xref: /titanic_44/usr/src/cmd/cmd-inet/usr.lib/in.mpathd/mpd_main.c (revision 136097cead98e48ee9260b7cc3b7edd749c5c808)
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 (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include "mpd_defs.h"
29 #include "mpd_tables.h"
30 
31 int debug = 0;				/* Debug flag */
32 static int pollfd_num = 0;		/* Num. of poll descriptors */
33 static struct pollfd *pollfds = NULL;	/* Array of poll descriptors */
34 
35 					/* All times below in ms */
36 int	user_failure_detection_time;	/* user specified failure detection */
37 					/* time (fdt) */
38 int	user_probe_interval;		/* derived from user specified fdt */
39 
40 static int	rtsock_v4;		/* AF_INET routing socket */
41 static int	rtsock_v6;		/* AF_INET6 routing socket */
42 int	ifsock_v4 = -1;			/* IPv4 socket for ioctls  */
43 int	ifsock_v6 = -1;			/* IPv6 socket for ioctls  */
44 static int	lsock_v4;		/* Listen socket to detect mpathd */
45 static int	lsock_v6;		/* Listen socket to detect mpathd */
46 static int	mibfd = -1;		/* fd to get mib info */
47 static boolean_t force_mcast = _B_FALSE; /* Only for test purposes */
48 
49 boolean_t	full_scan_required = _B_FALSE;
50 static uint_t	last_initifs_time;	/* Time when initifs was last run */
51 static	char **argv0;			/* Saved for re-exec on SIGHUP */
52 boolean_t handle_link_notifications = _B_TRUE;
53 
54 static void	initlog(void);
55 static void	run_timeouts(void);
56 static void	initifs(void);
57 static void	check_if_removed(struct phyint_instance *pii);
58 static void	select_test_ifs(void);
59 static void	ire_process_v4(mib2_ipRouteEntry_t *buf, size_t len);
60 static void	ire_process_v6(mib2_ipv6RouteEntry_t *buf, size_t len);
61 static void	router_add_v4(mib2_ipRouteEntry_t *rp1,
62     struct in_addr nexthop_v4);
63 static void	router_add_v6(mib2_ipv6RouteEntry_t *rp1,
64     struct in6_addr nexthop_v6);
65 static void	router_add_common(int af, char *ifname,
66     struct in6_addr nexthop);
67 static void	init_router_targets();
68 static void	cleanup(void);
69 static int	setup_listener(int af);
70 static void	check_config(void);
71 static void	check_addr_unique(int af, char *name);
72 static void	init_host_targets(void);
73 static void	dup_host_targets(struct phyint_instance *desired_pii);
74 static void	loopback_cmd(int sock, int family);
75 static int	poll_remove(int fd);
76 static boolean_t daemonize(void);
77 static int	closefunc(void *, int);
78 static unsigned int process_cmd(int newfd, union mi_commands *mpi);
79 static unsigned int process_query(int fd, mi_query_t *miq);
80 static unsigned int send_groupinfo(int fd, ipmp_groupinfo_t *grinfop);
81 static unsigned int send_grouplist(int fd, ipmp_grouplist_t *grlistp);
82 static unsigned int send_ifinfo(int fd, ipmp_ifinfo_t *ifinfop);
83 static unsigned int send_result(int fd, unsigned int error, int syserror);
84 
85 struct local_addr *laddr_list = NULL;
86 
87 /*
88  * Return the current time in milliseconds (from an arbitrary reference)
89  * truncated to fit into an int. Truncation is ok since we are interested
90  * only in differences and not the absolute values.
91  */
92 uint_t
93 getcurrenttime(void)
94 {
95 	uint_t	cur_time;	/* In ms */
96 
97 	/*
98 	 * Use of a non-user-adjustable source of time is
99 	 * required. However millisecond precision is sufficient.
100 	 * divide by 10^6
101 	 */
102 	cur_time = (uint_t)(gethrtime() / 1000000LL);
103 	return (cur_time);
104 }
105 
106 /*
107  * Add fd to the set being polled. Returns 0 if ok; -1 if failed.
108  */
109 int
110 poll_add(int fd)
111 {
112 	int i;
113 	int new_num;
114 	struct pollfd *newfds;
115 retry:
116 	/* Check if already present */
117 	for (i = 0; i < pollfd_num; i++) {
118 		if (pollfds[i].fd == fd)
119 			return (0);
120 	}
121 	/* Check for empty spot already present */
122 	for (i = 0; i < pollfd_num; i++) {
123 		if (pollfds[i].fd == -1) {
124 			pollfds[i].fd = fd;
125 			return (0);
126 		}
127 	}
128 
129 	/* Allocate space for 32 more fds and initialize to -1 */
130 	new_num = pollfd_num + 32;
131 	newfds = realloc(pollfds, new_num * sizeof (struct pollfd));
132 	if (newfds == NULL) {
133 		logperror("poll_add: realloc");
134 		return (-1);
135 	}
136 	for (i = pollfd_num; i < new_num; i++) {
137 		newfds[i].fd = -1;
138 		newfds[i].events = POLLIN;
139 	}
140 	pollfd_num = new_num;
141 	pollfds = newfds;
142 	goto retry;
143 }
144 
145 /*
146  * Remove fd from the set being polled. Returns 0 if ok; -1 if failed.
147  */
148 static int
149 poll_remove(int fd)
150 {
151 	int i;
152 
153 	/* Check if already present */
154 	for (i = 0; i < pollfd_num; i++) {
155 		if (pollfds[i].fd == fd) {
156 			pollfds[i].fd = -1;
157 			return (0);
158 		}
159 	}
160 	return (-1);
161 }
162 
163 /*
164  * Extract information about the phyint instance. If the phyint instance still
165  * exists in the kernel then set pii_in_use, else clear it. check_if_removed()
166  * will use it to detect phyint instances that don't exist any longer and
167  * remove them, from our database of phyint instances.
168  * Return value:
169  *	returns true if the phyint instance exists in the kernel,
170  *	returns false otherwise
171  */
172 static boolean_t
173 pii_process(int af, char *name, struct phyint_instance **pii_p)
174 {
175 	int err;
176 	struct phyint_instance *pii;
177 	struct phyint_instance *pii_other;
178 
179 	if (debug & D_PHYINT)
180 		logdebug("pii_process(%s %s)\n", AF_STR(af), name);
181 
182 	pii = phyint_inst_lookup(af, name);
183 	if (pii == NULL) {
184 		/*
185 		 * Phyint instance does not exist in our tables,
186 		 * create new phyint instance
187 		 */
188 		pii = phyint_inst_init_from_k(af, name);
189 	} else {
190 		/* Phyint exists in our tables */
191 		err = phyint_inst_update_from_k(pii);
192 
193 		switch (err) {
194 		case PI_IOCTL_ERROR:
195 			/* Some ioctl error. don't change anything */
196 			pii->pii_in_use = 1;
197 			break;
198 
199 		case PI_GROUP_CHANGED:
200 			/*
201 			 * The phyint has changed group.
202 			 */
203 			restore_phyint(pii->pii_phyint);
204 			/* FALLTHRU */
205 
206 		case PI_IFINDEX_CHANGED:
207 			/*
208 			 * Interface index has changed. Delete and
209 			 * recreate the phyint as it is quite likely
210 			 * the interface has been unplumbed and replumbed.
211 			 */
212 			pii_other = phyint_inst_other(pii);
213 			if (pii_other != NULL)
214 				phyint_inst_delete(pii_other);
215 			phyint_inst_delete(pii);
216 			pii = phyint_inst_init_from_k(af, name);
217 			break;
218 
219 		case PI_DELETED:
220 			/* Phyint instance has disappeared from kernel */
221 			pii->pii_in_use = 0;
222 			break;
223 
224 		case PI_OK:
225 			/* Phyint instance exists and is fine */
226 			pii->pii_in_use = 1;
227 			break;
228 
229 		default:
230 			/* Unknown status */
231 			logerr("pii_process: Unknown status %d\n", err);
232 			break;
233 		}
234 	}
235 
236 	*pii_p = pii;
237 	if (pii != NULL)
238 		return (pii->pii_in_use ? _B_TRUE : _B_FALSE);
239 	else
240 		return (_B_FALSE);
241 }
242 
243 /*
244  * This phyint is leaving the group. Try to restore the phyint to its
245  * initial state. Return the addresses that belong to other group members,
246  * to the group, and take back any addresses owned by this phyint
247  */
248 void
249 restore_phyint(struct phyint *pi)
250 {
251 	if (pi->pi_group == phyint_anongroup)
252 		return;
253 
254 	/*
255 	 * Move everthing to some other member in the group.
256 	 * The phyint has changed group in the kernel. But we
257 	 * have yet to do it in our tables.
258 	 */
259 	if (!pi->pi_empty)
260 		(void) try_failover(pi, FAILOVER_TO_ANY);
261 	/*
262 	 * Move all addresses owned by 'pi' back to pi, from each
263 	 * of the other members of the group
264 	 */
265 	(void) try_failback(pi, _B_FALSE);
266 }
267 
268 /*
269  * Scan all interfaces to detect changes as well as new and deleted interfaces
270  */
271 static void
272 initifs()
273 {
274 	int	n;
275 	int	af;
276 	char	*cp;
277 	char	*buf;
278 	int	numifs;
279 	struct lifnum	lifn;
280 	struct lifconf	lifc;
281 	struct lifreq	*lifr;
282 	struct logint	*li;
283 	struct phyint_instance *pii;
284 	struct phyint_instance *next_pii;
285 	char	pi_name[LIFNAMSIZ + 1];
286 	boolean_t exists;
287 	struct phyint	*pi;
288 	struct local_addr *next;
289 
290 	if (debug & D_PHYINT)
291 		logdebug("initifs: Scanning interfaces\n");
292 
293 	last_initifs_time = getcurrenttime();
294 
295 	/*
296 	 * Free the laddr_list before collecting the local addresses.
297 	 */
298 	while (laddr_list != NULL) {
299 		next = laddr_list->next;
300 		free(laddr_list);
301 		laddr_list = next;
302 	}
303 
304 	/*
305 	 * Mark the interfaces so that we can find phyints and logints
306 	 * which have disappeared from the kernel. pii_process() and
307 	 * logint_init_from_k() will set {pii,li}_in_use when they find
308 	 * the interface in the kernel. Also, clear dupaddr bit on probe
309 	 * logint. check_addr_unique() will set the dupaddr bit on the
310 	 * probe logint, if the testaddress is not unique.
311 	 */
312 	for (pii = phyint_instances; pii != NULL; pii = pii->pii_next) {
313 		pii->pii_in_use = 0;
314 		for (li = pii->pii_logint; li != NULL; li = li->li_next) {
315 			li->li_in_use = 0;
316 			if (pii->pii_probe_logint == li)
317 				li->li_dupaddr = 0;
318 		}
319 	}
320 
321 	lifn.lifn_family = AF_UNSPEC;
322 	lifn.lifn_flags = LIFC_ALLZONES;
323 	if (ioctl(ifsock_v4, SIOCGLIFNUM, (char *)&lifn) < 0) {
324 		logperror("initifs: ioctl (get interface numbers)");
325 		return;
326 	}
327 	numifs = lifn.lifn_count;
328 
329 	buf = (char *)calloc(numifs, sizeof (struct lifreq));
330 	if (buf == NULL) {
331 		logperror("initifs: calloc");
332 		return;
333 	}
334 
335 	lifc.lifc_family = AF_UNSPEC;
336 	lifc.lifc_flags = LIFC_ALLZONES;
337 	lifc.lifc_len = numifs * sizeof (struct lifreq);
338 	lifc.lifc_buf = buf;
339 
340 	if (ioctl(ifsock_v4, SIOCGLIFCONF, (char *)&lifc) < 0) {
341 		/*
342 		 * EINVAL is commonly encountered, when things change
343 		 * underneath us rapidly, (eg. at boot, when new interfaces
344 		 * are plumbed successively) and the kernel finds the buffer
345 		 * size we passed as too small. We will retry again
346 		 * when we see the next routing socket msg, or at worst after
347 		 * IF_SCAN_INTERVAL ms.
348 		 */
349 		if (errno != EINVAL) {
350 			logperror("initifs: ioctl"
351 			    " (get interface configuration)");
352 		}
353 		free(buf);
354 		return;
355 	}
356 
357 	lifr = (struct lifreq *)lifc.lifc_req;
358 
359 	/*
360 	 * For each lifreq returned by SIOGGLIFCONF, call pii_process()
361 	 * and get the state of the corresponding phyint_instance. If it is
362 	 * successful, then call logint_init_from_k() to get the state of the
363 	 * logint.
364 	 */
365 	for (n = lifc.lifc_len / sizeof (struct lifreq); n > 0; n--, lifr++) {
366 		int	sockfd;
367 		struct local_addr	*taddr;
368 		struct sockaddr_in	*sin;
369 		struct sockaddr_in6	*sin6;
370 		struct lifreq	lifreq;
371 
372 		af = lifr->lifr_addr.ss_family;
373 
374 		/*
375 		 * Collect all local addresses.
376 		 */
377 		sockfd = (af == AF_INET) ? ifsock_v4 : ifsock_v6;
378 		(void) memset(&lifreq, 0, sizeof (lifreq));
379 		(void) strlcpy(lifreq.lifr_name, lifr->lifr_name,
380 		    sizeof (lifreq.lifr_name));
381 
382 		if (ioctl(sockfd, SIOCGLIFFLAGS, &lifreq) == -1) {
383 			if (errno != ENXIO)
384 				logperror("initifs: ioctl (SIOCGLIFFLAGS)");
385 			continue;
386 		}
387 
388 		/*
389 		 * Add the interface address to laddr_list.
390 		 * Another node might have the same IP address which is up.
391 		 * In that case, it is appropriate  to use the address as a
392 		 * target, even though it is also configured (but not up) on
393 		 * the local system.
394 		 * Hence,the interface address is not added to laddr_list
395 		 * unless it is IFF_UP.
396 		 */
397 		if (lifreq.lifr_flags & IFF_UP) {
398 			taddr = malloc(sizeof (struct local_addr));
399 			if (taddr == NULL) {
400 				logperror("initifs: malloc");
401 				continue;
402 			}
403 			if (af == AF_INET) {
404 				sin = (struct sockaddr_in *)&lifr->lifr_addr;
405 				IN6_INADDR_TO_V4MAPPED(&sin->sin_addr,
406 				    &taddr->addr);
407 			} else {
408 				sin6 = (struct sockaddr_in6 *)&lifr->lifr_addr;
409 				taddr->addr = sin6->sin6_addr;
410 			}
411 			taddr->next = laddr_list;
412 			laddr_list = taddr;
413 		}
414 
415 		/*
416 		 * Need to pass a phyint name to pii_process. Insert the
417 		 * null where the ':' IF_SEPARATOR is found in the logical
418 		 * name.
419 		 */
420 		(void) strlcpy(pi_name, lifr->lifr_name, sizeof (pi_name));
421 		if ((cp = strchr(pi_name, IF_SEPARATOR)) != NULL)
422 			*cp = '\0';
423 
424 		exists = pii_process(af, pi_name, &pii);
425 		if (exists) {
426 			/* The phyint is fine. So process the logint */
427 			logint_init_from_k(pii, lifr->lifr_name);
428 		}
429 		check_addr_unique(af, lifr->lifr_name);
430 	}
431 
432 	free(buf);
433 
434 	/*
435 	 * If the test address is now unique, and if it was not unique
436 	 * previously,	clear the li_dupaddrmsg_printed flag and log a
437 	 * recovery message
438 	 */
439 	for (pii = phyint_instances; pii != NULL; pii = pii->pii_next) {
440 		struct logint *li;
441 		char abuf[INET6_ADDRSTRLEN];
442 
443 		li = pii->pii_probe_logint;
444 		if ((li != NULL) && !li->li_dupaddr &&
445 		    li->li_dupaddrmsg_printed) {
446 			logerr("Test address %s is unique; enabling probe-"
447 			    "based failure detection\n",
448 			    pr_addr(pii->pii_af, li->li_addr, abuf,
449 				sizeof (abuf)));
450 			li->li_dupaddrmsg_printed = 0;
451 		}
452 	}
453 
454 	/*
455 	 * Scan for phyints and logints that have disappeared from the
456 	 * kernel, and delete them.
457 	 */
458 	pii = phyint_instances;
459 
460 	while (pii != NULL) {
461 		next_pii = pii->pii_next;
462 		check_if_removed(pii);
463 		pii = next_pii;
464 	}
465 
466 	/*
467 	 * Select a test address for sending probes on each phyint instance
468 	 */
469 	select_test_ifs();
470 
471 	/*
472 	 * Handle link up/down notifications from the NICs.
473 	 */
474 	process_link_state_changes();
475 
476 	for (pi = phyints; pi != NULL; pi = pi->pi_next) {
477 		/*
478 		 * If this is a case of group failure, we don't have much
479 		 * to do until the group recovers again.
480 		 */
481 		if (GROUP_FAILED(pi->pi_group))
482 			continue;
483 
484 		/*
485 		 * Try/Retry any pending failovers / failbacks, that did not
486 		 * not complete, or that could not be initiated previously.
487 		 * This implements the 3 invariants described in the big block
488 		 * comment at the beginning of probe.c
489 		 */
490 		if (pi->pi_flags & IFF_INACTIVE) {
491 			if (!pi->pi_empty && (pi->pi_flags & IFF_STANDBY))
492 				(void) try_failover(pi, FAILOVER_TO_NONSTANDBY);
493 		} else {
494 			struct phyint_instance *pii;
495 
496 			pii = pi->pi_v4;
497 			if (LINK_UP(pi) && !PROBE_CAPABLE(pii))
498 				pii = pi->pi_v6;
499 			if (LINK_UP(pi) && !PROBE_CAPABLE(pii))
500 				continue;
501 			/*
502 			 * It is possible that the phyint has started
503 			 * receiving packets, after it has been marked
504 			 * PI_FAILED. Don't initiate failover, if the
505 			 * phyint has started recovering. failure_state()
506 			 * captures this check. A similar logic is used
507 			 * for failback/repair case.
508 			 */
509 			if (pi->pi_state == PI_FAILED && !pi->pi_empty &&
510 			    (failure_state(pii) == PHYINT_FAILURE)) {
511 				(void) try_failover(pi, FAILOVER_NORMAL);
512 			} else if (pi->pi_state == PI_RUNNING && !pi->pi_full) {
513 				if (try_failback(pi, _B_FALSE) !=
514 				    IPMP_FAILURE) {
515 					(void) change_lif_flags(pi, IFF_FAILED,
516 					    _B_FALSE);
517 					/* Per state diagram */
518 					pi->pi_empty = 0;
519 				}
520 			}
521 		}
522 	}
523 }
524 
525 /*
526  * Check that test/probe addresses are always unique. link-locals and
527  * ptp unnumbered may not be unique, and bind to such an (IFF_NOFAILOVER)
528  * address can produce unexpected results. Log an error and alert the user.
529  */
530 static void
531 check_addr_unique(int af, char *name)
532 {
533 	struct lifreq	lifr;
534 	struct phyint	*pi;
535 	struct in6_addr	addr;
536 	struct phyint_instance	*pii;
537 	struct sockaddr_in	*sin;
538 	struct sockaddr_in6	*sin6;
539 	int ifsock;
540 	char abuf[INET6_ADDRSTRLEN];
541 
542 	/* Get the socket for doing ioctls */
543 	ifsock = (af == AF_INET) ? ifsock_v4 : ifsock_v6;
544 
545 	(void) strncpy(lifr.lifr_name, name, sizeof (lifr.lifr_name));
546 	lifr.lifr_name[sizeof (lifr.lifr_name) - 1] = '\0';
547 	/*
548 	 * Get the address corresponding to 'name'. We cannot
549 	 * do a logint lookup in our tables, because, not all logints
550 	 * in the system are tracked by mpathd. (eg. things not in a group)
551 	 */
552 	if (ioctl(ifsock, SIOCGLIFADDR, (char *)&lifr) < 0) {
553 		if (errno == ENXIO) {
554 			/* Interface has vanished */
555 			return;
556 		} else {
557 			logperror("ioctl (get addr)");
558 			return;
559 		}
560 	}
561 
562 	if (af == AF_INET) {
563 		sin = (struct sockaddr_in *)&lifr.lifr_addr;
564 		IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &addr);
565 	} else {
566 		sin6 = (struct sockaddr_in6 *)&lifr.lifr_addr;
567 		addr = sin6->sin6_addr;
568 	}
569 
570 	/*
571 	 * Does the address 'addr' match any known test address ? If so
572 	 * it is a duplicate, unless we are looking at the same logint
573 	 */
574 	for (pi = phyints; pi != NULL; pi = pi->pi_next) {
575 		pii = PHYINT_INSTANCE(pi, af);
576 		if (pii == NULL || pii->pii_probe_logint == NULL)
577 			continue;
578 
579 		if (!IN6_ARE_ADDR_EQUAL(&addr,
580 		    &pii->pii_probe_logint->li_addr)) {
581 			continue;
582 		}
583 
584 		if (strncmp(pii->pii_probe_logint->li_name, name,
585 		    sizeof (pii->pii_probe_logint->li_name)) == 0) {
586 			continue;
587 		}
588 
589 		/*
590 		 * This test address is not unique. Set the dupaddr bit
591 		 */
592 		pii->pii_probe_logint->li_dupaddr = 1;
593 
594 		/*
595 		 * Log an error message if not already logged
596 		 */
597 		if (pii->pii_probe_logint->li_dupaddrmsg_printed)
598 			continue;
599 
600 		logerr("Test address %s is not unique; disabling "
601 		    "probe-based failure detection\n",
602 		    pr_addr(af, addr, abuf, sizeof (abuf)));
603 
604 		pii->pii_probe_logint->li_dupaddrmsg_printed = 1;
605 	}
606 }
607 
608 /*
609  * Stop probing an interface.  Called when an interface is offlined.
610  * The probe socket is closed on each interface instance, and the
611  * interface state set to PI_OFFLINE.
612  */
613 static void
614 stop_probing(struct phyint *pi)
615 {
616 	struct phyint_instance *pii;
617 
618 	pii = pi->pi_v4;
619 	if (pii != NULL) {
620 		if (pii->pii_probe_sock != -1)
621 			close_probe_socket(pii, _B_TRUE);
622 		pii->pii_probe_logint = NULL;
623 	}
624 
625 	pii = pi->pi_v6;
626 	if (pii != NULL) {
627 		if (pii->pii_probe_sock != -1)
628 			close_probe_socket(pii, _B_TRUE);
629 		pii->pii_probe_logint = NULL;
630 	}
631 
632 	phyint_chstate(pi, PI_OFFLINE);
633 }
634 
635 enum { BAD_TESTFLAGS, OK_TESTFLAGS, BEST_TESTFLAGS };
636 
637 /*
638  * Rate the provided test flags.  By definition, IFF_NOFAILOVER must be set.
639  * IFF_UP must also be set so that the associated address can be used as a
640  * source address.  Further, we must be able to exchange packets with local
641  * destinations, so IFF_NOXMIT and IFF_NOLOCAL must be clear.  For historical
642  * reasons, we have a proclivity for IFF_DEPRECATED IPv4 test addresses.
643  */
644 static int
645 rate_testflags(uint64_t flags)
646 {
647 	if ((flags & (IFF_NOFAILOVER | IFF_UP)) != (IFF_NOFAILOVER | IFF_UP))
648 		return (BAD_TESTFLAGS);
649 
650 	if ((flags & (IFF_NOXMIT | IFF_NOLOCAL)) != 0)
651 		return (BAD_TESTFLAGS);
652 
653 	if ((flags & (IFF_IPV6 | IFF_DEPRECATED)) == IFF_DEPRECATED)
654 		return (BEST_TESTFLAGS);
655 
656 	if ((flags & (IFF_IPV6 | IFF_DEPRECATED)) == IFF_IPV6)
657 		return (BEST_TESTFLAGS);
658 
659 	return (OK_TESTFLAGS);
660 }
661 
662 /*
663  * Attempt to select a test address for each phyint instance.
664  * Call phyint_inst_sockinit() to complete the initializations.
665  */
666 static void
667 select_test_ifs(void)
668 {
669 	struct phyint		*pi;
670 	struct phyint_instance	*pii;
671 	struct phyint_instance	*next_pii;
672 	struct logint		*li;
673 	struct logint  		*probe_logint;
674 	boolean_t		target_scan_reqd = _B_FALSE;
675 	struct target		*tg;
676 	int			rating;
677 
678 	if (debug & D_PHYINT)
679 		logdebug("select_test_ifs\n");
680 
681 	/*
682 	 * For each phyint instance, do the test address selection
683 	 */
684 	for (pii = phyint_instances; pii != NULL; pii = next_pii) {
685 		next_pii = pii->pii_next;
686 		probe_logint = NULL;
687 
688 		/*
689 		 * An interface that is offline, should not be probed.
690 		 * Offline interfaces should always in PI_OFFLINE state,
691 		 * unless some other entity has set the offline flag.
692 		 */
693 		if (pii->pii_phyint->pi_flags & IFF_OFFLINE) {
694 			if (pii->pii_phyint->pi_state != PI_OFFLINE) {
695 				logerr("shouldn't be probing offline"
696 					" interface %s (state is: %u)."
697 					" Stopping probes.\n",
698 					pii->pii_phyint->pi_name,
699 					pii->pii_phyint->pi_state);
700 				stop_probing(pii->pii_phyint);
701 			}
702 			continue;
703 		}
704 
705 		li = pii->pii_probe_logint;
706 		if (li != NULL) {
707 			/*
708 			 * We've already got a test address; only proceed
709 			 * if it's suboptimal.
710 			 */
711 			if (rate_testflags(li->li_flags) == BEST_TESTFLAGS)
712 				continue;
713 		}
714 
715 		/*
716 		 * Walk the logints of this phyint instance, and select
717 		 * the best available test address
718 		 */
719 		for (li = pii->pii_logint; li != NULL; li = li->li_next) {
720 			/*
721 			 * Skip any IPv6 logints that are not link-local,
722 			 * since we should always have a link-local address
723 			 * anyway and in6_data() expects link-local replies.
724 			 */
725 			if (pii->pii_af == AF_INET6 &&
726 			    !IN6_IS_ADDR_LINKLOCAL(&li->li_addr))
727 				continue;
728 
729 			/*
730 			 * Rate the testflags. If we've found an optimal
731 			 * match, then break out; otherwise, record the most
732 			 * recent OK one.
733 			 */
734 			rating = rate_testflags(li->li_flags);
735 			if (rating == BAD_TESTFLAGS)
736 				continue;
737 
738 			probe_logint = li;
739 			if (rating == BEST_TESTFLAGS)
740 				break;
741 		}
742 
743 		/*
744 		 * If the probe logint has changed, ditch the old one.
745 		 */
746 		if (pii->pii_probe_logint != NULL &&
747 		    pii->pii_probe_logint != probe_logint) {
748 			if (pii->pii_probe_sock != -1)
749 				close_probe_socket(pii, _B_TRUE);
750 			pii->pii_probe_logint = NULL;
751 		}
752 
753 		if (probe_logint == NULL) {
754 			/*
755 			 * We don't have a test address. Don't print an
756 			 * error message immediately. check_config() will
757 			 * take care of it. Zero out the probe stats array
758 			 * since it is no longer relevant. Optimize by
759 			 * checking if it is already zeroed out.
760 			 */
761 			int pr_ndx;
762 
763 			pr_ndx = PROBE_INDEX_PREV(pii->pii_probe_next);
764 			if (pii->pii_probes[pr_ndx].pr_status != PR_UNUSED) {
765 				clear_pii_probe_stats(pii);
766 				reset_crtt_all(pii->pii_phyint);
767 			}
768 			continue;
769 		} else if (probe_logint == pii->pii_probe_logint) {
770 			/*
771 			 * If we didn't find any new test addr, go to the
772 			 * next phyint.
773 			 */
774 			continue;
775 		}
776 
777 		/*
778 		 * The phyint is either being assigned a new testaddr
779 		 * or is being assigned a testaddr for the 1st time.
780 		 * Need to initialize the phyint socket
781 		 */
782 		pii->pii_probe_logint = probe_logint;
783 		if (!phyint_inst_sockinit(pii)) {
784 			if (debug & D_PHYINT) {
785 				logdebug("select_test_ifs: "
786 				    "phyint_sockinit failed\n");
787 			}
788 			phyint_inst_delete(pii);
789 			continue;
790 		}
791 
792 		/*
793 		 * This phyint instance is now enabled for probes; this
794 		 * impacts our state machine in two ways:
795 		 *
796 		 * 1. If we're probe *capable* as well (i.e., we have
797 		 *    probe targets) and the interface is in PI_NOTARGETS,
798 		 *    then transition to PI_RUNNING.
799 		 *
800 		 * 2. If we're not probe capable, and the other phyint
801 		 *    instance is also not probe capable, and we were in
802 		 *    PI_RUNNING, then transition to PI_NOTARGETS.
803 		 *
804 		 * Also see the state diagram in mpd_probe.c.
805 		 */
806 		if (PROBE_CAPABLE(pii)) {
807 			if (pii->pii_phyint->pi_state == PI_NOTARGETS)
808 				phyint_chstate(pii->pii_phyint, PI_RUNNING);
809 		} else if (!PROBE_CAPABLE(phyint_inst_other(pii))) {
810 			if (pii->pii_phyint->pi_state == PI_RUNNING)
811 				phyint_chstate(pii->pii_phyint, PI_NOTARGETS);
812 		}
813 
814 		if (pii->pii_phyint->pi_flags & IFF_POINTOPOINT) {
815 			tg = pii->pii_targets;
816 			if (tg != NULL)
817 				target_delete(tg);
818 			assert(pii->pii_targets == NULL);
819 			assert(pii->pii_target_next == NULL);
820 			assert(pii->pii_ntargets == 0);
821 			target_create(pii, probe_logint->li_dstaddr,
822 			    _B_TRUE);
823 		}
824 
825 		/*
826 		 * If no targets are currently known for this phyint
827 		 * we need to call init_router_targets. Since
828 		 * init_router_targets() initializes the list of targets
829 		 * for all phyints it is done below the loop.
830 		 */
831 		if (pii->pii_targets == NULL)
832 			target_scan_reqd = _B_TRUE;
833 
834 		/*
835 		 * Start the probe timer for this instance.
836 		 */
837 		if (!pii->pii_basetime_inited && pii->pii_probe_sock != -1) {
838 			start_timer(pii);
839 			pii->pii_basetime_inited = 1;
840 		}
841 	}
842 
843 	/*
844 	 * Check the interface list for any interfaces that are marked
845 	 * PI_FAILED but no longer enabled to send probes, and call
846 	 * phyint_check_for_repair() to see if the link now indicates that the
847 	 * interface should be repaired.  Also see the state diagram in
848 	 * mpd_probe.c.
849 	 */
850 	for (pi = phyints; pi != NULL; pi = pi->pi_next) {
851 		if (pi->pi_state == PI_FAILED &&
852 		    !PROBE_ENABLED(pi->pi_v4) && !PROBE_ENABLED(pi->pi_v6)) {
853 			phyint_check_for_repair(pi);
854 		}
855 	}
856 
857 	/*
858 	 * Try to populate the target list. init_router_targets populates
859 	 * the target list from the routing table. If our target list is
860 	 * still empty, init_host_targets adds host targets based on the
861 	 * host target list of other phyints in the group.
862 	 */
863 	if (target_scan_reqd) {
864 		init_router_targets();
865 		init_host_targets();
866 	}
867 }
868 
869 /*
870  * Check phyint group configuration, to detect any inconsistencies,
871  * and log an error message. This is called from runtimeouts every
872  * 20 secs. But the error message is displayed once. If the
873  * consistency is resolved by the admin, a recovery message is displayed
874  * once.
875  */
876 static void
877 check_config(void)
878 {
879 	struct phyint_group *pg;
880 	struct phyint *pi;
881 	boolean_t v4_in_group;
882 	boolean_t v6_in_group;
883 
884 	/*
885 	 * All phyints of a group must be homogenous to ensure that
886 	 * failover or failback can be done. If any phyint in a group
887 	 * has IPv4 plumbed, check that all phyints have IPv4 plumbed.
888 	 * Do a similar check for IPv6.
889 	 */
890 	for (pg = phyint_groups; pg != NULL; pg = pg->pg_next) {
891 		if (pg == phyint_anongroup)
892 			continue;
893 
894 		v4_in_group = _B_FALSE;
895 		v6_in_group = _B_FALSE;
896 		/*
897 		 * 1st pass. Determine if at least 1 phyint in the group
898 		 * has IPv4 plumbed and if so set v4_in_group to true.
899 		 * Repeat similarly for IPv6.
900 		 */
901 		for (pi = pg->pg_phyint; pi != NULL; pi = pi->pi_pgnext) {
902 			if (pi->pi_v4 != NULL)
903 				v4_in_group = _B_TRUE;
904 			if (pi->pi_v6 != NULL)
905 				v6_in_group = _B_TRUE;
906 		}
907 
908 		/*
909 		 * 2nd pass. If v4_in_group is true, check that phyint
910 		 * has IPv4 plumbed. Repeat similarly for IPv6. Print
911 		 * out a message the 1st time only.
912 		 */
913 		for (pi = pg->pg_phyint; pi != NULL; pi = pi->pi_pgnext) {
914 			if (pi->pi_flags & IFF_OFFLINE)
915 				continue;
916 
917 			if (v4_in_group == _B_TRUE && pi->pi_v4 == NULL) {
918 				if (!pi->pi_cfgmsg_printed) {
919 					logerr("NIC %s of group %s is"
920 					    " not plumbed for IPv4 and may"
921 					    " affect failover capability\n",
922 					    pi->pi_name,
923 					    pi->pi_group->pg_name);
924 					pi->pi_cfgmsg_printed = 1;
925 				}
926 			} else if (v6_in_group == _B_TRUE &&
927 			    pi->pi_v6 == NULL) {
928 				if (!pi->pi_cfgmsg_printed) {
929 					logerr("NIC %s of group %s is"
930 					    " not plumbed for IPv6 and may"
931 					    " affect failover capability\n",
932 					    pi->pi_name,
933 					    pi->pi_group->pg_name);
934 					pi->pi_cfgmsg_printed = 1;
935 				}
936 			} else {
937 				/*
938 				 * The phyint matches the group configuration,
939 				 * if we have reached this point. If it was
940 				 * improperly configured earlier, log an
941 				 * error recovery message
942 				 */
943 				if (pi->pi_cfgmsg_printed) {
944 					logerr("NIC %s is now consistent with "
945 					    "group %s and failover capability "
946 					    "is restored\n", pi->pi_name,
947 					    pi->pi_group->pg_name);
948 					pi->pi_cfgmsg_printed = 0;
949 				}
950 			}
951 
952 		}
953 	}
954 
955 	/*
956 	 * In order to perform probe-based failure detection, a phyint must
957 	 * have at least 1 test/probe address for sending and receiving probes
958 	 * (either on IPv4 or IPv6 instance or both).  If no test address has
959 	 * been configured, notify the administrator, but continue on since we
960 	 * can still perform load spreading, along with "link up/down" based
961 	 * failure detection.
962 	 */
963 	for (pi = phyints; pi != NULL; pi = pi->pi_next) {
964 		if (pi->pi_flags & IFF_OFFLINE)
965 			continue;
966 
967 		if ((pi->pi_v4 == NULL ||
968 		    pi->pi_v4->pii_probe_logint == NULL) &&
969 		    (pi->pi_v6 == NULL ||
970 		    pi->pi_v6->pii_probe_logint == NULL)) {
971 			if (!pi->pi_taddrmsg_printed) {
972 				logerr("No test address configured on "
973 				    "interface %s; disabling probe-based "
974 				    "failure detection on it\n", pi->pi_name);
975 				pi->pi_taddrmsg_printed = 1;
976 			}
977 		} else if (pi->pi_taddrmsg_printed) {
978 			logerr("Test address now configured on interface %s; "
979 			    "enabling probe-based failure detection on it\n",
980 			    pi->pi_name);
981 			pi->pi_taddrmsg_printed = 0;
982 		}
983 
984 	}
985 }
986 
987 /*
988  * Timer mechanism using relative time (in milliseconds) from the
989  * previous timer event. Timers exceeding TIMER_INFINITY milliseconds
990  * will fire after TIMER_INFINITY milliseconds.
991  * Unsigned arithmetic note: We assume a 32-bit circular sequence space for
992  * time values. Hence 2 consecutive timer events cannot be spaced farther
993  * than 0x7fffffff. We call this TIMER_INFINITY, and it is the maximum value
994  * that can be passed for the delay parameter of timer_schedule()
995  */
996 static uint_t timer_next;	/* Currently scheduled timeout */
997 static boolean_t timer_active = _B_FALSE; /* SIGALRM has not yet occurred */
998 
999 static void
1000 timer_init(void)
1001 {
1002 	timer_next = getcurrenttime() + TIMER_INFINITY;
1003 	/*
1004 	 * The call to run_timeouts() will get the timer started
1005 	 * Since there are no phyints at this point, the timer will
1006 	 * be set for IF_SCAN_INTERVAL ms.
1007 	 */
1008 	run_timeouts();
1009 }
1010 
1011 /*
1012  * Make sure the next SIGALRM occurs delay milliseconds from the current
1013  * time if not earlier. We are interested only in time differences.
1014  */
1015 void
1016 timer_schedule(uint_t delay)
1017 {
1018 	uint_t now;
1019 	struct itimerval itimerval;
1020 
1021 	if (debug & D_TIMER)
1022 		logdebug("timer_schedule(%u)\n", delay);
1023 
1024 	assert(delay <= TIMER_INFINITY);
1025 
1026 	now = getcurrenttime();
1027 	if (delay == 0) {
1028 		/* Minimum allowed delay */
1029 		delay = 1;
1030 	}
1031 	/* Will this timer occur before the currently scheduled SIGALRM? */
1032 	if (timer_active && TIME_GE(now + delay, timer_next)) {
1033 		if (debug & D_TIMER) {
1034 			logdebug("timer_schedule(%u) - no action: "
1035 			    "now %u next %u\n", delay, now, timer_next);
1036 		}
1037 		return;
1038 	}
1039 	timer_next = now + delay;
1040 
1041 	itimerval.it_value.tv_sec = delay / 1000;
1042 	itimerval.it_value.tv_usec = (delay % 1000) * 1000;
1043 	itimerval.it_interval.tv_sec = 0;
1044 	itimerval.it_interval.tv_usec = 0;
1045 	if (debug & D_TIMER) {
1046 		logdebug("timer_schedule(%u): sec %ld usec %ld\n",
1047 		    delay, itimerval.it_value.tv_sec,
1048 		    itimerval.it_value.tv_usec);
1049 	}
1050 	timer_active = _B_TRUE;
1051 	if (setitimer(ITIMER_REAL, &itimerval, NULL) < 0) {
1052 		logperror("timer_schedule: setitimer");
1053 		exit(2);
1054 	}
1055 }
1056 
1057 /*
1058  * Timer has fired. Determine when the next timer event will occur by asking
1059  * all the timer routines. Should not be called from a timer routine.
1060  */
1061 static void
1062 run_timeouts(void)
1063 {
1064 	uint_t next;
1065 	uint_t next_event_time;
1066 	struct phyint_instance *pii;
1067 	struct phyint_instance *next_pii;
1068 	static boolean_t timeout_running;
1069 
1070 	/* assert that recursive timeouts don't happen. */
1071 	assert(!timeout_running);
1072 
1073 	timeout_running = _B_TRUE;
1074 
1075 	if (debug & D_TIMER)
1076 		logdebug("run_timeouts()\n");
1077 
1078 	next = TIMER_INFINITY;
1079 
1080 	for (pii = phyint_instances; pii != NULL; pii = next_pii) {
1081 		next_pii = pii->pii_next;
1082 		next_event_time = phyint_inst_timer(pii);
1083 		if (next_event_time != TIMER_INFINITY && next_event_time < next)
1084 			next = next_event_time;
1085 
1086 		if (debug & D_TIMER) {
1087 			logdebug("run_timeouts(%s %s): next scheduled for"
1088 			    " this phyint inst %u, next scheduled global"
1089 			    " %u ms\n",
1090 			    AF_STR(pii->pii_af), pii->pii_phyint->pi_name,
1091 			    next_event_time, next);
1092 		}
1093 	}
1094 
1095 	/*
1096 	 * Make sure initifs() is called at least once every
1097 	 * IF_SCAN_INTERVAL, to make sure that we are in sync
1098 	 * with the kernel, in case we have missed any routing
1099 	 * socket messages.
1100 	 */
1101 	if (next > IF_SCAN_INTERVAL)
1102 		next = IF_SCAN_INTERVAL;
1103 
1104 	if ((getcurrenttime() - last_initifs_time) > IF_SCAN_INTERVAL) {
1105 		initifs();
1106 		check_config();
1107 	}
1108 
1109 	if (debug & D_TIMER)
1110 		logdebug("run_timeouts: %u ms\n", next);
1111 
1112 	timer_schedule(next);
1113 	timeout_running = _B_FALSE;
1114 }
1115 
1116 static int eventpipe_read = -1;	/* Used for synchronous signal delivery */
1117 static int eventpipe_write = -1;
1118 static boolean_t cleanup_started = _B_FALSE;
1119 				/* Don't write to eventpipe if in cleanup */
1120 /*
1121  * Ensure that signals are processed synchronously with the rest of
1122  * the code by just writing a one character signal number on the pipe.
1123  * The poll loop will pick this up and process the signal event.
1124  */
1125 static void
1126 sig_handler(int signo)
1127 {
1128 	uchar_t buf = (uchar_t)signo;
1129 
1130 	/*
1131 	 * Don't write to pipe if cleanup has already begun. cleanup()
1132 	 * might have closed the pipe already
1133 	 */
1134 	if (cleanup_started)
1135 		return;
1136 
1137 	if (eventpipe_write == -1) {
1138 		logerr("sig_handler: no pipe found\n");
1139 		return;
1140 	}
1141 	if (write(eventpipe_write, &buf, sizeof (buf)) < 0)
1142 		logperror("sig_handler: write");
1143 }
1144 
1145 extern struct probes_missed probes_missed;
1146 
1147 /*
1148  * Pick up a signal "byte" from the pipe and process it.
1149  */
1150 static void
1151 in_signal(int fd)
1152 {
1153 	uchar_t buf;
1154 	uint64_t  sent, acked, lost, unacked, unknown;
1155 	struct phyint_instance *pii;
1156 	int pr_ndx;
1157 
1158 	switch (read(fd, &buf, sizeof (buf))) {
1159 	case -1:
1160 		logperror("in_signal: read");
1161 		exit(1);
1162 		/* NOTREACHED */
1163 	case 1:
1164 		break;
1165 	case 0:
1166 		logerr("in_signal: read end of file\n");
1167 		exit(1);
1168 		/* NOTREACHED */
1169 	default:
1170 		logerr("in_signal: read > 1\n");
1171 		exit(1);
1172 	}
1173 
1174 	if (debug & D_TIMER)
1175 		logdebug("in_signal() got %d\n", buf);
1176 
1177 	switch (buf) {
1178 	case SIGALRM:
1179 		if (debug & D_TIMER) {
1180 			uint_t now = getcurrenttime();
1181 
1182 			logdebug("in_signal(SIGALRM) delta %u\n",
1183 			    now - timer_next);
1184 		}
1185 		timer_active = _B_FALSE;
1186 		run_timeouts();
1187 		break;
1188 	case SIGUSR1:
1189 		logdebug("Printing configuration:\n");
1190 		/* Print out the internal tables */
1191 		phyint_inst_print_all();
1192 
1193 		/*
1194 		 * Print out the accumulated statistics about missed
1195 		 * probes (happens due to scheduling delay).
1196 		 */
1197 		logerr("Missed sending total of %d probes spread over"
1198 		    " %d occurrences\n", probes_missed.pm_nprobes,
1199 		    probes_missed.pm_ntimes);
1200 
1201 		/*
1202 		 * Print out the accumulated statistics about probes
1203 		 * that were sent.
1204 		 */
1205 		for (pii = phyint_instances; pii != NULL;
1206 		    pii = pii->pii_next) {
1207 			unacked = 0;
1208 			acked = pii->pii_cum_stats.acked;
1209 			lost = pii->pii_cum_stats.lost;
1210 			sent = pii->pii_cum_stats.sent;
1211 			unknown = pii->pii_cum_stats.unknown;
1212 			for (pr_ndx = 0; pr_ndx < PROBE_STATS_COUNT; pr_ndx++) {
1213 				switch (pii->pii_probes[pr_ndx].pr_status) {
1214 				case PR_ACKED:
1215 					acked++;
1216 					break;
1217 				case PR_LOST:
1218 					lost++;
1219 					break;
1220 				case PR_UNACKED:
1221 					unacked++;
1222 					break;
1223 				}
1224 			}
1225 			logerr("\nProbe stats on (%s %s)\n"
1226 			    "Number of probes sent %lld\n"
1227 			    "Number of probe acks received %lld\n"
1228 			    "Number of probes/acks lost %lld\n"
1229 			    "Number of valid unacknowled probes %lld\n"
1230 			    "Number of ambiguous probe acks received %lld\n",
1231 			    AF_STR(pii->pii_af), pii->pii_name,
1232 			    sent, acked, lost, unacked, unknown);
1233 		}
1234 		break;
1235 	case SIGHUP:
1236 		logerr("SIGHUP: restart and reread config file\n");
1237 		cleanup();
1238 		(void) execv(argv0[0], argv0);
1239 		_exit(0177);
1240 		/* NOTREACHED */
1241 	case SIGINT:
1242 	case SIGTERM:
1243 	case SIGQUIT:
1244 		cleanup();
1245 		exit(0);
1246 		/* NOTREACHED */
1247 	default:
1248 		logerr("in_signal: unknown signal: %d\n", buf);
1249 	}
1250 }
1251 
1252 static void
1253 cleanup(void)
1254 {
1255 	struct phyint_instance *pii;
1256 	struct phyint_instance *next_pii;
1257 
1258 	/*
1259 	 * Make sure that we don't write to eventpipe in
1260 	 * sig_handler() if any signal notably SIGALRM,
1261 	 * occurs after we close the eventpipe descriptor below
1262 	 */
1263 	cleanup_started = _B_TRUE;
1264 
1265 	for (pii = phyint_instances; pii != NULL; pii = next_pii) {
1266 		next_pii = pii->pii_next;
1267 		phyint_inst_delete(pii);
1268 	}
1269 
1270 	(void) close(ifsock_v4);
1271 	(void) close(ifsock_v6);
1272 	(void) close(rtsock_v4);
1273 	(void) close(rtsock_v6);
1274 	(void) close(lsock_v4);
1275 	(void) close(lsock_v6);
1276 	(void) close(0);
1277 	(void) close(1);
1278 	(void) close(2);
1279 	(void) close(mibfd);
1280 	(void) close(eventpipe_read);
1281 	(void) close(eventpipe_write);
1282 }
1283 
1284 /*
1285  * Create pipe for signal delivery and set up signal handlers.
1286  */
1287 static void
1288 setup_eventpipe(void)
1289 {
1290 	int fds[2];
1291 	struct sigaction act;
1292 
1293 	if ((pipe(fds)) < 0) {
1294 		logperror("setup_eventpipe: pipe");
1295 		exit(1);
1296 	}
1297 	eventpipe_read = fds[0];
1298 	eventpipe_write = fds[1];
1299 	if (poll_add(eventpipe_read) == -1) {
1300 		exit(1);
1301 	}
1302 
1303 	act.sa_handler = sig_handler;
1304 	act.sa_flags = SA_RESTART;
1305 	(void) sigaction(SIGALRM, &act, NULL);
1306 
1307 	(void) sigset(SIGHUP, sig_handler);
1308 	(void) sigset(SIGUSR1, sig_handler);
1309 	(void) sigset(SIGTERM, sig_handler);
1310 	(void) sigset(SIGINT, sig_handler);
1311 	(void) sigset(SIGQUIT, sig_handler);
1312 }
1313 
1314 /*
1315  * Create a routing socket for receiving RTM_IFINFO messages.
1316  */
1317 static int
1318 setup_rtsock(int af)
1319 {
1320 	int	s;
1321 	int	flags;
1322 
1323 	s = socket(PF_ROUTE, SOCK_RAW, af);
1324 	if (s == -1) {
1325 		logperror("setup_rtsock: socket PF_ROUTE");
1326 		exit(1);
1327 	}
1328 	if ((flags = fcntl(s, F_GETFL, 0)) < 0) {
1329 		logperror("setup_rtsock: fcntl F_GETFL");
1330 		(void) close(s);
1331 		exit(1);
1332 	}
1333 	if ((fcntl(s, F_SETFL, flags | O_NONBLOCK)) < 0) {
1334 		logperror("setup_rtsock: fcntl F_SETFL");
1335 		(void) close(s);
1336 		exit(1);
1337 	}
1338 	if (poll_add(s) == -1) {
1339 		(void) close(s);
1340 		exit(1);
1341 	}
1342 	return (s);
1343 }
1344 
1345 /*
1346  * Process an RTM_IFINFO message received on a routing socket.
1347  * The return value indicates whether a full interface scan is required.
1348  * Link up/down notifications from the NICs are reflected in the
1349  * IFF_RUNNING flag.
1350  * If just the state of the IFF_RUNNING interface flag has changed, a
1351  * a full interface scan isn't required.
1352  */
1353 static boolean_t
1354 process_rtm_ifinfo(if_msghdr_t *ifm, int type)
1355 {
1356 	struct sockaddr_dl *sdl;
1357 	struct phyint *pi;
1358 	uint64_t old_flags;
1359 	struct phyint_instance *pii;
1360 
1361 	assert(ifm->ifm_type == RTM_IFINFO && ifm->ifm_addrs == RTA_IFP);
1362 
1363 	/*
1364 	 * Although the sockaddr_dl structure is directly after the
1365 	 * if_msghdr_t structure. At the time of writing, the size of the
1366 	 * if_msghdr_t structure is different on 32 and 64 bit kernels, due
1367 	 * to the presence of a timeval structure, which contains longs,
1368 	 * in the if_data structure.  Anyway, we know where the message ends,
1369 	 * so we work backwards to get the start of the sockaddr_dl structure.
1370 	 */
1371 	/*LINTED*/
1372 	sdl = (struct sockaddr_dl *)((char *)ifm + ifm->ifm_msglen -
1373 		sizeof (struct sockaddr_dl));
1374 
1375 	assert(sdl->sdl_family == AF_LINK);
1376 
1377 	/*
1378 	 * The interface name is in sdl_data.
1379 	 * RTM_IFINFO messages are only generated for logical interface
1380 	 * zero, so there is no colon and logical interface number to
1381 	 * strip from the name.	 The name is not null terminated, but
1382 	 * there should be enough space in sdl_data to add the null.
1383 	 */
1384 	if (sdl->sdl_nlen >= sizeof (sdl->sdl_data)) {
1385 		if (debug & D_LINKNOTE)
1386 			logdebug("process_rtm_ifinfo: "
1387 				"phyint name too long\n");
1388 		return (_B_TRUE);
1389 	}
1390 	sdl->sdl_data[sdl->sdl_nlen] = 0;
1391 
1392 	pi = phyint_lookup(sdl->sdl_data);
1393 	if (pi == NULL) {
1394 		if (debug & D_LINKNOTE)
1395 			logdebug("process_rtm_ifinfo: phyint lookup failed"
1396 				" for %s\n", sdl->sdl_data);
1397 		return (_B_TRUE);
1398 	}
1399 
1400 	/*
1401 	 * We want to try and avoid doing a full interface scan for
1402 	 * link state notifications from the NICs, as indicated
1403 	 * by the state of the IFF_RUNNING flag.  If just the
1404 	 * IFF_RUNNING flag has changed state, the link state changes
1405 	 * are processed without a full scan.
1406 	 * If there is both an IPv4 and IPv6 instance associated with
1407 	 * the physical interface, we will get an RTM_IFINFO message
1408 	 * for each instance.  If we just maintained a single copy of
1409 	 * the physical interface flags, it would appear that no flags
1410 	 * had changed when the second message is processed, leading us
1411 	 * to believe that the message wasn't generated by a flags change,
1412 	 * and that a full interface scan is required.
1413 	 * To get around this problem, two additional copies of the flags
1414 	 * are kept, one copy for each instance.  These are only used in
1415 	 * this routine.  At any one time, all three copies of the flags
1416 	 * should be identical except for the IFF_RUNNING flag.	 The
1417 	 * copy of the flags in the "phyint" structure is always up to
1418 	 * date.
1419 	 */
1420 	pii = (type == AF_INET) ? pi->pi_v4 : pi->pi_v6;
1421 	if (pii == NULL) {
1422 		if (debug & D_LINKNOTE)
1423 			logdebug("process_rtm_ifinfo: no instance of address "
1424 			    "family %s for %s\n", AF_STR(type), pi->pi_name);
1425 		return (_B_TRUE);
1426 	}
1427 
1428 	old_flags = pii->pii_flags;
1429 	pii->pii_flags = PHYINT_FLAGS(ifm->ifm_flags);
1430 	pi->pi_flags = pii->pii_flags;
1431 
1432 	if (debug & D_LINKNOTE) {
1433 		logdebug("process_rtm_ifinfo: %s address family: %s, "
1434 		    "old flags: %llx, new flags: %llx\n", pi->pi_name,
1435 		    AF_STR(type), old_flags, pi->pi_flags);
1436 	}
1437 
1438 	/*
1439 	 * If IFF_STANDBY has changed, indicate that the interface has changed
1440 	 * types.
1441 	 */
1442 	if ((old_flags ^ pii->pii_flags) & IFF_STANDBY)
1443 		phyint_newtype(pi);
1444 
1445 	/*
1446 	 * If IFF_INACTIVE has been set, then no data addresses should be
1447 	 * hosted on the interface.  If IFF_INACTIVE has been cleared, then
1448 	 * move previously failed-over addresses back to it, provided it is
1449 	 * not failed.	For details, see the state diagram in mpd_probe.c.
1450 	 */
1451 	if ((old_flags ^ pii->pii_flags) & IFF_INACTIVE) {
1452 		if (pii->pii_flags & IFF_INACTIVE) {
1453 			if (!pi->pi_empty && (pi->pi_flags & IFF_STANDBY))
1454 				(void) try_failover(pi, FAILOVER_TO_NONSTANDBY);
1455 		} else {
1456 			if (pi->pi_state == PI_RUNNING && !pi->pi_full) {
1457 				pi->pi_empty = 0;
1458 				(void) try_failback(pi, _B_FALSE);
1459 			}
1460 		}
1461 	}
1462 
1463 	/* Has just the IFF_RUNNING flag changed state ? */
1464 	if ((old_flags ^ pii->pii_flags) != IFF_RUNNING) {
1465 		struct phyint_instance *pii_other;
1466 		/*
1467 		 * It wasn't just a link state change.	Update
1468 		 * the other instance's copy of the flags.
1469 		 */
1470 		pii_other = phyint_inst_other(pii);
1471 		if (pii_other != NULL)
1472 			pii_other->pii_flags = pii->pii_flags;
1473 		return (_B_TRUE);
1474 	}
1475 
1476 	return (_B_FALSE);
1477 }
1478 
1479 /*
1480  * Retrieve as many routing socket messages as possible, and try to
1481  * empty the routing sockets. Initiate full scan of targets or interfaces
1482  * as needed.
1483  * We listen on separate IPv4 an IPv6 sockets so that we can accurately
1484  * detect changes in certain flags (see "process_rtm_ifinfo()" above).
1485  */
1486 static void
1487 process_rtsock(int rtsock_v4, int rtsock_v6)
1488 {
1489 	int	nbytes;
1490 	int64_t msg[2048 / 8];
1491 	struct rt_msghdr *rtm;
1492 	boolean_t need_if_scan = _B_FALSE;
1493 	boolean_t need_rt_scan = _B_FALSE;
1494 	boolean_t rtm_ifinfo_seen = _B_FALSE;
1495 	int type;
1496 
1497 	/* Read as many messages as possible and try to empty the sockets */
1498 	for (type = AF_INET; ; type = AF_INET6) {
1499 		for (;;) {
1500 			nbytes = read((type == AF_INET) ? rtsock_v4 :
1501 				rtsock_v6, msg, sizeof (msg));
1502 			if (nbytes <= 0) {
1503 				/* No more messages */
1504 				break;
1505 			}
1506 			rtm = (struct rt_msghdr *)msg;
1507 			if (rtm->rtm_version != RTM_VERSION) {
1508 				logerr("process_rtsock: version %d "
1509 				    "not understood\n", rtm->rtm_version);
1510 				break;
1511 			}
1512 
1513 			if (debug & D_PHYINT) {
1514 				logdebug("process_rtsock: message %d\n",
1515 				    rtm->rtm_type);
1516 			}
1517 
1518 			switch (rtm->rtm_type) {
1519 			case RTM_NEWADDR:
1520 			case RTM_DELADDR:
1521 				/*
1522 				 * Some logical interface has changed,
1523 				 * have to scan everything to determine
1524 				 * what actually changed.
1525 				 */
1526 				need_if_scan = _B_TRUE;
1527 				break;
1528 
1529 			case RTM_IFINFO:
1530 				rtm_ifinfo_seen = _B_TRUE;
1531 				need_if_scan |=
1532 					process_rtm_ifinfo((if_msghdr_t *)rtm,
1533 					type);
1534 				break;
1535 
1536 			case RTM_ADD:
1537 			case RTM_DELETE:
1538 			case RTM_CHANGE:
1539 			case RTM_OLDADD:
1540 			case RTM_OLDDEL:
1541 				need_rt_scan = _B_TRUE;
1542 				break;
1543 
1544 			default:
1545 				/* Not interesting */
1546 				break;
1547 			}
1548 		}
1549 		if (type == AF_INET6)
1550 			break;
1551 	}
1552 
1553 	if (need_if_scan) {
1554 		if (debug & D_LINKNOTE && rtm_ifinfo_seen)
1555 			logdebug("process_rtsock: synchronizing with kernel\n");
1556 		initifs();
1557 	} else if (rtm_ifinfo_seen) {
1558 		if (debug & D_LINKNOTE)
1559 			logdebug("process_rtsock: "
1560 			    "link up/down notification(s) seen\n");
1561 		process_link_state_changes();
1562 	}
1563 
1564 	if (need_rt_scan)
1565 		init_router_targets();
1566 }
1567 
1568 /*
1569  * Look if the phyint instance or one of its logints have been removed from
1570  * the kernel and take appropriate action.
1571  * Uses {pii,li}_in_use.
1572  */
1573 static void
1574 check_if_removed(struct phyint_instance *pii)
1575 {
1576 	struct logint *li;
1577 	struct logint *next_li;
1578 
1579 	/* Detect phyints that have been removed from the kernel. */
1580 	if (!pii->pii_in_use) {
1581 		logtrace("%s %s has been removed from kernel\n",
1582 		    AF_STR(pii->pii_af), pii->pii_phyint->pi_name);
1583 		phyint_inst_delete(pii);
1584 	} else {
1585 		/* Detect logints that have been removed. */
1586 		for (li = pii->pii_logint; li != NULL; li = next_li) {
1587 			next_li = li->li_next;
1588 			if (!li->li_in_use) {
1589 				logint_delete(li);
1590 			}
1591 		}
1592 	}
1593 }
1594 
1595 /*
1596  * Send down a T_OPTMGMT_REQ to ip asking for all data in the various
1597  * tables defined by mib2.h. Parse the returned data and extract
1598  * the 'routing' information table. Process the 'routing' table
1599  * to get the list of known onlink routers, and update our database.
1600  * These onlink routers will serve as our probe targets.
1601  * Returns false, if any system calls resulted in errors, true otherwise.
1602  */
1603 static boolean_t
1604 update_router_list(int fd)
1605 {
1606 	union {
1607 		char	ubuf[1024];
1608 		union T_primitives uprim;
1609 	} buf;
1610 
1611 	int			flags;
1612 	struct strbuf		ctlbuf;
1613 	struct strbuf		databuf;
1614 	struct T_optmgmt_req	*tor;
1615 	struct T_optmgmt_ack	*toa;
1616 	struct T_error_ack	*tea;
1617 	struct opthdr		*optp;
1618 	struct opthdr		*req;
1619 	int			status;
1620 	t_scalar_t		prim;
1621 
1622 	tor = (struct T_optmgmt_req *)&buf;
1623 
1624 	tor->PRIM_type = T_SVR4_OPTMGMT_REQ;
1625 	tor->OPT_offset = sizeof (struct T_optmgmt_req);
1626 	tor->OPT_length = sizeof (struct opthdr);
1627 	tor->MGMT_flags = T_CURRENT;
1628 
1629 	req = (struct opthdr *)&tor[1];
1630 	req->level = MIB2_IP;	/* any MIB2_xxx value ok here */
1631 	req->name  = 0;
1632 	req->len   = 0;
1633 
1634 	ctlbuf.buf = (char *)&buf;
1635 	ctlbuf.len = tor->OPT_length + tor->OPT_offset;
1636 	ctlbuf.maxlen = sizeof (buf);
1637 	flags = 0;
1638 	if (putmsg(fd, &ctlbuf, NULL, flags) == -1) {
1639 		logperror("update_router_list: putmsg(ctl)");
1640 		return (_B_FALSE);
1641 	}
1642 
1643 	/*
1644 	 * The response consists of multiple T_OPTMGMT_ACK msgs, 1 msg for
1645 	 * each table defined in mib2.h.  Each T_OPTMGMT_ACK msg contains
1646 	 * a control and data part. The control part contains a struct
1647 	 * T_optmgmt_ack followed by a struct opthdr. The 'opthdr' identifies
1648 	 * the level, name and length of the data in the data part. The
1649 	 * data part contains the actual table data. The last message
1650 	 * is an end-of-data (EOD), consisting of a T_OPTMGMT_ACK and a
1651 	 * single option with zero optlen.
1652 	 */
1653 
1654 	for (;;) {
1655 		/*
1656 		 * Go around this loop once for each table. Ignore
1657 		 * all tables except the routing information table.
1658 		 */
1659 		flags = 0;
1660 		status = getmsg(fd, &ctlbuf, NULL, &flags);
1661 		if (status < 0) {
1662 			if (errno == EINTR)
1663 				continue;
1664 			logperror("update_router_list: getmsg(ctl)");
1665 			return (_B_FALSE);
1666 		}
1667 		if (ctlbuf.len < sizeof (t_scalar_t)) {
1668 			logerr("update_router_list: ctlbuf.len %d\n",
1669 			    ctlbuf.len);
1670 			return (_B_FALSE);
1671 		}
1672 
1673 		prim = buf.uprim.type;
1674 
1675 		switch (prim) {
1676 
1677 		case T_ERROR_ACK:
1678 			tea = &buf.uprim.error_ack;
1679 			if (ctlbuf.len < sizeof (struct T_error_ack)) {
1680 				logerr("update_router_list: T_ERROR_ACK"
1681 				    " ctlbuf.len %d\n", ctlbuf.len);
1682 				return (_B_FALSE);
1683 			}
1684 			logerr("update_router_list: T_ERROR_ACK:"
1685 			    " TLI_error = 0x%lx, UNIX_error = 0x%lx\n",
1686 			    tea->TLI_error, tea->UNIX_error);
1687 			return (_B_FALSE);
1688 
1689 		case T_OPTMGMT_ACK:
1690 			toa = &buf.uprim.optmgmt_ack;
1691 			optp = (struct opthdr *)&toa[1];
1692 			if (ctlbuf.len < sizeof (struct T_optmgmt_ack)) {
1693 				logerr("update_router_list: ctlbuf.len %d\n",
1694 				    ctlbuf.len);
1695 				return (_B_FALSE);
1696 			}
1697 			if (toa->MGMT_flags != T_SUCCESS) {
1698 				logerr("update_router_list: MGMT_flags 0x%lx\n",
1699 				    toa->MGMT_flags);
1700 				return (_B_FALSE);
1701 			}
1702 			break;
1703 
1704 		default:
1705 			logerr("update_router_list: unknown primitive %ld\n",
1706 			    prim);
1707 			return (_B_FALSE);
1708 		}
1709 
1710 		/* Process the T_OPGMGMT_ACK below */
1711 		assert(prim == T_OPTMGMT_ACK);
1712 
1713 		switch (status) {
1714 		case 0:
1715 			/*
1716 			 * We have reached the end of this T_OPTMGMT_ACK
1717 			 * message. If this is the last message i.e EOD,
1718 			 * return, else process the next T_OPTMGMT_ACK msg.
1719 			 */
1720 			if ((ctlbuf.len == sizeof (struct T_optmgmt_ack) +
1721 			    sizeof (struct opthdr)) && optp->len == 0 &&
1722 			    optp->name == 0 && optp->level == 0) {
1723 				/*
1724 				 * This is the EOD message. Return
1725 				 */
1726 				return (_B_TRUE);
1727 			}
1728 			continue;
1729 
1730 		case MORECTL:
1731 		case MORECTL | MOREDATA:
1732 			/*
1733 			 * This should not happen. We should be able to read
1734 			 * the control portion in a single getmsg.
1735 			 */
1736 			logerr("update_router_list: MORECTL\n");
1737 			return (_B_FALSE);
1738 
1739 		case MOREDATA:
1740 			databuf.maxlen = optp->len;
1741 			/* malloc of 0 bytes is ok */
1742 			databuf.buf = malloc((size_t)optp->len);
1743 			if (databuf.maxlen != 0 && databuf.buf == NULL) {
1744 				logperror("update_router_list: malloc");
1745 				return (_B_FALSE);
1746 			}
1747 			databuf.len = 0;
1748 			flags = 0;
1749 			for (;;) {
1750 				status = getmsg(fd, NULL, &databuf, &flags);
1751 				if (status >= 0) {
1752 					break;
1753 				} else if (errno == EINTR) {
1754 					continue;
1755 				} else {
1756 					logperror("update_router_list:"
1757 					    " getmsg(data)");
1758 					free(databuf.buf);
1759 					return (_B_FALSE);
1760 				}
1761 			}
1762 
1763 			if (optp->level == MIB2_IP &&
1764 			    optp->name == MIB2_IP_ROUTE) {
1765 				/* LINTED */
1766 				ire_process_v4((mib2_ipRouteEntry_t *)
1767 				    databuf.buf, databuf.len);
1768 			} else if (optp->level == MIB2_IP6 &&
1769 			    optp->name == MIB2_IP6_ROUTE) {
1770 				/* LINTED */
1771 				ire_process_v6((mib2_ipv6RouteEntry_t *)
1772 				    databuf.buf, databuf.len);
1773 			}
1774 			free(databuf.buf);
1775 		}
1776 	}
1777 	/* NOTREACHED */
1778 }
1779 
1780 /*
1781  * Examine the IPv4 routing table, for default routers. For each default
1782  * router, populate the list of targets of each phyint that is on the same
1783  * link as the default router
1784  */
1785 static void
1786 ire_process_v4(mib2_ipRouteEntry_t *buf, size_t len)
1787 {
1788 	mib2_ipRouteEntry_t	*rp;
1789 	mib2_ipRouteEntry_t	*rp1;
1790 	struct	in_addr		nexthop_v4;
1791 	mib2_ipRouteEntry_t	*endp;
1792 
1793 	if (len == 0)
1794 		return;
1795 	assert((len % sizeof (mib2_ipRouteEntry_t)) == 0);
1796 
1797 	endp = buf + (len / sizeof (mib2_ipRouteEntry_t));
1798 
1799 	/*
1800 	 * Loop thru the routing table entries. Process any IRE_DEFAULT,
1801 	 * IRE_PREFIX, IRE_HOST, IRE_HOST_REDIRECT ire. Ignore the others.
1802 	 * For each such IRE_OFFSUBNET ire, get the nexthop gateway address.
1803 	 * This is a potential target for probing, which we try to add
1804 	 * to the list of probe targets.
1805 	 */
1806 	for (rp = buf; rp < endp; rp++) {
1807 		if (!(rp->ipRouteInfo.re_ire_type & IRE_OFFSUBNET))
1808 			continue;
1809 
1810 		/*  Get the nexthop address. */
1811 		nexthop_v4.s_addr = rp->ipRouteNextHop;
1812 
1813 		/*
1814 		 * Get the nexthop address. Then determine the outgoing
1815 		 * interface, by examining all interface IREs, and picking the
1816 		 * match. We don't look at the interface specified in the route
1817 		 * because we need to add the router target on all matching
1818 		 * interfaces anyway; the goal is to avoid falling back to
1819 		 * multicast when some interfaces are in the same subnet but
1820 		 * not in the same group.
1821 		 */
1822 		for (rp1 = buf; rp1 < endp; rp1++) {
1823 			if (!(rp1->ipRouteInfo.re_ire_type & IRE_INTERFACE)) {
1824 				continue;
1825 			}
1826 
1827 			/*
1828 			 * Determine the interface IRE that matches the nexthop.
1829 			 * i.e.	 (IRE addr & IRE mask) == (nexthop & IRE mask)
1830 			 */
1831 			if ((rp1->ipRouteDest & rp1->ipRouteMask) ==
1832 			    (nexthop_v4.s_addr & rp1->ipRouteMask)) {
1833 				/*
1834 				 * We found the interface ire
1835 				 */
1836 				router_add_v4(rp1, nexthop_v4);
1837 			}
1838 		}
1839 	}
1840 }
1841 
1842 void
1843 router_add_v4(mib2_ipRouteEntry_t *rp1, struct in_addr nexthop_v4)
1844 {
1845 	char *cp;
1846 	char ifname[LIFNAMSIZ + 1];
1847 	struct in6_addr	nexthop;
1848 	int len;
1849 
1850 	if (debug & D_TARGET)
1851 		logdebug("router_add_v4()\n");
1852 
1853 	len = MIN(rp1->ipRouteIfIndex.o_length, sizeof (ifname) - 1);
1854 	(void) memcpy(ifname, rp1->ipRouteIfIndex.o_bytes, len);
1855 	ifname[len] = '\0';
1856 
1857 	if (ifname[0] == '\0')
1858 		return;
1859 
1860 	cp = strchr(ifname, IF_SEPARATOR);
1861 	if (cp != NULL)
1862 		*cp = '\0';
1863 
1864 	IN6_INADDR_TO_V4MAPPED(&nexthop_v4, &nexthop);
1865 	router_add_common(AF_INET, ifname, nexthop);
1866 }
1867 
1868 void
1869 router_add_common(int af, char *ifname, struct in6_addr nexthop)
1870 {
1871 	struct phyint_instance *pii;
1872 	struct phyint *pi;
1873 
1874 	if (debug & D_TARGET)
1875 		logdebug("router_add_common(%s %s)\n", AF_STR(af), ifname);
1876 
1877 	/*
1878 	 * Retrieve the phyint instance; bail if it's not known to us yet.
1879 	 */
1880 	pii = phyint_inst_lookup(af, ifname);
1881 	if (pii == NULL)
1882 		return;
1883 
1884 	/*
1885 	 * Don't use our own addresses as targets.
1886 	 */
1887 	if (own_address(nexthop))
1888 		return;
1889 
1890 	/*
1891 	 * If the phyint is part a named group, then add the address to all
1892 	 * members of the group; note that this is suboptimal in the IPv4 case
1893 	 * as it has already been added to all matching interfaces in
1894 	 * ire_process_v4(). Otherwise, add the address only to the phyint
1895 	 * itself, since other phyints in the anongroup may not be on the same
1896 	 * subnet.
1897 	 */
1898 	pi = pii->pii_phyint;
1899 	if (pi->pi_group == phyint_anongroup) {
1900 		target_add(pii, nexthop, _B_TRUE);
1901 	} else {
1902 		pi = pi->pi_group->pg_phyint;
1903 		for (; pi != NULL; pi = pi->pi_pgnext)
1904 			target_add(PHYINT_INSTANCE(pi, af), nexthop, _B_TRUE);
1905 	}
1906 }
1907 
1908 /*
1909  * Examine the IPv6 routing table, for default routers. For each default
1910  * router, populate the list of targets of each phyint that is on the same
1911  * link as the default router
1912  */
1913 static void
1914 ire_process_v6(mib2_ipv6RouteEntry_t *buf, size_t len)
1915 {
1916 	mib2_ipv6RouteEntry_t	*rp;
1917 	mib2_ipv6RouteEntry_t	*endp;
1918 	struct	in6_addr nexthop_v6;
1919 
1920 	if (debug & D_TARGET)
1921 		logdebug("ire_process_v6(len %d)\n", len);
1922 
1923 	if (len == 0)
1924 		return;
1925 
1926 	assert((len % sizeof (mib2_ipv6RouteEntry_t)) == 0);
1927 	endp = buf + (len / sizeof (mib2_ipv6RouteEntry_t));
1928 
1929 	/*
1930 	 * Loop thru the routing table entries. Process any IRE_DEFAULT,
1931 	 * IRE_PREFIX, IRE_HOST, IRE_HOST_REDIRECT ire. Ignore the others.
1932 	 * For each such IRE_OFFSUBNET ire, get the nexthop gateway address.
1933 	 * This is a potential target for probing, which we try to add
1934 	 * to the list of probe targets.
1935 	 */
1936 	for (rp = buf; rp < endp; rp++) {
1937 		if (!(rp->ipv6RouteInfo.re_ire_type & IRE_OFFSUBNET))
1938 			continue;
1939 
1940 		/*
1941 		 * We have the outgoing interface in ipv6RouteIfIndex
1942 		 * if ipv6RouteIfindex.o_length is non-zero. The outgoing
1943 		 * interface must be present for link-local addresses. Since
1944 		 * we use only link-local addreses for probing, we don't
1945 		 * consider the case when the outgoing interface is not
1946 		 * known and we need to scan interface ires
1947 		 */
1948 		nexthop_v6 = rp->ipv6RouteNextHop;
1949 		if (rp->ipv6RouteIfIndex.o_length != 0) {
1950 			/*
1951 			 * We already have the outgoing interface
1952 			 * in ipv6RouteIfIndex.
1953 			 */
1954 			router_add_v6(rp, nexthop_v6);
1955 		}
1956 	}
1957 }
1958 
1959 
1960 void
1961 router_add_v6(mib2_ipv6RouteEntry_t *rp1, struct in6_addr nexthop_v6)
1962 {
1963 	char ifname[LIFNAMSIZ + 1];
1964 	char *cp;
1965 	int  len;
1966 
1967 	if (debug & D_TARGET)
1968 		logdebug("router_add_v6()\n");
1969 
1970 	len = MIN(rp1->ipv6RouteIfIndex.o_length, sizeof (ifname) - 1);
1971 	(void) memcpy(ifname, rp1->ipv6RouteIfIndex.o_bytes, len);
1972 	ifname[len] = '\0';
1973 
1974 	if (ifname[0] == '\0')
1975 		return;
1976 
1977 	cp = strchr(ifname, IF_SEPARATOR);
1978 	if (cp != NULL)
1979 		*cp = '\0';
1980 
1981 	router_add_common(AF_INET6, ifname, nexthop_v6);
1982 }
1983 
1984 
1985 
1986 /*
1987  * Build a list of target routers, by scanning the routing tables.
1988  * It is assumed that interface routes exist, to reach the routers.
1989  */
1990 static void
1991 init_router_targets(void)
1992 {
1993 	struct	target *tg;
1994 	struct	target *next_tg;
1995 	struct	phyint_instance *pii;
1996 	struct	phyint *pi;
1997 
1998 	if (force_mcast)
1999 		return;
2000 
2001 	for (pii = phyint_instances; pii != NULL; pii = pii->pii_next) {
2002 		pi = pii->pii_phyint;
2003 		/*
2004 		 * Exclude ptp and host targets. Set tg_in_use to false,
2005 		 * only for router targets.
2006 		 */
2007 		if (!pii->pii_targets_are_routers ||
2008 		    (pi->pi_flags & IFF_POINTOPOINT))
2009 			continue;
2010 
2011 		for (tg = pii->pii_targets; tg != NULL; tg = tg->tg_next)
2012 			tg->tg_in_use = 0;
2013 	}
2014 
2015 	if (mibfd < 0) {
2016 		mibfd = open("/dev/ip", O_RDWR);
2017 		if (mibfd < 0) {
2018 			logperror("mibopen: ip open");
2019 			exit(1);
2020 		}
2021 	}
2022 
2023 	if (!update_router_list(mibfd)) {
2024 		(void) close(mibfd);
2025 		mibfd = -1;
2026 	}
2027 
2028 	for (pii = phyint_instances; pii != NULL; pii = pii->pii_next) {
2029 		if (!pii->pii_targets_are_routers ||
2030 		    (pi->pi_flags & IFF_POINTOPOINT))
2031 			continue;
2032 
2033 		for (tg = pii->pii_targets; tg != NULL; tg = next_tg) {
2034 			next_tg = tg->tg_next;
2035 			if (!tg->tg_in_use) {
2036 				target_delete(tg);
2037 			}
2038 		}
2039 	}
2040 }
2041 
2042 /*
2043  * Attempt to assign host targets to any interfaces that do not currently
2044  * have probe targets by sharing targets with other interfaces in the group.
2045  */
2046 static void
2047 init_host_targets(void)
2048 {
2049 	struct phyint_instance *pii;
2050 	struct phyint_group *pg;
2051 
2052 	for (pii = phyint_instances; pii != NULL; pii = pii->pii_next) {
2053 		pg = pii->pii_phyint->pi_group;
2054 		if (pg != phyint_anongroup && pii->pii_targets == NULL)
2055 			dup_host_targets(pii);
2056 	}
2057 }
2058 
2059 /*
2060  * Duplicate host targets from other phyints of the group to
2061  * the phyint instance 'desired_pii'.
2062  */
2063 static void
2064 dup_host_targets(struct phyint_instance	 *desired_pii)
2065 {
2066 	int af;
2067 	struct phyint *pi;
2068 	struct phyint_instance *pii;
2069 	struct target *tg;
2070 
2071 	assert(desired_pii->pii_phyint->pi_group != phyint_anongroup);
2072 
2073 	af = desired_pii->pii_af;
2074 
2075 	/*
2076 	 * For every phyint in the same group as desired_pii, check if
2077 	 * it has any host targets. If so add them to desired_pii.
2078 	 */
2079 	for (pi = desired_pii->pii_phyint; pi != NULL; pi = pi->pi_pgnext) {
2080 		pii = PHYINT_INSTANCE(pi, af);
2081 		/*
2082 		 * We know that we don't have targets on this phyint instance
2083 		 * since we have been called. But we still check for
2084 		 * pii_targets_are_routers because another phyint instance
2085 		 * could have router targets, since IFF_NOFAILOVER addresses
2086 		 * on different phyint instances may belong to different
2087 		 * subnets.
2088 		 */
2089 		if ((pii == NULL) || (pii == desired_pii) ||
2090 		    pii->pii_targets_are_routers)
2091 			continue;
2092 		for (tg = pii->pii_targets; tg != NULL; tg = tg->tg_next) {
2093 			target_create(desired_pii, tg->tg_address, _B_FALSE);
2094 		}
2095 	}
2096 }
2097 
2098 static void
2099 usage(char *cmd)
2100 {
2101 	(void) fprintf(stderr, "usage: %s\n", cmd);
2102 }
2103 
2104 
2105 #define	MPATHD_DEFAULT_FILE	"/etc/default/mpathd"
2106 
2107 /* Get an option from the /etc/default/mpathd file */
2108 static char *
2109 getdefault(char *name)
2110 {
2111 	char namebuf[BUFSIZ];
2112 	char *value = NULL;
2113 
2114 	if (defopen(MPATHD_DEFAULT_FILE) == 0) {
2115 		char	*cp;
2116 		int	flags;
2117 
2118 		/*
2119 		 * ignore case
2120 		 */
2121 		flags = defcntl(DC_GETFLAGS, 0);
2122 		TURNOFF(flags, DC_CASE);
2123 		(void) defcntl(DC_SETFLAGS, flags);
2124 
2125 		/* Add "=" to the name */
2126 		(void) strncpy(namebuf, name, sizeof (namebuf) - 2);
2127 		(void) strncat(namebuf, "=", 2);
2128 
2129 		if ((cp = defread(namebuf)) != NULL)
2130 			value = strdup(cp);
2131 
2132 		/* close */
2133 		(void) defopen((char *)NULL);
2134 	}
2135 	return (value);
2136 }
2137 
2138 
2139 /*
2140  * Command line options below
2141  */
2142 boolean_t	failback_enabled = _B_TRUE;	/* failback enabled/disabled */
2143 boolean_t	track_all_phyints = _B_FALSE;	/* option to track all NICs */
2144 static boolean_t adopt = _B_FALSE;
2145 static boolean_t foreground = _B_FALSE;
2146 
2147 int
2148 main(int argc, char *argv[])
2149 {
2150 	int i;
2151 	int c;
2152 	struct phyint_instance *pii;
2153 	char *value;
2154 
2155 	argv0 = argv;		/* Saved for re-exec on SIGHUP */
2156 	srandom(gethostid());	/* Initialize the random number generator */
2157 
2158 	/*
2159 	 * NOTE: The messages output by in.mpathd are not suitable for
2160 	 * translation, so we do not call textdomain().
2161 	 */
2162 	(void) setlocale(LC_ALL, "");
2163 
2164 	/*
2165 	 * Get the user specified value of 'failure detection time'
2166 	 * from /etc/default/mpathd
2167 	 */
2168 	value = getdefault("FAILURE_DETECTION_TIME");
2169 	if (value != NULL) {
2170 		user_failure_detection_time =
2171 		    (int)strtol((char *)value, NULL, 0);
2172 
2173 		if (user_failure_detection_time <= 0) {
2174 			user_failure_detection_time = FAILURE_DETECTION_TIME;
2175 			logerr("Invalid failure detection time %s, assuming "
2176 			    "default %d\n", value, user_failure_detection_time);
2177 
2178 		} else if (user_failure_detection_time <
2179 		    MIN_FAILURE_DETECTION_TIME) {
2180 			user_failure_detection_time =
2181 			    MIN_FAILURE_DETECTION_TIME;
2182 			logerr("Too small failure detection time of %s, "
2183 			    "assuming minimum %d\n", value,
2184 			    user_failure_detection_time);
2185 		}
2186 		free(value);
2187 	} else {
2188 		/* User has not specified the parameter, Use default value */
2189 		user_failure_detection_time = FAILURE_DETECTION_TIME;
2190 	}
2191 
2192 	/*
2193 	 * This gives the frequency at which probes will be sent.
2194 	 * When fdt ms elapses, we should be able to determine
2195 	 * whether 5 consecutive probes have failed or not.
2196 	 * 1 probe will be sent in every user_probe_interval ms,
2197 	 * randomly anytime in the (0.5  - 1.0) 2nd half of every
2198 	 * user_probe_interval. Thus when we send out probe 'n' we
2199 	 * can be sure that probe 'n - 2' is lost, if we have not
2200 	 * got the ack. (since the probe interval is > crtt). But
2201 	 * probe 'n - 1' may be a valid unacked probe, since the
2202 	 * time between 2 successive probes could be as small as
2203 	 * 0.5 * user_probe_interval.  Hence the NUM_PROBE_FAILS + 2
2204 	 */
2205 	user_probe_interval = user_failure_detection_time /
2206 	    (NUM_PROBE_FAILS + 2);
2207 
2208 	/*
2209 	 * Get the user specified value of failback_enabled from
2210 	 * /etc/default/mpathd
2211 	 */
2212 	value = getdefault("FAILBACK");
2213 	if (value != NULL) {
2214 		if (strncasecmp(value, "yes", 3) == 0)
2215 			failback_enabled = _B_TRUE;
2216 		else if (strncasecmp(value, "no", 2) == 0)
2217 			failback_enabled = _B_FALSE;
2218 		else
2219 			logerr("Invalid value for FAILBACK %s\n", value);
2220 		free(value);
2221 	} else {
2222 		failback_enabled = _B_TRUE;
2223 	}
2224 
2225 	/*
2226 	 * Get the user specified value of track_all_phyints from
2227 	 * /etc/default/mpathd. The sense is reversed in
2228 	 * TRACK_INTERFACES_ONLY_WITH_GROUPS.
2229 	 */
2230 	value = getdefault("TRACK_INTERFACES_ONLY_WITH_GROUPS");
2231 	if (value != NULL) {
2232 		if (strncasecmp(value, "yes", 3) == 0)
2233 			track_all_phyints = _B_FALSE;
2234 		else if (strncasecmp(value, "no", 2) == 0)
2235 			track_all_phyints = _B_TRUE;
2236 		else
2237 			logerr("Invalid value for "
2238 			    "TRACK_INTERFACES_ONLY_WITH_GROUPS %s\n", value);
2239 		free(value);
2240 	} else {
2241 		track_all_phyints = _B_FALSE;
2242 	}
2243 
2244 	while ((c = getopt(argc, argv, "adD:ml")) != EOF) {
2245 		switch (c) {
2246 		case 'a':
2247 			adopt = _B_TRUE;
2248 			break;
2249 		case 'm':
2250 			force_mcast = _B_TRUE;
2251 			break;
2252 		case 'd':
2253 			debug = D_ALL;
2254 			foreground = _B_TRUE;
2255 			break;
2256 		case 'D':
2257 			i = (int)strtol(optarg, NULL, 0);
2258 			if (i == 0) {
2259 				(void) fprintf(stderr, "Bad debug flags: %s\n",
2260 				    optarg);
2261 				exit(1);
2262 			}
2263 			debug |= i;
2264 			foreground = _B_TRUE;
2265 			break;
2266 		case 'l':
2267 			/*
2268 			 * Turn off link state notification handling.
2269 			 * Undocumented command line flag, for debugging
2270 			 * purposes.
2271 			 */
2272 			handle_link_notifications = _B_FALSE;
2273 			break;
2274 		default:
2275 			usage(argv[0]);
2276 			exit(1);
2277 		}
2278 	}
2279 
2280 	/*
2281 	 * The sockets for the loopback command interface should be listening
2282 	 * before we fork and exit in daemonize(). This way, whoever started us
2283 	 * can use the loopback interface as soon as they get a zero exit
2284 	 * status.
2285 	 */
2286 	lsock_v4 = setup_listener(AF_INET);
2287 	lsock_v6 = setup_listener(AF_INET6);
2288 
2289 	if (lsock_v4 < 0 && lsock_v6 < 0) {
2290 		logerr("main: setup_listener failed for both IPv4 and IPv6\n");
2291 		exit(1);
2292 	}
2293 
2294 	if (!foreground) {
2295 		if (!daemonize()) {
2296 			logerr("cannot daemonize\n");
2297 			exit(EXIT_FAILURE);
2298 		}
2299 		initlog();
2300 	}
2301 
2302 	/*
2303 	 * Initializations:
2304 	 * 1. Create ifsock* sockets. These are used for performing SIOC*
2305 	 *    ioctls. We have 2 sockets 1 each for IPv4 and IPv6.
2306 	 * 2. Initialize a pipe for handling/recording signal events.
2307 	 * 3. Create the routing sockets,  used for listening
2308 	 *    to routing / interface changes.
2309 	 * 4. phyint_init() - Initialize physical interface state
2310 	 *    (in mpd_tables.c).  Must be done before creating interfaces,
2311 	 *    which timer_init() does indirectly.
2312 	 * 5. timer_init()  - Initialize timer related stuff
2313 	 * 6. initifs() - Initialize our database of all known interfaces
2314 	 * 7. init_router_targets() - Initialize our database of all known
2315 	 *    router targets.
2316 	 */
2317 	ifsock_v4 = socket(AF_INET, SOCK_DGRAM, 0);
2318 	if (ifsock_v4 < 0) {
2319 		logperror("main: IPv4 socket open");
2320 		exit(1);
2321 	}
2322 
2323 	ifsock_v6 = socket(AF_INET6, SOCK_DGRAM, 0);
2324 	if (ifsock_v6 < 0) {
2325 		logperror("main: IPv6 socket open");
2326 		exit(1);
2327 	}
2328 
2329 	setup_eventpipe();
2330 
2331 	rtsock_v4 = setup_rtsock(AF_INET);
2332 	rtsock_v6 = setup_rtsock(AF_INET6);
2333 
2334 	if (phyint_init() == -1) {
2335 		logerr("cannot initialize physical interface structures");
2336 		exit(1);
2337 	}
2338 
2339 	timer_init();
2340 
2341 	initifs();
2342 
2343 	/* Inform kernel whether failback is enabled or disabled */
2344 	if (ioctl(ifsock_v4, SIOCSIPMPFAILBACK, (int *)&failback_enabled) < 0) {
2345 		logperror("main: ioctl (SIOCSIPMPFAILBACK)");
2346 		exit(1);
2347 	}
2348 
2349 	/*
2350 	 * If we're operating in "adopt" mode and no interfaces need to be
2351 	 * tracked, shut down (ifconfig(1M) will restart us on demand if
2352 	 * interfaces are subsequently put into multipathing groups).
2353 	 */
2354 	if (adopt && phyint_instances == NULL)
2355 		exit(0);
2356 
2357 	/*
2358 	 * Main body. Keep listening for activity on any of the sockets
2359 	 * that we are monitoring and take appropriate action as necessary.
2360 	 * signals are also handled synchronously.
2361 	 */
2362 	for (;;) {
2363 		if (poll(pollfds, pollfd_num, -1) < 0) {
2364 			if (errno == EINTR)
2365 				continue;
2366 			logperror("main: poll");
2367 			exit(1);
2368 		}
2369 		for (i = 0; i < pollfd_num; i++) {
2370 			if ((pollfds[i].fd == -1) ||
2371 			    !(pollfds[i].revents & POLLIN))
2372 				continue;
2373 			if (pollfds[i].fd == eventpipe_read) {
2374 				in_signal(eventpipe_read);
2375 				break;
2376 			}
2377 			if (pollfds[i].fd == rtsock_v4 ||
2378 				pollfds[i].fd == rtsock_v6) {
2379 				process_rtsock(rtsock_v4, rtsock_v6);
2380 				break;
2381 			}
2382 			for (pii = phyint_instances; pii != NULL;
2383 			    pii = pii->pii_next) {
2384 				if (pollfds[i].fd == pii->pii_probe_sock) {
2385 					if (pii->pii_af == AF_INET)
2386 						in_data(pii);
2387 					else
2388 						in6_data(pii);
2389 					break;
2390 				}
2391 			}
2392 			if (pollfds[i].fd == lsock_v4)
2393 				loopback_cmd(lsock_v4, AF_INET);
2394 			else if (pollfds[i].fd == lsock_v6)
2395 				loopback_cmd(lsock_v6, AF_INET6);
2396 		}
2397 		if (full_scan_required) {
2398 			initifs();
2399 			full_scan_required = _B_FALSE;
2400 		}
2401 	}
2402 	/* NOTREACHED */
2403 	return (EXIT_SUCCESS);
2404 }
2405 
2406 static int
2407 setup_listener(int af)
2408 {
2409 	int sock;
2410 	int on;
2411 	int len;
2412 	int ret;
2413 	struct sockaddr_storage laddr;
2414 	struct sockaddr_in  *sin;
2415 	struct sockaddr_in6 *sin6;
2416 	struct in6_addr loopback_addr = IN6ADDR_LOOPBACK_INIT;
2417 
2418 	assert(af == AF_INET || af == AF_INET6);
2419 
2420 	sock = socket(af, SOCK_STREAM, 0);
2421 	if (sock < 0) {
2422 		logperror("setup_listener: socket");
2423 		exit(1);
2424 	}
2425 
2426 	on = 1;
2427 	if (setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, (char *)&on,
2428 	    sizeof (on)) < 0) {
2429 		logperror("setup_listener: setsockopt (SO_REUSEADDR)");
2430 		exit(1);
2431 	}
2432 
2433 	bzero(&laddr, sizeof (laddr));
2434 	laddr.ss_family = af;
2435 
2436 	if (af == AF_INET) {
2437 		sin = (struct sockaddr_in *)&laddr;
2438 		sin->sin_port = htons(MPATHD_PORT);
2439 		sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2440 		len = sizeof (struct sockaddr_in);
2441 	} else {
2442 		sin6 = (struct sockaddr_in6 *)&laddr;
2443 		sin6->sin6_port = htons(MPATHD_PORT);
2444 		sin6->sin6_addr = loopback_addr;
2445 		len = sizeof (struct sockaddr_in6);
2446 	}
2447 
2448 	ret = bind(sock, (struct sockaddr *)&laddr, len);
2449 	if (ret < 0) {
2450 		if (errno == EADDRINUSE) {
2451 			/*
2452 			 * Another instance of mpathd may be already active.
2453 			 */
2454 			logerr("main: is another instance of in.mpathd "
2455 			    "already active?\n");
2456 			exit(1);
2457 		} else {
2458 			(void) close(sock);
2459 			return (-1);
2460 		}
2461 	}
2462 	if (listen(sock, 30) < 0) {
2463 		logperror("main: listen");
2464 		exit(1);
2465 	}
2466 	if (poll_add(sock) == -1) {
2467 		(void) close(sock);
2468 		exit(1);
2469 	}
2470 
2471 	return (sock);
2472 }
2473 
2474 /*
2475  * Table of commands and their expected size; used by loopback_cmd().
2476  */
2477 static struct {
2478 	const char	*name;
2479 	unsigned int	size;
2480 } commands[] = {
2481 	{ "MI_PING",		sizeof (uint32_t)	},
2482 	{ "MI_OFFLINE",		sizeof (mi_offline_t)	},
2483 	{ "MI_UNDO_OFFLINE",	sizeof (mi_undo_offline_t) },
2484 	{ "MI_SETOINDEX",	sizeof (mi_setoindex_t) },
2485 	{ "MI_QUERY",		sizeof (mi_query_t)	}
2486 };
2487 
2488 /*
2489  * Commands received over the loopback interface come here. Currently
2490  * the agents that send commands are ifconfig, if_mpadm and the RCM IPMP
2491  * module. ifconfig only makes a connection, and closes it to check if
2492  * in.mpathd is running.
2493  * if_mpadm sends commands in the format specified by the mpathd_interface
2494  * structure.
2495  */
2496 static void
2497 loopback_cmd(int sock, int family)
2498 {
2499 	int newfd;
2500 	ssize_t len;
2501 	struct sockaddr_storage	peer;
2502 	struct sockaddr_in	*peer_sin;
2503 	struct sockaddr_in6	*peer_sin6;
2504 	socklen_t peerlen;
2505 	union mi_commands mpi;
2506 	struct in6_addr loopback_addr = IN6ADDR_LOOPBACK_INIT;
2507 	char abuf[INET6_ADDRSTRLEN];
2508 	uint_t cmd;
2509 	int retval;
2510 
2511 	peerlen = sizeof (peer);
2512 	newfd = accept(sock, (struct sockaddr *)&peer, &peerlen);
2513 	if (newfd < 0) {
2514 		logperror("loopback_cmd: accept");
2515 		return;
2516 	}
2517 
2518 	switch (family) {
2519 	case AF_INET:
2520 		/*
2521 		 * Validate the address and port to make sure that
2522 		 * non privileged processes don't connect and start
2523 		 * talking to us.
2524 		 */
2525 		if (peerlen != sizeof (struct sockaddr_in)) {
2526 			logerr("loopback_cmd: AF_INET peerlen %d\n", peerlen);
2527 			(void) close(newfd);
2528 			return;
2529 		}
2530 		peer_sin = (struct sockaddr_in *)&peer;
2531 		if ((ntohs(peer_sin->sin_port) >= IPPORT_RESERVED) ||
2532 		    (ntohl(peer_sin->sin_addr.s_addr) != INADDR_LOOPBACK)) {
2533 			(void) inet_ntop(AF_INET, &peer_sin->sin_addr.s_addr,
2534 			    abuf, sizeof (abuf));
2535 			logerr("Attempt to connect from addr %s port %d\n",
2536 			    abuf, ntohs(peer_sin->sin_port));
2537 			(void) close(newfd);
2538 			return;
2539 		}
2540 		break;
2541 
2542 	case AF_INET6:
2543 		if (peerlen != sizeof (struct sockaddr_in6)) {
2544 			logerr("loopback_cmd: AF_INET6 peerlen %d\n", peerlen);
2545 			(void) close(newfd);
2546 			return;
2547 		}
2548 		/*
2549 		 * Validate the address and port to make sure that
2550 		 * non privileged processes don't connect and start
2551 		 * talking to us.
2552 		 */
2553 		peer_sin6 = (struct sockaddr_in6 *)&peer;
2554 		if ((ntohs(peer_sin6->sin6_port) >= IPPORT_RESERVED) ||
2555 		    (!IN6_ARE_ADDR_EQUAL(&peer_sin6->sin6_addr,
2556 		    &loopback_addr))) {
2557 			(void) inet_ntop(AF_INET6, &peer_sin6->sin6_addr, abuf,
2558 			    sizeof (abuf));
2559 			logerr("Attempt to connect from addr %s port %d\n",
2560 			    abuf, ntohs(peer_sin6->sin6_port));
2561 			(void) close(newfd);
2562 			return;
2563 		}
2564 
2565 	default:
2566 		logdebug("loopback_cmd: family %d\n", family);
2567 		(void) close(newfd);
2568 		return;
2569 	}
2570 
2571 	/*
2572 	 * The sizeof the 'mpi' buffer corresponds to the maximum size of
2573 	 * all supported commands
2574 	 */
2575 	len = read(newfd, &mpi, sizeof (mpi));
2576 
2577 	/*
2578 	 * ifconfig does not send any data. Just tests to see if mpathd
2579 	 * is already running.
2580 	 */
2581 	if (len <= 0) {
2582 		(void) close(newfd);
2583 		return;
2584 	}
2585 
2586 	/*
2587 	 * In theory, we can receive any sized message for a stream socket,
2588 	 * but we don't expect that to happen for a small message over a
2589 	 * loopback connection.
2590 	 */
2591 	if (len < sizeof (uint32_t)) {
2592 		logerr("loopback_cmd: bad command format or read returns "
2593 		    "partial data %d\n", len);
2594 	}
2595 
2596 	cmd = mpi.mi_command;
2597 	if (cmd >= MI_NCMD) {
2598 		logerr("loopback_cmd: unknown command id `%d'\n", cmd);
2599 		(void) close(newfd);
2600 		return;
2601 	}
2602 
2603 	if (len < commands[cmd].size) {
2604 		logerr("loopback_cmd: short %s command (expected %d, got %d)\n",
2605 		    commands[cmd].name, commands[cmd].size, len);
2606 		(void) close(newfd);
2607 		return;
2608 	}
2609 
2610 	retval = process_cmd(newfd, &mpi);
2611 	if (retval != IPMP_SUCCESS) {
2612 		logerr("failed processing %s: %s\n", commands[cmd].name,
2613 		    ipmp_errmsg(retval));
2614 	}
2615 	(void) close(newfd);
2616 }
2617 
2618 extern int global_errno;	/* set by failover() or failback() */
2619 
2620 /*
2621  * Process the offline, undo offline and set original index commands,
2622  * received from if_mpadm(1M)
2623  */
2624 static unsigned int
2625 process_cmd(int newfd, union mi_commands *mpi)
2626 {
2627 	uint_t	nif = 0;
2628 	uint32_t cmd;
2629 	struct phyint *pi;
2630 	struct phyint *pi2;
2631 	struct phyint_group *pg;
2632 	boolean_t success;
2633 	int error;
2634 	struct mi_offline *mio;
2635 	struct mi_undo_offline *miu;
2636 	struct lifreq lifr;
2637 	int ifsock;
2638 	struct mi_setoindex *mis;
2639 
2640 	cmd = mpi->mi_command;
2641 
2642 	switch (cmd) {
2643 	case MI_OFFLINE:
2644 		mio = &mpi->mi_ocmd;
2645 		/*
2646 		 * Lookup the interface that needs to be offlined.
2647 		 * If it does not exist, return a suitable error.
2648 		 */
2649 		pi = phyint_lookup(mio->mio_ifname);
2650 		if (pi == NULL)
2651 			return (send_result(newfd, IPMP_FAILURE, EINVAL));
2652 
2653 		/*
2654 		 * Verify that the minimum redundancy requirements are met.
2655 		 * The multipathing group must have at least the specified
2656 		 * number of functional interfaces after offlining the
2657 		 * requested interface. Otherwise return a suitable error.
2658 		 */
2659 		pg = pi->pi_group;
2660 		nif = 0;
2661 		if (pg != phyint_anongroup) {
2662 			for (nif = 0, pi2 = pg->pg_phyint; pi2 != NULL;
2663 			    pi2 = pi2->pi_pgnext) {
2664 				if ((pi2->pi_state == PI_RUNNING) ||
2665 				    (pg->pg_groupfailed &&
2666 				    !(pi2->pi_flags & IFF_OFFLINE)))
2667 					nif++;
2668 			}
2669 		}
2670 		if (nif < mio->mio_min_redundancy)
2671 			return (send_result(newfd, IPMP_EMINRED, 0));
2672 
2673 		/*
2674 		 * The order of operation is to set IFF_OFFLINE, followed by
2675 		 * failover. Setting IFF_OFFLINE ensures that no new ipif's
2676 		 * can be created. Subsequent failover moves everything on
2677 		 * the OFFLINE interface to some other functional interface.
2678 		 */
2679 		success = change_lif_flags(pi, IFF_OFFLINE, _B_TRUE);
2680 		if (success) {
2681 			if (!pi->pi_empty) {
2682 				error = try_failover(pi, FAILOVER_NORMAL);
2683 				if (error != 0) {
2684 					if (!change_lif_flags(pi, IFF_OFFLINE,
2685 					    _B_FALSE)) {
2686 						logerr("process_cmd: couldn't"
2687 						    " clear OFFLINE flag on"
2688 						    " %s\n", pi->pi_name);
2689 						/*
2690 						 * Offline interfaces should
2691 						 * not be probed.
2692 						 */
2693 						stop_probing(pi);
2694 					}
2695 					return (send_result(newfd, error,
2696 					    global_errno));
2697 				}
2698 			}
2699 		} else {
2700 			return (send_result(newfd, IPMP_FAILURE, errno));
2701 		}
2702 
2703 		/*
2704 		 * The interface is now Offline, so stop probing it.
2705 		 * Note that if_mpadm(1M) will down the test addresses,
2706 		 * after receiving a success reply from us. The routing
2707 		 * socket message will then make us close the socket used
2708 		 * for sending probes. But it is more logical that an
2709 		 * offlined interface must not be probed, even if it has
2710 		 * test addresses.
2711 		 */
2712 		stop_probing(pi);
2713 		return (send_result(newfd, IPMP_SUCCESS, 0));
2714 
2715 	case MI_UNDO_OFFLINE:
2716 		miu = &mpi->mi_ucmd;
2717 		/*
2718 		 * Undo the offline command. As usual lookup the interface.
2719 		 * Send an error if it does not exist.
2720 		 */
2721 		pi = phyint_lookup(miu->miu_ifname);
2722 		if (pi == NULL)
2723 			return (send_result(newfd, IPMP_FAILURE, EINVAL));
2724 
2725 		/*
2726 		 * Inverse of the offline operation. Do a failback, and then
2727 		 * clear the IFF_OFFLINE flag.
2728 		 */
2729 		error = do_failback(pi, _B_TRUE);
2730 		if (error == IPMP_EFBPARTIAL)
2731 			return (send_result(newfd, IPMP_EFBPARTIAL, 0));
2732 		error = do_failback(pi, _B_FALSE);
2733 
2734 		switch (error) {
2735 		case IPMP_SUCCESS:
2736 			if (!change_lif_flags(pi, IFF_OFFLINE, _B_FALSE)) {
2737 				logdebug("undo error %X\n", global_errno);
2738 				error = IPMP_FAILURE;
2739 				break;
2740 			}
2741 			/* FALLTHROUGH */
2742 
2743 		case IPMP_EFBPARTIAL:
2744 			/*
2745 			 * Reset the state of the interface based on the
2746 			 * current link state; if this phyint subsequently
2747 			 * acquires a test address, the state will be changed
2748 			 * again later as a result of the probes.
2749 			 */
2750 			if (LINK_UP(pi))
2751 				phyint_chstate(pi, PI_RUNNING);
2752 			else
2753 				phyint_chstate(pi, PI_FAILED);
2754 			break;
2755 
2756 		case IPMP_FAILURE:
2757 			break;
2758 
2759 		default:
2760 			logdebug("do_failback: unexpected return value\n");
2761 			break;
2762 		}
2763 		return (send_result(newfd, error, global_errno));
2764 
2765 	case MI_SETOINDEX:
2766 		mis = &mpi->mi_scmd;
2767 
2768 		/* Get the socket for doing ioctls */
2769 		ifsock = (mis->mis_iftype == AF_INET) ? ifsock_v4 : ifsock_v6;
2770 
2771 		/*
2772 		 * Get index of new original interface.
2773 		 * The index is returned in lifr.lifr_index.
2774 		 */
2775 		(void) strlcpy(lifr.lifr_name, mis->mis_new_pifname,
2776 		    sizeof (lifr.lifr_name));
2777 
2778 		if (ioctl(ifsock, SIOCGLIFINDEX, (char *)&lifr) < 0)
2779 			return (send_result(newfd, IPMP_FAILURE, errno));
2780 
2781 		/*
2782 		 * Set new original interface index.
2783 		 * The new index was put into lifr.lifr_index by the
2784 		 * SIOCGLIFINDEX ioctl.
2785 		 */
2786 		(void) strlcpy(lifr.lifr_name, mis->mis_lifname,
2787 		    sizeof (lifr.lifr_name));
2788 
2789 		if (ioctl(ifsock, SIOCSLIFOINDEX, (char *)&lifr) < 0)
2790 			return (send_result(newfd, IPMP_FAILURE, errno));
2791 
2792 		return (send_result(newfd, IPMP_SUCCESS, 0));
2793 
2794 	case MI_QUERY:
2795 		return (process_query(newfd, &mpi->mi_qcmd));
2796 
2797 	default:
2798 		break;
2799 	}
2800 
2801 	return (send_result(newfd, IPMP_EPROTO, 0));
2802 }
2803 
2804 /*
2805  * Process the query request pointed to by `miq' and send a reply on file
2806  * descriptor `fd'.  Returns an IPMP error code.
2807  */
2808 static unsigned int
2809 process_query(int fd, mi_query_t *miq)
2810 {
2811 	ipmp_groupinfo_t	*grinfop;
2812 	ipmp_groupinfolist_t	*grlp;
2813 	ipmp_grouplist_t	*grlistp;
2814 	ipmp_ifinfo_t		*ifinfop;
2815 	ipmp_ifinfolist_t	*iflp;
2816 	ipmp_snap_t		*snap;
2817 	unsigned int		retval;
2818 
2819 	switch (miq->miq_inforeq) {
2820 	case IPMP_GROUPLIST:
2821 		retval = getgrouplist(&grlistp);
2822 		if (retval != IPMP_SUCCESS)
2823 			return (send_result(fd, retval, errno));
2824 
2825 		retval = send_result(fd, IPMP_SUCCESS, 0);
2826 		if (retval == IPMP_SUCCESS)
2827 			retval = send_grouplist(fd, grlistp);
2828 
2829 		ipmp_freegrouplist(grlistp);
2830 		return (retval);
2831 
2832 	case IPMP_GROUPINFO:
2833 		miq->miq_grname[LIFGRNAMSIZ - 1] = '\0';
2834 		retval = getgroupinfo(miq->miq_ifname, &grinfop);
2835 		if (retval != IPMP_SUCCESS)
2836 			return (send_result(fd, retval, errno));
2837 
2838 		retval = send_result(fd, IPMP_SUCCESS, 0);
2839 		if (retval == IPMP_SUCCESS)
2840 			retval = send_groupinfo(fd, grinfop);
2841 
2842 		ipmp_freegroupinfo(grinfop);
2843 		return (retval);
2844 
2845 	case IPMP_IFINFO:
2846 		miq->miq_ifname[LIFNAMSIZ - 1] = '\0';
2847 		retval = getifinfo(miq->miq_ifname, &ifinfop);
2848 		if (retval != IPMP_SUCCESS)
2849 			return (send_result(fd, retval, errno));
2850 
2851 		retval = send_result(fd, IPMP_SUCCESS, 0);
2852 		if (retval == IPMP_SUCCESS)
2853 			retval = send_ifinfo(fd, ifinfop);
2854 
2855 		ipmp_freeifinfo(ifinfop);
2856 		return (retval);
2857 
2858 	case IPMP_SNAP:
2859 		retval = getsnap(&snap);
2860 		if (retval != IPMP_SUCCESS)
2861 			return (send_result(fd, retval, errno));
2862 
2863 		retval = send_result(fd, IPMP_SUCCESS, 0);
2864 		if (retval != IPMP_SUCCESS)
2865 			goto out;
2866 
2867 		retval = ipmp_writetlv(fd, IPMP_SNAP, sizeof (*snap), snap);
2868 		if (retval != IPMP_SUCCESS)
2869 			goto out;
2870 
2871 		retval = send_grouplist(fd, snap->sn_grlistp);
2872 		if (retval != IPMP_SUCCESS)
2873 			goto out;
2874 
2875 		iflp = snap->sn_ifinfolistp;
2876 		for (; iflp != NULL; iflp = iflp->ifl_next) {
2877 			retval = send_ifinfo(fd, iflp->ifl_ifinfop);
2878 			if (retval != IPMP_SUCCESS)
2879 				goto out;
2880 		}
2881 
2882 		grlp = snap->sn_grinfolistp;
2883 		for (; grlp != NULL; grlp = grlp->grl_next) {
2884 			retval = send_groupinfo(fd, grlp->grl_grinfop);
2885 			if (retval != IPMP_SUCCESS)
2886 				goto out;
2887 		}
2888 	out:
2889 		ipmp_snap_free(snap);
2890 		return (retval);
2891 
2892 	default:
2893 		break;
2894 
2895 	}
2896 	return (send_result(fd, IPMP_EPROTO, 0));
2897 }
2898 
2899 /*
2900  * Send the group information pointed to by `grinfop' on file descriptor `fd'.
2901  * Returns an IPMP error code.
2902  */
2903 static unsigned int
2904 send_groupinfo(int fd, ipmp_groupinfo_t *grinfop)
2905 {
2906 	ipmp_iflist_t	*iflistp = grinfop->gr_iflistp;
2907 	unsigned int	retval;
2908 
2909 	retval = ipmp_writetlv(fd, IPMP_GROUPINFO, sizeof (*grinfop), grinfop);
2910 	if (retval != IPMP_SUCCESS)
2911 		return (retval);
2912 
2913 	return (ipmp_writetlv(fd, IPMP_IFLIST,
2914 	    IPMP_IFLIST_SIZE(iflistp->il_nif), iflistp));
2915 }
2916 
2917 /*
2918  * Send the interface information pointed to by `ifinfop' on file descriptor
2919  * `fd'.  Returns an IPMP error code.
2920  */
2921 static unsigned int
2922 send_ifinfo(int fd, ipmp_ifinfo_t *ifinfop)
2923 {
2924 	return (ipmp_writetlv(fd, IPMP_IFINFO, sizeof (*ifinfop), ifinfop));
2925 }
2926 
2927 /*
2928  * Send the group list pointed to by `grlistp' on file descriptor `fd'.
2929  * Returns an IPMP error code.
2930  */
2931 static unsigned int
2932 send_grouplist(int fd, ipmp_grouplist_t *grlistp)
2933 {
2934 	return (ipmp_writetlv(fd, IPMP_GROUPLIST,
2935 	    IPMP_GROUPLIST_SIZE(grlistp->gl_ngroup), grlistp));
2936 }
2937 
2938 /*
2939  * Initialize an mi_result_t structure using `error' and `syserror' and
2940  * send it on file descriptor `fd'.  Returns an IPMP error code.
2941  */
2942 static unsigned int
2943 send_result(int fd, unsigned int error, int syserror)
2944 {
2945 	mi_result_t me;
2946 
2947 	me.me_mpathd_error = error;
2948 	if (error == IPMP_FAILURE)
2949 		me.me_sys_error = syserror;
2950 	else
2951 		me.me_sys_error = 0;
2952 
2953 	return (ipmp_write(fd, &me, sizeof (me)));
2954 }
2955 
2956 /*
2957  * Daemonize the process.
2958  */
2959 static boolean_t
2960 daemonize(void)
2961 {
2962 	switch (fork()) {
2963 	case -1:
2964 		return (_B_FALSE);
2965 
2966 	case  0:
2967 		/*
2968 		 * Lose our controlling terminal, and become both a session
2969 		 * leader and a process group leader.
2970 		 */
2971 		if (setsid() == -1)
2972 			return (_B_FALSE);
2973 
2974 		/*
2975 		 * Under POSIX, a session leader can accidentally (through
2976 		 * open(2)) acquire a controlling terminal if it does not
2977 		 * have one.  Just to be safe, fork() again so we are not a
2978 		 * session leader.
2979 		 */
2980 		switch (fork()) {
2981 		case -1:
2982 			return (_B_FALSE);
2983 
2984 		case 0:
2985 			(void) chdir("/");
2986 			(void) umask(022);
2987 			(void) fdwalk(closefunc, NULL);
2988 			break;
2989 
2990 		default:
2991 			_exit(EXIT_SUCCESS);
2992 		}
2993 		break;
2994 
2995 	default:
2996 		_exit(EXIT_SUCCESS);
2997 	}
2998 
2999 	return (_B_TRUE);
3000 }
3001 
3002 /*
3003  * The parent has created some fds before forking on purpose, keep them open.
3004  */
3005 static int
3006 closefunc(void *not_used, int fd)
3007 /* ARGSUSED */
3008 {
3009 	if (fd != lsock_v4 && fd != lsock_v6)
3010 		(void) close(fd);
3011 	return (0);
3012 }
3013 
3014 /* LOGGER */
3015 
3016 #include <syslog.h>
3017 
3018 /*
3019  * Logging routines.  All routines log to syslog, unless the daemon is
3020  * running in the foreground, in which case the logging goes to stderr.
3021  *
3022  * The following routines are available:
3023  *
3024  *	logdebug(): A printf-like function for outputting debug messages
3025  *	(messages at LOG_DEBUG) that are only of use to developers.
3026  *
3027  *	logtrace(): A printf-like function for outputting tracing messages
3028  *	(messages at LOG_INFO) from the daemon.	 This is typically used
3029  *	to log the receipt of interesting network-related conditions.
3030  *
3031  *	logerr(): A printf-like function for outputting error messages
3032  *	(messages at LOG_ERR) from the daemon.
3033  *
3034  *	logperror*(): A set of functions used to output error messages
3035  *	(messages at LOG_ERR); these automatically append strerror(errno)
3036  *	and a newline to the message passed to them.
3037  *
3038  * NOTE: since the logging functions write to syslog, the messages passed
3039  *	 to them are not eligible for localization.  Thus, gettext() must
3040  *	 *not* be used.
3041  */
3042 
3043 static int logging = 0;
3044 
3045 static void
3046 initlog(void)
3047 {
3048 	logging++;
3049 	openlog("in.mpathd", LOG_PID | LOG_CONS, LOG_DAEMON);
3050 }
3051 
3052 /* PRINTFLIKE1 */
3053 void
3054 logerr(char *fmt, ...)
3055 {
3056 	va_list ap;
3057 
3058 	va_start(ap, fmt);
3059 
3060 	if (logging)
3061 		vsyslog(LOG_ERR, fmt, ap);
3062 	else
3063 		(void) vfprintf(stderr, fmt, ap);
3064 	va_end(ap);
3065 }
3066 
3067 /* PRINTFLIKE1 */
3068 void
3069 logtrace(char *fmt, ...)
3070 {
3071 	va_list ap;
3072 
3073 	va_start(ap, fmt);
3074 
3075 	if (logging)
3076 		vsyslog(LOG_INFO, fmt, ap);
3077 	else
3078 		(void) vfprintf(stderr, fmt, ap);
3079 	va_end(ap);
3080 }
3081 
3082 /* PRINTFLIKE1 */
3083 void
3084 logdebug(char *fmt, ...)
3085 {
3086 	va_list ap;
3087 
3088 	va_start(ap, fmt);
3089 
3090 	if (logging)
3091 		vsyslog(LOG_DEBUG, fmt, ap);
3092 	else
3093 		(void) vfprintf(stderr, fmt, ap);
3094 	va_end(ap);
3095 }
3096 
3097 /* PRINTFLIKE1 */
3098 void
3099 logperror(char *str)
3100 {
3101 	if (logging)
3102 		syslog(LOG_ERR, "%s: %m\n", str);
3103 	else
3104 		(void) fprintf(stderr, "%s: %s\n", str, strerror(errno));
3105 }
3106 
3107 void
3108 logperror_pii(struct phyint_instance *pii, char *str)
3109 {
3110 	if (logging) {
3111 		syslog(LOG_ERR, "%s (%s %s): %m\n",
3112 		    str, AF_STR(pii->pii_af), pii->pii_phyint->pi_name);
3113 	} else {
3114 		(void) fprintf(stderr, "%s (%s %s): %s\n",
3115 		    str, AF_STR(pii->pii_af), pii->pii_phyint->pi_name,
3116 		    strerror(errno));
3117 	}
3118 }
3119 
3120 void
3121 logperror_li(struct logint *li, char *str)
3122 {
3123 	struct	phyint_instance	*pii = li->li_phyint_inst;
3124 
3125 	if (logging) {
3126 		syslog(LOG_ERR, "%s (%s %s): %m\n",
3127 		    str, AF_STR(pii->pii_af), li->li_name);
3128 	} else {
3129 		(void) fprintf(stderr, "%s (%s %s): %s\n",
3130 		    str, AF_STR(pii->pii_af), li->li_name,
3131 		    strerror(errno));
3132 	}
3133 }
3134 
3135 void
3136 close_probe_socket(struct phyint_instance *pii, boolean_t polled)
3137 {
3138 	if (polled)
3139 		(void) poll_remove(pii->pii_probe_sock);
3140 	(void) close(pii->pii_probe_sock);
3141 	pii->pii_probe_sock = -1;
3142 	pii->pii_basetime_inited = 0;
3143 }
3144