port/rt/sigev_thread.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#pragma ident   "%Z%%M% %I% %E% SMI"

#include "lint.h"
#include "thr_uberdata.h"
#include <sys/types.h>
#include <pthread.h>
#include <unistd.h>
#include <stdlib.h>
#include <thread.h>
#include <pthread.h>
#include <synch.h>
#include <port.h>
#include <signal.h>
#include <stdio.h>
#include <errno.h>
#include <stdarg.h>
#include <string.h>
#include <sys/aiocb.h>
#include <time.h>
#include <signal.h>
#include <fcntl.h>
#include "sigev_thread.h"

/*
 * There is but one spawner for all aio operations.
 */
thread_communication_data_t *sigev_aio_tcd = NULL;

/*
 * Set non-zero via _RT_DEBUG to enable debugging printf's.
 */
static int _rt_debug = 0;

void
init_sigev_thread(void)
{
    char *ldebug;

    if ((ldebug = getenv("_RT_DEBUG")) != NULL)
        _rt_debug = atoi(ldebug);
}

/*
 * Routine to print debug messages:
 * If _rt_debug is set, printf the debug message to stderr
 * with an appropriate prefix.
 */
/*PRINTFLIKE1*/
static void
dprintf(const char *format, ...)
{
    if (_rt_debug) {
        va_list alist;

        va_start(alist, format);
        flockfile(stderr);
        pthread_cleanup_push(funlockfile, stderr);
        (void) fputs("DEBUG: ", stderr);
        (void) vfprintf(stderr, format, alist);
        pthread_cleanup_pop(1);     /* funlockfile(stderr) */
        va_end(alist);
    }
}

/*
 * The notify_thread() function can be used as the start function of a new
 * thread but it is normally called from notifier(), below, in the context
 * of a thread pool worker thread.  It is used as the start function of a
 * new thread only when individual pthread attributes differ from those
 * that are common to all workers.  This only occurs in the AIO case.
 */
static void *
notify_thread(void *arg)
{
    sigev_thread_data_t *stdp = arg;
    void (*function)(union sigval) = stdp->std_func;
    union sigval argument = stdp->std_arg;

    lfree(stdp, sizeof (*stdp));
    function(argument);
    return (NULL);
}

/*
 * Thread pool interface to call the user-supplied notification function.
 */
static void
notifier(void *arg)
{
    (void) notify_thread(arg);
}

/*
 * This routine adds a new work request, described by function
 * and argument, to the list of outstanding jobs.
 * It returns 0 indicating success.  A value != 0 indicates an error.
 */
static int
sigev_add_work(thread_communication_data_t *tcdp,
    void (*function)(union sigval), union sigval argument)
{
    tpool_t *tpool = tcdp->tcd_poolp;
    sigev_thread_data_t *stdp;

    if (tpool == NULL)
        return (EINVAL);
    if ((stdp = lmalloc(sizeof (*stdp))) == NULL)
        return (errno);
    stdp->std_func = function;
    stdp->std_arg = argument;
    if (tpool_dispatch(tpool, notifier, stdp) != 0) {
        lfree(stdp, sizeof (*stdp));
        return (errno);
    }
    return (0);
}

static void
sigev_destroy_pool(thread_communication_data_t *tcdp)
{
    if (tcdp->tcd_poolp != NULL)
        tpool_abandon(tcdp->tcd_poolp);
    tcdp->tcd_poolp = NULL;

    if (tcdp->tcd_subsystem == MQ) {
        /*
         * synchronize with del_sigev_mq()
         */
        sig_mutex_lock(&tcdp->tcd_lock);
        tcdp->tcd_server_id = 0;
        if (tcdp->tcd_msg_closing) {
            (void) cond_broadcast(&tcdp->tcd_cv);
            sig_mutex_unlock(&tcdp->tcd_lock);
            return;     /* del_sigev_mq() will free the tcd */
        }
        sig_mutex_unlock(&tcdp->tcd_lock);
    }

    /*
     * now delete everything
     */
    free_sigev_handler(tcdp);
}

/*
 * timer_spawner(), mqueue_spawner(), and aio_spawner() are the main
 * functions for the daemon threads that get the event(s) for the
 * respective SIGEV_THREAD subsystems.  There is one timer spawner for
 * each timer_create(), one mqueue spawner for every mq_open(), and
 * exactly one aio spawner for all aio requests.  These spawners add
 * work requests to be done by a pool of daemon worker threads.  In case
 * the event requires creation of a worker thread with different pthread
 * attributes than those from the pool of workers, a new daemon thread
 * with these attributes is spawned apart from the pool of workers.
 * If the spawner fails to add work or fails to create an additional
 * thread because of lacking resources, it puts the event back into
 * the kernel queue and re-tries some time later.
 */

void *
timer_spawner(void *arg)
{
    thread_communication_data_t *tcdp = (thread_communication_data_t *)arg;
    port_event_t port_event;

    /* destroy the pool if we are cancelled */
    pthread_cleanup_push(sigev_destroy_pool, tcdp);

    for (;;) {
        if (port_get(tcdp->tcd_port, &port_event, NULL) != 0) {
            dprintf("port_get on port %d failed with %d <%s>\n",
                tcdp->tcd_port, errno, strerror(errno));
            break;
        }
        switch (port_event.portev_source) {
        case PORT_SOURCE_TIMER:
            break;
        case PORT_SOURCE_ALERT:
            if (port_event.portev_events != SIGEV_THREAD_TERM)
                errno = EPROTO;
            goto out;
        default:
            dprintf("port_get on port %d returned %u "
                "(not PORT_SOURCE_TIMER)\n",
                tcdp->tcd_port, port_event.portev_source);
            errno = EPROTO;
            goto out;
        }

        tcdp->tcd_overruns = port_event.portev_events - 1;
        if (sigev_add_work(tcdp,
            tcdp->tcd_notif.sigev_notify_function,
            tcdp->tcd_notif.sigev_value) != 0)
            break;
        /* wait until job is done before looking for another */
        tpool_wait(tcdp->tcd_poolp);
    }
out:
    pthread_cleanup_pop(1);
    return (NULL);
}

void *
mqueue_spawner(void *arg)
{
    thread_communication_data_t *tcdp = (thread_communication_data_t *)arg;
    int ret = 0;
    int ntype;
    void (*function)(union sigval);
    union sigval argument;

    /* destroy the pool if we are cancelled */
    pthread_cleanup_push(sigev_destroy_pool, tcdp);

    while (ret == 0) {
        sig_mutex_lock(&tcdp->tcd_lock);
        pthread_cleanup_push(sig_mutex_unlock, &tcdp->tcd_lock);
        while ((ntype = tcdp->tcd_msg_enabled) == 0)
            (void) sig_cond_wait(&tcdp->tcd_cv, &tcdp->tcd_lock);
        pthread_cleanup_pop(1);

        while (sem_wait(tcdp->tcd_msg_avail) == -1)
            continue;

        sig_mutex_lock(&tcdp->tcd_lock);
        tcdp->tcd_msg_enabled = 0;
        sig_mutex_unlock(&tcdp->tcd_lock);

        /* ASSERT(ntype == SIGEV_THREAD || ntype == SIGEV_PORT); */
        if (ntype == SIGEV_THREAD) {
            function = tcdp->tcd_notif.sigev_notify_function;
            argument.sival_ptr = tcdp->tcd_msg_userval;
            ret = sigev_add_work(tcdp, function, argument);
        } else {    /* ntype == SIGEV_PORT */
            ret = _port_dispatch(tcdp->tcd_port, 0, PORT_SOURCE_MQ,
                0, (uintptr_t)tcdp->tcd_msg_object,
                tcdp->tcd_msg_userval);
        }
    }
    sig_mutex_unlock(&tcdp->tcd_lock);

    pthread_cleanup_pop(1);
    return (NULL);
}

void *
aio_spawner(void *arg)
{
    thread_communication_data_t *tcdp = (thread_communication_data_t *)arg;
    int error = 0;
    void (*function)(union sigval);
    union sigval argument;
    port_event_t port_event;
    struct sigevent *sigevp;
    timespec_t delta;
    pthread_attr_t *attrp;

    /* destroy the pool if we are cancelled */
    pthread_cleanup_push(sigev_destroy_pool, tcdp);

    while (error == 0) {
        if (port_get(tcdp->tcd_port, &port_event, NULL) != 0) {
            error = errno;
            dprintf("port_get on port %d failed with %d <%s>\n",
                tcdp->tcd_port, error, strerror(error));
            break;
        }
        switch (port_event.portev_source) {
        case PORT_SOURCE_AIO:
            break;
        case PORT_SOURCE_ALERT:
            if (port_event.portev_events != SIGEV_THREAD_TERM)
                errno = EPROTO;
            goto out;
        default:
            dprintf("port_get on port %d returned %u "
                "(not PORT_SOURCE_AIO)\n",
                tcdp->tcd_port, port_event.portev_source);
            errno = EPROTO;
            goto out;
        }
        argument.sival_ptr = port_event.portev_user;
        switch (port_event.portev_events) {
        case AIOLIO:
#if !defined(_LP64)
        case AIOLIO64:
#endif
            sigevp = (struct sigevent *)port_event.portev_object;
            function = sigevp->sigev_notify_function;
            attrp = sigevp->sigev_notify_attributes;
            break;
        case AIOAREAD:
        case AIOAWRITE:
        case AIOFSYNC:
            {
            aiocb_t *aiocbp =
                (aiocb_t *)port_event.portev_object;
            function = aiocbp->aio_sigevent.sigev_notify_function;
            attrp = aiocbp->aio_sigevent.sigev_notify_attributes;
            break;
            }
#if !defined(_LP64)
        case AIOAREAD64:
        case AIOAWRITE64:
        case AIOFSYNC64:
            {
            aiocb64_t *aiocbp =
                (aiocb64_t *)port_event.portev_object;
            function = aiocbp->aio_sigevent.sigev_notify_function;
            attrp = aiocbp->aio_sigevent.sigev_notify_attributes;
            break;
            }
#endif
        default:
            function = NULL;
            attrp = NULL;
            break;
        }

        if (function == NULL)
            error = EINVAL;
        else if (pthread_attr_equal(attrp, tcdp->tcd_attrp))
            error = sigev_add_work(tcdp, function, argument);
        else {
            /*
             * The attributes don't match.
             * Spawn a thread with the non-matching attributes.
             */
            pthread_attr_t local_attr;
            sigev_thread_data_t *stdp;

            if ((stdp = lmalloc(sizeof (*stdp))) == NULL)
                error = ENOMEM;
            else
                error = pthread_attr_clone(&local_attr, attrp);

            if (error == 0) {
                (void) pthread_attr_setdetachstate(
                    &local_attr, PTHREAD_CREATE_DETACHED);
                (void) pthread_attr_setdaemonstate_np(
                    &local_attr, PTHREAD_CREATE_DAEMON_NP);
                stdp->std_func = function;
                stdp->std_arg = argument;
                error = pthread_create(NULL, &local_attr,
                    notify_thread, stdp);
                (void) pthread_attr_destroy(&local_attr);
            }
            if (error && stdp != NULL)
                lfree(stdp, sizeof (*stdp));
        }

        if (error) {
            dprintf("Cannot add work, error=%d <%s>.\n",
                error, strerror(error));
            if (error == EAGAIN || error == ENOMEM) {
                /* (Temporary) no resources are available. */
                if (_port_dispatch(tcdp->tcd_port, 0,
                    PORT_SOURCE_AIO, port_event.portev_events,
                    port_event.portev_object,
                    port_event.portev_user) != 0)
                    break;
                error = 0;
                delta.tv_sec = 0;
                delta.tv_nsec = NANOSEC / 20;   /* 50 msec */
                (void) nanosleep(&delta, NULL);
            }
        }
    }
out:
    pthread_cleanup_pop(1);
    return (NULL);
}

/*
 * Allocate a thread_communication_data_t block.
 */
static thread_communication_data_t *
alloc_sigev_handler(subsystem_t caller)
{
    thread_communication_data_t *tcdp;

    if ((tcdp = lmalloc(sizeof (*tcdp))) != NULL) {
        tcdp->tcd_subsystem = caller;
        tcdp->tcd_port = -1;
        (void) mutex_init(&tcdp->tcd_lock, USYNC_THREAD, NULL);
        (void) cond_init(&tcdp->tcd_cv, USYNC_THREAD, NULL);
    }
    return (tcdp);
}

/*
 * Free a thread_communication_data_t block.
 */
void
free_sigev_handler(thread_communication_data_t *tcdp)
{
    if (tcdp->tcd_attrp) {
        (void) pthread_attr_destroy(tcdp->tcd_attrp);
        tcdp->tcd_attrp = NULL;
    }
    (void) memset(&tcdp->tcd_notif, 0, sizeof (tcdp->tcd_notif));

    switch (tcdp->tcd_subsystem) {
    case TIMER:
    case AIO:
        if (tcdp->tcd_port >= 0)
            (void) close(tcdp->tcd_port);
        break;
    case MQ:
        tcdp->tcd_msg_avail = NULL;
        tcdp->tcd_msg_object = NULL;
        tcdp->tcd_msg_userval = NULL;
        tcdp->tcd_msg_enabled = 0;
        break;
    }

    lfree(tcdp, sizeof (*tcdp));
}

/*
 * Initialize data structure and create the port.
 */
thread_communication_data_t *
setup_sigev_handler(const struct sigevent *sigevp, subsystem_t caller)
{
    thread_communication_data_t *tcdp;
    int error;

    if (sigevp == NULL) {
        errno = EINVAL;
        return (NULL);
    }

    if ((tcdp = alloc_sigev_handler(caller)) == NULL) {
        errno = ENOMEM;
        return (NULL);
    }

    if (sigevp->sigev_notify_attributes == NULL)
        tcdp->tcd_attrp = NULL;     /* default attributes */
    else {
        /*
         * We cannot just copy the sigevp->sigev_notify_attributes
         * pointer.  We need to initialize a new pthread_attr_t
         * structure with the values from the user-supplied
         * pthread_attr_t.
         */
        tcdp->tcd_attrp = &tcdp->tcd_user_attr;
        error = pthread_attr_clone(tcdp->tcd_attrp,
            sigevp->sigev_notify_attributes);
        if (error) {
            tcdp->tcd_attrp = NULL;
            free_sigev_handler(tcdp);
            errno = error;
            return (NULL);
        }
    }
    tcdp->tcd_notif = *sigevp;
    tcdp->tcd_notif.sigev_notify_attributes = tcdp->tcd_attrp;

    if (caller == TIMER || caller == AIO) {
        if ((tcdp->tcd_port = port_create()) < 0 ||
            fcntl(tcdp->tcd_port, FD_CLOEXEC) == -1) {
            free_sigev_handler(tcdp);
            errno = EBADF;
            return (NULL);
        }
    }
    return (tcdp);
}

/*
 * Create a thread pool and launch the spawner.
 */
int
launch_spawner(thread_communication_data_t *tcdp)
{
    int ret;
    int maxworkers;
    void *(*spawner)(void *);
    sigset_t set;
    sigset_t oset;

    switch (tcdp->tcd_subsystem) {
    case TIMER:
        spawner = timer_spawner;
        maxworkers = 1;
        break;
    case MQ:
        spawner = mqueue_spawner;
        maxworkers = 1;
        break;
    case AIO:
        spawner = aio_spawner;
        maxworkers = 100;
        break;
    default:
        return (-1);
    }
    tcdp->tcd_poolp = tpool_create(1, maxworkers, 20,
        tcdp->tcd_notif.sigev_notify_attributes);
    if (tcdp->tcd_poolp == NULL)
        return (-1);
    /* create the spawner with all signals blocked */
    (void) sigfillset(&set);
    (void) thr_sigsetmask(SIG_SETMASK, &set, &oset);
    ret = thr_create(NULL, 0, spawner, tcdp,
        THR_DETACHED | THR_DAEMON, &tcdp->tcd_server_id);
    (void) thr_sigsetmask(SIG_SETMASK, &oset, NULL);
    if (ret != 0) {
        tpool_destroy(tcdp->tcd_poolp);
        tcdp->tcd_poolp = NULL;
        return (-1);
    }
    return (0);
}

/*
 * Delete the data associated with the sigev_thread timer, if timer is
 * associated with such a notification option.
 * Destroy the timer_spawner thread.
 */
int
del_sigev_timer(timer_t timer)
{
    int rc = 0;
    thread_communication_data_t *tcdp;

    if ((uint_t)timer < timer_max && (tcdp = timer_tcd[timer]) != NULL) {
        sig_mutex_lock(&tcdp->tcd_lock);
        if (tcdp->tcd_port >= 0) {
            if ((rc = port_alert(tcdp->tcd_port,
                PORT_ALERT_SET, SIGEV_THREAD_TERM, NULL)) == 0) {
                dprintf("del_sigev_timer(%d) OK.\n", timer);
            }
        }
        timer_tcd[timer] = NULL;
        sig_mutex_unlock(&tcdp->tcd_lock);
    }
    return (rc);
}

int
sigev_timer_getoverrun(timer_t timer)
{
    thread_communication_data_t *tcdp;

    if ((uint_t)timer < timer_max && (tcdp = timer_tcd[timer]) != NULL)
        return (tcdp->tcd_overruns);
    return (0);
}

static void
del_sigev_mq_cleanup(thread_communication_data_t *tcdp)
{
    sig_mutex_unlock(&tcdp->tcd_lock);
    free_sigev_handler(tcdp);
}

/*
 * Delete the data associated with the sigev_thread message queue,
 * if the message queue is associated with such a notification option.
 * Destroy the mqueue_spawner thread.
 */
void
del_sigev_mq(thread_communication_data_t *tcdp)
{
    pthread_t server_id;
    int rc;

    sig_mutex_lock(&tcdp->tcd_lock);

    server_id = tcdp->tcd_server_id;
    tcdp->tcd_msg_closing = 1;
    if ((rc = pthread_cancel(server_id)) != 0) {    /* "can't happen" */
        sig_mutex_unlock(&tcdp->tcd_lock);
        dprintf("Fail to cancel %u with error %d <%s>.\n",
            server_id, rc, strerror(rc));
        return;
    }

    /*
     * wait for sigev_destroy_pool() to finish
     */
    pthread_cleanup_push(del_sigev_mq_cleanup, tcdp);
    while (tcdp->tcd_server_id == server_id)
        (void) sig_cond_wait(&tcdp->tcd_cv, &tcdp->tcd_lock);
    pthread_cleanup_pop(1);
}

/*
 * POSIX aio:
 * If the notification type is SIGEV_THREAD, set up
 * the port number for notifications.  Create the
 * thread pool and launch the spawner if necessary.
 * If the notification type is not SIGEV_THREAD, do nothing.
 */
int
_aio_sigev_thread_init(struct sigevent *sigevp)
{
    static mutex_t sigev_aio_lock = DEFAULTMUTEX;
    static cond_t sigev_aio_cv = DEFAULTCV;
    static int sigev_aio_busy = 0;

    thread_communication_data_t *tcdp;
    int port;
    int cancel_state;
    int rc = 0;

    if (sigevp == NULL ||
        sigevp->sigev_notify != SIGEV_THREAD ||
        sigevp->sigev_notify_function == NULL)
        return (0);

    lmutex_lock(&sigev_aio_lock);
    (void) pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &cancel_state);
    while (sigev_aio_busy)
        (void) cond_wait(&sigev_aio_cv, &sigev_aio_lock);
    (void) pthread_setcancelstate(cancel_state, NULL);
    if ((tcdp = sigev_aio_tcd) != NULL)
        port = tcdp->tcd_port;
    else {
        sigev_aio_busy = 1;
        lmutex_unlock(&sigev_aio_lock);

        tcdp = setup_sigev_handler(sigevp, AIO);
        if (tcdp == NULL) {
            port = -1;
            rc = -1;
        } else if (launch_spawner(tcdp) != 0) {
            free_sigev_handler(tcdp);
            tcdp = NULL;
            port = -1;
            rc = -1;
        } else {
            port = tcdp->tcd_port;
        }

        lmutex_lock(&sigev_aio_lock);
        sigev_aio_tcd = tcdp;
        sigev_aio_busy = 0;
        (void) cond_broadcast(&sigev_aio_cv);
    }
    lmutex_unlock(&sigev_aio_lock);
    sigevp->sigev_signo = port;
    return (rc);
}

int
_aio_sigev_thread(aiocb_t *aiocbp)
{
    if (aiocbp == NULL)
        return (0);
    return (_aio_sigev_thread_init(&aiocbp->aio_sigevent));
}

#if !defined(_LP64)
int
_aio_sigev_thread64(aiocb64_t *aiocbp)
{
    if (aiocbp == NULL)
        return (0);
    return (_aio_sigev_thread_init(&aiocbp->aio_sigevent));
}
#endif

/*
 * Cleanup POSIX aio after fork1() in the child process.
 */
void
postfork1_child_sigev_aio(void)
{
    thread_communication_data_t *tcdp;

    if ((tcdp = sigev_aio_tcd) != NULL) {
        sigev_aio_tcd = NULL;
        tcd_teardown(tcdp);
    }
}

/*
 * Utility function for the various postfork1_child_sigev_*() functions.
 * Clean up the tcdp data structure and close the port.
 */
void
tcd_teardown(thread_communication_data_t *tcdp)
{
    if (tcdp->tcd_poolp != NULL)
        tpool_abandon(tcdp->tcd_poolp);
    tcdp->tcd_poolp = NULL;
    tcdp->tcd_server_id = 0;
    free_sigev_handler(tcdp);
}