ossl-guide-tls-client-block.7ossl - Man Page

The first step is to create an SSL_CTX object for our client. We use the SSL_CTX_new(3) function for this purpose. We could alternatively use SSL_CTX_new_ex(3) if we want to associate the SSL_CTX with a particular OSSL_LIB_CTX (see ossl-guide-libraries-introduction(7) to learn about OSSL_LIB_CTX). We pass as an argument the return value of the function TLS_client_method(3). You should use this method whenever you are writing a TLS client. This method will automatically use TLS version negotiation to select the highest version of the protocol that is mutually supported by both the client and the server.

    /*
     * Create an SSL_CTX which we can use to create SSL objects from. We
     * want an SSL_CTX for creating clients so we use TLS_client_method()
     * here.
     */
    ctx = SSL_CTX_new(TLS_client_method());
    if (ctx == NULL) {
        printf("Failed to create the SSL_CTX\n");
        goto end;
    }

Since we are writing a client we must ensure that we verify the server's certificate. We do this by calling the SSL_CTX_set_verify(3) function and pass the SSL_VERIFY_PEER value to it. The final argument to this function is a callback that you can optionally supply to override the default handling for certificate verification. Most applications do not need to do this so this can safely be set to NULL to get the default handling.

    /*
     * Configure the client to abort the handshake if certificate
     * verification fails. Virtually all clients should do this unless you
     * really know what you are doing.
     */
    SSL_CTX_set_verify(ctx, SSL_VERIFY_PEER, NULL);

In order for certificate verification to be successful you must have configured where the trusted certificate store to be used is located (see ossl-guide-tls-introduction(7)). In most cases you just want to use the default store so we call SSL_CTX_set_default_verify_paths(3).

    /* Use the default trusted certificate store */
    if (!SSL_CTX_set_default_verify_paths(ctx)) {
        printf("Failed to set the default trusted certificate store\n");
        goto end;
    }

We would also like to restrict the TLS versions that we are willing to accept to TLSv1.2 or above. TLS protocol versions earlier than that are generally to be avoided where possible. We can do that using SSL_CTX_set_min_proto_version(3):

    /*
     * TLSv1.1 or earlier are deprecated by IETF and are generally to be
     * avoided if possible. We require a minimum TLS version of TLSv1.2.
     */
    if (!SSL_CTX_set_min_proto_version(ctx, TLS1_2_VERSION)) {
        printf("Failed to set the minimum TLS protocol version\n");
        goto end;
    }

That is all the setup that we need to do for the SSL_CTX, so next we need to create an SSL object to represent the TLS connection. In a real application we might expect to be creating more than one TLS connection over time. In that case we would expect to reuse the SSL_CTX that we already created each time. There is no need to repeat those steps. In fact it is best not to since certain internal resources are cached in the SSL_CTX. You will get better performance by reusing an existing SSL_CTX instead of creating a new one each time.

Creating the SSL object is a simple matter of calling the SSL_new(3) function and passing the SSL_CTX we created as an argument.

    /* Create an SSL object to represent the TLS connection */
    ssl = SSL_new(ctx);
    if (ssl == NULL) {
        printf("Failed to create the SSL object\n");
        goto end;
    }

TLS data is transmitted over an underlying transport layer. Normally a TCP socket. It is the application's responsibility for ensuring that the socket is created and associated with an SSL object (via a BIO).

Socket creation for use by a client is typically a 2 step process, i.e. constructing the socket; and connecting the socket.

How to construct a socket is platform specific - but most platforms (including Windows) provide a POSIX compatible interface via the socket function, e.g. to create an IPv4 TCP socket:

    int sock;

    sock = socket(AF_INET, SOCK_STREAM, 0);
    if (sock == -1)
        return NULL;

Once the socket is constructed it must be connected to the remote server. Again the details are platform specific but most platforms (including Windows) provide the POSIX compatible connect function. For example:

    struct sockaddr_in serveraddr;
    struct hostent *server;

    server = gethostbyname("www.openssl.org");
    if (server == NULL) {
        close(sock);
        return NULL;
    }

    memset(&serveraddr, 0, sizeof(serveraddr));
    serveraddr.sin_family = server->h_addrtype;
    serveraddr.sin_port = htons(443);
    memcpy(&serveraddr.sin_addr.s_addr, server->h_addr, server->h_length);

    if (connect(sock, (struct sockaddr *)&serveraddr,
                sizeof(serveraddr)) == -1) {
        close(sock);
        return NULL;
    }

OpenSSL provides portable helper functions to do these tasks which also integrate into the OpenSSL error system to log error data, e.g.

    int sock = -1;
    BIO_ADDRINFO *res;
    const BIO_ADDRINFO *ai = NULL;

    /*
     * Lookup IP address info for the server.
     */
    if (!BIO_lookup_ex(hostname, port, BIO_LOOKUP_CLIENT, family, SOCK_STREAM, 0,
                       &res))
        return NULL;

    /*
     * Loop through all the possible addresses for the server and find one
     * we can connect to.
     */
    for (ai = res; ai != NULL; ai = BIO_ADDRINFO_next(ai)) {
        /*
         * Create a TCP socket. We could equally use non-OpenSSL calls such
         * as "socket" here for this and the subsequent connect and close
         * functions. But for portability reasons and also so that we get
         * errors on the OpenSSL stack in the event of a failure we use
         * OpenSSL's versions of these functions.
         */
        sock = BIO_socket(BIO_ADDRINFO_family(ai), SOCK_STREAM, 0, 0);
        if (sock == -1)
            continue;

        /* Connect the socket to the server's address */
        if (!BIO_connect(sock, BIO_ADDRINFO_address(ai), BIO_SOCK_NODELAY)) {
            BIO_closesocket(sock);
            sock = -1;
            continue;
        }

        /* We have a connected socket so break out of the loop */
        break;
    }

    /* Free the address information resources we allocated earlier */
    BIO_ADDRINFO_free(res);

See BIO_lookup_ex(3), BIO_socket(3), BIO_connect(3), BIO_closesocket(3), BIO_ADDRINFO_next(3), BIO_ADDRINFO_address(3) and BIO_ADDRINFO_free(3) for further information on the functions used here. In the above example code the hostname and port variables are strings, e.g. "www.example.com" and "443".  Note also the use of the family variable, which can take the values of AF_INET or AF_INET6 based on the command line -6 option, to allow specific connections to an ipv4 or ipv6 enabled host.

Sockets created using the methods described above will automatically be blocking sockets - which is exactly what we want for this example.

Once the socket has been created and connected we need to associate it with a BIO object:

    BIO *bio;

    /* Create a BIO to wrap the socket */
    bio = BIO_new(BIO_s_socket());
    if (bio == NULL) {
        BIO_closesocket(sock);
        return NULL;
    }

    /*
     * Associate the newly created BIO with the underlying socket. By
     * passing BIO_CLOSE here the socket will be automatically closed when
     * the BIO is freed. Alternatively you can use BIO_NOCLOSE, in which
     * case you must close the socket explicitly when it is no longer
     * needed.
     */
    BIO_set_fd(bio, sock, BIO_CLOSE);

See BIO_new(3), BIO_s_socket(3) and BIO_set_fd(3) for further information on these functions.

Finally we associate the SSL object we created earlier with the BIO using the SSL_set_bio(3) function. Note that this passes ownership of the BIO object to the SSL object. Once ownership is passed the SSL object is responsible for its management and will free it automatically when the SSL is freed. So, once SSL_set_bio(3) has been been called, you should not call BIO_free(3) on the BIO.

    SSL_set_bio(ssl, bio, bio);

We have already connected our underlying socket to the server, but the client still needs to know the server's hostname. It uses this information for 2 key purposes and we need to set the hostname for each one.

Firstly, the server's hostname is included in the initial ClientHello message sent by the client. This is known as the Server Name Indication (SNI). This is important because it is common for multiple hostnames to be fronted by a single server that handles requests for all of them. In other words a single server may have multiple hostnames associated with it and it is important to indicate which one we want to connect to. Without this information we may get a handshake failure, or we may get connected to the "default" server which may not be the one we were expecting.

To set the SNI hostname data we call the SSL_set_tlsext_host_name(3) function like this:

    /*
     * Tell the server during the handshake which hostname we are attempting
     * to connect to in case the server supports multiple hosts.
     */
    if (!SSL_set_tlsext_host_name(ssl, hostname)) {
        printf("Failed to set the SNI hostname\n");
        goto end;
    }

Here the hostname argument is a string representing the hostname of the server, e.g. "www.example.com".

Secondly, we need to tell OpenSSL what hostname we expect to see in the certificate coming back from the server. This is almost always the same one that we asked for in the original request. This is important because, without this, we do not verify that the hostname in the certificate is what we expect it to be and any certificate is acceptable unless your application explicitly checks this itself. We do this via the SSL_set1_host(3) function:

    /*
     * Ensure we check during certificate verification that the server has
     * supplied a certificate for the hostname that we were expecting.
     * Virtually all clients should do this unless you really know what you
     * are doing.
     */
    if (!SSL_set1_host(ssl, hostname)) {
        printf("Failed to set the certificate verification hostname");
        goto end;
    }

All of the above steps must happen before we attempt to perform the handshake otherwise they will have no effect.

Before we can start sending or receiving application data over a TLS connection the TLS handshake must be performed. We can do this explicitly via the SSL_connect(3) function.

    /* Do the handshake with the server */
    if (SSL_connect(ssl) < 1) {
        printf("Failed to connect to the server\n");
        /*
         * If the failure is due to a verification error we can get more
         * information about it from SSL_get_verify_result().
         */
        if (SSL_get_verify_result(ssl) != X509_V_OK)
            printf("Verify error: %s\n",
                X509_verify_cert_error_string(SSL_get_verify_result(ssl)));
        goto end;
    }

The SSL_connect(3) function can return 1, 0 or less than 0. Only a return value of 1 is considered a success. For a simple blocking client we only need to concern ourselves with whether the call was successful or not. Anything else indicates that we have failed to connect to the server.

A common cause of failures at this stage is due to a problem verifying the server's certificate. For example if the certificate has expired, or it is not signed by a CA in our trusted certificate store. We can use the SSL_get_verify_result(3) function to find out more information about the verification failure. A return value of X509_V_OK indicates that the verification was successful (so the connection error must be due to some other cause). Otherwise we use the X509_verify_cert_error_string(3) function to get a human readable error message.

Once the handshake is complete we are able to send and receive application data. Exactly what data is sent and in what order is usually controlled by some application level protocol. In this example we are using HTTP 1.0 which is a very simple request and response protocol. The client sends a request to the server. The server sends the response data and then immediately closes down the connection.

To send data to the server we use the SSL_write_ex(3) function and to receive data from the server we use the SSL_read_ex(3) function. In HTTP 1.0 the client always writes data first. Our HTTP request will include the hostname that we are connecting to. For simplicity, we write the HTTP request in three chunks. First we write the start of the request. Secondly we write the hostname we are sending the request to. Finally we send the end of the request.

    size_t written;
    const char *request_start = "GET / HTTP/1.0\r\nConnection: close\r\nHost: ";
    const char *request_end = "\r\n\r\n";

    /* Write an HTTP GET request to the peer */
    if (!SSL_write_ex(ssl, request_start, strlen(request_start), &written)) {
        printf("Failed to write start of HTTP request\n");
        goto end;
    }
    if (!SSL_write_ex(ssl, hostname, strlen(hostname), &written)) {
        printf("Failed to write hostname in HTTP request\n");
        goto end;
    }
    if (!SSL_write_ex(ssl, request_end, strlen(request_end), &written)) {
        printf("Failed to write end of HTTP request\n");
        goto end;
    }

The SSL_write_ex(3) function returns 0 if it fails and 1 if it is successful. If it is successful then we can proceed to waiting for a response from the server.

    size_t readbytes;
    char buf[160];

    /*
     * Get up to sizeof(buf) bytes of the response. We keep reading until the
     * server closes the connection.
     */
    while (SSL_read_ex(ssl, buf, sizeof(buf), &readbytes)) {
        /*
        * OpenSSL does not guarantee that the returned data is a string or
        * that it is NUL terminated so we use fwrite() to write the exact
        * number of bytes that we read. The data could be non-printable or
        * have NUL characters in the middle of it. For this simple example
        * we're going to print it to stdout anyway.
        */
        fwrite(buf, 1, readbytes, stdout);
    }
    /* In case the response didn't finish with a newline we add one now */
    printf("\n");

We use the SSL_read_ex(3) function to read the response. We don't know exactly how much data we are going to receive back so we enter a loop reading blocks of data from the server and printing each block that we receive to the screen. The loop ends as soon as SSL_read_ex(3) returns 0 - meaning that it failed to read any data.

A failure to read data could mean that there has been some error, or it could simply mean that server has sent all the data that it wants to send and has indicated that it has finished by sending a "close_notify" alert. This alert is a TLS protocol level message indicating that the endpoint has finished sending all of its data and it will not send any more. Both of these conditions result in a 0 return value from SSL_read_ex(3) and we need to use the function SSL_get_error(3) to determine the cause of the 0 return value.

    /*
     * Check whether we finished the while loop above normally or as the
     * result of an error. The 0 argument to SSL_get_error() is the return
     * code we received from the SSL_read_ex() call. It must be 0 in order
     * to get here. Normal completion is indicated by SSL_ERROR_ZERO_RETURN.
     */
    if (SSL_get_error(ssl, 0) != SSL_ERROR_ZERO_RETURN) {
        /*
         * Some error occurred other than a graceful close down by the
         * peer
         */
        printf ("Failed reading remaining data\n");
        goto end;
    }

If SSL_get_error(3) returns SSL_ERROR_ZERO_RETURN then we know that the server has finished sending its data. Otherwise an error has occurred.

Once we have finished reading data from the server then we are ready to close the connection down. We do this via the SSL_shutdown(3) function which has the effect of sending a TLS protocol level message (a "close_notify" alert) to the server saying that we have finished writing data:

    /*
     * The peer already shutdown gracefully (we know this because of the
     * SSL_ERROR_ZERO_RETURN above). We should do the same back.
     */
    ret = SSL_shutdown(ssl);
    if (ret < 1) {
        /*
         * ret < 0 indicates an error. ret == 0 would be unexpected here
         * because that means "we've sent a close_notify and we're waiting
         * for one back". But we already know we got one from the peer
         * because of the SSL_ERROR_ZERO_RETURN above.
         */
        printf("Error shutting down\n");
        goto end;
    }

The SSL_shutdown(3) function will either return 1, 0, or less than 0. A return value of 1 is a success, and a return value less than 0 is an error. More precisely a return value of 1 means that we have sent a "close_notify" alert to the server, and that we have also received one back. A return value of 0 means that we have sent a "close_notify" alert to the server, but we have not yet received one back. Usually in this scenario you would call SSL_shutdown(3) again which (with a blocking socket) would block until the "close_notify" is received. However in this case we already know that the server has sent us a "close_notify" because of the SSL_ERROR_ZERO_RETURN that we received from the call to SSL_read_ex(3). So this scenario should never happen in practice. We just treat it as an error in this example.

Before the application exits we have to clean up some memory that we allocated. If we are exiting due to an error we might also want to display further information about that error if it is available to the user:

    /* Success! */
    res = EXIT_SUCCESS;
 end:
    /*
     * If something bad happened then we will dump the contents of the
     * OpenSSL error stack to stderr. There might be some useful diagnostic
     * information there.
     */
    if (res == EXIT_FAILURE)
        ERR_print_errors_fp(stderr);

    /*
     * Free the resources we allocated. We do not free the BIO object here
     * because ownership of it was immediately transferred to the SSL object
     * via SSL_set_bio(). The BIO will be freed when we free the SSL object.
     */
    SSL_free(ssl);
    SSL_CTX_free(ctx);
    return res;

To display errors we make use of the ERR_print_errors_fp(3) function which simply dumps out the contents of any errors on the OpenSSL error stack to the specified location (in this case stderr).

We need to free up the SSL object that we created for the connection via the SSL_free(3) function. Also, since we are not going to be creating any more TLS connections we must also free up the SSL_CTX via a call to SSL_CTX_free(3).

This could occur for numerous reasons. For example if there is a problem in the network route between the client and the server; or a firewall is blocking the communication; or the server is not in DNS. Check the network configuration.

A verification failure of the server certificate would result in a failure when running the SSL_connect(3) function. ERR_print_errors_fp(3) would display an error which would look something like this:

 Verify error: unable to get local issuer certificate
 40E74AF1F47F0000:error:0A000086:SSL routines:tls_post_process_server_certificate:certificate verify failed:ssl/statem/statem_clnt.c:2069:

A server certificate verification failure could be caused for a number of reasons. For example

Failure to correctly setup the trusted certificate store

See the page ossl-guide-tls-introduction(7) and check that your trusted certificate store is correctly configured

Unrecognised CA

If the CA used by the server's certificate is not in the trusted certificate store for the client then this will cause a verification failure during connection. Often this can occur if the server is using a self-signed certificate (i.e. a test certificate that has not been signed by a CA at all).

Missing intermediate CAs

This is a server misconfiguration where the client has the relevant root CA in its trust store, but the server has not supplied all of the intermediate CA certificates between that root CA and the server's own certificate. Therefore a trust chain cannot be established.

Mismatched hostname

If for some reason the hostname of the server that the client is expecting does not match the hostname in the certificate then this will cause verification to fail.

Expired certificate

The date that the server's certificate is valid to has passed.

The "unable to get local issuer certificate" we saw in the example above means that we have been unable to find the issuer of the server's certificate (or one of its intermediate CA certificates) in our trusted certificate store (e.g. because the trusted certificate store is misconfigured, or there are missing intermediate CAs, or the issuer is simply unrecognised).