1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * NOTES: To be expanded.
31  *
32  * The SMF inetd.
33  *
34  * Below are some high level notes of the operation of the SMF inetd. The
35  * notes don't go into any real detail, and the viewer of this file is
36  * encouraged to look at the code and its associated comments to better
37  * understand inetd's operation. This saves the potential for the code
38  * and these notes diverging over time.
39  *
40  * Inetd's major work is done from the context of event_loop(). Within this
41  * loop, inetd polls for events arriving from a number of different file
42  * descriptors, representing the following event types, and initiates
43  * any necessary event processing:
44  * - incoming network connections/datagrams.
45  * - notification of terminated processes (discovered via contract events).
46  * - instance specific events originating from the SMF master restarter.
47  * - stop/refresh requests from the inetd method processes (coming in on a
48  *   Unix Domain socket).
49  * There's also a timeout set for the poll, which is set to the nearest
50  * scheduled timer in a timer queue that inetd uses to perform delayed
51  * processing, such as bind retries.
52  * The SIGHUP and SIGINT signals can also interrupt the poll, and will
53  * result in inetd being refreshed or stopped respectively, as was the
54  * behavior with the old inetd.
55  *
56  * Inetd implements a state machine for each instance. The states within the
57  * machine are: offline, online, disabled, maintenance, uninitialized and
58  * specializations of the offline state for when an instance exceeds one of
59  * its DOS limits. The state of an instance can be changed as a
60  * result/side-effect of one of the above events occurring, or inetd being
61  * started up. The ongoing state of an instance is stored in the SMF
62  * repository, as required of SMF restarters. This enables an administrator
63  * to view the state of each instance, and, if inetd was to terminate
64  * unexpectedly, it could use the stored state to re-commence where it left off.
65  *
66  * Within the state machine a number of methods are run (if provided) as part
67  * of a state transition to aid/ effect a change in an instance's state. The
68  * supported methods are: offline, online, disable, refresh and start. The
69  * latter of these is the equivalent of the server program and its arguments
70  * in the old inetd.
71  *
72  * Events from the SMF master restarter come in on a number of threads
73  * created in the registration routine of librestart, the delegated restarter
74  * library. These threads call into the restart_event_proxy() function
75  * when an event arrives. To serialize the processing of instances, these events
76  * are then written down a pipe to the process's main thread, which listens
77  * for these events via a poll call, with the file descriptor of the other
78  * end of the pipe in its read set, and processes the event appropriately.
79  * When the event has been  processed (which may be delayed if the instance
80  * for which the event is for is in the process of executing one of its methods
81  * as part of a state transition) it writes an acknowledgement back down the
82  * pipe the event was received on. The thread in restart_event_proxy() that
83  * wrote the event will read the acknowledgement it was blocked upon, and will
84  * then be able to return to its caller, thus implicitly acknowledging the
85  * event, and allowing another event to be written down the pipe for the main
86  * thread to process.
87  */
88 
89 
90 #include <netdb.h>
91 #include <stdio.h>
92 #include <stdlib.h>
93 #include <strings.h>
94 #include <unistd.h>
95 #include <assert.h>
96 #include <sys/types.h>
97 #include <sys/socket.h>
98 #include <netinet/in.h>
99 #include <fcntl.h>
100 #include <signal.h>
101 #include <errno.h>
102 #include <locale.h>
103 #include <syslog.h>
104 #include <libintl.h>
105 #include <librestart.h>
106 #include <pthread.h>
107 #include <sys/stat.h>
108 #include <time.h>
109 #include <limits.h>
110 #include <libgen.h>
111 #include <tcpd.h>
112 #include <libscf.h>
113 #include <libuutil.h>
114 #include <stddef.h>
115 #include <bsm/adt_event.h>
116 #include "inetd_impl.h"
117 
118 /* path to inetd's binary */
119 #define	INETD_PATH	"/usr/lib/inet/inetd"
120 
121 /*
122  * inetd's default configuration file paths. /etc/inetd/inetd.conf is set
123  * be be the primary file, so it is checked before /etc/inetd.conf.
124  */
125 #define	PRIMARY_DEFAULT_CONF_FILE	"/etc/inet/inetd.conf"
126 #define	SECONDARY_DEFAULT_CONF_FILE	"/etc/inetd.conf"
127 
128 /* Arguments passed to this binary to request which method to execute. */
129 #define	START_METHOD_ARG	"start"
130 #define	STOP_METHOD_ARG		"stop"
131 #define	REFRESH_METHOD_ARG	"refresh"
132 
133 /* connection backlog for unix domain socket */
134 #define	UDS_BACKLOG	2
135 
136 /* number of retries to recv() a request on the UDS socket before giving up */
137 #define	UDS_RECV_RETRIES	10
138 
139 /* enumeration of the different ends of a pipe */
140 enum pipe_end {
141 	PE_CONSUMER,
142 	PE_PRODUCER
143 };
144 
145 typedef struct {
146 	internal_inst_state_t		istate;
147 	const char			*name;
148 	restarter_instance_state_t	smf_state;
149 	instance_method_t		method_running;
150 } state_info_t;
151 
152 
153 /*
154  * Collection of information for each state.
155  * NOTE:  This table is indexed into using the internal_inst_state_t
156  * enumeration, so the ordering needs to be kept in synch.
157  */
158 static state_info_t states[] = {
159 	{IIS_UNINITIALIZED, "uninitialized", RESTARTER_STATE_UNINIT,
160 	    IM_NONE},
161 	{IIS_ONLINE, "online", RESTARTER_STATE_ONLINE, IM_START},
162 	{IIS_IN_ONLINE_METHOD, "online_method", RESTARTER_STATE_OFFLINE,
163 	    IM_ONLINE},
164 	{IIS_OFFLINE, "offline", RESTARTER_STATE_OFFLINE, IM_NONE},
165 	{IIS_IN_OFFLINE_METHOD, "offline_method", RESTARTER_STATE_OFFLINE,
166 	    IM_OFFLINE},
167 	{IIS_DISABLED, "disabled", RESTARTER_STATE_DISABLED, IM_NONE},
168 	{IIS_IN_DISABLE_METHOD, "disabled_method", RESTARTER_STATE_OFFLINE,
169 	    IM_DISABLE},
170 	{IIS_IN_REFRESH_METHOD, "refresh_method", RESTARTER_STATE_ONLINE,
171 	    IM_REFRESH},
172 	{IIS_MAINTENANCE, "maintenance", RESTARTER_STATE_MAINT, IM_NONE},
173 	{IIS_OFFLINE_CONRATE, "cr_offline", RESTARTER_STATE_OFFLINE, IM_NONE},
174 	{IIS_OFFLINE_BIND, "bind_offline", RESTARTER_STATE_OFFLINE, IM_NONE},
175 	{IIS_OFFLINE_COPIES, "copies_offline", RESTARTER_STATE_OFFLINE,
176 	    IM_NONE},
177 	{IIS_DEGRADED, "degraded", RESTARTER_STATE_DEGRADED, IM_NONE},
178 	{IIS_NONE, "none", RESTARTER_STATE_NONE, IM_NONE}
179 };
180 
181 /*
182  * Pipe used to send events from the threads created by restarter_bind_handle()
183  * to the main thread of control.
184  */
185 static int			rst_event_pipe[] = {-1, -1};
186 /*
187  * Used to protect the critical section of code in restarter_event_proxy() that
188  * involves writing an event down the event pipe and reading an acknowledgement.
189  */
190 static pthread_mutex_t		rst_event_pipe_mtx = PTHREAD_MUTEX_INITIALIZER;
191 
192 /* handle used in communication with the master restarter */
193 static restarter_event_handle_t *rst_event_handle = NULL;
194 
195 /* set to indicate a refresh of inetd is requested */
196 static boolean_t		refresh_inetd_requested = B_FALSE;
197 
198 /* set by the SIGTERM handler to flag we got a SIGTERM */
199 static boolean_t		got_sigterm = B_FALSE;
200 
201 /*
202  * Timer queue used to store timers for delayed event processing, such as
203  * bind retries.
204  */
205 iu_tq_t				*timer_queue = NULL;
206 
207 /*
208  * fd of Unix Domain socket used to communicate stop and refresh requests
209  * to the inetd start method process.
210  */
211 static int			uds_fd = -1;
212 
213 /*
214  * List of inetd's currently managed instances; each containing its state,
215  * and in certain states its configuration.
216  */
217 static uu_list_pool_t		*instance_pool = NULL;
218 uu_list_t			*instance_list = NULL;
219 
220 /* set to indicate we're being stopped */
221 boolean_t			inetd_stopping = B_FALSE;
222 
223 /* TCP wrappers syslog globals. Consumed by libwrap. */
224 int				allow_severity = LOG_INFO;
225 int				deny_severity = LOG_WARNING;
226 
227 /* path of the configuration file being monitored by check_conf_file() */
228 static char			*conf_file = NULL;
229 
230 /* Auditing session handle */
231 static adt_session_data_t	*audit_handle;
232 
233 static void uds_fini(void);
234 static int uds_init(void);
235 static int run_method(instance_t *, instance_method_t, const proto_info_t *);
236 static void create_bound_fds(instance_t *);
237 static void destroy_bound_fds(instance_t *);
238 static void destroy_instance(instance_t *);
239 static void inetd_stop(void);
240 static void
241 exec_method(instance_t *instance, instance_method_t method, method_info_t *mi,
242     struct method_context *mthd_ctxt, const proto_info_t *pi) __NORETURN;
243 
244 /*
245  * The following two functions are callbacks that libumem uses to determine
246  * inetd's desired debugging/logging levels. The interface they consume is
247  * exported by FMA and is consolidation private. The comments in the two
248  * functions give the environment variable that will effectively be set to
249  * their returned value, and thus whose behavior for this value, described in
250  * umem_debug(3MALLOC), will be followed.
251  */
252 
253 const char *
254 _umem_debug_init(void)
255 {
256 	return ("default,verbose");	/* UMEM_DEBUG setting */
257 }
258 
259 const char *
260 _umem_logging_init(void)
261 {
262 	return ("fail,contents");	/* UMEM_LOGGING setting */
263 }
264 
265 static void
266 log_invalid_cfg(const char *fmri)
267 {
268 	error_msg(gettext(
269 	    "Invalid configuration for instance %s, placing in maintenance"),
270 	    fmri);
271 }
272 
273 /*
274  * Returns B_TRUE if the instance is in a suitable state for inetd to stop.
275  */
276 static boolean_t
277 instance_stopped(const instance_t *inst)
278 {
279 	return ((inst->cur_istate == IIS_OFFLINE) ||
280 	    (inst->cur_istate == IIS_MAINTENANCE) ||
281 	    (inst->cur_istate == IIS_DISABLED) ||
282 	    (inst->cur_istate == IIS_UNINITIALIZED));
283 }
284 
285 /*
286  * Updates the current and next repository states of instance 'inst'. If
287  * any errors occur an error message is output.
288  */
289 static void
290 update_instance_states(instance_t *inst, internal_inst_state_t new_cur_state,
291     internal_inst_state_t new_next_state, restarter_error_t err)
292 {
293 	internal_inst_state_t	old_cur = inst->cur_istate;
294 	internal_inst_state_t	old_next = inst->next_istate;
295 	scf_error_t		sret;
296 	int			ret;
297 
298 	debug_msg("Entering update_instance_states: oldcur: %s, newcur: %s "
299 	    "oldnext: %s, newnext: %s", states[old_cur].name,
300 	    states[new_cur_state].name, states[old_next].name,
301 	    states[new_next_state].name);
302 
303 
304 	/* update the repository/cached internal state */
305 	inst->cur_istate = new_cur_state;
306 	inst->next_istate = new_next_state;
307 	(void) set_single_rep_val(inst->cur_istate_rep,
308 	    (int64_t)new_cur_state);
309 	(void) set_single_rep_val(inst->next_istate_rep,
310 	    (int64_t)new_next_state);
311 
312 	if (((sret = store_rep_vals(inst->cur_istate_rep, inst->fmri,
313 	    PR_NAME_CUR_INT_STATE)) != 0) ||
314 	    ((sret = store_rep_vals(inst->next_istate_rep, inst->fmri,
315 	    PR_NAME_NEXT_INT_STATE)) != 0))
316 		error_msg(gettext("Failed to update state of instance %s in "
317 		    "repository: %s"), inst->fmri, scf_strerror(sret));
318 
319 	/* update the repository SMF state */
320 	if ((ret = restarter_set_states(rst_event_handle, inst->fmri,
321 	    states[old_cur].smf_state, states[new_cur_state].smf_state,
322 	    states[old_next].smf_state, states[new_next_state].smf_state,
323 	    err, 0)) != 0)
324 		error_msg(gettext("Failed to update state of instance %s in "
325 		    "repository: %s"), inst->fmri, strerror(ret));
326 
327 }
328 
329 void
330 update_state(instance_t *inst, internal_inst_state_t new_cur,
331     restarter_error_t err)
332 {
333 	update_instance_states(inst, new_cur, IIS_NONE, err);
334 }
335 
336 /*
337  * Sends a refresh event to the inetd start method process and returns
338  * SMF_EXIT_OK if it managed to send it. If it fails to send the request for
339  * some reason it returns SMF_EXIT_ERR_OTHER.
340  */
341 static int
342 refresh_method(void)
343 {
344 	uds_request_t   req = UR_REFRESH_INETD;
345 	int		fd;
346 
347 	debug_msg("Entering refresh_method");
348 
349 	if ((fd = connect_to_inetd()) < 0) {
350 		error_msg(gettext("Failed to connect to inetd: %s"),
351 		    strerror(errno));
352 		return (SMF_EXIT_ERR_OTHER);
353 	}
354 
355 	/* write the request and return success */
356 	if (safe_write(fd, &req, sizeof (req)) == -1) {
357 		error_msg(
358 		    gettext("Failed to send refresh request to inetd: %s"),
359 		    strerror(errno));
360 		(void) close(fd);
361 		return (SMF_EXIT_ERR_OTHER);
362 	}
363 
364 	(void) close(fd);
365 
366 	return (SMF_EXIT_OK);
367 }
368 
369 /*
370  * Sends a stop event to the inetd start method process and wait till it goes
371  * away. If inetd is determined to have stopped SMF_EXIT_OK is returned, else
372  * SMF_EXIT_ERR_OTHER is returned.
373  */
374 static int
375 stop_method(void)
376 {
377 	uds_request_t   req = UR_STOP_INETD;
378 	int		fd;
379 	char		c;
380 	ssize_t		ret;
381 
382 	debug_msg("Entering stop_method");
383 
384 	if ((fd = connect_to_inetd()) == -1) {
385 		debug_msg(gettext("Failed to connect to inetd: %s"),
386 		    strerror(errno));
387 		/*
388 		 * Assume connect_to_inetd() failed because inetd was already
389 		 * stopped, and return success.
390 		 */
391 		return (SMF_EXIT_OK);
392 	}
393 
394 	/*
395 	 * This is safe to do since we're fired off in a separate process
396 	 * than inetd and in the case we get wedged, the stop method timeout
397 	 * will occur and we'd be killed by our restarter.
398 	 */
399 	enable_blocking(fd);
400 
401 	/* write the stop request to inetd and wait till it goes away */
402 	if (safe_write(fd, &req, sizeof (req)) != 0) {
403 		error_msg(gettext("Failed to send stop request to inetd"));
404 		(void) close(fd);
405 		return (SMF_EXIT_ERR_OTHER);
406 	}
407 
408 	/* wait until remote end of socket is closed */
409 	while (((ret = recv(fd, &c, sizeof (c), 0)) != 0) && (errno == EINTR))
410 		;
411 
412 	(void) close(fd);
413 
414 	if (ret != 0) {
415 		error_msg(gettext("Failed to determine whether inetd stopped"));
416 		return (SMF_EXIT_ERR_OTHER);
417 	}
418 
419 	return (SMF_EXIT_OK);
420 }
421 
422 
423 /*
424  * This function is called to handle restarter events coming in from the
425  * master restarter. It is registered with the master restarter via
426  * restarter_bind_handle() and simply passes a pointer to the event down
427  * the event pipe, which will be discovered by the poll in the event loop
428  * and processed there. It waits for an acknowledgement to be written back down
429  * the pipe before returning.
430  * Writing a pointer to the function's 'event' parameter down the pipe will
431  * be safe, as the thread in restarter_event_proxy() doesn't return until
432  * the main thread has finished its processing of the passed event, thus
433  * the referenced event will remain around until the function returns.
434  * To impose the limit of only one event being in the pipe and processed
435  * at once, a lock is taken on entry to this function and returned on exit.
436  * Always returns 0.
437  */
438 static int
439 restarter_event_proxy(restarter_event_t *event)
440 {
441 	restarter_event_type_t  ev_type;
442 	boolean_t		processed;
443 
444 	debug_msg("Entering restarter_event_proxy");
445 	ev_type = restarter_event_get_type(event);
446 	debug_msg("event: %x, event type: %d", event, ev_type);
447 
448 	(void) pthread_mutex_lock(&rst_event_pipe_mtx);
449 
450 	/* write the event to the main worker thread down the pipe */
451 	if (safe_write(rst_event_pipe[PE_PRODUCER], &event,
452 	    sizeof (event)) != 0)
453 		goto pipe_error;
454 
455 	/*
456 	 * Wait for an acknowledgement that the event has been processed from
457 	 * the same pipe. In the case that inetd is stopping, any thread in
458 	 * this function will simply block on this read until inetd eventually
459 	 * exits. This will result in this function not returning success to
460 	 * its caller, and the event that was being processed when the
461 	 * function exited will be re-sent when inetd is next started.
462 	 */
463 	if (safe_read(rst_event_pipe[PE_PRODUCER], &processed,
464 	    sizeof (processed)) != 0)
465 		goto pipe_error;
466 
467 	(void) pthread_mutex_unlock(&rst_event_pipe_mtx);
468 
469 	return (processed ? 0 : EAGAIN);
470 
471 pipe_error:
472 	/*
473 	 * Something's seriously wrong with the event pipe. Notify the
474 	 * worker thread by closing this end of the event pipe and pause till
475 	 * inetd exits.
476 	 */
477 	error_msg(gettext("Can't process restarter events: %s"),
478 	    strerror(errno));
479 	(void) close(rst_event_pipe[PE_PRODUCER]);
480 	for (;;)
481 		(void) pause();
482 
483 	/* NOTREACHED */
484 }
485 
486 /*
487  * Let restarter_event_proxy() know we're finished with the event it's blocked
488  * upon. The 'processed' argument denotes whether we successfully processed the
489  * event.
490  */
491 static void
492 ack_restarter_event(boolean_t processed)
493 {
494 	debug_msg("Entering ack_restarter_event");
495 
496 	/*
497 	 * If safe_write returns -1 something's seriously wrong with the event
498 	 * pipe, so start the shutdown proceedings.
499 	 */
500 	if (safe_write(rst_event_pipe[PE_CONSUMER], &processed,
501 	    sizeof (processed)) == -1)
502 		inetd_stop();
503 }
504 
505 /*
506  * Switch the syslog identification string to 'ident'.
507  */
508 static void
509 change_syslog_ident(const char *ident)
510 {
511 	debug_msg("Entering change_syslog_ident: ident: %s", ident);
512 
513 	closelog();
514 	openlog(ident, LOG_PID|LOG_CONS, LOG_DAEMON);
515 }
516 
517 /*
518  * Perform TCP wrappers checks on this instance. Due to the fact that the
519  * current wrappers code used in Solaris is taken untouched from the open
520  * source version, we're stuck with using the daemon name for the checks, as
521  * opposed to making use of instance FMRIs. Sigh.
522  * Returns B_TRUE if the check passed, else B_FALSE.
523  */
524 static boolean_t
525 tcp_wrappers_ok(instance_t *instance)
526 {
527 	boolean_t		rval = B_TRUE;
528 	char			*daemon_name;
529 	basic_cfg_t		*cfg = instance->config->basic;
530 	struct request_info	req;
531 
532 	debug_msg("Entering tcp_wrappers_ok, instance: %s", instance->fmri);
533 
534 	/*
535 	 * Wrap the service using libwrap functions. The code below implements
536 	 * the functionality of tcpd. This is done only for stream,nowait
537 	 * services, following the convention of other vendors.  udp/dgram and
538 	 * stream/wait can NOT be wrapped with this libwrap, so be wary of
539 	 * changing the test below.
540 	 */
541 	if (cfg->do_tcp_wrappers && !cfg->iswait && !cfg->istlx) {
542 
543 		daemon_name = instance->config->methods[
544 		    IM_START]->exec_args_we.we_wordv[0];
545 		if (*daemon_name == '/')
546 			daemon_name = strrchr(daemon_name, '/') + 1;
547 
548 		/*
549 		 * Change the syslog message identity to the name of the
550 		 * daemon being wrapped, as opposed to "inetd".
551 		 */
552 		change_syslog_ident(daemon_name);
553 
554 		(void) request_init(&req, RQ_DAEMON, daemon_name, RQ_FILE,
555 		    instance->conn_fd, NULL);
556 		fromhost(&req);
557 
558 		if (strcasecmp(eval_hostname(req.client), paranoid) == 0) {
559 			syslog(deny_severity,
560 			    "refused connect from %s (name/address mismatch)",
561 			    eval_client(&req));
562 			if (req.sink != NULL)
563 				req.sink(instance->conn_fd);
564 			rval = B_FALSE;
565 		} else if (!hosts_access(&req)) {
566 			syslog(deny_severity,
567 			    "refused connect from %s (access denied)",
568 			    eval_client(&req));
569 			if (req.sink != NULL)
570 				req.sink(instance->conn_fd);
571 			rval = B_FALSE;
572 		} else {
573 			syslog(allow_severity, "connect from %s",
574 			    eval_client(&req));
575 		}
576 
577 		/* Revert syslog identity back to "inetd". */
578 		change_syslog_ident(SYSLOG_IDENT);
579 	}
580 	return (rval);
581 }
582 
583 /*
584  * Handler registered with the timer queue code to remove an instance from
585  * the connection rate offline state when it has been there for its allotted
586  * time.
587  */
588 /* ARGSUSED */
589 static void
590 conn_rate_online(iu_tq_t *tq, void *arg)
591 {
592 	instance_t *instance = arg;
593 
594 	debug_msg("Entering conn_rate_online, instance: %s",
595 	    instance->fmri);
596 
597 	assert(instance->cur_istate == IIS_OFFLINE_CONRATE);
598 	instance->timer_id = -1;
599 	update_state(instance, IIS_OFFLINE, RERR_RESTART);
600 	process_offline_inst(instance);
601 }
602 
603 /*
604  * Check whether this instance in the offline state is in transition to
605  * another state and do the work to continue this transition.
606  */
607 void
608 process_offline_inst(instance_t *inst)
609 {
610 	debug_msg("Entering process_offline_inst");
611 
612 	if (inst->disable_req) {
613 		inst->disable_req = B_FALSE;
614 		(void) run_method(inst, IM_DISABLE, NULL);
615 	} else if (inst->maintenance_req) {
616 		inst->maintenance_req = B_FALSE;
617 		update_state(inst, IIS_MAINTENANCE, RERR_RESTART);
618 	/*
619 	 * If inetd is in the process of stopping, we don't want to enter
620 	 * any states but offline, disabled and maintenance.
621 	 */
622 	} else if (!inetd_stopping) {
623 		if (inst->conn_rate_exceeded) {
624 			basic_cfg_t *cfg = inst->config->basic;
625 
626 			inst->conn_rate_exceeded = B_FALSE;
627 			update_state(inst, IIS_OFFLINE_CONRATE, RERR_RESTART);
628 			/*
629 			 * Schedule a timer to bring the instance out of the
630 			 * connection rate offline state.
631 			 */
632 			inst->timer_id = iu_schedule_timer(timer_queue,
633 			    cfg->conn_rate_offline, conn_rate_online,
634 			    inst);
635 			if (inst->timer_id == -1) {
636 				error_msg(gettext("%s unable to set timer, "
637 				    "won't be brought on line after %d "
638 				    "seconds."), inst->fmri,
639 				    cfg->conn_rate_offline);
640 			}
641 
642 		} else if (copies_limit_exceeded(inst)) {
643 			update_state(inst, IIS_OFFLINE_COPIES, RERR_RESTART);
644 		}
645 	}
646 }
647 
648 /*
649  * Create a socket bound to the instance's configured address. If the
650  * bind fails, returns -1, else the fd of the bound socket.
651  */
652 static int
653 create_bound_socket(const char *fmri, socket_info_t *sock_info)
654 {
655 	int		fd;
656 	int		on = 1;
657 	rpc_info_t	*rpc = sock_info->pr_info.ri;
658 	const char	*proto = sock_info->pr_info.proto;
659 
660 	debug_msg("Entering create_bound_socket");
661 
662 	fd = socket(sock_info->local_addr.ss_family, sock_info->type,
663 	    sock_info->protocol);
664 	if (fd < 0) {
665 		error_msg(gettext(
666 		    "Socket creation failure for instance %s, proto %s: %s"),
667 		    fmri, proto, strerror(errno));
668 		return (-1);
669 	}
670 
671 	if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof (on)) == -1) {
672 		error_msg(gettext("setsockopt SO_REUSEADDR failed for service "
673 		    "instance %s, proto %s: %s"), fmri, proto, strerror(errno));
674 		(void) close(fd);
675 		return (-1);
676 	}
677 	if (sock_info->pr_info.v6only) {
678 		/* restrict socket to IPv6 communications only */
679 		if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &on,
680 		    sizeof (on)) == -1) {
681 			error_msg(gettext("setsockopt IPV6_V6ONLY failed for "
682 			    "service instance %s, proto %s: %s"), fmri, proto,
683 			    strerror(errno));
684 			(void) close(fd);
685 			return (-1);
686 		}
687 	}
688 
689 	if (rpc != NULL)
690 		SS_SETPORT(sock_info->local_addr, 0);
691 
692 	if (bind(fd, (struct sockaddr *)&(sock_info->local_addr),
693 	    SS_ADDRLEN(sock_info->local_addr)) < 0) {
694 		error_msg(gettext(
695 		    "Failed to bind to the port of service instance %s, "
696 		    "proto %s: %s"), fmri, proto, strerror(errno));
697 		(void) close(fd);
698 		return (-1);
699 	}
700 
701 	/*
702 	 * Retrieve and store the address bound to for RPC services.
703 	 */
704 	if (rpc != NULL) {
705 		struct sockaddr_storage	ss;
706 		int			ss_size = sizeof (ss);
707 
708 		if (getsockname(fd, (struct sockaddr *)&ss, &ss_size) < 0) {
709 			error_msg(gettext("Failed getsockname for instance %s, "
710 			    "proto %s: %s"), fmri, proto, strerror(errno));
711 			(void) close(fd);
712 			return (-1);
713 		}
714 		(void) memcpy(rpc->netbuf.buf, &ss,
715 		    sizeof (struct sockaddr_storage));
716 		rpc->netbuf.len = SS_ADDRLEN(ss);
717 		rpc->netbuf.maxlen = SS_ADDRLEN(ss);
718 	}
719 
720 	if (sock_info->type == SOCK_STREAM)
721 		(void) listen(fd, CONNECTION_BACKLOG);
722 
723 	return (fd);
724 }
725 
726 /*
727  * Handler registered with the timer queue code to retry the creation
728  * of a bound fd.
729  */
730 /* ARGSUSED */
731 static void
732 retry_bind(iu_tq_t *tq, void *arg)
733 {
734 	instance_t *instance = arg;
735 
736 	debug_msg("Entering retry_bind, instance: %s", instance->fmri);
737 
738 	switch (instance->cur_istate) {
739 	case IIS_OFFLINE_BIND:
740 	case IIS_ONLINE:
741 	case IIS_DEGRADED:
742 	case IIS_IN_ONLINE_METHOD:
743 	case IIS_IN_REFRESH_METHOD:
744 		break;
745 	default:
746 #ifndef NDEBUG
747 		(void) fprintf(stderr, "%s:%d: Unknown instance state %d.\n",
748 		    __FILE__, __LINE__, instance->cur_istate);
749 #endif
750 		abort();
751 	}
752 
753 	instance->bind_timer_id = -1;
754 	create_bound_fds(instance);
755 }
756 
757 /*
758  * For each of the fds for the given instance that are bound, if 'listen' is
759  * set add them to the poll set, else remove them from it. If any additions
760  * fail, returns -1, else 0 on success.
761  */
762 int
763 poll_bound_fds(instance_t *instance, boolean_t listen)
764 {
765 	basic_cfg_t	*cfg = instance->config->basic;
766 	proto_info_t	*pi;
767 	int		ret = 0;
768 
769 	debug_msg("Entering poll_bound_fds: instance: %s, on: %d",
770 	    instance->fmri, listen);
771 
772 	for (pi = uu_list_first(cfg->proto_list); pi != NULL;
773 	    pi = uu_list_next(cfg->proto_list, pi)) {
774 		if (pi->listen_fd != -1) {	/* fd bound */
775 			if (!listen) {
776 				clear_pollfd(pi->listen_fd);
777 			} else if (set_pollfd(pi->listen_fd, POLLIN) == -1) {
778 				ret = -1;
779 			}
780 		}
781 	}
782 
783 	return (ret);
784 }
785 
786 /*
787  * Handle the case were we either fail to create a bound fd or we fail
788  * to add a bound fd to the poll set for the given instance.
789  */
790 static void
791 handle_bind_failure(instance_t *instance)
792 {
793 	basic_cfg_t *cfg = instance->config->basic;
794 
795 	debug_msg("Entering handle_bind_failure: instance: %s", instance);
796 
797 	/*
798 	 * We must be being called as a result of a failed poll_bound_fds()
799 	 * as a bind retry is already scheduled. Just return and let it do
800 	 * the work.
801 	 */
802 	if (instance->bind_timer_id != -1)
803 		return;
804 
805 	/*
806 	 * Check if the rebind retries limit is operative and if so,
807 	 * if it has been reached.
808 	 */
809 	if (((cfg->bind_fail_interval <= 0) ||		/* no retries */
810 	    ((cfg->bind_fail_max >= 0) &&		/* limit reached */
811 	    (++instance->bind_fail_count > cfg->bind_fail_max))) ||
812 	    ((instance->bind_timer_id = iu_schedule_timer(timer_queue,
813 	    cfg->bind_fail_interval, retry_bind, instance)) == -1)) {
814 		proto_info_t *pi;
815 
816 		instance->bind_fail_count = 0;
817 
818 		switch (instance->cur_istate) {
819 		case IIS_DEGRADED:
820 		case IIS_ONLINE:
821 			/* check if any of the fds are being poll'd upon */
822 			for (pi = uu_list_first(cfg->proto_list); pi != NULL;
823 			    pi = uu_list_next(cfg->proto_list, pi)) {
824 				if ((pi->listen_fd != -1) &&
825 				    (find_pollfd(pi->listen_fd) != NULL))
826 					break;
827 			}
828 			if (pi != NULL)	{	/* polling on > 0 fds */
829 				warn_msg(gettext("Failed to bind on "
830 				    "all protocols for instance %s, "
831 				    "transitioning to degraded"),
832 				    instance->fmri);
833 				update_state(instance, IIS_DEGRADED, RERR_NONE);
834 				instance->bind_retries_exceeded = B_TRUE;
835 				break;
836 			}
837 
838 			destroy_bound_fds(instance);
839 			/*
840 			 * In the case we failed the 'bind' because set_pollfd()
841 			 * failed on all bound fds, use the offline handling.
842 			 */
843 			/* FALLTHROUGH */
844 		case IIS_OFFLINE:
845 		case IIS_OFFLINE_BIND:
846 			error_msg(gettext("Too many bind failures for instance "
847 			"%s, transitioning to maintenance"), instance->fmri);
848 			update_state(instance, IIS_MAINTENANCE,
849 			    RERR_FAULT);
850 			break;
851 		case IIS_IN_ONLINE_METHOD:
852 		case IIS_IN_REFRESH_METHOD:
853 			warn_msg(gettext("Failed to bind on all "
854 			    "protocols for instance %s, instance will go to "
855 			    "degraded"), instance->fmri);
856 			/*
857 			 * Set the retries exceeded flag so when the method
858 			 * completes the instance goes to the degraded state.
859 			 */
860 			instance->bind_retries_exceeded = B_TRUE;
861 			break;
862 		default:
863 #ifndef NDEBUG
864 			(void) fprintf(stderr,
865 			    "%s:%d: Unknown instance state %d.\n",
866 			    __FILE__, __LINE__, instance->cur_istate);
867 #endif
868 			abort();
869 		}
870 	} else if (instance->cur_istate == IIS_OFFLINE) {
871 		/*
872 		 * bind re-scheduled, so if we're offline reflect this in the
873 		 * state.
874 		 */
875 		update_state(instance, IIS_OFFLINE_BIND, RERR_NONE);
876 	}
877 }
878 
879 
880 /*
881  * Check if two transport protocols for RPC conflict.
882  */
883 
884 boolean_t
885 is_rpc_proto_conflict(const char *proto0, const char *proto1) {
886 	if (strcmp(proto0, "tcp") == 0) {
887 		if (strcmp(proto1, "tcp") == 0)
888 			return (B_TRUE);
889 		if (strcmp(proto1, "tcp6") == 0)
890 			return (B_TRUE);
891 		return (B_FALSE);
892 	}
893 
894 	if (strcmp(proto0, "tcp6") == 0) {
895 		if (strcmp(proto1, "tcp") == 0)
896 			return (B_TRUE);
897 		if (strcmp(proto1, "tcp6only") == 0)
898 			return (B_TRUE);
899 		if (strcmp(proto1, "tcp6") == 0)
900 			return (B_TRUE);
901 		return (B_FALSE);
902 	}
903 
904 	if (strcmp(proto0, "tcp6only") == 0) {
905 		if (strcmp(proto1, "tcp6only") == 0)
906 			return (B_TRUE);
907 		if (strcmp(proto1, "tcp6") == 0)
908 			return (B_TRUE);
909 		return (B_FALSE);
910 	}
911 
912 	if (strcmp(proto0, "udp") == 0) {
913 		if (strcmp(proto1, "udp") == 0)
914 			return (B_TRUE);
915 		if (strcmp(proto1, "udp6") == 0)
916 			return (B_TRUE);
917 		return (B_FALSE);
918 	}
919 
920 	if (strcmp(proto0, "udp6") == 0) {
921 
922 		if (strcmp(proto1, "udp") == 0)
923 			return (B_TRUE);
924 		if (strcmp(proto1, "udp6only") == 0)
925 			return (B_TRUE);
926 		if (strcmp(proto1, "udp6") == 0)
927 			return (B_TRUE);
928 		return (B_FALSE);
929 	}
930 
931 	if (strcmp(proto0, "udp6only") == 0) {
932 
933 		if (strcmp(proto1, "udp6only") == 0)
934 			return (B_TRUE);
935 		if (strcmp(proto1, "udp6") == 0)
936 			return (B_TRUE);
937 		return (0);
938 	}
939 
940 	/*
941 	 * If the protocol isn't TCP/IP or UDP/IP assume that it has its own
942 	 * port namepsace and that conflicts can be detected by literal string
943 	 * comparison.
944 	 */
945 
946 	if (strcmp(proto0, proto1))
947 		return (FALSE);
948 
949 	return (B_TRUE);
950 }
951 
952 
953 /*
954  * Check if inetd thinks this RPC program number is already registered.
955  *
956  * An RPC protocol conflict occurs if
957  * 	a) the program numbers are the same and,
958  * 	b) the version numbers overlap,
959  * 	c) the protocols (TCP vs UDP vs tic*) are the same.
960  */
961 
962 boolean_t
963 is_rpc_num_in_use(int rpc_n, char *proto, int lowver, int highver) {
964 	instance_t *i;
965 	basic_cfg_t *cfg;
966 	proto_info_t *pi;
967 
968 	for (i = uu_list_first(instance_list); i != NULL;
969 	    i = uu_list_next(instance_list, i)) {
970 
971 		cfg = i->config->basic;
972 		if (i->cur_istate != IIS_ONLINE)
973 			continue;
974 
975 		for (pi = uu_list_first(cfg->proto_list); pi != NULL;
976 		    pi = uu_list_next(cfg->proto_list, pi)) {
977 
978 			if (pi->ri == NULL)
979 				continue;
980 			if (pi->ri->prognum != rpc_n)
981 				continue;
982 			if (!is_rpc_proto_conflict(pi->proto, proto))
983 				continue;
984 			if ((lowver < pi->ri->lowver &&
985 			    highver < pi->ri->lowver) ||
986 			    (lowver > pi->ri->highver &&
987 			    highver > pi->ri->highver))
988 				continue;
989 			return (B_TRUE);
990 		}
991 	}
992 	return (B_FALSE);
993 }
994 
995 
996 /*
997  * Independent of the transport, for each of the entries in the instance's
998  * proto list this function first attempts to create an associated network fd;
999  * for RPC services these are then bound to a kernel chosen port and the
1000  * fd is registered with rpcbind; for non-RPC services the fds are bound
1001  * to the port associated with the instance's service name. On any successful
1002  * binds the instance is taken online. Failed binds are handled by
1003  * handle_bind_failure().
1004  */
1005 void
1006 create_bound_fds(instance_t *instance)
1007 {
1008 	basic_cfg_t	*cfg = instance->config->basic;
1009 	boolean_t	failure = B_FALSE;
1010 	boolean_t	success = B_FALSE;
1011 	proto_info_t	*pi;
1012 
1013 	debug_msg("Entering create_bound_fd: instance: %s", instance->fmri);
1014 
1015 	/*
1016 	 * Loop through and try and bind any unbound protos.
1017 	 */
1018 	for (pi = uu_list_first(cfg->proto_list); pi != NULL;
1019 	    pi = uu_list_next(cfg->proto_list, pi)) {
1020 		if (pi->listen_fd != -1)
1021 			continue;
1022 		if (cfg->istlx) {
1023 			pi->listen_fd = create_bound_endpoint(instance->fmri,
1024 			    (tlx_info_t *)pi);
1025 		} else {
1026 			/*
1027 			 * We cast pi to a void so we can then go on to cast
1028 			 * it to a socket_info_t without lint complaining
1029 			 * about alignment. This is done because the x86
1030 			 * version of lint thinks a lint suppression directive
1031 			 * is unnecessary and flags it as such, yet the sparc
1032 			 * version complains if it's absent.
1033 			 */
1034 			void *p = pi;
1035 			pi->listen_fd = create_bound_socket(instance->fmri,
1036 			    (socket_info_t *)p);
1037 		}
1038 		if (pi->listen_fd == -1) {
1039 			failure = B_TRUE;
1040 			continue;
1041 		}
1042 
1043 		if (pi->ri != NULL) {
1044 
1045 			/*
1046 			 * Don't register the same RPC program number twice.
1047 			 * Doing so silently discards the old service
1048 			 * without causing an error.
1049 			 */
1050 			if (is_rpc_num_in_use(pi->ri->prognum, pi->proto,
1051 				pi->ri->lowver, pi->ri->highver)) {
1052 				failure = B_TRUE;
1053 				close_net_fd(instance, pi->listen_fd);
1054 				pi->listen_fd = -1;
1055 				continue;
1056 			}
1057 
1058 			unregister_rpc_service(instance->fmri, pi->ri);
1059 			if (register_rpc_service(instance->fmri, pi->ri) ==
1060 			    -1) {
1061 				close_net_fd(instance, pi->listen_fd);
1062 				pi->listen_fd = -1;
1063 				failure = B_TRUE;
1064 				continue;
1065 			}
1066 		}
1067 
1068 		success = B_TRUE;
1069 	}
1070 
1071 	switch (instance->cur_istate) {
1072 	case IIS_OFFLINE:
1073 	case IIS_OFFLINE_BIND:
1074 		/*
1075 		 * If we've managed to bind at least one proto lets run the
1076 		 * online method, so we can start listening for it.
1077 		 */
1078 		if (success && run_method(instance, IM_ONLINE, NULL) == -1)
1079 			return;	/* instance gone to maintenance */
1080 		break;
1081 	case IIS_ONLINE:
1082 	case IIS_IN_REFRESH_METHOD:
1083 		/*
1084 		 * We're 'online', so start polling on any bound fds we're
1085 		 * currently not.
1086 		 */
1087 		if (poll_bound_fds(instance, B_TRUE) != 0) {
1088 			failure = B_TRUE;
1089 		} else if (!failure) {
1090 			/*
1091 			 * We've successfully bound and poll'd upon all protos,
1092 			 * so reset the failure count.
1093 			 */
1094 			instance->bind_fail_count = 0;
1095 		}
1096 		break;
1097 	case IIS_IN_ONLINE_METHOD:
1098 		/*
1099 		 * Nothing to do here as the method completion code will start
1100 		 * listening for any successfully bound fds.
1101 		 */
1102 		break;
1103 	default:
1104 #ifndef NDEBUG
1105 		(void) fprintf(stderr, "%s:%d: Unknown instance state %d.\n",
1106 		    __FILE__, __LINE__, instance->cur_istate);
1107 #endif
1108 		abort();
1109 	}
1110 
1111 	if (failure)
1112 		handle_bind_failure(instance);
1113 }
1114 
1115 /*
1116  * Counter to create_bound_fds(), for each of the bound network fds this
1117  * function unregisters the instance from rpcbind if it's an RPC service,
1118  * stops listening for new connections for it and then closes the listening fd.
1119  */
1120 static void
1121 destroy_bound_fds(instance_t *instance)
1122 {
1123 	basic_cfg_t	*cfg = instance->config->basic;
1124 	proto_info_t	*pi;
1125 
1126 	debug_msg("Entering destroy_bound_fds: instance: %s", instance->fmri);
1127 
1128 	for (pi = uu_list_first(cfg->proto_list); pi != NULL;
1129 	    pi = uu_list_next(cfg->proto_list, pi)) {
1130 		if (pi->listen_fd != -1) {
1131 			if (pi->ri != NULL)
1132 				unregister_rpc_service(instance->fmri, pi->ri);
1133 			clear_pollfd(pi->listen_fd);
1134 			close_net_fd(instance, pi->listen_fd);
1135 			pi->listen_fd = -1;
1136 		}
1137 	}
1138 
1139 	/* cancel any bind retries */
1140 	if (instance->bind_timer_id != -1)
1141 		cancel_bind_timer(instance);
1142 
1143 	instance->bind_retries_exceeded = B_FALSE;
1144 }
1145 
1146 /*
1147  * Perform %A address expansion and return a pointer to a static string
1148  * array containing crafted arguments. This expansion is provided for
1149  * compatibility with 4.2BSD daemons, and as such we've copied the logic of
1150  * the legacy inetd to maintain this compatibility as much as possible. This
1151  * logic is a bit scatty, but it dates back at least as far as SunOS 4.x.
1152  */
1153 static char **
1154 expand_address(instance_t *inst, const proto_info_t *pi)
1155 {
1156 	static char	addrbuf[sizeof ("ffffffff.65536")];
1157 	static char	*ret[3];
1158 	instance_cfg_t	*cfg = inst->config;
1159 	/*
1160 	 * We cast pi to a void so we can then go on to cast it to a
1161 	 * socket_info_t without lint complaining about alignment. This
1162 	 * is done because the x86 version of lint thinks a lint suppression
1163 	 * directive is unnecessary and flags it as such, yet the sparc
1164 	 * version complains if it's absent.
1165 	 */
1166 	const void	*p = pi;
1167 
1168 	debug_msg("Entering expand_address");
1169 
1170 	/* set ret[0] to the basename of exec path */
1171 	if ((ret[0] = strrchr(cfg->methods[IM_START]->exec_path, '/'))
1172 	    != NULL) {
1173 		ret[0]++;
1174 	} else {
1175 		ret[0] = cfg->methods[IM_START]->exec_path;
1176 	}
1177 
1178 	if (!cfg->basic->istlx &&
1179 	    (((socket_info_t *)p)->type == SOCK_DGRAM)) {
1180 		ret[1] = NULL;
1181 	} else {
1182 		addrbuf[0] = '\0';
1183 		if (!cfg->basic->iswait &&
1184 		    (inst->remote_addr.ss_family == AF_INET)) {
1185 			struct sockaddr_in *sp;
1186 
1187 			sp = (struct sockaddr_in *)&(inst->remote_addr);
1188 			(void) snprintf(addrbuf, sizeof (addrbuf), "%x.%hu",
1189 			    ntohl(sp->sin_addr.s_addr), ntohs(sp->sin_port));
1190 		}
1191 		ret[1] = addrbuf;
1192 		ret[2] = NULL;
1193 	}
1194 
1195 	return (ret);
1196 }
1197 
1198 /*
1199  * Returns the state associated with the supplied method being run for an
1200  * instance.
1201  */
1202 static internal_inst_state_t
1203 get_method_state(instance_method_t method)
1204 {
1205 	state_info_t *sip;
1206 
1207 	for (sip = states; sip->istate != IIS_NONE; sip++) {
1208 		if (sip->method_running == method)
1209 			break;
1210 	}
1211 	assert(sip->istate != IIS_NONE);
1212 
1213 	return (sip->istate);
1214 }
1215 
1216 /*
1217  * Store the method's PID and CID in the repository. If the store fails
1218  * we ignore it and just drive on.
1219  */
1220 static void
1221 add_method_ids(instance_t *ins, pid_t pid, ctid_t cid, instance_method_t mthd)
1222 {
1223 	debug_msg("Entering add_method_ids");
1224 
1225 	if (cid != -1)
1226 		(void) add_remove_contract(ins->fmri, B_TRUE, cid);
1227 
1228 	if (mthd == IM_START) {
1229 		if (add_rep_val(ins->start_pids, (int64_t)pid) == 0) {
1230 			(void) store_rep_vals(ins->start_pids, ins->fmri,
1231 			    PR_NAME_START_PIDS);
1232 		}
1233 	} else {
1234 		if (add_rep_val(ins->non_start_pid, (int64_t)pid) == 0) {
1235 			(void) store_rep_vals(ins->non_start_pid, ins->fmri,
1236 			    PR_NAME_NON_START_PID);
1237 		}
1238 	}
1239 }
1240 
1241 /*
1242  * Remove the method's PID and CID from the repository. If the removal
1243  * fails we ignore it and drive on.
1244  */
1245 void
1246 remove_method_ids(instance_t *inst, pid_t pid, ctid_t cid,
1247     instance_method_t mthd)
1248 {
1249 	debug_msg("Entering remove_method_ids");
1250 
1251 	if (cid != -1)
1252 		(void) add_remove_contract(inst->fmri, B_FALSE, cid);
1253 
1254 	if (mthd == IM_START) {
1255 		remove_rep_val(inst->start_pids, (int64_t)pid);
1256 		(void) store_rep_vals(inst->start_pids, inst->fmri,
1257 		    PR_NAME_START_PIDS);
1258 	} else {
1259 		remove_rep_val(inst->non_start_pid, (int64_t)pid);
1260 		(void) store_rep_vals(inst->non_start_pid, inst->fmri,
1261 		    PR_NAME_NON_START_PID);
1262 	}
1263 }
1264 
1265 static instance_t *
1266 create_instance(const char *fmri)
1267 {
1268 	instance_t *ret;
1269 
1270 	debug_msg("Entering create_instance, instance: %s", fmri);
1271 
1272 	if (((ret = calloc(1, sizeof (instance_t))) == NULL) ||
1273 	    ((ret->fmri = strdup(fmri)) == NULL))
1274 		goto alloc_fail;
1275 
1276 	ret->conn_fd = -1;
1277 
1278 	ret->copies = 0;
1279 
1280 	ret->conn_rate_count = 0;
1281 	ret->fail_rate_count = 0;
1282 	ret->bind_fail_count = 0;
1283 
1284 	if (((ret->non_start_pid = create_rep_val_list()) == NULL) ||
1285 	    ((ret->start_pids = create_rep_val_list()) == NULL))
1286 		goto alloc_fail;
1287 
1288 	ret->cur_istate = IIS_NONE;
1289 	ret->next_istate = IIS_NONE;
1290 
1291 	if (((ret->cur_istate_rep = create_rep_val_list()) == NULL) ||
1292 	    ((ret->next_istate_rep = create_rep_val_list()) == NULL))
1293 		goto alloc_fail;
1294 
1295 	ret->config = NULL;
1296 	ret->new_config = NULL;
1297 
1298 	ret->timer_id = -1;
1299 	ret->bind_timer_id = -1;
1300 
1301 	ret->disable_req = B_FALSE;
1302 	ret->maintenance_req = B_FALSE;
1303 	ret->conn_rate_exceeded = B_FALSE;
1304 	ret->bind_retries_exceeded = B_FALSE;
1305 
1306 	ret->pending_rst_event = RESTARTER_EVENT_TYPE_INVALID;
1307 
1308 	return (ret);
1309 
1310 alloc_fail:
1311 	error_msg(strerror(errno));
1312 	destroy_instance(ret);
1313 	return (NULL);
1314 }
1315 
1316 static void
1317 destroy_instance(instance_t *inst)
1318 {
1319 	debug_msg("Entering destroy_instance");
1320 
1321 	if (inst == NULL)
1322 		return;
1323 
1324 	destroy_instance_cfg(inst->config);
1325 	destroy_instance_cfg(inst->new_config);
1326 
1327 	destroy_rep_val_list(inst->cur_istate_rep);
1328 	destroy_rep_val_list(inst->next_istate_rep);
1329 
1330 	destroy_rep_val_list(inst->start_pids);
1331 	destroy_rep_val_list(inst->non_start_pid);
1332 
1333 	free(inst->fmri);
1334 
1335 	free(inst);
1336 }
1337 
1338 /*
1339  * Retrieves the current and next states internal states. Returns 0 on success,
1340  * else returns one of the following on error:
1341  * SCF_ERROR_NO_MEMORY if memory allocation failed.
1342  * SCF_ERROR_CONNECTION_BROKEN if the connection to the repository was broken.
1343  * SCF_ERROR_TYPE_MISMATCH if the property was of an unexpected type.
1344  * SCF_ERROR_NO_RESOURCES if the server doesn't have adequate resources.
1345  * SCF_ERROR_NO_SERVER if the server isn't running.
1346  */
1347 static scf_error_t
1348 retrieve_instance_state(instance_t *inst)
1349 {
1350 	scf_error_t	ret;
1351 
1352 	debug_msg("Entering retrieve_instance_state: instance: %s",
1353 	    inst->fmri);
1354 
1355 	/* retrieve internal states */
1356 	if (((ret = retrieve_rep_vals(inst->cur_istate_rep, inst->fmri,
1357 	    PR_NAME_CUR_INT_STATE)) != 0) ||
1358 	    ((ret = retrieve_rep_vals(inst->next_istate_rep, inst->fmri,
1359 	    PR_NAME_NEXT_INT_STATE)) != 0)) {
1360 		if (ret != SCF_ERROR_NOT_FOUND) {
1361 			error_msg(gettext(
1362 			    "Failed to read state of instance %s: %s"),
1363 			    inst->fmri, scf_strerror(scf_error()));
1364 			return (ret);
1365 		}
1366 
1367 		debug_msg("instance with no previous int state - "
1368 		    "setting state to uninitialized");
1369 
1370 		if ((set_single_rep_val(inst->cur_istate_rep,
1371 		    (int64_t)IIS_UNINITIALIZED) == -1) ||
1372 		    (set_single_rep_val(inst->next_istate_rep,
1373 		    (int64_t)IIS_NONE) == -1)) {
1374 			return (SCF_ERROR_NO_MEMORY);
1375 		}
1376 	}
1377 
1378 	/* update convenience states */
1379 	inst->cur_istate = get_single_rep_val(inst->cur_istate_rep);
1380 	inst->next_istate = get_single_rep_val(inst->next_istate_rep);
1381 	debug_msg("previous states: cur: %d, next: %d", inst->cur_istate,
1382 	    inst->next_istate);
1383 
1384 	return (0);
1385 }
1386 
1387 /*
1388  * Retrieve stored process ids and register each of them so we process their
1389  * termination.
1390  */
1391 static int
1392 retrieve_method_pids(instance_t *inst)
1393 {
1394 	rep_val_t	*rv;
1395 
1396 	debug_msg("Entering remove_method_pids");
1397 
1398 	switch (retrieve_rep_vals(inst->start_pids, inst->fmri,
1399 	    PR_NAME_START_PIDS)) {
1400 	case 0:
1401 		break;
1402 	case SCF_ERROR_NOT_FOUND:
1403 		return (0);
1404 	default:
1405 		error_msg(gettext("Failed to retrieve the start pids of "
1406 		    "instance %s from repository: %s"), inst->fmri,
1407 		    scf_strerror(scf_error()));
1408 		return (-1);
1409 	}
1410 
1411 	rv = uu_list_first(inst->start_pids);
1412 	while (rv != NULL) {
1413 		if (register_method(inst, (pid_t)rv->val, (ctid_t)-1,
1414 		    IM_START) == 0) {
1415 			inst->copies++;
1416 			rv = uu_list_next(inst->start_pids, rv);
1417 		} else if (errno == ENOENT) {
1418 			pid_t pid = (pid_t)rv->val;
1419 
1420 			/*
1421 			 * The process must have already terminated. Remove
1422 			 * it from the list.
1423 			 */
1424 			rv = uu_list_next(inst->start_pids, rv);
1425 			remove_rep_val(inst->start_pids, pid);
1426 		} else {
1427 			error_msg(gettext("Failed to listen for the completion "
1428 			    "of %s method of instance %s"), START_METHOD_NAME,
1429 			    inst->fmri);
1430 			rv = uu_list_next(inst->start_pids, rv);
1431 		}
1432 	}
1433 
1434 	/* synch the repository pid list to remove any terminated pids */
1435 	(void) store_rep_vals(inst->start_pids, inst->fmri, PR_NAME_START_PIDS);
1436 
1437 	return (0);
1438 }
1439 
1440 /*
1441  * Remove the passed instance from inetd control.
1442  */
1443 static void
1444 remove_instance(instance_t *instance)
1445 {
1446 	debug_msg("Entering remove_instance");
1447 
1448 	switch (instance->cur_istate) {
1449 	case IIS_ONLINE:
1450 	case IIS_DEGRADED:
1451 		/* stop listening for network connections */
1452 		destroy_bound_fds(instance);
1453 		break;
1454 	case IIS_OFFLINE_BIND:
1455 		cancel_bind_timer(instance);
1456 		break;
1457 	case IIS_OFFLINE_CONRATE:
1458 		cancel_inst_timer(instance);
1459 		break;
1460 	}
1461 
1462 	/* stop listening for terminated methods */
1463 	unregister_instance_methods(instance);
1464 
1465 	uu_list_remove(instance_list, instance);
1466 	destroy_instance(instance);
1467 }
1468 
1469 /*
1470  * Refresh the configuration of instance 'inst'. This method gets called as
1471  * a result of a refresh event for the instance from the master restarter, so
1472  * we can rely upon the instance's running snapshot having been updated from
1473  * its configuration snapshot.
1474  */
1475 void
1476 refresh_instance(instance_t *inst)
1477 {
1478 	instance_cfg_t	*cfg;
1479 
1480 	debug_msg("Entering refresh_instance: inst: %s", inst->fmri);
1481 
1482 	switch (inst->cur_istate) {
1483 	case IIS_MAINTENANCE:
1484 	case IIS_DISABLED:
1485 	case IIS_UNINITIALIZED:
1486 		/*
1487 		 * Ignore any possible changes, we'll re-read the configuration
1488 		 * automatically when we exit these states.
1489 		 */
1490 		break;
1491 
1492 	case IIS_OFFLINE_COPIES:
1493 	case IIS_OFFLINE_BIND:
1494 	case IIS_OFFLINE:
1495 	case IIS_OFFLINE_CONRATE:
1496 		destroy_instance_cfg(inst->config);
1497 		if ((inst->config = read_instance_cfg(inst->fmri)) == NULL) {
1498 			log_invalid_cfg(inst->fmri);
1499 			if (inst->cur_istate == IIS_OFFLINE_BIND) {
1500 				cancel_bind_timer(inst);
1501 			} else if (inst->cur_istate == IIS_OFFLINE_CONRATE) {
1502 				cancel_inst_timer(inst);
1503 			}
1504 			update_state(inst, IIS_MAINTENANCE, RERR_FAULT);
1505 		} else {
1506 			switch (inst->cur_istate) {
1507 			case IIS_OFFLINE_BIND:
1508 				if (copies_limit_exceeded(inst)) {
1509 					/* Cancel scheduled bind retries. */
1510 					cancel_bind_timer(inst);
1511 
1512 					/*
1513 					 * Take the instance to the copies
1514 					 * offline state, via the offline
1515 					 * state.
1516 					 */
1517 					update_state(inst, IIS_OFFLINE,
1518 					    RERR_RESTART);
1519 					process_offline_inst(inst);
1520 				}
1521 				break;
1522 
1523 			case IIS_OFFLINE:
1524 				process_offline_inst(inst);
1525 				break;
1526 
1527 			case IIS_OFFLINE_CONRATE:
1528 				/*
1529 				 * Since we're already in a DOS state,
1530 				 * don't bother evaluating the copies
1531 				 * limit. This will be evaluated when
1532 				 * we leave this state in
1533 				 * process_offline_inst().
1534 				 */
1535 				break;
1536 
1537 			case IIS_OFFLINE_COPIES:
1538 				/*
1539 				 * Check if the copies limit has been increased
1540 				 * above the current count.
1541 				 */
1542 				if (!copies_limit_exceeded(inst)) {
1543 					update_state(inst, IIS_OFFLINE,
1544 					    RERR_RESTART);
1545 					process_offline_inst(inst);
1546 				}
1547 				break;
1548 
1549 			default:
1550 				assert(0);
1551 			}
1552 		}
1553 		break;
1554 
1555 	case IIS_DEGRADED:
1556 	case IIS_ONLINE:
1557 		if ((cfg = read_instance_cfg(inst->fmri)) != NULL) {
1558 			instance_cfg_t *ocfg = inst->config;
1559 
1560 			/*
1561 			 * Try to avoid the overhead of taking an instance
1562 			 * offline and back on again. We do this by limiting
1563 			 * this behavior to two eventualities:
1564 			 * - there needs to be a re-bind to listen on behalf
1565 			 *   of the instance with its new configuration. This
1566 			 *   could be because for example its service has been
1567 			 *   associated with a different port, or because the
1568 			 *   v6only protocol option has been newly applied to
1569 			 *   the instance.
1570 			 * - one or both of the start or online methods of the
1571 			 *   instance have changed in the new configuration.
1572 			 *   Without taking the instance offline when the
1573 			 *   start method changed the instance may be running
1574 			 *   with unwanted parameters (or event an unwanted
1575 			 *   binary); and without taking the instance offline
1576 			 *   if its online method was to change, some part of
1577 			 *   its running environment may have changed and would
1578 			 *   not be picked up until the instance next goes
1579 			 *   offline for another reason.
1580 			 */
1581 			if ((!bind_config_equal(ocfg->basic, cfg->basic)) ||
1582 			    !method_info_equal(ocfg->methods[IM_ONLINE],
1583 			    cfg->methods[IM_ONLINE]) ||
1584 			    !method_info_equal(ocfg->methods[IM_START],
1585 			    cfg->methods[IM_START])) {
1586 				destroy_bound_fds(inst);
1587 
1588 				assert(inst->new_config == NULL);
1589 				inst->new_config = cfg;
1590 
1591 				(void) run_method(inst, IM_OFFLINE, NULL);
1592 			} else {	/* no bind config / method changes */
1593 
1594 				/*
1595 				 * swap the proto list over from the old
1596 				 * configuration to the new, so we retain
1597 				 * our set of network fds.
1598 				 */
1599 				destroy_proto_list(cfg->basic);
1600 				cfg->basic->proto_list =
1601 				    ocfg->basic->proto_list;
1602 				ocfg->basic->proto_list = NULL;
1603 				destroy_instance_cfg(ocfg);
1604 				inst->config = cfg;
1605 
1606 				/* re-evaluate copies limits based on new cfg */
1607 				if (copies_limit_exceeded(inst)) {
1608 					destroy_bound_fds(inst);
1609 					(void) run_method(inst, IM_OFFLINE,
1610 					    NULL);
1611 				} else {
1612 					/*
1613 					 * Since the instance isn't being
1614 					 * taken offline, where we assume it
1615 					 * would pick-up any configuration
1616 					 * changes automatically when it goes
1617 					 * back online, run its refresh method
1618 					 * to allow it to pick-up any changes
1619 					 * whilst still online.
1620 					 */
1621 					(void) run_method(inst, IM_REFRESH,
1622 					    NULL);
1623 				}
1624 			}
1625 		} else {
1626 			log_invalid_cfg(inst->fmri);
1627 
1628 			destroy_bound_fds(inst);
1629 
1630 			inst->maintenance_req = B_TRUE;
1631 			(void) run_method(inst, IM_OFFLINE, NULL);
1632 		}
1633 		break;
1634 
1635 	default:
1636 		debug_msg("Unhandled current state %d for instance in "
1637 		    "refresh_instance", inst->cur_istate);
1638 		assert(0);
1639 	}
1640 }
1641 
1642 /*
1643  * Called by process_restarter_event() to handle a restarter event for an
1644  * instance.
1645  */
1646 static void
1647 handle_restarter_event(instance_t *instance, restarter_event_type_t event,
1648     boolean_t send_ack)
1649 {
1650 	debug_msg("Entering handle_restarter_event: inst: %s, event: %d, "
1651 	    "curr state: %d", instance->fmri, event, instance->cur_istate);
1652 
1653 	switch (event) {
1654 	case RESTARTER_EVENT_TYPE_ADMIN_REFRESH:
1655 		refresh_instance(instance);
1656 		goto done;
1657 	case RESTARTER_EVENT_TYPE_REMOVE_INSTANCE:
1658 		remove_instance(instance);
1659 		goto done;
1660 	case RESTARTER_EVENT_TYPE_STOP:
1661 		switch (instance->cur_istate) {
1662 		case IIS_OFFLINE_CONRATE:
1663 		case IIS_OFFLINE_BIND:
1664 		case IIS_OFFLINE_COPIES:
1665 			/*
1666 			 * inetd must be closing down as we wouldn't get this
1667 			 * event in one of these states from the master
1668 			 * restarter. Take the instance to the offline resting
1669 			 * state.
1670 			 */
1671 			if (instance->cur_istate == IIS_OFFLINE_BIND) {
1672 				cancel_bind_timer(instance);
1673 			} else if (instance->cur_istate ==
1674 			    IIS_OFFLINE_CONRATE) {
1675 				cancel_inst_timer(instance);
1676 			}
1677 			update_state(instance, IIS_OFFLINE, RERR_RESTART);
1678 			goto done;
1679 		}
1680 		break;
1681 	case RESTARTER_EVENT_TYPE_ADMIN_RESTART:
1682 		/*
1683 		 * We've got a restart event, so if the instance is online
1684 		 * in any way initiate taking it offline, and rely upon
1685 		 * our restarter to send us an online event to bring
1686 		 * it back online.
1687 		 */
1688 		switch (instance->cur_istate) {
1689 		case IIS_ONLINE:
1690 		case IIS_DEGRADED:
1691 			destroy_bound_fds(instance);
1692 			(void) run_method(instance, IM_OFFLINE, NULL);
1693 		}
1694 		goto done;
1695 	}
1696 
1697 	switch (instance->cur_istate) {
1698 	case IIS_OFFLINE:
1699 		switch (event) {
1700 		case RESTARTER_EVENT_TYPE_START:
1701 			/*
1702 			 * Dependencies are met, let's take the service online.
1703 			 * Only try and bind for a wait type service if
1704 			 * no process is running on its behalf. Otherwise, just
1705 			 * mark the service online and binding will be attempted
1706 			 * when the process exits.
1707 			 */
1708 			if (!(instance->config->basic->iswait &&
1709 			    (uu_list_first(instance->start_pids) != NULL))) {
1710 				create_bound_fds(instance);
1711 			} else {
1712 				update_state(instance, IIS_ONLINE, RERR_NONE);
1713 			}
1714 			break;
1715 		case RESTARTER_EVENT_TYPE_DISABLE:
1716 		case RESTARTER_EVENT_TYPE_ADMIN_DISABLE:
1717 			/*
1718 			 * The instance should be disabled, so run the
1719 			 * instance's disabled method that will do the work
1720 			 * to take it there.
1721 			 */
1722 			(void) run_method(instance, IM_DISABLE, NULL);
1723 			break;
1724 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON:
1725 		case RESTARTER_EVENT_TYPE_DEPENDENCY_CYCLE:
1726 		case RESTARTER_EVENT_TYPE_INVALID_DEPENDENCY:
1727 			/*
1728 			 * The master restarter has requested the instance
1729 			 * go to maintenance; since we're already offline
1730 			 * just update the state to the maintenance state.
1731 			 */
1732 			update_state(instance, IIS_MAINTENANCE, RERR_RESTART);
1733 			break;
1734 		}
1735 		break;
1736 
1737 	case IIS_OFFLINE_BIND:
1738 		switch (event) {
1739 		case RESTARTER_EVENT_TYPE_DISABLE:
1740 		case RESTARTER_EVENT_TYPE_ADMIN_DISABLE:
1741 			/*
1742 			 * The instance should be disabled. Firstly, as for
1743 			 * the above dependencies unmet comment, cancel
1744 			 * the bind retry timer and update the state to
1745 			 * offline. Then, run the disable method to do the
1746 			 * work to take the instance from offline to
1747 			 * disabled.
1748 			 */
1749 			cancel_bind_timer(instance);
1750 			update_state(instance, IIS_OFFLINE, RERR_RESTART);
1751 			(void) run_method(instance, IM_DISABLE, NULL);
1752 			break;
1753 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON:
1754 		case RESTARTER_EVENT_TYPE_DEPENDENCY_CYCLE:
1755 		case RESTARTER_EVENT_TYPE_INVALID_DEPENDENCY:
1756 			/*
1757 			 * The master restarter has requested the instance
1758 			 * be placed in the maintenance state. Cancel the
1759 			 * outstanding retry timer, and since we're already
1760 			 * offline, update the state to maintenance.
1761 			 */
1762 			cancel_bind_timer(instance);
1763 			update_state(instance, IIS_MAINTENANCE, RERR_RESTART);
1764 			break;
1765 		}
1766 		break;
1767 
1768 	case IIS_DEGRADED:
1769 	case IIS_ONLINE:
1770 		switch (event) {
1771 		case RESTARTER_EVENT_TYPE_DISABLE:
1772 		case RESTARTER_EVENT_TYPE_ADMIN_DISABLE:
1773 			/*
1774 			 * The instance needs to be disabled. Do the same work
1775 			 * as for the dependencies unmet event below to
1776 			 * take the instance offline.
1777 			 */
1778 			destroy_bound_fds(instance);
1779 			/*
1780 			 * Indicate that the offline method is being run
1781 			 * as part of going to the disabled state, and to
1782 			 * carry on this transition.
1783 			 */
1784 			instance->disable_req = B_TRUE;
1785 			(void) run_method(instance, IM_OFFLINE, NULL);
1786 			break;
1787 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON:
1788 		case RESTARTER_EVENT_TYPE_DEPENDENCY_CYCLE:
1789 		case RESTARTER_EVENT_TYPE_INVALID_DEPENDENCY:
1790 			/*
1791 			 * The master restarter has requested the instance be
1792 			 * placed in the maintenance state. This involves
1793 			 * firstly taking the service offline, so do the
1794 			 * same work as for the dependencies unmet event
1795 			 * below. We set the maintenance_req flag to
1796 			 * indicate that when we get to the offline state
1797 			 * we should be placed directly into the maintenance
1798 			 * state.
1799 			 */
1800 			instance->maintenance_req = B_TRUE;
1801 			/* FALLTHROUGH */
1802 		case RESTARTER_EVENT_TYPE_STOP:
1803 			/*
1804 			 * Dependencies have become unmet. Close and
1805 			 * stop listening on the instance's network file
1806 			 * descriptor, and run the offline method to do
1807 			 * any work required to take us to the offline state.
1808 			 */
1809 			destroy_bound_fds(instance);
1810 			(void) run_method(instance, IM_OFFLINE, NULL);
1811 		}
1812 		break;
1813 
1814 	case IIS_UNINITIALIZED:
1815 		if (event == RESTARTER_EVENT_TYPE_DISABLE ||
1816 		    event == RESTARTER_EVENT_TYPE_ADMIN_DISABLE) {
1817 			update_state(instance, IIS_DISABLED, RERR_NONE);
1818 			break;
1819 		} else if (event != RESTARTER_EVENT_TYPE_ENABLE) {
1820 			/*
1821 			 * Ignore other events until we know whether we're
1822 			 * enabled or not.
1823 			 */
1824 			break;
1825 		}
1826 
1827 		/*
1828 		 * We've got an enabled event; make use of the handling in the
1829 		 * disable case.
1830 		 */
1831 		/* FALLTHROUGH */
1832 
1833 	case IIS_DISABLED:
1834 		switch (event) {
1835 		case RESTARTER_EVENT_TYPE_ENABLE:
1836 			/*
1837 			 * The instance needs enabling. Commence reading its
1838 			 * configuration and if successful place the instance
1839 			 * in the offline state and let process_offline_inst()
1840 			 * take it from there.
1841 			 */
1842 			destroy_instance_cfg(instance->config);
1843 			instance->config = read_instance_cfg(instance->fmri);
1844 			if (instance->config != NULL) {
1845 				update_state(instance, IIS_OFFLINE,
1846 				    RERR_RESTART);
1847 				process_offline_inst(instance);
1848 			} else {
1849 				log_invalid_cfg(instance->fmri);
1850 				update_state(instance, IIS_MAINTENANCE,
1851 				    RERR_RESTART);
1852 			}
1853 
1854 			break;
1855 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON:
1856 		case RESTARTER_EVENT_TYPE_DEPENDENCY_CYCLE:
1857 		case RESTARTER_EVENT_TYPE_INVALID_DEPENDENCY:
1858 			/*
1859 			 * The master restarter has requested the instance be
1860 			 * placed in the maintenance state, so just update its
1861 			 * state to maintenance.
1862 			 */
1863 			update_state(instance, IIS_MAINTENANCE, RERR_RESTART);
1864 			break;
1865 		}
1866 		break;
1867 
1868 	case IIS_MAINTENANCE:
1869 		switch (event) {
1870 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_OFF:
1871 		case RESTARTER_EVENT_TYPE_ADMIN_DISABLE:
1872 			/*
1873 			 * The master restarter has requested that the instance
1874 			 * be taken out of maintenance. Read its configuration,
1875 			 * and if successful place the instance in the offline
1876 			 * state and call process_offline_inst() to take it
1877 			 * from there.
1878 			 */
1879 			destroy_instance_cfg(instance->config);
1880 			instance->config = read_instance_cfg(instance->fmri);
1881 			if (instance->config != NULL) {
1882 				update_state(instance, IIS_OFFLINE,
1883 				    RERR_RESTART);
1884 				process_offline_inst(instance);
1885 			} else {
1886 				boolean_t enabled;
1887 
1888 				/*
1889 				 * The configuration was invalid. If the
1890 				 * service has disabled requested, let's
1891 				 * just place the instance in disabled even
1892 				 * though we haven't been able to run its
1893 				 * disable method, as the slightly incorrect
1894 				 * state is likely to be less of an issue to
1895 				 * an administrator than refusing to move an
1896 				 * instance to disabled. If disable isn't
1897 				 * requested, re-mark the service's state
1898 				 * as maintenance, so the administrator can
1899 				 * see the request was processed.
1900 				 */
1901 				if ((read_enable_merged(instance->fmri,
1902 				    &enabled) == 0) && !enabled) {
1903 					update_state(instance, IIS_DISABLED,
1904 					    RERR_RESTART);
1905 				} else {
1906 					log_invalid_cfg(instance->fmri);
1907 					update_state(instance, IIS_MAINTENANCE,
1908 					    RERR_FAULT);
1909 				}
1910 			}
1911 			break;
1912 		}
1913 		break;
1914 
1915 	case IIS_OFFLINE_CONRATE:
1916 		switch (event) {
1917 		case RESTARTER_EVENT_TYPE_DISABLE:
1918 			/*
1919 			 * The instance wants disabling. Take the instance
1920 			 * offline as for the dependencies unmet event above,
1921 			 * and then from there run the disable method to do
1922 			 * the work to take the instance to the disabled state.
1923 			 */
1924 			cancel_inst_timer(instance);
1925 			update_state(instance, IIS_OFFLINE, RERR_RESTART);
1926 			(void) run_method(instance, IM_DISABLE, NULL);
1927 			break;
1928 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON:
1929 		case RESTARTER_EVENT_TYPE_DEPENDENCY_CYCLE:
1930 		case RESTARTER_EVENT_TYPE_INVALID_DEPENDENCY:
1931 			/*
1932 			 * The master restarter has requested the instance
1933 			 * be taken to maintenance. Cancel the timer setup
1934 			 * when we entered this state, and go directly to
1935 			 * maintenance.
1936 			 */
1937 			cancel_inst_timer(instance);
1938 			update_state(instance, IIS_MAINTENANCE, RERR_RESTART);
1939 			break;
1940 		}
1941 		break;
1942 
1943 	case IIS_OFFLINE_COPIES:
1944 		switch (event) {
1945 		case RESTARTER_EVENT_TYPE_DISABLE:
1946 			/*
1947 			 * The instance wants disabling. Update the state
1948 			 * to offline, and run the disable method to do the
1949 			 * work to take it to the disabled state.
1950 			 */
1951 			update_state(instance, IIS_OFFLINE, RERR_RESTART);
1952 			(void) run_method(instance, IM_DISABLE, NULL);
1953 			break;
1954 		case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON:
1955 		case RESTARTER_EVENT_TYPE_DEPENDENCY_CYCLE:
1956 		case RESTARTER_EVENT_TYPE_INVALID_DEPENDENCY:
1957 			/*
1958 			 * The master restarter has requested the instance be
1959 			 * placed in maintenance. Since it's already offline
1960 			 * simply update the state.
1961 			 */
1962 			update_state(instance, IIS_MAINTENANCE, RERR_RESTART);
1963 			break;
1964 		}
1965 		break;
1966 
1967 	default:
1968 		debug_msg("handle_restarter_event: instance in an "
1969 		    "unexpected state");
1970 		assert(0);
1971 	}
1972 
1973 done:
1974 	if (send_ack)
1975 		ack_restarter_event(B_TRUE);
1976 }
1977 
1978 /*
1979  * Tries to read and process an event from the event pipe. If there isn't one
1980  * or an error occurred processing the event it returns -1. Else, if the event
1981  * is for an instance we're not already managing we read its state, add it to
1982  * our list to manage, and if appropriate read its configuration. Whether it's
1983  * new to us or not, we then handle the specific event.
1984  * Returns 0 if an event was read and processed successfully, else -1.
1985  */
1986 static int
1987 process_restarter_event(void)
1988 {
1989 	char			*fmri;
1990 	size_t			fmri_size;
1991 	restarter_event_type_t  event_type;
1992 	instance_t		*instance;
1993 	restarter_event_t	*event;
1994 	ssize_t			sz;
1995 
1996 	debug_msg("Entering process_restarter_event");
1997 
1998 	/*
1999 	 * Try to read an event pointer from the event pipe.
2000 	 */
2001 	errno = 0;
2002 	switch (safe_read(rst_event_pipe[PE_CONSUMER], &event,
2003 	    sizeof (event))) {
2004 	case 0:
2005 		break;
2006 	case  1:
2007 		if (errno == EAGAIN)	/* no event to read */
2008 			return (-1);
2009 
2010 		/* other end of pipe closed */
2011 
2012 		/* FALLTHROUGH */
2013 	default:			/* unexpected read error */
2014 		/*
2015 		 * There's something wrong with the event pipe. Let's
2016 		 * shutdown and be restarted.
2017 		 */
2018 		inetd_stop();
2019 		return (-1);
2020 	}
2021 
2022 	/*
2023 	 * Check if we're currently managing the instance which the event
2024 	 * pertains to. If not, read its complete state and add it to our
2025 	 * list to manage.
2026 	 */
2027 
2028 	fmri_size = scf_limit(SCF_LIMIT_MAX_FMRI_LENGTH);
2029 	if ((fmri = malloc(fmri_size)) == NULL) {
2030 		error_msg(strerror(errno));
2031 		goto fail;
2032 	}
2033 	sz = restarter_event_get_instance(event, fmri, fmri_size);
2034 	if (sz >= fmri_size)
2035 		assert(0);
2036 
2037 	for (instance = uu_list_first(instance_list); instance != NULL;
2038 	    instance = uu_list_next(instance_list, instance)) {
2039 		if (strcmp(instance->fmri, fmri) == 0)
2040 			break;
2041 	}
2042 
2043 	if (instance == NULL) {
2044 		int err;
2045 
2046 		debug_msg("New instance to manage: %s", fmri);
2047 
2048 		if (((instance = create_instance(fmri)) == NULL) ||
2049 		    (retrieve_instance_state(instance) != 0) ||
2050 		    (retrieve_method_pids(instance) != 0)) {
2051 			destroy_instance(instance);
2052 			free(fmri);
2053 			goto fail;
2054 		}
2055 
2056 		if (((err = iterate_repository_contracts(instance->fmri, 0))
2057 		    != 0) && (err != ENOENT)) {
2058 			error_msg(gettext(
2059 			    "Failed to adopt contracts of instance %s: %s"),
2060 			    instance->fmri, strerror(err));
2061 			destroy_instance(instance);
2062 			free(fmri);
2063 			goto fail;
2064 		}
2065 
2066 		uu_list_node_init(instance, &instance->link, instance_pool);
2067 		(void) uu_list_insert_after(instance_list, NULL, instance);
2068 
2069 		/*
2070 		 * Only read configuration for instances that aren't in any of
2071 		 * the disabled, maintenance or uninitialized states, since
2072 		 * they'll read it on state exit.
2073 		 */
2074 		if ((instance->cur_istate != IIS_DISABLED) &&
2075 		    (instance->cur_istate != IIS_MAINTENANCE) &&
2076 		    (instance->cur_istate != IIS_UNINITIALIZED)) {
2077 			instance->config = read_instance_cfg(instance->fmri);
2078 			if (instance->config == NULL) {
2079 				log_invalid_cfg(instance->fmri);
2080 				update_state(instance, IIS_MAINTENANCE,
2081 				    RERR_FAULT);
2082 			}
2083 		}
2084 	}
2085 
2086 	free(fmri);
2087 
2088 	event_type = restarter_event_get_type(event);
2089 	debug_msg("Event type: %d for instance: %s", event_type,
2090 	    instance->fmri);
2091 
2092 	/*
2093 	 * If the instance is currently running a method, don't process the
2094 	 * event now, but attach it to the instance for processing when
2095 	 * the instance finishes its transition.
2096 	 */
2097 	if (INST_IN_TRANSITION(instance)) {
2098 		debug_msg("storing event %d for instance %s", event_type,
2099 		    instance->fmri);
2100 		instance->pending_rst_event = event_type;
2101 	} else {
2102 		handle_restarter_event(instance, event_type, B_TRUE);
2103 	}
2104 
2105 	return (0);
2106 
2107 fail:
2108 	ack_restarter_event(B_FALSE);
2109 	return (-1);
2110 }
2111 
2112 /*
2113  * Do the state machine processing associated with the termination of instance
2114  * 'inst''s start method.
2115  */
2116 void
2117 process_start_term(instance_t *inst)
2118 {
2119 	basic_cfg_t	*cfg;
2120 
2121 	debug_msg("Entering process_start_term: inst: %s", inst->fmri);
2122 
2123 	inst->copies--;
2124 
2125 	if ((inst->cur_istate == IIS_MAINTENANCE) ||
2126 	    (inst->cur_istate == IIS_DISABLED)) {
2127 		/* do any further processing/checks when we exit these states */
2128 		return;
2129 	}
2130 
2131 	cfg = inst->config->basic;
2132 
2133 	if (cfg->iswait) {
2134 		proto_info_t	*pi;
2135 
2136 		switch (inst->cur_istate) {
2137 		case IIS_ONLINE:
2138 		case IIS_DEGRADED:
2139 		case IIS_IN_REFRESH_METHOD:
2140 			/*
2141 			 * A wait type service's start method has exited.
2142 			 * Check if the method was fired off in this inetd's
2143 			 * lifetime, or a previous one; if the former,
2144 			 * re-commence listening on the service's behalf; if
2145 			 * the latter, mark the service offline and let bind
2146 			 * attempts commence.
2147 			 */
2148 			for (pi = uu_list_first(cfg->proto_list); pi != NULL;
2149 			    pi = uu_list_next(cfg->proto_list, pi)) {
2150 				/*
2151 				 * If a bound fd exists, the method was fired
2152 				 * off during this inetd's lifetime.
2153 				 */
2154 				if (pi->listen_fd != -1)
2155 					break;
2156 			}
2157 			if (pi != NULL) {
2158 				if (poll_bound_fds(inst, B_TRUE) != 0)
2159 					handle_bind_failure(inst);
2160 			} else {
2161 				update_state(inst, IIS_OFFLINE, RERR_RESTART);
2162 				create_bound_fds(inst);
2163 			}
2164 		}
2165 	} else {
2166 		/*
2167 		 * Check if a nowait service should be brought back online
2168 		 * after exceeding its copies limit.
2169 		 */
2170 		if ((inst->cur_istate == IIS_OFFLINE_COPIES) &&
2171 		    !copies_limit_exceeded(inst)) {
2172 			update_state(inst, IIS_OFFLINE, RERR_NONE);
2173 			process_offline_inst(inst);
2174 		}
2175 	}
2176 }
2177 
2178 /*
2179  * If the instance has a pending event process it and initiate the
2180  * acknowledgement.
2181  */
2182 static void
2183 process_pending_rst_event(instance_t *inst)
2184 {
2185 	if (inst->pending_rst_event != RESTARTER_EVENT_TYPE_INVALID) {
2186 		restarter_event_type_t re;
2187 
2188 		debug_msg("Injecting pending event %d for instance %s",
2189 		    inst->pending_rst_event, inst->fmri);
2190 		re = inst->pending_rst_event;
2191 		inst->pending_rst_event = RESTARTER_EVENT_TYPE_INVALID;
2192 		handle_restarter_event(inst, re, B_TRUE);
2193 	}
2194 }
2195 
2196 /*
2197  * Do the state machine processing associated with the termination
2198  * of the specified instance's non-start method with the specified status.
2199  * Once the processing of the termination is done, the function also picks up
2200  * any processing that was blocked on the method running.
2201  */
2202 void
2203 process_non_start_term(instance_t *inst, int status)
2204 {
2205 	boolean_t ran_online_method = B_FALSE;
2206 
2207 	debug_msg("Entering process_non_start_term: inst: %s, method: %s",
2208 	    inst->fmri, methods[states[inst->cur_istate].method_running].name);
2209 
2210 	if (status == IMRET_FAILURE) {
2211 		error_msg(gettext("The %s method of instance %s failed, "
2212 		    "transitioning to maintenance"),
2213 		    methods[states[inst->cur_istate].method_running].name,
2214 		    inst->fmri);
2215 
2216 		if ((inst->cur_istate == IIS_IN_ONLINE_METHOD) ||
2217 		    (inst->cur_istate == IIS_IN_REFRESH_METHOD))
2218 			destroy_bound_fds(inst);
2219 
2220 		update_state(inst, IIS_MAINTENANCE, RERR_FAULT);
2221 
2222 		inst->maintenance_req = B_FALSE;
2223 		inst->conn_rate_exceeded = B_FALSE;
2224 
2225 		if (inst->new_config != NULL) {
2226 			destroy_instance_cfg(inst->new_config);
2227 			inst->new_config = NULL;
2228 		}
2229 
2230 		if (!inetd_stopping)
2231 			process_pending_rst_event(inst);
2232 
2233 		return;
2234 	}
2235 
2236 	/* non-failure method return */
2237 
2238 	if (status != IMRET_SUCCESS) {
2239 		/*
2240 		 * An instance method never returned a supported return code.
2241 		 * We'll assume this means the method succeeded for now whilst
2242 		 * non-GL-cognizant methods are used - eg. pkill.
2243 		 */
2244 		debug_msg("The %s method of instance %s returned "
2245 		    "non-compliant exit code: %d, assuming success",
2246 		    methods[states[inst->cur_istate].method_running].name,
2247 		    inst->fmri, status);
2248 	}
2249 
2250 	/*
2251 	 * Update the state from the in-transition state.
2252 	 */
2253 	switch (inst->cur_istate) {
2254 	case IIS_IN_ONLINE_METHOD:
2255 		ran_online_method = B_TRUE;
2256 		/* FALLTHROUGH */
2257 	case IIS_IN_REFRESH_METHOD:
2258 		/*
2259 		 * If we've exhausted the bind retries, flag that by setting
2260 		 * the instance's state to degraded.
2261 		 */
2262 		if (inst->bind_retries_exceeded) {
2263 			update_state(inst, IIS_DEGRADED, RERR_NONE);
2264 			break;
2265 		}
2266 		/* FALLTHROUGH */
2267 	default:
2268 		update_state(inst,
2269 		    methods[states[inst->cur_istate].method_running].dst_state,
2270 		    RERR_NONE);
2271 	}
2272 
2273 	if (inst->cur_istate == IIS_OFFLINE) {
2274 		if (inst->new_config != NULL) {
2275 			/*
2276 			 * This instance was found during refresh to need
2277 			 * taking offline because its newly read configuration
2278 			 * was sufficiently different. Now we're offline,
2279 			 * activate this new configuration.
2280 			 */
2281 			destroy_instance_cfg(inst->config);
2282 			inst->config = inst->new_config;
2283 			inst->new_config = NULL;
2284 		}
2285 
2286 		/* continue/complete any transitions that are in progress */
2287 		process_offline_inst(inst);
2288 
2289 	} else if (ran_online_method) {
2290 		/*
2291 		 * We've just successfully executed the online method. We have
2292 		 * a set of bound network fds that were created before running
2293 		 * this method, so now we're online start listening for
2294 		 * connections on them.
2295 		 */
2296 		if (poll_bound_fds(inst, B_TRUE) != 0)
2297 			handle_bind_failure(inst);
2298 	}
2299 
2300 	/*
2301 	 * If we're now out of transition (process_offline_inst() could have
2302 	 * fired off another method), carry out any jobs that were blocked by
2303 	 * us being in transition.
2304 	 */
2305 	if (!INST_IN_TRANSITION(inst)) {
2306 		if (inetd_stopping) {
2307 			if (!instance_stopped(inst)) {
2308 				/*
2309 				 * inetd is stopping, and this instance hasn't
2310 				 * been stopped. Inject a stop event.
2311 				 */
2312 				handle_restarter_event(inst,
2313 				    RESTARTER_EVENT_TYPE_STOP, B_FALSE);
2314 			}
2315 		} else {
2316 			process_pending_rst_event(inst);
2317 		}
2318 	}
2319 }
2320 
2321 /*
2322  * Check if configuration file specified is readable. If not return B_FALSE,
2323  * else return B_TRUE.
2324  */
2325 static boolean_t
2326 can_read_file(const char *path)
2327 {
2328 	int	ret;
2329 	int	serrno;
2330 
2331 	debug_msg("Entering can_read_file");
2332 	do {
2333 		ret = access(path, R_OK);
2334 	} while ((ret < 0) && (errno == EINTR));
2335 	if (ret < 0) {
2336 		if (errno != ENOENT) {
2337 			serrno = errno;
2338 			error_msg(gettext("Failed to access configuration "
2339 			    "file %s for performing modification checks: %s"),
2340 			    path, strerror(errno));
2341 			errno = serrno;
2342 		}
2343 		return (B_FALSE);
2344 	}
2345 	return (B_TRUE);
2346 }
2347 
2348 /*
2349  * Check whether the configuration file has changed contents since inetd
2350  * was last started/refreshed, and if so, log a message indicating that
2351  * inetconv needs to be run.
2352  */
2353 static void
2354 check_conf_file(void)
2355 {
2356 	char		*new_hash;
2357 	char		*old_hash = NULL;
2358 	scf_error_t	ret;
2359 	const char	*file;
2360 
2361 	debug_msg("Entering check_conf_file");
2362 
2363 	if (conf_file == NULL) {
2364 		/*
2365 		 * No explicit config file specified, so see if one of the
2366 		 * default two are readable, checking the primary one first
2367 		 * followed by the secondary.
2368 		 */
2369 		if (can_read_file(PRIMARY_DEFAULT_CONF_FILE)) {
2370 			file = PRIMARY_DEFAULT_CONF_FILE;
2371 		} else if ((errno == ENOENT) &&
2372 		    can_read_file(SECONDARY_DEFAULT_CONF_FILE)) {
2373 			file = SECONDARY_DEFAULT_CONF_FILE;
2374 		} else {
2375 			return;
2376 		}
2377 	} else {
2378 		file = conf_file;
2379 		if (!can_read_file(file))
2380 			return;
2381 	}
2382 
2383 	if (calculate_hash(file, &new_hash) == 0) {
2384 		ret = retrieve_inetd_hash(&old_hash);
2385 		if (((ret == SCF_ERROR_NONE) &&
2386 		    (strcmp(old_hash, new_hash) != 0))) {
2387 			/* modified config file */
2388 			warn_msg(gettext(
2389 			    "Configuration file %s has been modified since "
2390 			    "inetconv was last run. \"inetconv -i %s\" must be "
2391 			    "run to apply any changes to the SMF"), file, file);
2392 		} else if ((ret != SCF_ERROR_NOT_FOUND) &&
2393 		    (ret != SCF_ERROR_NONE)) {
2394 			/* No message if hash not yet computed */
2395 			error_msg(gettext("Failed to check whether "
2396 			    "configuration file %s has been modified: %s"),
2397 			    file, scf_strerror(ret));
2398 		}
2399 		free(old_hash);
2400 		free(new_hash);
2401 	} else {
2402 		error_msg(gettext("Failed to check whether configuration file "
2403 		    "%s has been modified: %s"), file, strerror(errno));
2404 	}
2405 }
2406 
2407 /*
2408  * Refresh all inetd's managed instances and check the configuration file
2409  * for any updates since inetconv was last run, logging a message if there
2410  * are. We call the SMF refresh function to refresh each instance so that
2411  * the refresh request goes through the framework, and thus results in the
2412  * running snapshot of each instance being updated from the configuration
2413  * snapshot.
2414  */
2415 static void
2416 inetd_refresh(void)
2417 {
2418 	instance_t	*inst;
2419 
2420 	debug_msg("Entering inetd_refresh");
2421 
2422 	/* call libscf to send refresh requests for all managed instances */
2423 	for (inst = uu_list_first(instance_list); inst != NULL;
2424 	    inst = uu_list_next(instance_list, inst)) {
2425 		if (smf_refresh_instance(inst->fmri) < 0) {
2426 			error_msg(gettext("Failed to refresh instance %s: %s"),
2427 			    inst->fmri, scf_strerror(scf_error()));
2428 		}
2429 	}
2430 
2431 	/*
2432 	 * Log a message if the configuration file has changed since inetconv
2433 	 * was last run.
2434 	 */
2435 	check_conf_file();
2436 }
2437 
2438 /*
2439  * Initiate inetd's shutdown.
2440  */
2441 static void
2442 inetd_stop(void)
2443 {
2444 	instance_t *inst;
2445 
2446 	debug_msg("Entering inetd_stop");
2447 
2448 	/* Block handling signals for stop and refresh */
2449 	(void) sighold(SIGHUP);
2450 	(void) sighold(SIGTERM);
2451 
2452 	/* Indicate inetd is coming down */
2453 	inetd_stopping = B_TRUE;
2454 
2455 	/* Stop polling on restarter events. */
2456 	clear_pollfd(rst_event_pipe[PE_CONSUMER]);
2457 
2458 	/* Stop polling for any more stop/refresh requests. */
2459 	clear_pollfd(uds_fd);
2460 
2461 	/*
2462 	 * Send a stop event to all currently unstopped instances that
2463 	 * aren't in transition. For those that are in transition, the
2464 	 * event will get sent when the transition completes.
2465 	 */
2466 	for (inst = uu_list_first(instance_list); inst != NULL;
2467 	    inst = uu_list_next(instance_list, inst)) {
2468 		if (!instance_stopped(inst) && !INST_IN_TRANSITION(inst))
2469 			handle_restarter_event(inst,
2470 			    RESTARTER_EVENT_TYPE_STOP, B_FALSE);
2471 	}
2472 }
2473 
2474 /*
2475  * Sets up the intra-inetd-process Unix Domain Socket.
2476  * Returns -1 on error, else 0.
2477  */
2478 static int
2479 uds_init(void)
2480 {
2481 	struct sockaddr_un addr;
2482 
2483 	debug_msg("Entering uds_init");
2484 
2485 	if ((uds_fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0) {
2486 		error_msg("socket: %s", strerror(errno));
2487 		return (-1);
2488 	}
2489 
2490 	disable_blocking(uds_fd);
2491 
2492 	(void) unlink(INETD_UDS_PATH);  /* clean-up any stale files */
2493 
2494 	(void) memset(&addr, 0, sizeof (addr));
2495 	addr.sun_family = AF_UNIX;
2496 	/* CONSTCOND */
2497 	assert(sizeof (INETD_UDS_PATH) <= sizeof (addr.sun_path));
2498 	(void) strlcpy(addr.sun_path, INETD_UDS_PATH, sizeof (addr.sun_path));
2499 
2500 	if (bind(uds_fd, (struct sockaddr *)(&addr), sizeof (addr)) < 0) {
2501 		error_msg(gettext("Failed to bind socket to %s: %s"),
2502 		    INETD_UDS_PATH, strerror(errno));
2503 		(void) close(uds_fd);
2504 		return (-1);
2505 	}
2506 
2507 	(void) listen(uds_fd, UDS_BACKLOG);
2508 
2509 	if ((set_pollfd(uds_fd, POLLIN)) == -1) {
2510 		(void) close(uds_fd);
2511 		(void) unlink(INETD_UDS_PATH);
2512 		return (-1);
2513 	}
2514 
2515 	return (0);
2516 }
2517 
2518 static void
2519 uds_fini(void)
2520 {
2521 	if (uds_fd != -1)
2522 		(void) close(uds_fd);
2523 	(void) unlink(INETD_UDS_PATH);
2524 }
2525 
2526 /*
2527  * Handle an incoming request on the Unix Domain Socket. Returns -1 if there
2528  * was an error handling the event, else 0.
2529  */
2530 static int
2531 process_uds_event(void)
2532 {
2533 	uds_request_t		req;
2534 	int			fd;
2535 	struct sockaddr_un	addr;
2536 	socklen_t		len = sizeof (addr);
2537 	int			ret;
2538 	uint_t			retries = 0;
2539 
2540 	debug_msg("Entering process_uds_event");
2541 
2542 	do {
2543 		fd = accept(uds_fd, (struct sockaddr *)&addr, &len);
2544 	} while ((fd < 0) && (errno == EINTR));
2545 	if (fd < 0) {
2546 		if (errno != EWOULDBLOCK)
2547 			error_msg("accept failed: %s", strerror(errno));
2548 		return (-1);
2549 	}
2550 
2551 	for (retries = 0; retries < UDS_RECV_RETRIES; retries++) {
2552 		if (((ret = safe_read(fd, &req, sizeof (req))) != 1) ||
2553 		    (errno != EAGAIN))
2554 			break;
2555 
2556 		(void) poll(NULL, 0, 100);	/* 100ms pause */
2557 	}
2558 
2559 	if (ret != 0) {
2560 		error_msg(gettext("Failed read: %s"), strerror(errno));
2561 		(void) close(fd);
2562 		return (-1);
2563 	}
2564 
2565 	switch (req) {
2566 	case UR_REFRESH_INETD:
2567 		/* flag the request for event_loop() to process */
2568 		refresh_inetd_requested = B_TRUE;
2569 		(void) close(fd);
2570 		break;
2571 	case UR_STOP_INETD:
2572 		inetd_stop();
2573 		break;
2574 	default:
2575 		error_msg("unexpected UDS request");
2576 		(void) close(fd);
2577 		return (-1);
2578 	}
2579 
2580 	return (0);
2581 }
2582 
2583 /*
2584  * Perform checks for common exec string errors. We limit the checks to
2585  * whether the file exists, is a regular file, and has at least one execute
2586  * bit set. We leave the core security checks to exec() so as not to duplicate
2587  * and thus incur the associated drawbacks, but hope to catch the common
2588  * errors here.
2589  */
2590 static boolean_t
2591 passes_basic_exec_checks(const char *instance, const char *method,
2592     const char *path)
2593 {
2594 	struct stat	sbuf;
2595 
2596 	debug_msg("Entering passes_basic_exec_checks");
2597 
2598 	/* check the file exists */
2599 	while (stat(path, &sbuf) == -1) {
2600 		if (errno != EINTR) {
2601 			error_msg(gettext(
2602 			    "Can't stat the %s method of instance %s: %s"),
2603 			    method, instance, strerror(errno));
2604 			return (B_FALSE);
2605 		}
2606 	}
2607 
2608 	/*
2609 	 * Check if the file is a regular file and has at least one execute
2610 	 * bit set.
2611 	 */
2612 	if ((sbuf.st_mode & S_IFMT) != S_IFREG) {
2613 		error_msg(gettext(
2614 		    "The %s method of instance %s isn't a regular file"),
2615 		    method, instance);
2616 		return (B_FALSE);
2617 	} else if ((sbuf.st_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
2618 		error_msg(gettext("The %s method instance %s doesn't have "
2619 		    "any execute permissions set"), method, instance);
2620 		return (B_FALSE);
2621 	}
2622 
2623 	return (B_TRUE);
2624 }
2625 
2626 static void
2627 exec_method(instance_t *instance, instance_method_t method, method_info_t *mi,
2628     struct method_context *mthd_ctxt, const proto_info_t *pi)
2629 {
2630 	char		**args;
2631 	char 		**env;
2632 	const char	*errf;
2633 	int		serrno;
2634 	basic_cfg_t	*cfg = instance->config->basic;
2635 
2636 	if (method == IM_START) {
2637 		/*
2638 		 * If wrappers checks fail, pretend the method was exec'd and
2639 		 * failed.
2640 		 */
2641 		if (!tcp_wrappers_ok(instance))
2642 			exit(IMRET_FAILURE);
2643 	}
2644 
2645 	/*
2646 	 * Revert the disposition of handled signals and ignored signals to
2647 	 * their defaults, unblocking any blocked ones as a side effect.
2648 	 */
2649 	(void) sigset(SIGHUP, SIG_DFL);
2650 	(void) sigset(SIGTERM, SIG_DFL);
2651 	(void) sigset(SIGINT, SIG_DFL);
2652 
2653 	/*
2654 	 * Setup exec arguments. Do this before the fd setup below, so our
2655 	 * logging related file fd doesn't get taken over before we call
2656 	 * expand_address().
2657 	 */
2658 	if ((method == IM_START) &&
2659 	    (strcmp(mi->exec_args_we.we_wordv[0], "%A") == 0)) {
2660 		args = expand_address(instance, pi);
2661 	} else {
2662 		args = mi->exec_args_we.we_wordv;
2663 	}
2664 
2665 	/* Generate audit trail for start operations */
2666 	if (method == IM_START) {
2667 		adt_event_data_t *ae;
2668 		struct sockaddr_storage ss;
2669 		priv_set_t *privset;
2670 		socklen_t sslen = sizeof (ss);
2671 
2672 		if ((ae = adt_alloc_event(audit_handle, ADT_inetd_connect))
2673 		    == NULL) {
2674 			error_msg(gettext("Unable to allocate audit event for "
2675 			    "the %s method of instance %s"),
2676 			    methods[method].name, instance->fmri);
2677 			exit(IMRET_FAILURE);
2678 		}
2679 
2680 		/*
2681 		 * The inetd_connect audit record consists of:
2682 		 *	Service name
2683 		 *	Execution path
2684 		 *	Remote address and port
2685 		 *	Local port
2686 		 *	Process privileges
2687 		 */
2688 		ae->adt_inetd_connect.service_name = cfg->svc_name;
2689 		ae->adt_inetd_connect.cmd = mi->exec_path;
2690 
2691 		if (instance->remote_addr.ss_family == AF_INET) {
2692 			struct in_addr *in = SS_SINADDR(instance->remote_addr);
2693 			ae->adt_inetd_connect.ip_adr[0] = in->s_addr;
2694 			ae->adt_inetd_connect.ip_type = ADT_IPv4;
2695 		} else {
2696 			uint32_t *addr6;
2697 			int i;
2698 
2699 			ae->adt_inetd_connect.ip_type = ADT_IPv6;
2700 			addr6 = (uint32_t *)SS_SINADDR(instance->remote_addr);
2701 			for (i = 0; i < 4; ++i)
2702 				ae->adt_inetd_connect.ip_adr[i] = addr6[i];
2703 		}
2704 
2705 		ae->adt_inetd_connect.ip_remote_port =
2706 		    ntohs(SS_PORT(instance->remote_addr));
2707 
2708 		if (getsockname(instance->conn_fd, (struct sockaddr *)&ss,
2709 		    &sslen) == 0)
2710 			ae->adt_inetd_connect.ip_local_port =
2711 			    ntohs(SS_PORT(ss));
2712 
2713 		privset = mthd_ctxt->priv_set;
2714 		if (privset == NULL) {
2715 			privset = priv_allocset();
2716 			if (privset != NULL &&
2717 			    getppriv(PRIV_EFFECTIVE, privset) != 0) {
2718 				priv_freeset(privset);
2719 				privset = NULL;
2720 			}
2721 		}
2722 
2723 		ae->adt_inetd_connect.privileges = privset;
2724 
2725 		(void) adt_put_event(ae, ADT_SUCCESS, ADT_SUCCESS);
2726 		adt_free_event(ae);
2727 
2728 		if (privset != NULL && mthd_ctxt->priv_set == NULL)
2729 			priv_freeset(privset);
2730 	}
2731 
2732 	/*
2733 	 * Set method context before the fd setup below so we can output an
2734 	 * error message if it fails.
2735 	 */
2736 	if ((errno = restarter_set_method_context(mthd_ctxt, &errf)) != 0) {
2737 		const char *msg;
2738 
2739 		if (errno == -1) {
2740 			if (strcmp(errf, "core_set_process_path") == 0) {
2741 				msg = gettext("Failed to set the corefile path "
2742 				    "for the %s method of instance %s");
2743 			} else if (strcmp(errf, "setproject") == 0) {
2744 				msg = gettext("Failed to assign a resource "
2745 				    "control for the %s method of instance %s");
2746 			} else if (strcmp(errf, "pool_set_binding") == 0) {
2747 				msg = gettext("Failed to bind the %s method of "
2748 				    "instance %s to a pool due to a system "
2749 				    "error");
2750 			} else {
2751 				assert(0);
2752 				abort();
2753 			}
2754 
2755 			error_msg(msg, methods[method].name, instance->fmri);
2756 
2757 			exit(IMRET_FAILURE);
2758 		}
2759 
2760 		if (errf != NULL && strcmp(errf, "pool_set_binding") == 0) {
2761 			switch (errno) {
2762 			case ENOENT:
2763 				msg = gettext("Failed to find resource pool "
2764 				    "for the %s method of instance %s");
2765 				break;
2766 
2767 			case EBADF:
2768 				msg = gettext("Failed to bind the %s method of "
2769 				    "instance %s to a pool due to invalid "
2770 				    "configuration");
2771 				break;
2772 
2773 			default:
2774 				assert(0);
2775 				abort();
2776 			}
2777 
2778 			exit(IMRET_FAILURE);
2779 		}
2780 
2781 		if (errf != NULL) {
2782 			error_msg(gettext("Failed to set credentials for the "
2783 			    "%s method of instance %s (%s: %s)"),
2784 			    methods[method].name, instance->fmri, errf,
2785 			    strerror(errno));
2786 			exit(IMRET_FAILURE);
2787 		}
2788 
2789 		switch (errno) {
2790 		case ENOMEM:
2791 			msg = gettext("Failed to set credentials for the %s "
2792 			    "method of instance %s (out of memory)");
2793 			break;
2794 
2795 		case ENOENT:
2796 			msg = gettext("Failed to set credentials for the %s "
2797 			    "method of instance %s (no passwd or shadow "
2798 			    "entry for user)");
2799 			break;
2800 
2801 		default:
2802 			assert(0);
2803 			abort();
2804 		}
2805 
2806 		error_msg(msg, methods[method].name, instance->fmri);
2807 		exit(IMRET_FAILURE);
2808 	}
2809 
2810 	/* let exec() free mthd_ctxt */
2811 
2812 	/* setup standard fds */
2813 	if (method == IM_START) {
2814 		(void) dup2(instance->conn_fd, STDIN_FILENO);
2815 	} else {
2816 		(void) close(STDIN_FILENO);
2817 		(void) open("/dev/null", O_RDONLY);
2818 	}
2819 	(void) dup2(STDIN_FILENO, STDOUT_FILENO);
2820 	(void) dup2(STDIN_FILENO, STDERR_FILENO);
2821 
2822 	closefrom(STDERR_FILENO + 1);
2823 
2824 	method_preexec();
2825 
2826 	env = set_smf_env(mthd_ctxt, instance, methods[method].name);
2827 
2828 	if (env != NULL) {
2829 		do {
2830 			(void) execve(mi->exec_path, args, env);
2831 		} while (errno == EINTR);
2832 	}
2833 
2834 	serrno = errno;
2835 	/* start up logging again to report the error */
2836 	msg_init();
2837 	errno = serrno;
2838 
2839 	error_msg(
2840 	    gettext("Failed to exec %s method of instance %s: %s"),
2841 	    methods[method].name, instance->fmri, strerror(errno));
2842 
2843 	if ((method == IM_START) && (instance->config->basic->iswait)) {
2844 		/*
2845 		 * We couldn't exec the start method for a wait type service.
2846 		 * Eat up data from the endpoint, so that hopefully the
2847 		 * service's fd won't wake poll up on the next time round
2848 		 * event_loop(). This behavior is carried over from the old
2849 		 * inetd, and it seems somewhat arbitrary that it isn't
2850 		 * also done in the case of fork failures; but I guess
2851 		 * it assumes an exec failure is less likely to be the result
2852 		 * of a resource shortage, and is thus not worth retrying.
2853 		 */
2854 		consume_wait_data(instance, 0);
2855 	}
2856 
2857 	exit(IMRET_FAILURE);
2858 }
2859 
2860 static restarter_error_t
2861 get_method_error_success(instance_method_t method)
2862 {
2863 	switch (method) {
2864 	case IM_OFFLINE:
2865 		return (RERR_RESTART);
2866 	case IM_ONLINE:
2867 		return (RERR_RESTART);
2868 	case IM_DISABLE:
2869 		return (RERR_RESTART);
2870 	case IM_REFRESH:
2871 		return (RERR_REFRESH);
2872 	case IM_START:
2873 		return (RERR_RESTART);
2874 	}
2875 	(void) fprintf(stderr, gettext("Internal fatal error in inetd.\n"));
2876 
2877 	abort();
2878 	/* NOTREACHED */
2879 }
2880 
2881 /*
2882  * Runs the specified method of the specified service instance.
2883  * If the method was never specified, we handle it the same as if the
2884  * method was called and returned success, carrying on any transition the
2885  * instance may be in the midst of.
2886  * If the method isn't executable in its specified profile or an error occurs
2887  * forking a process to run the method in the function returns -1.
2888  * If a method binary is successfully executed, the function switches the
2889  * instance's cur state to the method's associated 'run' state and the next
2890  * state to the methods associated next state.
2891  * Returns -1 if there's an error before forking, else 0.
2892  */
2893 int
2894 run_method(instance_t *instance, instance_method_t method,
2895     const proto_info_t *start_info)
2896 {
2897 	pid_t			child_pid;
2898 	method_info_t		*mi;
2899 	struct method_context	*mthd_ctxt = NULL;
2900 	const char		*errstr;
2901 	int			sig;
2902 	int			ret;
2903 	instance_cfg_t		*cfg = instance->config;
2904 	ctid_t			cid;
2905 	boolean_t		trans_failure = B_TRUE;
2906 	int			serrno;
2907 
2908 	debug_msg("Entering run_method, instance: %s, method: %s",
2909 	    instance->fmri, methods[method].name);
2910 
2911 	/*
2912 	 * Don't bother updating the instance's state for the start method
2913 	 * as there isn't a separate start method state.
2914 	 */
2915 	if (method != IM_START)
2916 		update_instance_states(instance, get_method_state(method),
2917 		    methods[method].dst_state,
2918 		    get_method_error_success(method));
2919 
2920 	if ((mi = cfg->methods[method]) == NULL) {
2921 		/*
2922 		 * An unspecified method. Since the absence of this method
2923 		 * must be valid (otherwise it would have been caught
2924 		 * during configuration validation), simply pretend the method
2925 		 * ran and returned success.
2926 		 */
2927 		process_non_start_term(instance, IMRET_SUCCESS);
2928 		return (0);
2929 	}
2930 
2931 	/* Handle special method tokens, not allowed on start */
2932 	if (method != IM_START) {
2933 		if (restarter_is_null_method(mi->exec_path)) {
2934 			/* :true means nothing should be done */
2935 			process_non_start_term(instance, IMRET_SUCCESS);
2936 			return (0);
2937 		}
2938 
2939 		if ((sig = restarter_is_kill_method(mi->exec_path)) >= 0) {
2940 			/* Carry out contract assassination */
2941 			ret = iterate_repository_contracts(instance->fmri, sig);
2942 			/* ENOENT means we didn't find any contracts */
2943 			if (ret != 0 && ret != ENOENT) {
2944 				error_msg(gettext("Failed to send signal %d "
2945 				    "to contracts of instance %s: %s"), sig,
2946 				    instance->fmri, strerror(ret));
2947 				goto prefork_failure;
2948 			} else {
2949 				process_non_start_term(instance, IMRET_SUCCESS);
2950 				return (0);
2951 			}
2952 		}
2953 
2954 		if ((sig = restarter_is_kill_proc_method(mi->exec_path)) >= 0) {
2955 			/* Carry out process assassination */
2956 			rep_val_t	*rv;
2957 
2958 			ret = IMRET_SUCCESS;
2959 			for (rv = uu_list_first(instance->start_pids);
2960 			    rv != NULL;
2961 			    rv = uu_list_next(instance->start_pids, rv)) {
2962 				if ((kill((pid_t)rv->val, sig) != 0) &&
2963 				    (errno != ESRCH)) {
2964 					ret = IMRET_FAILURE;
2965 					error_msg(gettext("Unable to signal "
2966 					    "start process of instance %s: %s"),
2967 					    instance->fmri, strerror(errno));
2968 				}
2969 			}
2970 
2971 			process_non_start_term(instance, ret);
2972 			return (0);
2973 		}
2974 	}
2975 
2976 	/*
2977 	 * Get the associated method context before the fork so we can
2978 	 * modify the instances state if things go wrong.
2979 	 */
2980 	if ((mthd_ctxt = read_method_context(instance->fmri,
2981 	    methods[method].name, mi->exec_path, &errstr)) == NULL) {
2982 		error_msg(gettext("Failed to retrieve method context for the "
2983 		    "%s method of instance %s: %s"), methods[method].name,
2984 		    instance->fmri, errstr);
2985 		goto prefork_failure;
2986 	}
2987 
2988 	/*
2989 	 * Perform some basic checks before we fork to limit the possibility
2990 	 * of exec failures, so we can modify the instance state if necessary.
2991 	 */
2992 	if (!passes_basic_exec_checks(instance->fmri, methods[method].name,
2993 	    mi->exec_path)) {
2994 		trans_failure = B_FALSE;
2995 		goto prefork_failure;
2996 	}
2997 
2998 	if (contract_prefork() == -1)
2999 		goto prefork_failure;
3000 	child_pid = fork();
3001 	serrno = errno;
3002 	contract_postfork();
3003 
3004 	switch (child_pid) {
3005 	case -1:
3006 		error_msg(gettext(
3007 		    "Unable to fork %s method of instance %s: %s"),
3008 		    methods[method].name, instance->fmri, strerror(serrno));
3009 		if ((serrno != EAGAIN) && (serrno != ENOMEM))
3010 			trans_failure = B_FALSE;
3011 		goto prefork_failure;
3012 	case 0:				/* child */
3013 		exec_method(instance, method, mi, mthd_ctxt, start_info);
3014 		/* NOTREACHED */
3015 	default:			/* parent */
3016 		restarter_free_method_context(mthd_ctxt);
3017 		mthd_ctxt = NULL;
3018 
3019 		if (get_latest_contract(&cid) < 0)
3020 			cid = -1;
3021 
3022 		/*
3023 		 * Register this method so its termination is noticed and
3024 		 * the state transition this method participates in is
3025 		 * continued.
3026 		 */
3027 		if (register_method(instance, child_pid, cid, method) != 0) {
3028 			/*
3029 			 * Since we will never find out about the termination
3030 			 * of this method, if it's a non-start method treat
3031 			 * is as a failure so we don't block restarter event
3032 			 * processing on it whilst it languishes in a method
3033 			 * running state.
3034 			 */
3035 			error_msg(gettext("Failed to monitor status of "
3036 			    "%s method of instance %s"), methods[method].name,
3037 			    instance->fmri);
3038 			if (method != IM_START)
3039 				process_non_start_term(instance, IMRET_FAILURE);
3040 		}
3041 
3042 		add_method_ids(instance, child_pid, cid, method);
3043 
3044 		/* do tcp tracing for those nowait instances that request it */
3045 		if ((method == IM_START) && cfg->basic->do_tcp_trace &&
3046 		    !cfg->basic->iswait) {
3047 			char buf[INET6_ADDRSTRLEN];
3048 
3049 			syslog(LOG_NOTICE, "%s[%d] from %s %d",
3050 			    cfg->basic->svc_name, child_pid,
3051 			    inet_ntop_native(instance->remote_addr.ss_family,
3052 			    SS_SINADDR(instance->remote_addr), buf,
3053 			    sizeof (buf)),
3054 			    ntohs(SS_PORT(instance->remote_addr)));
3055 		}
3056 	}
3057 
3058 	return (0);
3059 
3060 prefork_failure:
3061 	if (mthd_ctxt != NULL) {
3062 		restarter_free_method_context(mthd_ctxt);
3063 		mthd_ctxt = NULL;
3064 	}
3065 
3066 	if (method == IM_START) {
3067 		/*
3068 		 * Only place a start method in maintenance if we're sure
3069 		 * that the failure was non-transient.
3070 		 */
3071 		if (!trans_failure) {
3072 			destroy_bound_fds(instance);
3073 			update_state(instance, IIS_MAINTENANCE, RERR_FAULT);
3074 		}
3075 	} else {
3076 		/* treat the failure as if the method ran and failed */
3077 		process_non_start_term(instance, IMRET_FAILURE);
3078 	}
3079 
3080 	return (-1);
3081 }
3082 
3083 static int
3084 accept_connection(instance_t *instance, proto_info_t *pi)
3085 {
3086 	int		fd;
3087 	socklen_t	size;
3088 
3089 	debug_msg("Entering accept_connection");
3090 
3091 	if (instance->config->basic->istlx) {
3092 		fd = tlx_accept(instance->fmri, (tlx_info_t *)pi,
3093 		    &(instance->remote_addr));
3094 	} else {
3095 		size = sizeof (instance->remote_addr);
3096 		fd = accept(pi->listen_fd,
3097 		    (struct sockaddr *)&(instance->remote_addr), &size);
3098 		if (fd < 0)
3099 			error_msg("accept: %s", strerror(errno));
3100 	}
3101 
3102 	return (fd);
3103 }
3104 
3105 /*
3106  * Handle an incoming connection request for a nowait service.
3107  * This involves accepting the incoming connection on a new fd. Connection
3108  * rate checks are then performed, transitioning the service to the
3109  * conrate offline state if these fail. Otherwise, the service's start method
3110  * is run (performing TCP wrappers checks if applicable as we do), and on
3111  * success concurrent copies checking is done, transitioning the service to the
3112  * copies offline state if this fails.
3113  */
3114 static void
3115 process_nowait_request(instance_t *instance, proto_info_t *pi)
3116 {
3117 	basic_cfg_t		*cfg = instance->config->basic;
3118 	int			ret;
3119 	adt_event_data_t	*ae;
3120 	char			buf[BUFSIZ];
3121 
3122 	debug_msg("Entering process_nowait_req");
3123 
3124 	/* accept nowait service connections on a new fd */
3125 	if ((instance->conn_fd = accept_connection(instance, pi)) == -1) {
3126 		/*
3127 		 * Failed accept. Return and allow the event loop to initiate
3128 		 * another attempt later if the request is still present.
3129 		 */
3130 		return;
3131 	}
3132 
3133 	/*
3134 	 * Limit connection rate of nowait services. If either conn_rate_max
3135 	 * or conn_rate_offline are <= 0, no connection rate limit checking
3136 	 * is done. If the configured rate is exceeded, the instance is taken
3137 	 * to the connrate_offline state and a timer scheduled to try and
3138 	 * bring the instance back online after the configured offline time.
3139 	 */
3140 	if ((cfg->conn_rate_max > 0) && (cfg->conn_rate_offline > 0)) {
3141 		if (instance->conn_rate_count++ == 0) {
3142 			instance->conn_rate_start = time(NULL);
3143 		} else if (instance->conn_rate_count >
3144 		    cfg->conn_rate_max) {
3145 			time_t now = time(NULL);
3146 
3147 			if ((now - instance->conn_rate_start) > 1) {
3148 				instance->conn_rate_start = now;
3149 				instance->conn_rate_count = 1;
3150 			} else {
3151 				/* Generate audit record */
3152 				if ((ae = adt_alloc_event(audit_handle,
3153 				    ADT_inetd_ratelimit)) == NULL) {
3154 					error_msg(gettext("Unable to allocate "
3155 					    "rate limit audit event"));
3156 				} else {
3157 					adt_inetd_ratelimit_t *rl =
3158 					    &ae->adt_inetd_ratelimit;
3159 					/*
3160 					 * The inetd_ratelimit audit
3161 					 * record consists of:
3162 					 * 	Service name
3163 					 *	Connection rate limit
3164 					 */
3165 					rl->service_name = cfg->svc_name;
3166 					(void) snprintf(buf, sizeof (buf),
3167 					    "limit=%lld", cfg->conn_rate_max);
3168 					rl->limit = buf;
3169 					(void) adt_put_event(ae, ADT_SUCCESS,
3170 					    ADT_SUCCESS);
3171 					adt_free_event(ae);
3172 				}
3173 
3174 				error_msg(gettext(
3175 				    "Instance %s has exceeded its configured "
3176 				    "connection rate, additional connections "
3177 				    "will not be accepted for %d seconds"),
3178 				    instance->fmri, cfg->conn_rate_offline);
3179 
3180 				close_net_fd(instance, instance->conn_fd);
3181 				instance->conn_fd = -1;
3182 
3183 				destroy_bound_fds(instance);
3184 
3185 				instance->conn_rate_count = 0;
3186 
3187 				instance->conn_rate_exceeded = B_TRUE;
3188 				(void) run_method(instance, IM_OFFLINE, NULL);
3189 
3190 				return;
3191 			}
3192 		}
3193 	}
3194 
3195 	ret = run_method(instance, IM_START, pi);
3196 
3197 	close_net_fd(instance, instance->conn_fd);
3198 	instance->conn_fd = -1;
3199 
3200 	if (ret == -1) /* the method wasn't forked  */
3201 		return;
3202 
3203 	instance->copies++;
3204 
3205 	/*
3206 	 * Limit concurrent connections of nowait services.
3207 	 */
3208 	if (copies_limit_exceeded(instance)) {
3209 		/* Generate audit record */
3210 		if ((ae = adt_alloc_event(audit_handle, ADT_inetd_copylimit))
3211 		    == NULL) {
3212 			error_msg(gettext("Unable to allocate copy limit "
3213 			    "audit event"));
3214 		} else {
3215 			/*
3216 			 * The inetd_copylimit audit record consists of:
3217 			 *	Service name
3218 			 * 	Copy limit
3219 			 */
3220 			ae->adt_inetd_copylimit.service_name = cfg->svc_name;
3221 			(void) snprintf(buf, sizeof (buf), "limit=%lld",
3222 			    cfg->max_copies);
3223 			ae->adt_inetd_copylimit.limit = buf;
3224 			(void) adt_put_event(ae, ADT_SUCCESS, ADT_SUCCESS);
3225 			adt_free_event(ae);
3226 		}
3227 
3228 		warn_msg(gettext("Instance %s has reached its maximum "
3229 		    "configured copies, no new connections will be accepted"),
3230 		    instance->fmri);
3231 		destroy_bound_fds(instance);
3232 		(void) run_method(instance, IM_OFFLINE, NULL);
3233 	}
3234 }
3235 
3236 /*
3237  * Handle an incoming request for a wait type service.
3238  * Failure rate checking is done first, taking the service to the maintenance
3239  * state if the checks fail. Following this, the service's start method is run,
3240  * and on success, we stop listening for new requests for this service.
3241  */
3242 static void
3243 process_wait_request(instance_t *instance, const proto_info_t *pi)
3244 {
3245 	basic_cfg_t		*cfg = instance->config->basic;
3246 	int			ret;
3247 	adt_event_data_t	*ae;
3248 	char			buf[BUFSIZ];
3249 
3250 	debug_msg("Entering process_wait_request");
3251 
3252 	instance->conn_fd = pi->listen_fd;
3253 
3254 	/*
3255 	 * Detect broken servers and transition them to maintenance. If a
3256 	 * wait type service exits without accepting the connection or
3257 	 * consuming (reading) the datagram, that service's descriptor will
3258 	 * select readable again, and inetd will fork another instance of
3259 	 * the server. If either wait_fail_cnt or wait_fail_interval are <= 0,
3260 	 * no failure rate detection is done.
3261 	 */
3262 	if ((cfg->wait_fail_cnt > 0) && (cfg->wait_fail_interval > 0)) {
3263 		if (instance->fail_rate_count++ == 0) {
3264 			instance->fail_rate_start = time(NULL);
3265 		} else if (instance->fail_rate_count > cfg->wait_fail_cnt) {
3266 			time_t now = time(NULL);
3267 
3268 			if ((now - instance->fail_rate_start) >
3269 			    cfg->wait_fail_interval) {
3270 				instance->fail_rate_start = now;
3271 				instance->fail_rate_count = 1;
3272 			} else {
3273 				/* Generate audit record */
3274 				if ((ae = adt_alloc_event(audit_handle,
3275 				    ADT_inetd_failrate)) == NULL) {
3276 					error_msg(gettext("Unable to allocate "
3277 					    "failure rate audit event"));
3278 				} else {
3279 					adt_inetd_failrate_t *fr =
3280 					    &ae->adt_inetd_failrate;
3281 					/*
3282 					 * The inetd_failrate audit record
3283 					 * consists of:
3284 					 * 	Service name
3285 					 * 	Failure rate
3286 					 *	Interval
3287 					 * Last two are expressed as k=v pairs
3288 					 * in the values field.
3289 					 */
3290 					fr->service_name = cfg->svc_name;
3291 					(void) snprintf(buf, sizeof (buf),
3292 					    "limit=%lld,interval=%d",
3293 					    cfg->wait_fail_cnt,
3294 					    cfg->wait_fail_interval);
3295 					fr->values = buf;
3296 					(void) adt_put_event(ae, ADT_SUCCESS,
3297 					    ADT_SUCCESS);
3298 					adt_free_event(ae);
3299 				}
3300 
3301 				error_msg(gettext(
3302 				    "Instance %s has exceeded its configured "
3303 				    "failure rate, transitioning to "
3304 				    "maintenance"), instance->fmri);
3305 				instance->fail_rate_count = 0;
3306 
3307 				destroy_bound_fds(instance);
3308 
3309 				instance->maintenance_req = B_TRUE;
3310 				(void) run_method(instance, IM_OFFLINE, NULL);
3311 				return;
3312 			}
3313 		}
3314 	}
3315 
3316 	ret = run_method(instance, IM_START, pi);
3317 
3318 	instance->conn_fd = -1;
3319 
3320 	if (ret == 0) {
3321 		/*
3322 		 * Stop listening for connections now we've fired off the
3323 		 * server for a wait type instance.
3324 		 */
3325 		(void) poll_bound_fds(instance, B_FALSE);
3326 	}
3327 }
3328 
3329 /*
3330  * Process any networks requests for each proto for each instance.
3331  */
3332 void
3333 process_network_events(void)
3334 {
3335 	instance_t	*instance;
3336 
3337 	debug_msg("Entering process_network_events");
3338 
3339 	for (instance = uu_list_first(instance_list); instance != NULL;
3340 	    instance = uu_list_next(instance_list, instance)) {
3341 		basic_cfg_t	*cfg;
3342 		proto_info_t	*pi;
3343 
3344 		/*
3345 		 * Ignore instances in states that definitely don't have any
3346 		 * listening fds.
3347 		 */
3348 		switch (instance->cur_istate) {
3349 		case IIS_ONLINE:
3350 		case IIS_DEGRADED:
3351 		case IIS_IN_REFRESH_METHOD:
3352 			break;
3353 		default:
3354 			continue;
3355 		}
3356 
3357 		cfg = instance->config->basic;
3358 
3359 		for (pi = uu_list_first(cfg->proto_list); pi != NULL;
3360 		    pi = uu_list_next(cfg->proto_list, pi)) {
3361 			if ((pi->listen_fd != -1) &&
3362 			    isset_pollfd(pi->listen_fd)) {
3363 				if (cfg->iswait) {
3364 					process_wait_request(instance, pi);
3365 				} else {
3366 					process_nowait_request(instance, pi);
3367 				}
3368 			}
3369 		}
3370 	}
3371 }
3372 
3373 /* ARGSUSED0 */
3374 static void
3375 sigterm_handler(int sig)
3376 {
3377 	debug_msg("Entering sigterm_handler");
3378 
3379 	got_sigterm = B_TRUE;
3380 }
3381 
3382 /* ARGSUSED0 */
3383 static void
3384 sighup_handler(int sig)
3385 {
3386 	debug_msg("Entering sighup_handler");
3387 
3388 	refresh_inetd_requested = B_TRUE;
3389 }
3390 
3391 /*
3392  * inetd's major work loop. This function sits in poll waiting for events
3393  * to occur, processing them when they do. The possible events are
3394  * master restarter requests, expired timer queue timers, stop/refresh signal
3395  * requests, contract events indicating process termination, stop/refresh
3396  * requests originating from one of the stop/refresh inetd processes and
3397  * network events.
3398  * The loop is exited when a stop request is received and processed, and
3399  * all the instances have reached a suitable 'stopping' state.
3400  */
3401 static void
3402 event_loop(void)
3403 {
3404 	instance_t		*instance;
3405 	int			timeout;
3406 
3407 	debug_msg("Entering event_loop");
3408 
3409 	for (;;) {
3410 		int	pret = -1;
3411 
3412 		timeout = iu_earliest_timer(timer_queue);
3413 
3414 		debug_msg("Doing signal check/poll");
3415 		if (!got_sigterm && !refresh_inetd_requested) {
3416 			pret = poll(poll_fds, num_pollfds, timeout);
3417 			if ((pret == -1) && (errno != EINTR)) {
3418 				error_msg(gettext("poll failure: %s"),
3419 				    strerror(errno));
3420 				continue;
3421 			}
3422 			debug_msg("Exiting poll, returned: %d", pret);
3423 		}
3424 
3425 		if (got_sigterm) {
3426 			msg_fini();
3427 			inetd_stop();
3428 			got_sigterm = B_FALSE;
3429 			goto check_if_stopped;
3430 		}
3431 
3432 		/*
3433 		 * Process any stop/refresh requests from the Unix Domain
3434 		 * Socket.
3435 		 */
3436 		if ((pret != -1) && isset_pollfd(uds_fd)) {
3437 			while (process_uds_event() == 0)
3438 				;
3439 		}
3440 
3441 		/*
3442 		 * Process refresh request. We do this check after the UDS
3443 		 * event check above, as it would be wasted processing if we
3444 		 * started refreshing inetd based on a SIGHUP, and then were
3445 		 * told to shut-down via a UDS event.
3446 		 */
3447 		if (refresh_inetd_requested) {
3448 			refresh_inetd_requested = B_FALSE;
3449 			if (!inetd_stopping)
3450 				inetd_refresh();
3451 		}
3452 
3453 		/*
3454 		 * We were interrupted by a signal. Don't waste any more
3455 		 * time processing a potentially inaccurate poll return.
3456 		 */
3457 		if (pret == -1)
3458 			continue;
3459 
3460 		/*
3461 		 * Process any instance restarter events.
3462 		 */
3463 		if (isset_pollfd(rst_event_pipe[PE_CONSUMER])) {
3464 			while (process_restarter_event() == 0)
3465 				;
3466 		}
3467 
3468 		/*
3469 		 * Process any expired timers (bind retry, con-rate offline,
3470 		 * method timeouts).
3471 		 */
3472 		(void) iu_expire_timers(timer_queue);
3473 
3474 		process_terminated_methods();
3475 
3476 		/*
3477 		 * If inetd is stopping, check whether all our managed
3478 		 * instances have been stopped and we can return.
3479 		 */
3480 		if (inetd_stopping) {
3481 check_if_stopped:
3482 			for (instance = uu_list_first(instance_list);
3483 			    instance != NULL;
3484 			    instance = uu_list_next(instance_list, instance)) {
3485 				if (!instance_stopped(instance)) {
3486 					debug_msg("%s not yet stopped",
3487 					    instance->fmri);
3488 					break;
3489 				}
3490 			}
3491 			/* if all instances are stopped, return */
3492 			if (instance == NULL)
3493 				return;
3494 		}
3495 
3496 		process_network_events();
3497 	}
3498 }
3499 
3500 static void
3501 fini(void)
3502 {
3503 	debug_msg("Entering fini");
3504 
3505 	method_fini();
3506 	uds_fini();
3507 	if (timer_queue != NULL)
3508 		iu_tq_destroy(timer_queue);
3509 
3510 
3511 	/*
3512 	 * We don't bother to undo the restarter interface at all.
3513 	 * Because of quirks in the interface, there is no way to
3514 	 * disconnect from the channel and cause any new events to be
3515 	 * queued.  However, any events which are received and not
3516 	 * acknowledged will be re-sent when inetd restarts as long as inetd
3517 	 * uses the same subscriber ID, which it does.
3518 	 *
3519 	 * By keeping the event pipe open but ignoring it, any events which
3520 	 * occur will cause restarter_event_proxy to hang without breaking
3521 	 * anything.
3522 	 */
3523 
3524 	if (instance_list != NULL) {
3525 		void		*cookie = NULL;
3526 		instance_t	*inst;
3527 
3528 		while ((inst = uu_list_teardown(instance_list, &cookie)) !=
3529 		    NULL)
3530 			destroy_instance(inst);
3531 		uu_list_destroy(instance_list);
3532 	}
3533 	if (instance_pool != NULL)
3534 		uu_list_pool_destroy(instance_pool);
3535 	tlx_fini();
3536 	config_fini();
3537 	repval_fini();
3538 	poll_fini();
3539 
3540 	/* Close audit session */
3541 	(void) adt_end_session(audit_handle);
3542 }
3543 
3544 static int
3545 init(void)
3546 {
3547 	int err;
3548 
3549 	debug_msg("Entering init");
3550 
3551 	if (repval_init() < 0)
3552 		goto failed;
3553 
3554 	if (config_init() < 0)
3555 		goto failed;
3556 
3557 	if (tlx_init() < 0)
3558 		goto failed;
3559 
3560 	/* Setup instance list. */
3561 	if ((instance_pool = uu_list_pool_create("instance_pool",
3562 	    sizeof (instance_t), offsetof(instance_t, link), NULL,
3563 	    UU_LIST_POOL_DEBUG)) == NULL) {
3564 		error_msg("%s: %s",
3565 		    gettext("Failed to create instance pool"),
3566 		    uu_strerror(uu_error()));
3567 		goto failed;
3568 	}
3569 	if ((instance_list = uu_list_create(instance_pool, NULL, 0)) == NULL) {
3570 		error_msg("%s: %s",
3571 		    gettext("Failed to create instance list"),
3572 		    uu_strerror(uu_error()));
3573 		goto failed;
3574 	}
3575 
3576 	/*
3577 	 * Create event pipe to communicate events with the main event
3578 	 * loop and add it to the event loop's fdset.
3579 	 */
3580 	if (pipe(rst_event_pipe) < 0) {
3581 		error_msg("pipe: %s", strerror(errno));
3582 		goto failed;
3583 	}
3584 	/*
3585 	 * We only leave the producer end to block on reads/writes as we
3586 	 * can't afford to block in the main thread, yet need to in
3587 	 * the restarter event thread, so it can sit and wait for an
3588 	 * acknowledgement to be written to the pipe.
3589 	 */
3590 	disable_blocking(rst_event_pipe[PE_CONSUMER]);
3591 	if ((set_pollfd(rst_event_pipe[PE_CONSUMER], POLLIN)) == -1)
3592 		goto failed;
3593 
3594 	/*
3595 	 * Register with master restarter for managed service events. This
3596 	 * will fail, amongst other reasons, if inetd is already running.
3597 	 */
3598 	if ((err = restarter_bind_handle(RESTARTER_EVENT_VERSION,
3599 	    INETD_INSTANCE_FMRI, restarter_event_proxy, 0,
3600 	    &rst_event_handle)) != 0) {
3601 		error_msg(gettext(
3602 		    "Failed to register for restarter events: %s"),
3603 		    strerror(err));
3604 		goto failed;
3605 	}
3606 
3607 	if (contract_init() < 0)
3608 		goto failed;
3609 
3610 	if ((timer_queue = iu_tq_create()) == NULL) {
3611 		error_msg(gettext("Failed to create timer queue."));
3612 		goto failed;
3613 	}
3614 
3615 	if (uds_init() < 0)
3616 		goto failed;
3617 
3618 	if (method_init() < 0)
3619 		goto failed;
3620 
3621 	/* Initialize auditing session */
3622 	if (adt_start_session(&audit_handle, NULL, ADT_USE_PROC_DATA) != 0) {
3623 		error_msg(gettext("Unable to start audit session"));
3624 	}
3625 
3626 	/*
3627 	 * Initialize signal dispositions/masks
3628 	 */
3629 	(void) sigset(SIGHUP, sighup_handler);
3630 	(void) sigset(SIGTERM, sigterm_handler);
3631 	(void) sigignore(SIGINT);
3632 
3633 	return (0);
3634 
3635 failed:
3636 	fini();
3637 	return (-1);
3638 }
3639 
3640 static int
3641 start_method(void)
3642 {
3643 	int	i;
3644 	int	pipe_fds[2];
3645 	int	child;
3646 
3647 	debug_msg("ENTERING START_METHOD:");
3648 
3649 	/* Create pipe for child to notify parent of initialization success. */
3650 	if (pipe(pipe_fds) < 0) {
3651 		debug_msg("pipe: %s", strerror(errno));
3652 		return (SMF_EXIT_ERR_OTHER);
3653 	}
3654 
3655 	if ((child = fork()) == -1) {
3656 		debug_msg("fork: %s", strerror(errno));
3657 		(void) close(pipe_fds[PE_CONSUMER]);
3658 		(void) close(pipe_fds[PE_PRODUCER]);
3659 		return (SMF_EXIT_ERR_OTHER);
3660 	} else if (child > 0) {			/* parent */
3661 
3662 		/* Wait on child to return success of initialization. */
3663 		(void) close(pipe_fds[PE_PRODUCER]);
3664 		if ((safe_read(pipe_fds[PE_CONSUMER], &i, sizeof (i)) != 0) ||
3665 		    (i < 0)) {
3666 			error_msg(gettext(
3667 			    "Initialization failed, unable to start"));
3668 			(void) close(pipe_fds[PE_CONSUMER]);
3669 			/*
3670 			 * Batch all initialization errors as 'other' errors,
3671 			 * resulting in retries being attempted.
3672 			 */
3673 			return (SMF_EXIT_ERR_OTHER);
3674 		} else {
3675 			(void) close(pipe_fds[PE_CONSUMER]);
3676 			return (SMF_EXIT_OK);
3677 		}
3678 	} else {				/* child */
3679 		/*
3680 		 * Perform initialization and return success code down
3681 		 * the pipe.
3682 		 */
3683 		(void) close(pipe_fds[PE_CONSUMER]);
3684 		i = init();
3685 		if ((safe_write(pipe_fds[PE_PRODUCER], &i, sizeof (i)) < 0) ||
3686 		    (i < 0)) {
3687 			error_msg(gettext("pipe write failure: %s"),
3688 			    strerror(errno));
3689 			exit(1);
3690 		}
3691 		(void) close(pipe_fds[PE_PRODUCER]);
3692 
3693 		(void) setsid();
3694 
3695 		/*
3696 		 * Log a message if the configuration file has changed since
3697 		 * inetconv was last run.
3698 		 */
3699 		check_conf_file();
3700 
3701 		event_loop();
3702 
3703 		fini();
3704 		debug_msg("inetd stopped");
3705 		msg_fini();
3706 		exit(0);
3707 	}
3708 	/* NOTREACHED */
3709 }
3710 
3711 /*
3712  * When inetd is run from outside the SMF, this message is output to provide
3713  * the person invoking inetd with further information that will help them
3714  * understand how to start and stop inetd, and to achieve the other
3715  * behaviors achievable with the legacy inetd command line interface, if
3716  * it is possible.
3717  */
3718 static void
3719 legacy_usage(void)
3720 {
3721 	(void) fprintf(stderr,
3722 	    "inetd is now an smf(5) managed service and can no longer be run "
3723 	    "from the\n"
3724 	    "command line. To enable or disable inetd refer to svcadm(1M) on\n"
3725 	    "how to enable \"%s\", the inetd instance.\n"
3726 	    "\n"
3727 	    "The traditional inetd command line option mappings are:\n"
3728 	    "\t-d : there is no supported debug output\n"
3729 	    "\t-s : inetd is only runnable from within the SMF\n"
3730 	    "\t-t : See inetadm(1M) on how to enable TCP tracing\n"
3731 	    "\t-r : See inetadm(1M) on how to set a failure rate\n"
3732 	    "\n"
3733 	    "To specify an alternative configuration file see svccfg(1M)\n"
3734 	    "for how to modify the \"%s/%s\" string type property of\n"
3735 	    "the inetd instance, and modify it according to the syntax:\n"
3736 	    "\"%s [alt_config_file] %%m\".\n"
3737 	    "\n"
3738 	    "For further information on inetd see inetd(1M).\n",
3739 	    INETD_INSTANCE_FMRI, START_METHOD_ARG, SCF_PROPERTY_EXEC,
3740 	    INETD_PATH);
3741 }
3742 
3743 /*
3744  * Usage message printed out for usage errors when running under the SMF.
3745  */
3746 static void
3747 smf_usage(const char *arg0)
3748 {
3749 	error_msg("Usage: %s [alt_conf_file] %s|%s|%s", arg0, START_METHOD_ARG,
3750 	    STOP_METHOD_ARG, REFRESH_METHOD_ARG);
3751 }
3752 
3753 /*
3754  * Returns B_TRUE if we're being run from within the SMF, else B_FALSE.
3755  */
3756 static boolean_t
3757 run_through_smf(void)
3758 {
3759 	char *fmri;
3760 
3761 	/*
3762 	 * check if the instance fmri environment variable has been set by
3763 	 * our restarter.
3764 	 */
3765 	return (((fmri = getenv("SMF_FMRI")) != NULL) &&
3766 	    (strcmp(fmri, INETD_INSTANCE_FMRI) == 0));
3767 }
3768 
3769 int
3770 main(int argc, char *argv[])
3771 {
3772 	char		*method;
3773 	int		ret;
3774 
3775 #if	!defined(TEXT_DOMAIN)
3776 #define	TEXT_DOMAIN "SYS_TEST"
3777 #endif
3778 	(void) textdomain(TEXT_DOMAIN);
3779 	(void) setlocale(LC_ALL, "");
3780 
3781 	if (!run_through_smf()) {
3782 		legacy_usage();
3783 		return (SMF_EXIT_ERR_NOSMF);
3784 	}
3785 
3786 	msg_init();	/* setup logging */
3787 
3788 	/* inetd invocation syntax is inetd [alt_conf_file] method_name */
3789 
3790 	switch (argc) {
3791 	case 2:
3792 		method = argv[1];
3793 		break;
3794 	case 3:
3795 		conf_file = argv[1];
3796 		method = argv[2];
3797 		break;
3798 	default:
3799 		smf_usage(argv[0]);
3800 		return (SMF_EXIT_ERR_CONFIG);
3801 
3802 	}
3803 
3804 	if (strcmp(method, START_METHOD_ARG) == 0) {
3805 		ret = start_method();
3806 	} else if (strcmp(method, STOP_METHOD_ARG) == 0) {
3807 		ret = stop_method();
3808 	} else if (strcmp(method, REFRESH_METHOD_ARG) == 0) {
3809 		ret = refresh_method();
3810 	} else {
3811 		smf_usage(argv[0]);
3812 		return (SMF_EXIT_ERR_CONFIG);
3813 	}
3814 
3815 	return (ret);
3816 }
3817