ERR_OK) {
OPENSSL_free(s->s3->alpn_selected);
s->s3->alpn_selected = OPENSSL_memdup(selected, selected_len);
if (s->s3->alpn_selected == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_HANDLE_ALPN,
ERR_R_INTERNAL_ERROR);
return 0;
}
s->s3->alpn_selected_len = selected_len;
#ifndef OPENSSL_NO_NEXTPROTONEG
/* ALPN takes precedence over NPN. */
s->s3->npn_seen = 0;
#endif
/* Check ALPN is consistent with session */
if (s->session->ext.alpn_selected == NULL
|| selected_len != s->session->ext.alpn_selected_len
|| memcmp(selected, s->session->ext.alpn_selected,
selected_len) != 0) {
/* Not consistent so can't be used for early_data */
s->ext.early_data_ok = 0;
if (!s->hit) {
/*
* This is a new session and so alpn_selected should have
* been initialised to NULL. We should update it with the
* selected ALPN.
*/
if (!ossl_assert(s->session->ext.alpn_selected == NULL)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_HANDLE_ALPN,
ERR_R_INTERNAL_ERROR);
return 0;
}
s->session->ext.alpn_selected = OPENSSL_memdup(selected,
selected_len);
if (s->session->ext.alpn_selected == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_HANDLE_ALPN,
ERR_R_INTERNAL_ERROR);
return 0;
}
s->session->ext.alpn_selected_len = selected_len;
}
}
return 1;
} else if (r != SSL_TLSEXT_ERR_NOACK) {
SSLfatal(s, SSL_AD_NO_APPLICATION_PROTOCOL, SSL_F_TLS_HANDLE_ALPN,
SSL_R_NO_APPLICATION_PROTOCOL);
return 0;
}
/*
* If r == SSL_TLSEXT_ERR_NOACK then behave as if no callback was
* present.
*/
}
/* Check ALPN is consistent with session */
if (s->session->ext.alpn_selected != NULL) {
/* Not consistent so can't be used for early_data */
s->ext.early_data_ok = 0;
}
return 1;
}
WORK_STATE tls_post_process_client_hello(SSL *s, WORK_STATE wst)
{
const SSL_CIPHER *cipher;
if (wst == WORK_MORE_A) {
int rv = tls_early_post_process_client_hello(s);
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (rv == 0) {
/* SSLfatal() was already called */
goto err;
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
}
if (rv < 0)
return WORK_MORE_A;
wst = WORK_MORE_B;
}
if (wst == WORK_MORE_B) {
if (!s->hit || SSL_IS_TLS13(s)) {
/* Let cert callback update server certificates if required */
if (!s->hit && s->cert->cert_cb != NULL) {
int rv = s->cert->cert_cb(s, s->cert->cert_cb_arg);
if (rv == 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_POST_PROCESS_CLIENT_HELLO,
SSL_R_CERT_CB_ERROR);
goto err;
}
if (rv < 0) {
s->rwstate = SSL_X509_LOOKUP;
return WORK_MORE_B;
}
s->rwstate = SSL_NOTHING;
}
/* In TLSv1.3 we selected the ciphersuite before resumption */
if (!SSL_IS_TLS13(s)) {
cipher =
ssl3_choose_cipher(s, s->peer_ciphers, SSL_get_ciphers(s));
if (cipher == NULL) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_POST_PROCESS_CLIENT_HELLO,
SSL_R_NO_SHARED_CIPHER);
goto err;
}
s->s3->tmp.new_cipher = cipher;
}
if (!s->hit) {
if (!tls_choose_sigalg(s, 1)) {
/* SSLfatal already called */
goto err;
}
/* check whether we should disable session resumption */
if (s->not_resumable_session_cb != NULL)
s->session->not_resumable =
s->not_resumable_session_cb(s,
((s->s3->tmp.new_cipher->algorithm_mkey
& (SSL_kDHE | SSL_kECDHE)) != 0));
if (s->session->not_resumable)
/* do not send a session ticket */
s->ext.ticket_expected = 0;
}
} else {
/* Session-id reuse */
s->s3->tmp.new_cipher = s->session->cipher;
}
/*-
* we now have the following setup.
* client_random
* cipher_list - our preferred list of ciphers
* ciphers - the clients preferred list of ciphers
* compression - basically ignored right now
* ssl version is set - sslv3
* s->session - The ssl session has been setup.
* s->hit - session reuse flag
* s->s3->tmp.new_cipher- the new cipher to use.
*/
/*
* Call status_request callback if needed. Has to be done after the
* certificate callbacks etc above.
*/
if (!tls_handle_status_request(s)) {
/* SSLfatal() already called */
goto err;
}
/*
* Call alpn_select callback if needed. Has to be done after SNI and
* cipher negotiation (HTTP/2 restricts permitted ciphers). In TLSv1.3
* we already did this because cipher negotiation happens earlier, and
* we must handle ALPN before we decide whether to accept early_data.
*/
if (!SSL_IS_TLS13(s) && !tls_handle_alpn(s)) {
/* SSLfatal() already called */
goto err;
}
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
wst = WORK_MORE_C;
}
#ifndef OPENSSL_NO_SRP
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
if (wst == WORK_MORE_C) {
int ret;
if ((ret = ssl_check_srp_ext_ClientHello(s)) == 0) {
/*
* callback indicates further work to be done
*/
s->rwstate = SSL_X509_LOOKUP;
Add SSL_CTX early callback Provide a callback interface that gives the application the ability to adjust the nascent SSL object at the earliest stage of ClientHello processing, immediately after extensions have been collected but before they have been processed. This is akin to BoringSSL's "select_certificate_cb" (though it is not API compatible), and as the name indicates, one major use is to examine the supplied server name indication and select what certificate to present to the client. However, it can also be used to make more sweeping configuration changes to the SSL object according to the selected server identity and configuration. That may include adjusting the permitted TLS versions, swapping out the SSL_CTX object (as is traditionally done in a tlsext_servername_callback), changing the server's cipher list, and more. We also wish to allow an early callback to indicate that it needs to perform additional work asynchronously and resume processing later. To that effect, refactor the second half of tls_process_client_hello() into a subroutine to be called at the post-processing stage (including the early callback itself), to allow the callback to result in remaining in the same work stage for a later call to succeed. This requires allocating for and storing the CLIENTHELLO_MSG in the SSL object to be preserved across such calls, but the storage is reclaimed after ClientHello processing finishes. Information about the CliehtHello is available to the callback by means of accessor functions that can only be used from the early callback. This allows extensions to make use of the existing internal parsing machinery without exposing structure internals (e.g., of PACKET), so that applications do not have to write fragile parsing code. Applications are encouraged to utilize an early callback and not use a servername_callback, in order to avoid unexpected behavior that occurs due to the relative order of processing between things like session resumption and the historical servername callback. Also tidy up nearby style by removing unnecessary braces around one-line conditional bodies. Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> (Merged from https://github.com/openssl/openssl/pull/2279)
2017-01-23 23:03:16 +00:00
return WORK_MORE_C;
}
if (ret < 0) {
/* SSLfatal() already called */
goto err;
}
}
#endif
return WORK_FINISHED_STOP;
err:
return WORK_ERROR;
}
int tls_construct_server_hello(SSL *s, WPACKET *pkt)
{
int compm;
size_t sl, len;
int version;
unsigned char *session_id;
int usetls13 = SSL_IS_TLS13(s) || s->hello_retry_request == SSL_HRR_PENDING;
version = usetls13 ? TLS1_2_VERSION : s->version;
if (!WPACKET_put_bytes_u16(pkt, version)
/*
* Random stuff. Filling of the server_random takes place in
* tls_process_client_hello()
*/
|| !WPACKET_memcpy(pkt,
s->hello_retry_request == SSL_HRR_PENDING
? hrrrandom : s->s3->server_random,
SSL3_RANDOM_SIZE)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_CONSTRUCT_SERVER_HELLO,
ERR_R_INTERNAL_ERROR);
return 0;
}
/*-
* There are several cases for the session ID to send
* back in the server hello:
* - For session reuse from the session cache,
* we send back the old session ID.
* - If stateless session reuse (using a session ticket)
* is successful, we send back the client's "session ID"
* (which doesn't actually identify the session).
* - If it is a new session, we send back the new
* session ID.
* - However, if we want the new session to be single-use,
* we send back a 0-length session ID.
* - In TLSv1.3 we echo back the session id sent to us by the client
* regardless
* s->hit is non-zero in either case of session reuse,
* so the following won't overwrite an ID that we're supposed
* to send back.
*/
if (s->session->not_resumable ||
(!(s->ctx->session_cache_mode & SSL_SESS_CACHE_SERVER)
&& !s->hit))
s->session->session_id_length = 0;
if (usetls13) {
sl = s->tmp_session_id_len;
session_id = s->tmp_session_id;
} else {
sl = s->session->session_id_length;
session_id = s->session->session_id;
}
if (sl > sizeof(s->session->session_id)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_CONSTRUCT_SERVER_HELLO,
ERR_R_INTERNAL_ERROR);
return 0;
}
/* set up the compression method */
2005-09-30 23:35:33 +00:00
#ifdef OPENSSL_NO_COMP
compm = 0;
2005-09-30 23:35:33 +00:00
#else
if (usetls13 || s->s3->tmp.new_compression == NULL)
compm = 0;
else
compm = s->s3->tmp.new_compression->id;
2005-09-30 23:35:33 +00:00
#endif
if (!WPACKET_sub_memcpy_u8(pkt, session_id, sl)
|| !s->method->put_cipher_by_char(s->s3->tmp.new_cipher, pkt, &len)
|| !WPACKET_put_bytes_u8(pkt, compm)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_CONSTRUCT_SERVER_HELLO,
ERR_R_INTERNAL_ERROR);
return 0;
}
if (!tls_construct_extensions(s, pkt,
s->hello_retry_request == SSL_HRR_PENDING
? SSL_EXT_TLS1_3_HELLO_RETRY_REQUEST
: (SSL_IS_TLS13(s)
? SSL_EXT_TLS1_3_SERVER_HELLO
: SSL_EXT_TLS1_2_SERVER_HELLO),
NULL, 0)) {
/* SSLfatal() already called */
return 0;
}
if (s->hello_retry_request == SSL_HRR_PENDING) {
/* Ditch the session. We'll create a new one next time around */
SSL_SESSION_free(s->session);
s->session = NULL;
s->hit = 0;
/*
* Re-initialise the Transcript Hash. We're going to prepopulate it with
* a synthetic message_hash in place of ClientHello1.
*/
if (!create_synthetic_message_hash(s, NULL, 0, NULL, 0)) {
/* SSLfatal() already called */
return 0;
}
} else if (!(s->verify_mode & SSL_VERIFY_PEER)
&& !ssl3_digest_cached_records(s, 0)) {
/* SSLfatal() already called */;
return 0;
}
return 1;
}
int tls_construct_server_done(SSL *s, WPACKET *pkt)
{
if (!s->s3->tmp.cert_request) {
if (!ssl3_digest_cached_records(s, 0)) {
/* SSLfatal() already called */
return 0;
}
}
return 1;
}
int tls_construct_server_key_exchange(SSL *s, WPACKET *pkt)
{
#ifndef OPENSSL_NO_DH
EVP_PKEY *pkdh = NULL;
#endif
#ifndef OPENSSL_NO_EC
unsigned char *encodedPoint = NULL;
size_t encodedlen = 0;
int curve_id = 0;
#endif
const SIGALG_LOOKUP *lu = s->s3->tmp.sigalg;
int i;
unsigned long type;
const BIGNUM *r[4];
EVP_MD_CTX *md_ctx = EVP_MD_CTX_new();
EVP_PKEY_CTX *pctx = NULL;
size_t paramlen, paramoffset;
if (!WPACKET_get_total_written(pkt, &paramoffset)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE, ERR_R_INTERNAL_ERROR);
goto err;
}
if (md_ctx == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE, ERR_R_MALLOC_FAILURE);
goto err;
}
type = s->s3->tmp.new_cipher->algorithm_mkey;
r[0] = r[1] = r[2] = r[3] = NULL;
#ifndef OPENSSL_NO_PSK
/* Plain PSK or RSAPSK nothing to do */
if (type & (SSL_kPSK | SSL_kRSAPSK)) {
} else
#endif /* !OPENSSL_NO_PSK */
#ifndef OPENSSL_NO_DH
if (type & (SSL_kDHE | SSL_kDHEPSK)) {
CERT *cert = s->cert;
EVP_PKEY *pkdhp = NULL;
DH *dh;
if (s->cert->dh_tmp_auto) {
DH *dhp = ssl_get_auto_dh(s);
pkdh = EVP_PKEY_new();
if (pkdh == NULL || dhp == NULL) {
DH_free(dhp);
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
EVP_PKEY_assign_DH(pkdh, dhp);
pkdhp = pkdh;
} else {
pkdhp = cert->dh_tmp;
}
if ((pkdhp == NULL) && (s->cert->dh_tmp_cb != NULL)) {
DH *dhp = s->cert->dh_tmp_cb(s, 0, 1024);
pkdh = ssl_dh_to_pkey(dhp);
if (pkdh == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
pkdhp = pkdh;
}
if (pkdhp == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
SSL_R_MISSING_TMP_DH_KEY);
goto err;
}
if (!ssl_security(s, SSL_SECOP_TMP_DH,
EVP_PKEY_security_bits(pkdhp), 0, pkdhp)) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
SSL_R_DH_KEY_TOO_SMALL);
goto err;
}
if (s->s3->tmp.pkey != NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
s->s3->tmp.pkey = ssl_generate_pkey(pkdhp);
if (s->s3->tmp.pkey == NULL) {
/* SSLfatal() already called */
goto err;
}
dh = EVP_PKEY_get0_DH(s->s3->tmp.pkey);
if (dh == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
EVP_PKEY_free(pkdh);
pkdh = NULL;
DH_get0_pqg(dh, &r[0], NULL, &r[1]);
DH_get0_key(dh, &r[2], NULL);
} else
#endif
#ifndef OPENSSL_NO_EC
if (type & (SSL_kECDHE | SSL_kECDHEPSK)) {
if (s->s3->tmp.pkey != NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
/* Get NID of appropriate shared curve */
curve_id = tls1_shared_group(s, -2);
if (curve_id == 0) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
SSL_R_UNSUPPORTED_ELLIPTIC_CURVE);
goto err;
}
s->s3->tmp.pkey = ssl_generate_pkey_group(s, curve_id);
/* Generate a new key for this curve */
if (s->s3->tmp.pkey == NULL) {
/* SSLfatal() already called */
goto err;
}
/* Encode the public key. */
encodedlen = EVP_PKEY_get1_tls_encodedpoint(s->s3->tmp.pkey,
&encodedPoint);
if (encodedlen == 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE, ERR_R_EC_LIB);
goto err;
}
/*
* We'll generate the serverKeyExchange message explicitly so we
* can set these to NULLs
*/
r[0] = NULL;
r[1] = NULL;
r[2] = NULL;
r[3] = NULL;
} else
#endif /* !OPENSSL_NO_EC */
2011-03-12 17:01:19 +00:00
#ifndef OPENSSL_NO_SRP
if (type & SSL_kSRP) {
if ((s->srp_ctx.N == NULL) ||
(s->srp_ctx.g == NULL) ||
(s->srp_ctx.s == NULL) || (s->srp_ctx.B == NULL)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
SSL_R_MISSING_SRP_PARAM);
goto err;
}
r[0] = s->srp_ctx.N;
r[1] = s->srp_ctx.g;
r[2] = s->srp_ctx.s;
r[3] = s->srp_ctx.B;
} else
#endif
{
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
SSL_R_UNKNOWN_KEY_EXCHANGE_TYPE);
goto err;
}
if (((s->s3->tmp.new_cipher->algorithm_auth & (SSL_aNULL | SSL_aSRP)) != 0)
|| ((s->s3->tmp.new_cipher->algorithm_mkey & SSL_PSK)) != 0) {
lu = NULL;
} else if (lu == NULL) {
SSLfatal(s, SSL_AD_DECODE_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE, ERR_R_INTERNAL_ERROR);
goto err;
}
#ifndef OPENSSL_NO_PSK
if (type & SSL_PSK) {
size_t len = (s->cert->psk_identity_hint == NULL)
? 0 : strlen(s->cert->psk_identity_hint);
/*
* It should not happen that len > PSK_MAX_IDENTITY_LEN - we already
* checked this when we set the identity hint - but just in case
*/
if (len > PSK_MAX_IDENTITY_LEN
|| !WPACKET_sub_memcpy_u16(pkt, s->cert->psk_identity_hint,
len)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
}
#endif
for (i = 0; i < 4 && r[i] != NULL; i++) {
unsigned char *binval;
int res;
2011-03-12 17:01:19 +00:00
#ifndef OPENSSL_NO_SRP
if ((i == 2) && (type & SSL_kSRP)) {
res = WPACKET_start_sub_packet_u8(pkt);
} else
#endif
res = WPACKET_start_sub_packet_u16(pkt);
if (!res) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
#ifndef OPENSSL_NO_DH
/*-
* for interoperability with some versions of the Microsoft TLS
* stack, we need to zero pad the DHE pub key to the same length
* as the prime
*/
if ((i == 2) && (type & (SSL_kDHE | SSL_kDHEPSK))) {
size_t len = BN_num_bytes(r[0]) - BN_num_bytes(r[2]);
if (len > 0) {
if (!WPACKET_allocate_bytes(pkt, len, &binval)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
memset(binval, 0, len);
}
}
2011-03-12 17:01:19 +00:00
#endif
if (!WPACKET_allocate_bytes(pkt, BN_num_bytes(r[i]), &binval)
|| !WPACKET_close(pkt)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
BN_bn2bin(r[i], binval);
}
#ifndef OPENSSL_NO_EC
if (type & (SSL_kECDHE | SSL_kECDHEPSK)) {
/*
* We only support named (not generic) curves. In this situation, the
* ServerKeyExchange message has: [1 byte CurveType], [2 byte CurveName]
* [1 byte length of encoded point], followed by the actual encoded
* point itself
*/
if (!WPACKET_put_bytes_u8(pkt, NAMED_CURVE_TYPE)
|| !WPACKET_put_bytes_u8(pkt, 0)
|| !WPACKET_put_bytes_u8(pkt, curve_id)
|| !WPACKET_sub_memcpy_u8(pkt, encodedPoint, encodedlen)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
OPENSSL_free(encodedPoint);
encodedPoint = NULL;
}
#endif
/* not anonymous */
if (lu != NULL) {
EVP_PKEY *pkey = s->s3->tmp.cert->privatekey;
const EVP_MD *md;
unsigned char *sigbytes1, *sigbytes2, *tbs;
size_t siglen, tbslen;
int rv;
if (pkey == NULL || !tls1_lookup_md(lu, &md)) {
/* Should never happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
/* Get length of the parameters we have written above */
if (!WPACKET_get_length(pkt, &paramlen)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
/* send signature algorithm */
if (SSL_USE_SIGALGS(s) && !WPACKET_put_bytes_u16(pkt, lu->sigalg)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
/*
* Create the signature. We don't know the actual length of the sig
* until after we've created it, so we reserve enough bytes for it
* up front, and then properly allocate them in the WPACKET
* afterwards.
*/
siglen = EVP_PKEY_size(pkey);
if (!WPACKET_sub_reserve_bytes_u16(pkt, siglen, &sigbytes1)
|| EVP_DigestSignInit(md_ctx, &pctx, md, NULL, pkey) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
if (lu->sig == EVP_PKEY_RSA_PSS) {
if (EVP_PKEY_CTX_set_rsa_padding(pctx, RSA_PKCS1_PSS_PADDING) <= 0
|| EVP_PKEY_CTX_set_rsa_pss_saltlen(pctx, RSA_PSS_SALTLEN_DIGEST) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_EVP_LIB);
goto err;
}
}
tbslen = construct_key_exchange_tbs(s, &tbs,
s->init_buf->data + paramoffset,
paramlen);
if (tbslen == 0) {
/* SSLfatal() already called */
goto err;
}
rv = EVP_DigestSign(md_ctx, sigbytes1, &siglen, tbs, tbslen);
OPENSSL_free(tbs);
if (rv <= 0 || !WPACKET_sub_allocate_bytes_u16(pkt, siglen, &sigbytes2)
|| sigbytes1 != sigbytes2) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
goto err;
}
}
EVP_MD_CTX_free(md_ctx);
return 1;
err:
#ifndef OPENSSL_NO_DH
EVP_PKEY_free(pkdh);
#endif
#ifndef OPENSSL_NO_EC
OPENSSL_free(encodedPoint);
#endif
EVP_MD_CTX_free(md_ctx);
return 0;
}
int tls_construct_certificate_request(SSL *s, WPACKET *pkt)
{
if (SSL_IS_TLS13(s)) {
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
/* Send random context when doing post-handshake auth */
if (s->post_handshake_auth == SSL_PHA_REQUEST_PENDING) {
OPENSSL_free(s->pha_context);
s->pha_context_len = 32;
if ((s->pha_context = OPENSSL_malloc(s->pha_context_len)) == NULL
|| RAND_bytes(s->pha_context, s->pha_context_len) <= 0
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
|| !WPACKET_sub_memcpy_u8(pkt, s->pha_context, s->pha_context_len)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_CERTIFICATE_REQUEST,
ERR_R_INTERNAL_ERROR);
return 0;
}
/* reset the handshake hash back to just after the ClientFinished */
if (!tls13_restore_handshake_digest_for_pha(s)) {
/* SSLfatal() already called */
return 0;
}
} else {
if (!WPACKET_put_bytes_u8(pkt, 0)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_CERTIFICATE_REQUEST,
ERR_R_INTERNAL_ERROR);
return 0;
}
}
if (!tls_construct_extensions(s, pkt,
SSL_EXT_TLS1_3_CERTIFICATE_REQUEST, NULL,
0)) {
/* SSLfatal() already called */
return 0;
}
goto done;
}
/* get the list of acceptable cert types */
if (!WPACKET_start_sub_packet_u8(pkt)
|| !ssl3_get_req_cert_type(s, pkt) || !WPACKET_close(pkt)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_CERTIFICATE_REQUEST, ERR_R_INTERNAL_ERROR);
return 0;
}
if (SSL_USE_SIGALGS(s)) {
const uint16_t *psigs;
size_t nl = tls12_get_psigalgs(s, 1, &psigs);
if (!WPACKET_start_sub_packet_u16(pkt)
|| !WPACKET_set_flags(pkt, WPACKET_FLAGS_NON_ZERO_LENGTH)
|| !tls12_copy_sigalgs(s, pkt, psigs, nl)
|| !WPACKET_close(pkt)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_CERTIFICATE_REQUEST,
ERR_R_INTERNAL_ERROR);
return 0;
}
}
if (!construct_ca_names(s, get_ca_names(s), pkt)) {
/* SSLfatal() already called */
return 0;
}
done:
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
s->certreqs_sent++;
s->s3->tmp.cert_request = 1;
return 1;
}
static int tls_process_cke_psk_preamble(SSL *s, PACKET *pkt)
{
#ifndef OPENSSL_NO_PSK
unsigned char psk[PSK_MAX_PSK_LEN];
size_t psklen;
PACKET psk_identity;
if (!PACKET_get_length_prefixed_2(pkt, &psk_identity)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CKE_PSK_PREAMBLE,
SSL_R_LENGTH_MISMATCH);
return 0;
}
if (PACKET_remaining(&psk_identity) > PSK_MAX_IDENTITY_LEN) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CKE_PSK_PREAMBLE,
SSL_R_DATA_LENGTH_TOO_LONG);
return 0;
}
if (s->psk_server_callback == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_PSK_PREAMBLE,
SSL_R_PSK_NO_SERVER_CB);
return 0;
}
if (!PACKET_strndup(&psk_identity, &s->session->psk_identity)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_PSK_PREAMBLE,
ERR_R_INTERNAL_ERROR);
return 0;
}
psklen = s->psk_server_callback(s, s->session->psk_identity,
psk, sizeof(psk));
if (psklen > PSK_MAX_PSK_LEN) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_PSK_PREAMBLE,
ERR_R_INTERNAL_ERROR);
return 0;
} else if (psklen == 0) {
/*
* PSK related to the given identity not found
*/
SSLfatal(s, SSL_AD_UNKNOWN_PSK_IDENTITY,
SSL_F_TLS_PROCESS_CKE_PSK_PREAMBLE,
SSL_R_PSK_IDENTITY_NOT_FOUND);
return 0;
}
OPENSSL_free(s->s3->tmp.psk);
s->s3->tmp.psk = OPENSSL_memdup(psk, psklen);
OPENSSL_cleanse(psk, psklen);
if (s->s3->tmp.psk == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_PROCESS_CKE_PSK_PREAMBLE, ERR_R_MALLOC_FAILURE);
return 0;
}
s->s3->tmp.psklen = psklen;
return 1;
#else
/* Should never happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_PSK_PREAMBLE,
ERR_R_INTERNAL_ERROR);
return 0;
#endif
}
static int tls_process_cke_rsa(SSL *s, PACKET *pkt)
{
#ifndef OPENSSL_NO_RSA
unsigned char rand_premaster_secret[SSL_MAX_MASTER_KEY_LENGTH];
int decrypt_len;
unsigned char decrypt_good, version_good;
size_t j, padding_len;
PACKET enc_premaster;
RSA *rsa = NULL;
unsigned char *rsa_decrypt = NULL;
int ret = 0;
rsa = EVP_PKEY_get0_RSA(s->cert->pkeys[SSL_PKEY_RSA].privatekey);
if (rsa == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_RSA,
SSL_R_MISSING_RSA_CERTIFICATE);
return 0;
}
/* SSLv3 and pre-standard DTLS omit the length bytes. */
if (s->version == SSL3_VERSION || s->version == DTLS1_BAD_VER) {
enc_premaster = *pkt;
} else {
if (!PACKET_get_length_prefixed_2(pkt, &enc_premaster)
|| PACKET_remaining(pkt) != 0) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CKE_RSA,
SSL_R_LENGTH_MISMATCH);
return 0;
}
}
/*
* We want to be sure that the plaintext buffer size makes it safe to
* iterate over the entire size of a premaster secret
* (SSL_MAX_MASTER_KEY_LENGTH). Reject overly short RSA keys because
* their ciphertext cannot accommodate a premaster secret anyway.
*/
if (RSA_size(rsa) < SSL_MAX_MASTER_KEY_LENGTH) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_RSA,
RSA_R_KEY_SIZE_TOO_SMALL);
return 0;
}
rsa_decrypt = OPENSSL_malloc(RSA_size(rsa));
if (rsa_decrypt == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_RSA,
ERR_R_MALLOC_FAILURE);
return 0;
}
/*
* We must not leak whether a decryption failure occurs because of
* Bleichenbacher's attack on PKCS #1 v1.5 RSA padding (see RFC 2246,
* section 7.4.7.1). The code follows that advice of the TLS RFC and
* generates a random premaster secret for the case that the decrypt
* fails. See https://tools.ietf.org/html/rfc5246#section-7.4.7.1
*/
if (RAND_priv_bytes(rand_premaster_secret,
sizeof(rand_premaster_secret)) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_RSA,
ERR_R_INTERNAL_ERROR);
goto err;
}
/*
* Decrypt with no padding. PKCS#1 padding will be removed as part of
* the timing-sensitive code below.
*/
/* TODO(size_t): Convert this function */
decrypt_len = (int)RSA_private_decrypt((int)PACKET_remaining(&enc_premaster),
PACKET_data(&enc_premaster),
rsa_decrypt, rsa, RSA_NO_PADDING);
if (decrypt_len < 0) {
SSLfatal(s, SSL_AD_DECRYPT_ERROR, SSL_F_TLS_PROCESS_CKE_RSA,
ERR_R_INTERNAL_ERROR);
goto err;
}
/* Check the padding. See RFC 3447, section 7.2.2. */
/*
* The smallest padded premaster is 11 bytes of overhead. Small keys
* are publicly invalid, so this may return immediately. This ensures
* PS is at least 8 bytes.
*/
if (decrypt_len < 11 + SSL_MAX_MASTER_KEY_LENGTH) {
SSLfatal(s, SSL_AD_DECRYPT_ERROR, SSL_F_TLS_PROCESS_CKE_RSA,
SSL_R_DECRYPTION_FAILED);
goto err;
}
padding_len = decrypt_len - SSL_MAX_MASTER_KEY_LENGTH;
decrypt_good = constant_time_eq_int_8(rsa_decrypt[0], 0) &
constant_time_eq_int_8(rsa_decrypt[1], 2);
for (j = 2; j < padding_len - 1; j++) {
decrypt_good &= ~constant_time_is_zero_8(rsa_decrypt[j]);
}
decrypt_good &= constant_time_is_zero_8(rsa_decrypt[padding_len - 1]);
/*
* If the version in the decrypted pre-master secret is correct then
* version_good will be 0xff, otherwise it'll be zero. The
* Klima-Pokorny-Rosa extension of Bleichenbacher's attack
* (http://eprint.iacr.org/2003/052/) exploits the version number
* check as a "bad version oracle". Thus version checks are done in
* constant time and are treated like any other decryption error.
*/
version_good =
constant_time_eq_8(rsa_decrypt[padding_len],
(unsigned)(s->client_version >> 8));
version_good &=
constant_time_eq_8(rsa_decrypt[padding_len + 1],
(unsigned)(s->client_version & 0xff));
/*
* The premaster secret must contain the same version number as the
* ClientHello to detect version rollback attacks (strangely, the
* protocol does not offer such protection for DH ciphersuites).
* However, buggy clients exist that send the negotiated protocol
* version instead if the server does not support the requested
* protocol version. If SSL_OP_TLS_ROLLBACK_BUG is set, tolerate such
* clients.
*/
if (s->options & SSL_OP_TLS_ROLLBACK_BUG) {
unsigned char workaround_good;
workaround_good = constant_time_eq_8(rsa_decrypt[padding_len],
(unsigned)(s->version >> 8));
workaround_good &=
constant_time_eq_8(rsa_decrypt[padding_len + 1],
(unsigned)(s->version & 0xff));
version_good |= workaround_good;
}
/*
* Both decryption and version must be good for decrypt_good to
* remain non-zero (0xff).
*/
decrypt_good &= version_good;
/*
* Now copy rand_premaster_secret over from p using
* decrypt_good_mask. If decryption failed, then p does not
* contain valid plaintext, however, a check above guarantees
* it is still sufficiently large to read from.
*/
for (j = 0; j < sizeof(rand_premaster_secret); j++) {
rsa_decrypt[padding_len + j] =
constant_time_select_8(decrypt_good,
rsa_decrypt[padding_len + j],
rand_premaster_secret[j]);
}
if (!ssl_generate_master_secret(s, rsa_decrypt + padding_len,
sizeof(rand_premaster_secret), 0)) {
/* SSLfatal() already called */
goto err;
}
ret = 1;
err:
OPENSSL_free(rsa_decrypt);
return ret;
#else
/* Should never happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_RSA,
ERR_R_INTERNAL_ERROR);
return 0;
#endif
}
static int tls_process_cke_dhe(SSL *s, PACKET *pkt)
{
#ifndef OPENSSL_NO_DH
EVP_PKEY *skey = NULL;
DH *cdh;
unsigned int i;
BIGNUM *pub_key;
const unsigned char *data;
EVP_PKEY *ckey = NULL;
int ret = 0;
if (!PACKET_get_net_2(pkt, &i) || PACKET_remaining(pkt) != i) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CKE_DHE,
SSL_R_DH_PUBLIC_VALUE_LENGTH_IS_WRONG);
goto err;
}
skey = s->s3->tmp.pkey;
if (skey == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_DHE,
SSL_R_MISSING_TMP_DH_KEY);
goto err;
}
if (PACKET_remaining(pkt) == 0L) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CKE_DHE,
SSL_R_MISSING_TMP_DH_KEY);
goto err;
}
if (!PACKET_get_bytes(pkt, &data, i)) {
/* We already checked we have enough data */
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_DHE,
ERR_R_INTERNAL_ERROR);
goto err;
}
ckey = EVP_PKEY_new();
if (ckey == NULL || EVP_PKEY_copy_parameters(ckey, skey) == 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_DHE,
SSL_R_BN_LIB);
goto err;
}
cdh = EVP_PKEY_get0_DH(ckey);
pub_key = BN_bin2bn(data, i, NULL);
if (pub_key == NULL || cdh == NULL || !DH_set0_key(cdh, pub_key, NULL)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_DHE,
ERR_R_INTERNAL_ERROR);
BN_free(pub_key);
goto err;
}
if (ssl_derive(s, skey, ckey, 1) == 0) {
/* SSLfatal() already called */
goto err;
}
ret = 1;
EVP_PKEY_free(s->s3->tmp.pkey);
s->s3->tmp.pkey = NULL;
err:
EVP_PKEY_free(ckey);
return ret;
#else
/* Should never happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_DHE,
ERR_R_INTERNAL_ERROR);
return 0;
#endif
}
static int tls_process_cke_ecdhe(SSL *s, PACKET *pkt)
{
#ifndef OPENSSL_NO_EC
EVP_PKEY *skey = s->s3->tmp.pkey;
EVP_PKEY *ckey = NULL;
int ret = 0;
if (PACKET_remaining(pkt) == 0L) {
/* We don't support ECDH client auth */
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE, SSL_F_TLS_PROCESS_CKE_ECDHE,
SSL_R_MISSING_TMP_ECDH_KEY);
goto err;
} else {
unsigned int i;
const unsigned char *data;
/*
* Get client's public key from encoded point in the
* ClientKeyExchange message.
*/
/* Get encoded point length */
if (!PACKET_get_1(pkt, &i) || !PACKET_get_bytes(pkt, &data, i)
|| PACKET_remaining(pkt) != 0) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CKE_ECDHE,
SSL_R_LENGTH_MISMATCH);
goto err;
}
if (skey == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_ECDHE,
SSL_R_MISSING_TMP_ECDH_KEY);
goto err;
}
ckey = EVP_PKEY_new();
if (ckey == NULL || EVP_PKEY_copy_parameters(ckey, skey) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_ECDHE,
ERR_R_EVP_LIB);
goto err;
}
if (EVP_PKEY_set1_tls_encodedpoint(ckey, data, i) == 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_ECDHE,
ERR_R_EC_LIB);
goto err;
}
}
if (ssl_derive(s, skey, ckey, 1) == 0) {
/* SSLfatal() already called */
goto err;
}
ret = 1;
EVP_PKEY_free(s->s3->tmp.pkey);
s->s3->tmp.pkey = NULL;
err:
EVP_PKEY_free(ckey);
return ret;
#else
/* Should never happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_ECDHE,
ERR_R_INTERNAL_ERROR);
return 0;
#endif
}
static int tls_process_cke_srp(SSL *s, PACKET *pkt)
{
#ifndef OPENSSL_NO_SRP
unsigned int i;
const unsigned char *data;
if (!PACKET_get_net_2(pkt, &i)
|| !PACKET_get_bytes(pkt, &data, i)) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CKE_SRP,
SSL_R_BAD_SRP_A_LENGTH);
return 0;
}
if ((s->srp_ctx.A = BN_bin2bn(data, i, NULL)) == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_SRP,
ERR_R_BN_LIB);
return 0;
}
if (BN_ucmp(s->srp_ctx.A, s->srp_ctx.N) >= 0 || BN_is_zero(s->srp_ctx.A)) {
SSLfatal(s, SSL_AD_ILLEGAL_PARAMETER, SSL_F_TLS_PROCESS_CKE_SRP,
SSL_R_BAD_SRP_PARAMETERS);
return 0;
}
OPENSSL_free(s->session->srp_username);
s->session->srp_username = OPENSSL_strdup(s->srp_ctx.login);
if (s->session->srp_username == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_SRP,
ERR_R_MALLOC_FAILURE);
return 0;
}
if (!srp_generate_server_master_secret(s)) {
/* SSLfatal() already called */
return 0;
}
return 1;
#else
/* Should never happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_SRP,
ERR_R_INTERNAL_ERROR);
return 0;
#endif
}
static int tls_process_cke_gost(SSL *s, PACKET *pkt)
{
#ifndef OPENSSL_NO_GOST
EVP_PKEY_CTX *pkey_ctx;
EVP_PKEY *client_pub_pkey = NULL, *pk = NULL;
unsigned char premaster_secret[32];
const unsigned char *start;
size_t outlen = 32, inlen;
unsigned long alg_a;
GOST_KX_MESSAGE *pKX = NULL;
const unsigned char *ptr;
int ret = 0;
/* Get our certificate private key */
alg_a = s->s3->tmp.new_cipher->algorithm_auth;
if (alg_a & SSL_aGOST12) {
/*
* New GOST ciphersuites have SSL_aGOST01 bit too
*/
pk = s->cert->pkeys[SSL_PKEY_GOST12_512].privatekey;
if (pk == NULL) {
pk = s->cert->pkeys[SSL_PKEY_GOST12_256].privatekey;
}
if (pk == NULL) {
pk = s->cert->pkeys[SSL_PKEY_GOST01].privatekey;
}
} else if (alg_a & SSL_aGOST01) {
pk = s->cert->pkeys[SSL_PKEY_GOST01].privatekey;
}
pkey_ctx = EVP_PKEY_CTX_new(pk, NULL);
if (pkey_ctx == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_GOST,
ERR_R_MALLOC_FAILURE);
return 0;
}
if (EVP_PKEY_decrypt_init(pkey_ctx) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_GOST,
ERR_R_INTERNAL_ERROR);
return 0;
}
/*
* If client certificate is present and is of the same type, maybe
* use it for key exchange. Don't mind errors from
* EVP_PKEY_derive_set_peer, because it is completely valid to use a
* client certificate for authorization only.
*/
client_pub_pkey = X509_get0_pubkey(s->session->peer);
if (client_pub_pkey) {
if (EVP_PKEY_derive_set_peer(pkey_ctx, client_pub_pkey) <= 0)
ERR_clear_error();
}
ptr = PACKET_data(pkt);
/* Some implementations provide extra data in the opaqueBlob
* We have nothing to do with this blob so we just skip it */
pKX = d2i_GOST_KX_MESSAGE(NULL, &ptr, PACKET_remaining(pkt));
if (pKX == NULL
|| pKX->kxBlob == NULL
|| ASN1_TYPE_get(pKX->kxBlob) != V_ASN1_SEQUENCE) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CKE_GOST,
SSL_R_DECRYPTION_FAILED);
goto err;
}
if (!PACKET_forward(pkt, ptr - PACKET_data(pkt))) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_GOST,
SSL_R_DECRYPTION_FAILED);
goto err;
}
if (PACKET_remaining(pkt) != 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_GOST,
SSL_R_DECRYPTION_FAILED);
goto err;
}
inlen = pKX->kxBlob->value.sequence->length;
start = pKX->kxBlob->value.sequence->data;
if (EVP_PKEY_decrypt(pkey_ctx, premaster_secret, &outlen, start,
inlen) <= 0) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CKE_GOST,
SSL_R_DECRYPTION_FAILED);
goto err;
}
/* Generate master secret */
if (!ssl_generate_master_secret(s, premaster_secret,
sizeof(premaster_secret), 0)) {
/* SSLfatal() already called */
goto err;
}
/* Check if pubkey from client certificate was used */
if (EVP_PKEY_CTX_ctrl(pkey_ctx, -1, -1, EVP_PKEY_CTRL_PEER_KEY, 2,
NULL) > 0)
s->statem.no_cert_verify = 1;
ret = 1;
err:
EVP_PKEY_CTX_free(pkey_ctx);
GOST_KX_MESSAGE_free(pKX);
return ret;
#else
/* Should never happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CKE_GOST,
ERR_R_INTERNAL_ERROR);
return 0;
#endif
}
MSG_PROCESS_RETURN tls_process_client_key_exchange(SSL *s, PACKET *pkt)
{
unsigned long alg_k;
alg_k = s->s3->tmp.new_cipher->algorithm_mkey;
/* For PSK parse and retrieve identity, obtain PSK key */
if ((alg_k & SSL_PSK) && !tls_process_cke_psk_preamble(s, pkt)) {
/* SSLfatal() already called */
goto err;
}
if (alg_k & SSL_kPSK) {
/* Identity extracted earlier: should be nothing left */
if (PACKET_remaining(pkt) != 0) {
SSLfatal(s, SSL_AD_DECODE_ERROR,
SSL_F_TLS_PROCESS_CLIENT_KEY_EXCHANGE,
SSL_R_LENGTH_MISMATCH);
goto err;
}
/* PSK handled by ssl_generate_master_secret */
if (!ssl_generate_master_secret(s, NULL, 0, 0)) {
/* SSLfatal() already called */
goto err;
}
} else if (alg_k & (SSL_kRSA | SSL_kRSAPSK)) {
if (!tls_process_cke_rsa(s, pkt)) {
/* SSLfatal() already called */
goto err;
}
} else if (alg_k & (SSL_kDHE | SSL_kDHEPSK)) {
if (!tls_process_cke_dhe(s, pkt)) {
/* SSLfatal() already called */
goto err;
}
} else if (alg_k & (SSL_kECDHE | SSL_kECDHEPSK)) {
if (!tls_process_cke_ecdhe(s, pkt)) {
/* SSLfatal() already called */
goto err;
}
} else if (alg_k & SSL_kSRP) {
if (!tls_process_cke_srp(s, pkt)) {
/* SSLfatal() already called */
goto err;
}
} else if (alg_k & SSL_kGOST) {
if (!tls_process_cke_gost(s, pkt)) {
/* SSLfatal() already called */
goto err;
}
} else {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_PROCESS_CLIENT_KEY_EXCHANGE,
SSL_R_UNKNOWN_CIPHER_TYPE);
goto err;
}
return MSG_PROCESS_CONTINUE_PROCESSING;
err:
#ifndef OPENSSL_NO_PSK
OPENSSL_clear_free(s->s3->tmp.psk, s->s3->tmp.psklen);
s->s3->tmp.psk = NULL;
#endif
return MSG_PROCESS_ERROR;
}
WORK_STATE tls_post_process_client_key_exchange(SSL *s, WORK_STATE wst)
{
#ifndef OPENSSL_NO_SCTP
if (wst == WORK_MORE_A) {
if (SSL_IS_DTLS(s)) {
unsigned char sctpauthkey[64];
char labelbuffer[sizeof(DTLS1_SCTP_AUTH_LABEL)];
size_t labellen;
/*
* Add new shared key for SCTP-Auth, will be ignored if no SCTP
* used.
*/
memcpy(labelbuffer, DTLS1_SCTP_AUTH_LABEL,
sizeof(DTLS1_SCTP_AUTH_LABEL));
/* Don't include the terminating zero. */
labellen = sizeof(labelbuffer) - 1;
if (s->mode & SSL_MODE_DTLS_SCTP_LABEL_LENGTH_BUG)
labellen += 1;
if (SSL_export_keying_material(s, sctpauthkey,
sizeof(sctpauthkey), labelbuffer,
labellen, NULL, 0,
0) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_POST_PROCESS_CLIENT_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
return WORK_ERROR;
}
BIO_ctrl(SSL_get_wbio(s), BIO_CTRL_DGRAM_SCTP_ADD_AUTH_KEY,
sizeof(sctpauthkey), sctpauthkey);
}
}
#endif
if (s->statem.no_cert_verify || !s->session->peer) {
/*
* No certificate verify or no peer certificate so we no longer need
* the handshake_buffer
*/
if (!ssl3_digest_cached_records(s, 0)) {
/* SSLfatal() already called */
return WORK_ERROR;
}
return WORK_FINISHED_CONTINUE;
} else {
if (!s->s3->handshake_buffer) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_POST_PROCESS_CLIENT_KEY_EXCHANGE,
ERR_R_INTERNAL_ERROR);
return WORK_ERROR;
}
/*
* For sigalgs freeze the handshake buffer. If we support
* extms we've done this already so this is a no-op
*/
if (!ssl3_digest_cached_records(s, 1)) {
/* SSLfatal() already called */
return WORK_ERROR;
}
}
return WORK_FINISHED_CONTINUE;
}
MSG_PROCESS_RETURN tls_process_client_certificate(SSL *s, PACKET *pkt)
{
int i;
MSG_PROCESS_RETURN ret = MSG_PROCESS_ERROR;
X509 *x = NULL;
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
unsigned long l;
const unsigned char *certstart, *certbytes;
STACK_OF(X509) *sk = NULL;
PACKET spkt, context;
size_t chainidx;
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
SSL_SESSION *new_sess = NULL;
/*
* To get this far we must have read encrypted data from the client. We no
* longer tolerate unencrypted alerts. This value is ignored if less than
* TLSv1.3
*/
s->statem.enc_read_state = ENC_READ_STATE_VALID;
if ((sk = sk_X509_new_null()) == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
ERR_R_MALLOC_FAILURE);
goto err;
}
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
if (SSL_IS_TLS13(s) && (!PACKET_get_length_prefixed_1(pkt, &context)
|| (s->pha_context == NULL && PACKET_remaining(&context) != 0)
|| (s->pha_context != NULL &&
!PACKET_equal(&context, s->pha_context, s->pha_context_len)))) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
SSL_R_INVALID_CONTEXT);
goto err;
}
if (!PACKET_get_length_prefixed_3(pkt, &spkt)
|| PACKET_remaining(pkt) != 0) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
SSL_R_LENGTH_MISMATCH);
goto err;
}
for (chainidx = 0; PACKET_remaining(&spkt) > 0; chainidx++) {
if (!PACKET_get_net_3(&spkt, &l)
|| !PACKET_get_bytes(&spkt, &certbytes, l)) {
SSLfatal(s, SSL_AD_DECODE_ERROR,
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
SSL_R_CERT_LENGTH_MISMATCH);
goto err;
}
certstart = certbytes;
x = d2i_X509(NULL, (const unsigned char **)&certbytes, l);
if (x == NULL) {
SSLfatal(s, SSL_AD_DECODE_ERROR,
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE, ERR_R_ASN1_LIB);
goto err;
}
if (certbytes != (certstart + l)) {
SSLfatal(s, SSL_AD_DECODE_ERROR,
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
SSL_R_CERT_LENGTH_MISMATCH);
goto err;
}
if (SSL_IS_TLS13(s)) {
RAW_EXTENSION *rawexts = NULL;
PACKET extensions;
if (!PACKET_get_length_prefixed_2(&spkt, &extensions)) {
SSLfatal(s, SSL_AD_DECODE_ERROR,
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
SSL_R_BAD_LENGTH);
goto err;
}
if (!tls_collect_extensions(s, &extensions,
SSL_EXT_TLS1_3_CERTIFICATE, &rawexts,
NULL, chainidx == 0)
|| !tls_parse_all_extensions(s, SSL_EXT_TLS1_3_CERTIFICATE,
rawexts, x, chainidx,
PACKET_remaining(&spkt) == 0)) {
OPENSSL_free(rawexts);
goto err;
}
OPENSSL_free(rawexts);
}
if (!sk_X509_push(sk, x)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
ERR_R_MALLOC_FAILURE);
goto err;
}
x = NULL;
}
if (sk_X509_num(sk) <= 0) {
/* TLS does not mind 0 certs returned */
if (s->version == SSL3_VERSION) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
SSL_R_NO_CERTIFICATES_RETURNED);
goto err;
}
/* Fail for TLS only if we required a certificate */
else if ((s->verify_mode & SSL_VERIFY_PEER) &&
(s->verify_mode & SSL_VERIFY_FAIL_IF_NO_PEER_CERT)) {
SSLfatal(s, SSL_AD_CERTIFICATE_REQUIRED,
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
SSL_R_PEER_DID_NOT_RETURN_A_CERTIFICATE);
goto err;
}
/* No client certificate so digest cached records */
if (s->s3->handshake_buffer && !ssl3_digest_cached_records(s, 0)) {
/* SSLfatal() already called */
goto err;
}
} else {
EVP_PKEY *pkey;
i = ssl_verify_cert_chain(s, sk);
if (i <= 0) {
SSLfatal(s, ssl_x509err2alert(s->verify_result),
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
SSL_R_CERTIFICATE_VERIFY_FAILED);
goto err;
}
if (i > 1) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE, i);
goto err;
}
pkey = X509_get0_pubkey(sk_X509_value(sk, 0));
if (pkey == NULL) {
SSLfatal(s, SSL_AD_HANDSHAKE_FAILURE,
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
SSL_R_UNKNOWN_CERTIFICATE_TYPE);
goto err;
}
}
Add TLSv1.3 post-handshake authentication (PHA) Add SSL_verify_client_post_handshake() for servers to initiate PHA Add SSL_force_post_handshake_auth() for clients that don't have certificates initially configured, but use a certificate callback. Update SSL_CTX_set_verify()/SSL_set_verify() mode: * Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after the initial handshake. * Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless of when the certificate authentication takes place; either initial handshake, re-negotiation, or post-handshake authentication. Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options Add support to s_client: * Enabled automatically when cert is configured * Can be forced enabled via -force_pha Add support to s_server: * Use 'c' to invoke PHA in s_server * Remove some dead code Update documentation Update unit tests: * Illegal use of PHA extension * TLSv1.3 certificate tests DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is implemented, it's PHA support state machine may need to be different. Add a TODO and a #error Update handshake context to deal with PHA. The handshake context for TLSv1.3 post-handshake auth is up through the ClientFinish message, plus the CertificateRequest message. Subsequent Certificate, CertificateVerify, and Finish messages are based on this handshake context (not the Certificate message per se, but it's included after the hash). KeyUpdate, NewSessionTicket, and prior Certificate Request messages are not included in post-handshake authentication. After the ClientFinished message is processed, save off the digest state for future post-handshake authentication. When post-handshake auth occurs, copy over the saved handshake context into the "main" handshake digest. This effectively discards the any KeyUpdate or NewSessionTicket messages and any prior post-handshake authentication. This, of course, assumes that the ID-22 did not mean to include any previous post-handshake authentication into the new handshake transcript. This is implied by section 4.4.1 that lists messages only up to the first ClientFinished. Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/4964)
2017-12-18 21:52:28 +00:00
/*
* Sessions must be immutable once they go into the session cache. Otherwise
* we can get multi-thread problems. Therefore we don't "update" sessions,
* we replace them with a duplicate. Here, we need to do this every time
* a new certificate is received via post-handshake authentication, as the
* session may have already gone into the session cache.
*/
if (s->post_handshake_auth == SSL_PHA_REQUESTED) {
if ((new_sess = ssl_session_dup(s->session, 0)) == 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_PROCESS_CLIENT_CERTIFICATE,
ERR_R_MALLOC_FAILURE);
goto err;
}
SSL_SESSION_free(s->session);
s->session = new_sess;
}
X509_free(s->session->peer);
s->session->peer = sk_X509_shift(sk);
s->session->verify_result = s->verify_result;
sk_X509_pop_free(s->session->peer_chain, X509_free);
s->session->peer_chain = sk;
/*
* Freeze the handshake buffer. For <TLS1.3 we do this after the CKE
* message
*/
if (SSL_IS_TLS13(s) && !ssl3_digest_cached_records(s, 1)) {
/* SSLfatal() already called */
goto err;
}
/*
* Inconsistency alert: cert_chain does *not* include the peer's own
* certificate, while we do include it in statem_clnt.c
*/
sk = NULL;
/* Save the current hash state for when we receive the CertificateVerify */
if (SSL_IS_TLS13(s)) {
if (!ssl_handshake_hash(s, s->cert_verify_hash,
sizeof(s->cert_verify_hash),
&s->cert_verify_hash_len)) {
/* SSLfatal() already called */
goto err;
}
/* Resend session tickets */
s->sent_tickets = 0;
}
ret = MSG_PROCESS_CONTINUE_READING;
err:
X509_free(x);
sk_X509_pop_free(sk, X509_free);
return ret;
}
int tls_construct_server_certificate(SSL *s, WPACKET *pkt)
{
CERT_PKEY *cpk = s->s3->tmp.cert;
if (cpk == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_CERTIFICATE, ERR_R_INTERNAL_ERROR);
return 0;
}
/*
* In TLSv1.3 the certificate chain is always preceded by a 0 length context
* for the server Certificate message
*/
if (SSL_IS_TLS13(s) && !WPACKET_put_bytes_u8(pkt, 0)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_SERVER_CERTIFICATE, ERR_R_INTERNAL_ERROR);
return 0;
}
if (!ssl3_output_cert_chain(s, pkt, cpk)) {
/* SSLfatal() already called */
return 0;
}
return 1;
}
static int create_ticket_prequel(SSL *s, WPACKET *pkt, uint32_t age_add,
unsigned char *tick_nonce)
{
/*
* Ticket lifetime hint: For TLSv1.2 this is advisory only and we leave this
* unspecified for resumed session (for simplicity).
* In TLSv1.3 we reset the "time" field above, and always specify the
* timeout.
*/
if (!WPACKET_put_bytes_u32(pkt,
(s->hit && !SSL_IS_TLS13(s))
? 0 : s->session->timeout)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CREATE_TICKET_PREQUEL,
ERR_R_INTERNAL_ERROR);
return 0;
}
if (SSL_IS_TLS13(s)) {
if (!WPACKET_put_bytes_u32(pkt, age_add)
|| !WPACKET_sub_memcpy_u8(pkt, tick_nonce, TICKET_NONCE_SIZE)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CREATE_TICKET_PREQUEL,
ERR_R_INTERNAL_ERROR);
return 0;
}
}
/* Start the sub-packet for the actual ticket data */
if (!WPACKET_start_sub_packet_u16(pkt)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CREATE_TICKET_PREQUEL,
ERR_R_INTERNAL_ERROR);
return 0;
}
return 1;
}
static int construct_stateless_ticket(SSL *s, WPACKET *pkt, uint32_t age_add,
unsigned char *tick_nonce)
{
unsigned char *senc = NULL;
EVP_CIPHER_CTX *ctx = NULL;
HMAC_CTX *hctx = NULL;
unsigned char *p, *encdata1, *encdata2, *macdata1, *macdata2;
const unsigned char *const_p;
int len, slen_full, slen, lenfinal;
SSL_SESSION *sess;
unsigned int hlen;
SSL_CTX *tctx = s->session_ctx;
unsigned char iv[EVP_MAX_IV_LENGTH];
unsigned char key_name[TLSEXT_KEYNAME_LENGTH];
int iv_len, ok = 0;
size_t macoffset, macendoffset;
/* get session encoding length */
slen_full = i2d_SSL_SESSION(s->session, NULL);
/*
* Some length values are 16 bits, so forget it if session is too
* long
*/
if (slen_full == 0 || slen_full > 0xFF00) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CONSTRUCT_STATELESS_TICKET,
ERR_R_INTERNAL_ERROR);
goto err;
}
senc = OPENSSL_malloc(slen_full);
if (senc == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_CONSTRUCT_STATELESS_TICKET, ERR_R_MALLOC_FAILURE);
goto err;
}
ctx = EVP_CIPHER_CTX_new();
hctx = HMAC_CTX_new();
if (ctx == NULL || hctx == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CONSTRUCT_STATELESS_TICKET,
ERR_R_MALLOC_FAILURE);
goto err;
}
p = senc;
if (!i2d_SSL_SESSION(s->session, &p)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CONSTRUCT_STATELESS_TICKET,
ERR_R_INTERNAL_ERROR);
goto err;
}
/*
* create a fresh copy (not shared with other threads) to clean up
*/
const_p = senc;
sess = d2i_SSL_SESSION(NULL, &const_p, slen_full);
if (sess == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CONSTRUCT_STATELESS_TICKET,
ERR_R_INTERNAL_ERROR);
goto err;
}
slen = i2d_SSL_SESSION(sess, NULL);
if (slen == 0 || slen > slen_full) {
/* shouldn't ever happen */
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CONSTRUCT_STATELESS_TICKET,
ERR_R_INTERNAL_ERROR);
SSL_SESSION_free(sess);
goto err;
}
p = senc;
if (!i2d_SSL_SESSION(sess, &p)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CONSTRUCT_STATELESS_TICKET,
ERR_R_INTERNAL_ERROR);
SSL_SESSION_free(sess);
goto err;
}
SSL_SESSION_free(sess);
/*
* Initialize HMAC and cipher contexts. If callback present it does
* all the work otherwise use generated values from parent ctx.
*/
if (tctx->ext.ticket_key_cb) {
/* if 0 is returned, write an empty ticket */
int ret = tctx->ext.ticket_key_cb(s, key_name, iv, ctx,
hctx, 1);
if (ret == 0) {
/* Put timeout and length */
if (!WPACKET_put_bytes_u32(pkt, 0)
|| !WPACKET_put_bytes_u16(pkt, 0)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_CONSTRUCT_STATELESS_TICKET,
ERR_R_INTERNAL_ERROR);
goto err;
}
OPENSSL_free(senc);
EVP_CIPHER_CTX_free(ctx);
HMAC_CTX_free(hctx);
return 1;
}
if (ret < 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CONSTRUCT_STATELESS_TICKET,
SSL_R_CALLBACK_FAILED);
goto err;
}
iv_len = EVP_CIPHER_CTX_iv_length(ctx);
} else {
const EVP_CIPHER *cipher = EVP_aes_256_cbc();
iv_len = EVP_CIPHER_iv_length(cipher);
if (RAND_bytes(iv, iv_len) <= 0
|| !EVP_EncryptInit_ex(ctx, cipher, NULL,
tctx->ext.secure->tick_aes_key, iv)
|| !HMAC_Init_ex(hctx, tctx->ext.secure->tick_hmac_key,
sizeof(tctx->ext.secure->tick_hmac_key),
EVP_sha256(), NULL)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CONSTRUCT_STATELESS_TICKET,
ERR_R_INTERNAL_ERROR);
goto err;
}
memcpy(key_name, tctx->ext.tick_key_name,
sizeof(tctx->ext.tick_key_name));
}
if (!create_ticket_prequel(s, pkt, age_add, tick_nonce)) {
/* SSLfatal() already called */
goto err;
}
if (!WPACKET_get_total_written(pkt, &macoffset)
/* Output key name */
|| !WPACKET_memcpy(pkt, key_name, sizeof(key_name))
/* output IV */
|| !WPACKET_memcpy(pkt, iv, iv_len)
|| !WPACKET_reserve_bytes(pkt, slen + EVP_MAX_BLOCK_LENGTH,
&encdata1)
/* Encrypt session data */
|| !EVP_EncryptUpdate(ctx, encdata1, &len, senc, slen)
|| !WPACKET_allocate_bytes(pkt, len, &encdata2)
|| encdata1 != encdata2
|| !EVP_EncryptFinal(ctx, encdata1 + len, &lenfinal)
|| !WPACKET_allocate_bytes(pkt, lenfinal, &encdata2)
|| encdata1 + len != encdata2
|| len + lenfinal > slen + EVP_MAX_BLOCK_LENGTH
|| !WPACKET_get_total_written(pkt, &macendoffset)
|| !HMAC_Update(hctx,
(unsigned char *)s->init_buf->data + macoffset,
macendoffset - macoffset)
|| !WPACKET_reserve_bytes(pkt, EVP_MAX_MD_SIZE, &macdata1)
|| !HMAC_Final(hctx, macdata1, &hlen)
|| hlen > EVP_MAX_MD_SIZE
|| !WPACKET_allocate_bytes(pkt, hlen, &macdata2)
|| macdata1 != macdata2) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_CONSTRUCT_STATELESS_TICKET, ERR_R_INTERNAL_ERROR);
goto err;
}
/* Close the sub-packet created by create_ticket_prequel() */
if (!WPACKET_close(pkt)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CONSTRUCT_STATELESS_TICKET,
ERR_R_INTERNAL_ERROR);
goto err;
}
ok = 1;
err:
OPENSSL_free(senc);
EVP_CIPHER_CTX_free(ctx);
HMAC_CTX_free(hctx);
return ok;
}
static int construct_stateful_ticket(SSL *s, WPACKET *pkt, uint32_t age_add,
unsigned char *tick_nonce)
{
if (!create_ticket_prequel(s, pkt, age_add, tick_nonce)) {
/* SSLfatal() already called */
return 0;
}
if (!WPACKET_memcpy(pkt, s->session->session_id,
s->session->session_id_length)
|| !WPACKET_close(pkt)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_CONSTRUCT_STATEFUL_TICKET,
ERR_R_INTERNAL_ERROR);
return 0;
}
return 1;
}
int tls_construct_new_session_ticket(SSL *s, WPACKET *pkt)
{
SSL_CTX *tctx = s->session_ctx;
unsigned char tick_nonce[TICKET_NONCE_SIZE];
union {
unsigned char age_add_c[sizeof(uint32_t)];
uint32_t age_add;
} age_add_u;
age_add_u.age_add = 0;
if (SSL_IS_TLS13(s)) {
size_t i, hashlen;
uint64_t nonce;
static const unsigned char nonce_label[] = "resumption";
const EVP_MD *md = ssl_handshake_md(s);
int hashleni = EVP_MD_size(md);
/* Ensure cast to size_t is safe */
if (!ossl_assert(hashleni >= 0)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_NEW_SESSION_TICKET,
ERR_R_INTERNAL_ERROR);
goto err;
}
hashlen = (size_t)hashleni;
/*
* If we already sent one NewSessionTicket, or we resumed then
* s->session may already be in a cache and so we must not modify it.
* Instead we need to take a copy of it and modify that.
*/
if (s->sent_tickets != 0 || s->hit) {
SSL_SESSION *new_sess = ssl_session_dup(s->session, 0);
if (new_sess == NULL) {
/* SSLfatal already called */
goto err;
}
SSL_SESSION_free(s->session);
s->session = new_sess;
}
if (!ssl_generate_session_id(s, s->session)) {
/* SSLfatal() already called */
goto err;
}
if (RAND_bytes(age_add_u.age_add_c, sizeof(age_add_u)) <= 0) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_NEW_SESSION_TICKET,
ERR_R_INTERNAL_ERROR);
goto err;
}
s->session->ext.tick_age_add = age_add_u.age_add;
nonce = s->next_ticket_nonce;
for (i = TICKET_NONCE_SIZE; i > 0; i--) {
tick_nonce[i - 1] = (unsigned char)(nonce & 0xff);
nonce >>= 8;
}
if (!tls13_hkdf_expand(s, md, s->resumption_master_secret,
nonce_label,
sizeof(nonce_label) - 1,
tick_nonce,
TICKET_NONCE_SIZE,
s->session->master_key,
hashlen, 1)) {
/* SSLfatal() already called */
goto err;
}
s->session->master_key_length = hashlen;
s->session->time = (long)time(NULL);
if (s->s3->alpn_selected != NULL) {
OPENSSL_free(s->session->ext.alpn_selected);
s->session->ext.alpn_selected =
OPENSSL_memdup(s->s3->alpn_selected, s->s3->alpn_selected_len);
if (s->session->ext.alpn_selected == NULL) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR,
SSL_F_TLS_CONSTRUCT_NEW_SESSION_TICKET,
ERR_R_MALLOC_FAILURE);
goto err;
}
s->session->ext.alpn_selected_len = s->s3->alpn_selected_len;
}
s->session->ext.max_early_data = s->max_early_data;
}
if (tctx->generate_ticket_cb != NULL &&
tctx->generate_ticket_cb(s, tctx->ticket_cb_data) == 0)
goto err;
/*
* If we are using anti-replay protection then we behave as if
* SSL_OP_NO_TICKET is set - we are caching tickets anyway so there
* is no point in using full stateless tickets.
*/
if (SSL_IS_TLS13(s)
&& ((s->options & SSL_OP_NO_TICKET) != 0
|| (s->max_early_data > 0
&& (s->options & SSL_OP_NO_ANTI_REPLAY) == 0))) {
if (!construct_stateful_ticket(s, pkt, age_add_u.age_add, tick_nonce)) {
/* SSLfatal() already called */
goto err;
}
} else if (!construct_stateless_ticket(s, pkt, age_add_u.age_add,
tick_nonce)) {
/* SSLfatal() already called */
goto err;
}
if (SSL_IS_TLS13(s)) {
if (!tls_construct_extensions(s, pkt,
SSL_EXT_TLS1_3_NEW_SESSION_TICKET,
NULL, 0)) {
/* SSLfatal() already called */
goto err;
}
/*
* Increment both |sent_tickets| and |next_ticket_nonce|. |sent_tickets|
* gets reset to 0 if we send more tickets following a post-handshake
* auth, but |next_ticket_nonce| does not.
*/
s->sent_tickets++;
s->next_ticket_nonce++;
ssl_update_cache(s, SSL_SESS_CACHE_SERVER);
}
return 1;
err:
return 0;
}
/*
* In TLSv1.3 this is called from the extensions code, otherwise it is used to
* create a separate message. Returns 1 on success or 0 on failure.
*/
int tls_construct_cert_status_body(SSL *s, WPACKET *pkt)
{
if (!WPACKET_put_bytes_u8(pkt, s->ext.status_type)
|| !WPACKET_sub_memcpy_u24(pkt, s->ext.ocsp.resp,
s->ext.ocsp.resp_len)) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_CONSTRUCT_CERT_STATUS_BODY,
ERR_R_INTERNAL_ERROR);
return 0;
}
return 1;
}
int tls_construct_cert_status(SSL *s, WPACKET *pkt)
{
if (!tls_construct_cert_status_body(s, pkt)) {
/* SSLfatal() already called */
return 0;
}
return 1;
}
#ifndef OPENSSL_NO_NEXTPROTONEG
/*
* tls_process_next_proto reads a Next Protocol Negotiation handshake message.
* It sets the next_proto member in s if found
*/
MSG_PROCESS_RETURN tls_process_next_proto(SSL *s, PACKET *pkt)
{
PACKET next_proto, padding;
size_t next_proto_len;
/*-
* The payload looks like:
* uint8 proto_len;
* uint8 proto[proto_len];
* uint8 padding_len;
* uint8 padding[padding_len];
*/
if (!PACKET_get_length_prefixed_1(pkt, &next_proto)
|| !PACKET_get_length_prefixed_1(pkt, &padding)
|| PACKET_remaining(pkt) > 0) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_NEXT_PROTO,
SSL_R_LENGTH_MISMATCH);
return MSG_PROCESS_ERROR;
}
if (!PACKET_memdup(&next_proto, &s->ext.npn, &next_proto_len)) {
s->ext.npn_len = 0;
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_NEXT_PROTO,
ERR_R_INTERNAL_ERROR);
return MSG_PROCESS_ERROR;
}
s->ext.npn_len = (unsigned char)next_proto_len;
return MSG_PROCESS_CONTINUE_READING;
}
#endif
static int tls_construct_encrypted_extensions(SSL *s, WPACKET *pkt)
{
if (!tls_construct_extensions(s, pkt, SSL_EXT_TLS1_3_ENCRYPTED_EXTENSIONS,
NULL, 0)) {
/* SSLfatal() already called */
return 0;
}
return 1;
}
MSG_PROCESS_RETURN tls_process_end_of_early_data(SSL *s, PACKET *pkt)
{
if (PACKET_remaining(pkt) != 0) {
SSLfatal(s, SSL_AD_DECODE_ERROR, SSL_F_TLS_PROCESS_END_OF_EARLY_DATA,
SSL_R_LENGTH_MISMATCH);
return MSG_PROCESS_ERROR;
}
if (s->early_data_state != SSL_EARLY_DATA_READING
&& s->early_data_state != SSL_EARLY_DATA_READ_RETRY) {
SSLfatal(s, SSL_AD_INTERNAL_ERROR, SSL_F_TLS_PROCESS_END_OF_EARLY_DATA,
ERR_R_INTERNAL_ERROR);
return MSG_PROCESS_ERROR;
}
/*
* EndOfEarlyData signals a key change so the end of the message must be on
* a record boundary.
*/
if (RECORD_LAYER_processed_read_pending(&s->rlayer)) {
SSLfatal(s, SSL_AD_UNEXPECTED_MESSAGE,
SSL_F_TLS_PROCESS_END_OF_EARLY_DATA,
SSL_R_NOT_ON_RECORD_BOUNDARY);
return MSG_PROCESS_ERROR;
}
s->early_data_state = SSL_EARLY_DATA_FINISHED_READING;
if (!s->method->ssl3_enc->change_cipher_state(s,
SSL3_CC_HANDSHAKE | SSL3_CHANGE_CIPHER_SERVER_READ)) {
/* SSLfatal() already called */
return MSG_PROCESS_ERROR;
}
return MSG_PROCESS_CONTINUE_READING;
}