xref: /onnv-gate/usr/src/cmd/svc/startd/restarter.c (revision 0:68f95e015346)
1*0Sstevel@tonic-gate /*
2*0Sstevel@tonic-gate  * CDDL HEADER START
3*0Sstevel@tonic-gate  *
4*0Sstevel@tonic-gate  * The contents of this file are subject to the terms of the
5*0Sstevel@tonic-gate  * Common Development and Distribution License, Version 1.0 only
6*0Sstevel@tonic-gate  * (the "License").  You may not use this file except in compliance
7*0Sstevel@tonic-gate  * with the License.
8*0Sstevel@tonic-gate  *
9*0Sstevel@tonic-gate  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10*0Sstevel@tonic-gate  * or http://www.opensolaris.org/os/licensing.
11*0Sstevel@tonic-gate  * See the License for the specific language governing permissions
12*0Sstevel@tonic-gate  * and limitations under the License.
13*0Sstevel@tonic-gate  *
14*0Sstevel@tonic-gate  * When distributing Covered Code, include this CDDL HEADER in each
15*0Sstevel@tonic-gate  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16*0Sstevel@tonic-gate  * If applicable, add the following below this CDDL HEADER, with the
17*0Sstevel@tonic-gate  * fields enclosed by brackets "[]" replaced with your own identifying
18*0Sstevel@tonic-gate  * information: Portions Copyright [yyyy] [name of copyright owner]
19*0Sstevel@tonic-gate  *
20*0Sstevel@tonic-gate  * CDDL HEADER END
21*0Sstevel@tonic-gate  */
22*0Sstevel@tonic-gate /*
23*0Sstevel@tonic-gate  * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
24*0Sstevel@tonic-gate  * Use is subject to license terms.
25*0Sstevel@tonic-gate  */
26*0Sstevel@tonic-gate 
27*0Sstevel@tonic-gate #pragma ident	"%Z%%M%	%I%	%E% SMI"
28*0Sstevel@tonic-gate 
29*0Sstevel@tonic-gate /*
30*0Sstevel@tonic-gate  * restarter.c - service manipulation
31*0Sstevel@tonic-gate  *
32*0Sstevel@tonic-gate  * This component manages services whose restarter is svc.startd, the standard
33*0Sstevel@tonic-gate  * restarter.  It translates restarter protocol events from the graph engine
34*0Sstevel@tonic-gate  * into actions on processes, as a delegated restarter would do.
35*0Sstevel@tonic-gate  *
36*0Sstevel@tonic-gate  * The master restarter manages a number of always-running threads:
37*0Sstevel@tonic-gate  *   - restarter event thread: events from the graph engine
38*0Sstevel@tonic-gate  *   - timeout thread: thread to fire queued timeouts
39*0Sstevel@tonic-gate  *   - contract thread: thread to handle contract events
40*0Sstevel@tonic-gate  *   - wait thread: thread to handle wait-based services
41*0Sstevel@tonic-gate  *
42*0Sstevel@tonic-gate  * The other threads are created as-needed:
43*0Sstevel@tonic-gate  *   - per-instance method threads
44*0Sstevel@tonic-gate  *   - per-instance event processing threads
45*0Sstevel@tonic-gate  *
46*0Sstevel@tonic-gate  * The interaction of all threads must result in the following conditions
47*0Sstevel@tonic-gate  * being satisfied (on a per-instance basis):
48*0Sstevel@tonic-gate  *   - restarter events must be processed in order
49*0Sstevel@tonic-gate  *   - method execution must be serialized
50*0Sstevel@tonic-gate  *   - instance delete must be held until outstanding methods are complete
51*0Sstevel@tonic-gate  *   - contract events shouldn't be processed while a method is running
52*0Sstevel@tonic-gate  *   - timeouts should fire even when a method is running
53*0Sstevel@tonic-gate  *
54*0Sstevel@tonic-gate  * Service instances are represented by restarter_inst_t's and are kept in the
55*0Sstevel@tonic-gate  * instance_list list.
56*0Sstevel@tonic-gate  *
57*0Sstevel@tonic-gate  * Service States
58*0Sstevel@tonic-gate  *   The current state of a service instance is kept in
59*0Sstevel@tonic-gate  *   restarter_inst_t->ri_i.i_state.  If transition to a new state could take
60*0Sstevel@tonic-gate  *   some time, then before we effect the transition we set
61*0Sstevel@tonic-gate  *   restarter_inst_t->ri_i.i_next_state to the target state, and afterwards we
62*0Sstevel@tonic-gate  *   rotate i_next_state to i_state and set i_next_state to
63*0Sstevel@tonic-gate  *   RESTARTER_STATE_NONE.  So usually i_next_state is _NONE when ri_lock is not
64*0Sstevel@tonic-gate  *   held.  The exception is when we launch methods, which are done with
65*0Sstevel@tonic-gate  *   a separate thread.  To keep any other threads from grabbing ri_lock before
66*0Sstevel@tonic-gate  *   method_thread() does, we set ri_method_thread to the thread id of the
67*0Sstevel@tonic-gate  *   method thread, and when it is nonzero any thread with a different thread id
68*0Sstevel@tonic-gate  *   waits on ri_method_cv.
69*0Sstevel@tonic-gate  *
70*0Sstevel@tonic-gate  * Method execution is serialized by blocking on ri_method_cv in
71*0Sstevel@tonic-gate  * inst_lookup_by_id() and waiting for a 0 value of ri_method_thread.  This
72*0Sstevel@tonic-gate  * also prevents the instance structure from being deleted until all
73*0Sstevel@tonic-gate  * outstanding operations such as method_thread() have finished.
74*0Sstevel@tonic-gate  *
75*0Sstevel@tonic-gate  * Lock ordering:
76*0Sstevel@tonic-gate  *
77*0Sstevel@tonic-gate  * dgraph_lock [can be held when taking:]
78*0Sstevel@tonic-gate  *   utmpx_lock
79*0Sstevel@tonic-gate  *   dictionary->dict_lock
80*0Sstevel@tonic-gate  *   st->st_load_lock
81*0Sstevel@tonic-gate  *   wait_info_lock
82*0Sstevel@tonic-gate  *   ru->restarter_update_lock
83*0Sstevel@tonic-gate  *     restarter_queue->rpeq_lock
84*0Sstevel@tonic-gate  *   instance_list.ril_lock
85*0Sstevel@tonic-gate  *     inst->ri_lock
86*0Sstevel@tonic-gate  *   st->st_configd_live_lock
87*0Sstevel@tonic-gate  *
88*0Sstevel@tonic-gate  * instance_list.ril_lock
89*0Sstevel@tonic-gate  *   graph_queue->gpeq_lock
90*0Sstevel@tonic-gate  *   gu->gu_lock
91*0Sstevel@tonic-gate  *   st->st_configd_live_lock
92*0Sstevel@tonic-gate  *   dictionary->dict_lock
93*0Sstevel@tonic-gate  *   inst->ri_lock
94*0Sstevel@tonic-gate  *     graph_queue->gpeq_lock
95*0Sstevel@tonic-gate  *     gu->gu_lock
96*0Sstevel@tonic-gate  *     tu->tu_lock
97*0Sstevel@tonic-gate  *     tq->tq_lock
98*0Sstevel@tonic-gate  *     inst->ri_queue_lock
99*0Sstevel@tonic-gate  *       wait_info_lock
100*0Sstevel@tonic-gate  *       bp->cb_lock
101*0Sstevel@tonic-gate  *     utmpx_lock
102*0Sstevel@tonic-gate  *
103*0Sstevel@tonic-gate  * single_user_thread_lock
104*0Sstevel@tonic-gate  *   wait_info_lock
105*0Sstevel@tonic-gate  *   utmpx_lock
106*0Sstevel@tonic-gate  *
107*0Sstevel@tonic-gate  * gu_freeze_lock
108*0Sstevel@tonic-gate  *
109*0Sstevel@tonic-gate  * logbuf_mutex nests inside pretty much everything.
110*0Sstevel@tonic-gate  */
111*0Sstevel@tonic-gate 
112*0Sstevel@tonic-gate #include <sys/contract/process.h>
113*0Sstevel@tonic-gate #include <sys/ctfs.h>
114*0Sstevel@tonic-gate #include <sys/stat.h>
115*0Sstevel@tonic-gate #include <sys/time.h>
116*0Sstevel@tonic-gate #include <sys/types.h>
117*0Sstevel@tonic-gate #include <sys/uio.h>
118*0Sstevel@tonic-gate #include <sys/wait.h>
119*0Sstevel@tonic-gate #include <assert.h>
120*0Sstevel@tonic-gate #include <errno.h>
121*0Sstevel@tonic-gate #include <fcntl.h>
122*0Sstevel@tonic-gate #include <libcontract.h>
123*0Sstevel@tonic-gate #include <libcontract_priv.h>
124*0Sstevel@tonic-gate #include <libintl.h>
125*0Sstevel@tonic-gate #include <librestart.h>
126*0Sstevel@tonic-gate #include <librestart_priv.h>
127*0Sstevel@tonic-gate #include <libuutil.h>
128*0Sstevel@tonic-gate #include <limits.h>
129*0Sstevel@tonic-gate #include <poll.h>
130*0Sstevel@tonic-gate #include <port.h>
131*0Sstevel@tonic-gate #include <pthread.h>
132*0Sstevel@tonic-gate #include <stdarg.h>
133*0Sstevel@tonic-gate #include <stdio.h>
134*0Sstevel@tonic-gate #include <strings.h>
135*0Sstevel@tonic-gate #include <unistd.h>
136*0Sstevel@tonic-gate 
137*0Sstevel@tonic-gate #include "startd.h"
138*0Sstevel@tonic-gate #include "protocol.h"
139*0Sstevel@tonic-gate 
140*0Sstevel@tonic-gate static uu_list_pool_t *restarter_instance_pool;
141*0Sstevel@tonic-gate static restarter_instance_list_t instance_list;
142*0Sstevel@tonic-gate 
143*0Sstevel@tonic-gate static uu_list_pool_t *restarter_queue_pool;
144*0Sstevel@tonic-gate 
145*0Sstevel@tonic-gate /*ARGSUSED*/
146*0Sstevel@tonic-gate static int
147*0Sstevel@tonic-gate restarter_instance_compare(const void *lc_arg, const void *rc_arg,
148*0Sstevel@tonic-gate     void *private)
149*0Sstevel@tonic-gate {
150*0Sstevel@tonic-gate 	int lc_id = ((const restarter_inst_t *)lc_arg)->ri_id;
151*0Sstevel@tonic-gate 	int rc_id = *(int *)rc_arg;
152*0Sstevel@tonic-gate 
153*0Sstevel@tonic-gate 	if (lc_id > rc_id)
154*0Sstevel@tonic-gate 		return (1);
155*0Sstevel@tonic-gate 	if (lc_id < rc_id)
156*0Sstevel@tonic-gate 		return (-1);
157*0Sstevel@tonic-gate 	return (0);
158*0Sstevel@tonic-gate }
159*0Sstevel@tonic-gate 
160*0Sstevel@tonic-gate static restarter_inst_t *
161*0Sstevel@tonic-gate inst_lookup_by_name(const char *name)
162*0Sstevel@tonic-gate {
163*0Sstevel@tonic-gate 	int id;
164*0Sstevel@tonic-gate 
165*0Sstevel@tonic-gate 	id = dict_lookup_byname(name);
166*0Sstevel@tonic-gate 	if (id == -1)
167*0Sstevel@tonic-gate 		return (NULL);
168*0Sstevel@tonic-gate 
169*0Sstevel@tonic-gate 	return (inst_lookup_by_id(id));
170*0Sstevel@tonic-gate }
171*0Sstevel@tonic-gate 
172*0Sstevel@tonic-gate restarter_inst_t *
173*0Sstevel@tonic-gate inst_lookup_by_id(int id)
174*0Sstevel@tonic-gate {
175*0Sstevel@tonic-gate 	restarter_inst_t *inst;
176*0Sstevel@tonic-gate 
177*0Sstevel@tonic-gate 	MUTEX_LOCK(&instance_list.ril_lock);
178*0Sstevel@tonic-gate 	inst = uu_list_find(instance_list.ril_instance_list, &id, NULL, NULL);
179*0Sstevel@tonic-gate 	if (inst != NULL)
180*0Sstevel@tonic-gate 		MUTEX_LOCK(&inst->ri_lock);
181*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&instance_list.ril_lock);
182*0Sstevel@tonic-gate 
183*0Sstevel@tonic-gate 	if (inst != NULL) {
184*0Sstevel@tonic-gate 		while (inst->ri_method_thread != 0 &&
185*0Sstevel@tonic-gate 		    !pthread_equal(inst->ri_method_thread, pthread_self())) {
186*0Sstevel@tonic-gate 			++inst->ri_method_waiters;
187*0Sstevel@tonic-gate 			(void) pthread_cond_wait(&inst->ri_method_cv,
188*0Sstevel@tonic-gate 			    &inst->ri_lock);
189*0Sstevel@tonic-gate 			assert(inst->ri_method_waiters > 0);
190*0Sstevel@tonic-gate 			--inst->ri_method_waiters;
191*0Sstevel@tonic-gate 		}
192*0Sstevel@tonic-gate 	}
193*0Sstevel@tonic-gate 
194*0Sstevel@tonic-gate 	return (inst);
195*0Sstevel@tonic-gate }
196*0Sstevel@tonic-gate 
197*0Sstevel@tonic-gate static restarter_inst_t *
198*0Sstevel@tonic-gate inst_lookup_queue(const char *name)
199*0Sstevel@tonic-gate {
200*0Sstevel@tonic-gate 	int id;
201*0Sstevel@tonic-gate 	restarter_inst_t *inst;
202*0Sstevel@tonic-gate 
203*0Sstevel@tonic-gate 	id = dict_lookup_byname(name);
204*0Sstevel@tonic-gate 	if (id == -1)
205*0Sstevel@tonic-gate 		return (NULL);
206*0Sstevel@tonic-gate 
207*0Sstevel@tonic-gate 	MUTEX_LOCK(&instance_list.ril_lock);
208*0Sstevel@tonic-gate 	inst = uu_list_find(instance_list.ril_instance_list, &id, NULL, NULL);
209*0Sstevel@tonic-gate 	if (inst != NULL)
210*0Sstevel@tonic-gate 		MUTEX_LOCK(&inst->ri_queue_lock);
211*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&instance_list.ril_lock);
212*0Sstevel@tonic-gate 
213*0Sstevel@tonic-gate 	return (inst);
214*0Sstevel@tonic-gate }
215*0Sstevel@tonic-gate 
216*0Sstevel@tonic-gate const char *
217*0Sstevel@tonic-gate service_style(int flags)
218*0Sstevel@tonic-gate {
219*0Sstevel@tonic-gate 	switch (flags & RINST_STYLE_MASK) {
220*0Sstevel@tonic-gate 	case RINST_CONTRACT:	return ("contract");
221*0Sstevel@tonic-gate 	case RINST_TRANSIENT:	return ("transient");
222*0Sstevel@tonic-gate 	case RINST_WAIT:	return ("wait");
223*0Sstevel@tonic-gate 
224*0Sstevel@tonic-gate 	default:
225*0Sstevel@tonic-gate #ifndef NDEBUG
226*0Sstevel@tonic-gate 		uu_warn("%s:%d: Bad flags 0x%x.\n", __FILE__, __LINE__, flags);
227*0Sstevel@tonic-gate #endif
228*0Sstevel@tonic-gate 		abort();
229*0Sstevel@tonic-gate 		/* NOTREACHED */
230*0Sstevel@tonic-gate 	}
231*0Sstevel@tonic-gate }
232*0Sstevel@tonic-gate 
233*0Sstevel@tonic-gate /*
234*0Sstevel@tonic-gate  * Fails with ECONNABORTED or ECANCELED.
235*0Sstevel@tonic-gate  */
236*0Sstevel@tonic-gate static int
237*0Sstevel@tonic-gate check_contract(restarter_inst_t *inst, boolean_t primary,
238*0Sstevel@tonic-gate     scf_instance_t *scf_inst)
239*0Sstevel@tonic-gate {
240*0Sstevel@tonic-gate 	ctid_t *ctidp;
241*0Sstevel@tonic-gate 	int fd, r;
242*0Sstevel@tonic-gate 
243*0Sstevel@tonic-gate 	ctidp = primary ? &inst->ri_i.i_primary_ctid :
244*0Sstevel@tonic-gate 	    &inst->ri_i.i_transient_ctid;
245*0Sstevel@tonic-gate 
246*0Sstevel@tonic-gate 	assert(*ctidp >= 1);
247*0Sstevel@tonic-gate 
248*0Sstevel@tonic-gate 	fd = contract_open(*ctidp, NULL, "status", O_RDONLY);
249*0Sstevel@tonic-gate 	if (fd >= 0) {
250*0Sstevel@tonic-gate 		r = close(fd);
251*0Sstevel@tonic-gate 		assert(r == 0);
252*0Sstevel@tonic-gate 		return (0);
253*0Sstevel@tonic-gate 	}
254*0Sstevel@tonic-gate 
255*0Sstevel@tonic-gate 	r = restarter_remove_contract(scf_inst, *ctidp, primary ?
256*0Sstevel@tonic-gate 	    RESTARTER_CONTRACT_PRIMARY : RESTARTER_CONTRACT_TRANSIENT);
257*0Sstevel@tonic-gate 	switch (r) {
258*0Sstevel@tonic-gate 	case 0:
259*0Sstevel@tonic-gate 	case ECONNABORTED:
260*0Sstevel@tonic-gate 	case ECANCELED:
261*0Sstevel@tonic-gate 		*ctidp = 0;
262*0Sstevel@tonic-gate 		return (r);
263*0Sstevel@tonic-gate 
264*0Sstevel@tonic-gate 	case ENOMEM:
265*0Sstevel@tonic-gate 		uu_die("Out of memory\n");
266*0Sstevel@tonic-gate 		/* NOTREACHED */
267*0Sstevel@tonic-gate 
268*0Sstevel@tonic-gate 	case EPERM:
269*0Sstevel@tonic-gate 		uu_die("Insufficient privilege.\n");
270*0Sstevel@tonic-gate 		/* NOTREACHED */
271*0Sstevel@tonic-gate 
272*0Sstevel@tonic-gate 	case EACCES:
273*0Sstevel@tonic-gate 		uu_die("Repository backend access denied.\n");
274*0Sstevel@tonic-gate 		/* NOTREACHED */
275*0Sstevel@tonic-gate 
276*0Sstevel@tonic-gate 	case EROFS:
277*0Sstevel@tonic-gate 		log_error(LOG_INFO, "Could not remove unusable contract id %ld "
278*0Sstevel@tonic-gate 		    "for %s from repository.\n", *ctidp, inst->ri_i.i_fmri);
279*0Sstevel@tonic-gate 		return (0);
280*0Sstevel@tonic-gate 
281*0Sstevel@tonic-gate 	case EINVAL:
282*0Sstevel@tonic-gate 	case EBADF:
283*0Sstevel@tonic-gate 	default:
284*0Sstevel@tonic-gate 		assert(0);
285*0Sstevel@tonic-gate 		abort();
286*0Sstevel@tonic-gate 		/* NOTREACHED */
287*0Sstevel@tonic-gate 	}
288*0Sstevel@tonic-gate }
289*0Sstevel@tonic-gate 
290*0Sstevel@tonic-gate static int stop_instance(scf_handle_t *, restarter_inst_t *, stop_cause_t);
291*0Sstevel@tonic-gate 
292*0Sstevel@tonic-gate /*
293*0Sstevel@tonic-gate  * int restarter_insert_inst(scf_handle_t *, char *)
294*0Sstevel@tonic-gate  *   If the inst is already in the restarter list, return its id.  If the inst
295*0Sstevel@tonic-gate  *   is not in the restarter list, initialize a restarter_inst_t, initialize its
296*0Sstevel@tonic-gate  *   states, insert it into the list, and return 0.
297*0Sstevel@tonic-gate  *
298*0Sstevel@tonic-gate  *   Fails with
299*0Sstevel@tonic-gate  *     ENOENT - name is not in the repository
300*0Sstevel@tonic-gate  */
301*0Sstevel@tonic-gate static int
302*0Sstevel@tonic-gate restarter_insert_inst(scf_handle_t *h, const char *name)
303*0Sstevel@tonic-gate {
304*0Sstevel@tonic-gate 	int id, r;
305*0Sstevel@tonic-gate 	restarter_inst_t *inst;
306*0Sstevel@tonic-gate 	uu_list_index_t idx;
307*0Sstevel@tonic-gate 	scf_service_t *scf_svc;
308*0Sstevel@tonic-gate 	scf_instance_t *scf_inst;
309*0Sstevel@tonic-gate 	scf_snapshot_t *snap;
310*0Sstevel@tonic-gate 	scf_propertygroup_t *pg;
311*0Sstevel@tonic-gate 	char *svc_name, *inst_name;
312*0Sstevel@tonic-gate 	char logfilebuf[PATH_MAX];
313*0Sstevel@tonic-gate 	char *c;
314*0Sstevel@tonic-gate 	boolean_t do_commit_states;
315*0Sstevel@tonic-gate 	restarter_instance_state_t state, next_state;
316*0Sstevel@tonic-gate 	protocol_states_t *ps;
317*0Sstevel@tonic-gate 	pid_t start_pid;
318*0Sstevel@tonic-gate 
319*0Sstevel@tonic-gate 	MUTEX_LOCK(&instance_list.ril_lock);
320*0Sstevel@tonic-gate 
321*0Sstevel@tonic-gate 	/*
322*0Sstevel@tonic-gate 	 * We don't use inst_lookup_by_name() here because we want the lookup
323*0Sstevel@tonic-gate 	 * & insert to be atomic.
324*0Sstevel@tonic-gate 	 */
325*0Sstevel@tonic-gate 	id = dict_lookup_byname(name);
326*0Sstevel@tonic-gate 	if (id != -1) {
327*0Sstevel@tonic-gate 		inst = uu_list_find(instance_list.ril_instance_list, &id, NULL,
328*0Sstevel@tonic-gate 		    &idx);
329*0Sstevel@tonic-gate 		if (inst != NULL) {
330*0Sstevel@tonic-gate 			MUTEX_UNLOCK(&instance_list.ril_lock);
331*0Sstevel@tonic-gate 			return (0);
332*0Sstevel@tonic-gate 		}
333*0Sstevel@tonic-gate 	}
334*0Sstevel@tonic-gate 
335*0Sstevel@tonic-gate 	/* Allocate an instance */
336*0Sstevel@tonic-gate 	inst = startd_zalloc(sizeof (restarter_inst_t));
337*0Sstevel@tonic-gate 	inst->ri_utmpx_prefix = startd_alloc(max_scf_value_size);
338*0Sstevel@tonic-gate 	inst->ri_utmpx_prefix[0] = '\0';
339*0Sstevel@tonic-gate 
340*0Sstevel@tonic-gate 	inst->ri_i.i_fmri = startd_alloc(strlen(name) + 1);
341*0Sstevel@tonic-gate 	(void) strcpy((char *)inst->ri_i.i_fmri, name);
342*0Sstevel@tonic-gate 
343*0Sstevel@tonic-gate 	inst->ri_queue = startd_list_create(restarter_queue_pool, inst, 0);
344*0Sstevel@tonic-gate 
345*0Sstevel@tonic-gate 	/*
346*0Sstevel@tonic-gate 	 * id shouldn't be -1 since we use the same dictionary as graph.c, but
347*0Sstevel@tonic-gate 	 * just in case.
348*0Sstevel@tonic-gate 	 */
349*0Sstevel@tonic-gate 	inst->ri_id = (id != -1 ? id : dict_insert(name));
350*0Sstevel@tonic-gate 
351*0Sstevel@tonic-gate 	special_online_hooks_get(name, &inst->ri_pre_online_hook,
352*0Sstevel@tonic-gate 	    &inst->ri_post_online_hook, &inst->ri_post_offline_hook);
353*0Sstevel@tonic-gate 
354*0Sstevel@tonic-gate 	scf_svc = safe_scf_service_create(h);
355*0Sstevel@tonic-gate 	scf_inst = safe_scf_instance_create(h);
356*0Sstevel@tonic-gate 	pg = safe_scf_pg_create(h);
357*0Sstevel@tonic-gate 	svc_name = startd_alloc(max_scf_name_size);
358*0Sstevel@tonic-gate 	inst_name = startd_alloc(max_scf_name_size);
359*0Sstevel@tonic-gate 
360*0Sstevel@tonic-gate rep_retry:
361*0Sstevel@tonic-gate 	if (scf_handle_decode_fmri(h, name, NULL, scf_svc, scf_inst, NULL,
362*0Sstevel@tonic-gate 	    NULL, SCF_DECODE_FMRI_EXACT) != 0) {
363*0Sstevel@tonic-gate 		switch (scf_error()) {
364*0Sstevel@tonic-gate 		case SCF_ERROR_CONNECTION_BROKEN:
365*0Sstevel@tonic-gate 			libscf_handle_rebind(h);
366*0Sstevel@tonic-gate 			goto rep_retry;
367*0Sstevel@tonic-gate 
368*0Sstevel@tonic-gate 		case SCF_ERROR_NOT_FOUND:
369*0Sstevel@tonic-gate deleted:
370*0Sstevel@tonic-gate 			MUTEX_UNLOCK(&instance_list.ril_lock);
371*0Sstevel@tonic-gate 			startd_free(inst_name, max_scf_name_size);
372*0Sstevel@tonic-gate 			startd_free(svc_name, max_scf_name_size);
373*0Sstevel@tonic-gate 			scf_pg_destroy(pg);
374*0Sstevel@tonic-gate 			scf_instance_destroy(scf_inst);
375*0Sstevel@tonic-gate 			scf_service_destroy(scf_svc);
376*0Sstevel@tonic-gate 			startd_free((void *)inst->ri_i.i_fmri,
377*0Sstevel@tonic-gate 			    strlen(inst->ri_i.i_fmri) + 1);
378*0Sstevel@tonic-gate 			startd_free(inst, sizeof (restarter_inst_t));
379*0Sstevel@tonic-gate 			return (ENOENT);
380*0Sstevel@tonic-gate 		}
381*0Sstevel@tonic-gate 
382*0Sstevel@tonic-gate 		uu_die("Can't decode FMRI %s: %s\n", name,
383*0Sstevel@tonic-gate 		    scf_strerror(scf_error()));
384*0Sstevel@tonic-gate 	}
385*0Sstevel@tonic-gate 
386*0Sstevel@tonic-gate 	/*
387*0Sstevel@tonic-gate 	 * If there's no running snapshot, then we execute using the editing
388*0Sstevel@tonic-gate 	 * snapshot.  Pending snapshots will be taken later.
389*0Sstevel@tonic-gate 	 */
390*0Sstevel@tonic-gate 	snap = libscf_get_running_snapshot(scf_inst);
391*0Sstevel@tonic-gate 
392*0Sstevel@tonic-gate 	if ((scf_service_get_name(scf_svc, svc_name, max_scf_name_size) < 0) ||
393*0Sstevel@tonic-gate 	    (scf_instance_get_name(scf_inst, inst_name, max_scf_name_size) <
394*0Sstevel@tonic-gate 	    0)) {
395*0Sstevel@tonic-gate 		switch (scf_error()) {
396*0Sstevel@tonic-gate 		case SCF_ERROR_NOT_SET:
397*0Sstevel@tonic-gate 			break;
398*0Sstevel@tonic-gate 
399*0Sstevel@tonic-gate 		case SCF_ERROR_CONNECTION_BROKEN:
400*0Sstevel@tonic-gate 			libscf_handle_rebind(h);
401*0Sstevel@tonic-gate 			goto rep_retry;
402*0Sstevel@tonic-gate 
403*0Sstevel@tonic-gate 		default:
404*0Sstevel@tonic-gate 			assert(0);
405*0Sstevel@tonic-gate 			abort();
406*0Sstevel@tonic-gate 		}
407*0Sstevel@tonic-gate 
408*0Sstevel@tonic-gate 		scf_snapshot_destroy(snap);
409*0Sstevel@tonic-gate 		goto deleted;
410*0Sstevel@tonic-gate 	}
411*0Sstevel@tonic-gate 
412*0Sstevel@tonic-gate 	/*
413*0Sstevel@tonic-gate 	 * If the restarter group is missing, use uninit/none.  Otherwise,
414*0Sstevel@tonic-gate 	 * we're probably being restarted & don't want to mess up the states
415*0Sstevel@tonic-gate 	 * that are there.
416*0Sstevel@tonic-gate 	 */
417*0Sstevel@tonic-gate 	state = RESTARTER_STATE_UNINIT;
418*0Sstevel@tonic-gate 	next_state = RESTARTER_STATE_NONE;
419*0Sstevel@tonic-gate 
420*0Sstevel@tonic-gate 	r = scf_instance_get_pg(scf_inst, SCF_PG_RESTARTER, pg);
421*0Sstevel@tonic-gate 	if (r != 0) {
422*0Sstevel@tonic-gate 		switch (scf_error()) {
423*0Sstevel@tonic-gate 		case SCF_ERROR_CONNECTION_BROKEN:
424*0Sstevel@tonic-gate 			libscf_handle_rebind(h);
425*0Sstevel@tonic-gate 			goto rep_retry;
426*0Sstevel@tonic-gate 
427*0Sstevel@tonic-gate 		case SCF_ERROR_NOT_SET:
428*0Sstevel@tonic-gate 			scf_snapshot_destroy(snap);
429*0Sstevel@tonic-gate 			goto deleted;
430*0Sstevel@tonic-gate 
431*0Sstevel@tonic-gate 		case SCF_ERROR_NOT_FOUND:
432*0Sstevel@tonic-gate 			/*
433*0Sstevel@tonic-gate 			 * This shouldn't happen since the graph engine should
434*0Sstevel@tonic-gate 			 * have initialized the state to uninitialized/none if
435*0Sstevel@tonic-gate 			 * there was no restarter pg.  In case somebody
436*0Sstevel@tonic-gate 			 * deleted it, though....
437*0Sstevel@tonic-gate 			 */
438*0Sstevel@tonic-gate 			do_commit_states = B_TRUE;
439*0Sstevel@tonic-gate 			break;
440*0Sstevel@tonic-gate 
441*0Sstevel@tonic-gate 		default:
442*0Sstevel@tonic-gate 			assert(0);
443*0Sstevel@tonic-gate 			abort();
444*0Sstevel@tonic-gate 		}
445*0Sstevel@tonic-gate 	} else {
446*0Sstevel@tonic-gate 		r = libscf_read_states(pg, &state, &next_state);
447*0Sstevel@tonic-gate 		if (r != 0) {
448*0Sstevel@tonic-gate 			do_commit_states = B_TRUE;
449*0Sstevel@tonic-gate 		} else {
450*0Sstevel@tonic-gate 			if (next_state != RESTARTER_STATE_NONE) {
451*0Sstevel@tonic-gate 				/*
452*0Sstevel@tonic-gate 				 * Force next_state to _NONE since we
453*0Sstevel@tonic-gate 				 * don't look for method processes.
454*0Sstevel@tonic-gate 				 */
455*0Sstevel@tonic-gate 				next_state = RESTARTER_STATE_NONE;
456*0Sstevel@tonic-gate 				do_commit_states = B_TRUE;
457*0Sstevel@tonic-gate 			} else {
458*0Sstevel@tonic-gate 				/*
459*0Sstevel@tonic-gate 				 * Inform the restarter of our state without
460*0Sstevel@tonic-gate 				 * changing the STIME in the repository.
461*0Sstevel@tonic-gate 				 */
462*0Sstevel@tonic-gate 				ps = startd_alloc(sizeof (*ps));
463*0Sstevel@tonic-gate 				inst->ri_i.i_state = ps->ps_state = state;
464*0Sstevel@tonic-gate 				inst->ri_i.i_next_state = ps->ps_state_next =
465*0Sstevel@tonic-gate 				    next_state;
466*0Sstevel@tonic-gate 
467*0Sstevel@tonic-gate 				graph_protocol_send_event(inst->ri_i.i_fmri,
468*0Sstevel@tonic-gate 				    GRAPH_UPDATE_STATE_CHANGE, ps);
469*0Sstevel@tonic-gate 
470*0Sstevel@tonic-gate 				do_commit_states = B_FALSE;
471*0Sstevel@tonic-gate 			}
472*0Sstevel@tonic-gate 		}
473*0Sstevel@tonic-gate 	}
474*0Sstevel@tonic-gate 
475*0Sstevel@tonic-gate 	switch (libscf_get_startd_properties(scf_inst, snap, &inst->ri_flags,
476*0Sstevel@tonic-gate 	    &inst->ri_utmpx_prefix)) {
477*0Sstevel@tonic-gate 	case 0:
478*0Sstevel@tonic-gate 		break;
479*0Sstevel@tonic-gate 
480*0Sstevel@tonic-gate 	case ECONNABORTED:
481*0Sstevel@tonic-gate 		libscf_handle_rebind(h);
482*0Sstevel@tonic-gate 		goto rep_retry;
483*0Sstevel@tonic-gate 
484*0Sstevel@tonic-gate 	case ECANCELED:
485*0Sstevel@tonic-gate 		scf_snapshot_destroy(snap);
486*0Sstevel@tonic-gate 		startd_free(inst->ri_utmpx_prefix, max_scf_value_size);
487*0Sstevel@tonic-gate 		goto deleted;
488*0Sstevel@tonic-gate 
489*0Sstevel@tonic-gate 	case ENOENT:
490*0Sstevel@tonic-gate 		/*
491*0Sstevel@tonic-gate 		 * This is odd, because the graph engine should have required
492*0Sstevel@tonic-gate 		 * the general property group.  So we'll just use default
493*0Sstevel@tonic-gate 		 * flags in anticipation of the graph engine sending us
494*0Sstevel@tonic-gate 		 * REMOVE_INSTANCE when it finds out that the general property
495*0Sstevel@tonic-gate 		 * group has been deleted.
496*0Sstevel@tonic-gate 		 */
497*0Sstevel@tonic-gate 		inst->ri_flags = RINST_CONTRACT;
498*0Sstevel@tonic-gate 		break;
499*0Sstevel@tonic-gate 
500*0Sstevel@tonic-gate 	default:
501*0Sstevel@tonic-gate 		assert(0);
502*0Sstevel@tonic-gate 		abort();
503*0Sstevel@tonic-gate 	}
504*0Sstevel@tonic-gate 
505*0Sstevel@tonic-gate 	switch (libscf_get_template_values(scf_inst, snap,
506*0Sstevel@tonic-gate 	    &inst->ri_common_name, &inst->ri_C_common_name)) {
507*0Sstevel@tonic-gate 	case 0:
508*0Sstevel@tonic-gate 		break;
509*0Sstevel@tonic-gate 
510*0Sstevel@tonic-gate 	case ECONNABORTED:
511*0Sstevel@tonic-gate 		libscf_handle_rebind(h);
512*0Sstevel@tonic-gate 		goto rep_retry;
513*0Sstevel@tonic-gate 
514*0Sstevel@tonic-gate 	case ECANCELED:
515*0Sstevel@tonic-gate 		scf_snapshot_destroy(snap);
516*0Sstevel@tonic-gate 		startd_free(inst->ri_common_name, max_scf_value_size);
517*0Sstevel@tonic-gate 		inst->ri_common_name = NULL;
518*0Sstevel@tonic-gate 		goto deleted;
519*0Sstevel@tonic-gate 
520*0Sstevel@tonic-gate 	case ECHILD:
521*0Sstevel@tonic-gate 	case ENOENT:
522*0Sstevel@tonic-gate 		break;
523*0Sstevel@tonic-gate 
524*0Sstevel@tonic-gate 	default:
525*0Sstevel@tonic-gate 		assert(0);
526*0Sstevel@tonic-gate 		abort();
527*0Sstevel@tonic-gate 	}
528*0Sstevel@tonic-gate 
529*0Sstevel@tonic-gate 	switch (libscf_read_method_ids(h, scf_inst, inst->ri_i.i_fmri,
530*0Sstevel@tonic-gate 	    &inst->ri_i.i_primary_ctid, &inst->ri_i.i_transient_ctid,
531*0Sstevel@tonic-gate 	    &start_pid)) {
532*0Sstevel@tonic-gate 	case 0:
533*0Sstevel@tonic-gate 		break;
534*0Sstevel@tonic-gate 
535*0Sstevel@tonic-gate 	case ECONNABORTED:
536*0Sstevel@tonic-gate 		libscf_handle_rebind(h);
537*0Sstevel@tonic-gate 		goto rep_retry;
538*0Sstevel@tonic-gate 
539*0Sstevel@tonic-gate 	case ECANCELED:
540*0Sstevel@tonic-gate 		scf_snapshot_destroy(snap);
541*0Sstevel@tonic-gate 		goto deleted;
542*0Sstevel@tonic-gate 
543*0Sstevel@tonic-gate 	default:
544*0Sstevel@tonic-gate 		assert(0);
545*0Sstevel@tonic-gate 		abort();
546*0Sstevel@tonic-gate 	}
547*0Sstevel@tonic-gate 
548*0Sstevel@tonic-gate 	if (inst->ri_i.i_primary_ctid >= 1) {
549*0Sstevel@tonic-gate 		contract_hash_store(inst->ri_i.i_primary_ctid, inst->ri_id);
550*0Sstevel@tonic-gate 
551*0Sstevel@tonic-gate 		switch (check_contract(inst, B_TRUE, scf_inst)) {
552*0Sstevel@tonic-gate 		case 0:
553*0Sstevel@tonic-gate 			break;
554*0Sstevel@tonic-gate 
555*0Sstevel@tonic-gate 		case ECONNABORTED:
556*0Sstevel@tonic-gate 			libscf_handle_rebind(h);
557*0Sstevel@tonic-gate 			goto rep_retry;
558*0Sstevel@tonic-gate 
559*0Sstevel@tonic-gate 		case ECANCELED:
560*0Sstevel@tonic-gate 			scf_snapshot_destroy(snap);
561*0Sstevel@tonic-gate 			goto deleted;
562*0Sstevel@tonic-gate 
563*0Sstevel@tonic-gate 		default:
564*0Sstevel@tonic-gate 			assert(0);
565*0Sstevel@tonic-gate 			abort();
566*0Sstevel@tonic-gate 		}
567*0Sstevel@tonic-gate 	}
568*0Sstevel@tonic-gate 
569*0Sstevel@tonic-gate 	if (inst->ri_i.i_transient_ctid >= 1) {
570*0Sstevel@tonic-gate 		switch (check_contract(inst, B_FALSE, scf_inst)) {
571*0Sstevel@tonic-gate 		case 0:
572*0Sstevel@tonic-gate 			break;
573*0Sstevel@tonic-gate 
574*0Sstevel@tonic-gate 		case ECONNABORTED:
575*0Sstevel@tonic-gate 			libscf_handle_rebind(h);
576*0Sstevel@tonic-gate 			goto rep_retry;
577*0Sstevel@tonic-gate 
578*0Sstevel@tonic-gate 		case ECANCELED:
579*0Sstevel@tonic-gate 			scf_snapshot_destroy(snap);
580*0Sstevel@tonic-gate 			goto deleted;
581*0Sstevel@tonic-gate 
582*0Sstevel@tonic-gate 		default:
583*0Sstevel@tonic-gate 			assert(0);
584*0Sstevel@tonic-gate 			abort();
585*0Sstevel@tonic-gate 		}
586*0Sstevel@tonic-gate 	}
587*0Sstevel@tonic-gate 
588*0Sstevel@tonic-gate 	/* No more failures we live through, so add it to the list. */
589*0Sstevel@tonic-gate 	(void) pthread_mutex_init(&inst->ri_lock, &mutex_attrs);
590*0Sstevel@tonic-gate 	(void) pthread_mutex_init(&inst->ri_queue_lock, &mutex_attrs);
591*0Sstevel@tonic-gate 	MUTEX_LOCK(&inst->ri_lock);
592*0Sstevel@tonic-gate 	MUTEX_LOCK(&inst->ri_queue_lock);
593*0Sstevel@tonic-gate 
594*0Sstevel@tonic-gate 	(void) pthread_cond_init(&inst->ri_method_cv, NULL);
595*0Sstevel@tonic-gate 
596*0Sstevel@tonic-gate 	uu_list_node_init(inst, &inst->ri_link, restarter_instance_pool);
597*0Sstevel@tonic-gate 	uu_list_insert(instance_list.ril_instance_list, inst, idx);
598*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&instance_list.ril_lock);
599*0Sstevel@tonic-gate 
600*0Sstevel@tonic-gate 	if (start_pid != -1 &&
601*0Sstevel@tonic-gate 	    (inst->ri_flags & RINST_STYLE_MASK) == RINST_WAIT) {
602*0Sstevel@tonic-gate 		int ret;
603*0Sstevel@tonic-gate 		ret = wait_register(start_pid, inst->ri_i.i_fmri, 0, 1);
604*0Sstevel@tonic-gate 		if (ret == -1) {
605*0Sstevel@tonic-gate 			/*
606*0Sstevel@tonic-gate 			 * Implication:  if we can't reregister the
607*0Sstevel@tonic-gate 			 * instance, we will start another one.  Two
608*0Sstevel@tonic-gate 			 * instances may or may not result in a resource
609*0Sstevel@tonic-gate 			 * conflict.
610*0Sstevel@tonic-gate 			 */
611*0Sstevel@tonic-gate 			log_error(LOG_WARNING,
612*0Sstevel@tonic-gate 			    "%s: couldn't reregister %ld for wait\n",
613*0Sstevel@tonic-gate 			    inst->ri_i.i_fmri, start_pid);
614*0Sstevel@tonic-gate 		} else if (ret == 1) {
615*0Sstevel@tonic-gate 			/*
616*0Sstevel@tonic-gate 			 * Leading PID has exited.
617*0Sstevel@tonic-gate 			 */
618*0Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_EXIT);
619*0Sstevel@tonic-gate 		}
620*0Sstevel@tonic-gate 	}
621*0Sstevel@tonic-gate 
622*0Sstevel@tonic-gate 
623*0Sstevel@tonic-gate 	scf_pg_destroy(pg);
624*0Sstevel@tonic-gate 
625*0Sstevel@tonic-gate 	if (do_commit_states)
626*0Sstevel@tonic-gate 		(void) restarter_instance_update_states(h, inst, state,
627*0Sstevel@tonic-gate 		    next_state, RERR_NONE, NULL);
628*0Sstevel@tonic-gate 
629*0Sstevel@tonic-gate 	(void) snprintf(logfilebuf, PATH_MAX, "%s:%s", svc_name, inst_name);
630*0Sstevel@tonic-gate 	for (c = logfilebuf; *c != '\0'; c++)
631*0Sstevel@tonic-gate 		if (*c == '/')
632*0Sstevel@tonic-gate 			*c = '-';
633*0Sstevel@tonic-gate 
634*0Sstevel@tonic-gate 	if ((inst->ri_logstem = uu_msprintf("%s%s", logfilebuf, LOG_SUFFIX)) ==
635*0Sstevel@tonic-gate 	    NULL)
636*0Sstevel@tonic-gate 		uu_die("Allocation failure\n");
637*0Sstevel@tonic-gate 
638*0Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s is a %s-style service\n", name,
639*0Sstevel@tonic-gate 	    service_style(inst->ri_flags));
640*0Sstevel@tonic-gate 
641*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&inst->ri_queue_lock);
642*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&inst->ri_lock);
643*0Sstevel@tonic-gate 
644*0Sstevel@tonic-gate 	startd_free(svc_name, max_scf_name_size);
645*0Sstevel@tonic-gate 	startd_free(inst_name, max_scf_name_size);
646*0Sstevel@tonic-gate 	scf_snapshot_destroy(snap);
647*0Sstevel@tonic-gate 	scf_instance_destroy(scf_inst);
648*0Sstevel@tonic-gate 	scf_service_destroy(scf_svc);
649*0Sstevel@tonic-gate 
650*0Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: inserted instance into restarter list\n",
651*0Sstevel@tonic-gate 	    name);
652*0Sstevel@tonic-gate 
653*0Sstevel@tonic-gate 	return (0);
654*0Sstevel@tonic-gate }
655*0Sstevel@tonic-gate 
656*0Sstevel@tonic-gate static void
657*0Sstevel@tonic-gate restarter_delete_inst(restarter_inst_t *ri)
658*0Sstevel@tonic-gate {
659*0Sstevel@tonic-gate 	int id;
660*0Sstevel@tonic-gate 	restarter_inst_t *rip;
661*0Sstevel@tonic-gate 	void *cookie = NULL;
662*0Sstevel@tonic-gate 	restarter_instance_qentry_t *e;
663*0Sstevel@tonic-gate 
664*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&ri->ri_lock));
665*0Sstevel@tonic-gate 
666*0Sstevel@tonic-gate 	/*
667*0Sstevel@tonic-gate 	 * Must drop the instance lock so we can pick up the instance_list
668*0Sstevel@tonic-gate 	 * lock & remove the instance.
669*0Sstevel@tonic-gate 	 */
670*0Sstevel@tonic-gate 	id = ri->ri_id;
671*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&ri->ri_lock);
672*0Sstevel@tonic-gate 
673*0Sstevel@tonic-gate 	MUTEX_LOCK(&instance_list.ril_lock);
674*0Sstevel@tonic-gate 
675*0Sstevel@tonic-gate 	rip = uu_list_find(instance_list.ril_instance_list, &id, NULL, NULL);
676*0Sstevel@tonic-gate 	if (rip == NULL) {
677*0Sstevel@tonic-gate 		MUTEX_UNLOCK(&instance_list.ril_lock);
678*0Sstevel@tonic-gate 		return;
679*0Sstevel@tonic-gate 	}
680*0Sstevel@tonic-gate 
681*0Sstevel@tonic-gate 	assert(ri == rip);
682*0Sstevel@tonic-gate 
683*0Sstevel@tonic-gate 	uu_list_remove(instance_list.ril_instance_list, ri);
684*0Sstevel@tonic-gate 
685*0Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: deleted instance from restarter list\n",
686*0Sstevel@tonic-gate 	    ri->ri_i.i_fmri);
687*0Sstevel@tonic-gate 
688*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&instance_list.ril_lock);
689*0Sstevel@tonic-gate 
690*0Sstevel@tonic-gate 	/*
691*0Sstevel@tonic-gate 	 * We can lock the instance without holding the instance_list lock
692*0Sstevel@tonic-gate 	 * since we removed the instance from the list.
693*0Sstevel@tonic-gate 	 */
694*0Sstevel@tonic-gate 	MUTEX_LOCK(&ri->ri_lock);
695*0Sstevel@tonic-gate 	MUTEX_LOCK(&ri->ri_queue_lock);
696*0Sstevel@tonic-gate 
697*0Sstevel@tonic-gate 	if (ri->ri_i.i_primary_ctid >= 1)
698*0Sstevel@tonic-gate 		contract_hash_remove(ri->ri_i.i_primary_ctid);
699*0Sstevel@tonic-gate 
700*0Sstevel@tonic-gate 	while (ri->ri_method_thread != 0 || ri->ri_method_waiters > 0)
701*0Sstevel@tonic-gate 		(void) pthread_cond_wait(&ri->ri_method_cv, &ri->ri_lock);
702*0Sstevel@tonic-gate 
703*0Sstevel@tonic-gate 	while ((e = uu_list_teardown(ri->ri_queue, &cookie)) != NULL)
704*0Sstevel@tonic-gate 		startd_free(e, sizeof (*e));
705*0Sstevel@tonic-gate 	uu_list_destroy(ri->ri_queue);
706*0Sstevel@tonic-gate 
707*0Sstevel@tonic-gate 	startd_free((void *)ri->ri_i.i_fmri, strlen(ri->ri_i.i_fmri) + 1);
708*0Sstevel@tonic-gate 	free(ri->ri_logstem);
709*0Sstevel@tonic-gate 	startd_free(ri->ri_utmpx_prefix, max_scf_value_size);
710*0Sstevel@tonic-gate 	(void) pthread_mutex_destroy(&ri->ri_lock);
711*0Sstevel@tonic-gate 	(void) pthread_mutex_destroy(&ri->ri_queue_lock);
712*0Sstevel@tonic-gate 	startd_free(ri, sizeof (restarter_inst_t));
713*0Sstevel@tonic-gate }
714*0Sstevel@tonic-gate 
715*0Sstevel@tonic-gate /*
716*0Sstevel@tonic-gate  * instance_is_wait_style()
717*0Sstevel@tonic-gate  *
718*0Sstevel@tonic-gate  *   Returns 1 if the given instance is a "wait-style" service instance.
719*0Sstevel@tonic-gate  */
720*0Sstevel@tonic-gate int
721*0Sstevel@tonic-gate instance_is_wait_style(restarter_inst_t *inst)
722*0Sstevel@tonic-gate {
723*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&inst->ri_lock));
724*0Sstevel@tonic-gate 	return ((inst->ri_flags & RINST_STYLE_MASK) == RINST_WAIT);
725*0Sstevel@tonic-gate }
726*0Sstevel@tonic-gate 
727*0Sstevel@tonic-gate /*
728*0Sstevel@tonic-gate  * instance_is_transient_style()
729*0Sstevel@tonic-gate  *
730*0Sstevel@tonic-gate  *   Returns 1 if the given instance is a transient service instance.
731*0Sstevel@tonic-gate  */
732*0Sstevel@tonic-gate int
733*0Sstevel@tonic-gate instance_is_transient_style(restarter_inst_t *inst)
734*0Sstevel@tonic-gate {
735*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&inst->ri_lock));
736*0Sstevel@tonic-gate 	return ((inst->ri_flags & RINST_STYLE_MASK) == RINST_TRANSIENT);
737*0Sstevel@tonic-gate }
738*0Sstevel@tonic-gate 
739*0Sstevel@tonic-gate /*
740*0Sstevel@tonic-gate  * instance_in_transition()
741*0Sstevel@tonic-gate  * Returns 1 if instance is in transition, 0 if not
742*0Sstevel@tonic-gate  */
743*0Sstevel@tonic-gate int
744*0Sstevel@tonic-gate instance_in_transition(restarter_inst_t *inst)
745*0Sstevel@tonic-gate {
746*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&inst->ri_lock));
747*0Sstevel@tonic-gate 	if (inst->ri_i.i_next_state == RESTARTER_STATE_NONE)
748*0Sstevel@tonic-gate 		return (0);
749*0Sstevel@tonic-gate 	return (1);
750*0Sstevel@tonic-gate }
751*0Sstevel@tonic-gate 
752*0Sstevel@tonic-gate /*
753*0Sstevel@tonic-gate  * Returns
754*0Sstevel@tonic-gate  *   0 - success
755*0Sstevel@tonic-gate  *   ECONNRESET - success, but h was rebound
756*0Sstevel@tonic-gate  */
757*0Sstevel@tonic-gate int
758*0Sstevel@tonic-gate restarter_instance_update_states(scf_handle_t *h, restarter_inst_t *ri,
759*0Sstevel@tonic-gate     restarter_instance_state_t new_state,
760*0Sstevel@tonic-gate     restarter_instance_state_t new_state_next, restarter_error_t err, char *aux)
761*0Sstevel@tonic-gate {
762*0Sstevel@tonic-gate 	protocol_states_t *states;
763*0Sstevel@tonic-gate 	int e;
764*0Sstevel@tonic-gate 	uint_t retry_count = 0, msecs = ALLOC_DELAY;
765*0Sstevel@tonic-gate 	boolean_t rebound = B_FALSE;
766*0Sstevel@tonic-gate 
767*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&ri->ri_lock));
768*0Sstevel@tonic-gate 
769*0Sstevel@tonic-gate retry:
770*0Sstevel@tonic-gate 	e = _restarter_commit_states(h, &ri->ri_i, new_state, new_state_next,
771*0Sstevel@tonic-gate 	    aux);
772*0Sstevel@tonic-gate 	switch (e) {
773*0Sstevel@tonic-gate 	case 0:
774*0Sstevel@tonic-gate 		break;
775*0Sstevel@tonic-gate 
776*0Sstevel@tonic-gate 	case ENOMEM:
777*0Sstevel@tonic-gate 		++retry_count;
778*0Sstevel@tonic-gate 		if (retry_count < ALLOC_RETRY) {
779*0Sstevel@tonic-gate 			(void) poll(NULL, 0, msecs);
780*0Sstevel@tonic-gate 			msecs *= ALLOC_DELAY_MULT;
781*0Sstevel@tonic-gate 			goto retry;
782*0Sstevel@tonic-gate 		}
783*0Sstevel@tonic-gate 
784*0Sstevel@tonic-gate 		/* Like startd_alloc(). */
785*0Sstevel@tonic-gate 		uu_die("Insufficient memory.\n");
786*0Sstevel@tonic-gate 		/* NOTREACHED */
787*0Sstevel@tonic-gate 
788*0Sstevel@tonic-gate 	case ECONNABORTED:
789*0Sstevel@tonic-gate 		libscf_handle_rebind(h);
790*0Sstevel@tonic-gate 		rebound = B_TRUE;
791*0Sstevel@tonic-gate 		goto retry;
792*0Sstevel@tonic-gate 
793*0Sstevel@tonic-gate 	case EPERM:
794*0Sstevel@tonic-gate 	case EACCES:
795*0Sstevel@tonic-gate 	case EROFS:
796*0Sstevel@tonic-gate 		log_error(LOG_NOTICE, "Could not commit state change for %s "
797*0Sstevel@tonic-gate 		    "to repository: %s.\n", ri->ri_i.i_fmri, strerror(e));
798*0Sstevel@tonic-gate 		/* FALLTHROUGH */
799*0Sstevel@tonic-gate 
800*0Sstevel@tonic-gate 	case ENOENT:
801*0Sstevel@tonic-gate 		ri->ri_i.i_state = new_state;
802*0Sstevel@tonic-gate 		ri->ri_i.i_next_state = new_state_next;
803*0Sstevel@tonic-gate 		break;
804*0Sstevel@tonic-gate 
805*0Sstevel@tonic-gate 	case EINVAL:
806*0Sstevel@tonic-gate 	default:
807*0Sstevel@tonic-gate 		bad_error("_restarter_commit_states", e);
808*0Sstevel@tonic-gate 	}
809*0Sstevel@tonic-gate 
810*0Sstevel@tonic-gate 	states = startd_alloc(sizeof (protocol_states_t));
811*0Sstevel@tonic-gate 	states->ps_state = new_state;
812*0Sstevel@tonic-gate 	states->ps_state_next = new_state_next;
813*0Sstevel@tonic-gate 	states->ps_err = err;
814*0Sstevel@tonic-gate 	graph_protocol_send_event(ri->ri_i.i_fmri, GRAPH_UPDATE_STATE_CHANGE,
815*0Sstevel@tonic-gate 	    (void *)states);
816*0Sstevel@tonic-gate 
817*0Sstevel@tonic-gate 	if (new_state == RESTARTER_STATE_ONLINE)
818*0Sstevel@tonic-gate 		ri->ri_post_online_hook();
819*0Sstevel@tonic-gate 
820*0Sstevel@tonic-gate 	return (rebound ? ECONNRESET : 0);
821*0Sstevel@tonic-gate }
822*0Sstevel@tonic-gate 
823*0Sstevel@tonic-gate void
824*0Sstevel@tonic-gate restarter_mark_pending_snapshot(const char *fmri, uint_t flag)
825*0Sstevel@tonic-gate {
826*0Sstevel@tonic-gate 	restarter_inst_t *inst;
827*0Sstevel@tonic-gate 
828*0Sstevel@tonic-gate 	assert(flag == RINST_RETAKE_RUNNING || flag == RINST_RETAKE_START);
829*0Sstevel@tonic-gate 
830*0Sstevel@tonic-gate 	inst = inst_lookup_by_name(fmri);
831*0Sstevel@tonic-gate 	if (inst == NULL)
832*0Sstevel@tonic-gate 		return;
833*0Sstevel@tonic-gate 
834*0Sstevel@tonic-gate 	inst->ri_flags |= flag;
835*0Sstevel@tonic-gate 
836*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&inst->ri_lock);
837*0Sstevel@tonic-gate }
838*0Sstevel@tonic-gate 
839*0Sstevel@tonic-gate static void
840*0Sstevel@tonic-gate restarter_take_pending_snapshots(scf_handle_t *h)
841*0Sstevel@tonic-gate {
842*0Sstevel@tonic-gate 	restarter_inst_t *inst;
843*0Sstevel@tonic-gate 	int r;
844*0Sstevel@tonic-gate 
845*0Sstevel@tonic-gate 	MUTEX_LOCK(&instance_list.ril_lock);
846*0Sstevel@tonic-gate 
847*0Sstevel@tonic-gate 	for (inst = uu_list_first(instance_list.ril_instance_list);
848*0Sstevel@tonic-gate 	    inst != NULL;
849*0Sstevel@tonic-gate 	    inst = uu_list_next(instance_list.ril_instance_list, inst)) {
850*0Sstevel@tonic-gate 		const char *fmri;
851*0Sstevel@tonic-gate 		scf_instance_t *sinst = NULL;
852*0Sstevel@tonic-gate 
853*0Sstevel@tonic-gate 		MUTEX_LOCK(&inst->ri_lock);
854*0Sstevel@tonic-gate 
855*0Sstevel@tonic-gate 		/*
856*0Sstevel@tonic-gate 		 * This is where we'd check inst->ri_method_thread and if it
857*0Sstevel@tonic-gate 		 * were nonzero we'd wait in anticipation of another thread
858*0Sstevel@tonic-gate 		 * executing a method for inst.  Doing so with the instance_list
859*0Sstevel@tonic-gate 		 * locked, though, leads to deadlock.  Since taking a snapshot
860*0Sstevel@tonic-gate 		 * during that window won't hurt anything, we'll just continue.
861*0Sstevel@tonic-gate 		 */
862*0Sstevel@tonic-gate 
863*0Sstevel@tonic-gate 		fmri = inst->ri_i.i_fmri;
864*0Sstevel@tonic-gate 
865*0Sstevel@tonic-gate 		if (inst->ri_flags & RINST_RETAKE_RUNNING) {
866*0Sstevel@tonic-gate 			scf_snapshot_t *rsnap;
867*0Sstevel@tonic-gate 
868*0Sstevel@tonic-gate 			(void) libscf_fmri_get_instance(h, fmri, &sinst);
869*0Sstevel@tonic-gate 
870*0Sstevel@tonic-gate 			rsnap = libscf_get_or_make_running_snapshot(sinst,
871*0Sstevel@tonic-gate 			    fmri, B_FALSE);
872*0Sstevel@tonic-gate 
873*0Sstevel@tonic-gate 			scf_instance_destroy(sinst);
874*0Sstevel@tonic-gate 
875*0Sstevel@tonic-gate 			if (rsnap != NULL)
876*0Sstevel@tonic-gate 				inst->ri_flags &= ~RINST_RETAKE_RUNNING;
877*0Sstevel@tonic-gate 
878*0Sstevel@tonic-gate 			scf_snapshot_destroy(rsnap);
879*0Sstevel@tonic-gate 		}
880*0Sstevel@tonic-gate 
881*0Sstevel@tonic-gate 		if (inst->ri_flags & RINST_RETAKE_START) {
882*0Sstevel@tonic-gate 			switch (r = libscf_snapshots_poststart(h, fmri,
883*0Sstevel@tonic-gate 			    B_FALSE)) {
884*0Sstevel@tonic-gate 			case 0:
885*0Sstevel@tonic-gate 			case ENOENT:
886*0Sstevel@tonic-gate 				inst->ri_flags &= ~RINST_RETAKE_START;
887*0Sstevel@tonic-gate 				break;
888*0Sstevel@tonic-gate 
889*0Sstevel@tonic-gate 			case ECONNABORTED:
890*0Sstevel@tonic-gate 				break;
891*0Sstevel@tonic-gate 
892*0Sstevel@tonic-gate 			case EACCES:
893*0Sstevel@tonic-gate 			default:
894*0Sstevel@tonic-gate 				bad_error("libscf_snapshots_poststart", r);
895*0Sstevel@tonic-gate 			}
896*0Sstevel@tonic-gate 		}
897*0Sstevel@tonic-gate 
898*0Sstevel@tonic-gate 		MUTEX_UNLOCK(&inst->ri_lock);
899*0Sstevel@tonic-gate 	}
900*0Sstevel@tonic-gate 
901*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&instance_list.ril_lock);
902*0Sstevel@tonic-gate }
903*0Sstevel@tonic-gate 
904*0Sstevel@tonic-gate /* ARGSUSED */
905*0Sstevel@tonic-gate void *
906*0Sstevel@tonic-gate restarter_post_fsminimal_thread(void *unused)
907*0Sstevel@tonic-gate {
908*0Sstevel@tonic-gate 	scf_handle_t *h;
909*0Sstevel@tonic-gate 	int r;
910*0Sstevel@tonic-gate 
911*0Sstevel@tonic-gate 	h = libscf_handle_create_bound_loop();
912*0Sstevel@tonic-gate 
913*0Sstevel@tonic-gate 	for (;;) {
914*0Sstevel@tonic-gate 		r = libscf_create_self(h);
915*0Sstevel@tonic-gate 		if (r == 0)
916*0Sstevel@tonic-gate 			break;
917*0Sstevel@tonic-gate 
918*0Sstevel@tonic-gate 		assert(r == ECONNABORTED);
919*0Sstevel@tonic-gate 		libscf_handle_rebind(h);
920*0Sstevel@tonic-gate 	}
921*0Sstevel@tonic-gate 
922*0Sstevel@tonic-gate 	restarter_take_pending_snapshots(h);
923*0Sstevel@tonic-gate 
924*0Sstevel@tonic-gate 	(void) scf_handle_unbind(h);
925*0Sstevel@tonic-gate 	scf_handle_destroy(h);
926*0Sstevel@tonic-gate 
927*0Sstevel@tonic-gate 	return (NULL);
928*0Sstevel@tonic-gate }
929*0Sstevel@tonic-gate 
930*0Sstevel@tonic-gate /*
931*0Sstevel@tonic-gate  * returns 1 if instance is already started, 0 if not
932*0Sstevel@tonic-gate  */
933*0Sstevel@tonic-gate static int
934*0Sstevel@tonic-gate instance_started(restarter_inst_t *inst)
935*0Sstevel@tonic-gate {
936*0Sstevel@tonic-gate 	int ret;
937*0Sstevel@tonic-gate 
938*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&inst->ri_lock));
939*0Sstevel@tonic-gate 
940*0Sstevel@tonic-gate 	if (inst->ri_i.i_state == RESTARTER_STATE_ONLINE ||
941*0Sstevel@tonic-gate 	    inst->ri_i.i_state == RESTARTER_STATE_DEGRADED)
942*0Sstevel@tonic-gate 		ret = 1;
943*0Sstevel@tonic-gate 	else
944*0Sstevel@tonic-gate 		ret = 0;
945*0Sstevel@tonic-gate 
946*0Sstevel@tonic-gate 	return (ret);
947*0Sstevel@tonic-gate }
948*0Sstevel@tonic-gate 
949*0Sstevel@tonic-gate /*
950*0Sstevel@tonic-gate  * int stop_instance()
951*0Sstevel@tonic-gate  *
952*0Sstevel@tonic-gate  *   Stop the instance identified by the instance given as the second argument,
953*0Sstevel@tonic-gate  *   for the cause stated.
954*0Sstevel@tonic-gate  *
955*0Sstevel@tonic-gate  *   Returns
956*0Sstevel@tonic-gate  *     0 - success
957*0Sstevel@tonic-gate  *     -1 - inst is in transition
958*0Sstevel@tonic-gate  */
959*0Sstevel@tonic-gate static int
960*0Sstevel@tonic-gate stop_instance(scf_handle_t *local_handle, restarter_inst_t *inst,
961*0Sstevel@tonic-gate     stop_cause_t cause)
962*0Sstevel@tonic-gate {
963*0Sstevel@tonic-gate 	fork_info_t *info;
964*0Sstevel@tonic-gate 	const char *cp;
965*0Sstevel@tonic-gate 	int err;
966*0Sstevel@tonic-gate 	restarter_error_t re;
967*0Sstevel@tonic-gate 
968*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&inst->ri_lock));
969*0Sstevel@tonic-gate 	assert(inst->ri_method_thread == 0);
970*0Sstevel@tonic-gate 
971*0Sstevel@tonic-gate 	switch (cause) {
972*0Sstevel@tonic-gate 	case RSTOP_EXIT:
973*0Sstevel@tonic-gate 		re = RERR_RESTART;
974*0Sstevel@tonic-gate 		cp = "all processes in service exited";
975*0Sstevel@tonic-gate 		break;
976*0Sstevel@tonic-gate 	case RSTOP_CORE:
977*0Sstevel@tonic-gate 		re = RERR_FAULT;
978*0Sstevel@tonic-gate 		cp = "process dumped core";
979*0Sstevel@tonic-gate 		break;
980*0Sstevel@tonic-gate 	case RSTOP_SIGNAL:
981*0Sstevel@tonic-gate 		re = RERR_FAULT;
982*0Sstevel@tonic-gate 		cp = "process received fatal signal from outside the service";
983*0Sstevel@tonic-gate 		break;
984*0Sstevel@tonic-gate 	case RSTOP_HWERR:
985*0Sstevel@tonic-gate 		re = RERR_FAULT;
986*0Sstevel@tonic-gate 		cp = "process killed due to uncorrectable hardware error";
987*0Sstevel@tonic-gate 		break;
988*0Sstevel@tonic-gate 	case RSTOP_DEPENDENCY:
989*0Sstevel@tonic-gate 		re = RERR_RESTART;
990*0Sstevel@tonic-gate 		cp = "dependency activity requires stop";
991*0Sstevel@tonic-gate 		break;
992*0Sstevel@tonic-gate 	case RSTOP_DISABLE:
993*0Sstevel@tonic-gate 		re = RERR_RESTART;
994*0Sstevel@tonic-gate 		cp = "service disabled";
995*0Sstevel@tonic-gate 		break;
996*0Sstevel@tonic-gate 	case RSTOP_RESTART:
997*0Sstevel@tonic-gate 		re = RERR_RESTART;
998*0Sstevel@tonic-gate 		cp = "service restarting";
999*0Sstevel@tonic-gate 		break;
1000*0Sstevel@tonic-gate 	default:
1001*0Sstevel@tonic-gate #ifndef NDEBUG
1002*0Sstevel@tonic-gate 		(void) fprintf(stderr, "Unknown cause %d at %s:%d.\n",
1003*0Sstevel@tonic-gate 		    cause, __FILE__, __LINE__);
1004*0Sstevel@tonic-gate #endif
1005*0Sstevel@tonic-gate 		abort();
1006*0Sstevel@tonic-gate 	}
1007*0Sstevel@tonic-gate 
1008*0Sstevel@tonic-gate 	/* Services in the disabled and maintenance state are ignored */
1009*0Sstevel@tonic-gate 	if (inst->ri_i.i_state == RESTARTER_STATE_MAINT ||
1010*0Sstevel@tonic-gate 	    inst->ri_i.i_state == RESTARTER_STATE_DISABLED) {
1011*0Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
1012*0Sstevel@tonic-gate 		    "%s: stop_instance -> is maint/disabled\n",
1013*0Sstevel@tonic-gate 		    inst->ri_i.i_fmri);
1014*0Sstevel@tonic-gate 		return (0);
1015*0Sstevel@tonic-gate 	}
1016*0Sstevel@tonic-gate 
1017*0Sstevel@tonic-gate 	/* Already stopped instances are left alone */
1018*0Sstevel@tonic-gate 	if (instance_started(inst) == 0) {
1019*0Sstevel@tonic-gate 		log_framework(LOG_DEBUG, "Restarter: %s is already stopped.\n",
1020*0Sstevel@tonic-gate 		    inst->ri_i.i_fmri);
1021*0Sstevel@tonic-gate 		return (0);
1022*0Sstevel@tonic-gate 	}
1023*0Sstevel@tonic-gate 
1024*0Sstevel@tonic-gate 	if (instance_in_transition(inst)) {
1025*0Sstevel@tonic-gate 		/* requeue event by returning -1 */
1026*0Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
1027*0Sstevel@tonic-gate 		    "Restarter: Not stopping %s, in transition.\n",
1028*0Sstevel@tonic-gate 		    inst->ri_i.i_fmri);
1029*0Sstevel@tonic-gate 		return (-1);
1030*0Sstevel@tonic-gate 	}
1031*0Sstevel@tonic-gate 
1032*0Sstevel@tonic-gate 	log_instance(inst, B_TRUE, "Stopping because %s.", cp);
1033*0Sstevel@tonic-gate 
1034*0Sstevel@tonic-gate 	log_framework(re == RERR_FAULT ? LOG_INFO : LOG_DEBUG,
1035*0Sstevel@tonic-gate 	    "%s: Instance stopping because %s.\n", inst->ri_i.i_fmri, cp);
1036*0Sstevel@tonic-gate 
1037*0Sstevel@tonic-gate 	if (instance_is_wait_style(inst) && cause == RSTOP_EXIT) {
1038*0Sstevel@tonic-gate 		/*
1039*0Sstevel@tonic-gate 		 * No need to stop instance, as child has exited; remove
1040*0Sstevel@tonic-gate 		 * contract and move the instance to the offline state.
1041*0Sstevel@tonic-gate 		 */
1042*0Sstevel@tonic-gate 		switch (err = restarter_instance_update_states(local_handle,
1043*0Sstevel@tonic-gate 		    inst, inst->ri_i.i_state, RESTARTER_STATE_OFFLINE, re,
1044*0Sstevel@tonic-gate 		    NULL)) {
1045*0Sstevel@tonic-gate 		case 0:
1046*0Sstevel@tonic-gate 		case ECONNRESET:
1047*0Sstevel@tonic-gate 			break;
1048*0Sstevel@tonic-gate 
1049*0Sstevel@tonic-gate 		default:
1050*0Sstevel@tonic-gate 			bad_error("restarter_instance_update_states", err);
1051*0Sstevel@tonic-gate 		}
1052*0Sstevel@tonic-gate 
1053*0Sstevel@tonic-gate 		(void) update_fault_count(inst, FAULT_COUNT_RESET);
1054*0Sstevel@tonic-gate 
1055*0Sstevel@tonic-gate 		if (inst->ri_i.i_primary_ctid != 0) {
1056*0Sstevel@tonic-gate 			inst->ri_m_inst =
1057*0Sstevel@tonic-gate 			    safe_scf_instance_create(local_handle);
1058*0Sstevel@tonic-gate 			inst->ri_mi_deleted = B_FALSE;
1059*0Sstevel@tonic-gate 
1060*0Sstevel@tonic-gate 			libscf_reget_instance(inst);
1061*0Sstevel@tonic-gate 			method_remove_contract(inst, B_TRUE, B_TRUE);
1062*0Sstevel@tonic-gate 
1063*0Sstevel@tonic-gate 			scf_instance_destroy(inst->ri_m_inst);
1064*0Sstevel@tonic-gate 			inst->ri_m_inst = NULL;
1065*0Sstevel@tonic-gate 		}
1066*0Sstevel@tonic-gate 
1067*0Sstevel@tonic-gate 		switch (err = restarter_instance_update_states(local_handle,
1068*0Sstevel@tonic-gate 		    inst, inst->ri_i.i_next_state, RESTARTER_STATE_NONE, re,
1069*0Sstevel@tonic-gate 		    NULL)) {
1070*0Sstevel@tonic-gate 		case 0:
1071*0Sstevel@tonic-gate 		case ECONNRESET:
1072*0Sstevel@tonic-gate 			break;
1073*0Sstevel@tonic-gate 
1074*0Sstevel@tonic-gate 		default:
1075*0Sstevel@tonic-gate 			bad_error("restarter_instance_update_states", err);
1076*0Sstevel@tonic-gate 		}
1077*0Sstevel@tonic-gate 
1078*0Sstevel@tonic-gate 		return (0);
1079*0Sstevel@tonic-gate 	}
1080*0Sstevel@tonic-gate 
1081*0Sstevel@tonic-gate 	switch (err = restarter_instance_update_states(local_handle, inst,
1082*0Sstevel@tonic-gate 	    inst->ri_i.i_state, inst->ri_i.i_enabled ? RESTARTER_STATE_OFFLINE :
1083*0Sstevel@tonic-gate 	    RESTARTER_STATE_DISABLED, RERR_NONE, NULL)) {
1084*0Sstevel@tonic-gate 	case 0:
1085*0Sstevel@tonic-gate 	case ECONNRESET:
1086*0Sstevel@tonic-gate 		break;
1087*0Sstevel@tonic-gate 
1088*0Sstevel@tonic-gate 	default:
1089*0Sstevel@tonic-gate 		bad_error("restarter_instance_update_states", err);
1090*0Sstevel@tonic-gate 	}
1091*0Sstevel@tonic-gate 
1092*0Sstevel@tonic-gate 	info = startd_zalloc(sizeof (fork_info_t));
1093*0Sstevel@tonic-gate 
1094*0Sstevel@tonic-gate 	info->sf_id = inst->ri_id;
1095*0Sstevel@tonic-gate 	info->sf_method_type = METHOD_STOP;
1096*0Sstevel@tonic-gate 	info->sf_event_type = re;
1097*0Sstevel@tonic-gate 	inst->ri_method_thread = startd_thread_create(method_thread, info);
1098*0Sstevel@tonic-gate 
1099*0Sstevel@tonic-gate 	return (0);
1100*0Sstevel@tonic-gate }
1101*0Sstevel@tonic-gate 
1102*0Sstevel@tonic-gate /*
1103*0Sstevel@tonic-gate  * Returns
1104*0Sstevel@tonic-gate  *   ENOENT - fmri is not in instance_list
1105*0Sstevel@tonic-gate  *   0 - success
1106*0Sstevel@tonic-gate  *   ECONNRESET - success, though handle was rebound
1107*0Sstevel@tonic-gate  *   -1 - instance is in transition
1108*0Sstevel@tonic-gate  */
1109*0Sstevel@tonic-gate int
1110*0Sstevel@tonic-gate stop_instance_fmri(scf_handle_t *h, const char *fmri, uint_t flags)
1111*0Sstevel@tonic-gate {
1112*0Sstevel@tonic-gate 	restarter_inst_t *rip;
1113*0Sstevel@tonic-gate 	int r;
1114*0Sstevel@tonic-gate 
1115*0Sstevel@tonic-gate 	rip = inst_lookup_by_name(fmri);
1116*0Sstevel@tonic-gate 	if (rip == NULL)
1117*0Sstevel@tonic-gate 		return (ENOENT);
1118*0Sstevel@tonic-gate 
1119*0Sstevel@tonic-gate 	r = stop_instance(h, rip, flags);
1120*0Sstevel@tonic-gate 
1121*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&rip->ri_lock);
1122*0Sstevel@tonic-gate 
1123*0Sstevel@tonic-gate 	return (r);
1124*0Sstevel@tonic-gate }
1125*0Sstevel@tonic-gate 
1126*0Sstevel@tonic-gate static void
1127*0Sstevel@tonic-gate unmaintain_instance(scf_handle_t *h, restarter_inst_t *rip,
1128*0Sstevel@tonic-gate     unmaint_cause_t cause)
1129*0Sstevel@tonic-gate {
1130*0Sstevel@tonic-gate 	ctid_t ctid;
1131*0Sstevel@tonic-gate 	scf_instance_t *inst;
1132*0Sstevel@tonic-gate 	int r;
1133*0Sstevel@tonic-gate 	uint_t tries = 0, msecs = ALLOC_DELAY;
1134*0Sstevel@tonic-gate 	const char *cp;
1135*0Sstevel@tonic-gate 
1136*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&rip->ri_lock));
1137*0Sstevel@tonic-gate 
1138*0Sstevel@tonic-gate 	if (rip->ri_i.i_state != RESTARTER_STATE_MAINT) {
1139*0Sstevel@tonic-gate 		log_error(LOG_DEBUG, "Restarter: "
1140*0Sstevel@tonic-gate 		    "Ignoring maintenance off command because %s is not in the "
1141*0Sstevel@tonic-gate 		    "maintenance state.\n", rip->ri_i.i_fmri);
1142*0Sstevel@tonic-gate 		return;
1143*0Sstevel@tonic-gate 	}
1144*0Sstevel@tonic-gate 
1145*0Sstevel@tonic-gate 	switch (cause) {
1146*0Sstevel@tonic-gate 	case RUNMAINT_CLEAR:
1147*0Sstevel@tonic-gate 		cp = "clear requested";
1148*0Sstevel@tonic-gate 		break;
1149*0Sstevel@tonic-gate 	case RUNMAINT_DISABLE:
1150*0Sstevel@tonic-gate 		cp = "disable requested";
1151*0Sstevel@tonic-gate 		break;
1152*0Sstevel@tonic-gate 	default:
1153*0Sstevel@tonic-gate #ifndef NDEBUG
1154*0Sstevel@tonic-gate 		(void) fprintf(stderr, "Uncaught case for %d at %s:%d.\n",
1155*0Sstevel@tonic-gate 		    cause, __FILE__, __LINE__);
1156*0Sstevel@tonic-gate #endif
1157*0Sstevel@tonic-gate 		abort();
1158*0Sstevel@tonic-gate 	}
1159*0Sstevel@tonic-gate 
1160*0Sstevel@tonic-gate 	log_instance(rip, B_TRUE, "Leaving maintenance because %s.",
1161*0Sstevel@tonic-gate 	    cp);
1162*0Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: Instance leaving maintenance because "
1163*0Sstevel@tonic-gate 	    "%s.\n", rip->ri_i.i_fmri, cp);
1164*0Sstevel@tonic-gate 
1165*0Sstevel@tonic-gate 	(void) restarter_instance_update_states(h, rip, RESTARTER_STATE_UNINIT,
1166*0Sstevel@tonic-gate 	    RESTARTER_STATE_NONE, RERR_RESTART, NULL);
1167*0Sstevel@tonic-gate 
1168*0Sstevel@tonic-gate 	/*
1169*0Sstevel@tonic-gate 	 * If we did ADMIN_MAINT_ON_IMMEDIATE, then there might still be
1170*0Sstevel@tonic-gate 	 * a primary contract.
1171*0Sstevel@tonic-gate 	 */
1172*0Sstevel@tonic-gate 	if (rip->ri_i.i_primary_ctid == 0)
1173*0Sstevel@tonic-gate 		return;
1174*0Sstevel@tonic-gate 
1175*0Sstevel@tonic-gate 	ctid = rip->ri_i.i_primary_ctid;
1176*0Sstevel@tonic-gate 	contract_abandon(ctid);
1177*0Sstevel@tonic-gate 	rip->ri_i.i_primary_ctid = 0;
1178*0Sstevel@tonic-gate 
1179*0Sstevel@tonic-gate rep_retry:
1180*0Sstevel@tonic-gate 	switch (r = libscf_fmri_get_instance(h, rip->ri_i.i_fmri, &inst)) {
1181*0Sstevel@tonic-gate 	case 0:
1182*0Sstevel@tonic-gate 		break;
1183*0Sstevel@tonic-gate 
1184*0Sstevel@tonic-gate 	case ECONNABORTED:
1185*0Sstevel@tonic-gate 		libscf_handle_rebind(h);
1186*0Sstevel@tonic-gate 		goto rep_retry;
1187*0Sstevel@tonic-gate 
1188*0Sstevel@tonic-gate 	case ENOENT:
1189*0Sstevel@tonic-gate 		/* Must have been deleted. */
1190*0Sstevel@tonic-gate 		return;
1191*0Sstevel@tonic-gate 
1192*0Sstevel@tonic-gate 	case EINVAL:
1193*0Sstevel@tonic-gate 	case ENOTSUP:
1194*0Sstevel@tonic-gate 	default:
1195*0Sstevel@tonic-gate 		bad_error("libscf_handle_rebind", r);
1196*0Sstevel@tonic-gate 	}
1197*0Sstevel@tonic-gate 
1198*0Sstevel@tonic-gate again:
1199*0Sstevel@tonic-gate 	r = restarter_remove_contract(inst, ctid, RESTARTER_CONTRACT_PRIMARY);
1200*0Sstevel@tonic-gate 	switch (r) {
1201*0Sstevel@tonic-gate 	case 0:
1202*0Sstevel@tonic-gate 		break;
1203*0Sstevel@tonic-gate 
1204*0Sstevel@tonic-gate 	case ENOMEM:
1205*0Sstevel@tonic-gate 		++tries;
1206*0Sstevel@tonic-gate 		if (tries < ALLOC_RETRY) {
1207*0Sstevel@tonic-gate 			(void) poll(NULL, 0, msecs);
1208*0Sstevel@tonic-gate 			msecs *= ALLOC_DELAY_MULT;
1209*0Sstevel@tonic-gate 			goto again;
1210*0Sstevel@tonic-gate 		}
1211*0Sstevel@tonic-gate 
1212*0Sstevel@tonic-gate 		uu_die("Insufficient memory.\n");
1213*0Sstevel@tonic-gate 		/* NOTREACHED */
1214*0Sstevel@tonic-gate 
1215*0Sstevel@tonic-gate 	case ECONNABORTED:
1216*0Sstevel@tonic-gate 		scf_instance_destroy(inst);
1217*0Sstevel@tonic-gate 		libscf_handle_rebind(h);
1218*0Sstevel@tonic-gate 		goto rep_retry;
1219*0Sstevel@tonic-gate 
1220*0Sstevel@tonic-gate 	case ECANCELED:
1221*0Sstevel@tonic-gate 		break;
1222*0Sstevel@tonic-gate 
1223*0Sstevel@tonic-gate 	case EPERM:
1224*0Sstevel@tonic-gate 	case EACCES:
1225*0Sstevel@tonic-gate 	case EROFS:
1226*0Sstevel@tonic-gate 		log_error(LOG_INFO,
1227*0Sstevel@tonic-gate 		    "Could not remove contract id %lu for %s (%s).\n", ctid,
1228*0Sstevel@tonic-gate 		    rip->ri_i.i_fmri, strerror(r));
1229*0Sstevel@tonic-gate 		break;
1230*0Sstevel@tonic-gate 
1231*0Sstevel@tonic-gate 	case EINVAL:
1232*0Sstevel@tonic-gate 	case EBADF:
1233*0Sstevel@tonic-gate 	default:
1234*0Sstevel@tonic-gate 		bad_error("restarter_remove_contract", r);
1235*0Sstevel@tonic-gate 	}
1236*0Sstevel@tonic-gate 
1237*0Sstevel@tonic-gate 	scf_instance_destroy(inst);
1238*0Sstevel@tonic-gate }
1239*0Sstevel@tonic-gate 
1240*0Sstevel@tonic-gate /*
1241*0Sstevel@tonic-gate  * enable_inst()
1242*0Sstevel@tonic-gate  *   Set inst->ri_i.i_enabled.  Expects 'e' to be _ENABLE, _DISABLE, or
1243*0Sstevel@tonic-gate  *   _ADMIN_DISABLE.  If the event is _ENABLE and inst is uninitialized or
1244*0Sstevel@tonic-gate  *   disabled, move it to offline.  If the event is _DISABLE or
1245*0Sstevel@tonic-gate  *   _ADMIN_DISABLE, make sure inst will move to disabled.
1246*0Sstevel@tonic-gate  *
1247*0Sstevel@tonic-gate  *   Returns
1248*0Sstevel@tonic-gate  *     0 - success
1249*0Sstevel@tonic-gate  *     ECONNRESET - h was rebound
1250*0Sstevel@tonic-gate  */
1251*0Sstevel@tonic-gate static int
1252*0Sstevel@tonic-gate enable_inst(scf_handle_t *h, restarter_inst_t *inst, restarter_event_type_t e)
1253*0Sstevel@tonic-gate {
1254*0Sstevel@tonic-gate 	restarter_instance_state_t state;
1255*0Sstevel@tonic-gate 	int r;
1256*0Sstevel@tonic-gate 
1257*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&inst->ri_lock));
1258*0Sstevel@tonic-gate 	assert(e == RESTARTER_EVENT_TYPE_ADMIN_DISABLE ||
1259*0Sstevel@tonic-gate 	    e == RESTARTER_EVENT_TYPE_DISABLE ||
1260*0Sstevel@tonic-gate 	    e == RESTARTER_EVENT_TYPE_ENABLE);
1261*0Sstevel@tonic-gate 	assert(instance_in_transition(inst) == 0);
1262*0Sstevel@tonic-gate 
1263*0Sstevel@tonic-gate 	state = inst->ri_i.i_state;
1264*0Sstevel@tonic-gate 
1265*0Sstevel@tonic-gate 	if (e == RESTARTER_EVENT_TYPE_ENABLE) {
1266*0Sstevel@tonic-gate 		inst->ri_i.i_enabled = 1;
1267*0Sstevel@tonic-gate 
1268*0Sstevel@tonic-gate 		if (state == RESTARTER_STATE_UNINIT ||
1269*0Sstevel@tonic-gate 		    state == RESTARTER_STATE_DISABLED) {
1270*0Sstevel@tonic-gate 			/*
1271*0Sstevel@tonic-gate 			 * B_FALSE: Don't log an error if the log_instance()
1272*0Sstevel@tonic-gate 			 * fails because it will fail on the miniroot before
1273*0Sstevel@tonic-gate 			 * install-discovery runs.
1274*0Sstevel@tonic-gate 			 */
1275*0Sstevel@tonic-gate 			log_instance(inst, B_FALSE, "Enabled.");
1276*0Sstevel@tonic-gate 			log_framework(LOG_DEBUG, "%s: Instance enabled.\n",
1277*0Sstevel@tonic-gate 			    inst->ri_i.i_fmri);
1278*0Sstevel@tonic-gate 			(void) restarter_instance_update_states(h, inst,
1279*0Sstevel@tonic-gate 			    RESTARTER_STATE_OFFLINE, RESTARTER_STATE_NONE,
1280*0Sstevel@tonic-gate 			    RERR_NONE, NULL);
1281*0Sstevel@tonic-gate 		} else {
1282*0Sstevel@tonic-gate 			log_framework(LOG_DEBUG, "Restarter: "
1283*0Sstevel@tonic-gate 			    "Not changing state of %s for enable command.\n",
1284*0Sstevel@tonic-gate 			    inst->ri_i.i_fmri);
1285*0Sstevel@tonic-gate 		}
1286*0Sstevel@tonic-gate 	} else {
1287*0Sstevel@tonic-gate 		inst->ri_i.i_enabled = 0;
1288*0Sstevel@tonic-gate 
1289*0Sstevel@tonic-gate 		switch (state) {
1290*0Sstevel@tonic-gate 		case RESTARTER_STATE_ONLINE:
1291*0Sstevel@tonic-gate 		case RESTARTER_STATE_DEGRADED:
1292*0Sstevel@tonic-gate 			r = stop_instance(h, inst, RSTOP_DISABLE);
1293*0Sstevel@tonic-gate 			return (r == ECONNRESET ? 0 : r);
1294*0Sstevel@tonic-gate 
1295*0Sstevel@tonic-gate 		case RESTARTER_STATE_OFFLINE:
1296*0Sstevel@tonic-gate 		case RESTARTER_STATE_UNINIT:
1297*0Sstevel@tonic-gate 			if (inst->ri_i.i_primary_ctid != 0) {
1298*0Sstevel@tonic-gate 				inst->ri_m_inst = safe_scf_instance_create(h);
1299*0Sstevel@tonic-gate 				inst->ri_mi_deleted = B_FALSE;
1300*0Sstevel@tonic-gate 
1301*0Sstevel@tonic-gate 				libscf_reget_instance(inst);
1302*0Sstevel@tonic-gate 				method_remove_contract(inst, B_TRUE, B_TRUE);
1303*0Sstevel@tonic-gate 
1304*0Sstevel@tonic-gate 				scf_instance_destroy(inst->ri_m_inst);
1305*0Sstevel@tonic-gate 			}
1306*0Sstevel@tonic-gate 			/* B_FALSE: See log_instance(..., "Enabled."); above */
1307*0Sstevel@tonic-gate 			log_instance(inst, B_FALSE, "Disabled.");
1308*0Sstevel@tonic-gate 			log_framework(LOG_DEBUG, "%s: Instance disabled.\n",
1309*0Sstevel@tonic-gate 			    inst->ri_i.i_fmri);
1310*0Sstevel@tonic-gate 			(void) restarter_instance_update_states(h, inst,
1311*0Sstevel@tonic-gate 			    RESTARTER_STATE_DISABLED, RESTARTER_STATE_NONE,
1312*0Sstevel@tonic-gate 			    RERR_RESTART, NULL);
1313*0Sstevel@tonic-gate 			return (0);
1314*0Sstevel@tonic-gate 
1315*0Sstevel@tonic-gate 		case RESTARTER_STATE_DISABLED:
1316*0Sstevel@tonic-gate 			break;
1317*0Sstevel@tonic-gate 
1318*0Sstevel@tonic-gate 		case RESTARTER_STATE_MAINT:
1319*0Sstevel@tonic-gate 			/*
1320*0Sstevel@tonic-gate 			 * We only want to pull the instance out of maintenance
1321*0Sstevel@tonic-gate 			 * if the disable is on adminstrative request.  The
1322*0Sstevel@tonic-gate 			 * graph engine sends _DISABLE events whenever a
1323*0Sstevel@tonic-gate 			 * service isn't in the disabled state, and we don't
1324*0Sstevel@tonic-gate 			 * want to pull the service out of maintenance if,
1325*0Sstevel@tonic-gate 			 * for example, it is there due to a dependency cycle.
1326*0Sstevel@tonic-gate 			 */
1327*0Sstevel@tonic-gate 			if (e == RESTARTER_EVENT_TYPE_ADMIN_DISABLE)
1328*0Sstevel@tonic-gate 				unmaintain_instance(h, inst, RUNMAINT_DISABLE);
1329*0Sstevel@tonic-gate 			break;
1330*0Sstevel@tonic-gate 
1331*0Sstevel@tonic-gate 		default:
1332*0Sstevel@tonic-gate #ifndef NDEBUG
1333*0Sstevel@tonic-gate 			(void) fprintf(stderr, "Restarter instance %s has "
1334*0Sstevel@tonic-gate 			    "unknown state %d.\n", inst->ri_i.i_fmri, state);
1335*0Sstevel@tonic-gate #endif
1336*0Sstevel@tonic-gate 			abort();
1337*0Sstevel@tonic-gate 		}
1338*0Sstevel@tonic-gate 	}
1339*0Sstevel@tonic-gate 
1340*0Sstevel@tonic-gate 	return (0);
1341*0Sstevel@tonic-gate }
1342*0Sstevel@tonic-gate 
1343*0Sstevel@tonic-gate static void
1344*0Sstevel@tonic-gate start_instance(scf_handle_t *local_handle, restarter_inst_t *inst)
1345*0Sstevel@tonic-gate {
1346*0Sstevel@tonic-gate 	fork_info_t *info;
1347*0Sstevel@tonic-gate 
1348*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&inst->ri_lock));
1349*0Sstevel@tonic-gate 	assert(instance_in_transition(inst) == 0);
1350*0Sstevel@tonic-gate 	assert(inst->ri_method_thread == 0);
1351*0Sstevel@tonic-gate 
1352*0Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: trying to start instance\n",
1353*0Sstevel@tonic-gate 	    inst->ri_i.i_fmri);
1354*0Sstevel@tonic-gate 
1355*0Sstevel@tonic-gate 	/* Services in the disabled and maintenance state are ignored */
1356*0Sstevel@tonic-gate 	if (inst->ri_i.i_state == RESTARTER_STATE_MAINT ||
1357*0Sstevel@tonic-gate 	    inst->ri_i.i_state == RESTARTER_STATE_DISABLED ||
1358*0Sstevel@tonic-gate 	    inst->ri_i.i_enabled == 0) {
1359*0Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
1360*0Sstevel@tonic-gate 		    "%s: start_instance -> is maint/disabled\n",
1361*0Sstevel@tonic-gate 		    inst->ri_i.i_fmri);
1362*0Sstevel@tonic-gate 		return;
1363*0Sstevel@tonic-gate 	}
1364*0Sstevel@tonic-gate 
1365*0Sstevel@tonic-gate 	/* Already started instances are left alone */
1366*0Sstevel@tonic-gate 	if (instance_started(inst) == 1) {
1367*0Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
1368*0Sstevel@tonic-gate 		    "%s: start_instance -> is already started\n",
1369*0Sstevel@tonic-gate 		    inst->ri_i.i_fmri);
1370*0Sstevel@tonic-gate 		return;
1371*0Sstevel@tonic-gate 	}
1372*0Sstevel@tonic-gate 
1373*0Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: starting instance.\n", inst->ri_i.i_fmri);
1374*0Sstevel@tonic-gate 
1375*0Sstevel@tonic-gate 	(void) restarter_instance_update_states(local_handle, inst,
1376*0Sstevel@tonic-gate 	    inst->ri_i.i_state, RESTARTER_STATE_ONLINE, RERR_NONE, NULL);
1377*0Sstevel@tonic-gate 
1378*0Sstevel@tonic-gate 	info = startd_zalloc(sizeof (fork_info_t));
1379*0Sstevel@tonic-gate 
1380*0Sstevel@tonic-gate 	info->sf_id = inst->ri_id;
1381*0Sstevel@tonic-gate 	info->sf_method_type = METHOD_START;
1382*0Sstevel@tonic-gate 	info->sf_event_type = RERR_NONE;
1383*0Sstevel@tonic-gate 	inst->ri_method_thread = startd_thread_create(method_thread, info);
1384*0Sstevel@tonic-gate }
1385*0Sstevel@tonic-gate 
1386*0Sstevel@tonic-gate static void
1387*0Sstevel@tonic-gate maintain_instance(scf_handle_t *h, restarter_inst_t *rip, int immediate,
1388*0Sstevel@tonic-gate     const char *aux)
1389*0Sstevel@tonic-gate {
1390*0Sstevel@tonic-gate 	fork_info_t *info;
1391*0Sstevel@tonic-gate 
1392*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&rip->ri_lock));
1393*0Sstevel@tonic-gate 	assert(aux != NULL);
1394*0Sstevel@tonic-gate 	assert(rip->ri_method_thread == 0);
1395*0Sstevel@tonic-gate 
1396*0Sstevel@tonic-gate 	log_instance(rip, B_TRUE, "Stopping for maintenance due to %s.", aux);
1397*0Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: stopping for maintenance due to %s.\n",
1398*0Sstevel@tonic-gate 	    rip->ri_i.i_fmri, aux);
1399*0Sstevel@tonic-gate 
1400*0Sstevel@tonic-gate 	/* Services in the maintenance state are ignored */
1401*0Sstevel@tonic-gate 	if (rip->ri_i.i_state == RESTARTER_STATE_MAINT) {
1402*0Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
1403*0Sstevel@tonic-gate 		    "%s: maintain_instance -> is already in maintenance\n",
1404*0Sstevel@tonic-gate 		    rip->ri_i.i_fmri);
1405*0Sstevel@tonic-gate 		return;
1406*0Sstevel@tonic-gate 	}
1407*0Sstevel@tonic-gate 
1408*0Sstevel@tonic-gate 	if (immediate || !instance_started(rip)) {
1409*0Sstevel@tonic-gate 		if (rip->ri_i.i_primary_ctid != 0) {
1410*0Sstevel@tonic-gate 			rip->ri_m_inst = safe_scf_instance_create(h);
1411*0Sstevel@tonic-gate 			rip->ri_mi_deleted = B_FALSE;
1412*0Sstevel@tonic-gate 
1413*0Sstevel@tonic-gate 			libscf_reget_instance(rip);
1414*0Sstevel@tonic-gate 			method_remove_contract(rip, B_TRUE, B_TRUE);
1415*0Sstevel@tonic-gate 
1416*0Sstevel@tonic-gate 			scf_instance_destroy(rip->ri_m_inst);
1417*0Sstevel@tonic-gate 		}
1418*0Sstevel@tonic-gate 
1419*0Sstevel@tonic-gate 		(void) restarter_instance_update_states(h, rip,
1420*0Sstevel@tonic-gate 		    RESTARTER_STATE_MAINT, RESTARTER_STATE_NONE, RERR_RESTART,
1421*0Sstevel@tonic-gate 		    (char *)aux);
1422*0Sstevel@tonic-gate 		return;
1423*0Sstevel@tonic-gate 	}
1424*0Sstevel@tonic-gate 
1425*0Sstevel@tonic-gate 	(void) restarter_instance_update_states(h, rip, rip->ri_i.i_state,
1426*0Sstevel@tonic-gate 	    RESTARTER_STATE_MAINT, RERR_NONE, (char *)aux);
1427*0Sstevel@tonic-gate 
1428*0Sstevel@tonic-gate 	info = startd_zalloc(sizeof (*info));
1429*0Sstevel@tonic-gate 	info->sf_id = rip->ri_id;
1430*0Sstevel@tonic-gate 	info->sf_method_type = METHOD_STOP;
1431*0Sstevel@tonic-gate 	info->sf_event_type = RERR_RESTART;
1432*0Sstevel@tonic-gate 	rip->ri_method_thread = startd_thread_create(method_thread, info);
1433*0Sstevel@tonic-gate }
1434*0Sstevel@tonic-gate 
1435*0Sstevel@tonic-gate static void
1436*0Sstevel@tonic-gate refresh_instance(scf_handle_t *h, restarter_inst_t *rip)
1437*0Sstevel@tonic-gate {
1438*0Sstevel@tonic-gate 	scf_instance_t *inst;
1439*0Sstevel@tonic-gate 	scf_snapshot_t *snap;
1440*0Sstevel@tonic-gate 	fork_info_t *info;
1441*0Sstevel@tonic-gate 	int r;
1442*0Sstevel@tonic-gate 
1443*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&rip->ri_lock));
1444*0Sstevel@tonic-gate 
1445*0Sstevel@tonic-gate 	log_instance(rip, B_TRUE, "Rereading configuration.");
1446*0Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "%s: rereading configuration.\n",
1447*0Sstevel@tonic-gate 	    rip->ri_i.i_fmri);
1448*0Sstevel@tonic-gate 
1449*0Sstevel@tonic-gate rep_retry:
1450*0Sstevel@tonic-gate 	r = libscf_fmri_get_instance(h, rip->ri_i.i_fmri, &inst);
1451*0Sstevel@tonic-gate 	switch (r) {
1452*0Sstevel@tonic-gate 	case 0:
1453*0Sstevel@tonic-gate 		break;
1454*0Sstevel@tonic-gate 
1455*0Sstevel@tonic-gate 	case ECONNABORTED:
1456*0Sstevel@tonic-gate 		libscf_handle_rebind(h);
1457*0Sstevel@tonic-gate 		goto rep_retry;
1458*0Sstevel@tonic-gate 
1459*0Sstevel@tonic-gate 	case ENOENT:
1460*0Sstevel@tonic-gate 		/* Must have been deleted. */
1461*0Sstevel@tonic-gate 		return;
1462*0Sstevel@tonic-gate 
1463*0Sstevel@tonic-gate 	case EINVAL:
1464*0Sstevel@tonic-gate 	case ENOTSUP:
1465*0Sstevel@tonic-gate 	default:
1466*0Sstevel@tonic-gate 		bad_error("libscf_fmri_get_instance", r);
1467*0Sstevel@tonic-gate 	}
1468*0Sstevel@tonic-gate 
1469*0Sstevel@tonic-gate 	snap = libscf_get_running_snapshot(inst);
1470*0Sstevel@tonic-gate 
1471*0Sstevel@tonic-gate 	r = libscf_get_startd_properties(inst, snap, &rip->ri_flags,
1472*0Sstevel@tonic-gate 	    &rip->ri_utmpx_prefix);
1473*0Sstevel@tonic-gate 	switch (r) {
1474*0Sstevel@tonic-gate 	case 0:
1475*0Sstevel@tonic-gate 		log_framework(LOG_DEBUG, "%s is a %s-style service\n",
1476*0Sstevel@tonic-gate 		    rip->ri_i.i_fmri, service_style(rip->ri_flags));
1477*0Sstevel@tonic-gate 		break;
1478*0Sstevel@tonic-gate 
1479*0Sstevel@tonic-gate 	case ECONNABORTED:
1480*0Sstevel@tonic-gate 		scf_instance_destroy(inst);
1481*0Sstevel@tonic-gate 		scf_snapshot_destroy(snap);
1482*0Sstevel@tonic-gate 		libscf_handle_rebind(h);
1483*0Sstevel@tonic-gate 		goto rep_retry;
1484*0Sstevel@tonic-gate 
1485*0Sstevel@tonic-gate 	case ECANCELED:
1486*0Sstevel@tonic-gate 	case ENOENT:
1487*0Sstevel@tonic-gate 		/* Succeed in anticipation of REMOVE_INSTANCE. */
1488*0Sstevel@tonic-gate 		break;
1489*0Sstevel@tonic-gate 
1490*0Sstevel@tonic-gate 	default:
1491*0Sstevel@tonic-gate 		bad_error("libscf_get_startd_properties", r);
1492*0Sstevel@tonic-gate 	}
1493*0Sstevel@tonic-gate 
1494*0Sstevel@tonic-gate 	if (instance_started(rip)) {
1495*0Sstevel@tonic-gate 		/* Refresh does not change the state. */
1496*0Sstevel@tonic-gate 		(void) restarter_instance_update_states(h, rip,
1497*0Sstevel@tonic-gate 		    rip->ri_i.i_state, rip->ri_i.i_state, RERR_NONE, NULL);
1498*0Sstevel@tonic-gate 
1499*0Sstevel@tonic-gate 		info = startd_zalloc(sizeof (*info));
1500*0Sstevel@tonic-gate 		info->sf_id = rip->ri_id;
1501*0Sstevel@tonic-gate 		info->sf_method_type = METHOD_REFRESH;
1502*0Sstevel@tonic-gate 		info->sf_event_type = RERR_REFRESH;
1503*0Sstevel@tonic-gate 
1504*0Sstevel@tonic-gate 		assert(rip->ri_method_thread == 0);
1505*0Sstevel@tonic-gate 		rip->ri_method_thread =
1506*0Sstevel@tonic-gate 		    startd_thread_create(method_thread, info);
1507*0Sstevel@tonic-gate 	}
1508*0Sstevel@tonic-gate 
1509*0Sstevel@tonic-gate 	scf_snapshot_destroy(snap);
1510*0Sstevel@tonic-gate 	scf_instance_destroy(inst);
1511*0Sstevel@tonic-gate }
1512*0Sstevel@tonic-gate 
1513*0Sstevel@tonic-gate const char *event_names[] = { "INVALID", "ADD_INSTANCE", "REMOVE_INSTANCE",
1514*0Sstevel@tonic-gate 	"ENABLE", "DISABLE", "ADMIN_DEGRADED", "ADMIN_REFRESH",
1515*0Sstevel@tonic-gate 	"ADMIN_RESTART", "ADMIN_MAINT_OFF", "ADMIN_MAINT_ON",
1516*0Sstevel@tonic-gate 	"ADMIN_MAINT_ON_IMMEDIATE", "STOP", "START", "DEPENDENCY_CYCLE",
1517*0Sstevel@tonic-gate 	"INVALID_DEPENDENCY", "ADMIN_DISABLE"
1518*0Sstevel@tonic-gate };
1519*0Sstevel@tonic-gate 
1520*0Sstevel@tonic-gate /*
1521*0Sstevel@tonic-gate  * void *restarter_process_events()
1522*0Sstevel@tonic-gate  *
1523*0Sstevel@tonic-gate  *   Called in a separate thread to process the events on an instance's
1524*0Sstevel@tonic-gate  *   queue.  Empties the queue completely, and tries to keep the thread
1525*0Sstevel@tonic-gate  *   around for a little while after the queue is empty to save on
1526*0Sstevel@tonic-gate  *   startup costs.
1527*0Sstevel@tonic-gate  */
1528*0Sstevel@tonic-gate static void *
1529*0Sstevel@tonic-gate restarter_process_events(void *arg)
1530*0Sstevel@tonic-gate {
1531*0Sstevel@tonic-gate 	scf_handle_t *h;
1532*0Sstevel@tonic-gate 	restarter_instance_qentry_t *event;
1533*0Sstevel@tonic-gate 	restarter_inst_t *rip;
1534*0Sstevel@tonic-gate 	char *fmri = (char *)arg;
1535*0Sstevel@tonic-gate 	struct timespec to;
1536*0Sstevel@tonic-gate 
1537*0Sstevel@tonic-gate 	assert(fmri != NULL);
1538*0Sstevel@tonic-gate 
1539*0Sstevel@tonic-gate 	h = libscf_handle_create_bound_loop();
1540*0Sstevel@tonic-gate 
1541*0Sstevel@tonic-gate 	/* grab the queue lock */
1542*0Sstevel@tonic-gate 	rip = inst_lookup_queue(fmri);
1543*0Sstevel@tonic-gate 	if (rip == NULL)
1544*0Sstevel@tonic-gate 		goto out;
1545*0Sstevel@tonic-gate 
1546*0Sstevel@tonic-gate again:
1547*0Sstevel@tonic-gate 
1548*0Sstevel@tonic-gate 	while ((event = uu_list_first(rip->ri_queue)) != NULL) {
1549*0Sstevel@tonic-gate 		restarter_inst_t *inst;
1550*0Sstevel@tonic-gate 
1551*0Sstevel@tonic-gate 		/* drop the queue lock */
1552*0Sstevel@tonic-gate 		MUTEX_UNLOCK(&rip->ri_queue_lock);
1553*0Sstevel@tonic-gate 
1554*0Sstevel@tonic-gate 		/*
1555*0Sstevel@tonic-gate 		 * Grab the inst lock -- this waits until any outstanding
1556*0Sstevel@tonic-gate 		 * method finishes running.
1557*0Sstevel@tonic-gate 		 */
1558*0Sstevel@tonic-gate 		inst = inst_lookup_by_name(fmri);
1559*0Sstevel@tonic-gate 		if (inst == NULL) {
1560*0Sstevel@tonic-gate 			/* Getting deleted in the middle isn't an error. */
1561*0Sstevel@tonic-gate 			goto cont;
1562*0Sstevel@tonic-gate 		}
1563*0Sstevel@tonic-gate 
1564*0Sstevel@tonic-gate 		assert(instance_in_transition(inst) == 0);
1565*0Sstevel@tonic-gate 
1566*0Sstevel@tonic-gate 		/* process the event */
1567*0Sstevel@tonic-gate 		switch (event->riq_type) {
1568*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ENABLE:
1569*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_DISABLE:
1570*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_DISABLE:
1571*0Sstevel@tonic-gate 			(void) enable_inst(h, inst, event->riq_type);
1572*0Sstevel@tonic-gate 			break;
1573*0Sstevel@tonic-gate 
1574*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_REMOVE_INSTANCE:
1575*0Sstevel@tonic-gate 			restarter_delete_inst(inst);
1576*0Sstevel@tonic-gate 			inst = NULL;
1577*0Sstevel@tonic-gate 			goto cont;
1578*0Sstevel@tonic-gate 
1579*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_STOP:
1580*0Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_DEPENDENCY);
1581*0Sstevel@tonic-gate 			break;
1582*0Sstevel@tonic-gate 
1583*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_START:
1584*0Sstevel@tonic-gate 			start_instance(h, inst);
1585*0Sstevel@tonic-gate 			break;
1586*0Sstevel@tonic-gate 
1587*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_DEPENDENCY_CYCLE:
1588*0Sstevel@tonic-gate 			maintain_instance(h, inst, 0, "dependency_cycle");
1589*0Sstevel@tonic-gate 			break;
1590*0Sstevel@tonic-gate 
1591*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_INVALID_DEPENDENCY:
1592*0Sstevel@tonic-gate 			maintain_instance(h, inst, 0, "invalid_dependency");
1593*0Sstevel@tonic-gate 			break;
1594*0Sstevel@tonic-gate 
1595*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON:
1596*0Sstevel@tonic-gate 			maintain_instance(h, inst, 0, "administrative_request");
1597*0Sstevel@tonic-gate 			break;
1598*0Sstevel@tonic-gate 
1599*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON_IMMEDIATE:
1600*0Sstevel@tonic-gate 			maintain_instance(h, inst, 1, "administrative_request");
1601*0Sstevel@tonic-gate 			break;
1602*0Sstevel@tonic-gate 
1603*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_OFF:
1604*0Sstevel@tonic-gate 			unmaintain_instance(h, inst, RUNMAINT_CLEAR);
1605*0Sstevel@tonic-gate 			break;
1606*0Sstevel@tonic-gate 
1607*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_REFRESH:
1608*0Sstevel@tonic-gate 			refresh_instance(h, inst);
1609*0Sstevel@tonic-gate 			break;
1610*0Sstevel@tonic-gate 
1611*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_DEGRADED:
1612*0Sstevel@tonic-gate 			log_framework(LOG_WARNING, "Restarter: "
1613*0Sstevel@tonic-gate 			    "%s command (for %s) unimplemented.\n",
1614*0Sstevel@tonic-gate 			    event_names[event->riq_type], inst->ri_i.i_fmri);
1615*0Sstevel@tonic-gate 			break;
1616*0Sstevel@tonic-gate 
1617*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADMIN_RESTART:
1618*0Sstevel@tonic-gate 			if (!instance_started(inst)) {
1619*0Sstevel@tonic-gate 				log_framework(LOG_DEBUG, "Restarter: "
1620*0Sstevel@tonic-gate 				    "Not restarting %s; not running.\n",
1621*0Sstevel@tonic-gate 				    inst->ri_i.i_fmri);
1622*0Sstevel@tonic-gate 			} else {
1623*0Sstevel@tonic-gate 				/*
1624*0Sstevel@tonic-gate 				 * Stop the instance.  If it can be restarted,
1625*0Sstevel@tonic-gate 				 * the graph engine will send a new event.
1626*0Sstevel@tonic-gate 				 */
1627*0Sstevel@tonic-gate 				(void) stop_instance(h, inst, RSTOP_RESTART);
1628*0Sstevel@tonic-gate 			}
1629*0Sstevel@tonic-gate 			break;
1630*0Sstevel@tonic-gate 
1631*0Sstevel@tonic-gate 		case RESTARTER_EVENT_TYPE_ADD_INSTANCE:
1632*0Sstevel@tonic-gate 		default:
1633*0Sstevel@tonic-gate #ifndef NDEBUG
1634*0Sstevel@tonic-gate 			uu_warn("%s:%d: Bad restarter event %d.  "
1635*0Sstevel@tonic-gate 			    "Aborting.\n", __FILE__, __LINE__, event->riq_type);
1636*0Sstevel@tonic-gate #endif
1637*0Sstevel@tonic-gate 			abort();
1638*0Sstevel@tonic-gate 		}
1639*0Sstevel@tonic-gate 
1640*0Sstevel@tonic-gate 		assert(inst != NULL);
1641*0Sstevel@tonic-gate 		MUTEX_UNLOCK(&inst->ri_lock);
1642*0Sstevel@tonic-gate 
1643*0Sstevel@tonic-gate cont:
1644*0Sstevel@tonic-gate 		/* grab the queue lock */
1645*0Sstevel@tonic-gate 		rip = inst_lookup_queue(fmri);
1646*0Sstevel@tonic-gate 		if (rip == NULL)
1647*0Sstevel@tonic-gate 			goto out;
1648*0Sstevel@tonic-gate 
1649*0Sstevel@tonic-gate 		/* delete the event */
1650*0Sstevel@tonic-gate 		uu_list_remove(rip->ri_queue, event);
1651*0Sstevel@tonic-gate 		startd_free(event, sizeof (restarter_instance_qentry_t));
1652*0Sstevel@tonic-gate 	}
1653*0Sstevel@tonic-gate 
1654*0Sstevel@tonic-gate 	assert(rip != NULL);
1655*0Sstevel@tonic-gate 
1656*0Sstevel@tonic-gate 	/*
1657*0Sstevel@tonic-gate 	 * Try to preserve the thread for a little while for future use.
1658*0Sstevel@tonic-gate 	 */
1659*0Sstevel@tonic-gate 	to.tv_sec = 3;
1660*0Sstevel@tonic-gate 	to.tv_nsec = 0;
1661*0Sstevel@tonic-gate 	(void) pthread_cond_reltimedwait_np(&rip->ri_queue_cv,
1662*0Sstevel@tonic-gate 	    &rip->ri_queue_lock, &to);
1663*0Sstevel@tonic-gate 
1664*0Sstevel@tonic-gate 	if (uu_list_first(rip->ri_queue) != NULL)
1665*0Sstevel@tonic-gate 		goto again;
1666*0Sstevel@tonic-gate 
1667*0Sstevel@tonic-gate 	rip->ri_queue_thread = 0;
1668*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&rip->ri_queue_lock);
1669*0Sstevel@tonic-gate out:
1670*0Sstevel@tonic-gate 	(void) scf_handle_unbind(h);
1671*0Sstevel@tonic-gate 	scf_handle_destroy(h);
1672*0Sstevel@tonic-gate 	free(fmri);
1673*0Sstevel@tonic-gate 	return (NULL);
1674*0Sstevel@tonic-gate }
1675*0Sstevel@tonic-gate 
1676*0Sstevel@tonic-gate static int
1677*0Sstevel@tonic-gate is_admin_event(restarter_event_type_t t) {
1678*0Sstevel@tonic-gate 
1679*0Sstevel@tonic-gate 	switch (t) {
1680*0Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON:
1681*0Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON_IMMEDIATE:
1682*0Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_MAINT_OFF:
1683*0Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_REFRESH:
1684*0Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_DEGRADED:
1685*0Sstevel@tonic-gate 	case RESTARTER_EVENT_TYPE_ADMIN_RESTART:
1686*0Sstevel@tonic-gate 		return (1);
1687*0Sstevel@tonic-gate 	default:
1688*0Sstevel@tonic-gate 		return (0);
1689*0Sstevel@tonic-gate 	}
1690*0Sstevel@tonic-gate }
1691*0Sstevel@tonic-gate 
1692*0Sstevel@tonic-gate static void
1693*0Sstevel@tonic-gate restarter_queue_event(restarter_inst_t *ri, restarter_protocol_event_t *e)
1694*0Sstevel@tonic-gate {
1695*0Sstevel@tonic-gate 	restarter_instance_qentry_t *qe;
1696*0Sstevel@tonic-gate 	int r;
1697*0Sstevel@tonic-gate 
1698*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&ri->ri_queue_lock));
1699*0Sstevel@tonic-gate 	assert(!PTHREAD_MUTEX_HELD(&ri->ri_lock));
1700*0Sstevel@tonic-gate 
1701*0Sstevel@tonic-gate 	qe = startd_zalloc(sizeof (restarter_instance_qentry_t));
1702*0Sstevel@tonic-gate 	qe->riq_type = e->rpe_type;
1703*0Sstevel@tonic-gate 
1704*0Sstevel@tonic-gate 	uu_list_node_init(qe, &qe->riq_link, restarter_queue_pool);
1705*0Sstevel@tonic-gate 	r = uu_list_insert_before(ri->ri_queue, NULL, qe);
1706*0Sstevel@tonic-gate 	assert(r == 0);
1707*0Sstevel@tonic-gate }
1708*0Sstevel@tonic-gate 
1709*0Sstevel@tonic-gate /*
1710*0Sstevel@tonic-gate  * void *restarter_event_thread()
1711*0Sstevel@tonic-gate  *
1712*0Sstevel@tonic-gate  *  Handle incoming graph events by placing them on a per-instance
1713*0Sstevel@tonic-gate  *  queue.  We can't lock the main part of the instance structure, so
1714*0Sstevel@tonic-gate  *  just modify the seprarately locked event queue portion.
1715*0Sstevel@tonic-gate  */
1716*0Sstevel@tonic-gate /*ARGSUSED*/
1717*0Sstevel@tonic-gate static void *
1718*0Sstevel@tonic-gate restarter_event_thread(void *unused)
1719*0Sstevel@tonic-gate {
1720*0Sstevel@tonic-gate 	scf_handle_t *h;
1721*0Sstevel@tonic-gate 
1722*0Sstevel@tonic-gate 	/*
1723*0Sstevel@tonic-gate 	 * This is a new thread, and thus, gets its own handle
1724*0Sstevel@tonic-gate 	 * to the repository.
1725*0Sstevel@tonic-gate 	 */
1726*0Sstevel@tonic-gate 	h = libscf_handle_create_bound_loop();
1727*0Sstevel@tonic-gate 
1728*0Sstevel@tonic-gate 	MUTEX_LOCK(&ru->restarter_update_lock);
1729*0Sstevel@tonic-gate 
1730*0Sstevel@tonic-gate 	/*CONSTCOND*/
1731*0Sstevel@tonic-gate 	while (1) {
1732*0Sstevel@tonic-gate 		restarter_protocol_event_t *e;
1733*0Sstevel@tonic-gate 
1734*0Sstevel@tonic-gate 		while (ru->restarter_update_wakeup == 0)
1735*0Sstevel@tonic-gate 			(void) pthread_cond_wait(&ru->restarter_update_cv,
1736*0Sstevel@tonic-gate 			    &ru->restarter_update_lock);
1737*0Sstevel@tonic-gate 
1738*0Sstevel@tonic-gate 		ru->restarter_update_wakeup = 0;
1739*0Sstevel@tonic-gate 
1740*0Sstevel@tonic-gate 		while ((e = restarter_event_dequeue()) != NULL) {
1741*0Sstevel@tonic-gate 			restarter_inst_t *rip;
1742*0Sstevel@tonic-gate 			char *fmri;
1743*0Sstevel@tonic-gate 
1744*0Sstevel@tonic-gate 			MUTEX_UNLOCK(&ru->restarter_update_lock);
1745*0Sstevel@tonic-gate 
1746*0Sstevel@tonic-gate 			/*
1747*0Sstevel@tonic-gate 			 * ADD_INSTANCE is special: there's likely no
1748*0Sstevel@tonic-gate 			 * instance structure yet, so we need to handle the
1749*0Sstevel@tonic-gate 			 * addition synchronously.
1750*0Sstevel@tonic-gate 			 */
1751*0Sstevel@tonic-gate 			switch (e->rpe_type) {
1752*0Sstevel@tonic-gate 			case RESTARTER_EVENT_TYPE_ADD_INSTANCE:
1753*0Sstevel@tonic-gate 				if (restarter_insert_inst(h, e->rpe_inst) != 0)
1754*0Sstevel@tonic-gate 					log_error(LOG_INFO, "Restarter: "
1755*0Sstevel@tonic-gate 					    "Could not add %s.\n", e->rpe_inst);
1756*0Sstevel@tonic-gate 
1757*0Sstevel@tonic-gate 				MUTEX_LOCK(&st->st_load_lock);
1758*0Sstevel@tonic-gate 				if (--st->st_load_instances == 0)
1759*0Sstevel@tonic-gate 					(void) pthread_cond_broadcast(
1760*0Sstevel@tonic-gate 					    &st->st_load_cv);
1761*0Sstevel@tonic-gate 				MUTEX_UNLOCK(&st->st_load_lock);
1762*0Sstevel@tonic-gate 
1763*0Sstevel@tonic-gate 				goto nolookup;
1764*0Sstevel@tonic-gate 			}
1765*0Sstevel@tonic-gate 
1766*0Sstevel@tonic-gate 			/*
1767*0Sstevel@tonic-gate 			 * Lookup the instance, locking only the event queue.
1768*0Sstevel@tonic-gate 			 * Can't grab ri_lock here because it might be held
1769*0Sstevel@tonic-gate 			 * by a long-running method.
1770*0Sstevel@tonic-gate 			 */
1771*0Sstevel@tonic-gate 			rip = inst_lookup_queue(e->rpe_inst);
1772*0Sstevel@tonic-gate 			if (rip == NULL) {
1773*0Sstevel@tonic-gate 				log_error(LOG_INFO, "Restarter: "
1774*0Sstevel@tonic-gate 				    "Ignoring %s command for unknown service "
1775*0Sstevel@tonic-gate 				    "%s.\n", event_names[e->rpe_type],
1776*0Sstevel@tonic-gate 				    e->rpe_inst);
1777*0Sstevel@tonic-gate 				goto nolookup;
1778*0Sstevel@tonic-gate 			}
1779*0Sstevel@tonic-gate 
1780*0Sstevel@tonic-gate 			/* Keep ADMIN events from filling up the queue. */
1781*0Sstevel@tonic-gate 			if (is_admin_event(e->rpe_type) &&
1782*0Sstevel@tonic-gate 			    uu_list_numnodes(rip->ri_queue) >
1783*0Sstevel@tonic-gate 			    RINST_QUEUE_THRESHOLD) {
1784*0Sstevel@tonic-gate 				MUTEX_UNLOCK(&rip->ri_queue_lock);
1785*0Sstevel@tonic-gate 				log_instance(rip, B_TRUE, "Instance event "
1786*0Sstevel@tonic-gate 				    "queue overflow.  Dropping administrative "
1787*0Sstevel@tonic-gate 				    "request.");
1788*0Sstevel@tonic-gate 				log_framework(LOG_DEBUG, "%s: Instance event "
1789*0Sstevel@tonic-gate 				    "queue overflow.  Dropping administrative "
1790*0Sstevel@tonic-gate 				    "request.\n", rip->ri_i.i_fmri);
1791*0Sstevel@tonic-gate 				goto nolookup;
1792*0Sstevel@tonic-gate 			}
1793*0Sstevel@tonic-gate 
1794*0Sstevel@tonic-gate 			/* Now add the event to the instance queue. */
1795*0Sstevel@tonic-gate 			restarter_queue_event(rip, e);
1796*0Sstevel@tonic-gate 
1797*0Sstevel@tonic-gate 			if (rip->ri_queue_thread == 0) {
1798*0Sstevel@tonic-gate 				/*
1799*0Sstevel@tonic-gate 				 * Start a thread if one isn't already
1800*0Sstevel@tonic-gate 				 * running.
1801*0Sstevel@tonic-gate 				 */
1802*0Sstevel@tonic-gate 				fmri = safe_strdup(e->rpe_inst);
1803*0Sstevel@tonic-gate 				rip->ri_queue_thread =  startd_thread_create(
1804*0Sstevel@tonic-gate 				    restarter_process_events, (void *)fmri);
1805*0Sstevel@tonic-gate 			} else {
1806*0Sstevel@tonic-gate 				/*
1807*0Sstevel@tonic-gate 				 * Signal the existing thread that there's
1808*0Sstevel@tonic-gate 				 * a new event.
1809*0Sstevel@tonic-gate 				 */
1810*0Sstevel@tonic-gate 				(void) pthread_cond_broadcast(
1811*0Sstevel@tonic-gate 				    &rip->ri_queue_cv);
1812*0Sstevel@tonic-gate 			}
1813*0Sstevel@tonic-gate 
1814*0Sstevel@tonic-gate 			MUTEX_UNLOCK(&rip->ri_queue_lock);
1815*0Sstevel@tonic-gate nolookup:
1816*0Sstevel@tonic-gate 			restarter_event_release(e);
1817*0Sstevel@tonic-gate 
1818*0Sstevel@tonic-gate 			MUTEX_LOCK(&ru->restarter_update_lock);
1819*0Sstevel@tonic-gate 		}
1820*0Sstevel@tonic-gate 	}
1821*0Sstevel@tonic-gate 
1822*0Sstevel@tonic-gate 	/*
1823*0Sstevel@tonic-gate 	 * Unreachable for now -- there's currently no graceful cleanup
1824*0Sstevel@tonic-gate 	 * called on exit().
1825*0Sstevel@tonic-gate 	 */
1826*0Sstevel@tonic-gate 	(void) scf_handle_unbind(h);
1827*0Sstevel@tonic-gate 	scf_handle_destroy(h);
1828*0Sstevel@tonic-gate 	return (NULL);
1829*0Sstevel@tonic-gate }
1830*0Sstevel@tonic-gate 
1831*0Sstevel@tonic-gate static restarter_inst_t *
1832*0Sstevel@tonic-gate contract_to_inst(ctid_t ctid)
1833*0Sstevel@tonic-gate {
1834*0Sstevel@tonic-gate 	restarter_inst_t *inst;
1835*0Sstevel@tonic-gate 	int id;
1836*0Sstevel@tonic-gate 
1837*0Sstevel@tonic-gate 	id = lookup_inst_by_contract(ctid);
1838*0Sstevel@tonic-gate 	if (id == -1)
1839*0Sstevel@tonic-gate 		return (NULL);
1840*0Sstevel@tonic-gate 
1841*0Sstevel@tonic-gate 	inst = inst_lookup_by_id(id);
1842*0Sstevel@tonic-gate 	if (inst != NULL) {
1843*0Sstevel@tonic-gate 		/*
1844*0Sstevel@tonic-gate 		 * Since ri_lock isn't held by the contract id lookup, this
1845*0Sstevel@tonic-gate 		 * instance may have been restarted and now be in a new
1846*0Sstevel@tonic-gate 		 * contract, making the old contract no longer valid for this
1847*0Sstevel@tonic-gate 		 * instance.
1848*0Sstevel@tonic-gate 		 */
1849*0Sstevel@tonic-gate 		if (ctid != inst->ri_i.i_primary_ctid) {
1850*0Sstevel@tonic-gate 			MUTEX_UNLOCK(&inst->ri_lock);
1851*0Sstevel@tonic-gate 			inst = NULL;
1852*0Sstevel@tonic-gate 		}
1853*0Sstevel@tonic-gate 	}
1854*0Sstevel@tonic-gate 	return (inst);
1855*0Sstevel@tonic-gate }
1856*0Sstevel@tonic-gate 
1857*0Sstevel@tonic-gate /*
1858*0Sstevel@tonic-gate  * void contract_action()
1859*0Sstevel@tonic-gate  *   Take action on contract events.
1860*0Sstevel@tonic-gate  */
1861*0Sstevel@tonic-gate static void
1862*0Sstevel@tonic-gate contract_action(scf_handle_t *h, restarter_inst_t *inst, ctid_t id,
1863*0Sstevel@tonic-gate     uint32_t type)
1864*0Sstevel@tonic-gate {
1865*0Sstevel@tonic-gate 	const char *fmri = inst->ri_i.i_fmri;
1866*0Sstevel@tonic-gate 
1867*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&inst->ri_lock));
1868*0Sstevel@tonic-gate 
1869*0Sstevel@tonic-gate 	/*
1870*0Sstevel@tonic-gate 	 * If startd has stopped this contract, there is no need to
1871*0Sstevel@tonic-gate 	 * stop it again.
1872*0Sstevel@tonic-gate 	 */
1873*0Sstevel@tonic-gate 	if (inst->ri_i.i_primary_ctid > 0 &&
1874*0Sstevel@tonic-gate 	    inst->ri_i.i_primary_ctid_stopped)
1875*0Sstevel@tonic-gate 		return;
1876*0Sstevel@tonic-gate 
1877*0Sstevel@tonic-gate 	if ((type & (CT_PR_EV_EMPTY | CT_PR_EV_CORE | CT_PR_EV_SIGNAL
1878*0Sstevel@tonic-gate 	    | CT_PR_EV_HWERR)) == 0) {
1879*0Sstevel@tonic-gate 		/*
1880*0Sstevel@tonic-gate 		 * There shouldn't be other events, since that's not how we set
1881*0Sstevel@tonic-gate 		 * the terms. Thus, just log an error and drive on.
1882*0Sstevel@tonic-gate 		 */
1883*0Sstevel@tonic-gate 		log_framework(LOG_NOTICE,
1884*0Sstevel@tonic-gate 		    "%s: contract %ld received unexpected critical event "
1885*0Sstevel@tonic-gate 		    "(%d)\n", fmri, id, type);
1886*0Sstevel@tonic-gate 		    return;
1887*0Sstevel@tonic-gate 	}
1888*0Sstevel@tonic-gate 
1889*0Sstevel@tonic-gate 	assert(instance_in_transition(inst) == 0);
1890*0Sstevel@tonic-gate 
1891*0Sstevel@tonic-gate 	if (instance_is_wait_style(inst)) {
1892*0Sstevel@tonic-gate 		/*
1893*0Sstevel@tonic-gate 		 * We ignore all events; if they impact the
1894*0Sstevel@tonic-gate 		 * process we're monitoring, then the
1895*0Sstevel@tonic-gate 		 * wait_thread will stop the instance.
1896*0Sstevel@tonic-gate 		 */
1897*0Sstevel@tonic-gate 		log_framework(LOG_DEBUG,
1898*0Sstevel@tonic-gate 		    "%s: ignoring contract event on wait-style service\n",
1899*0Sstevel@tonic-gate 		    fmri);
1900*0Sstevel@tonic-gate 	} else {
1901*0Sstevel@tonic-gate 		/*
1902*0Sstevel@tonic-gate 		 * A CT_PR_EV_EMPTY event is an RSTOP_EXIT request.
1903*0Sstevel@tonic-gate 		 */
1904*0Sstevel@tonic-gate 		switch (type) {
1905*0Sstevel@tonic-gate 		case CT_PR_EV_EMPTY:
1906*0Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_EXIT);
1907*0Sstevel@tonic-gate 			break;
1908*0Sstevel@tonic-gate 		case CT_PR_EV_CORE:
1909*0Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_CORE);
1910*0Sstevel@tonic-gate 			break;
1911*0Sstevel@tonic-gate 		case CT_PR_EV_SIGNAL:
1912*0Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_SIGNAL);
1913*0Sstevel@tonic-gate 			break;
1914*0Sstevel@tonic-gate 		case CT_PR_EV_HWERR:
1915*0Sstevel@tonic-gate 			(void) stop_instance(h, inst, RSTOP_HWERR);
1916*0Sstevel@tonic-gate 			break;
1917*0Sstevel@tonic-gate 		}
1918*0Sstevel@tonic-gate 	}
1919*0Sstevel@tonic-gate }
1920*0Sstevel@tonic-gate 
1921*0Sstevel@tonic-gate /*
1922*0Sstevel@tonic-gate  * void *restarter_contract_event_thread(void *)
1923*0Sstevel@tonic-gate  *   Listens to the process contract bundle for critical events, taking action
1924*0Sstevel@tonic-gate  *   on events from contracts we know we are responsible for.
1925*0Sstevel@tonic-gate  */
1926*0Sstevel@tonic-gate /*ARGSUSED*/
1927*0Sstevel@tonic-gate static void *
1928*0Sstevel@tonic-gate restarter_contracts_event_thread(void *unused)
1929*0Sstevel@tonic-gate {
1930*0Sstevel@tonic-gate 	int fd, err;
1931*0Sstevel@tonic-gate 	scf_handle_t *local_handle;
1932*0Sstevel@tonic-gate 
1933*0Sstevel@tonic-gate 	/*
1934*0Sstevel@tonic-gate 	 * Await graph load completion.  That is, stop here, until we've scanned
1935*0Sstevel@tonic-gate 	 * the repository for contract - instance associations.
1936*0Sstevel@tonic-gate 	 */
1937*0Sstevel@tonic-gate 	MUTEX_LOCK(&st->st_load_lock);
1938*0Sstevel@tonic-gate 	while (!(st->st_load_complete && st->st_load_instances == 0))
1939*0Sstevel@tonic-gate 		(void) pthread_cond_wait(&st->st_load_cv, &st->st_load_lock);
1940*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&st->st_load_lock);
1941*0Sstevel@tonic-gate 
1942*0Sstevel@tonic-gate 	/*
1943*0Sstevel@tonic-gate 	 * This is a new thread, and thus, gets its own handle
1944*0Sstevel@tonic-gate 	 * to the repository.
1945*0Sstevel@tonic-gate 	 */
1946*0Sstevel@tonic-gate 	if ((local_handle = libscf_handle_create_bound(SCF_VERSION)) == NULL)
1947*0Sstevel@tonic-gate 		uu_die("Unable to bind a new repository handle: %s\n",
1948*0Sstevel@tonic-gate 		    scf_strerror(scf_error()));
1949*0Sstevel@tonic-gate 
1950*0Sstevel@tonic-gate 	fd = open64(CTFS_ROOT "/process/pbundle", O_RDONLY);
1951*0Sstevel@tonic-gate 	if (fd == -1)
1952*0Sstevel@tonic-gate 		uu_die("process bundle open failed");
1953*0Sstevel@tonic-gate 
1954*0Sstevel@tonic-gate 	/*
1955*0Sstevel@tonic-gate 	 * Make sure we get all events (including those generated by configd
1956*0Sstevel@tonic-gate 	 * before this thread was started).
1957*0Sstevel@tonic-gate 	 */
1958*0Sstevel@tonic-gate 	err = ct_event_reset(fd);
1959*0Sstevel@tonic-gate 	assert(err == 0);
1960*0Sstevel@tonic-gate 
1961*0Sstevel@tonic-gate 	for (;;) {
1962*0Sstevel@tonic-gate 		int efd, sfd;
1963*0Sstevel@tonic-gate 		ct_evthdl_t ev;
1964*0Sstevel@tonic-gate 		uint32_t type;
1965*0Sstevel@tonic-gate 		ctevid_t evid;
1966*0Sstevel@tonic-gate 		ct_stathdl_t status;
1967*0Sstevel@tonic-gate 		ctid_t ctid;
1968*0Sstevel@tonic-gate 		restarter_inst_t *inst;
1969*0Sstevel@tonic-gate 		uint64_t cookie;
1970*0Sstevel@tonic-gate 
1971*0Sstevel@tonic-gate 		if (err = ct_event_read_critical(fd, &ev)) {
1972*0Sstevel@tonic-gate 			log_error(LOG_WARNING,
1973*0Sstevel@tonic-gate 			    "Error reading next contract event: %s",
1974*0Sstevel@tonic-gate 			    strerror(err));
1975*0Sstevel@tonic-gate 			continue;
1976*0Sstevel@tonic-gate 		}
1977*0Sstevel@tonic-gate 
1978*0Sstevel@tonic-gate 		evid = ct_event_get_evid(ev);
1979*0Sstevel@tonic-gate 		ctid = ct_event_get_ctid(ev);
1980*0Sstevel@tonic-gate 		type = ct_event_get_type(ev);
1981*0Sstevel@tonic-gate 
1982*0Sstevel@tonic-gate 		/* Fetch cookie. */
1983*0Sstevel@tonic-gate 		if ((sfd = contract_open(ctid, "process", "status", O_RDONLY))
1984*0Sstevel@tonic-gate 		    < 0) {
1985*0Sstevel@tonic-gate 			ct_event_free(ev);
1986*0Sstevel@tonic-gate 			continue;
1987*0Sstevel@tonic-gate 		}
1988*0Sstevel@tonic-gate 
1989*0Sstevel@tonic-gate 		if (err = ct_status_read(sfd, CTD_COMMON, &status)) {
1990*0Sstevel@tonic-gate 			log_framework(LOG_WARNING, "Could not get status for "
1991*0Sstevel@tonic-gate 			    "contract %ld: %s\n", ctid, strerror(err));
1992*0Sstevel@tonic-gate 
1993*0Sstevel@tonic-gate 			startd_close(sfd);
1994*0Sstevel@tonic-gate 			ct_event_free(ev);
1995*0Sstevel@tonic-gate 			continue;
1996*0Sstevel@tonic-gate 		}
1997*0Sstevel@tonic-gate 
1998*0Sstevel@tonic-gate 		cookie = ct_status_get_cookie(status);
1999*0Sstevel@tonic-gate 
2000*0Sstevel@tonic-gate 		ct_status_free(status);
2001*0Sstevel@tonic-gate 
2002*0Sstevel@tonic-gate 		startd_close(sfd);
2003*0Sstevel@tonic-gate 
2004*0Sstevel@tonic-gate 		/*
2005*0Sstevel@tonic-gate 		 * svc.configd(1M) restart handling performed by the
2006*0Sstevel@tonic-gate 		 * fork_configd_thread.  We don't acknowledge, as that thread
2007*0Sstevel@tonic-gate 		 * will do so.
2008*0Sstevel@tonic-gate 		 */
2009*0Sstevel@tonic-gate 		if (cookie == CONFIGD_COOKIE) {
2010*0Sstevel@tonic-gate 			ct_event_free(ev);
2011*0Sstevel@tonic-gate 			continue;
2012*0Sstevel@tonic-gate 		}
2013*0Sstevel@tonic-gate 
2014*0Sstevel@tonic-gate 		inst = contract_to_inst(ctid);
2015*0Sstevel@tonic-gate 		if (inst == NULL) {
2016*0Sstevel@tonic-gate 			/*
2017*0Sstevel@tonic-gate 			 * This can happen if we receive an EMPTY
2018*0Sstevel@tonic-gate 			 * event for an abandoned contract.
2019*0Sstevel@tonic-gate 			 */
2020*0Sstevel@tonic-gate 			log_framework(LOG_DEBUG,
2021*0Sstevel@tonic-gate 			    "Received event %d for unknown contract id "
2022*0Sstevel@tonic-gate 			    "%ld\n", type, ctid);
2023*0Sstevel@tonic-gate 		} else {
2024*0Sstevel@tonic-gate 			log_framework(LOG_DEBUG,
2025*0Sstevel@tonic-gate 			    "Received event %d for contract id "
2026*0Sstevel@tonic-gate 			    "%ld (%s)\n", type, ctid,
2027*0Sstevel@tonic-gate 			    inst->ri_i.i_fmri);
2028*0Sstevel@tonic-gate 
2029*0Sstevel@tonic-gate 			contract_action(local_handle, inst, ctid, type);
2030*0Sstevel@tonic-gate 
2031*0Sstevel@tonic-gate 			MUTEX_UNLOCK(&inst->ri_lock);
2032*0Sstevel@tonic-gate 		}
2033*0Sstevel@tonic-gate 
2034*0Sstevel@tonic-gate 		efd = contract_open(ct_event_get_ctid(ev), "process", "ctl",
2035*0Sstevel@tonic-gate 		    O_WRONLY);
2036*0Sstevel@tonic-gate 		if (efd != -1) {
2037*0Sstevel@tonic-gate 			(void) ct_ctl_ack(efd, evid);
2038*0Sstevel@tonic-gate 			startd_close(efd);
2039*0Sstevel@tonic-gate 		}
2040*0Sstevel@tonic-gate 
2041*0Sstevel@tonic-gate 		ct_event_free(ev);
2042*0Sstevel@tonic-gate 
2043*0Sstevel@tonic-gate 	}
2044*0Sstevel@tonic-gate 
2045*0Sstevel@tonic-gate 	/*NOTREACHED*/
2046*0Sstevel@tonic-gate 	return (NULL);
2047*0Sstevel@tonic-gate }
2048*0Sstevel@tonic-gate 
2049*0Sstevel@tonic-gate /*
2050*0Sstevel@tonic-gate  * Timeout queue, processed by restarter_timeouts_event_thread().
2051*0Sstevel@tonic-gate  */
2052*0Sstevel@tonic-gate timeout_queue_t *timeouts;
2053*0Sstevel@tonic-gate static uu_list_pool_t *timeout_pool;
2054*0Sstevel@tonic-gate 
2055*0Sstevel@tonic-gate typedef struct timeout_update {
2056*0Sstevel@tonic-gate 	pthread_mutex_t		tu_lock;
2057*0Sstevel@tonic-gate 	pthread_cond_t		tu_cv;
2058*0Sstevel@tonic-gate 	int			tu_wakeup;
2059*0Sstevel@tonic-gate } timeout_update_t;
2060*0Sstevel@tonic-gate 
2061*0Sstevel@tonic-gate timeout_update_t *tu;
2062*0Sstevel@tonic-gate 
2063*0Sstevel@tonic-gate static const char *timeout_ovr_svcs[] = {
2064*0Sstevel@tonic-gate 	"svc:/system/manifest-import:default",
2065*0Sstevel@tonic-gate 	"svc:/network/initial:default",
2066*0Sstevel@tonic-gate 	"svc:/network/service:default",
2067*0Sstevel@tonic-gate 	"svc:/system/rmtmpfiles:default",
2068*0Sstevel@tonic-gate 	"svc:/network/loopback:default",
2069*0Sstevel@tonic-gate 	"svc:/network/physical:default",
2070*0Sstevel@tonic-gate 	"svc:/system/device/local:default",
2071*0Sstevel@tonic-gate 	"svc:/system/metainit:default",
2072*0Sstevel@tonic-gate 	"svc:/system/filesystem/usr:default",
2073*0Sstevel@tonic-gate 	"svc:/system/filesystem/minimal:default",
2074*0Sstevel@tonic-gate 	"svc:/system/filesystem/local:default",
2075*0Sstevel@tonic-gate 	NULL
2076*0Sstevel@tonic-gate };
2077*0Sstevel@tonic-gate 
2078*0Sstevel@tonic-gate int
2079*0Sstevel@tonic-gate is_timeout_ovr(restarter_inst_t *inst)
2080*0Sstevel@tonic-gate {
2081*0Sstevel@tonic-gate 	int i;
2082*0Sstevel@tonic-gate 
2083*0Sstevel@tonic-gate 	for (i = 0; timeout_ovr_svcs[i] != NULL; ++i) {
2084*0Sstevel@tonic-gate 		if (strcmp(inst->ri_i.i_fmri, timeout_ovr_svcs[i]) == 0) {
2085*0Sstevel@tonic-gate 			log_instance(inst, B_TRUE, "Timeout override by "
2086*0Sstevel@tonic-gate 			    "svc.startd.  Using infinite timeout");
2087*0Sstevel@tonic-gate 			return (1);
2088*0Sstevel@tonic-gate 		}
2089*0Sstevel@tonic-gate 	}
2090*0Sstevel@tonic-gate 
2091*0Sstevel@tonic-gate 	return (0);
2092*0Sstevel@tonic-gate }
2093*0Sstevel@tonic-gate 
2094*0Sstevel@tonic-gate /*ARGSUSED*/
2095*0Sstevel@tonic-gate static int
2096*0Sstevel@tonic-gate timeout_compare(const void *lc_arg, const void *rc_arg, void *private)
2097*0Sstevel@tonic-gate {
2098*0Sstevel@tonic-gate 	hrtime_t t1 = ((const timeout_entry_t *)lc_arg)->te_timeout;
2099*0Sstevel@tonic-gate 	hrtime_t t2 = ((const timeout_entry_t *)rc_arg)->te_timeout;
2100*0Sstevel@tonic-gate 
2101*0Sstevel@tonic-gate 	if (t1 > t2)
2102*0Sstevel@tonic-gate 		return (1);
2103*0Sstevel@tonic-gate 	else if (t1 < t2)
2104*0Sstevel@tonic-gate 		return (-1);
2105*0Sstevel@tonic-gate 	return (0);
2106*0Sstevel@tonic-gate }
2107*0Sstevel@tonic-gate 
2108*0Sstevel@tonic-gate void
2109*0Sstevel@tonic-gate timeout_init()
2110*0Sstevel@tonic-gate {
2111*0Sstevel@tonic-gate 	timeouts = startd_zalloc(sizeof (timeout_queue_t));
2112*0Sstevel@tonic-gate 
2113*0Sstevel@tonic-gate 	(void) pthread_mutex_init(&timeouts->tq_lock, &mutex_attrs);
2114*0Sstevel@tonic-gate 
2115*0Sstevel@tonic-gate 	timeout_pool = startd_list_pool_create("timeouts",
2116*0Sstevel@tonic-gate 	    sizeof (timeout_entry_t), offsetof(timeout_entry_t, te_link),
2117*0Sstevel@tonic-gate 	    timeout_compare, UU_LIST_POOL_DEBUG);
2118*0Sstevel@tonic-gate 	assert(timeout_pool != NULL);
2119*0Sstevel@tonic-gate 
2120*0Sstevel@tonic-gate 	timeouts->tq_list = startd_list_create(timeout_pool,
2121*0Sstevel@tonic-gate 	    timeouts, UU_LIST_SORTED);
2122*0Sstevel@tonic-gate 	assert(timeouts->tq_list != NULL);
2123*0Sstevel@tonic-gate 
2124*0Sstevel@tonic-gate 	tu = startd_zalloc(sizeof (timeout_update_t));
2125*0Sstevel@tonic-gate 	(void) pthread_cond_init(&tu->tu_cv, NULL);
2126*0Sstevel@tonic-gate 	(void) pthread_mutex_init(&tu->tu_lock, &mutex_attrs);
2127*0Sstevel@tonic-gate }
2128*0Sstevel@tonic-gate 
2129*0Sstevel@tonic-gate void
2130*0Sstevel@tonic-gate timeout_insert(restarter_inst_t *inst, ctid_t cid, uint64_t timeout_sec)
2131*0Sstevel@tonic-gate {
2132*0Sstevel@tonic-gate 	hrtime_t now, timeout;
2133*0Sstevel@tonic-gate 	timeout_entry_t *entry;
2134*0Sstevel@tonic-gate 	uu_list_index_t idx;
2135*0Sstevel@tonic-gate 
2136*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&inst->ri_lock));
2137*0Sstevel@tonic-gate 
2138*0Sstevel@tonic-gate 	now = gethrtime();
2139*0Sstevel@tonic-gate 
2140*0Sstevel@tonic-gate 	/*
2141*0Sstevel@tonic-gate 	 * If we overflow LLONG_MAX, we're never timing out anyways, so
2142*0Sstevel@tonic-gate 	 * just return.
2143*0Sstevel@tonic-gate 	 */
2144*0Sstevel@tonic-gate 	if (timeout_sec >= (LLONG_MAX - now) / 1000000000LL) {
2145*0Sstevel@tonic-gate 		log_instance(inst, B_TRUE, "timeout_seconds too large, "
2146*0Sstevel@tonic-gate 		    "treating as infinite.");
2147*0Sstevel@tonic-gate 		return;
2148*0Sstevel@tonic-gate 	}
2149*0Sstevel@tonic-gate 
2150*0Sstevel@tonic-gate 	/* hrtime is in nanoseconds. Convert timeout_sec. */
2151*0Sstevel@tonic-gate 	timeout = now + (timeout_sec * 1000000000LL);
2152*0Sstevel@tonic-gate 
2153*0Sstevel@tonic-gate 	entry = startd_alloc(sizeof (timeout_entry_t));
2154*0Sstevel@tonic-gate 	entry->te_timeout = timeout;
2155*0Sstevel@tonic-gate 	entry->te_ctid = cid;
2156*0Sstevel@tonic-gate 	entry->te_fmri = safe_strdup(inst->ri_i.i_fmri);
2157*0Sstevel@tonic-gate 	entry->te_logstem = safe_strdup(inst->ri_logstem);
2158*0Sstevel@tonic-gate 	entry->te_fired = 0;
2159*0Sstevel@tonic-gate 	/* Insert the calculated timeout time onto the queue. */
2160*0Sstevel@tonic-gate 	MUTEX_LOCK(&timeouts->tq_lock);
2161*0Sstevel@tonic-gate 	(void) uu_list_find(timeouts->tq_list, entry, NULL, &idx);
2162*0Sstevel@tonic-gate 	uu_list_node_init(entry, &entry->te_link, timeout_pool);
2163*0Sstevel@tonic-gate 	uu_list_insert(timeouts->tq_list, entry, idx);
2164*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&timeouts->tq_lock);
2165*0Sstevel@tonic-gate 
2166*0Sstevel@tonic-gate 	assert(inst->ri_timeout == NULL);
2167*0Sstevel@tonic-gate 	inst->ri_timeout = entry;
2168*0Sstevel@tonic-gate 
2169*0Sstevel@tonic-gate 	MUTEX_LOCK(&tu->tu_lock);
2170*0Sstevel@tonic-gate 	tu->tu_wakeup = 1;
2171*0Sstevel@tonic-gate 	(void) pthread_cond_broadcast(&tu->tu_cv);
2172*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&tu->tu_lock);
2173*0Sstevel@tonic-gate }
2174*0Sstevel@tonic-gate 
2175*0Sstevel@tonic-gate 
2176*0Sstevel@tonic-gate void
2177*0Sstevel@tonic-gate timeout_remove(restarter_inst_t *inst, ctid_t cid)
2178*0Sstevel@tonic-gate {
2179*0Sstevel@tonic-gate 	assert(PTHREAD_MUTEX_HELD(&inst->ri_lock));
2180*0Sstevel@tonic-gate 
2181*0Sstevel@tonic-gate 	if (inst->ri_timeout == NULL)
2182*0Sstevel@tonic-gate 		return;
2183*0Sstevel@tonic-gate 
2184*0Sstevel@tonic-gate 	assert(inst->ri_timeout->te_ctid == cid);
2185*0Sstevel@tonic-gate 
2186*0Sstevel@tonic-gate 	MUTEX_LOCK(&timeouts->tq_lock);
2187*0Sstevel@tonic-gate 	uu_list_remove(timeouts->tq_list, inst->ri_timeout);
2188*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&timeouts->tq_lock);
2189*0Sstevel@tonic-gate 
2190*0Sstevel@tonic-gate 	free(inst->ri_timeout->te_fmri);
2191*0Sstevel@tonic-gate 	free(inst->ri_timeout->te_logstem);
2192*0Sstevel@tonic-gate 	startd_free(inst->ri_timeout, sizeof (timeout_entry_t));
2193*0Sstevel@tonic-gate 	inst->ri_timeout = NULL;
2194*0Sstevel@tonic-gate }
2195*0Sstevel@tonic-gate 
2196*0Sstevel@tonic-gate static int
2197*0Sstevel@tonic-gate timeout_now()
2198*0Sstevel@tonic-gate {
2199*0Sstevel@tonic-gate 	timeout_entry_t *e;
2200*0Sstevel@tonic-gate 	hrtime_t now;
2201*0Sstevel@tonic-gate 	int ret;
2202*0Sstevel@tonic-gate 
2203*0Sstevel@tonic-gate 	now = gethrtime();
2204*0Sstevel@tonic-gate 
2205*0Sstevel@tonic-gate 	/*
2206*0Sstevel@tonic-gate 	 * Walk through the (sorted) timeouts list.  While the timeout
2207*0Sstevel@tonic-gate 	 * at the head of the list is <= the current time, kill the
2208*0Sstevel@tonic-gate 	 * method.
2209*0Sstevel@tonic-gate 	 */
2210*0Sstevel@tonic-gate 	MUTEX_LOCK(&timeouts->tq_lock);
2211*0Sstevel@tonic-gate 
2212*0Sstevel@tonic-gate 	for (e = uu_list_first(timeouts->tq_list);
2213*0Sstevel@tonic-gate 	    e != NULL && e->te_timeout <= now;
2214*0Sstevel@tonic-gate 	    e = uu_list_next(timeouts->tq_list, e)) {
2215*0Sstevel@tonic-gate 		log_framework(LOG_WARNING, "%s: Method or service exit timed "
2216*0Sstevel@tonic-gate 		    "out.  Killing contract %ld.\n", e->te_fmri, e->te_ctid);
2217*0Sstevel@tonic-gate 		log_instance_fmri(e->te_fmri, e->te_logstem, B_TRUE,
2218*0Sstevel@tonic-gate 		    "Method or service exit timed out.  Killing contract %ld",
2219*0Sstevel@tonic-gate 		    e->te_ctid);
2220*0Sstevel@tonic-gate 		e->te_fired = 1;
2221*0Sstevel@tonic-gate 		(void) contract_kill(e->te_ctid, SIGKILL, e->te_fmri);
2222*0Sstevel@tonic-gate 	}
2223*0Sstevel@tonic-gate 
2224*0Sstevel@tonic-gate 	if (uu_list_numnodes(timeouts->tq_list) > 0)
2225*0Sstevel@tonic-gate 		ret = 0;
2226*0Sstevel@tonic-gate 	else
2227*0Sstevel@tonic-gate 		ret = -1;
2228*0Sstevel@tonic-gate 
2229*0Sstevel@tonic-gate 	MUTEX_UNLOCK(&timeouts->tq_lock);
2230*0Sstevel@tonic-gate 
2231*0Sstevel@tonic-gate 	return (ret);
2232*0Sstevel@tonic-gate }
2233*0Sstevel@tonic-gate 
2234*0Sstevel@tonic-gate /*
2235*0Sstevel@tonic-gate  * void *restarter_timeouts_event_thread(void *)
2236*0Sstevel@tonic-gate  *   Responsible for monitoring the method timeouts.  This thread must
2237*0Sstevel@tonic-gate  *   be started before any methods are called.
2238*0Sstevel@tonic-gate  */
2239*0Sstevel@tonic-gate /*ARGSUSED*/
2240*0Sstevel@tonic-gate static void *
2241*0Sstevel@tonic-gate restarter_timeouts_event_thread(void *unused)
2242*0Sstevel@tonic-gate {
2243*0Sstevel@tonic-gate 	/*
2244*0Sstevel@tonic-gate 	 * Timeouts are entered on a priority queue, which is processed by
2245*0Sstevel@tonic-gate 	 * this thread.  As timeouts are specified in seconds, we'll do
2246*0Sstevel@tonic-gate 	 * the necessary processing every second, as long as the queue
2247*0Sstevel@tonic-gate 	 * is not empty.
2248*0Sstevel@tonic-gate 	 */
2249*0Sstevel@tonic-gate 
2250*0Sstevel@tonic-gate 	/*CONSTCOND*/
2251*0Sstevel@tonic-gate 	while (1) {
2252*0Sstevel@tonic-gate 		/*
2253*0Sstevel@tonic-gate 		 * As long as the timeout list isn't empty, process it
2254*0Sstevel@tonic-gate 		 * every second.
2255*0Sstevel@tonic-gate 		 */
2256*0Sstevel@tonic-gate 		if (timeout_now() == 0) {
2257*0Sstevel@tonic-gate 			(void) sleep(1);
2258*0Sstevel@tonic-gate 			continue;
2259*0Sstevel@tonic-gate 		}
2260*0Sstevel@tonic-gate 
2261*0Sstevel@tonic-gate 		/* The list is empty, wait until we have more timeouts. */
2262*0Sstevel@tonic-gate 		MUTEX_LOCK(&tu->tu_lock);
2263*0Sstevel@tonic-gate 
2264*0Sstevel@tonic-gate 		while (tu->tu_wakeup == 0)
2265*0Sstevel@tonic-gate 			(void) pthread_cond_wait(&tu->tu_cv, &tu->tu_lock);
2266*0Sstevel@tonic-gate 
2267*0Sstevel@tonic-gate 		tu->tu_wakeup = 0;
2268*0Sstevel@tonic-gate 		MUTEX_UNLOCK(&tu->tu_lock);
2269*0Sstevel@tonic-gate 	}
2270*0Sstevel@tonic-gate 
2271*0Sstevel@tonic-gate 	return (NULL);
2272*0Sstevel@tonic-gate }
2273*0Sstevel@tonic-gate 
2274*0Sstevel@tonic-gate void
2275*0Sstevel@tonic-gate restarter_start()
2276*0Sstevel@tonic-gate {
2277*0Sstevel@tonic-gate 	(void) startd_thread_create(restarter_timeouts_event_thread, NULL);
2278*0Sstevel@tonic-gate 	(void) startd_thread_create(restarter_event_thread, NULL);
2279*0Sstevel@tonic-gate 	(void) startd_thread_create(restarter_contracts_event_thread, NULL);
2280*0Sstevel@tonic-gate 	(void) startd_thread_create(wait_thread, NULL);
2281*0Sstevel@tonic-gate }
2282*0Sstevel@tonic-gate 
2283*0Sstevel@tonic-gate 
2284*0Sstevel@tonic-gate void
2285*0Sstevel@tonic-gate restarter_init()
2286*0Sstevel@tonic-gate {
2287*0Sstevel@tonic-gate 	restarter_instance_pool = startd_list_pool_create("restarter_instances",
2288*0Sstevel@tonic-gate 	    sizeof (restarter_inst_t), offsetof(restarter_inst_t,
2289*0Sstevel@tonic-gate 		ri_link), restarter_instance_compare, UU_LIST_POOL_DEBUG);
2290*0Sstevel@tonic-gate 	(void) memset(&instance_list, 0, sizeof (instance_list));
2291*0Sstevel@tonic-gate 
2292*0Sstevel@tonic-gate 	(void) pthread_mutex_init(&instance_list.ril_lock, &mutex_attrs);
2293*0Sstevel@tonic-gate 	instance_list.ril_instance_list = startd_list_create(
2294*0Sstevel@tonic-gate 	    restarter_instance_pool, &instance_list, UU_LIST_SORTED);
2295*0Sstevel@tonic-gate 
2296*0Sstevel@tonic-gate 	restarter_queue_pool = startd_list_pool_create(
2297*0Sstevel@tonic-gate 	    "restarter_instance_queue", sizeof (restarter_instance_qentry_t),
2298*0Sstevel@tonic-gate 	    offsetof(restarter_instance_qentry_t,  riq_link), NULL,
2299*0Sstevel@tonic-gate 	    UU_LIST_POOL_DEBUG);
2300*0Sstevel@tonic-gate 
2301*0Sstevel@tonic-gate 	contract_list_pool = startd_list_pool_create(
2302*0Sstevel@tonic-gate 	    "contract_list", sizeof (contract_entry_t),
2303*0Sstevel@tonic-gate 	    offsetof(contract_entry_t,  ce_link), NULL,
2304*0Sstevel@tonic-gate 	    UU_LIST_POOL_DEBUG);
2305*0Sstevel@tonic-gate 	contract_hash_init();
2306*0Sstevel@tonic-gate 
2307*0Sstevel@tonic-gate 	log_framework(LOG_DEBUG, "Initialized restarter\n");
2308*0Sstevel@tonic-gate }
2309