xref: /titanic_52/usr/src/cmd/svc/startd/restarter.c (revision c238c833c812c2845a9e35d4599c8d2a85f73ba9)
17c478bd9Sstevel@tonic-gate /*
27c478bd9Sstevel@tonic-gate  * CDDL HEADER START
37c478bd9Sstevel@tonic-gate  *
47c478bd9Sstevel@tonic-gate  * The contents of this file are subject to the terms of the
521d7f835Sgm149974  * Common Development and Distribution License (the "License").
621d7f835Sgm149974  * You may not use this file except in compliance with the License.
77c478bd9Sstevel@tonic-gate  *
87c478bd9Sstevel@tonic-gate  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
97c478bd9Sstevel@tonic-gate  * or http://www.opensolaris.org/os/licensing.
107c478bd9Sstevel@tonic-gate  * See the License for the specific language governing permissions
117c478bd9Sstevel@tonic-gate  * and limitations under the License.
127c478bd9Sstevel@tonic-gate  *
137c478bd9Sstevel@tonic-gate  * When distributing Covered Code, include this CDDL HEADER in each
147c478bd9Sstevel@tonic-gate  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
157c478bd9Sstevel@tonic-gate  * If applicable, add the following below this CDDL HEADER, with the
167c478bd9Sstevel@tonic-gate  * fields enclosed by brackets "[]" replaced with your own identifying
177c478bd9Sstevel@tonic-gate  * information: Portions Copyright [yyyy] [name of copyright owner]
187c478bd9Sstevel@tonic-gate  *
197c478bd9Sstevel@tonic-gate  * CDDL HEADER END
207c478bd9Sstevel@tonic-gate  */
2153f3aea0SRoger A. Faulkner 
227c478bd9Sstevel@tonic-gate /*
2353f3aea0SRoger A. Faulkner  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
247c478bd9Sstevel@tonic-gate  * Use is subject to license terms.
257c478bd9Sstevel@tonic-gate  */
267c478bd9Sstevel@tonic-gate 
277c478bd9Sstevel@tonic-gate /*
287c478bd9Sstevel@tonic-gate  * restarter.c - service manipulation
297c478bd9Sstevel@tonic-gate  *
307c478bd9Sstevel@tonic-gate  * This component manages services whose restarter is svc.startd, the standard
317c478bd9Sstevel@tonic-gate  * restarter.  It translates restarter protocol events from the graph engine
327c478bd9Sstevel@tonic-gate  * into actions on processes, as a delegated restarter would do.
337c478bd9Sstevel@tonic-gate  *
347c478bd9Sstevel@tonic-gate  * The master restarter manages a number of always-running threads:
357c478bd9Sstevel@tonic-gate  *   - restarter event thread: events from the graph engine
367c478bd9Sstevel@tonic-gate  *   - timeout thread: thread to fire queued timeouts
377c478bd9Sstevel@tonic-gate  *   - contract thread: thread to handle contract events
387c478bd9Sstevel@tonic-gate  *   - wait thread: thread to handle wait-based services
397c478bd9Sstevel@tonic-gate  *
407c478bd9Sstevel@tonic-gate  * The other threads are created as-needed:
417c478bd9Sstevel@tonic-gate  *   - per-instance method threads
427c478bd9Sstevel@tonic-gate  *   - per-instance event processing threads
437c478bd9Sstevel@tonic-gate  *
447c478bd9Sstevel@tonic-gate  * The interaction of all threads must result in the following conditions
457c478bd9Sstevel@tonic-gate  * being satisfied (on a per-instance basis):
467c478bd9Sstevel@tonic-gate  *   - restarter events must be processed in order
477c478bd9Sstevel@tonic-gate  *   - method execution must be serialized
487c478bd9Sstevel@tonic-gate  *   - instance delete must be held until outstanding methods are complete
497c478bd9Sstevel@tonic-gate  *   - contract events shouldn't be processed while a method is running
507c478bd9Sstevel@tonic-gate  *   - timeouts should fire even when a method is running
517c478bd9Sstevel@tonic-gate  *
527c478bd9Sstevel@tonic-gate  * Service instances are represented by restarter_inst_t's and are kept in the
537c478bd9Sstevel@tonic-gate  * instance_list list.
547c478bd9Sstevel@tonic-gate  *
557c478bd9Sstevel@tonic-gate  * Service States
567c478bd9Sstevel@tonic-gate  *   The current state of a service instance is kept in
577c478bd9Sstevel@tonic-gate  *   restarter_inst_t->ri_i.i_state.  If transition to a new state could take
587c478bd9Sstevel@tonic-gate  *   some time, then before we effect the transition we set
597c478bd9Sstevel@tonic-gate  *   restarter_inst_t->ri_i.i_next_state to the target state, and afterwards we
607c478bd9Sstevel@tonic-gate  *   rotate i_next_state to i_state and set i_next_state to
617c478bd9Sstevel@tonic-gate  *   RESTARTER_STATE_NONE.  So usually i_next_state is _NONE when ri_lock is not
627c478bd9Sstevel@tonic-gate  *   held.  The exception is when we launch methods, which are done with
637c478bd9Sstevel@tonic-gate  *   a separate thread.  To keep any other threads from grabbing ri_lock before
647c478bd9Sstevel@tonic-gate  *   method_thread() does, we set ri_method_thread to the thread id of the
657c478bd9Sstevel@tonic-gate  *   method thread, and when it is nonzero any thread with a different thread id
667c478bd9Sstevel@tonic-gate  *   waits on ri_method_cv.
677c478bd9Sstevel@tonic-gate  *
687c478bd9Sstevel@tonic-gate  * Method execution is serialized by blocking on ri_method_cv in
697c478bd9Sstevel@tonic-gate  * inst_lookup_by_id() and waiting for a 0 value of ri_method_thread.  This
707c478bd9Sstevel@tonic-gate  * also prevents the instance structure from being deleted until all
717c478bd9Sstevel@tonic-gate  * outstanding operations such as method_thread() have finished.
727c478bd9Sstevel@tonic-gate  *
737c478bd9Sstevel@tonic-gate  * Lock ordering:
747c478bd9Sstevel@tonic-gate  *
757c478bd9Sstevel@tonic-gate  * dgraph_lock [can be held when taking:]
767c478bd9Sstevel@tonic-gate  *   utmpx_lock
777c478bd9Sstevel@tonic-gate  *   dictionary->dict_lock
787c478bd9Sstevel@tonic-gate  *   st->st_load_lock
797c478bd9Sstevel@tonic-gate  *   wait_info_lock
807c478bd9Sstevel@tonic-gate  *   ru->restarter_update_lock
817c478bd9Sstevel@tonic-gate  *     restarter_queue->rpeq_lock
827c478bd9Sstevel@tonic-gate  *   instance_list.ril_lock
837c478bd9Sstevel@tonic-gate  *     inst->ri_lock
847c478bd9Sstevel@tonic-gate  *   st->st_configd_live_lock
857c478bd9Sstevel@tonic-gate  *
867c478bd9Sstevel@tonic-gate  * instance_list.ril_lock
877c478bd9Sstevel@tonic-gate  *   graph_queue->gpeq_lock
887c478bd9Sstevel@tonic-gate  *   gu->gu_lock
897c478bd9Sstevel@tonic-gate  *   st->st_configd_live_lock
907c478bd9Sstevel@tonic-gate  *   dictionary->dict_lock
917c478bd9Sstevel@tonic-gate  *   inst->ri_lock
927c478bd9Sstevel@tonic-gate  *     graph_queue->gpeq_lock
937c478bd9Sstevel@tonic-gate  *     gu->gu_lock
947c478bd9Sstevel@tonic-gate  *     tu->tu_lock
957c478bd9Sstevel@tonic-gate  *     tq->tq_lock
967c478bd9Sstevel@tonic-gate  *     inst->ri_queue_lock
977c478bd9Sstevel@tonic-gate  *       wait_info_lock
987c478bd9Sstevel@tonic-gate  *       bp->cb_lock
997c478bd9Sstevel@tonic-gate  *     utmpx_lock
1007c478bd9Sstevel@tonic-gate  *
1017c478bd9Sstevel@tonic-gate  * single_user_thread_lock
1027c478bd9Sstevel@tonic-gate  *   wait_info_lock
1037c478bd9Sstevel@tonic-gate  *   utmpx_lock
1047c478bd9Sstevel@tonic-gate  *
1057c478bd9Sstevel@tonic-gate  * gu_freeze_lock
1067c478bd9Sstevel@tonic-gate  *
1077c478bd9Sstevel@tonic-gate  * logbuf_mutex nests inside pretty much everything.
1087c478bd9Sstevel@tonic-gate  */
1097c478bd9Sstevel@tonic-gate 
1107c478bd9Sstevel@tonic-gate #include <sys/contract/process.h>
1117c478bd9Sstevel@tonic-gate #include <sys/ctfs.h>
1127c478bd9Sstevel@tonic-gate #include <sys/stat.h>
1137c478bd9Sstevel@tonic-gate #include <sys/time.h>
1147c478bd9Sstevel@tonic-gate #include <sys/types.h>
1157c478bd9Sstevel@tonic-gate #include <sys/uio.h>
1167c478bd9Sstevel@tonic-gate #include <sys/wait.h>
1177c478bd9Sstevel@tonic-gate #include <assert.h>
1187c478bd9Sstevel@tonic-gate #include <errno.h>
1197c478bd9Sstevel@tonic-gate #include <fcntl.h>
1207c478bd9Sstevel@tonic-gate #include <libcontract.h>
1217c478bd9Sstevel@tonic-gate #include <libcontract_priv.h>
1227c478bd9Sstevel@tonic-gate #include <libintl.h>
1237c478bd9Sstevel@tonic-gate #include <librestart.h>
1247c478bd9Sstevel@tonic-gate #include <librestart_priv.h>
1257c478bd9Sstevel@tonic-gate #include <libuutil.h>
1267c478bd9Sstevel@tonic-gate #include <limits.h>
1277c478bd9Sstevel@tonic-gate #include <poll.h>
1287c478bd9Sstevel@tonic-gate #include <port.h>
1297c478bd9Sstevel@tonic-gate #include <pthread.h>
1307c478bd9Sstevel@tonic-gate #include <stdarg.h>
1317c478bd9Sstevel@tonic-gate #include <stdio.h>
1327c478bd9Sstevel@tonic-gate #include <strings.h>
1337c478bd9Sstevel@tonic-gate #include <unistd.h>
1347c478bd9Sstevel@tonic-gate 
1357c478bd9Sstevel@tonic-gate #include "startd.h"
1367c478bd9Sstevel@tonic-gate #include "protocol.h"
1377c478bd9Sstevel@tonic-gate 
1387c478bd9Sstevel@tonic-gate static uu_list_pool_t *restarter_instance_pool;
1397c478bd9Sstevel@tonic-gate static restarter_instance_list_t instance_list;
1407c478bd9Sstevel@tonic-gate 
1417c478bd9Sstevel@tonic-gate static uu_list_pool_t *restarter_queue_pool;
1427c478bd9Sstevel@tonic-gate 
14316ba0facSSean Wilcox /*
14416ba0facSSean Wilcox  * Function used to reset the restart times for an instance, when
14516ba0facSSean Wilcox  * an administrative task comes along and essentially makes the times
14616ba0facSSean Wilcox  * in this array ineffective.
14716ba0facSSean Wilcox  */
14816ba0facSSean Wilcox static void
14916ba0facSSean Wilcox reset_start_times(restarter_inst_t *inst)
15016ba0facSSean Wilcox {
15116ba0facSSean Wilcox 	inst->ri_start_index = 0;
15216ba0facSSean Wilcox 	bzero(inst->ri_start_time, sizeof (inst->ri_start_time));
15316ba0facSSean Wilcox }
15416ba0facSSean Wilcox 
1557c478bd9Sstevel@tonic-gate /*ARGSUSED*/
1567c478bd9Sstevel@tonic-gate static int
1577c478bd9Sstevel@tonic-gate restarter_instance_compare(const void *lc_arg, const void *rc_arg,
1587c478bd9Sstevel@tonic-gate     void *private)
1597c478bd9Sstevel@tonic-gate {
1607c478bd9Sstevel@tonic-gate 	int lc_id = ((const restarter_inst_t *)lc_arg)->ri_id;
1617c478bd9Sstevel@tonic-gate 	int rc_id = *(int *)rc_arg;
1627c478bd9Sstevel@tonic-gate 
1637c478bd9Sstevel@tonic-gate 	if (lc_id > rc_id)
1647c478bd9Sstevel@tonic-gate 		return (1);
1657c478bd9Sstevel@tonic-gate 	if (lc_id < rc_id)
1667c478bd9Sstevel@tonic-gate 		return (-1);
1677c478bd9Sstevel@tonic-gate 	return (0);
1687c478bd9Sstevel@tonic-gate }
1697c478bd9Sstevel@tonic-gate 
1707c478bd9Sstevel@tonic-gate static restarter_inst_t *
1717c478bd9Sstevel@tonic-gate inst_lookup_by_name(const char *name)
1727c478bd9Sstevel@tonic-gate {
1737c478bd9Sstevel@tonic-gate 	int id;
1747c478bd9Sstevel@tonic-gate 
1757c478bd9Sstevel@tonic-gate 	id = dict_lookup_byname(name);
1767c478bd9Sstevel@tonic-gate 	if (id == -1)
1777c478bd9Sstevel@tonic-gate 		return (NULL);
1787c478bd9Sstevel@tonic-gate 
1797c478bd9Sstevel@tonic-gate 	return (inst_lookup_by_id(id));
1807c478bd9Sstevel@tonic-gate }
1817c478bd9Sstevel@tonic-gate 
1827c478bd9Sstevel@tonic-gate restarter_inst_t *
1837c478bd9Sstevel@tonic-gate inst_lookup_by_id(int id)
1847c478bd9Sstevel@tonic-gate {
1857c478bd9Sstevel@tonic-gate 	restarter_inst_t *inst;
1867c478bd9Sstevel@tonic-gate 
1877c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&instance_list.ril_lock);
1887c478bd9Sstevel@tonic-gate 	inst = uu_list_find(instance_list.ril_instance_list, &id, NULL, NULL);
1897c478bd9Sstevel@tonic-gate 	if (inst != NULL)
1907c478bd9Sstevel@tonic-gate 		MUTEX_LOCK(&inst->ri_lock);
1917c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&instance_list.ril_lock);
1927c478bd9Sstevel@tonic-gate 
1937c478bd9Sstevel@tonic-gate 	if (inst != NULL) {
1947c478bd9Sstevel@tonic-gate 		while (inst->ri_method_thread != 0 &&
1957c478bd9Sstevel@tonic-gate 		    !pthread_equal(inst->ri_method_thread, pthread_self())) {
1967c478bd9Sstevel@tonic-gate 			++inst->ri_method_waiters;
1977c478bd9Sstevel@tonic-gate 			(void) pthread_cond_wait(&inst->ri_method_cv,
1987c478bd9Sstevel@tonic-gate 			    &inst->ri_lock);
1997c478bd9Sstevel@tonic-gate 			assert(inst->ri_method_waiters > 0);
2007c478bd9Sstevel@tonic-gate 			--inst->ri_method_waiters;
2017c478bd9Sstevel@tonic-gate 		}
2027c478bd9Sstevel@tonic-gate 	}
2037c478bd9Sstevel@tonic-gate 
2047c478bd9Sstevel@tonic-gate 	return (inst);
2057c478bd9Sstevel@tonic-gate }
2067c478bd9Sstevel@tonic-gate 
2077c478bd9Sstevel@tonic-gate static restarter_inst_t *
2087c478bd9Sstevel@tonic-gate inst_lookup_queue(const char *name)
2097c478bd9Sstevel@tonic-gate {
2107c478bd9Sstevel@tonic-gate 	int id;
2117c478bd9Sstevel@tonic-gate 	restarter_inst_t *inst;
2127c478bd9Sstevel@tonic-gate 
2137c478bd9Sstevel@tonic-gate 	id = dict_lookup_byname(name);
2147c478bd9Sstevel@tonic-gate 	if (id == -1)
2157c478bd9Sstevel@tonic-gate 		return (NULL);
2167c478bd9Sstevel@tonic-gate 
2177c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&instance_list.ril_lock);
2187c478bd9Sstevel@tonic-gate 	inst = uu_list_find(instance_list.ril_instance_list, &id, NULL, NULL);
2197c478bd9Sstevel@tonic-gate 	if (inst != NULL)
2207c478bd9Sstevel@tonic-gate 		MUTEX_LOCK(&inst->ri_queue_lock);
2217c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&instance_list.ril_lock);
2227c478bd9Sstevel@tonic-gate 
2237c478bd9Sstevel@tonic-gate 	return (inst);
2247c478bd9Sstevel@tonic-gate }
2257c478bd9Sstevel@tonic-gate 
2267c478bd9Sstevel@tonic-gate const char *
2277c478bd9Sstevel@tonic-gate service_style(int flags)
2287c478bd9Sstevel@tonic-gate {
2297c478bd9Sstevel@tonic-gate 	switch (flags & RINST_STYLE_MASK) {
2307c478bd9Sstevel@tonic-gate 	case RINST_CONTRACT:	return ("contract");
2317c478bd9Sstevel@tonic-gate 	case RINST_TRANSIENT:	return ("transient");
2327c478bd9Sstevel@tonic-gate 	case RINST_WAIT:	return ("wait");
2337c478bd9Sstevel@tonic-gate 
2347c478bd9Sstevel@tonic-gate 	default:
2357c478bd9Sstevel@tonic-gate #ifndef NDEBUG
2367c478bd9Sstevel@tonic-gate 		uu_warn("%s:%d: Bad flags 0x%x.\n", __FILE__, __LINE__, flags);
2377c478bd9Sstevel@tonic-gate #endif
2387c478bd9Sstevel@tonic-gate 		abort();
2397c478bd9Sstevel@tonic-gate 		/* NOTREACHED */
2407c478bd9Sstevel@tonic-gate 	}
2417c478bd9Sstevel@tonic-gate }
2427c478bd9Sstevel@tonic-gate 
2437c478bd9Sstevel@tonic-gate /*
2447c478bd9Sstevel@tonic-gate  * Fails with ECONNABORTED or ECANCELED.
2457c478bd9Sstevel@tonic-gate  */
2467c478bd9Sstevel@tonic-gate static int
2477c478bd9Sstevel@tonic-gate check_contract(restarter_inst_t *inst, boolean_t primary,
2487c478bd9Sstevel@tonic-gate     scf_instance_t *scf_inst)
2497c478bd9Sstevel@tonic-gate {
2507c478bd9Sstevel@tonic-gate 	ctid_t *ctidp;
2517c478bd9Sstevel@tonic-gate 	int fd, r;
2527c478bd9Sstevel@tonic-gate 
2537c478bd9Sstevel@tonic-gate 	ctidp = primary ? &inst->ri_i.i_primary_ctid :
2547c478bd9Sstevel@tonic-gate 	    &inst->ri_i.i_transient_ctid;
2557c478bd9Sstevel@tonic-gate 
2567c478bd9Sstevel@tonic-gate 	assert(*ctidp >= 1);
2577c478bd9Sstevel@tonic-gate 
2587c478bd9Sstevel@tonic-gate 	fd = contract_open(*ctidp, NULL, "status", O_RDONLY);
2597c478bd9Sstevel@tonic-gate 	if (fd >= 0) {
2607c478bd9Sstevel@tonic-gate 		r = close(fd);
2617c478bd9Sstevel@tonic-gate 		assert(r == 0);
2627c478bd9Sstevel@tonic-gate 		return (0);
2637c478bd9Sstevel@tonic-gate 	}
2647c478bd9Sstevel@tonic-gate 
2657c478bd9Sstevel@tonic-gate 	r = restarter_remove_contract(scf_inst, *ctidp, primary ?
2667c478bd9Sstevel@tonic-gate 	    RESTARTER_CONTRACT_PRIMARY : RESTARTER_CONTRACT_TRANSIENT);
2677c478bd9Sstevel@tonic-gate 	switch (r) {
2687c478bd9Sstevel@tonic-gate 	case 0:
2697c478bd9Sstevel@tonic-gate 	case ECONNABORTED:
2707c478bd9Sstevel@tonic-gate 	case ECANCELED:
2717c478bd9Sstevel@tonic-gate 		*ctidp = 0;
2727c478bd9Sstevel@tonic-gate 		return (r);
2737c478bd9Sstevel@tonic-gate 
2747c478bd9Sstevel@tonic-gate 	case ENOMEM:
2757c478bd9Sstevel@tonic-gate 		uu_die("Out of memory\n");
2767c478bd9Sstevel@tonic-gate 		/* NOTREACHED */
2777c478bd9Sstevel@tonic-gate 
2787c478bd9Sstevel@tonic-gate 	case EPERM:
2797c478bd9Sstevel@tonic-gate 		uu_die("Insufficient privilege.\n");
2807c478bd9Sstevel@tonic-gate 		/* NOTREACHED */
2817c478bd9Sstevel@tonic-gate 
2827c478bd9Sstevel@tonic-gate 	case EACCES:
2837c478bd9Sstevel@tonic-gate 		uu_die("Repository backend access denied.\n");
2847c478bd9Sstevel@tonic-gate 		/* NOTREACHED */
2857c478bd9Sstevel@tonic-gate 
2867c478bd9Sstevel@tonic-gate 	case EROFS:
2877c478bd9Sstevel@tonic-gate 		log_error(LOG_INFO, "Could not remove unusable contract id %ld "
2887c478bd9Sstevel@tonic-gate 		    "for %s from repository.\n", *ctidp, inst->ri_i.i_fmri);
2897c478bd9Sstevel@tonic-gate 		return (0);
2907c478bd9Sstevel@tonic-gate 
2917c478bd9Sstevel@tonic-gate 	case EINVAL:
2927c478bd9Sstevel@tonic-gate 	case EBADF:
2937c478bd9Sstevel@tonic-gate 	default:
2947c478bd9Sstevel@tonic-gate 		assert(0);
2957c478bd9Sstevel@tonic-gate 		abort();
2967c478bd9Sstevel@tonic-gate 		/* NOTREACHED */
2977c478bd9Sstevel@tonic-gate 	}
2987c478bd9Sstevel@tonic-gate }
2997c478bd9Sstevel@tonic-gate 
3007c478bd9Sstevel@tonic-gate static int stop_instance(scf_handle_t *, restarter_inst_t *, stop_cause_t);
3017c478bd9Sstevel@tonic-gate 
3027c478bd9Sstevel@tonic-gate /*
3037c478bd9Sstevel@tonic-gate  * int restarter_insert_inst(scf_handle_t *, char *)
3047c478bd9Sstevel@tonic-gate  *   If the inst is already in the restarter list, return its id.  If the inst
3057c478bd9Sstevel@tonic-gate  *   is not in the restarter list, initialize a restarter_inst_t, initialize its
3067c478bd9Sstevel@tonic-gate  *   states, insert it into the list, and return 0.
3077c478bd9Sstevel@tonic-gate  *
3087c478bd9Sstevel@tonic-gate  *   Fails with
3097c478bd9Sstevel@tonic-gate  *     ENOENT - name is not in the repository
3107c478bd9Sstevel@tonic-gate  */
3117c478bd9Sstevel@tonic-gate static int
3127c478bd9Sstevel@tonic-gate restarter_insert_inst(scf_handle_t *h, const char *name)
3137c478bd9Sstevel@tonic-gate {
3147c478bd9Sstevel@tonic-gate 	int id, r;
3157c478bd9Sstevel@tonic-gate 	restarter_inst_t *inst;
3167c478bd9Sstevel@tonic-gate 	uu_list_index_t idx;
3177c478bd9Sstevel@tonic-gate 	scf_service_t *scf_svc;
3187c478bd9Sstevel@tonic-gate 	scf_instance_t *scf_inst;
31992175b8eSrm88369 	scf_snapshot_t *snap = NULL;
3207c478bd9Sstevel@tonic-gate 	scf_propertygroup_t *pg;
3217c478bd9Sstevel@tonic-gate 	char *svc_name, *inst_name;
3227c478bd9Sstevel@tonic-gate 	char logfilebuf[PATH_MAX];
3237c478bd9Sstevel@tonic-gate 	char *c;
3247c478bd9Sstevel@tonic-gate 	boolean_t do_commit_states;
3257c478bd9Sstevel@tonic-gate 	restarter_instance_state_t state, next_state;
3267c478bd9Sstevel@tonic-gate 	protocol_states_t *ps;
3277c478bd9Sstevel@tonic-gate 	pid_t start_pid;
3287c478bd9Sstevel@tonic-gate 
3297c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&instance_list.ril_lock);
3307c478bd9Sstevel@tonic-gate 
3317c478bd9Sstevel@tonic-gate 	/*
3327c478bd9Sstevel@tonic-gate 	 * We don't use inst_lookup_by_name() here because we want the lookup
3337c478bd9Sstevel@tonic-gate 	 * & insert to be atomic.
3347c478bd9Sstevel@tonic-gate 	 */
3357c478bd9Sstevel@tonic-gate 	id = dict_lookup_byname(name);
3367c478bd9Sstevel@tonic-gate 	if (id != -1) {
3377c478bd9Sstevel@tonic-gate 		inst = uu_list_find(instance_list.ril_instance_list, &id, NULL,
3387c478bd9Sstevel@tonic-gate 		    &idx);
3397c478bd9Sstevel@tonic-gate 		if (inst != NULL) {
3407c478bd9Sstevel@tonic-gate 			MUTEX_UNLOCK(&instance_list.ril_lock);
3417c478bd9Sstevel@tonic-gate 			return (0);
3427c478bd9Sstevel@tonic-gate 		}
3437c478bd9Sstevel@tonic-gate 	}
3447c478bd9Sstevel@tonic-gate 
3457c478bd9Sstevel@tonic-gate 	/* Allocate an instance */
3467c478bd9Sstevel@tonic-gate 	inst = startd_zalloc(sizeof (restarter_inst_t));
3477c478bd9Sstevel@tonic-gate 	inst->ri_utmpx_prefix = startd_alloc(max_scf_value_size);
3487c478bd9Sstevel@tonic-gate 	inst->ri_utmpx_prefix[0] = '\0';
3497c478bd9Sstevel@tonic-gate 
3507c478bd9Sstevel@tonic-gate 	inst->ri_i.i_fmri = startd_alloc(strlen(name) + 1);
3517c478bd9Sstevel@tonic-gate 	(void) strcpy((char *)inst->ri_i.i_fmri, name);
3527c478bd9Sstevel@tonic-gate 
3537c478bd9Sstevel@tonic-gate 	inst->ri_queue = startd_list_create(restarter_queue_pool, inst, 0);
3547c478bd9Sstevel@tonic-gate 
3557c478bd9Sstevel@tonic-gate 	/*
3567c478bd9Sstevel@tonic-gate 	 * id shouldn't be -1 since we use the same dictionary as graph.c, but
3577c478bd9Sstevel@tonic-gate 	 * just in case.
3587c478bd9Sstevel@tonic-gate 	 */
3597c478bd9Sstevel@tonic-gate 	inst->ri_id = (id != -1 ? id : dict_insert(name));
3607c478bd9Sstevel@tonic-gate 
3617c478bd9Sstevel@tonic-gate 	special_online_hooks_get(name, &inst->ri_pre_online_hook,
3627c478bd9Sstevel@tonic-gate 	    &inst->ri_post_online_hook, &inst->ri_post_offline_hook);
3637c478bd9Sstevel@tonic-gate 
3647c478bd9Sstevel@tonic-gate 	scf_svc = safe_scf_service_create(h);
3657c478bd9Sstevel@tonic-gate 	scf_inst = safe_scf_instance_create(h);
3667c478bd9Sstevel@tonic-gate 	pg = safe_scf_pg_create(h);
3677c478bd9Sstevel@tonic-gate 	svc_name = startd_alloc(max_scf_name_size);
3687c478bd9Sstevel@tonic-gate 	inst_name = startd_alloc(max_scf_name_size);
3697c478bd9Sstevel@tonic-gate 
3707c478bd9Sstevel@tonic-gate rep_retry:
37192175b8eSrm88369 	if (snap != NULL)
37292175b8eSrm88369 		scf_snapshot_destroy(snap);
37392175b8eSrm88369 	if (inst->ri_logstem != NULL)
37492175b8eSrm88369 		startd_free(inst->ri_logstem, PATH_MAX);
37592175b8eSrm88369 	if (inst->ri_common_name != NULL)
37692175b8eSrm88369 		startd_free(inst->ri_common_name, max_scf_value_size);
37792175b8eSrm88369 	if (inst->ri_C_common_name != NULL)
37892175b8eSrm88369 		startd_free(inst->ri_C_common_name, max_scf_value_size);
37992175b8eSrm88369 	snap = NULL;
38092175b8eSrm88369 	inst->ri_logstem = NULL;
38192175b8eSrm88369 	inst->ri_common_name = NULL;
38292175b8eSrm88369 	inst->ri_C_common_name = NULL;
38392175b8eSrm88369 
3847c478bd9Sstevel@tonic-gate 	if (scf_handle_decode_fmri(h, name, NULL, scf_svc, scf_inst, NULL,
3857c478bd9Sstevel@tonic-gate 	    NULL, SCF_DECODE_FMRI_EXACT) != 0) {
3867c478bd9Sstevel@tonic-gate 		switch (scf_error()) {
3877c478bd9Sstevel@tonic-gate 		case SCF_ERROR_CONNECTION_BROKEN:
3887c478bd9Sstevel@tonic-gate 			libscf_handle_rebind(h);
3897c478bd9Sstevel@tonic-gate 			goto rep_retry;
3907c478bd9Sstevel@tonic-gate 
3917c478bd9Sstevel@tonic-gate 		case SCF_ERROR_NOT_FOUND:
39292175b8eSrm88369 			goto deleted;
3937c478bd9Sstevel@tonic-gate 		}
3947c478bd9Sstevel@tonic-gate 
3957c478bd9Sstevel@tonic-gate 		uu_die("Can't decode FMRI %s: %s\n", name,
3967c478bd9Sstevel@tonic-gate 		    scf_strerror(scf_error()));
3977c478bd9Sstevel@tonic-gate 	}
3987c478bd9Sstevel@tonic-gate 
3997c478bd9Sstevel@tonic-gate 	/*
4007c478bd9Sstevel@tonic-gate 	 * If there's no running snapshot, then we execute using the editing
4017c478bd9Sstevel@tonic-gate 	 * snapshot.  Pending snapshots will be taken later.
4027c478bd9Sstevel@tonic-gate 	 */
4037c478bd9Sstevel@tonic-gate 	snap = libscf_get_running_snapshot(scf_inst);
4047c478bd9Sstevel@tonic-gate 
4057c478bd9Sstevel@tonic-gate 	if ((scf_service_get_name(scf_svc, svc_name, max_scf_name_size) < 0) ||
4067c478bd9Sstevel@tonic-gate 	    (scf_instance_get_name(scf_inst, inst_name, max_scf_name_size) <
4077c478bd9Sstevel@tonic-gate 	    0)) {
4087c478bd9Sstevel@tonic-gate 		switch (scf_error()) {
4097c478bd9Sstevel@tonic-gate 		case SCF_ERROR_NOT_SET:
4107c478bd9Sstevel@tonic-gate 			break;
4117c478bd9Sstevel@tonic-gate 
4127c478bd9Sstevel@tonic-gate 		case SCF_ERROR_CONNECTION_BROKEN:
4137c478bd9Sstevel@tonic-gate 			libscf_handle_rebind(h);
4147c478bd9Sstevel@tonic-gate 			goto rep_retry;
4157c478bd9Sstevel@tonic-gate 
4167c478bd9Sstevel@tonic-gate 		default:
4177c478bd9Sstevel@tonic-gate 			assert(0);
4187c478bd9Sstevel@tonic-gate 			abort();
4197c478bd9Sstevel@tonic-gate 		}
4207c478bd9Sstevel@tonic-gate 
4217c478bd9Sstevel@tonic-gate 		goto deleted;
4227c478bd9Sstevel@tonic-gate 	}
4237c478bd9Sstevel@tonic-gate 
4241855af6bSlianep 	(void) snprintf(logfilebuf, PATH_MAX, "%s:%s", svc_name, inst_name);
4251855af6bSlianep 	for (c = logfilebuf; *c != '\0'; c++)
4261855af6bSlianep 		if (*c == '/')
4271855af6bSlianep 			*c = '-';
4281855af6bSlianep 
4291855af6bSlianep 	inst->ri_logstem = startd_alloc(PATH_MAX);
4301855af6bSlianep 	(void) snprintf(inst->ri_logstem, PATH_MAX, "%s%s", logfilebuf,
4311855af6bSlianep 	    LOG_SUFFIX);
4321855af6bSlianep 
4337c478bd9Sstevel@tonic-gate 	/*
4347c478bd9Sstevel@tonic-gate 	 * If the restarter group is missing, use uninit/none.  Otherwise,
4357c478bd9Sstevel@tonic-gate 	 * we're probably being restarted & don't want to mess up the states
4367c478bd9Sstevel@tonic-gate 	 * that are there.
4377c478bd9Sstevel@tonic-gate 	 */
4387c478bd9Sstevel@tonic-gate 	state = RESTARTER_STATE_UNINIT;
4397c478bd9Sstevel@tonic-gate 	next_state = RESTARTER_STATE_NONE;
4407c478bd9Sstevel@tonic-gate 
4417c478bd9Sstevel@tonic-gate 	r = scf_instance_get_pg(scf_inst, SCF_PG_RESTARTER, pg);
4427c478bd9Sstevel@tonic-gate 	if (r != 0) {
4437c478bd9Sstevel@tonic-gate 		switch (scf_error()) {
4447c478bd9Sstevel@tonic-gate 		case SCF_ERROR_CONNECTION_BROKEN:
4457c478bd9Sstevel@tonic-gate 			libscf_handle_rebind(h);
4467c478bd9Sstevel@tonic-gate 			goto rep_retry;
4477c478bd9Sstevel@tonic-gate 
4487c478bd9Sstevel@tonic-gate 		case SCF_ERROR_NOT_SET:
4497c478bd9Sstevel@tonic-gate 			goto deleted;
4507c478bd9Sstevel@tonic-gate 
4517c478bd9Sstevel@tonic-gate 		case SCF_ERROR_NOT_FOUND:
4527c478bd9Sstevel@tonic-gate 			/*
4537c478bd9Sstevel@tonic-gate 			 * This shouldn't happen since the graph engine should
4547c478bd9Sstevel@tonic-gate 			 * have initialized the state to uninitialized/none if
4557c478bd9Sstevel@tonic-gate 			 * there was no restarter pg.  In case somebody
4567c478bd9Sstevel@tonic-gate 			 * deleted it, though....
4577c478bd9Sstevel@tonic-gate 			 */
4587c478bd9Sstevel@tonic-gate 			do_commit_states = B_TRUE;
4597c478bd9Sstevel@tonic-gate 			break;
4607c478bd9Sstevel@tonic-gate 
4617c478bd9Sstevel@tonic-gate 		default:
4627c478bd9Sstevel@tonic-gate 			assert(0);
4637c478bd9Sstevel@tonic-gate 			abort();
4647c478bd9Sstevel@tonic-gate 		}
4657c478bd9Sstevel@tonic-gate 	} else {
4667c478bd9Sstevel@tonic-gate 		r = libscf_read_states(pg, &state, &next_state);
4677c478bd9Sstevel@tonic-gate 		if (r != 0) {
4687c478bd9Sstevel@tonic-gate 			do_commit_states = B_TRUE;
4697c478bd9Sstevel@tonic-gate 		} else {
4707c478bd9Sstevel@tonic-gate 			if (next_state != RESTARTER_STATE_NONE) {
4717c478bd9Sstevel@tonic-gate 				/*
4727c478bd9Sstevel@tonic-gate 				 * Force next_state to _NONE since we
4737c478bd9Sstevel@tonic-gate 				 * don't look for method processes.
4747c478bd9Sstevel@tonic-gate 				 */
4757c478bd9Sstevel@tonic-gate 				next_state = RESTARTER_STATE_NONE;
4767c478bd9Sstevel@tonic-gate 				do_commit_states = B_TRUE;
4777c478bd9Sstevel@tonic-gate 			} else {
4787c478bd9Sstevel@tonic-gate 				/*
4797c478bd9Sstevel@tonic-gate 				 * Inform the restarter of our state without
4807c478bd9Sstevel@tonic-gate 				 * changing the STIME in the repository.
4817c478bd9Sstevel@tonic-gate 				 */
4827c478bd9Sstevel@tonic-gate 				ps = startd_alloc(sizeof (*ps));
4837c478bd9Sstevel@tonic-gate 				inst->ri_i.i_state = ps->ps_state = state;
4847c478bd9Sstevel@tonic-gate 				inst->ri_i.i_next_state = ps->ps_state_next =
4857c478bd9Sstevel@tonic-gate 				    next_state;
4867c478bd9Sstevel@tonic-gate 
4877c478bd9Sstevel@tonic-gate 				graph_protocol_send_event(inst->ri_i.i_fmri,
4887c478bd9Sstevel@tonic-gate 				    GRAPH_UPDATE_STATE_CHANGE, ps);
4897c478bd9Sstevel@tonic-gate 
4907c478bd9Sstevel@tonic-gate 				do_commit_states = B_FALSE;
4917c478bd9Sstevel@tonic-gate 			}
4927c478bd9Sstevel@tonic-gate 		}
4937c478bd9Sstevel@tonic-gate 	}
4947c478bd9Sstevel@tonic-gate 
4957c478bd9Sstevel@tonic-gate 	switch (libscf_get_startd_properties(scf_inst, snap, &inst->ri_flags,
4967c478bd9Sstevel@tonic-gate 	    &inst->ri_utmpx_prefix)) {
4977c478bd9Sstevel@tonic-gate 	case 0:
4987c478bd9Sstevel@tonic-gate 		break;
4997c478bd9Sstevel@tonic-gate 
5007c478bd9Sstevel@tonic-gate 	case ECONNABORTED:
5017c478bd9Sstevel@tonic-gate 		libscf_handle_rebind(h);
5027c478bd9Sstevel@tonic-gate 		goto rep_retry;
5037c478bd9Sstevel@tonic-gate 
5047c478bd9Sstevel@tonic-gate 	case ECANCELED:
5057c478bd9Sstevel@tonic-gate 		goto deleted;
5067c478bd9Sstevel@tonic-gate 
5077c478bd9Sstevel@tonic-gate 	case ENOENT:
5087c478bd9Sstevel@tonic-gate 		/*
5097c478bd9Sstevel@tonic-gate 		 * This is odd, because the graph engine should have required
5107c478bd9Sstevel@tonic-gate 		 * the general property group.  So we'll just use default
5117c478bd9Sstevel@tonic-gate 		 * flags in anticipation of the graph engine sending us
5127c478bd9Sstevel@tonic-gate 		 * REMOVE_INSTANCE when it finds out that the general property
5137c478bd9Sstevel@tonic-gate 		 * group has been deleted.
5147c478bd9Sstevel@tonic-gate 		 */
5157c478bd9Sstevel@tonic-gate 		inst->ri_flags = RINST_CONTRACT;
5167c478bd9Sstevel@tonic-gate 		break;
5177c478bd9Sstevel@tonic-gate 
5187c478bd9Sstevel@tonic-gate 	default:
5197c478bd9Sstevel@tonic-gate 		assert(0);
5207c478bd9Sstevel@tonic-gate 		abort();
5217c478bd9Sstevel@tonic-gate 	}
5227c478bd9Sstevel@tonic-gate 
5237c478bd9Sstevel@tonic-gate 	switch (libscf_get_template_values(scf_inst, snap,
5247c478bd9Sstevel@tonic-gate 	    &inst->ri_common_name, &inst->ri_C_common_name)) {
5257c478bd9Sstevel@tonic-gate 	case 0:
5267c478bd9Sstevel@tonic-gate 		break;
5277c478bd9Sstevel@tonic-gate 
5287c478bd9Sstevel@tonic-gate 	case ECONNABORTED:
5297c478bd9Sstevel@tonic-gate 		libscf_handle_rebind(h);
5307c478bd9Sstevel@tonic-gate 		goto rep_retry;
5317c478bd9Sstevel@tonic-gate 
5327c478bd9Sstevel@tonic-gate 	case ECANCELED:
5337c478bd9Sstevel@tonic-gate 		goto deleted;
5347c478bd9Sstevel@tonic-gate 
5357c478bd9Sstevel@tonic-gate 	case ECHILD:
5367c478bd9Sstevel@tonic-gate 	case ENOENT:
5377c478bd9Sstevel@tonic-gate 		break;
5387c478bd9Sstevel@tonic-gate 
5397c478bd9Sstevel@tonic-gate 	default:
5407c478bd9Sstevel@tonic-gate 		assert(0);
5417c478bd9Sstevel@tonic-gate 		abort();
5427c478bd9Sstevel@tonic-gate 	}
5437c478bd9Sstevel@tonic-gate 
5447c478bd9Sstevel@tonic-gate 	switch (libscf_read_method_ids(h, scf_inst, inst->ri_i.i_fmri,
5457c478bd9Sstevel@tonic-gate 	    &inst->ri_i.i_primary_ctid, &inst->ri_i.i_transient_ctid,
5467c478bd9Sstevel@tonic-gate 	    &start_pid)) {
5477c478bd9Sstevel@tonic-gate 	case 0:
5487c478bd9Sstevel@tonic-gate 		break;
5497c478bd9Sstevel@tonic-gate 
5507c478bd9Sstevel@tonic-gate 	case ECONNABORTED:
5517c478bd9Sstevel@tonic-gate 		libscf_handle_rebind(h);
5527c478bd9Sstevel@tonic-gate 		goto rep_retry;
5537c478bd9Sstevel@tonic-gate 
5547c478bd9Sstevel@tonic-gate 	case ECANCELED:
5557c478bd9Sstevel@tonic-gate 		goto deleted;
5567c478bd9Sstevel@tonic-gate 
5577c478bd9Sstevel@tonic-gate 	default:
5587c478bd9Sstevel@tonic-gate 		assert(0);
5597c478bd9Sstevel@tonic-gate 		abort();
5607c478bd9Sstevel@tonic-gate 	}
5617c478bd9Sstevel@tonic-gate 
5627c478bd9Sstevel@tonic-gate 	if (inst->ri_i.i_primary_ctid >= 1) {
5637c478bd9Sstevel@tonic-gate 		contract_hash_store(inst->ri_i.i_primary_ctid, inst->ri_id);
5647c478bd9Sstevel@tonic-gate 
5657c478bd9Sstevel@tonic-gate 		switch (check_contract(inst, B_TRUE, scf_inst)) {
5667c478bd9Sstevel@tonic-gate 		case 0:
5677c478bd9Sstevel@tonic-gate 			break;
5687c478bd9Sstevel@tonic-gate 
5697c478bd9Sstevel@tonic-gate 		case ECONNABORTED:
5707c478bd9Sstevel@tonic-gate 			libscf_handle_rebind(h);
5717c478bd9Sstevel@tonic-gate 			goto rep_retry;
5727c478bd9Sstevel@tonic-gate 
5737c478bd9Sstevel@tonic-gate 		case ECANCELED:
5747c478bd9Sstevel@tonic-gate 			goto deleted;
5757c478bd9Sstevel@tonic-gate 
5767c478bd9Sstevel@tonic-gate 		default:
5777c478bd9Sstevel@tonic-gate 			assert(0);
5787c478bd9Sstevel@tonic-gate 			abort();
5797c478bd9Sstevel@tonic-gate 		}
5807c478bd9Sstevel@tonic-gate 	}
5817c478bd9Sstevel@tonic-gate 
5827c478bd9Sstevel@tonic-gate 	if (inst->ri_i.i_transient_ctid >= 1) {
5837c478bd9Sstevel@tonic-gate 		switch (check_contract(inst, B_FALSE, scf_inst)) {
5847c478bd9Sstevel@tonic-gate 		case 0:
5857c478bd9Sstevel@tonic-gate 			break;
5867c478bd9Sstevel@tonic-gate 
5877c478bd9Sstevel@tonic-gate 		case ECONNABORTED:
5887c478bd9Sstevel@tonic-gate 			libscf_handle_rebind(h);
5897c478bd9Sstevel@tonic-gate 			goto rep_retry;
5907c478bd9Sstevel@tonic-gate 
5917c478bd9Sstevel@tonic-gate 		case ECANCELED:
5927c478bd9Sstevel@tonic-gate 			goto deleted;
5937c478bd9Sstevel@tonic-gate 
5947c478bd9Sstevel@tonic-gate 		default:
5957c478bd9Sstevel@tonic-gate 			assert(0);
5967c478bd9Sstevel@tonic-gate 			abort();
5977c478bd9Sstevel@tonic-gate 		}
5987c478bd9Sstevel@tonic-gate 	}
5997c478bd9Sstevel@tonic-gate 
6007c478bd9Sstevel@tonic-gate 	/* No more failures we live through, so add it to the list. */
6017c478bd9Sstevel@tonic-gate 	(void) pthread_mutex_init(&inst->ri_lock, &mutex_attrs);
6027c478bd9Sstevel@tonic-gate 	(void) pthread_mutex_init(&inst->ri_queue_lock, &mutex_attrs);
6037c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&inst->ri_lock);
6047c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&inst->ri_queue_lock);
6057c478bd9Sstevel@tonic-gate 
6067c478bd9Sstevel@tonic-gate 	(void) pthread_cond_init(&inst->ri_method_cv, NULL);
6077c478bd9Sstevel@tonic-gate 
6087c478bd9Sstevel@tonic-gate 	uu_list_node_init(inst, &inst->ri_link, restarter_instance_pool);
6097c478bd9Sstevel@tonic-gate 	uu_list_insert(instance_list.ril_instance_list, inst, idx);
6107c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&instance_list.ril_lock);
6117c478bd9Sstevel@tonic-gate 
6127c478bd9Sstevel@tonic-gate 	if (start_pid != -1 &&
6137c478bd9Sstevel@tonic-gate 	    (inst->ri_flags & RINST_STYLE_MASK) == RINST_WAIT) {
6147c478bd9Sstevel@tonic-gate 		int ret;
6157c478bd9Sstevel@tonic-gate 		ret = wait_register(start_pid, inst->ri_i.i_fmri, 0, 1);
6167c478bd9Sstevel@tonic-gate 		if (ret == -1) {
6177c478bd9Sstevel@tonic-gate 			/*
6187c478bd9Sstevel@tonic-gate 			 * Implication:  if we can't reregister the
6197c478bd9Sstevel@tonic-gate 			 * instance, we will start another one.  Two
6207c478bd9Sstevel@tonic-gate 			 * instances may or may not result in a resource
6217c478bd9Sstevel@tonic-gate 			 * conflict.
6227c478bd9Sstevel@tonic-gate 			 */
6237c478bd9Sstevel@tonic-gate 			log_error(LOG_WARNING,
6247c478bd9Sstevel@tonic-gate 			    "%s: couldn't reregister %ld for wait\n",
6257c478bd9Sstevel@tonic-gate 			    inst->ri_i.i_fmri, start_pid);
6267c478bd9Sstevel@tonic-gate 		} else if (ret == 1) {
6277c478bd9Sstevel@tonic-gate 			/*
6287c478bd9Sstevel@tonic-gate 			 * Leading PID has exited.
6297c478bd9Sstevel@tonic-gate 			 */
6307c478bd9Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_EXIT);
6317c478bd9Sstevel@tonic-gate 		}
6327c478bd9Sstevel@tonic-gate 	}
6337c478bd9Sstevel@tonic-gate 
6347c478bd9Sstevel@tonic-gate 
6357c478bd9Sstevel@tonic-gate 	scf_pg_destroy(pg);
6367c478bd9Sstevel@tonic-gate 
6377c478bd9Sstevel@tonic-gate 	if (do_commit_states)
6387c478bd9Sstevel@tonic-gate 		(void) restarter_instance_update_states(h, inst, state,
6397c478bd9Sstevel@tonic-gate 		    next_state, RERR_NONE, NULL);
6407c478bd9Sstevel@tonic-gate 
6417c478bd9Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s is a %s-style service\n", name,
6427c478bd9Sstevel@tonic-gate 	    service_style(inst->ri_flags));
6437c478bd9Sstevel@tonic-gate 
6447c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&inst->ri_queue_lock);
6457c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&inst->ri_lock);
6467c478bd9Sstevel@tonic-gate 
6477c478bd9Sstevel@tonic-gate 	startd_free(svc_name, max_scf_name_size);
6487c478bd9Sstevel@tonic-gate 	startd_free(inst_name, max_scf_name_size);
6497c478bd9Sstevel@tonic-gate 	scf_snapshot_destroy(snap);
6507c478bd9Sstevel@tonic-gate 	scf_instance_destroy(scf_inst);
6517c478bd9Sstevel@tonic-gate 	scf_service_destroy(scf_svc);
6527c478bd9Sstevel@tonic-gate 
6537c478bd9Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: inserted instance into restarter list\n",
6547c478bd9Sstevel@tonic-gate 	    name);
6557c478bd9Sstevel@tonic-gate 
6567c478bd9Sstevel@tonic-gate 	return (0);
65792175b8eSrm88369 
65892175b8eSrm88369 deleted:
65992175b8eSrm88369 	MUTEX_UNLOCK(&instance_list.ril_lock);
66092175b8eSrm88369 	startd_free(inst_name, max_scf_name_size);
66192175b8eSrm88369 	startd_free(svc_name, max_scf_name_size);
66292175b8eSrm88369 	if (snap != NULL)
66392175b8eSrm88369 		scf_snapshot_destroy(snap);
66492175b8eSrm88369 	scf_pg_destroy(pg);
66592175b8eSrm88369 	scf_instance_destroy(scf_inst);
66692175b8eSrm88369 	scf_service_destroy(scf_svc);
66792175b8eSrm88369 	startd_free((void *)inst->ri_i.i_fmri, strlen(inst->ri_i.i_fmri) + 1);
66892175b8eSrm88369 	uu_list_destroy(inst->ri_queue);
66992175b8eSrm88369 	if (inst->ri_logstem != NULL)
67092175b8eSrm88369 		startd_free(inst->ri_logstem, PATH_MAX);
67192175b8eSrm88369 	if (inst->ri_common_name != NULL)
67292175b8eSrm88369 		startd_free(inst->ri_common_name, max_scf_value_size);
67392175b8eSrm88369 	if (inst->ri_C_common_name != NULL)
67492175b8eSrm88369 		startd_free(inst->ri_C_common_name, max_scf_value_size);
67592175b8eSrm88369 	startd_free(inst->ri_utmpx_prefix, max_scf_value_size);
67692175b8eSrm88369 	startd_free(inst, sizeof (restarter_inst_t));
67792175b8eSrm88369 	return (ENOENT);
6787c478bd9Sstevel@tonic-gate }
6797c478bd9Sstevel@tonic-gate 
6807c478bd9Sstevel@tonic-gate static void
6817c478bd9Sstevel@tonic-gate restarter_delete_inst(restarter_inst_t *ri)
6827c478bd9Sstevel@tonic-gate {
6837c478bd9Sstevel@tonic-gate 	int id;
6847c478bd9Sstevel@tonic-gate 	restarter_inst_t *rip;
6857c478bd9Sstevel@tonic-gate 	void *cookie = NULL;
6867c478bd9Sstevel@tonic-gate 	restarter_instance_qentry_t *e;
6877c478bd9Sstevel@tonic-gate 
68853f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&ri->ri_lock));
6897c478bd9Sstevel@tonic-gate 
6907c478bd9Sstevel@tonic-gate 	/*
6917c478bd9Sstevel@tonic-gate 	 * Must drop the instance lock so we can pick up the instance_list
6927c478bd9Sstevel@tonic-gate 	 * lock & remove the instance.
6937c478bd9Sstevel@tonic-gate 	 */
6947c478bd9Sstevel@tonic-gate 	id = ri->ri_id;
6957c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&ri->ri_lock);
6967c478bd9Sstevel@tonic-gate 
6977c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&instance_list.ril_lock);
6987c478bd9Sstevel@tonic-gate 
6997c478bd9Sstevel@tonic-gate 	rip = uu_list_find(instance_list.ril_instance_list, &id, NULL, NULL);
7007c478bd9Sstevel@tonic-gate 	if (rip == NULL) {
7017c478bd9Sstevel@tonic-gate 		MUTEX_UNLOCK(&instance_list.ril_lock);
7027c478bd9Sstevel@tonic-gate 		return;
7037c478bd9Sstevel@tonic-gate 	}
7047c478bd9Sstevel@tonic-gate 
7057c478bd9Sstevel@tonic-gate 	assert(ri == rip);
7067c478bd9Sstevel@tonic-gate 
7077c478bd9Sstevel@tonic-gate 	uu_list_remove(instance_list.ril_instance_list, ri);
7087c478bd9Sstevel@tonic-gate 
7097c478bd9Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: deleted instance from restarter list\n",
7107c478bd9Sstevel@tonic-gate 	    ri->ri_i.i_fmri);
7117c478bd9Sstevel@tonic-gate 
7127c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&instance_list.ril_lock);
7137c478bd9Sstevel@tonic-gate 
7147c478bd9Sstevel@tonic-gate 	/*
7157c478bd9Sstevel@tonic-gate 	 * We can lock the instance without holding the instance_list lock
7167c478bd9Sstevel@tonic-gate 	 * since we removed the instance from the list.
7177c478bd9Sstevel@tonic-gate 	 */
7187c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&ri->ri_lock);
7197c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&ri->ri_queue_lock);
7207c478bd9Sstevel@tonic-gate 
7217c478bd9Sstevel@tonic-gate 	if (ri->ri_i.i_primary_ctid >= 1)
7227c478bd9Sstevel@tonic-gate 		contract_hash_remove(ri->ri_i.i_primary_ctid);
7237c478bd9Sstevel@tonic-gate 
7247c478bd9Sstevel@tonic-gate 	while (ri->ri_method_thread != 0 || ri->ri_method_waiters > 0)
7257c478bd9Sstevel@tonic-gate 		(void) pthread_cond_wait(&ri->ri_method_cv, &ri->ri_lock);
7267c478bd9Sstevel@tonic-gate 
7277c478bd9Sstevel@tonic-gate 	while ((e = uu_list_teardown(ri->ri_queue, &cookie)) != NULL)
7287c478bd9Sstevel@tonic-gate 		startd_free(e, sizeof (*e));
7297c478bd9Sstevel@tonic-gate 	uu_list_destroy(ri->ri_queue);
7307c478bd9Sstevel@tonic-gate 
7317c478bd9Sstevel@tonic-gate 	startd_free((void *)ri->ri_i.i_fmri, strlen(ri->ri_i.i_fmri) + 1);
7321855af6bSlianep 	startd_free(ri->ri_logstem, PATH_MAX);
7333ba60fe0Srm88369 	if (ri->ri_common_name != NULL)
7343ba60fe0Srm88369 		startd_free(ri->ri_common_name, max_scf_value_size);
7353ba60fe0Srm88369 	if (ri->ri_C_common_name != NULL)
7363ba60fe0Srm88369 		startd_free(ri->ri_C_common_name, max_scf_value_size);
7377c478bd9Sstevel@tonic-gate 	startd_free(ri->ri_utmpx_prefix, max_scf_value_size);
7387c478bd9Sstevel@tonic-gate 	(void) pthread_mutex_destroy(&ri->ri_lock);
7397c478bd9Sstevel@tonic-gate 	(void) pthread_mutex_destroy(&ri->ri_queue_lock);
7407c478bd9Sstevel@tonic-gate 	startd_free(ri, sizeof (restarter_inst_t));
7417c478bd9Sstevel@tonic-gate }
7427c478bd9Sstevel@tonic-gate 
7437c478bd9Sstevel@tonic-gate /*
7447c478bd9Sstevel@tonic-gate  * instance_is_wait_style()
7457c478bd9Sstevel@tonic-gate  *
7467c478bd9Sstevel@tonic-gate  *   Returns 1 if the given instance is a "wait-style" service instance.
7477c478bd9Sstevel@tonic-gate  */
7487c478bd9Sstevel@tonic-gate int
7497c478bd9Sstevel@tonic-gate instance_is_wait_style(restarter_inst_t *inst)
7507c478bd9Sstevel@tonic-gate {
75153f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&inst->ri_lock));
7527c478bd9Sstevel@tonic-gate 	return ((inst->ri_flags & RINST_STYLE_MASK) == RINST_WAIT);
7537c478bd9Sstevel@tonic-gate }
7547c478bd9Sstevel@tonic-gate 
7557c478bd9Sstevel@tonic-gate /*
7567c478bd9Sstevel@tonic-gate  * instance_is_transient_style()
7577c478bd9Sstevel@tonic-gate  *
7587c478bd9Sstevel@tonic-gate  *   Returns 1 if the given instance is a transient service instance.
7597c478bd9Sstevel@tonic-gate  */
7607c478bd9Sstevel@tonic-gate int
7617c478bd9Sstevel@tonic-gate instance_is_transient_style(restarter_inst_t *inst)
7627c478bd9Sstevel@tonic-gate {
76353f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&inst->ri_lock));
7647c478bd9Sstevel@tonic-gate 	return ((inst->ri_flags & RINST_STYLE_MASK) == RINST_TRANSIENT);
7657c478bd9Sstevel@tonic-gate }
7667c478bd9Sstevel@tonic-gate 
7677c478bd9Sstevel@tonic-gate /*
7687c478bd9Sstevel@tonic-gate  * instance_in_transition()
7697c478bd9Sstevel@tonic-gate  * Returns 1 if instance is in transition, 0 if not
7707c478bd9Sstevel@tonic-gate  */
7717c478bd9Sstevel@tonic-gate int
7727c478bd9Sstevel@tonic-gate instance_in_transition(restarter_inst_t *inst)
7737c478bd9Sstevel@tonic-gate {
77453f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&inst->ri_lock));
7757c478bd9Sstevel@tonic-gate 	if (inst->ri_i.i_next_state == RESTARTER_STATE_NONE)
7767c478bd9Sstevel@tonic-gate 		return (0);
7777c478bd9Sstevel@tonic-gate 	return (1);
7787c478bd9Sstevel@tonic-gate }
7797c478bd9Sstevel@tonic-gate 
7807c478bd9Sstevel@tonic-gate /*
78121d7f835Sgm149974  * returns 1 if instance is already started, 0 if not
78221d7f835Sgm149974  */
78321d7f835Sgm149974 static int
78421d7f835Sgm149974 instance_started(restarter_inst_t *inst)
78521d7f835Sgm149974 {
78621d7f835Sgm149974 	int ret;
78721d7f835Sgm149974 
78853f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&inst->ri_lock));
78921d7f835Sgm149974 
79021d7f835Sgm149974 	if (inst->ri_i.i_state == RESTARTER_STATE_ONLINE ||
79121d7f835Sgm149974 	    inst->ri_i.i_state == RESTARTER_STATE_DEGRADED)
79221d7f835Sgm149974 		ret = 1;
79321d7f835Sgm149974 	else
79421d7f835Sgm149974 		ret = 0;
79521d7f835Sgm149974 
79621d7f835Sgm149974 	return (ret);
79721d7f835Sgm149974 }
79821d7f835Sgm149974 
79921d7f835Sgm149974 /*
8007c478bd9Sstevel@tonic-gate  * Returns
8017c478bd9Sstevel@tonic-gate  *   0 - success
8027c478bd9Sstevel@tonic-gate  *   ECONNRESET - success, but h was rebound
8037c478bd9Sstevel@tonic-gate  */
8047c478bd9Sstevel@tonic-gate int
8057c478bd9Sstevel@tonic-gate restarter_instance_update_states(scf_handle_t *h, restarter_inst_t *ri,
8067c478bd9Sstevel@tonic-gate     restarter_instance_state_t new_state,
8077c478bd9Sstevel@tonic-gate     restarter_instance_state_t new_state_next, restarter_error_t err, char *aux)
8087c478bd9Sstevel@tonic-gate {
8097c478bd9Sstevel@tonic-gate 	protocol_states_t *states;
8107c478bd9Sstevel@tonic-gate 	int e;
8117c478bd9Sstevel@tonic-gate 	uint_t retry_count = 0, msecs = ALLOC_DELAY;
8127c478bd9Sstevel@tonic-gate 	boolean_t rebound = B_FALSE;
81321d7f835Sgm149974 	int prev_state_online;
81421d7f835Sgm149974 	int state_online;
8157c478bd9Sstevel@tonic-gate 
81653f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&ri->ri_lock));
8177c478bd9Sstevel@tonic-gate 
81821d7f835Sgm149974 	prev_state_online = instance_started(ri);
81921d7f835Sgm149974 
8207c478bd9Sstevel@tonic-gate retry:
8217c478bd9Sstevel@tonic-gate 	e = _restarter_commit_states(h, &ri->ri_i, new_state, new_state_next,
8227c478bd9Sstevel@tonic-gate 	    aux);
8237c478bd9Sstevel@tonic-gate 	switch (e) {
8247c478bd9Sstevel@tonic-gate 	case 0:
8257c478bd9Sstevel@tonic-gate 		break;
8267c478bd9Sstevel@tonic-gate 
8277c478bd9Sstevel@tonic-gate 	case ENOMEM:
8287c478bd9Sstevel@tonic-gate 		++retry_count;
8297c478bd9Sstevel@tonic-gate 		if (retry_count < ALLOC_RETRY) {
8307c478bd9Sstevel@tonic-gate 			(void) poll(NULL, 0, msecs);
8317c478bd9Sstevel@tonic-gate 			msecs *= ALLOC_DELAY_MULT;
8327c478bd9Sstevel@tonic-gate 			goto retry;
8337c478bd9Sstevel@tonic-gate 		}
8347c478bd9Sstevel@tonic-gate 
8357c478bd9Sstevel@tonic-gate 		/* Like startd_alloc(). */
8367c478bd9Sstevel@tonic-gate 		uu_die("Insufficient memory.\n");
8377c478bd9Sstevel@tonic-gate 		/* NOTREACHED */
8387c478bd9Sstevel@tonic-gate 
8397c478bd9Sstevel@tonic-gate 	case ECONNABORTED:
8407c478bd9Sstevel@tonic-gate 		libscf_handle_rebind(h);
8417c478bd9Sstevel@tonic-gate 		rebound = B_TRUE;
8427c478bd9Sstevel@tonic-gate 		goto retry;
8437c478bd9Sstevel@tonic-gate 
8447c478bd9Sstevel@tonic-gate 	case EPERM:
8457c478bd9Sstevel@tonic-gate 	case EACCES:
8467c478bd9Sstevel@tonic-gate 	case EROFS:
8477c478bd9Sstevel@tonic-gate 		log_error(LOG_NOTICE, "Could not commit state change for %s "
8487c478bd9Sstevel@tonic-gate 		    "to repository: %s.\n", ri->ri_i.i_fmri, strerror(e));
8497c478bd9Sstevel@tonic-gate 		/* FALLTHROUGH */
8507c478bd9Sstevel@tonic-gate 
8517c478bd9Sstevel@tonic-gate 	case ENOENT:
8527c478bd9Sstevel@tonic-gate 		ri->ri_i.i_state = new_state;
8537c478bd9Sstevel@tonic-gate 		ri->ri_i.i_next_state = new_state_next;
8547c478bd9Sstevel@tonic-gate 		break;
8557c478bd9Sstevel@tonic-gate 
8567c478bd9Sstevel@tonic-gate 	case EINVAL:
8577c478bd9Sstevel@tonic-gate 	default:
8587c478bd9Sstevel@tonic-gate 		bad_error("_restarter_commit_states", e);
8597c478bd9Sstevel@tonic-gate 	}
8607c478bd9Sstevel@tonic-gate 
8617c478bd9Sstevel@tonic-gate 	states = startd_alloc(sizeof (protocol_states_t));
8627c478bd9Sstevel@tonic-gate 	states->ps_state = new_state;
8637c478bd9Sstevel@tonic-gate 	states->ps_state_next = new_state_next;
8647c478bd9Sstevel@tonic-gate 	states->ps_err = err;
8657c478bd9Sstevel@tonic-gate 	graph_protocol_send_event(ri->ri_i.i_fmri, GRAPH_UPDATE_STATE_CHANGE,
8667c478bd9Sstevel@tonic-gate 	    (void *)states);
8677c478bd9Sstevel@tonic-gate 
86821d7f835Sgm149974 	state_online = instance_started(ri);
86921d7f835Sgm149974 
87021d7f835Sgm149974 	if (prev_state_online && !state_online)
87121d7f835Sgm149974 		ri->ri_post_offline_hook();
87221d7f835Sgm149974 	else if (!prev_state_online && state_online)
8737c478bd9Sstevel@tonic-gate 		ri->ri_post_online_hook();
8747c478bd9Sstevel@tonic-gate 
8757c478bd9Sstevel@tonic-gate 	return (rebound ? ECONNRESET : 0);
8767c478bd9Sstevel@tonic-gate }
8777c478bd9Sstevel@tonic-gate 
8787c478bd9Sstevel@tonic-gate void
8797c478bd9Sstevel@tonic-gate restarter_mark_pending_snapshot(const char *fmri, uint_t flag)
8807c478bd9Sstevel@tonic-gate {
8817c478bd9Sstevel@tonic-gate 	restarter_inst_t *inst;
8827c478bd9Sstevel@tonic-gate 
8837c478bd9Sstevel@tonic-gate 	assert(flag == RINST_RETAKE_RUNNING || flag == RINST_RETAKE_START);
8847c478bd9Sstevel@tonic-gate 
8857c478bd9Sstevel@tonic-gate 	inst = inst_lookup_by_name(fmri);
8867c478bd9Sstevel@tonic-gate 	if (inst == NULL)
8877c478bd9Sstevel@tonic-gate 		return;
8887c478bd9Sstevel@tonic-gate 
8897c478bd9Sstevel@tonic-gate 	inst->ri_flags |= flag;
8907c478bd9Sstevel@tonic-gate 
8917c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&inst->ri_lock);
8927c478bd9Sstevel@tonic-gate }
8937c478bd9Sstevel@tonic-gate 
8947c478bd9Sstevel@tonic-gate static void
8957c478bd9Sstevel@tonic-gate restarter_take_pending_snapshots(scf_handle_t *h)
8967c478bd9Sstevel@tonic-gate {
8977c478bd9Sstevel@tonic-gate 	restarter_inst_t *inst;
8987c478bd9Sstevel@tonic-gate 	int r;
8997c478bd9Sstevel@tonic-gate 
9007c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&instance_list.ril_lock);
9017c478bd9Sstevel@tonic-gate 
9027c478bd9Sstevel@tonic-gate 	for (inst = uu_list_first(instance_list.ril_instance_list);
9037c478bd9Sstevel@tonic-gate 	    inst != NULL;
9047c478bd9Sstevel@tonic-gate 	    inst = uu_list_next(instance_list.ril_instance_list, inst)) {
9057c478bd9Sstevel@tonic-gate 		const char *fmri;
9067c478bd9Sstevel@tonic-gate 		scf_instance_t *sinst = NULL;
9077c478bd9Sstevel@tonic-gate 
9087c478bd9Sstevel@tonic-gate 		MUTEX_LOCK(&inst->ri_lock);
9097c478bd9Sstevel@tonic-gate 
9107c478bd9Sstevel@tonic-gate 		/*
9117c478bd9Sstevel@tonic-gate 		 * This is where we'd check inst->ri_method_thread and if it
9127c478bd9Sstevel@tonic-gate 		 * were nonzero we'd wait in anticipation of another thread
9137c478bd9Sstevel@tonic-gate 		 * executing a method for inst.  Doing so with the instance_list
9147c478bd9Sstevel@tonic-gate 		 * locked, though, leads to deadlock.  Since taking a snapshot
9157c478bd9Sstevel@tonic-gate 		 * during that window won't hurt anything, we'll just continue.
9167c478bd9Sstevel@tonic-gate 		 */
9177c478bd9Sstevel@tonic-gate 
9187c478bd9Sstevel@tonic-gate 		fmri = inst->ri_i.i_fmri;
9197c478bd9Sstevel@tonic-gate 
9207c478bd9Sstevel@tonic-gate 		if (inst->ri_flags & RINST_RETAKE_RUNNING) {
9217c478bd9Sstevel@tonic-gate 			scf_snapshot_t *rsnap;
9227c478bd9Sstevel@tonic-gate 
9237c478bd9Sstevel@tonic-gate 			(void) libscf_fmri_get_instance(h, fmri, &sinst);
9247c478bd9Sstevel@tonic-gate 
9257c478bd9Sstevel@tonic-gate 			rsnap = libscf_get_or_make_running_snapshot(sinst,
9267c478bd9Sstevel@tonic-gate 			    fmri, B_FALSE);
9277c478bd9Sstevel@tonic-gate 
9287c478bd9Sstevel@tonic-gate 			scf_instance_destroy(sinst);
9297c478bd9Sstevel@tonic-gate 
9307c478bd9Sstevel@tonic-gate 			if (rsnap != NULL)
9317c478bd9Sstevel@tonic-gate 				inst->ri_flags &= ~RINST_RETAKE_RUNNING;
9327c478bd9Sstevel@tonic-gate 
9337c478bd9Sstevel@tonic-gate 			scf_snapshot_destroy(rsnap);
9347c478bd9Sstevel@tonic-gate 		}
9357c478bd9Sstevel@tonic-gate 
9367c478bd9Sstevel@tonic-gate 		if (inst->ri_flags & RINST_RETAKE_START) {
9377c478bd9Sstevel@tonic-gate 			switch (r = libscf_snapshots_poststart(h, fmri,
9387c478bd9Sstevel@tonic-gate 			    B_FALSE)) {
9397c478bd9Sstevel@tonic-gate 			case 0:
9407c478bd9Sstevel@tonic-gate 			case ENOENT:
9417c478bd9Sstevel@tonic-gate 				inst->ri_flags &= ~RINST_RETAKE_START;
9427c478bd9Sstevel@tonic-gate 				break;
9437c478bd9Sstevel@tonic-gate 
9447c478bd9Sstevel@tonic-gate 			case ECONNABORTED:
9457c478bd9Sstevel@tonic-gate 				break;
9467c478bd9Sstevel@tonic-gate 
9477c478bd9Sstevel@tonic-gate 			case EACCES:
9487c478bd9Sstevel@tonic-gate 			default:
9497c478bd9Sstevel@tonic-gate 				bad_error("libscf_snapshots_poststart", r);
9507c478bd9Sstevel@tonic-gate 			}
9517c478bd9Sstevel@tonic-gate 		}
9527c478bd9Sstevel@tonic-gate 
9537c478bd9Sstevel@tonic-gate 		MUTEX_UNLOCK(&inst->ri_lock);
9547c478bd9Sstevel@tonic-gate 	}
9557c478bd9Sstevel@tonic-gate 
9567c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&instance_list.ril_lock);
9577c478bd9Sstevel@tonic-gate }
9587c478bd9Sstevel@tonic-gate 
9597c478bd9Sstevel@tonic-gate /* ARGSUSED */
9607c478bd9Sstevel@tonic-gate void *
9617c478bd9Sstevel@tonic-gate restarter_post_fsminimal_thread(void *unused)
9627c478bd9Sstevel@tonic-gate {
9637c478bd9Sstevel@tonic-gate 	scf_handle_t *h;
9647c478bd9Sstevel@tonic-gate 	int r;
9657c478bd9Sstevel@tonic-gate 
9667c478bd9Sstevel@tonic-gate 	h = libscf_handle_create_bound_loop();
9677c478bd9Sstevel@tonic-gate 
9687c478bd9Sstevel@tonic-gate 	for (;;) {
9697c478bd9Sstevel@tonic-gate 		r = libscf_create_self(h);
9707c478bd9Sstevel@tonic-gate 		if (r == 0)
9717c478bd9Sstevel@tonic-gate 			break;
9727c478bd9Sstevel@tonic-gate 
9737c478bd9Sstevel@tonic-gate 		assert(r == ECONNABORTED);
9747c478bd9Sstevel@tonic-gate 		libscf_handle_rebind(h);
9757c478bd9Sstevel@tonic-gate 	}
9767c478bd9Sstevel@tonic-gate 
9777c478bd9Sstevel@tonic-gate 	restarter_take_pending_snapshots(h);
9787c478bd9Sstevel@tonic-gate 
9797c478bd9Sstevel@tonic-gate 	(void) scf_handle_unbind(h);
9807c478bd9Sstevel@tonic-gate 	scf_handle_destroy(h);
9817c478bd9Sstevel@tonic-gate 
9827c478bd9Sstevel@tonic-gate 	return (NULL);
9837c478bd9Sstevel@tonic-gate }
9847c478bd9Sstevel@tonic-gate 
9857c478bd9Sstevel@tonic-gate /*
9867c478bd9Sstevel@tonic-gate  * int stop_instance()
9877c478bd9Sstevel@tonic-gate  *
9887c478bd9Sstevel@tonic-gate  *   Stop the instance identified by the instance given as the second argument,
9897c478bd9Sstevel@tonic-gate  *   for the cause stated.
9907c478bd9Sstevel@tonic-gate  *
9917c478bd9Sstevel@tonic-gate  *   Returns
9927c478bd9Sstevel@tonic-gate  *     0 - success
9937c478bd9Sstevel@tonic-gate  *     -1 - inst is in transition
9947c478bd9Sstevel@tonic-gate  */
9957c478bd9Sstevel@tonic-gate static int
9967c478bd9Sstevel@tonic-gate stop_instance(scf_handle_t *local_handle, restarter_inst_t *inst,
9977c478bd9Sstevel@tonic-gate     stop_cause_t cause)
9987c478bd9Sstevel@tonic-gate {
9997c478bd9Sstevel@tonic-gate 	fork_info_t *info;
10007c478bd9Sstevel@tonic-gate 	const char *cp;
10017c478bd9Sstevel@tonic-gate 	int err;
10027c478bd9Sstevel@tonic-gate 	restarter_error_t re;
10037c478bd9Sstevel@tonic-gate 
100453f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&inst->ri_lock));
10057c478bd9Sstevel@tonic-gate 	assert(inst->ri_method_thread == 0);
10067c478bd9Sstevel@tonic-gate 
10077c478bd9Sstevel@tonic-gate 	switch (cause) {
10087c478bd9Sstevel@tonic-gate 	case RSTOP_EXIT:
10097c478bd9Sstevel@tonic-gate 		re = RERR_RESTART;
10107c478bd9Sstevel@tonic-gate 		cp = "all processes in service exited";
10117c478bd9Sstevel@tonic-gate 		break;
10127c478bd9Sstevel@tonic-gate 	case RSTOP_CORE:
10137c478bd9Sstevel@tonic-gate 		re = RERR_FAULT;
10147c478bd9Sstevel@tonic-gate 		cp = "process dumped core";
10157c478bd9Sstevel@tonic-gate 		break;
10167c478bd9Sstevel@tonic-gate 	case RSTOP_SIGNAL:
10177c478bd9Sstevel@tonic-gate 		re = RERR_FAULT;
10187c478bd9Sstevel@tonic-gate 		cp = "process received fatal signal from outside the service";
10197c478bd9Sstevel@tonic-gate 		break;
10207c478bd9Sstevel@tonic-gate 	case RSTOP_HWERR:
10217c478bd9Sstevel@tonic-gate 		re = RERR_FAULT;
10227c478bd9Sstevel@tonic-gate 		cp = "process killed due to uncorrectable hardware error";
10237c478bd9Sstevel@tonic-gate 		break;
10247c478bd9Sstevel@tonic-gate 	case RSTOP_DEPENDENCY:
10257c478bd9Sstevel@tonic-gate 		re = RERR_RESTART;
10267c478bd9Sstevel@tonic-gate 		cp = "dependency activity requires stop";
10277c478bd9Sstevel@tonic-gate 		break;
10287c478bd9Sstevel@tonic-gate 	case RSTOP_DISABLE:
10297c478bd9Sstevel@tonic-gate 		re = RERR_RESTART;
10307c478bd9Sstevel@tonic-gate 		cp = "service disabled";
10317c478bd9Sstevel@tonic-gate 		break;
10327c478bd9Sstevel@tonic-gate 	case RSTOP_RESTART:
10337c478bd9Sstevel@tonic-gate 		re = RERR_RESTART;
10347c478bd9Sstevel@tonic-gate 		cp = "service restarting";
10357c478bd9Sstevel@tonic-gate 		break;
10367c478bd9Sstevel@tonic-gate 	default:
10377c478bd9Sstevel@tonic-gate #ifndef NDEBUG
10387c478bd9Sstevel@tonic-gate 		(void) fprintf(stderr, "Unknown cause %d at %s:%d.\n",
10397c478bd9Sstevel@tonic-gate 		    cause, __FILE__, __LINE__);
10407c478bd9Sstevel@tonic-gate #endif
10417c478bd9Sstevel@tonic-gate 		abort();
10427c478bd9Sstevel@tonic-gate 	}
10437c478bd9Sstevel@tonic-gate 
10447c478bd9Sstevel@tonic-gate 	/* Services in the disabled and maintenance state are ignored */
10457c478bd9Sstevel@tonic-gate 	if (inst->ri_i.i_state == RESTARTER_STATE_MAINT ||
10467c478bd9Sstevel@tonic-gate 	    inst->ri_i.i_state == RESTARTER_STATE_DISABLED) {
10477c478bd9Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
10487c478bd9Sstevel@tonic-gate 		    "%s: stop_instance -> is maint/disabled\n",
10497c478bd9Sstevel@tonic-gate 		    inst->ri_i.i_fmri);
10507c478bd9Sstevel@tonic-gate 		return (0);
10517c478bd9Sstevel@tonic-gate 	}
10527c478bd9Sstevel@tonic-gate 
10537c478bd9Sstevel@tonic-gate 	/* Already stopped instances are left alone */
10547c478bd9Sstevel@tonic-gate 	if (instance_started(inst) == 0) {
10557c478bd9Sstevel@tonic-gate 		log_framework(LOG_DEBUG, "Restarter: %s is already stopped.\n",
10567c478bd9Sstevel@tonic-gate 		    inst->ri_i.i_fmri);
10577c478bd9Sstevel@tonic-gate 		return (0);
10587c478bd9Sstevel@tonic-gate 	}
10597c478bd9Sstevel@tonic-gate 
10607c478bd9Sstevel@tonic-gate 	if (instance_in_transition(inst)) {
10617c478bd9Sstevel@tonic-gate 		/* requeue event by returning -1 */
10627c478bd9Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
10637c478bd9Sstevel@tonic-gate 		    "Restarter: Not stopping %s, in transition.\n",
10647c478bd9Sstevel@tonic-gate 		    inst->ri_i.i_fmri);
10657c478bd9Sstevel@tonic-gate 		return (-1);
10667c478bd9Sstevel@tonic-gate 	}
10677c478bd9Sstevel@tonic-gate 
10687c478bd9Sstevel@tonic-gate 	log_instance(inst, B_TRUE, "Stopping because %s.", cp);
10697c478bd9Sstevel@tonic-gate 
10707c478bd9Sstevel@tonic-gate 	log_framework(re == RERR_FAULT ? LOG_INFO : LOG_DEBUG,
10717c478bd9Sstevel@tonic-gate 	    "%s: Instance stopping because %s.\n", inst->ri_i.i_fmri, cp);
10727c478bd9Sstevel@tonic-gate 
10737c478bd9Sstevel@tonic-gate 	if (instance_is_wait_style(inst) && cause == RSTOP_EXIT) {
10747c478bd9Sstevel@tonic-gate 		/*
10757c478bd9Sstevel@tonic-gate 		 * No need to stop instance, as child has exited; remove
10767c478bd9Sstevel@tonic-gate 		 * contract and move the instance to the offline state.
10777c478bd9Sstevel@tonic-gate 		 */
10787c478bd9Sstevel@tonic-gate 		switch (err = restarter_instance_update_states(local_handle,
10797c478bd9Sstevel@tonic-gate 		    inst, inst->ri_i.i_state, RESTARTER_STATE_OFFLINE, re,
10807c478bd9Sstevel@tonic-gate 		    NULL)) {
10817c478bd9Sstevel@tonic-gate 		case 0:
10827c478bd9Sstevel@tonic-gate 		case ECONNRESET:
10837c478bd9Sstevel@tonic-gate 			break;
10847c478bd9Sstevel@tonic-gate 
10857c478bd9Sstevel@tonic-gate 		default:
10867c478bd9Sstevel@tonic-gate 			bad_error("restarter_instance_update_states", err);
10877c478bd9Sstevel@tonic-gate 		}
10887c478bd9Sstevel@tonic-gate 
10897c478bd9Sstevel@tonic-gate 		(void) update_fault_count(inst, FAULT_COUNT_RESET);
1090*c238c833SSean Wilcox 		reset_start_times(inst);
10917c478bd9Sstevel@tonic-gate 
10927c478bd9Sstevel@tonic-gate 		if (inst->ri_i.i_primary_ctid != 0) {
10937c478bd9Sstevel@tonic-gate 			inst->ri_m_inst =
10947c478bd9Sstevel@tonic-gate 			    safe_scf_instance_create(local_handle);
10957c478bd9Sstevel@tonic-gate 			inst->ri_mi_deleted = B_FALSE;
10967c478bd9Sstevel@tonic-gate 
10977c478bd9Sstevel@tonic-gate 			libscf_reget_instance(inst);
10987c478bd9Sstevel@tonic-gate 			method_remove_contract(inst, B_TRUE, B_TRUE);
10997c478bd9Sstevel@tonic-gate 
11007c478bd9Sstevel@tonic-gate 			scf_instance_destroy(inst->ri_m_inst);
11017c478bd9Sstevel@tonic-gate 			inst->ri_m_inst = NULL;
11027c478bd9Sstevel@tonic-gate 		}
11037c478bd9Sstevel@tonic-gate 
11047c478bd9Sstevel@tonic-gate 		switch (err = restarter_instance_update_states(local_handle,
11057c478bd9Sstevel@tonic-gate 		    inst, inst->ri_i.i_next_state, RESTARTER_STATE_NONE, re,
11067c478bd9Sstevel@tonic-gate 		    NULL)) {
11077c478bd9Sstevel@tonic-gate 		case 0:
11087c478bd9Sstevel@tonic-gate 		case ECONNRESET:
11097c478bd9Sstevel@tonic-gate 			break;
11107c478bd9Sstevel@tonic-gate 
11117c478bd9Sstevel@tonic-gate 		default:
11127c478bd9Sstevel@tonic-gate 			bad_error("restarter_instance_update_states", err);
11137c478bd9Sstevel@tonic-gate 		}
11147c478bd9Sstevel@tonic-gate 
11157c478bd9Sstevel@tonic-gate 		return (0);
111663c7f71bSrm88369 	} else if (instance_is_wait_style(inst) && re == RERR_RESTART) {
111763c7f71bSrm88369 		/*
111863c7f71bSrm88369 		 * Stopping a wait service through means other than the pid
111963c7f71bSrm88369 		 * exiting should keep wait_thread() from restarting the
112063c7f71bSrm88369 		 * service, by removing it from the wait list.
112163c7f71bSrm88369 		 * We cannot remove it right now otherwise the process will
112263c7f71bSrm88369 		 * end up <defunct> so mark it to be ignored.
112363c7f71bSrm88369 		 */
112463c7f71bSrm88369 		wait_ignore_by_fmri(inst->ri_i.i_fmri);
11257c478bd9Sstevel@tonic-gate 	}
11267c478bd9Sstevel@tonic-gate 
11277c478bd9Sstevel@tonic-gate 	switch (err = restarter_instance_update_states(local_handle, inst,
11287c478bd9Sstevel@tonic-gate 	    inst->ri_i.i_state, inst->ri_i.i_enabled ? RESTARTER_STATE_OFFLINE :
11297c478bd9Sstevel@tonic-gate 	    RESTARTER_STATE_DISABLED, RERR_NONE, NULL)) {
11307c478bd9Sstevel@tonic-gate 	case 0:
11317c478bd9Sstevel@tonic-gate 	case ECONNRESET:
11327c478bd9Sstevel@tonic-gate 		break;
11337c478bd9Sstevel@tonic-gate 
11347c478bd9Sstevel@tonic-gate 	default:
11357c478bd9Sstevel@tonic-gate 		bad_error("restarter_instance_update_states", err);
11367c478bd9Sstevel@tonic-gate 	}
11377c478bd9Sstevel@tonic-gate 
11387c478bd9Sstevel@tonic-gate 	info = startd_zalloc(sizeof (fork_info_t));
11397c478bd9Sstevel@tonic-gate 
11407c478bd9Sstevel@tonic-gate 	info->sf_id = inst->ri_id;
11417c478bd9Sstevel@tonic-gate 	info->sf_method_type = METHOD_STOP;
11427c478bd9Sstevel@tonic-gate 	info->sf_event_type = re;
11437c478bd9Sstevel@tonic-gate 	inst->ri_method_thread = startd_thread_create(method_thread, info);
11447c478bd9Sstevel@tonic-gate 
11457c478bd9Sstevel@tonic-gate 	return (0);
11467c478bd9Sstevel@tonic-gate }
11477c478bd9Sstevel@tonic-gate 
11487c478bd9Sstevel@tonic-gate /*
11497c478bd9Sstevel@tonic-gate  * Returns
11507c478bd9Sstevel@tonic-gate  *   ENOENT - fmri is not in instance_list
11517c478bd9Sstevel@tonic-gate  *   0 - success
11527c478bd9Sstevel@tonic-gate  *   ECONNRESET - success, though handle was rebound
11537c478bd9Sstevel@tonic-gate  *   -1 - instance is in transition
11547c478bd9Sstevel@tonic-gate  */
11557c478bd9Sstevel@tonic-gate int
11567c478bd9Sstevel@tonic-gate stop_instance_fmri(scf_handle_t *h, const char *fmri, uint_t flags)
11577c478bd9Sstevel@tonic-gate {
11587c478bd9Sstevel@tonic-gate 	restarter_inst_t *rip;
11597c478bd9Sstevel@tonic-gate 	int r;
11607c478bd9Sstevel@tonic-gate 
11617c478bd9Sstevel@tonic-gate 	rip = inst_lookup_by_name(fmri);
11627c478bd9Sstevel@tonic-gate 	if (rip == NULL)
11637c478bd9Sstevel@tonic-gate 		return (ENOENT);
11647c478bd9Sstevel@tonic-gate 
11657c478bd9Sstevel@tonic-gate 	r = stop_instance(h, rip, flags);
11667c478bd9Sstevel@tonic-gate 
11677c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&rip->ri_lock);
11687c478bd9Sstevel@tonic-gate 
11697c478bd9Sstevel@tonic-gate 	return (r);
11707c478bd9Sstevel@tonic-gate }
11717c478bd9Sstevel@tonic-gate 
11727c478bd9Sstevel@tonic-gate static void
11737c478bd9Sstevel@tonic-gate unmaintain_instance(scf_handle_t *h, restarter_inst_t *rip,
11747c478bd9Sstevel@tonic-gate     unmaint_cause_t cause)
11757c478bd9Sstevel@tonic-gate {
11767c478bd9Sstevel@tonic-gate 	ctid_t ctid;
11777c478bd9Sstevel@tonic-gate 	scf_instance_t *inst;
11787c478bd9Sstevel@tonic-gate 	int r;
11797c478bd9Sstevel@tonic-gate 	uint_t tries = 0, msecs = ALLOC_DELAY;
11807c478bd9Sstevel@tonic-gate 	const char *cp;
11817c478bd9Sstevel@tonic-gate 
118253f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&rip->ri_lock));
11837c478bd9Sstevel@tonic-gate 
11847c478bd9Sstevel@tonic-gate 	if (rip->ri_i.i_state != RESTARTER_STATE_MAINT) {
11857c478bd9Sstevel@tonic-gate 		log_error(LOG_DEBUG, "Restarter: "
11867c478bd9Sstevel@tonic-gate 		    "Ignoring maintenance off command because %s is not in the "
11877c478bd9Sstevel@tonic-gate 		    "maintenance state.\n", rip->ri_i.i_fmri);
11887c478bd9Sstevel@tonic-gate 		return;
11897c478bd9Sstevel@tonic-gate 	}
11907c478bd9Sstevel@tonic-gate 
11917c478bd9Sstevel@tonic-gate 	switch (cause) {
11927c478bd9Sstevel@tonic-gate 	case RUNMAINT_CLEAR:
11937c478bd9Sstevel@tonic-gate 		cp = "clear requested";
11947c478bd9Sstevel@tonic-gate 		break;
11957c478bd9Sstevel@tonic-gate 	case RUNMAINT_DISABLE:
11967c478bd9Sstevel@tonic-gate 		cp = "disable requested";
11977c478bd9Sstevel@tonic-gate 		break;
11987c478bd9Sstevel@tonic-gate 	default:
11997c478bd9Sstevel@tonic-gate #ifndef NDEBUG
12007c478bd9Sstevel@tonic-gate 		(void) fprintf(stderr, "Uncaught case for %d at %s:%d.\n",
12017c478bd9Sstevel@tonic-gate 		    cause, __FILE__, __LINE__);
12027c478bd9Sstevel@tonic-gate #endif
12037c478bd9Sstevel@tonic-gate 		abort();
12047c478bd9Sstevel@tonic-gate 	}
12057c478bd9Sstevel@tonic-gate 
12067c478bd9Sstevel@tonic-gate 	log_instance(rip, B_TRUE, "Leaving maintenance because %s.",
12077c478bd9Sstevel@tonic-gate 	    cp);
12087c478bd9Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: Instance leaving maintenance because "
12097c478bd9Sstevel@tonic-gate 	    "%s.\n", rip->ri_i.i_fmri, cp);
12107c478bd9Sstevel@tonic-gate 
12117c478bd9Sstevel@tonic-gate 	(void) restarter_instance_update_states(h, rip, RESTARTER_STATE_UNINIT,
1212eb1a3463STruong Nguyen 	    RESTARTER_STATE_NONE, RERR_RESTART, "none");
12137c478bd9Sstevel@tonic-gate 
12147c478bd9Sstevel@tonic-gate 	/*
12157c478bd9Sstevel@tonic-gate 	 * If we did ADMIN_MAINT_ON_IMMEDIATE, then there might still be
12167c478bd9Sstevel@tonic-gate 	 * a primary contract.
12177c478bd9Sstevel@tonic-gate 	 */
12187c478bd9Sstevel@tonic-gate 	if (rip->ri_i.i_primary_ctid == 0)
12197c478bd9Sstevel@tonic-gate 		return;
12207c478bd9Sstevel@tonic-gate 
12217c478bd9Sstevel@tonic-gate 	ctid = rip->ri_i.i_primary_ctid;
12227c478bd9Sstevel@tonic-gate 	contract_abandon(ctid);
12237c478bd9Sstevel@tonic-gate 	rip->ri_i.i_primary_ctid = 0;
12247c478bd9Sstevel@tonic-gate 
12257c478bd9Sstevel@tonic-gate rep_retry:
12267c478bd9Sstevel@tonic-gate 	switch (r = libscf_fmri_get_instance(h, rip->ri_i.i_fmri, &inst)) {
12277c478bd9Sstevel@tonic-gate 	case 0:
12287c478bd9Sstevel@tonic-gate 		break;
12297c478bd9Sstevel@tonic-gate 
12307c478bd9Sstevel@tonic-gate 	case ECONNABORTED:
12317c478bd9Sstevel@tonic-gate 		libscf_handle_rebind(h);
12327c478bd9Sstevel@tonic-gate 		goto rep_retry;
12337c478bd9Sstevel@tonic-gate 
12347c478bd9Sstevel@tonic-gate 	case ENOENT:
12357c478bd9Sstevel@tonic-gate 		/* Must have been deleted. */
12367c478bd9Sstevel@tonic-gate 		return;
12377c478bd9Sstevel@tonic-gate 
12387c478bd9Sstevel@tonic-gate 	case EINVAL:
12397c478bd9Sstevel@tonic-gate 	case ENOTSUP:
12407c478bd9Sstevel@tonic-gate 	default:
12417c478bd9Sstevel@tonic-gate 		bad_error("libscf_handle_rebind", r);
12427c478bd9Sstevel@tonic-gate 	}
12437c478bd9Sstevel@tonic-gate 
12447c478bd9Sstevel@tonic-gate again:
12457c478bd9Sstevel@tonic-gate 	r = restarter_remove_contract(inst, ctid, RESTARTER_CONTRACT_PRIMARY);
12467c478bd9Sstevel@tonic-gate 	switch (r) {
12477c478bd9Sstevel@tonic-gate 	case 0:
12487c478bd9Sstevel@tonic-gate 		break;
12497c478bd9Sstevel@tonic-gate 
12507c478bd9Sstevel@tonic-gate 	case ENOMEM:
12517c478bd9Sstevel@tonic-gate 		++tries;
12527c478bd9Sstevel@tonic-gate 		if (tries < ALLOC_RETRY) {
12537c478bd9Sstevel@tonic-gate 			(void) poll(NULL, 0, msecs);
12547c478bd9Sstevel@tonic-gate 			msecs *= ALLOC_DELAY_MULT;
12557c478bd9Sstevel@tonic-gate 			goto again;
12567c478bd9Sstevel@tonic-gate 		}
12577c478bd9Sstevel@tonic-gate 
12587c478bd9Sstevel@tonic-gate 		uu_die("Insufficient memory.\n");
12597c478bd9Sstevel@tonic-gate 		/* NOTREACHED */
12607c478bd9Sstevel@tonic-gate 
12617c478bd9Sstevel@tonic-gate 	case ECONNABORTED:
12627c478bd9Sstevel@tonic-gate 		scf_instance_destroy(inst);
12637c478bd9Sstevel@tonic-gate 		libscf_handle_rebind(h);
12647c478bd9Sstevel@tonic-gate 		goto rep_retry;
12657c478bd9Sstevel@tonic-gate 
12667c478bd9Sstevel@tonic-gate 	case ECANCELED:
12677c478bd9Sstevel@tonic-gate 		break;
12687c478bd9Sstevel@tonic-gate 
12697c478bd9Sstevel@tonic-gate 	case EPERM:
12707c478bd9Sstevel@tonic-gate 	case EACCES:
12717c478bd9Sstevel@tonic-gate 	case EROFS:
12727c478bd9Sstevel@tonic-gate 		log_error(LOG_INFO,
12737c478bd9Sstevel@tonic-gate 		    "Could not remove contract id %lu for %s (%s).\n", ctid,
12747c478bd9Sstevel@tonic-gate 		    rip->ri_i.i_fmri, strerror(r));
12757c478bd9Sstevel@tonic-gate 		break;
12767c478bd9Sstevel@tonic-gate 
12777c478bd9Sstevel@tonic-gate 	case EINVAL:
12787c478bd9Sstevel@tonic-gate 	case EBADF:
12797c478bd9Sstevel@tonic-gate 	default:
12807c478bd9Sstevel@tonic-gate 		bad_error("restarter_remove_contract", r);
12817c478bd9Sstevel@tonic-gate 	}
12827c478bd9Sstevel@tonic-gate 
12837c478bd9Sstevel@tonic-gate 	scf_instance_destroy(inst);
12847c478bd9Sstevel@tonic-gate }
12857c478bd9Sstevel@tonic-gate 
12867c478bd9Sstevel@tonic-gate /*
12877c478bd9Sstevel@tonic-gate  * enable_inst()
12887c478bd9Sstevel@tonic-gate  *   Set inst->ri_i.i_enabled.  Expects 'e' to be _ENABLE, _DISABLE, or
12897c478bd9Sstevel@tonic-gate  *   _ADMIN_DISABLE.  If the event is _ENABLE and inst is uninitialized or
12907c478bd9Sstevel@tonic-gate  *   disabled, move it to offline.  If the event is _DISABLE or
12917c478bd9Sstevel@tonic-gate  *   _ADMIN_DISABLE, make sure inst will move to disabled.
12927c478bd9Sstevel@tonic-gate  *
12937c478bd9Sstevel@tonic-gate  *   Returns
12947c478bd9Sstevel@tonic-gate  *     0 - success
12957c478bd9Sstevel@tonic-gate  *     ECONNRESET - h was rebound
12967c478bd9Sstevel@tonic-gate  */
12977c478bd9Sstevel@tonic-gate static int
12987c478bd9Sstevel@tonic-gate enable_inst(scf_handle_t *h, restarter_inst_t *inst, restarter_event_type_t e)
12997c478bd9Sstevel@tonic-gate {
13007c478bd9Sstevel@tonic-gate 	restarter_instance_state_t state;
13017c478bd9Sstevel@tonic-gate 	int r;
13027c478bd9Sstevel@tonic-gate 
130353f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&inst->ri_lock));
13047c478bd9Sstevel@tonic-gate 	assert(e == RESTARTER_EVENT_TYPE_ADMIN_DISABLE ||
13057c478bd9Sstevel@tonic-gate 	    e == RESTARTER_EVENT_TYPE_DISABLE ||
13067c478bd9Sstevel@tonic-gate 	    e == RESTARTER_EVENT_TYPE_ENABLE);
13077c478bd9Sstevel@tonic-gate 	assert(instance_in_transition(inst) == 0);
13087c478bd9Sstevel@tonic-gate 
13097c478bd9Sstevel@tonic-gate 	state = inst->ri_i.i_state;
13107c478bd9Sstevel@tonic-gate 
13117c478bd9Sstevel@tonic-gate 	if (e == RESTARTER_EVENT_TYPE_ENABLE) {
13127c478bd9Sstevel@tonic-gate 		inst->ri_i.i_enabled = 1;
13137c478bd9Sstevel@tonic-gate 
13147c478bd9Sstevel@tonic-gate 		if (state == RESTARTER_STATE_UNINIT ||
13157c478bd9Sstevel@tonic-gate 		    state == RESTARTER_STATE_DISABLED) {
13167c478bd9Sstevel@tonic-gate 			/*
13177c478bd9Sstevel@tonic-gate 			 * B_FALSE: Don't log an error if the log_instance()
13187c478bd9Sstevel@tonic-gate 			 * fails because it will fail on the miniroot before
13197c478bd9Sstevel@tonic-gate 			 * install-discovery runs.
13207c478bd9Sstevel@tonic-gate 			 */
13217c478bd9Sstevel@tonic-gate 			log_instance(inst, B_FALSE, "Enabled.");
13227c478bd9Sstevel@tonic-gate 			log_framework(LOG_DEBUG, "%s: Instance enabled.\n",
13237c478bd9Sstevel@tonic-gate 			    inst->ri_i.i_fmri);
13247c478bd9Sstevel@tonic-gate 			(void) restarter_instance_update_states(h, inst,
13257c478bd9Sstevel@tonic-gate 			    RESTARTER_STATE_OFFLINE, RESTARTER_STATE_NONE,
13267c478bd9Sstevel@tonic-gate 			    RERR_NONE, NULL);
13277c478bd9Sstevel@tonic-gate 		} else {
13287c478bd9Sstevel@tonic-gate 			log_framework(LOG_DEBUG, "Restarter: "
13297c478bd9Sstevel@tonic-gate 			    "Not changing state of %s for enable command.\n",
13307c478bd9Sstevel@tonic-gate 			    inst->ri_i.i_fmri);
13317c478bd9Sstevel@tonic-gate 		}
13327c478bd9Sstevel@tonic-gate 	} else {
13337c478bd9Sstevel@tonic-gate 		inst->ri_i.i_enabled = 0;
13347c478bd9Sstevel@tonic-gate 
13357c478bd9Sstevel@tonic-gate 		switch (state) {
13367c478bd9Sstevel@tonic-gate 		case RESTARTER_STATE_ONLINE:
13377c478bd9Sstevel@tonic-gate 		case RESTARTER_STATE_DEGRADED:
13387c478bd9Sstevel@tonic-gate 			r = stop_instance(h, inst, RSTOP_DISABLE);
13397c478bd9Sstevel@tonic-gate 			return (r == ECONNRESET ? 0 : r);
13407c478bd9Sstevel@tonic-gate 
13417c478bd9Sstevel@tonic-gate 		case RESTARTER_STATE_OFFLINE:
13427c478bd9Sstevel@tonic-gate 		case RESTARTER_STATE_UNINIT:
13437c478bd9Sstevel@tonic-gate 			if (inst->ri_i.i_primary_ctid != 0) {
13447c478bd9Sstevel@tonic-gate 				inst->ri_m_inst = safe_scf_instance_create(h);
13457c478bd9Sstevel@tonic-gate 				inst->ri_mi_deleted = B_FALSE;
13467c478bd9Sstevel@tonic-gate 
13477c478bd9Sstevel@tonic-gate 				libscf_reget_instance(inst);
13487c478bd9Sstevel@tonic-gate 				method_remove_contract(inst, B_TRUE, B_TRUE);
13497c478bd9Sstevel@tonic-gate 
13507c478bd9Sstevel@tonic-gate 				scf_instance_destroy(inst->ri_m_inst);
13517c478bd9Sstevel@tonic-gate 			}
13527c478bd9Sstevel@tonic-gate 			/* B_FALSE: See log_instance(..., "Enabled."); above */
13537c478bd9Sstevel@tonic-gate 			log_instance(inst, B_FALSE, "Disabled.");
13547c478bd9Sstevel@tonic-gate 			log_framework(LOG_DEBUG, "%s: Instance disabled.\n",
13557c478bd9Sstevel@tonic-gate 			    inst->ri_i.i_fmri);
13567c478bd9Sstevel@tonic-gate 			(void) restarter_instance_update_states(h, inst,
13577c478bd9Sstevel@tonic-gate 			    RESTARTER_STATE_DISABLED, RESTARTER_STATE_NONE,
13587c478bd9Sstevel@tonic-gate 			    RERR_RESTART, NULL);
13597c478bd9Sstevel@tonic-gate 			return (0);
13607c478bd9Sstevel@tonic-gate 
13617c478bd9Sstevel@tonic-gate 		case RESTARTER_STATE_DISABLED:
13627c478bd9Sstevel@tonic-gate 			break;
13637c478bd9Sstevel@tonic-gate 
13647c478bd9Sstevel@tonic-gate 		case RESTARTER_STATE_MAINT:
13657c478bd9Sstevel@tonic-gate 			/*
13667c478bd9Sstevel@tonic-gate 			 * We only want to pull the instance out of maintenance
13677c478bd9Sstevel@tonic-gate 			 * if the disable is on adminstrative request.  The
13687c478bd9Sstevel@tonic-gate 			 * graph engine sends _DISABLE events whenever a
13697c478bd9Sstevel@tonic-gate 			 * service isn't in the disabled state, and we don't
13707c478bd9Sstevel@tonic-gate 			 * want to pull the service out of maintenance if,
13717c478bd9Sstevel@tonic-gate 			 * for example, it is there due to a dependency cycle.
13727c478bd9Sstevel@tonic-gate 			 */
13737c478bd9Sstevel@tonic-gate 			if (e == RESTARTER_EVENT_TYPE_ADMIN_DISABLE)
13747c478bd9Sstevel@tonic-gate 				unmaintain_instance(h, inst, RUNMAINT_DISABLE);
13757c478bd9Sstevel@tonic-gate 			break;
13767c478bd9Sstevel@tonic-gate 
13777c478bd9Sstevel@tonic-gate 		default:
13787c478bd9Sstevel@tonic-gate #ifndef NDEBUG
13797c478bd9Sstevel@tonic-gate 			(void) fprintf(stderr, "Restarter instance %s has "
13807c478bd9Sstevel@tonic-gate 			    "unknown state %d.\n", inst->ri_i.i_fmri, state);
13817c478bd9Sstevel@tonic-gate #endif
13827c478bd9Sstevel@tonic-gate 			abort();
13837c478bd9Sstevel@tonic-gate 		}
13847c478bd9Sstevel@tonic-gate 	}
13857c478bd9Sstevel@tonic-gate 
13867c478bd9Sstevel@tonic-gate 	return (0);
13877c478bd9Sstevel@tonic-gate }
13887c478bd9Sstevel@tonic-gate 
13897c478bd9Sstevel@tonic-gate static void
13907c478bd9Sstevel@tonic-gate start_instance(scf_handle_t *local_handle, restarter_inst_t *inst)
13917c478bd9Sstevel@tonic-gate {
13927c478bd9Sstevel@tonic-gate 	fork_info_t *info;
13937c478bd9Sstevel@tonic-gate 
139453f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&inst->ri_lock));
13957c478bd9Sstevel@tonic-gate 	assert(instance_in_transition(inst) == 0);
13967c478bd9Sstevel@tonic-gate 	assert(inst->ri_method_thread == 0);
13977c478bd9Sstevel@tonic-gate 
13987c478bd9Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: trying to start instance\n",
13997c478bd9Sstevel@tonic-gate 	    inst->ri_i.i_fmri);
14007c478bd9Sstevel@tonic-gate 
14017c478bd9Sstevel@tonic-gate 	/* Services in the disabled and maintenance state are ignored */
14027c478bd9Sstevel@tonic-gate 	if (inst->ri_i.i_state == RESTARTER_STATE_MAINT ||
14037c478bd9Sstevel@tonic-gate 	    inst->ri_i.i_state == RESTARTER_STATE_DISABLED ||
14047c478bd9Sstevel@tonic-gate 	    inst->ri_i.i_enabled == 0) {
14057c478bd9Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
14067c478bd9Sstevel@tonic-gate 		    "%s: start_instance -> is maint/disabled\n",
14077c478bd9Sstevel@tonic-gate 		    inst->ri_i.i_fmri);
14087c478bd9Sstevel@tonic-gate 		return;
14097c478bd9Sstevel@tonic-gate 	}
14107c478bd9Sstevel@tonic-gate 
14117c478bd9Sstevel@tonic-gate 	/* Already started instances are left alone */
14127c478bd9Sstevel@tonic-gate 	if (instance_started(inst) == 1) {
14137c478bd9Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
14147c478bd9Sstevel@tonic-gate 		    "%s: start_instance -> is already started\n",
14157c478bd9Sstevel@tonic-gate 		    inst->ri_i.i_fmri);
14167c478bd9Sstevel@tonic-gate 		return;
14177c478bd9Sstevel@tonic-gate 	}
14187c478bd9Sstevel@tonic-gate 
14197c478bd9Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: starting instance.\n", inst->ri_i.i_fmri);
14207c478bd9Sstevel@tonic-gate 
14217c478bd9Sstevel@tonic-gate 	(void) restarter_instance_update_states(local_handle, inst,
1422eb1a3463STruong Nguyen 	    inst->ri_i.i_state, RESTARTER_STATE_ONLINE, RERR_NONE, "none");
14237c478bd9Sstevel@tonic-gate 
14247c478bd9Sstevel@tonic-gate 	info = startd_zalloc(sizeof (fork_info_t));
14257c478bd9Sstevel@tonic-gate 
14267c478bd9Sstevel@tonic-gate 	info->sf_id = inst->ri_id;
14277c478bd9Sstevel@tonic-gate 	info->sf_method_type = METHOD_START;
14287c478bd9Sstevel@tonic-gate 	info->sf_event_type = RERR_NONE;
14297c478bd9Sstevel@tonic-gate 	inst->ri_method_thread = startd_thread_create(method_thread, info);
14307c478bd9Sstevel@tonic-gate }
14317c478bd9Sstevel@tonic-gate 
1432eb1a3463STruong Nguyen static int
1433eb1a3463STruong Nguyen event_from_tty(scf_handle_t *h, restarter_inst_t *rip)
1434eb1a3463STruong Nguyen {
1435eb1a3463STruong Nguyen 	scf_instance_t *inst;
1436eb1a3463STruong Nguyen 	int ret = 0;
1437eb1a3463STruong Nguyen 
1438eb1a3463STruong Nguyen 	if (libscf_fmri_get_instance(h, rip->ri_i.i_fmri, &inst))
1439eb1a3463STruong Nguyen 		return (-1);
1440eb1a3463STruong Nguyen 
1441eb1a3463STruong Nguyen 	ret = restarter_inst_ractions_from_tty(inst);
1442eb1a3463STruong Nguyen 
1443eb1a3463STruong Nguyen 	scf_instance_destroy(inst);
1444eb1a3463STruong Nguyen 	return (ret);
1445eb1a3463STruong Nguyen }
1446eb1a3463STruong Nguyen 
14477c478bd9Sstevel@tonic-gate static void
14487c478bd9Sstevel@tonic-gate maintain_instance(scf_handle_t *h, restarter_inst_t *rip, int immediate,
14497c478bd9Sstevel@tonic-gate     const char *aux)
14507c478bd9Sstevel@tonic-gate {
14517c478bd9Sstevel@tonic-gate 	fork_info_t *info;
1452eb1a3463STruong Nguyen 	scf_instance_t *scf_inst = NULL;
14537c478bd9Sstevel@tonic-gate 
145453f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&rip->ri_lock));
14557c478bd9Sstevel@tonic-gate 	assert(aux != NULL);
14567c478bd9Sstevel@tonic-gate 	assert(rip->ri_method_thread == 0);
14577c478bd9Sstevel@tonic-gate 
14587c478bd9Sstevel@tonic-gate 	log_instance(rip, B_TRUE, "Stopping for maintenance due to %s.", aux);
14597c478bd9Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: stopping for maintenance due to %s.\n",
14607c478bd9Sstevel@tonic-gate 	    rip->ri_i.i_fmri, aux);
14617c478bd9Sstevel@tonic-gate 
14627c478bd9Sstevel@tonic-gate 	/* Services in the maintenance state are ignored */
14637c478bd9Sstevel@tonic-gate 	if (rip->ri_i.i_state == RESTARTER_STATE_MAINT) {
14647c478bd9Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
14657c478bd9Sstevel@tonic-gate 		    "%s: maintain_instance -> is already in maintenance\n",
14667c478bd9Sstevel@tonic-gate 		    rip->ri_i.i_fmri);
14677c478bd9Sstevel@tonic-gate 		return;
14687c478bd9Sstevel@tonic-gate 	}
14697c478bd9Sstevel@tonic-gate 
1470eb1a3463STruong Nguyen 	/*
1471eb1a3463STruong Nguyen 	 * If aux state is "service_request" and
1472eb1a3463STruong Nguyen 	 * restarter_actions/auxiliary_fmri property is set with a valid fmri,
1473eb1a3463STruong Nguyen 	 * copy the fmri to restarter/auxiliary_fmri so svcs -x can use.
1474eb1a3463STruong Nguyen 	 */
1475eb1a3463STruong Nguyen 	if (strcmp(aux, "service_request") == 0 && libscf_fmri_get_instance(h,
1476eb1a3463STruong Nguyen 	    rip->ri_i.i_fmri, &scf_inst) == 0) {
1477eb1a3463STruong Nguyen 		if (restarter_inst_validate_ractions_aux_fmri(scf_inst) == 0) {
1478eb1a3463STruong Nguyen 			if (restarter_inst_set_aux_fmri(scf_inst))
1479eb1a3463STruong Nguyen 				log_framework(LOG_DEBUG, "%s: "
1480eb1a3463STruong Nguyen 				    "restarter_inst_set_aux_fmri failed: ",
1481eb1a3463STruong Nguyen 				    rip->ri_i.i_fmri);
1482eb1a3463STruong Nguyen 		} else {
1483eb1a3463STruong Nguyen 			log_framework(LOG_DEBUG, "%s: "
1484eb1a3463STruong Nguyen 			    "restarter_inst_validate_ractions_aux_fmri "
1485eb1a3463STruong Nguyen 			    "failed: ", rip->ri_i.i_fmri);
1486eb1a3463STruong Nguyen 
1487eb1a3463STruong Nguyen 			if (restarter_inst_reset_aux_fmri(scf_inst))
1488eb1a3463STruong Nguyen 				log_framework(LOG_DEBUG, "%s: "
1489eb1a3463STruong Nguyen 				    "restarter_inst_reset_aux_fmri failed: ",
1490eb1a3463STruong Nguyen 				    rip->ri_i.i_fmri);
1491eb1a3463STruong Nguyen 		}
1492eb1a3463STruong Nguyen 		scf_instance_destroy(scf_inst);
1493eb1a3463STruong Nguyen 	}
1494eb1a3463STruong Nguyen 
14957c478bd9Sstevel@tonic-gate 	if (immediate || !instance_started(rip)) {
14967c478bd9Sstevel@tonic-gate 		if (rip->ri_i.i_primary_ctid != 0) {
14977c478bd9Sstevel@tonic-gate 			rip->ri_m_inst = safe_scf_instance_create(h);
14987c478bd9Sstevel@tonic-gate 			rip->ri_mi_deleted = B_FALSE;
14997c478bd9Sstevel@tonic-gate 
15007c478bd9Sstevel@tonic-gate 			libscf_reget_instance(rip);
15017c478bd9Sstevel@tonic-gate 			method_remove_contract(rip, B_TRUE, B_TRUE);
15027c478bd9Sstevel@tonic-gate 
15037c478bd9Sstevel@tonic-gate 			scf_instance_destroy(rip->ri_m_inst);
15047c478bd9Sstevel@tonic-gate 		}
15057c478bd9Sstevel@tonic-gate 
15067c478bd9Sstevel@tonic-gate 		(void) restarter_instance_update_states(h, rip,
15077c478bd9Sstevel@tonic-gate 		    RESTARTER_STATE_MAINT, RESTARTER_STATE_NONE, RERR_RESTART,
15087c478bd9Sstevel@tonic-gate 		    (char *)aux);
15097c478bd9Sstevel@tonic-gate 		return;
15107c478bd9Sstevel@tonic-gate 	}
15117c478bd9Sstevel@tonic-gate 
15127c478bd9Sstevel@tonic-gate 	(void) restarter_instance_update_states(h, rip, rip->ri_i.i_state,
15137c478bd9Sstevel@tonic-gate 	    RESTARTER_STATE_MAINT, RERR_NONE, (char *)aux);
15147c478bd9Sstevel@tonic-gate 
151599b44c3bSlianep 	log_transition(rip, MAINT_REQUESTED);
151699b44c3bSlianep 
15177c478bd9Sstevel@tonic-gate 	info = startd_zalloc(sizeof (*info));
15187c478bd9Sstevel@tonic-gate 	info->sf_id = rip->ri_id;
15197c478bd9Sstevel@tonic-gate 	info->sf_method_type = METHOD_STOP;
15207c478bd9Sstevel@tonic-gate 	info->sf_event_type = RERR_RESTART;
15217c478bd9Sstevel@tonic-gate 	rip->ri_method_thread = startd_thread_create(method_thread, info);
15227c478bd9Sstevel@tonic-gate }
15237c478bd9Sstevel@tonic-gate 
15247c478bd9Sstevel@tonic-gate static void
15257c478bd9Sstevel@tonic-gate refresh_instance(scf_handle_t *h, restarter_inst_t *rip)
15267c478bd9Sstevel@tonic-gate {
15277c478bd9Sstevel@tonic-gate 	scf_instance_t *inst;
15287c478bd9Sstevel@tonic-gate 	scf_snapshot_t *snap;
15297c478bd9Sstevel@tonic-gate 	fork_info_t *info;
15307c478bd9Sstevel@tonic-gate 	int r;
15317c478bd9Sstevel@tonic-gate 
153253f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&rip->ri_lock));
15337c478bd9Sstevel@tonic-gate 
15347c478bd9Sstevel@tonic-gate 	log_instance(rip, B_TRUE, "Rereading configuration.");
15357c478bd9Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: rereading configuration.\n",
15367c478bd9Sstevel@tonic-gate 	    rip->ri_i.i_fmri);
15377c478bd9Sstevel@tonic-gate 
15387c478bd9Sstevel@tonic-gate rep_retry:
15397c478bd9Sstevel@tonic-gate 	r = libscf_fmri_get_instance(h, rip->ri_i.i_fmri, &inst);
15407c478bd9Sstevel@tonic-gate 	switch (r) {
15417c478bd9Sstevel@tonic-gate 	case 0:
15427c478bd9Sstevel@tonic-gate 		break;
15437c478bd9Sstevel@tonic-gate 
15447c478bd9Sstevel@tonic-gate 	case ECONNABORTED:
15457c478bd9Sstevel@tonic-gate 		libscf_handle_rebind(h);
15467c478bd9Sstevel@tonic-gate 		goto rep_retry;
15477c478bd9Sstevel@tonic-gate 
15487c478bd9Sstevel@tonic-gate 	case ENOENT:
15497c478bd9Sstevel@tonic-gate 		/* Must have been deleted. */
15507c478bd9Sstevel@tonic-gate 		return;
15517c478bd9Sstevel@tonic-gate 
15527c478bd9Sstevel@tonic-gate 	case EINVAL:
15537c478bd9Sstevel@tonic-gate 	case ENOTSUP:
15547c478bd9Sstevel@tonic-gate 	default:
15557c478bd9Sstevel@tonic-gate 		bad_error("libscf_fmri_get_instance", r);
15567c478bd9Sstevel@tonic-gate 	}
15577c478bd9Sstevel@tonic-gate 
15587c478bd9Sstevel@tonic-gate 	snap = libscf_get_running_snapshot(inst);
15597c478bd9Sstevel@tonic-gate 
15607c478bd9Sstevel@tonic-gate 	r = libscf_get_startd_properties(inst, snap, &rip->ri_flags,
15617c478bd9Sstevel@tonic-gate 	    &rip->ri_utmpx_prefix);
15627c478bd9Sstevel@tonic-gate 	switch (r) {
15637c478bd9Sstevel@tonic-gate 	case 0:
15647c478bd9Sstevel@tonic-gate 		log_framework(LOG_DEBUG, "%s is a %s-style service\n",
15657c478bd9Sstevel@tonic-gate 		    rip->ri_i.i_fmri, service_style(rip->ri_flags));
15667c478bd9Sstevel@tonic-gate 		break;
15677c478bd9Sstevel@tonic-gate 
15687c478bd9Sstevel@tonic-gate 	case ECONNABORTED:
15697c478bd9Sstevel@tonic-gate 		scf_instance_destroy(inst);
15707c478bd9Sstevel@tonic-gate 		scf_snapshot_destroy(snap);
15717c478bd9Sstevel@tonic-gate 		libscf_handle_rebind(h);
15727c478bd9Sstevel@tonic-gate 		goto rep_retry;
15737c478bd9Sstevel@tonic-gate 
15747c478bd9Sstevel@tonic-gate 	case ECANCELED:
15757c478bd9Sstevel@tonic-gate 	case ENOENT:
15767c478bd9Sstevel@tonic-gate 		/* Succeed in anticipation of REMOVE_INSTANCE. */
15777c478bd9Sstevel@tonic-gate 		break;
15787c478bd9Sstevel@tonic-gate 
15797c478bd9Sstevel@tonic-gate 	default:
15807c478bd9Sstevel@tonic-gate 		bad_error("libscf_get_startd_properties", r);
15817c478bd9Sstevel@tonic-gate 	}
15827c478bd9Sstevel@tonic-gate 
15837c478bd9Sstevel@tonic-gate 	if (instance_started(rip)) {
15847c478bd9Sstevel@tonic-gate 		/* Refresh does not change the state. */
15857c478bd9Sstevel@tonic-gate 		(void) restarter_instance_update_states(h, rip,
15867c478bd9Sstevel@tonic-gate 		    rip->ri_i.i_state, rip->ri_i.i_state, RERR_NONE, NULL);
15877c478bd9Sstevel@tonic-gate 
15887c478bd9Sstevel@tonic-gate 		info = startd_zalloc(sizeof (*info));
15897c478bd9Sstevel@tonic-gate 		info->sf_id = rip->ri_id;
15907c478bd9Sstevel@tonic-gate 		info->sf_method_type = METHOD_REFRESH;
15917c478bd9Sstevel@tonic-gate 		info->sf_event_type = RERR_REFRESH;
15927c478bd9Sstevel@tonic-gate 
15937c478bd9Sstevel@tonic-gate 		assert(rip->ri_method_thread == 0);
15947c478bd9Sstevel@tonic-gate 		rip->ri_method_thread =
15957c478bd9Sstevel@tonic-gate 		    startd_thread_create(method_thread, info);
15967c478bd9Sstevel@tonic-gate 	}
15977c478bd9Sstevel@tonic-gate 
15987c478bd9Sstevel@tonic-gate 	scf_snapshot_destroy(snap);
15997c478bd9Sstevel@tonic-gate 	scf_instance_destroy(inst);
16007c478bd9Sstevel@tonic-gate }
16017c478bd9Sstevel@tonic-gate 
16027c478bd9Sstevel@tonic-gate const char *event_names[] = { "INVALID", "ADD_INSTANCE", "REMOVE_INSTANCE",
16037c478bd9Sstevel@tonic-gate 	"ENABLE", "DISABLE", "ADMIN_DEGRADED", "ADMIN_REFRESH",
16047c478bd9Sstevel@tonic-gate 	"ADMIN_RESTART", "ADMIN_MAINT_OFF", "ADMIN_MAINT_ON",
16057c478bd9Sstevel@tonic-gate 	"ADMIN_MAINT_ON_IMMEDIATE", "STOP", "START", "DEPENDENCY_CYCLE",
1606*c238c833SSean Wilcox 	"INVALID_DEPENDENCY", "ADMIN_DISABLE", "STOP_RESET"
16077c478bd9Sstevel@tonic-gate };
16087c478bd9Sstevel@tonic-gate 
16097c478bd9Sstevel@tonic-gate /*
16107c478bd9Sstevel@tonic-gate  * void *restarter_process_events()
16117c478bd9Sstevel@tonic-gate  *
16127c478bd9Sstevel@tonic-gate  *   Called in a separate thread to process the events on an instance's
16137c478bd9Sstevel@tonic-gate  *   queue.  Empties the queue completely, and tries to keep the thread
16147c478bd9Sstevel@tonic-gate  *   around for a little while after the queue is empty to save on
16157c478bd9Sstevel@tonic-gate  *   startup costs.
16167c478bd9Sstevel@tonic-gate  */
16177c478bd9Sstevel@tonic-gate static void *
16187c478bd9Sstevel@tonic-gate restarter_process_events(void *arg)
16197c478bd9Sstevel@tonic-gate {
16207c478bd9Sstevel@tonic-gate 	scf_handle_t *h;
16217c478bd9Sstevel@tonic-gate 	restarter_instance_qentry_t *event;
16227c478bd9Sstevel@tonic-gate 	restarter_inst_t *rip;
16237c478bd9Sstevel@tonic-gate 	char *fmri = (char *)arg;
16247c478bd9Sstevel@tonic-gate 	struct timespec to;
16257c478bd9Sstevel@tonic-gate 
16267c478bd9Sstevel@tonic-gate 	assert(fmri != NULL);
16277c478bd9Sstevel@tonic-gate 
16287c478bd9Sstevel@tonic-gate 	h = libscf_handle_create_bound_loop();
16297c478bd9Sstevel@tonic-gate 
16307c478bd9Sstevel@tonic-gate 	/* grab the queue lock */
16317c478bd9Sstevel@tonic-gate 	rip = inst_lookup_queue(fmri);
16327c478bd9Sstevel@tonic-gate 	if (rip == NULL)
16337c478bd9Sstevel@tonic-gate 		goto out;
16347c478bd9Sstevel@tonic-gate 
16357c478bd9Sstevel@tonic-gate again:
16367c478bd9Sstevel@tonic-gate 
16377c478bd9Sstevel@tonic-gate 	while ((event = uu_list_first(rip->ri_queue)) != NULL) {
16387c478bd9Sstevel@tonic-gate 		restarter_inst_t *inst;
16397c478bd9Sstevel@tonic-gate 
16407c478bd9Sstevel@tonic-gate 		/* drop the queue lock */
16417c478bd9Sstevel@tonic-gate 		MUTEX_UNLOCK(&rip->ri_queue_lock);
16427c478bd9Sstevel@tonic-gate 
16437c478bd9Sstevel@tonic-gate 		/*
16447c478bd9Sstevel@tonic-gate 		 * Grab the inst lock -- this waits until any outstanding
16457c478bd9Sstevel@tonic-gate 		 * method finishes running.
16467c478bd9Sstevel@tonic-gate 		 */
16477c478bd9Sstevel@tonic-gate 		inst = inst_lookup_by_name(fmri);
16487c478bd9Sstevel@tonic-gate 		if (inst == NULL) {
16497c478bd9Sstevel@tonic-gate 			/* Getting deleted in the middle isn't an error. */
16507c478bd9Sstevel@tonic-gate 			goto cont;
16517c478bd9Sstevel@tonic-gate 		}
16527c478bd9Sstevel@tonic-gate 
16537c478bd9Sstevel@tonic-gate 		assert(instance_in_transition(inst) == 0);
16547c478bd9Sstevel@tonic-gate 
16557c478bd9Sstevel@tonic-gate 		/* process the event */
16567c478bd9Sstevel@tonic-gate 		switch (event->riq_type) {
16577c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ENABLE:
16587c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_DISABLE:
16597c478bd9Sstevel@tonic-gate 			(void) enable_inst(h, inst, event->riq_type);
16607c478bd9Sstevel@tonic-gate 			break;
16617c478bd9Sstevel@tonic-gate 
166216ba0facSSean Wilcox 		case RESTARTER_EVENT_TYPE_ADMIN_DISABLE:
166316ba0facSSean Wilcox 			if (enable_inst(h, inst, event->riq_type) == 0)
166416ba0facSSean Wilcox 				reset_start_times(inst);
166516ba0facSSean Wilcox 			break;
166616ba0facSSean Wilcox 
16677c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_REMOVE_INSTANCE:
16687c478bd9Sstevel@tonic-gate 			restarter_delete_inst(inst);
16697c478bd9Sstevel@tonic-gate 			inst = NULL;
16707c478bd9Sstevel@tonic-gate 			goto cont;
16717c478bd9Sstevel@tonic-gate 
167216ba0facSSean Wilcox 		case RESTARTER_EVENT_TYPE_STOP_RESET:
167316ba0facSSean Wilcox 			reset_start_times(inst);
167416ba0facSSean Wilcox 			/* FALLTHROUGH */
16757c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_STOP:
16767c478bd9Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_DEPENDENCY);
16777c478bd9Sstevel@tonic-gate 			break;
16787c478bd9Sstevel@tonic-gate 
16797c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_START:
16807c478bd9Sstevel@tonic-gate 			start_instance(h, inst);
16817c478bd9Sstevel@tonic-gate 			break;
16827c478bd9Sstevel@tonic-gate 
16837c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_DEPENDENCY_CYCLE:
16847c478bd9Sstevel@tonic-gate 			maintain_instance(h, inst, 0, "dependency_cycle");
16857c478bd9Sstevel@tonic-gate 			break;
16867c478bd9Sstevel@tonic-gate 
16877c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_INVALID_DEPENDENCY:
16887c478bd9Sstevel@tonic-gate 			maintain_instance(h, inst, 0, "invalid_dependency");
16897c478bd9Sstevel@tonic-gate 			break;
16907c478bd9Sstevel@tonic-gate 
16917c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON:
1692eb1a3463STruong Nguyen 			if (event_from_tty(h, inst) == 0)
1693eb1a3463STruong Nguyen 				maintain_instance(h, inst, 0,
1694eb1a3463STruong Nguyen 				    "service_request");
1695eb1a3463STruong Nguyen 			else
1696eb1a3463STruong Nguyen 				maintain_instance(h, inst, 0,
1697eb1a3463STruong Nguyen 				    "administrative_request");
16987c478bd9Sstevel@tonic-gate 			break;
16997c478bd9Sstevel@tonic-gate 
17007c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON_IMMEDIATE:
1701eb1a3463STruong Nguyen 			if (event_from_tty(h, inst) == 0)
1702eb1a3463STruong Nguyen 				maintain_instance(h, inst, 1,
1703eb1a3463STruong Nguyen 				    "service_request");
1704eb1a3463STruong Nguyen 			else
1705eb1a3463STruong Nguyen 				maintain_instance(h, inst, 1,
1706eb1a3463STruong Nguyen 				    "administrative_request");
17077c478bd9Sstevel@tonic-gate 			break;
17087c478bd9Sstevel@tonic-gate 
17097c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_OFF:
17107c478bd9Sstevel@tonic-gate 			unmaintain_instance(h, inst, RUNMAINT_CLEAR);
17117c478bd9Sstevel@tonic-gate 			break;
17127c478bd9Sstevel@tonic-gate 
17137c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_REFRESH:
17147c478bd9Sstevel@tonic-gate 			refresh_instance(h, inst);
17157c478bd9Sstevel@tonic-gate 			break;
17167c478bd9Sstevel@tonic-gate 
17177c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_DEGRADED:
17187c478bd9Sstevel@tonic-gate 			log_framework(LOG_WARNING, "Restarter: "
17197c478bd9Sstevel@tonic-gate 			    "%s command (for %s) unimplemented.\n",
17207c478bd9Sstevel@tonic-gate 			    event_names[event->riq_type], inst->ri_i.i_fmri);
17217c478bd9Sstevel@tonic-gate 			break;
17227c478bd9Sstevel@tonic-gate 
17237c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_RESTART:
17247c478bd9Sstevel@tonic-gate 			if (!instance_started(inst)) {
17257c478bd9Sstevel@tonic-gate 				log_framework(LOG_DEBUG, "Restarter: "
17267c478bd9Sstevel@tonic-gate 				    "Not restarting %s; not running.\n",
17277c478bd9Sstevel@tonic-gate 				    inst->ri_i.i_fmri);
17287c478bd9Sstevel@tonic-gate 			} else {
17297c478bd9Sstevel@tonic-gate 				/*
17307c478bd9Sstevel@tonic-gate 				 * Stop the instance.  If it can be restarted,
17317c478bd9Sstevel@tonic-gate 				 * the graph engine will send a new event.
17327c478bd9Sstevel@tonic-gate 				 */
173316ba0facSSean Wilcox 				if (stop_instance(h, inst, RSTOP_RESTART) == 0)
173416ba0facSSean Wilcox 					reset_start_times(inst);
17357c478bd9Sstevel@tonic-gate 			}
17367c478bd9Sstevel@tonic-gate 			break;
17377c478bd9Sstevel@tonic-gate 
17387c478bd9Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADD_INSTANCE:
17397c478bd9Sstevel@tonic-gate 		default:
17407c478bd9Sstevel@tonic-gate #ifndef NDEBUG
17417c478bd9Sstevel@tonic-gate 			uu_warn("%s:%d: Bad restarter event %d.  "
17427c478bd9Sstevel@tonic-gate 			    "Aborting.\n", __FILE__, __LINE__, event->riq_type);
17437c478bd9Sstevel@tonic-gate #endif
17447c478bd9Sstevel@tonic-gate 			abort();
17457c478bd9Sstevel@tonic-gate 		}
17467c478bd9Sstevel@tonic-gate 
17477c478bd9Sstevel@tonic-gate 		assert(inst != NULL);
17487c478bd9Sstevel@tonic-gate 		MUTEX_UNLOCK(&inst->ri_lock);
17497c478bd9Sstevel@tonic-gate 
17507c478bd9Sstevel@tonic-gate cont:
17517c478bd9Sstevel@tonic-gate 		/* grab the queue lock */
17527c478bd9Sstevel@tonic-gate 		rip = inst_lookup_queue(fmri);
17537c478bd9Sstevel@tonic-gate 		if (rip == NULL)
17547c478bd9Sstevel@tonic-gate 			goto out;
17557c478bd9Sstevel@tonic-gate 
17567c478bd9Sstevel@tonic-gate 		/* delete the event */
17577c478bd9Sstevel@tonic-gate 		uu_list_remove(rip->ri_queue, event);
17587c478bd9Sstevel@tonic-gate 		startd_free(event, sizeof (restarter_instance_qentry_t));
17597c478bd9Sstevel@tonic-gate 	}
17607c478bd9Sstevel@tonic-gate 
17617c478bd9Sstevel@tonic-gate 	assert(rip != NULL);
17627c478bd9Sstevel@tonic-gate 
17637c478bd9Sstevel@tonic-gate 	/*
17647c478bd9Sstevel@tonic-gate 	 * Try to preserve the thread for a little while for future use.
17657c478bd9Sstevel@tonic-gate 	 */
17667c478bd9Sstevel@tonic-gate 	to.tv_sec = 3;
17677c478bd9Sstevel@tonic-gate 	to.tv_nsec = 0;
17687c478bd9Sstevel@tonic-gate 	(void) pthread_cond_reltimedwait_np(&rip->ri_queue_cv,
17697c478bd9Sstevel@tonic-gate 	    &rip->ri_queue_lock, &to);
17707c478bd9Sstevel@tonic-gate 
17717c478bd9Sstevel@tonic-gate 	if (uu_list_first(rip->ri_queue) != NULL)
17727c478bd9Sstevel@tonic-gate 		goto again;
17737c478bd9Sstevel@tonic-gate 
17747c478bd9Sstevel@tonic-gate 	rip->ri_queue_thread = 0;
17757c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&rip->ri_queue_lock);
17767c478bd9Sstevel@tonic-gate out:
17777c478bd9Sstevel@tonic-gate 	(void) scf_handle_unbind(h);
17787c478bd9Sstevel@tonic-gate 	scf_handle_destroy(h);
17797c478bd9Sstevel@tonic-gate 	free(fmri);
17807c478bd9Sstevel@tonic-gate 	return (NULL);
17817c478bd9Sstevel@tonic-gate }
17827c478bd9Sstevel@tonic-gate 
17837c478bd9Sstevel@tonic-gate static int
17847c478bd9Sstevel@tonic-gate is_admin_event(restarter_event_type_t t) {
17857c478bd9Sstevel@tonic-gate 
17867c478bd9Sstevel@tonic-gate 	switch (t) {
17877c478bd9Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON:
17887c478bd9Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON_IMMEDIATE:
17897c478bd9Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_MAINT_OFF:
17907c478bd9Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_REFRESH:
17917c478bd9Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_DEGRADED:
17927c478bd9Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_RESTART:
17937c478bd9Sstevel@tonic-gate 		return (1);
17947c478bd9Sstevel@tonic-gate 	default:
17957c478bd9Sstevel@tonic-gate 		return (0);
17967c478bd9Sstevel@tonic-gate 	}
17977c478bd9Sstevel@tonic-gate }
17987c478bd9Sstevel@tonic-gate 
17997c478bd9Sstevel@tonic-gate static void
18007c478bd9Sstevel@tonic-gate restarter_queue_event(restarter_inst_t *ri, restarter_protocol_event_t *e)
18017c478bd9Sstevel@tonic-gate {
18027c478bd9Sstevel@tonic-gate 	restarter_instance_qentry_t *qe;
18037c478bd9Sstevel@tonic-gate 	int r;
18047c478bd9Sstevel@tonic-gate 
180553f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&ri->ri_queue_lock));
180653f3aea0SRoger A. Faulkner 	assert(!MUTEX_HELD(&ri->ri_lock));
18077c478bd9Sstevel@tonic-gate 
18087c478bd9Sstevel@tonic-gate 	qe = startd_zalloc(sizeof (restarter_instance_qentry_t));
18097c478bd9Sstevel@tonic-gate 	qe->riq_type = e->rpe_type;
18107c478bd9Sstevel@tonic-gate 
18117c478bd9Sstevel@tonic-gate 	uu_list_node_init(qe, &qe->riq_link, restarter_queue_pool);
18127c478bd9Sstevel@tonic-gate 	r = uu_list_insert_before(ri->ri_queue, NULL, qe);
18137c478bd9Sstevel@tonic-gate 	assert(r == 0);
18147c478bd9Sstevel@tonic-gate }
18157c478bd9Sstevel@tonic-gate 
18167c478bd9Sstevel@tonic-gate /*
18177c478bd9Sstevel@tonic-gate  * void *restarter_event_thread()
18187c478bd9Sstevel@tonic-gate  *
18197c478bd9Sstevel@tonic-gate  *  Handle incoming graph events by placing them on a per-instance
18207c478bd9Sstevel@tonic-gate  *  queue.  We can't lock the main part of the instance structure, so
18217c478bd9Sstevel@tonic-gate  *  just modify the seprarately locked event queue portion.
18227c478bd9Sstevel@tonic-gate  */
18237c478bd9Sstevel@tonic-gate /*ARGSUSED*/
18247c478bd9Sstevel@tonic-gate static void *
18257c478bd9Sstevel@tonic-gate restarter_event_thread(void *unused)
18267c478bd9Sstevel@tonic-gate {
18277c478bd9Sstevel@tonic-gate 	scf_handle_t *h;
18287c478bd9Sstevel@tonic-gate 
18297c478bd9Sstevel@tonic-gate 	/*
18307c478bd9Sstevel@tonic-gate 	 * This is a new thread, and thus, gets its own handle
18317c478bd9Sstevel@tonic-gate 	 * to the repository.
18327c478bd9Sstevel@tonic-gate 	 */
18337c478bd9Sstevel@tonic-gate 	h = libscf_handle_create_bound_loop();
18347c478bd9Sstevel@tonic-gate 
18357c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&ru->restarter_update_lock);
18367c478bd9Sstevel@tonic-gate 
18377c478bd9Sstevel@tonic-gate 	/*CONSTCOND*/
18387c478bd9Sstevel@tonic-gate 	while (1) {
18397c478bd9Sstevel@tonic-gate 		restarter_protocol_event_t *e;
18407c478bd9Sstevel@tonic-gate 
18417c478bd9Sstevel@tonic-gate 		while (ru->restarter_update_wakeup == 0)
18427c478bd9Sstevel@tonic-gate 			(void) pthread_cond_wait(&ru->restarter_update_cv,
18437c478bd9Sstevel@tonic-gate 			    &ru->restarter_update_lock);
18447c478bd9Sstevel@tonic-gate 
18457c478bd9Sstevel@tonic-gate 		ru->restarter_update_wakeup = 0;
18467c478bd9Sstevel@tonic-gate 
18477c478bd9Sstevel@tonic-gate 		while ((e = restarter_event_dequeue()) != NULL) {
18487c478bd9Sstevel@tonic-gate 			restarter_inst_t *rip;
18497c478bd9Sstevel@tonic-gate 			char *fmri;
18507c478bd9Sstevel@tonic-gate 
18517c478bd9Sstevel@tonic-gate 			MUTEX_UNLOCK(&ru->restarter_update_lock);
18527c478bd9Sstevel@tonic-gate 
18537c478bd9Sstevel@tonic-gate 			/*
18547c478bd9Sstevel@tonic-gate 			 * ADD_INSTANCE is special: there's likely no
18557c478bd9Sstevel@tonic-gate 			 * instance structure yet, so we need to handle the
18567c478bd9Sstevel@tonic-gate 			 * addition synchronously.
18577c478bd9Sstevel@tonic-gate 			 */
18587c478bd9Sstevel@tonic-gate 			switch (e->rpe_type) {
18597c478bd9Sstevel@tonic-gate 			case RESTARTER_EVENT_TYPE_ADD_INSTANCE:
18607c478bd9Sstevel@tonic-gate 				if (restarter_insert_inst(h, e->rpe_inst) != 0)
18617c478bd9Sstevel@tonic-gate 					log_error(LOG_INFO, "Restarter: "
18627c478bd9Sstevel@tonic-gate 					    "Could not add %s.\n", e->rpe_inst);
18637c478bd9Sstevel@tonic-gate 
18647c478bd9Sstevel@tonic-gate 				MUTEX_LOCK(&st->st_load_lock);
18657c478bd9Sstevel@tonic-gate 				if (--st->st_load_instances == 0)
18667c478bd9Sstevel@tonic-gate 					(void) pthread_cond_broadcast(
18677c478bd9Sstevel@tonic-gate 					    &st->st_load_cv);
18687c478bd9Sstevel@tonic-gate 				MUTEX_UNLOCK(&st->st_load_lock);
18697c478bd9Sstevel@tonic-gate 
18707c478bd9Sstevel@tonic-gate 				goto nolookup;
18717c478bd9Sstevel@tonic-gate 			}
18727c478bd9Sstevel@tonic-gate 
18737c478bd9Sstevel@tonic-gate 			/*
18747c478bd9Sstevel@tonic-gate 			 * Lookup the instance, locking only the event queue.
18757c478bd9Sstevel@tonic-gate 			 * Can't grab ri_lock here because it might be held
18767c478bd9Sstevel@tonic-gate 			 * by a long-running method.
18777c478bd9Sstevel@tonic-gate 			 */
18787c478bd9Sstevel@tonic-gate 			rip = inst_lookup_queue(e->rpe_inst);
18797c478bd9Sstevel@tonic-gate 			if (rip == NULL) {
18807c478bd9Sstevel@tonic-gate 				log_error(LOG_INFO, "Restarter: "
18817c478bd9Sstevel@tonic-gate 				    "Ignoring %s command for unknown service "
18827c478bd9Sstevel@tonic-gate 				    "%s.\n", event_names[e->rpe_type],
18837c478bd9Sstevel@tonic-gate 				    e->rpe_inst);
18847c478bd9Sstevel@tonic-gate 				goto nolookup;
18857c478bd9Sstevel@tonic-gate 			}
18867c478bd9Sstevel@tonic-gate 
18877c478bd9Sstevel@tonic-gate 			/* Keep ADMIN events from filling up the queue. */
18887c478bd9Sstevel@tonic-gate 			if (is_admin_event(e->rpe_type) &&
18897c478bd9Sstevel@tonic-gate 			    uu_list_numnodes(rip->ri_queue) >
18907c478bd9Sstevel@tonic-gate 			    RINST_QUEUE_THRESHOLD) {
18917c478bd9Sstevel@tonic-gate 				MUTEX_UNLOCK(&rip->ri_queue_lock);
18927c478bd9Sstevel@tonic-gate 				log_instance(rip, B_TRUE, "Instance event "
18937c478bd9Sstevel@tonic-gate 				    "queue overflow.  Dropping administrative "
18947c478bd9Sstevel@tonic-gate 				    "request.");
18957c478bd9Sstevel@tonic-gate 				log_framework(LOG_DEBUG, "%s: Instance event "
18967c478bd9Sstevel@tonic-gate 				    "queue overflow.  Dropping administrative "
18977c478bd9Sstevel@tonic-gate 				    "request.\n", rip->ri_i.i_fmri);
18987c478bd9Sstevel@tonic-gate 				goto nolookup;
18997c478bd9Sstevel@tonic-gate 			}
19007c478bd9Sstevel@tonic-gate 
19017c478bd9Sstevel@tonic-gate 			/* Now add the event to the instance queue. */
19027c478bd9Sstevel@tonic-gate 			restarter_queue_event(rip, e);
19037c478bd9Sstevel@tonic-gate 
19047c478bd9Sstevel@tonic-gate 			if (rip->ri_queue_thread == 0) {
19057c478bd9Sstevel@tonic-gate 				/*
19067c478bd9Sstevel@tonic-gate 				 * Start a thread if one isn't already
19077c478bd9Sstevel@tonic-gate 				 * running.
19087c478bd9Sstevel@tonic-gate 				 */
19097c478bd9Sstevel@tonic-gate 				fmri = safe_strdup(e->rpe_inst);
19107c478bd9Sstevel@tonic-gate 				rip->ri_queue_thread =  startd_thread_create(
19117c478bd9Sstevel@tonic-gate 				    restarter_process_events, (void *)fmri);
19127c478bd9Sstevel@tonic-gate 			} else {
19137c478bd9Sstevel@tonic-gate 				/*
19147c478bd9Sstevel@tonic-gate 				 * Signal the existing thread that there's
19157c478bd9Sstevel@tonic-gate 				 * a new event.
19167c478bd9Sstevel@tonic-gate 				 */
19177c478bd9Sstevel@tonic-gate 				(void) pthread_cond_broadcast(
19187c478bd9Sstevel@tonic-gate 				    &rip->ri_queue_cv);
19197c478bd9Sstevel@tonic-gate 			}
19207c478bd9Sstevel@tonic-gate 
19217c478bd9Sstevel@tonic-gate 			MUTEX_UNLOCK(&rip->ri_queue_lock);
19227c478bd9Sstevel@tonic-gate nolookup:
19237c478bd9Sstevel@tonic-gate 			restarter_event_release(e);
19247c478bd9Sstevel@tonic-gate 
19257c478bd9Sstevel@tonic-gate 			MUTEX_LOCK(&ru->restarter_update_lock);
19267c478bd9Sstevel@tonic-gate 		}
19277c478bd9Sstevel@tonic-gate 	}
19287c478bd9Sstevel@tonic-gate 
19297c478bd9Sstevel@tonic-gate 	/*
19307c478bd9Sstevel@tonic-gate 	 * Unreachable for now -- there's currently no graceful cleanup
19317c478bd9Sstevel@tonic-gate 	 * called on exit().
19327c478bd9Sstevel@tonic-gate 	 */
19337c478bd9Sstevel@tonic-gate 	(void) scf_handle_unbind(h);
19347c478bd9Sstevel@tonic-gate 	scf_handle_destroy(h);
19357c478bd9Sstevel@tonic-gate 	return (NULL);
19367c478bd9Sstevel@tonic-gate }
19377c478bd9Sstevel@tonic-gate 
19387c478bd9Sstevel@tonic-gate static restarter_inst_t *
19397c478bd9Sstevel@tonic-gate contract_to_inst(ctid_t ctid)
19407c478bd9Sstevel@tonic-gate {
19417c478bd9Sstevel@tonic-gate 	restarter_inst_t *inst;
19427c478bd9Sstevel@tonic-gate 	int id;
19437c478bd9Sstevel@tonic-gate 
19447c478bd9Sstevel@tonic-gate 	id = lookup_inst_by_contract(ctid);
19457c478bd9Sstevel@tonic-gate 	if (id == -1)
19467c478bd9Sstevel@tonic-gate 		return (NULL);
19477c478bd9Sstevel@tonic-gate 
19487c478bd9Sstevel@tonic-gate 	inst = inst_lookup_by_id(id);
19497c478bd9Sstevel@tonic-gate 	if (inst != NULL) {
19507c478bd9Sstevel@tonic-gate 		/*
19517c478bd9Sstevel@tonic-gate 		 * Since ri_lock isn't held by the contract id lookup, this
19527c478bd9Sstevel@tonic-gate 		 * instance may have been restarted and now be in a new
19537c478bd9Sstevel@tonic-gate 		 * contract, making the old contract no longer valid for this
19547c478bd9Sstevel@tonic-gate 		 * instance.
19557c478bd9Sstevel@tonic-gate 		 */
19567c478bd9Sstevel@tonic-gate 		if (ctid != inst->ri_i.i_primary_ctid) {
19577c478bd9Sstevel@tonic-gate 			MUTEX_UNLOCK(&inst->ri_lock);
19587c478bd9Sstevel@tonic-gate 			inst = NULL;
19597c478bd9Sstevel@tonic-gate 		}
19607c478bd9Sstevel@tonic-gate 	}
19617c478bd9Sstevel@tonic-gate 	return (inst);
19627c478bd9Sstevel@tonic-gate }
19637c478bd9Sstevel@tonic-gate 
19647c478bd9Sstevel@tonic-gate /*
19657c478bd9Sstevel@tonic-gate  * void contract_action()
19667c478bd9Sstevel@tonic-gate  *   Take action on contract events.
19677c478bd9Sstevel@tonic-gate  */
19687c478bd9Sstevel@tonic-gate static void
19697c478bd9Sstevel@tonic-gate contract_action(scf_handle_t *h, restarter_inst_t *inst, ctid_t id,
19707c478bd9Sstevel@tonic-gate     uint32_t type)
19717c478bd9Sstevel@tonic-gate {
19727c478bd9Sstevel@tonic-gate 	const char *fmri = inst->ri_i.i_fmri;
19737c478bd9Sstevel@tonic-gate 
197453f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&inst->ri_lock));
19757c478bd9Sstevel@tonic-gate 
19767c478bd9Sstevel@tonic-gate 	/*
19777c478bd9Sstevel@tonic-gate 	 * If startd has stopped this contract, there is no need to
19787c478bd9Sstevel@tonic-gate 	 * stop it again.
19797c478bd9Sstevel@tonic-gate 	 */
19807c478bd9Sstevel@tonic-gate 	if (inst->ri_i.i_primary_ctid > 0 &&
19817c478bd9Sstevel@tonic-gate 	    inst->ri_i.i_primary_ctid_stopped)
19827c478bd9Sstevel@tonic-gate 		return;
19837c478bd9Sstevel@tonic-gate 
19847c478bd9Sstevel@tonic-gate 	if ((type & (CT_PR_EV_EMPTY | CT_PR_EV_CORE | CT_PR_EV_SIGNAL
19857c478bd9Sstevel@tonic-gate 	    | CT_PR_EV_HWERR)) == 0) {
19867c478bd9Sstevel@tonic-gate 		/*
19877c478bd9Sstevel@tonic-gate 		 * There shouldn't be other events, since that's not how we set
19887c478bd9Sstevel@tonic-gate 		 * the terms. Thus, just log an error and drive on.
19897c478bd9Sstevel@tonic-gate 		 */
19907c478bd9Sstevel@tonic-gate 		log_framework(LOG_NOTICE,
19917c478bd9Sstevel@tonic-gate 		    "%s: contract %ld received unexpected critical event "
19927c478bd9Sstevel@tonic-gate 		    "(%d)\n", fmri, id, type);
19937c478bd9Sstevel@tonic-gate 		return;
19947c478bd9Sstevel@tonic-gate 	}
19957c478bd9Sstevel@tonic-gate 
19967c478bd9Sstevel@tonic-gate 	assert(instance_in_transition(inst) == 0);
19977c478bd9Sstevel@tonic-gate 
19987c478bd9Sstevel@tonic-gate 	if (instance_is_wait_style(inst)) {
19997c478bd9Sstevel@tonic-gate 		/*
20007c478bd9Sstevel@tonic-gate 		 * We ignore all events; if they impact the
20017c478bd9Sstevel@tonic-gate 		 * process we're monitoring, then the
20027c478bd9Sstevel@tonic-gate 		 * wait_thread will stop the instance.
20037c478bd9Sstevel@tonic-gate 		 */
20047c478bd9Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
20057c478bd9Sstevel@tonic-gate 		    "%s: ignoring contract event on wait-style service\n",
20067c478bd9Sstevel@tonic-gate 		    fmri);
20077c478bd9Sstevel@tonic-gate 	} else {
20087c478bd9Sstevel@tonic-gate 		/*
20097c478bd9Sstevel@tonic-gate 		 * A CT_PR_EV_EMPTY event is an RSTOP_EXIT request.
20107c478bd9Sstevel@tonic-gate 		 */
20117c478bd9Sstevel@tonic-gate 		switch (type) {
20127c478bd9Sstevel@tonic-gate 		case CT_PR_EV_EMPTY:
20137c478bd9Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_EXIT);
20147c478bd9Sstevel@tonic-gate 			break;
20157c478bd9Sstevel@tonic-gate 		case CT_PR_EV_CORE:
20167c478bd9Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_CORE);
20177c478bd9Sstevel@tonic-gate 			break;
20187c478bd9Sstevel@tonic-gate 		case CT_PR_EV_SIGNAL:
20197c478bd9Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_SIGNAL);
20207c478bd9Sstevel@tonic-gate 			break;
20217c478bd9Sstevel@tonic-gate 		case CT_PR_EV_HWERR:
20227c478bd9Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_HWERR);
20237c478bd9Sstevel@tonic-gate 			break;
20247c478bd9Sstevel@tonic-gate 		}
20257c478bd9Sstevel@tonic-gate 	}
20267c478bd9Sstevel@tonic-gate }
20277c478bd9Sstevel@tonic-gate 
20287c478bd9Sstevel@tonic-gate /*
20297c478bd9Sstevel@tonic-gate  * void *restarter_contract_event_thread(void *)
20307c478bd9Sstevel@tonic-gate  *   Listens to the process contract bundle for critical events, taking action
20317c478bd9Sstevel@tonic-gate  *   on events from contracts we know we are responsible for.
20327c478bd9Sstevel@tonic-gate  */
20337c478bd9Sstevel@tonic-gate /*ARGSUSED*/
20347c478bd9Sstevel@tonic-gate static void *
20357c478bd9Sstevel@tonic-gate restarter_contracts_event_thread(void *unused)
20367c478bd9Sstevel@tonic-gate {
20377c478bd9Sstevel@tonic-gate 	int fd, err;
20387c478bd9Sstevel@tonic-gate 	scf_handle_t *local_handle;
20397c478bd9Sstevel@tonic-gate 
20407c478bd9Sstevel@tonic-gate 	/*
20417c478bd9Sstevel@tonic-gate 	 * Await graph load completion.  That is, stop here, until we've scanned
20427c478bd9Sstevel@tonic-gate 	 * the repository for contract - instance associations.
20437c478bd9Sstevel@tonic-gate 	 */
20447c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&st->st_load_lock);
20457c478bd9Sstevel@tonic-gate 	while (!(st->st_load_complete && st->st_load_instances == 0))
20467c478bd9Sstevel@tonic-gate 		(void) pthread_cond_wait(&st->st_load_cv, &st->st_load_lock);
20477c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&st->st_load_lock);
20487c478bd9Sstevel@tonic-gate 
20497c478bd9Sstevel@tonic-gate 	/*
20507c478bd9Sstevel@tonic-gate 	 * This is a new thread, and thus, gets its own handle
20517c478bd9Sstevel@tonic-gate 	 * to the repository.
20527c478bd9Sstevel@tonic-gate 	 */
20537c478bd9Sstevel@tonic-gate 	if ((local_handle = libscf_handle_create_bound(SCF_VERSION)) == NULL)
20547c478bd9Sstevel@tonic-gate 		uu_die("Unable to bind a new repository handle: %s\n",
20557c478bd9Sstevel@tonic-gate 		    scf_strerror(scf_error()));
20567c478bd9Sstevel@tonic-gate 
20577c478bd9Sstevel@tonic-gate 	fd = open64(CTFS_ROOT "/process/pbundle", O_RDONLY);
20587c478bd9Sstevel@tonic-gate 	if (fd == -1)
20597c478bd9Sstevel@tonic-gate 		uu_die("process bundle open failed");
20607c478bd9Sstevel@tonic-gate 
20617c478bd9Sstevel@tonic-gate 	/*
20627c478bd9Sstevel@tonic-gate 	 * Make sure we get all events (including those generated by configd
20637c478bd9Sstevel@tonic-gate 	 * before this thread was started).
20647c478bd9Sstevel@tonic-gate 	 */
20657c478bd9Sstevel@tonic-gate 	err = ct_event_reset(fd);
20667c478bd9Sstevel@tonic-gate 	assert(err == 0);
20677c478bd9Sstevel@tonic-gate 
20687c478bd9Sstevel@tonic-gate 	for (;;) {
20697c478bd9Sstevel@tonic-gate 		int efd, sfd;
20707c478bd9Sstevel@tonic-gate 		ct_evthdl_t ev;
20717c478bd9Sstevel@tonic-gate 		uint32_t type;
20727c478bd9Sstevel@tonic-gate 		ctevid_t evid;
20737c478bd9Sstevel@tonic-gate 		ct_stathdl_t status;
20747c478bd9Sstevel@tonic-gate 		ctid_t ctid;
20757c478bd9Sstevel@tonic-gate 		restarter_inst_t *inst;
20767c478bd9Sstevel@tonic-gate 		uint64_t cookie;
20777c478bd9Sstevel@tonic-gate 
20787c478bd9Sstevel@tonic-gate 		if (err = ct_event_read_critical(fd, &ev)) {
20797c478bd9Sstevel@tonic-gate 			log_error(LOG_WARNING,
20807c478bd9Sstevel@tonic-gate 			    "Error reading next contract event: %s",
20817c478bd9Sstevel@tonic-gate 			    strerror(err));
20827c478bd9Sstevel@tonic-gate 			continue;
20837c478bd9Sstevel@tonic-gate 		}
20847c478bd9Sstevel@tonic-gate 
20857c478bd9Sstevel@tonic-gate 		evid = ct_event_get_evid(ev);
20867c478bd9Sstevel@tonic-gate 		ctid = ct_event_get_ctid(ev);
20877c478bd9Sstevel@tonic-gate 		type = ct_event_get_type(ev);
20887c478bd9Sstevel@tonic-gate 
20897c478bd9Sstevel@tonic-gate 		/* Fetch cookie. */
20907c478bd9Sstevel@tonic-gate 		if ((sfd = contract_open(ctid, "process", "status", O_RDONLY))
20917c478bd9Sstevel@tonic-gate 		    < 0) {
20927c478bd9Sstevel@tonic-gate 			ct_event_free(ev);
20937c478bd9Sstevel@tonic-gate 			continue;
20947c478bd9Sstevel@tonic-gate 		}
20957c478bd9Sstevel@tonic-gate 
20967c478bd9Sstevel@tonic-gate 		if (err = ct_status_read(sfd, CTD_COMMON, &status)) {
20977c478bd9Sstevel@tonic-gate 			log_framework(LOG_WARNING, "Could not get status for "
20987c478bd9Sstevel@tonic-gate 			    "contract %ld: %s\n", ctid, strerror(err));
20997c478bd9Sstevel@tonic-gate 
21007c478bd9Sstevel@tonic-gate 			startd_close(sfd);
21017c478bd9Sstevel@tonic-gate 			ct_event_free(ev);
21027c478bd9Sstevel@tonic-gate 			continue;
21037c478bd9Sstevel@tonic-gate 		}
21047c478bd9Sstevel@tonic-gate 
21057c478bd9Sstevel@tonic-gate 		cookie = ct_status_get_cookie(status);
21067c478bd9Sstevel@tonic-gate 
2107dfe57350Sjeanm 		log_framework(LOG_DEBUG, "Received event %d for ctid %ld "
2108dfe57350Sjeanm 		    "cookie %lld\n", type, ctid, cookie);
2109dfe57350Sjeanm 
21107c478bd9Sstevel@tonic-gate 		ct_status_free(status);
21117c478bd9Sstevel@tonic-gate 
21127c478bd9Sstevel@tonic-gate 		startd_close(sfd);
21137c478bd9Sstevel@tonic-gate 
21147c478bd9Sstevel@tonic-gate 		/*
21157c478bd9Sstevel@tonic-gate 		 * svc.configd(1M) restart handling performed by the
21167c478bd9Sstevel@tonic-gate 		 * fork_configd_thread.  We don't acknowledge, as that thread
21177c478bd9Sstevel@tonic-gate 		 * will do so.
21187c478bd9Sstevel@tonic-gate 		 */
21197c478bd9Sstevel@tonic-gate 		if (cookie == CONFIGD_COOKIE) {
21207c478bd9Sstevel@tonic-gate 			ct_event_free(ev);
21217c478bd9Sstevel@tonic-gate 			continue;
21227c478bd9Sstevel@tonic-gate 		}
21237c478bd9Sstevel@tonic-gate 
2124dfe57350Sjeanm 		inst = NULL;
2125dfe57350Sjeanm 		if (storing_contract != 0 &&
2126dfe57350Sjeanm 		    (inst = contract_to_inst(ctid)) == NULL) {
2127dfe57350Sjeanm 			/*
2128dfe57350Sjeanm 			 * This can happen for two reasons:
2129dfe57350Sjeanm 			 * - method_run() has not yet stored the
2130dfe57350Sjeanm 			 *    the contract into the internal hash table.
2131dfe57350Sjeanm 			 * - we receive an EMPTY event for an abandoned
2132dfe57350Sjeanm 			 *    contract.
2133dfe57350Sjeanm 			 * If there is any contract in the process of
2134dfe57350Sjeanm 			 * being stored into the hash table then re-read
2135dfe57350Sjeanm 			 * the event later.
2136dfe57350Sjeanm 			 */
2137dfe57350Sjeanm 			log_framework(LOG_DEBUG,
2138dfe57350Sjeanm 			    "Reset event %d for unknown "
2139dfe57350Sjeanm 			    "contract id %ld\n", type, ctid);
2140dfe57350Sjeanm 
2141dfe57350Sjeanm 			/* don't go too fast */
2142dfe57350Sjeanm 			(void) poll(NULL, 0, 100);
2143dfe57350Sjeanm 
2144dfe57350Sjeanm 			(void) ct_event_reset(fd);
2145dfe57350Sjeanm 			ct_event_free(ev);
2146dfe57350Sjeanm 			continue;
2147dfe57350Sjeanm 		}
2148dfe57350Sjeanm 
2149dfe57350Sjeanm 		/*
2150dfe57350Sjeanm 		 * Do not call contract_to_inst() again if first
2151dfe57350Sjeanm 		 * call succeeded.
2152dfe57350Sjeanm 		 */
2153dfe57350Sjeanm 		if (inst == NULL)
21547c478bd9Sstevel@tonic-gate 			inst = contract_to_inst(ctid);
21557c478bd9Sstevel@tonic-gate 		if (inst == NULL) {
21567c478bd9Sstevel@tonic-gate 			/*
21577c478bd9Sstevel@tonic-gate 			 * This can happen if we receive an EMPTY
21587c478bd9Sstevel@tonic-gate 			 * event for an abandoned contract.
21597c478bd9Sstevel@tonic-gate 			 */
21607c478bd9Sstevel@tonic-gate 			log_framework(LOG_DEBUG,
21617c478bd9Sstevel@tonic-gate 			    "Received event %d for unknown contract id "
21627c478bd9Sstevel@tonic-gate 			    "%ld\n", type, ctid);
21637c478bd9Sstevel@tonic-gate 		} else {
21647c478bd9Sstevel@tonic-gate 			log_framework(LOG_DEBUG,
21657c478bd9Sstevel@tonic-gate 			    "Received event %d for contract id "
21667c478bd9Sstevel@tonic-gate 			    "%ld (%s)\n", type, ctid,
21677c478bd9Sstevel@tonic-gate 			    inst->ri_i.i_fmri);
21687c478bd9Sstevel@tonic-gate 
21697c478bd9Sstevel@tonic-gate 			contract_action(local_handle, inst, ctid, type);
21707c478bd9Sstevel@tonic-gate 
21717c478bd9Sstevel@tonic-gate 			MUTEX_UNLOCK(&inst->ri_lock);
21727c478bd9Sstevel@tonic-gate 		}
21737c478bd9Sstevel@tonic-gate 
21747c478bd9Sstevel@tonic-gate 		efd = contract_open(ct_event_get_ctid(ev), "process", "ctl",
21757c478bd9Sstevel@tonic-gate 		    O_WRONLY);
21767c478bd9Sstevel@tonic-gate 		if (efd != -1) {
21777c478bd9Sstevel@tonic-gate 			(void) ct_ctl_ack(efd, evid);
21787c478bd9Sstevel@tonic-gate 			startd_close(efd);
21797c478bd9Sstevel@tonic-gate 		}
21807c478bd9Sstevel@tonic-gate 
21817c478bd9Sstevel@tonic-gate 		ct_event_free(ev);
21827c478bd9Sstevel@tonic-gate 
21837c478bd9Sstevel@tonic-gate 	}
21847c478bd9Sstevel@tonic-gate 
21857c478bd9Sstevel@tonic-gate 	/*NOTREACHED*/
21867c478bd9Sstevel@tonic-gate 	return (NULL);
21877c478bd9Sstevel@tonic-gate }
21887c478bd9Sstevel@tonic-gate 
21897c478bd9Sstevel@tonic-gate /*
21907c478bd9Sstevel@tonic-gate  * Timeout queue, processed by restarter_timeouts_event_thread().
21917c478bd9Sstevel@tonic-gate  */
21927c478bd9Sstevel@tonic-gate timeout_queue_t *timeouts;
21937c478bd9Sstevel@tonic-gate static uu_list_pool_t *timeout_pool;
21947c478bd9Sstevel@tonic-gate 
21957c478bd9Sstevel@tonic-gate typedef struct timeout_update {
21967c478bd9Sstevel@tonic-gate 	pthread_mutex_t		tu_lock;
21977c478bd9Sstevel@tonic-gate 	pthread_cond_t		tu_cv;
21987c478bd9Sstevel@tonic-gate 	int			tu_wakeup;
21997c478bd9Sstevel@tonic-gate } timeout_update_t;
22007c478bd9Sstevel@tonic-gate 
22017c478bd9Sstevel@tonic-gate timeout_update_t *tu;
22027c478bd9Sstevel@tonic-gate 
22037c478bd9Sstevel@tonic-gate static const char *timeout_ovr_svcs[] = {
22047c478bd9Sstevel@tonic-gate 	"svc:/system/manifest-import:default",
22057c478bd9Sstevel@tonic-gate 	"svc:/network/initial:default",
22067c478bd9Sstevel@tonic-gate 	"svc:/network/service:default",
22077c478bd9Sstevel@tonic-gate 	"svc:/system/rmtmpfiles:default",
22087c478bd9Sstevel@tonic-gate 	"svc:/network/loopback:default",
22097c478bd9Sstevel@tonic-gate 	"svc:/network/physical:default",
22107c478bd9Sstevel@tonic-gate 	"svc:/system/device/local:default",
22117c478bd9Sstevel@tonic-gate 	"svc:/system/metainit:default",
22127c478bd9Sstevel@tonic-gate 	"svc:/system/filesystem/usr:default",
22137c478bd9Sstevel@tonic-gate 	"svc:/system/filesystem/minimal:default",
22147c478bd9Sstevel@tonic-gate 	"svc:/system/filesystem/local:default",
22157c478bd9Sstevel@tonic-gate 	NULL
22167c478bd9Sstevel@tonic-gate };
22177c478bd9Sstevel@tonic-gate 
22187c478bd9Sstevel@tonic-gate int
22197c478bd9Sstevel@tonic-gate is_timeout_ovr(restarter_inst_t *inst)
22207c478bd9Sstevel@tonic-gate {
22217c478bd9Sstevel@tonic-gate 	int i;
22227c478bd9Sstevel@tonic-gate 
22237c478bd9Sstevel@tonic-gate 	for (i = 0; timeout_ovr_svcs[i] != NULL; ++i) {
22247c478bd9Sstevel@tonic-gate 		if (strcmp(inst->ri_i.i_fmri, timeout_ovr_svcs[i]) == 0) {
22257c478bd9Sstevel@tonic-gate 			log_instance(inst, B_TRUE, "Timeout override by "
22265ca87c7fSlianep 			    "svc.startd.  Using infinite timeout.");
22277c478bd9Sstevel@tonic-gate 			return (1);
22287c478bd9Sstevel@tonic-gate 		}
22297c478bd9Sstevel@tonic-gate 	}
22307c478bd9Sstevel@tonic-gate 
22317c478bd9Sstevel@tonic-gate 	return (0);
22327c478bd9Sstevel@tonic-gate }
22337c478bd9Sstevel@tonic-gate 
22347c478bd9Sstevel@tonic-gate /*ARGSUSED*/
22357c478bd9Sstevel@tonic-gate static int
22367c478bd9Sstevel@tonic-gate timeout_compare(const void *lc_arg, const void *rc_arg, void *private)
22377c478bd9Sstevel@tonic-gate {
22387c478bd9Sstevel@tonic-gate 	hrtime_t t1 = ((const timeout_entry_t *)lc_arg)->te_timeout;
22397c478bd9Sstevel@tonic-gate 	hrtime_t t2 = ((const timeout_entry_t *)rc_arg)->te_timeout;
22407c478bd9Sstevel@tonic-gate 
22417c478bd9Sstevel@tonic-gate 	if (t1 > t2)
22427c478bd9Sstevel@tonic-gate 		return (1);
22437c478bd9Sstevel@tonic-gate 	else if (t1 < t2)
22447c478bd9Sstevel@tonic-gate 		return (-1);
22457c478bd9Sstevel@tonic-gate 	return (0);
22467c478bd9Sstevel@tonic-gate }
22477c478bd9Sstevel@tonic-gate 
22487c478bd9Sstevel@tonic-gate void
22497c478bd9Sstevel@tonic-gate timeout_init()
22507c478bd9Sstevel@tonic-gate {
22517c478bd9Sstevel@tonic-gate 	timeouts = startd_zalloc(sizeof (timeout_queue_t));
22527c478bd9Sstevel@tonic-gate 
22537c478bd9Sstevel@tonic-gate 	(void) pthread_mutex_init(&timeouts->tq_lock, &mutex_attrs);
22547c478bd9Sstevel@tonic-gate 
22557c478bd9Sstevel@tonic-gate 	timeout_pool = startd_list_pool_create("timeouts",
22567c478bd9Sstevel@tonic-gate 	    sizeof (timeout_entry_t), offsetof(timeout_entry_t, te_link),
22577c478bd9Sstevel@tonic-gate 	    timeout_compare, UU_LIST_POOL_DEBUG);
22587c478bd9Sstevel@tonic-gate 	assert(timeout_pool != NULL);
22597c478bd9Sstevel@tonic-gate 
22607c478bd9Sstevel@tonic-gate 	timeouts->tq_list = startd_list_create(timeout_pool,
22617c478bd9Sstevel@tonic-gate 	    timeouts, UU_LIST_SORTED);
22627c478bd9Sstevel@tonic-gate 	assert(timeouts->tq_list != NULL);
22637c478bd9Sstevel@tonic-gate 
22647c478bd9Sstevel@tonic-gate 	tu = startd_zalloc(sizeof (timeout_update_t));
22657c478bd9Sstevel@tonic-gate 	(void) pthread_cond_init(&tu->tu_cv, NULL);
22667c478bd9Sstevel@tonic-gate 	(void) pthread_mutex_init(&tu->tu_lock, &mutex_attrs);
22677c478bd9Sstevel@tonic-gate }
22687c478bd9Sstevel@tonic-gate 
22697c478bd9Sstevel@tonic-gate void
22707c478bd9Sstevel@tonic-gate timeout_insert(restarter_inst_t *inst, ctid_t cid, uint64_t timeout_sec)
22717c478bd9Sstevel@tonic-gate {
22727c478bd9Sstevel@tonic-gate 	hrtime_t now, timeout;
22737c478bd9Sstevel@tonic-gate 	timeout_entry_t *entry;
22747c478bd9Sstevel@tonic-gate 	uu_list_index_t idx;
22757c478bd9Sstevel@tonic-gate 
227653f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&inst->ri_lock));
22777c478bd9Sstevel@tonic-gate 
22787c478bd9Sstevel@tonic-gate 	now = gethrtime();
22797c478bd9Sstevel@tonic-gate 
22807c478bd9Sstevel@tonic-gate 	/*
22817c478bd9Sstevel@tonic-gate 	 * If we overflow LLONG_MAX, we're never timing out anyways, so
22827c478bd9Sstevel@tonic-gate 	 * just return.
22837c478bd9Sstevel@tonic-gate 	 */
22847c478bd9Sstevel@tonic-gate 	if (timeout_sec >= (LLONG_MAX - now) / 1000000000LL) {
22857c478bd9Sstevel@tonic-gate 		log_instance(inst, B_TRUE, "timeout_seconds too large, "
22867c478bd9Sstevel@tonic-gate 		    "treating as infinite.");
22877c478bd9Sstevel@tonic-gate 		return;
22887c478bd9Sstevel@tonic-gate 	}
22897c478bd9Sstevel@tonic-gate 
22907c478bd9Sstevel@tonic-gate 	/* hrtime is in nanoseconds. Convert timeout_sec. */
22917c478bd9Sstevel@tonic-gate 	timeout = now + (timeout_sec * 1000000000LL);
22927c478bd9Sstevel@tonic-gate 
22937c478bd9Sstevel@tonic-gate 	entry = startd_alloc(sizeof (timeout_entry_t));
22947c478bd9Sstevel@tonic-gate 	entry->te_timeout = timeout;
22957c478bd9Sstevel@tonic-gate 	entry->te_ctid = cid;
22967c478bd9Sstevel@tonic-gate 	entry->te_fmri = safe_strdup(inst->ri_i.i_fmri);
22977c478bd9Sstevel@tonic-gate 	entry->te_logstem = safe_strdup(inst->ri_logstem);
22987c478bd9Sstevel@tonic-gate 	entry->te_fired = 0;
22997c478bd9Sstevel@tonic-gate 	/* Insert the calculated timeout time onto the queue. */
23007c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&timeouts->tq_lock);
23017c478bd9Sstevel@tonic-gate 	(void) uu_list_find(timeouts->tq_list, entry, NULL, &idx);
23027c478bd9Sstevel@tonic-gate 	uu_list_node_init(entry, &entry->te_link, timeout_pool);
23037c478bd9Sstevel@tonic-gate 	uu_list_insert(timeouts->tq_list, entry, idx);
23047c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&timeouts->tq_lock);
23057c478bd9Sstevel@tonic-gate 
23067c478bd9Sstevel@tonic-gate 	assert(inst->ri_timeout == NULL);
23077c478bd9Sstevel@tonic-gate 	inst->ri_timeout = entry;
23087c478bd9Sstevel@tonic-gate 
23097c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&tu->tu_lock);
23107c478bd9Sstevel@tonic-gate 	tu->tu_wakeup = 1;
23117c478bd9Sstevel@tonic-gate 	(void) pthread_cond_broadcast(&tu->tu_cv);
23127c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&tu->tu_lock);
23137c478bd9Sstevel@tonic-gate }
23147c478bd9Sstevel@tonic-gate 
23157c478bd9Sstevel@tonic-gate 
23167c478bd9Sstevel@tonic-gate void
23177c478bd9Sstevel@tonic-gate timeout_remove(restarter_inst_t *inst, ctid_t cid)
23187c478bd9Sstevel@tonic-gate {
231953f3aea0SRoger A. Faulkner 	assert(MUTEX_HELD(&inst->ri_lock));
23207c478bd9Sstevel@tonic-gate 
23217c478bd9Sstevel@tonic-gate 	if (inst->ri_timeout == NULL)
23227c478bd9Sstevel@tonic-gate 		return;
23237c478bd9Sstevel@tonic-gate 
23247c478bd9Sstevel@tonic-gate 	assert(inst->ri_timeout->te_ctid == cid);
23257c478bd9Sstevel@tonic-gate 
23267c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&timeouts->tq_lock);
23277c478bd9Sstevel@tonic-gate 	uu_list_remove(timeouts->tq_list, inst->ri_timeout);
23287c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&timeouts->tq_lock);
23297c478bd9Sstevel@tonic-gate 
23307c478bd9Sstevel@tonic-gate 	free(inst->ri_timeout->te_fmri);
23317c478bd9Sstevel@tonic-gate 	free(inst->ri_timeout->te_logstem);
23327c478bd9Sstevel@tonic-gate 	startd_free(inst->ri_timeout, sizeof (timeout_entry_t));
23337c478bd9Sstevel@tonic-gate 	inst->ri_timeout = NULL;
23347c478bd9Sstevel@tonic-gate }
23357c478bd9Sstevel@tonic-gate 
23367c478bd9Sstevel@tonic-gate static int
23377c478bd9Sstevel@tonic-gate timeout_now()
23387c478bd9Sstevel@tonic-gate {
23397c478bd9Sstevel@tonic-gate 	timeout_entry_t *e;
23407c478bd9Sstevel@tonic-gate 	hrtime_t now;
23417c478bd9Sstevel@tonic-gate 	int ret;
23427c478bd9Sstevel@tonic-gate 
23437c478bd9Sstevel@tonic-gate 	now = gethrtime();
23447c478bd9Sstevel@tonic-gate 
23457c478bd9Sstevel@tonic-gate 	/*
23467c478bd9Sstevel@tonic-gate 	 * Walk through the (sorted) timeouts list.  While the timeout
23477c478bd9Sstevel@tonic-gate 	 * at the head of the list is <= the current time, kill the
23487c478bd9Sstevel@tonic-gate 	 * method.
23497c478bd9Sstevel@tonic-gate 	 */
23507c478bd9Sstevel@tonic-gate 	MUTEX_LOCK(&timeouts->tq_lock);
23517c478bd9Sstevel@tonic-gate 
23527c478bd9Sstevel@tonic-gate 	for (e = uu_list_first(timeouts->tq_list);
23537c478bd9Sstevel@tonic-gate 	    e != NULL && e->te_timeout <= now;
23547c478bd9Sstevel@tonic-gate 	    e = uu_list_next(timeouts->tq_list, e)) {
23557c478bd9Sstevel@tonic-gate 		log_framework(LOG_WARNING, "%s: Method or service exit timed "
23567c478bd9Sstevel@tonic-gate 		    "out.  Killing contract %ld.\n", e->te_fmri, e->te_ctid);
23577c478bd9Sstevel@tonic-gate 		log_instance_fmri(e->te_fmri, e->te_logstem, B_TRUE,
23585ca87c7fSlianep 		    "Method or service exit timed out.  Killing contract %ld.",
23597c478bd9Sstevel@tonic-gate 		    e->te_ctid);
23607c478bd9Sstevel@tonic-gate 		e->te_fired = 1;
23617c478bd9Sstevel@tonic-gate 		(void) contract_kill(e->te_ctid, SIGKILL, e->te_fmri);
23627c478bd9Sstevel@tonic-gate 	}
23637c478bd9Sstevel@tonic-gate 
23647c478bd9Sstevel@tonic-gate 	if (uu_list_numnodes(timeouts->tq_list) > 0)
23657c478bd9Sstevel@tonic-gate 		ret = 0;
23667c478bd9Sstevel@tonic-gate 	else
23677c478bd9Sstevel@tonic-gate 		ret = -1;
23687c478bd9Sstevel@tonic-gate 
23697c478bd9Sstevel@tonic-gate 	MUTEX_UNLOCK(&timeouts->tq_lock);
23707c478bd9Sstevel@tonic-gate 
23717c478bd9Sstevel@tonic-gate 	return (ret);
23727c478bd9Sstevel@tonic-gate }
23737c478bd9Sstevel@tonic-gate 
23747c478bd9Sstevel@tonic-gate /*
23757c478bd9Sstevel@tonic-gate  * void *restarter_timeouts_event_thread(void *)
23767c478bd9Sstevel@tonic-gate  *   Responsible for monitoring the method timeouts.  This thread must
23777c478bd9Sstevel@tonic-gate  *   be started before any methods are called.
23787c478bd9Sstevel@tonic-gate  */
23797c478bd9Sstevel@tonic-gate /*ARGSUSED*/
23807c478bd9Sstevel@tonic-gate static void *
23817c478bd9Sstevel@tonic-gate restarter_timeouts_event_thread(void *unused)
23827c478bd9Sstevel@tonic-gate {
23837c478bd9Sstevel@tonic-gate 	/*
23847c478bd9Sstevel@tonic-gate 	 * Timeouts are entered on a priority queue, which is processed by
23857c478bd9Sstevel@tonic-gate 	 * this thread.  As timeouts are specified in seconds, we'll do
23867c478bd9Sstevel@tonic-gate 	 * the necessary processing every second, as long as the queue
23877c478bd9Sstevel@tonic-gate 	 * is not empty.
23887c478bd9Sstevel@tonic-gate 	 */
23897c478bd9Sstevel@tonic-gate 
23907c478bd9Sstevel@tonic-gate 	/*CONSTCOND*/
23917c478bd9Sstevel@tonic-gate 	while (1) {
23927c478bd9Sstevel@tonic-gate 		/*
23937c478bd9Sstevel@tonic-gate 		 * As long as the timeout list isn't empty, process it
23947c478bd9Sstevel@tonic-gate 		 * every second.
23957c478bd9Sstevel@tonic-gate 		 */
23967c478bd9Sstevel@tonic-gate 		if (timeout_now() == 0) {
23977c478bd9Sstevel@tonic-gate 			(void) sleep(1);
23987c478bd9Sstevel@tonic-gate 			continue;
23997c478bd9Sstevel@tonic-gate 		}
24007c478bd9Sstevel@tonic-gate 
24017c478bd9Sstevel@tonic-gate 		/* The list is empty, wait until we have more timeouts. */
24027c478bd9Sstevel@tonic-gate 		MUTEX_LOCK(&tu->tu_lock);
24037c478bd9Sstevel@tonic-gate 
24047c478bd9Sstevel@tonic-gate 		while (tu->tu_wakeup == 0)
24057c478bd9Sstevel@tonic-gate 			(void) pthread_cond_wait(&tu->tu_cv, &tu->tu_lock);
24067c478bd9Sstevel@tonic-gate 
24077c478bd9Sstevel@tonic-gate 		tu->tu_wakeup = 0;
24087c478bd9Sstevel@tonic-gate 		MUTEX_UNLOCK(&tu->tu_lock);
24097c478bd9Sstevel@tonic-gate 	}
24107c478bd9Sstevel@tonic-gate 
24117c478bd9Sstevel@tonic-gate 	return (NULL);
24127c478bd9Sstevel@tonic-gate }
24137c478bd9Sstevel@tonic-gate 
24147c478bd9Sstevel@tonic-gate void
24157c478bd9Sstevel@tonic-gate restarter_start()
24167c478bd9Sstevel@tonic-gate {
24177c478bd9Sstevel@tonic-gate 	(void) startd_thread_create(restarter_timeouts_event_thread, NULL);
24187c478bd9Sstevel@tonic-gate 	(void) startd_thread_create(restarter_event_thread, NULL);
24197c478bd9Sstevel@tonic-gate 	(void) startd_thread_create(restarter_contracts_event_thread, NULL);
24207c478bd9Sstevel@tonic-gate 	(void) startd_thread_create(wait_thread, NULL);
24217c478bd9Sstevel@tonic-gate }
24227c478bd9Sstevel@tonic-gate 
24237c478bd9Sstevel@tonic-gate 
24247c478bd9Sstevel@tonic-gate void
24257c478bd9Sstevel@tonic-gate restarter_init()
24267c478bd9Sstevel@tonic-gate {
24277c478bd9Sstevel@tonic-gate 	restarter_instance_pool = startd_list_pool_create("restarter_instances",
24287c478bd9Sstevel@tonic-gate 	    sizeof (restarter_inst_t), offsetof(restarter_inst_t,
24297c478bd9Sstevel@tonic-gate 	    ri_link), restarter_instance_compare, UU_LIST_POOL_DEBUG);
24307c478bd9Sstevel@tonic-gate 	(void) memset(&instance_list, 0, sizeof (instance_list));
24317c478bd9Sstevel@tonic-gate 
24327c478bd9Sstevel@tonic-gate 	(void) pthread_mutex_init(&instance_list.ril_lock, &mutex_attrs);
24337c478bd9Sstevel@tonic-gate 	instance_list.ril_instance_list = startd_list_create(
24347c478bd9Sstevel@tonic-gate 	    restarter_instance_pool, &instance_list, UU_LIST_SORTED);
24357c478bd9Sstevel@tonic-gate 
24367c478bd9Sstevel@tonic-gate 	restarter_queue_pool = startd_list_pool_create(
24377c478bd9Sstevel@tonic-gate 	    "restarter_instance_queue", sizeof (restarter_instance_qentry_t),
24387c478bd9Sstevel@tonic-gate 	    offsetof(restarter_instance_qentry_t,  riq_link), NULL,
24397c478bd9Sstevel@tonic-gate 	    UU_LIST_POOL_DEBUG);
24407c478bd9Sstevel@tonic-gate 
24417c478bd9Sstevel@tonic-gate 	contract_list_pool = startd_list_pool_create(
24427c478bd9Sstevel@tonic-gate 	    "contract_list", sizeof (contract_entry_t),
24437c478bd9Sstevel@tonic-gate 	    offsetof(contract_entry_t,  ce_link), NULL,
24447c478bd9Sstevel@tonic-gate 	    UU_LIST_POOL_DEBUG);
24457c478bd9Sstevel@tonic-gate 	contract_hash_init();
24467c478bd9Sstevel@tonic-gate 
24477c478bd9Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "Initialized restarter\n");
24487c478bd9Sstevel@tonic-gate }
2449