xref: /titanic_51/usr/src/uts/common/inet/ip/ip6_ire.c (revision 80feccabd0bc0fca1df2b60bfa50449d9a6e3e09)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /*
26  * Copyright (c) 1990 Mentat Inc.
27  */
28 
29 #pragma ident	"%Z%%M%	%I%	%E% SMI"
30 
31 /*
32  * This file contains routines that manipulate Internet Routing Entries (IREs).
33  */
34 #include <sys/types.h>
35 #include <sys/stream.h>
36 #include <sys/stropts.h>
37 #include <sys/ddi.h>
38 #include <sys/cmn_err.h>
39 
40 #include <sys/systm.h>
41 #include <sys/param.h>
42 #include <sys/socket.h>
43 #include <net/if.h>
44 #include <net/route.h>
45 #include <netinet/in.h>
46 #include <net/if_dl.h>
47 #include <netinet/ip6.h>
48 #include <netinet/icmp6.h>
49 
50 #include <inet/common.h>
51 #include <inet/mi.h>
52 #include <inet/ip.h>
53 #include <inet/ip6.h>
54 #include <inet/ip_ndp.h>
55 #include <inet/ip_if.h>
56 #include <inet/ip_ire.h>
57 #include <inet/ipclassifier.h>
58 #include <inet/nd.h>
59 #include <sys/kmem.h>
60 #include <sys/zone.h>
61 
62 #include <sys/tsol/label.h>
63 #include <sys/tsol/tnet.h>
64 
65 static	ire_t	ire_null;
66 
67 static ire_t	*ire_ihandle_lookup_onlink_v6(ire_t *cire);
68 static boolean_t ire_match_args_v6(ire_t *ire, const in6_addr_t *addr,
69     const in6_addr_t *mask, const in6_addr_t *gateway, int type,
70     const ipif_t *ipif, zoneid_t zoneid, uint32_t ihandle,
71     const ts_label_t *tsl, int match_flags);
72 static	ire_t	*ire_init_v6(ire_t *, const in6_addr_t *, const in6_addr_t *,
73     const in6_addr_t *, const in6_addr_t *, uint_t *, queue_t *, queue_t *,
74     ushort_t, ipif_t *, const in6_addr_t *, uint32_t, uint32_t, uint_t,
75     const iulp_t *, tsol_gc_t *, tsol_gcgrp_t *, ip_stack_t *);
76 
77 
78 /*
79  * Initialize the ire that is specific to IPv6 part and call
80  * ire_init_common to finish it.
81  */
82 static ire_t *
83 ire_init_v6(ire_t *ire, const in6_addr_t *v6addr, const in6_addr_t *v6mask,
84     const in6_addr_t *v6src_addr, const in6_addr_t *v6gateway,
85     uint_t *max_fragp, queue_t *rfq, queue_t *stq, ushort_t type,
86     ipif_t *ipif, const in6_addr_t *v6cmask, uint32_t phandle,
87     uint32_t ihandle, uint_t flags, const iulp_t *ulp_info, tsol_gc_t *gc,
88     tsol_gcgrp_t *gcgrp, ip_stack_t *ipst)
89 {
90 
91 	/*
92 	 * Reject IRE security attribute creation/initialization
93 	 * if system is not running in Trusted mode.
94 	 */
95 	if ((gc != NULL || gcgrp != NULL) && !is_system_labeled())
96 		return (NULL);
97 
98 
99 	BUMP_IRE_STATS(ipst->ips_ire_stats_v6, ire_stats_alloced);
100 	ire->ire_addr_v6 = *v6addr;
101 
102 	if (v6src_addr != NULL)
103 		ire->ire_src_addr_v6 = *v6src_addr;
104 	if (v6mask != NULL) {
105 		ire->ire_mask_v6 = *v6mask;
106 		ire->ire_masklen = ip_mask_to_plen_v6(&ire->ire_mask_v6);
107 	}
108 	if (v6gateway != NULL)
109 		ire->ire_gateway_addr_v6 = *v6gateway;
110 
111 	if (type == IRE_CACHE && v6cmask != NULL)
112 		ire->ire_cmask_v6 = *v6cmask;
113 
114 	/*
115 	 * Multirouted packets need to have a fragment header added so that
116 	 * the receiver is able to discard duplicates according to their
117 	 * fragment identifier.
118 	 */
119 	if (type == IRE_CACHE && (flags & RTF_MULTIRT)) {
120 		ire->ire_frag_flag = IPH_FRAG_HDR;
121 	}
122 
123 	/* ire_init_common will free the mblks upon encountering any failure */
124 	if (!ire_init_common(ire, max_fragp, NULL, rfq, stq, type, ipif,
125 	    phandle, ihandle, flags, IPV6_VERSION, ulp_info, gc, gcgrp, ipst))
126 		return (NULL);
127 
128 	return (ire);
129 }
130 
131 /*
132  * Similar to ire_create_v6 except that it is called only when
133  * we want to allocate ire as an mblk e.g. we have a external
134  * resolver. Do we need this in IPv6 ?
135  *
136  * IPv6 initializes the ire_nce in ire_add_v6, which expects to
137  * find the ire_nce to be null when it is called. So, although
138  * we have a src_nce parameter (in the interest of matching up with
139  * the argument list of the v4 version), we ignore the src_nce
140  * argument here.
141  */
142 /* ARGSUSED */
143 ire_t *
144 ire_create_mp_v6(const in6_addr_t *v6addr, const in6_addr_t *v6mask,
145     const in6_addr_t *v6src_addr, const in6_addr_t *v6gateway,
146     nce_t *src_nce, queue_t *rfq, queue_t *stq, ushort_t type,
147     ipif_t *ipif, const in6_addr_t *v6cmask,
148     uint32_t phandle, uint32_t ihandle, uint_t flags, const iulp_t *ulp_info,
149     tsol_gc_t *gc, tsol_gcgrp_t *gcgrp, ip_stack_t *ipst)
150 {
151 	ire_t	*ire;
152 	ire_t	*ret_ire;
153 	mblk_t	*mp;
154 
155 	ASSERT(!IN6_IS_ADDR_V4MAPPED(v6addr));
156 
157 	/* Allocate the new IRE. */
158 	mp = allocb(sizeof (ire_t), BPRI_MED);
159 	if (mp == NULL) {
160 		ip1dbg(("ire_create_mp_v6: alloc failed\n"));
161 		return (NULL);
162 	}
163 
164 	ire = (ire_t *)mp->b_rptr;
165 	mp->b_wptr = (uchar_t *)&ire[1];
166 
167 	/* Start clean. */
168 	*ire = ire_null;
169 	ire->ire_mp = mp;
170 	mp->b_datap->db_type = IRE_DB_TYPE;
171 
172 	ret_ire = ire_init_v6(ire, v6addr, v6mask, v6src_addr, v6gateway,
173 	    NULL, rfq, stq, type, ipif, v6cmask, phandle,
174 	    ihandle, flags, ulp_info, gc, gcgrp, ipst);
175 
176 	if (ret_ire == NULL) {
177 		freeb(ire->ire_mp);
178 		return (NULL);
179 	}
180 	return (ire);
181 }
182 
183 /*
184  * ire_create_v6 is called to allocate and initialize a new IRE.
185  *
186  * NOTE : This is called as writer sometimes though not required
187  * by this function.
188  *
189  * See comments above ire_create_mp_v6() for the rationale behind the
190  * unused src_nce argument.
191  */
192 /* ARGSUSED */
193 ire_t *
194 ire_create_v6(const in6_addr_t *v6addr, const in6_addr_t *v6mask,
195     const in6_addr_t *v6src_addr, const in6_addr_t *v6gateway,
196     uint_t *max_fragp, nce_t *src_nce, queue_t *rfq, queue_t *stq,
197     ushort_t type, ipif_t *ipif, const in6_addr_t *v6cmask,
198     uint32_t phandle, uint32_t ihandle, uint_t flags, const iulp_t *ulp_info,
199     tsol_gc_t *gc, tsol_gcgrp_t *gcgrp, ip_stack_t *ipst)
200 {
201 	ire_t	*ire;
202 	ire_t	*ret_ire;
203 
204 	ASSERT(!IN6_IS_ADDR_V4MAPPED(v6addr));
205 
206 	ire = kmem_cache_alloc(ire_cache, KM_NOSLEEP);
207 	if (ire == NULL) {
208 		ip1dbg(("ire_create_v6: alloc failed\n"));
209 		return (NULL);
210 	}
211 	*ire = ire_null;
212 
213 	ret_ire = ire_init_v6(ire, v6addr, v6mask, v6src_addr, v6gateway,
214 	    max_fragp, rfq, stq, type, ipif, v6cmask, phandle,
215 	    ihandle, flags, ulp_info, gc, gcgrp, ipst);
216 
217 	if (ret_ire == NULL) {
218 		kmem_cache_free(ire_cache, ire);
219 		return (NULL);
220 	}
221 	ASSERT(ret_ire == ire);
222 	return (ire);
223 }
224 
225 /*
226  * Find an IRE_INTERFACE for the multicast group.
227  * Allows different routes for multicast addresses
228  * in the unicast routing table (akin to FF::0/8 but could be more specific)
229  * which point at different interfaces. This is used when IPV6_MULTICAST_IF
230  * isn't specified (when sending) and when IPV6_JOIN_GROUP doesn't
231  * specify the interface to join on.
232  *
233  * Supports link-local addresses by following the ipif/ill when recursing.
234  */
235 ire_t *
236 ire_lookup_multi_v6(const in6_addr_t *group, zoneid_t zoneid, ip_stack_t *ipst)
237 {
238 	ire_t	*ire;
239 	ipif_t	*ipif = NULL;
240 	int	match_flags = MATCH_IRE_TYPE;
241 	in6_addr_t gw_addr_v6;
242 
243 	ire = ire_ftable_lookup_v6(group, 0, 0, 0, NULL, NULL,
244 	    zoneid, 0, NULL, MATCH_IRE_DEFAULT, ipst);
245 
246 	/* We search a resolvable ire in case of multirouting. */
247 	if ((ire != NULL) && (ire->ire_flags & RTF_MULTIRT)) {
248 		ire_t *cire = NULL;
249 		/*
250 		 * If the route is not resolvable, the looked up ire
251 		 * may be changed here. In that case, ire_multirt_lookup()
252 		 * IRE_REFRELE the original ire and change it.
253 		 */
254 		(void) ire_multirt_lookup_v6(&cire, &ire, MULTIRT_CACHEGW,
255 		    NULL, ipst);
256 		if (cire != NULL)
257 			ire_refrele(cire);
258 	}
259 	if (ire == NULL)
260 		return (NULL);
261 	/*
262 	 * Make sure we follow ire_ipif.
263 	 *
264 	 * We need to determine the interface route through
265 	 * which the gateway will be reached. We don't really
266 	 * care which interface is picked if the interface is
267 	 * part of a group.
268 	 */
269 	if (ire->ire_ipif != NULL) {
270 		ipif = ire->ire_ipif;
271 		match_flags |= MATCH_IRE_ILL_GROUP;
272 	}
273 
274 	switch (ire->ire_type) {
275 	case IRE_DEFAULT:
276 	case IRE_PREFIX:
277 	case IRE_HOST:
278 		mutex_enter(&ire->ire_lock);
279 		gw_addr_v6 = ire->ire_gateway_addr_v6;
280 		mutex_exit(&ire->ire_lock);
281 		ire_refrele(ire);
282 		ire = ire_ftable_lookup_v6(&gw_addr_v6, 0, 0,
283 		    IRE_INTERFACE, ipif, NULL, zoneid, 0,
284 		    NULL, match_flags, ipst);
285 		return (ire);
286 	case IRE_IF_NORESOLVER:
287 	case IRE_IF_RESOLVER:
288 		return (ire);
289 	default:
290 		ire_refrele(ire);
291 		return (NULL);
292 	}
293 }
294 
295 /*
296  * Return any local address.  We use this to target ourselves
297  * when the src address was specified as 'default'.
298  * Preference for IRE_LOCAL entries.
299  */
300 ire_t *
301 ire_lookup_local_v6(zoneid_t zoneid, ip_stack_t *ipst)
302 {
303 	ire_t	*ire;
304 	irb_t	*irb;
305 	ire_t	*maybe = NULL;
306 	int i;
307 
308 	for (i = 0; i < ipst->ips_ip6_cache_table_size;  i++) {
309 		irb = &ipst->ips_ip_cache_table_v6[i];
310 		if (irb->irb_ire == NULL)
311 			continue;
312 		rw_enter(&irb->irb_lock, RW_READER);
313 		for (ire = irb->irb_ire; ire; ire = ire->ire_next) {
314 			if ((ire->ire_marks & IRE_MARK_CONDEMNED) ||
315 			    ire->ire_zoneid != zoneid &&
316 			    ire->ire_zoneid != ALL_ZONES)
317 				continue;
318 			switch (ire->ire_type) {
319 			case IRE_LOOPBACK:
320 				if (maybe == NULL) {
321 					IRE_REFHOLD(ire);
322 					maybe = ire;
323 				}
324 				break;
325 			case IRE_LOCAL:
326 				if (maybe != NULL) {
327 					ire_refrele(maybe);
328 				}
329 				IRE_REFHOLD(ire);
330 				rw_exit(&irb->irb_lock);
331 				return (ire);
332 			}
333 		}
334 		rw_exit(&irb->irb_lock);
335 	}
336 	return (maybe);
337 }
338 
339 /*
340  * This function takes a mask and returns number of bits set in the
341  * mask (the represented prefix length).  Assumes a contiguous mask.
342  */
343 int
344 ip_mask_to_plen_v6(const in6_addr_t *v6mask)
345 {
346 	int		bits;
347 	int		plen = IPV6_ABITS;
348 	int		i;
349 
350 	for (i = 3; i >= 0; i--) {
351 		if (v6mask->s6_addr32[i] == 0) {
352 			plen -= 32;
353 			continue;
354 		}
355 		bits = ffs(ntohl(v6mask->s6_addr32[i])) - 1;
356 		if (bits == 0)
357 			break;
358 		plen -= bits;
359 	}
360 
361 	return (plen);
362 }
363 
364 /*
365  * Convert a prefix length to the mask for that prefix.
366  * Returns the argument bitmask.
367  */
368 in6_addr_t *
369 ip_plen_to_mask_v6(uint_t plen, in6_addr_t *bitmask)
370 {
371 	uint32_t *ptr;
372 
373 	if (plen < 0 || plen > IPV6_ABITS)
374 		return (NULL);
375 	*bitmask = ipv6_all_zeros;
376 
377 	ptr = (uint32_t *)bitmask;
378 	while (plen > 32) {
379 		*ptr++ = 0xffffffffU;
380 		plen -= 32;
381 	}
382 	*ptr = htonl(0xffffffffU << (32 - plen));
383 	return (bitmask);
384 }
385 
386 /*
387  * Add a fully initialized IRE to an appropriate
388  * table based on ire_type.
389  *
390  * The forward table contains IRE_PREFIX/IRE_HOST/IRE_HOST and
391  * IRE_IF_RESOLVER/IRE_IF_NORESOLVER and IRE_DEFAULT.
392  *
393  * The cache table contains IRE_BROADCAST/IRE_LOCAL/IRE_LOOPBACK
394  * and IRE_CACHE.
395  *
396  * NOTE : This function is called as writer though not required
397  * by this function.
398  */
399 int
400 ire_add_v6(ire_t **ire_p, queue_t *q, mblk_t *mp, ipsq_func_t func)
401 {
402 	ire_t	*ire1;
403 	int	mask_table_index;
404 	irb_t	*irb_ptr;
405 	ire_t	**irep;
406 	int	flags;
407 	ire_t	*pire = NULL;
408 	ill_t	*stq_ill;
409 	boolean_t	ndp_g_lock_held = B_FALSE;
410 	ire_t	*ire = *ire_p;
411 	int	error;
412 	ip_stack_t	*ipst = ire->ire_ipst;
413 
414 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
415 	ASSERT(ire->ire_mp == NULL); /* Calls should go through ire_add */
416 	ASSERT(ire->ire_nce == NULL);
417 
418 	/* Find the appropriate list head. */
419 	switch (ire->ire_type) {
420 	case IRE_HOST:
421 		ire->ire_mask_v6 = ipv6_all_ones;
422 		ire->ire_masklen = IPV6_ABITS;
423 		if ((ire->ire_flags & RTF_SETSRC) == 0)
424 			ire->ire_src_addr_v6 = ipv6_all_zeros;
425 		break;
426 	case IRE_CACHE:
427 	case IRE_LOCAL:
428 	case IRE_LOOPBACK:
429 		ire->ire_mask_v6 = ipv6_all_ones;
430 		ire->ire_masklen = IPV6_ABITS;
431 		break;
432 	case IRE_PREFIX:
433 		if ((ire->ire_flags & RTF_SETSRC) == 0)
434 			ire->ire_src_addr_v6 = ipv6_all_zeros;
435 		break;
436 	case IRE_DEFAULT:
437 		if ((ire->ire_flags & RTF_SETSRC) == 0)
438 			ire->ire_src_addr_v6 = ipv6_all_zeros;
439 		break;
440 	case IRE_IF_RESOLVER:
441 	case IRE_IF_NORESOLVER:
442 		break;
443 	default:
444 		printf("ire_add_v6: ire %p has unrecognized IRE type (%d)\n",
445 		    (void *)ire, ire->ire_type);
446 		ire_delete(ire);
447 		*ire_p = NULL;
448 		return (EINVAL);
449 	}
450 
451 	/* Make sure the address is properly masked. */
452 	V6_MASK_COPY(ire->ire_addr_v6, ire->ire_mask_v6, ire->ire_addr_v6);
453 
454 	if ((ire->ire_type & IRE_CACHETABLE) == 0) {
455 		/* IRE goes into Forward Table */
456 		mask_table_index = ip_mask_to_plen_v6(&ire->ire_mask_v6);
457 		if ((ipst->ips_ip_forwarding_table_v6[mask_table_index]) ==
458 		    NULL) {
459 			irb_t *ptr;
460 			int i;
461 
462 			ptr = (irb_t *)mi_zalloc((
463 			    ipst->ips_ip6_ftable_hash_size * sizeof (irb_t)));
464 			if (ptr == NULL) {
465 				ire_delete(ire);
466 				*ire_p = NULL;
467 				return (ENOMEM);
468 			}
469 			for (i = 0; i < ipst->ips_ip6_ftable_hash_size; i++) {
470 				rw_init(&ptr[i].irb_lock, NULL,
471 				    RW_DEFAULT, NULL);
472 			}
473 			mutex_enter(&ipst->ips_ire_ft_init_lock);
474 			if (ipst->ips_ip_forwarding_table_v6[
475 			    mask_table_index] == NULL) {
476 				ipst->ips_ip_forwarding_table_v6[
477 				    mask_table_index] = ptr;
478 				mutex_exit(&ipst->ips_ire_ft_init_lock);
479 			} else {
480 				/*
481 				 * Some other thread won the race in
482 				 * initializing the forwarding table at the
483 				 * same index.
484 				 */
485 				mutex_exit(&ipst->ips_ire_ft_init_lock);
486 				for (i = 0; i < ipst->ips_ip6_ftable_hash_size;
487 				    i++) {
488 					rw_destroy(&ptr[i].irb_lock);
489 				}
490 				mi_free(ptr);
491 			}
492 		}
493 		irb_ptr = &(ipst->ips_ip_forwarding_table_v6[mask_table_index][
494 		    IRE_ADDR_MASK_HASH_V6(ire->ire_addr_v6, ire->ire_mask_v6,
495 		    ipst->ips_ip6_ftable_hash_size)]);
496 	} else {
497 		irb_ptr = &(ipst->ips_ip_cache_table_v6[IRE_ADDR_HASH_V6(
498 		    ire->ire_addr_v6, ipst->ips_ip6_cache_table_size)]);
499 	}
500 	/*
501 	 * For xresolv interfaces (v6 interfaces with an external
502 	 * address resolver), ip_newroute_v6/ip_newroute_ipif_v6
503 	 * are unable to prevent the deletion of the interface route
504 	 * while adding an IRE_CACHE for an on-link destination
505 	 * in the IRE_IF_RESOLVER case, since the ire has to go to
506 	 * the external resolver and return. We can't do a REFHOLD on the
507 	 * associated interface ire for fear of the message being freed
508 	 * if the external resolver can't resolve the address.
509 	 * Here we look up the interface ire in the forwarding table
510 	 * and make sure that the interface route has not been deleted.
511 	 */
512 	if (ire->ire_type == IRE_CACHE &&
513 	    IN6_IS_ADDR_UNSPECIFIED(&ire->ire_gateway_addr_v6) &&
514 	    (((ill_t *)ire->ire_stq->q_ptr)->ill_net_type == IRE_IF_RESOLVER) &&
515 	    (((ill_t *)ire->ire_stq->q_ptr)->ill_flags & ILLF_XRESOLV)) {
516 
517 		pire = ire_ihandle_lookup_onlink_v6(ire);
518 		if (pire == NULL) {
519 			ire_delete(ire);
520 			*ire_p = NULL;
521 			return (EINVAL);
522 		}
523 		/* Prevent pire from getting deleted */
524 		IRB_REFHOLD(pire->ire_bucket);
525 		/* Has it been removed already? */
526 		if (pire->ire_marks & IRE_MARK_CONDEMNED) {
527 			IRB_REFRELE(pire->ire_bucket);
528 			ire_refrele(pire);
529 			ire_delete(ire);
530 			*ire_p = NULL;
531 			return (EINVAL);
532 		}
533 	}
534 
535 	flags = (MATCH_IRE_MASK | MATCH_IRE_TYPE | MATCH_IRE_GW);
536 	/*
537 	 * For IRE_CACHES, MATCH_IRE_IPIF is not enough to check
538 	 * for duplicates because :
539 	 *
540 	 * 1) ire_ipif->ipif_ill and ire_stq->q_ptr could be
541 	 *    pointing at different ills. A real duplicate is
542 	 *    a match on both ire_ipif and ire_stq.
543 	 *
544 	 * 2) We could have multiple packets trying to create
545 	 *    an IRE_CACHE for the same ill.
546 	 *
547 	 * Moreover, IPIF_NOFAILOVER and IPV6_BOUND_PIF endpoints wants
548 	 * to go out on a particular ill. Rather than looking at the
549 	 * packet, we depend on the above for MATCH_IRE_ILL here.
550 	 *
551 	 * Unlike IPv4, MATCH_IRE_IPIF is needed here as we could have
552 	 * multiple IRE_CACHES for an ill for the same destination
553 	 * with various scoped addresses i.e represented by ipifs.
554 	 *
555 	 * MATCH_IRE_ILL is done implicitly below for IRE_CACHES.
556 	 */
557 	if (ire->ire_ipif != NULL)
558 		flags |= MATCH_IRE_IPIF;
559 	/*
560 	 * If we are creating hidden ires, make sure we search on
561 	 * this ill (MATCH_IRE_ILL) and a hidden ire, while we are
562 	 * searching for duplicates below. Otherwise we could
563 	 * potentially find an IRE on some other interface
564 	 * and it may not be a IRE marked with IRE_MARK_HIDDEN. We
565 	 * shouldn't do this as this will lead to an infinite loop as
566 	 * eventually we need an hidden ire for this packet to go
567 	 * out. MATCH_IRE_ILL is already marked above.
568 	 */
569 	if (ire->ire_marks & IRE_MARK_HIDDEN) {
570 		ASSERT(ire->ire_type == IRE_CACHE);
571 		flags |= MATCH_IRE_MARK_HIDDEN;
572 	}
573 
574 	/*
575 	 * Start the atomic add of the ire. Grab the ill locks,
576 	 * ill_g_usesrc_lock and the bucket lock. Check for condemned.
577 	 * To avoid lock order problems, get the ndp6.ndp_g_lock now itself.
578 	 */
579 	if (ire->ire_type == IRE_CACHE) {
580 		mutex_enter(&ipst->ips_ndp6->ndp_g_lock);
581 		ndp_g_lock_held = B_TRUE;
582 	}
583 
584 	/*
585 	 * If ipif or ill is changing ire_atomic_start() may queue the
586 	 * request and return EINPROGRESS.
587 	 */
588 
589 	error = ire_atomic_start(irb_ptr, ire, q, mp, func);
590 	if (error != 0) {
591 		if (ndp_g_lock_held)
592 			mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
593 		/*
594 		 * We don't know whether it is a valid ipif or not.
595 		 * So, set it to NULL. This assumes that the ire has not added
596 		 * a reference to the ipif.
597 		 */
598 		ire->ire_ipif = NULL;
599 		ire_delete(ire);
600 		if (pire != NULL) {
601 			IRB_REFRELE(pire->ire_bucket);
602 			ire_refrele(pire);
603 		}
604 		*ire_p = NULL;
605 		return (error);
606 	}
607 	/*
608 	 * To avoid creating ires having stale values for the ire_max_frag
609 	 * we get the latest value atomically here. For more details
610 	 * see the block comment in ip_sioctl_mtu and in DL_NOTE_SDU_CHANGE
611 	 * in ip_rput_dlpi_writer
612 	 */
613 	if (ire->ire_max_fragp == NULL) {
614 		if (IN6_IS_ADDR_MULTICAST(&ire->ire_addr_v6))
615 			ire->ire_max_frag = ire->ire_ipif->ipif_mtu;
616 		else
617 			ire->ire_max_frag = pire->ire_max_frag;
618 	} else {
619 		uint_t  max_frag;
620 
621 		max_frag = *ire->ire_max_fragp;
622 		ire->ire_max_fragp = NULL;
623 		ire->ire_max_frag = max_frag;
624 	}
625 
626 	/*
627 	 * Atomically check for duplicate and insert in the table.
628 	 */
629 	for (ire1 = irb_ptr->irb_ire; ire1 != NULL; ire1 = ire1->ire_next) {
630 		if (ire1->ire_marks & IRE_MARK_CONDEMNED)
631 			continue;
632 
633 		if (ire->ire_type == IRE_CACHE) {
634 			/*
635 			 * We do MATCH_IRE_ILL implicitly here for IRE_CACHES.
636 			 * As ire_ipif and ire_stq could point to two
637 			 * different ills, we can't pass just ire_ipif to
638 			 * ire_match_args and get a match on both ills.
639 			 * This is just needed for duplicate checks here and
640 			 * so we don't add an extra argument to
641 			 * ire_match_args for this. Do it locally.
642 			 *
643 			 * NOTE : Currently there is no part of the code
644 			 * that asks for both MATH_IRE_IPIF and MATCH_IRE_ILL
645 			 * match for IRE_CACHEs. Thus we don't want to
646 			 * extend the arguments to ire_match_args_v6.
647 			 */
648 			if (ire1->ire_stq != ire->ire_stq)
649 				continue;
650 			/*
651 			 * Multiroute IRE_CACHEs for a given destination can
652 			 * have the same ire_ipif, typically if their source
653 			 * address is forced using RTF_SETSRC, and the same
654 			 * send-to queue. We differentiate them using the parent
655 			 * handle.
656 			 */
657 			if ((ire1->ire_flags & RTF_MULTIRT) &&
658 			    (ire->ire_flags & RTF_MULTIRT) &&
659 			    (ire1->ire_phandle != ire->ire_phandle))
660 				continue;
661 		}
662 		if (ire1->ire_zoneid != ire->ire_zoneid)
663 			continue;
664 		if (ire_match_args_v6(ire1, &ire->ire_addr_v6,
665 		    &ire->ire_mask_v6, &ire->ire_gateway_addr_v6,
666 		    ire->ire_type, ire->ire_ipif, ire->ire_zoneid, 0, NULL,
667 		    flags)) {
668 			/*
669 			 * Return the old ire after doing a REFHOLD.
670 			 * As most of the callers continue to use the IRE
671 			 * after adding, we return a held ire. This will
672 			 * avoid a lookup in the caller again. If the callers
673 			 * don't want to use it, they need to do a REFRELE.
674 			 */
675 			ip1dbg(("found dup ire existing %p new %p",
676 			    (void *)ire1, (void *)ire));
677 			IRE_REFHOLD(ire1);
678 			if (ndp_g_lock_held)
679 				mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
680 			ire_atomic_end(irb_ptr, ire);
681 			ire_delete(ire);
682 			if (pire != NULL) {
683 				/*
684 				 * Assert that it is
685 				 * not yet removed from the list.
686 				 */
687 				ASSERT(pire->ire_ptpn != NULL);
688 				IRB_REFRELE(pire->ire_bucket);
689 				ire_refrele(pire);
690 			}
691 			*ire_p = ire1;
692 			return (0);
693 		}
694 	}
695 	if (ire->ire_type == IRE_CACHE) {
696 		in6_addr_t gw_addr_v6;
697 		ill_t	*ill = ire_to_ill(ire);
698 		char	buf[INET6_ADDRSTRLEN];
699 		nce_t	*nce;
700 
701 		/*
702 		 * All IRE_CACHE types must have a nce.  If this is
703 		 * not the case the entry will not be added. We need
704 		 * to make sure that if somebody deletes the nce
705 		 * after we looked up, they will find this ire and
706 		 * delete the ire. To delete this ire one needs the
707 		 * bucket lock which we are still holding here. So,
708 		 * even if the nce gets deleted after we looked up,
709 		 * this ire  will get deleted.
710 		 *
711 		 * NOTE : Don't need the ire_lock for accessing
712 		 * ire_gateway_addr_v6 as it is appearing first
713 		 * time on the list and rts_setgwr_v6 could not
714 		 * be changing this.
715 		 */
716 		gw_addr_v6 = ire->ire_gateway_addr_v6;
717 		if (IN6_IS_ADDR_UNSPECIFIED(&gw_addr_v6)) {
718 			nce = ndp_lookup_v6(ill, &ire->ire_addr_v6, B_TRUE);
719 		} else {
720 			nce = ndp_lookup_v6(ill, &gw_addr_v6, B_TRUE);
721 		}
722 		if (nce == NULL)
723 			goto failed;
724 
725 		/* Pair of refhold, refrele just to get the tracing right */
726 		NCE_REFHOLD_TO_REFHOLD_NOTR(nce);
727 		/*
728 		 * Atomically make sure that new IREs don't point
729 		 * to an NCE that is logically deleted (CONDEMNED).
730 		 * ndp_delete() first marks the NCE CONDEMNED.
731 		 * This ensures that the nce_refcnt won't increase
732 		 * due to new nce_lookups or due to addition of new IREs
733 		 * pointing to this NCE. Then ndp_delete() cleans up
734 		 * existing references. If we don't do it atomically here,
735 		 * ndp_delete() -> nce_ire_delete() will not be able to
736 		 * clean up the IRE list completely, and the nce_refcnt
737 		 * won't go down to zero.
738 		 */
739 		mutex_enter(&nce->nce_lock);
740 		if (ill->ill_flags & ILLF_XRESOLV) {
741 			/*
742 			 * If we used an external resolver, we may not
743 			 * have gone through neighbor discovery to get here.
744 			 * Must update the nce_state before the next check.
745 			 */
746 			if (nce->nce_state == ND_INCOMPLETE)
747 				nce->nce_state = ND_REACHABLE;
748 		}
749 		if (nce->nce_state == ND_INCOMPLETE ||
750 		    (nce->nce_flags & NCE_F_CONDEMNED) ||
751 		    (nce->nce_state == ND_UNREACHABLE)) {
752 failed:
753 			if (ndp_g_lock_held)
754 				mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
755 			if (nce != NULL)
756 				mutex_exit(&nce->nce_lock);
757 			ire_atomic_end(irb_ptr, ire);
758 			ip1dbg(("ire_add_v6: No nce for dst %s \n",
759 			    inet_ntop(AF_INET6, &ire->ire_addr_v6,
760 			    buf, sizeof (buf))));
761 			ire_delete(ire);
762 			if (pire != NULL) {
763 				/*
764 				 * Assert that it is
765 				 * not yet removed from the list.
766 				 */
767 				ASSERT(pire->ire_ptpn != NULL);
768 				IRB_REFRELE(pire->ire_bucket);
769 				ire_refrele(pire);
770 			}
771 			if (nce != NULL)
772 				NCE_REFRELE_NOTR(nce);
773 			*ire_p = NULL;
774 			return (EINVAL);
775 		} else {
776 			ire->ire_nce = nce;
777 		}
778 		mutex_exit(&nce->nce_lock);
779 	}
780 	/*
781 	 * Find the first entry that matches ire_addr - provides
782 	 * tail insertion. *irep will be null if no match.
783 	 */
784 	irep = (ire_t **)irb_ptr;
785 	while ((ire1 = *irep) != NULL &&
786 	    !IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, &ire1->ire_addr_v6))
787 		irep = &ire1->ire_next;
788 	ASSERT(!(ire->ire_type & IRE_BROADCAST));
789 
790 	if (*irep != NULL) {
791 		/*
792 		 * Find the last ire which matches ire_addr_v6.
793 		 * Needed to do tail insertion among entries with the same
794 		 * ire_addr_v6.
795 		 */
796 		while (IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6,
797 		    &ire1->ire_addr_v6)) {
798 			irep = &ire1->ire_next;
799 			ire1 = *irep;
800 			if (ire1 == NULL)
801 				break;
802 		}
803 	}
804 
805 	if (ire->ire_type == IRE_DEFAULT) {
806 		/*
807 		 * We keep a count of default gateways which is used when
808 		 * assigning them as routes.
809 		 */
810 		ipst->ips_ipv6_ire_default_count++;
811 		ASSERT(ipst->ips_ipv6_ire_default_count != 0); /* Wraparound */
812 	}
813 	/* Insert at *irep */
814 	ire1 = *irep;
815 	if (ire1 != NULL)
816 		ire1->ire_ptpn = &ire->ire_next;
817 	ire->ire_next = ire1;
818 	/* Link the new one in. */
819 	ire->ire_ptpn = irep;
820 	/*
821 	 * ire_walk routines de-reference ire_next without holding
822 	 * a lock. Before we point to the new ire, we want to make
823 	 * sure the store that sets the ire_next of the new ire
824 	 * reaches global visibility, so that ire_walk routines
825 	 * don't see a truncated list of ires i.e if the ire_next
826 	 * of the new ire gets set after we do "*irep = ire" due
827 	 * to re-ordering, the ire_walk thread will see a NULL
828 	 * once it accesses the ire_next of the new ire.
829 	 * membar_producer() makes sure that the following store
830 	 * happens *after* all of the above stores.
831 	 */
832 	membar_producer();
833 	*irep = ire;
834 	ire->ire_bucket = irb_ptr;
835 	/*
836 	 * We return a bumped up IRE above. Keep it symmetrical
837 	 * so that the callers will always have to release. This
838 	 * helps the callers of this function because they continue
839 	 * to use the IRE after adding and hence they don't have to
840 	 * lookup again after we return the IRE.
841 	 *
842 	 * NOTE : We don't have to use atomics as this is appearing
843 	 * in the list for the first time and no one else can bump
844 	 * up the reference count on this yet.
845 	 */
846 	IRE_REFHOLD_LOCKED(ire);
847 	BUMP_IRE_STATS(ipst->ips_ire_stats_v6, ire_stats_inserted);
848 	irb_ptr->irb_ire_cnt++;
849 	if (ire->ire_marks & IRE_MARK_TEMPORARY)
850 		irb_ptr->irb_tmp_ire_cnt++;
851 
852 	if (ire->ire_ipif != NULL) {
853 		ire->ire_ipif->ipif_ire_cnt++;
854 		if (ire->ire_stq != NULL) {
855 			stq_ill = (ill_t *)ire->ire_stq->q_ptr;
856 			stq_ill->ill_ire_cnt++;
857 		}
858 	} else {
859 		ASSERT(ire->ire_stq == NULL);
860 	}
861 
862 	if (ndp_g_lock_held)
863 		mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
864 	ire_atomic_end(irb_ptr, ire);
865 
866 	if (pire != NULL) {
867 		/* Assert that it is not removed from the list yet */
868 		ASSERT(pire->ire_ptpn != NULL);
869 		IRB_REFRELE(pire->ire_bucket);
870 		ire_refrele(pire);
871 	}
872 
873 	if (ire->ire_type != IRE_CACHE) {
874 		/*
875 		 * For ire's with with host mask see if there is an entry
876 		 * in the cache. If there is one flush the whole cache as
877 		 * there might be multiple entries due to RTF_MULTIRT (CGTP).
878 		 * If no entry is found than there is no need to flush the
879 		 * cache.
880 		 */
881 
882 		if (ip_mask_to_plen_v6(&ire->ire_mask_v6) == IPV6_ABITS) {
883 			ire_t *lire;
884 			lire = ire_ctable_lookup_v6(&ire->ire_addr_v6, NULL,
885 			    IRE_CACHE, NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE,
886 			    ipst);
887 			if (lire != NULL) {
888 				ire_refrele(lire);
889 				ire_flush_cache_v6(ire, IRE_FLUSH_ADD);
890 			}
891 		} else {
892 			ire_flush_cache_v6(ire, IRE_FLUSH_ADD);
893 		}
894 	}
895 
896 	*ire_p = ire;
897 	return (0);
898 }
899 
900 /*
901  * Search for all HOST REDIRECT routes that are
902  * pointing at the specified gateway and
903  * delete them. This routine is called only
904  * when a default gateway is going away.
905  */
906 static void
907 ire_delete_host_redirects_v6(const in6_addr_t *gateway, ip_stack_t *ipst)
908 {
909 	irb_t *irb_ptr;
910 	irb_t *irb;
911 	ire_t *ire;
912 	in6_addr_t gw_addr_v6;
913 	int i;
914 
915 	/* get the hash table for HOST routes */
916 	irb_ptr = ipst->ips_ip_forwarding_table_v6[(IP6_MASK_TABLE_SIZE - 1)];
917 	if (irb_ptr == NULL)
918 		return;
919 	for (i = 0; (i < ipst->ips_ip6_ftable_hash_size); i++) {
920 		irb = &irb_ptr[i];
921 		IRB_REFHOLD(irb);
922 		for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
923 			if (!(ire->ire_flags & RTF_DYNAMIC))
924 				continue;
925 			mutex_enter(&ire->ire_lock);
926 			gw_addr_v6 = ire->ire_gateway_addr_v6;
927 			mutex_exit(&ire->ire_lock);
928 			if (IN6_ARE_ADDR_EQUAL(&gw_addr_v6, gateway))
929 				ire_delete(ire);
930 		}
931 		IRB_REFRELE(irb);
932 	}
933 }
934 
935 /*
936  * Delete all the cache entries with this 'addr'. This is the IPv6 counterpart
937  * of ip_ire_clookup_and_delete. The difference being this function does not
938  * return any value. IPv6 processing of a gratuitous ARP, as it stands, is
939  * different than IPv4 in that, regardless of the presence of a cache entry
940  * for this address, an ire_walk_v6 is done. Another difference is that unlike
941  * in the case of IPv4 this does not take an ipif_t argument, since it is only
942  * called by ip_arp_news and the match is always only on the address.
943  */
944 void
945 ip_ire_clookup_and_delete_v6(const in6_addr_t *addr, ip_stack_t *ipst)
946 {
947 	irb_t		*irb;
948 	ire_t		*cire;
949 	boolean_t	found = B_FALSE;
950 
951 	irb = &ipst->ips_ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr,
952 	    ipst->ips_ip6_cache_table_size)];
953 	IRB_REFHOLD(irb);
954 	for (cire = irb->irb_ire; cire != NULL; cire = cire->ire_next) {
955 		if (cire->ire_marks & IRE_MARK_CONDEMNED)
956 			continue;
957 		if (IN6_ARE_ADDR_EQUAL(&cire->ire_addr_v6, addr)) {
958 
959 			/* This signifies start of a match */
960 			if (!found)
961 				found = B_TRUE;
962 			if (cire->ire_type == IRE_CACHE) {
963 				if (cire->ire_nce != NULL)
964 					ndp_delete(cire->ire_nce);
965 				ire_delete_v6(cire);
966 			}
967 		/* End of the match */
968 		} else if (found)
969 			break;
970 	}
971 	IRB_REFRELE(irb);
972 }
973 
974 /*
975  * Delete the specified IRE.
976  * All calls should use ire_delete().
977  * Sometimes called as writer though not required by this function.
978  *
979  * NOTE : This function is called only if the ire was added
980  * in the list.
981  */
982 void
983 ire_delete_v6(ire_t *ire)
984 {
985 	in6_addr_t gw_addr_v6;
986 	ip_stack_t	*ipst = ire->ire_ipst;
987 
988 	ASSERT(ire->ire_refcnt >= 1);
989 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
990 
991 	if (ire->ire_type != IRE_CACHE)
992 		ire_flush_cache_v6(ire, IRE_FLUSH_DELETE);
993 	if (ire->ire_type == IRE_DEFAULT) {
994 		/*
995 		 * when a default gateway is going away
996 		 * delete all the host redirects pointing at that
997 		 * gateway.
998 		 */
999 		mutex_enter(&ire->ire_lock);
1000 		gw_addr_v6 = ire->ire_gateway_addr_v6;
1001 		mutex_exit(&ire->ire_lock);
1002 		ire_delete_host_redirects_v6(&gw_addr_v6, ipst);
1003 	}
1004 }
1005 
1006 /*
1007  * ire_walk routine to delete all IRE_CACHE and IRE_HOST type redirect
1008  * entries.
1009  */
1010 /*ARGSUSED1*/
1011 void
1012 ire_delete_cache_v6(ire_t *ire, char *arg)
1013 {
1014 	char    addrstr1[INET6_ADDRSTRLEN];
1015 	char    addrstr2[INET6_ADDRSTRLEN];
1016 
1017 	if ((ire->ire_type & IRE_CACHE) ||
1018 	    (ire->ire_flags & RTF_DYNAMIC)) {
1019 		ip1dbg(("ire_delete_cache_v6: deleted %s type %d through %s\n",
1020 		    inet_ntop(AF_INET6, &ire->ire_addr_v6,
1021 		    addrstr1, sizeof (addrstr1)),
1022 		    ire->ire_type,
1023 		    inet_ntop(AF_INET6, &ire->ire_gateway_addr_v6,
1024 		    addrstr2, sizeof (addrstr2))));
1025 		ire_delete(ire);
1026 	}
1027 
1028 }
1029 
1030 /*
1031  * ire_walk routine to delete all IRE_CACHE/IRE_HOST type redirect entries
1032  * that have a given gateway address.
1033  */
1034 void
1035 ire_delete_cache_gw_v6(ire_t *ire, char *addr)
1036 {
1037 	in6_addr_t	*gw_addr = (in6_addr_t *)addr;
1038 	char		buf1[INET6_ADDRSTRLEN];
1039 	char		buf2[INET6_ADDRSTRLEN];
1040 	in6_addr_t	ire_gw_addr_v6;
1041 
1042 	if (!(ire->ire_type & IRE_CACHE) &&
1043 	    !(ire->ire_flags & RTF_DYNAMIC))
1044 		return;
1045 
1046 	mutex_enter(&ire->ire_lock);
1047 	ire_gw_addr_v6 = ire->ire_gateway_addr_v6;
1048 	mutex_exit(&ire->ire_lock);
1049 
1050 	if (IN6_ARE_ADDR_EQUAL(&ire_gw_addr_v6, gw_addr)) {
1051 		ip1dbg(("ire_delete_cache_gw_v6: deleted %s type %d to %s\n",
1052 		    inet_ntop(AF_INET6, &ire->ire_src_addr_v6,
1053 		    buf1, sizeof (buf1)),
1054 		    ire->ire_type,
1055 		    inet_ntop(AF_INET6, &ire_gw_addr_v6,
1056 		    buf2, sizeof (buf2))));
1057 		ire_delete(ire);
1058 	}
1059 }
1060 
1061 /*
1062  * Remove all IRE_CACHE entries that match
1063  * the ire specified.  (Sometimes called
1064  * as writer though not required by this function.)
1065  *
1066  * The flag argument indicates if the
1067  * flush request is due to addition
1068  * of new route (IRE_FLUSH_ADD) or deletion of old
1069  * route (IRE_FLUSH_DELETE).
1070  *
1071  * This routine takes only the IREs from the forwarding
1072  * table and flushes the corresponding entries from
1073  * the cache table.
1074  *
1075  * When flushing due to the deletion of an old route, it
1076  * just checks the cache handles (ire_phandle and ire_ihandle) and
1077  * deletes the ones that match.
1078  *
1079  * When flushing due to the creation of a new route, it checks
1080  * if a cache entry's address matches the one in the IRE and
1081  * that the cache entry's parent has a less specific mask than the
1082  * one in IRE. The destination of such a cache entry could be the
1083  * gateway for other cache entries, so we need to flush those as
1084  * well by looking for gateway addresses matching the IRE's address.
1085  */
1086 void
1087 ire_flush_cache_v6(ire_t *ire, int flag)
1088 {
1089 	int i;
1090 	ire_t *cire;
1091 	irb_t *irb;
1092 	ip_stack_t	*ipst = ire->ire_ipst;
1093 
1094 	if (ire->ire_type & IRE_CACHE)
1095 		return;
1096 
1097 	/*
1098 	 * If a default is just created, there is no point
1099 	 * in going through the cache, as there will not be any
1100 	 * cached ires.
1101 	 */
1102 	if (ire->ire_type == IRE_DEFAULT && flag == IRE_FLUSH_ADD)
1103 		return;
1104 	if (flag == IRE_FLUSH_ADD) {
1105 		/*
1106 		 * This selective flush is
1107 		 * due to the addition of
1108 		 * new IRE.
1109 		 */
1110 		for (i = 0; i < ipst->ips_ip6_cache_table_size; i++) {
1111 			irb = &ipst->ips_ip_cache_table_v6[i];
1112 			if ((cire = irb->irb_ire) == NULL)
1113 				continue;
1114 			IRB_REFHOLD(irb);
1115 			for (cire = irb->irb_ire; cire != NULL;
1116 			    cire = cire->ire_next) {
1117 				if (cire->ire_type != IRE_CACHE)
1118 					continue;
1119 				/*
1120 				 * If 'cire' belongs to the same subnet
1121 				 * as the new ire being added, and 'cire'
1122 				 * is derived from a prefix that is less
1123 				 * specific than the new ire being added,
1124 				 * we need to flush 'cire'; for instance,
1125 				 * when a new interface comes up.
1126 				 */
1127 				if ((V6_MASK_EQ_2(cire->ire_addr_v6,
1128 				    ire->ire_mask_v6, ire->ire_addr_v6) &&
1129 				    (ip_mask_to_plen_v6(&cire->ire_cmask_v6) <=
1130 				    ire->ire_masklen))) {
1131 					ire_delete(cire);
1132 					continue;
1133 				}
1134 				/*
1135 				 * This is the case when the ire_gateway_addr
1136 				 * of 'cire' belongs to the same subnet as
1137 				 * the new ire being added.
1138 				 * Flushing such ires is sometimes required to
1139 				 * avoid misrouting: say we have a machine with
1140 				 * two interfaces (I1 and I2), a default router
1141 				 * R on the I1 subnet, and a host route to an
1142 				 * off-link destination D with a gateway G on
1143 				 * the I2 subnet.
1144 				 * Under normal operation, we will have an
1145 				 * on-link cache entry for G and an off-link
1146 				 * cache entry for D with G as ire_gateway_addr,
1147 				 * traffic to D will reach its destination
1148 				 * through gateway G.
1149 				 * If the administrator does 'ifconfig I2 down',
1150 				 * the cache entries for D and G will be
1151 				 * flushed. However, G will now be resolved as
1152 				 * an off-link destination using R (the default
1153 				 * router) as gateway. Then D will also be
1154 				 * resolved as an off-link destination using G
1155 				 * as gateway - this behavior is due to
1156 				 * compatibility reasons, see comment in
1157 				 * ire_ihandle_lookup_offlink(). Traffic to D
1158 				 * will go to the router R and probably won't
1159 				 * reach the destination.
1160 				 * The administrator then does 'ifconfig I2 up'.
1161 				 * Since G is on the I2 subnet, this routine
1162 				 * will flush its cache entry. It must also
1163 				 * flush the cache entry for D, otherwise
1164 				 * traffic will stay misrouted until the IRE
1165 				 * times out.
1166 				 */
1167 				if (V6_MASK_EQ_2(cire->ire_gateway_addr_v6,
1168 				    ire->ire_mask_v6, ire->ire_addr_v6)) {
1169 					ire_delete(cire);
1170 					continue;
1171 				}
1172 			}
1173 			IRB_REFRELE(irb);
1174 		}
1175 	} else {
1176 		/*
1177 		 * delete the cache entries based on
1178 		 * handle in the IRE as this IRE is
1179 		 * being deleted/changed.
1180 		 */
1181 		for (i = 0; i < ipst->ips_ip6_cache_table_size; i++) {
1182 			irb = &ipst->ips_ip_cache_table_v6[i];
1183 			if ((cire = irb->irb_ire) == NULL)
1184 				continue;
1185 			IRB_REFHOLD(irb);
1186 			for (cire = irb->irb_ire; cire != NULL;
1187 			    cire = cire->ire_next) {
1188 				if (cire->ire_type != IRE_CACHE)
1189 					continue;
1190 				if ((cire->ire_phandle == 0 ||
1191 				    cire->ire_phandle != ire->ire_phandle) &&
1192 				    (cire->ire_ihandle == 0 ||
1193 				    cire->ire_ihandle != ire->ire_ihandle))
1194 					continue;
1195 				ire_delete(cire);
1196 			}
1197 			IRB_REFRELE(irb);
1198 		}
1199 	}
1200 }
1201 
1202 /*
1203  * Matches the arguments passed with the values in the ire.
1204  *
1205  * Note: for match types that match using "ipif" passed in, ipif
1206  * must be checked for non-NULL before calling this routine.
1207  */
1208 static boolean_t
1209 ire_match_args_v6(ire_t *ire, const in6_addr_t *addr, const in6_addr_t *mask,
1210     const in6_addr_t *gateway, int type, const ipif_t *ipif, zoneid_t zoneid,
1211     uint32_t ihandle, const ts_label_t *tsl, int match_flags)
1212 {
1213 	in6_addr_t masked_addr;
1214 	in6_addr_t gw_addr_v6;
1215 	ill_t *ire_ill = NULL, *dst_ill;
1216 	ill_t *ipif_ill = NULL;
1217 	ill_group_t *ire_ill_group = NULL;
1218 	ill_group_t *ipif_ill_group = NULL;
1219 	ipif_t	*src_ipif;
1220 
1221 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
1222 	ASSERT(addr != NULL);
1223 	ASSERT(mask != NULL);
1224 	ASSERT((!(match_flags & MATCH_IRE_GW)) || gateway != NULL);
1225 	ASSERT((!(match_flags & (MATCH_IRE_ILL|MATCH_IRE_ILL_GROUP))) ||
1226 	    (ipif != NULL && ipif->ipif_isv6));
1227 
1228 	/*
1229 	 * HIDDEN cache entries have to be looked up specifically with
1230 	 * MATCH_IRE_MARK_HIDDEN. MATCH_IRE_MARK_HIDDEN is usually set
1231 	 * when the interface is FAILED or INACTIVE. In that case,
1232 	 * any IRE_CACHES that exists should be marked with
1233 	 * IRE_MARK_HIDDEN. So, we don't really need to match below
1234 	 * for IRE_MARK_HIDDEN. But we do so for consistency.
1235 	 */
1236 	if (!(match_flags & MATCH_IRE_MARK_HIDDEN) &&
1237 	    (ire->ire_marks & IRE_MARK_HIDDEN))
1238 		return (B_FALSE);
1239 
1240 	if (zoneid != ALL_ZONES && zoneid != ire->ire_zoneid &&
1241 	    ire->ire_zoneid != ALL_ZONES) {
1242 		/*
1243 		 * If MATCH_IRE_ZONEONLY has been set and the supplied zoneid is
1244 		 * valid and does not match that of ire_zoneid, a failure to
1245 		 * match is reported at this point. Otherwise, since some IREs
1246 		 * that are available in the global zone can be used in local
1247 		 * zones, additional checks need to be performed:
1248 		 *
1249 		 *	IRE_CACHE and IRE_LOOPBACK entries should
1250 		 *	never be matched in this situation.
1251 		 *
1252 		 *	IRE entries that have an interface associated with them
1253 		 *	should in general not match unless they are an IRE_LOCAL
1254 		 *	or in the case when MATCH_IRE_DEFAULT has been set in
1255 		 *	the caller.  In the case of the former, checking of the
1256 		 *	other fields supplied should take place.
1257 		 *
1258 		 *	In the case where MATCH_IRE_DEFAULT has been set,
1259 		 *	all of the ipif's associated with the IRE's ill are
1260 		 *	checked to see if there is a matching zoneid.  If any
1261 		 *	one ipif has a matching zoneid, this IRE is a
1262 		 *	potential candidate so checking of the other fields
1263 		 *	takes place.
1264 		 *
1265 		 *	In the case where the IRE_INTERFACE has a usable source
1266 		 *	address (indicated by ill_usesrc_ifindex) in the
1267 		 *	correct zone then it's permitted to return this IRE
1268 		 */
1269 		if (match_flags & MATCH_IRE_ZONEONLY)
1270 			return (B_FALSE);
1271 		if (ire->ire_type & (IRE_CACHE | IRE_LOOPBACK))
1272 			return (B_FALSE);
1273 		/*
1274 		 * Note, IRE_INTERFACE can have the stq as NULL. For
1275 		 * example, if the default multicast route is tied to
1276 		 * the loopback address.
1277 		 */
1278 		if ((ire->ire_type & IRE_INTERFACE) &&
1279 		    (ire->ire_stq != NULL)) {
1280 			dst_ill = (ill_t *)ire->ire_stq->q_ptr;
1281 			/*
1282 			 * If there is a usable source address in the
1283 			 * zone, then it's ok to return an
1284 			 * IRE_INTERFACE
1285 			 */
1286 			if ((dst_ill->ill_usesrc_ifindex != 0) &&
1287 			    (src_ipif = ipif_select_source_v6(dst_ill, addr,
1288 			    RESTRICT_TO_NONE, IPV6_PREFER_SRC_DEFAULT, zoneid))
1289 			    != NULL) {
1290 				ip3dbg(("ire_match_args: src_ipif %p"
1291 				    " dst_ill %p", (void *)src_ipif,
1292 				    (void *)dst_ill));
1293 				ipif_refrele(src_ipif);
1294 			} else {
1295 				ip3dbg(("ire_match_args: src_ipif NULL"
1296 				    " dst_ill %p\n", (void *)dst_ill));
1297 				return (B_FALSE);
1298 			}
1299 		}
1300 		if (ire->ire_ipif != NULL && ire->ire_type != IRE_LOCAL &&
1301 		    !(ire->ire_type & IRE_INTERFACE)) {
1302 			ipif_t	*tipif;
1303 
1304 			if ((match_flags & MATCH_IRE_DEFAULT) == 0)
1305 				return (B_FALSE);
1306 			mutex_enter(&ire->ire_ipif->ipif_ill->ill_lock);
1307 			for (tipif = ire->ire_ipif->ipif_ill->ill_ipif;
1308 			    tipif != NULL; tipif = tipif->ipif_next) {
1309 				if (IPIF_CAN_LOOKUP(tipif) &&
1310 				    (tipif->ipif_flags & IPIF_UP) &&
1311 				    (tipif->ipif_zoneid == zoneid ||
1312 				    tipif->ipif_zoneid == ALL_ZONES))
1313 					break;
1314 			}
1315 			mutex_exit(&ire->ire_ipif->ipif_ill->ill_lock);
1316 			if (tipif == NULL)
1317 				return (B_FALSE);
1318 		}
1319 	}
1320 
1321 	if (match_flags & MATCH_IRE_GW) {
1322 		mutex_enter(&ire->ire_lock);
1323 		gw_addr_v6 = ire->ire_gateway_addr_v6;
1324 		mutex_exit(&ire->ire_lock);
1325 	}
1326 	/*
1327 	 * For IRE_CACHES, MATCH_IRE_ILL/ILL_GROUP really means that
1328 	 * somebody wants to send out on a particular interface which
1329 	 * is given by ire_stq and hence use ire_stq to derive the ill
1330 	 * value. ire_ipif for IRE_CACHES is just the
1331 	 * means of getting a source address i.e ire_src_addr_v6 =
1332 	 * ire->ire_ipif->ipif_src_addr_v6.
1333 	 */
1334 	if (match_flags & (MATCH_IRE_ILL|MATCH_IRE_ILL_GROUP)) {
1335 		ire_ill = ire_to_ill(ire);
1336 		if (ire_ill != NULL)
1337 			ire_ill_group = ire_ill->ill_group;
1338 		ipif_ill = ipif->ipif_ill;
1339 		ipif_ill_group = ipif_ill->ill_group;
1340 	}
1341 
1342 	/* No ire_addr_v6 bits set past the mask */
1343 	ASSERT(V6_MASK_EQ(ire->ire_addr_v6, ire->ire_mask_v6,
1344 	    ire->ire_addr_v6));
1345 	V6_MASK_COPY(*addr, *mask, masked_addr);
1346 
1347 	if (V6_MASK_EQ(*addr, *mask, ire->ire_addr_v6) &&
1348 	    ((!(match_flags & MATCH_IRE_GW)) ||
1349 	    IN6_ARE_ADDR_EQUAL(&gw_addr_v6, gateway)) &&
1350 	    ((!(match_flags & MATCH_IRE_TYPE)) ||
1351 	    (ire->ire_type & type)) &&
1352 	    ((!(match_flags & MATCH_IRE_SRC)) ||
1353 	    IN6_ARE_ADDR_EQUAL(&ire->ire_src_addr_v6,
1354 	    &ipif->ipif_v6src_addr)) &&
1355 	    ((!(match_flags & MATCH_IRE_IPIF)) ||
1356 	    (ire->ire_ipif == ipif)) &&
1357 	    ((!(match_flags & MATCH_IRE_MARK_HIDDEN)) ||
1358 	    (ire->ire_type != IRE_CACHE ||
1359 	    ire->ire_marks & IRE_MARK_HIDDEN)) &&
1360 	    ((!(match_flags & MATCH_IRE_ILL)) ||
1361 	    (ire_ill == ipif_ill)) &&
1362 	    ((!(match_flags & MATCH_IRE_IHANDLE)) ||
1363 	    (ire->ire_ihandle == ihandle)) &&
1364 	    ((!(match_flags & MATCH_IRE_ILL_GROUP)) ||
1365 	    (ire_ill == ipif_ill) ||
1366 	    (ire_ill_group != NULL &&
1367 	    ire_ill_group == ipif_ill_group)) &&
1368 	    ((!(match_flags & MATCH_IRE_SECATTR)) ||
1369 	    (!is_system_labeled()) ||
1370 	    (tsol_ire_match_gwattr(ire, tsl) == 0))) {
1371 		/* We found the matched IRE */
1372 		return (B_TRUE);
1373 	}
1374 	return (B_FALSE);
1375 }
1376 
1377 /*
1378  * Lookup for a route in all the tables
1379  */
1380 ire_t *
1381 ire_route_lookup_v6(const in6_addr_t *addr, const in6_addr_t *mask,
1382     const in6_addr_t *gateway, int type, const ipif_t *ipif, ire_t **pire,
1383     zoneid_t zoneid, const ts_label_t *tsl, int flags, ip_stack_t *ipst)
1384 {
1385 	ire_t *ire = NULL;
1386 
1387 	/*
1388 	 * ire_match_args_v6() will dereference ipif MATCH_IRE_SRC or
1389 	 * MATCH_IRE_ILL is set.
1390 	 */
1391 	if ((flags & (MATCH_IRE_SRC | MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) &&
1392 	    (ipif == NULL))
1393 		return (NULL);
1394 
1395 	/*
1396 	 * might be asking for a cache lookup,
1397 	 * This is not best way to lookup cache,
1398 	 * user should call ire_cache_lookup directly.
1399 	 *
1400 	 * If MATCH_IRE_TYPE was set, first lookup in the cache table and then
1401 	 * in the forwarding table, if the applicable type flags were set.
1402 	 */
1403 	if ((flags & MATCH_IRE_TYPE) == 0 || (type & IRE_CACHETABLE) != 0) {
1404 		ire = ire_ctable_lookup_v6(addr, gateway, type, ipif, zoneid,
1405 		    tsl, flags, ipst);
1406 		if (ire != NULL)
1407 			return (ire);
1408 	}
1409 	if ((flags & MATCH_IRE_TYPE) == 0 || (type & IRE_FORWARDTABLE) != 0) {
1410 		ire = ire_ftable_lookup_v6(addr, mask, gateway, type, ipif,
1411 		    pire, zoneid, 0, tsl, flags, ipst);
1412 	}
1413 	return (ire);
1414 }
1415 
1416 /*
1417  * Lookup a route in forwarding table.
1418  * specific lookup is indicated by passing the
1419  * required parameters and indicating the
1420  * match required in flag field.
1421  *
1422  * Looking for default route can be done in three ways
1423  * 1) pass mask as ipv6_all_zeros and set MATCH_IRE_MASK in flags field
1424  *    along with other matches.
1425  * 2) pass type as IRE_DEFAULT and set MATCH_IRE_TYPE in flags
1426  *    field along with other matches.
1427  * 3) if the destination and mask are passed as zeros.
1428  *
1429  * A request to return a default route if no route
1430  * is found, can be specified by setting MATCH_IRE_DEFAULT
1431  * in flags.
1432  *
1433  * It does not support recursion more than one level. It
1434  * will do recursive lookup only when the lookup maps to
1435  * a prefix or default route and MATCH_IRE_RECURSIVE flag is passed.
1436  *
1437  * If the routing table is setup to allow more than one level
1438  * of recursion, the cleaning up cache table will not work resulting
1439  * in invalid routing.
1440  *
1441  * Supports link-local addresses by following the ipif/ill when recursing.
1442  *
1443  * NOTE : When this function returns NULL, pire has already been released.
1444  *	  pire is valid only when this function successfully returns an
1445  *	  ire.
1446  */
1447 ire_t *
1448 ire_ftable_lookup_v6(const in6_addr_t *addr, const in6_addr_t *mask,
1449     const in6_addr_t *gateway, int type, const ipif_t *ipif, ire_t **pire,
1450     zoneid_t zoneid, uint32_t ihandle, const ts_label_t *tsl, int flags,
1451     ip_stack_t *ipst)
1452 {
1453 	irb_t *irb_ptr;
1454 	ire_t	*rire;
1455 	ire_t *ire = NULL;
1456 	ire_t	*saved_ire;
1457 	nce_t	*nce;
1458 	int i;
1459 	in6_addr_t gw_addr_v6;
1460 
1461 	ASSERT(addr != NULL);
1462 	ASSERT((!(flags & MATCH_IRE_MASK)) || mask != NULL);
1463 	ASSERT((!(flags & MATCH_IRE_GW)) || gateway != NULL);
1464 	ASSERT(ipif == NULL || ipif->ipif_isv6);
1465 
1466 	/*
1467 	 * When we return NULL from this function, we should make
1468 	 * sure that *pire is NULL so that the callers will not
1469 	 * wrongly REFRELE the pire.
1470 	 */
1471 	if (pire != NULL)
1472 		*pire = NULL;
1473 	/*
1474 	 * ire_match_args_v6() will dereference ipif MATCH_IRE_SRC or
1475 	 * MATCH_IRE_ILL is set.
1476 	 */
1477 	if ((flags & (MATCH_IRE_SRC | MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) &&
1478 	    (ipif == NULL))
1479 		return (NULL);
1480 
1481 	/*
1482 	 * If the mask is known, the lookup
1483 	 * is simple, if the mask is not known
1484 	 * we need to search.
1485 	 */
1486 	if (flags & MATCH_IRE_MASK) {
1487 		uint_t masklen;
1488 
1489 		masklen = ip_mask_to_plen_v6(mask);
1490 		if (ipst->ips_ip_forwarding_table_v6[masklen] == NULL)
1491 			return (NULL);
1492 		irb_ptr = &(ipst->ips_ip_forwarding_table_v6[masklen][
1493 		    IRE_ADDR_MASK_HASH_V6(*addr, *mask,
1494 		    ipst->ips_ip6_ftable_hash_size)]);
1495 		rw_enter(&irb_ptr->irb_lock, RW_READER);
1496 		for (ire = irb_ptr->irb_ire; ire != NULL;
1497 		    ire = ire->ire_next) {
1498 			if (ire->ire_marks & IRE_MARK_CONDEMNED)
1499 				continue;
1500 			if (ire_match_args_v6(ire, addr, mask, gateway, type,
1501 			    ipif, zoneid, ihandle, tsl, flags))
1502 				goto found_ire;
1503 		}
1504 		rw_exit(&irb_ptr->irb_lock);
1505 	} else {
1506 		/*
1507 		 * In this case we don't know the mask, we need to
1508 		 * search the table assuming different mask sizes.
1509 		 * we start with 128 bit mask, we don't allow default here.
1510 		 */
1511 		for (i = (IP6_MASK_TABLE_SIZE - 1); i > 0; i--) {
1512 			in6_addr_t tmpmask;
1513 
1514 			if ((ipst->ips_ip_forwarding_table_v6[i]) == NULL)
1515 				continue;
1516 			(void) ip_plen_to_mask_v6(i, &tmpmask);
1517 			irb_ptr = &ipst->ips_ip_forwarding_table_v6[i][
1518 			    IRE_ADDR_MASK_HASH_V6(*addr, tmpmask,
1519 			    ipst->ips_ip6_ftable_hash_size)];
1520 			rw_enter(&irb_ptr->irb_lock, RW_READER);
1521 			for (ire = irb_ptr->irb_ire; ire != NULL;
1522 			    ire = ire->ire_next) {
1523 				if (ire->ire_marks & IRE_MARK_CONDEMNED)
1524 					continue;
1525 				if (ire_match_args_v6(ire, addr,
1526 				    &ire->ire_mask_v6, gateway, type, ipif,
1527 				    zoneid, ihandle, tsl, flags))
1528 					goto found_ire;
1529 			}
1530 			rw_exit(&irb_ptr->irb_lock);
1531 		}
1532 	}
1533 
1534 	/*
1535 	 * We come here if no route has yet been found.
1536 	 *
1537 	 * Handle the case where default route is
1538 	 * requested by specifying type as one of the possible
1539 	 * types for that can have a zero mask (IRE_DEFAULT and IRE_INTERFACE).
1540 	 *
1541 	 * If MATCH_IRE_MASK is specified, then the appropriate default route
1542 	 * would have been found above if it exists so it isn't looked up here.
1543 	 * If MATCH_IRE_DEFAULT was also specified, then a default route will be
1544 	 * searched for later.
1545 	 */
1546 	if ((flags & (MATCH_IRE_TYPE | MATCH_IRE_MASK)) == MATCH_IRE_TYPE &&
1547 	    (type & (IRE_DEFAULT | IRE_INTERFACE))) {
1548 		if (ipst->ips_ip_forwarding_table_v6[0] != NULL) {
1549 			/* addr & mask is zero for defaults */
1550 			irb_ptr = &ipst->ips_ip_forwarding_table_v6[0][
1551 			    IRE_ADDR_HASH_V6(ipv6_all_zeros,
1552 			    ipst->ips_ip6_ftable_hash_size)];
1553 			rw_enter(&irb_ptr->irb_lock, RW_READER);
1554 			for (ire = irb_ptr->irb_ire; ire != NULL;
1555 			    ire = ire->ire_next) {
1556 
1557 				if (ire->ire_marks & IRE_MARK_CONDEMNED)
1558 					continue;
1559 
1560 				if (ire_match_args_v6(ire, addr,
1561 				    &ipv6_all_zeros, gateway, type, ipif,
1562 				    zoneid, ihandle, tsl, flags))
1563 					goto found_ire;
1564 			}
1565 			rw_exit(&irb_ptr->irb_lock);
1566 		}
1567 	}
1568 	/*
1569 	 * We come here only if no route is found.
1570 	 * see if the default route can be used which is allowed
1571 	 * only if the default matching criteria is specified.
1572 	 * The ipv6_ire_default_count tracks the number of IRE_DEFAULT
1573 	 * entries. However, the ip_forwarding_table_v6[0] also contains
1574 	 * interface routes thus the count can be zero.
1575 	 */
1576 	saved_ire = NULL;
1577 	if ((flags & (MATCH_IRE_DEFAULT | MATCH_IRE_MASK)) ==
1578 	    MATCH_IRE_DEFAULT) {
1579 		ire_t	*ire_origin;
1580 		uint_t	g_index;
1581 		uint_t	index;
1582 
1583 		if (ipst->ips_ip_forwarding_table_v6[0] == NULL)
1584 			return (NULL);
1585 		irb_ptr = &(ipst->ips_ip_forwarding_table_v6[0])[0];
1586 
1587 		/*
1588 		 * Keep a tab on the bucket while looking the IRE_DEFAULT
1589 		 * entries. We need to keep track of a particular IRE
1590 		 * (ire_origin) so this ensures that it will not be unlinked
1591 		 * from the hash list during the recursive lookup below.
1592 		 */
1593 		IRB_REFHOLD(irb_ptr);
1594 		ire = irb_ptr->irb_ire;
1595 		if (ire == NULL) {
1596 			IRB_REFRELE(irb_ptr);
1597 			return (NULL);
1598 		}
1599 
1600 		/*
1601 		 * Get the index first, since it can be changed by other
1602 		 * threads. Then get to the right default route skipping
1603 		 * default interface routes if any. As we hold a reference on
1604 		 * the IRE bucket, ipv6_ire_default_count can only increase so
1605 		 * we can't reach the end of the hash list unexpectedly.
1606 		 */
1607 		if (ipst->ips_ipv6_ire_default_count != 0) {
1608 			g_index = ipst->ips_ipv6_ire_default_index++;
1609 			index = g_index % ipst->ips_ipv6_ire_default_count;
1610 			while (index != 0) {
1611 				if (!(ire->ire_type & IRE_INTERFACE))
1612 					index--;
1613 				ire = ire->ire_next;
1614 			}
1615 			ASSERT(ire != NULL);
1616 		} else {
1617 			/*
1618 			 * No default route, so we only have default interface
1619 			 * routes: don't enter the first loop.
1620 			 */
1621 			ire = NULL;
1622 		}
1623 
1624 		/*
1625 		 * Round-robin the default routers list looking for a neighbor
1626 		 * that matches the passed in parameters and is reachable.  If
1627 		 * none found, just return a route from the default router list
1628 		 * if it exists. If we can't find a default route (IRE_DEFAULT),
1629 		 * look for interface default routes.
1630 		 * We start with the ire we found above and we walk the hash
1631 		 * list until we're back where we started, see
1632 		 * ire_get_next_default_ire(). It doesn't matter if default
1633 		 * routes are added or deleted by other threads - we know this
1634 		 * ire will stay in the list because we hold a reference on the
1635 		 * ire bucket.
1636 		 * NB: if we only have interface default routes, ire is NULL so
1637 		 * we don't even enter this loop (see above).
1638 		 */
1639 		ire_origin = ire;
1640 		for (; ire != NULL;
1641 		    ire = ire_get_next_default_ire(ire, ire_origin)) {
1642 
1643 			if (ire_match_args_v6(ire, addr,
1644 			    &ipv6_all_zeros, gateway, type, ipif,
1645 			    zoneid, ihandle, tsl, flags)) {
1646 				int match_flags;
1647 
1648 				/*
1649 				 * We have something to work with.
1650 				 * If we can find a resolved/reachable
1651 				 * entry, we will use this. Otherwise
1652 				 * we'll try to find an entry that has
1653 				 * a resolved cache entry. We will fallback
1654 				 * on this if we don't find anything else.
1655 				 */
1656 				if (saved_ire == NULL)
1657 					saved_ire = ire;
1658 				mutex_enter(&ire->ire_lock);
1659 				gw_addr_v6 = ire->ire_gateway_addr_v6;
1660 				mutex_exit(&ire->ire_lock);
1661 				match_flags = MATCH_IRE_ILL_GROUP |
1662 				    MATCH_IRE_SECATTR;
1663 				rire = ire_ctable_lookup_v6(&gw_addr_v6, NULL,
1664 				    0, ire->ire_ipif, zoneid, tsl, match_flags,
1665 				    ipst);
1666 				if (rire != NULL) {
1667 					nce = rire->ire_nce;
1668 					if (nce != NULL &&
1669 					    NCE_ISREACHABLE(nce) &&
1670 					    nce->nce_flags & NCE_F_ISROUTER) {
1671 						ire_refrele(rire);
1672 						IRE_REFHOLD(ire);
1673 						IRB_REFRELE(irb_ptr);
1674 						goto found_ire_held;
1675 					} else if (nce != NULL &&
1676 					    !(nce->nce_flags &
1677 					    NCE_F_ISROUTER)) {
1678 						/*
1679 						 * Make sure we don't use
1680 						 * this ire
1681 						 */
1682 						if (saved_ire == ire)
1683 							saved_ire = NULL;
1684 					}
1685 					ire_refrele(rire);
1686 				} else if (ipst->
1687 				    ips_ipv6_ire_default_count > 1 &&
1688 				    zoneid != GLOBAL_ZONEID) {
1689 					/*
1690 					 * When we're in a local zone, we're
1691 					 * only interested in default routers
1692 					 * that are reachable through ipifs
1693 					 * within our zone.
1694 					 * The potentially expensive call to
1695 					 * ire_route_lookup_v6() is avoided when
1696 					 * we have only one default route.
1697 					 */
1698 					int ire_match_flags = MATCH_IRE_TYPE |
1699 					    MATCH_IRE_SECATTR;
1700 
1701 					if (ire->ire_ipif != NULL) {
1702 						ire_match_flags |=
1703 						    MATCH_IRE_ILL_GROUP;
1704 					}
1705 					rire = ire_route_lookup_v6(&gw_addr_v6,
1706 					    NULL, NULL, IRE_INTERFACE,
1707 					    ire->ire_ipif, NULL,
1708 					    zoneid, tsl, ire_match_flags, ipst);
1709 					if (rire != NULL) {
1710 						ire_refrele(rire);
1711 						saved_ire = ire;
1712 					} else if (saved_ire == ire) {
1713 						/*
1714 						 * Make sure we don't use
1715 						 * this ire
1716 						 */
1717 						saved_ire = NULL;
1718 					}
1719 				}
1720 			}
1721 		}
1722 		if (saved_ire != NULL) {
1723 			ire = saved_ire;
1724 			IRE_REFHOLD(ire);
1725 			IRB_REFRELE(irb_ptr);
1726 			goto found_ire_held;
1727 		} else {
1728 			/*
1729 			 * Look for a interface default route matching the
1730 			 * args passed in. No round robin here. Just pick
1731 			 * the right one.
1732 			 */
1733 			for (ire = irb_ptr->irb_ire; ire != NULL;
1734 			    ire = ire->ire_next) {
1735 
1736 				if (!(ire->ire_type & IRE_INTERFACE))
1737 					continue;
1738 
1739 				if (ire->ire_marks & IRE_MARK_CONDEMNED)
1740 					continue;
1741 
1742 				if (ire_match_args_v6(ire, addr,
1743 				    &ipv6_all_zeros, gateway, type, ipif,
1744 				    zoneid, ihandle, tsl, flags)) {
1745 					IRE_REFHOLD(ire);
1746 					IRB_REFRELE(irb_ptr);
1747 					goto found_ire_held;
1748 				}
1749 			}
1750 			IRB_REFRELE(irb_ptr);
1751 		}
1752 	}
1753 	ASSERT(ire == NULL);
1754 	ip1dbg(("ire_ftable_lookup_v6: returning NULL ire"));
1755 	return (NULL);
1756 found_ire:
1757 	ASSERT((ire->ire_marks & IRE_MARK_CONDEMNED) == 0);
1758 	IRE_REFHOLD(ire);
1759 	rw_exit(&irb_ptr->irb_lock);
1760 
1761 found_ire_held:
1762 	if ((flags & MATCH_IRE_RJ_BHOLE) &&
1763 	    (ire->ire_flags & (RTF_BLACKHOLE | RTF_REJECT))) {
1764 		return (ire);
1765 	}
1766 	/*
1767 	 * At this point, IRE that was found must be an IRE_FORWARDTABLE
1768 	 * or IRE_CACHETABLE type.  If this is a recursive lookup and an
1769 	 * IRE_INTERFACE type was found, return that.  If it was some other
1770 	 * IRE_FORWARDTABLE type of IRE (one of the prefix types), then it
1771 	 * is necessary to fill in the  parent IRE pointed to by pire, and
1772 	 * then lookup the gateway address of  the parent.  For backwards
1773 	 * compatiblity, if this lookup returns an
1774 	 * IRE other than a IRE_CACHETABLE or IRE_INTERFACE, then one more level
1775 	 * of lookup is done.
1776 	 */
1777 	if (flags & MATCH_IRE_RECURSIVE) {
1778 		const ipif_t *gw_ipif;
1779 		int match_flags = MATCH_IRE_DSTONLY;
1780 
1781 		if (ire->ire_type & IRE_INTERFACE)
1782 			return (ire);
1783 		if (pire != NULL)
1784 			*pire = ire;
1785 		/*
1786 		 * If we can't find an IRE_INTERFACE or the caller has not
1787 		 * asked for pire, we need to REFRELE the saved_ire.
1788 		 */
1789 		saved_ire = ire;
1790 
1791 		/*
1792 		 * Currently MATCH_IRE_ILL is never used with
1793 		 * (MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT) while
1794 		 * sending out packets as MATCH_IRE_ILL is used only
1795 		 * for communicating with on-link hosts. We can't assert
1796 		 * that here as RTM_GET calls this function with
1797 		 * MATCH_IRE_ILL | MATCH_IRE_DEFAULT | MATCH_IRE_RECURSIVE.
1798 		 * We have already used the MATCH_IRE_ILL in determining
1799 		 * the right prefix route at this point. To match the
1800 		 * behavior of how we locate routes while sending out
1801 		 * packets, we don't want to use MATCH_IRE_ILL below
1802 		 * while locating the interface route.
1803 		 */
1804 		if (ire->ire_ipif != NULL)
1805 			match_flags |= MATCH_IRE_ILL_GROUP;
1806 
1807 		mutex_enter(&ire->ire_lock);
1808 		gw_addr_v6 = ire->ire_gateway_addr_v6;
1809 		mutex_exit(&ire->ire_lock);
1810 
1811 		ire = ire_route_lookup_v6(&gw_addr_v6, NULL, NULL, 0,
1812 		    ire->ire_ipif, NULL, zoneid, tsl, match_flags, ipst);
1813 		if (ire == NULL) {
1814 			/*
1815 			 * In this case we have to deal with the
1816 			 * MATCH_IRE_PARENT flag, which means the
1817 			 * parent has to be returned if ire is NULL.
1818 			 * The aim of this is to have (at least) a starting
1819 			 * ire when we want to look at all of the ires in a
1820 			 * bucket aimed at a single destination (as is the
1821 			 * case in ip_newroute_v6 for the RTF_MULTIRT
1822 			 * flagged routes).
1823 			 */
1824 			if (flags & MATCH_IRE_PARENT) {
1825 				if (pire != NULL) {
1826 					/*
1827 					 * Need an extra REFHOLD, if the
1828 					 * parent ire is returned via both
1829 					 * ire and pire.
1830 					 */
1831 					IRE_REFHOLD(saved_ire);
1832 				}
1833 				ire = saved_ire;
1834 			} else {
1835 				ire_refrele(saved_ire);
1836 				if (pire != NULL)
1837 					*pire = NULL;
1838 			}
1839 			return (ire);
1840 		}
1841 		if (ire->ire_type & (IRE_CACHETABLE | IRE_INTERFACE)) {
1842 			/*
1843 			 * If the caller did not ask for pire, release
1844 			 * it now.
1845 			 */
1846 			if (pire == NULL) {
1847 				ire_refrele(saved_ire);
1848 			}
1849 			return (ire);
1850 		}
1851 		match_flags |= MATCH_IRE_TYPE;
1852 		mutex_enter(&ire->ire_lock);
1853 		gw_addr_v6 = ire->ire_gateway_addr_v6;
1854 		mutex_exit(&ire->ire_lock);
1855 		gw_ipif = ire->ire_ipif;
1856 		ire_refrele(ire);
1857 		ire = ire_route_lookup_v6(&gw_addr_v6, NULL, NULL,
1858 		    (IRE_CACHETABLE | IRE_INTERFACE), gw_ipif, NULL, zoneid,
1859 		    NULL, match_flags, ipst);
1860 		if (ire == NULL) {
1861 			/*
1862 			 * In this case we have to deal with the
1863 			 * MATCH_IRE_PARENT flag, which means the
1864 			 * parent has to be returned if ire is NULL.
1865 			 * The aim of this is to have (at least) a starting
1866 			 * ire when we want to look at all of the ires in a
1867 			 * bucket aimed at a single destination (as is the
1868 			 * case in ip_newroute_v6 for the RTF_MULTIRT
1869 			 * flagged routes).
1870 			 */
1871 			if (flags & MATCH_IRE_PARENT) {
1872 				if (pire != NULL) {
1873 					/*
1874 					 * Need an extra REFHOLD, if the
1875 					 * parent ire is returned via both
1876 					 * ire and pire.
1877 					 */
1878 					IRE_REFHOLD(saved_ire);
1879 				}
1880 				ire = saved_ire;
1881 			} else {
1882 				ire_refrele(saved_ire);
1883 				if (pire != NULL)
1884 					*pire = NULL;
1885 			}
1886 			return (ire);
1887 		} else if (pire == NULL) {
1888 			/*
1889 			 * If the caller did not ask for pire, release
1890 			 * it now.
1891 			 */
1892 			ire_refrele(saved_ire);
1893 		}
1894 		return (ire);
1895 	}
1896 
1897 	ASSERT(pire == NULL || *pire == NULL);
1898 	return (ire);
1899 }
1900 
1901 /*
1902  * Delete the IRE cache for the gateway and all IRE caches whose
1903  * ire_gateway_addr_v6 points to this gateway, and allow them to
1904  * be created on demand by ip_newroute_v6.
1905  */
1906 void
1907 ire_clookup_delete_cache_gw_v6(const in6_addr_t *addr, zoneid_t zoneid,
1908 	ip_stack_t *ipst)
1909 {
1910 	irb_t *irb;
1911 	ire_t *ire;
1912 
1913 	irb = &ipst->ips_ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr,
1914 	    ipst->ips_ip6_cache_table_size)];
1915 	IRB_REFHOLD(irb);
1916 	for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
1917 		if (ire->ire_marks & IRE_MARK_CONDEMNED)
1918 			continue;
1919 
1920 		ASSERT(IN6_ARE_ADDR_EQUAL(&ire->ire_mask_v6, &ipv6_all_ones));
1921 		if (ire_match_args_v6(ire, addr, &ire->ire_mask_v6, 0,
1922 		    IRE_CACHE, NULL, zoneid, 0, NULL, MATCH_IRE_TYPE)) {
1923 			ire_delete(ire);
1924 		}
1925 	}
1926 	IRB_REFRELE(irb);
1927 
1928 	ire_walk_v6(ire_delete_cache_gw_v6, (char *)addr, zoneid, ipst);
1929 }
1930 
1931 /*
1932  * Looks up cache table for a route.
1933  * specific lookup can be indicated by
1934  * passing the MATCH_* flags and the
1935  * necessary parameters.
1936  */
1937 ire_t *
1938 ire_ctable_lookup_v6(const in6_addr_t *addr, const in6_addr_t *gateway,
1939     int type, const ipif_t *ipif, zoneid_t zoneid, const ts_label_t *tsl,
1940     int flags, ip_stack_t *ipst)
1941 {
1942 	ire_t *ire;
1943 	irb_t *irb_ptr;
1944 	ASSERT(addr != NULL);
1945 	ASSERT((!(flags & MATCH_IRE_GW)) || gateway != NULL);
1946 
1947 	/*
1948 	 * ire_match_args_v6() will dereference ipif MATCH_IRE_SRC or
1949 	 * MATCH_IRE_ILL is set.
1950 	 */
1951 	if ((flags & (MATCH_IRE_SRC |  MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) &&
1952 	    (ipif == NULL))
1953 		return (NULL);
1954 
1955 	irb_ptr = &ipst->ips_ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr,
1956 	    ipst->ips_ip6_cache_table_size)];
1957 	rw_enter(&irb_ptr->irb_lock, RW_READER);
1958 	for (ire = irb_ptr->irb_ire; ire; ire = ire->ire_next) {
1959 		if (ire->ire_marks & IRE_MARK_CONDEMNED)
1960 			continue;
1961 
1962 		ASSERT(IN6_ARE_ADDR_EQUAL(&ire->ire_mask_v6, &ipv6_all_ones));
1963 		if (ire_match_args_v6(ire, addr, &ire->ire_mask_v6, gateway,
1964 		    type, ipif, zoneid, 0, tsl, flags)) {
1965 			IRE_REFHOLD(ire);
1966 			rw_exit(&irb_ptr->irb_lock);
1967 			return (ire);
1968 		}
1969 	}
1970 	rw_exit(&irb_ptr->irb_lock);
1971 	return (NULL);
1972 }
1973 
1974 /*
1975  * Lookup cache. Don't return IRE_MARK_HIDDEN entries. Callers
1976  * should use ire_ctable_lookup with MATCH_IRE_MARK_HIDDEN to get
1977  * to the hidden ones.
1978  *
1979  * In general the zoneid has to match (where ALL_ZONES match all of them).
1980  * But for IRE_LOCAL we also need to handle the case where L2 should
1981  * conceptually loop back the packet. This is necessary since neither
1982  * Ethernet drivers nor Ethernet hardware loops back packets sent to their
1983  * own MAC address. This loopback is needed when the normal
1984  * routes (ignoring IREs with different zoneids) would send out the packet on
1985  * the same ill (or ill group) as the ill with which this IRE_LOCAL is
1986  * associated.
1987  *
1988  * Earlier versions of this code always matched an IRE_LOCAL independently of
1989  * the zoneid. We preserve that earlier behavior when
1990  * ip_restrict_interzone_loopback is turned off.
1991  */
1992 ire_t *
1993 ire_cache_lookup_v6(const in6_addr_t *addr, zoneid_t zoneid,
1994     const ts_label_t *tsl, ip_stack_t *ipst)
1995 {
1996 	irb_t *irb_ptr;
1997 	ire_t *ire;
1998 
1999 	irb_ptr = &ipst->ips_ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr,
2000 	    ipst->ips_ip6_cache_table_size)];
2001 	rw_enter(&irb_ptr->irb_lock, RW_READER);
2002 	for (ire = irb_ptr->irb_ire; ire; ire = ire->ire_next) {
2003 		if (ire->ire_marks & (IRE_MARK_CONDEMNED|IRE_MARK_HIDDEN))
2004 			continue;
2005 		if (IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, addr)) {
2006 			/*
2007 			 * Finally, check if the security policy has any
2008 			 * restriction on using this route for the specified
2009 			 * message.
2010 			 */
2011 			if (tsl != NULL &&
2012 			    ire->ire_gw_secattr != NULL &&
2013 			    tsol_ire_match_gwattr(ire, tsl) != 0) {
2014 				continue;
2015 			}
2016 
2017 			if (zoneid == ALL_ZONES || ire->ire_zoneid == zoneid ||
2018 			    ire->ire_zoneid == ALL_ZONES) {
2019 				IRE_REFHOLD(ire);
2020 				rw_exit(&irb_ptr->irb_lock);
2021 				return (ire);
2022 			}
2023 
2024 			if (ire->ire_type == IRE_LOCAL) {
2025 				if (ipst->ips_ip_restrict_interzone_loopback &&
2026 				    !ire_local_ok_across_zones(ire, zoneid,
2027 				    (void *)addr, tsl, ipst))
2028 					continue;
2029 
2030 				IRE_REFHOLD(ire);
2031 				rw_exit(&irb_ptr->irb_lock);
2032 				return (ire);
2033 			}
2034 		}
2035 	}
2036 	rw_exit(&irb_ptr->irb_lock);
2037 	return (NULL);
2038 }
2039 
2040 /*
2041  * Locate the interface ire that is tied to the cache ire 'cire' via
2042  * cire->ire_ihandle.
2043  *
2044  * We are trying to create the cache ire for an onlink destn. or
2045  * gateway in 'cire'. We are called from ire_add_v6() in the IRE_IF_RESOLVER
2046  * case for xresolv interfaces, after the ire has come back from
2047  * an external resolver.
2048  */
2049 static ire_t *
2050 ire_ihandle_lookup_onlink_v6(ire_t *cire)
2051 {
2052 	ire_t	*ire;
2053 	int	match_flags;
2054 	int	i;
2055 	int	j;
2056 	irb_t	*irb_ptr;
2057 	ip_stack_t	*ipst = cire->ire_ipst;
2058 
2059 	ASSERT(cire != NULL);
2060 
2061 	match_flags =  MATCH_IRE_TYPE | MATCH_IRE_IHANDLE | MATCH_IRE_MASK;
2062 	/*
2063 	 * We know that the mask of the interface ire equals cire->ire_cmask.
2064 	 * (When ip_newroute_v6() created 'cire' for an on-link destn.
2065 	 * it set its cmask from the interface ire's mask)
2066 	 */
2067 	ire = ire_ftable_lookup_v6(&cire->ire_addr_v6, &cire->ire_cmask_v6,
2068 	    NULL, IRE_INTERFACE, NULL, NULL, ALL_ZONES, cire->ire_ihandle,
2069 	    NULL, match_flags, ipst);
2070 	if (ire != NULL)
2071 		return (ire);
2072 	/*
2073 	 * If we didn't find an interface ire above, we can't declare failure.
2074 	 * For backwards compatibility, we need to support prefix routes
2075 	 * pointing to next hop gateways that are not on-link.
2076 	 *
2077 	 * In the resolver/noresolver case, ip_newroute_v6() thinks
2078 	 * it is creating the cache ire for an onlink destination in 'cire'.
2079 	 * But 'cire' is not actually onlink, because ire_ftable_lookup_v6()
2080 	 * cheated it, by doing ire_route_lookup_v6() twice and returning an
2081 	 * interface ire.
2082 	 *
2083 	 * Eg. default	-	gw1			(line 1)
2084 	 *	gw1	-	gw2			(line 2)
2085 	 *	gw2	-	hme0			(line 3)
2086 	 *
2087 	 * In the above example, ip_newroute_v6() tried to create the cache ire
2088 	 * 'cire' for gw1, based on the interface route in line 3. The
2089 	 * ire_ftable_lookup_v6() above fails, because there is
2090 	 * no interface route to reach gw1. (it is gw2). We fall thru below.
2091 	 *
2092 	 * Do a brute force search based on the ihandle in a subset of the
2093 	 * forwarding tables, corresponding to cire->ire_cmask_v6. Otherwise
2094 	 * things become very complex, since we don't have 'pire' in this
2095 	 * case. (Also note that this method is not possible in the offlink
2096 	 * case because we don't know the mask)
2097 	 */
2098 	i = ip_mask_to_plen_v6(&cire->ire_cmask_v6);
2099 	if ((ipst->ips_ip_forwarding_table_v6[i]) == NULL)
2100 		return (NULL);
2101 	for (j = 0; j < ipst->ips_ip6_ftable_hash_size; j++) {
2102 		irb_ptr = &ipst->ips_ip_forwarding_table_v6[i][j];
2103 		rw_enter(&irb_ptr->irb_lock, RW_READER);
2104 		for (ire = irb_ptr->irb_ire; ire != NULL;
2105 		    ire = ire->ire_next) {
2106 			if (ire->ire_marks & IRE_MARK_CONDEMNED)
2107 				continue;
2108 			if ((ire->ire_type & IRE_INTERFACE) &&
2109 			    (ire->ire_ihandle == cire->ire_ihandle)) {
2110 				IRE_REFHOLD(ire);
2111 				rw_exit(&irb_ptr->irb_lock);
2112 				return (ire);
2113 			}
2114 		}
2115 		rw_exit(&irb_ptr->irb_lock);
2116 	}
2117 	return (NULL);
2118 }
2119 
2120 
2121 /*
2122  * Locate the interface ire that is tied to the cache ire 'cire' via
2123  * cire->ire_ihandle.
2124  *
2125  * We are trying to create the cache ire for an offlink destn based
2126  * on the cache ire of the gateway in 'cire'. 'pire' is the prefix ire
2127  * as found by ip_newroute_v6(). We are called from ip_newroute_v6() in
2128  * the IRE_CACHE case.
2129  */
2130 ire_t *
2131 ire_ihandle_lookup_offlink_v6(ire_t *cire, ire_t *pire)
2132 {
2133 	ire_t	*ire;
2134 	int	match_flags;
2135 	in6_addr_t	gw_addr;
2136 	ipif_t		*gw_ipif;
2137 	ip_stack_t	*ipst = cire->ire_ipst;
2138 
2139 	ASSERT(cire != NULL && pire != NULL);
2140 
2141 	match_flags =  MATCH_IRE_TYPE | MATCH_IRE_IHANDLE | MATCH_IRE_MASK;
2142 	/*
2143 	 * ip_newroute_v6 calls ire_ftable_lookup with MATCH_IRE_ILL only
2144 	 * for on-link hosts. We should never be here for onlink.
2145 	 * Thus, use MATCH_IRE_ILL_GROUP.
2146 	 */
2147 	if (pire->ire_ipif != NULL)
2148 		match_flags |= MATCH_IRE_ILL_GROUP;
2149 	/*
2150 	 * We know that the mask of the interface ire equals cire->ire_cmask.
2151 	 * (When ip_newroute_v6() created 'cire' for an on-link destn. it set
2152 	 * its cmask from the interface ire's mask)
2153 	 */
2154 	ire = ire_ftable_lookup_v6(&cire->ire_addr_v6, &cire->ire_cmask_v6, 0,
2155 	    IRE_INTERFACE, pire->ire_ipif, NULL, ALL_ZONES, cire->ire_ihandle,
2156 	    NULL, match_flags, ipst);
2157 	if (ire != NULL)
2158 		return (ire);
2159 	/*
2160 	 * If we didn't find an interface ire above, we can't declare failure.
2161 	 * For backwards compatibility, we need to support prefix routes
2162 	 * pointing to next hop gateways that are not on-link.
2163 	 *
2164 	 * Assume we are trying to ping some offlink destn, and we have the
2165 	 * routing table below.
2166 	 *
2167 	 * Eg.	default	- gw1		<--- pire	(line 1)
2168 	 *	gw1	- gw2				(line 2)
2169 	 *	gw2	- hme0				(line 3)
2170 	 *
2171 	 * If we already have a cache ire for gw1 in 'cire', the
2172 	 * ire_ftable_lookup_v6 above would have failed, since there is no
2173 	 * interface ire to reach gw1. We will fallthru below.
2174 	 *
2175 	 * Here we duplicate the steps that ire_ftable_lookup_v6() did in
2176 	 * getting 'cire' from 'pire', in the MATCH_IRE_RECURSIVE case.
2177 	 * The differences are the following
2178 	 * i.   We want the interface ire only, so we call
2179 	 *	ire_ftable_lookup_v6() instead of ire_route_lookup_v6()
2180 	 * ii.  We look for only prefix routes in the 1st call below.
2181 	 * ii.  We want to match on the ihandle in the 2nd call below.
2182 	 */
2183 	match_flags =  MATCH_IRE_TYPE;
2184 	if (pire->ire_ipif != NULL)
2185 		match_flags |= MATCH_IRE_ILL_GROUP;
2186 
2187 	mutex_enter(&pire->ire_lock);
2188 	gw_addr = pire->ire_gateway_addr_v6;
2189 	mutex_exit(&pire->ire_lock);
2190 	ire = ire_ftable_lookup_v6(&gw_addr, 0, 0, IRE_OFFSUBNET,
2191 	    pire->ire_ipif, NULL, ALL_ZONES, 0, NULL, match_flags, ipst);
2192 	if (ire == NULL)
2193 		return (NULL);
2194 	/*
2195 	 * At this point 'ire' corresponds to the entry shown in line 2.
2196 	 * gw_addr is 'gw2' in the example above.
2197 	 */
2198 	mutex_enter(&ire->ire_lock);
2199 	gw_addr = ire->ire_gateway_addr_v6;
2200 	mutex_exit(&ire->ire_lock);
2201 	gw_ipif = ire->ire_ipif;
2202 	ire_refrele(ire);
2203 
2204 	match_flags |= MATCH_IRE_IHANDLE;
2205 	ire = ire_ftable_lookup_v6(&gw_addr, 0, 0, IRE_INTERFACE,
2206 	    gw_ipif, NULL, ALL_ZONES, cire->ire_ihandle,
2207 	    NULL, match_flags, ipst);
2208 	return (ire);
2209 }
2210 
2211 /*
2212  * Return the IRE_LOOPBACK, IRE_IF_RESOLVER or IRE_IF_NORESOLVER
2213  * ire associated with the specified ipif.
2214  *
2215  * This might occasionally be called when IPIF_UP is not set since
2216  * the IPV6_MULTICAST_IF as well as creating interface routes
2217  * allows specifying a down ipif (ipif_lookup* match ipifs that are down).
2218  *
2219  * Note that if IPIF_NOLOCAL, IPIF_NOXMIT, or IPIF_DEPRECATED is set on
2220  * the ipif this routine might return NULL.
2221  * (Sometimes called as writer though not required by this function.)
2222  */
2223 ire_t *
2224 ipif_to_ire_v6(const ipif_t *ipif)
2225 {
2226 	ire_t	*ire;
2227 	ip_stack_t	*ipst = ipif->ipif_ill->ill_ipst;
2228 
2229 	ASSERT(ipif->ipif_isv6);
2230 	if (ipif->ipif_ire_type == IRE_LOOPBACK) {
2231 		ire = ire_ctable_lookup_v6(&ipif->ipif_v6lcl_addr, NULL,
2232 		    IRE_LOOPBACK, ipif, ALL_ZONES, NULL,
2233 		    (MATCH_IRE_TYPE | MATCH_IRE_IPIF), ipst);
2234 	} else if (ipif->ipif_flags & IPIF_POINTOPOINT) {
2235 		/* In this case we need to lookup destination address. */
2236 		ire = ire_ftable_lookup_v6(&ipif->ipif_v6pp_dst_addr,
2237 		    &ipv6_all_ones, NULL, IRE_INTERFACE, ipif, NULL, ALL_ZONES,
2238 		    0, NULL, (MATCH_IRE_TYPE | MATCH_IRE_IPIF |
2239 		    MATCH_IRE_MASK), ipst);
2240 	} else {
2241 		ire = ire_ftable_lookup_v6(&ipif->ipif_v6subnet,
2242 		    &ipif->ipif_v6net_mask, NULL, IRE_INTERFACE, ipif, NULL,
2243 		    ALL_ZONES, 0, NULL, (MATCH_IRE_TYPE | MATCH_IRE_IPIF |
2244 		    MATCH_IRE_MASK), ipst);
2245 	}
2246 	return (ire);
2247 }
2248 
2249 /*
2250  * Return B_TRUE if a multirt route is resolvable
2251  * (or if no route is resolved yet), B_FALSE otherwise.
2252  * This only works in the global zone.
2253  */
2254 boolean_t
2255 ire_multirt_need_resolve_v6(const in6_addr_t *v6dstp, const ts_label_t *tsl,
2256     ip_stack_t *ipst)
2257 {
2258 	ire_t	*first_fire;
2259 	ire_t	*first_cire;
2260 	ire_t	*fire;
2261 	ire_t	*cire;
2262 	irb_t	*firb;
2263 	irb_t	*cirb;
2264 	int	unres_cnt = 0;
2265 	boolean_t resolvable = B_FALSE;
2266 
2267 	/* Retrieve the first IRE_HOST that matches the destination */
2268 	first_fire = ire_ftable_lookup_v6(v6dstp, &ipv6_all_ones, 0, IRE_HOST,
2269 	    NULL, NULL, ALL_ZONES, 0, tsl, MATCH_IRE_MASK | MATCH_IRE_TYPE |
2270 	    MATCH_IRE_SECATTR, ipst);
2271 
2272 	/* No route at all */
2273 	if (first_fire == NULL) {
2274 		return (B_TRUE);
2275 	}
2276 
2277 	firb = first_fire->ire_bucket;
2278 	ASSERT(firb);
2279 
2280 	/* Retrieve the first IRE_CACHE ire for that destination. */
2281 	first_cire = ire_cache_lookup_v6(v6dstp, GLOBAL_ZONEID, tsl, ipst);
2282 
2283 	/* No resolved route. */
2284 	if (first_cire == NULL) {
2285 		ire_refrele(first_fire);
2286 		return (B_TRUE);
2287 	}
2288 
2289 	/* At least one route is resolved. */
2290 
2291 	cirb = first_cire->ire_bucket;
2292 	ASSERT(cirb);
2293 
2294 	/* Count the number of routes to that dest that are declared. */
2295 	IRB_REFHOLD(firb);
2296 	for (fire = first_fire; fire != NULL; fire = fire->ire_next) {
2297 		if (!(fire->ire_flags & RTF_MULTIRT))
2298 			continue;
2299 		if (!IN6_ARE_ADDR_EQUAL(&fire->ire_addr_v6, v6dstp))
2300 			continue;
2301 		unres_cnt++;
2302 	}
2303 	IRB_REFRELE(firb);
2304 
2305 
2306 	/* Then subtract the number of routes to that dst that are resolved */
2307 	IRB_REFHOLD(cirb);
2308 	for (cire = first_cire; cire != NULL; cire = cire->ire_next) {
2309 		if (!(cire->ire_flags & RTF_MULTIRT))
2310 			continue;
2311 		if (!IN6_ARE_ADDR_EQUAL(&cire->ire_addr_v6, v6dstp))
2312 			continue;
2313 		if (cire->ire_marks & (IRE_MARK_CONDEMNED|IRE_MARK_HIDDEN))
2314 			continue;
2315 		unres_cnt--;
2316 	}
2317 	IRB_REFRELE(cirb);
2318 
2319 	/* At least one route is unresolved; search for a resolvable route. */
2320 	if (unres_cnt > 0)
2321 		resolvable = ire_multirt_lookup_v6(&first_cire, &first_fire,
2322 		    MULTIRT_USESTAMP|MULTIRT_CACHEGW, tsl, ipst);
2323 
2324 	if (first_fire)
2325 		ire_refrele(first_fire);
2326 
2327 	if (first_cire)
2328 		ire_refrele(first_cire);
2329 
2330 	return (resolvable);
2331 }
2332 
2333 
2334 /*
2335  * Return B_TRUE and update *ire_arg and *fire_arg
2336  * if at least one resolvable route is found.
2337  * Return B_FALSE otherwise (all routes are resolved or
2338  * the remaining unresolved routes are all unresolvable).
2339  * This only works in the global zone.
2340  */
2341 boolean_t
2342 ire_multirt_lookup_v6(ire_t **ire_arg, ire_t **fire_arg, uint32_t flags,
2343     const ts_label_t *tsl, ip_stack_t *ipst)
2344 {
2345 	clock_t	delta;
2346 	ire_t	*best_fire = NULL;
2347 	ire_t	*best_cire = NULL;
2348 	ire_t	*first_fire;
2349 	ire_t	*first_cire;
2350 	ire_t	*fire;
2351 	ire_t	*cire;
2352 	irb_t	*firb = NULL;
2353 	irb_t	*cirb = NULL;
2354 	ire_t	*gw_ire;
2355 	boolean_t	already_resolved;
2356 	boolean_t	res;
2357 	in6_addr_t	v6dst;
2358 	in6_addr_t	v6gw;
2359 
2360 	ip2dbg(("ire_multirt_lookup_v6: *ire_arg %p, *fire_arg %p, "
2361 	    "flags %04x\n", (void *)*ire_arg, (void *)*fire_arg, flags));
2362 
2363 	ASSERT(ire_arg);
2364 	ASSERT(fire_arg);
2365 
2366 	/* Not an IRE_HOST ire; give up. */
2367 	if ((*fire_arg == NULL) ||
2368 	    ((*fire_arg)->ire_type != IRE_HOST)) {
2369 		return (B_FALSE);
2370 	}
2371 
2372 	/* This is the first IRE_HOST ire for that destination. */
2373 	first_fire = *fire_arg;
2374 	firb = first_fire->ire_bucket;
2375 	ASSERT(firb);
2376 
2377 	mutex_enter(&first_fire->ire_lock);
2378 	v6dst = first_fire->ire_addr_v6;
2379 	mutex_exit(&first_fire->ire_lock);
2380 
2381 	ip2dbg(("ire_multirt_lookup_v6: dst %08x\n",
2382 	    ntohl(V4_PART_OF_V6(v6dst))));
2383 
2384 	/*
2385 	 * Retrieve the first IRE_CACHE ire for that destination;
2386 	 * if we don't find one, no route for that dest is
2387 	 * resolved yet.
2388 	 */
2389 	first_cire = ire_cache_lookup_v6(&v6dst, GLOBAL_ZONEID, tsl, ipst);
2390 	if (first_cire) {
2391 		cirb = first_cire->ire_bucket;
2392 	}
2393 
2394 	ip2dbg(("ire_multirt_lookup_v6: first_cire %p\n", (void *)first_cire));
2395 
2396 	/*
2397 	 * Search for a resolvable route, giving the top priority
2398 	 * to routes that can be resolved without any call to the resolver.
2399 	 */
2400 	IRB_REFHOLD(firb);
2401 
2402 	if (!IN6_IS_ADDR_MULTICAST(&v6dst)) {
2403 		/*
2404 		 * For all multiroute IRE_HOST ires for that destination,
2405 		 * check if the route via the IRE_HOST's gateway is
2406 		 * resolved yet.
2407 		 */
2408 		for (fire = first_fire; fire != NULL; fire = fire->ire_next) {
2409 
2410 			if (!(fire->ire_flags & RTF_MULTIRT))
2411 				continue;
2412 			if (!IN6_ARE_ADDR_EQUAL(&fire->ire_addr_v6, &v6dst))
2413 				continue;
2414 
2415 			if (fire->ire_gw_secattr != NULL &&
2416 			    tsol_ire_match_gwattr(fire, tsl) != 0) {
2417 				continue;
2418 			}
2419 
2420 			mutex_enter(&fire->ire_lock);
2421 			v6gw = fire->ire_gateway_addr_v6;
2422 			mutex_exit(&fire->ire_lock);
2423 
2424 			ip2dbg(("ire_multirt_lookup_v6: fire %p, "
2425 			    "ire_addr %08x, ire_gateway_addr %08x\n",
2426 			    (void *)fire,
2427 			    ntohl(V4_PART_OF_V6(fire->ire_addr_v6)),
2428 			    ntohl(V4_PART_OF_V6(v6gw))));
2429 
2430 			already_resolved = B_FALSE;
2431 
2432 			if (first_cire) {
2433 				ASSERT(cirb);
2434 
2435 				IRB_REFHOLD(cirb);
2436 				/*
2437 				 * For all IRE_CACHE ires for that
2438 				 * destination.
2439 				 */
2440 				for (cire = first_cire;
2441 				    cire != NULL;
2442 				    cire = cire->ire_next) {
2443 
2444 					if (!(cire->ire_flags & RTF_MULTIRT))
2445 						continue;
2446 					if (!IN6_ARE_ADDR_EQUAL(
2447 					    &cire->ire_addr_v6, &v6dst))
2448 						continue;
2449 					if (cire->ire_marks &
2450 					    (IRE_MARK_CONDEMNED|
2451 					    IRE_MARK_HIDDEN))
2452 						continue;
2453 
2454 					if (cire->ire_gw_secattr != NULL &&
2455 					    tsol_ire_match_gwattr(cire,
2456 					    tsl) != 0) {
2457 						continue;
2458 					}
2459 
2460 					/*
2461 					 * Check if the IRE_CACHE's gateway
2462 					 * matches the IRE_HOST's gateway.
2463 					 */
2464 					if (IN6_ARE_ADDR_EQUAL(
2465 					    &cire->ire_gateway_addr_v6,
2466 					    &v6gw)) {
2467 						already_resolved = B_TRUE;
2468 						break;
2469 					}
2470 				}
2471 				IRB_REFRELE(cirb);
2472 			}
2473 
2474 			/*
2475 			 * This route is already resolved;
2476 			 * proceed with next one.
2477 			 */
2478 			if (already_resolved) {
2479 				ip2dbg(("ire_multirt_lookup_v6: found cire %p, "
2480 				    "already resolved\n", (void *)cire));
2481 				continue;
2482 			}
2483 
2484 			/*
2485 			 * The route is unresolved; is it actually
2486 			 * resolvable, i.e. is there a cache or a resolver
2487 			 * for the gateway?
2488 			 */
2489 			gw_ire = ire_route_lookup_v6(&v6gw, 0, 0, 0, NULL, NULL,
2490 			    ALL_ZONES, tsl, MATCH_IRE_RECURSIVE |
2491 			    MATCH_IRE_SECATTR, ipst);
2492 
2493 			ip2dbg(("ire_multirt_lookup_v6: looked up gw_ire %p\n",
2494 			    (void *)gw_ire));
2495 
2496 			/*
2497 			 * This route can be resolved without any call to the
2498 			 * resolver; if the MULTIRT_CACHEGW flag is set,
2499 			 * give the top priority to this ire and exit the
2500 			 * loop.
2501 			 * This occurs when an resolver reply is processed
2502 			 * through ip_wput_nondata()
2503 			 */
2504 			if ((flags & MULTIRT_CACHEGW) &&
2505 			    (gw_ire != NULL) &&
2506 			    (gw_ire->ire_type & IRE_CACHETABLE)) {
2507 				/*
2508 				 * Release the resolver associated to the
2509 				 * previous candidate best ire, if any.
2510 				 */
2511 				if (best_cire) {
2512 					ire_refrele(best_cire);
2513 					ASSERT(best_fire);
2514 				}
2515 
2516 				best_fire = fire;
2517 				best_cire = gw_ire;
2518 
2519 				ip2dbg(("ire_multirt_lookup_v6: found top prio "
2520 				    "best_fire %p, best_cire %p\n",
2521 				    (void *)best_fire, (void *)best_cire));
2522 				break;
2523 			}
2524 
2525 			/*
2526 			 * Compute the time elapsed since our preceding
2527 			 * attempt to  resolve that route.
2528 			 * If the MULTIRT_USESTAMP flag is set, we take that
2529 			 * route into account only if this time interval
2530 			 * exceeds ip_multirt_resolution_interval;
2531 			 * this prevents us from attempting to resolve a
2532 			 * broken route upon each sending of a packet.
2533 			 */
2534 			delta = lbolt - fire->ire_last_used_time;
2535 			delta = TICK_TO_MSEC(delta);
2536 
2537 			res = (boolean_t)
2538 			    ((delta > ipst->
2539 			    ips_ip_multirt_resolution_interval) ||
2540 			    (!(flags & MULTIRT_USESTAMP)));
2541 
2542 			ip2dbg(("ire_multirt_lookup_v6: fire %p, delta %lu, "
2543 			    "res %d\n",
2544 			    (void *)fire, delta, res));
2545 
2546 			if (res) {
2547 				/*
2548 				 * A resolver exists for the gateway: save
2549 				 * the current IRE_HOST ire as a candidate
2550 				 * best ire. If we later discover that a
2551 				 * top priority ire exists (i.e. no need to
2552 				 * call the resolver), then this new ire
2553 				 * will be preferred to the current one.
2554 				 */
2555 				if (gw_ire != NULL) {
2556 					if (best_fire == NULL) {
2557 						ASSERT(best_cire == NULL);
2558 
2559 						best_fire = fire;
2560 						best_cire = gw_ire;
2561 
2562 						ip2dbg(("ire_multirt_lookup_v6:"
2563 						    "found candidate "
2564 						    "best_fire %p, "
2565 						    "best_cire %p\n",
2566 						    (void *)best_fire,
2567 						    (void *)best_cire));
2568 
2569 						/*
2570 						 * If MULTIRT_CACHEGW is not
2571 						 * set, we ignore the top
2572 						 * priority ires that can
2573 						 * be resolved without any
2574 						 * call to the resolver;
2575 						 * In that case, there is
2576 						 * actually no need
2577 						 * to continue the loop.
2578 						 */
2579 						if (!(flags &
2580 						    MULTIRT_CACHEGW)) {
2581 							break;
2582 						}
2583 						continue;
2584 					}
2585 				} else {
2586 					/*
2587 					 * No resolver for the gateway: the
2588 					 * route is not resolvable.
2589 					 * If the MULTIRT_SETSTAMP flag is
2590 					 * set, we stamp the IRE_HOST ire,
2591 					 * so we will not select it again
2592 					 * during this resolution interval.
2593 					 */
2594 					if (flags & MULTIRT_SETSTAMP)
2595 						fire->ire_last_used_time =
2596 						    lbolt;
2597 				}
2598 			}
2599 
2600 			if (gw_ire != NULL)
2601 				ire_refrele(gw_ire);
2602 		}
2603 	} else { /* IN6_IS_ADDR_MULTICAST(&v6dst) */
2604 
2605 		for (fire = first_fire;
2606 		    fire != NULL;
2607 		    fire = fire->ire_next) {
2608 
2609 			if (!(fire->ire_flags & RTF_MULTIRT))
2610 				continue;
2611 			if (!IN6_ARE_ADDR_EQUAL(&fire->ire_addr_v6, &v6dst))
2612 				continue;
2613 
2614 			if (fire->ire_gw_secattr != NULL &&
2615 			    tsol_ire_match_gwattr(fire, tsl) != 0) {
2616 				continue;
2617 			}
2618 
2619 			already_resolved = B_FALSE;
2620 
2621 			mutex_enter(&fire->ire_lock);
2622 			v6gw = fire->ire_gateway_addr_v6;
2623 			mutex_exit(&fire->ire_lock);
2624 
2625 			gw_ire = ire_ftable_lookup_v6(&v6gw, 0, 0,
2626 			    IRE_INTERFACE, NULL, NULL, ALL_ZONES, 0, tsl,
2627 			    MATCH_IRE_RECURSIVE | MATCH_IRE_TYPE |
2628 			    MATCH_IRE_SECATTR, ipst);
2629 
2630 			/* No resolver for the gateway; we skip this ire. */
2631 			if (gw_ire == NULL) {
2632 				continue;
2633 			}
2634 
2635 			if (first_cire) {
2636 
2637 				IRB_REFHOLD(cirb);
2638 				/*
2639 				 * For all IRE_CACHE ires for that
2640 				 * destination.
2641 				 */
2642 				for (cire = first_cire;
2643 				    cire != NULL;
2644 				    cire = cire->ire_next) {
2645 
2646 					if (!(cire->ire_flags & RTF_MULTIRT))
2647 						continue;
2648 					if (!IN6_ARE_ADDR_EQUAL(
2649 					    &cire->ire_addr_v6, &v6dst))
2650 						continue;
2651 					if (cire->ire_marks &
2652 					    (IRE_MARK_CONDEMNED|
2653 					    IRE_MARK_HIDDEN))
2654 						continue;
2655 
2656 					if (cire->ire_gw_secattr != NULL &&
2657 					    tsol_ire_match_gwattr(cire,
2658 					    tsl) != 0) {
2659 						continue;
2660 					}
2661 
2662 					/*
2663 					 * Cache entries are linked to the
2664 					 * parent routes using the parent handle
2665 					 * (ire_phandle). If no cache entry has
2666 					 * the same handle as fire, fire is
2667 					 * still unresolved.
2668 					 */
2669 					ASSERT(cire->ire_phandle != 0);
2670 					if (cire->ire_phandle ==
2671 					    fire->ire_phandle) {
2672 						already_resolved = B_TRUE;
2673 						break;
2674 					}
2675 				}
2676 				IRB_REFRELE(cirb);
2677 			}
2678 
2679 			/*
2680 			 * This route is already resolved; proceed with
2681 			 * next one.
2682 			 */
2683 			if (already_resolved) {
2684 				ire_refrele(gw_ire);
2685 				continue;
2686 			}
2687 
2688 			/*
2689 			 * Compute the time elapsed since our preceding
2690 			 * attempt to resolve that route.
2691 			 * If the MULTIRT_USESTAMP flag is set, we take
2692 			 * that route into account only if this time
2693 			 * interval exceeds ip_multirt_resolution_interval;
2694 			 * this prevents us from attempting to resolve a
2695 			 * broken route upon each sending of a packet.
2696 			 */
2697 			delta = lbolt - fire->ire_last_used_time;
2698 			delta = TICK_TO_MSEC(delta);
2699 
2700 			res = (boolean_t)
2701 			    ((delta > ipst->
2702 			    ips_ip_multirt_resolution_interval) ||
2703 			    (!(flags & MULTIRT_USESTAMP)));
2704 
2705 			ip3dbg(("ire_multirt_lookup_v6: fire %p, delta %lx, "
2706 			    "flags %04x, res %d\n",
2707 			    (void *)fire, delta, flags, res));
2708 
2709 			if (res) {
2710 				if (best_cire) {
2711 					/*
2712 					 * Release the resolver associated
2713 					 * to the preceding candidate best
2714 					 * ire, if any.
2715 					 */
2716 					ire_refrele(best_cire);
2717 					ASSERT(best_fire);
2718 				}
2719 				best_fire = fire;
2720 				best_cire = gw_ire;
2721 				continue;
2722 			}
2723 
2724 			ire_refrele(gw_ire);
2725 		}
2726 	}
2727 
2728 	if (best_fire) {
2729 		IRE_REFHOLD(best_fire);
2730 	}
2731 	IRB_REFRELE(firb);
2732 
2733 	/* Release the first IRE_CACHE we initially looked up, if any. */
2734 	if (first_cire)
2735 		ire_refrele(first_cire);
2736 
2737 	/* Found a resolvable route. */
2738 	if (best_fire) {
2739 		ASSERT(best_cire);
2740 
2741 		if (*fire_arg)
2742 			ire_refrele(*fire_arg);
2743 		if (*ire_arg)
2744 			ire_refrele(*ire_arg);
2745 
2746 		/*
2747 		 * Update the passed arguments with the
2748 		 * resolvable multirt route we found
2749 		 */
2750 		*fire_arg = best_fire;
2751 		*ire_arg = best_cire;
2752 
2753 		ip2dbg(("ire_multirt_lookup_v6: returning B_TRUE, "
2754 		    "*fire_arg %p, *ire_arg %p\n",
2755 		    (void *)best_fire, (void *)best_cire));
2756 
2757 		return (B_TRUE);
2758 	}
2759 
2760 	ASSERT(best_cire == NULL);
2761 
2762 	ip2dbg(("ire_multirt_lookup_v6: returning B_FALSE, *fire_arg %p, "
2763 	    "*ire_arg %p\n",
2764 	    (void *)*fire_arg, (void *)*ire_arg));
2765 
2766 	/* No resolvable route. */
2767 	return (B_FALSE);
2768 }
2769 
2770 
2771 /*
2772  * Find an IRE_OFFSUBNET IRE entry for the multicast address 'v6dstp'
2773  * that goes through 'ipif'. As a fallback, a route that goes through
2774  * ipif->ipif_ill can be returned.
2775  */
2776 ire_t *
2777 ipif_lookup_multi_ire_v6(ipif_t *ipif, const in6_addr_t *v6dstp)
2778 {
2779 	ire_t	*ire;
2780 	ire_t	*save_ire = NULL;
2781 	ire_t   *gw_ire;
2782 	irb_t   *irb;
2783 	in6_addr_t v6gw;
2784 	int	match_flags = MATCH_IRE_TYPE | MATCH_IRE_ILL;
2785 	ip_stack_t	*ipst = ipif->ipif_ill->ill_ipst;
2786 
2787 	ire = ire_ftable_lookup_v6(v6dstp, 0, 0, 0, NULL, NULL, ALL_ZONES, 0,
2788 	    NULL, MATCH_IRE_DEFAULT, ipst);
2789 
2790 	if (ire == NULL)
2791 		return (NULL);
2792 
2793 	irb = ire->ire_bucket;
2794 	ASSERT(irb);
2795 
2796 	IRB_REFHOLD(irb);
2797 	ire_refrele(ire);
2798 	for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
2799 		if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6dstp) ||
2800 		    (ipif->ipif_zoneid != ire->ire_zoneid &&
2801 		    ire->ire_zoneid != ALL_ZONES)) {
2802 			continue;
2803 		}
2804 
2805 		switch (ire->ire_type) {
2806 		case IRE_DEFAULT:
2807 		case IRE_PREFIX:
2808 		case IRE_HOST:
2809 			mutex_enter(&ire->ire_lock);
2810 			v6gw = ire->ire_gateway_addr_v6;
2811 			mutex_exit(&ire->ire_lock);
2812 			gw_ire = ire_ftable_lookup_v6(&v6gw, 0, 0,
2813 			    IRE_INTERFACE, ipif, NULL, ALL_ZONES, 0,
2814 			    NULL, match_flags, ipst);
2815 
2816 			if (gw_ire != NULL) {
2817 				if (save_ire != NULL) {
2818 					ire_refrele(save_ire);
2819 				}
2820 				IRE_REFHOLD(ire);
2821 				if (gw_ire->ire_ipif == ipif) {
2822 					ire_refrele(gw_ire);
2823 
2824 					IRB_REFRELE(irb);
2825 					return (ire);
2826 				}
2827 				ire_refrele(gw_ire);
2828 				save_ire = ire;
2829 			}
2830 			break;
2831 		case IRE_IF_NORESOLVER:
2832 		case IRE_IF_RESOLVER:
2833 			if (ire->ire_ipif == ipif) {
2834 				if (save_ire != NULL) {
2835 					ire_refrele(save_ire);
2836 				}
2837 				IRE_REFHOLD(ire);
2838 
2839 				IRB_REFRELE(irb);
2840 				return (ire);
2841 			}
2842 			break;
2843 		}
2844 	}
2845 	IRB_REFRELE(irb);
2846 
2847 	return (save_ire);
2848 }
2849