xref: /titanic_44/usr/src/cmd/cmd-inet/sbin/dhcpagent/agent.c (revision 2ec7cc7fc084163eaed884efee9bbd322cc8951b)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/types.h>
27 #include <stdlib.h>
28 #include <assert.h>
29 #include <errno.h>
30 #include <locale.h>
31 #include <string.h>
32 #include <unistd.h>
33 #include <signal.h>
34 #include <stdio.h>
35 #include <stdio_ext.h>
36 #include <dhcp_hostconf.h>
37 #include <dhcpagent_ipc.h>
38 #include <dhcpagent_util.h>
39 #include <dhcpmsg.h>
40 #include <netinet/dhcp.h>
41 #include <net/route.h>
42 #include <sys/sockio.h>
43 #include <sys/stat.h>
44 #include <stropts.h>
45 #include <fcntl.h>
46 #include <sys/scsi/adapters/iscsi_if.h>
47 
48 #include "async.h"
49 #include "agent.h"
50 #include "script_handler.h"
51 #include "util.h"
52 #include "class_id.h"
53 #include "states.h"
54 #include "packet.h"
55 #include "interface.h"
56 #include "defaults.h"
57 
58 #ifndef	TEXT_DOMAIN
59 #define	TEXT_DOMAIN	"SYS_TEST"
60 #endif
61 
62 iu_timer_id_t		inactivity_id;
63 int			class_id_len = 0;
64 char			*class_id;
65 iu_eh_t			*eh;
66 iu_tq_t			*tq;
67 pid_t			grandparent;
68 int			rtsock_fd;
69 
70 static boolean_t	shutdown_started = B_FALSE;
71 static boolean_t	do_adopt = B_FALSE;
72 static unsigned int	debug_level = 0;
73 static iu_eh_callback_t	accept_event, ipc_event, rtsock_event;
74 
75 /*
76  * The ipc_cmd_allowed[] table indicates which IPC commands are allowed in
77  * which states; a non-zero value indicates the command is permitted.
78  *
79  * START is permitted if the state machine is fresh, or if we are in the
80  * process of trying to obtain a lease (as a convenience to save the
81  * administrator from having to do an explicit DROP).  EXTEND, RELEASE, and
82  * GET_TAG require a lease to be obtained in order to make sense.  INFORM is
83  * permitted if the interface is fresh or has an INFORM in progress or
84  * previously done on it -- otherwise a DROP or RELEASE is first required.
85  * PING and STATUS always make sense and thus are always permitted, as is DROP
86  * in order to permit the administrator to always bail out.
87  */
88 static int ipc_cmd_allowed[DHCP_NSTATES][DHCP_NIPC] = {
89 	/*			  D  E	P  R  S	 S  I  G */
90 	/*			  R  X	I  E  T	 T  N  E */
91 	/*			  O  T	N  L  A	 A  F  T */
92 	/*			  P  E	G  E  R	 T  O  _ */
93 	/*			  .  N  .  A  T  U  R  T */
94 	/*			  .  D	.  S  .  S  M  A */
95 	/*			  .  .  .  E  .  .  .  G */
96 	/* INIT		*/	{ 1, 0, 1, 0, 1, 1, 1, 0 },
97 	/* SELECTING	*/	{ 1, 0, 1, 0, 1, 1, 0, 0 },
98 	/* REQUESTING	*/	{ 1, 0, 1, 0, 1, 1, 0, 0 },
99 	/* PRE_BOUND	*/	{ 1, 1, 1, 1, 0, 1, 0, 1 },
100 	/* BOUND	*/	{ 1, 1, 1, 1, 0, 1, 0, 1 },
101 	/* RENEWING	*/	{ 1, 1, 1, 1, 0, 1, 0, 1 },
102 	/* REBINDING	*/	{ 1, 1, 1, 1, 0, 1, 0, 1 },
103 	/* INFORMATION  */	{ 1, 0, 1, 0, 1, 1, 1, 1 },
104 	/* INIT_REBOOT  */	{ 1, 0, 1, 1, 1, 1, 0, 0 },
105 	/* ADOPTING	*/	{ 1, 0, 1, 1, 0, 1, 0, 0 },
106 	/* INFORM_SENT  */	{ 1, 0, 1, 0, 1, 1, 1, 0 },
107 	/* DECLINING	*/	{ 1, 1, 1, 1, 0, 1, 0, 1 },
108 	/* RELEASING	*/	{ 1, 0, 1, 0, 0, 1, 0, 1 },
109 };
110 
111 #define	CMD_ISPRIV	0x1	/* Command requires privileges */
112 #define	CMD_CREATE	0x2	/* Command creates an interface */
113 #define	CMD_BOOTP	0x4	/* Command is valid with BOOTP */
114 #define	CMD_IMMED	0x8	/* Reply is immediate (no BUSY state) */
115 
116 static uint_t ipc_cmd_flags[DHCP_NIPC] = {
117 	/* DHCP_DROP */		CMD_ISPRIV|CMD_BOOTP,
118 	/* DHCP_EXTEND */	CMD_ISPRIV,
119 	/* DHCP_PING */		CMD_BOOTP|CMD_IMMED,
120 	/* DHCP_RELEASE */	CMD_ISPRIV,
121 	/* DHCP_START */	CMD_CREATE|CMD_ISPRIV|CMD_BOOTP,
122 	/* DHCP_STATUS */	CMD_BOOTP|CMD_IMMED,
123 	/* DHCP_INFORM */	CMD_CREATE|CMD_ISPRIV,
124 	/* DHCP_GET_TAG */	CMD_BOOTP|CMD_IMMED
125 };
126 
127 static boolean_t is_iscsi_active(void);
128 
129 int
130 main(int argc, char **argv)
131 {
132 	boolean_t	is_daemon  = B_TRUE;
133 	boolean_t	is_verbose;
134 	int		ipc_fd;
135 	int		c;
136 	int		aware = RTAW_UNDER_IPMP;
137 	struct rlimit	rl;
138 
139 	debug_level = df_get_int("", B_FALSE, DF_DEBUG_LEVEL);
140 	is_verbose = df_get_bool("", B_FALSE, DF_VERBOSE);
141 
142 	/*
143 	 * -l is ignored for compatibility with old agent.
144 	 */
145 
146 	while ((c = getopt(argc, argv, "vd:l:fa")) != EOF) {
147 
148 		switch (c) {
149 
150 		case 'a':
151 			do_adopt = B_TRUE;
152 			grandparent = getpid();
153 			break;
154 
155 		case 'd':
156 			debug_level = strtoul(optarg, NULL, 0);
157 			break;
158 
159 		case 'f':
160 			is_daemon = B_FALSE;
161 			break;
162 
163 		case 'v':
164 			is_verbose = B_TRUE;
165 			break;
166 
167 		case '?':
168 			(void) fprintf(stderr, "usage: %s [-a] [-d n] [-f] [-v]"
169 			    "\n", argv[0]);
170 			return (EXIT_FAILURE);
171 
172 		default:
173 			break;
174 		}
175 	}
176 
177 	(void) setlocale(LC_ALL, "");
178 	(void) textdomain(TEXT_DOMAIN);
179 
180 	if (geteuid() != 0) {
181 		dhcpmsg_init(argv[0], B_FALSE, is_verbose, debug_level);
182 		dhcpmsg(MSG_ERROR, "must be super-user");
183 		dhcpmsg_fini();
184 		return (EXIT_FAILURE);
185 	}
186 
187 	if (is_daemon && daemonize() == 0) {
188 		dhcpmsg_init(argv[0], B_FALSE, is_verbose, debug_level);
189 		dhcpmsg(MSG_ERR, "cannot become daemon, exiting");
190 		dhcpmsg_fini();
191 		return (EXIT_FAILURE);
192 	}
193 
194 	/*
195 	 * Seed the random number generator, since we're going to need it
196 	 * to set transaction id's and for exponential backoff.
197 	 */
198 	srand48(gethrtime() ^ gethostid() ^ getpid());
199 
200 	dhcpmsg_init(argv[0], is_daemon, is_verbose, debug_level);
201 	(void) atexit(dhcpmsg_fini);
202 
203 	tq = iu_tq_create();
204 	eh = iu_eh_create();
205 
206 	if (eh == NULL || tq == NULL) {
207 		errno = ENOMEM;
208 		dhcpmsg(MSG_ERR, "cannot create timer queue or event handler");
209 		return (EXIT_FAILURE);
210 	}
211 
212 	/*
213 	 * ignore most signals that could be reasonably generated.
214 	 */
215 
216 	(void) signal(SIGTERM, graceful_shutdown);
217 	(void) signal(SIGQUIT, graceful_shutdown);
218 	(void) signal(SIGPIPE, SIG_IGN);
219 	(void) signal(SIGUSR1, SIG_IGN);
220 	(void) signal(SIGUSR2, SIG_IGN);
221 	(void) signal(SIGINT,  SIG_IGN);
222 	(void) signal(SIGHUP,  SIG_IGN);
223 	(void) signal(SIGCHLD, SIG_IGN);
224 
225 	/*
226 	 * upon SIGTHAW we need to refresh any non-infinite leases.
227 	 */
228 
229 	(void) iu_eh_register_signal(eh, SIGTHAW, refresh_smachs, NULL);
230 
231 	class_id = get_class_id();
232 	if (class_id != NULL)
233 		class_id_len = strlen(class_id);
234 	else
235 		dhcpmsg(MSG_WARNING, "get_class_id failed, continuing "
236 		    "with no vendor class id");
237 
238 	/*
239 	 * the inactivity timer is enabled any time there are no
240 	 * interfaces under DHCP control.  if DHCP_INACTIVITY_WAIT
241 	 * seconds transpire without an interface under DHCP control,
242 	 * the agent shuts down.
243 	 */
244 
245 	inactivity_id = iu_schedule_timer(tq, DHCP_INACTIVITY_WAIT,
246 	    inactivity_shutdown, NULL);
247 
248 	/*
249 	 * max out the number available descriptors, just in case..
250 	 */
251 
252 	rl.rlim_cur = RLIM_INFINITY;
253 	rl.rlim_max = RLIM_INFINITY;
254 	if (setrlimit(RLIMIT_NOFILE, &rl) == -1)
255 		dhcpmsg(MSG_ERR, "setrlimit failed");
256 
257 	(void) enable_extended_FILE_stdio(-1, -1);
258 
259 	/*
260 	 * Create and bind default IP sockets used to control interfaces and to
261 	 * catch stray packets.
262 	 */
263 
264 	if (!dhcp_ip_default())
265 		return (EXIT_FAILURE);
266 
267 	/*
268 	 * create the ipc channel that the agent will listen for
269 	 * requests on, and register it with the event handler so that
270 	 * `accept_event' will be called back.
271 	 */
272 
273 	switch (dhcp_ipc_init(&ipc_fd)) {
274 
275 	case 0:
276 		break;
277 
278 	case DHCP_IPC_E_BIND:
279 		dhcpmsg(MSG_ERROR, "dhcp_ipc_init: cannot bind to port "
280 		    "%i (agent already running?)", IPPORT_DHCPAGENT);
281 		return (EXIT_FAILURE);
282 
283 	default:
284 		dhcpmsg(MSG_ERROR, "dhcp_ipc_init failed");
285 		return (EXIT_FAILURE);
286 	}
287 
288 	if (iu_register_event(eh, ipc_fd, POLLIN, accept_event, 0) == -1) {
289 		dhcpmsg(MSG_ERR, "cannot register ipc fd for messages");
290 		return (EXIT_FAILURE);
291 	}
292 
293 	/*
294 	 * Create the global routing socket.  This is used for monitoring
295 	 * interface transitions, so that we learn about the kernel's Duplicate
296 	 * Address Detection status, and for inserting and removing default
297 	 * routes as learned from DHCP servers.  Both v4 and v6 are handed
298 	 * with this one socket.
299 	 */
300 	rtsock_fd = socket(PF_ROUTE, SOCK_RAW, 0);
301 	if (rtsock_fd == -1) {
302 		dhcpmsg(MSG_ERR, "cannot open routing socket");
303 		return (EXIT_FAILURE);
304 	}
305 
306 	/*
307 	 * We're IPMP-aware and can manage IPMP test addresses, so issue
308 	 * RT_AWARE to get routing socket messages for interfaces under IPMP.
309 	 */
310 	if (setsockopt(rtsock_fd, SOL_ROUTE, RT_AWARE, &aware,
311 	    sizeof (aware)) == -1) {
312 		dhcpmsg(MSG_ERR, "cannot set RT_AWARE on routing socket");
313 		return (EXIT_FAILURE);
314 	}
315 
316 	if (iu_register_event(eh, rtsock_fd, POLLIN, rtsock_event, 0) == -1) {
317 		dhcpmsg(MSG_ERR, "cannot register routing socket for messages");
318 		return (EXIT_FAILURE);
319 	}
320 
321 	/*
322 	 * if the -a (adopt) option was specified, try to adopt the
323 	 * kernel-managed interface before we start.
324 	 */
325 
326 	if (do_adopt && !dhcp_adopt())
327 		return (EXIT_FAILURE);
328 
329 	/*
330 	 * For DHCPv6, we own all of the interfaces marked DHCPRUNNING.  As
331 	 * we're starting operation here, if there are any of those interfaces
332 	 * lingering around, they're strays, and need to be removed.
333 	 *
334 	 * It might be nice to save these addresses off somewhere -- for both
335 	 * v4 and v6 -- and use them as hints for later negotiation.
336 	 */
337 	remove_v6_strays();
338 
339 	/*
340 	 * enter the main event loop; this is where all the real work
341 	 * takes place (through registering events and scheduling timers).
342 	 * this function only returns when the agent is shutting down.
343 	 */
344 
345 	switch (iu_handle_events(eh, tq)) {
346 
347 	case -1:
348 		dhcpmsg(MSG_WARNING, "iu_handle_events exited abnormally");
349 		break;
350 
351 	case DHCP_REASON_INACTIVITY:
352 		dhcpmsg(MSG_INFO, "no interfaces to manage, shutting down...");
353 		break;
354 
355 	case DHCP_REASON_TERMINATE:
356 		dhcpmsg(MSG_INFO, "received SIGTERM, shutting down...");
357 		break;
358 
359 	case DHCP_REASON_SIGNAL:
360 		dhcpmsg(MSG_WARNING, "received unexpected signal, shutting "
361 		    "down...");
362 		break;
363 	}
364 
365 	(void) iu_eh_unregister_signal(eh, SIGTHAW, NULL);
366 
367 	iu_eh_destroy(eh);
368 	iu_tq_destroy(tq);
369 
370 	return (EXIT_SUCCESS);
371 }
372 
373 /*
374  * drain_script(): event loop callback during shutdown
375  *
376  *   input: eh_t *: unused
377  *	    void *: unused
378  *  output: boolean_t: B_TRUE if event loop should exit; B_FALSE otherwise
379  */
380 
381 /* ARGSUSED */
382 boolean_t
383 drain_script(iu_eh_t *ehp, void *arg)
384 {
385 	if (shutdown_started == B_FALSE) {
386 		shutdown_started = B_TRUE;
387 		/*
388 		 * Check if the system is diskless client and/or
389 		 * there are active iSCSI sessions
390 		 *
391 		 * Do not drop the lease, or the system will be
392 		 * unable to sync(dump) through nfs/iSCSI driver
393 		 */
394 		if (!do_adopt && !is_iscsi_active()) {
395 			nuke_smach_list();
396 		}
397 	}
398 	return (script_count == 0);
399 }
400 
401 /*
402  * accept_event(): accepts a new connection on the ipc socket and registers
403  *		   to receive its messages with the event handler
404  *
405  *   input: iu_eh_t *: unused
406  *	    int: the file descriptor in the iu_eh_t * the connection came in on
407  *	    (other arguments unused)
408  *  output: void
409  */
410 
411 /* ARGSUSED */
412 static void
413 accept_event(iu_eh_t *ehp, int fd, short events, iu_event_id_t id, void *arg)
414 {
415 	int	client_fd;
416 	int	is_priv;
417 
418 	if (dhcp_ipc_accept(fd, &client_fd, &is_priv) != 0) {
419 		dhcpmsg(MSG_ERR, "accept_event: accept on ipc socket");
420 		return;
421 	}
422 
423 	if (iu_register_event(eh, client_fd, POLLIN, ipc_event,
424 	    (void *)is_priv) == -1) {
425 		dhcpmsg(MSG_ERROR, "accept_event: cannot register ipc socket "
426 		    "for callback");
427 	}
428 }
429 
430 /*
431  * ipc_event(): processes incoming ipc requests
432  *
433  *   input: iu_eh_t *: unused
434  *	    int: the file descriptor in the iu_eh_t * the request came in on
435  *	    short: unused
436  *	    iu_event_id_t: event ID
437  *	    void *: indicates whether the request is from a privileged client
438  *  output: void
439  */
440 
441 /* ARGSUSED */
442 static void
443 ipc_event(iu_eh_t *ehp, int fd, short events, iu_event_id_t id, void *arg)
444 {
445 	ipc_action_t		ia, *iap;
446 	dhcp_smach_t		*dsmp;
447 	int			error, is_priv = (int)arg;
448 	const char		*ifname;
449 	boolean_t		isv6;
450 
451 	ipc_action_init(&ia);
452 	error = dhcp_ipc_recv_request(fd, &ia.ia_request,
453 	    DHCP_IPC_REQUEST_WAIT);
454 	if (error != DHCP_IPC_SUCCESS) {
455 		if (error != DHCP_IPC_E_EOF) {
456 			dhcpmsg(MSG_ERROR,
457 			    "ipc_event: dhcp_ipc_recv_request failed: %s",
458 			    dhcp_ipc_strerror(error));
459 		} else {
460 			dhcpmsg(MSG_DEBUG, "ipc_event: connection closed");
461 		}
462 		if ((dsmp = lookup_smach_by_event(id)) != NULL) {
463 			ipc_action_finish(dsmp, error);
464 		} else {
465 			(void) iu_unregister_event(eh, id, NULL);
466 			(void) dhcp_ipc_close(fd);
467 		}
468 		return;
469 	}
470 
471 	/* Fill in temporary ipc_action structure for utility functions */
472 	ia.ia_cmd = DHCP_IPC_CMD(ia.ia_request->message_type);
473 	ia.ia_fd = fd;
474 	ia.ia_eid = id;
475 
476 	if (ia.ia_cmd >= DHCP_NIPC) {
477 		dhcpmsg(MSG_ERROR,
478 		    "ipc_event: invalid command (%s) attempted on %s",
479 		    dhcp_ipc_type_to_string(ia.ia_cmd), ia.ia_request->ifname);
480 		send_error_reply(&ia, DHCP_IPC_E_CMD_UNKNOWN);
481 		return;
482 	}
483 
484 	/* return EPERM for any of the privileged actions */
485 
486 	if (!is_priv && (ipc_cmd_flags[ia.ia_cmd] & CMD_ISPRIV)) {
487 		dhcpmsg(MSG_WARNING,
488 		    "ipc_event: privileged ipc command (%s) attempted on %s",
489 		    dhcp_ipc_type_to_string(ia.ia_cmd), ia.ia_request->ifname);
490 		send_error_reply(&ia, DHCP_IPC_E_PERM);
491 		return;
492 	}
493 
494 	/*
495 	 * Try to locate the state machine associated with this command.  If
496 	 * the command is DHCP_START or DHCP_INFORM and there isn't a state
497 	 * machine already, make one (there may already be one from a previous
498 	 * failed attempt to START or INFORM).  Otherwise, verify the reference
499 	 * is still valid.
500 	 *
501 	 * The interface name may be blank.  In that case, we look up the
502 	 * primary interface, and the requested type (v4 or v6) doesn't matter.
503 	 */
504 
505 	isv6 = (ia.ia_request->message_type & DHCP_V6) != 0;
506 	ifname = ia.ia_request->ifname;
507 	if (*ifname == '\0')
508 		dsmp = primary_smach(isv6);
509 	else
510 		dsmp = lookup_smach(ifname, isv6);
511 
512 	if (dsmp != NULL) {
513 		/* Note that verify_smach drops a reference */
514 		hold_smach(dsmp);
515 		if (!verify_smach(dsmp))
516 			dsmp = NULL;
517 	}
518 
519 	if (dsmp == NULL) {
520 		/*
521 		 * If the user asked for the primary DHCP interface by giving
522 		 * an empty string and there is no primary, then check if we're
523 		 * handling dhcpinfo.  If so, then simulate primary selection.
524 		 * Otherwise, report failure.
525 		 */
526 		if (ifname[0] == '\0') {
527 			if (ia.ia_cmd == DHCP_GET_TAG)
528 				dsmp = info_primary_smach(isv6);
529 			if (dsmp == NULL)
530 				error = DHCP_IPC_E_NOPRIMARY;
531 
532 		/*
533 		 * If there's no interface, and we're starting up, then create
534 		 * it now, along with a state machine for it.  Note that if
535 		 * insert_smach fails, it discards the LIF reference.
536 		 */
537 		} else if (ipc_cmd_flags[ia.ia_cmd] & CMD_CREATE) {
538 			dhcp_lif_t *lif;
539 
540 			lif = attach_lif(ifname, isv6, &error);
541 			if (lif != NULL &&
542 			    (dsmp = insert_smach(lif, &error)) != NULL) {
543 				/*
544 				 * Get client ID and set "DHCPRUNNING" flag on
545 				 * logical interface.  (V4 only, because V6
546 				 * plumbs its own interfaces.)
547 				 */
548 				error = get_smach_cid(dsmp);
549 				if (error == DHCP_IPC_SUCCESS)
550 					error = set_lif_dhcp(lif, B_FALSE);
551 				if (error != DHCP_IPC_SUCCESS) {
552 					remove_smach(dsmp);
553 					dsmp = NULL;
554 				}
555 			}
556 
557 		/*
558 		 * Otherwise, this is an operation on an unknown interface.
559 		 */
560 		} else {
561 			error = DHCP_IPC_E_UNKIF;
562 		}
563 		if (dsmp == NULL) {
564 			send_error_reply(&ia, error);
565 			return;
566 		}
567 	}
568 
569 	if ((dsmp->dsm_dflags & DHCP_IF_BOOTP) &&
570 	    !(ipc_cmd_flags[ia.ia_cmd] & CMD_BOOTP)) {
571 		dhcpmsg(MSG_ERROR, "command %s not valid for BOOTP on %s",
572 		    dhcp_ipc_type_to_string(ia.ia_cmd), dsmp->dsm_name);
573 		send_error_reply(&ia, DHCP_IPC_E_BOOTP);
574 		return;
575 	}
576 
577 	/*
578 	 * verify that the state machine is in a state which will allow the
579 	 * command.  we do this up front so that we can return an error
580 	 * *before* needlessly cancelling an in-progress transaction.
581 	 */
582 
583 	if (!check_cmd_allowed(dsmp->dsm_state, ia.ia_cmd)) {
584 		dhcpmsg(MSG_DEBUG,
585 		    "in state %s; not allowing %s command on %s",
586 		    dhcp_state_to_string(dsmp->dsm_state),
587 		    dhcp_ipc_type_to_string(ia.ia_cmd), dsmp->dsm_name);
588 		send_error_reply(&ia,
589 		    ia.ia_cmd == DHCP_START && dsmp->dsm_state != INIT ?
590 		    DHCP_IPC_E_RUNNING : DHCP_IPC_E_OUTSTATE);
591 		return;
592 	}
593 
594 	dhcpmsg(MSG_DEBUG, "in state %s; allowing %s command on %s",
595 	    dhcp_state_to_string(dsmp->dsm_state),
596 	    dhcp_ipc_type_to_string(ia.ia_cmd), dsmp->dsm_name);
597 
598 	if ((ia.ia_request->message_type & DHCP_PRIMARY) && is_priv)
599 		make_primary(dsmp);
600 
601 	/*
602 	 * The current design dictates that there can be only one outstanding
603 	 * transaction per state machine -- this simplifies the code
604 	 * considerably and also fits well with RFCs 2131 and 3315.  It is
605 	 * worth classifying the different DHCP commands into synchronous
606 	 * (those which we will handle now and reply to immediately) and
607 	 * asynchronous (those which require transactions and will be completed
608 	 * at an indeterminate time in the future):
609 	 *
610 	 *    DROP: removes the agent's management of a state machine.
611 	 *	    asynchronous as the script program may be invoked.
612 	 *
613 	 *    PING: checks to see if the agent has a named state machine.
614 	 *	    synchronous, since no packets need to be sent
615 	 *	    to the DHCP server.
616 	 *
617 	 *  STATUS: returns information about a state machine.
618 	 *	    synchronous, since no packets need to be sent
619 	 *	    to the DHCP server.
620 	 *
621 	 * RELEASE: releases the agent's management of a state machine
622 	 *	    and brings the associated interfaces down.  asynchronous
623 	 *	    as the script program may be invoked.
624 	 *
625 	 *  EXTEND: renews a lease.  asynchronous, since the agent
626 	 *	    needs to wait for an ACK, etc.
627 	 *
628 	 *   START: starts DHCP on a named state machine.  asynchronous since
629 	 *	    the agent needs to wait for OFFERs, ACKs, etc.
630 	 *
631 	 *  INFORM: obtains configuration parameters for the system using
632 	 *	    externally configured interface.  asynchronous, since the
633 	 *	    agent needs to wait for an ACK.
634 	 *
635 	 * Notice that EXTEND, INFORM, START, DROP and RELEASE are
636 	 * asynchronous.  Notice also that asynchronous commands may occur from
637 	 * within the agent -- for instance, the agent will need to do implicit
638 	 * EXTENDs to extend the lease. In order to make the code simpler, the
639 	 * following rules apply for asynchronous commands:
640 	 *
641 	 * There can only be one asynchronous command at a time per state
642 	 * machine.  The current asynchronous command is managed by the async_*
643 	 * api: async_start(), async_finish(), and async_cancel().
644 	 * async_start() starts management of a new asynchronous command on an
645 	 * state machine, which should only be done after async_cancel() to
646 	 * terminate a previous command.  When the command is completed,
647 	 * async_finish() should be called.
648 	 *
649 	 * Asynchronous commands started by a user command have an associated
650 	 * ipc_action which provides the agent with information for how to get
651 	 * in touch with the user command when the action completes.  These
652 	 * ipc_action records also have an associated timeout which may be
653 	 * infinite.  ipc_action_start() should be called when starting an
654 	 * asynchronous command requested by a user, which sets up the timer
655 	 * and keeps track of the ipc information (file descriptor, request
656 	 * type).  When the asynchronous command completes, ipc_action_finish()
657 	 * should be called to return a command status code to the user and
658 	 * close the ipc connection).  If the command does not complete before
659 	 * the timer fires, ipc_action_timeout() is called which closes the ipc
660 	 * connection and returns DHCP_IPC_E_TIMEOUT to the user.  Note that
661 	 * independent of ipc_action_timeout(), ipc_action_finish() should be
662 	 * called.
663 	 *
664 	 * on a case-by-case basis, here is what happens (per state machine):
665 	 *
666 	 *    o When an asynchronous command is requested, then
667 	 *	async_cancel() is called to terminate any non-user
668 	 *	action in progress.  If there's a user action running,
669 	 *	the user command is sent DHCP_IPC_E_PEND.
670 	 *
671 	 *    o otherwise, the the transaction is started with
672 	 *	async_start().  if the transaction is on behalf
673 	 *	of a user, ipc_action_start() is called to keep
674 	 *	track of the ipc information and set up the
675 	 *	ipc_action timer.
676 	 *
677 	 *    o if the command completes normally and before a
678 	 *	timeout fires, then async_finish() is called.
679 	 *	if there was an associated ipc_action,
680 	 *	ipc_action_finish() is called to complete it.
681 	 *
682 	 *    o if the command fails before a timeout fires, then
683 	 *	async_finish() is called, and the state machine is
684 	 *	is returned to a known state based on the command.
685 	 *	if there was an associated ipc_action,
686 	 *	ipc_action_finish() is called to complete it.
687 	 *
688 	 *    o if the ipc_action timer fires before command
689 	 *	completion, then DHCP_IPC_E_TIMEOUT is returned to
690 	 *	the user.  however, the transaction continues to
691 	 *	be carried out asynchronously.
692 	 */
693 
694 	if (ipc_cmd_flags[ia.ia_cmd] & CMD_IMMED) {
695 		/*
696 		 * Only immediate commands (ping, status, get_tag) need to
697 		 * worry about freeing ia through one of the reply functions
698 		 * before returning.
699 		 */
700 		iap = &ia;
701 	} else {
702 		/*
703 		 * if shutdown request has been received, send back an error.
704 		 */
705 		if (shutdown_started) {
706 			send_error_reply(&ia, DHCP_IPC_E_OUTSTATE);
707 			return;
708 		}
709 
710 		if (dsmp->dsm_dflags & DHCP_IF_BUSY) {
711 			send_error_reply(&ia, DHCP_IPC_E_PEND);
712 			return;
713 		}
714 
715 		if (!ipc_action_start(dsmp, &ia)) {
716 			dhcpmsg(MSG_WARNING, "ipc_event: ipc_action_start "
717 			    "failed for %s", dsmp->dsm_name);
718 			send_error_reply(&ia, DHCP_IPC_E_MEMORY);
719 			return;
720 		}
721 
722 		/* Action structure consumed by above function */
723 		iap = &dsmp->dsm_ia;
724 	}
725 
726 	switch (iap->ia_cmd) {
727 
728 	case DHCP_DROP:
729 		if (dsmp->dsm_droprelease)
730 			break;
731 		dsmp->dsm_droprelease = B_TRUE;
732 
733 		/*
734 		 * Ensure that a timer associated with the existing state
735 		 * doesn't pop while we're waiting for the script to complete.
736 		 * (If so, chaos can result -- e.g., a timer causes us to end
737 		 * up in dhcp_selecting() would start acquiring a new lease on
738 		 * dsmp while our DHCP_DROP dismantling is ongoing.)
739 		 */
740 		cancel_smach_timers(dsmp);
741 		(void) script_start(dsmp, isv6 ? EVENT_DROP6 : EVENT_DROP,
742 		    dhcp_drop, NULL, NULL);
743 		break;		/* not an immediate function */
744 
745 	case DHCP_EXTEND:
746 		(void) dhcp_extending(dsmp);
747 		break;
748 
749 	case DHCP_GET_TAG: {
750 		dhcp_optnum_t	optnum;
751 		void		*opt = NULL;
752 		uint_t		optlen;
753 		boolean_t	did_alloc = B_FALSE;
754 		PKT_LIST	*ack = dsmp->dsm_ack;
755 		int		i;
756 
757 		/*
758 		 * verify the request makes sense.
759 		 */
760 
761 		if (iap->ia_request->data_type   != DHCP_TYPE_OPTNUM ||
762 		    iap->ia_request->data_length != sizeof (dhcp_optnum_t)) {
763 			send_error_reply(iap, DHCP_IPC_E_PROTO);
764 			break;
765 		}
766 
767 		(void) memcpy(&optnum, iap->ia_request->buffer,
768 		    sizeof (dhcp_optnum_t));
769 
770 load_option:
771 		switch (optnum.category) {
772 
773 		case DSYM_SITE:			/* FALLTHRU */
774 		case DSYM_STANDARD:
775 			for (i = 0; i < dsmp->dsm_pillen; i++) {
776 				if (dsmp->dsm_pil[i] == optnum.code)
777 					break;
778 			}
779 			if (i < dsmp->dsm_pillen)
780 				break;
781 			if (isv6) {
782 				opt = dhcpv6_pkt_option(ack, NULL, optnum.code,
783 				    NULL);
784 			} else {
785 				if (optnum.code <= DHCP_LAST_OPT)
786 					opt = ack->opts[optnum.code];
787 			}
788 			break;
789 
790 		case DSYM_VENDOR:
791 			if (isv6) {
792 				dhcpv6_option_t *d6o;
793 				uint32_t ent;
794 
795 				/*
796 				 * Look through vendor options to find our
797 				 * enterprise number.
798 				 */
799 				d6o = NULL;
800 				for (;;) {
801 					d6o = dhcpv6_pkt_option(ack, d6o,
802 					    DHCPV6_OPT_VENDOR_OPT, &optlen);
803 					if (d6o == NULL)
804 						break;
805 					optlen -= sizeof (*d6o);
806 					if (optlen < sizeof (ent))
807 						continue;
808 					(void) memcpy(&ent, d6o + 1,
809 					    sizeof (ent));
810 					if (ntohl(ent) != DHCPV6_SUN_ENT)
811 						continue;
812 					break;
813 				}
814 				if (d6o != NULL) {
815 					/*
816 					 * Now find the requested vendor option
817 					 * within the vendor options block.
818 					 */
819 					opt = dhcpv6_find_option(
820 					    (char *)(d6o + 1) + sizeof (ent),
821 					    optlen - sizeof (ent), NULL,
822 					    optnum.code, NULL);
823 				}
824 			} else {
825 				/*
826 				 * the test against VS_OPTION_START is broken
827 				 * up into two tests to avoid compiler warnings
828 				 * under intel.
829 				 */
830 				if ((optnum.code > VS_OPTION_START ||
831 				    optnum.code == VS_OPTION_START) &&
832 				    optnum.code <= VS_OPTION_END)
833 					opt = ack->vs[optnum.code];
834 			}
835 			break;
836 
837 		case DSYM_FIELD:
838 			if (isv6) {
839 				dhcpv6_message_t *d6m =
840 				    (dhcpv6_message_t *)ack->pkt;
841 				dhcpv6_option_t *d6o;
842 
843 				/* Validate the packet field the user wants */
844 				optlen = optnum.code + optnum.size;
845 				if (d6m->d6m_msg_type ==
846 				    DHCPV6_MSG_RELAY_FORW ||
847 				    d6m->d6m_msg_type ==
848 				    DHCPV6_MSG_RELAY_REPL) {
849 					if (optlen > sizeof (dhcpv6_relay_t))
850 						break;
851 				} else {
852 					if (optlen > sizeof (*d6m))
853 						break;
854 				}
855 
856 				opt = malloc(sizeof (*d6o) + optnum.size);
857 				if (opt != NULL) {
858 					d6o = opt;
859 					d6o->d6o_code = htons(optnum.code);
860 					d6o->d6o_len = htons(optnum.size);
861 					(void) memcpy(d6o + 1, (caddr_t)d6m +
862 					    optnum.code, optnum.size);
863 				}
864 			} else {
865 				if (optnum.code + optnum.size > sizeof (PKT))
866 					break;
867 
868 				/*
869 				 * + 2 to account for option code and length
870 				 * byte
871 				 */
872 				opt = malloc(optnum.size + 2);
873 				if (opt != NULL) {
874 					DHCP_OPT *v4opt = opt;
875 
876 					v4opt->len  = optnum.size;
877 					v4opt->code = optnum.code;
878 					(void) memcpy(v4opt->value,
879 					    (caddr_t)ack->pkt + optnum.code,
880 					    optnum.size);
881 				}
882 			}
883 
884 			if (opt == NULL) {
885 				send_error_reply(iap, DHCP_IPC_E_MEMORY);
886 				return;
887 			}
888 			did_alloc = B_TRUE;
889 			break;
890 
891 		default:
892 			send_error_reply(iap, DHCP_IPC_E_PROTO);
893 			return;
894 		}
895 
896 		/*
897 		 * return the option payload, if there was one.  the "+ 2"
898 		 * accounts for the option code number and length byte.
899 		 */
900 
901 		if (opt != NULL) {
902 			if (isv6) {
903 				dhcpv6_option_t d6ov;
904 
905 				(void) memcpy(&d6ov, opt, sizeof (d6ov));
906 				optlen = ntohs(d6ov.d6o_len) + sizeof (d6ov);
907 			} else {
908 				optlen = ((DHCP_OPT *)opt)->len + 2;
909 			}
910 			send_data_reply(iap, 0, DHCP_TYPE_OPTION, opt, optlen);
911 
912 			if (did_alloc)
913 				free(opt);
914 			break;
915 		} else if (ack != dsmp->dsm_orig_ack) {
916 			/*
917 			 * There wasn't any definition for the option in the
918 			 * current ack, so now retry with the original ack if
919 			 * the original ack is not the current ack.
920 			 */
921 			ack = dsmp->dsm_orig_ack;
922 			goto load_option;
923 		}
924 
925 		/*
926 		 * note that an "okay" response is returned either in
927 		 * the case of an unknown option or a known option
928 		 * with no payload.  this is okay (for now) since
929 		 * dhcpinfo checks whether an option is valid before
930 		 * ever performing ipc with the agent.
931 		 */
932 
933 		send_ok_reply(iap);
934 		break;
935 	}
936 
937 	case DHCP_INFORM:
938 		dhcp_inform(dsmp);
939 		/* next destination: dhcp_acknak() */
940 		break;		/* not an immediate function */
941 
942 	case DHCP_PING:
943 		if (dsmp->dsm_dflags & DHCP_IF_FAILED)
944 			send_error_reply(iap, DHCP_IPC_E_FAILEDIF);
945 		else
946 			send_ok_reply(iap);
947 		break;
948 
949 	case DHCP_RELEASE:
950 		if (dsmp->dsm_droprelease)
951 			break;
952 		dsmp->dsm_droprelease = B_TRUE;
953 		cancel_smach_timers(dsmp); /* see comment in DHCP_DROP above */
954 		(void) script_start(dsmp, isv6 ? EVENT_RELEASE6 :
955 		    EVENT_RELEASE, dhcp_release, "Finished with lease.", NULL);
956 		break;		/* not an immediate function */
957 
958 	case DHCP_START: {
959 		PKT_LIST *ack, *oack;
960 		PKT_LIST *plp[2];
961 
962 		deprecate_leases(dsmp);
963 
964 		/*
965 		 * if we have a valid hostconf lying around, then jump
966 		 * into INIT_REBOOT.  if it fails, we'll end up going
967 		 * through the whole selecting() procedure again.
968 		 */
969 
970 		error = read_hostconf(dsmp->dsm_name, plp, 2, dsmp->dsm_isv6);
971 		ack = error > 0 ? plp[0] : NULL;
972 		oack = error > 1 ? plp[1] : NULL;
973 
974 		/*
975 		 * If the allocation of the old ack fails, that's fine;
976 		 * continue without it.
977 		 */
978 		if (oack == NULL)
979 			oack = ack;
980 
981 		/*
982 		 * As long as we've allocated something, start using it.
983 		 */
984 		if (ack != NULL) {
985 			dsmp->dsm_orig_ack = oack;
986 			dsmp->dsm_ack = ack;
987 			dhcp_init_reboot(dsmp);
988 			/* next destination: dhcp_acknak() */
989 			break;
990 		}
991 
992 		/*
993 		 * if not debugging, wait for a few seconds before
994 		 * going into SELECTING.
995 		 */
996 
997 		if (debug_level == 0 && set_start_timer(dsmp)) {
998 			/* next destination: dhcp_start() */
999 			break;
1000 		} else {
1001 			dhcp_selecting(dsmp);
1002 			/* next destination: dhcp_requesting() */
1003 			break;
1004 		}
1005 	}
1006 
1007 	case DHCP_STATUS: {
1008 		dhcp_status_t	status;
1009 		dhcp_lease_t	*dlp;
1010 
1011 		status.if_began = monosec_to_time(dsmp->dsm_curstart_monosec);
1012 
1013 		/*
1014 		 * We return information on just the first lease as being
1015 		 * representative of the lot.  A better status mechanism is
1016 		 * needed.
1017 		 */
1018 		dlp = dsmp->dsm_leases;
1019 
1020 		if (dlp == NULL ||
1021 		    dlp->dl_lifs->lif_expire.dt_start == DHCP_PERM) {
1022 			status.if_t1	= DHCP_PERM;
1023 			status.if_t2	= DHCP_PERM;
1024 			status.if_lease	= DHCP_PERM;
1025 		} else {
1026 			status.if_t1	= status.if_began +
1027 			    dlp->dl_t1.dt_start;
1028 			status.if_t2	= status.if_began +
1029 			    dlp->dl_t2.dt_start;
1030 			status.if_lease	= status.if_began +
1031 			    dlp->dl_lifs->lif_expire.dt_start;
1032 		}
1033 
1034 		status.version		= DHCP_STATUS_VER;
1035 		status.if_state		= dsmp->dsm_state;
1036 		status.if_dflags	= dsmp->dsm_dflags;
1037 		status.if_sent		= dsmp->dsm_sent;
1038 		status.if_recv		= dsmp->dsm_received;
1039 		status.if_bad_offers	= dsmp->dsm_bad_offers;
1040 
1041 		(void) strlcpy(status.if_name, dsmp->dsm_name, LIFNAMSIZ);
1042 
1043 		send_data_reply(iap, 0, DHCP_TYPE_STATUS, &status,
1044 		    sizeof (dhcp_status_t));
1045 		break;
1046 	}
1047 	}
1048 }
1049 
1050 /*
1051  * check_rtm_addr(): determine if routing socket message matches interface
1052  *		     address
1053  *
1054  *   input: const struct if_msghdr *: pointer to routing socket message
1055  *	    int: routing socket message length
1056  *	    boolean_t: set to B_TRUE if IPv6
1057  *	    const in6_addr_t *: pointer to IP address
1058  *  output: boolean_t: B_TRUE if address is a match
1059  */
1060 
1061 static boolean_t
1062 check_rtm_addr(const struct ifa_msghdr *ifam, int msglen, boolean_t isv6,
1063     const in6_addr_t *addr)
1064 {
1065 	const char *cp, *lim;
1066 	uint_t flag;
1067 	const struct sockaddr *sa;
1068 
1069 	if (!(ifam->ifam_addrs & RTA_IFA))
1070 		return (B_FALSE);
1071 
1072 	cp = (const char *)(ifam + 1);
1073 	lim = (const char *)ifam + msglen;
1074 	for (flag = 1; flag < RTA_IFA; flag <<= 1) {
1075 		if (ifam->ifam_addrs & flag) {
1076 			/* LINTED: alignment */
1077 			sa = (const struct sockaddr *)cp;
1078 			if ((const char *)(sa + 1) > lim)
1079 				return (B_FALSE);
1080 			switch (sa->sa_family) {
1081 			case AF_INET:
1082 				cp += sizeof (struct sockaddr_in);
1083 				break;
1084 			case AF_LINK:
1085 				cp += sizeof (struct sockaddr_dl);
1086 				break;
1087 			case AF_INET6:
1088 				cp += sizeof (struct sockaddr_in6);
1089 				break;
1090 			default:
1091 				cp += sizeof (struct sockaddr);
1092 				break;
1093 			}
1094 		}
1095 	}
1096 	if (isv6) {
1097 		const struct sockaddr_in6 *sin6;
1098 
1099 		/* LINTED: alignment */
1100 		sin6 = (const struct sockaddr_in6 *)cp;
1101 		if ((const char *)(sin6 + 1) > lim)
1102 			return (B_FALSE);
1103 		if (sin6->sin6_family != AF_INET6)
1104 			return (B_FALSE);
1105 		return (IN6_ARE_ADDR_EQUAL(&sin6->sin6_addr, addr));
1106 	} else {
1107 		const struct sockaddr_in *sinp;
1108 		ipaddr_t v4addr;
1109 
1110 		/* LINTED: alignment */
1111 		sinp = (const struct sockaddr_in *)cp;
1112 		if ((const char *)(sinp + 1) > lim)
1113 			return (B_FALSE);
1114 		if (sinp->sin_family != AF_INET)
1115 			return (B_FALSE);
1116 		IN6_V4MAPPED_TO_IPADDR(addr, v4addr);
1117 		return (sinp->sin_addr.s_addr == v4addr);
1118 	}
1119 }
1120 
1121 /*
1122  * is_rtm_v6(): determine if routing socket message is IPv6
1123  *
1124  *   input: struct ifa_msghdr *: pointer to routing socket message
1125  *	    int: message length
1126  *  output: boolean_t
1127  */
1128 
1129 static boolean_t
1130 is_rtm_v6(const struct ifa_msghdr *ifam, int msglen)
1131 {
1132 	const char *cp, *lim;
1133 	uint_t flag;
1134 	const struct sockaddr *sa;
1135 
1136 	cp = (const char *)(ifam + 1);
1137 	lim = (const char *)ifam + msglen;
1138 	for (flag = ifam->ifam_addrs; flag != 0; flag &= flag - 1) {
1139 		/* LINTED: alignment */
1140 		sa = (const struct sockaddr *)cp;
1141 		if ((const char *)(sa + 1) > lim)
1142 			return (B_FALSE);
1143 		switch (sa->sa_family) {
1144 		case AF_INET:
1145 			return (B_FALSE);
1146 		case AF_LINK:
1147 			cp += sizeof (struct sockaddr_dl);
1148 			break;
1149 		case AF_INET6:
1150 			return (B_TRUE);
1151 		default:
1152 			cp += sizeof (struct sockaddr);
1153 			break;
1154 		}
1155 	}
1156 	return (B_FALSE);
1157 }
1158 
1159 /*
1160  * check_lif(): check the state of a given logical interface and its DHCP
1161  *		lease.  We've been told by the routing socket that the
1162  *		corresponding ifIndex has changed.  This may mean that DAD has
1163  *		completed or failed.
1164  *
1165  *   input: dhcp_lif_t *: pointer to the LIF
1166  *	    const struct ifa_msghdr *: routing socket message
1167  *	    int: size of routing socket message
1168  *  output: boolean_t: B_TRUE if DAD has completed on this interface
1169  */
1170 
1171 static boolean_t
1172 check_lif(dhcp_lif_t *lif, const struct ifa_msghdr *ifam, int msglen)
1173 {
1174 	boolean_t isv6, dad_wait, unplumb;
1175 	int fd;
1176 	struct lifreq lifr;
1177 
1178 	isv6 = lif->lif_pif->pif_isv6;
1179 	fd = isv6 ? v6_sock_fd : v4_sock_fd;
1180 
1181 	/*
1182 	 * Get the real (64 bit) logical interface flags.  Note that the
1183 	 * routing socket message has flags, but these are just the lower 32
1184 	 * bits.
1185 	 */
1186 	unplumb = B_FALSE;
1187 	(void) memset(&lifr, 0, sizeof (lifr));
1188 	(void) strlcpy(lifr.lifr_name, lif->lif_name, sizeof (lifr.lifr_name));
1189 	if (ioctl(fd, SIOCGLIFFLAGS, &lifr) == -1) {
1190 		/*
1191 		 * Failing to retrieve flags means that the interface is gone.
1192 		 * It hasn't failed to verify with DAD, but we still have to
1193 		 * give up on it.
1194 		 */
1195 		lifr.lifr_flags = 0;
1196 		if (errno == ENXIO) {
1197 			lif->lif_plumbed = B_FALSE;
1198 			dhcpmsg(MSG_INFO, "%s has been removed; abandoning",
1199 			    lif->lif_name);
1200 			if (!isv6)
1201 				discard_default_routes(lif->lif_smachs);
1202 		} else {
1203 			dhcpmsg(MSG_ERR,
1204 			    "unable to retrieve interface flags on %s",
1205 			    lif->lif_name);
1206 		}
1207 		unplumb = B_TRUE;
1208 	} else if (!check_rtm_addr(ifam, msglen, isv6, &lif->lif_v6addr)) {
1209 		/*
1210 		 * If the message is not about this logical interface,
1211 		 * then just ignore it.
1212 		 */
1213 		return (B_FALSE);
1214 	} else if (lifr.lifr_flags & IFF_DUPLICATE) {
1215 		dhcpmsg(MSG_ERROR, "interface %s has duplicate address",
1216 		    lif->lif_name);
1217 		lif_mark_decline(lif, "duplicate address");
1218 		close_ip_lif(lif);
1219 		(void) open_ip_lif(lif, INADDR_ANY, B_TRUE);
1220 	}
1221 
1222 	dad_wait = lif->lif_dad_wait;
1223 	if (dad_wait) {
1224 		dhcpmsg(MSG_VERBOSE, "check_lif: %s has finished DAD",
1225 		    lif->lif_name);
1226 		lif->lif_dad_wait = B_FALSE;
1227 	}
1228 
1229 	if (unplumb)
1230 		unplumb_lif(lif);
1231 
1232 	return (dad_wait);
1233 }
1234 
1235 /*
1236  * check_main_lif(): check the state of a main logical interface for a state
1237  *		     machine.  This is used only for DHCPv6.
1238  *
1239  *   input: dhcp_smach_t *: pointer to the state machine
1240  *	    const struct ifa_msghdr *: routing socket message
1241  *	    int: size of routing socket message
1242  *  output: boolean_t: B_TRUE if LIF is ok.
1243  */
1244 
1245 static boolean_t
1246 check_main_lif(dhcp_smach_t *dsmp, const struct ifa_msghdr *ifam, int msglen)
1247 {
1248 	dhcp_lif_t *lif = dsmp->dsm_lif;
1249 	struct lifreq lifr;
1250 
1251 	/*
1252 	 * Get the real (64 bit) logical interface flags.  Note that the
1253 	 * routing socket message has flags, but these are just the lower 32
1254 	 * bits.
1255 	 */
1256 	(void) memset(&lifr, 0, sizeof (lifr));
1257 	(void) strlcpy(lifr.lifr_name, lif->lif_name, sizeof (lifr.lifr_name));
1258 	if (ioctl(v6_sock_fd, SIOCGLIFFLAGS, &lifr) == -1) {
1259 		/*
1260 		 * Failing to retrieve flags means that the interface is gone.
1261 		 * Our state machine is now trash.
1262 		 */
1263 		if (errno == ENXIO) {
1264 			dhcpmsg(MSG_INFO, "%s has been removed; abandoning",
1265 			    lif->lif_name);
1266 		} else {
1267 			dhcpmsg(MSG_ERR,
1268 			    "unable to retrieve interface flags on %s",
1269 			    lif->lif_name);
1270 		}
1271 		return (B_FALSE);
1272 	} else if (!check_rtm_addr(ifam, msglen, B_TRUE, &lif->lif_v6addr)) {
1273 		/*
1274 		 * If the message is not about this logical interface,
1275 		 * then just ignore it.
1276 		 */
1277 		return (B_TRUE);
1278 	} else if (lifr.lifr_flags & IFF_DUPLICATE) {
1279 		dhcpmsg(MSG_ERROR, "interface %s has duplicate address",
1280 		    lif->lif_name);
1281 		return (B_FALSE);
1282 	} else {
1283 		return (B_TRUE);
1284 	}
1285 }
1286 
1287 /*
1288  * process_link_up_down(): check the state of a physical interface for up/down
1289  *			   transitions; must go through INIT_REBOOT state if
1290  *			   the link flaps.
1291  *
1292  *   input: dhcp_pif_t *: pointer to the physical interface to check
1293  *	    const struct if_msghdr *: routing socket message
1294  *  output: none
1295  */
1296 
1297 static void
1298 process_link_up_down(dhcp_pif_t *pif, const struct if_msghdr *ifm)
1299 {
1300 	struct lifreq lifr;
1301 	boolean_t isv6;
1302 	int fd;
1303 
1304 	/*
1305 	 * If the message implies no change of flags, then we're done; no need
1306 	 * to check further.  Note that if we have multiple state machines on a
1307 	 * single physical interface, this test keeps us from issuing an ioctl
1308 	 * for each one.
1309 	 */
1310 	if ((ifm->ifm_flags & IFF_RUNNING) && pif->pif_running ||
1311 	    !(ifm->ifm_flags & IFF_RUNNING) && !pif->pif_running)
1312 		return;
1313 
1314 	/*
1315 	 * We don't know what the real interface flags are, because the
1316 	 * if_index number is only 16 bits; we must go ask.
1317 	 */
1318 	isv6 = pif->pif_isv6;
1319 	fd = isv6 ? v6_sock_fd : v4_sock_fd;
1320 	(void) memset(&lifr, 0, sizeof (lifr));
1321 	(void) strlcpy(lifr.lifr_name, pif->pif_name, sizeof (lifr.lifr_name));
1322 
1323 	if (ioctl(fd, SIOCGLIFFLAGS, &lifr) == -1 ||
1324 	    !(lifr.lifr_flags & IFF_RUNNING)) {
1325 		/*
1326 		 * If we've lost the interface or it has gone down, then
1327 		 * nothing special to do; just turn off the running flag.
1328 		 */
1329 		pif_status(pif, B_FALSE);
1330 	} else {
1331 		/*
1332 		 * Interface has come back up: go through verification process.
1333 		 */
1334 		pif_status(pif, B_TRUE);
1335 	}
1336 }
1337 
1338 /*
1339  * rtsock_event(): fetches routing socket messages and updates internal
1340  *		   interface state based on those messages.
1341  *
1342  *   input: iu_eh_t *: unused
1343  *	    int: the routing socket file descriptor
1344  *	    (other arguments unused)
1345  *  output: void
1346  */
1347 
1348 /* ARGSUSED */
1349 static void
1350 rtsock_event(iu_eh_t *ehp, int fd, short events, iu_event_id_t id, void *arg)
1351 {
1352 	dhcp_smach_t *dsmp, *dsmnext;
1353 	union {
1354 		struct ifa_msghdr ifam;
1355 		struct if_msghdr ifm;
1356 		char buf[1024];
1357 	} msg;
1358 	uint16_t ifindex;
1359 	int msglen;
1360 	boolean_t isv6;
1361 
1362 	if ((msglen = read(fd, &msg, sizeof (msg))) <= 0)
1363 		return;
1364 
1365 	/* Note that the routing socket interface index is just 16 bits */
1366 	if (msg.ifm.ifm_type == RTM_IFINFO) {
1367 		ifindex = msg.ifm.ifm_index;
1368 		isv6 = (msg.ifm.ifm_flags & IFF_IPV6) ? B_TRUE : B_FALSE;
1369 	} else if (msg.ifam.ifam_type == RTM_DELADDR ||
1370 	    msg.ifam.ifam_type == RTM_NEWADDR) {
1371 		ifindex = msg.ifam.ifam_index;
1372 		isv6 = is_rtm_v6(&msg.ifam, msglen);
1373 	} else {
1374 		return;
1375 	}
1376 
1377 	for (dsmp = lookup_smach_by_uindex(ifindex, NULL, isv6);
1378 	    dsmp != NULL; dsmp = dsmnext) {
1379 		DHCPSTATE oldstate;
1380 		boolean_t lif_finished;
1381 		boolean_t lease_removed;
1382 		dhcp_lease_t *dlp, *dlnext;
1383 
1384 		/*
1385 		 * Note that script_start can call dhcp_drop directly, and
1386 		 * that will do release_smach.
1387 		 */
1388 		dsmnext = lookup_smach_by_uindex(ifindex, dsmp, isv6);
1389 		oldstate = dsmp->dsm_state;
1390 
1391 		/*
1392 		 * Ignore state machines that are currently processing drop or
1393 		 * release; there is nothing more we can do for them.
1394 		 */
1395 		if (dsmp->dsm_droprelease)
1396 			continue;
1397 
1398 		/*
1399 		 * Look for link up/down notifications.  These occur on a
1400 		 * physical interface basis.
1401 		 */
1402 		if (msg.ifm.ifm_type == RTM_IFINFO) {
1403 			process_link_up_down(dsmp->dsm_lif->lif_pif, &msg.ifm);
1404 			continue;
1405 		}
1406 
1407 		/*
1408 		 * Since we cannot trust the flags reported by the routing
1409 		 * socket (they're just 32 bits -- and thus never include
1410 		 * IFF_DUPLICATE), and we can't trust the ifindex (it's only 16
1411 		 * bits and also doesn't reflect the alias in use), we get
1412 		 * flags on all matching interfaces, and go by that.
1413 		 */
1414 		lif_finished = B_FALSE;
1415 		lease_removed = B_FALSE;
1416 		for (dlp = dsmp->dsm_leases; dlp != NULL; dlp = dlnext) {
1417 			dhcp_lif_t *lif, *lifnext;
1418 			uint_t nlifs = dlp->dl_nlifs;
1419 
1420 			dlnext = dlp->dl_next;
1421 			for (lif = dlp->dl_lifs; lif != NULL && nlifs > 0;
1422 			    lif = lifnext, nlifs--) {
1423 				lifnext = lif->lif_next;
1424 				if (check_lif(lif, &msg.ifam, msglen)) {
1425 					dsmp->dsm_lif_wait--;
1426 					lif_finished = B_TRUE;
1427 				}
1428 			}
1429 			if (dlp->dl_nlifs == 0) {
1430 				remove_lease(dlp);
1431 				lease_removed = B_TRUE;
1432 			}
1433 		}
1434 
1435 		if ((isv6 && !check_main_lif(dsmp, &msg.ifam, msglen)) ||
1436 		    (!isv6 && !verify_lif(dsmp->dsm_lif))) {
1437 			dsmp->dsm_droprelease = B_TRUE;
1438 			(void) script_start(dsmp, isv6 ? EVENT_DROP6 :
1439 			    EVENT_DROP, dhcp_drop, NULL, NULL);
1440 			continue;
1441 		}
1442 
1443 		/*
1444 		 * Ignore this state machine if nothing interesting has
1445 		 * happened.
1446 		 */
1447 		if (!lif_finished && dsmp->dsm_lif_down == 0 &&
1448 		    (dsmp->dsm_leases != NULL || !lease_removed))
1449 			continue;
1450 
1451 		/*
1452 		 * If we're still waiting for DAD to complete on some of the
1453 		 * configured LIFs, then don't send a response.
1454 		 */
1455 		if (dsmp->dsm_lif_wait != 0) {
1456 			dhcpmsg(MSG_VERBOSE, "rtsock_event: %s still has %d "
1457 			    "LIFs waiting on DAD", dsmp->dsm_name,
1458 			    dsmp->dsm_lif_wait);
1459 			continue;
1460 		}
1461 
1462 		/*
1463 		 * If we have some failed LIFs, then handle them now.  We'll
1464 		 * remove them from the list.  Any leases that become empty are
1465 		 * also removed as part of the decline-generation process.
1466 		 */
1467 		if (dsmp->dsm_lif_down != 0)
1468 			send_declines(dsmp);
1469 
1470 		if (dsmp->dsm_leases == NULL) {
1471 			dsmp->dsm_bad_offers++;
1472 			/*
1473 			 * For DHCPv6, we'll process the restart once we're
1474 			 * done sending Decline messages, because these are
1475 			 * supposed to be acknowledged.  With DHCPv4, there's
1476 			 * no acknowledgment for a DECLINE, so after sending
1477 			 * it, we just restart right away.
1478 			 */
1479 			if (!dsmp->dsm_isv6) {
1480 				dhcpmsg(MSG_VERBOSE, "rtsock_event: %s has no "
1481 				    "LIFs left", dsmp->dsm_name);
1482 				dhcp_restart(dsmp);
1483 			}
1484 		} else {
1485 			/*
1486 			 * If we're now up on at least some of the leases and
1487 			 * we were waiting for that, then kick off the rest of
1488 			 * configuration.  Lease validation and DAD are done.
1489 			 */
1490 			dhcpmsg(MSG_VERBOSE, "rtsock_event: all LIFs verified "
1491 			    "on %s in %s state", dsmp->dsm_name,
1492 			    dhcp_state_to_string(oldstate));
1493 			if (oldstate == PRE_BOUND ||
1494 			    oldstate == ADOPTING)
1495 				dhcp_bound_complete(dsmp);
1496 			if (oldstate == ADOPTING)
1497 				dhcp_adopt_complete(dsmp);
1498 		}
1499 	}
1500 }
1501 
1502 /*
1503  * check_cmd_allowed(): check whether the requested command is allowed in the
1504  *			state specified.
1505  *
1506  *   input: DHCPSTATE: current state
1507  *	    dhcp_ipc_type_t: requested command
1508  *  output: boolean_t: B_TRUE if command is allowed in this state
1509  */
1510 
1511 boolean_t
1512 check_cmd_allowed(DHCPSTATE state, dhcp_ipc_type_t cmd)
1513 {
1514 	return (ipc_cmd_allowed[state][cmd] != 0);
1515 }
1516 
1517 static boolean_t
1518 is_iscsi_active(void)
1519 {
1520 	int fd;
1521 	int active = 0;
1522 
1523 	if ((fd = open(ISCSI_DRIVER_DEVCTL, O_RDONLY)) != -1) {
1524 		if (ioctl(fd, ISCSI_IS_ACTIVE, &active) != 0)
1525 			active = 0;
1526 		(void) close(fd);
1527 	}
1528 
1529 	return (active != 0);
1530 }
1531