xref: /titanic_52/usr/src/uts/common/inet/ip/ip6_ire.c (revision 038dc6b3d39a14038605fbb3fc6a9bb76dd27fa8)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /*
26  * Copyright (c) 1990 Mentat Inc.
27  */
28 
29 #pragma ident	"%Z%%M%	%I%	%E% SMI"
30 
31 /*
32  * This file contains routines that manipulate Internet Routing Entries (IREs).
33  */
34 #include <sys/types.h>
35 #include <sys/stream.h>
36 #include <sys/stropts.h>
37 #include <sys/ddi.h>
38 #include <sys/cmn_err.h>
39 
40 #include <sys/systm.h>
41 #include <sys/param.h>
42 #include <sys/socket.h>
43 #include <net/if.h>
44 #include <net/route.h>
45 #include <netinet/in.h>
46 #include <net/if_dl.h>
47 #include <netinet/ip6.h>
48 #include <netinet/icmp6.h>
49 
50 #include <inet/common.h>
51 #include <inet/mi.h>
52 #include <inet/ip.h>
53 #include <inet/ip6.h>
54 #include <inet/ip_ndp.h>
55 #include <inet/ip_if.h>
56 #include <inet/ip_ire.h>
57 #include <inet/ipclassifier.h>
58 #include <inet/nd.h>
59 #include <sys/kmem.h>
60 #include <sys/zone.h>
61 
62 #include <sys/tsol/label.h>
63 #include <sys/tsol/tnet.h>
64 
65 static	ire_t	ire_null;
66 
67 static ire_t	*ire_ihandle_lookup_onlink_v6(ire_t *cire);
68 static	void	ire_report_ftable_v6(ire_t *ire, char *mp);
69 static	void	ire_report_ctable_v6(ire_t *ire, char *mp);
70 static boolean_t ire_match_args_v6(ire_t *ire, const in6_addr_t *addr,
71     const in6_addr_t *mask, const in6_addr_t *gateway, int type,
72     const ipif_t *ipif, zoneid_t zoneid, uint32_t ihandle,
73     const ts_label_t *tsl, int match_flags);
74 static	ire_t	*ire_init_v6(ire_t *, const in6_addr_t *, const in6_addr_t *,
75     const in6_addr_t *, const in6_addr_t *, uint_t *, queue_t *, queue_t *,
76     ushort_t, ipif_t *, const in6_addr_t *, uint32_t, uint32_t, uint_t,
77     const iulp_t *, tsol_gc_t *, tsol_gcgrp_t *, ip_stack_t *);
78 
79 /*
80  * Named Dispatch routine to produce a formatted report on all IREs.
81  * This report is accessed by using the ndd utility to "get" ND variable
82  * "ip_ire_status_v6".
83  */
84 /* ARGSUSED */
85 int
86 ip_ire_report_v6(queue_t *q, mblk_t *mp, caddr_t arg, cred_t *ioc_cr)
87 {
88 	zoneid_t zoneid;
89 	ip_stack_t *ipst;
90 
91 	(void) mi_mpprintf(mp,
92 	    "IRE      " MI_COL_HDRPAD_STR
93 	    "rfq      " MI_COL_HDRPAD_STR
94 	    "stq      " MI_COL_HDRPAD_STR
95 	    " zone mxfrg rtt   rtt_sd ssthresh ref "
96 	    "rtomax tstamp_ok wscale_ok ecn_ok pmtud_ok sack sendpipe recvpipe "
97 	    "in/out/forward type    addr         mask         "
98 	    "src             gateway");
99 	/*
100 	 *   01234567 01234567 01234567 12345 12345 12345 12345  12345678 123
101 	 *   123456 123456789 123456789 123456 12345678 1234 12345678 12345678
102 	 *   in/out/forward xxxxxxxxxx
103 	 *   xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx
104 	 *   xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx
105 	 *   xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx
106 	 *   xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx:xxxx
107 	 */
108 
109 	/*
110 	 * Because of the ndd constraint, at most we can have 64K buffer
111 	 * to put in all IRE info.  So to be more efficient, just
112 	 * allocate a 64K buffer here, assuming we need that large buffer.
113 	 * This should be OK as only root can do ndd /dev/ip.
114 	 */
115 	if ((mp->b_cont = allocb(ND_MAX_BUF_LEN, BPRI_HI)) == NULL) {
116 		/* The following may work even if we cannot get a large buf. */
117 		(void) mi_mpprintf(mp, "<< Out of buffer >>\n");
118 		return (0);
119 	}
120 	zoneid = Q_TO_CONN(q)->conn_zoneid;
121 	if (zoneid == GLOBAL_ZONEID)
122 		zoneid = ALL_ZONES;
123 	ipst = CONNQ_TO_IPST(q);
124 
125 	ire_walk_v6(ire_report_ftable_v6, (char *)mp->b_cont, zoneid, ipst);
126 	ire_walk_v6(ire_report_ctable_v6, (char *)mp->b_cont, zoneid, ipst);
127 	return (0);
128 }
129 
130 /*
131  * ire_walk routine invoked for ip_ire_report_v6 for each IRE.
132  */
133 static void
134 ire_report_ftable_v6(ire_t *ire, char *mp)
135 {
136 	char	buf1[INET6_ADDRSTRLEN];
137 	char	buf2[INET6_ADDRSTRLEN];
138 	char	buf3[INET6_ADDRSTRLEN];
139 	char	buf4[INET6_ADDRSTRLEN];
140 	uint_t	fo_pkt_count;
141 	uint_t	ib_pkt_count;
142 	int	ref;
143 	in6_addr_t gw_addr_v6;
144 	uint_t	print_len, buf_len;
145 
146 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
147 	if (ire->ire_type & IRE_CACHETABLE)
148 		return;
149 	buf_len = ((mblk_t *)mp)->b_datap->db_lim - ((mblk_t *)mp)->b_wptr;
150 	if (buf_len <= 0)
151 		return;
152 
153 	/* Number of active references of this ire */
154 	ref = ire->ire_refcnt;
155 	/* "inbound" to a non local address is a forward */
156 	ib_pkt_count = ire->ire_ib_pkt_count;
157 	fo_pkt_count = 0;
158 	ASSERT(!(ire->ire_type & IRE_BROADCAST));
159 	if (!(ire->ire_type & (IRE_LOCAL|IRE_BROADCAST))) {
160 		fo_pkt_count = ib_pkt_count;
161 		ib_pkt_count = 0;
162 	}
163 
164 	mutex_enter(&ire->ire_lock);
165 	gw_addr_v6 = ire->ire_gateway_addr_v6;
166 	mutex_exit(&ire->ire_lock);
167 
168 	print_len = snprintf((char *)((mblk_t *)mp)->b_wptr, buf_len,
169 	    MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR "%5d "
170 	    "%05d %05ld %06ld %08d %03d %06d %09d %09d %06d %08d "
171 	    "%04d %08d %08d %d/%d/%d %s\n\t%s\n\t%s\n\t%s\n\t%s\n",
172 	    (void *)ire, (void *)ire->ire_rfq, (void *)ire->ire_stq,
173 	    (int)ire->ire_zoneid,
174 	    ire->ire_max_frag, ire->ire_uinfo.iulp_rtt,
175 	    ire->ire_uinfo.iulp_rtt_sd,
176 	    ire->ire_uinfo.iulp_ssthresh, ref,
177 	    ire->ire_uinfo.iulp_rtomax,
178 	    (ire->ire_uinfo.iulp_tstamp_ok ? 1: 0),
179 	    (ire->ire_uinfo.iulp_wscale_ok ? 1: 0),
180 	    (ire->ire_uinfo.iulp_ecn_ok ? 1: 0),
181 	    (ire->ire_uinfo.iulp_pmtud_ok ? 1: 0),
182 	    ire->ire_uinfo.iulp_sack,
183 	    ire->ire_uinfo.iulp_spipe, ire->ire_uinfo.iulp_rpipe,
184 	    ib_pkt_count, ire->ire_ob_pkt_count, fo_pkt_count,
185 	    ip_nv_lookup(ire_nv_tbl, (int)ire->ire_type),
186 	    inet_ntop(AF_INET6, &ire->ire_addr_v6, buf1, sizeof (buf1)),
187 	    inet_ntop(AF_INET6, &ire->ire_mask_v6, buf2, sizeof (buf2)),
188 	    inet_ntop(AF_INET6, &ire->ire_src_addr_v6, buf3, sizeof (buf3)),
189 	    inet_ntop(AF_INET6, &gw_addr_v6, buf4, sizeof (buf4)));
190 	if (print_len < buf_len) {
191 		((mblk_t *)mp)->b_wptr += print_len;
192 	} else {
193 		((mblk_t *)mp)->b_wptr += buf_len;
194 	}
195 }
196 
197 /* ire_walk routine invoked for ip_ire_report_v6 for each IRE. */
198 static void
199 ire_report_ctable_v6(ire_t *ire, char *mp)
200 {
201 	char	buf1[INET6_ADDRSTRLEN];
202 	char	buf2[INET6_ADDRSTRLEN];
203 	char	buf3[INET6_ADDRSTRLEN];
204 	char	buf4[INET6_ADDRSTRLEN];
205 	uint_t	fo_pkt_count;
206 	uint_t	ib_pkt_count;
207 	int	ref;
208 	in6_addr_t gw_addr_v6;
209 	uint_t	print_len, buf_len;
210 
211 	if ((ire->ire_type & IRE_CACHETABLE) == 0)
212 		return;
213 	buf_len = ((mblk_t *)mp)->b_datap->db_lim - ((mblk_t *)mp)->b_wptr;
214 	if (buf_len <= 0)
215 		return;
216 
217 	/* Number of active references of this ire */
218 	ref = ire->ire_refcnt;
219 	/* "inbound" to a non local address is a forward */
220 	ib_pkt_count = ire->ire_ib_pkt_count;
221 	fo_pkt_count = 0;
222 	ASSERT(!(ire->ire_type & IRE_BROADCAST));
223 	if (ire->ire_type & IRE_LOCAL) {
224 		fo_pkt_count = ib_pkt_count;
225 		ib_pkt_count = 0;
226 	}
227 
228 	mutex_enter(&ire->ire_lock);
229 	gw_addr_v6 = ire->ire_gateway_addr_v6;
230 	mutex_exit(&ire->ire_lock);
231 
232 	print_len =  snprintf((char *)((mblk_t *)mp)->b_wptr, buf_len,
233 	    MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR "%5d "
234 	    "%05d %05ld %06ld %08d %03d %06d %09d %09d %06d %08d "
235 	    "%04d %08d %08d %d/%d/%d %s\n\t%s\n\t%s\n\t%s\n\t%s\n",
236 	    (void *)ire, (void *)ire->ire_rfq, (void *)ire->ire_stq,
237 	    (int)ire->ire_zoneid,
238 	    ire->ire_max_frag, ire->ire_uinfo.iulp_rtt,
239 	    ire->ire_uinfo.iulp_rtt_sd, ire->ire_uinfo.iulp_ssthresh, ref,
240 	    ire->ire_uinfo.iulp_rtomax,
241 	    (ire->ire_uinfo.iulp_tstamp_ok ? 1: 0),
242 	    (ire->ire_uinfo.iulp_wscale_ok ? 1: 0),
243 	    (ire->ire_uinfo.iulp_ecn_ok ? 1: 0),
244 	    (ire->ire_uinfo.iulp_pmtud_ok ? 1: 0),
245 	    ire->ire_uinfo.iulp_sack,
246 	    ire->ire_uinfo.iulp_spipe, ire->ire_uinfo.iulp_rpipe,
247 	    ib_pkt_count, ire->ire_ob_pkt_count,
248 	    fo_pkt_count, ip_nv_lookup(ire_nv_tbl, (int)ire->ire_type),
249 	    inet_ntop(AF_INET6, &ire->ire_addr_v6, buf1, sizeof (buf1)),
250 	    inet_ntop(AF_INET6, &ire->ire_mask_v6, buf2, sizeof (buf2)),
251 	    inet_ntop(AF_INET6, &ire->ire_src_addr_v6, buf3, sizeof (buf3)),
252 	    inet_ntop(AF_INET6, &gw_addr_v6, buf4, sizeof (buf4)));
253 	if (print_len < buf_len) {
254 		((mblk_t *)mp)->b_wptr += print_len;
255 	} else {
256 		((mblk_t *)mp)->b_wptr += buf_len;
257 	}
258 }
259 
260 
261 /*
262  * Initialize the ire that is specific to IPv6 part and call
263  * ire_init_common to finish it.
264  */
265 static ire_t *
266 ire_init_v6(ire_t *ire, const in6_addr_t *v6addr, const in6_addr_t *v6mask,
267     const in6_addr_t *v6src_addr, const in6_addr_t *v6gateway,
268     uint_t *max_fragp, queue_t *rfq, queue_t *stq, ushort_t type,
269     ipif_t *ipif, const in6_addr_t *v6cmask, uint32_t phandle,
270     uint32_t ihandle, uint_t flags, const iulp_t *ulp_info, tsol_gc_t *gc,
271     tsol_gcgrp_t *gcgrp, ip_stack_t *ipst)
272 {
273 
274 	/*
275 	 * Reject IRE security attribute creation/initialization
276 	 * if system is not running in Trusted mode.
277 	 */
278 	if ((gc != NULL || gcgrp != NULL) && !is_system_labeled())
279 		return (NULL);
280 
281 
282 	BUMP_IRE_STATS(ipst->ips_ire_stats_v6, ire_stats_alloced);
283 	ire->ire_addr_v6 = *v6addr;
284 
285 	if (v6src_addr != NULL)
286 		ire->ire_src_addr_v6 = *v6src_addr;
287 	if (v6mask != NULL) {
288 		ire->ire_mask_v6 = *v6mask;
289 		ire->ire_masklen = ip_mask_to_plen_v6(&ire->ire_mask_v6);
290 	}
291 	if (v6gateway != NULL)
292 		ire->ire_gateway_addr_v6 = *v6gateway;
293 
294 	if (type == IRE_CACHE && v6cmask != NULL)
295 		ire->ire_cmask_v6 = *v6cmask;
296 
297 	/*
298 	 * Multirouted packets need to have a fragment header added so that
299 	 * the receiver is able to discard duplicates according to their
300 	 * fragment identifier.
301 	 */
302 	if (type == IRE_CACHE && (flags & RTF_MULTIRT)) {
303 		ire->ire_frag_flag = IPH_FRAG_HDR;
304 	}
305 
306 	/* ire_init_common will free the mblks upon encountering any failure */
307 	if (!ire_init_common(ire, max_fragp, NULL, rfq, stq, type,
308 	    ipif, NULL, phandle, ihandle, flags, IPV6_VERSION, ulp_info,
309 	    gc, gcgrp, ipst))
310 		return (NULL);
311 
312 	return (ire);
313 }
314 
315 /*
316  * Similar to ire_create_v6 except that it is called only when
317  * we want to allocate ire as an mblk e.g. we have a external
318  * resolver. Do we need this in IPv6 ?
319  *
320  * IPv6 initializes the ire_nce in ire_add_v6, which expects to
321  * find the ire_nce to be null when it is called. So, although
322  * we have a src_nce parameter (in the interest of matching up with
323  * the argument list of the v4 version), we ignore the src_nce
324  * argument here.
325  */
326 /* ARGSUSED */
327 ire_t *
328 ire_create_mp_v6(const in6_addr_t *v6addr, const in6_addr_t *v6mask,
329     const in6_addr_t *v6src_addr, const in6_addr_t *v6gateway,
330     nce_t *src_nce, queue_t *rfq, queue_t *stq, ushort_t type,
331     ipif_t *ipif, const in6_addr_t *v6cmask,
332     uint32_t phandle, uint32_t ihandle, uint_t flags, const iulp_t *ulp_info,
333     tsol_gc_t *gc, tsol_gcgrp_t *gcgrp, ip_stack_t *ipst)
334 {
335 	ire_t	*ire;
336 	ire_t	*ret_ire;
337 	mblk_t	*mp;
338 
339 	ASSERT(!IN6_IS_ADDR_V4MAPPED(v6addr));
340 
341 	/* Allocate the new IRE. */
342 	mp = allocb(sizeof (ire_t), BPRI_MED);
343 	if (mp == NULL) {
344 		ip1dbg(("ire_create_mp_v6: alloc failed\n"));
345 		return (NULL);
346 	}
347 
348 	ire = (ire_t *)mp->b_rptr;
349 	mp->b_wptr = (uchar_t *)&ire[1];
350 
351 	/* Start clean. */
352 	*ire = ire_null;
353 	ire->ire_mp = mp;
354 	mp->b_datap->db_type = IRE_DB_TYPE;
355 
356 	ret_ire = ire_init_v6(ire, v6addr, v6mask, v6src_addr, v6gateway,
357 	    NULL, rfq, stq, type, ipif, v6cmask, phandle,
358 	    ihandle, flags, ulp_info, gc, gcgrp, ipst);
359 
360 	if (ret_ire == NULL) {
361 		freeb(ire->ire_mp);
362 		return (NULL);
363 	}
364 	return (ire);
365 }
366 
367 /*
368  * ire_create_v6 is called to allocate and initialize a new IRE.
369  *
370  * NOTE : This is called as writer sometimes though not required
371  * by this function.
372  *
373  * See comments above ire_create_mp_v6() for the rationale behind the
374  * unused src_nce argument.
375  */
376 /* ARGSUSED */
377 ire_t *
378 ire_create_v6(const in6_addr_t *v6addr, const in6_addr_t *v6mask,
379     const in6_addr_t *v6src_addr, const in6_addr_t *v6gateway,
380     uint_t *max_fragp, nce_t *src_nce, queue_t *rfq, queue_t *stq,
381     ushort_t type, ipif_t *ipif, const in6_addr_t *v6cmask,
382     uint32_t phandle, uint32_t ihandle, uint_t flags, const iulp_t *ulp_info,
383     tsol_gc_t *gc, tsol_gcgrp_t *gcgrp, ip_stack_t *ipst)
384 {
385 	ire_t	*ire;
386 	ire_t	*ret_ire;
387 
388 	ASSERT(!IN6_IS_ADDR_V4MAPPED(v6addr));
389 
390 	ire = kmem_cache_alloc(ire_cache, KM_NOSLEEP);
391 	if (ire == NULL) {
392 		ip1dbg(("ire_create_v6: alloc failed\n"));
393 		return (NULL);
394 	}
395 	*ire = ire_null;
396 
397 	ret_ire = ire_init_v6(ire, v6addr, v6mask, v6src_addr, v6gateway,
398 	    max_fragp, rfq, stq, type, ipif, v6cmask, phandle,
399 	    ihandle, flags, ulp_info, gc, gcgrp, ipst);
400 
401 	if (ret_ire == NULL) {
402 		kmem_cache_free(ire_cache, ire);
403 		return (NULL);
404 	}
405 	ASSERT(ret_ire == ire);
406 	return (ire);
407 }
408 
409 /*
410  * Find an IRE_INTERFACE for the multicast group.
411  * Allows different routes for multicast addresses
412  * in the unicast routing table (akin to FF::0/8 but could be more specific)
413  * which point at different interfaces. This is used when IPV6_MULTICAST_IF
414  * isn't specified (when sending) and when IPV6_JOIN_GROUP doesn't
415  * specify the interface to join on.
416  *
417  * Supports link-local addresses by following the ipif/ill when recursing.
418  */
419 ire_t *
420 ire_lookup_multi_v6(const in6_addr_t *group, zoneid_t zoneid, ip_stack_t *ipst)
421 {
422 	ire_t	*ire;
423 	ipif_t	*ipif = NULL;
424 	int	match_flags = MATCH_IRE_TYPE;
425 	in6_addr_t gw_addr_v6;
426 
427 	ire = ire_ftable_lookup_v6(group, 0, 0, 0, NULL, NULL,
428 	    zoneid, 0, NULL, MATCH_IRE_DEFAULT, ipst);
429 
430 	/* We search a resolvable ire in case of multirouting. */
431 	if ((ire != NULL) && (ire->ire_flags & RTF_MULTIRT)) {
432 		ire_t *cire = NULL;
433 		/*
434 		 * If the route is not resolvable, the looked up ire
435 		 * may be changed here. In that case, ire_multirt_lookup()
436 		 * IRE_REFRELE the original ire and change it.
437 		 */
438 		(void) ire_multirt_lookup_v6(&cire, &ire, MULTIRT_CACHEGW,
439 		    NULL, ipst);
440 		if (cire != NULL)
441 			ire_refrele(cire);
442 	}
443 	if (ire == NULL)
444 		return (NULL);
445 	/*
446 	 * Make sure we follow ire_ipif.
447 	 *
448 	 * We need to determine the interface route through
449 	 * which the gateway will be reached. We don't really
450 	 * care which interface is picked if the interface is
451 	 * part of a group.
452 	 */
453 	if (ire->ire_ipif != NULL) {
454 		ipif = ire->ire_ipif;
455 		match_flags |= MATCH_IRE_ILL_GROUP;
456 	}
457 
458 	switch (ire->ire_type) {
459 	case IRE_DEFAULT:
460 	case IRE_PREFIX:
461 	case IRE_HOST:
462 		mutex_enter(&ire->ire_lock);
463 		gw_addr_v6 = ire->ire_gateway_addr_v6;
464 		mutex_exit(&ire->ire_lock);
465 		ire_refrele(ire);
466 		ire = ire_ftable_lookup_v6(&gw_addr_v6, 0, 0,
467 		    IRE_INTERFACE, ipif, NULL, zoneid, 0,
468 		    NULL, match_flags, ipst);
469 		return (ire);
470 	case IRE_IF_NORESOLVER:
471 	case IRE_IF_RESOLVER:
472 		return (ire);
473 	default:
474 		ire_refrele(ire);
475 		return (NULL);
476 	}
477 }
478 
479 /*
480  * Return any local address.  We use this to target ourselves
481  * when the src address was specified as 'default'.
482  * Preference for IRE_LOCAL entries.
483  */
484 ire_t *
485 ire_lookup_local_v6(zoneid_t zoneid, ip_stack_t *ipst)
486 {
487 	ire_t	*ire;
488 	irb_t	*irb;
489 	ire_t	*maybe = NULL;
490 	int i;
491 
492 	for (i = 0; i < ipst->ips_ip6_cache_table_size;  i++) {
493 		irb = &ipst->ips_ip_cache_table_v6[i];
494 		if (irb->irb_ire == NULL)
495 			continue;
496 		rw_enter(&irb->irb_lock, RW_READER);
497 		for (ire = irb->irb_ire; ire; ire = ire->ire_next) {
498 			if ((ire->ire_marks & IRE_MARK_CONDEMNED) ||
499 			    ire->ire_zoneid != zoneid &&
500 			    ire->ire_zoneid != ALL_ZONES)
501 				continue;
502 			switch (ire->ire_type) {
503 			case IRE_LOOPBACK:
504 				if (maybe == NULL) {
505 					IRE_REFHOLD(ire);
506 					maybe = ire;
507 				}
508 				break;
509 			case IRE_LOCAL:
510 				if (maybe != NULL) {
511 					ire_refrele(maybe);
512 				}
513 				IRE_REFHOLD(ire);
514 				rw_exit(&irb->irb_lock);
515 				return (ire);
516 			}
517 		}
518 		rw_exit(&irb->irb_lock);
519 	}
520 	return (maybe);
521 }
522 
523 /*
524  * This function takes a mask and returns number of bits set in the
525  * mask (the represented prefix length).  Assumes a contiguous mask.
526  */
527 int
528 ip_mask_to_plen_v6(const in6_addr_t *v6mask)
529 {
530 	int		bits;
531 	int		plen = IPV6_ABITS;
532 	int		i;
533 
534 	for (i = 3; i >= 0; i--) {
535 		if (v6mask->s6_addr32[i] == 0) {
536 			plen -= 32;
537 			continue;
538 		}
539 		bits = ffs(ntohl(v6mask->s6_addr32[i])) - 1;
540 		if (bits == 0)
541 			break;
542 		plen -= bits;
543 	}
544 
545 	return (plen);
546 }
547 
548 /*
549  * Convert a prefix length to the mask for that prefix.
550  * Returns the argument bitmask.
551  */
552 in6_addr_t *
553 ip_plen_to_mask_v6(uint_t plen, in6_addr_t *bitmask)
554 {
555 	uint32_t *ptr;
556 
557 	if (plen < 0 || plen > IPV6_ABITS)
558 		return (NULL);
559 	*bitmask = ipv6_all_zeros;
560 
561 	ptr = (uint32_t *)bitmask;
562 	while (plen > 32) {
563 		*ptr++ = 0xffffffffU;
564 		plen -= 32;
565 	}
566 	*ptr = htonl(0xffffffffU << (32 - plen));
567 	return (bitmask);
568 }
569 
570 /*
571  * Add a fully initialized IRE to an appropriate
572  * table based on ire_type.
573  *
574  * The forward table contains IRE_PREFIX/IRE_HOST/IRE_HOST and
575  * IRE_IF_RESOLVER/IRE_IF_NORESOLVER and IRE_DEFAULT.
576  *
577  * The cache table contains IRE_BROADCAST/IRE_LOCAL/IRE_LOOPBACK
578  * and IRE_CACHE.
579  *
580  * NOTE : This function is called as writer though not required
581  * by this function.
582  */
583 int
584 ire_add_v6(ire_t **ire_p, queue_t *q, mblk_t *mp, ipsq_func_t func)
585 {
586 	ire_t	*ire1;
587 	int	mask_table_index;
588 	irb_t	*irb_ptr;
589 	ire_t	**irep;
590 	int	flags;
591 	ire_t	*pire = NULL;
592 	ill_t	*stq_ill;
593 	boolean_t	ndp_g_lock_held = B_FALSE;
594 	ire_t	*ire = *ire_p;
595 	int	error;
596 	ip_stack_t	*ipst = ire->ire_ipst;
597 
598 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
599 	ASSERT(ire->ire_mp == NULL); /* Calls should go through ire_add */
600 	ASSERT(ire->ire_nce == NULL);
601 
602 	/* Find the appropriate list head. */
603 	switch (ire->ire_type) {
604 	case IRE_HOST:
605 		ire->ire_mask_v6 = ipv6_all_ones;
606 		ire->ire_masklen = IPV6_ABITS;
607 		if ((ire->ire_flags & RTF_SETSRC) == 0)
608 			ire->ire_src_addr_v6 = ipv6_all_zeros;
609 		break;
610 	case IRE_CACHE:
611 	case IRE_LOCAL:
612 	case IRE_LOOPBACK:
613 		ire->ire_mask_v6 = ipv6_all_ones;
614 		ire->ire_masklen = IPV6_ABITS;
615 		break;
616 	case IRE_PREFIX:
617 		if ((ire->ire_flags & RTF_SETSRC) == 0)
618 			ire->ire_src_addr_v6 = ipv6_all_zeros;
619 		break;
620 	case IRE_DEFAULT:
621 		if ((ire->ire_flags & RTF_SETSRC) == 0)
622 			ire->ire_src_addr_v6 = ipv6_all_zeros;
623 		break;
624 	case IRE_IF_RESOLVER:
625 	case IRE_IF_NORESOLVER:
626 		break;
627 	default:
628 		printf("ire_add_v6: ire %p has unrecognized IRE type (%d)\n",
629 		    (void *)ire, ire->ire_type);
630 		ire_delete(ire);
631 		*ire_p = NULL;
632 		return (EINVAL);
633 	}
634 
635 	/* Make sure the address is properly masked. */
636 	V6_MASK_COPY(ire->ire_addr_v6, ire->ire_mask_v6, ire->ire_addr_v6);
637 
638 	if ((ire->ire_type & IRE_CACHETABLE) == 0) {
639 		/* IRE goes into Forward Table */
640 		mask_table_index = ip_mask_to_plen_v6(&ire->ire_mask_v6);
641 		if ((ipst->ips_ip_forwarding_table_v6[mask_table_index]) ==
642 		    NULL) {
643 			irb_t *ptr;
644 			int i;
645 
646 			ptr = (irb_t *)mi_zalloc((
647 			    ipst->ips_ip6_ftable_hash_size * sizeof (irb_t)));
648 			if (ptr == NULL) {
649 				ire_delete(ire);
650 				*ire_p = NULL;
651 				return (ENOMEM);
652 			}
653 			for (i = 0; i < ipst->ips_ip6_ftable_hash_size; i++) {
654 				rw_init(&ptr[i].irb_lock, NULL,
655 				    RW_DEFAULT, NULL);
656 			}
657 			mutex_enter(&ipst->ips_ire_ft_init_lock);
658 			if (ipst->ips_ip_forwarding_table_v6[
659 			    mask_table_index] == NULL) {
660 				ipst->ips_ip_forwarding_table_v6[
661 				    mask_table_index] = ptr;
662 				mutex_exit(&ipst->ips_ire_ft_init_lock);
663 			} else {
664 				/*
665 				 * Some other thread won the race in
666 				 * initializing the forwarding table at the
667 				 * same index.
668 				 */
669 				mutex_exit(&ipst->ips_ire_ft_init_lock);
670 				for (i = 0; i < ipst->ips_ip6_ftable_hash_size;
671 				    i++) {
672 					rw_destroy(&ptr[i].irb_lock);
673 				}
674 				mi_free(ptr);
675 			}
676 		}
677 		irb_ptr = &(ipst->ips_ip_forwarding_table_v6[mask_table_index][
678 		    IRE_ADDR_MASK_HASH_V6(ire->ire_addr_v6, ire->ire_mask_v6,
679 		    ipst->ips_ip6_ftable_hash_size)]);
680 	} else {
681 		irb_ptr = &(ipst->ips_ip_cache_table_v6[IRE_ADDR_HASH_V6(
682 		    ire->ire_addr_v6, ipst->ips_ip6_cache_table_size)]);
683 	}
684 	/*
685 	 * For xresolv interfaces (v6 interfaces with an external
686 	 * address resolver), ip_newroute_v6/ip_newroute_ipif_v6
687 	 * are unable to prevent the deletion of the interface route
688 	 * while adding an IRE_CACHE for an on-link destination
689 	 * in the IRE_IF_RESOLVER case, since the ire has to go to
690 	 * the external resolver and return. We can't do a REFHOLD on the
691 	 * associated interface ire for fear of the message being freed
692 	 * if the external resolver can't resolve the address.
693 	 * Here we look up the interface ire in the forwarding table
694 	 * and make sure that the interface route has not been deleted.
695 	 */
696 	if (ire->ire_type == IRE_CACHE &&
697 	    IN6_IS_ADDR_UNSPECIFIED(&ire->ire_gateway_addr_v6) &&
698 	    (((ill_t *)ire->ire_stq->q_ptr)->ill_net_type == IRE_IF_RESOLVER) &&
699 	    (((ill_t *)ire->ire_stq->q_ptr)->ill_flags & ILLF_XRESOLV)) {
700 
701 		pire = ire_ihandle_lookup_onlink_v6(ire);
702 		if (pire == NULL) {
703 			ire_delete(ire);
704 			*ire_p = NULL;
705 			return (EINVAL);
706 		}
707 		/* Prevent pire from getting deleted */
708 		IRB_REFHOLD(pire->ire_bucket);
709 		/* Has it been removed already? */
710 		if (pire->ire_marks & IRE_MARK_CONDEMNED) {
711 			IRB_REFRELE(pire->ire_bucket);
712 			ire_refrele(pire);
713 			ire_delete(ire);
714 			*ire_p = NULL;
715 			return (EINVAL);
716 		}
717 	}
718 
719 	flags = (MATCH_IRE_MASK | MATCH_IRE_TYPE | MATCH_IRE_GW);
720 	/*
721 	 * For IRE_CACHES, MATCH_IRE_IPIF is not enough to check
722 	 * for duplicates because :
723 	 *
724 	 * 1) ire_ipif->ipif_ill and ire_stq->q_ptr could be
725 	 *    pointing at different ills. A real duplicate is
726 	 *    a match on both ire_ipif and ire_stq.
727 	 *
728 	 * 2) We could have multiple packets trying to create
729 	 *    an IRE_CACHE for the same ill.
730 	 *
731 	 * Moreover, IPIF_NOFAILOVER and IPV6_BOUND_PIF endpoints wants
732 	 * to go out on a particular ill. Rather than looking at the
733 	 * packet, we depend on the above for MATCH_IRE_ILL here.
734 	 *
735 	 * Unlike IPv4, MATCH_IRE_IPIF is needed here as we could have
736 	 * multiple IRE_CACHES for an ill for the same destination
737 	 * with various scoped addresses i.e represented by ipifs.
738 	 *
739 	 * MATCH_IRE_ILL is done implicitly below for IRE_CACHES.
740 	 */
741 	if (ire->ire_ipif != NULL)
742 		flags |= MATCH_IRE_IPIF;
743 	/*
744 	 * If we are creating hidden ires, make sure we search on
745 	 * this ill (MATCH_IRE_ILL) and a hidden ire, while we are
746 	 * searching for duplicates below. Otherwise we could
747 	 * potentially find an IRE on some other interface
748 	 * and it may not be a IRE marked with IRE_MARK_HIDDEN. We
749 	 * shouldn't do this as this will lead to an infinite loop as
750 	 * eventually we need an hidden ire for this packet to go
751 	 * out. MATCH_IRE_ILL is already marked above.
752 	 */
753 	if (ire->ire_marks & IRE_MARK_HIDDEN) {
754 		ASSERT(ire->ire_type == IRE_CACHE);
755 		flags |= MATCH_IRE_MARK_HIDDEN;
756 	}
757 
758 	/*
759 	 * Start the atomic add of the ire. Grab the ill locks,
760 	 * ill_g_usesrc_lock and the bucket lock. Check for condemned.
761 	 * To avoid lock order problems, get the ndp6.ndp_g_lock now itself.
762 	 */
763 	if (ire->ire_type == IRE_CACHE) {
764 		mutex_enter(&ipst->ips_ndp6->ndp_g_lock);
765 		ndp_g_lock_held = B_TRUE;
766 	}
767 
768 	/*
769 	 * If ipif or ill is changing ire_atomic_start() may queue the
770 	 * request and return EINPROGRESS.
771 	 */
772 
773 	error = ire_atomic_start(irb_ptr, ire, q, mp, func);
774 	if (error != 0) {
775 		if (ndp_g_lock_held)
776 			mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
777 		/*
778 		 * We don't know whether it is a valid ipif or not.
779 		 * So, set it to NULL. This assumes that the ire has not added
780 		 * a reference to the ipif.
781 		 */
782 		ire->ire_ipif = NULL;
783 		ire_delete(ire);
784 		if (pire != NULL) {
785 			IRB_REFRELE(pire->ire_bucket);
786 			ire_refrele(pire);
787 		}
788 		*ire_p = NULL;
789 		return (error);
790 	}
791 	/*
792 	 * To avoid creating ires having stale values for the ire_max_frag
793 	 * we get the latest value atomically here. For more details
794 	 * see the block comment in ip_sioctl_mtu and in DL_NOTE_SDU_CHANGE
795 	 * in ip_rput_dlpi_writer
796 	 */
797 	if (ire->ire_max_fragp == NULL) {
798 		if (IN6_IS_ADDR_MULTICAST(&ire->ire_addr_v6))
799 			ire->ire_max_frag = ire->ire_ipif->ipif_mtu;
800 		else
801 			ire->ire_max_frag = pire->ire_max_frag;
802 	} else {
803 		uint_t  max_frag;
804 
805 		max_frag = *ire->ire_max_fragp;
806 		ire->ire_max_fragp = NULL;
807 		ire->ire_max_frag = max_frag;
808 	}
809 
810 	/*
811 	 * Atomically check for duplicate and insert in the table.
812 	 */
813 	for (ire1 = irb_ptr->irb_ire; ire1 != NULL; ire1 = ire1->ire_next) {
814 		if (ire1->ire_marks & IRE_MARK_CONDEMNED)
815 			continue;
816 
817 		if (ire->ire_type == IRE_CACHE) {
818 			/*
819 			 * We do MATCH_IRE_ILL implicitly here for IRE_CACHES.
820 			 * As ire_ipif and ire_stq could point to two
821 			 * different ills, we can't pass just ire_ipif to
822 			 * ire_match_args and get a match on both ills.
823 			 * This is just needed for duplicate checks here and
824 			 * so we don't add an extra argument to
825 			 * ire_match_args for this. Do it locally.
826 			 *
827 			 * NOTE : Currently there is no part of the code
828 			 * that asks for both MATH_IRE_IPIF and MATCH_IRE_ILL
829 			 * match for IRE_CACHEs. Thus we don't want to
830 			 * extend the arguments to ire_match_args_v6.
831 			 */
832 			if (ire1->ire_stq != ire->ire_stq)
833 				continue;
834 			/*
835 			 * Multiroute IRE_CACHEs for a given destination can
836 			 * have the same ire_ipif, typically if their source
837 			 * address is forced using RTF_SETSRC, and the same
838 			 * send-to queue. We differentiate them using the parent
839 			 * handle.
840 			 */
841 			if ((ire1->ire_flags & RTF_MULTIRT) &&
842 			    (ire->ire_flags & RTF_MULTIRT) &&
843 			    (ire1->ire_phandle != ire->ire_phandle))
844 				continue;
845 		}
846 		if (ire1->ire_zoneid != ire->ire_zoneid)
847 			continue;
848 		if (ire_match_args_v6(ire1, &ire->ire_addr_v6,
849 		    &ire->ire_mask_v6, &ire->ire_gateway_addr_v6,
850 		    ire->ire_type, ire->ire_ipif, ire->ire_zoneid, 0, NULL,
851 		    flags)) {
852 			/*
853 			 * Return the old ire after doing a REFHOLD.
854 			 * As most of the callers continue to use the IRE
855 			 * after adding, we return a held ire. This will
856 			 * avoid a lookup in the caller again. If the callers
857 			 * don't want to use it, they need to do a REFRELE.
858 			 */
859 			ip1dbg(("found dup ire existing %p new %p",
860 			    (void *)ire1, (void *)ire));
861 			IRE_REFHOLD(ire1);
862 			if (ndp_g_lock_held)
863 				mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
864 			ire_atomic_end(irb_ptr, ire);
865 			ire_delete(ire);
866 			if (pire != NULL) {
867 				/*
868 				 * Assert that it is
869 				 * not yet removed from the list.
870 				 */
871 				ASSERT(pire->ire_ptpn != NULL);
872 				IRB_REFRELE(pire->ire_bucket);
873 				ire_refrele(pire);
874 			}
875 			*ire_p = ire1;
876 			return (0);
877 		}
878 	}
879 	if (ire->ire_type == IRE_CACHE) {
880 		in6_addr_t gw_addr_v6;
881 		ill_t	*ill = ire_to_ill(ire);
882 		char	buf[INET6_ADDRSTRLEN];
883 		nce_t	*nce;
884 
885 		/*
886 		 * All IRE_CACHE types must have a nce.  If this is
887 		 * not the case the entry will not be added. We need
888 		 * to make sure that if somebody deletes the nce
889 		 * after we looked up, they will find this ire and
890 		 * delete the ire. To delete this ire one needs the
891 		 * bucket lock which we are still holding here. So,
892 		 * even if the nce gets deleted after we looked up,
893 		 * this ire  will get deleted.
894 		 *
895 		 * NOTE : Don't need the ire_lock for accessing
896 		 * ire_gateway_addr_v6 as it is appearing first
897 		 * time on the list and rts_setgwr_v6 could not
898 		 * be changing this.
899 		 */
900 		gw_addr_v6 = ire->ire_gateway_addr_v6;
901 		if (IN6_IS_ADDR_UNSPECIFIED(&gw_addr_v6)) {
902 			nce = ndp_lookup_v6(ill, &ire->ire_addr_v6, B_TRUE);
903 		} else {
904 			nce = ndp_lookup_v6(ill, &gw_addr_v6, B_TRUE);
905 		}
906 		if (nce == NULL)
907 			goto failed;
908 
909 		/* Pair of refhold, refrele just to get the tracing right */
910 		NCE_REFHOLD_TO_REFHOLD_NOTR(nce);
911 		/*
912 		 * Atomically make sure that new IREs don't point
913 		 * to an NCE that is logically deleted (CONDEMNED).
914 		 * ndp_delete() first marks the NCE CONDEMNED.
915 		 * This ensures that the nce_refcnt won't increase
916 		 * due to new nce_lookups or due to addition of new IREs
917 		 * pointing to this NCE. Then ndp_delete() cleans up
918 		 * existing references. If we don't do it atomically here,
919 		 * ndp_delete() -> nce_ire_delete() will not be able to
920 		 * clean up the IRE list completely, and the nce_refcnt
921 		 * won't go down to zero.
922 		 */
923 		mutex_enter(&nce->nce_lock);
924 		if (ill->ill_flags & ILLF_XRESOLV) {
925 			/*
926 			 * If we used an external resolver, we may not
927 			 * have gone through neighbor discovery to get here.
928 			 * Must update the nce_state before the next check.
929 			 */
930 			if (nce->nce_state == ND_INCOMPLETE)
931 				nce->nce_state = ND_REACHABLE;
932 		}
933 		if (nce->nce_state == ND_INCOMPLETE ||
934 		    (nce->nce_flags & NCE_F_CONDEMNED) ||
935 		    (nce->nce_state == ND_UNREACHABLE)) {
936 failed:
937 			if (ndp_g_lock_held)
938 				mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
939 			if (nce != NULL)
940 				mutex_exit(&nce->nce_lock);
941 			ire_atomic_end(irb_ptr, ire);
942 			ip1dbg(("ire_add_v6: No nce for dst %s \n",
943 			    inet_ntop(AF_INET6, &ire->ire_addr_v6,
944 			    buf, sizeof (buf))));
945 			ire_delete(ire);
946 			if (pire != NULL) {
947 				/*
948 				 * Assert that it is
949 				 * not yet removed from the list.
950 				 */
951 				ASSERT(pire->ire_ptpn != NULL);
952 				IRB_REFRELE(pire->ire_bucket);
953 				ire_refrele(pire);
954 			}
955 			if (nce != NULL)
956 				NCE_REFRELE_NOTR(nce);
957 			*ire_p = NULL;
958 			return (EINVAL);
959 		} else {
960 			ire->ire_nce = nce;
961 		}
962 		mutex_exit(&nce->nce_lock);
963 	}
964 	/*
965 	 * Find the first entry that matches ire_addr - provides
966 	 * tail insertion. *irep will be null if no match.
967 	 */
968 	irep = (ire_t **)irb_ptr;
969 	while ((ire1 = *irep) != NULL &&
970 	    !IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, &ire1->ire_addr_v6))
971 		irep = &ire1->ire_next;
972 	ASSERT(!(ire->ire_type & IRE_BROADCAST));
973 
974 	if (*irep != NULL) {
975 		/*
976 		 * Find the last ire which matches ire_addr_v6.
977 		 * Needed to do tail insertion among entries with the same
978 		 * ire_addr_v6.
979 		 */
980 		while (IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6,
981 		    &ire1->ire_addr_v6)) {
982 			irep = &ire1->ire_next;
983 			ire1 = *irep;
984 			if (ire1 == NULL)
985 				break;
986 		}
987 	}
988 
989 	if (ire->ire_type == IRE_DEFAULT) {
990 		/*
991 		 * We keep a count of default gateways which is used when
992 		 * assigning them as routes.
993 		 */
994 		ipst->ips_ipv6_ire_default_count++;
995 		ASSERT(ipst->ips_ipv6_ire_default_count != 0); /* Wraparound */
996 	}
997 	/* Insert at *irep */
998 	ire1 = *irep;
999 	if (ire1 != NULL)
1000 		ire1->ire_ptpn = &ire->ire_next;
1001 	ire->ire_next = ire1;
1002 	/* Link the new one in. */
1003 	ire->ire_ptpn = irep;
1004 	/*
1005 	 * ire_walk routines de-reference ire_next without holding
1006 	 * a lock. Before we point to the new ire, we want to make
1007 	 * sure the store that sets the ire_next of the new ire
1008 	 * reaches global visibility, so that ire_walk routines
1009 	 * don't see a truncated list of ires i.e if the ire_next
1010 	 * of the new ire gets set after we do "*irep = ire" due
1011 	 * to re-ordering, the ire_walk thread will see a NULL
1012 	 * once it accesses the ire_next of the new ire.
1013 	 * membar_producer() makes sure that the following store
1014 	 * happens *after* all of the above stores.
1015 	 */
1016 	membar_producer();
1017 	*irep = ire;
1018 	ire->ire_bucket = irb_ptr;
1019 	/*
1020 	 * We return a bumped up IRE above. Keep it symmetrical
1021 	 * so that the callers will always have to release. This
1022 	 * helps the callers of this function because they continue
1023 	 * to use the IRE after adding and hence they don't have to
1024 	 * lookup again after we return the IRE.
1025 	 *
1026 	 * NOTE : We don't have to use atomics as this is appearing
1027 	 * in the list for the first time and no one else can bump
1028 	 * up the reference count on this yet.
1029 	 */
1030 	IRE_REFHOLD_LOCKED(ire);
1031 	BUMP_IRE_STATS(ipst->ips_ire_stats_v6, ire_stats_inserted);
1032 	irb_ptr->irb_ire_cnt++;
1033 	if (ire->ire_marks & IRE_MARK_TEMPORARY)
1034 		irb_ptr->irb_tmp_ire_cnt++;
1035 
1036 	if (ire->ire_ipif != NULL) {
1037 		ire->ire_ipif->ipif_ire_cnt++;
1038 		if (ire->ire_stq != NULL) {
1039 			stq_ill = (ill_t *)ire->ire_stq->q_ptr;
1040 			stq_ill->ill_ire_cnt++;
1041 		}
1042 	} else {
1043 		ASSERT(ire->ire_stq == NULL);
1044 	}
1045 
1046 	if (ndp_g_lock_held)
1047 		mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
1048 	ire_atomic_end(irb_ptr, ire);
1049 
1050 	if (pire != NULL) {
1051 		/* Assert that it is not removed from the list yet */
1052 		ASSERT(pire->ire_ptpn != NULL);
1053 		IRB_REFRELE(pire->ire_bucket);
1054 		ire_refrele(pire);
1055 	}
1056 
1057 	if (ire->ire_type != IRE_CACHE) {
1058 		/*
1059 		 * For ire's with with host mask see if there is an entry
1060 		 * in the cache. If there is one flush the whole cache as
1061 		 * there might be multiple entries due to RTF_MULTIRT (CGTP).
1062 		 * If no entry is found than there is no need to flush the
1063 		 * cache.
1064 		 */
1065 
1066 		if (ip_mask_to_plen_v6(&ire->ire_mask_v6) == IPV6_ABITS) {
1067 			ire_t *lire;
1068 			lire = ire_ctable_lookup_v6(&ire->ire_addr_v6, NULL,
1069 			    IRE_CACHE, NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE,
1070 			    ipst);
1071 			if (lire != NULL) {
1072 				ire_refrele(lire);
1073 				ire_flush_cache_v6(ire, IRE_FLUSH_ADD);
1074 			}
1075 		} else {
1076 			ire_flush_cache_v6(ire, IRE_FLUSH_ADD);
1077 		}
1078 	}
1079 
1080 	*ire_p = ire;
1081 	return (0);
1082 }
1083 
1084 /*
1085  * Search for all HOST REDIRECT routes that are
1086  * pointing at the specified gateway and
1087  * delete them. This routine is called only
1088  * when a default gateway is going away.
1089  */
1090 static void
1091 ire_delete_host_redirects_v6(const in6_addr_t *gateway, ip_stack_t *ipst)
1092 {
1093 	irb_t *irb_ptr;
1094 	irb_t *irb;
1095 	ire_t *ire;
1096 	in6_addr_t gw_addr_v6;
1097 	int i;
1098 
1099 	/* get the hash table for HOST routes */
1100 	irb_ptr = ipst->ips_ip_forwarding_table_v6[(IP6_MASK_TABLE_SIZE - 1)];
1101 	if (irb_ptr == NULL)
1102 		return;
1103 	for (i = 0; (i < ipst->ips_ip6_ftable_hash_size); i++) {
1104 		irb = &irb_ptr[i];
1105 		IRB_REFHOLD(irb);
1106 		for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
1107 			if (!(ire->ire_flags & RTF_DYNAMIC))
1108 				continue;
1109 			mutex_enter(&ire->ire_lock);
1110 			gw_addr_v6 = ire->ire_gateway_addr_v6;
1111 			mutex_exit(&ire->ire_lock);
1112 			if (IN6_ARE_ADDR_EQUAL(&gw_addr_v6, gateway))
1113 				ire_delete(ire);
1114 		}
1115 		IRB_REFRELE(irb);
1116 	}
1117 }
1118 
1119 /*
1120  * Delete all the cache entries with this 'addr'. This is the IPv6 counterpart
1121  * of ip_ire_clookup_and_delete. The difference being this function does not
1122  * return any value. IPv6 processing of a gratuitous ARP, as it stands, is
1123  * different than IPv4 in that, regardless of the presence of a cache entry
1124  * for this address, an ire_walk_v6 is done. Another difference is that unlike
1125  * in the case of IPv4 this does not take an ipif_t argument, since it is only
1126  * called by ip_arp_news and the match is always only on the address.
1127  */
1128 void
1129 ip_ire_clookup_and_delete_v6(const in6_addr_t *addr, ip_stack_t *ipst)
1130 {
1131 	irb_t		*irb;
1132 	ire_t		*cire;
1133 	boolean_t	found = B_FALSE;
1134 
1135 	irb = &ipst->ips_ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr,
1136 	    ipst->ips_ip6_cache_table_size)];
1137 	IRB_REFHOLD(irb);
1138 	for (cire = irb->irb_ire; cire != NULL; cire = cire->ire_next) {
1139 		if (cire->ire_marks & IRE_MARK_CONDEMNED)
1140 			continue;
1141 		if (IN6_ARE_ADDR_EQUAL(&cire->ire_addr_v6, addr)) {
1142 
1143 			/* This signifies start of a match */
1144 			if (!found)
1145 				found = B_TRUE;
1146 			if (cire->ire_type == IRE_CACHE) {
1147 				if (cire->ire_nce != NULL)
1148 					ndp_delete(cire->ire_nce);
1149 				ire_delete_v6(cire);
1150 			}
1151 		/* End of the match */
1152 		} else if (found)
1153 			break;
1154 	}
1155 	IRB_REFRELE(irb);
1156 }
1157 
1158 /*
1159  * Delete the specified IRE.
1160  * All calls should use ire_delete().
1161  * Sometimes called as writer though not required by this function.
1162  *
1163  * NOTE : This function is called only if the ire was added
1164  * in the list.
1165  */
1166 void
1167 ire_delete_v6(ire_t *ire)
1168 {
1169 	in6_addr_t gw_addr_v6;
1170 	ip_stack_t	*ipst = ire->ire_ipst;
1171 
1172 	ASSERT(ire->ire_refcnt >= 1);
1173 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
1174 
1175 	if (ire->ire_type != IRE_CACHE)
1176 		ire_flush_cache_v6(ire, IRE_FLUSH_DELETE);
1177 	if (ire->ire_type == IRE_DEFAULT) {
1178 		/*
1179 		 * when a default gateway is going away
1180 		 * delete all the host redirects pointing at that
1181 		 * gateway.
1182 		 */
1183 		mutex_enter(&ire->ire_lock);
1184 		gw_addr_v6 = ire->ire_gateway_addr_v6;
1185 		mutex_exit(&ire->ire_lock);
1186 		ire_delete_host_redirects_v6(&gw_addr_v6, ipst);
1187 	}
1188 }
1189 
1190 /*
1191  * ire_walk routine to delete all IRE_CACHE and IRE_HOST type redirect
1192  * entries.
1193  */
1194 /*ARGSUSED1*/
1195 void
1196 ire_delete_cache_v6(ire_t *ire, char *arg)
1197 {
1198 	char    addrstr1[INET6_ADDRSTRLEN];
1199 	char    addrstr2[INET6_ADDRSTRLEN];
1200 
1201 	if ((ire->ire_type & IRE_CACHE) ||
1202 	    (ire->ire_flags & RTF_DYNAMIC)) {
1203 		ip1dbg(("ire_delete_cache_v6: deleted %s type %d through %s\n",
1204 		    inet_ntop(AF_INET6, &ire->ire_addr_v6,
1205 		    addrstr1, sizeof (addrstr1)),
1206 		    ire->ire_type,
1207 		    inet_ntop(AF_INET6, &ire->ire_gateway_addr_v6,
1208 		    addrstr2, sizeof (addrstr2))));
1209 		ire_delete(ire);
1210 	}
1211 
1212 }
1213 
1214 /*
1215  * ire_walk routine to delete all IRE_CACHE/IRE_HOST type redirect entries
1216  * that have a given gateway address.
1217  */
1218 void
1219 ire_delete_cache_gw_v6(ire_t *ire, char *addr)
1220 {
1221 	in6_addr_t	*gw_addr = (in6_addr_t *)addr;
1222 	char		buf1[INET6_ADDRSTRLEN];
1223 	char		buf2[INET6_ADDRSTRLEN];
1224 	in6_addr_t	ire_gw_addr_v6;
1225 
1226 	if (!(ire->ire_type & IRE_CACHE) &&
1227 	    !(ire->ire_flags & RTF_DYNAMIC))
1228 		return;
1229 
1230 	mutex_enter(&ire->ire_lock);
1231 	ire_gw_addr_v6 = ire->ire_gateway_addr_v6;
1232 	mutex_exit(&ire->ire_lock);
1233 
1234 	if (IN6_ARE_ADDR_EQUAL(&ire_gw_addr_v6, gw_addr)) {
1235 		ip1dbg(("ire_delete_cache_gw_v6: deleted %s type %d to %s\n",
1236 		    inet_ntop(AF_INET6, &ire->ire_src_addr_v6,
1237 		    buf1, sizeof (buf1)),
1238 		    ire->ire_type,
1239 		    inet_ntop(AF_INET6, &ire_gw_addr_v6,
1240 		    buf2, sizeof (buf2))));
1241 		ire_delete(ire);
1242 	}
1243 }
1244 
1245 /*
1246  * Remove all IRE_CACHE entries that match
1247  * the ire specified.  (Sometimes called
1248  * as writer though not required by this function.)
1249  *
1250  * The flag argument indicates if the
1251  * flush request is due to addition
1252  * of new route (IRE_FLUSH_ADD) or deletion of old
1253  * route (IRE_FLUSH_DELETE).
1254  *
1255  * This routine takes only the IREs from the forwarding
1256  * table and flushes the corresponding entries from
1257  * the cache table.
1258  *
1259  * When flushing due to the deletion of an old route, it
1260  * just checks the cache handles (ire_phandle and ire_ihandle) and
1261  * deletes the ones that match.
1262  *
1263  * When flushing due to the creation of a new route, it checks
1264  * if a cache entry's address matches the one in the IRE and
1265  * that the cache entry's parent has a less specific mask than the
1266  * one in IRE. The destination of such a cache entry could be the
1267  * gateway for other cache entries, so we need to flush those as
1268  * well by looking for gateway addresses matching the IRE's address.
1269  */
1270 void
1271 ire_flush_cache_v6(ire_t *ire, int flag)
1272 {
1273 	int i;
1274 	ire_t *cire;
1275 	irb_t *irb;
1276 	ip_stack_t	*ipst = ire->ire_ipst;
1277 
1278 	if (ire->ire_type & IRE_CACHE)
1279 		return;
1280 
1281 	/*
1282 	 * If a default is just created, there is no point
1283 	 * in going through the cache, as there will not be any
1284 	 * cached ires.
1285 	 */
1286 	if (ire->ire_type == IRE_DEFAULT && flag == IRE_FLUSH_ADD)
1287 		return;
1288 	if (flag == IRE_FLUSH_ADD) {
1289 		/*
1290 		 * This selective flush is
1291 		 * due to the addition of
1292 		 * new IRE.
1293 		 */
1294 		for (i = 0; i < ipst->ips_ip6_cache_table_size; i++) {
1295 			irb = &ipst->ips_ip_cache_table_v6[i];
1296 			if ((cire = irb->irb_ire) == NULL)
1297 				continue;
1298 			IRB_REFHOLD(irb);
1299 			for (cire = irb->irb_ire; cire != NULL;
1300 			    cire = cire->ire_next) {
1301 				if (cire->ire_type != IRE_CACHE)
1302 					continue;
1303 				/*
1304 				 * If 'cire' belongs to the same subnet
1305 				 * as the new ire being added, and 'cire'
1306 				 * is derived from a prefix that is less
1307 				 * specific than the new ire being added,
1308 				 * we need to flush 'cire'; for instance,
1309 				 * when a new interface comes up.
1310 				 */
1311 				if ((V6_MASK_EQ_2(cire->ire_addr_v6,
1312 				    ire->ire_mask_v6, ire->ire_addr_v6) &&
1313 				    (ip_mask_to_plen_v6(&cire->ire_cmask_v6) <=
1314 				    ire->ire_masklen))) {
1315 					ire_delete(cire);
1316 					continue;
1317 				}
1318 				/*
1319 				 * This is the case when the ire_gateway_addr
1320 				 * of 'cire' belongs to the same subnet as
1321 				 * the new ire being added.
1322 				 * Flushing such ires is sometimes required to
1323 				 * avoid misrouting: say we have a machine with
1324 				 * two interfaces (I1 and I2), a default router
1325 				 * R on the I1 subnet, and a host route to an
1326 				 * off-link destination D with a gateway G on
1327 				 * the I2 subnet.
1328 				 * Under normal operation, we will have an
1329 				 * on-link cache entry for G and an off-link
1330 				 * cache entry for D with G as ire_gateway_addr,
1331 				 * traffic to D will reach its destination
1332 				 * through gateway G.
1333 				 * If the administrator does 'ifconfig I2 down',
1334 				 * the cache entries for D and G will be
1335 				 * flushed. However, G will now be resolved as
1336 				 * an off-link destination using R (the default
1337 				 * router) as gateway. Then D will also be
1338 				 * resolved as an off-link destination using G
1339 				 * as gateway - this behavior is due to
1340 				 * compatibility reasons, see comment in
1341 				 * ire_ihandle_lookup_offlink(). Traffic to D
1342 				 * will go to the router R and probably won't
1343 				 * reach the destination.
1344 				 * The administrator then does 'ifconfig I2 up'.
1345 				 * Since G is on the I2 subnet, this routine
1346 				 * will flush its cache entry. It must also
1347 				 * flush the cache entry for D, otherwise
1348 				 * traffic will stay misrouted until the IRE
1349 				 * times out.
1350 				 */
1351 				if (V6_MASK_EQ_2(cire->ire_gateway_addr_v6,
1352 				    ire->ire_mask_v6, ire->ire_addr_v6)) {
1353 					ire_delete(cire);
1354 					continue;
1355 				}
1356 			}
1357 			IRB_REFRELE(irb);
1358 		}
1359 	} else {
1360 		/*
1361 		 * delete the cache entries based on
1362 		 * handle in the IRE as this IRE is
1363 		 * being deleted/changed.
1364 		 */
1365 		for (i = 0; i < ipst->ips_ip6_cache_table_size; i++) {
1366 			irb = &ipst->ips_ip_cache_table_v6[i];
1367 			if ((cire = irb->irb_ire) == NULL)
1368 				continue;
1369 			IRB_REFHOLD(irb);
1370 			for (cire = irb->irb_ire; cire != NULL;
1371 			    cire = cire->ire_next) {
1372 				if (cire->ire_type != IRE_CACHE)
1373 					continue;
1374 				if ((cire->ire_phandle == 0 ||
1375 				    cire->ire_phandle != ire->ire_phandle) &&
1376 				    (cire->ire_ihandle == 0 ||
1377 				    cire->ire_ihandle != ire->ire_ihandle))
1378 					continue;
1379 				ire_delete(cire);
1380 			}
1381 			IRB_REFRELE(irb);
1382 		}
1383 	}
1384 }
1385 
1386 /*
1387  * Matches the arguments passed with the values in the ire.
1388  *
1389  * Note: for match types that match using "ipif" passed in, ipif
1390  * must be checked for non-NULL before calling this routine.
1391  */
1392 static boolean_t
1393 ire_match_args_v6(ire_t *ire, const in6_addr_t *addr, const in6_addr_t *mask,
1394     const in6_addr_t *gateway, int type, const ipif_t *ipif, zoneid_t zoneid,
1395     uint32_t ihandle, const ts_label_t *tsl, int match_flags)
1396 {
1397 	in6_addr_t masked_addr;
1398 	in6_addr_t gw_addr_v6;
1399 	ill_t *ire_ill = NULL, *dst_ill;
1400 	ill_t *ipif_ill = NULL;
1401 	ill_group_t *ire_ill_group = NULL;
1402 	ill_group_t *ipif_ill_group = NULL;
1403 	ipif_t	*src_ipif;
1404 
1405 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
1406 	ASSERT(addr != NULL);
1407 	ASSERT(mask != NULL);
1408 	ASSERT((!(match_flags & MATCH_IRE_GW)) || gateway != NULL);
1409 	ASSERT((!(match_flags & (MATCH_IRE_ILL|MATCH_IRE_ILL_GROUP))) ||
1410 	    (ipif != NULL && ipif->ipif_isv6));
1411 	ASSERT(!(match_flags & MATCH_IRE_WQ));
1412 
1413 	/*
1414 	 * HIDDEN cache entries have to be looked up specifically with
1415 	 * MATCH_IRE_MARK_HIDDEN. MATCH_IRE_MARK_HIDDEN is usually set
1416 	 * when the interface is FAILED or INACTIVE. In that case,
1417 	 * any IRE_CACHES that exists should be marked with
1418 	 * IRE_MARK_HIDDEN. So, we don't really need to match below
1419 	 * for IRE_MARK_HIDDEN. But we do so for consistency.
1420 	 */
1421 	if (!(match_flags & MATCH_IRE_MARK_HIDDEN) &&
1422 	    (ire->ire_marks & IRE_MARK_HIDDEN))
1423 		return (B_FALSE);
1424 
1425 	if (zoneid != ALL_ZONES && zoneid != ire->ire_zoneid &&
1426 	    ire->ire_zoneid != ALL_ZONES) {
1427 		/*
1428 		 * If MATCH_IRE_ZONEONLY has been set and the supplied zoneid is
1429 		 * valid and does not match that of ire_zoneid, a failure to
1430 		 * match is reported at this point. Otherwise, since some IREs
1431 		 * that are available in the global zone can be used in local
1432 		 * zones, additional checks need to be performed:
1433 		 *
1434 		 *	IRE_CACHE and IRE_LOOPBACK entries should
1435 		 *	never be matched in this situation.
1436 		 *
1437 		 *	IRE entries that have an interface associated with them
1438 		 *	should in general not match unless they are an IRE_LOCAL
1439 		 *	or in the case when MATCH_IRE_DEFAULT has been set in
1440 		 *	the caller.  In the case of the former, checking of the
1441 		 *	other fields supplied should take place.
1442 		 *
1443 		 *	In the case where MATCH_IRE_DEFAULT has been set,
1444 		 *	all of the ipif's associated with the IRE's ill are
1445 		 *	checked to see if there is a matching zoneid.  If any
1446 		 *	one ipif has a matching zoneid, this IRE is a
1447 		 *	potential candidate so checking of the other fields
1448 		 *	takes place.
1449 		 *
1450 		 *	In the case where the IRE_INTERFACE has a usable source
1451 		 *	address (indicated by ill_usesrc_ifindex) in the
1452 		 *	correct zone then it's permitted to return this IRE
1453 		 */
1454 		if (match_flags & MATCH_IRE_ZONEONLY)
1455 			return (B_FALSE);
1456 		if (ire->ire_type & (IRE_CACHE | IRE_LOOPBACK))
1457 			return (B_FALSE);
1458 		/*
1459 		 * Note, IRE_INTERFACE can have the stq as NULL. For
1460 		 * example, if the default multicast route is tied to
1461 		 * the loopback address.
1462 		 */
1463 		if ((ire->ire_type & IRE_INTERFACE) &&
1464 		    (ire->ire_stq != NULL)) {
1465 			dst_ill = (ill_t *)ire->ire_stq->q_ptr;
1466 			/*
1467 			 * If there is a usable source address in the
1468 			 * zone, then it's ok to return an
1469 			 * IRE_INTERFACE
1470 			 */
1471 			if ((dst_ill->ill_usesrc_ifindex != 0) &&
1472 			    (src_ipif = ipif_select_source_v6(dst_ill, addr,
1473 			    RESTRICT_TO_NONE, IPV6_PREFER_SRC_DEFAULT, zoneid))
1474 			    != NULL) {
1475 				ip3dbg(("ire_match_args: src_ipif %p"
1476 				    " dst_ill %p", (void *)src_ipif,
1477 				    (void *)dst_ill));
1478 				ipif_refrele(src_ipif);
1479 			} else {
1480 				ip3dbg(("ire_match_args: src_ipif NULL"
1481 				    " dst_ill %p\n", (void *)dst_ill));
1482 				return (B_FALSE);
1483 			}
1484 		}
1485 		if (ire->ire_ipif != NULL && ire->ire_type != IRE_LOCAL &&
1486 		    !(ire->ire_type & IRE_INTERFACE)) {
1487 			ipif_t	*tipif;
1488 
1489 			if ((match_flags & MATCH_IRE_DEFAULT) == 0)
1490 				return (B_FALSE);
1491 			mutex_enter(&ire->ire_ipif->ipif_ill->ill_lock);
1492 			for (tipif = ire->ire_ipif->ipif_ill->ill_ipif;
1493 			    tipif != NULL; tipif = tipif->ipif_next) {
1494 				if (IPIF_CAN_LOOKUP(tipif) &&
1495 				    (tipif->ipif_flags & IPIF_UP) &&
1496 				    (tipif->ipif_zoneid == zoneid ||
1497 				    tipif->ipif_zoneid == ALL_ZONES))
1498 					break;
1499 			}
1500 			mutex_exit(&ire->ire_ipif->ipif_ill->ill_lock);
1501 			if (tipif == NULL)
1502 				return (B_FALSE);
1503 		}
1504 	}
1505 
1506 	if (match_flags & MATCH_IRE_GW) {
1507 		mutex_enter(&ire->ire_lock);
1508 		gw_addr_v6 = ire->ire_gateway_addr_v6;
1509 		mutex_exit(&ire->ire_lock);
1510 	}
1511 	/*
1512 	 * For IRE_CACHES, MATCH_IRE_ILL/ILL_GROUP really means that
1513 	 * somebody wants to send out on a particular interface which
1514 	 * is given by ire_stq and hence use ire_stq to derive the ill
1515 	 * value. ire_ipif for IRE_CACHES is just the
1516 	 * means of getting a source address i.e ire_src_addr_v6 =
1517 	 * ire->ire_ipif->ipif_src_addr_v6.
1518 	 */
1519 	if (match_flags & (MATCH_IRE_ILL|MATCH_IRE_ILL_GROUP)) {
1520 		ire_ill = ire_to_ill(ire);
1521 		if (ire_ill != NULL)
1522 			ire_ill_group = ire_ill->ill_group;
1523 		ipif_ill = ipif->ipif_ill;
1524 		ipif_ill_group = ipif_ill->ill_group;
1525 	}
1526 
1527 	/* No ire_addr_v6 bits set past the mask */
1528 	ASSERT(V6_MASK_EQ(ire->ire_addr_v6, ire->ire_mask_v6,
1529 	    ire->ire_addr_v6));
1530 	V6_MASK_COPY(*addr, *mask, masked_addr);
1531 
1532 	if (V6_MASK_EQ(*addr, *mask, ire->ire_addr_v6) &&
1533 	    ((!(match_flags & MATCH_IRE_GW)) ||
1534 	    IN6_ARE_ADDR_EQUAL(&gw_addr_v6, gateway)) &&
1535 	    ((!(match_flags & MATCH_IRE_TYPE)) ||
1536 	    (ire->ire_type & type)) &&
1537 	    ((!(match_flags & MATCH_IRE_SRC)) ||
1538 	    IN6_ARE_ADDR_EQUAL(&ire->ire_src_addr_v6,
1539 	    &ipif->ipif_v6src_addr)) &&
1540 	    ((!(match_flags & MATCH_IRE_IPIF)) ||
1541 	    (ire->ire_ipif == ipif)) &&
1542 	    ((!(match_flags & MATCH_IRE_MARK_HIDDEN)) ||
1543 	    (ire->ire_type != IRE_CACHE ||
1544 	    ire->ire_marks & IRE_MARK_HIDDEN)) &&
1545 	    ((!(match_flags & MATCH_IRE_ILL)) ||
1546 	    (ire_ill == ipif_ill)) &&
1547 	    ((!(match_flags & MATCH_IRE_IHANDLE)) ||
1548 	    (ire->ire_ihandle == ihandle)) &&
1549 	    ((!(match_flags & MATCH_IRE_ILL_GROUP)) ||
1550 	    (ire_ill == ipif_ill) ||
1551 	    (ire_ill_group != NULL &&
1552 	    ire_ill_group == ipif_ill_group)) &&
1553 	    ((!(match_flags & MATCH_IRE_SECATTR)) ||
1554 	    (!is_system_labeled()) ||
1555 	    (tsol_ire_match_gwattr(ire, tsl) == 0))) {
1556 		/* We found the matched IRE */
1557 		return (B_TRUE);
1558 	}
1559 	return (B_FALSE);
1560 }
1561 
1562 /*
1563  * Lookup for a route in all the tables
1564  */
1565 ire_t *
1566 ire_route_lookup_v6(const in6_addr_t *addr, const in6_addr_t *mask,
1567     const in6_addr_t *gateway, int type, const ipif_t *ipif, ire_t **pire,
1568     zoneid_t zoneid, const ts_label_t *tsl, int flags, ip_stack_t *ipst)
1569 {
1570 	ire_t *ire = NULL;
1571 
1572 	/*
1573 	 * ire_match_args_v6() will dereference ipif MATCH_IRE_SRC or
1574 	 * MATCH_IRE_ILL is set.
1575 	 */
1576 	if ((flags & (MATCH_IRE_SRC | MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) &&
1577 	    (ipif == NULL))
1578 		return (NULL);
1579 
1580 	/*
1581 	 * might be asking for a cache lookup,
1582 	 * This is not best way to lookup cache,
1583 	 * user should call ire_cache_lookup directly.
1584 	 *
1585 	 * If MATCH_IRE_TYPE was set, first lookup in the cache table and then
1586 	 * in the forwarding table, if the applicable type flags were set.
1587 	 */
1588 	if ((flags & MATCH_IRE_TYPE) == 0 || (type & IRE_CACHETABLE) != 0) {
1589 		ire = ire_ctable_lookup_v6(addr, gateway, type, ipif, zoneid,
1590 		    tsl, flags, ipst);
1591 		if (ire != NULL)
1592 			return (ire);
1593 	}
1594 	if ((flags & MATCH_IRE_TYPE) == 0 || (type & IRE_FORWARDTABLE) != 0) {
1595 		ire = ire_ftable_lookup_v6(addr, mask, gateway, type, ipif,
1596 		    pire, zoneid, 0, tsl, flags, ipst);
1597 	}
1598 	return (ire);
1599 }
1600 
1601 /*
1602  * Lookup a route in forwarding table.
1603  * specific lookup is indicated by passing the
1604  * required parameters and indicating the
1605  * match required in flag field.
1606  *
1607  * Looking for default route can be done in three ways
1608  * 1) pass mask as ipv6_all_zeros and set MATCH_IRE_MASK in flags field
1609  *    along with other matches.
1610  * 2) pass type as IRE_DEFAULT and set MATCH_IRE_TYPE in flags
1611  *    field along with other matches.
1612  * 3) if the destination and mask are passed as zeros.
1613  *
1614  * A request to return a default route if no route
1615  * is found, can be specified by setting MATCH_IRE_DEFAULT
1616  * in flags.
1617  *
1618  * It does not support recursion more than one level. It
1619  * will do recursive lookup only when the lookup maps to
1620  * a prefix or default route and MATCH_IRE_RECURSIVE flag is passed.
1621  *
1622  * If the routing table is setup to allow more than one level
1623  * of recursion, the cleaning up cache table will not work resulting
1624  * in invalid routing.
1625  *
1626  * Supports link-local addresses by following the ipif/ill when recursing.
1627  *
1628  * NOTE : When this function returns NULL, pire has already been released.
1629  *	  pire is valid only when this function successfully returns an
1630  *	  ire.
1631  */
1632 ire_t *
1633 ire_ftable_lookup_v6(const in6_addr_t *addr, const in6_addr_t *mask,
1634     const in6_addr_t *gateway, int type, const ipif_t *ipif, ire_t **pire,
1635     zoneid_t zoneid, uint32_t ihandle, const ts_label_t *tsl, int flags,
1636     ip_stack_t *ipst)
1637 {
1638 	irb_t *irb_ptr;
1639 	ire_t	*rire;
1640 	ire_t *ire = NULL;
1641 	ire_t	*saved_ire;
1642 	nce_t	*nce;
1643 	int i;
1644 	in6_addr_t gw_addr_v6;
1645 
1646 	ASSERT(addr != NULL);
1647 	ASSERT((!(flags & MATCH_IRE_MASK)) || mask != NULL);
1648 	ASSERT((!(flags & MATCH_IRE_GW)) || gateway != NULL);
1649 	ASSERT(ipif == NULL || ipif->ipif_isv6);
1650 	ASSERT(!(flags & MATCH_IRE_WQ));
1651 
1652 	/*
1653 	 * When we return NULL from this function, we should make
1654 	 * sure that *pire is NULL so that the callers will not
1655 	 * wrongly REFRELE the pire.
1656 	 */
1657 	if (pire != NULL)
1658 		*pire = NULL;
1659 	/*
1660 	 * ire_match_args_v6() will dereference ipif MATCH_IRE_SRC or
1661 	 * MATCH_IRE_ILL is set.
1662 	 */
1663 	if ((flags & (MATCH_IRE_SRC | MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) &&
1664 	    (ipif == NULL))
1665 		return (NULL);
1666 
1667 	/*
1668 	 * If the mask is known, the lookup
1669 	 * is simple, if the mask is not known
1670 	 * we need to search.
1671 	 */
1672 	if (flags & MATCH_IRE_MASK) {
1673 		uint_t masklen;
1674 
1675 		masklen = ip_mask_to_plen_v6(mask);
1676 		if (ipst->ips_ip_forwarding_table_v6[masklen] == NULL)
1677 			return (NULL);
1678 		irb_ptr = &(ipst->ips_ip_forwarding_table_v6[masklen][
1679 		    IRE_ADDR_MASK_HASH_V6(*addr, *mask,
1680 		    ipst->ips_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 ((ipst->ips_ip_forwarding_table_v6[i]) == NULL)
1701 				continue;
1702 			(void) ip_plen_to_mask_v6(i, &tmpmask);
1703 			irb_ptr = &ipst->ips_ip_forwarding_table_v6[i][
1704 			    IRE_ADDR_MASK_HASH_V6(*addr, tmpmask,
1705 			    ipst->ips_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 (ipst->ips_ip_forwarding_table_v6[0] != NULL) {
1735 			/* addr & mask is zero for defaults */
1736 			irb_ptr = &ipst->ips_ip_forwarding_table_v6[0][
1737 			    IRE_ADDR_HASH_V6(ipv6_all_zeros,
1738 			    ipst->ips_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 (ipst->ips_ip_forwarding_table_v6[0] == NULL)
1770 			return (NULL);
1771 		irb_ptr = &(ipst->ips_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 (ipst->ips_ipv6_ire_default_count != 0) {
1794 			g_index = ipst->ips_ipv6_ire_default_index++;
1795 			index = g_index % ipst->ips_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 				    ipst);
1852 				if (rire != NULL) {
1853 					nce = rire->ire_nce;
1854 					if (nce != NULL &&
1855 					    NCE_ISREACHABLE(nce) &&
1856 					    nce->nce_flags & NCE_F_ISROUTER) {
1857 						ire_refrele(rire);
1858 						IRE_REFHOLD(ire);
1859 						IRB_REFRELE(irb_ptr);
1860 						goto found_ire_held;
1861 					} else if (nce != NULL &&
1862 					    !(nce->nce_flags &
1863 					    NCE_F_ISROUTER)) {
1864 						/*
1865 						 * Make sure we don't use
1866 						 * this ire
1867 						 */
1868 						if (saved_ire == ire)
1869 							saved_ire = NULL;
1870 					}
1871 					ire_refrele(rire);
1872 				} else if (ipst->
1873 				    ips_ipv6_ire_default_count > 1 &&
1874 				    zoneid != GLOBAL_ZONEID) {
1875 					/*
1876 					 * When we're in a local zone, we're
1877 					 * only interested in default routers
1878 					 * that are reachable through ipifs
1879 					 * within our zone.
1880 					 * The potentially expensive call to
1881 					 * ire_route_lookup_v6() is avoided when
1882 					 * we have only one default route.
1883 					 */
1884 					int ire_match_flags = MATCH_IRE_TYPE |
1885 					    MATCH_IRE_SECATTR;
1886 
1887 					if (ire->ire_ipif != NULL) {
1888 						ire_match_flags |=
1889 						    MATCH_IRE_ILL_GROUP;
1890 					}
1891 					rire = ire_route_lookup_v6(&gw_addr_v6,
1892 					    NULL, NULL, IRE_INTERFACE,
1893 					    ire->ire_ipif, NULL,
1894 					    zoneid, tsl, ire_match_flags, ipst);
1895 					if (rire != NULL) {
1896 						ire_refrele(rire);
1897 						saved_ire = ire;
1898 					} else if (saved_ire == ire) {
1899 						/*
1900 						 * Make sure we don't use
1901 						 * this ire
1902 						 */
1903 						saved_ire = NULL;
1904 					}
1905 				}
1906 			}
1907 		}
1908 		if (saved_ire != NULL) {
1909 			ire = saved_ire;
1910 			IRE_REFHOLD(ire);
1911 			IRB_REFRELE(irb_ptr);
1912 			goto found_ire_held;
1913 		} else {
1914 			/*
1915 			 * Look for a interface default route matching the
1916 			 * args passed in. No round robin here. Just pick
1917 			 * the right one.
1918 			 */
1919 			for (ire = irb_ptr->irb_ire; ire != NULL;
1920 			    ire = ire->ire_next) {
1921 
1922 				if (!(ire->ire_type & IRE_INTERFACE))
1923 					continue;
1924 
1925 				if (ire->ire_marks & IRE_MARK_CONDEMNED)
1926 					continue;
1927 
1928 				if (ire_match_args_v6(ire, addr,
1929 				    &ipv6_all_zeros, gateway, type, ipif,
1930 				    zoneid, ihandle, tsl, flags)) {
1931 					IRE_REFHOLD(ire);
1932 					IRB_REFRELE(irb_ptr);
1933 					goto found_ire_held;
1934 				}
1935 			}
1936 			IRB_REFRELE(irb_ptr);
1937 		}
1938 	}
1939 	ASSERT(ire == NULL);
1940 	ip1dbg(("ire_ftable_lookup_v6: returning NULL ire"));
1941 	return (NULL);
1942 found_ire:
1943 	ASSERT((ire->ire_marks & IRE_MARK_CONDEMNED) == 0);
1944 	IRE_REFHOLD(ire);
1945 	rw_exit(&irb_ptr->irb_lock);
1946 
1947 found_ire_held:
1948 	if ((flags & MATCH_IRE_RJ_BHOLE) &&
1949 	    (ire->ire_flags & (RTF_BLACKHOLE | RTF_REJECT))) {
1950 		return (ire);
1951 	}
1952 	/*
1953 	 * At this point, IRE that was found must be an IRE_FORWARDTABLE
1954 	 * or IRE_CACHETABLE type.  If this is a recursive lookup and an
1955 	 * IRE_INTERFACE type was found, return that.  If it was some other
1956 	 * IRE_FORWARDTABLE type of IRE (one of the prefix types), then it
1957 	 * is necessary to fill in the  parent IRE pointed to by pire, and
1958 	 * then lookup the gateway address of  the parent.  For backwards
1959 	 * compatiblity, if this lookup returns an
1960 	 * IRE other than a IRE_CACHETABLE or IRE_INTERFACE, then one more level
1961 	 * of lookup is done.
1962 	 */
1963 	if (flags & MATCH_IRE_RECURSIVE) {
1964 		const ipif_t *gw_ipif;
1965 		int match_flags = MATCH_IRE_DSTONLY;
1966 
1967 		if (ire->ire_type & IRE_INTERFACE)
1968 			return (ire);
1969 		if (pire != NULL)
1970 			*pire = ire;
1971 		/*
1972 		 * If we can't find an IRE_INTERFACE or the caller has not
1973 		 * asked for pire, we need to REFRELE the saved_ire.
1974 		 */
1975 		saved_ire = ire;
1976 
1977 		/*
1978 		 * Currently MATCH_IRE_ILL is never used with
1979 		 * (MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT) while
1980 		 * sending out packets as MATCH_IRE_ILL is used only
1981 		 * for communicating with on-link hosts. We can't assert
1982 		 * that here as RTM_GET calls this function with
1983 		 * MATCH_IRE_ILL | MATCH_IRE_DEFAULT | MATCH_IRE_RECURSIVE.
1984 		 * We have already used the MATCH_IRE_ILL in determining
1985 		 * the right prefix route at this point. To match the
1986 		 * behavior of how we locate routes while sending out
1987 		 * packets, we don't want to use MATCH_IRE_ILL below
1988 		 * while locating the interface route.
1989 		 */
1990 		if (ire->ire_ipif != NULL)
1991 			match_flags |= MATCH_IRE_ILL_GROUP;
1992 
1993 		mutex_enter(&ire->ire_lock);
1994 		gw_addr_v6 = ire->ire_gateway_addr_v6;
1995 		mutex_exit(&ire->ire_lock);
1996 
1997 		ire = ire_route_lookup_v6(&gw_addr_v6, NULL, NULL, 0,
1998 		    ire->ire_ipif, NULL, zoneid, tsl, match_flags, ipst);
1999 		if (ire == NULL) {
2000 			/*
2001 			 * In this case we have to deal with the
2002 			 * MATCH_IRE_PARENT flag, which means the
2003 			 * parent has to be returned if ire is NULL.
2004 			 * The aim of this is to have (at least) a starting
2005 			 * ire when we want to look at all of the ires in a
2006 			 * bucket aimed at a single destination (as is the
2007 			 * case in ip_newroute_v6 for the RTF_MULTIRT
2008 			 * flagged routes).
2009 			 */
2010 			if (flags & MATCH_IRE_PARENT) {
2011 				if (pire != NULL) {
2012 					/*
2013 					 * Need an extra REFHOLD, if the
2014 					 * parent ire is returned via both
2015 					 * ire and pire.
2016 					 */
2017 					IRE_REFHOLD(saved_ire);
2018 				}
2019 				ire = saved_ire;
2020 			} else {
2021 				ire_refrele(saved_ire);
2022 				if (pire != NULL)
2023 					*pire = NULL;
2024 			}
2025 			return (ire);
2026 		}
2027 		if (ire->ire_type & (IRE_CACHETABLE | IRE_INTERFACE)) {
2028 			/*
2029 			 * If the caller did not ask for pire, release
2030 			 * it now.
2031 			 */
2032 			if (pire == NULL) {
2033 				ire_refrele(saved_ire);
2034 			}
2035 			return (ire);
2036 		}
2037 		match_flags |= MATCH_IRE_TYPE;
2038 		mutex_enter(&ire->ire_lock);
2039 		gw_addr_v6 = ire->ire_gateway_addr_v6;
2040 		mutex_exit(&ire->ire_lock);
2041 		gw_ipif = ire->ire_ipif;
2042 		ire_refrele(ire);
2043 		ire = ire_route_lookup_v6(&gw_addr_v6, NULL, NULL,
2044 		    (IRE_CACHETABLE | IRE_INTERFACE), gw_ipif, NULL, zoneid,
2045 		    NULL, match_flags, ipst);
2046 		if (ire == NULL) {
2047 			/*
2048 			 * In this case we have to deal with the
2049 			 * MATCH_IRE_PARENT flag, which means the
2050 			 * parent has to be returned if ire is NULL.
2051 			 * The aim of this is to have (at least) a starting
2052 			 * ire when we want to look at all of the ires in a
2053 			 * bucket aimed at a single destination (as is the
2054 			 * case in ip_newroute_v6 for the RTF_MULTIRT
2055 			 * flagged routes).
2056 			 */
2057 			if (flags & MATCH_IRE_PARENT) {
2058 				if (pire != NULL) {
2059 					/*
2060 					 * Need an extra REFHOLD, if the
2061 					 * parent ire is returned via both
2062 					 * ire and pire.
2063 					 */
2064 					IRE_REFHOLD(saved_ire);
2065 				}
2066 				ire = saved_ire;
2067 			} else {
2068 				ire_refrele(saved_ire);
2069 				if (pire != NULL)
2070 					*pire = NULL;
2071 			}
2072 			return (ire);
2073 		} else if (pire == NULL) {
2074 			/*
2075 			 * If the caller did not ask for pire, release
2076 			 * it now.
2077 			 */
2078 			ire_refrele(saved_ire);
2079 		}
2080 		return (ire);
2081 	}
2082 
2083 	ASSERT(pire == NULL || *pire == NULL);
2084 	return (ire);
2085 }
2086 
2087 /*
2088  * Delete the IRE cache for the gateway and all IRE caches whose
2089  * ire_gateway_addr_v6 points to this gateway, and allow them to
2090  * be created on demand by ip_newroute_v6.
2091  */
2092 void
2093 ire_clookup_delete_cache_gw_v6(const in6_addr_t *addr, zoneid_t zoneid,
2094 	ip_stack_t *ipst)
2095 {
2096 	irb_t *irb;
2097 	ire_t *ire;
2098 
2099 	irb = &ipst->ips_ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr,
2100 	    ipst->ips_ip6_cache_table_size)];
2101 	IRB_REFHOLD(irb);
2102 	for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
2103 		if (ire->ire_marks & IRE_MARK_CONDEMNED)
2104 			continue;
2105 
2106 		ASSERT(IN6_ARE_ADDR_EQUAL(&ire->ire_mask_v6, &ipv6_all_ones));
2107 		if (ire_match_args_v6(ire, addr, &ire->ire_mask_v6, 0,
2108 		    IRE_CACHE, NULL, zoneid, 0, NULL, MATCH_IRE_TYPE)) {
2109 			ire_delete(ire);
2110 		}
2111 	}
2112 	IRB_REFRELE(irb);
2113 
2114 	ire_walk_v6(ire_delete_cache_gw_v6, (char *)addr, zoneid, ipst);
2115 }
2116 
2117 /*
2118  * Looks up cache table for a route.
2119  * specific lookup can be indicated by
2120  * passing the MATCH_* flags and the
2121  * necessary parameters.
2122  */
2123 ire_t *
2124 ire_ctable_lookup_v6(const in6_addr_t *addr, const in6_addr_t *gateway,
2125     int type, const ipif_t *ipif, zoneid_t zoneid, const ts_label_t *tsl,
2126     int flags, ip_stack_t *ipst)
2127 {
2128 	ire_t *ire;
2129 	irb_t *irb_ptr;
2130 	ASSERT(addr != NULL);
2131 	ASSERT((!(flags & MATCH_IRE_GW)) || gateway != NULL);
2132 
2133 	/*
2134 	 * ire_match_args_v6() will dereference ipif MATCH_IRE_SRC or
2135 	 * MATCH_IRE_ILL is set.
2136 	 */
2137 	if ((flags & (MATCH_IRE_SRC |  MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) &&
2138 	    (ipif == NULL))
2139 		return (NULL);
2140 
2141 	irb_ptr = &ipst->ips_ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr,
2142 	    ipst->ips_ip6_cache_table_size)];
2143 	rw_enter(&irb_ptr->irb_lock, RW_READER);
2144 	for (ire = irb_ptr->irb_ire; ire; ire = ire->ire_next) {
2145 		if (ire->ire_marks & IRE_MARK_CONDEMNED)
2146 			continue;
2147 
2148 		ASSERT(IN6_ARE_ADDR_EQUAL(&ire->ire_mask_v6, &ipv6_all_ones));
2149 		if (ire_match_args_v6(ire, addr, &ire->ire_mask_v6, gateway,
2150 		    type, ipif, zoneid, 0, tsl, flags)) {
2151 			IRE_REFHOLD(ire);
2152 			rw_exit(&irb_ptr->irb_lock);
2153 			return (ire);
2154 		}
2155 	}
2156 	rw_exit(&irb_ptr->irb_lock);
2157 	return (NULL);
2158 }
2159 
2160 /*
2161  * Lookup cache. Don't return IRE_MARK_HIDDEN entries. Callers
2162  * should use ire_ctable_lookup with MATCH_IRE_MARK_HIDDEN to get
2163  * to the hidden ones.
2164  *
2165  * In general the zoneid has to match (where ALL_ZONES match all of them).
2166  * But for IRE_LOCAL we also need to handle the case where L2 should
2167  * conceptually loop back the packet. This is necessary since neither
2168  * Ethernet drivers nor Ethernet hardware loops back packets sent to their
2169  * own MAC address. This loopback is needed when the normal
2170  * routes (ignoring IREs with different zoneids) would send out the packet on
2171  * the same ill (or ill group) as the ill with which this IRE_LOCAL is
2172  * associated.
2173  *
2174  * Earlier versions of this code always matched an IRE_LOCAL independently of
2175  * the zoneid. We preserve that earlier behavior when
2176  * ip_restrict_interzone_loopback is turned off.
2177  */
2178 ire_t *
2179 ire_cache_lookup_v6(const in6_addr_t *addr, zoneid_t zoneid,
2180     const ts_label_t *tsl, ip_stack_t *ipst)
2181 {
2182 	irb_t *irb_ptr;
2183 	ire_t *ire;
2184 
2185 	irb_ptr = &ipst->ips_ip_cache_table_v6[IRE_ADDR_HASH_V6(*addr,
2186 	    ipst->ips_ip6_cache_table_size)];
2187 	rw_enter(&irb_ptr->irb_lock, RW_READER);
2188 	for (ire = irb_ptr->irb_ire; ire; ire = ire->ire_next) {
2189 		if (ire->ire_marks & (IRE_MARK_CONDEMNED|IRE_MARK_HIDDEN))
2190 			continue;
2191 		if (IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, addr)) {
2192 			/*
2193 			 * Finally, check if the security policy has any
2194 			 * restriction on using this route for the specified
2195 			 * message.
2196 			 */
2197 			if (tsl != NULL &&
2198 			    ire->ire_gw_secattr != NULL &&
2199 			    tsol_ire_match_gwattr(ire, tsl) != 0) {
2200 				continue;
2201 			}
2202 
2203 			if (zoneid == ALL_ZONES || ire->ire_zoneid == zoneid ||
2204 			    ire->ire_zoneid == ALL_ZONES) {
2205 				IRE_REFHOLD(ire);
2206 				rw_exit(&irb_ptr->irb_lock);
2207 				return (ire);
2208 			}
2209 
2210 			if (ire->ire_type == IRE_LOCAL) {
2211 				if (ipst->ips_ip_restrict_interzone_loopback &&
2212 				    !ire_local_ok_across_zones(ire, zoneid,
2213 				    (void *)addr, tsl, ipst))
2214 					continue;
2215 
2216 				IRE_REFHOLD(ire);
2217 				rw_exit(&irb_ptr->irb_lock);
2218 				return (ire);
2219 			}
2220 		}
2221 	}
2222 	rw_exit(&irb_ptr->irb_lock);
2223 	return (NULL);
2224 }
2225 
2226 /*
2227  * Locate the interface ire that is tied to the cache ire 'cire' via
2228  * cire->ire_ihandle.
2229  *
2230  * We are trying to create the cache ire for an onlink destn. or
2231  * gateway in 'cire'. We are called from ire_add_v6() in the IRE_IF_RESOLVER
2232  * case for xresolv interfaces, after the ire has come back from
2233  * an external resolver.
2234  */
2235 static ire_t *
2236 ire_ihandle_lookup_onlink_v6(ire_t *cire)
2237 {
2238 	ire_t	*ire;
2239 	int	match_flags;
2240 	int	i;
2241 	int	j;
2242 	irb_t	*irb_ptr;
2243 	ip_stack_t	*ipst = cire->ire_ipst;
2244 
2245 	ASSERT(cire != NULL);
2246 
2247 	match_flags =  MATCH_IRE_TYPE | MATCH_IRE_IHANDLE | MATCH_IRE_MASK;
2248 	/*
2249 	 * We know that the mask of the interface ire equals cire->ire_cmask.
2250 	 * (When ip_newroute_v6() created 'cire' for an on-link destn.
2251 	 * it set its cmask from the interface ire's mask)
2252 	 */
2253 	ire = ire_ftable_lookup_v6(&cire->ire_addr_v6, &cire->ire_cmask_v6,
2254 	    NULL, IRE_INTERFACE, NULL, NULL, ALL_ZONES, cire->ire_ihandle,
2255 	    NULL, match_flags, ipst);
2256 	if (ire != NULL)
2257 		return (ire);
2258 	/*
2259 	 * If we didn't find an interface ire above, we can't declare failure.
2260 	 * For backwards compatibility, we need to support prefix routes
2261 	 * pointing to next hop gateways that are not on-link.
2262 	 *
2263 	 * In the resolver/noresolver case, ip_newroute_v6() thinks
2264 	 * it is creating the cache ire for an onlink destination in 'cire'.
2265 	 * But 'cire' is not actually onlink, because ire_ftable_lookup_v6()
2266 	 * cheated it, by doing ire_route_lookup_v6() twice and returning an
2267 	 * interface ire.
2268 	 *
2269 	 * Eg. default	-	gw1			(line 1)
2270 	 *	gw1	-	gw2			(line 2)
2271 	 *	gw2	-	hme0			(line 3)
2272 	 *
2273 	 * In the above example, ip_newroute_v6() tried to create the cache ire
2274 	 * 'cire' for gw1, based on the interface route in line 3. The
2275 	 * ire_ftable_lookup_v6() above fails, because there is
2276 	 * no interface route to reach gw1. (it is gw2). We fall thru below.
2277 	 *
2278 	 * Do a brute force search based on the ihandle in a subset of the
2279 	 * forwarding tables, corresponding to cire->ire_cmask_v6. Otherwise
2280 	 * things become very complex, since we don't have 'pire' in this
2281 	 * case. (Also note that this method is not possible in the offlink
2282 	 * case because we don't know the mask)
2283 	 */
2284 	i = ip_mask_to_plen_v6(&cire->ire_cmask_v6);
2285 	if ((ipst->ips_ip_forwarding_table_v6[i]) == NULL)
2286 		return (NULL);
2287 	for (j = 0; j < ipst->ips_ip6_ftable_hash_size; j++) {
2288 		irb_ptr = &ipst->ips_ip_forwarding_table_v6[i][j];
2289 		rw_enter(&irb_ptr->irb_lock, RW_READER);
2290 		for (ire = irb_ptr->irb_ire; ire != NULL;
2291 		    ire = ire->ire_next) {
2292 			if (ire->ire_marks & IRE_MARK_CONDEMNED)
2293 				continue;
2294 			if ((ire->ire_type & IRE_INTERFACE) &&
2295 			    (ire->ire_ihandle == cire->ire_ihandle)) {
2296 				IRE_REFHOLD(ire);
2297 				rw_exit(&irb_ptr->irb_lock);
2298 				return (ire);
2299 			}
2300 		}
2301 		rw_exit(&irb_ptr->irb_lock);
2302 	}
2303 	return (NULL);
2304 }
2305 
2306 
2307 /*
2308  * Locate the interface ire that is tied to the cache ire 'cire' via
2309  * cire->ire_ihandle.
2310  *
2311  * We are trying to create the cache ire for an offlink destn based
2312  * on the cache ire of the gateway in 'cire'. 'pire' is the prefix ire
2313  * as found by ip_newroute_v6(). We are called from ip_newroute_v6() in
2314  * the IRE_CACHE case.
2315  */
2316 ire_t *
2317 ire_ihandle_lookup_offlink_v6(ire_t *cire, ire_t *pire)
2318 {
2319 	ire_t	*ire;
2320 	int	match_flags;
2321 	in6_addr_t	gw_addr;
2322 	ipif_t		*gw_ipif;
2323 	ip_stack_t	*ipst = cire->ire_ipst;
2324 
2325 	ASSERT(cire != NULL && pire != NULL);
2326 
2327 	match_flags =  MATCH_IRE_TYPE | MATCH_IRE_IHANDLE | MATCH_IRE_MASK;
2328 	/*
2329 	 * ip_newroute_v6 calls ire_ftable_lookup with MATCH_IRE_ILL only
2330 	 * for on-link hosts. We should never be here for onlink.
2331 	 * Thus, use MATCH_IRE_ILL_GROUP.
2332 	 */
2333 	if (pire->ire_ipif != NULL)
2334 		match_flags |= MATCH_IRE_ILL_GROUP;
2335 	/*
2336 	 * We know that the mask of the interface ire equals cire->ire_cmask.
2337 	 * (When ip_newroute_v6() created 'cire' for an on-link destn. it set
2338 	 * its cmask from the interface ire's mask)
2339 	 */
2340 	ire = ire_ftable_lookup_v6(&cire->ire_addr_v6, &cire->ire_cmask_v6, 0,
2341 	    IRE_INTERFACE, pire->ire_ipif, NULL, ALL_ZONES, cire->ire_ihandle,
2342 	    NULL, match_flags, ipst);
2343 	if (ire != NULL)
2344 		return (ire);
2345 	/*
2346 	 * If we didn't find an interface ire above, we can't declare failure.
2347 	 * For backwards compatibility, we need to support prefix routes
2348 	 * pointing to next hop gateways that are not on-link.
2349 	 *
2350 	 * Assume we are trying to ping some offlink destn, and we have the
2351 	 * routing table below.
2352 	 *
2353 	 * Eg.	default	- gw1		<--- pire	(line 1)
2354 	 *	gw1	- gw2				(line 2)
2355 	 *	gw2	- hme0				(line 3)
2356 	 *
2357 	 * If we already have a cache ire for gw1 in 'cire', the
2358 	 * ire_ftable_lookup_v6 above would have failed, since there is no
2359 	 * interface ire to reach gw1. We will fallthru below.
2360 	 *
2361 	 * Here we duplicate the steps that ire_ftable_lookup_v6() did in
2362 	 * getting 'cire' from 'pire', in the MATCH_IRE_RECURSIVE case.
2363 	 * The differences are the following
2364 	 * i.   We want the interface ire only, so we call
2365 	 *	ire_ftable_lookup_v6() instead of ire_route_lookup_v6()
2366 	 * ii.  We look for only prefix routes in the 1st call below.
2367 	 * ii.  We want to match on the ihandle in the 2nd call below.
2368 	 */
2369 	match_flags =  MATCH_IRE_TYPE;
2370 	if (pire->ire_ipif != NULL)
2371 		match_flags |= MATCH_IRE_ILL_GROUP;
2372 
2373 	mutex_enter(&pire->ire_lock);
2374 	gw_addr = pire->ire_gateway_addr_v6;
2375 	mutex_exit(&pire->ire_lock);
2376 	ire = ire_ftable_lookup_v6(&gw_addr, 0, 0, IRE_OFFSUBNET,
2377 	    pire->ire_ipif, NULL, ALL_ZONES, 0, NULL, match_flags, ipst);
2378 	if (ire == NULL)
2379 		return (NULL);
2380 	/*
2381 	 * At this point 'ire' corresponds to the entry shown in line 2.
2382 	 * gw_addr is 'gw2' in the example above.
2383 	 */
2384 	mutex_enter(&ire->ire_lock);
2385 	gw_addr = ire->ire_gateway_addr_v6;
2386 	mutex_exit(&ire->ire_lock);
2387 	gw_ipif = ire->ire_ipif;
2388 	ire_refrele(ire);
2389 
2390 	match_flags |= MATCH_IRE_IHANDLE;
2391 	ire = ire_ftable_lookup_v6(&gw_addr, 0, 0, IRE_INTERFACE,
2392 	    gw_ipif, NULL, ALL_ZONES, cire->ire_ihandle,
2393 	    NULL, match_flags, ipst);
2394 	return (ire);
2395 }
2396 
2397 /*
2398  * Return the IRE_LOOPBACK, IRE_IF_RESOLVER or IRE_IF_NORESOLVER
2399  * ire associated with the specified ipif.
2400  *
2401  * This might occasionally be called when IPIF_UP is not set since
2402  * the IPV6_MULTICAST_IF as well as creating interface routes
2403  * allows specifying a down ipif (ipif_lookup* match ipifs that are down).
2404  *
2405  * Note that if IPIF_NOLOCAL, IPIF_NOXMIT, or IPIF_DEPRECATED is set on
2406  * the ipif this routine might return NULL.
2407  * (Sometimes called as writer though not required by this function.)
2408  */
2409 ire_t *
2410 ipif_to_ire_v6(const ipif_t *ipif)
2411 {
2412 	ire_t	*ire;
2413 	ip_stack_t	*ipst = ipif->ipif_ill->ill_ipst;
2414 
2415 	ASSERT(ipif->ipif_isv6);
2416 	if (ipif->ipif_ire_type == IRE_LOOPBACK) {
2417 		ire = ire_ctable_lookup_v6(&ipif->ipif_v6lcl_addr, NULL,
2418 		    IRE_LOOPBACK, ipif, ALL_ZONES, NULL,
2419 		    (MATCH_IRE_TYPE | MATCH_IRE_IPIF), ipst);
2420 	} else if (ipif->ipif_flags & IPIF_POINTOPOINT) {
2421 		/* In this case we need to lookup destination address. */
2422 		ire = ire_ftable_lookup_v6(&ipif->ipif_v6pp_dst_addr,
2423 		    &ipv6_all_ones, NULL, IRE_INTERFACE, ipif, NULL, ALL_ZONES,
2424 		    0, NULL, (MATCH_IRE_TYPE | MATCH_IRE_IPIF |
2425 		    MATCH_IRE_MASK), ipst);
2426 	} else {
2427 		ire = ire_ftable_lookup_v6(&ipif->ipif_v6subnet,
2428 		    &ipif->ipif_v6net_mask, NULL, IRE_INTERFACE, ipif, NULL,
2429 		    ALL_ZONES, 0, NULL, (MATCH_IRE_TYPE | MATCH_IRE_IPIF |
2430 		    MATCH_IRE_MASK), ipst);
2431 	}
2432 	return (ire);
2433 }
2434 
2435 /*
2436  * Return B_TRUE if a multirt route is resolvable
2437  * (or if no route is resolved yet), B_FALSE otherwise.
2438  * This only works in the global zone.
2439  */
2440 boolean_t
2441 ire_multirt_need_resolve_v6(const in6_addr_t *v6dstp, const ts_label_t *tsl,
2442     ip_stack_t *ipst)
2443 {
2444 	ire_t	*first_fire;
2445 	ire_t	*first_cire;
2446 	ire_t	*fire;
2447 	ire_t	*cire;
2448 	irb_t	*firb;
2449 	irb_t	*cirb;
2450 	int	unres_cnt = 0;
2451 	boolean_t resolvable = B_FALSE;
2452 
2453 	/* Retrieve the first IRE_HOST that matches the destination */
2454 	first_fire = ire_ftable_lookup_v6(v6dstp, &ipv6_all_ones, 0, IRE_HOST,
2455 	    NULL, NULL, ALL_ZONES, 0, tsl, MATCH_IRE_MASK | MATCH_IRE_TYPE |
2456 	    MATCH_IRE_SECATTR, ipst);
2457 
2458 	/* No route at all */
2459 	if (first_fire == NULL) {
2460 		return (B_TRUE);
2461 	}
2462 
2463 	firb = first_fire->ire_bucket;
2464 	ASSERT(firb);
2465 
2466 	/* Retrieve the first IRE_CACHE ire for that destination. */
2467 	first_cire = ire_cache_lookup_v6(v6dstp, GLOBAL_ZONEID, tsl, ipst);
2468 
2469 	/* No resolved route. */
2470 	if (first_cire == NULL) {
2471 		ire_refrele(first_fire);
2472 		return (B_TRUE);
2473 	}
2474 
2475 	/* At least one route is resolved. */
2476 
2477 	cirb = first_cire->ire_bucket;
2478 	ASSERT(cirb);
2479 
2480 	/* Count the number of routes to that dest that are declared. */
2481 	IRB_REFHOLD(firb);
2482 	for (fire = first_fire; fire != NULL; fire = fire->ire_next) {
2483 		if (!(fire->ire_flags & RTF_MULTIRT))
2484 			continue;
2485 		if (!IN6_ARE_ADDR_EQUAL(&fire->ire_addr_v6, v6dstp))
2486 			continue;
2487 		unres_cnt++;
2488 	}
2489 	IRB_REFRELE(firb);
2490 
2491 
2492 	/* Then subtract the number of routes to that dst that are resolved */
2493 	IRB_REFHOLD(cirb);
2494 	for (cire = first_cire; cire != NULL; cire = cire->ire_next) {
2495 		if (!(cire->ire_flags & RTF_MULTIRT))
2496 			continue;
2497 		if (!IN6_ARE_ADDR_EQUAL(&cire->ire_addr_v6, v6dstp))
2498 			continue;
2499 		if (cire->ire_marks & (IRE_MARK_CONDEMNED|IRE_MARK_HIDDEN))
2500 			continue;
2501 		unres_cnt--;
2502 	}
2503 	IRB_REFRELE(cirb);
2504 
2505 	/* At least one route is unresolved; search for a resolvable route. */
2506 	if (unres_cnt > 0)
2507 		resolvable = ire_multirt_lookup_v6(&first_cire, &first_fire,
2508 		    MULTIRT_USESTAMP|MULTIRT_CACHEGW, tsl, ipst);
2509 
2510 	if (first_fire)
2511 		ire_refrele(first_fire);
2512 
2513 	if (first_cire)
2514 		ire_refrele(first_cire);
2515 
2516 	return (resolvable);
2517 }
2518 
2519 
2520 /*
2521  * Return B_TRUE and update *ire_arg and *fire_arg
2522  * if at least one resolvable route is found.
2523  * Return B_FALSE otherwise (all routes are resolved or
2524  * the remaining unresolved routes are all unresolvable).
2525  * This only works in the global zone.
2526  */
2527 boolean_t
2528 ire_multirt_lookup_v6(ire_t **ire_arg, ire_t **fire_arg, uint32_t flags,
2529     const ts_label_t *tsl, ip_stack_t *ipst)
2530 {
2531 	clock_t	delta;
2532 	ire_t	*best_fire = NULL;
2533 	ire_t	*best_cire = NULL;
2534 	ire_t	*first_fire;
2535 	ire_t	*first_cire;
2536 	ire_t	*fire;
2537 	ire_t	*cire;
2538 	irb_t	*firb = NULL;
2539 	irb_t	*cirb = NULL;
2540 	ire_t	*gw_ire;
2541 	boolean_t	already_resolved;
2542 	boolean_t	res;
2543 	in6_addr_t	v6dst;
2544 	in6_addr_t	v6gw;
2545 
2546 	ip2dbg(("ire_multirt_lookup_v6: *ire_arg %p, *fire_arg %p, "
2547 	    "flags %04x\n", (void *)*ire_arg, (void *)*fire_arg, flags));
2548 
2549 	ASSERT(ire_arg);
2550 	ASSERT(fire_arg);
2551 
2552 	/* Not an IRE_HOST ire; give up. */
2553 	if ((*fire_arg == NULL) ||
2554 	    ((*fire_arg)->ire_type != IRE_HOST)) {
2555 		return (B_FALSE);
2556 	}
2557 
2558 	/* This is the first IRE_HOST ire for that destination. */
2559 	first_fire = *fire_arg;
2560 	firb = first_fire->ire_bucket;
2561 	ASSERT(firb);
2562 
2563 	mutex_enter(&first_fire->ire_lock);
2564 	v6dst = first_fire->ire_addr_v6;
2565 	mutex_exit(&first_fire->ire_lock);
2566 
2567 	ip2dbg(("ire_multirt_lookup_v6: dst %08x\n",
2568 	    ntohl(V4_PART_OF_V6(v6dst))));
2569 
2570 	/*
2571 	 * Retrieve the first IRE_CACHE ire for that destination;
2572 	 * if we don't find one, no route for that dest is
2573 	 * resolved yet.
2574 	 */
2575 	first_cire = ire_cache_lookup_v6(&v6dst, GLOBAL_ZONEID, tsl, ipst);
2576 	if (first_cire) {
2577 		cirb = first_cire->ire_bucket;
2578 	}
2579 
2580 	ip2dbg(("ire_multirt_lookup_v6: first_cire %p\n", (void *)first_cire));
2581 
2582 	/*
2583 	 * Search for a resolvable route, giving the top priority
2584 	 * to routes that can be resolved without any call to the resolver.
2585 	 */
2586 	IRB_REFHOLD(firb);
2587 
2588 	if (!IN6_IS_ADDR_MULTICAST(&v6dst)) {
2589 		/*
2590 		 * For all multiroute IRE_HOST ires for that destination,
2591 		 * check if the route via the IRE_HOST's gateway is
2592 		 * resolved yet.
2593 		 */
2594 		for (fire = first_fire; fire != NULL; fire = fire->ire_next) {
2595 
2596 			if (!(fire->ire_flags & RTF_MULTIRT))
2597 				continue;
2598 			if (!IN6_ARE_ADDR_EQUAL(&fire->ire_addr_v6, &v6dst))
2599 				continue;
2600 
2601 			if (fire->ire_gw_secattr != NULL &&
2602 			    tsol_ire_match_gwattr(fire, tsl) != 0) {
2603 				continue;
2604 			}
2605 
2606 			mutex_enter(&fire->ire_lock);
2607 			v6gw = fire->ire_gateway_addr_v6;
2608 			mutex_exit(&fire->ire_lock);
2609 
2610 			ip2dbg(("ire_multirt_lookup_v6: fire %p, "
2611 			    "ire_addr %08x, ire_gateway_addr %08x\n",
2612 			    (void *)fire,
2613 			    ntohl(V4_PART_OF_V6(fire->ire_addr_v6)),
2614 			    ntohl(V4_PART_OF_V6(v6gw))));
2615 
2616 			already_resolved = B_FALSE;
2617 
2618 			if (first_cire) {
2619 				ASSERT(cirb);
2620 
2621 				IRB_REFHOLD(cirb);
2622 				/*
2623 				 * For all IRE_CACHE ires for that
2624 				 * destination.
2625 				 */
2626 				for (cire = first_cire;
2627 				    cire != NULL;
2628 				    cire = cire->ire_next) {
2629 
2630 					if (!(cire->ire_flags & RTF_MULTIRT))
2631 						continue;
2632 					if (!IN6_ARE_ADDR_EQUAL(
2633 					    &cire->ire_addr_v6, &v6dst))
2634 						continue;
2635 					if (cire->ire_marks &
2636 					    (IRE_MARK_CONDEMNED|
2637 					    IRE_MARK_HIDDEN))
2638 						continue;
2639 
2640 					if (cire->ire_gw_secattr != NULL &&
2641 					    tsol_ire_match_gwattr(cire,
2642 					    tsl) != 0) {
2643 						continue;
2644 					}
2645 
2646 					/*
2647 					 * Check if the IRE_CACHE's gateway
2648 					 * matches the IRE_HOST's gateway.
2649 					 */
2650 					if (IN6_ARE_ADDR_EQUAL(
2651 					    &cire->ire_gateway_addr_v6,
2652 					    &v6gw)) {
2653 						already_resolved = B_TRUE;
2654 						break;
2655 					}
2656 				}
2657 				IRB_REFRELE(cirb);
2658 			}
2659 
2660 			/*
2661 			 * This route is already resolved;
2662 			 * proceed with next one.
2663 			 */
2664 			if (already_resolved) {
2665 				ip2dbg(("ire_multirt_lookup_v6: found cire %p, "
2666 				    "already resolved\n", (void *)cire));
2667 				continue;
2668 			}
2669 
2670 			/*
2671 			 * The route is unresolved; is it actually
2672 			 * resolvable, i.e. is there a cache or a resolver
2673 			 * for the gateway?
2674 			 */
2675 			gw_ire = ire_route_lookup_v6(&v6gw, 0, 0, 0, NULL, NULL,
2676 			    ALL_ZONES, tsl, MATCH_IRE_RECURSIVE |
2677 			    MATCH_IRE_SECATTR, ipst);
2678 
2679 			ip2dbg(("ire_multirt_lookup_v6: looked up gw_ire %p\n",
2680 			    (void *)gw_ire));
2681 
2682 			/*
2683 			 * This route can be resolved without any call to the
2684 			 * resolver; if the MULTIRT_CACHEGW flag is set,
2685 			 * give the top priority to this ire and exit the
2686 			 * loop.
2687 			 * This occurs when an resolver reply is processed
2688 			 * through ip_wput_nondata()
2689 			 */
2690 			if ((flags & MULTIRT_CACHEGW) &&
2691 			    (gw_ire != NULL) &&
2692 			    (gw_ire->ire_type & IRE_CACHETABLE)) {
2693 				/*
2694 				 * Release the resolver associated to the
2695 				 * previous candidate best ire, if any.
2696 				 */
2697 				if (best_cire) {
2698 					ire_refrele(best_cire);
2699 					ASSERT(best_fire);
2700 				}
2701 
2702 				best_fire = fire;
2703 				best_cire = gw_ire;
2704 
2705 				ip2dbg(("ire_multirt_lookup_v6: found top prio "
2706 				    "best_fire %p, best_cire %p\n",
2707 				    (void *)best_fire, (void *)best_cire));
2708 				break;
2709 			}
2710 
2711 			/*
2712 			 * Compute the time elapsed since our preceding
2713 			 * attempt to  resolve that route.
2714 			 * If the MULTIRT_USESTAMP flag is set, we take that
2715 			 * route into account only if this time interval
2716 			 * exceeds ip_multirt_resolution_interval;
2717 			 * this prevents us from attempting to resolve a
2718 			 * broken route upon each sending of a packet.
2719 			 */
2720 			delta = lbolt - fire->ire_last_used_time;
2721 			delta = TICK_TO_MSEC(delta);
2722 
2723 			res = (boolean_t)
2724 			    ((delta > ipst->
2725 			    ips_ip_multirt_resolution_interval) ||
2726 			    (!(flags & MULTIRT_USESTAMP)));
2727 
2728 			ip2dbg(("ire_multirt_lookup_v6: fire %p, delta %lu, "
2729 			    "res %d\n",
2730 			    (void *)fire, delta, res));
2731 
2732 			if (res) {
2733 				/*
2734 				 * A resolver exists for the gateway: save
2735 				 * the current IRE_HOST ire as a candidate
2736 				 * best ire. If we later discover that a
2737 				 * top priority ire exists (i.e. no need to
2738 				 * call the resolver), then this new ire
2739 				 * will be preferred to the current one.
2740 				 */
2741 				if (gw_ire != NULL) {
2742 					if (best_fire == NULL) {
2743 						ASSERT(best_cire == NULL);
2744 
2745 						best_fire = fire;
2746 						best_cire = gw_ire;
2747 
2748 						ip2dbg(("ire_multirt_lookup_v6:"
2749 						    "found candidate "
2750 						    "best_fire %p, "
2751 						    "best_cire %p\n",
2752 						    (void *)best_fire,
2753 						    (void *)best_cire));
2754 
2755 						/*
2756 						 * If MULTIRT_CACHEGW is not
2757 						 * set, we ignore the top
2758 						 * priority ires that can
2759 						 * be resolved without any
2760 						 * call to the resolver;
2761 						 * In that case, there is
2762 						 * actually no need
2763 						 * to continue the loop.
2764 						 */
2765 						if (!(flags &
2766 						    MULTIRT_CACHEGW)) {
2767 							break;
2768 						}
2769 						continue;
2770 					}
2771 				} else {
2772 					/*
2773 					 * No resolver for the gateway: the
2774 					 * route is not resolvable.
2775 					 * If the MULTIRT_SETSTAMP flag is
2776 					 * set, we stamp the IRE_HOST ire,
2777 					 * so we will not select it again
2778 					 * during this resolution interval.
2779 					 */
2780 					if (flags & MULTIRT_SETSTAMP)
2781 						fire->ire_last_used_time =
2782 						    lbolt;
2783 				}
2784 			}
2785 
2786 			if (gw_ire != NULL)
2787 				ire_refrele(gw_ire);
2788 		}
2789 	} else { /* IN6_IS_ADDR_MULTICAST(&v6dst) */
2790 
2791 		for (fire = first_fire;
2792 		    fire != NULL;
2793 		    fire = fire->ire_next) {
2794 
2795 			if (!(fire->ire_flags & RTF_MULTIRT))
2796 				continue;
2797 			if (!IN6_ARE_ADDR_EQUAL(&fire->ire_addr_v6, &v6dst))
2798 				continue;
2799 
2800 			if (fire->ire_gw_secattr != NULL &&
2801 			    tsol_ire_match_gwattr(fire, tsl) != 0) {
2802 				continue;
2803 			}
2804 
2805 			already_resolved = B_FALSE;
2806 
2807 			mutex_enter(&fire->ire_lock);
2808 			v6gw = fire->ire_gateway_addr_v6;
2809 			mutex_exit(&fire->ire_lock);
2810 
2811 			gw_ire = ire_ftable_lookup_v6(&v6gw, 0, 0,
2812 			    IRE_INTERFACE, NULL, NULL, ALL_ZONES, 0, tsl,
2813 			    MATCH_IRE_RECURSIVE | MATCH_IRE_TYPE |
2814 			    MATCH_IRE_SECATTR, ipst);
2815 
2816 			/* No resolver for the gateway; we skip this ire. */
2817 			if (gw_ire == NULL) {
2818 				continue;
2819 			}
2820 
2821 			if (first_cire) {
2822 
2823 				IRB_REFHOLD(cirb);
2824 				/*
2825 				 * For all IRE_CACHE ires for that
2826 				 * destination.
2827 				 */
2828 				for (cire = first_cire;
2829 				    cire != NULL;
2830 				    cire = cire->ire_next) {
2831 
2832 					if (!(cire->ire_flags & RTF_MULTIRT))
2833 						continue;
2834 					if (!IN6_ARE_ADDR_EQUAL(
2835 					    &cire->ire_addr_v6, &v6dst))
2836 						continue;
2837 					if (cire->ire_marks &
2838 					    (IRE_MARK_CONDEMNED|
2839 					    IRE_MARK_HIDDEN))
2840 						continue;
2841 
2842 					if (cire->ire_gw_secattr != NULL &&
2843 					    tsol_ire_match_gwattr(cire,
2844 					    tsl) != 0) {
2845 						continue;
2846 					}
2847 
2848 					/*
2849 					 * Cache entries are linked to the
2850 					 * parent routes using the parent handle
2851 					 * (ire_phandle). If no cache entry has
2852 					 * the same handle as fire, fire is
2853 					 * still unresolved.
2854 					 */
2855 					ASSERT(cire->ire_phandle != 0);
2856 					if (cire->ire_phandle ==
2857 					    fire->ire_phandle) {
2858 						already_resolved = B_TRUE;
2859 						break;
2860 					}
2861 				}
2862 				IRB_REFRELE(cirb);
2863 			}
2864 
2865 			/*
2866 			 * This route is already resolved; proceed with
2867 			 * next one.
2868 			 */
2869 			if (already_resolved) {
2870 				ire_refrele(gw_ire);
2871 				continue;
2872 			}
2873 
2874 			/*
2875 			 * Compute the time elapsed since our preceding
2876 			 * attempt to resolve that route.
2877 			 * If the MULTIRT_USESTAMP flag is set, we take
2878 			 * that route into account only if this time
2879 			 * interval exceeds ip_multirt_resolution_interval;
2880 			 * this prevents us from attempting to resolve a
2881 			 * broken route upon each sending of a packet.
2882 			 */
2883 			delta = lbolt - fire->ire_last_used_time;
2884 			delta = TICK_TO_MSEC(delta);
2885 
2886 			res = (boolean_t)
2887 			    ((delta > ipst->
2888 			    ips_ip_multirt_resolution_interval) ||
2889 			    (!(flags & MULTIRT_USESTAMP)));
2890 
2891 			ip3dbg(("ire_multirt_lookup_v6: fire %p, delta %lx, "
2892 			    "flags %04x, res %d\n",
2893 			    (void *)fire, delta, flags, res));
2894 
2895 			if (res) {
2896 				if (best_cire) {
2897 					/*
2898 					 * Release the resolver associated
2899 					 * to the preceding candidate best
2900 					 * ire, if any.
2901 					 */
2902 					ire_refrele(best_cire);
2903 					ASSERT(best_fire);
2904 				}
2905 				best_fire = fire;
2906 				best_cire = gw_ire;
2907 				continue;
2908 			}
2909 
2910 			ire_refrele(gw_ire);
2911 		}
2912 	}
2913 
2914 	if (best_fire) {
2915 		IRE_REFHOLD(best_fire);
2916 	}
2917 	IRB_REFRELE(firb);
2918 
2919 	/* Release the first IRE_CACHE we initially looked up, if any. */
2920 	if (first_cire)
2921 		ire_refrele(first_cire);
2922 
2923 	/* Found a resolvable route. */
2924 	if (best_fire) {
2925 		ASSERT(best_cire);
2926 
2927 		if (*fire_arg)
2928 			ire_refrele(*fire_arg);
2929 		if (*ire_arg)
2930 			ire_refrele(*ire_arg);
2931 
2932 		/*
2933 		 * Update the passed arguments with the
2934 		 * resolvable multirt route we found
2935 		 */
2936 		*fire_arg = best_fire;
2937 		*ire_arg = best_cire;
2938 
2939 		ip2dbg(("ire_multirt_lookup_v6: returning B_TRUE, "
2940 		    "*fire_arg %p, *ire_arg %p\n",
2941 		    (void *)best_fire, (void *)best_cire));
2942 
2943 		return (B_TRUE);
2944 	}
2945 
2946 	ASSERT(best_cire == NULL);
2947 
2948 	ip2dbg(("ire_multirt_lookup_v6: returning B_FALSE, *fire_arg %p, "
2949 	    "*ire_arg %p\n",
2950 	    (void *)*fire_arg, (void *)*ire_arg));
2951 
2952 	/* No resolvable route. */
2953 	return (B_FALSE);
2954 }
2955 
2956 
2957 /*
2958  * Find an IRE_OFFSUBNET IRE entry for the multicast address 'v6dstp'
2959  * that goes through 'ipif'. As a fallback, a route that goes through
2960  * ipif->ipif_ill can be returned.
2961  */
2962 ire_t *
2963 ipif_lookup_multi_ire_v6(ipif_t *ipif, const in6_addr_t *v6dstp)
2964 {
2965 	ire_t	*ire;
2966 	ire_t	*save_ire = NULL;
2967 	ire_t   *gw_ire;
2968 	irb_t   *irb;
2969 	in6_addr_t v6gw;
2970 	int	match_flags = MATCH_IRE_TYPE | MATCH_IRE_ILL;
2971 	ip_stack_t	*ipst = ipif->ipif_ill->ill_ipst;
2972 
2973 	ire = ire_ftable_lookup_v6(v6dstp, 0, 0, 0, NULL, NULL, ALL_ZONES, 0,
2974 	    NULL, MATCH_IRE_DEFAULT, ipst);
2975 
2976 	if (ire == NULL)
2977 		return (NULL);
2978 
2979 	irb = ire->ire_bucket;
2980 	ASSERT(irb);
2981 
2982 	IRB_REFHOLD(irb);
2983 	ire_refrele(ire);
2984 	for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
2985 		if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6dstp) ||
2986 		    (ipif->ipif_zoneid != ire->ire_zoneid &&
2987 		    ire->ire_zoneid != ALL_ZONES)) {
2988 			continue;
2989 		}
2990 
2991 		switch (ire->ire_type) {
2992 		case IRE_DEFAULT:
2993 		case IRE_PREFIX:
2994 		case IRE_HOST:
2995 			mutex_enter(&ire->ire_lock);
2996 			v6gw = ire->ire_gateway_addr_v6;
2997 			mutex_exit(&ire->ire_lock);
2998 			gw_ire = ire_ftable_lookup_v6(&v6gw, 0, 0,
2999 			    IRE_INTERFACE, ipif, NULL, ALL_ZONES, 0,
3000 			    NULL, match_flags, ipst);
3001 
3002 			if (gw_ire != NULL) {
3003 				if (save_ire != NULL) {
3004 					ire_refrele(save_ire);
3005 				}
3006 				IRE_REFHOLD(ire);
3007 				if (gw_ire->ire_ipif == ipif) {
3008 					ire_refrele(gw_ire);
3009 
3010 					IRB_REFRELE(irb);
3011 					return (ire);
3012 				}
3013 				ire_refrele(gw_ire);
3014 				save_ire = ire;
3015 			}
3016 			break;
3017 		case IRE_IF_NORESOLVER:
3018 		case IRE_IF_RESOLVER:
3019 			if (ire->ire_ipif == ipif) {
3020 				if (save_ire != NULL) {
3021 					ire_refrele(save_ire);
3022 				}
3023 				IRE_REFHOLD(ire);
3024 
3025 				IRB_REFRELE(irb);
3026 				return (ire);
3027 			}
3028 			break;
3029 		}
3030 	}
3031 	IRB_REFRELE(irb);
3032 
3033 	return (save_ire);
3034 }
3035