QUIC FIN Support: Various fixes
[openssl.git] / ssl / quic / quic_impl.c
index 1c673d23b66bbc4b22e863e08aff510b30cc9206..5448e32e73faa4cca8259b52eb849b8ab885d8d6 100644 (file)
 
 #include <openssl/macros.h>
 #include <openssl/objects.h>
+#include <openssl/sslerr.h>
+#include <crypto/rand.h>
 #include "quic_local.h"
+#include "internal/quic_tls.h"
+#include "internal/quic_rx_depack.h"
+#include "internal/quic_error.h"
+#include "internal/time.h"
 
-int ossl_quic_new(SSL *s)
+static void aon_write_finish(QUIC_CONNECTION *qc);
+static int ensure_channel(QUIC_CONNECTION *qc);
+
+/*
+ * QUIC Front-End I/O API: Common Utilities
+ * ========================================
+ */
+
+/*
+ * Block until a predicate is met.
+ *
+ * Precondition: Must have a channel.
+ */
+static int block_until_pred(QUIC_CONNECTION *qc,
+                            int (*pred)(void *arg), void *pred_arg,
+                            uint32_t flags)
+{
+    QUIC_REACTOR *rtor;
+
+    assert(qc->ch != NULL);
+
+    rtor = ossl_quic_channel_get_reactor(qc->ch);
+    return ossl_quic_reactor_block_until_pred(rtor, pred, pred_arg, flags);
+}
+
+/*
+ * Raise a 'normal' error, meaning one that can be reported via SSL_get_error()
+ * rather than via ERR.
+ */
+static int quic_raise_normal_error(QUIC_CONNECTION *qc,
+                                   int err)
 {
-    return s->method->ssl_clear(s);
+    qc->last_error = err;
+    return 0;
 }
 
+/*
+ * Raise a 'non-normal' error, meaning any error that is not reported via
+ * SSL_get_error() and must be reported via ERR.
+ */
+static int quic_raise_non_normal_error(QUIC_CONNECTION *qc,
+                                       const char *file,
+                                       int line,
+                                       const char *func,
+                                       int reason,
+                                       const char *fmt,
+                                       ...)
+{
+    va_list args;
+
+    ERR_new();
+    ERR_set_debug(file, line, func);
+
+    va_start(args, fmt);
+    ERR_vset_error(ERR_LIB_SSL, reason, fmt, args);
+    va_end(args);
+
+    qc->last_error = SSL_ERROR_SSL;
+    return 0;
+}
+
+#define QUIC_RAISE_NORMAL_ERROR(qc, err)                        \
+    quic_raise_normal_error((qc), (err))
+
+#define QUIC_RAISE_NON_NORMAL_ERROR(qc, reason, msg)            \
+    quic_raise_non_normal_error((qc),                           \
+                                OPENSSL_FILE, OPENSSL_LINE,     \
+                                OPENSSL_FUNC,                   \
+                                (reason),                       \
+                                (msg))
+
+/*
+ * Should be called at entry of every public function to confirm we have a valid
+ * QUIC_CONNECTION.
+ */
+static ossl_inline int expect_quic_conn(const QUIC_CONNECTION *qc)
+{
+    if (!ossl_assert(qc != NULL))
+        return QUIC_RAISE_NON_NORMAL_ERROR(NULL, ERR_R_INTERNAL_ERROR, NULL);
+
+    return 1;
+
+}
+
+/*
+ * QUIC Front-End I/O API: Initialization
+ * ======================================
+ *
+ *         SSL_new                  => ossl_quic_new
+ *                                     ossl_quic_init
+ *         SSL_reset                => ossl_quic_reset
+ *         SSL_clear                => ossl_quic_clear
+ *                                     ossl_quic_deinit
+ *         SSL_free                 => ossl_quic_free
+ *
+ */
+
+/* SSL_new */
+SSL *ossl_quic_new(SSL_CTX *ctx)
+{
+    QUIC_CONNECTION *qc = NULL;
+    SSL *ssl_base = NULL;
+    SSL_CONNECTION *sc = NULL;
+
+    qc = OPENSSL_zalloc(sizeof(*qc));
+    if (qc == NULL)
+        goto err;
+
+    /* Initialise the QUIC_CONNECTION's stub header. */
+    ssl_base = &qc->ssl;
+    if (!ossl_ssl_init(ssl_base, ctx, ctx->method, SSL_TYPE_QUIC_CONNECTION)) {
+        ssl_base = NULL;
+        goto err;
+    }
+
+    qc->tls = ossl_ssl_connection_new_int(ctx, TLS_method());
+    if (qc->tls == NULL || (sc = SSL_CONNECTION_FROM_SSL(qc->tls)) == NULL)
+         goto err;
+
+    /* Channel is not created yet. */
+    qc->ssl_mode   = qc->ssl.ctx->mode;
+    qc->last_error = SSL_ERROR_NONE;
+    qc->blocking   = 1;
+
+    return ssl_base;
+
+err:
+    OPENSSL_free(qc);
+    return NULL;
+}
+
+/* SSL_free */
 void ossl_quic_free(SSL *s)
 {
-    return;
+    QUIC_CONNECTION *qc = QUIC_CONNECTION_FROM_SSL(s);
+
+    /* We should never be called on anything but a QUIC_CONNECTION. */
+    if (!expect_quic_conn(qc))
+        return;
+
+    ossl_quic_channel_free(qc->ch);
+
+    BIO_free(qc->net_rbio);
+    BIO_free(qc->net_wbio);
+
+    /* Note: SSL_free calls OPENSSL_free(qc) for us */
+
+    SSL_free(qc->tls);
 }
 
+/* SSL method init */
+int ossl_quic_init(SSL *s)
+{
+    QUIC_CONNECTION *qc = QUIC_CONNECTION_FROM_SSL(s);
+
+    if (!expect_quic_conn(qc))
+        return 0;
+
+    /* Same op as SSL_clear, forward the call. */
+    return ossl_quic_clear(s);
+}
+
+/* SSL method deinit */
+void ossl_quic_deinit(SSL *s)
+{
+    /* No-op. */
+}
+
+/* SSL_reset */
+int ossl_quic_reset(SSL *s)
+{
+    QUIC_CONNECTION *qc = QUIC_CONNECTION_FROM_SSL(s);
+
+    if (!expect_quic_conn(qc))
+        return 0;
+
+    /* TODO(QUIC); Currently a no-op. */
+    return 1;
+}
+
+/* SSL_clear */
 int ossl_quic_clear(SSL *s)
 {
+    QUIC_CONNECTION *qc = QUIC_CONNECTION_FROM_SSL(s);
+
+    if (!expect_quic_conn(qc))
+        return 0;
+
+    /* TODO(QUIC): Currently a no-op. */
     return 1;
 }
 
-int ossl_quic_accept(SSL *s)
+/*
+ * QUIC Front-End I/O API: Network BIO Configuration
+ * =================================================
+ *
+ * Handling the different BIOs is difficult:
+ *
+ *   - It is more or less a requirement that we use non-blocking network I/O;
+ *     we need to be able to have timeouts on recv() calls, and make best effort
+ *     (non blocking) send() and recv() calls.
+ *
+ *     The only sensible way to do this is to configure the socket into
+ *     non-blocking mode. We could try to do select() before calling send() or
+ *     recv() to get a guarantee that the call will not block, but this will
+ *     probably run into issues with buggy OSes which generate spurious socket
+ *     readiness events. In any case, relying on this to work reliably does not
+ *     seem sane.
+ *
+ *     Timeouts could be handled via setsockopt() socket timeout options, but
+ *     this depends on OS support and adds another syscall to every network I/O
+ *     operation. It also has obvious thread safety concerns if we want to move
+ *     to concurrent use of a single socket at some later date.
+ *
+ *     Some OSes support a MSG_DONTWAIT flag which allows a single I/O option to
+ *     be made non-blocking. However some OSes (e.g. Windows) do not support
+ *     this, so we cannot rely on this.
+ *
+ *     As such, we need to configure any FD in non-blocking mode. This may
+ *     confound users who pass a blocking socket to libssl. However, in practice
+ *     it would be extremely strange for a user of QUIC to pass an FD to us,
+ *     then also try and send receive traffic on the same socket(!). Thus the
+ *     impact of this should be limited, and can be documented.
+ *
+ *   - We support both blocking and non-blocking operation in terms of the API
+ *     presented to the user. One prospect is to set the blocking mode based on
+ *     whether the socket passed to us was already in blocking mode. However,
+ *     Windows has no API for determining if a socket is in blocking mode (!),
+ *     therefore this cannot be done portably. Currently therefore we expose an
+ *     explicit API call to set this, and default to blocking mode.
+ *
+ *   - We need to determine our initial destination UDP address. The "natural"
+ *     way for a user to do this is to set the peer variable on a BIO_dgram.
+ *     However, this has problems because BIO_dgram's peer variable is used for
+ *     both transmission and reception. This means it can be constantly being
+ *     changed to a malicious value (e.g. if some random unrelated entity on the
+ *     network starts sending traffic to us) on every read call. This is not a
+ *     direct issue because we use the 'stateless' BIO_sendmmsg and BIO_recvmmsg
+ *     calls only, which do not use this variable. However, we do need to let
+ *     the user specify the peer in a 'normal' manner. The compromise here is
+ *     that we grab the current peer value set at the time the write BIO is set
+ *     and do not read the value again.
+ *
+ *   - We also need to support memory BIOs (e.g. BIO_dgram_pair) or custom BIOs.
+ *     Currently we do this by only supporting non-blocking mode.
+ *
+ */
+
+/*
+ * Determines what initial destination UDP address we should use, if possible.
+ * If this fails the client must set the destination address manually, or use a
+ * BIO which does not need a destination address.
+ */
+static int csm_analyse_init_peer_addr(BIO *net_wbio, BIO_ADDR *peer)
 {
+    if (BIO_dgram_get_peer(net_wbio, peer) <= 0)
+        return 0;
+
     return 1;
 }
 
-int ossl_quic_connect(SSL *s)
+void ossl_quic_conn_set0_net_rbio(QUIC_CONNECTION *qc, BIO *net_rbio)
+{
+    if (qc->net_rbio == net_rbio)
+        return;
+
+    if (qc->ch != NULL && !ossl_quic_channel_set_net_rbio(qc->ch, net_rbio))
+        return;
+
+    BIO_free(qc->net_rbio);
+    qc->net_rbio = net_rbio;
+
+    /*
+     * If what we have is not pollable (e.g. a BIO_dgram_pair) disable blocking
+     * mode as we do not support it for non-pollable BIOs.
+     */
+    if (net_rbio != NULL) {
+        BIO_POLL_DESCRIPTOR d = {0};
+
+        if (!BIO_get_rpoll_descriptor(net_rbio, &d)
+            || d.type != BIO_POLL_DESCRIPTOR_TYPE_SOCK_FD) {
+            qc->blocking = 0;
+            qc->can_poll_net_rbio = 0;
+        } else {
+            qc->can_poll_net_rbio = 1;
+        }
+    }
+}
+
+void ossl_quic_conn_set0_net_wbio(QUIC_CONNECTION *qc, BIO *net_wbio)
+{
+    if (qc->net_wbio == net_wbio)
+        return;
+
+    if (qc->ch != NULL && !ossl_quic_channel_set_net_wbio(qc->ch, net_wbio))
+        return;
+
+    BIO_free(qc->net_wbio);
+    qc->net_wbio = net_wbio;
+
+    if (net_wbio != NULL) {
+        BIO_POLL_DESCRIPTOR d = {0};
+
+        if (!BIO_get_wpoll_descriptor(net_wbio, &d)
+            || d.type != BIO_POLL_DESCRIPTOR_TYPE_SOCK_FD) {
+            qc->blocking = 0;
+            qc->can_poll_net_wbio = 0;
+        } else {
+            qc->can_poll_net_wbio = 1;
+        }
+
+        /*
+         * If we do not have a peer address yet, and we have not started trying
+         * to connect yet, try to autodetect one.
+         */
+        if (BIO_ADDR_family(&qc->init_peer_addr) == AF_UNSPEC
+            && !qc->started) {
+            if (!csm_analyse_init_peer_addr(net_wbio, &qc->init_peer_addr))
+                /* best effort */
+                BIO_ADDR_clear(&qc->init_peer_addr);
+
+            if (qc->ch != NULL)
+                ossl_quic_channel_set_peer_addr(qc->ch, &qc->init_peer_addr);
+        }
+    }
+}
+
+BIO *ossl_quic_conn_get_net_rbio(const QUIC_CONNECTION *qc)
+{
+    return qc->net_rbio;
+}
+
+BIO *ossl_quic_conn_get_net_wbio(const QUIC_CONNECTION *qc)
+{
+    return qc->net_wbio;
+}
+
+int ossl_quic_conn_get_blocking_mode(const QUIC_CONNECTION *qc)
+{
+    return qc->blocking;
+}
+
+int ossl_quic_conn_set_blocking_mode(QUIC_CONNECTION *qc, int blocking)
 {
+    /* Cannot enable blocking mode if we do not have pollable FDs. */
+    if (blocking != 0 &&
+        (!qc->can_poll_net_rbio || !qc->can_poll_net_wbio))
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_UNSUPPORTED, NULL);
+
+    qc->blocking = (blocking != 0);
     return 1;
 }
 
-int ossl_quic_read(SSL *s, void *buf, size_t len, size_t *readbytes)
+int ossl_quic_conn_set_initial_peer_addr(QUIC_CONNECTION *qc,
+                                         const BIO_ADDR *peer_addr)
 {
-    BIO *rbio = SSL_get_rbio(s);
+    if (qc->started)
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED,
+                                           NULL);
 
-    if (rbio == NULL)
-        return 0;
+    if (peer_addr == NULL) {
+        BIO_ADDR_clear(&qc->init_peer_addr);
+        return 1;
+    }
+
+    qc->init_peer_addr = *peer_addr;
+    return 1;
+}
 
-    return BIO_read_ex(rbio, buf, len, readbytes);
+/*
+ * QUIC Front-End I/O API: Asynchronous I/O Management
+ * ===================================================
+ *
+ *   (BIO/)SSL_tick                 => ossl_quic_tick
+ *   (BIO/)SSL_get_tick_timeout     => ossl_quic_get_tick_timeout
+ *   (BIO/)SSL_get_poll_fd          => ossl_quic_get_poll_fd
+ *
+ */
+
+/* Returns 1 if the connection is being used in blocking mode. */
+static int blocking_mode(const QUIC_CONNECTION *qc)
+{
+    return qc->blocking;
 }
 
-int ossl_quic_peek(SSL *s, void *buf, size_t len, size_t *readbytes)
+/* SSL_tick; ticks the reactor. */
+int ossl_quic_tick(QUIC_CONNECTION *qc)
 {
+    if (qc->ch == NULL)
+        return 1;
+
+    ossl_quic_reactor_tick(ossl_quic_channel_get_reactor(qc->ch));
     return 1;
 }
 
-int ossl_quic_write(SSL *s, const void *buf, size_t len, size_t *written)
+/*
+ * SSL_get_tick_timeout. Get the time in milliseconds until the SSL object
+ * should be ticked by the application by calling SSL_tick(). tv is set to 0 if
+ * the object should be ticked immediately and tv->tv_sec is set to -1 if no
+ * timeout is currently active.
+ */
+int ossl_quic_get_tick_timeout(QUIC_CONNECTION *qc, struct timeval *tv)
 {
-    BIO *wbio = SSL_get_wbio(s);
+    OSSL_TIME deadline = ossl_time_infinite();
 
-    if (wbio == NULL)
+    if (qc->ch != NULL)
+        deadline
+            = ossl_quic_reactor_get_tick_deadline(ossl_quic_channel_get_reactor(qc->ch));
+
+    if (ossl_time_is_infinite(deadline)) {
+        tv->tv_sec  = -1;
+        tv->tv_usec = 0;
+        return 1;
+    }
+
+    *tv = ossl_time_to_timeval(ossl_time_subtract(deadline, ossl_time_now()));
+    return 1;
+}
+
+/* SSL_get_rpoll_descriptor */
+int ossl_quic_get_rpoll_descriptor(QUIC_CONNECTION *qc, BIO_POLL_DESCRIPTOR *desc)
+{
+    if (desc == NULL || qc->net_rbio == NULL)
         return 0;
 
-    return BIO_write_ex(wbio, buf, len, written);
+    return BIO_get_rpoll_descriptor(qc->net_rbio, desc);
 }
 
-int ossl_quic_shutdown(SSL *s)
+/* SSL_get_wpoll_descriptor */
+int ossl_quic_get_wpoll_descriptor(QUIC_CONNECTION *qc, BIO_POLL_DESCRIPTOR *desc)
 {
-    return 1;
+    if (desc == NULL || qc->net_wbio == NULL)
+        return 0;
+
+    return BIO_get_wpoll_descriptor(qc->net_wbio, desc);
 }
 
+/* SSL_net_read_desired */
+int ossl_quic_get_net_read_desired(QUIC_CONNECTION *qc)
+{
+    if (qc->ch == NULL)
+        return 0;
+
+    return ossl_quic_reactor_net_read_desired(ossl_quic_channel_get_reactor(qc->ch));
+}
+
+/* SSL_net_write_desired */
+int ossl_quic_get_net_write_desired(QUIC_CONNECTION *qc)
+{
+    if (qc->ch == NULL)
+        return 0;
+
+    return ossl_quic_reactor_net_write_desired(ossl_quic_channel_get_reactor(qc->ch));
+}
+
+/*
+ * QUIC Front-End I/O API: Connection Lifecycle Operations
+ * =======================================================
+ *
+ *         SSL_do_handshake         => ossl_quic_do_handshake
+ *         SSL_set_connect_state    => ossl_quic_set_connect_state
+ *         SSL_set_accept_state     => ossl_quic_set_accept_state
+ *         SSL_shutdown             => ossl_quic_shutdown
+ *         SSL_ctrl                 => ossl_quic_ctrl
+ *   (BIO/)SSL_connect              => ossl_quic_connect
+ *   (BIO/)SSL_accept               => ossl_quic_accept
+ *
+ */
+
+/* SSL_shutdown */
+static int quic_shutdown_wait(void *arg)
+{
+    QUIC_CONNECTION *qc = arg;
+
+    return qc->ch == NULL || ossl_quic_channel_is_terminated(qc->ch);
+}
+
+int ossl_quic_conn_shutdown(QUIC_CONNECTION *qc, uint64_t flags,
+                            const SSL_SHUTDOWN_EX_ARGS *args,
+                            size_t args_len)
+{
+    if (!ensure_channel(qc))
+        return -1;
+
+    ossl_quic_channel_local_close(qc->ch,
+                                  args != NULL ? args->quic_error_code : 0);
+
+    /* TODO(QUIC): !SSL_SHUTDOWN_FLAG_NO_STREAM_FLUSH */
+
+    if (ossl_quic_channel_is_terminated(qc->ch))
+        return 1;
+
+    if (blocking_mode(qc) && (flags & SSL_SHUTDOWN_FLAG_RAPID) == 0)
+        block_until_pred(qc, quic_shutdown_wait, NULL, 0);
+    else
+        ossl_quic_reactor_tick(ossl_quic_channel_get_reactor(qc->ch));
+
+    return ossl_quic_channel_is_terminated(qc->ch);
+}
+
+/* SSL_ctrl */
 long ossl_quic_ctrl(SSL *s, int cmd, long larg, void *parg)
 {
-    return 0;
+    QUIC_CONNECTION *qc = QUIC_CONNECTION_FROM_SSL(s);
+
+    if (!expect_quic_conn(qc))
+        return 0;
+
+    switch (cmd) {
+    case SSL_CTRL_MODE:
+        /* Cannot enable EPW while AON write in progress. */
+        if (qc->aon_write_in_progress)
+            larg &= ~SSL_MODE_ENABLE_PARTIAL_WRITE;
+
+        qc->ssl_mode |= (uint32_t)larg;
+        return qc->ssl_mode;
+    case SSL_CTRL_CLEAR_MODE:
+        qc->ssl_mode &= ~(uint32_t)larg;
+        return qc->ssl_mode;
+    default:
+        return 0;
+    }
 }
 
-long ossl_quic_ctx_ctrl(SSL_CTX *s, int cmd, long larg, void *parg)
+/* SSL_set_connect_state */
+void ossl_quic_set_connect_state(QUIC_CONNECTION *qc)
 {
-    return 0;
+    /* Cannot be changed after handshake started */
+    if (qc->started)
+        return;
+
+    qc->as_server = 0;
 }
 
-long ossl_quic_callback_ctrl(SSL *s, int cmd, void (*fp) (void))
+/* SSL_set_accept_state */
+void ossl_quic_set_accept_state(QUIC_CONNECTION *qc)
 {
-    return 0;
+    /* Cannot be changed after handshake started */
+    if (qc->started)
+        return;
+
+    qc->as_server = 1;
 }
 
-long ossl_quic_ctx_callback_ctrl(SSL_CTX *s, int cmd, void (*fp) (void))
+/* SSL_do_handshake */
+struct quic_handshake_wait_args {
+    QUIC_CONNECTION     *qc;
+};
+
+static int quic_handshake_wait(void *arg)
 {
+    struct quic_handshake_wait_args *args = arg;
+
+    if (!ossl_quic_channel_is_active(args->qc->ch))
+        return -1;
+
+    if (ossl_quic_channel_is_handshake_complete(args->qc->ch))
+        return 1;
+
     return 0;
 }
 
-size_t ossl_quic_pending(const SSL *s)
+static int configure_channel(QUIC_CONNECTION *qc)
 {
-    return 0;
+    assert(qc->ch != NULL);
+
+    if (!ossl_quic_channel_set_net_rbio(qc->ch, qc->net_rbio)
+        || !ossl_quic_channel_set_net_wbio(qc->ch, qc->net_wbio)
+        || !ossl_quic_channel_set_peer_addr(qc->ch, &qc->init_peer_addr))
+        return 0;
+
+    return 1;
+}
+
+static int ensure_channel(QUIC_CONNECTION *qc)
+{
+    QUIC_CHANNEL_ARGS args = {0};
+
+    if (qc->ch != NULL)
+        return 1;
+
+    args.libctx     = qc->ssl.ctx->libctx;
+    args.propq      = qc->ssl.ctx->propq;
+    args.is_server  = 0;
+    args.tls        = qc->tls;
+
+    qc->ch = ossl_quic_channel_new(&args);
+    if (qc->ch == NULL)
+        return 0;
+
+    return 1;
+}
+
+/*
+ * Creates a channel and configures it with the information we have accumulated
+ * via calls made to us from the application prior to starting a handshake
+ * attempt.
+ */
+static int ensure_channel_and_start(QUIC_CONNECTION *qc)
+{
+    if (!ensure_channel(qc))
+        return 0;
+
+    if (!configure_channel(qc)
+        || !ossl_quic_channel_start(qc->ch)) {
+        ossl_quic_channel_free(qc->ch);
+        qc->ch = NULL;
+        return 0;
+    }
+
+    qc->stream0 = ossl_quic_channel_get_stream_by_id(qc->ch, 0);
+    if (qc->stream0 == NULL) {
+        ossl_quic_channel_free(qc->ch);
+        qc->ch = NULL;
+        return 0;
+    }
+
+    qc->started = 1;
+    return 1;
+}
+
+int ossl_quic_do_handshake(QUIC_CONNECTION *qc)
+{
+    int ret;
+
+    if (qc->ch != NULL && ossl_quic_channel_is_handshake_complete(qc->ch))
+        /* Handshake already completed. */
+        return 1;
+
+    if (qc->ch != NULL && ossl_quic_channel_is_term_any(qc->ch))
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, SSL_R_PROTOCOL_IS_SHUTDOWN, NULL);
+
+    if (BIO_ADDR_family(&qc->init_peer_addr) == AF_UNSPEC) {
+        /* Peer address must have been set. */
+        QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_PASSED_INVALID_ARGUMENT, NULL);
+        return -1; /* Non-protocol error */
+    }
+
+    if (qc->as_server) {
+        /* TODO(QUIC): Server mode not currently supported */
+        QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_PASSED_INVALID_ARGUMENT, NULL);
+        return -1; /* Non-protocol error */
+    }
+
+    if (qc->net_rbio == NULL || qc->net_wbio == NULL) {
+        /* Need read and write BIOs. */
+        QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_PASSED_INVALID_ARGUMENT, NULL);
+        return -1; /* Non-protocol error */
+    }
+
+    /*
+     * Start connection process. Note we may come here multiple times in
+     * non-blocking mode, which is fine.
+     */
+    if (!ensure_channel_and_start(qc)) {
+        QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+        return -1; /* Non-protocol error */
+    }
+
+    if (ossl_quic_channel_is_handshake_complete(qc->ch))
+        /* The handshake is now done. */
+        return 1;
+
+    if (blocking_mode(qc)) {
+        /* In blocking mode, wait for the handshake to complete. */
+        struct quic_handshake_wait_args args;
+
+        args.qc     = qc;
+
+        ret = block_until_pred(qc, quic_handshake_wait, &args, 0);
+        if (!ossl_quic_channel_is_active(qc->ch)) {
+            QUIC_RAISE_NON_NORMAL_ERROR(qc, SSL_R_PROTOCOL_IS_SHUTDOWN, NULL);
+            return 0; /* Shutdown before completion */
+        } else if (ret <= 0) {
+            QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+            return -1; /* Non-protocol error */
+        }
+
+        assert(ossl_quic_channel_is_handshake_complete(qc->ch));
+        return 1;
+    } else {
+        /* Try to advance the reactor. */
+        ossl_quic_reactor_tick(ossl_quic_channel_get_reactor(qc->ch));
+
+        if (ossl_quic_channel_is_handshake_complete(qc->ch))
+            /* The handshake is now done. */
+            return 1;
+
+        /* Otherwise, indicate that the handshake isn't done yet. */
+        QUIC_RAISE_NORMAL_ERROR(qc, SSL_ERROR_WANT_READ);
+        return -1; /* Non-protocol error */
+    }
 }
 
-long ossl_quic_default_timeout(void)
+/* SSL_connect */
+int ossl_quic_connect(SSL *s)
 {
+    QUIC_CONNECTION *qc = QUIC_CONNECTION_FROM_SSL(s);
+
+    if (!expect_quic_conn(qc))
+        return 0;
+
+    /* Ensure we are in connect state (no-op if non-idle). */
+    ossl_quic_set_connect_state(qc);
+
+    /* Begin or continue the handshake */
+    return ossl_quic_do_handshake(qc);
+}
+
+/* SSL_accept */
+int ossl_quic_accept(SSL *s)
+{
+    QUIC_CONNECTION *qc = QUIC_CONNECTION_FROM_SSL(s);
+
+    if (!expect_quic_conn(qc))
+        return 0;
+
+    /* Ensure we are in accept state (no-op if non-idle). */
+    ossl_quic_set_accept_state(qc);
+
+    /* Begin or continue the handshake */
+    return ossl_quic_do_handshake(qc);
+}
+
+/*
+ * QUIC Front-End I/O API: Steady-State Operations
+ * ===============================================
+ *
+ * Here we dispatch calls to the steady-state front-end I/O API functions; that
+ * is, the functions used during the established phase of a QUIC connection
+ * (e.g. SSL_read, SSL_write).
+ *
+ * Each function must handle both blocking and non-blocking modes. As discussed
+ * above, all QUIC I/O is implemented using non-blocking mode internally.
+ *
+ *         SSL_get_error        => partially implemented by ossl_quic_get_error
+ *   (BIO/)SSL_read             => ossl_quic_read
+ *   (BIO/)SSL_write            => ossl_quic_write
+ *         SSL_pending          => ossl_quic_pending
+ *         SSL_stream_conclude  => ossl_quic_conn_stream_conclude
+ */
+
+/* SSL_get_error */
+int ossl_quic_get_error(const QUIC_CONNECTION *qc, int i)
+{
+    return qc->last_error;
+}
+
+/*
+ * SSL_write
+ * ---------
+ *
+ * The set of functions below provide the implementation of the public SSL_write
+ * function. We must handle:
+ *
+ *   - both blocking and non-blocking operation at the application level,
+ *     depending on how we are configured;
+ *
+ *   - SSL_MODE_ENABLE_PARTIAL_WRITE being on or off;
+ *
+ *   - SSL_MODE_ACCEPT_MOVING_WRITE_BUFFER.
+ *
+ */
+static void quic_post_write(QUIC_CONNECTION *qc, int did_append, int do_tick)
+{
+    /*
+     * We have appended at least one byte to the stream.
+     * Potentially mark stream as active, depending on FC.
+     */
+    if (did_append)
+        ossl_quic_stream_map_update_state(ossl_quic_channel_get_qsm(qc->ch),
+                                          qc->stream0);
+
+    /*
+     * Try and send.
+     *
+     * TODO(QUIC): It is probably inefficient to try and do this immediately,
+     * plus we should eventually consider Nagle's algorithm.
+     */
+    if (do_tick)
+        ossl_quic_reactor_tick(ossl_quic_channel_get_reactor(qc->ch));
+}
+
+struct quic_write_again_args {
+    QUIC_CONNECTION     *qc;
+    const unsigned char *buf;
+    size_t              len;
+    size_t              total_written;
+};
+
+static int quic_write_again(void *arg)
+{
+    struct quic_write_again_args *args = arg;
+    size_t actual_written = 0;
+
+    if (!ossl_quic_channel_is_active(args->qc->ch))
+        /* If connection is torn down due to an error while blocking, stop. */
+        return -2;
+
+    if (!ossl_quic_sstream_append(args->qc->stream0->sstream,
+                                  args->buf, args->len, &actual_written))
+        return -2;
+
+    quic_post_write(args->qc, actual_written > 0, 0);
+
+    args->buf           += actual_written;
+    args->len           -= actual_written;
+    args->total_written += actual_written;
+
+    if (args->len == 0)
+        /* Written everything, done. */
+        return 1;
+
+    /* Not written everything yet, keep trying. */
     return 0;
 }
 
-int ossl_quic_num_ciphers(void)
+static int quic_write_blocking(QUIC_CONNECTION *qc, const void *buf, size_t len,
+                               size_t *written)
 {
+    int res;
+    struct quic_write_again_args args;
+    size_t actual_written = 0;
+
+    /* First make a best effort to append as much of the data as possible. */
+    if (!ossl_quic_sstream_append(qc->stream0->sstream, buf, len,
+                                  &actual_written)) {
+        /* Stream already finished or allocation error. */
+        *written = 0;
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+    }
+
+    quic_post_write(qc, actual_written > 0, 1);
+
+    if (actual_written == len) {
+        /* Managed to append everything on the first try. */
+        *written = actual_written;
+        return 1;
+    }
+
+    /*
+     * We did not manage to append all of the data immediately, so the stream
+     * buffer has probably filled up. This means we need to block until some of
+     * it is freed up.
+     */
+    args.qc             = qc;
+    args.buf            = (const unsigned char *)buf + actual_written;
+    args.len            = len - actual_written;
+    args.total_written  = 0;
+
+    res = block_until_pred(qc, quic_write_again, &args, 0);
+    if (res <= 0) {
+        if (!ossl_quic_channel_is_active(qc->ch))
+            return QUIC_RAISE_NON_NORMAL_ERROR(qc, SSL_R_PROTOCOL_IS_SHUTDOWN, NULL);
+        else
+            return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+    }
+
+    *written = args.total_written;
     return 1;
 }
 
-const SSL_CIPHER *ossl_quic_get_cipher(unsigned int u)
+/*
+ * Functions to manage All-or-Nothing (AON) (that is, non-ENABLE_PARTIAL_WRITE)
+ * write semantics.
+ */
+static void aon_write_begin(QUIC_CONNECTION *qc, const unsigned char *buf,
+                            size_t buf_len, size_t already_sent)
 {
-    static const SSL_CIPHER ciph = { 0 };
+    assert(!qc->aon_write_in_progress);
 
-    return &ciph;
+    qc->aon_write_in_progress = 1;
+    qc->aon_buf_base          = buf;
+    qc->aon_buf_pos           = already_sent;
+    qc->aon_buf_len           = buf_len;
 }
 
-int ossl_quic_renegotiate_check(SSL *ssl, int initok)
+static void aon_write_finish(QUIC_CONNECTION *qc)
+{
+    qc->aon_write_in_progress   = 0;
+    qc->aon_buf_base            = NULL;
+    qc->aon_buf_pos             = 0;
+    qc->aon_buf_len             = 0;
+}
+
+static int quic_write_nonblocking_aon(QUIC_CONNECTION *qc, const void *buf,
+                                      size_t len, size_t *written)
+{
+    const void *actual_buf;
+    size_t actual_len, actual_written = 0;
+    int accept_moving_buffer
+        = ((qc->ssl_mode & SSL_MODE_ACCEPT_MOVING_WRITE_BUFFER) != 0);
+
+    if (qc->aon_write_in_progress) {
+        /*
+         * We are in the middle of an AON write (i.e., a previous write did not
+         * manage to append all data to the SSTREAM and we have Enable Partial
+         * Write (EPW) mode disabled.)
+         */
+        if ((!accept_moving_buffer && qc->aon_buf_base != buf)
+            || len != qc->aon_buf_len)
+            /*
+             * Pointer must not have changed if we are not in accept moving
+             * buffer mode. Length must never change.
+             */
+            return QUIC_RAISE_NON_NORMAL_ERROR(qc, SSL_R_BAD_WRITE_RETRY, NULL);
+
+        actual_buf = (unsigned char *)buf + qc->aon_buf_pos;
+        actual_len = len - qc->aon_buf_pos;
+        assert(actual_len > 0);
+    } else {
+        actual_buf = buf;
+        actual_len = len;
+    }
+
+    /* First make a best effort to append as much of the data as possible. */
+    if (!ossl_quic_sstream_append(qc->stream0->sstream, actual_buf, actual_len,
+                                  &actual_written)) {
+        /* Stream already finished or allocation error. */
+        *written = 0;
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+    }
+
+    quic_post_write(qc, actual_written > 0, 1);
+
+    if (actual_written == actual_len) {
+        /* We have sent everything. */
+        if (qc->aon_write_in_progress) {
+            /*
+             * We have sent everything, and we were in the middle of an AON
+             * write. The output write length is the total length of the AON
+             * buffer, not however many bytes we managed to write to the stream
+             * in this call.
+             */
+            *written = qc->aon_buf_len;
+            aon_write_finish(qc);
+        } else {
+            *written = actual_written;
+        }
+
+        return 1;
+    }
+
+    if (qc->aon_write_in_progress) {
+        /*
+         * AON write is in progress but we have not written everything yet. We
+         * may have managed to send zero bytes, or some number of bytes less
+         * than the total remaining which need to be appended during this
+         * AON operation.
+         */
+        qc->aon_buf_pos += actual_written;
+        assert(qc->aon_buf_pos < qc->aon_buf_len);
+        return QUIC_RAISE_NORMAL_ERROR(qc, SSL_ERROR_WANT_WRITE);
+    }
+
+    /*
+     * Not in an existing AON operation but partial write is not enabled, so we
+     * need to begin a new AON operation. However we needn't bother if we didn't
+     * actually append anything.
+     */
+    if (actual_written > 0)
+        aon_write_begin(qc, buf, len, actual_written);
+
+    /*
+     * AON - We do not publicly admit to having appended anything until AON
+     * completes.
+     */
+    *written = 0;
+    return QUIC_RAISE_NORMAL_ERROR(qc, SSL_ERROR_WANT_WRITE);
+}
+
+static int quic_write_nonblocking_epw(QUIC_CONNECTION *qc, const void *buf, size_t len,
+                                      size_t *written)
+{
+    /* Simple best effort operation. */
+    if (!ossl_quic_sstream_append(qc->stream0->sstream, buf, len, written)) {
+        /* Stream already finished or allocation error. */
+        *written = 0;
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+    }
+
+    quic_post_write(qc, *written > 0, 1);
+    return 1;
+}
+
+int ossl_quic_write(SSL *s, const void *buf, size_t len, size_t *written)
+{
+    QUIC_CONNECTION *qc = QUIC_CONNECTION_FROM_SSL(s);
+    int partial_write = ((qc->ssl_mode & SSL_MODE_ENABLE_PARTIAL_WRITE) != 0);
+
+    *written = 0;
+
+    if (!expect_quic_conn(qc))
+        return 0;
+
+    if (qc->ch != NULL && ossl_quic_channel_is_term_any(qc->ch))
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, SSL_R_PROTOCOL_IS_SHUTDOWN, NULL);
+
+    /*
+     * If we haven't finished the handshake, try to advance it.
+     * We don't accept writes until the handshake is completed.
+     */
+    if (ossl_quic_do_handshake(qc) < 1)
+        return 0;
+
+    if (qc->stream0 == NULL || qc->stream0->sstream == NULL)
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+
+    if (blocking_mode(qc))
+        return quic_write_blocking(qc, buf, len, written);
+    else if (partial_write)
+        return quic_write_nonblocking_epw(qc, buf, len, written);
+    else
+        return quic_write_nonblocking_aon(qc, buf, len, written);
+}
+
+/*
+ * SSL_read
+ * --------
+ */
+struct quic_read_again_args {
+    QUIC_CONNECTION *qc;
+    QUIC_STREAM     *stream;
+    void            *buf;
+    size_t          len;
+    size_t          *bytes_read;
+    int             peek;
+};
+
+static int quic_read_actual(QUIC_CONNECTION *qc,
+                            QUIC_STREAM *stream,
+                            void *buf, size_t buf_len,
+                            size_t *bytes_read,
+                            int peek)
 {
+    int is_fin = 0;
+
+    /* If the receive part of the stream is over, issue EOF. */
+    if (stream->recv_fin_retired)
+        return QUIC_RAISE_NORMAL_ERROR(qc, SSL_ERROR_ZERO_RETURN);
+
+    if (stream->rstream == NULL)
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+
+    if (peek) {
+        if (!ossl_quic_rstream_peek(stream->rstream, buf, buf_len,
+                                    bytes_read, &is_fin))
+            return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+
+    } else {
+        if (!ossl_quic_rstream_read(stream->rstream, buf, buf_len,
+                                    bytes_read, &is_fin))
+            return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+    }
+
+    if (!peek) {
+        if (*bytes_read > 0) {
+            /*
+             * We have read at least one byte from the stream. Inform stream-level
+             * RXFC of the retirement of controlled bytes. Update the active stream
+             * status (the RXFC may now want to emit a frame granting more credit to
+             * the peer).
+             */
+            OSSL_RTT_INFO rtt_info;
+
+            ossl_statm_get_rtt_info(ossl_quic_channel_get_statm(qc->ch), &rtt_info);
+
+            if (!ossl_quic_rxfc_on_retire(&qc->stream0->rxfc, *bytes_read,
+                                          rtt_info.smoothed_rtt))
+                return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+        }
+
+        if (is_fin)
+            stream->recv_fin_retired = 1;
+
+        if (*bytes_read > 0)
+            ossl_quic_stream_map_update_state(ossl_quic_channel_get_qsm(qc->ch),
+                                              qc->stream0);
+    }
+
     return 1;
 }
+
+static int quic_read_again(void *arg)
+{
+    struct quic_read_again_args *args = arg;
+
+    if (!ossl_quic_channel_is_active(args->qc->ch)) {
+        /* If connection is torn down due to an error while blocking, stop. */
+        QUIC_RAISE_NON_NORMAL_ERROR(args->qc, SSL_R_PROTOCOL_IS_SHUTDOWN, NULL);
+        return -1;
+    }
+
+    if (!quic_read_actual(args->qc, args->stream,
+                          args->buf, args->len, args->bytes_read,
+                          args->peek))
+        return -1;
+
+    if (*args->bytes_read > 0)
+        /* got at least one byte, the SSL_read op can finish now */
+        return 1;
+
+    return 0; /* did not read anything, keep trying */
+}
+
+static int quic_read(SSL *s, void *buf, size_t len, size_t *bytes_read, int peek)
+{
+    int res;
+    QUIC_CONNECTION *qc = QUIC_CONNECTION_FROM_SSL(s);
+    struct quic_read_again_args args;
+
+    *bytes_read = 0;
+
+    if (!expect_quic_conn(qc))
+        return 0;
+
+    if (qc->ch != NULL && ossl_quic_channel_is_term_any(qc->ch))
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, SSL_R_PROTOCOL_IS_SHUTDOWN, NULL);
+
+    /* If we haven't finished the handshake, try to advance it. */
+    if (ossl_quic_do_handshake(qc) < 1)
+        return 0; /* ossl_quic_do_handshake raised error here */
+
+    if (qc->stream0 == NULL)
+        return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+
+    if (!quic_read_actual(qc, qc->stream0, buf, len, bytes_read, peek))
+        return 0; /* quic_read_actual raised error here */
+
+    if (*bytes_read > 0) {
+        /*
+         * Even though we succeeded, tick the reactor here to ensure we are
+         * handling other aspects of the QUIC connection.
+         */
+        ossl_quic_reactor_tick(ossl_quic_channel_get_reactor(qc->ch));
+        return 1;
+    } else if (blocking_mode(qc)) {
+        /*
+         * We were not able to read anything immediately, so our stream
+         * buffer is empty. This means we need to block until we get
+         * at least one byte.
+         */
+        args.qc         = qc;
+        args.stream     = qc->stream0;
+        args.buf        = buf;
+        args.len        = len;
+        args.bytes_read = bytes_read;
+        args.peek       = peek;
+
+        res = block_until_pred(qc, quic_read_again, &args, 0);
+        if (res == 0)
+            return QUIC_RAISE_NON_NORMAL_ERROR(qc, ERR_R_INTERNAL_ERROR, NULL);
+        else if (res < 0)
+            return 0; /* quic_read_again raised error here */
+
+        return 1;
+    } else {
+        /* We did not get any bytes and are not in blocking mode. */
+        return QUIC_RAISE_NORMAL_ERROR(qc, SSL_ERROR_WANT_READ);
+    }
+}
+
+int ossl_quic_read(SSL *s, void *buf, size_t len, size_t *bytes_read)
+{
+    return quic_read(s, buf, len, bytes_read, 0);
+}
+
+int ossl_quic_peek(SSL *s, void *buf, size_t len, size_t *bytes_read)
+{
+    return quic_read(s, buf, len, bytes_read, 1);
+}
+
+/*
+ * SSL_pending
+ * -----------
+ */
+size_t ossl_quic_pending(const SSL *s)
+{
+    const QUIC_CONNECTION *qc = QUIC_CONNECTION_FROM_CONST_SSL(s);
+    size_t avail = 0;
+    int fin = 0;
+
+    if (!expect_quic_conn(qc))
+        return 0;
+
+    if (qc->stream0 == NULL || qc->stream0->rstream == NULL)
+        /* Cannot raise errors here because we are const, just fail. */
+        return 0;
+
+    if (!ossl_quic_rstream_available(qc->stream0->rstream, &avail, &fin))
+        return 0;
+
+    return avail;
+}
+
+/*
+ * SSL_stream_conclude
+ * -------------------
+ */
+int ossl_quic_conn_stream_conclude(QUIC_CONNECTION *qc)
+{
+    QUIC_STREAM *qs = qc->stream0;
+
+    if (qs == NULL || qs->sstream == NULL)
+        return 0;
+
+    if (!ossl_quic_channel_is_active(qc->ch)
+        || ossl_quic_sstream_get_final_size(qs->sstream, NULL))
+        return 1;
+
+    ossl_quic_sstream_fin(qs->sstream);
+    quic_post_write(qc, 1, 1);
+    return 1;
+}
+
+/*
+ * QUIC Front-End I/O API: SSL_CTX Management
+ * ==========================================
+ */
+
+long ossl_quic_ctx_ctrl(SSL_CTX *ctx, int cmd, long larg, void *parg)
+{
+    switch (cmd) {
+    default:
+        return ssl3_ctx_ctrl(ctx, cmd, larg, parg);
+    }
+}
+
+long ossl_quic_callback_ctrl(SSL *s, int cmd, void (*fp) (void))
+{
+    return ssl3_callback_ctrl(s, cmd, fp);
+}
+
+long ossl_quic_ctx_callback_ctrl(SSL_CTX *ctx, int cmd, void (*fp) (void))
+{
+    return ssl3_ctx_callback_ctrl(ctx, cmd, fp);
+}
+
+int ossl_quic_renegotiate_check(SSL *ssl, int initok)
+{
+    /* We never do renegotiation. */
+    return 0;
+}
+
+/*
+ * This is the subset of TLS1.3 ciphers which can be used with QUIC and which we
+ * actually support.
+ *
+ * TODO(QUIC): CCM support
+ */
+static SSL_CIPHER tls13_quic_ciphers[] = {
+    {
+        1,
+        TLS1_3_RFC_AES_128_GCM_SHA256,
+        TLS1_3_RFC_AES_128_GCM_SHA256,
+        TLS1_3_CK_AES_128_GCM_SHA256,
+        SSL_kANY,
+        SSL_aANY,
+        SSL_AES128GCM,
+        SSL_AEAD,
+        TLS1_3_VERSION, TLS1_3_VERSION,
+        0, 0,
+        SSL_HIGH,
+        SSL_HANDSHAKE_MAC_SHA256,
+        128,
+        128,
+    }, {
+        1,
+        TLS1_3_RFC_AES_256_GCM_SHA384,
+        TLS1_3_RFC_AES_256_GCM_SHA384,
+        TLS1_3_CK_AES_256_GCM_SHA384,
+        SSL_kANY,
+        SSL_aANY,
+        SSL_AES256GCM,
+        SSL_AEAD,
+        TLS1_3_VERSION, TLS1_3_VERSION,
+        0, 0,
+        SSL_HIGH,
+        SSL_HANDSHAKE_MAC_SHA384,
+        256,
+        256,
+    },
+    {
+        1,
+        TLS1_3_RFC_CHACHA20_POLY1305_SHA256,
+        TLS1_3_RFC_CHACHA20_POLY1305_SHA256,
+        TLS1_3_CK_CHACHA20_POLY1305_SHA256,
+        SSL_kANY,
+        SSL_aANY,
+        SSL_CHACHA20POLY1305,
+        SSL_AEAD,
+        TLS1_3_VERSION, TLS1_3_VERSION,
+        0, 0,
+        SSL_HIGH,
+        SSL_HANDSHAKE_MAC_SHA256,
+        256,
+        256,
+    }
+};
+
+int ossl_quic_num_ciphers(void)
+{
+    return OSSL_NELEM(tls13_quic_ciphers);
+}
+
+const SSL_CIPHER *ossl_quic_get_cipher(unsigned int u)
+{
+    if (u >= OSSL_NELEM(tls13_quic_ciphers))
+        return NULL;
+
+    return &tls13_quic_ciphers[u];
+}