--- libaitsched/src/hooks.c 2011/10/04 12:34:33 1.2 +++ libaitsched/src/hooks.c 2013/06/03 20:52:53 1.18 @@ -3,7 +3,7 @@ * by Michael Pounov * * $Author: misho $ -* $Id: hooks.c,v 1.2 2011/10/04 12:34:33 misho Exp $ +* $Id: hooks.c,v 1.18 2013/06/03 20:52:53 misho Exp $ * ************************************************************************** The ELWIX and AITNET software is distributed under the following @@ -12,7 +12,7 @@ terms: All of the documentation and software included in the ELWIX and AITNET Releases is copyrighted by ELWIX - Sofia/Bulgaria -Copyright 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011 +Copyright 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013 by Michael Pounov . All rights reserved. Redistribution and use in source and binary forms, with or without @@ -49,27 +49,19 @@ SUCH DAMAGE. /* * sched_hook_init() - Default INIT hook + * * @root = root task - * @data = optional data if !=NULL + * @arg = unused * return: <0 errors and 0 ok */ void * -sched_hook_init(void *root, void *data) +sched_hook_init(void *root, void *arg __unused) { sched_root_task_t *r = root; - if (!r || r->root_data.iov_base || r->root_data.iov_len) + if (!r) return (void*) -1; - r->root_data.iov_base = malloc(sizeof(struct sched_IO)); - if (!r->root_data.iov_base) { - LOGERR; - return (void*) -1; - } else { - r->root_data.iov_len = sizeof(struct sched_IO); - memset(r->root_data.iov_base, 0, r->root_data.iov_len); - } - r->root_kq = kqueue(); if (r->root_kq == -1) { LOGERR; @@ -81,6 +73,7 @@ sched_hook_init(void *root, void *data) /* * sched_hook_fini() - Default FINI hook + * * @root = root task * @arg = unused * return: <0 errors and 0 ok @@ -98,17 +91,12 @@ sched_hook_fini(void *root, void *arg __unused) r->root_kq = 0; } - if (r->root_data.iov_base && r->root_data.iov_len) { - free(r->root_data.iov_base); - r->root_data.iov_base = NULL; - r->root_data.iov_len = 0; - } - return NULL; } /* * sched_hook_cancel() - Default CANCEL hook + * * @task = current task * @arg = unused * return: <0 errors and 0 ok @@ -116,25 +104,27 @@ sched_hook_fini(void *root, void *arg __unused) void * sched_hook_cancel(void *task, void *arg __unused) { - struct sched_IO *io; sched_task_t *t = task; struct kevent chg[1]; struct timespec timeout = { 0, 0 }; +#ifdef AIO_SUPPORT + struct aiocb *acb; +#ifdef EVFILT_LIO + register int i = 0; + struct aiocb **acbs; +#endif /* EVFILT_LIO */ +#endif /* AIO_SUPPORT */ - if (!t || !t->task_root || !ROOT_DATA(t->task_root) || !ROOT_DATLEN(t->task_root)) + if (!t || !TASK_ROOT(t)) return (void*) -1; - else - io = ROOT_DATA(t->task_root); - switch (t->task_type) { + switch (TASK_TYPE(t)) { case taskREAD: #ifdef __NetBSD__ EV_SET(&chg[0], TASK_FD(t), EVFILT_READ, EV_DELETE, 0, 0, (intptr_t) TASK_FD(t)); #else EV_SET(&chg[0], TASK_FD(t), EVFILT_READ, EV_DELETE, 0, 0, (void*) TASK_FD(t)); #endif - kevent(t->task_root->root_kq, chg, 1, NULL, 0, &timeout); - FD_CLR(TASK_FD(t), &io->rfd); break; case taskWRITE: #ifdef __NetBSD__ @@ -142,18 +132,134 @@ sched_hook_cancel(void *task, void *arg __unused) #else EV_SET(&chg[0], TASK_FD(t), EVFILT_WRITE, EV_DELETE, 0, 0, (void*) TASK_FD(t)); #endif - kevent(t->task_root->root_kq, chg, 1, NULL, 0, &timeout); - FD_CLR(TASK_FD(t), &io->wfd); break; - default: + case taskALARM: +#ifdef __NetBSD__ + EV_SET(&chg[0], (uintptr_t) TASK_DATA(t), EVFILT_TIMER, EV_DELETE, + 0, 0, (intptr_t) TASK_DATA(t)); +#else + EV_SET(&chg[0], (uintptr_t) TASK_DATA(t), EVFILT_TIMER, EV_DELETE, + 0, 0, (void*) TASK_DATA(t)); +#endif break; + case taskNODE: +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_FD(t), EVFILT_VNODE, EV_DELETE, 0, 0, (intptr_t) TASK_FD(t)); +#else + EV_SET(&chg[0], TASK_FD(t), EVFILT_VNODE, EV_DELETE, 0, 0, (void*) TASK_FD(t)); +#endif + break; + case taskPROC: +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_VAL(t), EVFILT_PROC, EV_DELETE, 0, 0, (intptr_t) TASK_VAL(t)); +#else + EV_SET(&chg[0], TASK_VAL(t), EVFILT_PROC, EV_DELETE, 0, 0, (void*) TASK_VAL(t)); +#endif + break; + case taskSIGNAL: +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_VAL(t), EVFILT_SIGNAL, EV_DELETE, 0, 0, (intptr_t) TASK_VAL(t)); +#else + EV_SET(&chg[0], TASK_VAL(t), EVFILT_SIGNAL, EV_DELETE, 0, 0, (void*) TASK_VAL(t)); +#endif + /* restore signal */ + signal(TASK_VAL(t), SIG_DFL); + break; +#ifdef AIO_SUPPORT + case taskAIO: +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_VAL(t), EVFILT_AIO, EV_DELETE, 0, 0, (intptr_t) TASK_VAL(t)); +#else + EV_SET(&chg[0], TASK_VAL(t), EVFILT_AIO, EV_DELETE, 0, 0, (void*) TASK_VAL(t)); +#endif + acb = (struct aiocb*) TASK_VAL(t); + if (acb) { + if (aio_cancel(acb->aio_fildes, acb) == AIO_CANCELED) + aio_return(acb); + free(acb); + TASK_VAL(t) = 0; + } + break; +#ifdef EVFILT_LIO + case taskLIO: +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_VAL(t), EVFILT_LIO, EV_DELETE, 0, 0, (intptr_t) TASK_VAL(t)); +#else + EV_SET(&chg[0], TASK_VAL(t), EVFILT_LIO, EV_DELETE, 0, 0, (void*) TASK_VAL(t)); +#endif + acbs = (struct aiocb**) TASK_VAL(t); + if (acbs) { + for (i = 0; i < TASK_DATLEN(t); i++) { + if (aio_cancel(acbs[i]->aio_fildes, acbs[i]) == AIO_CANCELED) + aio_return(acbs[i]); + free(acbs[i]); + } + free(acbs); + TASK_VAL(t) = 0; + } + break; +#endif /* EVFILT_LIO */ +#endif /* AIO_SUPPORT */ +#ifdef EVFILT_USER + case taskUSER: +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_VAL(t), EVFILT_USER, EV_DELETE, 0, 0, (intptr_t) TASK_VAL(t)); +#else + EV_SET(&chg[0], TASK_VAL(t), EVFILT_USER, EV_DELETE, 0, 0, (void*) TASK_VAL(t)); +#endif + break; +#endif + case taskTHREAD: +#ifdef HAVE_LIBPTHREAD + pthread_cancel((pthread_t) TASK_VAL(t)); +#endif + default: + return NULL; } + kevent(TASK_ROOT(t)->root_kq, chg, 1, NULL, 0, &timeout); return NULL; } +#ifdef HAVE_LIBPTHREAD /* + * sched_hook_thread() - Default THREAD hook + * + * @task = current task + * @arg = pthread attributes + * return: <0 errors and 0 ok + */ +void * +sched_hook_thread(void *task, void *arg) +{ + sched_task_t *t = task; + pthread_t tid; + sigset_t s, o; + + if (!t || !TASK_ROOT(t)) + return (void*) -1; + + sigfillset(&s); + pthread_sigmask(SIG_BLOCK, &s, &o); + if ((errno = pthread_create(&tid, (pthread_attr_t*) arg, + (void *(*)(void*)) _sched_threadWrapper, t))) { + LOGERR; + pthread_sigmask(SIG_SETMASK, &o, NULL); + return (void*) -1; + } else + TASK_VAL(t) = (u_long) tid; + + if (!TASK_ISLOCKED(t)) + TASK_LOCK(t); + + pthread_sigmask(SIG_SETMASK, &o, NULL); + return NULL; +} +#endif + +/* * sched_hook_read() - Default READ hook + * * @task = current task * @arg = unused * return: <0 errors and 0 ok @@ -161,28 +267,23 @@ sched_hook_cancel(void *task, void *arg __unused) void * sched_hook_read(void *task, void *arg __unused) { - struct sched_IO *io; sched_task_t *t = task; struct kevent chg[1]; struct timespec timeout = { 0, 0 }; - if (!t || !t->task_root || !ROOT_DATA(t->task_root) || !ROOT_DATLEN(t->task_root)) + if (!t || !TASK_ROOT(t)) return (void*) -1; - else - io = ROOT_DATA(t->task_root); - if (FD_ISSET(TASK_FD(t), &io->rfd)) - return NULL; - else - FD_SET(TASK_FD(t), &io->rfd); - #ifdef __NetBSD__ - EV_SET(&chg[0], TASK_FD(t), EVFILT_READ, EV_ADD, 0, 0, (intptr_t) TASK_FD(t)); + EV_SET(&chg[0], TASK_FD(t), EVFILT_READ, EV_ADD | EV_CLEAR, 0, 0, (intptr_t) TASK_FD(t)); #else - EV_SET(&chg[0], TASK_FD(t), EVFILT_READ, EV_ADD, 0, 0, (void*) TASK_FD(t)); + EV_SET(&chg[0], TASK_FD(t), EVFILT_READ, EV_ADD | EV_CLEAR, 0, 0, (void*) TASK_FD(t)); #endif - if (kevent(t->task_root->root_kq, chg, 1, NULL, 0, &timeout) == -1) { - LOGERR; + if (kevent(TASK_ROOT(t)->root_kq, chg, 1, NULL, 0, &timeout) == -1) { + if (TASK_ROOT(t)->root_hooks.hook_exec.exception) + TASK_ROOT(t)->root_hooks.hook_exec.exception(TASK_ROOT(t), NULL); + else + LOGERR; return (void*) -1; } @@ -191,6 +292,7 @@ sched_hook_read(void *task, void *arg __unused) /* * sched_hook_write() - Default WRITE hook + * * @task = current task * @arg = unused * return: <0 errors and 0 ok @@ -198,28 +300,60 @@ sched_hook_read(void *task, void *arg __unused) void * sched_hook_write(void *task, void *arg __unused) { - struct sched_IO *io; sched_task_t *t = task; struct kevent chg[1]; struct timespec timeout = { 0, 0 }; - if (!t || !t->task_root || !ROOT_DATA(t->task_root) || !ROOT_DATLEN(t->task_root)) + if (!t || !TASK_ROOT(t)) return (void*) -1; - else - io = ROOT_DATA(t->task_root); - if (FD_ISSET(TASK_FD(t), &io->wfd)) - return NULL; - else - FD_SET(TASK_FD(t), &io->wfd); +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_FD(t), EVFILT_WRITE, EV_ADD | EV_CLEAR, 0, 0, (intptr_t) TASK_FD(t)); +#else + EV_SET(&chg[0], TASK_FD(t), EVFILT_WRITE, EV_ADD | EV_CLEAR, 0, 0, (void*) TASK_FD(t)); +#endif + if (kevent(TASK_ROOT(t)->root_kq, chg, 1, NULL, 0, &timeout) == -1) { + if (TASK_ROOT(t)->root_hooks.hook_exec.exception) + TASK_ROOT(t)->root_hooks.hook_exec.exception(TASK_ROOT(t), NULL); + else + LOGERR; + return (void*) -1; + } + return NULL; +} + +/* + * sched_hook_alarm() - Default ALARM hook + * + * @task = current task + * @arg = unused + * return: <0 errors and 0 ok + */ +void * +sched_hook_alarm(void *task, void *arg __unused) +{ + sched_task_t *t = task; + struct kevent chg[1]; + struct timespec timeout = { 0, 0 }; + + if (!t || !TASK_ROOT(t)) + return (void*) -1; + #ifdef __NetBSD__ - EV_SET(&chg[0], TASK_FD(t), EVFILT_WRITE, EV_ADD, 0, 0, (intptr_t) TASK_FD(t)); + EV_SET(&chg[0], (uintptr_t) TASK_DATA(t), EVFILT_TIMER, EV_ADD | EV_ONESHOT, 0, + t->task_val.ts.tv_sec * 1000 + t->task_val.ts.tv_nsec / 1000000, + (intptr_t) TASK_DATA(t)); #else - EV_SET(&chg[0], TASK_FD(t), EVFILT_WRITE, EV_ADD, 0, 0, (void*) TASK_FD(t)); + EV_SET(&chg[0], (uintptr_t) TASK_DATA(t), EVFILT_TIMER, EV_ADD | EV_ONESHOT, 0, + t->task_val.ts.tv_sec * 1000 + t->task_val.ts.tv_nsec / 1000000, + (void*) TASK_DATA(t)); #endif - if (kevent(t->task_root->root_kq, chg, 1, NULL, 0, &timeout) == -1) { - LOGERR; + if (kevent(TASK_ROOT(t)->root_kq, chg, 1, NULL, 0, &timeout) == -1) { + if (TASK_ROOT(t)->root_hooks.hook_exec.exception) + TASK_ROOT(t)->root_hooks.hook_exec.exception(TASK_ROOT(t), NULL); + else + LOGERR; return (void*) -1; } @@ -227,7 +361,153 @@ sched_hook_write(void *task, void *arg __unused) } /* + * sched_hook_node() - Default NODE hook + * + * @task = current task + * @arg = unused + * return: <0 errors and 0 ok + */ +void * +sched_hook_node(void *task, void *arg __unused) +{ + sched_task_t *t = task; + struct kevent chg[1]; + struct timespec timeout = { 0, 0 }; + + if (!t || !TASK_ROOT(t)) + return (void*) -1; + +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_FD(t), EVFILT_VNODE, EV_ADD | EV_CLEAR, + NOTE_DELETE | NOTE_WRITE | NOTE_EXTEND | NOTE_ATTRIB | + NOTE_LINK | NOTE_RENAME | NOTE_REVOKE, 0, (intptr_t) TASK_FD(t)); +#else + EV_SET(&chg[0], TASK_FD(t), EVFILT_VNODE, EV_ADD | EV_CLEAR, + NOTE_DELETE | NOTE_WRITE | NOTE_EXTEND | NOTE_ATTRIB | + NOTE_LINK | NOTE_RENAME | NOTE_REVOKE, 0, (void*) TASK_FD(t)); +#endif + if (kevent(TASK_ROOT(t)->root_kq, chg, 1, NULL, 0, &timeout) == -1) { + if (TASK_ROOT(t)->root_hooks.hook_exec.exception) + TASK_ROOT(t)->root_hooks.hook_exec.exception(TASK_ROOT(t), NULL); + else + LOGERR; + return (void*) -1; + } + + return NULL; +} + +/* + * sched_hook_proc() - Default PROC hook + * + * @task = current task + * @arg = unused + * return: <0 errors and 0 ok + */ +void * +sched_hook_proc(void *task, void *arg __unused) +{ + sched_task_t *t = task; + struct kevent chg[1]; + struct timespec timeout = { 0, 0 }; + + if (!t || !TASK_ROOT(t)) + return (void*) -1; + +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_VAL(t), EVFILT_PROC, EV_ADD | EV_CLEAR, + NOTE_EXIT | NOTE_FORK | NOTE_EXEC | NOTE_TRACK, 0, (intptr_t) TASK_VAL(t)); +#else + EV_SET(&chg[0], TASK_VAL(t), EVFILT_PROC, EV_ADD | EV_CLEAR, + NOTE_EXIT | NOTE_FORK | NOTE_EXEC | NOTE_TRACK, 0, (void*) TASK_VAL(t)); +#endif + if (kevent(TASK_ROOT(t)->root_kq, chg, 1, NULL, 0, &timeout) == -1) { + if (TASK_ROOT(t)->root_hooks.hook_exec.exception) + TASK_ROOT(t)->root_hooks.hook_exec.exception(TASK_ROOT(t), NULL); + else + LOGERR; + return (void*) -1; + } + + return NULL; +} + +/* + * sched_hook_signal() - Default SIGNAL hook + * + * @task = current task + * @arg = unused + * return: <0 errors and 0 ok + */ +void * +sched_hook_signal(void *task, void *arg __unused) +{ + sched_task_t *t = task; + struct kevent chg[1]; + struct timespec timeout = { 0, 0 }; + + if (!t || !TASK_ROOT(t)) + return (void*) -1; + + /* ignore signal */ + signal(TASK_VAL(t), SIG_IGN); + +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_VAL(t), EVFILT_SIGNAL, EV_ADD, 0, 0, (intptr_t) TASK_VAL(t)); +#else + EV_SET(&chg[0], TASK_VAL(t), EVFILT_SIGNAL, EV_ADD, 0, 0, (void*) TASK_VAL(t)); +#endif + if (kevent(TASK_ROOT(t)->root_kq, chg, 1, NULL, 0, &timeout) == -1) { + if (TASK_ROOT(t)->root_hooks.hook_exec.exception) + TASK_ROOT(t)->root_hooks.hook_exec.exception(TASK_ROOT(t), NULL); + else + LOGERR; + return (void*) -1; + } + + return NULL; +} + +/* + * sched_hook_user() - Default USER hook + * + * @task = current task + * @arg = unused + * return: <0 errors and 0 ok + */ +#ifdef EVFILT_USER +void * +sched_hook_user(void *task, void *arg __unused) +{ + sched_task_t *t = task; + struct kevent chg[1]; + struct timespec timeout = { 0, 0 }; + + if (!t || !TASK_ROOT(t)) + return (void*) -1; + +#ifdef __NetBSD__ + EV_SET(&chg[0], TASK_VAL(t), EVFILT_USER, EV_ADD | EV_CLEAR, TASK_DATLEN(t), + 0, (intptr_t) TASK_VAL(t)); +#else + EV_SET(&chg[0], TASK_VAL(t), EVFILT_USER, EV_ADD | EV_CLEAR, TASK_DATLEN(t), + 0, (void*) TASK_VAL(t)); +#endif + if (kevent(TASK_ROOT(t)->root_kq, chg, 1, NULL, 0, &timeout) == -1) { + if (TASK_ROOT(t)->root_hooks.hook_exec.exception) + TASK_ROOT(t)->root_hooks.hook_exec.exception(TASK_ROOT(t), NULL); + else + LOGERR; + return (void*) -1; + } + + return NULL; +} +#endif + +/* * sched_hook_fetch() - Default FETCH hook + * * @root = root task * @arg = unused * return: NULL error or !=NULL fetched task @@ -235,90 +515,117 @@ sched_hook_write(void *task, void *arg __unused) void * sched_hook_fetch(void *root, void *arg __unused) { - struct sched_IO *io; sched_root_task_t *r = root; - sched_task_t *task; - struct timeval now, m, mtmp; - struct timespec nw, *timeout; + sched_task_t *task, *tmp; + struct timespec now, m, mtmp; + struct timespec *timeout; struct kevent evt[1], res[KQ_EVENTS]; - register int i; + register int i, flg; int en; +#ifdef AIO_SUPPORT + int len, fd; + struct aiocb *acb; +#ifdef EVFILT_LIO + int l; + register int j; + off_t off; + struct aiocb **acbs; + struct iovec *iv; +#endif /* EVFILT_LIO */ +#endif /* AIO_SUPPORT */ - if (!r || !ROOT_DATA(r) || !ROOT_DATLEN(r)) + if (!r) return NULL; /* get new task by queue priority */ -retry: while ((task = TAILQ_FIRST(&r->root_event))) { +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskEVENT]); +#endif TAILQ_REMOVE(&r->root_event, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskEVENT]); +#endif task->task_type = taskUNUSE; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskUNUSE]); +#endif TAILQ_INSERT_TAIL(&r->root_unuse, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskUNUSE]); +#endif return task; } while ((task = TAILQ_FIRST(&r->root_ready))) { +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif TAILQ_REMOVE(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif task->task_type = taskUNUSE; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskUNUSE]); +#endif TAILQ_INSERT_TAIL(&r->root_unuse, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskUNUSE]); +#endif return task; } #ifdef TIMER_WITHOUT_SORT - clock_gettime(CLOCK_MONOTONIC, &nw); - now.tv_sec = nw.tv_sec; - now.tv_usec = nw.tv_nsec / 1000; + clock_gettime(CLOCK_MONOTONIC, &now); - timerclear(&r->root_wait); + sched_timespecclear(&r->root_wait); TAILQ_FOREACH(task, &r->root_timer, task_node) { - if (!timerisset(&r->root_wait)) - r->root_wait = TASK_TV(task); - else if (timercmp(&TASK_TV(task), &r->root_wait, -) < 0) - r->root_wait = TASK_TV(task); + if (!sched_timespecisset(&r->root_wait)) + r->root_wait = TASK_TS(task); + else if (sched_timespeccmp(&TASK_TS(task), &r->root_wait, -) < 0) + r->root_wait = TASK_TS(task); } if (TAILQ_FIRST(&r->root_timer)) { m = r->root_wait; - timersub(&m, &now, &mtmp); + sched_timespecsub(&m, &now, &mtmp); r->root_wait = mtmp; } else { /* set wait INFTIM */ - r->root_wait.tv_sec = r->root_wait.tv_usec = -1; + sched_timespecinf(&r->root_wait); } #else - if (!TAILQ_FIRST(&r->root_eventlo) && (task = TAILQ_FIRST(&r->root_timer))) { - clock_gettime(CLOCK_MONOTONIC, &nw); - now.tv_sec = nw.tv_sec; - now.tv_usec = nw.tv_nsec / 1000; + if (!TAILQ_FIRST(&r->root_task) && (task = TAILQ_FIRST(&r->root_timer))) { + clock_gettime(CLOCK_MONOTONIC, &now); - m = TASK_TV(task); - timersub(&m, &now, &mtmp); + m = TASK_TS(task); + sched_timespecsub(&m, &now, &mtmp); r->root_wait = mtmp; } else { /* set wait INFTIM */ - r->root_wait.tv_sec = r->root_wait.tv_usec = -1; + sched_timespecinf(&r->root_wait); } #endif - /* if present member of eventLo, set NOWAIT */ - if (TAILQ_FIRST(&r->root_eventlo)) - timerclear(&r->root_wait); + /* if present member of task, set NOWAIT */ + if (TAILQ_FIRST(&r->root_task)) + sched_timespecclear(&r->root_wait); - if (r->root_wait.tv_sec != -1 && r->root_wait.tv_usec != -1) { - nw.tv_sec = r->root_wait.tv_sec; - nw.tv_nsec = r->root_wait.tv_usec * 1000; - timeout = &nw; - } else /* wait INFTIM */ + if (r->root_wait.tv_sec != -1 && r->root_wait.tv_nsec != -1) + timeout = &r->root_wait; + else if (sched_timespecisinf(&r->root_poll)) timeout = NULL; + else + timeout = &r->root_poll; if ((en = kevent(r->root_kq, NULL, 0, res, KQ_EVENTS, timeout)) == -1) { - LOGERR; -#ifdef NDEBUG - /* kevent no exit by error, if non-debug version */ - goto retry; -#else - /* diagnostic exit from scheduler if kevent error occur */ + if (r->root_hooks.hook_exec.exception) { + if (r->root_hooks.hook_exec.exception(r, NULL)) + return NULL; + } else if (errno != EINTR) + LOGERR; return NULL; -#endif } - nw.tv_sec = nw.tv_nsec = 0; + now.tv_sec = now.tv_nsec = 0; /* Go and catch the cat into pipes ... */ for (i = 0; i < en; i++) { memcpy(evt, &res[i], sizeof evt); @@ -326,79 +633,487 @@ retry: /* Put read/write task to ready queue */ switch (res[i].filter) { case EVFILT_READ: - TAILQ_FOREACH(task, &r->root_read, task_node) { - if (TASK_FD(task) != ((int) res[i].udata)) + flg = 0; + TAILQ_FOREACH_SAFE(task, &r->root_read, task_node, tmp) { + if (TASK_FD(task) != ((intptr_t) res[i].udata)) continue; + else { + flg++; + TASK_RET(task) = res[i].data; + TASK_FLAG(task) = res[i].fflags; + } /* remove read handle */ - io = ROOT_DATA(task->task_root); - FD_CLR(TASK_FD(task), &io->rfd); - +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREAD]); +#endif TAILQ_REMOVE(&r->root_read, task, task_node); -/* if (res[i].flags & EV_EOF) { - task->task_type = taskUNUSE; - TAILQ_INSERT_TAIL(&r->root_unuse, task, task_node); - } else { */ +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREAD]); +#endif + if (r->root_hooks.hook_exec.exception && res[i].flags & EV_EOF) { + if (r->root_hooks.hook_exec.exception(r, (void*) EV_EOF)) { + task->task_type = taskUNUSE; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskUNUSE]); +#endif + TAILQ_INSERT_TAIL(&r->root_unuse, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskUNUSE]); +#endif + } else { + task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif + TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + } + } else { task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); -/* } */ - break; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + } } + /* if match at least 2, don't remove resouce of event */ + if (flg > 1) + evt->flags ^= evt->flags; break; case EVFILT_WRITE: - TAILQ_FOREACH(task, &r->root_write, task_node) { - if (TASK_FD(task) != ((int) res[i].udata)) + flg = 0; + TAILQ_FOREACH_SAFE(task, &r->root_write, task_node, tmp) { + if (TASK_FD(task) != ((intptr_t) res[i].udata)) continue; + else { + flg++; + TASK_RET(task) = res[i].data; + TASK_FLAG(task) = res[i].fflags; + } /* remove write handle */ - io = ROOT_DATA(task->task_root); - FD_CLR(TASK_FD(task), &io->wfd); - +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskWRITE]); +#endif TAILQ_REMOVE(&r->root_write, task, task_node); -/* if (res[i].flags & EV_EOF) { - task->task_type = taskUNUSE; - TAILQ_INSERT_TAIL(&r->root_unuse, task, task_node); - } else { */ +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskWRITE]); +#endif + if (r->root_hooks.hook_exec.exception && res[i].flags & EV_EOF) { + if (r->root_hooks.hook_exec.exception(r, (void*) EV_EOF)) { + task->task_type = taskUNUSE; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskUNUSE]); +#endif + TAILQ_INSERT_TAIL(&r->root_unuse, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskUNUSE]); +#endif + } else { + task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif + TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + } + } else { task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); -/* } */ - break; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + } } + /* if match at least 2, don't remove resouce of event */ + if (flg > 1) + evt->flags ^= evt->flags; break; + case EVFILT_TIMER: + flg = 0; + TAILQ_FOREACH_SAFE(task, &r->root_alarm, task_node, tmp) { + if ((uintptr_t) TASK_DATA(task) != ((uintptr_t) res[i].udata)) + continue; + else { + flg++; + TASK_RET(task) = res[i].data; + TASK_FLAG(task) = res[i].fflags; + } + /* remove alarm handle */ +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskALARM]); +#endif + TAILQ_REMOVE(&r->root_alarm, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskALARM]); +#endif + task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif + TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + } + /* if match at least 2, don't remove resouce of event */ + if (flg > 1) + evt->flags ^= evt->flags; + break; + case EVFILT_VNODE: + flg = 0; + TAILQ_FOREACH_SAFE(task, &r->root_node, task_node, tmp) { + if (TASK_FD(task) != ((intptr_t) res[i].udata)) + continue; + else { + flg++; + TASK_RET(task) = res[i].data; + TASK_FLAG(task) = res[i].fflags; + } + /* remove node handle */ +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskNODE]); +#endif + TAILQ_REMOVE(&r->root_node, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskNODE]); +#endif + task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif + TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + } + /* if match at least 2, don't remove resouce of event */ + if (flg > 1) + evt->flags ^= evt->flags; + break; + case EVFILT_PROC: + flg = 0; + TAILQ_FOREACH_SAFE(task, &r->root_proc, task_node, tmp) { + if (TASK_VAL(task) != ((uintptr_t) res[i].udata)) + continue; + else { + flg++; + TASK_RET(task) = res[i].data; + TASK_FLAG(task) = res[i].fflags; + } + /* remove proc handle */ +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskPROC]); +#endif + TAILQ_REMOVE(&r->root_proc, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskPROC]); +#endif + task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif + TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + } + /* if match at least 2, don't remove resouce of event */ + if (flg > 1) + evt->flags ^= evt->flags; + break; + case EVFILT_SIGNAL: + flg = 0; + TAILQ_FOREACH_SAFE(task, &r->root_signal, task_node, tmp) { + if (TASK_VAL(task) != ((uintptr_t) res[i].udata)) + continue; + else { + flg++; + TASK_RET(task) = res[i].data; + TASK_FLAG(task) = res[i].fflags; + } + /* remove signal handle */ +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskSIGNAL]); +#endif + TAILQ_REMOVE(&r->root_signal, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskSIGNAL]); +#endif + task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif + TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + } + /* if match at least 2, don't remove resouce of event */ + if (flg > 1) + evt->flags ^= evt->flags; + break; +#ifdef AIO_SUPPORT + case EVFILT_AIO: + flg = 0; + TAILQ_FOREACH_SAFE(task, &r->root_aio, task_node, tmp) { + acb = (struct aiocb*) TASK_VAL(task); + if (acb != ((struct aiocb*) res[i].udata)) + continue; + else { + flg++; + TASK_RET(task) = res[i].data; + TASK_FLAG(task) = res[i].fflags; + } + /* remove user handle */ +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskAIO]); +#endif + TAILQ_REMOVE(&r->root_aio, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskAIO]); +#endif + task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif + TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + fd = acb->aio_fildes; + if ((len = aio_return(acb)) != -1) { + if (lseek(fd, acb->aio_offset + len, SEEK_CUR) == -1) + LOGERR; + } else + LOGERR; + free(acb); + TASK_DATLEN(task) = (u_long) len; + TASK_FD(task) = fd; + } + /* if match at least 2, don't remove resouce of event */ + if (flg > 1) + evt->flags ^= evt->flags; + break; +#ifdef EVFILT_LIO + case EVFILT_LIO: + flg = 0; + TAILQ_FOREACH_SAFE(task, &r->root_lio, task_node, tmp) { + acbs = (struct aiocb**) TASK_VAL(task); + if (acbs != ((struct aiocb**) res[i].udata)) + continue; + else { + flg++; + TASK_RET(task) = res[i].data; + TASK_FLAG(task) = res[i].fflags; + } + /* remove user handle */ +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskLIO]); +#endif + TAILQ_REMOVE(&r->root_lio, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskLIO]); +#endif + task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif + TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + iv = (struct iovec*) TASK_DATA(task); + fd = acbs[0]->aio_fildes; + off = acbs[0]->aio_offset; + for (j = len = 0; i < TASK_DATLEN(task); len += l, i++) { + if ((iv[i].iov_len = aio_return(acbs[i])) == -1) + l = 0; + else + l = iv[i].iov_len; + free(acbs[i]); + } + free(acbs); + TASK_DATLEN(task) = (u_long) len; + TASK_FD(task) = fd; + + if (lseek(fd, off + len, SEEK_CUR) == -1) + LOGERR; + } + /* if match at least 2, don't remove resouce of event */ + if (flg > 1) + evt->flags ^= evt->flags; + break; +#endif /* EVFILT_LIO */ +#endif /* AIO_SUPPORT */ +#ifdef EVFILT_USER + case EVFILT_USER: + flg = 0; + TAILQ_FOREACH_SAFE(task, &r->root_user, task_node, tmp) { + if (TASK_VAL(task) != ((uintptr_t) res[i].udata)) + continue; + else { + flg++; + TASK_RET(task) = res[i].data; + TASK_FLAG(task) = res[i].fflags; + } + /* remove user handle */ +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskUSER]); +#endif + TAILQ_REMOVE(&r->root_user, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskUSER]); +#endif + task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif + TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif + } + /* if match at least 2, don't remove resouce of event */ + if (flg > 1) + evt->flags ^= evt->flags; + break; +#endif /* EVFILT_USER */ } - if (kevent(r->root_kq, evt, 1, NULL, 0, &nw) == -1) - LOGERR; + if (kevent(r->root_kq, evt, 1, NULL, 0, &now) == -1) { + if (r->root_hooks.hook_exec.exception) { + if (r->root_hooks.hook_exec.exception(r, NULL)) + return NULL; + } else + LOGERR; + } } /* timer update & put in ready queue */ - clock_gettime(CLOCK_MONOTONIC, &nw); - now.tv_sec = nw.tv_sec; - now.tv_usec = nw.tv_nsec / 1000; + clock_gettime(CLOCK_MONOTONIC, &now); - TAILQ_FOREACH(task, &r->root_timer, task_node) - if (timercmp(&now, &TASK_TV(task), -) >= 0) { + TAILQ_FOREACH_SAFE(task, &r->root_timer, task_node, tmp) + if (sched_timespeccmp(&now, &TASK_TS(task), -) >= 0) { +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskTIMER]); +#endif TAILQ_REMOVE(&r->root_timer, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskTIMER]); +#endif task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif } - /* put eventlo priority task to ready queue, if there is no ready task or - reach max missed fetch-rotate */ - if ((task = TAILQ_FIRST(&r->root_eventlo))) { - if (!TAILQ_FIRST(&r->root_ready) || r->root_eventlo_miss > MAX_EVENTLO_MISS) { - r->root_eventlo_miss = 0; + /* put regular task priority task to ready queue, + if there is no ready task or reach max missing hit for regular task */ + if ((task = TAILQ_FIRST(&r->root_task))) { + if (!TAILQ_FIRST(&r->root_ready) || r->root_miss >= TASK_VAL(task)) { + r->root_miss ^= r->root_miss; - TAILQ_REMOVE(&r->root_eventlo, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskTASK]); +#endif + TAILQ_REMOVE(&r->root_task, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskTASK]); +#endif task->task_type = taskREADY; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif TAILQ_INSERT_TAIL(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif } else - r->root_eventlo_miss++; + r->root_miss++; } else - r->root_eventlo_miss = 0; + r->root_miss ^= r->root_miss; /* OK, lets get ready task !!! */ - if (!(task = TAILQ_FIRST(&r->root_ready))) - goto retry; + task = TAILQ_FIRST(&r->root_ready); + if (!(task)) + return NULL; + +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskREADY]); +#endif TAILQ_REMOVE(&r->root_ready, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskREADY]); +#endif task->task_type = taskUNUSE; +#ifdef HAVE_LIBPTHREAD + pthread_mutex_lock(&r->root_mtx[taskUNUSE]); +#endif TAILQ_INSERT_TAIL(&r->root_unuse, task, task_node); +#ifdef HAVE_LIBPTHREAD + pthread_mutex_unlock(&r->root_mtx[taskUNUSE]); +#endif return task; +} + +/* + * sched_hook_exception() - Default EXCEPTION hook + * + * @root = root task + * @arg = custom handling: if arg == EV_EOF or other value; default: arg == NULL log errno + * return: <0 errors and 0 ok + */ +void * +sched_hook_exception(void *root, void *arg) +{ + sched_root_task_t *r = root; + + if (!r) + return NULL; + + /* custom exception handling ... */ + if (arg) { + if (arg == (void*) EV_EOF) + return NULL; + return (void*) -1; /* raise scheduler error!!! */ + } + + /* if error hook exists */ + if (r->root_hooks.hook_root.error) + return (r->root_hooks.hook_root.error(root, (void*) ((intptr_t) errno))); + + /* default case! */ + LOGERR; + return NULL; +} + +/* + * sched_hook_condition() - Default CONDITION hook + * + * @root = root task + * @arg = killState from schedRun() + * return: NULL kill scheduler loop or !=NULL ok + */ +void * +sched_hook_condition(void *root, void *arg) +{ + sched_root_task_t *r = root; + + if (!r) + return NULL; + + return (void*) (r->root_cond - *(intptr_t*) arg); }