xref: /titanic_44/usr/src/uts/common/inet/ip/ip_ftable.c (revision 3d0479833b8db89da4bca3e8e8e88f996686aa8e)
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 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * This file contains consumer routines of the IPv4 forwarding engine
30  */
31 
32 #include <sys/types.h>
33 #include <sys/stream.h>
34 #include <sys/stropts.h>
35 #include <sys/strlog.h>
36 #include <sys/dlpi.h>
37 #include <sys/ddi.h>
38 #include <sys/cmn_err.h>
39 #include <sys/policy.h>
40 
41 #include <sys/systm.h>
42 #include <sys/strsun.h>
43 #include <sys/kmem.h>
44 #include <sys/param.h>
45 #include <sys/socket.h>
46 #include <sys/strsubr.h>
47 #include <sys/pattr.h>
48 #include <net/if.h>
49 #include <net/route.h>
50 #include <netinet/in.h>
51 #include <net/if_dl.h>
52 #include <netinet/ip6.h>
53 #include <netinet/icmp6.h>
54 
55 #include <inet/common.h>
56 #include <inet/mi.h>
57 #include <inet/mib2.h>
58 #include <inet/ip.h>
59 #include <inet/ip_impl.h>
60 #include <inet/ip6.h>
61 #include <inet/ip_ndp.h>
62 #include <inet/arp.h>
63 #include <inet/ip_if.h>
64 #include <inet/ip_ire.h>
65 #include <inet/ip_ftable.h>
66 #include <inet/ip_rts.h>
67 #include <inet/nd.h>
68 
69 #include <net/pfkeyv2.h>
70 #include <inet/ipsec_info.h>
71 #include <inet/sadb.h>
72 #include <sys/kmem.h>
73 #include <inet/tcp.h>
74 #include <inet/ipclassifier.h>
75 #include <sys/zone.h>
76 #include <net/radix.h>
77 #include <sys/tsol/label.h>
78 #include <sys/tsol/tnet.h>
79 
80 #define	IS_DEFAULT_ROUTE(ire)	\
81 	(((ire)->ire_type & IRE_DEFAULT) || \
82 	    (((ire)->ire_type & IRE_INTERFACE) && ((ire)->ire_addr == 0)))
83 
84 /*
85  * structure for passing args between ire_ftable_lookup and ire_find_best_route
86  */
87 typedef struct ire_ftable_args_s {
88 	ipaddr_t	ift_addr;
89 	ipaddr_t	ift_mask;
90 	ipaddr_t	ift_gateway;
91 	int		ift_type;
92 	const ipif_t		*ift_ipif;
93 	zoneid_t	ift_zoneid;
94 	uint32_t	ift_ihandle;
95 	const ts_label_t	*ift_tsl;
96 	int		ift_flags;
97 	ire_t		*ift_best_ire;
98 } ire_ftable_args_t;
99 
100 static ire_t	*route_to_dst(const struct sockaddr *, zoneid_t, ip_stack_t *);
101 static ire_t   	*ire_round_robin(irb_t *, zoneid_t, ire_ftable_args_t *,
102     ip_stack_t *);
103 static void		ire_del_host_redir(ire_t *, char *);
104 static boolean_t	ire_find_best_route(struct radix_node *, void *);
105 static int	ip_send_align_hcksum_flags(mblk_t *, ill_t *);
106 
107 /*
108  * Lookup a route in forwarding table. A specific lookup is indicated by
109  * passing the required parameters and indicating the match required in the
110  * flag field.
111  *
112  * Looking for default route can be done in three ways
113  * 1) pass mask as 0 and set MATCH_IRE_MASK in flags field
114  *    along with other matches.
115  * 2) pass type as IRE_DEFAULT and set MATCH_IRE_TYPE in flags
116  *    field along with other matches.
117  * 3) if the destination and mask are passed as zeros.
118  *
119  * A request to return a default route if no route
120  * is found, can be specified by setting MATCH_IRE_DEFAULT
121  * in flags.
122  *
123  * It does not support recursion more than one level. It
124  * will do recursive lookup only when the lookup maps to
125  * a prefix or default route and MATCH_IRE_RECURSIVE flag is passed.
126  *
127  * If the routing table is setup to allow more than one level
128  * of recursion, the cleaning up cache table will not work resulting
129  * in invalid routing.
130  *
131  * Supports IP_BOUND_IF by following the ipif/ill when recursing.
132  *
133  * NOTE : When this function returns NULL, pire has already been released.
134  *	  pire is valid only when this function successfully returns an
135  *	  ire.
136  */
137 ire_t *
138 ire_ftable_lookup(ipaddr_t addr, ipaddr_t mask, ipaddr_t gateway,
139     int type, const ipif_t *ipif, ire_t **pire, zoneid_t zoneid,
140     uint32_t ihandle, const ts_label_t *tsl, int flags, ip_stack_t *ipst)
141 {
142 	ire_t *ire = NULL;
143 	ipaddr_t gw_addr;
144 	struct rt_sockaddr rdst, rmask;
145 	struct rt_entry *rt;
146 	ire_ftable_args_t margs;
147 	boolean_t found_incomplete = B_FALSE;
148 
149 	ASSERT(ipif == NULL || !ipif->ipif_isv6);
150 	ASSERT(!(flags & MATCH_IRE_WQ));
151 
152 	/*
153 	 * When we return NULL from this function, we should make
154 	 * sure that *pire is NULL so that the callers will not
155 	 * wrongly REFRELE the pire.
156 	 */
157 	if (pire != NULL)
158 		*pire = NULL;
159 	/*
160 	 * ire_match_args() will dereference ipif MATCH_IRE_SRC or
161 	 * MATCH_IRE_ILL is set.
162 	 */
163 	if ((flags & (MATCH_IRE_SRC | MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) &&
164 	    (ipif == NULL))
165 		return (NULL);
166 
167 	(void) memset(&rdst, 0, sizeof (rdst));
168 	rdst.rt_sin_len = sizeof (rdst);
169 	rdst.rt_sin_family = AF_INET;
170 	rdst.rt_sin_addr.s_addr = addr;
171 
172 	(void) memset(&rmask, 0, sizeof (rmask));
173 	rmask.rt_sin_len = sizeof (rmask);
174 	rmask.rt_sin_family = AF_INET;
175 	rmask.rt_sin_addr.s_addr = mask;
176 
177 	(void) memset(&margs, 0, sizeof (margs));
178 	margs.ift_addr = addr;
179 	margs.ift_mask = mask;
180 	margs.ift_gateway = gateway;
181 	margs.ift_type = type;
182 	margs.ift_ipif = ipif;
183 	margs.ift_zoneid = zoneid;
184 	margs.ift_ihandle = ihandle;
185 	margs.ift_tsl = tsl;
186 	margs.ift_flags = flags;
187 
188 	/*
189 	 * The flags argument passed to ire_ftable_lookup may cause the
190 	 * search to return, not the longest matching prefix, but the
191 	 * "best matching prefix", i.e., the longest prefix that also
192 	 * satisfies constraints imposed via the permutation of flags
193 	 * passed in. To achieve this, we invoke ire_match_args() on
194 	 * each matching leaf in the  radix tree. ire_match_args is
195 	 * invoked by the callback function ire_find_best_route()
196 	 * We hold the global tree lock in read mode when calling
197 	 * rn_match_args.Before dropping the global tree lock, ensure
198 	 * that the radix node can't be deleted by incrementing ire_refcnt.
199 	 */
200 	RADIX_NODE_HEAD_RLOCK(ipst->ips_ip_ftable);
201 	rt = (struct rt_entry *)ipst->ips_ip_ftable->rnh_matchaddr_args(&rdst,
202 	    ipst->ips_ip_ftable, ire_find_best_route, &margs);
203 	ire = margs.ift_best_ire;
204 	RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable);
205 
206 	if (rt == NULL) {
207 		return (NULL);
208 	} else {
209 		ASSERT(ire != NULL);
210 	}
211 
212 	DTRACE_PROBE2(ire__found, ire_ftable_args_t *, &margs, ire_t *, ire);
213 
214 	if (!IS_DEFAULT_ROUTE(ire))
215 		goto found_ire_held;
216 	/*
217 	 * If default route is found, see if default matching criteria
218 	 * are satisfied.
219 	 */
220 	if (flags & MATCH_IRE_MASK) {
221 		/*
222 		 * we were asked to match a 0 mask, and came back with
223 		 * a default route. Ok to return it.
224 		 */
225 		goto found_default_ire;
226 	}
227 	if ((flags & MATCH_IRE_TYPE) &&
228 	    (type & (IRE_DEFAULT | IRE_INTERFACE))) {
229 		/*
230 		 * we were asked to match a default ire type. Ok to return it.
231 		 */
232 		goto found_default_ire;
233 	}
234 	if (flags & MATCH_IRE_DEFAULT) {
235 		goto found_default_ire;
236 	}
237 	/*
238 	 * we found a default route, but default matching criteria
239 	 * are not specified and we are not explicitly looking for
240 	 * default.
241 	 */
242 	IRE_REFRELE(ire);
243 	return (NULL);
244 found_default_ire:
245 	/*
246 	 * round-robin only if we have more than one route in the bucket.
247 	 */
248 	if ((ire->ire_bucket->irb_ire_cnt > 1) &&
249 	    IS_DEFAULT_ROUTE(ire) &&
250 	    ((flags & (MATCH_IRE_DEFAULT | MATCH_IRE_MASK)) ==
251 	    MATCH_IRE_DEFAULT)) {
252 		ire_t *next_ire;
253 
254 		next_ire = ire_round_robin(ire->ire_bucket, zoneid, &margs,
255 		    ipst);
256 		IRE_REFRELE(ire);
257 		if (next_ire != NULL) {
258 			ire = next_ire;
259 		} else {
260 			/* no route */
261 			return (NULL);
262 		}
263 	}
264 found_ire_held:
265 	if ((flags & MATCH_IRE_RJ_BHOLE) &&
266 	    (ire->ire_flags & (RTF_BLACKHOLE | RTF_REJECT))) {
267 		return (ire);
268 	}
269 	/*
270 	 * At this point, IRE that was found must be an IRE_FORWARDTABLE
271 	 * type.  If this is a recursive lookup and an IRE_INTERFACE type was
272 	 * found, return that.  If it was some other IRE_FORWARDTABLE type of
273 	 * IRE (one of the prefix types), then it is necessary to fill in the
274 	 * parent IRE pointed to by pire, and then lookup the gateway address of
275 	 * the parent.  For backwards compatiblity, if this lookup returns an
276 	 * IRE other than a IRE_CACHETABLE or IRE_INTERFACE, then one more level
277 	 * of lookup is done.
278 	 */
279 	if (flags & MATCH_IRE_RECURSIVE) {
280 		ipif_t	*gw_ipif;
281 		int match_flags = MATCH_IRE_DSTONLY;
282 		ire_t *save_ire;
283 
284 		if (ire->ire_type & IRE_INTERFACE)
285 			return (ire);
286 		if (pire != NULL)
287 			*pire = ire;
288 		/*
289 		 * If we can't find an IRE_INTERFACE or the caller has not
290 		 * asked for pire, we need to REFRELE the save_ire.
291 		 */
292 		save_ire = ire;
293 
294 		/*
295 		 * Currently MATCH_IRE_ILL is never used with
296 		 * (MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT) while
297 		 * sending out packets as MATCH_IRE_ILL is used only
298 		 * for communicating with on-link hosts. We can't assert
299 		 * that here as RTM_GET calls this function with
300 		 * MATCH_IRE_ILL | MATCH_IRE_DEFAULT | MATCH_IRE_RECURSIVE.
301 		 * We have already used the MATCH_IRE_ILL in determining
302 		 * the right prefix route at this point. To match the
303 		 * behavior of how we locate routes while sending out
304 		 * packets, we don't want to use MATCH_IRE_ILL below
305 		 * while locating the interface route.
306 		 *
307 		 * ire_ftable_lookup may end up with an incomplete IRE_CACHE
308 		 * entry for the gateway (i.e., one for which the
309 		 * ire_nce->nce_state is not yet ND_REACHABLE). If the caller
310 		 * has specified MATCH_IRE_COMPLETE, such entries will not
311 		 * be returned; instead, we return the IF_RESOLVER ire.
312 		 */
313 		if (ire->ire_ipif != NULL)
314 			match_flags |= MATCH_IRE_ILL_GROUP;
315 
316 		ire = ire_route_lookup(ire->ire_gateway_addr, 0, 0, 0,
317 		    ire->ire_ipif, NULL, zoneid, tsl, match_flags, ipst);
318 		DTRACE_PROBE2(ftable__route__lookup1, (ire_t *), ire,
319 		    (ire_t *), save_ire);
320 		if (ire == NULL ||
321 		    ((ire->ire_type & IRE_CACHE) && ire->ire_nce &&
322 		    ire->ire_nce->nce_state != ND_REACHABLE &&
323 		    (flags & MATCH_IRE_COMPLETE))) {
324 			/*
325 			 * Do not release the parent ire if MATCH_IRE_PARENT
326 			 * is set. Also return it via ire.
327 			 */
328 			if (ire != NULL) {
329 				ire_refrele(ire);
330 				ire = NULL;
331 				found_incomplete = B_TRUE;
332 			}
333 			if (flags & MATCH_IRE_PARENT) {
334 				if (pire != NULL) {
335 					/*
336 					 * Need an extra REFHOLD, if the parent
337 					 * ire is returned via both ire and
338 					 * pire.
339 					 */
340 					IRE_REFHOLD(save_ire);
341 				}
342 				ire = save_ire;
343 			} else {
344 				ire_refrele(save_ire);
345 				if (pire != NULL)
346 					*pire = NULL;
347 			}
348 			if (!found_incomplete)
349 				return (ire);
350 		}
351 		if (ire->ire_type & (IRE_CACHETABLE | IRE_INTERFACE)) {
352 			/*
353 			 * If the caller did not ask for pire, release
354 			 * it now.
355 			 */
356 			if (pire == NULL) {
357 				ire_refrele(save_ire);
358 			}
359 			return (ire);
360 		}
361 		match_flags |= MATCH_IRE_TYPE;
362 		gw_addr = ire->ire_gateway_addr;
363 		gw_ipif = ire->ire_ipif;
364 		ire_refrele(ire);
365 		ire = ire_route_lookup(gw_addr, 0, 0,
366 		    (found_incomplete? IRE_INTERFACE :
367 		    (IRE_CACHETABLE | IRE_INTERFACE)),
368 		    gw_ipif, NULL, zoneid, tsl, match_flags, ipst);
369 		DTRACE_PROBE2(ftable__route__lookup2, (ire_t *), ire,
370 		    (ire_t *), save_ire);
371 		if (ire == NULL ||
372 		    ((ire->ire_type & IRE_CACHE) && ire->ire_nce &&
373 		    ire->ire_nce->nce_state != ND_REACHABLE &&
374 		    (flags & MATCH_IRE_COMPLETE))) {
375 			/*
376 			 * Do not release the parent ire if MATCH_IRE_PARENT
377 			 * is set. Also return it via ire.
378 			 */
379 			if (ire != NULL) {
380 				ire_refrele(ire);
381 				ire = NULL;
382 			}
383 			if (flags & MATCH_IRE_PARENT) {
384 				if (pire != NULL) {
385 					/*
386 					 * Need an extra REFHOLD, if the
387 					 * parent ire is returned via both
388 					 * ire and pire.
389 					 */
390 					IRE_REFHOLD(save_ire);
391 				}
392 				ire = save_ire;
393 			} else {
394 				ire_refrele(save_ire);
395 				if (pire != NULL)
396 					*pire = NULL;
397 			}
398 			return (ire);
399 		} else if (pire == NULL) {
400 			/*
401 			 * If the caller did not ask for pire, release
402 			 * it now.
403 			 */
404 			ire_refrele(save_ire);
405 		}
406 		return (ire);
407 	}
408 	ASSERT(pire == NULL || *pire == NULL);
409 	return (ire);
410 }
411 
412 
413 /*
414  * Find an IRE_OFFSUBNET IRE entry for the multicast address 'group'
415  * that goes through 'ipif'. As a fallback, a route that goes through
416  * ipif->ipif_ill can be returned.
417  */
418 ire_t *
419 ipif_lookup_multi_ire(ipif_t *ipif, ipaddr_t group)
420 {
421 	ire_t	*ire;
422 	ire_t	*save_ire = NULL;
423 	ire_t   *gw_ire;
424 	irb_t   *irb;
425 	ipaddr_t gw_addr;
426 	int	match_flags = MATCH_IRE_TYPE | MATCH_IRE_ILL;
427 	ip_stack_t *ipst = ipif->ipif_ill->ill_ipst;
428 
429 	ASSERT(CLASSD(group));
430 
431 	ire = ire_ftable_lookup(group, 0, 0, 0, NULL, NULL, ALL_ZONES, 0,
432 	    NULL, MATCH_IRE_DEFAULT, ipst);
433 
434 	if (ire == NULL)
435 		return (NULL);
436 
437 	irb = ire->ire_bucket;
438 	ASSERT(irb);
439 
440 	IRB_REFHOLD(irb);
441 	ire_refrele(ire);
442 	for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
443 		if (ire->ire_addr != group ||
444 		    ipif->ipif_zoneid != ire->ire_zoneid &&
445 		    ire->ire_zoneid != ALL_ZONES) {
446 			continue;
447 		}
448 
449 		switch (ire->ire_type) {
450 		case IRE_DEFAULT:
451 		case IRE_PREFIX:
452 		case IRE_HOST:
453 			gw_addr = ire->ire_gateway_addr;
454 			gw_ire = ire_ftable_lookup(gw_addr, 0, 0, IRE_INTERFACE,
455 			    ipif, NULL, ALL_ZONES, 0, NULL, match_flags, ipst);
456 
457 			if (gw_ire != NULL) {
458 				if (save_ire != NULL) {
459 					ire_refrele(save_ire);
460 				}
461 				IRE_REFHOLD(ire);
462 				if (gw_ire->ire_ipif == ipif) {
463 					ire_refrele(gw_ire);
464 
465 					IRB_REFRELE(irb);
466 					return (ire);
467 				}
468 				ire_refrele(gw_ire);
469 				save_ire = ire;
470 			}
471 			break;
472 		case IRE_IF_NORESOLVER:
473 		case IRE_IF_RESOLVER:
474 			if (ire->ire_ipif == ipif) {
475 				if (save_ire != NULL) {
476 					ire_refrele(save_ire);
477 				}
478 				IRE_REFHOLD(ire);
479 
480 				IRB_REFRELE(irb);
481 				return (ire);
482 			}
483 			break;
484 		}
485 	}
486 	IRB_REFRELE(irb);
487 
488 	return (save_ire);
489 }
490 
491 /*
492  * Find an IRE_INTERFACE for the multicast group.
493  * Allows different routes for multicast addresses
494  * in the unicast routing table (akin to 224.0.0.0 but could be more specific)
495  * which point at different interfaces. This is used when IP_MULTICAST_IF
496  * isn't specified (when sending) and when IP_ADD_MEMBERSHIP doesn't
497  * specify the interface to join on.
498  *
499  * Supports IP_BOUND_IF by following the ipif/ill when recursing.
500  */
501 ire_t *
502 ire_lookup_multi(ipaddr_t group, zoneid_t zoneid, ip_stack_t *ipst)
503 {
504 	ire_t	*ire;
505 	ipif_t	*ipif = NULL;
506 	int	match_flags = MATCH_IRE_TYPE;
507 	ipaddr_t gw_addr;
508 
509 	ire = ire_ftable_lookup(group, 0, 0, 0, NULL, NULL, zoneid,
510 	    0, NULL, MATCH_IRE_DEFAULT, ipst);
511 
512 	/* We search a resolvable ire in case of multirouting. */
513 	if ((ire != NULL) && (ire->ire_flags & RTF_MULTIRT)) {
514 		ire_t *cire = NULL;
515 		/*
516 		 * If the route is not resolvable, the looked up ire
517 		 * may be changed here. In that case, ire_multirt_lookup()
518 		 * IRE_REFRELE the original ire and change it.
519 		 */
520 		(void) ire_multirt_lookup(&cire, &ire, MULTIRT_CACHEGW,
521 		    NULL, ipst);
522 		if (cire != NULL)
523 			ire_refrele(cire);
524 	}
525 	if (ire == NULL)
526 		return (NULL);
527 	/*
528 	 * Make sure we follow ire_ipif.
529 	 *
530 	 * We need to determine the interface route through
531 	 * which the gateway will be reached. We don't really
532 	 * care which interface is picked if the interface is
533 	 * part of a group.
534 	 */
535 	if (ire->ire_ipif != NULL) {
536 		ipif = ire->ire_ipif;
537 		match_flags |= MATCH_IRE_ILL_GROUP;
538 	}
539 
540 	switch (ire->ire_type) {
541 	case IRE_DEFAULT:
542 	case IRE_PREFIX:
543 	case IRE_HOST:
544 		gw_addr = ire->ire_gateway_addr;
545 		ire_refrele(ire);
546 		ire = ire_ftable_lookup(gw_addr, 0, 0,
547 		    IRE_INTERFACE, ipif, NULL, zoneid, 0,
548 		    NULL, match_flags, ipst);
549 		return (ire);
550 	case IRE_IF_NORESOLVER:
551 	case IRE_IF_RESOLVER:
552 		return (ire);
553 	default:
554 		ire_refrele(ire);
555 		return (NULL);
556 	}
557 }
558 
559 /*
560  * Delete the passed in ire if the gateway addr matches
561  */
562 void
563 ire_del_host_redir(ire_t *ire, char *gateway)
564 {
565 	if ((ire->ire_flags & RTF_DYNAMIC) &&
566 	    (ire->ire_gateway_addr == *(ipaddr_t *)gateway))
567 		ire_delete(ire);
568 }
569 
570 /*
571  * Search for all HOST REDIRECT routes that are
572  * pointing at the specified gateway and
573  * delete them. This routine is called only
574  * when a default gateway is going away.
575  */
576 void
577 ire_delete_host_redirects(ipaddr_t gateway, ip_stack_t *ipst)
578 {
579 	struct rtfuncarg rtfarg;
580 
581 	(void) memset(&rtfarg, 0, sizeof (rtfarg));
582 	rtfarg.rt_func = ire_del_host_redir;
583 	rtfarg.rt_arg = (void *)&gateway;
584 	(void) ipst->ips_ip_ftable->rnh_walktree_mt(ipst->ips_ip_ftable,
585 	    rtfunc, &rtfarg, irb_refhold_rn, irb_refrele_rn);
586 }
587 
588 struct ihandle_arg {
589 	uint32_t ihandle;
590 	ire_t	 *ire;
591 };
592 
593 static int
594 ire_ihandle_onlink_match(struct radix_node *rn, void *arg)
595 {
596 	struct rt_entry *rt;
597 	irb_t *irb;
598 	ire_t *ire;
599 	struct ihandle_arg *ih = arg;
600 
601 	rt = (struct rt_entry *)rn;
602 	ASSERT(rt != NULL);
603 	irb = &rt->rt_irb;
604 	for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) {
605 		if ((ire->ire_type & IRE_INTERFACE) &&
606 		    (ire->ire_ihandle == ih->ihandle)) {
607 			ih->ire = ire;
608 			IRE_REFHOLD(ire);
609 			return (1);
610 		}
611 	}
612 	return (0);
613 }
614 
615 /*
616  * Locate the interface ire that is tied to the cache ire 'cire' via
617  * cire->ire_ihandle.
618  *
619  * We are trying to create the cache ire for an onlink destn. or
620  * gateway in 'cire'. We are called from ire_add_v4() in the IRE_IF_RESOLVER
621  * case, after the ire has come back from ARP.
622  */
623 ire_t *
624 ire_ihandle_lookup_onlink(ire_t *cire)
625 {
626 	ire_t	*ire;
627 	int	match_flags;
628 	struct ihandle_arg ih;
629 	ip_stack_t *ipst;
630 
631 	ASSERT(cire != NULL);
632 	ipst = cire->ire_ipst;
633 
634 	/*
635 	 * We don't need to specify the zoneid to ire_ftable_lookup() below
636 	 * because the ihandle refers to an ipif which can be in only one zone.
637 	 */
638 	match_flags =  MATCH_IRE_TYPE | MATCH_IRE_IHANDLE | MATCH_IRE_MASK;
639 	/*
640 	 * We know that the mask of the interface ire equals cire->ire_cmask.
641 	 * (When ip_newroute() created 'cire' for an on-link destn. it set its
642 	 * cmask from the interface ire's mask)
643 	 */
644 	ire = ire_ftable_lookup(cire->ire_addr, cire->ire_cmask, 0,
645 	    IRE_INTERFACE, NULL, NULL, ALL_ZONES, cire->ire_ihandle,
646 	    NULL, match_flags, ipst);
647 	if (ire != NULL)
648 		return (ire);
649 	/*
650 	 * If we didn't find an interface ire above, we can't declare failure.
651 	 * For backwards compatibility, we need to support prefix routes
652 	 * pointing to next hop gateways that are not on-link.
653 	 *
654 	 * In the resolver/noresolver case, ip_newroute() thinks it is creating
655 	 * the cache ire for an onlink destination in 'cire'. But 'cire' is
656 	 * not actually onlink, because ire_ftable_lookup() cheated it, by
657 	 * doing ire_route_lookup() twice and returning an interface ire.
658 	 *
659 	 * Eg. default	-	gw1			(line 1)
660 	 *	gw1	-	gw2			(line 2)
661 	 *	gw2	-	hme0			(line 3)
662 	 *
663 	 * In the above example, ip_newroute() tried to create the cache ire
664 	 * 'cire' for gw1, based on the interface route in line 3. The
665 	 * ire_ftable_lookup() above fails, because there is no interface route
666 	 * to reach gw1. (it is gw2). We fall thru below.
667 	 *
668 	 * Do a brute force search based on the ihandle in a subset of the
669 	 * forwarding tables, corresponding to cire->ire_cmask. Otherwise
670 	 * things become very complex, since we don't have 'pire' in this
671 	 * case. (Also note that this method is not possible in the offlink
672 	 * case because we don't know the mask)
673 	 */
674 	(void) memset(&ih, 0, sizeof (ih));
675 	ih.ihandle = cire->ire_ihandle;
676 	(void) ipst->ips_ip_ftable->rnh_walktree_mt(ipst->ips_ip_ftable,
677 	    ire_ihandle_onlink_match, &ih, irb_refhold_rn, irb_refrele_rn);
678 	return (ih.ire);
679 }
680 
681 /*
682  * IRE iterator used by ire_ftable_lookup[_v6]() to process multiple default
683  * routes. Given a starting point in the hash list (ire_origin), walk the IREs
684  * in the bucket skipping default interface routes and deleted entries.
685  * Returns the next IRE (unheld), or NULL when we're back to the starting point.
686  * Assumes that the caller holds a reference on the IRE bucket.
687  */
688 ire_t *
689 ire_get_next_default_ire(ire_t *ire, ire_t *ire_origin)
690 {
691 	ASSERT(ire_origin->ire_bucket != NULL);
692 	ASSERT(ire != NULL);
693 
694 	do {
695 		ire = ire->ire_next;
696 		if (ire == NULL)
697 			ire = ire_origin->ire_bucket->irb_ire;
698 		if (ire == ire_origin)
699 			return (NULL);
700 	} while ((ire->ire_type & IRE_INTERFACE) ||
701 	    (ire->ire_marks & IRE_MARK_CONDEMNED));
702 	ASSERT(ire != NULL);
703 	return (ire);
704 }
705 
706 static ipif_t *
707 ire_forward_src_ipif(ipaddr_t dst, ire_t *sire, ire_t *ire, ill_t *dst_ill,
708     int zoneid, ushort_t *marks)
709 {
710 	ipif_t *src_ipif;
711 	ip_stack_t *ipst = dst_ill->ill_ipst;
712 
713 	/*
714 	 * Pick the best source address from dst_ill.
715 	 *
716 	 * 1) If it is part of a multipathing group, we would
717 	 *    like to spread the inbound packets across different
718 	 *    interfaces. ipif_select_source picks a random source
719 	 *    across the different ills in the group.
720 	 *
721 	 * 2) If it is not part of a multipathing group, we try
722 	 *    to pick the source address from the destination
723 	 *    route. Clustering assumes that when we have multiple
724 	 *    prefixes hosted on an interface, the prefix of the
725 	 *    source address matches the prefix of the destination
726 	 *    route. We do this only if the address is not
727 	 *    DEPRECATED.
728 	 *
729 	 * 3) If the conn is in a different zone than the ire, we
730 	 *    need to pick a source address from the right zone.
731 	 *
732 	 * NOTE : If we hit case (1) above, the prefix of the source
733 	 *	  address picked may not match the prefix of the
734 	 *	  destination routes prefix as ipif_select_source
735 	 *	  does not look at "dst" while picking a source
736 	 *	  address.
737 	 *	  If we want the same behavior as (2), we will need
738 	 *	  to change the behavior of ipif_select_source.
739 	 */
740 
741 	if ((sire != NULL) && (sire->ire_flags & RTF_SETSRC)) {
742 		/*
743 		 * The RTF_SETSRC flag is set in the parent ire (sire).
744 		 * Check that the ipif matching the requested source
745 		 * address still exists.
746 		 */
747 		src_ipif = ipif_lookup_addr(sire->ire_src_addr, NULL,
748 		    zoneid, NULL, NULL, NULL, NULL, ipst);
749 		return (src_ipif);
750 	}
751 	*marks |= IRE_MARK_USESRC_CHECK;
752 	if ((dst_ill->ill_group != NULL) ||
753 	    (ire->ire_ipif->ipif_flags & IPIF_DEPRECATED) ||
754 	    (dst_ill->ill_usesrc_ifindex != 0)) {
755 		src_ipif = ipif_select_source(dst_ill, dst, zoneid);
756 		if (src_ipif == NULL)
757 			return (NULL);
758 
759 	} else {
760 		src_ipif = ire->ire_ipif;
761 		ASSERT(src_ipif != NULL);
762 		/* hold src_ipif for uniformity */
763 		ipif_refhold(src_ipif);
764 	}
765 	return (src_ipif);
766 }
767 
768 /*
769  * This function is called by ip_rput_noire() and ip_fast_forward()
770  * to resolve the route of incoming packet that needs to be forwarded.
771  * If the ire of the nexthop is not already in the cachetable, this
772  * routine will insert it to the table, but won't trigger ARP resolution yet.
773  * Thus unlike ip_newroute, this function adds incomplete ires to
774  * the cachetable. ARP resolution for these ires are  delayed until
775  * after all of the packet processing is completed and its ready to
776  * be sent out on the wire, Eventually, the packet transmit routine
777  * ip_xmit_v4() attempts to send a packet  to the driver. If it finds
778  * that there is no link layer information, it will do the arp
779  * resolution and queue the packet in ire->ire_nce->nce_qd_mp and
780  * then send it out once the arp resolution is over
781  * (see ip_xmit_v4()->ire_arpresolve()). This scheme is similar to
782  * the model of BSD/SunOS 4
783  *
784  * In future, the insertion of incomplete ires in the cachetable should
785  * be implemented in hostpath as well, as doing so will greatly reduce
786  * the existing complexity for code paths that depend on the context of
787  * the sender (such as IPsec).
788  *
789  * Thus this scheme of adding incomplete ires in cachetable in forwarding
790  * path can be used as a template for simplifying the hostpath.
791  */
792 
793 ire_t *
794 ire_forward(ipaddr_t dst, boolean_t *check_multirt, ire_t *supplied_ire,
795     ire_t *supplied_sire, const struct ts_label_s *tsl, ip_stack_t *ipst)
796 {
797 	ipaddr_t gw = 0;
798 	ire_t	*ire = NULL;
799 	ire_t   *sire = NULL, *save_ire;
800 	ill_t *dst_ill = NULL;
801 	int error;
802 	zoneid_t zoneid;
803 	ipif_t *src_ipif = NULL;
804 	mblk_t *res_mp;
805 	ushort_t ire_marks = 0;
806 	tsol_gcgrp_t *gcgrp = NULL;
807 	tsol_gcgrp_addr_t ga;
808 
809 	zoneid = GLOBAL_ZONEID;
810 
811 	if (supplied_ire != NULL) {
812 		/* We have arrived here from ipfil_sendpkt */
813 		ire = supplied_ire;
814 		sire = supplied_sire;
815 		goto create_irecache;
816 	}
817 
818 	ire = ire_ftable_lookup(dst, 0, 0, 0, NULL, &sire, zoneid, 0,
819 	    tsl, MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT |
820 	    MATCH_IRE_RJ_BHOLE | MATCH_IRE_PARENT|MATCH_IRE_SECATTR, ipst);
821 
822 	if (ire == NULL) {
823 		ip_rts_change(RTM_MISS, dst, 0, 0, 0, 0, 0, 0, RTA_DST, ipst);
824 		goto icmp_err_ret;
825 	}
826 
827 	/*
828 	 * If we encounter CGTP, we should  have the caller use
829 	 * ip_newroute to resolve multirt instead of this function.
830 	 * CGTP specs explicitly state that it can't be used with routers.
831 	 * This essentially prevents insertion of incomplete RTF_MULTIRT
832 	 * ires in cachetable.
833 	 */
834 	if (ip_cgtp_filter &&
835 	    ((ire->ire_flags & RTF_MULTIRT) ||
836 	    ((sire != NULL) && (sire->ire_flags & RTF_MULTIRT)))) {
837 		ip3dbg(("ire_forward: packet is to be multirouted- "
838 		    "handing it to ip_newroute\n"));
839 		if (sire != NULL)
840 			ire_refrele(sire);
841 		ire_refrele(ire);
842 		/*
843 		 * Inform caller about encountering of multirt so that
844 		 * ip_newroute() can be called.
845 		 */
846 		*check_multirt = B_TRUE;
847 		return (NULL);
848 	}
849 
850 	*check_multirt = B_FALSE;
851 
852 	/*
853 	 * Verify that the returned IRE does not have either
854 	 * the RTF_REJECT or RTF_BLACKHOLE flags set and that the IRE is
855 	 * either an IRE_CACHE, IRE_IF_NORESOLVER or IRE_IF_RESOLVER.
856 	 */
857 	if ((ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE)) ||
858 	    (ire->ire_type & (IRE_CACHE | IRE_INTERFACE)) == 0) {
859 		ip3dbg(("ire 0x%p is not cache/resolver/noresolver\n",
860 		    (void *)ire));
861 		goto icmp_err_ret;
862 	}
863 
864 	/*
865 	 * If we already have a fully resolved IRE CACHE of the
866 	 * nexthop router, just hand over the cache entry
867 	 * and we are done.
868 	 */
869 
870 	if (ire->ire_type & IRE_CACHE) {
871 
872 		/*
873 		 * If we are using this ire cache entry as a
874 		 * gateway to forward packets, chances are we
875 		 * will be using it again. So turn off
876 		 * the temporary flag, thus reducing its
877 		 * chances of getting deleted frequently.
878 		 */
879 		if (ire->ire_marks & IRE_MARK_TEMPORARY) {
880 			irb_t *irb = ire->ire_bucket;
881 			rw_enter(&irb->irb_lock, RW_WRITER);
882 			ire->ire_marks &= ~IRE_MARK_TEMPORARY;
883 			irb->irb_tmp_ire_cnt--;
884 			rw_exit(&irb->irb_lock);
885 		}
886 
887 		if (sire != NULL) {
888 			UPDATE_OB_PKT_COUNT(sire);
889 			sire->ire_last_used_time = lbolt;
890 			ire_refrele(sire);
891 		}
892 		return (ire);
893 	}
894 create_irecache:
895 	/*
896 	 * Increment the ire_ob_pkt_count field for ire if it is an
897 	 * INTERFACE (IF_RESOLVER or IF_NORESOLVER) IRE type, and
898 	 * increment the same for the parent IRE, sire, if it is some
899 	 * sort of prefix IRE (which includes DEFAULT, PREFIX, and HOST).
900 	 */
901 	if ((ire->ire_type & IRE_INTERFACE) != 0) {
902 		UPDATE_OB_PKT_COUNT(ire);
903 		ire->ire_last_used_time = lbolt;
904 	}
905 
906 	/*
907 	 * sire must be either IRE_CACHETABLE OR IRE_INTERFACE type
908 	 */
909 	if (sire != NULL) {
910 		gw = sire->ire_gateway_addr;
911 		ASSERT((sire->ire_type &
912 		    (IRE_CACHETABLE | IRE_INTERFACE)) == 0);
913 		UPDATE_OB_PKT_COUNT(sire);
914 		sire->ire_last_used_time = lbolt;
915 	}
916 
917 	/* Obtain dst_ill */
918 	dst_ill = ip_newroute_get_dst_ill(ire->ire_ipif->ipif_ill);
919 	if (dst_ill == NULL) {
920 		ip2dbg(("ire_forward no dst ill; ire 0x%p\n", (void *)ire));
921 		goto icmp_err_ret;
922 	}
923 
924 	ASSERT(src_ipif == NULL);
925 	/* Now obtain the src_ipif */
926 	src_ipif = ire_forward_src_ipif(dst, sire, ire, dst_ill,
927 	    zoneid, &ire_marks);
928 	if (src_ipif == NULL)
929 		goto icmp_err_ret;
930 
931 	switch (ire->ire_type) {
932 	case IRE_IF_NORESOLVER:
933 		/* create ire_cache for ire_addr endpoint */
934 	case IRE_IF_RESOLVER:
935 		/*
936 		 * We have the IRE_IF_RESOLVER of the nexthop gateway
937 		 * and now need to build a IRE_CACHE for it.
938 		 * In this case, we have the following :
939 		 *
940 		 * 1) src_ipif - used for getting a source address.
941 		 *
942 		 * 2) dst_ill - from which we derive ire_stq/ire_rfq. This
943 		 *    means packets using the IRE_CACHE that we will build
944 		 *    here will go out on dst_ill.
945 		 *
946 		 * 3) sire may or may not be NULL. But, the IRE_CACHE that is
947 		 *    to be created will only be tied to the IRE_INTERFACE
948 		 *    that was derived from the ire_ihandle field.
949 		 *
950 		 *    If sire is non-NULL, it means the destination is
951 		 *    off-link and we will first create the IRE_CACHE for the
952 		 *    gateway.
953 		 */
954 		res_mp = dst_ill->ill_resolver_mp;
955 		if (ire->ire_type == IRE_IF_RESOLVER &&
956 		    (!OK_RESOLVER_MP(res_mp))) {
957 			ire_refrele(ire);
958 			ire = NULL;
959 			goto out;
960 		}
961 		/*
962 		 * To be at this point in the code with a non-zero gw
963 		 * means that dst is reachable through a gateway that
964 		 * we have never resolved.  By changing dst to the gw
965 		 * addr we resolve the gateway first.
966 		 */
967 		if (gw != INADDR_ANY) {
968 			/*
969 			 * The source ipif that was determined above was
970 			 * relative to the destination address, not the
971 			 * gateway's. If src_ipif was not taken out of
972 			 * the IRE_IF_RESOLVER entry, we'll need to call
973 			 * ipif_select_source() again.
974 			 */
975 			if (src_ipif != ire->ire_ipif) {
976 				ipif_refrele(src_ipif);
977 				src_ipif = ipif_select_source(dst_ill,
978 				    gw, zoneid);
979 				if (src_ipif == NULL)
980 					goto icmp_err_ret;
981 			}
982 			dst = gw;
983 			gw = INADDR_ANY;
984 		}
985 		/*
986 		 * dst has been set to the address of the nexthop.
987 		 *
988 		 * TSol note: get security attributes of the nexthop;
989 		 * Note that the nexthop may either be a gateway, or the
990 		 * packet destination itself; Detailed explanation of
991 		 * issues involved is  provided in the  IRE_IF_NORESOLVER
992 		 * logic in ip_newroute().
993 		 */
994 		ga.ga_af = AF_INET;
995 		IN6_IPADDR_TO_V4MAPPED(dst, &ga.ga_addr);
996 		gcgrp = gcgrp_lookup(&ga, B_FALSE);
997 
998 		if (ire->ire_type == IRE_IF_NORESOLVER)
999 			dst = ire->ire_addr; /* ire_cache for tunnel endpoint */
1000 
1001 		save_ire = ire;
1002 		/*
1003 		 * create an incomplete IRE_CACHE.
1004 		 * An areq_mp will be generated in ire_arpresolve() for
1005 		 * RESOLVER interfaces.
1006 		 */
1007 		ire = ire_create(
1008 		    (uchar_t *)&dst,		/* dest address */
1009 		    (uchar_t *)&ip_g_all_ones,	/* mask */
1010 		    (uchar_t *)&src_ipif->ipif_src_addr, /* src addr */
1011 		    (uchar_t *)&gw,		/* gateway address */
1012 		    (save_ire->ire_type == IRE_IF_RESOLVER ?  NULL:
1013 		    &save_ire->ire_max_frag),
1014 		    NULL,
1015 		    dst_ill->ill_rq,		/* recv-from queue */
1016 		    dst_ill->ill_wq,		/* send-to queue */
1017 		    IRE_CACHE,			/* IRE type */
1018 		    src_ipif,
1019 		    ire->ire_mask,		/* Parent mask */
1020 		    0,
1021 		    ire->ire_ihandle,	/* Interface handle */
1022 		    0,
1023 		    &(ire->ire_uinfo),
1024 		    NULL,
1025 		    gcgrp,
1026 		    ipst);
1027 		ip1dbg(("incomplete ire_cache 0x%p\n", (void *)ire));
1028 		if (ire != NULL) {
1029 			gcgrp = NULL; /* reference now held by IRE */
1030 			ire->ire_marks |= ire_marks;
1031 			/* add the incomplete ire: */
1032 			error = ire_add(&ire, NULL, NULL, NULL, B_TRUE);
1033 			if (error == 0 && ire != NULL) {
1034 				ire->ire_max_frag = save_ire->ire_max_frag;
1035 				ip1dbg(("setting max_frag to %d in ire 0x%p\n",
1036 				    ire->ire_max_frag, (void *)ire));
1037 			} else {
1038 				ire_refrele(save_ire);
1039 				goto icmp_err_ret;
1040 			}
1041 		} else {
1042 			if (gcgrp != NULL) {
1043 				GCGRP_REFRELE(gcgrp);
1044 				gcgrp = NULL;
1045 			}
1046 		}
1047 
1048 		ire_refrele(save_ire);
1049 		break;
1050 	default:
1051 		break;
1052 	}
1053 
1054 out:
1055 	if (sire != NULL)
1056 		ire_refrele(sire);
1057 	if (dst_ill != NULL)
1058 		ill_refrele(dst_ill);
1059 	if (src_ipif != NULL)
1060 		ipif_refrele(src_ipif);
1061 	return (ire);
1062 icmp_err_ret:
1063 	if (src_ipif != NULL)
1064 		ipif_refrele(src_ipif);
1065 	if (dst_ill != NULL)
1066 		ill_refrele(dst_ill);
1067 	if (sire != NULL)
1068 		ire_refrele(sire);
1069 	if (ire != NULL) {
1070 		ire_refrele(ire);
1071 	}
1072 	/* caller needs to send icmp error message */
1073 	return (NULL);
1074 
1075 }
1076 
1077 /*
1078  * Obtain the rt_entry and rt_irb for the route to be added to
1079  * the ips_ip_ftable.
1080  * First attempt to add a node to the radix tree via rn_addroute. If the
1081  * route already exists, return the bucket for the existing route.
1082  *
1083  * Locking notes: Need to hold the global radix tree lock in write mode to
1084  * add a radix node. To prevent the node from being deleted, ire_get_bucket()
1085  * returns with a ref'ed irb_t. The ire itself is added in ire_add_v4()
1086  * while holding the irb_lock, but not the radix tree lock.
1087  */
1088 irb_t *
1089 ire_get_bucket(ire_t *ire)
1090 {
1091 	struct radix_node *rn;
1092 	struct rt_entry *rt;
1093 	struct rt_sockaddr rmask, rdst;
1094 	irb_t *irb = NULL;
1095 	ip_stack_t *ipst = ire->ire_ipst;
1096 
1097 	ASSERT(ipst->ips_ip_ftable != NULL);
1098 
1099 	/* first try to see if route exists (based on rtalloc1) */
1100 	(void) memset(&rdst, 0, sizeof (rdst));
1101 	rdst.rt_sin_len = sizeof (rdst);
1102 	rdst.rt_sin_family = AF_INET;
1103 	rdst.rt_sin_addr.s_addr = ire->ire_addr;
1104 
1105 	(void) memset(&rmask, 0, sizeof (rmask));
1106 	rmask.rt_sin_len = sizeof (rmask);
1107 	rmask.rt_sin_family = AF_INET;
1108 	rmask.rt_sin_addr.s_addr = ire->ire_mask;
1109 
1110 	/*
1111 	 * add the route. based on BSD's rtrequest1(RTM_ADD)
1112 	 */
1113 	R_Malloc(rt, rt_entry_cache,  sizeof (*rt));
1114 	(void) memset(rt, 0, sizeof (*rt));
1115 	rt->rt_nodes->rn_key = (char *)&rt->rt_dst;
1116 	rt->rt_dst = rdst;
1117 	irb = &rt->rt_irb;
1118 	irb->irb_marks |= IRB_MARK_FTABLE; /* dynamically allocated/freed */
1119 	irb->irb_ipst = ipst;
1120 	rw_init(&irb->irb_lock, NULL, RW_DEFAULT, NULL);
1121 	RADIX_NODE_HEAD_WLOCK(ipst->ips_ip_ftable);
1122 	rn = ipst->ips_ip_ftable->rnh_addaddr(&rt->rt_dst, &rmask,
1123 	    ipst->ips_ip_ftable, (struct radix_node *)rt);
1124 	if (rn == NULL) {
1125 		RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable);
1126 		Free(rt, rt_entry_cache);
1127 		rt = NULL;
1128 		irb = NULL;
1129 		RADIX_NODE_HEAD_RLOCK(ipst->ips_ip_ftable);
1130 		rn = ipst->ips_ip_ftable->rnh_lookup(&rdst, &rmask,
1131 		    ipst->ips_ip_ftable);
1132 		if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) {
1133 			/* found a non-root match */
1134 			rt = (struct rt_entry *)rn;
1135 		}
1136 	}
1137 	if (rt != NULL) {
1138 		irb = &rt->rt_irb;
1139 		IRB_REFHOLD(irb);
1140 	}
1141 	RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable);
1142 	return (irb);
1143 }
1144 
1145 /*
1146  * This function is used when the caller wants to know the outbound
1147  * interface for a packet given only the address.
1148  * If this is a offlink IP address and there are multiple
1149  * routes to this destination, this routine will utilise the
1150  * first route it finds to IP address
1151  * Return values:
1152  * 	0	- FAILURE
1153  *	nonzero	- ifindex
1154  */
1155 uint_t
1156 ifindex_lookup(const struct sockaddr *ipaddr, zoneid_t zoneid)
1157 {
1158 	uint_t ifindex = 0;
1159 	ire_t *ire;
1160 	ill_t *ill;
1161 	netstack_t *ns;
1162 	ip_stack_t *ipst;
1163 
1164 	if (zoneid == ALL_ZONES)
1165 		ns = netstack_find_by_zoneid(GLOBAL_ZONEID);
1166 	else
1167 		ns = netstack_find_by_zoneid(zoneid);
1168 	ASSERT(ns != NULL);
1169 
1170 	/*
1171 	 * For exclusive stacks we set the zoneid to zero
1172 	 * since IP uses the global zoneid in the exclusive stacks.
1173 	 */
1174 	if (ns->netstack_stackid != GLOBAL_NETSTACKID)
1175 		zoneid = GLOBAL_ZONEID;
1176 	ipst = ns->netstack_ip;
1177 
1178 	ASSERT(ipaddr->sa_family == AF_INET || ipaddr->sa_family == AF_INET6);
1179 
1180 	if ((ire =  route_to_dst(ipaddr, zoneid, ipst)) != NULL) {
1181 		ill = ire_to_ill(ire);
1182 		if (ill != NULL)
1183 			ifindex = ill->ill_phyint->phyint_ifindex;
1184 		ire_refrele(ire);
1185 	}
1186 	netstack_rele(ns);
1187 	return (ifindex);
1188 }
1189 
1190 /*
1191  * Routine to find the route to a destination. If a ifindex is supplied
1192  * it tries to match the the route to the corresponding ipif for the ifindex
1193  */
1194 static	ire_t *
1195 route_to_dst(const struct sockaddr *dst_addr, zoneid_t zoneid, ip_stack_t *ipst)
1196 {
1197 	ire_t *ire = NULL;
1198 	int match_flags;
1199 
1200 	match_flags = (MATCH_IRE_DSTONLY | MATCH_IRE_DEFAULT |
1201 	    MATCH_IRE_RECURSIVE | MATCH_IRE_RJ_BHOLE);
1202 
1203 	/* XXX pass NULL tsl for now */
1204 
1205 	if (dst_addr->sa_family == AF_INET) {
1206 		ire = ire_route_lookup(
1207 		    ((struct sockaddr_in *)dst_addr)->sin_addr.s_addr,
1208 		    0, 0, 0, NULL, NULL, zoneid, NULL, match_flags, ipst);
1209 	} else {
1210 		ire = ire_route_lookup_v6(
1211 		    &((struct sockaddr_in6 *)dst_addr)->sin6_addr,
1212 		    0, 0, 0, NULL, NULL, zoneid, NULL, match_flags, ipst);
1213 	}
1214 	return (ire);
1215 }
1216 
1217 /*
1218  * This routine is called by IP Filter to send a packet out on the wire
1219  * to a specified V4 dst (which may be onlink or offlink). The ifindex may or
1220  * may not be 0. A non-null ifindex indicates IP Filter has stipulated
1221  * an outgoing interface and requires the nexthop to be on that interface.
1222  * IP WILL NOT DO the following to the data packet before sending it out:
1223  *	a. manipulate ttl
1224  *	b. ipsec work
1225  *	c. fragmentation
1226  *
1227  * If the packet has been prepared for hardware checksum then it will be
1228  * passed off to ip_send_align_cksum() to check that the flags set on the
1229  * packet are in alignment with the capabilities of the new outgoing NIC.
1230  *
1231  * Return values:
1232  *	0:		IP was able to send of the data pkt
1233  *	ECOMM:		Could not send packet
1234  *	ENONET		No route to dst. It is up to the caller
1235  *			to send icmp unreachable error message,
1236  *	EINPROGRESS	The macaddr of the onlink dst or that
1237  *			of the offlink dst's nexthop needs to get
1238  *			resolved before packet can be sent to dst.
1239  *			Thus transmission is not guaranteed.
1240  *
1241  */
1242 
1243 int
1244 ipfil_sendpkt(const struct sockaddr *dst_addr, mblk_t *mp, uint_t ifindex,
1245     zoneid_t zoneid)
1246 {
1247 	ire_t *ire = NULL, *sire = NULL;
1248 	ire_t *ire_cache = NULL;
1249 	boolean_t   check_multirt = B_FALSE;
1250 	int value;
1251 	int match_flags;
1252 	ipaddr_t dst;
1253 	netstack_t *ns;
1254 	ip_stack_t *ipst;
1255 
1256 	ASSERT(mp != NULL);
1257 
1258 	if (zoneid == ALL_ZONES)
1259 		ns = netstack_find_by_zoneid(GLOBAL_ZONEID);
1260 	else
1261 		ns = netstack_find_by_zoneid(zoneid);
1262 	ASSERT(ns != NULL);
1263 
1264 	/*
1265 	 * For exclusive stacks we set the zoneid to zero
1266 	 * since IP uses the global zoneid in the exclusive stacks.
1267 	 */
1268 	if (ns->netstack_stackid != GLOBAL_NETSTACKID)
1269 		zoneid = GLOBAL_ZONEID;
1270 	ipst = ns->netstack_ip;
1271 
1272 	ASSERT(dst_addr->sa_family == AF_INET ||
1273 	    dst_addr->sa_family == AF_INET6);
1274 
1275 	if (dst_addr->sa_family == AF_INET) {
1276 		dst = ((struct sockaddr_in *)dst_addr)->sin_addr.s_addr;
1277 	} else {
1278 		/*
1279 		 * We dont have support for V6 yet. It will be provided
1280 		 * once RFE  6399103  has been delivered.
1281 		 * Until then, for V6 dsts, IP Filter will not call
1282 		 * this function. Instead the netinfo framework provides
1283 		 * its own code path, in ip_inject_impl(), to achieve
1284 		 * what it needs to do, for the time being.
1285 		 */
1286 		ip1dbg(("ipfil_sendpkt: no V6 support \n"));
1287 		value = ECOMM;
1288 		freemsg(mp);
1289 		goto discard;
1290 	}
1291 
1292 	/*
1293 	 * Lets get the ire. We might get the ire cache entry,
1294 	 * or the ire,sire pair needed to create the cache entry.
1295 	 * XXX pass NULL tsl for now.
1296 	 */
1297 
1298 	if (ifindex == 0) {
1299 		/* There is no supplied index. So use the FIB info */
1300 
1301 		match_flags = (MATCH_IRE_DSTONLY | MATCH_IRE_DEFAULT |
1302 		    MATCH_IRE_RECURSIVE | MATCH_IRE_RJ_BHOLE);
1303 		ire = ire_route_lookup(dst,
1304 		    0, 0, 0, NULL, &sire, zoneid, MBLK_GETLABEL(mp),
1305 		    match_flags, ipst);
1306 	} else {
1307 		ipif_t *supplied_ipif;
1308 		ill_t *ill;
1309 
1310 		match_flags = (MATCH_IRE_DSTONLY | MATCH_IRE_DEFAULT |
1311 		    MATCH_IRE_RECURSIVE| MATCH_IRE_RJ_BHOLE|
1312 		    MATCH_IRE_SECATTR);
1313 
1314 		/*
1315 		 * If supplied ifindex is non-null, the only valid
1316 		 * nexthop is one off of the interface or group corresponding
1317 		 * to the specified ifindex.
1318 		 */
1319 		ill = ill_lookup_on_ifindex(ifindex, B_FALSE,
1320 		    NULL, NULL, NULL, NULL, ipst);
1321 		if (ill != NULL) {
1322 			match_flags |= MATCH_IRE_ILL;
1323 		} else {
1324 			/* Fallback to group names if hook_emulation set */
1325 			if (ipst->ips_ipmp_hook_emulation) {
1326 				ill = ill_group_lookup_on_ifindex(ifindex,
1327 				    B_FALSE, ipst);
1328 			}
1329 			if (ill == NULL) {
1330 				ip1dbg(("ipfil_sendpkt: Could not find"
1331 				    " route to dst\n"));
1332 				value = ECOMM;
1333 				freemsg(mp);
1334 				goto discard;
1335 			}
1336 			match_flags |= MATCH_IRE_ILL_GROUP;
1337 		}
1338 		supplied_ipif = ipif_get_next_ipif(NULL, ill);
1339 
1340 		ire = ire_route_lookup(dst, 0, 0, 0, supplied_ipif,
1341 		    &sire, zoneid, MBLK_GETLABEL(mp), match_flags, ipst);
1342 		ipif_refrele(supplied_ipif);
1343 		ill_refrele(ill);
1344 	}
1345 
1346 	/*
1347 	 * Verify that the returned IRE is non-null and does
1348 	 * not have either the RTF_REJECT or RTF_BLACKHOLE
1349 	 * flags set and that the IRE is  either an IRE_CACHE,
1350 	 * IRE_IF_NORESOLVER or IRE_IF_RESOLVER.
1351 	 */
1352 	if (ire == NULL ||
1353 	    ((ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE)) ||
1354 	    (ire->ire_type & (IRE_CACHE | IRE_INTERFACE)) == 0)) {
1355 		/*
1356 		 * Either ire could not be found or we got
1357 		 * an invalid one
1358 		 */
1359 		ip1dbg(("ipfil_sendpkt: Could not find route to dst\n"));
1360 		value = ENONET;
1361 		freemsg(mp);
1362 		goto discard;
1363 	}
1364 
1365 	/* IP Filter and CGTP dont mix. So bail out if CGTP is on */
1366 	if (ip_cgtp_filter &&
1367 	    ((ire->ire_flags & RTF_MULTIRT) ||
1368 	    ((sire != NULL) && (sire->ire_flags & RTF_MULTIRT)))) {
1369 		ip1dbg(("ipfil_sendpkt: IPFilter does not work with CGTP\n"));
1370 		value = ECOMM;
1371 		freemsg(mp);
1372 		goto discard;
1373 	}
1374 
1375 	ASSERT(ire->ire_type != IRE_CACHE || ire->ire_nce != NULL);
1376 
1377 	/*
1378 	 * If needed, we will create the ire cache entry for the
1379 	 * nexthop, resolve its link-layer address and then send
1380 	 * the packet out without ttl or IPSec processing.
1381 	 */
1382 	switch (ire->ire_type) {
1383 	case IRE_IF_NORESOLVER:
1384 	case IRE_CACHE:
1385 		if (sire != NULL) {
1386 			UPDATE_OB_PKT_COUNT(sire);
1387 			sire->ire_last_used_time = lbolt;
1388 			ire_refrele(sire);
1389 		}
1390 		ire_cache = ire;
1391 		break;
1392 	case IRE_IF_RESOLVER:
1393 		/*
1394 		 * Call ire_forward(). This function
1395 		 * will, create the ire cache entry of the
1396 		 * the nexthop and adds this incomplete ire
1397 		 * to the ire cache table
1398 		 */
1399 		ire_cache = ire_forward(dst, &check_multirt, ire, sire,
1400 		    MBLK_GETLABEL(mp), ipst);
1401 		if (ire_cache == NULL) {
1402 			ip1dbg(("ipfil_sendpkt: failed to create the"
1403 			    " ire cache entry \n"));
1404 			value = ENONET;
1405 			freemsg(mp);
1406 			sire = NULL;
1407 			ire = NULL;
1408 			goto discard;
1409 		}
1410 		break;
1411 	}
1412 
1413 	if (DB_CKSUMFLAGS(mp)) {
1414 		if (ip_send_align_hcksum_flags(mp, ire_to_ill(ire_cache)))
1415 			goto cleanup;
1416 	}
1417 
1418 	/*
1419 	 * Now that we have the ire cache entry of the nexthop, call
1420 	 * ip_xmit_v4() to trigger mac addr resolution
1421 	 * if necessary and send it once ready.
1422 	 */
1423 
1424 	value = ip_xmit_v4(mp, ire_cache, NULL, B_FALSE);
1425 cleanup:
1426 	ire_refrele(ire_cache);
1427 	/*
1428 	 * At this point, the reference for these have already been
1429 	 * released within ire_forward() and/or ip_xmit_v4(). So we set
1430 	 * them to NULL to make sure we dont drop the references
1431 	 * again in case ip_xmit_v4() returns with either SEND_FAILED
1432 	 * or LLHDR_RESLV_FAILED
1433 	 */
1434 	sire = NULL;
1435 	ire = NULL;
1436 
1437 	switch (value) {
1438 	case SEND_FAILED:
1439 		ip1dbg(("ipfil_sendpkt: Send failed\n"));
1440 		value = ECOMM;
1441 		break;
1442 	case LLHDR_RESLV_FAILED:
1443 		ip1dbg(("ipfil_sendpkt: Link-layer resolution"
1444 		    "  failed\n"));
1445 		value = ECOMM;
1446 		break;
1447 	case LOOKUP_IN_PROGRESS:
1448 		netstack_rele(ns);
1449 		return (EINPROGRESS);
1450 	case SEND_PASSED:
1451 		netstack_rele(ns);
1452 		return (0);
1453 	}
1454 discard:
1455 	if (dst_addr->sa_family == AF_INET) {
1456 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
1457 	} else {
1458 		BUMP_MIB(&ipst->ips_ip6_mib, ipIfStatsOutDiscards);
1459 	}
1460 	if (ire != NULL)
1461 		ire_refrele(ire);
1462 	if (sire != NULL)
1463 		ire_refrele(sire);
1464 	netstack_rele(ns);
1465 	return (value);
1466 }
1467 
1468 
1469 /*
1470  * We don't check for dohwcksum in here because it should be being used
1471  * elsewhere to control what flags are being set on the mblk.  That is,
1472  * if DB_CKSUMFLAGS() is non-zero then we assume dohwcksum to be true
1473  * for this packet.
1474  *
1475  * This function assumes that it is *only* being called for TCP or UDP
1476  * packets and nothing else.
1477  */
1478 static int
1479 ip_send_align_hcksum_flags(mblk_t *mp, ill_t *ill)
1480 {
1481 	int illhckflags;
1482 	int mbhckflags;
1483 	uint16_t *up;
1484 	uint32_t cksum;
1485 	ipha_t *ipha;
1486 	ip6_t *ip6;
1487 	int proto;
1488 	int ipversion;
1489 	int length;
1490 	int start;
1491 	ip6_pkt_t ipp;
1492 
1493 	mbhckflags = DB_CKSUMFLAGS(mp);
1494 	ASSERT(mbhckflags != 0);
1495 	ASSERT(mp->b_datap->db_type == M_DATA);
1496 	/*
1497 	 * Since this function only knows how to manage the hardware checksum
1498 	 * issue, reject and packets that have flags set on the aside from
1499 	 * checksum related attributes as we cannot necessarily safely map
1500 	 * that packet onto the new NIC.  Packets that can be potentially
1501 	 * dropped here include those marked for LSO.
1502 	 */
1503 	if ((mbhckflags &
1504 	    ~(HCK_FULLCKSUM|HCK_PARTIALCKSUM|HCK_IPV4_HDRCKSUM)) != 0) {
1505 		DTRACE_PROBE2(pbr__incapable, (mblk_t *), mp, (ill_t *), ill);
1506 		freemsg(mp);
1507 		return (-1);
1508 	}
1509 
1510 	ipha = (ipha_t *)mp->b_rptr;
1511 
1512 	/*
1513 	 * Find out what the new NIC is capable of, if anything, and
1514 	 * only allow it to be used with M_DATA mblks being sent out.
1515 	 */
1516 	if (ILL_HCKSUM_CAPABLE(ill)) {
1517 		illhckflags = ill->ill_hcksum_capab->ill_hcksum_txflags;
1518 	} else {
1519 		/*
1520 		 * No capabilities, so turn off everything.
1521 		 */
1522 		illhckflags = 0;
1523 		(void) hcksum_assoc(mp, NULL, NULL, 0, 0, 0, 0, 0, 0);
1524 		mp->b_datap->db_struioflag &= ~STRUIO_IP;
1525 	}
1526 
1527 	DTRACE_PROBE4(pbr__info__a, (mblk_t *), mp, (ill_t *), ill,
1528 	    uint32_t, illhckflags, uint32_t, mbhckflags);
1529 	/*
1530 	 * This block of code that looks for the position of the TCP/UDP
1531 	 * checksum is early in this function because we need to know
1532 	 * what needs to be blanked out for the hardware checksum case.
1533 	 *
1534 	 * That we're in this function implies that the packet is either
1535 	 * TCP or UDP on Solaris, so checks are made for one protocol and
1536 	 * if that fails, the other is therefore implied.
1537 	 */
1538 	ipversion = IPH_HDR_VERSION(ipha);
1539 
1540 	if (ipversion == IPV4_VERSION) {
1541 		proto = ipha->ipha_protocol;
1542 		if (proto == IPPROTO_TCP) {
1543 			up = IPH_TCPH_CHECKSUMP(ipha, IP_SIMPLE_HDR_LENGTH);
1544 		} else {
1545 			up = IPH_UDPH_CHECKSUMP(ipha, IP_SIMPLE_HDR_LENGTH);
1546 		}
1547 	} else {
1548 		uint8_t lasthdr;
1549 
1550 		/*
1551 		 * Nothing I've seen indicates that IPv6 checksum'ing
1552 		 * precludes the presence of extension headers, so we
1553 		 * can't just look at the next header value in the IPv6
1554 		 * packet header to see if it is TCP/UDP.
1555 		 */
1556 		ip6 = (ip6_t *)ipha;
1557 		(void) memset(&ipp, 0, sizeof (ipp));
1558 		start = ip_find_hdr_v6(mp, ip6, &ipp, &lasthdr);
1559 		proto = lasthdr;
1560 
1561 		if (proto == IPPROTO_TCP) {
1562 			up = IPH_TCPH_CHECKSUMP(ipha, start);
1563 		} else {
1564 			up = IPH_UDPH_CHECKSUMP(ipha, start);
1565 		}
1566 	}
1567 
1568 	/*
1569 	 * The first case here is easiest:
1570 	 * mblk hasn't asked for full checksum, but the card supports it.
1571 	 *
1572 	 * In addition, check for IPv4 header capability.  Note that only
1573 	 * the mblk flag is checked and not ipversion.
1574 	 */
1575 	if ((((illhckflags & HCKSUM_INET_FULL_V4) && (ipversion == 4)) ||
1576 	    (((illhckflags & HCKSUM_INET_FULL_V6) && (ipversion == 6)))) &&
1577 	    ((mbhckflags & (HCK_FULLCKSUM|HCK_PARTIALCKSUM)) != 0)) {
1578 		int newflags = HCK_FULLCKSUM;
1579 
1580 		if ((mbhckflags & HCK_IPV4_HDRCKSUM) != 0) {
1581 			if ((illhckflags & HCKSUM_IPHDRCKSUM) != 0) {
1582 				newflags |= HCK_IPV4_HDRCKSUM;
1583 			} else {
1584 				/*
1585 				 * Rather than call a function, just inline
1586 				 * the computation of the basic IPv4 header.
1587 				 */
1588 				cksum = (ipha->ipha_dst >> 16) +
1589 				    (ipha->ipha_dst & 0xFFFF) +
1590 				    (ipha->ipha_src >> 16) +
1591 				    (ipha->ipha_src & 0xFFFF);
1592 				IP_HDR_CKSUM(ipha, cksum,
1593 				    ((uint32_t *)ipha)[0],
1594 				    ((uint16_t *)ipha)[4]);
1595 			}
1596 		}
1597 
1598 		*up = 0;
1599 		(void) hcksum_assoc(mp, NULL, NULL, 0, 0, 0, 0,
1600 		    newflags, 0);
1601 		return (0);
1602 	}
1603 
1604 	DTRACE_PROBE2(pbr__info__b, int, ipversion, int, proto);
1605 
1606 	/*
1607 	 * Start calculating the pseudo checksum over the IP packet header.
1608 	 * Although the final pseudo checksum used by TCP/UDP consists of
1609 	 * more than just the address fields, we can use the result of
1610 	 * adding those together a little bit further down for IPv4.
1611 	 */
1612 	if (ipversion == IPV4_VERSION) {
1613 		cksum = (ipha->ipha_dst >> 16) + (ipha->ipha_dst & 0xFFFF) +
1614 		    (ipha->ipha_src >> 16) + (ipha->ipha_src & 0xFFFF);
1615 		start = IP_SIMPLE_HDR_LENGTH;
1616 		length = ntohs(ipha->ipha_length);
1617 		DTRACE_PROBE3(pbr__info__e, uint32_t, ipha->ipha_src,
1618 		    uint32_t, ipha->ipha_dst, int, cksum);
1619 	} else {
1620 		uint16_t *pseudo;
1621 
1622 		pseudo = (uint16_t *)&ip6->ip6_src;
1623 
1624 		/* calculate pseudo-header checksum */
1625 		cksum = pseudo[0] + pseudo[1] + pseudo[2] + pseudo[3] +
1626 		    pseudo[4] + pseudo[5] + pseudo[6] + pseudo[7] +
1627 		    pseudo[8] + pseudo[9] + pseudo[10] + pseudo[11] +
1628 		    pseudo[12] + pseudo[13] + pseudo[14] + pseudo[15];
1629 
1630 		length = ntohs(ip6->ip6_plen) + sizeof (ip6_t);
1631 	}
1632 
1633 	/* Fold the initial sum */
1634 	cksum = (cksum & 0xffff) + (cksum >> 16);
1635 
1636 	/*
1637 	 * If the packet was asking for an IPv4 header checksum to be
1638 	 * calculated but the interface doesn't support that, fill it in
1639 	 * using our pseudo checksum as a starting point.
1640 	 */
1641 	if (((mbhckflags & HCK_IPV4_HDRCKSUM) != 0) &&
1642 	    ((illhckflags & HCKSUM_IPHDRCKSUM) == 0)) {
1643 		/*
1644 		 * IP_HDR_CKSUM uses the 2rd arg to the macro in a destructive
1645 		 * way so pass in a copy of the checksum calculated thus far.
1646 		 */
1647 		uint32_t ipsum = cksum;
1648 
1649 		DB_CKSUMFLAGS(mp) &= ~HCK_IPV4_HDRCKSUM;
1650 
1651 		IP_HDR_CKSUM(ipha, ipsum, ((uint32_t *)ipha)[0],
1652 		    ((uint16_t *)ipha)[4]);
1653 	}
1654 
1655 	DTRACE_PROBE3(pbr__info__c, int, start, int, length, int, cksum);
1656 
1657 	if (proto == IPPROTO_TCP) {
1658 		cksum += IP_TCP_CSUM_COMP;
1659 	} else {
1660 		cksum += IP_UDP_CSUM_COMP;
1661 	}
1662 	cksum += htons(length - start);
1663 	cksum = (cksum & 0xffff) + (cksum >> 16);
1664 
1665 	/*
1666 	 * For TCP/UDP, we either want to setup the packet for partial
1667 	 * checksum or we want to do it all ourselves because the NIC
1668 	 * offers no support for either partial or full checksum.
1669 	 */
1670 	if ((illhckflags & HCKSUM_INET_PARTIAL) != 0) {
1671 		/*
1672 		 * The only case we care about here is if the mblk was
1673 		 * previously set for full checksum offload.  If it was
1674 		 * marked for partial (and the NIC does partial), then
1675 		 * we have nothing to do.  Similarly if the packet was
1676 		 * not set for partial or full, we do nothing as this
1677 		 * is cheaper than more work to set something up.
1678 		 */
1679 		if ((mbhckflags & HCK_FULLCKSUM) != 0) {
1680 			uint32_t offset;
1681 
1682 			if (proto == IPPROTO_TCP) {
1683 				offset = TCP_CHECKSUM_OFFSET;
1684 			} else {
1685 				offset = UDP_CHECKSUM_OFFSET;
1686 			}
1687 			*up = cksum;
1688 
1689 			DTRACE_PROBE3(pbr__info__f, int, length - start, int,
1690 			    cksum, int, offset);
1691 
1692 			(void) hcksum_assoc(mp, NULL, NULL, start,
1693 			    start + offset, length, 0,
1694 			    DB_CKSUMFLAGS(mp) | HCK_PARTIALCKSUM, 0);
1695 		}
1696 
1697 	} else if (mbhckflags & (HCK_FULLCKSUM|HCK_PARTIALCKSUM)) {
1698 		DB_CKSUMFLAGS(mp) &= ~(HCK_PARTIALCKSUM|HCK_FULLCKSUM);
1699 
1700 		*up = 0;
1701 		*up = IP_CSUM(mp, start, cksum);
1702 	}
1703 
1704 	DTRACE_PROBE4(pbr__info__d, (mblk_t *), mp, (ipha_t *), ipha,
1705 	    (uint16_t *), up, int, cksum);
1706 	return (0);
1707 }
1708 
1709 
1710 /* ire_walk routine invoked for ip_ire_report for each IRE. */
1711 void
1712 ire_report_ftable(ire_t *ire, char *m)
1713 {
1714 	char	buf1[16];
1715 	char	buf2[16];
1716 	char	buf3[16];
1717 	char	buf4[16];
1718 	uint_t	fo_pkt_count;
1719 	uint_t	ib_pkt_count;
1720 	int	ref;
1721 	uint_t	print_len, buf_len;
1722 	mblk_t 	*mp = (mblk_t *)m;
1723 
1724 	if (ire->ire_type & IRE_CACHETABLE)
1725 		return;
1726 	buf_len = mp->b_datap->db_lim - mp->b_wptr;
1727 	if (buf_len <= 0)
1728 		return;
1729 
1730 	/* Number of active references of this ire */
1731 	ref = ire->ire_refcnt;
1732 	/* "inbound" to a non local address is a forward */
1733 	ib_pkt_count = ire->ire_ib_pkt_count;
1734 	fo_pkt_count = 0;
1735 	if (!(ire->ire_type & (IRE_LOCAL|IRE_BROADCAST))) {
1736 		fo_pkt_count = ib_pkt_count;
1737 		ib_pkt_count = 0;
1738 	}
1739 	print_len = snprintf((char *)mp->b_wptr, buf_len,
1740 	    MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR "%5d "
1741 	    "%s %s %s %s %05d %05ld %06ld %08d %03d %06d %09d %09d %06d %08d "
1742 	    "%04d %08d %08d %d/%d/%d %s\n",
1743 	    (void *)ire, (void *)ire->ire_rfq, (void *)ire->ire_stq,
1744 	    (int)ire->ire_zoneid,
1745 	    ip_dot_addr(ire->ire_addr, buf1), ip_dot_addr(ire->ire_mask, buf2),
1746 	    ip_dot_addr(ire->ire_src_addr, buf3),
1747 	    ip_dot_addr(ire->ire_gateway_addr, buf4),
1748 	    ire->ire_max_frag, ire->ire_uinfo.iulp_rtt,
1749 	    ire->ire_uinfo.iulp_rtt_sd,
1750 	    ire->ire_uinfo.iulp_ssthresh, ref,
1751 	    ire->ire_uinfo.iulp_rtomax,
1752 	    (ire->ire_uinfo.iulp_tstamp_ok ? 1: 0),
1753 	    (ire->ire_uinfo.iulp_wscale_ok ? 1: 0),
1754 	    (ire->ire_uinfo.iulp_ecn_ok ? 1: 0),
1755 	    (ire->ire_uinfo.iulp_pmtud_ok ? 1: 0),
1756 	    ire->ire_uinfo.iulp_sack,
1757 	    ire->ire_uinfo.iulp_spipe, ire->ire_uinfo.iulp_rpipe,
1758 	    ib_pkt_count, ire->ire_ob_pkt_count, fo_pkt_count,
1759 	    ip_nv_lookup(ire_nv_tbl, (int)ire->ire_type));
1760 	if (print_len < buf_len) {
1761 		mp->b_wptr += print_len;
1762 	} else {
1763 		mp->b_wptr += buf_len;
1764 	}
1765 }
1766 
1767 /*
1768  * callback function provided by ire_ftable_lookup when calling
1769  * rn_match_args(). Invoke ire_match_args on each matching leaf node in
1770  * the radix tree.
1771  */
1772 boolean_t
1773 ire_find_best_route(struct radix_node *rn, void *arg)
1774 {
1775 	struct rt_entry *rt = (struct rt_entry *)rn;
1776 	irb_t *irb_ptr;
1777 	ire_t *ire;
1778 	ire_ftable_args_t *margs = arg;
1779 	ipaddr_t match_mask;
1780 
1781 	ASSERT(rt != NULL);
1782 
1783 	irb_ptr = &rt->rt_irb;
1784 
1785 	if (irb_ptr->irb_ire_cnt == 0)
1786 		return (B_FALSE);
1787 
1788 	rw_enter(&irb_ptr->irb_lock, RW_READER);
1789 	for (ire = irb_ptr->irb_ire; ire != NULL; ire = ire->ire_next) {
1790 		if (ire->ire_marks & IRE_MARK_CONDEMNED)
1791 			continue;
1792 		if (margs->ift_flags & MATCH_IRE_MASK)
1793 			match_mask = margs->ift_mask;
1794 		else
1795 			match_mask = ire->ire_mask;
1796 
1797 		if (ire_match_args(ire, margs->ift_addr, match_mask,
1798 		    margs->ift_gateway, margs->ift_type, margs->ift_ipif,
1799 		    margs->ift_zoneid, margs->ift_ihandle, margs->ift_tsl,
1800 		    margs->ift_flags)) {
1801 			IRE_REFHOLD(ire);
1802 			rw_exit(&irb_ptr->irb_lock);
1803 			margs->ift_best_ire = ire;
1804 			return (B_TRUE);
1805 		}
1806 	}
1807 	rw_exit(&irb_ptr->irb_lock);
1808 	return (B_FALSE);
1809 }
1810 
1811 /*
1812  * ftable irb_t structures are dynamically allocated, and we need to
1813  * check if the irb_t (and associated ftable tree attachment) needs to
1814  * be cleaned up when the irb_refcnt goes to 0. The conditions that need
1815  * be verified are:
1816  * - no other walkers of the irebucket, i.e., quiescent irb_refcnt,
1817  * - no other threads holding references to ire's in the bucket,
1818  *   i.e., irb_nire == 0
1819  * - no active ire's in the bucket, i.e., irb_ire_cnt == 0
1820  * - need to hold the global tree lock and irb_lock in write mode.
1821  */
1822 void
1823 irb_refrele_ftable(irb_t *irb)
1824 {
1825 	for (;;) {
1826 		rw_enter(&irb->irb_lock, RW_WRITER);
1827 		ASSERT(irb->irb_refcnt != 0);
1828 		if (irb->irb_refcnt != 1) {
1829 			/*
1830 			 * Someone has a reference to this radix node
1831 			 * or there is some bucket walker.
1832 			 */
1833 			irb->irb_refcnt--;
1834 			rw_exit(&irb->irb_lock);
1835 			return;
1836 		} else {
1837 			/*
1838 			 * There is no other walker, nor is there any
1839 			 * other thread that holds a direct ref to this
1840 			 * radix node. Do the clean up if needed. Call
1841 			 * to ire_unlink will clear the IRB_MARK_CONDEMNED flag
1842 			 */
1843 			if (irb->irb_marks & IRB_MARK_CONDEMNED)  {
1844 				ire_t *ire_list;
1845 
1846 				ire_list = ire_unlink(irb);
1847 				rw_exit(&irb->irb_lock);
1848 
1849 				if (ire_list != NULL)
1850 					ire_cleanup(ire_list);
1851 				/*
1852 				 * more CONDEMNED entries could have
1853 				 * been added while we dropped the lock,
1854 				 * so we have to re-check.
1855 				 */
1856 				continue;
1857 			}
1858 
1859 			/*
1860 			 * Now check if there are still any ires
1861 			 * associated with this radix node.
1862 			 */
1863 			if (irb->irb_nire != 0) {
1864 				/*
1865 				 * someone is still holding on
1866 				 * to ires in this bucket
1867 				 */
1868 				irb->irb_refcnt--;
1869 				rw_exit(&irb->irb_lock);
1870 				return;
1871 			} else {
1872 				/*
1873 				 * Everything is clear. Zero walkers,
1874 				 * Zero threads with a ref to this
1875 				 * radix node, Zero ires associated with
1876 				 * this radix node. Due to lock order,
1877 				 * check the above conditions again
1878 				 * after grabbing all locks in the right order
1879 				 */
1880 				rw_exit(&irb->irb_lock);
1881 				if (irb_inactive(irb))
1882 					return;
1883 				/*
1884 				 * irb_inactive could not free the irb.
1885 				 * See if there are any walkers, if not
1886 				 * try to clean up again.
1887 				 */
1888 			}
1889 		}
1890 	}
1891 }
1892 
1893 /*
1894  * IRE iterator used by ire_ftable_lookup() to process multiple default
1895  * routes. Given a starting point in the hash list (ire_origin), walk the IREs
1896  * in the bucket skipping default interface routes and deleted entries.
1897  * Returns the next IRE (unheld), or NULL when we're back to the starting point.
1898  * Assumes that the caller holds a reference on the IRE bucket.
1899  *
1900  * In the absence of good IRE_DEFAULT routes, this function will return
1901  * the first IRE_INTERFACE route found (if any).
1902  */
1903 ire_t *
1904 ire_round_robin(irb_t *irb_ptr, zoneid_t zoneid, ire_ftable_args_t *margs,
1905 	ip_stack_t *ipst)
1906 {
1907 	ire_t	*ire_origin;
1908 	ire_t	*ire, *maybe_ire = NULL;
1909 
1910 	rw_enter(&irb_ptr->irb_lock, RW_WRITER);
1911 	ire_origin = irb_ptr->irb_rr_origin;
1912 	if (ire_origin != NULL) {
1913 		ire_origin = ire_origin->ire_next;
1914 		IRE_FIND_NEXT_ORIGIN(ire_origin);
1915 	}
1916 
1917 	if (ire_origin == NULL) {
1918 		/*
1919 		 * first time through routine, or we dropped off the end
1920 		 * of list.
1921 		 */
1922 		ire_origin = irb_ptr->irb_ire;
1923 		IRE_FIND_NEXT_ORIGIN(ire_origin);
1924 	}
1925 	irb_ptr->irb_rr_origin = ire_origin;
1926 	IRB_REFHOLD_LOCKED(irb_ptr);
1927 	rw_exit(&irb_ptr->irb_lock);
1928 
1929 	DTRACE_PROBE2(ire__rr__origin, (irb_t *), irb_ptr,
1930 	    (ire_t *), ire_origin);
1931 
1932 	/*
1933 	 * Round-robin the routers list looking for a route that
1934 	 * matches the passed in parameters.
1935 	 * We start with the ire we found above and we walk the hash
1936 	 * list until we're back where we started. It doesn't matter if
1937 	 * routes are added or deleted by other threads - we know this
1938 	 * ire will stay in the list because we hold a reference on the
1939 	 * ire bucket.
1940 	 */
1941 	ire = ire_origin;
1942 	while (ire != NULL) {
1943 		int match_flags = MATCH_IRE_TYPE | MATCH_IRE_SECATTR;
1944 		ire_t *rire;
1945 
1946 		if (ire->ire_marks & IRE_MARK_CONDEMNED)
1947 			goto next_ire;
1948 
1949 		if (!ire_match_args(ire, margs->ift_addr, (ipaddr_t)0,
1950 		    margs->ift_gateway, margs->ift_type, margs->ift_ipif,
1951 		    margs->ift_zoneid, margs->ift_ihandle, margs->ift_tsl,
1952 		    margs->ift_flags))
1953 			goto next_ire;
1954 
1955 		if (ire->ire_type & IRE_INTERFACE) {
1956 			/*
1957 			 * keep looking to see if there is a non-interface
1958 			 * default ire, but save this one as a last resort.
1959 			 */
1960 			if (maybe_ire == NULL)
1961 				maybe_ire = ire;
1962 			goto next_ire;
1963 		}
1964 
1965 		if (zoneid == ALL_ZONES) {
1966 			IRE_REFHOLD(ire);
1967 			IRB_REFRELE(irb_ptr);
1968 			return (ire);
1969 		}
1970 		/*
1971 		 * When we're in a non-global zone, we're only
1972 		 * interested in routers that are
1973 		 * reachable through ipifs within our zone.
1974 		 */
1975 		if (ire->ire_ipif != NULL) {
1976 			match_flags |= MATCH_IRE_ILL_GROUP;
1977 		}
1978 		rire = ire_route_lookup(ire->ire_gateway_addr, 0, 0,
1979 		    IRE_INTERFACE, ire->ire_ipif, NULL, zoneid, margs->ift_tsl,
1980 		    match_flags, ipst);
1981 		if (rire != NULL) {
1982 			ire_refrele(rire);
1983 			IRE_REFHOLD(ire);
1984 			IRB_REFRELE(irb_ptr);
1985 			return (ire);
1986 		}
1987 next_ire:
1988 		ire = (ire->ire_next ?  ire->ire_next : irb_ptr->irb_ire);
1989 		if (ire == ire_origin)
1990 			break;
1991 	}
1992 	if (maybe_ire != NULL)
1993 		IRE_REFHOLD(maybe_ire);
1994 	IRB_REFRELE(irb_ptr);
1995 	return (maybe_ire);
1996 }
1997 
1998 void
1999 irb_refhold_rn(struct radix_node *rn)
2000 {
2001 	if ((rn->rn_flags & RNF_ROOT) == 0)
2002 		IRB_REFHOLD(&((rt_t *)(rn))->rt_irb);
2003 }
2004 
2005 void
2006 irb_refrele_rn(struct radix_node *rn)
2007 {
2008 	if ((rn->rn_flags & RNF_ROOT) == 0)
2009 		irb_refrele_ftable(&((rt_t *)(rn))->rt_irb);
2010 }
2011