18  * Support for the signalfd facility, a Linux-borne facility for
19  * file descriptor-based synchronous signal consumption.
54  * arriving signals should wake only process-local pollers.  Additionally,
59  * poll machinery can act upon them without risk of use-after-free.  When a
81  * Poller entries in sigfd_proc_state_t`sigfd_list are cleaned up under two
91  * fork-shared signalfd descriptors in conjuction with fork-shared caching poll
93  * wake-ups.  This is caused by the pollhead identity of signalfd descriptors
95  * thread-local cache, poll(2) is unaffected by this limitation.
108  * Process exit, cleaning up signalfd pollers:
140 /* Per-instance signalfd device state: */
178 	kcondvar_t	sp_cv;		/* CV for cleaning up */
193 	sigfd_proc_state_t *pstate = p->p_sigfd;  in signalfd_proc_clean()
195 	ASSERT(MUTEX_HELD(&p->p_lock));  in signalfd_proc_clean()
197 	VERIFY(list_is_empty(&pstate->sigfd_list));  in signalfd_proc_clean()
199 	p->p_sigfd = NULL;  in signalfd_proc_clean()
200 	list_destroy(&pstate->sigfd_list);  in signalfd_proc_clean()
209 	mutex_enter(&sp->sp_lock);  in signalfd_wake_task()
210 	VERIFY(sp->sp_pollev != 0);  in signalfd_wake_task()
211 	VERIFY(sp->sp_pending);  in signalfd_wake_task()
213 		const short pollev = sp->sp_pollev;  in signalfd_wake_task()
215 		sp->sp_pollev = 0;  in signalfd_wake_task()
216 		mutex_exit(&sp->sp_lock);  in signalfd_wake_task()
224 		pollwakeup(&sp->sp_pollhead, pollev);  in signalfd_wake_task()
226 			pollhead_clean(&sp->sp_pollhead);  in signalfd_wake_task()
229 		mutex_enter(&sp->sp_lock);  in signalfd_wake_task()
234 		 * underlying resource is undergoing clean-up.  in signalfd_wake_task()
239 	} while (sp->sp_pollev != 0);  in signalfd_wake_task()
242 	 * Indicate that wake task processing is complete.  in signalfd_wake_task()
244 	 * Wake any thread waiting for event delivery to complete if this poller  in signalfd_wake_task()
247 	sp->sp_pending = false;  in signalfd_wake_task()
248 	cv_signal(&sp->sp_cv);  in signalfd_wake_task()
249 	mutex_exit(&sp->sp_lock);  in signalfd_wake_task()
255 	ASSERT(MUTEX_HELD(&sp->sp_lock));  in signalfd_poller_wake()
257 	sp->sp_pollev |= ev;  in signalfd_poller_wake()
258 	if (!sp->sp_pending) {  in signalfd_poller_wake()
259 		sp->sp_pending = true;  in signalfd_poller_wake()
261 		    &sp->sp_taskent);  in signalfd_poller_wake()
273 	sigfd_proc_state_t *pstate = (sigfd_proc_state_t *)p->p_sigfd;  in signalfd_pollwake_cb()
275 	ASSERT(MUTEX_HELD(&p->p_lock));  in signalfd_pollwake_cb()
278 	list_t *pollers = &pstate->sigfd_list;  in signalfd_pollwake_cb()
281 		mutex_enter(&sp->sp_lock);  in signalfd_pollwake_cb()
282 		if (sigismember(&sp->sp_mask, sig)) {  in signalfd_pollwake_cb()
285 		mutex_exit(&sp->sp_lock);  in signalfd_pollwake_cb()
298 	ASSERT(MUTEX_HELD(&p->p_lock));  in signalfd_proc_pstate()
300 	sigfd_proc_state_t *pstate = p->p_sigfd;  in signalfd_proc_pstate()
302 		mutex_exit(&p->p_lock);  in signalfd_proc_pstate()
304 		list_create(&pstate->sigfd_list,  in signalfd_proc_pstate()
307 		pstate->sigfd_pollwake_cb = signalfd_pollwake_cb;  in signalfd_proc_pstate()
310 		mutex_enter(&p->p_lock);  in signalfd_proc_pstate()
311 		if (p->p_sigfd == NULL) {  in signalfd_proc_pstate()
312 			p->p_sigfd = pstate;  in signalfd_proc_pstate()
315 			list_destroy(&pstate->sigfd_list);  in signalfd_proc_pstate()
317 			pstate = p->p_sigfd;  in signalfd_proc_pstate()
331 	ASSERT(MUTEX_HELD(&state->sfd_lock));  in signalfd_poller_associate()
333 	mutex_enter(&p->p_lock);  in signalfd_poller_associate()
336 	pollers = &pstate->sigfd_list;  in signalfd_poller_associate()
343 		if (sp->sp_state == state) {  in signalfd_poller_associate()
344 			mutex_exit(&p->p_lock);  in signalfd_poller_associate()
354 	mutex_exit(&p->p_lock);  in signalfd_poller_associate()
357 	mutex_init(&sp->sp_lock, NULL, MUTEX_DEFAULT, NULL);  in signalfd_poller_associate()
358 	cv_init(&sp->sp_cv, NULL, CV_DEFAULT, NULL);  in signalfd_poller_associate()
359 	sigorset(&sp->sp_mask, &state->sfd_mask);  in signalfd_poller_associate()
360 	sp->sp_state = state;  in signalfd_poller_associate()
361 	sp->sp_proc = p;  in signalfd_poller_associate()
363 	mutex_enter(&p->p_lock);  in signalfd_poller_associate()
370 	list_insert_tail(&pstate->sigfd_list, sp);  in signalfd_poller_associate()
371 	list_insert_tail(&state->sfd_pollers, sp);  in signalfd_poller_associate()
372 	mutex_exit(&p->p_lock);  in signalfd_poller_associate()
380 	ASSERT(MUTEX_HELD(&state->sfd_lock));  in signalfd_pollers_dissociate()
385 	list_t *pollers = &state->sfd_pollers;  in signalfd_pollers_dissociate()
387 		proc_t *p = sp->sp_proc;  in signalfd_pollers_dissociate()
394 		 * Even if the process in question is racing us to clean-up in  in signalfd_pollers_dissociate()
399 		mutex_enter(&p->p_lock);  in signalfd_pollers_dissociate()
400 		if (sp->sp_proc == NULL) {  in signalfd_pollers_dissociate()
401 			mutex_exit(&p->p_lock);  in signalfd_pollers_dissociate()
404 		VERIFY3P(sp->sp_proc, ==, p);  in signalfd_pollers_dissociate()
405 		VERIFY3P(sp->sp_state, ==, state);  in signalfd_pollers_dissociate()
406 		VERIFY3P(p->p_sigfd, !=, NULL);  in signalfd_pollers_dissociate()
408 		sigfd_proc_state_t *pstate = p->p_sigfd;  in signalfd_pollers_dissociate()
409 		list_remove(&pstate->sigfd_list, sp);  in signalfd_pollers_dissociate()
410 		sp->sp_proc = NULL;  in signalfd_pollers_dissociate()
412 		/* Wake any lingering pollers referencing the pollhead */  in signalfd_pollers_dissociate()
413 		mutex_enter(&sp->sp_lock);  in signalfd_pollers_dissociate()
415 		mutex_exit(&sp->sp_lock);  in signalfd_pollers_dissociate()
417 		if (list_is_empty(&pstate->sigfd_list)) {  in signalfd_pollers_dissociate()
420 			 * process, then clean up its state as well.  in signalfd_pollers_dissociate()
424 		mutex_exit(&p->p_lock);  in signalfd_pollers_dissociate()
432 	ASSERT(MUTEX_HELD(&state->sfd_lock));  in signalfd_pollers_free()
435 	while ((sp = list_remove_head(&state->sfd_pollers)) != NULL) {  in signalfd_pollers_free()
436 		ASSERT3P(sp->sp_proc, ==, NULL);  in signalfd_pollers_free()
438 		mutex_enter(&sp->sp_lock);  in signalfd_pollers_free()
439 		while (sp->sp_pending) {  in signalfd_pollers_free()
440 			cv_wait(&sp->sp_cv, &sp->sp_lock);  in signalfd_pollers_free()
446 		ASSERT3P(sp->sp_pollhead.ph_list, ==, NULL);  in signalfd_pollers_free()
448 		cv_destroy(&sp->sp_cv);  in signalfd_pollers_free()
449 		mutex_destroy(&sp->sp_lock);  in signalfd_pollers_free()
455  * Callback for cleaning up signalfd state from a process during proc_exit().
462 	mutex_enter(&p->p_lock);  in signalfd_exit_helper()
464 	sigfd_proc_state_t *pstate = p->p_sigfd;  in signalfd_exit_helper()
466 		mutex_exit(&p->p_lock);  in signalfd_exit_helper()
471 	while ((sp = list_remove_head(&pstate->sigfd_list)) != NULL) {  in signalfd_exit_helper()
476 		sp->sp_proc = NULL;  in signalfd_exit_helper()
478 		/* Wake any lingering pollers referencing the pollhead */  in signalfd_exit_helper()
479 		mutex_enter(&sp->sp_lock);  in signalfd_exit_helper()
481 		mutex_exit(&sp->sp_lock);  in signalfd_exit_helper()
484 	mutex_exit(&p->p_lock);  in signalfd_exit_helper()
496 	if (minor == -1) {  in signalfd_open()
506 	mutex_init(&state->sfd_lock, NULL, MUTEX_DEFAULT, NULL);  in signalfd_open()
507 	list_create(&state->sfd_pollers, sizeof (signalfd_poller_t),  in signalfd_open()
509 	state->sfd_minor = minor;  in signalfd_open()
537 	t->t_sigwait = set;  in signalfd_consume_signal()
539 	mutex_enter(&p->p_lock);  in signalfd_consume_signal()
543 	const k_sigset_t oldmask = t->t_hold;  in signalfd_consume_signal()
544 	sigdiffset(&t->t_hold, &t->t_sigwait);  in signalfd_consume_signal()
548 			ret = cv_waituntil_sig(&t->t_delay_cv, &p->p_lock,  in signalfd_consume_signal()
552 		mutex_exit(&p->p_lock);  in signalfd_consume_signal()
554 			ret = -1;  in signalfd_consume_signal()
556 		mutex_enter(&p->p_lock);  in signalfd_consume_signal()
563 	t->t_hold = oldmask;  in signalfd_consume_signal()
564 	t->t_sig_check = 1;  in signalfd_consume_signal()
566 	if (ret == -1) {  in signalfd_consume_signal()
568 		mutex_exit(&p->p_lock);  in signalfd_consume_signal()
569 		sigemptyset(&t->t_sigwait);  in signalfd_consume_signal()
574 	if (lwp->lwp_cursig == 0 ||  in signalfd_consume_signal()
575 	    !sigismember(&t->t_sigwait, lwp->lwp_cursig)) {  in signalfd_consume_signal()
581 		mutex_exit(&p->p_lock);  in signalfd_consume_signal()
582 		sigemptyset(&t->t_sigwait);  in signalfd_consume_signal()
589 	if (lwp->lwp_curinfo != NULL) {  in signalfd_consume_signal()
590 		k_siginfo_t *infop = &lwp->lwp_curinfo->sq_info;  in signalfd_consume_signal()
592 		ssi.ssi_signo	= infop->si_signo;  in signalfd_consume_signal()
593 		ssi.ssi_errno	= infop->si_errno;  in signalfd_consume_signal()
594 		ssi.ssi_code	= infop->si_code;  in signalfd_consume_signal()
595 		ssi.ssi_pid	= infop->si_pid;  in signalfd_consume_signal()
596 		ssi.ssi_uid	= infop->si_uid;  in signalfd_consume_signal()
597 		ssi.ssi_fd	= infop->si_fd;  in signalfd_consume_signal()
598 		ssi.ssi_band	= infop->si_band;  in signalfd_consume_signal()
599 		ssi.ssi_trapno	= infop->si_trapno;  in signalfd_consume_signal()
600 		ssi.ssi_status	= infop->si_status;  in signalfd_consume_signal()
601 		ssi.ssi_utime	= infop->si_utime;  in signalfd_consume_signal()
602 		ssi.ssi_stime	= infop->si_stime;  in signalfd_consume_signal()
603 		ssi.ssi_addr	= (uint64_t)(intptr_t)infop->si_addr;  in signalfd_consume_signal()
609 			.si_signo = lwp->lwp_cursig,  in signalfd_consume_signal()
619 	lwp->lwp_ru.nsignals++;  in signalfd_consume_signal()
620 	lwp->lwp_cursig = 0;  in signalfd_consume_signal()
621 	lwp->lwp_extsig = 0;  in signalfd_consume_signal()
622 	if (lwp->lwp_curinfo != NULL) {  in signalfd_consume_signal()
623 		siginfofree(lwp->lwp_curinfo);  in signalfd_consume_signal()
624 		lwp->lwp_curinfo = NULL;  in signalfd_consume_signal()
626 	mutex_exit(&p->p_lock);  in signalfd_consume_signal()
629 	sigemptyset(&t->t_sigwait);  in signalfd_consume_signal()
652 	if (uio->uio_resid < sizeof (signalfd_siginfo_t)) {  in signalfd_read()
656 	if (uio->uio_fmode & (FNDELAY|FNONBLOCK)) {  in signalfd_read()
660 	mutex_enter(&state->sfd_lock);  in signalfd_read()
661 	set = state->sfd_mask;  in signalfd_read()
662 	mutex_exit(&state->sfd_lock);  in signalfd_read()
682 			mutex_enter(&state->sfd_lock);  in signalfd_read()
683 			set = state->sfd_mask;  in signalfd_read()
684 			mutex_exit(&state->sfd_lock);  in signalfd_read()
688 	} while (res == 0 && uio->uio_resid >= sizeof (signalfd_siginfo_t));  in signalfd_read()
698  *      return (((p->p_sig | t->t_sig) & set) & fillset);
703 	return (((p->p_sig.__sigbits[0] | t->t_sig.__sigbits[0]) &  in signalfd_sig_pending()
705 	    ((p->p_sig.__sigbits[1] | t->t_sig.__sigbits[1]) &  in signalfd_sig_pending()
707 	    (((p->p_sig.__sigbits[2] | t->t_sig.__sigbits[2]) &  in signalfd_sig_pending()
725 	mutex_enter(&state->sfd_lock);  in signalfd_poll()
726 	if (signalfd_sig_pending(p, t, state->sfd_mask) != 0) {  in signalfd_poll()
735 		*phpp = &sp->sp_pollhead;  in signalfd_poll()
737 	mutex_exit(&state->sfd_lock);  in signalfd_poll()
749 	mutex_enter(&state->sfd_lock);  in signalfd_set_mask()
750 	state->sfd_mask = kmask;  in signalfd_set_mask()
751 	list_t *pollers = &state->sfd_pollers;  in signalfd_set_mask()
754 		mutex_enter(&sp->sp_lock);  in signalfd_set_mask()
755 		sp->sp_mask = kmask;  in signalfd_set_mask()
756 		mutex_exit(&sp->sp_lock);  in signalfd_set_mask()
758 	mutex_exit(&state->sfd_lock);  in signalfd_set_mask()
804 	mutex_enter(&state->sfd_lock);  in signalfd_close()
809 	/* ... and free all those (now-dissociated) pollers */  in signalfd_close()
811 	ASSERT(list_is_empty(&state->sfd_pollers));  in signalfd_close()
813 	mutex_destroy(&state->sfd_lock);  in signalfd_close()