openssl/test/handshake_helper.c

1773 lines
60 KiB
C
Raw Normal View History

/*
* Copyright 2016-2019 The OpenSSL Project Authors. 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 <string.h>
#include <openssl/bio.h>
#include <openssl/x509_vfy.h>
#include <openssl/ssl.h>
#ifndef OPENSSL_NO_SRP
#include <openssl/srp.h>
#endif
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
#include "../ssl/ssl_locl.h"
#include "internal/sockets.h"
#include "internal/nelem.h"
#include "handshake_helper.h"
#include "testutil.h"
#if !defined(OPENSSL_NO_SCTP) && !defined(OPENSSL_NO_SOCK)
#include <netinet/sctp.h>
#endif
HANDSHAKE_RESULT *HANDSHAKE_RESULT_new(void)
{
HANDSHAKE_RESULT *ret;
TEST_ptr(ret = OPENSSL_zalloc(sizeof(*ret)));
return ret;
}
void HANDSHAKE_RESULT_free(HANDSHAKE_RESULT *result)
{
if (result == NULL)
return;
OPENSSL_free(result->client_npn_negotiated);
OPENSSL_free(result->server_npn_negotiated);
OPENSSL_free(result->client_alpn_negotiated);
OPENSSL_free(result->server_alpn_negotiated);
OPENSSL_free(result->result_session_ticket_app_data);
sk_X509_NAME_pop_free(result->server_ca_names, X509_NAME_free);
sk_X509_NAME_pop_free(result->client_ca_names, X509_NAME_free);
OPENSSL_free(result->cipher);
OPENSSL_free(result);
}
/*
* Since there appears to be no way to extract the sent/received alert
* from the SSL object directly, we use the info callback and stash
* the result in ex_data.
*/
typedef struct handshake_ex_data_st {
int alert_sent;
int num_fatal_alerts_sent;
int alert_received;
int session_ticket_do_not_call;
ssl_servername_t servername;
} HANDSHAKE_EX_DATA;
typedef struct ctx_data_st {
unsigned char *npn_protocols;
size_t npn_protocols_len;
unsigned char *alpn_protocols;
size_t alpn_protocols_len;
char *srp_user;
char *srp_password;
char *session_ticket_app_data;
} CTX_DATA;
/* |ctx_data| itself is stack-allocated. */
static void ctx_data_free_data(CTX_DATA *ctx_data)
{
OPENSSL_free(ctx_data->npn_protocols);
ctx_data->npn_protocols = NULL;
OPENSSL_free(ctx_data->alpn_protocols);
ctx_data->alpn_protocols = NULL;
OPENSSL_free(ctx_data->srp_user);
ctx_data->srp_user = NULL;
OPENSSL_free(ctx_data->srp_password);
ctx_data->srp_password = NULL;
OPENSSL_free(ctx_data->session_ticket_app_data);
ctx_data->session_ticket_app_data = NULL;
}
static int ex_data_idx;
static void info_cb(const SSL *s, int where, int ret)
{
if (where & SSL_CB_ALERT) {
HANDSHAKE_EX_DATA *ex_data =
(HANDSHAKE_EX_DATA*)(SSL_get_ex_data(s, ex_data_idx));
if (where & SSL_CB_WRITE) {
ex_data->alert_sent = ret;
if (strcmp(SSL_alert_type_string(ret), "F") == 0
|| strcmp(SSL_alert_desc_string(ret), "CN") == 0)
ex_data->num_fatal_alerts_sent++;
} else {
ex_data->alert_received = ret;
}
}
}
/* Select the appropriate server CTX.
* Returns SSL_TLSEXT_ERR_OK if a match was found.
* If |ignore| is 1, returns SSL_TLSEXT_ERR_NOACK on mismatch.
* Otherwise, returns SSL_TLSEXT_ERR_ALERT_FATAL on mismatch.
* An empty SNI extension also returns SSL_TSLEXT_ERR_NOACK.
*/
static int select_server_ctx(SSL *s, void *arg, int ignore)
{
const char *servername = SSL_get_servername(s, TLSEXT_NAMETYPE_host_name);
HANDSHAKE_EX_DATA *ex_data =
(HANDSHAKE_EX_DATA*)(SSL_get_ex_data(s, ex_data_idx));
if (servername == NULL) {
ex_data->servername = SSL_TEST_SERVERNAME_SERVER1;
return SSL_TLSEXT_ERR_NOACK;
}
if (strcmp(servername, "server2") == 0) {
SSL_CTX *new_ctx = (SSL_CTX*)arg;
SSL_set_SSL_CTX(s, new_ctx);
/*
* Copy over all the SSL_CTX options - reasonable behavior
* allows testing of cases where the options between two
* contexts differ/conflict
*/
SSL_clear_options(s, 0xFFFFFFFFL);
SSL_set_options(s, SSL_CTX_get_options(new_ctx));
ex_data->servername = SSL_TEST_SERVERNAME_SERVER2;
return SSL_TLSEXT_ERR_OK;
} else if (strcmp(servername, "server1") == 0) {
ex_data->servername = SSL_TEST_SERVERNAME_SERVER1;
return SSL_TLSEXT_ERR_OK;
} else if (ignore) {
ex_data->servername = SSL_TEST_SERVERNAME_SERVER1;
return SSL_TLSEXT_ERR_NOACK;
} else {
/* Don't set an explicit alert, to test library defaults. */
return SSL_TLSEXT_ERR_ALERT_FATAL;
}
}
static int client_hello_select_server_ctx(SSL *s, void *arg, int ignore)
{
const char *servername;
const unsigned char *p;
size_t len, remaining;
HANDSHAKE_EX_DATA *ex_data =
(HANDSHAKE_EX_DATA*)(SSL_get_ex_data(s, ex_data_idx));
/*
* The server_name extension was given too much extensibility when it
* was written, so parsing the normal case is a bit complex.
*/
if (!SSL_client_hello_get0_ext(s, TLSEXT_TYPE_server_name, &p,
&remaining) ||
remaining <= 2)
return 0;
/* Extract the length of the supplied list of names. */
len = (*(p++) << 8);
len += *(p++);
if (len + 2 != remaining)
return 0;
remaining = len;
/*
* The list in practice only has a single element, so we only consider
* the first one.
*/
if (remaining == 0 || *p++ != TLSEXT_NAMETYPE_host_name)
return 0;
remaining--;
/* Now we can finally pull out the byte array with the actual hostname. */
if (remaining <= 2)
return 0;
len = (*(p++) << 8);
len += *(p++);
if (len + 2 > remaining)
return 0;
remaining = len;
servername = (const char *)p;
if (len == strlen("server2") && strncmp(servername, "server2", len) == 0) {
SSL_CTX *new_ctx = arg;
SSL_set_SSL_CTX(s, new_ctx);
/*
* Copy over all the SSL_CTX options - reasonable behavior
* allows testing of cases where the options between two
* contexts differ/conflict
*/
SSL_clear_options(s, 0xFFFFFFFFL);
SSL_set_options(s, SSL_CTX_get_options(new_ctx));
ex_data->servername = SSL_TEST_SERVERNAME_SERVER2;
return 1;
} else if (len == strlen("server1") &&
strncmp(servername, "server1", len) == 0) {
ex_data->servername = SSL_TEST_SERVERNAME_SERVER1;
return 1;
} else if (ignore) {
ex_data->servername = SSL_TEST_SERVERNAME_SERVER1;
return 1;
}
return 0;
}
/*
* (RFC 6066):
* If the server understood the ClientHello extension but
* does not recognize the server name, the server SHOULD take one of two
* actions: either abort the handshake by sending a fatal-level
* unrecognized_name(112) alert or continue the handshake.
*
* This behaviour is up to the application to configure; we test both
* configurations to ensure the state machine propagates the result
* correctly.
*/
static int servername_ignore_cb(SSL *s, int *ad, void *arg)
{
return select_server_ctx(s, arg, 1);
}
static int servername_reject_cb(SSL *s, int *ad, void *arg)
{
return select_server_ctx(s, arg, 0);
}
static int client_hello_ignore_cb(SSL *s, int *al, void *arg)
{
if (!client_hello_select_server_ctx(s, arg, 1)) {
*al = SSL_AD_UNRECOGNIZED_NAME;
return SSL_CLIENT_HELLO_ERROR;
}
return SSL_CLIENT_HELLO_SUCCESS;
}
static int client_hello_reject_cb(SSL *s, int *al, void *arg)
{
if (!client_hello_select_server_ctx(s, arg, 0)) {
*al = SSL_AD_UNRECOGNIZED_NAME;
return SSL_CLIENT_HELLO_ERROR;
}
return SSL_CLIENT_HELLO_SUCCESS;
}
static int client_hello_nov12_cb(SSL *s, int *al, void *arg)
{
int ret;
unsigned int v;
const unsigned char *p;
v = SSL_client_hello_get0_legacy_version(s);
if (v > TLS1_2_VERSION || v < SSL3_VERSION) {
*al = SSL_AD_PROTOCOL_VERSION;
return SSL_CLIENT_HELLO_ERROR;
}
(void)SSL_client_hello_get0_session_id(s, &p);
if (p == NULL ||
SSL_client_hello_get0_random(s, &p) == 0 ||
SSL_client_hello_get0_ciphers(s, &p) == 0 ||
SSL_client_hello_get0_compression_methods(s, &p) == 0) {
*al = SSL_AD_INTERNAL_ERROR;
return SSL_CLIENT_HELLO_ERROR;
}
ret = client_hello_select_server_ctx(s, arg, 0);
SSL_set_max_proto_version(s, TLS1_1_VERSION);
if (!ret) {
*al = SSL_AD_UNRECOGNIZED_NAME;
return SSL_CLIENT_HELLO_ERROR;
}
return SSL_CLIENT_HELLO_SUCCESS;
}
static unsigned char dummy_ocsp_resp_good_val = 0xff;
static unsigned char dummy_ocsp_resp_bad_val = 0xfe;
static int server_ocsp_cb(SSL *s, void *arg)
{
unsigned char *resp;
resp = OPENSSL_malloc(1);
if (resp == NULL)
return SSL_TLSEXT_ERR_ALERT_FATAL;
/*
* For the purposes of testing we just send back a dummy OCSP response
*/
*resp = *(unsigned char *)arg;
if (!SSL_set_tlsext_status_ocsp_resp(s, resp, 1))
return SSL_TLSEXT_ERR_ALERT_FATAL;
return SSL_TLSEXT_ERR_OK;
}
static int client_ocsp_cb(SSL *s, void *arg)
{
const unsigned char *resp;
int len;
len = SSL_get_tlsext_status_ocsp_resp(s, &resp);
if (len != 1 || *resp != dummy_ocsp_resp_good_val)
return 0;
return 1;
}
static int verify_reject_cb(X509_STORE_CTX *ctx, void *arg) {
X509_STORE_CTX_set_error(ctx, X509_V_ERR_APPLICATION_VERIFICATION);
return 0;
}
static int verify_accept_cb(X509_STORE_CTX *ctx, void *arg) {
return 1;
}
static int broken_session_ticket_cb(SSL *s, unsigned char *key_name, unsigned char *iv,
EVP_CIPHER_CTX *ctx, HMAC_CTX *hctx, int enc)
{
return 0;
}
static int do_not_call_session_ticket_cb(SSL *s, unsigned char *key_name,
unsigned char *iv,
EVP_CIPHER_CTX *ctx,
HMAC_CTX *hctx, int enc)
{
HANDSHAKE_EX_DATA *ex_data =
(HANDSHAKE_EX_DATA*)(SSL_get_ex_data(s, ex_data_idx));
ex_data->session_ticket_do_not_call = 1;
return 0;
}
/* Parse the comma-separated list into TLS format. */
static int parse_protos(const char *protos, unsigned char **out, size_t *outlen)
{
size_t len, i, prefix;
len = strlen(protos);
/* Should never have reuse. */
if (!TEST_ptr_null(*out)
/* Test values are small, so we omit length limit checks. */
|| !TEST_ptr(*out = OPENSSL_malloc(len + 1)))
return 0;
*outlen = len + 1;
/*
* foo => '3', 'f', 'o', 'o'
* foo,bar => '3', 'f', 'o', 'o', '3', 'b', 'a', 'r'
*/
memcpy(*out + 1, protos, len);
prefix = 0;
i = prefix + 1;
while (i <= len) {
if ((*out)[i] == ',') {
if (!TEST_int_gt(i - 1, prefix))
goto err;
(*out)[prefix] = (unsigned char)(i - 1 - prefix);
prefix = i;
}
i++;
}
if (!TEST_int_gt(len, prefix))
goto err;
(*out)[prefix] = (unsigned char)(len - prefix);
return 1;
err:
OPENSSL_free(*out);
*out = NULL;
return 0;
}
#ifndef OPENSSL_NO_NEXTPROTONEG
/*
* The client SHOULD select the first protocol advertised by the server that it
* also supports. In the event that the client doesn't support any of server's
* protocols, or the server doesn't advertise any, it SHOULD select the first
* protocol that it supports.
*/
static int client_npn_cb(SSL *s, unsigned char **out, unsigned char *outlen,
const unsigned char *in, unsigned int inlen,
void *arg)
{
CTX_DATA *ctx_data = (CTX_DATA*)(arg);
int ret;
ret = SSL_select_next_proto(out, outlen, in, inlen,
ctx_data->npn_protocols,
ctx_data->npn_protocols_len);
/* Accept both OPENSSL_NPN_NEGOTIATED and OPENSSL_NPN_NO_OVERLAP. */
return TEST_true(ret == OPENSSL_NPN_NEGOTIATED || ret == OPENSSL_NPN_NO_OVERLAP)
? SSL_TLSEXT_ERR_OK : SSL_TLSEXT_ERR_ALERT_FATAL;
}
static int server_npn_cb(SSL *s, const unsigned char **data,
unsigned int *len, void *arg)
{
CTX_DATA *ctx_data = (CTX_DATA*)(arg);
*data = ctx_data->npn_protocols;
*len = ctx_data->npn_protocols_len;
return SSL_TLSEXT_ERR_OK;
}
#endif
/*
* The server SHOULD select the most highly preferred protocol that it supports
* and that is also advertised by the client. In the event that the server
* supports no protocols that the client advertises, then the server SHALL
* respond with a fatal "no_application_protocol" alert.
*/
static int server_alpn_cb(SSL *s, const unsigned char **out,
unsigned char *outlen, const unsigned char *in,
unsigned int inlen, void *arg)
{
CTX_DATA *ctx_data = (CTX_DATA*)(arg);
int ret;
/* SSL_select_next_proto isn't const-correct... */
unsigned char *tmp_out;
/*
* The result points either to |in| or to |ctx_data->alpn_protocols|.
* The callback is allowed to point to |in| or to a long-lived buffer,
* so we can return directly without storing a copy.
*/
ret = SSL_select_next_proto(&tmp_out, outlen,
ctx_data->alpn_protocols,
ctx_data->alpn_protocols_len, in, inlen);
*out = tmp_out;
/* Unlike NPN, we don't tolerate a mismatch. */
return ret == OPENSSL_NPN_NEGOTIATED ? SSL_TLSEXT_ERR_OK
2017-02-07 22:23:16 +00:00
: SSL_TLSEXT_ERR_ALERT_FATAL;
}
#ifndef OPENSSL_NO_SRP
static char *client_srp_cb(SSL *s, void *arg)
{
CTX_DATA *ctx_data = (CTX_DATA*)(arg);
return OPENSSL_strdup(ctx_data->srp_password);
}
static int server_srp_cb(SSL *s, int *ad, void *arg)
{
CTX_DATA *ctx_data = (CTX_DATA*)(arg);
if (strcmp(ctx_data->srp_user, SSL_get_srp_username(s)) != 0)
return SSL3_AL_FATAL;
if (SSL_set_srp_server_param_pw(s, ctx_data->srp_user,
ctx_data->srp_password,
"2048" /* known group */) < 0) {
*ad = SSL_AD_INTERNAL_ERROR;
return SSL3_AL_FATAL;
}
return SSL_ERROR_NONE;
}
#endif /* !OPENSSL_NO_SRP */
static int generate_session_ticket_cb(SSL *s, void *arg)
{
CTX_DATA *server_ctx_data = arg;
SSL_SESSION *ss = SSL_get_session(s);
char *app_data = server_ctx_data->session_ticket_app_data;
if (ss == NULL || app_data == NULL)
return 0;
return SSL_SESSION_set1_ticket_appdata(ss, app_data, strlen(app_data));
}
static int decrypt_session_ticket_cb(SSL *s, SSL_SESSION *ss,
const unsigned char *keyname,
size_t keyname_len,
SSL_TICKET_STATUS status,
void *arg)
{
switch (status) {
case SSL_TICKET_EMPTY:
case SSL_TICKET_NO_DECRYPT:
return SSL_TICKET_RETURN_IGNORE_RENEW;
case SSL_TICKET_SUCCESS:
return SSL_TICKET_RETURN_USE;
case SSL_TICKET_SUCCESS_RENEW:
return SSL_TICKET_RETURN_USE_RENEW;
default:
break;
}
return SSL_TICKET_RETURN_ABORT;
}
/*
* Configure callbacks and other properties that can't be set directly
* in the server/client CONF.
*/
static int configure_handshake_ctx(SSL_CTX *server_ctx, SSL_CTX *server2_ctx,
SSL_CTX *client_ctx,
const SSL_TEST_CTX *test,
const SSL_TEST_EXTRA_CONF *extra,
CTX_DATA *server_ctx_data,
CTX_DATA *server2_ctx_data,
CTX_DATA *client_ctx_data)
{
unsigned char *ticket_keys;
size_t ticket_key_len;
if (!TEST_int_eq(SSL_CTX_set_max_send_fragment(server_ctx,
test->max_fragment_size), 1))
goto err;
if (server2_ctx != NULL) {
if (!TEST_int_eq(SSL_CTX_set_max_send_fragment(server2_ctx,
test->max_fragment_size),
1))
goto err;
}
if (!TEST_int_eq(SSL_CTX_set_max_send_fragment(client_ctx,
test->max_fragment_size), 1))
goto err;
switch (extra->client.verify_callback) {
case SSL_TEST_VERIFY_ACCEPT_ALL:
SSL_CTX_set_cert_verify_callback(client_ctx, &verify_accept_cb, NULL);
break;
case SSL_TEST_VERIFY_REJECT_ALL:
SSL_CTX_set_cert_verify_callback(client_ctx, &verify_reject_cb, NULL);
break;
case SSL_TEST_VERIFY_NONE:
break;
}
switch (extra->client.max_fragment_len_mode) {
case TLSEXT_max_fragment_length_512:
case TLSEXT_max_fragment_length_1024:
case TLSEXT_max_fragment_length_2048:
case TLSEXT_max_fragment_length_4096:
case TLSEXT_max_fragment_length_DISABLED:
SSL_CTX_set_tlsext_max_fragment_length(
client_ctx, extra->client.max_fragment_len_mode);
break;
}
/*
* Link the two contexts for SNI purposes.
* Also do ClientHello callbacks here, as setting both ClientHello and SNI
* is bad.
*/
switch (extra->server.servername_callback) {
case SSL_TEST_SERVERNAME_IGNORE_MISMATCH:
SSL_CTX_set_tlsext_servername_callback(server_ctx, servername_ignore_cb);
SSL_CTX_set_tlsext_servername_arg(server_ctx, server2_ctx);
break;
case SSL_TEST_SERVERNAME_REJECT_MISMATCH:
SSL_CTX_set_tlsext_servername_callback(server_ctx, servername_reject_cb);
SSL_CTX_set_tlsext_servername_arg(server_ctx, server2_ctx);
break;
case SSL_TEST_SERVERNAME_CB_NONE:
break;
case SSL_TEST_SERVERNAME_CLIENT_HELLO_IGNORE_MISMATCH:
SSL_CTX_set_client_hello_cb(server_ctx, client_hello_ignore_cb, server2_ctx);
break;
case SSL_TEST_SERVERNAME_CLIENT_HELLO_REJECT_MISMATCH:
SSL_CTX_set_client_hello_cb(server_ctx, client_hello_reject_cb, server2_ctx);
break;
case SSL_TEST_SERVERNAME_CLIENT_HELLO_NO_V12:
SSL_CTX_set_client_hello_cb(server_ctx, client_hello_nov12_cb, server2_ctx);
}
if (extra->server.cert_status != SSL_TEST_CERT_STATUS_NONE) {
SSL_CTX_set_tlsext_status_type(client_ctx, TLSEXT_STATUSTYPE_ocsp);
SSL_CTX_set_tlsext_status_cb(client_ctx, client_ocsp_cb);
SSL_CTX_set_tlsext_status_arg(client_ctx, NULL);
SSL_CTX_set_tlsext_status_cb(server_ctx, server_ocsp_cb);
SSL_CTX_set_tlsext_status_arg(server_ctx,
((extra->server.cert_status == SSL_TEST_CERT_STATUS_GOOD_RESPONSE)
? &dummy_ocsp_resp_good_val : &dummy_ocsp_resp_bad_val));
}
/*
* The initial_ctx/session_ctx always handles the encrypt/decrypt of the
* session ticket. This ticket_key callback is assigned to the second
* session (assigned via SNI), and should never be invoked
*/
if (server2_ctx != NULL)
SSL_CTX_set_tlsext_ticket_key_cb(server2_ctx,
do_not_call_session_ticket_cb);
if (extra->server.broken_session_ticket) {
SSL_CTX_set_tlsext_ticket_key_cb(server_ctx, broken_session_ticket_cb);
}
#ifndef OPENSSL_NO_NEXTPROTONEG
if (extra->server.npn_protocols != NULL) {
if (!TEST_true(parse_protos(extra->server.npn_protocols,
&server_ctx_data->npn_protocols,
&server_ctx_data->npn_protocols_len)))
goto err;
SSL_CTX_set_npn_advertised_cb(server_ctx, server_npn_cb,
server_ctx_data);
}
if (extra->server2.npn_protocols != NULL) {
if (!TEST_true(parse_protos(extra->server2.npn_protocols,
&server2_ctx_data->npn_protocols,
&server2_ctx_data->npn_protocols_len))
|| !TEST_ptr(server2_ctx))
goto err;
SSL_CTX_set_npn_advertised_cb(server2_ctx, server_npn_cb,
server2_ctx_data);
}
if (extra->client.npn_protocols != NULL) {
if (!TEST_true(parse_protos(extra->client.npn_protocols,
&client_ctx_data->npn_protocols,
&client_ctx_data->npn_protocols_len)))
goto err;
SSL_CTX_set_next_proto_select_cb(client_ctx, client_npn_cb,
client_ctx_data);
}
#endif
if (extra->server.alpn_protocols != NULL) {
if (!TEST_true(parse_protos(extra->server.alpn_protocols,
&server_ctx_data->alpn_protocols,
&server_ctx_data->alpn_protocols_len)))
goto err;
SSL_CTX_set_alpn_select_cb(server_ctx, server_alpn_cb, server_ctx_data);
}
if (extra->server2.alpn_protocols != NULL) {
if (!TEST_ptr(server2_ctx)
|| !TEST_true(parse_protos(extra->server2.alpn_protocols,
&server2_ctx_data->alpn_protocols,
&server2_ctx_data->alpn_protocols_len
)))
goto err;
SSL_CTX_set_alpn_select_cb(server2_ctx, server_alpn_cb,
server2_ctx_data);
}
if (extra->client.alpn_protocols != NULL) {
unsigned char *alpn_protos = NULL;
size_t alpn_protos_len;
if (!TEST_true(parse_protos(extra->client.alpn_protocols,
&alpn_protos, &alpn_protos_len))
/* Reversed return value convention... */
|| !TEST_int_eq(SSL_CTX_set_alpn_protos(client_ctx, alpn_protos,
alpn_protos_len), 0))
goto err;
OPENSSL_free(alpn_protos);
}
if (extra->server.session_ticket_app_data != NULL) {
server_ctx_data->session_ticket_app_data =
OPENSSL_strdup(extra->server.session_ticket_app_data);
SSL_CTX_set_session_ticket_cb(server_ctx, generate_session_ticket_cb,
decrypt_session_ticket_cb, server_ctx_data);
}
if (extra->server2.session_ticket_app_data != NULL) {
if (!TEST_ptr(server2_ctx))
goto err;
server2_ctx_data->session_ticket_app_data =
OPENSSL_strdup(extra->server2.session_ticket_app_data);
SSL_CTX_set_session_ticket_cb(server2_ctx, NULL,
decrypt_session_ticket_cb, server2_ctx_data);
}
/*
* Use fixed session ticket keys so that we can decrypt a ticket created with
* one CTX in another CTX. Don't address server2 for the moment.
*/
ticket_key_len = SSL_CTX_set_tlsext_ticket_keys(server_ctx, NULL, 0);
if (!TEST_ptr(ticket_keys = OPENSSL_zalloc(ticket_key_len))
|| !TEST_int_eq(SSL_CTX_set_tlsext_ticket_keys(server_ctx,
ticket_keys,
ticket_key_len), 1)) {
OPENSSL_free(ticket_keys);
goto err;
}
OPENSSL_free(ticket_keys);
/* The default log list includes EC keys, so CT can't work without EC. */
#if !defined(OPENSSL_NO_CT) && !defined(OPENSSL_NO_EC)
if (!TEST_true(SSL_CTX_set_default_ctlog_list_file(client_ctx)))
goto err;
switch (extra->client.ct_validation) {
case SSL_TEST_CT_VALIDATION_PERMISSIVE:
if (!TEST_true(SSL_CTX_enable_ct(client_ctx,
SSL_CT_VALIDATION_PERMISSIVE)))
goto err;
break;
case SSL_TEST_CT_VALIDATION_STRICT:
if (!TEST_true(SSL_CTX_enable_ct(client_ctx, SSL_CT_VALIDATION_STRICT)))
goto err;
break;
case SSL_TEST_CT_VALIDATION_NONE:
break;
}
#endif
#ifndef OPENSSL_NO_SRP
if (extra->server.srp_user != NULL) {
SSL_CTX_set_srp_username_callback(server_ctx, server_srp_cb);
server_ctx_data->srp_user = OPENSSL_strdup(extra->server.srp_user);
server_ctx_data->srp_password = OPENSSL_strdup(extra->server.srp_password);
SSL_CTX_set_srp_cb_arg(server_ctx, server_ctx_data);
}
if (extra->server2.srp_user != NULL) {
if (!TEST_ptr(server2_ctx))
goto err;
SSL_CTX_set_srp_username_callback(server2_ctx, server_srp_cb);
server2_ctx_data->srp_user = OPENSSL_strdup(extra->server2.srp_user);
server2_ctx_data->srp_password = OPENSSL_strdup(extra->server2.srp_password);
SSL_CTX_set_srp_cb_arg(server2_ctx, server2_ctx_data);
}
if (extra->client.srp_user != NULL) {
if (!TEST_true(SSL_CTX_set_srp_username(client_ctx,
extra->client.srp_user)))
goto err;
SSL_CTX_set_srp_client_pwd_callback(client_ctx, client_srp_cb);
client_ctx_data->srp_password = OPENSSL_strdup(extra->client.srp_password);
SSL_CTX_set_srp_cb_arg(client_ctx, client_ctx_data);
}
#endif /* !OPENSSL_NO_SRP */
return 1;
err:
return 0;
}
/* Configure per-SSL callbacks and other properties. */
static void configure_handshake_ssl(SSL *server, SSL *client,
const SSL_TEST_EXTRA_CONF *extra)
{
if (extra->client.servername != SSL_TEST_SERVERNAME_NONE)
SSL_set_tlsext_host_name(client,
ssl_servername_name(extra->client.servername));
if (extra->client.enable_pha)
SSL_set_post_handshake_auth(client, 1);
}
/* The status for each connection phase. */
typedef enum {
PEER_SUCCESS,
PEER_RETRY,
PEER_ERROR,
PEER_WAITING,
PEER_TEST_FAILURE
} peer_status_t;
/* An SSL object and associated read-write buffers. */
typedef struct peer_st {
SSL *ssl;
/* Buffer lengths are int to match the SSL read/write API. */
unsigned char *write_buf;
int write_buf_len;
unsigned char *read_buf;
int read_buf_len;
int bytes_to_write;
int bytes_to_read;
peer_status_t status;
} PEER;
static int create_peer(PEER *peer, SSL_CTX *ctx)
{
static const int peer_buffer_size = 64 * 1024;
SSL *ssl = NULL;
unsigned char *read_buf = NULL, *write_buf = NULL;
if (!TEST_ptr(ssl = SSL_new(ctx))
|| !TEST_ptr(write_buf = OPENSSL_zalloc(peer_buffer_size))
|| !TEST_ptr(read_buf = OPENSSL_zalloc(peer_buffer_size)))
goto err;
peer->ssl = ssl;
peer->write_buf = write_buf;
peer->read_buf = read_buf;
peer->write_buf_len = peer->read_buf_len = peer_buffer_size;
return 1;
err:
SSL_free(ssl);
OPENSSL_free(write_buf);
OPENSSL_free(read_buf);
return 0;
}
static void peer_free_data(PEER *peer)
{
SSL_free(peer->ssl);
OPENSSL_free(peer->write_buf);
OPENSSL_free(peer->read_buf);
}
/*
* Note that we could do the handshake transparently under an SSL_write,
* but separating the steps is more helpful for debugging test failures.
*/
static void do_handshake_step(PEER *peer)
{
if (!TEST_int_eq(peer->status, PEER_RETRY)) {
peer->status = PEER_TEST_FAILURE;
} else {
int ret = SSL_do_handshake(peer->ssl);
if (ret == 1) {
peer->status = PEER_SUCCESS;
} else if (ret == 0) {
peer->status = PEER_ERROR;
} else {
int error = SSL_get_error(peer->ssl, ret);
/* Memory bios should never block with SSL_ERROR_WANT_WRITE. */
if (error != SSL_ERROR_WANT_READ)
peer->status = PEER_ERROR;
}
}
}
/*-
* Send/receive some application data. The read-write sequence is
* Peer A: (R) W - first read will yield no data
* Peer B: R W
* ...
* Peer A: R W
* Peer B: R W
* Peer A: R
*/
static void do_app_data_step(PEER *peer)
{
int ret = 1, write_bytes;
if (!TEST_int_eq(peer->status, PEER_RETRY)) {
peer->status = PEER_TEST_FAILURE;
return;
}
/* We read everything available... */
while (ret > 0 && peer->bytes_to_read) {
ret = SSL_read(peer->ssl, peer->read_buf, peer->read_buf_len);
if (ret > 0) {
if (!TEST_int_le(ret, peer->bytes_to_read)) {
peer->status = PEER_TEST_FAILURE;
return;
}
peer->bytes_to_read -= ret;
} else if (ret == 0) {
peer->status = PEER_ERROR;
return;
} else {
int error = SSL_get_error(peer->ssl, ret);
if (error != SSL_ERROR_WANT_READ) {
peer->status = PEER_ERROR;
return;
} /* Else continue with write. */
}
}
/* ... but we only write one write-buffer-full of data. */
write_bytes = peer->bytes_to_write < peer->write_buf_len ? peer->bytes_to_write :
peer->write_buf_len;
if (write_bytes) {
ret = SSL_write(peer->ssl, peer->write_buf, write_bytes);
if (ret > 0) {
/* SSL_write will only succeed with a complete write. */
if (!TEST_int_eq(ret, write_bytes)) {
peer->status = PEER_TEST_FAILURE;
return;
}
peer->bytes_to_write -= ret;
} else {
/*
* We should perhaps check for SSL_ERROR_WANT_READ/WRITE here
* but this doesn't yet occur with current app data sizes.
*/
peer->status = PEER_ERROR;
return;
}
}
/*
* We could simply finish when there was nothing to read, and we have
* nothing left to write. But keeping track of the expected number of bytes
* to read gives us somewhat better guarantees that all data sent is in fact
* received.
*/
if (!peer->bytes_to_write && !peer->bytes_to_read) {
peer->status = PEER_SUCCESS;
}
}
static void do_reneg_setup_step(const SSL_TEST_CTX *test_ctx, PEER *peer)
{
int ret;
char buf;
if (peer->status == PEER_SUCCESS) {
/*
* We are a client that succeeded this step previously, but the server
* wanted to retry. Probably there is a no_renegotiation warning alert
* waiting for us. Attempt to continue the handshake.
*/
peer->status = PEER_RETRY;
do_handshake_step(peer);
return;
}
if (!TEST_int_eq(peer->status, PEER_RETRY)
|| !TEST_true(test_ctx->handshake_mode
== SSL_TEST_HANDSHAKE_RENEG_SERVER
|| test_ctx->handshake_mode
== SSL_TEST_HANDSHAKE_RENEG_CLIENT
|| test_ctx->handshake_mode
== SSL_TEST_HANDSHAKE_KEY_UPDATE_SERVER
|| test_ctx->handshake_mode
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_TEST_HANDSHAKE_KEY_UPDATE_CLIENT
|| test_ctx->handshake_mode
== SSL_TEST_HANDSHAKE_POST_HANDSHAKE_AUTH)) {
peer->status = PEER_TEST_FAILURE;
return;
}
/* Reset the count of the amount of app data we need to read/write */
peer->bytes_to_write = peer->bytes_to_read = test_ctx->app_data_size;
/* Check if we are the peer that is going to initiate */
if ((test_ctx->handshake_mode == SSL_TEST_HANDSHAKE_RENEG_SERVER
&& SSL_is_server(peer->ssl))
|| (test_ctx->handshake_mode == SSL_TEST_HANDSHAKE_RENEG_CLIENT
&& !SSL_is_server(peer->ssl))) {
/*
* If we already asked for a renegotiation then fall through to the
* SSL_read() below.
*/
if (!SSL_renegotiate_pending(peer->ssl)) {
/*
* If we are the client we will always attempt to resume the
* session. The server may or may not resume dependent on the
* setting of SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION
*/
if (SSL_is_server(peer->ssl)) {
ret = SSL_renegotiate(peer->ssl);
} else {
if (test_ctx->extra.client.reneg_ciphers != NULL) {
if (!SSL_set_cipher_list(peer->ssl,
test_ctx->extra.client.reneg_ciphers)) {
peer->status = PEER_ERROR;
return;
}
ret = SSL_renegotiate(peer->ssl);
} else {
ret = SSL_renegotiate_abbreviated(peer->ssl);
}
}
if (!ret) {
peer->status = PEER_ERROR;
return;
}
do_handshake_step(peer);
/*
* If status is PEER_RETRY it means we're waiting on the peer to
* continue the handshake. As far as setting up the renegotiation is
* concerned that is a success. The next step will continue the
* handshake to its conclusion.
*
* If status is PEER_SUCCESS then we are the server and we have
* successfully sent the HelloRequest. We need to continue to wait
* until the handshake arrives from the client.
*/
if (peer->status == PEER_RETRY)
peer->status = PEER_SUCCESS;
else if (peer->status == PEER_SUCCESS)
peer->status = PEER_RETRY;
return;
}
} else if (test_ctx->handshake_mode == SSL_TEST_HANDSHAKE_KEY_UPDATE_SERVER
|| test_ctx->handshake_mode
== SSL_TEST_HANDSHAKE_KEY_UPDATE_CLIENT) {
if (SSL_is_server(peer->ssl)
!= (test_ctx->handshake_mode
== SSL_TEST_HANDSHAKE_KEY_UPDATE_SERVER)) {
peer->status = PEER_SUCCESS;
return;
}
ret = SSL_key_update(peer->ssl, test_ctx->key_update_type);
if (!ret) {
peer->status = PEER_ERROR;
return;
}
do_handshake_step(peer);
/*
* This is a one step handshake. We shouldn't get anything other than
* PEER_SUCCESS
*/
if (peer->status != PEER_SUCCESS)
peer->status = PEER_ERROR;
return;
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
} else if (test_ctx->handshake_mode == SSL_TEST_HANDSHAKE_POST_HANDSHAKE_AUTH) {
if (SSL_is_server(peer->ssl)) {
/* Make the server believe it's received the extension */
if (test_ctx->extra.server.force_pha)
peer->ssl->post_handshake_auth = SSL_PHA_EXT_RECEIVED;
ret = SSL_verify_client_post_handshake(peer->ssl);
if (!ret) {
peer->status = PEER_ERROR;
return;
}
}
do_handshake_step(peer);
/*
* This is a one step handshake. We shouldn't get anything other than
* PEER_SUCCESS
*/
if (peer->status != PEER_SUCCESS)
peer->status = PEER_ERROR;
return;
}
/*
* The SSL object is still expecting app data, even though it's going to
* get a handshake message. We try to read, and it should fail - after which
* we should be in a handshake
*/
ret = SSL_read(peer->ssl, &buf, sizeof(buf));
if (ret >= 0) {
/*
* We're not actually expecting data - we're expecting a reneg to
* start
*/
peer->status = PEER_ERROR;
return;
} else {
int error = SSL_get_error(peer->ssl, ret);
if (error != SSL_ERROR_WANT_READ) {
peer->status = PEER_ERROR;
return;
}
/* If we're not in init yet then we're not done with setup yet */
if (!SSL_in_init(peer->ssl))
return;
}
peer->status = PEER_SUCCESS;
}
/*
* RFC 5246 says:
*
* Note that as of TLS 1.1,
* failure to properly close a connection no longer requires that a
* session not be resumed. This is a change from TLS 1.0 to conform
* with widespread implementation practice.
*
* However,
* (a) OpenSSL requires that a connection be shutdown for all protocol versions.
* (b) We test lower versions, too.
* So we just implement shutdown. We do a full bidirectional shutdown so that we
* can compare sent and received close_notify alerts and get some test coverage
* for SSL_shutdown as a bonus.
*/
static void do_shutdown_step(PEER *peer)
{
int ret;
if (!TEST_int_eq(peer->status, PEER_RETRY)) {
peer->status = PEER_TEST_FAILURE;
return;
}
ret = SSL_shutdown(peer->ssl);
if (ret == 1) {
peer->status = PEER_SUCCESS;
} else if (ret < 0) { /* On 0, we retry. */
int error = SSL_get_error(peer->ssl, ret);
if (error != SSL_ERROR_WANT_READ && error != SSL_ERROR_WANT_WRITE)
peer->status = PEER_ERROR;
}
}
typedef enum {
HANDSHAKE,
RENEG_APPLICATION_DATA,
RENEG_SETUP,
RENEG_HANDSHAKE,
APPLICATION_DATA,
SHUTDOWN,
CONNECTION_DONE
} connect_phase_t;
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
static int renegotiate_op(const SSL_TEST_CTX *test_ctx)
{
switch (test_ctx->handshake_mode) {
case SSL_TEST_HANDSHAKE_RENEG_SERVER:
case SSL_TEST_HANDSHAKE_RENEG_CLIENT:
return 1;
default:
return 0;
}
}
static int post_handshake_op(const SSL_TEST_CTX *test_ctx)
{
switch (test_ctx->handshake_mode) {
case SSL_TEST_HANDSHAKE_KEY_UPDATE_CLIENT:
case SSL_TEST_HANDSHAKE_KEY_UPDATE_SERVER:
case SSL_TEST_HANDSHAKE_POST_HANDSHAKE_AUTH:
return 1;
default:
return 0;
}
}
static connect_phase_t next_phase(const SSL_TEST_CTX *test_ctx,
connect_phase_t phase)
{
switch (phase) {
case HANDSHAKE:
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 (renegotiate_op(test_ctx) || post_handshake_op(test_ctx))
return RENEG_APPLICATION_DATA;
return APPLICATION_DATA;
case RENEG_APPLICATION_DATA:
return RENEG_SETUP;
case RENEG_SETUP:
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 (post_handshake_op(test_ctx))
return APPLICATION_DATA;
return RENEG_HANDSHAKE;
case RENEG_HANDSHAKE:
return APPLICATION_DATA;
case APPLICATION_DATA:
return SHUTDOWN;
case SHUTDOWN:
return CONNECTION_DONE;
case CONNECTION_DONE:
TEST_error("Trying to progress after connection done");
break;
}
return -1;
}
static void do_connect_step(const SSL_TEST_CTX *test_ctx, PEER *peer,
connect_phase_t phase)
{
switch (phase) {
case HANDSHAKE:
do_handshake_step(peer);
break;
case RENEG_APPLICATION_DATA:
do_app_data_step(peer);
break;
case RENEG_SETUP:
do_reneg_setup_step(test_ctx, peer);
break;
case RENEG_HANDSHAKE:
do_handshake_step(peer);
break;
case APPLICATION_DATA:
do_app_data_step(peer);
break;
case SHUTDOWN:
do_shutdown_step(peer);
break;
case CONNECTION_DONE:
TEST_error("Action after connection done");
break;
}
}
typedef enum {
/* Both parties succeeded. */
HANDSHAKE_SUCCESS,
/* Client errored. */
CLIENT_ERROR,
/* Server errored. */
SERVER_ERROR,
/* Peers are in inconsistent state. */
INTERNAL_ERROR,
/* One or both peers not done. */
HANDSHAKE_RETRY
} handshake_status_t;
/*
* Determine the handshake outcome.
* last_status: the status of the peer to have acted last.
* previous_status: the status of the peer that didn't act last.
* client_spoke_last: 1 if the client went last.
*/
static handshake_status_t handshake_status(peer_status_t last_status,
peer_status_t previous_status,
int client_spoke_last)
{
switch (last_status) {
case PEER_TEST_FAILURE:
return INTERNAL_ERROR;
case PEER_WAITING:
/* Shouldn't ever happen */
return INTERNAL_ERROR;
case PEER_SUCCESS:
switch (previous_status) {
case PEER_TEST_FAILURE:
return INTERNAL_ERROR;
case PEER_SUCCESS:
/* Both succeeded. */
return HANDSHAKE_SUCCESS;
case PEER_WAITING:
case PEER_RETRY:
/* Let the first peer finish. */
return HANDSHAKE_RETRY;
case PEER_ERROR:
/*
* Second peer succeeded despite the fact that the first peer
* already errored. This shouldn't happen.
*/
return INTERNAL_ERROR;
}
break;
case PEER_RETRY:
return HANDSHAKE_RETRY;
case PEER_ERROR:
switch (previous_status) {
case PEER_TEST_FAILURE:
return INTERNAL_ERROR;
case PEER_WAITING:
/* The client failed immediately before sending the ClientHello */
return client_spoke_last ? CLIENT_ERROR : INTERNAL_ERROR;
case PEER_SUCCESS:
/*
* First peer succeeded but second peer errored.
* TODO(emilia): we should be able to continue here (with some
* application data?) to ensure the first peer receives the
* alert / close_notify.
* (No tests currently exercise this branch.)
*/
return client_spoke_last ? CLIENT_ERROR : SERVER_ERROR;
case PEER_RETRY:
/* We errored; let the peer finish. */
return HANDSHAKE_RETRY;
case PEER_ERROR:
/* Both peers errored. Return the one that errored first. */
return client_spoke_last ? SERVER_ERROR : CLIENT_ERROR;
}
}
/* Control should never reach here. */
return INTERNAL_ERROR;
}
/* Convert unsigned char buf's that shouldn't contain any NUL-bytes to char. */
static char *dup_str(const unsigned char *in, size_t len)
{
char *ret = NULL;
if (len == 0)
return NULL;
/* Assert that the string does not contain NUL-bytes. */
if (TEST_size_t_eq(OPENSSL_strnlen((const char*)(in), len), len))
TEST_ptr(ret = OPENSSL_strndup((const char*)(in), len));
return ret;
}
static int pkey_type(EVP_PKEY *pkey)
{
int nid = EVP_PKEY_id(pkey);
#ifndef OPENSSL_NO_EC
if (nid == EVP_PKEY_EC) {
const EC_KEY *ec = EVP_PKEY_get0_EC_KEY(pkey);
return EC_GROUP_get_curve_name(EC_KEY_get0_group(ec));
}
#endif
return nid;
}
static int peer_pkey_type(SSL *s)
{
X509 *x = SSL_get_peer_certificate(s);
if (x != NULL) {
int nid = pkey_type(X509_get0_pubkey(x));
X509_free(x);
return nid;
}
return NID_undef;
}
#if !defined(OPENSSL_NO_SCTP) && !defined(OPENSSL_NO_SOCK)
static int set_sock_as_sctp(int sock)
{
struct sctp_assocparams assocparams;
struct sctp_rtoinfo rto_info;
BIO *tmpbio;
/*
* To allow tests to fail fast (within a second or so), reduce the
* retransmission timeouts and the number of retransmissions.
*/
memset(&rto_info, 0, sizeof(struct sctp_rtoinfo));
rto_info.srto_initial = 100;
rto_info.srto_max = 200;
rto_info.srto_min = 50;
(void)setsockopt(sock, IPPROTO_SCTP, SCTP_RTOINFO,
(const void *)&rto_info, sizeof(struct sctp_rtoinfo));
memset(&assocparams, 0, sizeof(struct sctp_assocparams));
assocparams.sasoc_asocmaxrxt = 2;
(void)setsockopt(sock, IPPROTO_SCTP, SCTP_ASSOCINFO,
(const void *)&assocparams,
sizeof(struct sctp_assocparams));
/*
* For SCTP we have to set various options on the socket prior to
* connecting. This is done automatically by BIO_new_dgram_sctp().
* We don't actually need the created BIO though so we free it again
* immediately.
*/
tmpbio = BIO_new_dgram_sctp(sock, BIO_NOCLOSE);
if (tmpbio == NULL)
return 0;
BIO_free(tmpbio);
return 1;
}
static int create_sctp_socks(int *ssock, int *csock)
{
BIO_ADDRINFO *res = NULL;
const BIO_ADDRINFO *ai = NULL;
int lsock = INVALID_SOCKET, asock = INVALID_SOCKET;
int consock = INVALID_SOCKET;
int ret = 0;
int family = 0;
if (BIO_sock_init() != 1)
return 0;
/*
* Port is 4463. It could be anything. It will fail if it's already being
* used for some other SCTP service. It seems unlikely though so we don't
* worry about it here.
*/
if (!BIO_lookup_ex(NULL, "4463", BIO_LOOKUP_SERVER, family, SOCK_STREAM,
IPPROTO_SCTP, &res))
return 0;
for (ai = res; ai != NULL; ai = BIO_ADDRINFO_next(ai)) {
family = BIO_ADDRINFO_family(ai);
lsock = BIO_socket(family, SOCK_STREAM, IPPROTO_SCTP, 0);
if (lsock == INVALID_SOCKET) {
/* Maybe the kernel doesn't support the socket family, even if
* BIO_lookup() added it in the returned result...
*/
continue;
}
if (!set_sock_as_sctp(lsock)
|| !BIO_listen(lsock, BIO_ADDRINFO_address(ai),
BIO_SOCK_REUSEADDR)) {
BIO_closesocket(lsock);
lsock = INVALID_SOCKET;
continue;
}
/* Success, don't try any more addresses */
break;
}
if (lsock == INVALID_SOCKET)
goto err;
BIO_ADDRINFO_free(res);
res = NULL;
if (!BIO_lookup_ex(NULL, "4463", BIO_LOOKUP_CLIENT, family, SOCK_STREAM,
IPPROTO_SCTP, &res))
goto err;
consock = BIO_socket(family, SOCK_STREAM, IPPROTO_SCTP, 0);
if (consock == INVALID_SOCKET)
goto err;
if (!set_sock_as_sctp(consock)
|| !BIO_connect(consock, BIO_ADDRINFO_address(res), 0)
|| !BIO_socket_nbio(consock, 1))
goto err;
asock = BIO_accept_ex(lsock, NULL, BIO_SOCK_NONBLOCK);
if (asock == INVALID_SOCKET)
goto err;
*csock = consock;
*ssock = asock;
consock = asock = INVALID_SOCKET;
ret = 1;
err:
BIO_ADDRINFO_free(res);
if (consock != INVALID_SOCKET)
BIO_closesocket(consock);
if (lsock != INVALID_SOCKET)
BIO_closesocket(lsock);
if (asock != INVALID_SOCKET)
BIO_closesocket(asock);
return ret;
}
#endif
/*
* Note that |extra| points to the correct client/server configuration
* within |test_ctx|. When configuring the handshake, general mode settings
* are taken from |test_ctx|, and client/server-specific settings should be
* taken from |extra|.
*
* The configuration code should never reach into |test_ctx->extra| or
* |test_ctx->resume_extra| directly.
*
* (We could refactor test mode settings into a substructure. This would result
* in cleaner argument passing but would complicate the test configuration
* parsing.)
*/
static HANDSHAKE_RESULT *do_handshake_internal(
SSL_CTX *server_ctx, SSL_CTX *server2_ctx, SSL_CTX *client_ctx,
const SSL_TEST_CTX *test_ctx, const SSL_TEST_EXTRA_CONF *extra,
SSL_SESSION *session_in, SSL_SESSION *serv_sess_in,
SSL_SESSION **session_out, SSL_SESSION **serv_sess_out)
{
PEER server, client;
BIO *client_to_server = NULL, *server_to_client = NULL;
HANDSHAKE_EX_DATA server_ex_data, client_ex_data;
CTX_DATA client_ctx_data, server_ctx_data, server2_ctx_data;
HANDSHAKE_RESULT *ret = HANDSHAKE_RESULT_new();
int client_turn = 1, client_turn_count = 0, client_wait_count = 0;
connect_phase_t phase = HANDSHAKE;
handshake_status_t status = HANDSHAKE_RETRY;
const unsigned char* tick = NULL;
size_t tick_len = 0;
const unsigned char* sess_id = NULL;
unsigned int sess_id_len = 0;
SSL_SESSION* sess = NULL;
const unsigned char *proto = NULL;
/* API dictates unsigned int rather than size_t. */
unsigned int proto_len = 0;
EVP_PKEY *tmp_key;
const STACK_OF(X509_NAME) *names;
time_t start;
const char* cipher;
if (ret == NULL)
return NULL;
memset(&server_ctx_data, 0, sizeof(server_ctx_data));
memset(&server2_ctx_data, 0, sizeof(server2_ctx_data));
memset(&client_ctx_data, 0, sizeof(client_ctx_data));
memset(&server, 0, sizeof(server));
memset(&client, 0, sizeof(client));
memset(&server_ex_data, 0, sizeof(server_ex_data));
memset(&client_ex_data, 0, sizeof(client_ex_data));
if (!configure_handshake_ctx(server_ctx, server2_ctx, client_ctx,
test_ctx, extra, &server_ctx_data,
&server2_ctx_data, &client_ctx_data)) {
TEST_note("configure_handshake_ctx");
return NULL;
}
#if !defined(OPENSSL_NO_SCTP) && !defined(OPENSSL_NO_SOCK)
if (test_ctx->enable_client_sctp_label_bug)
SSL_CTX_set_mode(client_ctx, SSL_MODE_DTLS_SCTP_LABEL_LENGTH_BUG);
if (test_ctx->enable_server_sctp_label_bug)
SSL_CTX_set_mode(server_ctx, SSL_MODE_DTLS_SCTP_LABEL_LENGTH_BUG);
#endif
/* Setup SSL and buffers; additional configuration happens below. */
if (!create_peer(&server, server_ctx)) {
TEST_note("creating server context");
goto err;
}
if (!create_peer(&client, client_ctx)) {
TEST_note("creating client context");
goto err;
}
server.bytes_to_write = client.bytes_to_read = test_ctx->app_data_size;
client.bytes_to_write = server.bytes_to_read = test_ctx->app_data_size;
configure_handshake_ssl(server.ssl, client.ssl, extra);
if (session_in != NULL) {
SSL_SESSION_get_id(serv_sess_in, &sess_id_len);
/* In case we're testing resumption without tickets. */
if ((sess_id_len > 0
&& !TEST_true(SSL_CTX_add_session(server_ctx,
serv_sess_in)))
|| !TEST_true(SSL_set_session(client.ssl, session_in)))
goto err;
sess_id_len = 0;
}
ret->result = SSL_TEST_INTERNAL_ERROR;
if (test_ctx->use_sctp) {
#if !defined(OPENSSL_NO_SCTP) && !defined(OPENSSL_NO_SOCK)
int csock, ssock;
if (create_sctp_socks(&ssock, &csock)) {
client_to_server = BIO_new_dgram_sctp(csock, BIO_CLOSE);
server_to_client = BIO_new_dgram_sctp(ssock, BIO_CLOSE);
}
#endif
} else {
client_to_server = BIO_new(BIO_s_mem());
server_to_client = BIO_new(BIO_s_mem());
}
if (!TEST_ptr(client_to_server)
|| !TEST_ptr(server_to_client))
goto err;
/* Non-blocking bio. */
BIO_set_nbio(client_to_server, 1);
BIO_set_nbio(server_to_client, 1);
SSL_set_connect_state(client.ssl);
SSL_set_accept_state(server.ssl);
/* The bios are now owned by the SSL object. */
if (test_ctx->use_sctp) {
SSL_set_bio(client.ssl, client_to_server, client_to_server);
SSL_set_bio(server.ssl, server_to_client, server_to_client);
} else {
SSL_set_bio(client.ssl, server_to_client, client_to_server);
if (!TEST_int_gt(BIO_up_ref(server_to_client), 0)
|| !TEST_int_gt(BIO_up_ref(client_to_server), 0))
goto err;
SSL_set_bio(server.ssl, client_to_server, server_to_client);
}
ex_data_idx = SSL_get_ex_new_index(0, "ex data", NULL, NULL, NULL);
if (!TEST_int_ge(ex_data_idx, 0)
|| !TEST_int_eq(SSL_set_ex_data(server.ssl, ex_data_idx, &server_ex_data), 1)
|| !TEST_int_eq(SSL_set_ex_data(client.ssl, ex_data_idx, &client_ex_data), 1))
goto err;
SSL_set_info_callback(server.ssl, &info_cb);
SSL_set_info_callback(client.ssl, &info_cb);
client.status = PEER_RETRY;
server.status = PEER_WAITING;
start = time(NULL);
/*
* Half-duplex handshake loop.
* Client and server speak to each other synchronously in the same process.
* We use non-blocking BIOs, so whenever one peer blocks for read, it
* returns PEER_RETRY to indicate that it's the other peer's turn to write.
* The handshake succeeds once both peers have succeeded. If one peer
* errors out, we also let the other peer retry (and presumably fail).
*/
for(;;) {
if (client_turn) {
do_connect_step(test_ctx, &client, phase);
status = handshake_status(client.status, server.status,
1 /* client went last */);
if (server.status == PEER_WAITING)
server.status = PEER_RETRY;
} else {
do_connect_step(test_ctx, &server, phase);
status = handshake_status(server.status, client.status,
0 /* server went last */);
}
switch (status) {
case HANDSHAKE_SUCCESS:
client_turn_count = 0;
phase = next_phase(test_ctx, phase);
if (phase == CONNECTION_DONE) {
ret->result = SSL_TEST_SUCCESS;
goto err;
} else {
client.status = server.status = PEER_RETRY;
/*
* For now, client starts each phase. Since each phase is
* started separately, we can later control this more
* precisely, for example, to test client-initiated and
* server-initiated shutdown.
*/
client_turn = 1;
break;
}
case CLIENT_ERROR:
ret->result = SSL_TEST_CLIENT_FAIL;
goto err;
case SERVER_ERROR:
ret->result = SSL_TEST_SERVER_FAIL;
goto err;
case INTERNAL_ERROR:
ret->result = SSL_TEST_INTERNAL_ERROR;
goto err;
case HANDSHAKE_RETRY:
if (test_ctx->use_sctp) {
if (time(NULL) - start > 3) {
/*
* We've waited for too long. Give up.
*/
ret->result = SSL_TEST_INTERNAL_ERROR;
goto err;
}
/*
* With "real" sockets we only swap to processing the peer
* if they are expecting to retry. Otherwise we just retry the
* same endpoint again.
*/
if ((client_turn && server.status == PEER_RETRY)
|| (!client_turn && client.status == PEER_RETRY))
client_turn ^= 1;
} else {
if (client_turn_count++ >= 2000) {
/*
* At this point, there's been so many PEER_RETRY in a row
* that it's likely both sides are stuck waiting for a read.
* It's time to give up.
*/
ret->result = SSL_TEST_INTERNAL_ERROR;
goto err;
}
if (client_turn && server.status == PEER_SUCCESS) {
/*
* The server may finish before the client because the
* client spends some turns processing NewSessionTickets.
*/
if (client_wait_count++ >= 2) {
ret->result = SSL_TEST_INTERNAL_ERROR;
goto err;
}
} else {
/* Continue. */
client_turn ^= 1;
}
}
break;
}
}
err:
ret->server_alert_sent = server_ex_data.alert_sent;
ret->server_num_fatal_alerts_sent = server_ex_data.num_fatal_alerts_sent;
ret->server_alert_received = client_ex_data.alert_received;
ret->client_alert_sent = client_ex_data.alert_sent;
ret->client_num_fatal_alerts_sent = client_ex_data.num_fatal_alerts_sent;
ret->client_alert_received = server_ex_data.alert_received;
ret->server_protocol = SSL_version(server.ssl);
ret->client_protocol = SSL_version(client.ssl);
ret->servername = server_ex_data.servername;
if ((sess = SSL_get0_session(client.ssl)) != NULL) {
SSL_SESSION_get0_ticket(sess, &tick, &tick_len);
sess_id = SSL_SESSION_get_id(sess, &sess_id_len);
}
if (tick == NULL || tick_len == 0)
ret->session_ticket = SSL_TEST_SESSION_TICKET_NO;
else
ret->session_ticket = SSL_TEST_SESSION_TICKET_YES;
ret->compression = (SSL_get_current_compression(client.ssl) == NULL)
? SSL_TEST_COMPRESSION_NO
: SSL_TEST_COMPRESSION_YES;
if (sess_id == NULL || sess_id_len == 0)
ret->session_id = SSL_TEST_SESSION_ID_NO;
else
ret->session_id = SSL_TEST_SESSION_ID_YES;
ret->session_ticket_do_not_call = server_ex_data.session_ticket_do_not_call;
#ifndef OPENSSL_NO_NEXTPROTONEG
SSL_get0_next_proto_negotiated(client.ssl, &proto, &proto_len);
ret->client_npn_negotiated = dup_str(proto, proto_len);
SSL_get0_next_proto_negotiated(server.ssl, &proto, &proto_len);
ret->server_npn_negotiated = dup_str(proto, proto_len);
#endif
SSL_get0_alpn_selected(client.ssl, &proto, &proto_len);
ret->client_alpn_negotiated = dup_str(proto, proto_len);
SSL_get0_alpn_selected(server.ssl, &proto, &proto_len);
ret->server_alpn_negotiated = dup_str(proto, proto_len);
if ((sess = SSL_get0_session(server.ssl)) != NULL) {
SSL_SESSION_get0_ticket_appdata(sess, (void**)&tick, &tick_len);
ret->result_session_ticket_app_data = OPENSSL_strndup((const char*)tick, tick_len);
}
ret->client_resumed = SSL_session_reused(client.ssl);
ret->server_resumed = SSL_session_reused(server.ssl);
cipher = SSL_CIPHER_get_name(SSL_get_current_cipher(client.ssl));
ret->cipher = dup_str((const unsigned char*)cipher, strlen(cipher));
if (session_out != NULL)
*session_out = SSL_get1_session(client.ssl);
if (serv_sess_out != NULL) {
SSL_SESSION *tmp = SSL_get_session(server.ssl);
/*
* We create a fresh copy that is not in the server session ctx linked
* list.
*/
if (tmp != NULL)
*serv_sess_out = SSL_SESSION_dup(tmp);
}
if (SSL_get_peer_tmp_key(client.ssl, &tmp_key)) {
ret->tmp_key_type = pkey_type(tmp_key);
EVP_PKEY_free(tmp_key);
}
SSL_get_peer_signature_nid(client.ssl, &ret->server_sign_hash);
SSL_get_peer_signature_nid(server.ssl, &ret->client_sign_hash);
SSL_get_peer_signature_type_nid(client.ssl, &ret->server_sign_type);
SSL_get_peer_signature_type_nid(server.ssl, &ret->client_sign_type);
names = SSL_get0_peer_CA_list(client.ssl);
if (names == NULL)
ret->client_ca_names = NULL;
else
ret->client_ca_names = SSL_dup_CA_list(names);
names = SSL_get0_peer_CA_list(server.ssl);
if (names == NULL)
ret->server_ca_names = NULL;
else
ret->server_ca_names = SSL_dup_CA_list(names);
ret->server_cert_type = peer_pkey_type(client.ssl);
ret->client_cert_type = peer_pkey_type(server.ssl);
ctx_data_free_data(&server_ctx_data);
ctx_data_free_data(&server2_ctx_data);
ctx_data_free_data(&client_ctx_data);
peer_free_data(&server);
peer_free_data(&client);
return ret;
}
HANDSHAKE_RESULT *do_handshake(SSL_CTX *server_ctx, SSL_CTX *server2_ctx,
SSL_CTX *client_ctx, SSL_CTX *resume_server_ctx,
SSL_CTX *resume_client_ctx,
const SSL_TEST_CTX *test_ctx)
{
HANDSHAKE_RESULT *result;
SSL_SESSION *session = NULL, *serv_sess = NULL;
result = do_handshake_internal(server_ctx, server2_ctx, client_ctx,
test_ctx, &test_ctx->extra,
NULL, NULL, &session, &serv_sess);
if (result == NULL
|| test_ctx->handshake_mode != SSL_TEST_HANDSHAKE_RESUME
|| result->result == SSL_TEST_INTERNAL_ERROR)
goto end;
if (result->result != SSL_TEST_SUCCESS) {
result->result = SSL_TEST_FIRST_HANDSHAKE_FAILED;
goto end;
}
HANDSHAKE_RESULT_free(result);
/* We don't support SNI on second handshake yet, so server2_ctx is NULL. */
result = do_handshake_internal(resume_server_ctx, NULL, resume_client_ctx,
test_ctx, &test_ctx->resume_extra,
session, serv_sess, NULL, NULL);
end:
SSL_SESSION_free(session);
SSL_SESSION_free(serv_sess);
return result;
}