QUIC_CHANNEL: Implementation
authorHugo Landau <hlandau@openssl.org>
Thu, 17 Nov 2022 15:00:41 +0000 (15:00 +0000)
committerHugo Landau <hlandau@openssl.org>
Fri, 13 Jan 2023 13:20:14 +0000 (13:20 +0000)
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19703)

include/internal/quic_channel.h [new file with mode: 0644]
ssl/quic/build.info
ssl/quic/quic_channel.c [new file with mode: 0644]
ssl/quic/quic_channel_local.h [new file with mode: 0644]

diff --git a/include/internal/quic_channel.h b/include/internal/quic_channel.h
new file mode 100644 (file)
index 0000000..6c2364a
--- /dev/null
@@ -0,0 +1,159 @@
+/*
+ * Copyright 2022 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
+ * https://www.openssl.org/source/license.html
+ */
+
+#ifndef OSSL_QUIC_CHANNEL_H
+# define OSSL_QUIC_CHANNEL_H
+
+# include <openssl/ssl.h>
+# include "internal/quic_types.h"
+# include "internal/quic_stream_map.h"
+# include "internal/quic_reactor.h"
+# include "internal/quic_statm.h"
+# include "internal/time.h"
+
+/*
+ * QUIC Channel
+ * ============
+ *
+ * A QUIC channel (QUIC_CHANNEL) is an object which binds together all of the
+ * various pieces of QUIC into a single top-level object, and handles connection
+ * state which is not specific to the client or server roles. In particular, it
+ * is strictly separated from the libssl front end I/O API personality layer,
+ * and is not an SSL object.
+ *
+ * The name QUIC_CHANNEL is chosen because QUIC_CONNECTION is already in use,
+ * but functionally these relate to the same thing (a QUIC connection). The use
+ * of two separate objects ensures clean separation between the API personality
+ * layer and common code for handling connections, and between the functionality
+ * which is specific to clients and which is specific to servers, and the
+ * functionality which is common to both.
+ *
+ * The API personality layer provides SSL objects (e.g. a QUIC_CONNECTION) which
+ * consume a QUIC channel and implement a specific public API. Things which are
+ * handled by the API personality layer include emulation of blocking semantics,
+ * handling of SSL object mode flags like non-partial write mode, etc.
+ *
+ * Where the QUIC_CHANNEL is used in a server role, there is one QUIC_CHANNEL
+ * per connection. In the future a QUIC Channel Manager will probably be defined
+ * to handle ownership of resources which are shared between connections (e.g.
+ * demuxers). Since we only use server-side functionality for dummy test servers
+ * for now, which only need to handle one connection at a time, this is not
+ * currently modelled.
+ */
+
+#define QUIC_CHANNEL_STATE_IDLE                        0
+#define QUIC_CHANNEL_STATE_ACTIVE                      1
+#define QUIC_CHANNEL_STATE_TERMINATING_CLOSING         2
+#define QUIC_CHANNEL_STATE_TERMINATING_DRAINING        3
+#define QUIC_CHANNEL_STATE_TERMINATED                  4
+
+typedef struct quic_channel_args_st {
+    OSSL_LIB_CTX *libctx;
+    const char *propq;
+    int is_server;
+} QUIC_CHANNEL_ARGS;
+
+typedef struct quic_channel_st QUIC_CHANNEL;
+
+/*
+ * Create a new QUIC channel using the given arguments. The argument structure
+ * does not need to remain allocated. Returns NULL on failure.
+ */
+QUIC_CHANNEL *ossl_quic_channel_new(const QUIC_CHANNEL_ARGS *args);
+
+/* No-op if ch is NULL. */
+void ossl_quic_channel_free(QUIC_CHANNEL *ch);
+
+/*
+ * Connection Lifecycle Events
+ * ===========================
+ *
+ * Various events that can be raised on the channel by other parts of the QUIC
+ * implementation. Some of these are suitable for general use by any part of the
+ * code (e.g. ossl_quic_channel_raise_protocol_error), others are for very
+ * specific use by particular components only (e.g.
+ * ossl_quic_channel_on_handshake_confirmed).
+ *
+ */
+
+/*
+ * To be used by a QUIC connection. Starts the channel. For a client-mode
+ * channel, this starts sending the first handshake layer message, etc. Can only
+ * be called in the idle state; successive calls are ignored.
+ */
+int ossl_quic_channel_start(QUIC_CHANNEL *ch);
+
+/* Start a locally initiated connection shutdown. */
+void ossl_quic_channel_local_close(QUIC_CHANNEL *ch);
+
+/*
+ * Called when the handshake is confirmed.
+ */
+int ossl_quic_channel_on_handshake_confirmed(QUIC_CHANNEL *ch);
+
+/*
+ * Raises a protocol error. This is intended to be the universal call suitable
+ * for handling of all peer-triggered protocol violations or errors detected by
+ * us. We specify a QUIC transport-scope error code and optional frame type
+ * which was responsible. If a frame type is not applicable, specify zero. The
+ * reason string is not currently handled, but should be a string of static
+ * storage duration. If the connection has already terminated due to a previous
+ * protocol error, this is a no-op; first error wins.
+ */
+void ossl_quic_channel_raise_protocol_error(QUIC_CHANNEL *ch,
+                                            uint64_t error_code,
+                                            uint64_t frame_type,
+                                            const char *reason);
+
+/* For RXDP use. */
+void ossl_quic_channel_on_remote_conn_close(QUIC_CHANNEL *ch,
+                                            OSSL_QUIC_FRAME_CONN_CLOSE *f);
+
+/*
+ * Queries and Accessors
+ * =====================
+ */
+
+/* Gets the reactor which can be used to tick/poll on the channel. */
+QUIC_REACTOR *ossl_quic_channel_get_reactor(QUIC_CHANNEL *ch);
+
+/* Gets the QSM used with the channel. */
+QUIC_STREAM_MAP *ossl_quic_channel_get_qsm(QUIC_CHANNEL *ch);
+
+/* Gets the statistics manager used with the channel. */
+OSSL_STATM *ossl_quic_channel_get_statm(QUIC_CHANNEL *ch);
+
+/*
+ * Gets/sets the current peer address. Generally this should be used before
+ * starting a channel in client mode.
+ */
+int ossl_quic_channel_get_peer_addr(QUIC_CHANNEL *ch, BIO_ADDR *peer_addr);
+int ossl_quic_channel_set_peer_addr(QUIC_CHANNEL *ch, const BIO_ADDR *peer_addr);
+
+/* Gets/sets the underlying network read and write BIOs. */
+BIO *ossl_quic_channel_get_net_rbio(QUIC_CHANNEL *ch);
+BIO *ossl_quic_channel_get_net_wbio(QUIC_CHANNEL *ch);
+int ossl_quic_channel_set0_net_rbio(QUIC_CHANNEL *ch, BIO *net_rbio);
+int ossl_quic_channel_set0_net_wbio(QUIC_CHANNEL *ch, BIO *net_wbio);
+
+/*
+ * Returns an existing stream by stream ID. Returns NULL if the stream does not
+ * exist.
+ */
+QUIC_STREAM *ossl_quic_channel_get_stream_by_id(QUIC_CHANNEL *ch,
+                                                uint64_t stream_id);
+
+/* Returns 1 if channel is terminating or terminated. */
+int ossl_quic_channel_is_term_any(const QUIC_CHANNEL *ch);
+int ossl_quic_channel_is_terminating(const QUIC_CHANNEL *ch);
+int ossl_quic_channel_is_terminated(const QUIC_CHANNEL *ch);
+int ossl_quic_channel_is_active(const QUIC_CHANNEL *ch);
+int ossl_quic_channel_is_handshake_complete(const QUIC_CHANNEL *ch);
+
+#endif
index 15aa53a359e3e1ba5ea00d3d44c1df0d8217c881..5f4d761b26e9e59408d60a8f26b13a689a77f2a4 100644 (file)
@@ -10,3 +10,4 @@ SOURCE[$LIBSSL]=quic_stream_map.c
 SOURCE[$LIBSSL]=quic_sf_list.c quic_rstream.c quic_sstream.c
 SOURCE[$LIBSSL]=quic_dummy_handshake.c
 SOURCE[$LIBSSL]=quic_reactor.c
+SOURCE[$LIBSSL]=quic_channel.c
diff --git a/ssl/quic/quic_channel.c b/ssl/quic/quic_channel.c
new file mode 100644 (file)
index 0000000..64ccf16
--- /dev/null
@@ -0,0 +1,1627 @@
+/*
+ * Copyright 2022 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
+ * https://www.openssl.org/source/license.html
+ */
+
+#include "internal/quic_channel.h"
+#include "internal/quic_error.h"
+#include "internal/quic_rx_depack.h"
+#include "../ssl_local.h"
+#include "quic_channel_local.h"
+#include <openssl/rand.h>
+
+#define INIT_DCID_LEN           8
+#define INIT_CRYPTO_BUF_LEN     8192
+#define INIT_APP_BUF_LEN        8192
+
+static int ch_rx(QUIC_CHANNEL *ch);
+static int ch_tx(QUIC_CHANNEL *ch);
+static void ch_tick(QUIC_TICK_RESULT *res, void *arg);
+static void ch_rx_handle_packet(QUIC_CHANNEL *ch);
+static OSSL_TIME ch_determine_next_tick_deadline(QUIC_CHANNEL *ch);
+static int ch_retry(QUIC_CHANNEL *ch,
+                    const unsigned char *retry_token,
+                    size_t retry_token_len,
+                    const QUIC_CONN_ID *retry_scid);
+static void ch_cleanup(QUIC_CHANNEL *ch);
+static int ch_generate_transport_params(QUIC_CHANNEL *ch);
+static int ch_on_transport_params(const unsigned char *params,
+                                  size_t params_len,
+                                  void *arg);
+static int ch_on_handshake_alert(void *arg, unsigned char alert_code);
+static int ch_on_handshake_complete(void *arg);
+static int ch_on_handshake_yield_secret(uint32_t enc_level, int direction,
+                                        uint32_t suite_id, EVP_MD *md,
+                                        const unsigned char *secret,
+                                        size_t secret_len,
+                                        void *arg);
+static int ch_on_crypto_recv(unsigned char *buf, size_t buf_len,
+                             size_t *bytes_read, void *arg);
+static int crypto_ensure_empty(QUIC_RSTREAM *rstream);
+static int ch_on_crypto_send(const unsigned char *buf, size_t buf_len,
+                             size_t *consumed, void *arg);
+static OSSL_TIME get_time(void *arg);
+static uint64_t get_stream_limit(int uni, void *arg);
+static int rx_early_validate(QUIC_PN pn, int pn_space, void *arg);
+static int ch_retry(QUIC_CHANNEL *ch,
+                    const unsigned char *retry_token,
+                    size_t retry_token_len,
+                    const QUIC_CONN_ID *retry_scid);
+static void ch_update_idle(QUIC_CHANNEL *ch);
+static int ch_discard_el(QUIC_CHANNEL *ch,
+                         uint32_t enc_level);
+static void ch_on_idle_timeout(QUIC_CHANNEL *ch);
+static void ch_update_idle(QUIC_CHANNEL *ch);
+static void ch_on_terminating_timeout(QUIC_CHANNEL *ch);
+static void ch_start_terminating(QUIC_CHANNEL *ch,
+                                 const QUIC_TERMINATE_CAUSE *tcause);
+
+static int gen_rand_conn_id(OSSL_LIB_CTX *libctx, size_t len, QUIC_CONN_ID *cid)
+{
+    if (len > QUIC_MAX_CONN_ID_LEN)
+        return 0;
+
+    cid->id_len = (unsigned char)len;
+
+    if (RAND_bytes_ex(libctx, cid->id, len, len * 8) != 1) {
+        cid->id_len = 0;
+        return 0;
+    }
+
+    return 1;
+}
+
+/*
+ * QUIC Channel Initialization and Teardown
+ * ========================================
+ */
+static int ch_init(QUIC_CHANNEL *ch)
+{
+    OSSL_QUIC_TX_PACKETISER_ARGS txp_args = {0};
+    OSSL_QTX_ARGS qtx_args = {0};
+    OSSL_QRX_ARGS qrx_args = {0};
+    QUIC_DHS_ARGS dhs_args = {0};
+    uint32_t pn_space;
+
+    // TODO CLIENT ONLY
+    if (!gen_rand_conn_id(ch->libctx, INIT_DCID_LEN, &ch->init_dcid))
+        goto err;
+
+    /* We plug in a network write BIO to the QTX later when we get one. */
+    qtx_args.mdpl = QUIC_MIN_INITIAL_DGRAM_LEN;
+    ch->rx_max_udp_payload_size = qtx_args.mdpl;
+
+    ch->qtx = ossl_qtx_new(&qtx_args);
+    if (ch->qtx == NULL)
+        goto err;
+
+    ch->txpim = ossl_quic_txpim_new();
+    if (ch->txpim == NULL)
+        goto err;
+
+    ch->cfq = ossl_quic_cfq_new();
+    if (ch->cfq == NULL)
+        goto err;
+
+    if (!ossl_quic_txfc_init(&ch->conn_txfc, NULL))
+        goto err;
+
+    if (!ossl_quic_rxfc_init(&ch->conn_rxfc, NULL,
+                             2  * 1024 * 1024,
+                             10 * 1024 * 1024,
+                             get_time, NULL))
+        goto err;
+
+    if (!ossl_statm_init(&ch->statm))
+        goto err;
+
+    ch->have_statm = 1;
+    ch->cc_method = &ossl_cc_dummy_method;
+    if ((ch->cc_data = ch->cc_method->new(NULL, NULL, NULL)) == NULL)
+        goto err;
+
+    if ((ch->ackm = ossl_ackm_new(get_time, NULL, &ch->statm,
+                                  ch->cc_method, ch->cc_data)) == NULL)
+        goto err;
+
+    if (!ossl_quic_stream_map_init(&ch->qsm, get_stream_limit, ch))
+        goto err;
+
+    ch->have_qsm = 1;
+
+    /* We use a zero-length SCID. */
+    txp_args.cur_dcid           = ch->init_dcid;
+    txp_args.ack_delay_exponent = 3;
+    txp_args.qtx                = ch->qtx;
+    txp_args.txpim              = ch->txpim;
+    txp_args.cfq                = ch->cfq;
+    txp_args.ackm               = ch->ackm;
+    txp_args.qsm                = &ch->qsm;
+    txp_args.conn_txfc          = &ch->conn_txfc;
+    txp_args.conn_rxfc          = &ch->conn_rxfc;
+    txp_args.cc_method          = ch->cc_method;
+    txp_args.cc_data            = ch->cc_data;
+    txp_args.now                = get_time;
+    for (pn_space = QUIC_PN_SPACE_INITIAL; pn_space < QUIC_PN_SPACE_NUM; ++pn_space) {
+        ch->crypto_send[pn_space] = ossl_quic_sstream_new(INIT_CRYPTO_BUF_LEN);
+        if (ch->crypto_send[pn_space] == NULL)
+            goto err;
+
+        txp_args.crypto[pn_space] = ch->crypto_send[pn_space];
+    }
+
+    ch->txp = ossl_quic_tx_packetiser_new(&txp_args);
+    if (ch->txp == NULL)
+        goto err;
+
+    if ((ch->demux = ossl_quic_demux_new(/*BIO=*/NULL, /*Short CID Len=*/0,
+                                         1200, get_time, NULL)) == NULL)
+        goto err;
+
+    qrx_args.demux              = ch->demux;
+    qrx_args.short_conn_id_len  = 0; /* We use a zero-length SCID. */
+    qrx_args.max_deferred       = 32;
+
+    if ((ch->qrx = ossl_qrx_new(&qrx_args)) == NULL)
+        goto err;
+
+    if (!ossl_qrx_set_early_validation_cb(ch->qrx,
+                                          rx_early_validate,
+                                          ch))
+        goto err;
+
+    if (!ossl_qrx_add_dst_conn_id(ch->qrx, &txp_args.cur_scid))
+        goto err;
+
+    for (pn_space = QUIC_PN_SPACE_INITIAL; pn_space < QUIC_PN_SPACE_NUM; ++pn_space) {
+        ch->crypto_recv[pn_space] = ossl_quic_rstream_new(NULL, NULL);
+        if (ch->crypto_recv[pn_space] == NULL)
+            goto err;
+    }
+
+    if ((ch->stream0 = ossl_quic_stream_map_alloc(&ch->qsm, 0,
+                                                  QUIC_STREAM_INITIATOR_CLIENT
+                                                  | QUIC_STREAM_DIR_BIDI)) == NULL)
+        goto err;
+
+    if ((ch->stream0->sstream = ossl_quic_sstream_new(INIT_APP_BUF_LEN)) == NULL)
+        goto err;
+
+    if ((ch->stream0->rstream = ossl_quic_rstream_new(NULL, NULL)) == NULL)
+        goto err;
+
+    if (!ossl_quic_txfc_init(&ch->stream0->txfc, &ch->conn_txfc))
+        goto err;
+
+    if (!ossl_quic_rxfc_init(&ch->stream0->rxfc, &ch->conn_rxfc,
+                             1 * 1024 * 1024,
+                             5 * 1024 * 1024,
+                             get_time, NULL))
+        goto err;
+
+    /* Plug in the dummy handshake layer. */
+    dhs_args.crypto_send_cb             = ch_on_crypto_send;
+    dhs_args.crypto_send_cb_arg         = ch;
+    dhs_args.crypto_recv_cb             = ch_on_crypto_recv;
+    dhs_args.crypto_recv_cb_arg         = ch;
+    dhs_args.yield_secret_cb            = ch_on_handshake_yield_secret;
+    dhs_args.yield_secret_cb_arg        = ch;
+    dhs_args.got_transport_params_cb    = ch_on_transport_params;
+    dhs_args.got_transport_params_cb_arg= ch;
+    dhs_args.handshake_complete_cb      = ch_on_handshake_complete;
+    dhs_args.handshake_complete_cb_arg  = ch;
+    dhs_args.alert_cb                   = ch_on_handshake_alert;
+    dhs_args.alert_cb_arg               = ch;
+
+    if ((ch->dhs = ossl_quic_dhs_new(&dhs_args)) == NULL)
+        goto err;
+
+    /*
+     * Determine the QUIC Transport Parameters and serialize the transport
+     * parameters block. (For servers, we do this later as we must defer
+     * generation until we have received the client's transport parameters.)
+     */
+    if (!ch->is_server && !ch_generate_transport_params(ch))
+        goto err;
+
+    ch->rx_max_ack_delay        = QUIC_DEFAULT_MAX_ACK_DELAY;
+    ch->rx_ack_delay_exp        = QUIC_DEFAULT_ACK_DELAY_EXP;
+    ch->rx_active_conn_id_limit = QUIC_MIN_ACTIVE_CONN_ID_LIMIT;
+    ch->max_idle_timeout        = QUIC_DEFAULT_IDLE_TIMEOUT;
+    ch->tx_enc_level            = QUIC_ENC_LEVEL_INITIAL;
+    ch_update_idle(ch);
+    ossl_quic_reactor_init(&ch->rtor, ch_tick, ch,
+                           ch_determine_next_tick_deadline(ch));
+    return 1;
+
+err:
+    ch_cleanup(ch);
+    return 0;
+}
+
+static void ch_cleanup(QUIC_CHANNEL *ch)
+{
+    uint32_t pn_space;
+
+    if (ch->ackm != NULL)
+        for (pn_space = QUIC_PN_SPACE_INITIAL;
+             pn_space < QUIC_PN_SPACE_NUM;
+             ++pn_space)
+            ossl_ackm_on_pkt_space_discarded(ch->ackm, pn_space);
+
+    ossl_quic_tx_packetiser_free(ch->txp);
+    ossl_quic_txpim_free(ch->txpim);
+    ossl_quic_cfq_free(ch->cfq);
+    ossl_qtx_free(ch->qtx);
+    if (ch->cc_data != NULL)
+        ch->cc_method->free(ch->cc_data);
+    if (ch->have_statm)
+        ossl_statm_destroy(&ch->statm);
+    ossl_ackm_free(ch->ackm);
+
+    if (ch->stream0 != NULL) {
+        assert(ch->have_qsm);
+        ossl_quic_stream_map_release(&ch->qsm, ch->stream0); /* frees sstream */
+    }
+
+    if (ch->have_qsm)
+        ossl_quic_stream_map_cleanup(&ch->qsm);
+
+    for (pn_space = QUIC_PN_SPACE_INITIAL; pn_space < QUIC_PN_SPACE_NUM; ++pn_space) {
+        ossl_quic_sstream_free(ch->crypto_send[pn_space]);
+        ossl_quic_rstream_free(ch->crypto_recv[pn_space]);
+    }
+
+    ossl_qrx_pkt_release(ch->qrx_pkt);
+    ch->qrx_pkt = NULL;
+
+    ossl_quic_dhs_free(ch->dhs);
+    ossl_qrx_free(ch->qrx);
+    ossl_quic_demux_free(ch->demux);
+    OPENSSL_free(ch->local_transport_params);
+    BIO_free(ch->net_rbio);
+    BIO_free(ch->net_wbio);
+}
+
+QUIC_CHANNEL *ossl_quic_channel_new(const QUIC_CHANNEL_ARGS *args)
+{
+    QUIC_CHANNEL *ch = NULL;
+
+    if ((ch = OPENSSL_zalloc(sizeof(*ch))) == NULL)
+        return NULL;
+
+    ch->libctx      = args->libctx;
+    ch->propq       = args->propq;
+    ch->is_server   = args->is_server;
+
+    if (!ch_init(ch)) {
+        OPENSSL_free(ch);
+        return NULL;
+    }
+
+    return ch;
+}
+
+void ossl_quic_channel_free(QUIC_CHANNEL *ch)
+{
+    if (ch == NULL)
+        return;
+
+    ch_cleanup(ch);
+    OPENSSL_free(ch);
+}
+
+int ossl_quic_channel_get_peer_addr(QUIC_CHANNEL *ch, BIO_ADDR *peer_addr)
+{
+    *peer_addr = ch->cur_peer_addr;
+    return 1;
+}
+
+int ossl_quic_channel_set_peer_addr(QUIC_CHANNEL *ch, const BIO_ADDR *peer_addr)
+{
+    ch->cur_peer_addr = *peer_addr;
+    return 1;
+}
+
+QUIC_REACTOR *ossl_quic_channel_get_reactor(QUIC_CHANNEL *ch)
+{
+    return &ch->rtor;
+}
+
+QUIC_STREAM_MAP *ossl_quic_channel_get_qsm(QUIC_CHANNEL *ch)
+{
+    return &ch->qsm;
+}
+
+OSSL_STATM *ossl_quic_channel_get_statm(QUIC_CHANNEL *ch)
+{
+    return &ch->statm;
+}
+
+QUIC_STREAM *ossl_quic_channel_get_stream_by_id(QUIC_CHANNEL *ch,
+                                                uint64_t stream_id)
+{
+    return ossl_quic_stream_map_get_by_id(&ch->qsm, stream_id);
+}
+
+int ossl_quic_channel_is_active(const QUIC_CHANNEL *ch)
+{
+    return ch != NULL && ch->state == QUIC_CHANNEL_STATE_ACTIVE;
+}
+
+int ossl_quic_channel_is_terminating(const QUIC_CHANNEL *ch)
+{
+    return ch->state == QUIC_CHANNEL_STATE_TERMINATING_CLOSING
+        || ch->state == QUIC_CHANNEL_STATE_TERMINATING_DRAINING;
+}
+
+int ossl_quic_channel_is_terminated(const QUIC_CHANNEL *ch)
+{
+    return ch->state == QUIC_CHANNEL_STATE_TERMINATED;
+}
+
+int ossl_quic_channel_is_term_any(const QUIC_CHANNEL *ch)
+{
+    return ossl_quic_channel_is_terminating(ch)
+        || ossl_quic_channel_is_terminated(ch);
+}
+
+int ossl_quic_channel_is_handshake_complete(const QUIC_CHANNEL *ch)
+{
+    return ch->handshake_complete;
+}
+
+/*
+ * QUIC Channel: Callbacks from Miscellaneous Subsidiary Components
+ * ================================================================
+ */
+
+/* Used by various components. */
+static OSSL_TIME get_time(void *arg)
+{
+    return ossl_time_now();
+}
+
+/* Used by QSM. */
+static uint64_t get_stream_limit(int uni, void *arg)
+{
+    QUIC_CHANNEL *ch = arg;
+
+    return uni ? ch->max_local_streams_uni : ch->max_local_streams_bidi;
+}
+
+/*
+ * Called by QRX to determine if a packet is potentially invalid before trying
+ * to decrypt it.
+ */
+static int rx_early_validate(QUIC_PN pn, int pn_space, void *arg)
+{
+    QUIC_CHANNEL *ch = arg;
+
+    /* Potential duplicates should not be processed. */
+    if (!ossl_ackm_is_rx_pn_processable(ch->ackm, pn, pn_space))
+        return 0;
+
+    return 1;
+}
+
+/*
+ * QUIC Channel: Handshake Layer Event Handling
+ * ============================================
+ */
+static int ch_on_crypto_send(const unsigned char *buf, size_t buf_len,
+                             size_t *consumed, void *arg)
+{
+    int ret;
+    QUIC_CHANNEL *ch = arg;
+    uint32_t enc_level = ch->tx_enc_level;
+    uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
+    QUIC_SSTREAM *sstream = ch->crypto_send[pn_space];
+
+    if (!ossl_assert(sstream != NULL))
+        return 0;
+
+    ret = ossl_quic_sstream_append(sstream, buf, buf_len, consumed);
+    return ret;
+}
+
+static int crypto_ensure_empty(QUIC_RSTREAM *rstream)
+{
+    size_t avail = 0;
+    int is_fin = 0;
+
+    if (rstream == NULL)
+        return 1;
+
+    if (!ossl_quic_rstream_available(rstream, &avail, &is_fin))
+        return 0;
+
+    return avail == 0;
+}
+
+static int ch_on_crypto_recv(unsigned char *buf, size_t buf_len,
+                             size_t *bytes_read, void *arg)
+{
+    QUIC_CHANNEL *ch = arg;
+    QUIC_RSTREAM *rstream;
+    int is_fin = 0; /* crypto stream is never finished, so we don't use this */
+    uint32_t i;
+
+    /*
+     * After we move to a later EL we must not allow our peer to send any new
+     * bytes in the crypto stream on a previous EL. Retransmissions of old bytes
+     * are allowed.
+     *
+     * In practice we will only move to a new EL when we have consumed all bytes
+     * which should be sent on the crypto stream at a previous EL. For example,
+     * the Handshake EL should not be provisioned until we have completely
+     * consumed a TLS 1.3 ServerHello. Thus when we provision an EL the output
+     * of ossl_quic_rstream_available() should be 0 for all lower ELs. Thus if a
+     * given EL is available we simply ensure we have not received any further
+     * bytes at a lower EL.
+     */
+    for (i = QUIC_ENC_LEVEL_INITIAL; i < ch->tx_enc_level; ++i)
+        if (i != QUIC_ENC_LEVEL_0RTT &&
+            !crypto_ensure_empty(ch->crypto_recv[ossl_quic_enc_level_to_pn_space(i)])) {
+            /* Protocol violation (RFC 9001 s. 4.1.3) */
+            ossl_quic_channel_raise_protocol_error(ch, QUIC_ERR_PROTOCOL_VIOLATION,
+                                                   OSSL_QUIC_FRAME_TYPE_CRYPTO,
+                                                   "crypto stream data in wrong EL");
+            return 0;
+        }
+
+    rstream = ch->crypto_recv[ossl_quic_enc_level_to_pn_space(ch->tx_enc_level)];
+    if (rstream == NULL)
+        return 0;
+
+    return ossl_quic_rstream_read(rstream, buf, buf_len, bytes_read,
+                                  &is_fin);
+}
+
+static int ch_on_handshake_yield_secret(uint32_t enc_level, int direction,
+                                        uint32_t suite_id, EVP_MD *md,
+                                        const unsigned char *secret,
+                                        size_t secret_len,
+                                        void *arg)
+{
+    QUIC_CHANNEL *ch = arg;
+    uint32_t i;
+
+    if (enc_level < QUIC_ENC_LEVEL_HANDSHAKE || enc_level >= QUIC_ENC_LEVEL_NUM)
+        /* Invalid EL. */
+        return 0;
+
+    if (enc_level <= ch->tx_enc_level)
+        /*
+         * Does not make sense for us to try and provision an EL we have already
+         * attained.
+         */
+        return 0;
+
+    /*
+     * Ensure all crypto streams for previous ELs are now empty of available
+     * data.
+     */
+    for (i = QUIC_ENC_LEVEL_INITIAL; i < enc_level; ++i)
+        if (!crypto_ensure_empty(ch->crypto_recv[i])) {
+            /* Protocol violation (RFC 9001 s. 4.1.3) */
+            ossl_quic_channel_raise_protocol_error(ch, QUIC_ERR_PROTOCOL_VIOLATION,
+                                                   OSSL_QUIC_FRAME_TYPE_CRYPTO,
+                                                   "crypto stream data in wrong EL");
+            return 0;
+        }
+
+    if (direction) {
+        /* TX */
+        if (!ossl_qtx_provide_secret(ch->qtx, enc_level,
+                                     suite_id, md,
+                                     secret, secret_len))
+            return 0;
+
+        ch->tx_enc_level = enc_level;
+    } else {
+        /* RX */
+        if (!ossl_qrx_provide_secret(ch->qrx, enc_level,
+                                     suite_id, md,
+                                     secret, secret_len))
+            return 0;
+    }
+
+    return 1;
+}
+
+static int ch_on_handshake_complete(void *arg)
+{
+    QUIC_CHANNEL *ch = arg;
+
+    if (ch->handshake_complete)
+        return 0; /* this should not happen twice */
+
+    if (!ossl_assert(ch->tx_enc_level == QUIC_ENC_LEVEL_1RTT))
+        return 0;
+
+    if (!ch->got_remote_transport_params)
+        /*
+         * Was not a valid QUIC handshake if we did not get valid transport
+         * params.
+         */
+        return 0;
+
+    /* Don't need transport parameters anymore. */
+    OPENSSL_free(ch->local_transport_params);
+    ch->local_transport_params = NULL;
+
+    /* Tell TXP the handshake is complete. */
+    ossl_quic_tx_packetiser_notify_handshake_complete(ch->txp);
+
+    ch->handshake_complete = 1;
+    return 1;
+}
+
+static int ch_on_handshake_alert(void *arg, unsigned char alert_code)
+{
+    QUIC_CHANNEL *ch = arg;
+
+    ossl_quic_channel_raise_protocol_error(ch, QUIC_ERR_CRYPTO_ERR_BEGIN + alert_code,
+                                           0, "handshake alert");
+    return 1;
+}
+
+/*
+ * QUIC Channel: Transport Parameter Handling
+ * ==========================================
+ */
+
+/*
+ * Called by handshake layer when we receive QUIC Transport Parameters from the
+ * peer. Note that these are not authenticated until the handshake is marked
+ * as complete.
+ */
+static int ch_on_transport_params(const unsigned char *params,
+                                  size_t params_len,
+                                  void *arg)
+{
+    QUIC_CHANNEL *ch = arg;
+    PACKET pkt;
+    uint64_t id, v;
+    size_t len;
+    const unsigned char *body;
+    int got_orig_dcid = 0;
+    int got_initial_scid = 0;
+    int got_retry_scid = 0;
+    int got_initial_max_data = 0;
+    int got_initial_max_stream_data_bidi_local = 0;
+    int got_initial_max_stream_data_bidi_remote = 0;
+    int got_initial_max_stream_data_uni = 0;
+    int got_initial_max_streams_bidi = 0;
+    int got_initial_max_streams_uni = 0;
+    int got_ack_delay_exp = 0;
+    int got_max_ack_delay = 0;
+    int got_max_udp_payload_size = 0;
+    int got_max_idle_timeout = 0;
+    int got_active_conn_id_limit = 0;
+    QUIC_CONN_ID cid;
+
+    if (ch->got_remote_transport_params)
+        goto malformed;
+
+    if (!PACKET_buf_init(&pkt, params, params_len))
+        return 0;
+
+    while (PACKET_remaining(&pkt) > 0) {
+        if (!ossl_quic_wire_peek_transport_param(&pkt, &id))
+            goto malformed;
+
+        switch (id) {
+            case QUIC_TPARAM_ORIG_DCID:
+                if (got_orig_dcid)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_cid(&pkt, NULL, &cid))
+                    goto malformed;
+
+                /* Must match our initial DCID. */
+                if (!ossl_quic_conn_id_eq(&ch->init_dcid, &cid))
+                    goto malformed;
+
+                got_orig_dcid = 1;
+                break;
+
+            case QUIC_TPARAM_RETRY_SCID:
+                if (got_retry_scid || !ch->doing_retry)
+                    /* must not appear more than once or if retry not done */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_cid(&pkt, NULL, &cid))
+                    goto malformed;
+
+                /* Must match Retry packet SCID. */
+                if (!ossl_quic_conn_id_eq(&ch->retry_scid, &cid))
+                    goto malformed;
+
+                got_retry_scid = 1;
+                break;
+
+            case QUIC_TPARAM_INITIAL_SCID:
+                if (got_initial_scid)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_cid(&pkt, NULL, &cid))
+                    goto malformed;
+
+                /* Must match SCID of first Initial packet from server. */
+                if (!ossl_quic_conn_id_eq(&ch->init_scid, &cid))
+                    goto malformed;
+
+                got_initial_scid = 1;
+                break;
+
+            case QUIC_TPARAM_INITIAL_MAX_DATA:
+                if (got_initial_max_data)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v))
+                    goto malformed;
+
+                ossl_quic_txfc_bump_cwm(&ch->conn_txfc, v);
+                got_initial_max_data = 1;
+                break;
+
+            case QUIC_TPARAM_INITIAL_MAX_STREAM_DATA_BIDI_LOCAL:
+                if (got_initial_max_stream_data_bidi_local)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v))
+                    goto malformed;
+
+                /*
+                 * This is correct; the BIDI_LOCAL TP governs streams created by
+                 * the endpoint which sends the TP, i.e., our peer.
+                 */
+                ch->init_max_stream_data_bidi_remote = v;
+                got_initial_max_stream_data_bidi_local = 1;
+                break;
+
+            case QUIC_TPARAM_INITIAL_MAX_STREAM_DATA_BIDI_REMOTE:
+                if (got_initial_max_stream_data_bidi_remote)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v))
+                    goto malformed;
+
+                /*
+                 * This is correct; the BIDI_REMOTE TP governs streams created
+                 * by the endpoint which receives the TP, i.e., us.
+                 */
+                ch->init_max_stream_data_bidi_local = v;
+
+                /* Apply to stream 0. */
+                ossl_quic_txfc_bump_cwm(&ch->stream0->txfc, v);
+                got_initial_max_stream_data_bidi_remote = 1;
+                break;
+
+            case QUIC_TPARAM_INITIAL_MAX_STREAM_DATA_UNI:
+                if (got_initial_max_stream_data_uni)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v))
+                    goto malformed;
+
+                ch->init_max_stream_data_uni_remote = v;
+                got_initial_max_stream_data_uni = 1;
+                break;
+
+            case QUIC_TPARAM_ACK_DELAY_EXP:
+                if (got_ack_delay_exp)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v)
+                    || v > QUIC_MAX_ACK_DELAY_EXP)
+                    goto malformed;
+
+                ch->rx_ack_delay_exp = (unsigned char)v;
+                got_ack_delay_exp = 1;
+                break;
+
+            case QUIC_TPARAM_MAX_ACK_DELAY:
+                if (got_max_ack_delay)
+                    /* must not appear more than once */
+                    return 0;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v)
+                    || v >= (((uint64_t)1) << 14))
+                    goto malformed;
+
+                ch->rx_max_ack_delay = v;
+                got_max_ack_delay = 1;
+                break;
+
+            case QUIC_TPARAM_INITIAL_MAX_STREAMS_BIDI:
+                if (got_initial_max_streams_bidi)
+                    /* must not appear more than once */
+                    return 0;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v)
+                    || v > (((uint64_t)1) << 60))
+                    goto malformed;
+
+                assert(ch->max_local_streams_bidi == 0);
+                ch->max_local_streams_bidi = v;
+                got_initial_max_streams_bidi = 1;
+                break;
+
+            case QUIC_TPARAM_INITIAL_MAX_STREAMS_UNI:
+                if (got_initial_max_streams_uni)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v)
+                    || v > (((uint64_t)1) << 60))
+                    goto malformed;
+
+                assert(ch->max_local_streams_uni == 0);
+                ch->max_local_streams_uni = v;
+                got_initial_max_streams_uni = 1;
+                break;
+
+            case QUIC_TPARAM_MAX_IDLE_TIMEOUT:
+                if (got_max_idle_timeout)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v))
+                    goto malformed;
+
+                if (v < ch->max_idle_timeout)
+                    ch->max_idle_timeout = v;
+
+                ch_update_idle(ch);
+                got_max_idle_timeout = 1;
+                break;
+
+            case QUIC_TPARAM_MAX_UDP_PAYLOAD_SIZE:
+                if (got_max_udp_payload_size)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v)
+                    || v < QUIC_MIN_INITIAL_DGRAM_LEN)
+                    goto malformed;
+
+                ch->rx_max_udp_payload_size = v;
+                got_max_udp_payload_size    = 1;
+                break;
+
+            case QUIC_TPARAM_ACTIVE_CONN_ID_LIMIT:
+                if (got_active_conn_id_limit)
+                    /* must not appear more than once */
+                    goto malformed;
+
+                if (!ossl_quic_wire_decode_transport_param_int(&pkt, &id, &v)
+                    || v < QUIC_MIN_ACTIVE_CONN_ID_LIMIT)
+                    goto malformed;
+
+                ch->rx_active_conn_id_limit = v;
+                got_active_conn_id_limit = 1;
+                break;
+
+            /*
+             * TODO(QUIC): Handle:
+             *   QUIC_TPARAM_STATELESS_RESET_TOKEN
+             *   QUIC_TPARAM_PREFERRED_ADDR
+             */
+
+            case QUIC_TPARAM_DISABLE_ACTIVE_MIGRATION:
+                /* We do not currently handle migration, so nothing to do. */
+            default:
+                /* Skip over and ignore. */
+                body = ossl_quic_wire_decode_transport_param_bytes(&pkt, &id,
+                                                                   &len);
+                if (body == NULL)
+                    goto malformed;
+
+                break;
+        }
+    }
+
+    if (!got_orig_dcid || !got_initial_scid || got_retry_scid != ch->doing_retry)
+        /* Transport parameters were not valid. */
+        goto malformed;
+
+    ch->got_remote_transport_params = 1;
+
+    if (got_initial_max_data || got_initial_max_stream_data_bidi_remote
+        || got_initial_max_streams_bidi || got_initial_max_streams_uni)
+        /* If FC credit was bumped, we may now be able to send. */
+        ossl_quic_stream_map_update_state(&ch->qsm, ch->stream0);
+
+    /* If we are a server, we now generate our own transport parameters. */
+    if (ch->is_server && !ch_generate_transport_params(ch)) {
+        ossl_quic_channel_raise_protocol_error(ch, QUIC_ERR_INTERNAL_ERROR, 0,
+                                               "internal error");
+        return 0;
+    }
+
+    return 1;
+
+malformed:
+    ossl_quic_channel_raise_protocol_error(ch, QUIC_ERR_TRANSPORT_PARAMETER_ERROR,
+                                           0, "bad transport parameter");
+    return 0;
+}
+
+/*
+ * Called when we want to generate transport parameters. This is called
+ * immediately at instantiation time for a client and after we receive the
+ * client's transport parameters for a server.
+ */
+static int ch_generate_transport_params(QUIC_CHANNEL *ch)
+{
+    int ok = 0;
+    BUF_MEM *buf_mem = NULL;
+    WPACKET wpkt;
+    int wpkt_valid = 0;
+    size_t buf_len = 0;
+
+    if (ch->local_transport_params != NULL)
+        goto err;
+
+    if ((buf_mem = BUF_MEM_new()) == NULL)
+        goto err;
+
+    if (!WPACKET_init(&wpkt, buf_mem))
+        goto err;
+
+    wpkt_valid = 1;
+
+    if (ossl_quic_wire_encode_transport_param_bytes(&wpkt, QUIC_TPARAM_DISABLE_ACTIVE_MIGRATION,
+                                                    NULL, 0) == NULL)
+        goto err;
+
+    if (ossl_quic_wire_encode_transport_param_bytes(&wpkt, QUIC_TPARAM_INITIAL_SCID,
+                                                    NULL, 0) == NULL)
+        goto err;
+
+    if (!ossl_quic_wire_encode_transport_param_int(&wpkt, QUIC_TPARAM_MAX_IDLE_TIMEOUT,
+                                                   ch->max_idle_timeout))
+        goto err;
+
+    if (!ossl_quic_wire_encode_transport_param_int(&wpkt, QUIC_TPARAM_MAX_UDP_PAYLOAD_SIZE,
+                                                   QUIC_MIN_INITIAL_DGRAM_LEN))
+        goto err;
+
+    if (!ossl_quic_wire_encode_transport_param_int(&wpkt, QUIC_TPARAM_ACTIVE_CONN_ID_LIMIT,
+                                                   4))
+        goto err;
+
+    if (!ossl_quic_wire_encode_transport_param_int(&wpkt, QUIC_TPARAM_INITIAL_MAX_DATA,
+                                                   ossl_quic_rxfc_get_cwm(&ch->conn_rxfc)))
+        goto err;
+
+    /*
+     * We actually want the default CWM for a new RXFC, but here we just use
+     * stream0 as a representative specimen. TODO(QUIC): revisit this when we
+     * support multiple streams.
+     */
+    if (!ossl_quic_wire_encode_transport_param_int(&wpkt, QUIC_TPARAM_INITIAL_MAX_STREAM_DATA_BIDI_LOCAL,
+                                                   ossl_quic_rxfc_get_cwm(&ch->stream0->rxfc)))
+        goto err;
+
+    if (!ossl_quic_wire_encode_transport_param_int(&wpkt, QUIC_TPARAM_INITIAL_MAX_STREAM_DATA_BIDI_REMOTE,
+                                                   ossl_quic_rxfc_get_cwm(&ch->stream0->rxfc)))
+        goto err;
+
+    if (!ossl_quic_wire_encode_transport_param_int(&wpkt, QUIC_TPARAM_INITIAL_MAX_STREAM_DATA_UNI,
+                                                   ossl_quic_rxfc_get_cwm(&ch->stream0->rxfc)))
+        goto err;
+
+    if (!ossl_quic_wire_encode_transport_param_int(&wpkt, QUIC_TPARAM_INITIAL_MAX_STREAMS_BIDI,
+                                                   0))
+        goto err;
+
+    if (!ossl_quic_wire_encode_transport_param_int(&wpkt, QUIC_TPARAM_INITIAL_MAX_STREAMS_UNI,
+                                                   0))
+        goto err;
+
+    if (!WPACKET_get_total_written(&wpkt, &buf_len))
+        goto err;
+
+    ch->local_transport_params = (unsigned char *)buf_mem->data;
+    buf_mem->data = NULL;
+
+    if (!WPACKET_finish(&wpkt))
+        goto err;
+
+    wpkt_valid = 0;
+
+    if (!ossl_quic_dhs_set_transport_params(ch->dhs, ch->local_transport_params,
+                                            buf_len))
+        goto err;
+
+    ok = 1;
+err:
+    if (wpkt_valid)
+        WPACKET_cleanup(&wpkt);
+    BUF_MEM_free(buf_mem);
+    return ok;
+}
+
+/*
+ * QUIC Channel: Ticker-Mutator
+ * ============================
+ */
+
+/*
+ * The central ticker function called by the reactor. This does everything, or
+ * at least everything network I/O related. Best effort - not allowed to fail
+ * "loudly".
+ */
+static void ch_tick(QUIC_TICK_RESULT *res, void *arg)
+{
+    OSSL_TIME now, deadline;
+    QUIC_CHANNEL *ch = arg;
+
+    /*
+     * When we tick the QUIC connection, we do everything we need to do
+     * periodically. In order, we:
+     *
+     *   - handle any incoming data from the network;
+     *   - handle any timer events which are due to fire (ACKM, etc.)
+     *   - write any data to the network due to be sent, to the extent
+     *     possible;
+     *   - determine the time at which we should next be ticked.
+     */
+
+    /* If we are in the TERMINATED state, there is nothing to do. */
+    if (ossl_quic_channel_is_terminated(ch)) {
+        res->want_net_read  = 0;
+        res->want_net_write = 0;
+        res->tick_deadline  = ossl_time_infinite();
+        return;
+    }
+
+    /*
+     * If we are in the TERMINATING state, check if the terminating timer has
+     * expired.
+     */
+    if (ossl_quic_channel_is_terminating(ch)) {
+        now = ossl_time_now();
+
+        if (ossl_time_compare(now, ch->terminate_deadline) >= 0) {
+            ch_on_terminating_timeout(ch);
+            res->want_net_read  = 0;
+            res->want_net_write = 0;
+            res->tick_deadline  = ossl_time_infinite();
+            return; /* abort normal processing, nothing to do */
+        }
+    }
+
+    /* Handle any incoming data from the network. */
+    ch_rx(ch);
+
+    /*
+     * Allow the handshake layer to check for any new incoming data and generate
+     * new outgoing data.
+     */
+    ossl_quic_dhs_tick(ch->dhs);
+
+    /*
+     * Handle any timer events which are due to fire; namely, the loss detection
+     * deadline and the idle timeout.
+     *
+     * ACKM ACK generation deadline is polled by TXP, so we don't need to handle
+     * it here.
+     */
+    now = ossl_time_now();
+    if (ossl_time_compare(now, ch->idle_deadline) >= 0) {
+        /*
+         * Idle timeout differs from normal protocol violation because we do not
+         * send a CONN_CLOSE frame; go straight to TERMINATED.
+         */
+        ch_on_idle_timeout(ch);
+        res->want_net_read  = 0;
+        res->want_net_write = 0;
+        res->tick_deadline  = ossl_time_infinite();
+        return;
+    }
+
+    deadline = ossl_ackm_get_loss_detection_deadline(ch->ackm);
+    if (!ossl_time_is_zero(deadline) && ossl_time_compare(now, deadline) >= 0)
+        ossl_ackm_on_timeout(ch->ackm);
+
+    /* Write any data to the network due to be sent. */
+    ch_tx(ch);
+
+    /* Determine the time at which we should next be ticked. */
+    res->tick_deadline = ch_determine_next_tick_deadline(ch);
+
+    /* Always process network input. */
+    res->want_net_read = 1;
+
+    /* We want to write to the network if we have any in our queue. */
+    res->want_net_write = (ossl_qtx_get_queue_len_datagrams(ch->qtx) > 0);
+}
+
+/* Process incoming packets and handle frames, if any. */
+static int ch_rx(QUIC_CHANNEL *ch)
+{
+    int handled_any = 0;
+
+    if (!ch->have_sent_any_pkt)
+        /*
+         * We have not sent anything yet, therefore there is no need to check
+         * for incoming data. TODO SERVER
+         */
+        return 1;
+
+    /*
+     * Get DEMUX to BIO_recvmmsg from the network and queue incoming datagrams
+     * to the appropriate QRX instance.
+     */
+    ossl_quic_demux_pump(ch->demux); /* best effort */
+
+    for (;;) {
+        assert(ch->qrx_pkt == NULL);
+
+        if (!ossl_qrx_read_pkt(ch->qrx, &ch->qrx_pkt))
+            break;
+
+        if (!handled_any)
+            ch_update_idle(ch);
+
+        ch_rx_handle_packet(ch); /* best effort */
+
+        /*
+         * Regardless of the outcome of frame handling, unref the packet.
+         * This will free the packet unless something added another
+         * reference to it during frame processing.
+         */
+        ossl_qrx_pkt_release(ch->qrx_pkt);
+        ch->qrx_pkt = NULL;
+
+        handled_any = 1;
+    }
+
+    /*
+     * When in TERMINATING - CLOSING, generate a CONN_CLOSE frame whenever we
+     * process one or more incoming packets.
+     */
+    if (handled_any && ch->state == QUIC_CHANNEL_STATE_TERMINATING_CLOSING)
+        ch->conn_close_queued = 1;
+
+    return 1;
+}
+
+/* Handles the packet currently in ch->qrx_pkt->hdr. */
+static void ch_rx_handle_packet(QUIC_CHANNEL *ch)
+{
+    uint32_t enc_level;
+
+    assert(ch->qrx_pkt != NULL);
+
+    if (ossl_quic_pkt_type_is_encrypted(ch->qrx_pkt->hdr->type)) {
+        if (!ch->have_received_enc_pkt) {
+            ch->init_scid = ch->qrx_pkt->hdr->src_conn_id;
+            ch->have_received_enc_pkt = 1;
+
+            /*
+             * We change to using the SCID in the first Initial packet as the
+             * DCID.
+             */
+            ossl_quic_tx_packetiser_set_cur_dcid(ch->txp, &ch->init_scid);
+        }
+
+        enc_level = ossl_quic_pkt_type_to_enc_level(ch->qrx_pkt->hdr->type);
+        if ((ch->el_discarded & (1U << enc_level)) != 0)
+            /* Do not process packets from ELs we have already discarded. */
+            return;
+    }
+
+    /* Handle incoming packet. */
+    switch (ch->qrx_pkt->hdr->type) {
+        case QUIC_PKT_TYPE_RETRY:
+            if (ch->doing_retry)
+                /* It is not allowed to ask a client to do a retry more than
+                 * once. */
+                return;
+
+            /* TODO if server */
+
+            if (ch->qrx_pkt->hdr->len <= QUIC_RETRY_INTEGRITY_TAG_LEN)
+                /* Packets with zero-length Retry Tokens are invalid. */
+                return;
+
+            /*
+             * TODO(QUIC): Theoretically this should probably be in the QRX.
+             * However because validation is dependent on context (namely the
+             * client's initial DCID) we can't do this cleanly. In the future we
+             * should probably add a callback to the QRX to let it call us (via
+             * the DEMUX) and ask us about the correct original DCID, rather
+             * than allow the QRX to emit a potentially malformed packet to the
+             * upper layers. However, special casing this will do for now.
+             */
+            if (!ossl_quic_validate_retry_integrity_tag(ch->libctx,
+                                                        ch->propq,
+                                                        ch->qrx_pkt->hdr,
+                                                        &ch->init_dcid))
+                /* Malformed retry packet, ignore. */
+                return;
+
+            ch_retry(ch, ch->qrx_pkt->hdr->data,
+                     ch->qrx_pkt->hdr->len - QUIC_RETRY_INTEGRITY_TAG_LEN,
+                     &ch->qrx_pkt->hdr->src_conn_id);
+            break;
+
+        case QUIC_PKT_TYPE_VERSION_NEG:
+            /* TODO(QUIC): Implement version negotiation */
+            break;
+
+        case QUIC_PKT_TYPE_0RTT:
+            /* TODO if server */
+            /* Clients should never receive 0-RTT packets */
+            break;
+
+        default:
+            if (ch->qrx_pkt->hdr->type == QUIC_PKT_TYPE_HANDSHAKE)
+                /*
+                 * We automatically drop INITIAL EL keys when first successfully
+                 * decrypting a HANDSHAKE packet, as per the RFC.
+                 */
+                ch_discard_el(ch, QUIC_ENC_LEVEL_INITIAL);
+
+            /* This packet contains frames, pass to the RXDP. */
+            ossl_quic_handle_frames(ch, ch->qrx_pkt); /* best effort */
+            break;
+    }
+}
+
+/* Try to generate packets and if possible, flush them to the network. */
+static int ch_tx(QUIC_CHANNEL *ch)
+{
+    if (ch->state == QUIC_CHANNEL_STATE_TERMINATING_CLOSING) {
+        /*
+         * While closing, only send CONN_CLOSE if we've received more traffic
+         * from the peer. Once we tell the TXP to generate CONN_CLOSE, all
+         * future calls to it generate CONN_CLOSE frames, so otherwise we would
+         * just constantly generate CONN_CLOSE frames.
+         */
+        if (!ch->conn_close_queued)
+            return 0;
+
+        ch->conn_close_queued = 0;
+    }
+
+    /*
+     * Send a packet, if we need to. Best effort. The TXP consults the CC and
+     * applies any limitations imposed by it, so we don't need to do it here.
+     *
+     * Best effort. In particular if TXP fails for some reason we should still
+     * flush any queued packets which we already generated.
+     */
+    if (ossl_quic_tx_packetiser_generate(ch->txp,
+                                         TX_PACKETISER_ARCHETYPE_NORMAL)
+        == TX_PACKETISER_RES_SENT_PKT)
+        ch->have_sent_any_pkt = 1;
+
+    ossl_qtx_flush_net(ch->qtx); /* best effort */
+    return 1;
+}
+
+/* Determine next tick deadline. */
+static OSSL_TIME ch_determine_next_tick_deadline(QUIC_CHANNEL *ch)
+{
+    OSSL_TIME deadline;
+    uint32_t pn_space;
+
+    deadline = ossl_ackm_get_loss_detection_deadline(ch->ackm);
+    if (ossl_time_is_zero(deadline))
+        deadline = ossl_time_infinite();
+
+    for (pn_space = QUIC_PN_SPACE_INITIAL; pn_space < QUIC_PN_SPACE_NUM; ++pn_space)
+        deadline = ossl_time_min(deadline,
+                                 ossl_ackm_get_ack_deadline(ch->ackm, pn_space));
+
+    /* When will CC let us send more? */
+    if (ossl_quic_tx_packetiser_has_pending(ch->txp, TX_PACKETISER_ARCHETYPE_NORMAL,
+                                            TX_PACKETISER_BYPASS_CC))
+        deadline = ossl_time_min(deadline,
+                                 ch->cc_method->get_next_credit_time(ch->cc_data));
+
+    /* Is the terminating timer armed? */
+    if (ossl_quic_channel_is_terminating(ch))
+        deadline = ossl_time_min(deadline,
+                                 ch->terminate_deadline);
+    else if (!ossl_time_is_infinite(ch->idle_deadline))
+        deadline = ossl_time_min(deadline,
+                                 ch->idle_deadline);
+
+    return deadline;
+}
+
+/*
+ * QUIC Channel: Network BIO Configuration
+ * =======================================
+ */
+
+/* Determines whether we can support a given poll descriptor. */
+static int validate_poll_descriptor(const BIO_POLL_DESCRIPTOR *d)
+{
+    if (d->type == BIO_POLL_DESCRIPTOR_TYPE_SOCK_FD && d->value.fd < 0)
+        return 0;
+
+    return 1;
+}
+
+BIO *ossl_quic_channel_get_net_rbio(QUIC_CHANNEL *ch)
+{
+    return ch->net_rbio;
+}
+
+BIO *ossl_quic_channel_get_net_wbio(QUIC_CHANNEL *ch)
+{
+    return ch->net_wbio;
+}
+
+int ossl_quic_channel_set0_net_rbio(QUIC_CHANNEL *ch, BIO *net_rbio)
+{
+    BIO_POLL_DESCRIPTOR d = {0};
+
+    if (ch->net_rbio == net_rbio)
+        return 1;
+
+    if (net_rbio != NULL) {
+        if (!BIO_get_rpoll_descriptor(net_rbio, &d))
+            /* Non-pollable BIO */
+            d.type = BIO_POLL_DESCRIPTOR_TYPE_NONE;
+
+        if (!validate_poll_descriptor(&d))
+            return 0;
+    }
+
+    ossl_quic_reactor_set_poll_r(&ch->rtor, &d);
+    BIO_free(ch->net_rbio);
+    ossl_quic_demux_set_bio(ch->demux, net_rbio);
+    ch->net_rbio = net_rbio;
+    return 1;
+}
+
+int ossl_quic_channel_set0_net_wbio(QUIC_CHANNEL *ch, BIO *net_wbio)
+{
+    BIO_POLL_DESCRIPTOR d = {0};
+
+    if (ch->net_wbio == net_wbio)
+        return 1;
+
+    if (net_wbio != NULL) {
+        if (!BIO_get_wpoll_descriptor(net_wbio, &d))
+            /* Non-pollable BIO */
+            d.type = BIO_POLL_DESCRIPTOR_TYPE_NONE;
+
+        if (!validate_poll_descriptor(&d))
+            return 0;
+    }
+
+    ossl_quic_reactor_set_poll_w(&ch->rtor, &d);
+    BIO_free(ch->net_wbio);
+    ossl_qtx_set_bio(ch->qtx, net_wbio);
+    ch->net_wbio = net_wbio;
+    return 1;
+}
+
+/*
+ * QUIC Channel: Lifecycle Events
+ * ==============================
+ */
+
+int ossl_quic_channel_start(QUIC_CHANNEL *ch)
+{
+    /* TODO SERVER */
+    if (ch->state != QUIC_CHANNEL_STATE_IDLE)
+        /* Calls to connect are idempotent */
+        return 1;
+
+    /* Inform QTX of peer address. */
+    if (!ossl_quic_tx_packetiser_set_peer(ch->txp, &ch->cur_peer_addr))
+        return 0;
+
+    /* Plug in secrets for the Initial EL. */
+    if (!ossl_quic_provide_initial_secret(ch->libctx,
+                                          ch->propq,
+                                          &ch->init_dcid,
+                                          /*is_server=*/0,
+                                          ch->qrx, ch->qtx))
+        return 0;
+
+    /* Change state. */
+    ch->state                   = QUIC_CHANNEL_STATE_ACTIVE;
+    ch->doing_proactive_ver_neg = 0; /* not currently supported */
+
+    /* Handshake layer: start (e.g. send CH). */
+    if (!ossl_quic_dhs_tick(ch->dhs))
+        return 0;
+
+    ossl_quic_reactor_tick(&ch->rtor); /* best effort */
+    return 1;
+}
+
+/* Start a locally initiated connection shutdown. */
+void ossl_quic_channel_local_close(QUIC_CHANNEL *ch)
+{
+    QUIC_TERMINATE_CAUSE tcause = {0};
+
+    if (ossl_quic_channel_is_term_any(ch))
+        return;
+
+    tcause.app = 1;
+    ch_start_terminating(ch, &tcause);
+}
+
+static void free_token(const unsigned char *buf, size_t buf_len, void *arg)
+{
+    OPENSSL_free((unsigned char *)buf);
+}
+
+/* Called when a server asks us to do a retry. */
+static int ch_retry(QUIC_CHANNEL *ch,
+                    const unsigned char *retry_token,
+                    size_t retry_token_len,
+                    const QUIC_CONN_ID *retry_scid)
+{
+    void *buf;
+
+    /* We change to using the SCID in the Retry packet as the DCID. */
+    if (!ossl_quic_tx_packetiser_set_cur_dcid(ch->txp, retry_scid))
+        return 0;
+
+    /*
+     * Now we retry. We will release the Retry packet immediately, so copy
+     * the token.
+     */
+    if ((buf = OPENSSL_malloc(retry_token_len)) == NULL)
+        return 0;
+
+    memcpy(buf, retry_token, retry_token_len);
+
+    ossl_quic_tx_packetiser_set_initial_token(ch->txp, buf, retry_token_len,
+                                              free_token, NULL);
+
+    ch->retry_scid  = *retry_scid;
+    ch->doing_retry = 1;
+
+    /*
+     * We need to stimulate the Initial EL to generate the first CRYPTO frame
+     * again. We can do this most cleanly by simply forcing the ACKM to consider
+     * the first Initial packet as lost, which it effectively was as the server
+     * hasn't processed it. This also maintains the desired behaviour with e.g.
+     * PNs not resetting and so on.
+     *
+     * The PN we used initially is always zero, because QUIC does not allow
+     * repeated retries.
+     */
+    if (!ossl_ackm_mark_packet_pseudo_lost(ch->ackm, QUIC_PN_SPACE_INITIAL,
+                                           /*PN=*/0))
+        return 0;
+
+    /*
+     * Plug in new secrets for the Initial EL. This is the only time we change
+     * the secrets for an EL after we already provisioned it.
+     */
+    if (!ossl_quic_provide_initial_secret(ch->libctx,
+                                          ch->propq,
+                                          &ch->retry_scid,
+                                          /*is_server=*/0,
+                                          ch->qrx, ch->qtx))
+        return 0;
+
+    return 1;
+}
+
+/* Called when an EL is to be discarded. */
+static int ch_discard_el(QUIC_CHANNEL *ch,
+                         uint32_t enc_level)
+{
+    if (!ossl_assert(enc_level < QUIC_ENC_LEVEL_1RTT))
+        return 0;
+
+    if ((ch->el_discarded & (1U << enc_level)) != 0)
+        /* Already done. */
+        return 1;
+
+    /* Best effort for all of these. */
+    ossl_quic_tx_packetiser_discard_enc_level(ch->txp, enc_level);
+    ossl_qrx_discard_enc_level(ch->qrx, enc_level);
+    ossl_qtx_discard_enc_level(ch->qtx, enc_level);
+
+    if (enc_level != QUIC_ENC_LEVEL_0RTT) {
+        uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
+
+        ossl_ackm_on_pkt_space_discarded(ch->ackm, pn_space);
+
+        /* We should still have crypto streams at this point. */
+        assert(ch->crypto_send[pn_space] != NULL);
+        assert(ch->crypto_recv[pn_space] != NULL);
+
+        /* Get rid of the crypto stream state for the EL. */
+        ossl_quic_sstream_free(ch->crypto_send[pn_space]);
+        ch->crypto_send[pn_space] = NULL;
+
+        ossl_quic_rstream_free(ch->crypto_recv[pn_space]);
+        ch->crypto_recv[pn_space] = NULL;
+    }
+
+    ch->el_discarded |= (1U << enc_level);
+    return 1;
+}
+
+/* Intended to be called by the RXDP. */
+int ossl_quic_channel_on_handshake_confirmed(QUIC_CHANNEL *ch)
+{
+    if (ch->handshake_confirmed)
+        return 1;
+
+    if (!ch->handshake_complete) {
+        /*
+         * Does not make sense for handshake to be confirmed before it is
+         * completed.
+         */
+        ossl_quic_channel_raise_protocol_error(ch, QUIC_ERR_PROTOCOL_VIOLATION,
+                                               OSSL_QUIC_FRAME_TYPE_HANDSHAKE_DONE,
+                                               "handshake cannot be confirmed "
+                                               "before it is completed");
+        return 0;
+    }
+
+    ch_discard_el(ch, QUIC_ENC_LEVEL_HANDSHAKE);
+    ch->handshake_confirmed = 1;
+    return 1;
+}
+
+/*
+ * Master function used when we want to start tearing down a connection:
+ *
+ *   - If the connection is still IDLE we can go straight to TERMINATED;
+ *
+ *   - If we are already TERMINATED this is a no-op.
+ *
+ *   - If we are TERMINATING - CLOSING and we have now got a CONNECTION_CLOSE
+ *     from the peer (tcause->remote == 1), we move to TERMINATING - CLOSING.
+ *
+ *   - If we are TERMINATING - DRAINING, we remain here until the terminating
+ *     timer expires.
+ *
+ *   - Otherwise, we are in ACTIVE and move to TERMINATING - CLOSING.
+ *     if we caused the termination (e.g. we have sent a CONNECTION_CLOSE). Note
+ *     that we are considered to have caused a termination if we sent the first
+ *     CONNECTION_CLOSE frame, even if it is caused by a peer protocol
+ *     violation. If the peer sent the first CONNECTION_CLOSE frame, we move to
+ *     TERMINATING - DRAINING.
+ *
+ * We record the termination cause structure passed on the first call only.
+ * Any successive calls have their termination cause data discarded;
+ * once we start sending a CONNECTION_CLOSE frame, we don't change the details
+ * in it.
+ */
+static void ch_start_terminating(QUIC_CHANNEL *ch,
+                                 const QUIC_TERMINATE_CAUSE *tcause)
+{
+    switch (ch->state) {
+        default:
+        case QUIC_CHANNEL_STATE_IDLE:
+            ch->terminate_cause = *tcause;
+            ch_on_terminating_timeout(ch);
+            break;
+
+        case QUIC_CHANNEL_STATE_ACTIVE:
+            ch->state = tcause->remote ? QUIC_CHANNEL_STATE_TERMINATING_DRAINING
+                                       : QUIC_CHANNEL_STATE_TERMINATING_CLOSING;
+            ch->terminate_cause = *tcause;
+            ch->terminate_deadline
+                = ossl_time_add(ossl_time_now(),
+                                ossl_time_multiply(ossl_ackm_get_pto_duration(ch->ackm),
+                                                   3));
+
+            if (!tcause->remote) {
+                OSSL_QUIC_FRAME_CONN_CLOSE f = {0};
+
+                /* best effort */
+                f.error_code = ch->terminate_cause.error_code;
+                f.frame_type = ch->terminate_cause.frame_type;
+                f.is_app     = ch->terminate_cause.app;
+                ossl_quic_tx_packetiser_schedule_conn_close(ch->txp, &f);
+                ch->conn_close_queued = 1;
+            }
+            break;
+
+        case QUIC_CHANNEL_STATE_TERMINATING_CLOSING:
+            if (tcause->remote)
+                ch->state = QUIC_CHANNEL_STATE_TERMINATING_DRAINING;
+
+            break;
+
+        case QUIC_CHANNEL_STATE_TERMINATING_DRAINING:
+            /* We remain here until the timout expires. */
+            break;
+
+        case QUIC_CHANNEL_STATE_TERMINATED:
+            /* No-op. */
+            break;
+    }
+}
+
+/* For RXDP use. */
+void ossl_quic_channel_on_remote_conn_close(QUIC_CHANNEL *ch,
+                                            OSSL_QUIC_FRAME_CONN_CLOSE *f)
+{
+    QUIC_TERMINATE_CAUSE tcause = {0};
+
+    if (!ossl_quic_channel_is_active(ch))
+        return;
+
+    tcause.remote     = 1;
+    tcause.app        = f->is_app;
+    tcause.error_code = f->error_code;
+    tcause.frame_type = f->frame_type;
+
+    ch_start_terminating(ch, &tcause);
+}
+
+void ossl_quic_channel_raise_protocol_error(QUIC_CHANNEL *ch,
+                                            uint64_t error_code,
+                                            uint64_t frame_type,
+                                            const char *reason)
+{
+    QUIC_TERMINATE_CAUSE tcause = {0};
+
+    tcause.error_code = error_code;
+    tcause.frame_type = frame_type;
+
+    ch_start_terminating(ch, &tcause);
+}
+
+/*
+ * Called once the terminating timer expires, meaning we move from TERMINATING
+ * to TERMINATED.
+ */
+static void ch_on_terminating_timeout(QUIC_CHANNEL *ch)
+{
+    ch->state = QUIC_CHANNEL_STATE_TERMINATED;
+}
+
+/*
+ * Updates our idle deadline. Called when an event happens which should bump the
+ * idle timeout.
+ */
+static void ch_update_idle(QUIC_CHANNEL *ch)
+{
+    if (ch->max_idle_timeout == 0)
+        ch->idle_deadline = ossl_time_infinite();
+    else
+        ch->idle_deadline = ossl_time_add(ossl_time_now(),
+            ossl_ms2time(ch->max_idle_timeout));
+}
+
+/* Called when the idle timeout expires. */
+static void ch_on_idle_timeout(QUIC_CHANNEL *ch)
+{
+    /*
+     * Idle timeout does not have an error code associated with it because a
+     * CONN_CLOSE is never sent for it. We shouldn't use this data once we reach
+     * TERMINATED anyway.
+     */
+    ch->terminate_cause.app         = 0;
+    ch->terminate_cause.error_code  = UINT64_MAX;
+    ch->terminate_cause.frame_type  = 0;
+
+    ch->state = QUIC_CHANNEL_STATE_TERMINATED;
+}
diff --git a/ssl/quic/quic_channel_local.h b/ssl/quic/quic_channel_local.h
new file mode 100644 (file)
index 0000000..5859c03
--- /dev/null
@@ -0,0 +1,270 @@
+#ifndef OSSL_QUIC_CHANNEL_LOCAL_H
+# define OSSL_QUIC_CHANNEL_LOCAL_H
+
+# include "internal/quic_channel.h"
+
+# ifndef OPENSSL_NO_QUIC
+
+/* Represents the cause for a connection's termination. */
+typedef struct quic_terminate_cause_st {
+    /*
+     * If we are in a TERMINATING or TERMINATED state, this is the error code
+     * associated with the error. This field is valid iff we are in the
+     * TERMINATING or TERMINATED states.
+     */
+    uint64_t                        error_code;
+
+    /*
+     * If terminate_app is set and this is nonzero, this is the frame type which
+     * caused the connection to be terminated.
+     */
+    uint64_t                        frame_type;
+
+    /* Is this error code in the transport (0) or application (1) space? */
+    unsigned int                    app : 1;
+
+    /*
+     * If set, the cause of the termination is a received CONNECTION_CLOSE
+     * frame. Otherwise, we decided to terminate ourselves and sent a
+     * CONNECTION_CLOSE frame (regardless of whether the peer later also sends
+     * one).
+     */
+    unsigned int                    remote : 1;
+} QUIC_TERMINATE_CAUSE;
+
+/*
+ * QUIC Channel Structure
+ * ======================
+ *
+ * QUIC channel internals. It is intended that only the QUIC_CHANNEL
+ * implementation and the RX depacketiser be allowed to access this structure
+ * directly. As the RX depacketiser has no state of its own and computes over a
+ * QUIC_CHANNEL structure, it can be viewed as an extention of the QUIC_CHANNEL
+ * implementation. While the RX depacketiser could be provided with adequate
+ * accessors to do what it needs, this would weaken the abstraction provided by
+ * the QUIC_CHANNEL to other components; moreover the coupling of the RX
+ * depacketiser to QUIC_CHANNEL internals is too deep and bespoke to make this
+ * desirable.
+ *
+ * Other components should not include this header.
+ */
+struct quic_channel_st {
+    OSSL_LIB_CTX                    *libctx;
+    const char                      *propq;
+
+    /*
+     * The associated TLS 1.3 connection data. Used to provide the handshake
+     * layer; its 'network' side is plugged into the crypto stream for each EL
+     * (other than the 0-RTT EL).
+     */
+    QUIC_DHS                        *dhs;
+    /* TODO(QUIC): Replace this with a QUIC_TLS instance when ready. */
+
+    /*
+     * The transport parameter block we will send or have sent.
+     * Freed after sending or when connection is freed.
+     */
+    unsigned char                   *local_transport_params;
+
+    /* Asynchronous I/O reactor. */
+    QUIC_REACTOR                    rtor;
+
+    /* Our current L4 peer address, if any. */
+    BIO_ADDR                        cur_peer_addr;
+
+    /* Network-side read and write BIOs. */
+    BIO                             *net_rbio, *net_wbio;
+
+    /*
+     * Subcomponents of the connection. All of these components are instantiated
+     * and owned by us.
+     */
+    OSSL_QUIC_TX_PACKETISER         *txp;
+    QUIC_TXPIM                      *txpim;
+    QUIC_CFQ                        *cfq;
+    /* Connection level FC. */
+    QUIC_TXFC                       conn_txfc;
+    QUIC_RXFC                       conn_rxfc;
+    QUIC_STREAM_MAP                 qsm;
+    OSSL_STATM                      statm;
+    OSSL_CC_DATA                    *cc_data;
+    const OSSL_CC_METHOD            *cc_method;
+    OSSL_ACKM                       *ackm;
+
+    /*
+     * RX demuxer. We register incoming DCIDs with this. Since we currently only
+     * support client operation and use one L4 port per connection, we own the
+     * demuxer and register a single zero-length DCID with it.
+     */
+    QUIC_DEMUX                      *demux;
+
+    /* Record layers in the TX and RX directions, plus the RX demuxer. */
+    OSSL_QTX                        *qtx;
+    OSSL_QRX                        *qrx;
+
+    /*
+     * Send and receive parts of the crypto streams.
+     * crypto_send[QUIC_PN_SPACE_APP] is the 1-RTT crypto stream. There is no
+     * 0-RTT crypto stream.
+     */
+    QUIC_SSTREAM                    *crypto_send[QUIC_PN_SPACE_NUM];
+    QUIC_RSTREAM                    *crypto_recv[QUIC_PN_SPACE_NUM];
+
+    /*
+     * Our (currently only) application data stream. This is a bidirectional
+     * client-initiated stream and thus (in QUICv1) always has a stream ID of 0.
+     */
+    QUIC_STREAM                     *stream0;
+
+    /* Internal state. */
+    /*
+     * The DCID used in the first Initial packet we transmit as a client.
+     * Randomly generated and required by RFC to be at least 8 bytes.
+     */
+    QUIC_CONN_ID                    init_dcid;
+
+    /*
+     * The SCID found in the first Initial packet from the server.
+     * Valid if have_received_enc_pkt is set.
+     */
+    QUIC_CONN_ID                    init_scid;
+
+    /* The SCID found in an incoming Retry packet we handled. */
+    QUIC_CONN_ID                    retry_scid;
+
+    /* Transport parameter values received from server. */
+    uint64_t                        init_max_stream_data_bidi_local;
+    uint64_t                        init_max_stream_data_bidi_remote;
+    uint64_t                        init_max_stream_data_uni_remote;
+    uint64_t                        rx_max_ack_delay; /* ms */
+    unsigned char                   rx_ack_delay_exp;
+
+    /*
+     * Temporary staging area to store information about the incoming packet we
+     * are currently processing.
+     */
+    OSSL_QRX_PKT                    *qrx_pkt;
+
+    /*
+     * Current limit on number of streams we may create. Set by transport
+     * parameters initially and then by MAX_STREAMS frames.
+     */
+    uint64_t                        max_local_streams_bidi;
+    uint64_t                        max_local_streams_uni;
+
+    /* The negotiated maximum idle timeout in milliseconds. */
+    uint64_t                        max_idle_timeout;
+
+    /*
+     * Maximum payload size in bytes for datagrams sent to our peer, as
+     * negotiated by transport parameters.
+     */
+    uint64_t                        rx_max_udp_payload_size;
+    /* Maximum active CID limit, as negotiated by transport parameters. */
+    uint64_t                        rx_active_conn_id_limit;
+
+    /* Valid if we are in the TERMINATING or TERMINATED states. */
+    QUIC_TERMINATE_CAUSE            terminate_cause;
+
+    /*
+     * Deadline at which we move to TERMINATING state. Valid if in the
+     * TERMINATING state.
+     */
+    OSSL_TIME                       terminate_deadline;
+
+    /*
+     * Deadline at which connection dies due to idle timeout if no further
+     * events occur.
+     */
+    OSSL_TIME                       idle_deadline;
+
+    /*
+     * State tracking. QUIC connection-level state is best represented based on
+     * whether various things have happened yet or not, rather than as an
+     * explicit FSM. We do have a coarse state variable which tracks the basic
+     * state of the connection's lifecycle, but more fine-grained conditions of
+     * the Active state are tracked via flags below. For more details, see
+     * doc/designs/quic-design/connection-state-machine.md. We are in the Open
+     * state if the state is QUIC_CSM_STATE_ACTIVE and handshake_confirmed is
+     * set.
+     */
+    unsigned int                    state                   : 3;
+
+    /*
+     * Have we received at least one encrypted packet from the peer?
+     * (If so, Retry and Version Negotiation messages should no longer
+     *  be received and should be ignored if they do occur.)
+     */
+    unsigned int                    have_received_enc_pkt   : 1;
+
+    /*
+     * Have we sent literally any packet yet? If not, there is no point polling
+     * RX.
+     */
+    unsigned int                    have_sent_any_pkt       : 1;
+
+    /*
+     * Are we currently doing proactive version negotiation?
+     */
+    unsigned int                    doing_proactive_ver_neg : 1;
+
+    /* We have received transport parameters from the peer. */
+    unsigned int                    got_remote_transport_params    : 1;
+
+    /*
+     * This monotonically transitions to 1 once the TLS state machine is
+     * 'complete', meaning that it has both sent a Finished and successfully
+     * verified the peer's Finished (see RFC 9001 s. 4.1.1). Note that it
+     * does not transition to 1 at both peers simultaneously.
+     *
+     * Handshake completion is not the same as handshake confirmation (see
+     * below).
+     */
+    unsigned int                    handshake_complete      : 1;
+
+    /*
+     * This monotonically transitions to 1 once the handshake is confirmed.
+     * This happens on the client when we receive a HANDSHAKE_DONE frame.
+     * At our option, we may also take acknowledgement of any 1-RTT packet
+     * we sent as a handshake confirmation.
+     */
+    unsigned int                    handshake_confirmed     : 1;
+
+    /*
+     * We are sending Initial packets based on a Retry. This means we definitely
+     * should not receive another Retry, and if we do it is an error.
+     */
+    unsigned int                    doing_retry             : 1;
+
+    /*
+     * We don't store the current EL here; the TXP asks the QTX which ELs
+     * are provisioned to determine which ELs to use.
+     */
+
+    /* Have statm, qsm been initialised? Used to track cleanup. */
+    unsigned int                    have_statm              : 1;
+    unsigned int                    have_qsm                : 1;
+
+    /*
+     * Preferred EL for transmission. This is not strictly needed as it can be
+     * inferred from what keys we have provisioned, but makes determining the
+     * current EL simpler and faster.
+     */
+    unsigned int                    tx_enc_level            : 3;
+
+    /* If bit n is set, EL n has been discarded. */
+    unsigned int                    el_discarded            : 4;
+
+    /*
+     * While in TERMINATING - CLOSING, set when we should generate a connection
+     * close frame.
+     */
+    unsigned int                    conn_close_queued       : 1;
+
+    /* Are we in server mode? Never changes after instantiation. */
+    unsigned int                    is_server               : 1;
+};
+
+# endif
+
+#endif