xref: /titanic_41/usr/src/uts/common/inet/ip/ip6_ire.c (revision bb25c06cca41ca78e5fb87fbb8e81d55beb18c95)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /*
26  * 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 irb_t *ip_forwarding_table_v6[IP6_MASK_TABLE_SIZE];
66 /* This is dynamically allocated in ip_ire_init */
67 irb_t *ip_cache_table_v6;
68 static	ire_t	ire_null;
69 
70 static ire_t	*ire_ihandle_lookup_onlink_v6(ire_t *cire);
71 static	void	ire_report_ftable_v6(ire_t *ire, char *mp);
72 static	void	ire_report_ctable_v6(ire_t *ire, char *mp);
73 static boolean_t ire_match_args_v6(ire_t *ire, const in6_addr_t *addr,
74     const in6_addr_t *mask, const in6_addr_t *gateway, int type,
75     const ipif_t *ipif, zoneid_t zoneid, uint32_t ihandle,
76     const ts_label_t *tsl, int match_flags);
77 
78 /*
79  * Named Dispatch routine to produce a formatted report on all IREs.
80  * This report is accessed by using the ndd utility to "get" ND variable
81  * "ip_ire_status_v6".
82  */
83 /* ARGSUSED */
84 int
85 ip_ire_report_v6(queue_t *q, mblk_t *mp, caddr_t arg, cred_t *ioc_cr)
86 {
87 	zoneid_t zoneid;
88 
89 	(void) mi_mpprintf(mp,
90 	    "IRE      " MI_COL_HDRPAD_STR
91 	    "rfq      " MI_COL_HDRPAD_STR
92 	    "stq      " MI_COL_HDRPAD_STR
93 	    " zone mxfrg rtt   rtt_sd ssthresh ref "
94 	    "rtomax tstamp_ok wscale_ok ecn_ok pmtud_ok sack sendpipe recvpipe "
95 	    "in/out/forward type    addr         mask         "
96 	    "src             gateway");
97 	/*
98 	 *   01234567 01234567 01234567 12345 12345 12345 12345  12345678 123
99 	 *   123456 123456789 123456789 123456 12345678 1234 12345678 12345678
100 	 *   in/out/forward xxxxxxxxxx
101 	 *   xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx
102 	 *   xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx
103 	 *   xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx
104 	 *   xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx
105 	 */
106 
107 	/*
108 	 * Because of the ndd constraint, at most we can have 64K buffer
109 	 * to put in all IRE info.  So to be more efficient, just
110 	 * allocate a 64K buffer here, assuming we need that large buffer.
111 	 * This should be OK as only root can do ndd /dev/ip.
112 	 */
113 	if ((mp->b_cont = allocb(ND_MAX_BUF_LEN, BPRI_HI)) == NULL) {
114 		/* The following may work even if we cannot get a large buf. */
115 		(void) mi_mpprintf(mp, "<< Out of buffer >>\n");
116 		return (0);
117 	}
118 	zoneid = Q_TO_CONN(q)->conn_zoneid;
119 	if (zoneid == GLOBAL_ZONEID)
120 		zoneid = ALL_ZONES;
121 
122 	ire_walk_v6(ire_report_ftable_v6, (char *)mp->b_cont, zoneid);
123 	ire_walk_v6(ire_report_ctable_v6, (char *)mp->b_cont, zoneid);
124 	return (0);
125 }
126 
127 /*
128  * ire_walk routine invoked for ip_ire_report_v6 for each IRE.
129  */
130 static void
131 ire_report_ftable_v6(ire_t *ire, char *mp)
132 {
133 	char	buf1[INET6_ADDRSTRLEN];
134 	char	buf2[INET6_ADDRSTRLEN];
135 	char	buf3[INET6_ADDRSTRLEN];
136 	char	buf4[INET6_ADDRSTRLEN];
137 	uint_t	fo_pkt_count;
138 	uint_t	ib_pkt_count;
139 	int	ref;
140 	in6_addr_t gw_addr_v6;
141 	uint_t	print_len, buf_len;
142 
143 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
144 	if (ire->ire_type & IRE_CACHETABLE)
145 	    return;
146 	buf_len = ((mblk_t *)mp)->b_datap->db_lim - ((mblk_t *)mp)->b_wptr;
147 	if (buf_len <= 0)
148 		return;
149 
150 	/* Number of active references of this ire */
151 	ref = ire->ire_refcnt;
152 	/* "inbound" to a non local address is a forward */
153 	ib_pkt_count = ire->ire_ib_pkt_count;
154 	fo_pkt_count = 0;
155 	ASSERT(!(ire->ire_type & IRE_BROADCAST));
156 	if (!(ire->ire_type & (IRE_LOCAL|IRE_BROADCAST))) {
157 		fo_pkt_count = ib_pkt_count;
158 		ib_pkt_count = 0;
159 	}
160 
161 	mutex_enter(&ire->ire_lock);
162 	gw_addr_v6 = ire->ire_gateway_addr_v6;
163 	mutex_exit(&ire->ire_lock);
164 
165 	print_len = snprintf((char *)((mblk_t *)mp)->b_wptr, buf_len,
166 	    MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR "%5d "
167 	    "%05d %05ld %06ld %08d %03d %06d %09d %09d %06d %08d "
168 	    "%04d %08d %08d %d/%d/%d %s\n\t%s\n\t%s\n\t%s\n\t%s\n",
169 	    (void *)ire, (void *)ire->ire_rfq, (void *)ire->ire_stq,
170 	    (int)ire->ire_zoneid,
171 	    ire->ire_max_frag, ire->ire_uinfo.iulp_rtt,
172 	    ire->ire_uinfo.iulp_rtt_sd,
173 	    ire->ire_uinfo.iulp_ssthresh, ref,
174 	    ire->ire_uinfo.iulp_rtomax,
175 	    (ire->ire_uinfo.iulp_tstamp_ok ? 1: 0),
176 	    (ire->ire_uinfo.iulp_wscale_ok ? 1: 0),
177 	    (ire->ire_uinfo.iulp_ecn_ok ? 1: 0),
178 	    (ire->ire_uinfo.iulp_pmtud_ok ? 1: 0),
179 	    ire->ire_uinfo.iulp_sack,
180 	    ire->ire_uinfo.iulp_spipe, ire->ire_uinfo.iulp_rpipe,
181 	    ib_pkt_count, ire->ire_ob_pkt_count, fo_pkt_count,
182 	    ip_nv_lookup(ire_nv_tbl, (int)ire->ire_type),
183 	    inet_ntop(AF_INET6, &ire->ire_addr_v6, buf1, sizeof (buf1)),
184 	    inet_ntop(AF_INET6, &ire->ire_mask_v6, buf2, sizeof (buf2)),
185 	    inet_ntop(AF_INET6, &ire->ire_src_addr_v6, buf3, sizeof (buf3)),
186 	    inet_ntop(AF_INET6, &gw_addr_v6, buf4, sizeof (buf4)));
187 	if (print_len < buf_len) {
188 		((mblk_t *)mp)->b_wptr += print_len;
189 	} else {
190 		((mblk_t *)mp)->b_wptr += buf_len;
191 	}
192 }
193 
194 /* ire_walk routine invoked for ip_ire_report_v6 for each IRE. */
195 static void
196 ire_report_ctable_v6(ire_t *ire, char *mp)
197 {
198 	char	buf1[INET6_ADDRSTRLEN];
199 	char	buf2[INET6_ADDRSTRLEN];
200 	char	buf3[INET6_ADDRSTRLEN];
201 	char	buf4[INET6_ADDRSTRLEN];
202 	uint_t	fo_pkt_count;
203 	uint_t	ib_pkt_count;
204 	int	ref;
205 	in6_addr_t gw_addr_v6;
206 	uint_t	print_len, buf_len;
207 
208 	if ((ire->ire_type & IRE_CACHETABLE) == 0)
209 		return;
210 	buf_len = ((mblk_t *)mp)->b_datap->db_lim - ((mblk_t *)mp)->b_wptr;
211 	if (buf_len <= 0)
212 		return;
213 
214 	/* Number of active references of this ire */
215 	ref = ire->ire_refcnt;
216 	/* "inbound" to a non local address is a forward */
217 	ib_pkt_count = ire->ire_ib_pkt_count;
218 	fo_pkt_count = 0;
219 	ASSERT(!(ire->ire_type & IRE_BROADCAST));
220 	if (ire->ire_type & IRE_LOCAL) {
221 		fo_pkt_count = ib_pkt_count;
222 		ib_pkt_count = 0;
223 	}
224 
225 	mutex_enter(&ire->ire_lock);
226 	gw_addr_v6 = ire->ire_gateway_addr_v6;
227 	mutex_exit(&ire->ire_lock);
228 
229 	print_len =  snprintf((char *)((mblk_t *)mp)->b_wptr, buf_len,
230 	    MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR "%5d "
231 	    "%05d %05ld %06ld %08d %03d %06d %09d %09d %06d %08d "
232 	    "%04d %08d %08d %d/%d/%d %s\n\t%s\n\t%s\n\t%s\n\t%s\n",
233 	    (void *)ire, (void *)ire->ire_rfq, (void *)ire->ire_stq,
234 	    (int)ire->ire_zoneid,
235 	    ire->ire_max_frag, ire->ire_uinfo.iulp_rtt,
236 	    ire->ire_uinfo.iulp_rtt_sd, ire->ire_uinfo.iulp_ssthresh, ref,
237 	    ire->ire_uinfo.iulp_rtomax,
238 	    (ire->ire_uinfo.iulp_tstamp_ok ? 1: 0),
239 	    (ire->ire_uinfo.iulp_wscale_ok ? 1: 0),
240 	    (ire->ire_uinfo.iulp_ecn_ok ? 1: 0),
241 	    (ire->ire_uinfo.iulp_pmtud_ok ? 1: 0),
242 	    ire->ire_uinfo.iulp_sack,
243 	    ire->ire_uinfo.iulp_spipe, ire->ire_uinfo.iulp_rpipe,
244 	    ib_pkt_count, ire->ire_ob_pkt_count,
245 	    fo_pkt_count, ip_nv_lookup(ire_nv_tbl, (int)ire->ire_type),
246 	    inet_ntop(AF_INET6, &ire->ire_addr_v6, buf1, sizeof (buf1)),
247 	    inet_ntop(AF_INET6, &ire->ire_mask_v6, buf2, sizeof (buf2)),
248 	    inet_ntop(AF_INET6, &ire->ire_src_addr_v6, buf3, sizeof (buf3)),
249 	    inet_ntop(AF_INET6, &gw_addr_v6, buf4, sizeof (buf4)));
250 	if (print_len < buf_len) {
251 		((mblk_t *)mp)->b_wptr += print_len;
252 	} else {
253 		((mblk_t *)mp)->b_wptr += buf_len;
254 	}
255 }
256 
257 
258 /*
259  * Initialize the ire that is specific to IPv6 part and call
260  * ire_init_common to finish it.
261  */
262 ire_t *
263 ire_init_v6(ire_t *ire, const in6_addr_t *v6addr,
264     const in6_addr_t *v6mask, const in6_addr_t *v6src_addr,
265     const in6_addr_t *v6gateway, uint_t *max_fragp,
266     mblk_t *fp_mp, queue_t *rfq, queue_t *stq, ushort_t type,
267     mblk_t *dlureq_mp, ipif_t *ipif, const in6_addr_t *v6cmask,
268     uint32_t phandle, uint32_t ihandle, uint_t flags, const iulp_t *ulp_info,
269     tsol_gc_t *gc, tsol_gcgrp_t *gcgrp)
270 {
271 
272 	/*
273 	 * Reject IRE security attribute creation/initialization
274 	 * if system is not running in Trusted mode.
275 	 */
276 	if ((gc != NULL || gcgrp != NULL) && !is_system_labeled())
277 		return (NULL);
278 
279 	if (fp_mp != NULL) {
280 		/*
281 		 * We can't dupb() here as multiple threads could be
282 		 * calling dupb on the same mp which is incorrect.
283 		 * First dupb() should be called only by one thread.
284 		 */
285 		fp_mp = copyb(fp_mp);
286 		if (fp_mp == NULL)
287 			return (NULL);
288 	}
289 
290 	if (dlureq_mp != NULL) {
291 		/*
292 		 * We can't dupb() here as multiple threads could be
293 		 * calling dupb on the same mp which is incorrect.
294 		 * First dupb() should be called only by one thread.
295 		 */
296 		dlureq_mp = copyb(dlureq_mp);
297 		if (dlureq_mp == NULL) {
298 			if (fp_mp != NULL)
299 				freeb(fp_mp);
300 			return (NULL);
301 		}
302 	}
303 
304 	BUMP_IRE_STATS(ire_stats_v6, ire_stats_alloced);
305 	ire->ire_addr_v6 = *v6addr;
306 
307 	if (v6src_addr != NULL)
308 		ire->ire_src_addr_v6 = *v6src_addr;
309 	if (v6mask != NULL) {
310 		ire->ire_mask_v6 = *v6mask;
311 		ire->ire_masklen = ip_mask_to_plen_v6(&ire->ire_mask_v6);
312 	}
313 	if (v6gateway != NULL)
314 		ire->ire_gateway_addr_v6 = *v6gateway;
315 
316 	if (type == IRE_CACHE && v6cmask != NULL)
317 		ire->ire_cmask_v6 = *v6cmask;
318 
319 	/*
320 	 * Multirouted packets need to have a fragment header added so that
321 	 * the receiver is able to discard duplicates according to their
322 	 * fragment identifier.
323 	 */
324 	if (type == IRE_CACHE && (flags & RTF_MULTIRT)) {
325 		ire->ire_frag_flag = IPH_FRAG_HDR;
326 	}
327 
328 	/* ire_init_common will free the mblks upon encountering any failure */
329 	if (!ire_init_common(ire, max_fragp, fp_mp, rfq, stq, type, dlureq_mp,
330 	    ipif, NULL, phandle, ihandle, flags, IPV6_VERSION, ulp_info,
331 	    gc, gcgrp))
332 		return (NULL);
333 
334 	return (ire);
335 }
336 
337 /*
338  * Similar to ire_create_v6 except that it is called only when
339  * we want to allocate ire as an mblk e.g. we have a external
340  * resolver. Do we need this in IPv6 ?
341  */
342 ire_t *
343 ire_create_mp_v6(const in6_addr_t *v6addr, const in6_addr_t *v6mask,
344     const in6_addr_t *v6src_addr, const in6_addr_t *v6gateway,
345     mblk_t *fp_mp, queue_t *rfq, queue_t *stq, ushort_t type,
346     mblk_t *dlureq_mp, ipif_t *ipif, const in6_addr_t *v6cmask,
347     uint32_t phandle, uint32_t ihandle, uint_t flags, const iulp_t *ulp_info,
348     tsol_gc_t *gc, tsol_gcgrp_t *gcgrp)
349 {
350 	ire_t	*ire;
351 	ire_t	*ret_ire;
352 	mblk_t	*mp;
353 
354 	ASSERT(!IN6_IS_ADDR_V4MAPPED(v6addr));
355 
356 	/* Allocate the new IRE. */
357 	mp = allocb(sizeof (ire_t), BPRI_MED);
358 	if (mp == NULL) {
359 		ip1dbg(("ire_create_mp_v6: alloc failed\n"));
360 		return (NULL);
361 	}
362 
363 	ire = (ire_t *)mp->b_rptr;
364 	mp->b_wptr = (uchar_t *)&ire[1];
365 
366 	/* Start clean. */
367 	*ire = ire_null;
368 	ire->ire_mp = mp;
369 	mp->b_datap->db_type = IRE_DB_TYPE;
370 
371 	ret_ire = ire_init_v6(ire, v6addr, v6mask, v6src_addr, v6gateway,
372 	    NULL, fp_mp, rfq, stq, type, dlureq_mp, ipif, v6cmask, phandle,
373 	    ihandle, flags, ulp_info, gc, gcgrp);
374 
375 	if (ret_ire == NULL) {
376 		freeb(ire->ire_mp);
377 		return (NULL);
378 	}
379 	return (ire);
380 }
381 
382 /*
383  * ire_create_v6 is called to allocate and initialize a new IRE.
384  *
385  * NOTE : This is called as writer sometimes though not required
386  * by this function.
387  */
388 ire_t *
389 ire_create_v6(const in6_addr_t *v6addr, const in6_addr_t *v6mask,
390     const in6_addr_t *v6src_addr, const in6_addr_t *v6gateway,
391     uint_t *max_fragp, mblk_t *fp_mp, queue_t *rfq, queue_t *stq, ushort_t type,
392     mblk_t *dlureq_mp, ipif_t *ipif, const in6_addr_t *v6cmask,
393     uint32_t phandle, uint32_t ihandle, uint_t flags, const iulp_t *ulp_info,
394     tsol_gc_t *gc, tsol_gcgrp_t *gcgrp)
395 {
396 	ire_t	*ire;
397 	ire_t	*ret_ire;
398 
399 	ASSERT(!IN6_IS_ADDR_V4MAPPED(v6addr));
400 
401 	ire = kmem_cache_alloc(ire_cache, KM_NOSLEEP);
402 	if (ire == NULL) {
403 		ip1dbg(("ire_create_v6: alloc failed\n"));
404 		return (NULL);
405 	}
406 	*ire = ire_null;
407 
408 	ret_ire = ire_init_v6(ire, v6addr, v6mask, v6src_addr, v6gateway,
409 	    max_fragp, fp_mp, rfq, stq, type, dlureq_mp, ipif, v6cmask, phandle,
410 	    ihandle, flags, ulp_info, gc, gcgrp);
411 
412 	if (ret_ire == NULL) {
413 		kmem_cache_free(ire_cache, ire);
414 		return (NULL);
415 	}
416 	ASSERT(ret_ire == ire);
417 	return (ire);
418 }
419 
420 /*
421  * Find an IRE_INTERFACE for the multicast group.
422  * Allows different routes for multicast addresses
423  * in the unicast routing table (akin to FF::0/8 but could be more specific)
424  * which point at different interfaces. This is used when IPV6_MULTICAST_IF
425  * isn't specified (when sending) and when IPV6_JOIN_GROUP doesn't
426  * specify the interface to join on.
427  *
428  * Supports link-local addresses by following the ipif/ill when recursing.
429  */
430 ire_t *
431 ire_lookup_multi_v6(const in6_addr_t *group, zoneid_t zoneid)
432 {
433 	ire_t	*ire;
434 	ipif_t	*ipif = NULL;
435 	int	match_flags = MATCH_IRE_TYPE;
436 	in6_addr_t gw_addr_v6;
437 
438 	ire = ire_ftable_lookup_v6(group, 0, 0, 0, NULL, NULL,
439 	    zoneid, 0, NULL, MATCH_IRE_DEFAULT);
440 
441 	/* We search a resolvable ire in case of multirouting. */
442 	if ((ire != NULL) && (ire->ire_flags & RTF_MULTIRT)) {
443 		ire_t *cire = NULL;
444 		/*
445 		 * If the route is not resolvable, the looked up ire
446 		 * may be changed here. In that case, ire_multirt_lookup()
447 		 * IRE_REFRELE the original ire and change it.
448 		 */
449 		(void) ire_multirt_lookup_v6(&cire, &ire, MULTIRT_CACHEGW,
450 		    NULL);
451 		if (cire != NULL)
452 			ire_refrele(cire);
453 	}
454 	if (ire == NULL)
455 		return (NULL);
456 	/*
457 	 * Make sure we follow ire_ipif.
458 	 *
459 	 * We need to determine the interface route through
460 	 * which the gateway will be reached. We don't really
461 	 * care which interface is picked if the interface is
462 	 * part of a group.
463 	 */
464 	if (ire->ire_ipif != NULL) {
465 		ipif = ire->ire_ipif;
466 		match_flags |= MATCH_IRE_ILL_GROUP;
467 	}
468 
469 	switch (ire->ire_type) {
470 	case IRE_DEFAULT:
471 	case IRE_PREFIX:
472 	case IRE_HOST:
473 		mutex_enter(&ire->ire_lock);
474 		gw_addr_v6 = ire->ire_gateway_addr_v6;
475 		mutex_exit(&ire->ire_lock);
476 		ire_refrele(ire);
477 		ire = ire_ftable_lookup_v6(&gw_addr_v6, 0, 0,
478 		    IRE_INTERFACE, ipif, NULL, zoneid, 0,
479 		    NULL, match_flags);
480 		return (ire);
481 	case IRE_IF_NORESOLVER:
482 	case IRE_IF_RESOLVER:
483 		return (ire);
484 	default:
485 		ire_refrele(ire);
486 		return (NULL);
487 	}
488 }
489 
490 /*
491  * Return any local address.  We use this to target ourselves
492  * when the src address was specified as 'default'.
493  * Preference for IRE_LOCAL entries.
494  */
495 ire_t *
496 ire_lookup_local_v6(zoneid_t zoneid)
497 {
498 	ire_t	*ire;
499 	irb_t	*irb;
500 	ire_t	*maybe = NULL;
501 	int i;
502 
503 	for (i = 0; i < ip6_cache_table_size;  i++) {
504 		irb = &ip_cache_table_v6[i];
505 		if (irb->irb_ire == NULL)
506 			continue;
507 		rw_enter(&irb->irb_lock, RW_READER);
508 		for (ire = irb->irb_ire; ire; ire = ire->ire_next) {
509 			if ((ire->ire_marks & IRE_MARK_CONDEMNED) ||
510 			    ire->ire_zoneid != zoneid &&
511 			    ire->ire_zoneid != ALL_ZONES)
512 				continue;
513 			switch (ire->ire_type) {
514 			case IRE_LOOPBACK:
515 				if (maybe == NULL) {
516 					IRE_REFHOLD(ire);
517 					maybe = ire;
518 				}
519 				break;
520 			case IRE_LOCAL:
521 				if (maybe != NULL) {
522 					ire_refrele(maybe);
523 				}
524 				IRE_REFHOLD(ire);
525 				rw_exit(&irb->irb_lock);
526 				return (ire);
527 			}
528 		}
529 		rw_exit(&irb->irb_lock);
530 	}
531 	return (maybe);
532 }
533 
534 /*
535  * This function takes a mask and returns number of bits set in the
536  * mask (the represented prefix length).  Assumes a contiguous mask.
537  */
538 int
539 ip_mask_to_plen_v6(const in6_addr_t *v6mask)
540 {
541 	int		bits;
542 	int		plen = IPV6_ABITS;
543 	int		i;
544 
545 	for (i = 3; i >= 0; i--) {
546 		if (v6mask->s6_addr32[i] == 0) {
547 			plen -= 32;
548 			continue;
549 		}
550 		bits = ffs(ntohl(v6mask->s6_addr32[i])) - 1;
551 		if (bits == 0)
552 			break;
553 		plen -= bits;
554 	}
555 
556 	return (plen);
557 }
558 
559 /*
560  * Convert a prefix length to the mask for that prefix.
561  * Returns the argument bitmask.
562  */
563 in6_addr_t *
564 ip_plen_to_mask_v6(uint_t plen, in6_addr_t *bitmask)
565 {
566 	uint32_t *ptr;
567 
568 	if (plen < 0 || plen > IPV6_ABITS)
569 		return (NULL);
570 	*bitmask = ipv6_all_zeros;
571 
572 	ptr = (uint32_t *)bitmask;
573 	while (plen > 32) {
574 		*ptr++ = 0xffffffffU;
575 		plen -= 32;
576 	}
577 	*ptr = htonl(0xffffffffU << (32 - plen));
578 	return (bitmask);
579 }
580 
581 /*
582  * Add a fully initialized IRE to an appropriate
583  * table based on ire_type.
584  *
585  * The forward table contains IRE_PREFIX/IRE_HOST/IRE_HOST and
586  * IRE_IF_RESOLVER/IRE_IF_NORESOLVER and IRE_DEFAULT.
587  *
588  * The cache table contains IRE_BROADCAST/IRE_LOCAL/IRE_LOOPBACK
589  * and IRE_CACHE.
590  *
591  * NOTE : This function is called as writer though not required
592  * by this function.
593  */
594 int
595 ire_add_v6(ire_t **ire_p, queue_t *q, mblk_t *mp, ipsq_func_t func)
596 {
597 	ire_t	*ire1;
598 	int	mask_table_index;
599 	irb_t	*irb_ptr;
600 	ire_t	**irep;
601 	int	flags;
602 	ire_t	*pire = NULL;
603 	ill_t	*stq_ill;
604 	boolean_t	ndp_g_lock_held = B_FALSE;
605 	ire_t	*ire = *ire_p;
606 	int	error;
607 
608 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
609 	ASSERT(ire->ire_mp == NULL); /* Calls should go through ire_add */
610 	ASSERT(ire->ire_nce == NULL);
611 
612 	/* Find the appropriate list head. */
613 	switch (ire->ire_type) {
614 	case IRE_HOST:
615 		ire->ire_mask_v6 = ipv6_all_ones;
616 		ire->ire_masklen = IPV6_ABITS;
617 		if ((ire->ire_flags & RTF_SETSRC) == 0)
618 			ire->ire_src_addr_v6 = ipv6_all_zeros;
619 		break;
620 	case IRE_CACHE:
621 	case IRE_LOCAL:
622 	case IRE_LOOPBACK:
623 		ire->ire_mask_v6 = ipv6_all_ones;
624 		ire->ire_masklen = IPV6_ABITS;
625 		break;
626 	case IRE_PREFIX:
627 		if ((ire->ire_flags & RTF_SETSRC) == 0)
628 			ire->ire_src_addr_v6 = ipv6_all_zeros;
629 		break;
630 	case IRE_DEFAULT:
631 		if ((ire->ire_flags & RTF_SETSRC) == 0)
632 			ire->ire_src_addr_v6 = ipv6_all_zeros;
633 		break;
634 	case IRE_IF_RESOLVER:
635 	case IRE_IF_NORESOLVER:
636 		break;
637 	default:
638 		printf("ire_add_v6: ire %p has unrecognized IRE type (%d)\n",
639 		    (void *)ire, ire->ire_type);
640 		ire_delete(ire);
641 		*ire_p = NULL;
642 		return (EINVAL);
643 	}
644 
645 	/* Make sure the address is properly masked. */
646 	V6_MASK_COPY(ire->ire_addr_v6, ire->ire_mask_v6, ire->ire_addr_v6);
647 
648 	if ((ire->ire_type & IRE_CACHETABLE) == 0) {
649 		/* IRE goes into Forward Table */
650 		mask_table_index = ip_mask_to_plen_v6(&ire->ire_mask_v6);
651 		if ((ip_forwarding_table_v6[mask_table_index]) == NULL) {
652 			irb_t *ptr;
653 			int i;
654 
655 			ptr = (irb_t *)mi_zalloc((ip6_ftable_hash_size *
656 			    sizeof (irb_t)));
657 			if (ptr == NULL) {
658 				ire_delete(ire);
659 				*ire_p = NULL;
660 				return (ENOMEM);
661 			}
662 			for (i = 0; i < ip6_ftable_hash_size; i++) {
663 				rw_init(&ptr[i].irb_lock, NULL,
664 				    RW_DEFAULT, NULL);
665 			}
666 			mutex_enter(&ire_ft_init_lock);
667 			if (ip_forwarding_table_v6[mask_table_index] == NULL) {
668 				ip_forwarding_table_v6[mask_table_index] = ptr;
669 				mutex_exit(&ire_ft_init_lock);
670 			} else {
671 				/*
672 				 * Some other thread won the race in
673 				 * initializing the forwarding table at the
674 				 * same index.
675 				 */
676 				mutex_exit(&ire_ft_init_lock);
677 				for (i = 0; i < ip6_ftable_hash_size; i++) {
678 					rw_destroy(&ptr[i].irb_lock);
679 				}
680 				mi_free(ptr);
681 			}
682 		}
683 		irb_ptr = &(ip_forwarding_table_v6[mask_table_index][
684 		    IRE_ADDR_MASK_HASH_V6(ire->ire_addr_v6, ire->ire_mask_v6,
685 		    ip6_ftable_hash_size)]);
686 	} else {
687 		irb_ptr = &(ip_cache_table_v6[IRE_ADDR_HASH_V6(
688 		    ire->ire_addr_v6, ip6_cache_table_size)]);
689 	}
690 	/*
691 	 * For xresolv interfaces (v6 interfaces with an external
692 	 * address resolver), ip_newroute_v6/ip_newroute_ipif_v6
693 	 * are unable to prevent the deletion of the interface route
694 	 * while adding an IRE_CACHE for an on-link destination
695 	 * in the IRE_IF_RESOLVER case, since the ire has to go to
696 	 * the external resolver and return. We can't do a REFHOLD on the
697 	 * associated interface ire for fear of the message being freed
698 	 * if the external resolver can't resolve the address.
699 	 * Here we look up the interface ire in the forwarding table
700 	 * and make sure that the interface route has not been deleted.
701 	 */
702 	if (ire->ire_type == IRE_CACHE &&
703 	    IN6_IS_ADDR_UNSPECIFIED(&ire->ire_gateway_addr_v6) &&
704 	    (((ill_t *)ire->ire_stq->q_ptr)->ill_net_type == IRE_IF_RESOLVER) &&
705 	    (((ill_t *)ire->ire_stq->q_ptr)->ill_flags & ILLF_XRESOLV)) {
706 
707 		pire = ire_ihandle_lookup_onlink_v6(ire);
708 		if (pire == NULL) {
709 			ire_delete(ire);
710 			*ire_p = NULL;
711 			return (EINVAL);
712 		}
713 		/* Prevent pire from getting deleted */
714 		IRB_REFHOLD(pire->ire_bucket);
715 		/* Has it been removed already? */
716 		if (pire->ire_marks & IRE_MARK_CONDEMNED) {
717 			IRB_REFRELE(pire->ire_bucket);
718 			ire_refrele(pire);
719 			ire_delete(ire);
720 			*ire_p = NULL;
721 			return (EINVAL);
722 		}
723 	}
724 
725 	flags = (MATCH_IRE_MASK | MATCH_IRE_TYPE | MATCH_IRE_GW);
726 	/*
727 	 * For IRE_CACHES, MATCH_IRE_IPIF is not enough to check
728 	 * for duplicates because :
729 	 *
730 	 * 1) ire_ipif->ipif_ill and ire_stq->q_ptr could be
731 	 *    pointing at different ills. A real duplicate is
732 	 *    a match on both ire_ipif and ire_stq.
733 	 *
734 	 * 2) We could have multiple packets trying to create
735 	 *    an IRE_CACHE for the same ill.
736 	 *
737 	 * Moreover, IPIF_NOFAILOVER and IPV6_BOUND_PIF endpoints wants
738 	 * to go out on a particular ill. Rather than looking at the
739 	 * packet, we depend on the above for MATCH_IRE_ILL here.
740 	 *
741 	 * Unlike IPv4, MATCH_IRE_IPIF is needed here as we could have
742 	 * multiple IRE_CACHES for an ill for the same destination
743 	 * with various scoped addresses i.e represented by ipifs.
744 	 *
745 	 * MATCH_IRE_ILL is done implicitly below for IRE_CACHES.
746 	 */
747 	if (ire->ire_ipif != NULL)
748 		flags |= MATCH_IRE_IPIF;
749 	/*
750 	 * If we are creating hidden ires, make sure we search on
751 	 * this ill (MATCH_IRE_ILL) and a hidden ire, while we are
752 	 * searching for duplicates below. Otherwise we could
753 	 * potentially find an IRE on some other interface
754 	 * and it may not be a IRE marked with IRE_MARK_HIDDEN. We
755 	 * shouldn't do this as this will lead to an infinite loop as
756 	 * eventually we need an hidden ire for this packet to go
757 	 * out. MATCH_IRE_ILL is already marked above.
758 	 */
759 	if (ire->ire_marks & IRE_MARK_HIDDEN) {
760 		ASSERT(ire->ire_type == IRE_CACHE);
761 		flags |= MATCH_IRE_MARK_HIDDEN;
762 	}
763 
764 	/*
765 	 * Start the atomic add of the ire. Grab the ill locks,
766 	 * ill_g_usesrc_lock and the bucket lock. Check for condemned.
767 	 * To avoid lock order problems, get the ndp6.ndp_g_lock now itself.
768 	 */
769 	if (ire->ire_type == IRE_CACHE) {
770 		mutex_enter(&ndp6.ndp_g_lock);
771 		ndp_g_lock_held = B_TRUE;
772 	}
773 
774 	/*
775 	 * If ipif or ill is changing ire_atomic_start() may queue the
776 	 * request and return EINPROGRESS.
777 	 */
778 
779 	error = ire_atomic_start(irb_ptr, ire, q, mp, func);
780 	if (error != 0) {
781 		if (ndp_g_lock_held)
782 			mutex_exit(&ndp6.ndp_g_lock);
783 		/*
784 		 * We don't know whether it is a valid ipif or not.
785 		 * So, set it to NULL. This assumes that the ire has not added
786 		 * a reference to the ipif.
787 		 */
788 		ire->ire_ipif = NULL;
789 		ire_delete(ire);
790 		if (pire != NULL) {
791 			IRB_REFRELE(pire->ire_bucket);
792 			ire_refrele(pire);
793 		}
794 		*ire_p = NULL;
795 		return (error);
796 	}
797 	/*
798 	 * To avoid creating ires having stale values for the ire_max_frag
799 	 * we get the latest value atomically here. For more details
800 	 * see the block comment in ip_sioctl_mtu and in DL_NOTE_SDU_CHANGE
801 	 * in ip_rput_dlpi_writer
802 	 */
803 	if (ire->ire_max_fragp == NULL) {
804 		if (IN6_IS_ADDR_MULTICAST(&ire->ire_addr_v6))
805 			ire->ire_max_frag = ire->ire_ipif->ipif_mtu;
806 		else
807 			ire->ire_max_frag = pire->ire_max_frag;
808 	} else {
809 		uint_t  max_frag;
810 
811 		max_frag = *ire->ire_max_fragp;
812 		ire->ire_max_fragp = NULL;
813 		ire->ire_max_frag = max_frag;
814 	}
815 
816 	/*
817 	 * Atomically check for duplicate and insert in the table.
818 	 */
819 	for (ire1 = irb_ptr->irb_ire; ire1 != NULL; ire1 = ire1->ire_next) {
820 		if (ire1->ire_marks & IRE_MARK_CONDEMNED)
821 			continue;
822 
823 		if (ire->ire_type == IRE_CACHE) {
824 			/*
825 			 * We do MATCH_IRE_ILL implicitly here for IRE_CACHES.
826 			 * As ire_ipif and ire_stq could point to two
827 			 * different ills, we can't pass just ire_ipif to
828 			 * ire_match_args and get a match on both ills.
829 			 * This is just needed for duplicate checks here and
830 			 * so we don't add an extra argument to
831 			 * ire_match_args for this. Do it locally.
832 			 *
833 			 * NOTE : Currently there is no part of the code
834 			 * that asks for both MATH_IRE_IPIF and MATCH_IRE_ILL
835 			 * match for IRE_CACHEs. Thus we don't want to
836 			 * extend the arguments to ire_match_args_v6.
837 			 */
838 			if (ire1->ire_stq != ire->ire_stq)
839 				continue;
840 			/*
841 			 * Multiroute IRE_CACHEs for a given destination can
842 			 * have the same ire_ipif, typically if their source
843 			 * address is forced using RTF_SETSRC, and the same
844 			 * send-to queue. We differentiate them using the parent
845 			 * handle.
846 			 */
847 			if ((ire1->ire_flags & RTF_MULTIRT) &&
848 			    (ire->ire_flags & RTF_MULTIRT) &&
849 			    (ire1->ire_phandle != ire->ire_phandle))
850 				continue;
851 		}
852 		if (ire1->ire_zoneid != ire->ire_zoneid)
853 			continue;
854 		if (ire_match_args_v6(ire1, &ire->ire_addr_v6,
855 		    &ire->ire_mask_v6, &ire->ire_gateway_addr_v6,
856 		    ire->ire_type, ire->ire_ipif, ire->ire_zoneid, 0, NULL,
857 		    flags)) {
858 			/*
859 			 * Return the old ire after doing a REFHOLD.
860 			 * As most of the callers continue to use the IRE
861 			 * after adding, we return a held ire. This will
862 			 * avoid a lookup in the caller again. If the callers
863 			 * don't want to use it, they need to do a REFRELE.
864 			 */
865 			ip1dbg(("found dup ire existing %p new %p",
866 			    (void *)ire1, (void *)ire));
867 			IRE_REFHOLD(ire1);
868 			if (ndp_g_lock_held)
869 				mutex_exit(&ndp6.ndp_g_lock);
870 			ire_atomic_end(irb_ptr, ire);
871 			ire_delete(ire);
872 			if (pire != NULL) {
873 				/*
874 				 * Assert that it is
875 				 * not yet removed from the list.
876 				 */
877 				ASSERT(pire->ire_ptpn != NULL);
878 				IRB_REFRELE(pire->ire_bucket);
879 				ire_refrele(pire);
880 			}
881 			*ire_p = ire1;
882 			return (0);
883 		}
884 	}
885 	if (ire->ire_type == IRE_CACHE) {
886 		in6_addr_t gw_addr_v6;
887 		ill_t	*ill = ire_to_ill(ire);
888 		char	buf[INET6_ADDRSTRLEN];
889 		nce_t	*nce;
890 
891 		/*
892 		 * All IRE_CACHE types must have a nce.  If this is
893 		 * not the case the entry will not be added. We need
894 		 * to make sure that if somebody deletes the nce
895 		 * after we looked up, they will find this ire and
896 		 * delete the ire. To delete this ire one needs the
897 		 * bucket lock which we are still holding here. So,
898 		 * even if the nce gets deleted after we looked up,
899 		 * this ire  will get deleted.
900 		 *
901 		 * NOTE : Don't need the ire_lock for accessing
902 		 * ire_gateway_addr_v6 as it is appearing first
903 		 * time on the list and rts_setgwr_v6 could not
904 		 * be changing this.
905 		 */
906 		gw_addr_v6 = ire->ire_gateway_addr_v6;
907 		if (IN6_IS_ADDR_UNSPECIFIED(&gw_addr_v6)) {
908 			nce = ndp_lookup_v6(ill, &ire->ire_addr_v6, B_TRUE);
909 		} else {
910 			nce = ndp_lookup_v6(ill, &gw_addr_v6, B_TRUE);
911 		}
912 		if (nce == NULL)
913 			goto failed;
914 
915 		/* Pair of refhold, refrele just to get the tracing right */
916 		NCE_REFHOLD_TO_REFHOLD_NOTR(nce);
917 		/*
918 		 * Atomically make sure that new IREs don't point
919 		 * to an NCE that is logically deleted (CONDEMNED).
920 		 * ndp_delete() first marks the NCE CONDEMNED.
921 		 * This ensures that the nce_refcnt won't increase
922 		 * due to new nce_lookups or due to addition of new IREs
923 		 * pointing to this NCE. Then ndp_delete() cleans up
924 		 * existing references. If we don't do it atomically here,
925 		 * ndp_delete() -> nce_ire_delete() will not be able to
926 		 * clean up the IRE list completely, and the nce_refcnt
927 		 * won't go down to zero.
928 		 */
929 		mutex_enter(&nce->nce_lock);
930 		if (ill->ill_flags & ILLF_XRESOLV) {
931 			/*
932 			 * If we used an external resolver, we may not
933 			 * have gone through neighbor discovery to get here.
934 			 * Must update the nce_state before the next check.
935 			 */
936 			if (nce->nce_state == ND_INCOMPLETE)
937 				nce->nce_state = ND_REACHABLE;
938 		}
939 		if (nce->nce_state == ND_INCOMPLETE ||
940 		    (nce->nce_flags & NCE_F_CONDEMNED) ||
941 		    (nce->nce_state == ND_UNREACHABLE)) {
942 failed:
943 			if (ndp_g_lock_held)
944 				mutex_exit(&ndp6.ndp_g_lock);
945 			if (nce != NULL)
946 				mutex_exit(&nce->nce_lock);
947 			ire_atomic_end(irb_ptr, ire);
948 			ip1dbg(("ire_add_v6: No nce for dst %s \n",
949 			    inet_ntop(AF_INET6, &ire->ire_addr_v6,
950 			    buf, sizeof (buf))));
951 			ire_delete(ire);
952 			if (pire != NULL) {
953 				/*
954 				 * Assert that it is
955 				 * not yet removed from the list.
956 				 */
957 				ASSERT(pire->ire_ptpn != NULL);
958 				IRB_REFRELE(pire->ire_bucket);
959 				ire_refrele(pire);
960 			}
961 			if (nce != NULL)
962 				NCE_REFRELE_NOTR(nce);
963 			*ire_p = NULL;
964 			return (EINVAL);
965 		} else {
966 			ire->ire_nce = nce;
967 		}
968 		mutex_exit(&nce->nce_lock);
969 	}
970 	/*
971 	 * Find the first entry that matches ire_addr - provides
972 	 * tail insertion. *irep will be null if no match.
973 	 */
974 	irep = (ire_t **)irb_ptr;
975 	while ((ire1 = *irep) != NULL &&
976 	    !IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, &ire1->ire_addr_v6))
977 		irep = &ire1->ire_next;
978 	ASSERT(!(ire->ire_type & IRE_BROADCAST));
979 
980 	if (*irep != NULL) {
981 		/*
982 		 * Find the last ire which matches ire_addr_v6.
983 		 * Needed to do tail insertion among entries with the same
984 		 * ire_addr_v6.
985 		 */
986 		while (IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6,
987 		    &ire1->ire_addr_v6)) {
988 			irep = &ire1->ire_next;
989 			ire1 = *irep;
990 			if (ire1 == NULL)
991 				break;
992 		}
993 	}
994 
995 	if (ire->ire_type == IRE_DEFAULT) {
996 		/*
997 		 * We keep a count of default gateways which is used when
998 		 * assigning them as routes.
999 		 */
1000 		ipv6_ire_default_count++;
1001 		ASSERT(ipv6_ire_default_count != 0); /* Wraparound */
1002 	}
1003 	/* Insert at *irep */
1004 	ire1 = *irep;
1005 	if (ire1 != NULL)
1006 		ire1->ire_ptpn = &ire->ire_next;
1007 	ire->ire_next = ire1;
1008 	/* Link the new one in. */
1009 	ire->ire_ptpn = irep;
1010 	/*
1011 	 * ire_walk routines de-reference ire_next without holding
1012 	 * a lock. Before we point to the new ire, we want to make
1013 	 * sure the store that sets the ire_next of the new ire
1014 	 * reaches global visibility, so that ire_walk routines
1015 	 * don't see a truncated list of ires i.e if the ire_next
1016 	 * of the new ire gets set after we do "*irep = ire" due
1017 	 * to re-ordering, the ire_walk thread will see a NULL
1018 	 * once it accesses the ire_next of the new ire.
1019 	 * membar_producer() makes sure that the following store
1020 	 * happens *after* all of the above stores.
1021 	 */
1022 	membar_producer();
1023 	*irep = ire;
1024 	ire->ire_bucket = irb_ptr;
1025 	/*
1026 	 * We return a bumped up IRE above. Keep it symmetrical
1027 	 * so that the callers will always have to release. This
1028 	 * helps the callers of this function because they continue
1029 	 * to use the IRE after adding and hence they don't have to
1030 	 * lookup again after we return the IRE.
1031 	 *
1032 	 * NOTE : We don't have to use atomics as this is appearing
1033 	 * in the list for the first time and no one else can bump
1034 	 * up the reference count on this yet.
1035 	 */
1036 	IRE_REFHOLD_LOCKED(ire);
1037 	BUMP_IRE_STATS(ire_stats_v6, ire_stats_inserted);
1038 	irb_ptr->irb_ire_cnt++;
1039 	if (ire->ire_marks & IRE_MARK_TEMPORARY)
1040 		irb_ptr->irb_tmp_ire_cnt++;
1041 
1042 	if (ire->ire_ipif != NULL) {
1043 		ire->ire_ipif->ipif_ire_cnt++;
1044 		if (ire->ire_stq != NULL) {
1045 			stq_ill = (ill_t *)ire->ire_stq->q_ptr;
1046 			stq_ill->ill_ire_cnt++;
1047 		}
1048 	} else {
1049 		ASSERT(ire->ire_stq == NULL);
1050 	}
1051 
1052 	if (ndp_g_lock_held)
1053 		mutex_exit(&ndp6.ndp_g_lock);
1054 	ire_atomic_end(irb_ptr, ire);
1055 
1056 	if (pire != NULL) {
1057 		/* Assert that it is not removed from the list yet */
1058 		ASSERT(pire->ire_ptpn != NULL);
1059 		IRB_REFRELE(pire->ire_bucket);
1060 		ire_refrele(pire);
1061 	}
1062 
1063 	if (ire->ire_type != IRE_CACHE) {
1064 		/*
1065 		 * For ire's with with host mask see if there is an entry
1066 		 * in the cache. If there is one flush the whole cache as
1067 		 * there might be multiple entries due to RTF_MULTIRT (CGTP).
1068 		 * If no entry is found than there is no need to flush the
1069 		 * cache.
1070 		 */
1071 
1072 		if (ip_mask_to_plen_v6(&ire->ire_mask_v6) == IPV6_ABITS) {
1073 			ire_t *lire;
1074 			lire = ire_ctable_lookup_v6(&ire->ire_addr_v6, NULL,
1075 			    IRE_CACHE, NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE);
1076 			if (lire != NULL) {
1077 				ire_refrele(lire);
1078 				ire_flush_cache_v6(ire, IRE_FLUSH_ADD);
1079 			}
1080 		} else {
1081 			ire_flush_cache_v6(ire, IRE_FLUSH_ADD);
1082 		}
1083 	}
1084 
1085 	*ire_p = ire;
1086 	return (0);
1087 }
1088 
1089 /*
1090  * Search for all HOST REDIRECT routes that are
1091  * pointing at the specified gateway and
1092  * delete them. This routine is called only
1093  * when a default gateway is going away.
1094  */
1095 static void
1096 ire_delete_host_redirects_v6(const in6_addr_t *gateway)
1097 {
1098 	irb_t *irb_ptr;
1099 	irb_t *irb;
1100 	ire_t *ire;
1101 	in6_addr_t gw_addr_v6;
1102 	int i;
1103 
1104 	/* get the hash table for HOST routes */
1105 	irb_ptr = ip_forwarding_table_v6[(IP6_MASK_TABLE_SIZE - 1)];
1106 	if (irb_ptr == NULL)
1107 		return;
1108 	for (i = 0; (i < ip6_ftable_hash_size); i++) {
1109 		irb = &irb_ptr[i];
1110 		IRB_REFHOLD(irb);
1111 		for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
1112 			if (!(ire->ire_flags & RTF_DYNAMIC))
1113 				continue;
1114 			mutex_enter(&ire->ire_lock);
1115 			gw_addr_v6 = ire->ire_gateway_addr_v6;
1116 			mutex_exit(&ire->ire_lock);
1117 			if (IN6_ARE_ADDR_EQUAL(&gw_addr_v6, gateway))
1118 				ire_delete(ire);
1119 		}
1120 		IRB_REFRELE(irb);
1121 	}
1122 }
1123 
1124 /*
1125  * Delete all the cache entries with this 'addr'. This is the IPv6 counterpart
1126  * of ip_ire_clookup_and_delete. The difference being this function does not
1127  * return any value. IPv6 processing of a gratuitous ARP, as it stands, is
1128  * different than IPv4 in that, regardless of the presence of a cache entry
1129  * for this address, an ire_walk_v6 is done. Another difference is that unlike
1130  * in the case of IPv4 this does not take an ipif_t argument, since it is only
1131  * called by ip_arp_news and the match is always only on the address.
1132  */
1133 void
1134 ip_ire_clookup_and_delete_v6(const in6_addr_t *addr)
1135 {
1136 	irb_t		*irb;
1137 	ire_t		*cire;
1138 	boolean_t	found = B_FALSE;
1139 
1140 	irb = &ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr, ip6_cache_table_size)];
1141 	IRB_REFHOLD(irb);
1142 	for (cire = irb->irb_ire; cire != NULL; cire = cire->ire_next) {
1143 		if (cire->ire_marks == IRE_MARK_CONDEMNED)
1144 			continue;
1145 		if (IN6_ARE_ADDR_EQUAL(&cire->ire_addr_v6, addr)) {
1146 
1147 			/* This signifies start of a match */
1148 			if (!found)
1149 				found = B_TRUE;
1150 			if (cire->ire_type == IRE_CACHE) {
1151 				if (cire->ire_nce != NULL)
1152 					ndp_delete(cire->ire_nce);
1153 				ire_delete_v6(cire);
1154 			}
1155 		/* End of the match */
1156 		} else if (found)
1157 			break;
1158 	}
1159 	IRB_REFRELE(irb);
1160 }
1161 
1162 /*
1163  * Delete the specified IRE.
1164  * All calls should use ire_delete().
1165  * Sometimes called as writer though not required by this function.
1166  *
1167  * NOTE : This function is called only if the ire was added
1168  * in the list.
1169  */
1170 void
1171 ire_delete_v6(ire_t *ire)
1172 {
1173 	in6_addr_t gw_addr_v6;
1174 
1175 	ASSERT(ire->ire_refcnt >= 1);
1176 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
1177 
1178 	if (ire->ire_type != IRE_CACHE)
1179 		ire_flush_cache_v6(ire, IRE_FLUSH_DELETE);
1180 	if (ire->ire_type == IRE_DEFAULT) {
1181 		/*
1182 		 * when a default gateway is going away
1183 		 * delete all the host redirects pointing at that
1184 		 * gateway.
1185 		 */
1186 		mutex_enter(&ire->ire_lock);
1187 		gw_addr_v6 = ire->ire_gateway_addr_v6;
1188 		mutex_exit(&ire->ire_lock);
1189 		ire_delete_host_redirects_v6(&gw_addr_v6);
1190 	}
1191 }
1192 
1193 /*
1194  * ire_walk routine to delete all IRE_CACHE and IRE_HOST type redirect
1195  * entries.
1196  */
1197 /*ARGSUSED1*/
1198 void
1199 ire_delete_cache_v6(ire_t *ire, char *arg)
1200 {
1201 	char    addrstr1[INET6_ADDRSTRLEN];
1202 	char    addrstr2[INET6_ADDRSTRLEN];
1203 
1204 	if ((ire->ire_type & IRE_CACHE) ||
1205 	    (ire->ire_flags & RTF_DYNAMIC)) {
1206 		ip1dbg(("ire_delete_cache_v6: deleted %s type %d through %s\n",
1207 		    inet_ntop(AF_INET6, &ire->ire_addr_v6,
1208 			addrstr1, sizeof (addrstr1)),
1209 		    ire->ire_type,
1210 		    inet_ntop(AF_INET6, &ire->ire_gateway_addr_v6,
1211 			addrstr2, sizeof (addrstr2))));
1212 		ire_delete(ire);
1213 	}
1214 
1215 }
1216 
1217 /*
1218  * ire_walk routine to delete all IRE_CACHE/IRE_HOST type redirect entries
1219  * that have a given gateway address.
1220  */
1221 void
1222 ire_delete_cache_gw_v6(ire_t *ire, char *addr)
1223 {
1224 	in6_addr_t	*gw_addr = (in6_addr_t *)addr;
1225 	char		buf1[INET6_ADDRSTRLEN];
1226 	char		buf2[INET6_ADDRSTRLEN];
1227 	in6_addr_t	ire_gw_addr_v6;
1228 
1229 	if (!(ire->ire_type & IRE_CACHE) &&
1230 	    !(ire->ire_flags & RTF_DYNAMIC))
1231 		return;
1232 
1233 	mutex_enter(&ire->ire_lock);
1234 	ire_gw_addr_v6 = ire->ire_gateway_addr_v6;
1235 	mutex_exit(&ire->ire_lock);
1236 
1237 	if (IN6_ARE_ADDR_EQUAL(&ire_gw_addr_v6, gw_addr)) {
1238 		ip1dbg(("ire_delete_cache_gw_v6: deleted %s type %d to %s\n",
1239 		    inet_ntop(AF_INET6, &ire->ire_src_addr_v6,
1240 		    buf1, sizeof (buf1)),
1241 		    ire->ire_type,
1242 		    inet_ntop(AF_INET6, &ire_gw_addr_v6,
1243 		    buf2, sizeof (buf2))));
1244 		ire_delete(ire);
1245 	}
1246 }
1247 
1248 /*
1249  * Remove all IRE_CACHE entries that match
1250  * the ire specified.  (Sometimes called
1251  * as writer though not required by this function.)
1252  *
1253  * The flag argument indicates if the
1254  * flush request is due to addition
1255  * of new route (IRE_FLUSH_ADD) or deletion of old
1256  * route (IRE_FLUSH_DELETE).
1257  *
1258  * This routine takes only the IREs from the forwarding
1259  * table and flushes the corresponding entries from
1260  * the cache table.
1261  *
1262  * When flushing due to the deletion of an old route, it
1263  * just checks the cache handles (ire_phandle and ire_ihandle) and
1264  * deletes the ones that match.
1265  *
1266  * When flushing due to the creation of a new route, it checks
1267  * if a cache entry's address matches the one in the IRE and
1268  * that the cache entry's parent has a less specific mask than the
1269  * one in IRE. The destination of such a cache entry could be the
1270  * gateway for other cache entries, so we need to flush those as
1271  * well by looking for gateway addresses matching the IRE's address.
1272  */
1273 void
1274 ire_flush_cache_v6(ire_t *ire, int flag)
1275 {
1276 	int i;
1277 	ire_t *cire;
1278 	irb_t *irb;
1279 
1280 	if (ire->ire_type & IRE_CACHE)
1281 	    return;
1282 
1283 	/*
1284 	 * If a default is just created, there is no point
1285 	 * in going through the cache, as there will not be any
1286 	 * cached ires.
1287 	 */
1288 	if (ire->ire_type == IRE_DEFAULT && flag == IRE_FLUSH_ADD)
1289 		return;
1290 	if (flag == IRE_FLUSH_ADD) {
1291 		/*
1292 		 * This selective flush is
1293 		 * due to the addition of
1294 		 * new IRE.
1295 		 */
1296 		for (i = 0; i < ip6_cache_table_size; i++) {
1297 			irb = &ip_cache_table_v6[i];
1298 			if ((cire = irb->irb_ire) == NULL)
1299 				continue;
1300 			IRB_REFHOLD(irb);
1301 			for (cire = irb->irb_ire; cire != NULL;
1302 			    cire = cire->ire_next) {
1303 				if (cire->ire_type != IRE_CACHE)
1304 					continue;
1305 				/*
1306 				 * If 'cire' belongs to the same subnet
1307 				 * as the new ire being added, and 'cire'
1308 				 * is derived from a prefix that is less
1309 				 * specific than the new ire being added,
1310 				 * we need to flush 'cire'; for instance,
1311 				 * when a new interface comes up.
1312 				 */
1313 				if ((V6_MASK_EQ_2(cire->ire_addr_v6,
1314 				    ire->ire_mask_v6, ire->ire_addr_v6) &&
1315 				    (ip_mask_to_plen_v6(&cire->ire_cmask_v6) <=
1316 				    ire->ire_masklen))) {
1317 					ire_delete(cire);
1318 					continue;
1319 				}
1320 				/*
1321 				 * This is the case when the ire_gateway_addr
1322 				 * of 'cire' belongs to the same subnet as
1323 				 * the new ire being added.
1324 				 * Flushing such ires is sometimes required to
1325 				 * avoid misrouting: say we have a machine with
1326 				 * two interfaces (I1 and I2), a default router
1327 				 * R on the I1 subnet, and a host route to an
1328 				 * off-link destination D with a gateway G on
1329 				 * the I2 subnet.
1330 				 * Under normal operation, we will have an
1331 				 * on-link cache entry for G and an off-link
1332 				 * cache entry for D with G as ire_gateway_addr,
1333 				 * traffic to D will reach its destination
1334 				 * through gateway G.
1335 				 * If the administrator does 'ifconfig I2 down',
1336 				 * the cache entries for D and G will be
1337 				 * flushed. However, G will now be resolved as
1338 				 * an off-link destination using R (the default
1339 				 * router) as gateway. Then D will also be
1340 				 * resolved as an off-link destination using G
1341 				 * as gateway - this behavior is due to
1342 				 * compatibility reasons, see comment in
1343 				 * ire_ihandle_lookup_offlink(). Traffic to D
1344 				 * will go to the router R and probably won't
1345 				 * reach the destination.
1346 				 * The administrator then does 'ifconfig I2 up'.
1347 				 * Since G is on the I2 subnet, this routine
1348 				 * will flush its cache entry. It must also
1349 				 * flush the cache entry for D, otherwise
1350 				 * traffic will stay misrouted until the IRE
1351 				 * times out.
1352 				 */
1353 				if (V6_MASK_EQ_2(cire->ire_gateway_addr_v6,
1354 				    ire->ire_mask_v6, ire->ire_addr_v6)) {
1355 					ire_delete(cire);
1356 					continue;
1357 				}
1358 			}
1359 			IRB_REFRELE(irb);
1360 		}
1361 	} else {
1362 		/*
1363 		 * delete the cache entries based on
1364 		 * handle in the IRE as this IRE is
1365 		 * being deleted/changed.
1366 		 */
1367 		for (i = 0; i < ip6_cache_table_size; i++) {
1368 			irb = &ip_cache_table_v6[i];
1369 			if ((cire = irb->irb_ire) == NULL)
1370 				continue;
1371 			IRB_REFHOLD(irb);
1372 			for (cire = irb->irb_ire; cire != NULL;
1373 			    cire = cire->ire_next) {
1374 				if (cire->ire_type != IRE_CACHE)
1375 					continue;
1376 				if ((cire->ire_phandle == 0 ||
1377 				    cire->ire_phandle != ire->ire_phandle) &&
1378 				    (cire->ire_ihandle == 0 ||
1379 				    cire->ire_ihandle != ire->ire_ihandle))
1380 					continue;
1381 				ire_delete(cire);
1382 			}
1383 			IRB_REFRELE(irb);
1384 		}
1385 	}
1386 }
1387 
1388 /*
1389  * Matches the arguments passed with the values in the ire.
1390  *
1391  * Note: for match types that match using "ipif" passed in, ipif
1392  * must be checked for non-NULL before calling this routine.
1393  */
1394 static boolean_t
1395 ire_match_args_v6(ire_t *ire, const in6_addr_t *addr, const in6_addr_t *mask,
1396     const in6_addr_t *gateway, int type, const ipif_t *ipif, zoneid_t zoneid,
1397     uint32_t ihandle, const ts_label_t *tsl, int match_flags)
1398 {
1399 	in6_addr_t masked_addr;
1400 	in6_addr_t gw_addr_v6;
1401 	ill_t *ire_ill = NULL, *dst_ill;
1402 	ill_t *ipif_ill = NULL;
1403 	ill_group_t *ire_ill_group = NULL;
1404 	ill_group_t *ipif_ill_group = NULL;
1405 	ipif_t	*src_ipif;
1406 
1407 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
1408 	ASSERT(addr != NULL);
1409 	ASSERT(mask != NULL);
1410 	ASSERT((!(match_flags & MATCH_IRE_GW)) || gateway != NULL);
1411 	ASSERT((!(match_flags & (MATCH_IRE_ILL|MATCH_IRE_ILL_GROUP))) ||
1412 	    (ipif != NULL && ipif->ipif_isv6));
1413 	ASSERT(!(match_flags & MATCH_IRE_WQ));
1414 
1415 	/*
1416 	 * HIDDEN cache entries have to be looked up specifically with
1417 	 * MATCH_IRE_MARK_HIDDEN. MATCH_IRE_MARK_HIDDEN is usually set
1418 	 * when the interface is FAILED or INACTIVE. In that case,
1419 	 * any IRE_CACHES that exists should be marked with
1420 	 * IRE_MARK_HIDDEN. So, we don't really need to match below
1421 	 * for IRE_MARK_HIDDEN. But we do so for consistency.
1422 	 */
1423 	if (!(match_flags & MATCH_IRE_MARK_HIDDEN) &&
1424 	    (ire->ire_marks & IRE_MARK_HIDDEN))
1425 		return (B_FALSE);
1426 
1427 	if (zoneid != ALL_ZONES && zoneid != ire->ire_zoneid &&
1428 	    ire->ire_zoneid != ALL_ZONES) {
1429 		/*
1430 		 * If MATCH_IRE_ZONEONLY has been set and the supplied zoneid is
1431 		 * valid and does not match that of ire_zoneid, a failure to
1432 		 * match is reported at this point. Otherwise, since some IREs
1433 		 * that are available in the global zone can be used in local
1434 		 * zones, additional checks need to be performed:
1435 		 *
1436 		 *	IRE_CACHE and IRE_LOOPBACK entries should
1437 		 *	never be matched in this situation.
1438 		 *
1439 		 *	IRE entries that have an interface associated with them
1440 		 *	should in general not match unless they are an IRE_LOCAL
1441 		 *	or in the case when MATCH_IRE_DEFAULT has been set in
1442 		 *	the caller.  In the case of the former, checking of the
1443 		 *	other fields supplied should take place.
1444 		 *
1445 		 *	In the case where MATCH_IRE_DEFAULT has been set,
1446 		 *	all of the ipif's associated with the IRE's ill are
1447 		 *	checked to see if there is a matching zoneid.  If any
1448 		 *	one ipif has a matching zoneid, this IRE is a
1449 		 *	potential candidate so checking of the other fields
1450 		 *	takes place.
1451 		 *
1452 		 *	In the case where the IRE_INTERFACE has a usable source
1453 		 *	address (indicated by ill_usesrc_ifindex) in the
1454 		 *	correct zone then it's permitted to return this IRE
1455 		 */
1456 		if (match_flags & MATCH_IRE_ZONEONLY)
1457 			return (B_FALSE);
1458 		if (ire->ire_type & (IRE_CACHE | IRE_LOOPBACK))
1459 			return (B_FALSE);
1460 		/*
1461 		 * Note, IRE_INTERFACE can have the stq as NULL. For
1462 		 * example, if the default multicast route is tied to
1463 		 * the loopback address.
1464 		 */
1465 		if ((ire->ire_type & IRE_INTERFACE) &&
1466 		    (ire->ire_stq != NULL)) {
1467 			dst_ill = (ill_t *)ire->ire_stq->q_ptr;
1468 			/*
1469 			 * If there is a usable source address in the
1470 			 * zone, then it's ok to return an
1471 			 * IRE_INTERFACE
1472 			 */
1473 			if ((dst_ill->ill_usesrc_ifindex != 0) &&
1474 			    (src_ipif = ipif_select_source_v6(dst_ill, addr,
1475 			    RESTRICT_TO_NONE, IPV6_PREFER_SRC_DEFAULT, zoneid))
1476 			    != NULL) {
1477 				ip3dbg(("ire_match_args: src_ipif %p"
1478 				    " dst_ill %p", (void *)src_ipif,
1479 				    (void *)dst_ill));
1480 				ipif_refrele(src_ipif);
1481 			} else {
1482 				ip3dbg(("ire_match_args: src_ipif NULL"
1483 				    " dst_ill %p\n", (void *)dst_ill));
1484 				return (B_FALSE);
1485 			}
1486 		}
1487 		if (ire->ire_ipif != NULL && ire->ire_type != IRE_LOCAL &&
1488 		    !(ire->ire_type & IRE_INTERFACE)) {
1489 			ipif_t	*tipif;
1490 
1491 			if ((match_flags & MATCH_IRE_DEFAULT) == 0)
1492 				return (B_FALSE);
1493 			mutex_enter(&ire->ire_ipif->ipif_ill->ill_lock);
1494 			for (tipif = ire->ire_ipif->ipif_ill->ill_ipif;
1495 			    tipif != NULL; tipif = tipif->ipif_next) {
1496 				if (IPIF_CAN_LOOKUP(tipif) &&
1497 				    (tipif->ipif_flags & IPIF_UP) &&
1498 				    (tipif->ipif_zoneid == zoneid ||
1499 				    tipif->ipif_zoneid == ALL_ZONES))
1500 					break;
1501 			}
1502 			mutex_exit(&ire->ire_ipif->ipif_ill->ill_lock);
1503 			if (tipif == NULL)
1504 				return (B_FALSE);
1505 		}
1506 	}
1507 
1508 	if (match_flags & MATCH_IRE_GW) {
1509 		mutex_enter(&ire->ire_lock);
1510 		gw_addr_v6 = ire->ire_gateway_addr_v6;
1511 		mutex_exit(&ire->ire_lock);
1512 	}
1513 	/*
1514 	 * For IRE_CACHES, MATCH_IRE_ILL/ILL_GROUP really means that
1515 	 * somebody wants to send out on a particular interface which
1516 	 * is given by ire_stq and hence use ire_stq to derive the ill
1517 	 * value. ire_ipif for IRE_CACHES is just the
1518 	 * means of getting a source address i.e ire_src_addr_v6 =
1519 	 * ire->ire_ipif->ipif_src_addr_v6.
1520 	 */
1521 	if (match_flags & (MATCH_IRE_ILL|MATCH_IRE_ILL_GROUP)) {
1522 		ire_ill = ire_to_ill(ire);
1523 		if (ire_ill != NULL)
1524 			ire_ill_group = ire_ill->ill_group;
1525 		ipif_ill = ipif->ipif_ill;
1526 		ipif_ill_group = ipif_ill->ill_group;
1527 	}
1528 
1529 	/* No ire_addr_v6 bits set past the mask */
1530 	ASSERT(V6_MASK_EQ(ire->ire_addr_v6, ire->ire_mask_v6,
1531 	    ire->ire_addr_v6));
1532 	V6_MASK_COPY(*addr, *mask, masked_addr);
1533 
1534 	if (V6_MASK_EQ(*addr, *mask, ire->ire_addr_v6) &&
1535 	    ((!(match_flags & MATCH_IRE_GW)) ||
1536 		IN6_ARE_ADDR_EQUAL(&gw_addr_v6, gateway)) &&
1537 	    ((!(match_flags & MATCH_IRE_TYPE)) ||
1538 		(ire->ire_type & type)) &&
1539 	    ((!(match_flags & MATCH_IRE_SRC)) ||
1540 		IN6_ARE_ADDR_EQUAL(&ire->ire_src_addr_v6,
1541 		&ipif->ipif_v6src_addr)) &&
1542 	    ((!(match_flags & MATCH_IRE_IPIF)) ||
1543 		(ire->ire_ipif == ipif)) &&
1544 	    ((!(match_flags & MATCH_IRE_MARK_HIDDEN)) ||
1545 		(ire->ire_type != IRE_CACHE ||
1546 		ire->ire_marks & IRE_MARK_HIDDEN)) &&
1547 	    ((!(match_flags & MATCH_IRE_ILL)) ||
1548 		(ire_ill == ipif_ill)) &&
1549 	    ((!(match_flags & MATCH_IRE_IHANDLE)) ||
1550 		(ire->ire_ihandle == ihandle)) &&
1551 	    ((!(match_flags & MATCH_IRE_ILL_GROUP)) ||
1552 		(ire_ill == ipif_ill) ||
1553 		(ire_ill_group != NULL &&
1554 		ire_ill_group == ipif_ill_group)) &&
1555 	    ((!(match_flags & MATCH_IRE_SECATTR)) ||
1556 		(!is_system_labeled()) ||
1557 		(tsol_ire_match_gwattr(ire, tsl) == 0))) {
1558 		/* We found the matched IRE */
1559 		return (B_TRUE);
1560 	}
1561 	return (B_FALSE);
1562 }
1563 
1564 /*
1565  * Lookup for a route in all the tables
1566  */
1567 ire_t *
1568 ire_route_lookup_v6(const in6_addr_t *addr, const in6_addr_t *mask,
1569     const in6_addr_t *gateway, int type, const ipif_t *ipif, ire_t **pire,
1570     zoneid_t zoneid, const ts_label_t *tsl, int flags)
1571 {
1572 	ire_t *ire = NULL;
1573 
1574 	/*
1575 	 * ire_match_args_v6() will dereference ipif MATCH_IRE_SRC or
1576 	 * MATCH_IRE_ILL is set.
1577 	 */
1578 	if ((flags & (MATCH_IRE_SRC | MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) &&
1579 	    (ipif == NULL))
1580 		return (NULL);
1581 
1582 	/*
1583 	 * might be asking for a cache lookup,
1584 	 * This is not best way to lookup cache,
1585 	 * user should call ire_cache_lookup directly.
1586 	 *
1587 	 * If MATCH_IRE_TYPE was set, first lookup in the cache table and then
1588 	 * in the forwarding table, if the applicable type flags were set.
1589 	 */
1590 	if ((flags & MATCH_IRE_TYPE) == 0 || (type & IRE_CACHETABLE) != 0) {
1591 		ire = ire_ctable_lookup_v6(addr, gateway, type, ipif, zoneid,
1592 		    tsl, flags);
1593 		if (ire != NULL)
1594 			return (ire);
1595 	}
1596 	if ((flags & MATCH_IRE_TYPE) == 0 || (type & IRE_FORWARDTABLE) != 0) {
1597 		ire = ire_ftable_lookup_v6(addr, mask, gateway, type, ipif,
1598 		    pire, zoneid, 0, tsl, flags);
1599 	}
1600 	return (ire);
1601 }
1602 
1603 /*
1604  * Lookup a route in forwarding table.
1605  * specific lookup is indicated by passing the
1606  * required parameters and indicating the
1607  * match required in flag field.
1608  *
1609  * Looking for default route can be done in three ways
1610  * 1) pass mask as ipv6_all_zeros and set MATCH_IRE_MASK in flags field
1611  *    along with other matches.
1612  * 2) pass type as IRE_DEFAULT and set MATCH_IRE_TYPE in flags
1613  *    field along with other matches.
1614  * 3) if the destination and mask are passed as zeros.
1615  *
1616  * A request to return a default route if no route
1617  * is found, can be specified by setting MATCH_IRE_DEFAULT
1618  * in flags.
1619  *
1620  * It does not support recursion more than one level. It
1621  * will do recursive lookup only when the lookup maps to
1622  * a prefix or default route and MATCH_IRE_RECURSIVE flag is passed.
1623  *
1624  * If the routing table is setup to allow more than one level
1625  * of recursion, the cleaning up cache table will not work resulting
1626  * in invalid routing.
1627  *
1628  * Supports link-local addresses by following the ipif/ill when recursing.
1629  *
1630  * NOTE : When this function returns NULL, pire has already been released.
1631  *	  pire is valid only when this function successfully returns an
1632  *	  ire.
1633  */
1634 ire_t *
1635 ire_ftable_lookup_v6(const in6_addr_t *addr, const in6_addr_t *mask,
1636     const in6_addr_t *gateway, int type, const ipif_t *ipif, ire_t **pire,
1637     zoneid_t zoneid, uint32_t ihandle, const ts_label_t *tsl, int flags)
1638 {
1639 	irb_t *irb_ptr;
1640 	ire_t	*rire;
1641 	ire_t *ire = NULL;
1642 	ire_t	*saved_ire;
1643 	nce_t	*nce;
1644 	int i;
1645 	in6_addr_t gw_addr_v6;
1646 
1647 	ASSERT(addr != NULL);
1648 	ASSERT((!(flags & MATCH_IRE_MASK)) || mask != NULL);
1649 	ASSERT((!(flags & MATCH_IRE_GW)) || gateway != NULL);
1650 	ASSERT(ipif == NULL || ipif->ipif_isv6);
1651 	ASSERT(!(flags & MATCH_IRE_WQ));
1652 
1653 	/*
1654 	 * When we return NULL from this function, we should make
1655 	 * sure that *pire is NULL so that the callers will not
1656 	 * wrongly REFRELE the pire.
1657 	 */
1658 	if (pire != NULL)
1659 		*pire = NULL;
1660 	/*
1661 	 * ire_match_args_v6() will dereference ipif MATCH_IRE_SRC or
1662 	 * MATCH_IRE_ILL is set.
1663 	 */
1664 	if ((flags & (MATCH_IRE_SRC | MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) &&
1665 	    (ipif == NULL))
1666 		return (NULL);
1667 
1668 	/*
1669 	 * If the mask is known, the lookup
1670 	 * is simple, if the mask is not known
1671 	 * we need to search.
1672 	 */
1673 	if (flags & MATCH_IRE_MASK) {
1674 		uint_t masklen;
1675 
1676 		masklen = ip_mask_to_plen_v6(mask);
1677 		if (ip_forwarding_table_v6[masklen] == NULL)
1678 			return (NULL);
1679 		irb_ptr = &(ip_forwarding_table_v6[masklen][
1680 		    IRE_ADDR_MASK_HASH_V6(*addr, *mask, ip6_ftable_hash_size)]);
1681 		rw_enter(&irb_ptr->irb_lock, RW_READER);
1682 		for (ire = irb_ptr->irb_ire; ire != NULL;
1683 		    ire = ire->ire_next) {
1684 			if (ire->ire_marks & IRE_MARK_CONDEMNED)
1685 				continue;
1686 			if (ire_match_args_v6(ire, addr, mask, gateway, type,
1687 			    ipif, zoneid, ihandle, tsl, flags))
1688 				goto found_ire;
1689 		}
1690 		rw_exit(&irb_ptr->irb_lock);
1691 	} else {
1692 		/*
1693 		 * In this case we don't know the mask, we need to
1694 		 * search the table assuming different mask sizes.
1695 		 * we start with 128 bit mask, we don't allow default here.
1696 		 */
1697 		for (i = (IP6_MASK_TABLE_SIZE - 1); i > 0; i--) {
1698 			in6_addr_t tmpmask;
1699 
1700 			if ((ip_forwarding_table_v6[i]) == NULL)
1701 				continue;
1702 			(void) ip_plen_to_mask_v6(i, &tmpmask);
1703 			irb_ptr = &ip_forwarding_table_v6[i][
1704 			    IRE_ADDR_MASK_HASH_V6(*addr, tmpmask,
1705 			    ip6_ftable_hash_size)];
1706 			rw_enter(&irb_ptr->irb_lock, RW_READER);
1707 			for (ire = irb_ptr->irb_ire; ire != NULL;
1708 			    ire = ire->ire_next) {
1709 				if (ire->ire_marks & IRE_MARK_CONDEMNED)
1710 					continue;
1711 				if (ire_match_args_v6(ire, addr,
1712 				    &ire->ire_mask_v6, gateway, type, ipif,
1713 				    zoneid, ihandle, tsl, flags))
1714 					goto found_ire;
1715 			}
1716 			rw_exit(&irb_ptr->irb_lock);
1717 		}
1718 	}
1719 
1720 	/*
1721 	 * We come here if no route has yet been found.
1722 	 *
1723 	 * Handle the case where default route is
1724 	 * requested by specifying type as one of the possible
1725 	 * types for that can have a zero mask (IRE_DEFAULT and IRE_INTERFACE).
1726 	 *
1727 	 * If MATCH_IRE_MASK is specified, then the appropriate default route
1728 	 * would have been found above if it exists so it isn't looked up here.
1729 	 * If MATCH_IRE_DEFAULT was also specified, then a default route will be
1730 	 * searched for later.
1731 	 */
1732 	if ((flags & (MATCH_IRE_TYPE | MATCH_IRE_MASK)) == MATCH_IRE_TYPE &&
1733 	    (type & (IRE_DEFAULT | IRE_INTERFACE))) {
1734 		if (ip_forwarding_table_v6[0] != NULL) {
1735 			/* addr & mask is zero for defaults */
1736 			irb_ptr = &ip_forwarding_table_v6[0][
1737 			    IRE_ADDR_HASH_V6(ipv6_all_zeros,
1738 			    ip6_ftable_hash_size)];
1739 			rw_enter(&irb_ptr->irb_lock, RW_READER);
1740 			for (ire = irb_ptr->irb_ire; ire != NULL;
1741 			    ire = ire->ire_next) {
1742 
1743 				if (ire->ire_marks & IRE_MARK_CONDEMNED)
1744 					continue;
1745 
1746 				if (ire_match_args_v6(ire, addr,
1747 				    &ipv6_all_zeros, gateway, type, ipif,
1748 				    zoneid, ihandle, tsl, flags))
1749 					goto found_ire;
1750 			}
1751 			rw_exit(&irb_ptr->irb_lock);
1752 		}
1753 	}
1754 	/*
1755 	 * We come here only if no route is found.
1756 	 * see if the default route can be used which is allowed
1757 	 * only if the default matching criteria is specified.
1758 	 * The ipv6_ire_default_count tracks the number of IRE_DEFAULT
1759 	 * entries. However, the ip_forwarding_table_v6[0] also contains
1760 	 * interface routes thus the count can be zero.
1761 	 */
1762 	saved_ire = NULL;
1763 	if ((flags & (MATCH_IRE_DEFAULT | MATCH_IRE_MASK)) ==
1764 	    MATCH_IRE_DEFAULT) {
1765 		ire_t	*ire_origin;
1766 		uint_t	g_index;
1767 		uint_t	index;
1768 
1769 		if (ip_forwarding_table_v6[0] == NULL)
1770 			return (NULL);
1771 		irb_ptr = &(ip_forwarding_table_v6[0])[0];
1772 
1773 		/*
1774 		 * Keep a tab on the bucket while looking the IRE_DEFAULT
1775 		 * entries. We need to keep track of a particular IRE
1776 		 * (ire_origin) so this ensures that it will not be unlinked
1777 		 * from the hash list during the recursive lookup below.
1778 		 */
1779 		IRB_REFHOLD(irb_ptr);
1780 		ire = irb_ptr->irb_ire;
1781 		if (ire == NULL) {
1782 			IRB_REFRELE(irb_ptr);
1783 			return (NULL);
1784 		}
1785 
1786 		/*
1787 		 * Get the index first, since it can be changed by other
1788 		 * threads. Then get to the right default route skipping
1789 		 * default interface routes if any. As we hold a reference on
1790 		 * the IRE bucket, ipv6_ire_default_count can only increase so
1791 		 * we can't reach the end of the hash list unexpectedly.
1792 		 */
1793 		if (ipv6_ire_default_count != 0) {
1794 			g_index = ipv6_ire_default_index++;
1795 			index = g_index % ipv6_ire_default_count;
1796 			while (index != 0) {
1797 				if (!(ire->ire_type & IRE_INTERFACE))
1798 					index--;
1799 				ire = ire->ire_next;
1800 			}
1801 			ASSERT(ire != NULL);
1802 		} else {
1803 			/*
1804 			 * No default route, so we only have default interface
1805 			 * routes: don't enter the first loop.
1806 			 */
1807 			ire = NULL;
1808 		}
1809 
1810 		/*
1811 		 * Round-robin the default routers list looking for a neighbor
1812 		 * that matches the passed in parameters and is reachable.  If
1813 		 * none found, just return a route from the default router list
1814 		 * if it exists. If we can't find a default route (IRE_DEFAULT),
1815 		 * look for interface default routes.
1816 		 * We start with the ire we found above and we walk the hash
1817 		 * list until we're back where we started, see
1818 		 * ire_get_next_default_ire(). It doesn't matter if default
1819 		 * routes are added or deleted by other threads - we know this
1820 		 * ire will stay in the list because we hold a reference on the
1821 		 * ire bucket.
1822 		 * NB: if we only have interface default routes, ire is NULL so
1823 		 * we don't even enter this loop (see above).
1824 		 */
1825 		ire_origin = ire;
1826 		for (; ire != NULL;
1827 		    ire = ire_get_next_default_ire(ire, ire_origin)) {
1828 
1829 			if (ire_match_args_v6(ire, addr,
1830 			    &ipv6_all_zeros, gateway, type, ipif,
1831 			    zoneid, ihandle, tsl, flags)) {
1832 				int match_flags;
1833 
1834 				/*
1835 				 * We have something to work with.
1836 				 * If we can find a resolved/reachable
1837 				 * entry, we will use this. Otherwise
1838 				 * we'll try to find an entry that has
1839 				 * a resolved cache entry. We will fallback
1840 				 * on this if we don't find anything else.
1841 				 */
1842 				if (saved_ire == NULL)
1843 					saved_ire = ire;
1844 				mutex_enter(&ire->ire_lock);
1845 				gw_addr_v6 = ire->ire_gateway_addr_v6;
1846 				mutex_exit(&ire->ire_lock);
1847 				match_flags = MATCH_IRE_ILL_GROUP |
1848 				    MATCH_IRE_SECATTR;
1849 				rire = ire_ctable_lookup_v6(&gw_addr_v6, NULL,
1850 				    0, ire->ire_ipif, zoneid, tsl, match_flags);
1851 				if (rire != NULL) {
1852 					nce = rire->ire_nce;
1853 					if (nce != NULL &&
1854 					    NCE_ISREACHABLE(nce) &&
1855 					    nce->nce_flags & NCE_F_ISROUTER) {
1856 						ire_refrele(rire);
1857 						IRE_REFHOLD(ire);
1858 						IRB_REFRELE(irb_ptr);
1859 						goto found_ire_held;
1860 					} else if (nce != NULL &&
1861 					    !(nce->nce_flags &
1862 					    NCE_F_ISROUTER)) {
1863 						/*
1864 						 * Make sure we don't use
1865 						 * this ire
1866 						 */
1867 						if (saved_ire == ire)
1868 							saved_ire = NULL;
1869 					}
1870 					ire_refrele(rire);
1871 				} else if (ipv6_ire_default_count > 1 &&
1872 				    zoneid != ALL_ZONES) {
1873 					/*
1874 					 * When we're in a local zone, we're
1875 					 * only interested in default routers
1876 					 * that are reachable through ipifs
1877 					 * within our zone.
1878 					 * The potentially expensive call to
1879 					 * ire_route_lookup_v6() is avoided when
1880 					 * we have only one default route.
1881 					 */
1882 					match_flags |= MATCH_IRE_TYPE;
1883 					rire = ire_route_lookup_v6(&gw_addr_v6,
1884 					    NULL, NULL, IRE_INTERFACE,
1885 					    ire->ire_ipif, NULL,
1886 					    zoneid, tsl, match_flags);
1887 					if (rire != NULL) {
1888 						ire_refrele(rire);
1889 						saved_ire = ire;
1890 					} else if (saved_ire == ire) {
1891 						/*
1892 						 * Make sure we don't use
1893 						 * this ire
1894 						 */
1895 						saved_ire = NULL;
1896 					}
1897 				}
1898 			}
1899 		}
1900 		if (saved_ire != NULL) {
1901 			ire = saved_ire;
1902 			IRE_REFHOLD(ire);
1903 			IRB_REFRELE(irb_ptr);
1904 			goto found_ire_held;
1905 		} else {
1906 			/*
1907 			 * Look for a interface default route matching the
1908 			 * args passed in. No round robin here. Just pick
1909 			 * the right one.
1910 			 */
1911 			for (ire = irb_ptr->irb_ire; ire != NULL;
1912 			    ire = ire->ire_next) {
1913 
1914 				if (!(ire->ire_type & IRE_INTERFACE))
1915 					continue;
1916 
1917 				if (ire->ire_marks & IRE_MARK_CONDEMNED)
1918 					continue;
1919 
1920 				if (ire_match_args_v6(ire, addr,
1921 				    &ipv6_all_zeros, gateway, type, ipif,
1922 				    zoneid, ihandle, tsl, flags)) {
1923 					IRE_REFHOLD(ire);
1924 					IRB_REFRELE(irb_ptr);
1925 					goto found_ire_held;
1926 				}
1927 			}
1928 			IRB_REFRELE(irb_ptr);
1929 		}
1930 	}
1931 	ASSERT(ire == NULL);
1932 	ip1dbg(("ire_ftable_lookup_v6: returning NULL ire"));
1933 	return (NULL);
1934 found_ire:
1935 	ASSERT((ire->ire_marks & IRE_MARK_CONDEMNED) == 0);
1936 	IRE_REFHOLD(ire);
1937 	rw_exit(&irb_ptr->irb_lock);
1938 
1939 found_ire_held:
1940 	if ((flags & MATCH_IRE_RJ_BHOLE) &&
1941 	    (ire->ire_flags & (RTF_BLACKHOLE | RTF_REJECT))) {
1942 		return (ire);
1943 	}
1944 	/*
1945 	 * At this point, IRE that was found must be an IRE_FORWARDTABLE
1946 	 * or IRE_CACHETABLE type.  If this is a recursive lookup and an
1947 	 * IRE_INTERFACE type was found, return that.  If it was some other
1948 	 * IRE_FORWARDTABLE type of IRE (one of the prefix types), then it
1949 	 * is necessary to fill in the  parent IRE pointed to by pire, and
1950 	 * then lookup the gateway address of  the parent.  For backwards
1951 	 * compatiblity, if this lookup returns an
1952 	 * IRE other than a IRE_CACHETABLE or IRE_INTERFACE, then one more level
1953 	 * of lookup is done.
1954 	 */
1955 	if (flags & MATCH_IRE_RECURSIVE) {
1956 		const ipif_t *gw_ipif;
1957 		int match_flags = MATCH_IRE_DSTONLY;
1958 
1959 		if (ire->ire_type & IRE_INTERFACE)
1960 			return (ire);
1961 		if (pire != NULL)
1962 			*pire = ire;
1963 		/*
1964 		 * If we can't find an IRE_INTERFACE or the caller has not
1965 		 * asked for pire, we need to REFRELE the saved_ire.
1966 		 */
1967 		saved_ire = ire;
1968 
1969 		/*
1970 		 * Currently MATCH_IRE_ILL is never used with
1971 		 * (MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT) while
1972 		 * sending out packets as MATCH_IRE_ILL is used only
1973 		 * for communicating with on-link hosts. We can't assert
1974 		 * that here as RTM_GET calls this function with
1975 		 * MATCH_IRE_ILL | MATCH_IRE_DEFAULT | MATCH_IRE_RECURSIVE.
1976 		 * We have already used the MATCH_IRE_ILL in determining
1977 		 * the right prefix route at this point. To match the
1978 		 * behavior of how we locate routes while sending out
1979 		 * packets, we don't want to use MATCH_IRE_ILL below
1980 		 * while locating the interface route.
1981 		 */
1982 		if (ire->ire_ipif != NULL)
1983 			match_flags |= MATCH_IRE_ILL_GROUP;
1984 
1985 		mutex_enter(&ire->ire_lock);
1986 		gw_addr_v6 = ire->ire_gateway_addr_v6;
1987 		mutex_exit(&ire->ire_lock);
1988 
1989 		ire = ire_route_lookup_v6(&gw_addr_v6, NULL, NULL, 0,
1990 		    ire->ire_ipif, NULL, zoneid, tsl, match_flags);
1991 		if (ire == NULL) {
1992 			/*
1993 			 * In this case we have to deal with the
1994 			 * MATCH_IRE_PARENT flag, which means the
1995 			 * parent has to be returned if ire is NULL.
1996 			 * The aim of this is to have (at least) a starting
1997 			 * ire when we want to look at all of the ires in a
1998 			 * bucket aimed at a single destination (as is the
1999 			 * case in ip_newroute_v6 for the RTF_MULTIRT
2000 			 * flagged routes).
2001 			 */
2002 			if (flags & MATCH_IRE_PARENT) {
2003 				if (pire != NULL) {
2004 					/*
2005 					 * Need an extra REFHOLD, if the
2006 					 * parent ire is returned via both
2007 					 * ire and pire.
2008 					 */
2009 					IRE_REFHOLD(saved_ire);
2010 				}
2011 				ire = saved_ire;
2012 			} else {
2013 				ire_refrele(saved_ire);
2014 				if (pire != NULL)
2015 					*pire = NULL;
2016 			}
2017 			return (ire);
2018 		}
2019 		if (ire->ire_type & (IRE_CACHETABLE | IRE_INTERFACE)) {
2020 			/*
2021 			 * If the caller did not ask for pire, release
2022 			 * it now.
2023 			 */
2024 			if (pire == NULL) {
2025 				ire_refrele(saved_ire);
2026 			}
2027 			return (ire);
2028 		}
2029 		match_flags |= MATCH_IRE_TYPE;
2030 		mutex_enter(&ire->ire_lock);
2031 		gw_addr_v6 = ire->ire_gateway_addr_v6;
2032 		mutex_exit(&ire->ire_lock);
2033 		gw_ipif = ire->ire_ipif;
2034 		ire_refrele(ire);
2035 		ire = ire_route_lookup_v6(&gw_addr_v6, NULL, NULL,
2036 		    (IRE_CACHETABLE | IRE_INTERFACE), gw_ipif, NULL, zoneid,
2037 		    NULL, match_flags);
2038 		if (ire == NULL) {
2039 			/*
2040 			 * In this case we have to deal with the
2041 			 * MATCH_IRE_PARENT flag, which means the
2042 			 * parent has to be returned if ire is NULL.
2043 			 * The aim of this is to have (at least) a starting
2044 			 * ire when we want to look at all of the ires in a
2045 			 * bucket aimed at a single destination (as is the
2046 			 * case in ip_newroute_v6 for the RTF_MULTIRT
2047 			 * flagged routes).
2048 			 */
2049 			if (flags & MATCH_IRE_PARENT) {
2050 				if (pire != NULL) {
2051 					/*
2052 					 * Need an extra REFHOLD, if the
2053 					 * parent ire is returned via both
2054 					 * ire and pire.
2055 					 */
2056 					IRE_REFHOLD(saved_ire);
2057 				}
2058 				ire = saved_ire;
2059 			} else {
2060 				ire_refrele(saved_ire);
2061 				if (pire != NULL)
2062 					*pire = NULL;
2063 			}
2064 			return (ire);
2065 		} else if (pire == NULL) {
2066 			/*
2067 			 * If the caller did not ask for pire, release
2068 			 * it now.
2069 			 */
2070 			ire_refrele(saved_ire);
2071 		}
2072 		return (ire);
2073 	}
2074 
2075 	ASSERT(pire == NULL || *pire == NULL);
2076 	return (ire);
2077 }
2078 
2079 /*
2080  * Delete the IRE cache for the gateway and all IRE caches whose
2081  * ire_gateway_addr_v6 points to this gateway, and allow them to
2082  * be created on demand by ip_newroute_v6.
2083  */
2084 void
2085 ire_clookup_delete_cache_gw_v6(const in6_addr_t *addr, zoneid_t zoneid)
2086 {
2087 	irb_t *irb;
2088 	ire_t *ire;
2089 
2090 	irb = &ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr, ip6_cache_table_size)];
2091 	IRB_REFHOLD(irb);
2092 	for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
2093 		if (ire->ire_marks & IRE_MARK_CONDEMNED)
2094 			continue;
2095 
2096 		ASSERT(IN6_ARE_ADDR_EQUAL(&ire->ire_mask_v6, &ipv6_all_ones));
2097 		if (ire_match_args_v6(ire, addr, &ire->ire_mask_v6, 0,
2098 		    IRE_CACHE, NULL, zoneid, 0, NULL, MATCH_IRE_TYPE)) {
2099 			ire_delete(ire);
2100 		}
2101 	}
2102 	IRB_REFRELE(irb);
2103 
2104 	ire_walk_v6(ire_delete_cache_gw_v6, (char *)addr, zoneid);
2105 }
2106 
2107 /*
2108  * Looks up cache table for a route.
2109  * specific lookup can be indicated by
2110  * passing the MATCH_* flags and the
2111  * necessary parameters.
2112  */
2113 ire_t *
2114 ire_ctable_lookup_v6(const in6_addr_t *addr, const in6_addr_t *gateway,
2115     int type, const ipif_t *ipif, zoneid_t zoneid, const ts_label_t *tsl,
2116     int flags)
2117 {
2118 	ire_t *ire;
2119 	irb_t *irb_ptr;
2120 	ASSERT(addr != NULL);
2121 	ASSERT((!(flags & MATCH_IRE_GW)) || gateway != NULL);
2122 
2123 	/*
2124 	 * ire_match_args_v6() will dereference ipif MATCH_IRE_SRC or
2125 	 * MATCH_IRE_ILL is set.
2126 	 */
2127 	if ((flags & (MATCH_IRE_SRC |  MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) &&
2128 	    (ipif == NULL))
2129 		return (NULL);
2130 
2131 	irb_ptr = &ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr,
2132 	    ip6_cache_table_size)];
2133 	rw_enter(&irb_ptr->irb_lock, RW_READER);
2134 	for (ire = irb_ptr->irb_ire; ire; ire = ire->ire_next) {
2135 		if (ire->ire_marks & IRE_MARK_CONDEMNED)
2136 			continue;
2137 
2138 		ASSERT(IN6_ARE_ADDR_EQUAL(&ire->ire_mask_v6, &ipv6_all_ones));
2139 		if (ire_match_args_v6(ire, addr, &ire->ire_mask_v6, gateway,
2140 		    type, ipif, zoneid, 0, tsl, flags)) {
2141 			IRE_REFHOLD(ire);
2142 			rw_exit(&irb_ptr->irb_lock);
2143 			return (ire);
2144 		}
2145 	}
2146 	rw_exit(&irb_ptr->irb_lock);
2147 	return (NULL);
2148 }
2149 
2150 /*
2151  * Lookup cache. Don't return IRE_MARK_HIDDEN entries. Callers
2152  * should use ire_ctable_lookup with MATCH_IRE_MARK_HIDDEN to get
2153  * to the hidden ones.
2154  *
2155  * In general the zoneid has to match (where ALL_ZONES match all of them).
2156  * But for IRE_LOCAL we also need to handle the case where L2 should
2157  * conceptually loop back the packet. This is necessary since neither
2158  * Ethernet drivers nor Ethernet hardware loops back packets sent to their
2159  * own MAC address. This loopback is needed when the normal
2160  * routes (ignoring IREs with different zoneids) would send out the packet on
2161  * the same ill (or ill group) as the ill with which this IRE_LOCAL is
2162  * associated.
2163  *
2164  * Earlier versions of this code always matched an IRE_LOCAL independently of
2165  * the zoneid. We preserve that earlier behavior when
2166  * ip_restrict_interzone_loopback is turned off.
2167  */
2168 ire_t *
2169 ire_cache_lookup_v6(const in6_addr_t *addr, zoneid_t zoneid,
2170     const ts_label_t *tsl)
2171 {
2172 	irb_t *irb_ptr;
2173 	ire_t *ire;
2174 
2175 	irb_ptr = &ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr,
2176 	    ip6_cache_table_size)];
2177 	rw_enter(&irb_ptr->irb_lock, RW_READER);
2178 	for (ire = irb_ptr->irb_ire; ire; ire = ire->ire_next) {
2179 		if (ire->ire_marks & (IRE_MARK_CONDEMNED|IRE_MARK_HIDDEN))
2180 			continue;
2181 		if (IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, addr)) {
2182 			/*
2183 			 * Finally, check if the security policy has any
2184 			 * restriction on using this route for the specified
2185 			 * message.
2186 			 */
2187 			if (tsl != NULL &&
2188 			    ire->ire_gw_secattr != NULL &&
2189 			    tsol_ire_match_gwattr(ire, tsl) != 0) {
2190 				continue;
2191 			}
2192 
2193 			if (zoneid == ALL_ZONES || ire->ire_zoneid == zoneid ||
2194 			    ire->ire_zoneid == ALL_ZONES) {
2195 				IRE_REFHOLD(ire);
2196 				rw_exit(&irb_ptr->irb_lock);
2197 				return (ire);
2198 			}
2199 
2200 			if (ire->ire_type == IRE_LOCAL) {
2201 				if (ip_restrict_interzone_loopback &&
2202 				    !ire_local_ok_across_zones(ire, zoneid,
2203 				    (void *)addr, tsl))
2204 					continue;
2205 
2206 				IRE_REFHOLD(ire);
2207 				rw_exit(&irb_ptr->irb_lock);
2208 				return (ire);
2209 			}
2210 		}
2211 	}
2212 	rw_exit(&irb_ptr->irb_lock);
2213 	return (NULL);
2214 }
2215 
2216 /*
2217  * Locate the interface ire that is tied to the cache ire 'cire' via
2218  * cire->ire_ihandle.
2219  *
2220  * We are trying to create the cache ire for an onlink destn. or
2221  * gateway in 'cire'. We are called from ire_add_v6() in the IRE_IF_RESOLVER
2222  * case for xresolv interfaces, after the ire has come back from
2223  * an external resolver.
2224  */
2225 static ire_t *
2226 ire_ihandle_lookup_onlink_v6(ire_t *cire)
2227 {
2228 	ire_t	*ire;
2229 	int	match_flags;
2230 	int	i;
2231 	int	j;
2232 	irb_t	*irb_ptr;
2233 
2234 	ASSERT(cire != NULL);
2235 
2236 	match_flags =  MATCH_IRE_TYPE | MATCH_IRE_IHANDLE | MATCH_IRE_MASK;
2237 	/*
2238 	 * We know that the mask of the interface ire equals cire->ire_cmask.
2239 	 * (When ip_newroute_v6() created 'cire' for an on-link destn.
2240 	 * it set its cmask from the interface ire's mask)
2241 	 */
2242 	ire = ire_ftable_lookup_v6(&cire->ire_addr_v6, &cire->ire_cmask_v6,
2243 	    NULL, IRE_INTERFACE, NULL, NULL, ALL_ZONES, cire->ire_ihandle,
2244 	    NULL, match_flags);
2245 	if (ire != NULL)
2246 		return (ire);
2247 	/*
2248 	 * If we didn't find an interface ire above, we can't declare failure.
2249 	 * For backwards compatibility, we need to support prefix routes
2250 	 * pointing to next hop gateways that are not on-link.
2251 	 *
2252 	 * In the resolver/noresolver case, ip_newroute_v6() thinks
2253 	 * it is creating the cache ire for an onlink destination in 'cire'.
2254 	 * But 'cire' is not actually onlink, because ire_ftable_lookup_v6()
2255 	 * cheated it, by doing ire_route_lookup_v6() twice and returning an
2256 	 * interface ire.
2257 	 *
2258 	 * Eg. default	-	gw1			(line 1)
2259 	 *	gw1	-	gw2			(line 2)
2260 	 *	gw2	-	hme0			(line 3)
2261 	 *
2262 	 * In the above example, ip_newroute_v6() tried to create the cache ire
2263 	 * 'cire' for gw1, based on the interface route in line 3. The
2264 	 * ire_ftable_lookup_v6() above fails, because there is
2265 	 * no interface route to reach gw1. (it is gw2). We fall thru below.
2266 	 *
2267 	 * Do a brute force search based on the ihandle in a subset of the
2268 	 * forwarding tables, corresponding to cire->ire_cmask_v6. Otherwise
2269 	 * things become very complex, since we don't have 'pire' in this
2270 	 * case. (Also note that this method is not possible in the offlink
2271 	 * case because we don't know the mask)
2272 	 */
2273 	i = ip_mask_to_plen_v6(&cire->ire_cmask_v6);
2274 	if ((ip_forwarding_table_v6[i]) == NULL)
2275 		return (NULL);
2276 	for (j = 0; j < ip6_ftable_hash_size; j++) {
2277 		irb_ptr = &ip_forwarding_table_v6[i][j];
2278 		rw_enter(&irb_ptr->irb_lock, RW_READER);
2279 		for (ire = irb_ptr->irb_ire; ire != NULL;
2280 		    ire = ire->ire_next) {
2281 			if (ire->ire_marks & IRE_MARK_CONDEMNED)
2282 				continue;
2283 			if ((ire->ire_type & IRE_INTERFACE) &&
2284 			    (ire->ire_ihandle == cire->ire_ihandle)) {
2285 				IRE_REFHOLD(ire);
2286 				rw_exit(&irb_ptr->irb_lock);
2287 				return (ire);
2288 			}
2289 		}
2290 		rw_exit(&irb_ptr->irb_lock);
2291 	}
2292 	return (NULL);
2293 }
2294 
2295 
2296 /*
2297  * Locate the interface ire that is tied to the cache ire 'cire' via
2298  * cire->ire_ihandle.
2299  *
2300  * We are trying to create the cache ire for an offlink destn based
2301  * on the cache ire of the gateway in 'cire'. 'pire' is the prefix ire
2302  * as found by ip_newroute_v6(). We are called from ip_newroute_v6() in
2303  * the IRE_CACHE case.
2304  */
2305 ire_t *
2306 ire_ihandle_lookup_offlink_v6(ire_t *cire, ire_t *pire)
2307 {
2308 	ire_t	*ire;
2309 	int	match_flags;
2310 	in6_addr_t	gw_addr;
2311 	ipif_t		*gw_ipif;
2312 
2313 	ASSERT(cire != NULL && pire != NULL);
2314 
2315 	match_flags =  MATCH_IRE_TYPE | MATCH_IRE_IHANDLE | MATCH_IRE_MASK;
2316 	/*
2317 	 * ip_newroute_v6 calls ire_ftable_lookup with MATCH_IRE_ILL only
2318 	 * for on-link hosts. We should never be here for onlink.
2319 	 * Thus, use MATCH_IRE_ILL_GROUP.
2320 	 */
2321 	if (pire->ire_ipif != NULL)
2322 		match_flags |= MATCH_IRE_ILL_GROUP;
2323 	/*
2324 	 * We know that the mask of the interface ire equals cire->ire_cmask.
2325 	 * (When ip_newroute_v6() created 'cire' for an on-link destn. it set
2326 	 * its cmask from the interface ire's mask)
2327 	 */
2328 	ire = ire_ftable_lookup_v6(&cire->ire_addr_v6, &cire->ire_cmask_v6, 0,
2329 	    IRE_INTERFACE, pire->ire_ipif, NULL, ALL_ZONES, cire->ire_ihandle,
2330 	    NULL, match_flags);
2331 	if (ire != NULL)
2332 		return (ire);
2333 	/*
2334 	 * If we didn't find an interface ire above, we can't declare failure.
2335 	 * For backwards compatibility, we need to support prefix routes
2336 	 * pointing to next hop gateways that are not on-link.
2337 	 *
2338 	 * Assume we are trying to ping some offlink destn, and we have the
2339 	 * routing table below.
2340 	 *
2341 	 * Eg.	default	- gw1		<--- pire	(line 1)
2342 	 *	gw1	- gw2				(line 2)
2343 	 *	gw2	- hme0				(line 3)
2344 	 *
2345 	 * If we already have a cache ire for gw1 in 'cire', the
2346 	 * ire_ftable_lookup_v6 above would have failed, since there is no
2347 	 * interface ire to reach gw1. We will fallthru below.
2348 	 *
2349 	 * Here we duplicate the steps that ire_ftable_lookup_v6() did in
2350 	 * getting 'cire' from 'pire', in the MATCH_IRE_RECURSIVE case.
2351 	 * The differences are the following
2352 	 * i.   We want the interface ire only, so we call
2353 	 *	ire_ftable_lookup_v6() instead of ire_route_lookup_v6()
2354 	 * ii.  We look for only prefix routes in the 1st call below.
2355 	 * ii.  We want to match on the ihandle in the 2nd call below.
2356 	 */
2357 	match_flags =  MATCH_IRE_TYPE;
2358 	if (pire->ire_ipif != NULL)
2359 		match_flags |= MATCH_IRE_ILL_GROUP;
2360 
2361 	mutex_enter(&pire->ire_lock);
2362 	gw_addr = pire->ire_gateway_addr_v6;
2363 	mutex_exit(&pire->ire_lock);
2364 	ire = ire_ftable_lookup_v6(&gw_addr, 0, 0, IRE_OFFSUBNET,
2365 	    pire->ire_ipif, NULL, ALL_ZONES, 0, NULL, match_flags);
2366 	if (ire == NULL)
2367 		return (NULL);
2368 	/*
2369 	 * At this point 'ire' corresponds to the entry shown in line 2.
2370 	 * gw_addr is 'gw2' in the example above.
2371 	 */
2372 	mutex_enter(&ire->ire_lock);
2373 	gw_addr = ire->ire_gateway_addr_v6;
2374 	mutex_exit(&ire->ire_lock);
2375 	gw_ipif = ire->ire_ipif;
2376 	ire_refrele(ire);
2377 
2378 	match_flags |= MATCH_IRE_IHANDLE;
2379 	ire = ire_ftable_lookup_v6(&gw_addr, 0, 0, IRE_INTERFACE,
2380 	    gw_ipif, NULL, ALL_ZONES, cire->ire_ihandle,
2381 	    NULL, match_flags);
2382 	return (ire);
2383 }
2384 
2385 /*
2386  * Return the IRE_LOOPBACK, IRE_IF_RESOLVER or IRE_IF_NORESOLVER
2387  * ire associated with the specified ipif.
2388  *
2389  * This might occasionally be called when IPIF_UP is not set since
2390  * the IPV6_MULTICAST_IF as well as creating interface routes
2391  * allows specifying a down ipif (ipif_lookup* match ipifs that are down).
2392  *
2393  * Note that if IPIF_NOLOCAL, IPIF_NOXMIT, or IPIF_DEPRECATED is set on
2394  * the ipif this routine might return NULL.
2395  * (Sometimes called as writer though not required by this function.)
2396  */
2397 ire_t *
2398 ipif_to_ire_v6(const ipif_t *ipif)
2399 {
2400 	ire_t	*ire;
2401 
2402 	ASSERT(ipif->ipif_isv6);
2403 	if (ipif->ipif_ire_type == IRE_LOOPBACK) {
2404 		ire = ire_ctable_lookup_v6(&ipif->ipif_v6lcl_addr, NULL,
2405 		    IRE_LOOPBACK, ipif, ALL_ZONES, NULL,
2406 		    (MATCH_IRE_TYPE | MATCH_IRE_IPIF));
2407 	} else if (ipif->ipif_flags & IPIF_POINTOPOINT) {
2408 		/* In this case we need to lookup destination address. */
2409 		ire = ire_ftable_lookup_v6(&ipif->ipif_v6pp_dst_addr,
2410 		    &ipv6_all_ones, NULL, IRE_INTERFACE, ipif, NULL, ALL_ZONES,
2411 		    0, NULL, (MATCH_IRE_TYPE | MATCH_IRE_IPIF |
2412 		    MATCH_IRE_MASK));
2413 	} else {
2414 		ire = ire_ftable_lookup_v6(&ipif->ipif_v6subnet,
2415 		    &ipif->ipif_v6net_mask, NULL, IRE_INTERFACE, ipif, NULL,
2416 		    ALL_ZONES, 0, NULL, (MATCH_IRE_TYPE | MATCH_IRE_IPIF |
2417 		    MATCH_IRE_MASK));
2418 	}
2419 	return (ire);
2420 }
2421 
2422 /*
2423  * Return B_TRUE if a multirt route is resolvable
2424  * (or if no route is resolved yet), B_FALSE otherwise.
2425  * This only works in the global zone.
2426  */
2427 boolean_t
2428 ire_multirt_need_resolve_v6(const in6_addr_t *v6dstp, const ts_label_t *tsl)
2429 {
2430 	ire_t	*first_fire;
2431 	ire_t	*first_cire;
2432 	ire_t	*fire;
2433 	ire_t	*cire;
2434 	irb_t	*firb;
2435 	irb_t	*cirb;
2436 	int	unres_cnt = 0;
2437 	boolean_t resolvable = B_FALSE;
2438 
2439 	/* Retrieve the first IRE_HOST that matches the destination */
2440 	first_fire = ire_ftable_lookup_v6(v6dstp, &ipv6_all_ones, 0, IRE_HOST,
2441 	    NULL, NULL, ALL_ZONES, 0, tsl, MATCH_IRE_MASK | MATCH_IRE_TYPE |
2442 	    MATCH_IRE_SECATTR);
2443 
2444 	/* No route at all */
2445 	if (first_fire == NULL) {
2446 		return (B_TRUE);
2447 	}
2448 
2449 	firb = first_fire->ire_bucket;
2450 	ASSERT(firb);
2451 
2452 	/* Retrieve the first IRE_CACHE ire for that destination. */
2453 	first_cire = ire_cache_lookup_v6(v6dstp, GLOBAL_ZONEID, tsl);
2454 
2455 	/* No resolved route. */
2456 	if (first_cire == NULL) {
2457 		ire_refrele(first_fire);
2458 		return (B_TRUE);
2459 	}
2460 
2461 	/* At least one route is resolved. */
2462 
2463 	cirb = first_cire->ire_bucket;
2464 	ASSERT(cirb);
2465 
2466 	/* Count the number of routes to that dest that are declared. */
2467 	IRB_REFHOLD(firb);
2468 	for (fire = first_fire; fire != NULL; fire = fire->ire_next) {
2469 		if (!(fire->ire_flags & RTF_MULTIRT))
2470 			continue;
2471 		if (!IN6_ARE_ADDR_EQUAL(&fire->ire_addr_v6, v6dstp))
2472 			continue;
2473 		unres_cnt++;
2474 	}
2475 	IRB_REFRELE(firb);
2476 
2477 
2478 	/* Then subtract the number of routes to that dst that are resolved */
2479 	IRB_REFHOLD(cirb);
2480 	for (cire = first_cire; cire != NULL; cire = cire->ire_next) {
2481 	    if (!(cire->ire_flags & RTF_MULTIRT))
2482 		continue;
2483 	    if (!IN6_ARE_ADDR_EQUAL(&cire->ire_addr_v6, v6dstp))
2484 		continue;
2485 	    if (cire->ire_marks & (IRE_MARK_CONDEMNED|IRE_MARK_HIDDEN))
2486 		continue;
2487 	    unres_cnt--;
2488 	}
2489 	IRB_REFRELE(cirb);
2490 
2491 	/* At least one route is unresolved; search for a resolvable route. */
2492 	if (unres_cnt > 0)
2493 		resolvable = ire_multirt_lookup_v6(&first_cire, &first_fire,
2494 		    MULTIRT_USESTAMP|MULTIRT_CACHEGW, tsl);
2495 
2496 	if (first_fire)
2497 		ire_refrele(first_fire);
2498 
2499 	if (first_cire)
2500 		ire_refrele(first_cire);
2501 
2502 	return (resolvable);
2503 }
2504 
2505 
2506 /*
2507  * Return B_TRUE and update *ire_arg and *fire_arg
2508  * if at least one resolvable route is found.
2509  * Return B_FALSE otherwise (all routes are resolved or
2510  * the remaining unresolved routes are all unresolvable).
2511  * This only works in the global zone.
2512  */
2513 boolean_t
2514 ire_multirt_lookup_v6(ire_t **ire_arg, ire_t **fire_arg, uint32_t flags,
2515     const ts_label_t *tsl)
2516 {
2517 	clock_t	delta;
2518 	ire_t	*best_fire = NULL;
2519 	ire_t	*best_cire = NULL;
2520 	ire_t	*first_fire;
2521 	ire_t	*first_cire;
2522 	ire_t	*fire;
2523 	ire_t	*cire;
2524 	irb_t	*firb = NULL;
2525 	irb_t	*cirb = NULL;
2526 	ire_t	*gw_ire;
2527 	boolean_t	already_resolved;
2528 	boolean_t	res;
2529 	in6_addr_t	v6dst;
2530 	in6_addr_t	v6gw;
2531 
2532 	ip2dbg(("ire_multirt_lookup_v6: *ire_arg %p, *fire_arg %p, "
2533 	    "flags %04x\n", (void *)*ire_arg, (void *)*fire_arg, flags));
2534 
2535 	ASSERT(ire_arg);
2536 	ASSERT(fire_arg);
2537 
2538 	/* Not an IRE_HOST ire; give up. */
2539 	if ((*fire_arg == NULL) ||
2540 	    ((*fire_arg)->ire_type != IRE_HOST)) {
2541 		return (B_FALSE);
2542 	}
2543 
2544 	/* This is the first IRE_HOST ire for that destination. */
2545 	first_fire = *fire_arg;
2546 	firb = first_fire->ire_bucket;
2547 	ASSERT(firb);
2548 
2549 	mutex_enter(&first_fire->ire_lock);
2550 	v6dst = first_fire->ire_addr_v6;
2551 	mutex_exit(&first_fire->ire_lock);
2552 
2553 	ip2dbg(("ire_multirt_lookup_v6: dst %08x\n",
2554 	    ntohl(V4_PART_OF_V6(v6dst))));
2555 
2556 	/*
2557 	 * Retrieve the first IRE_CACHE ire for that destination;
2558 	 * if we don't find one, no route for that dest is
2559 	 * resolved yet.
2560 	 */
2561 	first_cire = ire_cache_lookup_v6(&v6dst, GLOBAL_ZONEID, tsl);
2562 	if (first_cire) {
2563 		cirb = first_cire->ire_bucket;
2564 	}
2565 
2566 	ip2dbg(("ire_multirt_lookup_v6: first_cire %p\n", (void *)first_cire));
2567 
2568 	/*
2569 	 * Search for a resolvable route, giving the top priority
2570 	 * to routes that can be resolved without any call to the resolver.
2571 	 */
2572 	IRB_REFHOLD(firb);
2573 
2574 	if (!IN6_IS_ADDR_MULTICAST(&v6dst)) {
2575 		/*
2576 		 * For all multiroute IRE_HOST ires for that destination,
2577 		 * check if the route via the IRE_HOST's gateway is
2578 		 * resolved yet.
2579 		 */
2580 		for (fire = first_fire; fire != NULL; fire = fire->ire_next) {
2581 
2582 			if (!(fire->ire_flags & RTF_MULTIRT))
2583 				continue;
2584 			if (!IN6_ARE_ADDR_EQUAL(&fire->ire_addr_v6, &v6dst))
2585 				continue;
2586 
2587 			if (fire->ire_gw_secattr != NULL &&
2588 			    tsol_ire_match_gwattr(fire, tsl) != 0) {
2589 				continue;
2590 			}
2591 
2592 			mutex_enter(&fire->ire_lock);
2593 			v6gw = fire->ire_gateway_addr_v6;
2594 			mutex_exit(&fire->ire_lock);
2595 
2596 			ip2dbg(("ire_multirt_lookup_v6: fire %p, "
2597 			    "ire_addr %08x, ire_gateway_addr %08x\n",
2598 			    (void *)fire,
2599 			    ntohl(V4_PART_OF_V6(fire->ire_addr_v6)),
2600 			    ntohl(V4_PART_OF_V6(v6gw))));
2601 
2602 			already_resolved = B_FALSE;
2603 
2604 			if (first_cire) {
2605 				ASSERT(cirb);
2606 
2607 				IRB_REFHOLD(cirb);
2608 				/*
2609 				 * For all IRE_CACHE ires for that
2610 				 * destination.
2611 				 */
2612 				for (cire = first_cire;
2613 				    cire != NULL;
2614 				    cire = cire->ire_next) {
2615 
2616 					if (!(cire->ire_flags & RTF_MULTIRT))
2617 						continue;
2618 					if (!IN6_ARE_ADDR_EQUAL(
2619 					    &cire->ire_addr_v6, &v6dst))
2620 						continue;
2621 					if (cire->ire_marks &
2622 					    (IRE_MARK_CONDEMNED|
2623 						IRE_MARK_HIDDEN))
2624 						continue;
2625 
2626 					if (cire->ire_gw_secattr != NULL &&
2627 					    tsol_ire_match_gwattr(cire,
2628 					    tsl) != 0) {
2629 						continue;
2630 					}
2631 
2632 					/*
2633 					 * Check if the IRE_CACHE's gateway
2634 					 * matches the IRE_HOST's gateway.
2635 					 */
2636 					if (IN6_ARE_ADDR_EQUAL(
2637 					    &cire->ire_gateway_addr_v6,
2638 					    &v6gw)) {
2639 						already_resolved = B_TRUE;
2640 						break;
2641 					}
2642 				}
2643 				IRB_REFRELE(cirb);
2644 			}
2645 
2646 			/*
2647 			 * This route is already resolved;
2648 			 * proceed with next one.
2649 			 */
2650 			if (already_resolved) {
2651 				ip2dbg(("ire_multirt_lookup_v6: found cire %p, "
2652 				    "already resolved\n", (void *)cire));
2653 				continue;
2654 			}
2655 
2656 			/*
2657 			 * The route is unresolved; is it actually
2658 			 * resolvable, i.e. is there a cache or a resolver
2659 			 * for the gateway?
2660 			 */
2661 			gw_ire = ire_route_lookup_v6(&v6gw, 0, 0, 0, NULL, NULL,
2662 			    ALL_ZONES, tsl, MATCH_IRE_RECURSIVE |
2663 			    MATCH_IRE_SECATTR);
2664 
2665 			ip2dbg(("ire_multirt_lookup_v6: looked up gw_ire %p\n",
2666 			    (void *)gw_ire));
2667 
2668 			/*
2669 			 * This route can be resolved without any call to the
2670 			 * resolver; if the MULTIRT_CACHEGW flag is set,
2671 			 * give the top priority to this ire and exit the
2672 			 * loop.
2673 			 * This occurs when an resolver reply is processed
2674 			 * through ip_wput_nondata()
2675 			 */
2676 			if ((flags & MULTIRT_CACHEGW) &&
2677 			    (gw_ire != NULL) &&
2678 			    (gw_ire->ire_type & IRE_CACHETABLE)) {
2679 				/*
2680 				 * Release the resolver associated to the
2681 				 * previous candidate best ire, if any.
2682 				 */
2683 				if (best_cire) {
2684 					ire_refrele(best_cire);
2685 					ASSERT(best_fire);
2686 				}
2687 
2688 				best_fire = fire;
2689 				best_cire = gw_ire;
2690 
2691 				ip2dbg(("ire_multirt_lookup_v6: found top prio "
2692 				    "best_fire %p, best_cire %p\n",
2693 				    (void *)best_fire, (void *)best_cire));
2694 				break;
2695 			}
2696 
2697 			/*
2698 			 * Compute the time elapsed since our preceding
2699 			 * attempt to  resolve that route.
2700 			 * If the MULTIRT_USESTAMP flag is set, we take that
2701 			 * route into account only if this time interval
2702 			 * exceeds ip_multirt_resolution_interval;
2703 			 * this prevents us from attempting to resolve a
2704 			 * broken route upon each sending of a packet.
2705 			 */
2706 			delta = lbolt - fire->ire_last_used_time;
2707 			delta = TICK_TO_MSEC(delta);
2708 
2709 			res = (boolean_t)
2710 			    ((delta > ip_multirt_resolution_interval) ||
2711 				(!(flags & MULTIRT_USESTAMP)));
2712 
2713 			ip2dbg(("ire_multirt_lookup_v6: fire %p, delta %lu, "
2714 			    "res %d\n",
2715 			    (void *)fire, delta, res));
2716 
2717 			if (res) {
2718 				/*
2719 				 * A resolver exists for the gateway: save
2720 				 * the current IRE_HOST ire as a candidate
2721 				 * best ire. If we later discover that a
2722 				 * top priority ire exists (i.e. no need to
2723 				 * call the resolver), then this new ire
2724 				 * will be preferred to the current one.
2725 				 */
2726 				if (gw_ire != NULL) {
2727 					if (best_fire == NULL) {
2728 						ASSERT(best_cire == NULL);
2729 
2730 						best_fire = fire;
2731 						best_cire = gw_ire;
2732 
2733 						ip2dbg(("ire_multirt_lookup_v6:"
2734 						    "found candidate "
2735 						    "best_fire %p, "
2736 						    "best_cire %p\n",
2737 						    (void *)best_fire,
2738 						    (void *)best_cire));
2739 
2740 						/*
2741 						 * If MULTIRT_CACHEGW is not
2742 						 * set, we ignore the top
2743 						 * priority ires that can
2744 						 * be resolved without any
2745 						 * call to the resolver;
2746 						 * In that case, there is
2747 						 * actually no need
2748 						 * to continue the loop.
2749 						 */
2750 						if (!(flags &
2751 						    MULTIRT_CACHEGW)) {
2752 							break;
2753 						}
2754 						continue;
2755 					}
2756 				} else {
2757 					/*
2758 					 * No resolver for the gateway: the
2759 					 * route is not resolvable.
2760 					 * If the MULTIRT_SETSTAMP flag is
2761 					 * set, we stamp the IRE_HOST ire,
2762 					 * so we will not select it again
2763 					 * during this resolution interval.
2764 					 */
2765 					if (flags & MULTIRT_SETSTAMP)
2766 						fire->ire_last_used_time =
2767 						    lbolt;
2768 				}
2769 			}
2770 
2771 			if (gw_ire != NULL)
2772 				ire_refrele(gw_ire);
2773 		}
2774 	} else { /* IN6_IS_ADDR_MULTICAST(&v6dst) */
2775 
2776 		for (fire = first_fire;
2777 		    fire != NULL;
2778 		    fire = fire->ire_next) {
2779 
2780 			if (!(fire->ire_flags & RTF_MULTIRT))
2781 				continue;
2782 			if (!IN6_ARE_ADDR_EQUAL(&fire->ire_addr_v6, &v6dst))
2783 				continue;
2784 
2785 			if (fire->ire_gw_secattr != NULL &&
2786 			    tsol_ire_match_gwattr(fire, tsl) != 0) {
2787 				continue;
2788 			}
2789 
2790 			already_resolved = B_FALSE;
2791 
2792 			mutex_enter(&fire->ire_lock);
2793 			v6gw = fire->ire_gateway_addr_v6;
2794 			mutex_exit(&fire->ire_lock);
2795 
2796 			gw_ire = ire_ftable_lookup_v6(&v6gw, 0, 0,
2797 			    IRE_INTERFACE, NULL, NULL, ALL_ZONES, 0, tsl,
2798 			    MATCH_IRE_RECURSIVE | MATCH_IRE_TYPE |
2799 			    MATCH_IRE_SECATTR);
2800 
2801 			/* No resolver for the gateway; we skip this ire. */
2802 			if (gw_ire == NULL) {
2803 				continue;
2804 			}
2805 
2806 			if (first_cire) {
2807 
2808 				IRB_REFHOLD(cirb);
2809 				/*
2810 				 * For all IRE_CACHE ires for that
2811 				 * destination.
2812 				 */
2813 				for (cire = first_cire;
2814 				    cire != NULL;
2815 				    cire = cire->ire_next) {
2816 
2817 					if (!(cire->ire_flags & RTF_MULTIRT))
2818 						continue;
2819 					if (!IN6_ARE_ADDR_EQUAL(
2820 					    &cire->ire_addr_v6, &v6dst))
2821 						continue;
2822 					if (cire->ire_marks &
2823 					    (IRE_MARK_CONDEMNED|
2824 						IRE_MARK_HIDDEN))
2825 						continue;
2826 
2827 					if (cire->ire_gw_secattr != NULL &&
2828 					    tsol_ire_match_gwattr(cire,
2829 					    tsl) != 0) {
2830 						continue;
2831 					}
2832 
2833 					/*
2834 					 * Cache entries are linked to the
2835 					 * parent routes using the parent handle
2836 					 * (ire_phandle). If no cache entry has
2837 					 * the same handle as fire, fire is
2838 					 * still unresolved.
2839 					 */
2840 					ASSERT(cire->ire_phandle != 0);
2841 					if (cire->ire_phandle ==
2842 					    fire->ire_phandle) {
2843 						already_resolved = B_TRUE;
2844 						break;
2845 					}
2846 				}
2847 				IRB_REFRELE(cirb);
2848 			}
2849 
2850 			/*
2851 			 * This route is already resolved; proceed with
2852 			 * next one.
2853 			 */
2854 			if (already_resolved) {
2855 				ire_refrele(gw_ire);
2856 				continue;
2857 			}
2858 
2859 			/*
2860 			 * Compute the time elapsed since our preceding
2861 			 * attempt to resolve that route.
2862 			 * If the MULTIRT_USESTAMP flag is set, we take
2863 			 * that route into account only if this time
2864 			 * interval exceeds ip_multirt_resolution_interval;
2865 			 * this prevents us from attempting to resolve a
2866 			 * broken route upon each sending of a packet.
2867 			 */
2868 			delta = lbolt - fire->ire_last_used_time;
2869 			delta = TICK_TO_MSEC(delta);
2870 
2871 			res = (boolean_t)
2872 			    ((delta > ip_multirt_resolution_interval) ||
2873 			    (!(flags & MULTIRT_USESTAMP)));
2874 
2875 			ip3dbg(("ire_multirt_lookup_v6: fire %p, delta %lx, "
2876 			    "flags %04x, res %d\n",
2877 			    (void *)fire, delta, flags, res));
2878 
2879 			if (res) {
2880 				if (best_cire) {
2881 					/*
2882 					 * Release the resolver associated
2883 					 * to the preceding candidate best
2884 					 * ire, if any.
2885 					 */
2886 					ire_refrele(best_cire);
2887 					ASSERT(best_fire);
2888 				}
2889 				best_fire = fire;
2890 				best_cire = gw_ire;
2891 				continue;
2892 			}
2893 
2894 			ire_refrele(gw_ire);
2895 		}
2896 	}
2897 
2898 	if (best_fire) {
2899 		IRE_REFHOLD(best_fire);
2900 	}
2901 	IRB_REFRELE(firb);
2902 
2903 	/* Release the first IRE_CACHE we initially looked up, if any. */
2904 	if (first_cire)
2905 		ire_refrele(first_cire);
2906 
2907 	/* Found a resolvable route. */
2908 	if (best_fire) {
2909 		ASSERT(best_cire);
2910 
2911 		if (*fire_arg)
2912 			ire_refrele(*fire_arg);
2913 		if (*ire_arg)
2914 			ire_refrele(*ire_arg);
2915 
2916 		/*
2917 		 * Update the passed arguments with the
2918 		 * resolvable multirt route we found
2919 		 */
2920 		*fire_arg = best_fire;
2921 		*ire_arg = best_cire;
2922 
2923 		ip2dbg(("ire_multirt_lookup_v6: returning B_TRUE, "
2924 		    "*fire_arg %p, *ire_arg %p\n",
2925 		    (void *)best_fire, (void *)best_cire));
2926 
2927 		return (B_TRUE);
2928 	}
2929 
2930 	ASSERT(best_cire == NULL);
2931 
2932 	ip2dbg(("ire_multirt_lookup_v6: returning B_FALSE, *fire_arg %p, "
2933 	    "*ire_arg %p\n",
2934 	    (void *)*fire_arg, (void *)*ire_arg));
2935 
2936 	/* No resolvable route. */
2937 	return (B_FALSE);
2938 }
2939 
2940 
2941 /*
2942  * Find an IRE_OFFSUBNET IRE entry for the multicast address 'v6dstp'
2943  * that goes through 'ipif'. As a fallback, a route that goes through
2944  * ipif->ipif_ill can be returned.
2945  */
2946 ire_t *
2947 ipif_lookup_multi_ire_v6(ipif_t *ipif, const in6_addr_t *v6dstp)
2948 {
2949 	ire_t	*ire;
2950 	ire_t	*save_ire = NULL;
2951 	ire_t   *gw_ire;
2952 	irb_t   *irb;
2953 	in6_addr_t v6gw;
2954 	int	match_flags = MATCH_IRE_TYPE | MATCH_IRE_ILL;
2955 
2956 	ire = ire_ftable_lookup_v6(v6dstp, 0, 0, 0, NULL, NULL, ALL_ZONES, 0,
2957 	    NULL, MATCH_IRE_DEFAULT);
2958 
2959 	if (ire == NULL)
2960 		return (NULL);
2961 
2962 	irb = ire->ire_bucket;
2963 	ASSERT(irb);
2964 
2965 	IRB_REFHOLD(irb);
2966 	ire_refrele(ire);
2967 	for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
2968 		if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6dstp) ||
2969 		    (ipif->ipif_zoneid != ire->ire_zoneid &&
2970 		    ire->ire_zoneid != ALL_ZONES)) {
2971 			continue;
2972 		}
2973 
2974 		switch (ire->ire_type) {
2975 		case IRE_DEFAULT:
2976 		case IRE_PREFIX:
2977 		case IRE_HOST:
2978 			mutex_enter(&ire->ire_lock);
2979 			v6gw = ire->ire_gateway_addr_v6;
2980 			mutex_exit(&ire->ire_lock);
2981 			gw_ire = ire_ftable_lookup_v6(&v6gw, 0, 0,
2982 			    IRE_INTERFACE, ipif, NULL, ALL_ZONES, 0,
2983 			    NULL, match_flags);
2984 
2985 			if (gw_ire != NULL) {
2986 				if (save_ire != NULL) {
2987 					ire_refrele(save_ire);
2988 				}
2989 				IRE_REFHOLD(ire);
2990 				if (gw_ire->ire_ipif == ipif) {
2991 					ire_refrele(gw_ire);
2992 
2993 					IRB_REFRELE(irb);
2994 					return (ire);
2995 				}
2996 				ire_refrele(gw_ire);
2997 				save_ire = ire;
2998 			}
2999 			break;
3000 		case IRE_IF_NORESOLVER:
3001 		case IRE_IF_RESOLVER:
3002 			if (ire->ire_ipif == ipif) {
3003 				if (save_ire != NULL) {
3004 					ire_refrele(save_ire);
3005 				}
3006 				IRE_REFHOLD(ire);
3007 
3008 				IRB_REFRELE(irb);
3009 				return (ire);
3010 			}
3011 			break;
3012 		}
3013 	}
3014 	IRB_REFRELE(irb);
3015 
3016 	return (save_ire);
3017 }
3018