openssl/doc/man7/ossl-guide-quic-server-non-block.pod

=pod

=begin comment

NB: Changes to the source code samples in this file should also be reflected in
demos/guide/quic-server-non-block.c

=end comment

=head1 NAME

ossl-guide-quic-server-non-block
- OpenSSL Guide: Writing a simple nonblocking QUIC server

=head1 SIMPLE NONBLOCKING QUIC SERVER EXAMPLE

This page presents various source code samples demonstrating how to write a
simple, non-concurrent, QUIC "echo" server application which accepts one client
connection at a time, echoing input from the client back to the same client.
Once the current client disconnects, the next client connection is accepted.

The server only accepts C<http/1.0> and C<hq-interop> ALPN's and doesn't actually
implement HTTP but only does a simple echo.  This is non-standard and will not
be supported by real world servers.  This is for demonstration purposes only.

There are various methods to test this server: B<quic-client-block.c> and
B<quic-client-non-block.c> will send a basic HTTP/1.0 request, which the server
will echo back.  You can also test this server by running
C<openssl s_client -connect localhost:4443 -4 -quic -alpn http/1.0> and entering
text that will be echoed back by the server.

Both the listening socket and connected socket are "nonblocking".  However,
we use select() to make the listening socket block when it cannot read/write.
Rather than stopping and waiting, your application may need to go and do other
tasks whilst the B<SSL> object is unable to read/write.  For example: updating a
GUI or performing operations on some other connection or stream.

The complete source code for this example nonblocking QUIC server is available
in the B<demos/guide> directory of the OpenSSL source distribution in the file
B<quic-server-non-block.c>.  It is also available online at
L<https://github.com/openssl/openssl/blob/master/demos/guide/quic-server-non-block.c>.

We assume that you already have OpenSSL installed on your system; that you
already have some fundamental understanding of OpenSSL concepts and QUIC (see
L<ossl-guide-libraries-introduction(7)> and L<ossl-guide-quic-introduction(7)>);
and that you know how to write and build C code and link it against the
libcrypto and libssl libraries that are provided by OpenSSL.  It also assumes
that you have a basic understanding of UDP/IP and sockets.

=head2 Creating the SSL_CTX and SSL objects

The first step is to create an B<SSL_CTX> object for our server.  We use the
L<SSL_CTX_new(3)> function for this purpose.  We pass as an argument the return
value of the function L<OSSL_QUIC_server_method(3)>.  You should use this method
whenever you are writing a QUIC server.

    /*
     * An SSL_CTX holds shared configuration information for multiple
     * subsequent per-client SSL connections. We specifically load a QUIC
     * server method here.
     */
    ctx = SSL_CTX_new(OSSL_QUIC_server_method());
    if (ctx == NULL)
        goto err;

Servers need a private key and certificate.  Intermediate issuer CA
certificates are often required, and both the server (end-entity or EE)
certificate and the issuer ("chain") certificates are most easily configured in
a single "chain file".  Below we load such a chain file (the EE certificate
must appear first), and then load the corresponding private key, checking that
it matches the server certificate.  No checks are performed to check the
integrity of the chain (CA signatures or certificate expiration dates, for
example), but we do verify the consistency of the private key with the
corresponding certificate.

    /*
     * Load the server's certificate *chain* file (PEM format), which includes
     * not only the leaf (end-entity) server certificate, but also any
     * intermediate issuer-CA certificates.  The leaf certificate must be the
     * first certificate in the file.
     *
     * In advanced use-cases this can be called multiple times, once per public
     * key algorithm for which the server has a corresponding certificate.
     * However, the corresponding private key (see below) must be loaded first,
     * *before* moving on to the next chain file.
     */
    if (SSL_CTX_use_certificate_chain_file(ctx, cert_path) <= 0) {
        fprintf(stderr, "couldn't load certificate file: %s\n", cert_path);
        goto err;
    }

    /*
     * Load the corresponding private key, this also checks that the private
     * key matches the just loaded end-entity certificate.  It does not check
     * whether the certificate chain is valid, the certificates could be
     * expired, or may otherwise fail to form a chain that a client can
     * validate.
     */
    if (SSL_CTX_use_PrivateKey_file(ctx, key_path, SSL_FILETYPE_PEM) <= 0) {
        fprintf(stderr, "couldn't load key file: %s\n", key_path);
        goto err;
    }

Most servers, including this one, do not solicit client certificates.  We
therefore do not need a "trust store" and allow the handshake to complete even
when the client does not present a certificate.  Note: Even if a client did
present a trusted certificate, for it to be useful, the server application
would still need custom code to use the verified identity to grant nondefault
access to that particular client.  Some servers grant access to all clients
with certificates from a private CA, this then requires processing of
certificate revocation lists to deauthorise a client.  It is often simpler and
more secure to instead keep a list of authorised public keys.

Though this is the default setting, we explicitly call the
L<SSL_CTX_set_verify(3)> function and pass the B<SSL_VERIFY_NONE> 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.

    /*
     * Clients rarely employ certificate-based authentication, and so we don't
     * require "mutual" TLS authentication (indeed there's no way to know
     * whether or how the client authenticated the server, so the term "mutual"
     * is potentially misleading).
     *
     * Since we're not soliciting or processing client certificates, we don't
     * need to configure a trusted-certificate store, so no call to
     * SSL_CTX_set_default_verify_paths() is needed.  The server's own
     * certificate chain is assumed valid.
     */
    SSL_CTX_set_verify(ctx, SSL_VERIFY_NONE, NULL);


QUIC also dictates using Application-Layer Protocol Negotiation (ALPN) to select
an application protocol.  We use L<SSL_CTX_set_alpn_select_cb(3)> for this
purpose.  We can pass a callback which will be called for each connection to
select an ALPN the server considers acceptable.

    /* Setup ALPN negotiation callback to decide which ALPN is accepted. */
    SSL_CTX_set_alpn_select_cb(ctx, select_alpn, NULL);

In this case, we only accept "http/1.0" and "hq-interop".

    /*
    * ALPN strings for TLS handshake. Only 'http/1.0' and 'hq-interop'
    * are accepted.
    */
    static const unsigned char alpn_ossltest[] = {
        8,  'h', 't', 't', 'p', '/', '1', '.', '0',
        10, 'h', 'q', '-', 'i', 'n', 't', 'e', 'r', 'o', 'p',
    };

    static int select_alpn(SSL *ssl, const unsigned char **out,
                           unsigned char *out_len, const unsigned char *in,
                           unsigned int in_len, void *arg)
    {
        if (SSL_select_next_proto((unsigned char **)out, out_len, alpn_ossltest,
                                  sizeof(alpn_ossltest), in,
                                  in_len) == OPENSSL_NPN_NEGOTIATED)
            return SSL_TLSEXT_ERR_OK;
        return SSL_TLSEXT_ERR_ALERT_FATAL;
    }

That is all the setup that we need to do for the B<SSL_CTX>.  Next, we create a
UDP socket and bind to it on localhost.

    /* Retrieve the file descriptor for a new UDP socket */
    if ((fd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP)) < 0) {
        fprintf(stderr, "cannot create socket");
        return -1;
    }

    sa.sin_family = AF_INET;
    sa.sin_port = htons(port);

    /* Bind to the new UDP socket on localhost */
    if (bind(fd, (const struct sockaddr *)&sa, sizeof(sa)) < 0) {
        fprintf(stderr, "cannot bind to %u\n", port);
        BIO_closesocket(fd);
        return -1;
    }

    /* Set port to nonblocking mode */
    if (BIO_socket_nbio(fd, 1) <= 0) {
        fprintf(stderr, "Unable to set port to nonblocking mode");
        BIO_closesocket(fd);
        return -1;
    }

To run the QUIC server, we create an B<SSL_LISTENER> to listen for incoming
connections.  We provide it with the bound UDP port to then explicitly begin
listening for new connections.

    /* Create a new QUIC listener */
    if ((listener = SSL_new_listener(ctx, 0)) == NULL)
        goto err;

    /* Provide the listener with our UDP socket. */
    if (!SSL_set_fd(listener, fd))
        goto err;

    /* Set the listener mode to nonblocking, which is inherited by
     * child objects.
     */
    if (!SSL_set_blocking_mode(listener, 0))
        goto err;

    /*
     * Begin listening. Note that is not usually needed as SSL_accept_connection
     * will implicitly start listening. It is only needed if a server wishes to
     * ensure it has started to accept incoming connections but does not wish to
     * actually call SSL_accept_connection yet.
     */
    if (!SSL_listen(listener))
        goto err;

=head2 Server loop

The server now enters a "forever" loop, handling one client connection at a
time.  Before each connection, we clear the OpenSSL error stack so that any
error reports are related to just the new connection.

    /* Pristine error stack for each new connection */
    ERR_clear_error();

We then wait until a connection is ready for reading.
It uses the select function to wait until the socket is either readable
or writable, depending on what the SSL connection requires.

We then accept a new connection in which the handshake will have already
occurred. However, since we are in nonblocking mode, L<SSL_accept_connection(3)>
will return immediately. Therefore, we use a helper function to essentially
block until a connection is established.

    printf("Waiting for connection\n");
    while ((conn = SSL_accept_connection(listener, 0)) == NULL) {
        wait_for_activity(listener);
    }
    printf("Accepted new connection\n");

The helper function wait_for_activity uses select() to block until the file
descriptor belonging to the passed SSL object is readable. As mentioned earlier,
a more real-world application would likely use this time to perform other tasks.

    /* Initialize the fd_set structure */
    FD_ZERO(&read_fd);
    FD_ZERO(&write_fd);

    /*
     * Determine if we would like to write to the socket, read from it, or both.
     */
    if (SSL_net_write_desired(ssl))
        FD_SET(sock, &write_fd);
    if (SSL_net_read_desired(ssl))
        FD_SET(sock, &read_fd);

    /*
     * Find out when OpenSSL would next like to be called, regardless of
     * whether the state of the underlying socket has changed or not.
     */
    if (SSL_get_event_timeout(ssl, &tv, &isinfinite) && !isinfinite)
        tvp = &tv;

    /*
     * Wait until the socket is writeable or readable. We use select here
     * for the sake of simplicity and portability, but you could equally use
     * poll/epoll or similar functions
     *
     * NOTE: For the purposes of this demonstration code this effectively
     * makes this demo block until it has something more useful to do. In a
     * real application you probably want to go and do other work here (e.g.
     * update a GUI, or service other connections).
     *
     * Let's say for example that you want to update the progress counter on
     * a GUI every 100ms. One way to do that would be to use the timeout in
     * the last parameter to "select" below. If the tvp value is greater
     * than 100ms then use 100ms instead. Then, when select returns, you
     * check if it did so because of activity on the file descriptors or
     * because of the timeout. If the 100ms GUI timeout has expired but the
     * tvp timeout has not then go and update the GUI and then restart the
     * "select" (with updated timeouts).
     */

    select(sock + 1, &read_fd, &write_fd, NULL, tvp);

With the handshake complete, the server reads all the client input.

    /* Read from client until the client sends a end of stream packet */
    while (!eof) {
        ret = SSL_read_ex(conn, buf + total_read, sizeof(buf) - total_read,
                          &nread);
        total_read += nread;
        if (total_read >= 8192) {
            fprintf(stderr, "Could not fit all data into buffer\n");
            goto err;
        }
        switch (handle_io_failure(conn, ret)) {
        case 1:
            continue; /* Retry */
        case 0:
            /* Reached end of stream */
            if (!SSL_has_pending(conn))
                eof = 1;
            break;
        default:
            fprintf(stderr, "Failed reading remaining data\n");
            goto err;
        }
    }

Finally, we echo the received data back to the client.  We can use
L<SSL_write_ex2(3)> to pass in a special flag SSL_WRITE_FLAG_CONCLUDE that will
send a FIN packet once the write has successfully finished writing all the data
to the peer.

    /* Echo client input */
    while (!SSL_write_ex2(conn, buf,
                          total_read,
                          SSL_WRITE_FLAG_CONCLUDE, &total_written)) {
        if (handle_io_failure(conn, 0) == 1)
            continue;
        fprintf(stderr, "Failed to write data\n");
        goto err;
    }

We then shut down the connection with L<SSL_shutdown(3)>, which may need
to be called multiple times to ensure the connection is shutdown completely.

    /*
     * Shut down the connection. We may need to call this multiple times
     * to ensure the connection is shutdown completely.
     */
    while ((ret = SSL_shutdown(conn)) != 1) {
        if (ret < 0 && handle_io_failure(conn, ret) == 1)
            continue; /* Retry */
    }

Finally, we free the SSL connection, and the server is now ready to accept the
next client connection.

    SSL_free(conn);

=head2 Final clean up

If the server somehow manages to break out of the infinite loop and
be ready to exit, it would deallocate the constructed B<SSL>.

    SSL_free(listener);

And in the main function, it would deallocate the constructed B<SSL_CTX>.

    SSL_CTX_free(ctx);
    BIO_closesocket(fd);

=head1 SEE ALSO

L<ossl-guide-introduction(7)>, L<ossl-guide-libraries-introduction(7)>,
L<ossl-guide-libssl-introduction(7)>, L<ossl-guide-quic-introduction(7)>,
L<ossl-guide-quic-client-non-block(7)>, L<ossl-guide-quic-client-block(7)>,
L<ossl-guide-tls-server-block(7)>, L<ossl-guide-quic-server-block(7)>

=head1 COPYRIGHT

Copyright 2024 The OpenSSL Project Authors. All Rights Reserved.

Licensed under the Apache License 2.0 (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
L<https://www.openssl.org/source/license.html>.

=cut