source: mainline/uspace/lib/c/generic/async/server.c@ 5c38838

/*
 * Copyright (c) 2006 Ondrej Palkovsky
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * - Redistributions of source code must retain the above copyright
 *   notice, this list of conditions and the following disclaimer.
 * - Redistributions in binary form must reproduce the above copyright
 *   notice, this list of conditions and the following disclaimer in the
 *   documentation and/or other materials provided with the distribution.
 * - The name of the author may not be used to endorse or promote products
 *   derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

/** @addtogroup libc
 * @{
 */
/** @file
 */

/**
 * Asynchronous library
 *
 * The aim of this library is to provide a facility for writing programs which
 * utilize the asynchronous nature of HelenOS IPC, yet using a normal way of
 * programming.
 *
 * You should be able to write very simple multithreaded programs. The async
 * framework will automatically take care of most of the synchronization
 * problems.
 *
 * Example of use (pseudo C):
 *
 * 1) Multithreaded client application
 *
 *   fibril_create(fibril1, ...);
 *   fibril_create(fibril2, ...);
 *   ...
 *
 *   int fibril1(void *arg)
 *   {
 *     conn = async_connect_me_to(...);
 *
 *     exch = async_exchange_begin(conn);
 *     c1 = async_send(exch);
 *     async_exchange_end(exch);
 *
 *     exch = async_exchange_begin(conn);
 *     c2 = async_send(exch);
 *     async_exchange_end(exch);
 *
 *     async_wait_for(c1);
 *     async_wait_for(c2);
 *     ...
 *   }
 *
 *
 * 2) Multithreaded server application
 *
 *   main()
 *   {
 *     async_manager();
 *   }
 *
 *   port_handler(ipc_call_t *icall)
 *   {
 *     if (want_refuse) {
 *       async_answer_0(icall, ELIMIT);
 *       return;
 *     }
 *
 *     async_answer_0(icall, EOK);
 *
 *     async_get_call(&call);
 *     somehow_handle_the_call(&call);
 *     async_answer_2(&call, 1, 2, 3);
 *
 *     async_get_call(&call);
 *     ...
 *   }
 *
 */

#define LIBC_ASYNC_C_
#include <ipc/ipc.h>
#include <async.h>
#include "../private/async.h"
#undef LIBC_ASYNC_C_

#include <ipc/irq.h>
#include <ipc/event.h>
#include <fibril.h>
#include <adt/hash_table.h>
#include <adt/hash.h>
#include <adt/list.h>
#include <assert.h>
#include <errno.h>
#include <time.h>
#include <stdbool.h>
#include <stdlib.h>
#include <mem.h>
#include <stdlib.h>
#include <macros.h>
#include <str_error.h>
#include <as.h>
#include <abi/mm/as.h>
#include "../private/libc.h"
#include "../private/fibril.h"

#define DPRINTF(...)  ((void) 0)

/* Client connection data */
typedef struct {
        ht_link_t link;

        task_id_t in_task_id;
        int refcnt;
        void *data;
} client_t;

/* Server connection data */
typedef struct {
        /** Fibril handling the connection. */
        fid_t fid;

        /** Hash table link. */
        ht_link_t link;

        /** Incoming client task ID. */
        task_id_t in_task_id;

        /** Incoming phone hash. */
        sysarg_t in_phone_hash;

        /** Link to the client tracking structure. */
        client_t *client;

        /** Channel for messages that should be delivered to this fibril. */
        mpsc_t *msg_channel;

        /** Call data of the opening call. */
        ipc_call_t call;

        /** Fibril function that will be used to handle the connection. */
        async_port_handler_t handler;

        /** Client data */
        void *data;
} connection_t;

/* Member of notification_t::msg_list. */
typedef struct {
        link_t link;
        ipc_call_t calldata;
} notification_msg_t;

/* Notification data */
typedef struct {
        /** notification_hash_table link */
        ht_link_t htlink;

        /** notification_queue link */
        link_t qlink;

        /** Notification method */
        sysarg_t imethod;

        /** Notification handler */
        async_notification_handler_t handler;

        /** Notification handler argument */
        void *arg;

        /** List of arrived notifications. */
        list_t msg_list;
} notification_t;

/** Identifier of the incoming connection handled by the current fibril. */
static fibril_local connection_t *fibril_connection;

static void *default_client_data_constructor(void)
{
        return NULL;
}

static void default_client_data_destructor(void *data)
{
}

static async_client_data_ctor_t async_client_data_create =
    default_client_data_constructor;
static async_client_data_dtor_t async_client_data_destroy =
    default_client_data_destructor;

void async_set_client_data_constructor(async_client_data_ctor_t ctor)
{
        assert(async_client_data_create == default_client_data_constructor);
        async_client_data_create = ctor;
}

void async_set_client_data_destructor(async_client_data_dtor_t dtor)
{
        assert(async_client_data_destroy == default_client_data_destructor);
        async_client_data_destroy = dtor;
}

static FIBRIL_RMUTEX_INITIALIZE(client_mutex);
static hash_table_t client_hash_table;

// TODO: lockfree notification_queue?
static FIBRIL_RMUTEX_INITIALIZE(notification_mutex);
static hash_table_t notification_hash_table;
static LIST_INITIALIZE(notification_queue);
static FIBRIL_SEMAPHORE_INITIALIZE(notification_semaphore, 0);

static LIST_INITIALIZE(notification_freelist);
static long notification_freelist_total = 0;
static long notification_freelist_used = 0;

static sysarg_t notification_avail = 0;

static FIBRIL_RMUTEX_INITIALIZE(conn_mutex);
static hash_table_t conn_hash_table;

static size_t client_key_hash(void *key)
{
        task_id_t in_task_id = *(task_id_t *) key;
        return in_task_id;
}

static size_t client_hash(const ht_link_t *item)
{
        client_t *client = hash_table_get_inst(item, client_t, link);
        return client_key_hash(&client->in_task_id);
}

static bool client_key_equal(void *key, const ht_link_t *item)
{
        task_id_t in_task_id = *(task_id_t *) key;
        client_t *client = hash_table_get_inst(item, client_t, link);
        return in_task_id == client->in_task_id;
}

/** Operations for the client hash table. */
static hash_table_ops_t client_hash_table_ops = {
        .hash = client_hash,
        .key_hash = client_key_hash,
        .key_equal = client_key_equal,
        .equal = NULL,
        .remove_callback = NULL
};

typedef struct {
        task_id_t task_id;
        sysarg_t phone_hash;
} conn_key_t;

/** Compute hash into the connection hash table
 *
 * The hash is based on the source task ID and the source phone hash. The task
 * ID is included in the hash because a phone hash alone might not be unique
 * while we still track connections for killed tasks due to kernel's recycling
 * of phone structures.
 *
 * @param key Pointer to the connection key structure.
 *
 * @return Index into the connection hash table.
 *
 */
static size_t conn_key_hash(void *key)
{
        conn_key_t *ck = (conn_key_t *) key;

        size_t hash = 0;
        hash = hash_combine(hash, LOWER32(ck->task_id));
        hash = hash_combine(hash, UPPER32(ck->task_id));
        hash = hash_combine(hash, ck->phone_hash);
        return hash;
}

static size_t conn_hash(const ht_link_t *item)
{
        connection_t *conn = hash_table_get_inst(item, connection_t, link);
        return conn_key_hash(&(conn_key_t){
                .task_id = conn->in_task_id,
                .phone_hash = conn->in_phone_hash
        });
}

static bool conn_key_equal(void *key, const ht_link_t *item)
{
        conn_key_t *ck = (conn_key_t *) key;
        connection_t *conn = hash_table_get_inst(item, connection_t, link);
        return ((ck->task_id == conn->in_task_id) &&
            (ck->phone_hash == conn->in_phone_hash));
}

/** Operations for the connection hash table. */
static hash_table_ops_t conn_hash_table_ops = {
        .hash = conn_hash,
        .key_hash = conn_key_hash,
        .key_equal = conn_key_equal,
        .equal = NULL,
        .remove_callback = NULL
};

static client_t *async_client_get(task_id_t client_id, bool create)
{
        client_t *client = NULL;

        fibril_rmutex_lock(&client_mutex);
        ht_link_t *link = hash_table_find(&client_hash_table, &client_id);
        if (link) {
                client = hash_table_get_inst(link, client_t, link);
                client->refcnt++;
        } else if (create) {
                // TODO: move the malloc out of critical section
                /* malloc() is rmutex safe. */
                client = malloc(sizeof(client_t));
                if (client) {
                        client->in_task_id = client_id;
                        client->data = async_client_data_create();

                        client->refcnt = 1;
                        hash_table_insert(&client_hash_table, &client->link);
                }
        }

        fibril_rmutex_unlock(&client_mutex);
        return client;
}

static void async_client_put(client_t *client)
{
        bool destroy;

        fibril_rmutex_lock(&client_mutex);

        if (--client->refcnt == 0) {
                hash_table_remove(&client_hash_table, &client->in_task_id);
                destroy = true;
        } else
                destroy = false;

        fibril_rmutex_unlock(&client_mutex);

        if (destroy) {
                if (client->data)
                        async_client_data_destroy(client->data);

                free(client);
        }
}

/** Wrapper for client connection fibril.
 *
 * When a new connection arrives, a fibril with this implementing
 * function is created.
 *
 * @param arg Connection structure pointer.
 *
 * @return Always zero.
 *
 */
static errno_t connection_fibril(void *arg)
{
        assert(arg);

        /*
         * Setup fibril-local connection pointer.
         */
        fibril_connection = (connection_t *) arg;

        /*
         * Add our reference for the current connection in the client task
         * tracking structure. If this is the first reference, create and
         * hash in a new tracking structure.
         */

        client_t *client = async_client_get(fibril_connection->in_task_id, true);
        if (!client) {
                ipc_answer_0(fibril_connection->call.cap_handle, ENOMEM);
                return 0;
        }

        fibril_connection->client = client;

        /*
         * Call the connection handler function.
         */
        fibril_connection->handler(&fibril_connection->call,
            fibril_connection->data);

        /*
         * Remove the reference for this client task connection.
         */
        async_client_put(client);

        fibril_rmutex_lock(&conn_mutex);

        /*
         * Remove myself from the connection hash table.
         */
        hash_table_remove(&conn_hash_table, &(conn_key_t){
                .task_id = fibril_connection->in_task_id,
                .phone_hash = fibril_connection->in_phone_hash
        });

        /*
         * Close the channel, if it isn't closed already.
         */
        mpsc_t *c = fibril_connection->msg_channel;
        mpsc_close(c);

        fibril_rmutex_unlock(&conn_mutex);

        /*
         * Answer all remaining messages with EHANGUP.
         */
        ipc_call_t call;
        while (mpsc_receive(c, &call, NULL) == EOK)
                ipc_answer_0(call.cap_handle, EHANGUP);

        /*
         * Clean up memory.
         */
        mpsc_destroy(c);
        free(fibril_connection);
        return EOK;
}

/** Create a new fibril for a new connection.
 *
 * Create new fibril for connection, fill in connection structures and insert it
 * into the hash table, so that later we can easily do routing of messages to
 * particular fibrils.
 *
 * @param in_task_id     Identification of the incoming connection.
 * @param in_phone_hash  Identification of the incoming connection.
 * @param call           Call data of the opening call. If call is NULL,
 *                       the connection was opened by accepting the
 *                       IPC_M_CONNECT_TO_ME call and this function is
 *                       called directly by the server.
 * @param handler        Connection handler.
 * @param data           Client argument to pass to the connection handler.
 *
 * @return  New fibril id or NULL on failure.
 *
 */
static fid_t async_new_connection(task_id_t in_task_id, sysarg_t in_phone_hash,
    ipc_call_t *call, async_port_handler_t handler, void *data)
{
        connection_t *conn = malloc(sizeof(*conn));
        if (!conn) {
                if (call)
                        ipc_answer_0(call->cap_handle, ENOMEM);

                return (fid_t) NULL;
        }

        conn->in_task_id = in_task_id;
        conn->in_phone_hash = in_phone_hash;
        conn->msg_channel = mpsc_create(sizeof(ipc_call_t));
        conn->handler = handler;
        conn->data = data;

        if (call)
                conn->call = *call;
        else
                conn->call.cap_handle = CAP_NIL;

        /* We will activate the fibril ASAP */
        conn->fid = fibril_create(connection_fibril, conn);

        if (conn->fid == 0) {
                mpsc_destroy(conn->msg_channel);
                free(conn);

                if (call)
                        ipc_answer_0(call->cap_handle, ENOMEM);

                return (fid_t) NULL;
        }

        /* Add connection to the connection hash table */

        fibril_rmutex_lock(&conn_mutex);
        hash_table_insert(&conn_hash_table, &conn->link);
        fibril_rmutex_unlock(&conn_mutex);

        fibril_start(conn->fid);

        return conn->fid;
}

/** Wrapper for making IPC_M_CONNECT_TO_ME calls using the async framework.
 *
 * Ask through phone for a new connection to some service.
 *
 * @param exch    Exchange for sending the message.
 * @param iface   Callback interface.
 * @param arg1    User defined argument.
 * @param arg2    User defined argument.
 * @param handler Callback handler.
 * @param data    Handler data.
 * @param port_id ID of the newly created port.
 *
 * @return Zero on success or an error code.
 *
 */
errno_t async_create_callback_port(async_exch_t *exch, iface_t iface, sysarg_t arg1,
    sysarg_t arg2, async_port_handler_t handler, void *data, port_id_t *port_id)
{
        if ((iface & IFACE_MOD_CALLBACK) != IFACE_MOD_CALLBACK)
                return EINVAL;

        if (exch == NULL)
                return ENOENT;

        ipc_call_t answer;
        aid_t req = async_send_3(exch, IPC_M_CONNECT_TO_ME, iface, arg1, arg2,
            &answer);

        errno_t rc;
        async_wait_for(req, &rc);
        if (rc != EOK)
                return rc;

        rc = async_create_port_internal(iface, handler, data, port_id);
        if (rc != EOK)
                return rc;

        sysarg_t phone_hash = IPC_GET_ARG5(answer);
        fid_t fid = async_new_connection(answer.in_task_id, phone_hash,
            NULL, handler, data);
        if (fid == (fid_t) NULL)
                return ENOMEM;

        return EOK;
}

static size_t notification_key_hash(void *key)
{
        sysarg_t id = *(sysarg_t *) key;
        return id;
}

static size_t notification_hash(const ht_link_t *item)
{
        notification_t *notification =
            hash_table_get_inst(item, notification_t, htlink);
        return notification_key_hash(&notification->imethod);
}

static bool notification_key_equal(void *key, const ht_link_t *item)
{
        sysarg_t id = *(sysarg_t *) key;
        notification_t *notification =
            hash_table_get_inst(item, notification_t, htlink);
        return id == notification->imethod;
}

/** Operations for the notification hash table. */
static hash_table_ops_t notification_hash_table_ops = {
        .hash = notification_hash,
        .key_hash = notification_key_hash,
        .key_equal = notification_key_equal,
        .equal = NULL,
        .remove_callback = NULL
};

/** Try to route a call to an appropriate connection fibril.
 *
 * If the proper connection fibril is found, a message with the call is added to
 * its message queue. If the fibril was not active, it is activated and all
 * timeouts are unregistered.
 *
 * @param call Data of the incoming call.
 *
 * @return EOK if the call was successfully passed to the respective fibril.
 * @return ENOENT if the call doesn't match any connection.
 * @return Other error code if routing failed for other reasons.
 *
 */
static errno_t route_call(ipc_call_t *call)
{
        assert(call);

        fibril_rmutex_lock(&conn_mutex);

        ht_link_t *link = hash_table_find(&conn_hash_table, &(conn_key_t){
                .task_id = call->in_task_id,
                .phone_hash = call->in_phone_hash
        });
        if (!link) {
                fibril_rmutex_unlock(&conn_mutex);
                return ENOENT;
        }

        connection_t *conn = hash_table_get_inst(link, connection_t, link);

        errno_t rc = mpsc_send(conn->msg_channel, call);

        if (IPC_GET_IMETHOD(*call) == IPC_M_PHONE_HUNGUP) {
                /* Close the channel, but let the connection fibril answer. */
                mpsc_close(conn->msg_channel);
                // FIXME: Ideally, we should be able to discard/answer the
                //        hungup message here and just close the channel without
                //        passing it out. Unfortunatelly, somehow that breaks
                //        handling of CPU exceptions.
                async_answer_0(call, EOK);
        }

        fibril_rmutex_unlock(&conn_mutex);
        return rc;
}

/** Function implementing the notification handler fibril. Never returns. */
static errno_t notification_fibril_func(void *arg)
{
        (void) arg;

        while (true) {
                fibril_semaphore_down(&notification_semaphore);

                fibril_rmutex_lock(&notification_mutex);

                /*
                 * The semaphore ensures that if we get this far,
                 * the queue must be non-empty.
                 */
                assert(!list_empty(&notification_queue));

                notification_t *notification = list_get_instance(
                    list_first(&notification_queue), notification_t, qlink);

                async_notification_handler_t handler = notification->handler;
                void *arg = notification->arg;

                notification_msg_t *m = list_pop(&notification->msg_list,
                    notification_msg_t, link);
                assert(m);
                ipc_call_t calldata = m->calldata;

                notification_freelist_used--;

                if (notification_freelist_total > 64 &&
                    notification_freelist_total > 2 * notification_freelist_used) {
                        /* Going to free the structure if we have too much. */
                        notification_freelist_total--;
                } else {
                        /* Otherwise add to freelist. */
                        list_append(&m->link, &notification_freelist);
                        m = NULL;
                }

                if (list_empty(&notification->msg_list))
                        list_remove(&notification->qlink);

                fibril_rmutex_unlock(&notification_mutex);

                if (handler)
                        handler(&calldata, arg);

                free(m);
        }

        /* Not reached. */
        return EOK;
}

/**
 * Creates a new dedicated fibril for handling notifications.
 * By default, there is one such fibril. This function can be used to
 * create more in order to increase the number of notification that can
 * be processed concurrently.
 *
 * Currently, there is no way to destroy those fibrils after they are created.
 */
errno_t async_spawn_notification_handler(void)
{
        fid_t f = fibril_create(notification_fibril_func, NULL);
        if (f == 0)
                return ENOMEM;

        fibril_add_ready(f);
        return EOK;
}

/** Queue notification.
 *
 * @param call   Data of the incoming call.
 *
 */
static void queue_notification(ipc_call_t *call)
{
        assert(call);

        fibril_rmutex_lock(&notification_mutex);

        notification_msg_t *m = list_pop(&notification_freelist,
            notification_msg_t, link);

        if (!m) {
                fibril_rmutex_unlock(&notification_mutex);
                m = malloc(sizeof(notification_msg_t));
                if (!m) {
                        DPRINTF("Out of memory.\n");
                        abort();
                }

                fibril_rmutex_lock(&notification_mutex);
                notification_freelist_total++;
        }

        ht_link_t *link = hash_table_find(&notification_hash_table,
            &IPC_GET_IMETHOD(*call));
        if (!link) {
                /* Invalid notification. */
                // TODO: Make sure this can't happen and turn it into assert.
                notification_freelist_total--;
                fibril_rmutex_unlock(&notification_mutex);
                free(m);
                return;
        }

        notification_t *notification =
            hash_table_get_inst(link, notification_t, htlink);

        notification_freelist_used++;
        m->calldata = *call;
        list_append(&m->link, &notification->msg_list);

        if (!link_in_use(&notification->qlink))
                list_append(&notification->qlink, &notification_queue);

        fibril_rmutex_unlock(&notification_mutex);

        fibril_semaphore_up(&notification_semaphore);
}

/**
 * Creates a new notification structure and inserts it into the hash table.
 *
 * @param handler  Function to call when notification is received.
 * @param arg      Argument for the handler function.
 * @return         The newly created notification structure.
 */
static notification_t *notification_create(async_notification_handler_t handler, void *arg)
{
        notification_t *notification = calloc(1, sizeof(notification_t));
        if (!notification)
                return NULL;

        notification->handler = handler;
        notification->arg = arg;

        list_initialize(&notification->msg_list);

        fid_t fib = 0;

        fibril_rmutex_lock(&notification_mutex);

        if (notification_avail == 0) {
                /* Attempt to create the first handler fibril. */
                fib = fibril_create(notification_fibril_func, NULL);
                if (fib == 0) {
                        fibril_rmutex_unlock(&notification_mutex);
                        free(notification);
                        return NULL;
                }
        }

        sysarg_t imethod = notification_avail;
        notification_avail++;

        notification->imethod = imethod;
        hash_table_insert(&notification_hash_table, &notification->htlink);

        fibril_rmutex_unlock(&notification_mutex);

        if (imethod == 0) {
                assert(fib);
                fibril_add_ready(fib);
        }

        return notification;
}

/** Subscribe to IRQ notification.
 *
 * @param inr     IRQ number.
 * @param handler Notification handler.
 * @param data    Notification handler client data.
 * @param ucode   Top-half pseudocode handler.
 *
 * @param[out] handle  IRQ capability handle on success.
 *
 * @return An error code.
 *
 */
errno_t async_irq_subscribe(int inr, async_notification_handler_t handler,
    void *data, const irq_code_t *ucode, cap_irq_handle_t *handle)
{
        notification_t *notification = notification_create(handler, data);
        if (!notification)
                return ENOMEM;

        cap_irq_handle_t ihandle;
        errno_t rc = ipc_irq_subscribe(inr, notification->imethod, ucode,
            &ihandle);
        if (rc == EOK && handle != NULL) {
                *handle = ihandle;
        }
        return rc;
}

/** Unsubscribe from IRQ notification.
 *
 * @param handle  IRQ capability handle.
 *
 * @return Zero on success or an error code.
 *
 */
errno_t async_irq_unsubscribe(cap_irq_handle_t ihandle)
{
        // TODO: Remove entry from hash table
        //       to avoid memory leak

        return ipc_irq_unsubscribe(ihandle);
}

/** Subscribe to event notifications.
 *
 * @param evno    Event type to subscribe.
 * @param handler Notification handler.
 * @param data    Notification handler client data.
 *
 * @return Zero on success or an error code.
 *
 */
errno_t async_event_subscribe(event_type_t evno,
    async_notification_handler_t handler, void *data)
{
        notification_t *notification = notification_create(handler, data);
        if (!notification)
                return ENOMEM;

        return ipc_event_subscribe(evno, notification->imethod);
}

/** Subscribe to task event notifications.
 *
 * @param evno    Event type to subscribe.
 * @param handler Notification handler.
 * @param data    Notification handler client data.
 *
 * @return Zero on success or an error code.
 *
 */
errno_t async_event_task_subscribe(event_task_type_t evno,
    async_notification_handler_t handler, void *data)
{
        notification_t *notification = notification_create(handler, data);
        if (!notification)
                return ENOMEM;

        return ipc_event_task_subscribe(evno, notification->imethod);
}

/** Unmask event notifications.
 *
 * @param evno Event type to unmask.
 *
 * @return Value returned by the kernel.
 *
 */
errno_t async_event_unmask(event_type_t evno)
{
        return ipc_event_unmask(evno);
}

/** Unmask task event notifications.
 *
 * @param evno Event type to unmask.
 *
 * @return Value returned by the kernel.
 *
 */
errno_t async_event_task_unmask(event_task_type_t evno)
{
        return ipc_event_task_unmask(evno);
}

/** Return new incoming message for the current (fibril-local) connection.
 *
 * @param call  Storage where the incoming call data will be stored.
 * @param usecs Timeout in microseconds. Zero denotes no timeout.
 *
 * @return If no timeout was specified, then true is returned.
 * @return If a timeout is specified, then true is returned unless
 *         the timeout expires prior to receiving a message.
 *
 */
bool async_get_call_timeout(ipc_call_t *call, usec_t usecs)
{
        assert(call);
        assert(fibril_connection);

        struct timespec ts;
        struct timespec *expires = NULL;
        if (usecs) {
                getuptime(&ts);
                ts_add_diff(&ts, USEC2NSEC(usecs));
                expires = &ts;
        }

        errno_t rc = mpsc_receive(fibril_connection->msg_channel,
            call, expires);

        if (rc == ETIMEOUT)
                return false;

        if (rc != EOK) {
                /*
                 * The async_get_call_timeout() interface doesn't support
                 * propagating errors. Return a null call instead.
                 */

                memset(call, 0, sizeof(ipc_call_t));
        }

        return true;
}

void *async_get_client_data(void)
{
        assert(fibril_connection);
        return fibril_connection->client->data;
}

void *async_get_client_data_by_id(task_id_t client_id)
{
        client_t *client = async_client_get(client_id, false);
        if (!client)
                return NULL;

        if (!client->data) {
                async_client_put(client);
                return NULL;
        }

        return client->data;
}

void async_put_client_data_by_id(task_id_t client_id)
{
        client_t *client = async_client_get(client_id, false);

        assert(client);
        assert(client->data);

        /* Drop the reference we got in async_get_client_data_by_hash(). */
        async_client_put(client);

        /* Drop our own reference we got at the beginning of this function. */
        async_client_put(client);
}

/** Handle a call that was received.
 *
 * If the call has the IPC_M_CONNECT_ME_TO method, a new connection is created.
 * Otherwise the call is routed to its connection fibril.
 *
 * @param call Data of the incoming call.
 *
 */
static void handle_call(ipc_call_t *call)
{
        assert(call);

        if (call->flags & IPC_CALL_ANSWERED) {
                /* Answer to a call made by us. */
                async_reply_received(call);
                return;
        }

        if (call->cap_handle == CAP_NIL) {
                if (call->flags & IPC_CALL_NOTIF) {
                        /* Kernel notification */
                        queue_notification(call);
                }
                return;
        }

        /* New connection */
        if (IPC_GET_IMETHOD(*call) == IPC_M_CONNECT_ME_TO) {
                iface_t iface = (iface_t) IPC_GET_ARG1(*call);
                sysarg_t in_phone_hash = IPC_GET_ARG5(*call);

                // TODO: Currently ignores all ports but the first one.
                void *data;
                async_port_handler_t handler =
                    async_get_port_handler(iface, 0, &data);

                async_new_connection(call->in_task_id, in_phone_hash, call,
                    handler, data);
                return;
        }

        /* Try to route the call through the connection hash table */
        errno_t rc = route_call(call);
        if (rc == EOK)
                return;

        // TODO: Log the error.

        if (call->cap_handle != CAP_NIL)
                /* Unknown call from unknown phone - hang it up */
                ipc_answer_0(call->cap_handle, EHANGUP);
}

/** Endless loop dispatching incoming calls and answers.
 *
 * @return Never returns.
 *
 */
static errno_t async_manager_worker(void)
{
        ipc_call_t call;
        errno_t rc;

        while (true) {
                rc = fibril_ipc_wait(&call, NULL);
                if (rc == EOK)
                        handle_call(&call);
        }

        return 0;
}

/** Function to start async_manager as a standalone fibril.
 *
 * When more kernel threads are used, one async manager should exist per thread.
 *
 * @param arg Unused.
 * @return Never returns.
 *
 */
static errno_t async_manager_fibril(void *arg)
{
        async_manager_worker();
        return 0;
}

/** Add one manager to manager list. */
fid_t async_create_manager(void)
{
        fid_t fid = fibril_create_generic(async_manager_fibril, NULL, PAGE_SIZE);
        fibril_start(fid);
        return fid;
}

/** Initialize the async framework.
 *
 */
void __async_server_init(void)
{
        if (!hash_table_create(&client_hash_table, 0, 0, &client_hash_table_ops))
                abort();

        if (!hash_table_create(&conn_hash_table, 0, 0, &conn_hash_table_ops))
                abort();

        if (!hash_table_create(&notification_hash_table, 0, 0,
            &notification_hash_table_ops))
                abort();

        async_create_manager();
}

errno_t async_answer_0(ipc_call_t *call, errno_t retval)
{
        return ipc_answer_0(call->cap_handle, retval);
}

errno_t async_answer_1(ipc_call_t *call, errno_t retval, sysarg_t arg1)
{
        return ipc_answer_1(call->cap_handle, retval, arg1);
}

errno_t async_answer_2(ipc_call_t *call, errno_t retval, sysarg_t arg1,
    sysarg_t arg2)
{
        return ipc_answer_2(call->cap_handle, retval, arg1, arg2);
}

errno_t async_answer_3(ipc_call_t *call, errno_t retval, sysarg_t arg1,
    sysarg_t arg2, sysarg_t arg3)
{
        return ipc_answer_3(call->cap_handle, retval, arg1, arg2, arg3);
}

errno_t async_answer_4(ipc_call_t *call, errno_t retval, sysarg_t arg1,
    sysarg_t arg2, sysarg_t arg3, sysarg_t arg4)
{
        return ipc_answer_4(call->cap_handle, retval, arg1, arg2, arg3, arg4);
}

errno_t async_answer_5(ipc_call_t *call, errno_t retval, sysarg_t arg1,
    sysarg_t arg2, sysarg_t arg3, sysarg_t arg4, sysarg_t arg5)
{
        return ipc_answer_5(call->cap_handle, retval, arg1, arg2, arg3, arg4,
            arg5);
}

errno_t async_forward_fast(ipc_call_t *call, async_exch_t *exch,
    sysarg_t imethod, sysarg_t arg1, sysarg_t arg2, unsigned int mode)
{
        assert(call);

        if (exch == NULL)
                return ENOENT;

        return ipc_forward_fast(call->cap_handle, exch->phone, imethod, arg1,
            arg2, mode);
}

errno_t async_forward_slow(ipc_call_t *call, async_exch_t *exch,
    sysarg_t imethod, sysarg_t arg1, sysarg_t arg2, sysarg_t arg3,
    sysarg_t arg4, sysarg_t arg5, unsigned int mode)
{
        assert(call);

        if (exch == NULL)
                return ENOENT;

        return ipc_forward_slow(call->cap_handle, exch->phone, imethod, arg1,
            arg2, arg3, arg4, arg5, mode);
}

/** Wrapper for making IPC_M_CONNECT_TO_ME calls using the async framework.
 *
 * Ask through phone for a new connection to some service.
 *
 * @param exch  Exchange for sending the message.
 * @param iface Callback interface.
 * @param arg2  User defined argument.
 * @param arg3  User defined argument.
 *
 * @return Zero on success or an error code.
 *
 */
errno_t async_connect_to_me(async_exch_t *exch, iface_t iface, sysarg_t arg2,
    sysarg_t arg3)
{
        if (exch == NULL)
                return ENOENT;

        ipc_call_t answer;
        aid_t req = async_send_3(exch, IPC_M_CONNECT_TO_ME, iface, arg2, arg3,
            &answer);

        errno_t rc;
        async_wait_for(req, &rc);
        if (rc != EOK)
                return (errno_t) rc;

        return EOK;
}

/** Wrapper for receiving the IPC_M_SHARE_IN calls using the async framework.
 *
 * This wrapper only makes it more comfortable to receive IPC_M_SHARE_IN
 * calls so that the user doesn't have to remember the meaning of each IPC
 * argument.
 *
 * So far, this wrapper is to be used from within a connection fibril.
 *
 * @param call Storage for the data of the IPC_M_SHARE_IN call.
 * @param size Destination address space area size.
 *
 * @return True on success, false on failure.
 *
 */
bool async_share_in_receive(ipc_call_t *call, size_t *size)
{
        assert(call);
        assert(size);

        async_get_call(call);

        if (IPC_GET_IMETHOD(*call) != IPC_M_SHARE_IN)
                return false;

        *size = (size_t) IPC_GET_ARG1(*call);
        return true;
}

/** Wrapper for answering the IPC_M_SHARE_IN calls using the async framework.
 *
 * This wrapper only makes it more comfortable to answer IPC_M_SHARE_IN
 * calls so that the user doesn't have to remember the meaning of each IPC
 * argument.
 *
 * @param call  IPC_M_SHARE_IN call to answer.
 * @param src   Source address space base.
 * @param flags Flags to be used for sharing. Bits can be only cleared.
 *
 * @return Zero on success or a value from @ref errno.h on failure.
 *
 */
errno_t async_share_in_finalize(ipc_call_t *call, void *src, unsigned int flags)
{
        assert(call);

        return ipc_answer_2(call->cap_handle, EOK, (sysarg_t) src, (sysarg_t) flags);
}

/** Wrapper for receiving the IPC_M_SHARE_OUT calls using the async framework.
 *
 * This wrapper only makes it more comfortable to receive IPC_M_SHARE_OUT
 * calls so that the user doesn't have to remember the meaning of each IPC
 * argument.
 *
 * So far, this wrapper is to be used from within a connection fibril.
 *
 * @param call  Storage for the data of the IPC_M_SHARE_OUT call.
 * @param size  Storage for the source address space area size.
 * @param flags Storage for the sharing flags.
 *
 * @return True on success, false on failure.
 *
 */
bool async_share_out_receive(ipc_call_t *call, size_t *size,
    unsigned int *flags)
{
        assert(call);
        assert(size);
        assert(flags);

        async_get_call(call);

        if (IPC_GET_IMETHOD(*call) != IPC_M_SHARE_OUT)
                return false;

        *size = (size_t) IPC_GET_ARG2(*call);
        *flags = (unsigned int) IPC_GET_ARG3(*call);
        return true;
}

/** Wrapper for answering the IPC_M_SHARE_OUT calls using the async framework.
 *
 * This wrapper only makes it more comfortable to answer IPC_M_SHARE_OUT
 * calls so that the user doesn't have to remember the meaning of each IPC
 * argument.
 *
 * @param call IPC_M_SHARE_OUT call to answer.
 * @param dst  Address of the storage for the destination address space area
 *             base address.
 *
 * @return  Zero on success or a value from @ref errno.h on failure.
 *
 */
errno_t async_share_out_finalize(ipc_call_t *call, void **dst)
{
        assert(call);

        return ipc_answer_2(call->cap_handle, EOK, (sysarg_t) __progsymbols.end,
            (sysarg_t) dst);
}

/** Wrapper for receiving the IPC_M_DATA_READ calls using the async framework.
 *
 * This wrapper only makes it more comfortable to receive IPC_M_DATA_READ
 * calls so that the user doesn't have to remember the meaning of each IPC
 * argument.
 *
 * So far, this wrapper is to be used from within a connection fibril.
 *
 * @param call Storage for the data of the IPC_M_DATA_READ.
 * @param size Storage for the maximum size. Can be NULL.
 *
 * @return True on success, false on failure.
 *
 */
bool async_data_read_receive(ipc_call_t *call, size_t *size)
{
        assert(call);

        async_get_call(call);

        if (IPC_GET_IMETHOD(*call) != IPC_M_DATA_READ)
                return false;

        if (size)
                *size = (size_t) IPC_GET_ARG2(*call);

        return true;
}

/** Wrapper for answering the IPC_M_DATA_READ calls using the async framework.
 *
 * This wrapper only makes it more comfortable to answer IPC_M_DATA_READ
 * calls so that the user doesn't have to remember the meaning of each IPC
 * argument.
 *
 * @param call IPC_M_DATA_READ call to answer.
 * @param src  Source address for the IPC_M_DATA_READ call.
 * @param size Size for the IPC_M_DATA_READ call. Can be smaller than
 *             the maximum size announced by the sender.
 *
 * @return  Zero on success or a value from @ref errno.h on failure.
 *
 */
errno_t async_data_read_finalize(ipc_call_t *call, const void *src, size_t size)
{
        assert(call);

        return ipc_answer_2(call->cap_handle, EOK, (sysarg_t) src,
            (sysarg_t) size);
}

/** Wrapper for forwarding any read request
 *
 */
errno_t async_data_read_forward_fast(async_exch_t *exch, sysarg_t imethod,
    sysarg_t arg1, sysarg_t arg2, sysarg_t arg3, sysarg_t arg4,
    ipc_call_t *dataptr)
{
        if (exch == NULL)
                return ENOENT;

        ipc_call_t call;
        if (!async_data_read_receive(&call, NULL)) {
                async_answer_0(&call, EINVAL);
                return EINVAL;
        }

        aid_t msg = async_send_fast(exch, imethod, arg1, arg2, arg3, arg4,
            dataptr);
        if (msg == 0) {
                async_answer_0(&call, EINVAL);
                return EINVAL;
        }

        errno_t retval = ipc_forward_fast(call.cap_handle, exch->phone, 0, 0, 0,
            IPC_FF_ROUTE_FROM_ME);
        if (retval != EOK) {
                async_forget(msg);
                async_answer_0(&call, retval);
                return retval;
        }

        errno_t rc;
        async_wait_for(msg, &rc);

        return (errno_t) rc;
}

/** Wrapper for receiving the IPC_M_DATA_WRITE calls using the async framework.
 *
 * This wrapper only makes it more comfortable to receive IPC_M_DATA_WRITE
 * calls so that the user doesn't have to remember the meaning of each IPC
 * argument.
 *
 * So far, this wrapper is to be used from within a connection fibril.
 *
 * @param call Storage for the data of the IPC_M_DATA_WRITE.
 * @param size Storage for the suggested size. May be NULL.
 *
 * @return True on success, false on failure.
 *
 */
bool async_data_write_receive(ipc_call_t *call, size_t *size)
{
        assert(call);

        async_get_call(call);

        if (IPC_GET_IMETHOD(*call) != IPC_M_DATA_WRITE)
                return false;

        if (size)
                *size = (size_t) IPC_GET_ARG2(*call);

        return true;
}

/** Wrapper for answering the IPC_M_DATA_WRITE calls using the async framework.
 *
 * This wrapper only makes it more comfortable to answer IPC_M_DATA_WRITE
 * calls so that the user doesn't have to remember the meaning of each IPC
 * argument.
 *
 * @param call IPC_M_DATA_WRITE call to answer.
 * @param dst  Final destination address for the IPC_M_DATA_WRITE call.
 * @param size Final size for the IPC_M_DATA_WRITE call.
 *
 * @return  Zero on success or a value from @ref errno.h on failure.
 *
 */
errno_t async_data_write_finalize(ipc_call_t *call, void *dst, size_t size)
{
        assert(call);

        return async_answer_2(call, EOK, (sysarg_t) dst, (sysarg_t) size);
}

/** Wrapper for receiving binary data or strings
 *
 * This wrapper only makes it more comfortable to use async_data_write_*
 * functions to receive binary data or strings.
 *
 * @param data       Pointer to data pointer (which should be later disposed
 *                   by free()). If the operation fails, the pointer is not
 *                   touched.
 * @param nullterm   If true then the received data is always zero terminated.
 *                   This also causes to allocate one extra byte beyond the
 *                   raw transmitted data.
 * @param min_size   Minimum size (in bytes) of the data to receive.
 * @param max_size   Maximum size (in bytes) of the data to receive. 0 means
 *                   no limit.
 * @param granulariy If non-zero then the size of the received data has to
 *                   be divisible by this value.
 * @param received   If not NULL, the size of the received data is stored here.
 *
 * @return Zero on success or a value from @ref errno.h on failure.
 *
 */
errno_t async_data_write_accept(void **data, const bool nullterm,
    const size_t min_size, const size_t max_size, const size_t granularity,
    size_t *received)
{
        assert(data);

        ipc_call_t call;
        size_t size;
        if (!async_data_write_receive(&call, &size)) {
                async_answer_0(&call, EINVAL);
                return EINVAL;
        }

        if (size < min_size) {
                async_answer_0(&call, EINVAL);
                return EINVAL;
        }

        if ((max_size > 0) && (size > max_size)) {
                async_answer_0(&call, EINVAL);
                return EINVAL;
        }

        if ((granularity > 0) && ((size % granularity) != 0)) {
                async_answer_0(&call, EINVAL);
                return EINVAL;
        }

        void *arg_data;

        if (nullterm)
                arg_data = malloc(size + 1);
        else
                arg_data = malloc(size);

        if (arg_data == NULL) {
                async_answer_0(&call, ENOMEM);
                return ENOMEM;
        }

        errno_t rc = async_data_write_finalize(&call, arg_data, size);
        if (rc != EOK) {
                free(arg_data);
                return rc;
        }

        if (nullterm)
                ((char *) arg_data)[size] = 0;

        *data = arg_data;
        if (received != NULL)
                *received = size;

        return EOK;
}

/** Wrapper for voiding any data that is about to be received
 *
 * This wrapper can be used to void any pending data
 *
 * @param retval Error value from @ref errno.h to be returned to the caller.
 *
 */
void async_data_write_void(errno_t retval)
{
        ipc_call_t call;
        async_data_write_receive(&call, NULL);
        async_answer_0(&call, retval);
}

/** Wrapper for forwarding any data that is about to be received
 *
 */
errno_t async_data_write_forward_fast(async_exch_t *exch, sysarg_t imethod,
    sysarg_t arg1, sysarg_t arg2, sysarg_t arg3, sysarg_t arg4,
    ipc_call_t *dataptr)
{
        if (exch == NULL)
                return ENOENT;

        ipc_call_t call;
        if (!async_data_write_receive(&call, NULL)) {
                async_answer_0(&call, EINVAL);
                return EINVAL;
        }

        aid_t msg = async_send_fast(exch, imethod, arg1, arg2, arg3, arg4,
            dataptr);
        if (msg == 0) {
                async_answer_0(&call, EINVAL);
                return EINVAL;
        }

        errno_t retval = ipc_forward_fast(call.cap_handle, exch->phone, 0, 0, 0,
            IPC_FF_ROUTE_FROM_ME);
        if (retval != EOK) {
                async_forget(msg);
                async_answer_0(&call, retval);
                return retval;
        }

        errno_t rc;
        async_wait_for(msg, &rc);

        return (errno_t) rc;
}

/** Wrapper for receiving the IPC_M_CONNECT_TO_ME calls.
 *
 * If the current call is IPC_M_CONNECT_TO_ME then a new
 * async session is created for the accepted phone.
 *
 * @param mgmt Exchange management style.
 *
 * @return New async session.
 * @return NULL on failure.
 *
 */
async_sess_t *async_callback_receive(exch_mgmt_t mgmt)
{
        /* Accept the phone */
        ipc_call_t call;
        async_get_call(&call);

        cap_phone_handle_t phandle = (cap_handle_t) IPC_GET_ARG5(call);

        if ((IPC_GET_IMETHOD(call) != IPC_M_CONNECT_TO_ME) ||
            !CAP_HANDLE_VALID((phandle))) {
                async_answer_0(&call, EINVAL);
                return NULL;
        }

        async_sess_t *sess = calloc(1, sizeof(async_sess_t));
        if (sess == NULL) {
                async_answer_0(&call, ENOMEM);
                return NULL;
        }

        sess->iface = 0;
        sess->mgmt = mgmt;
        sess->phone = phandle;

        fibril_mutex_initialize(&sess->remote_state_mtx);
        list_initialize(&sess->exch_list);
        fibril_mutex_initialize(&sess->mutex);

        /* Acknowledge the connected phone */
        async_answer_0(&call, EOK);

        return sess;
}

/** Wrapper for receiving the IPC_M_CONNECT_TO_ME calls.
 *
 * If the call is IPC_M_CONNECT_TO_ME then a new
 * async session is created. However, the phone is
 * not accepted automatically.
 *
 * @param mgmt   Exchange management style.
 * @param call   Call data.
 *
 * @return New async session.
 * @return NULL on failure.
 * @return NULL if the call is not IPC_M_CONNECT_TO_ME.
 *
 */
async_sess_t *async_callback_receive_start(exch_mgmt_t mgmt, ipc_call_t *call)
{
        cap_phone_handle_t phandle = (cap_handle_t) IPC_GET_ARG5(*call);

        if ((IPC_GET_IMETHOD(*call) != IPC_M_CONNECT_TO_ME) ||
            !CAP_HANDLE_VALID((phandle)))
                return NULL;

        async_sess_t *sess = calloc(1, sizeof(async_sess_t));
        if (sess == NULL)
                return NULL;

        sess->iface = 0;
        sess->mgmt = mgmt;
        sess->phone = phandle;

        fibril_mutex_initialize(&sess->remote_state_mtx);
        list_initialize(&sess->exch_list);
        fibril_mutex_initialize(&sess->mutex);

        return sess;
}

bool async_state_change_receive(ipc_call_t *call)
{
        assert(call);

        async_get_call(call);

        if (IPC_GET_IMETHOD(*call) != IPC_M_STATE_CHANGE_AUTHORIZE)
                return false;

        return true;
}

errno_t async_state_change_finalize(ipc_call_t *call, async_exch_t *other_exch)
{
        assert(call);

        return async_answer_1(call, EOK, CAP_HANDLE_RAW(other_exch->phone));
}

__noreturn void async_manager(void)
{
        fibril_event_t ever = FIBRIL_EVENT_INIT;
        fibril_wait_for(&ever);
        __builtin_unreachable();
}

/** @}
 */