xref: /titanic_41/usr/src/uts/common/inet/ip/ip_mroute.c (revision 5bbb4db2c3f208d12bf0fd11769728f9e5ba66a2)
1 /*
2  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
3  * Use is subject to license terms.
4  */
5 /*
6  * CDDL HEADER START
7  *
8  * The contents of this file are subject to the terms of the
9  * Common Development and Distribution License (the "License").
10  * You may not use this file except in compliance with the License.
11  *
12  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
13  * or http://www.opensolaris.org/os/licensing.
14  * See the License for the specific language governing permissions
15  * and limitations under the License.
16  *
17  * When distributing Covered Code, include this CDDL HEADER in each
18  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
19  * If applicable, add the following below this CDDL HEADER, with the
20  * fields enclosed by brackets "[]" replaced with your own identifying
21  * information: Portions Copyright [yyyy] [name of copyright owner]
22  *
23  * CDDL HEADER END
24  */
25 /*
26  * Copyright 2008 Sun Microsystems, Inc.
27  * All rights reserved.  Use is subject to license terms.
28  */
29 /* Copyright (c) 1990 Mentat Inc. */
30 
31 /*
32  * Procedures for the kernel part of DVMRP,
33  * a Distance-Vector Multicast Routing Protocol.
34  * (See RFC-1075)
35  * Written by David Waitzman, BBN Labs, August 1988.
36  * Modified by Steve Deering, Stanford, February 1989.
37  * Modified by Mark J. Steiglitz, Stanford, May, 1991
38  * Modified by Van Jacobson, LBL, January 1993
39  * Modified by Ajit Thyagarajan, PARC, August 1993
40  * Modified by Bill Fenner, PARC, April 1995
41  *
42  * MROUTING 3.5
43  */
44 
45 /*
46  * TODO
47  * - function pointer field in vif, void *vif_sendit()
48  */
49 
50 #include <sys/types.h>
51 #include <sys/stream.h>
52 #include <sys/stropts.h>
53 #include <sys/strlog.h>
54 #include <sys/systm.h>
55 #include <sys/ddi.h>
56 #include <sys/cmn_err.h>
57 #include <sys/zone.h>
58 
59 #include <sys/param.h>
60 #include <sys/socket.h>
61 #include <sys/vtrace.h>
62 #include <sys/debug.h>
63 #include <net/if.h>
64 #include <sys/sockio.h>
65 #include <netinet/in.h>
66 #include <net/if_dl.h>
67 
68 #include <inet/common.h>
69 #include <inet/mi.h>
70 #include <inet/nd.h>
71 #include <inet/mib2.h>
72 #include <netinet/ip6.h>
73 #include <inet/ip.h>
74 #include <inet/snmpcom.h>
75 
76 #include <netinet/igmp.h>
77 #include <netinet/igmp_var.h>
78 #include <netinet/udp.h>
79 #include <netinet/ip_mroute.h>
80 #include <inet/ip_multi.h>
81 #include <inet/ip_ire.h>
82 #include <inet/ip_if.h>
83 #include <inet/ipclassifier.h>
84 
85 #include <netinet/pim.h>
86 
87 
88 /*
89  * MT Design:
90  *
91  * There are three main data structures viftable, mfctable and tbftable that
92  * need to be protected against MT races.
93  *
94  * vitable is a fixed length array of vif structs. There is no lock to protect
95  * the whole array, instead each struct is protected by its own indiviual lock.
96  * The value of v_marks in conjuction with the value of v_refcnt determines the
97  * current state of a vif structure. One special state that needs mention
98  * is when the vif is marked VIF_MARK_NOTINUSE but refcnt != 0. This indicates
99  * that vif is being initalized.
100  * Each structure is freed when the refcnt goes down to zero. If a delete comes
101  * in when the the recfnt is > 1, the vif structure is marked VIF_MARK_CONDEMNED
102  * which prevents the struct from further use.  When the refcnt goes to zero
103  * the struct is freed and is marked VIF_MARK_NOTINUSE.
104  * vif struct stores a pointer to the ipif in v_ipif, to prevent ipif/ill
105  * from  going away a refhold is put on the ipif before using it. see
106  * lock_good_vif() and unlock_good_vif().
107  *
108  * VIF_REFHOLD and VIF_REFRELE macros have been provided to manipulate refcnts
109  * of the vif struct.
110  *
111  * tbftable is also a fixed length array of tbf structs and is only accessed
112  * via v_tbf.  It is protected by its own lock tbf_lock.
113  *
114  * Lock Ordering is
115  * v_lock --> tbf_lock
116  * v_lock --> ill_locK
117  *
118  * mfctable is a fixed size hash table of mfc buckets strcuts (struct mfcb).
119  * Each mfc bucket struct (struct mfcb) maintains a refcnt for each walker,
120  * it also maintains a state. These fields are protected by a lock (mfcb_lock).
121  * mfc structs only maintain a state and have no refcnt. mfc_mutex is used to
122  * protect the struct elements.
123  *
124  * mfc structs are dynamically allocated and are singly linked
125  * at the head of the chain. When an mfc structure is to be deleted
126  * it is marked condemned and so is the state in the bucket struct.
127  * When the last walker of the hash bucket exits all the mfc structs
128  * marked condemed are freed.
129  *
130  * Locking Hierarchy:
131  * The bucket lock should be acquired before the mfc struct lock.
132  * MFCB_REFHOLD and MFCB_REFRELE macros are provided for locking
133  * operations on the bucket struct.
134  *
135  * last_encap_lock and numvifs_mutex should be acquired after
136  * acquring vif or mfc locks. These locks protect some global variables.
137  *
138  * The statistics are not currently protected by a lock
139  * causing the stats be be approximate, not exact.
140  */
141 
142 #define	NO_VIF	MAXVIFS 	/* from mrouted, no route for src */
143 
144 /*
145  * Timeouts:
146  * 	Upcall timeouts - BSD uses boolean_t mfc->expire and
147  *	nexpire[MFCTBLSIZE], the number of times expire has been called.
148  *	SunOS 5.x uses mfc->timeout for each mfc.
149  *	Some Unixes are limited in the number of simultaneous timeouts
150  * 	that can be run, SunOS 5.x does not have this restriction.
151  */
152 
153 /*
154  * In BSD, EXPIRE_TIMEOUT is how often expire_upcalls() is called and
155  * UPCALL_EXPIRE is the nmber of timeouts before a particular upcall
156  * expires. Thus the time till expiration is EXPIRE_TIMEOUT * UPCALL_EXPIRE
157  */
158 #define		EXPIRE_TIMEOUT	(hz/4)	/* 4x / second	*/
159 #define		UPCALL_EXPIRE	6	/* number of timeouts	*/
160 
161 /*
162  * Hash function for a source, group entry
163  */
164 #define	MFCHASH(a, g) MFCHASHMOD(((a) >> 20) ^ ((a) >> 10) ^ (a) ^ \
165 	((g) >> 20) ^ ((g) >> 10) ^ (g))
166 
167 #define			TBF_REPROCESS	(hz / 100)	/* 100x /second	*/
168 
169 /* Identify PIM packet that came on a Register interface */
170 #define	PIM_REGISTER_MARKER	0xffffffff
171 
172 /* Function declarations */
173 static int	add_mfc(struct mfcctl *, ip_stack_t *);
174 static int	add_vif(struct vifctl *, conn_t *, mblk_t *, ip_stack_t *);
175 static int	del_mfc(struct mfcctl *, ip_stack_t *);
176 static int	del_vif(vifi_t *, conn_t *, mblk_t *, ip_stack_t *);
177 static void	del_vifp(struct vif *);
178 static void	encap_send(ipha_t *, mblk_t *, struct vif *, ipaddr_t);
179 static void	expire_upcalls(void *);
180 static void	fill_route(struct mfc *, struct mfcctl *, ip_stack_t *);
181 static void	free_queue(struct mfc *);
182 static int	get_assert(uchar_t *, ip_stack_t *);
183 static int	get_lsg_cnt(struct sioc_lsg_req *, ip_stack_t *);
184 static int	get_sg_cnt(struct sioc_sg_req *, ip_stack_t *);
185 static int	get_version(uchar_t *);
186 static int	get_vif_cnt(struct sioc_vif_req *, ip_stack_t *);
187 static int	ip_mdq(mblk_t *, ipha_t *, ill_t *,
188 		    ipaddr_t, struct mfc *);
189 static int	ip_mrouter_init(conn_t *, uchar_t *, int, ip_stack_t *);
190 static void	phyint_send(ipha_t *, mblk_t *, struct vif *, ipaddr_t);
191 static int	register_mforward(queue_t *, mblk_t *, ill_t *);
192 static void	register_send(ipha_t *, mblk_t *, struct vif *, ipaddr_t);
193 static int	set_assert(int *, ip_stack_t *);
194 
195 /*
196  * Token Bucket Filter functions
197  */
198 static int  priority(struct vif *, ipha_t *);
199 static void tbf_control(struct vif *, mblk_t *, ipha_t *);
200 static int  tbf_dq_sel(struct vif *, ipha_t *);
201 static void tbf_process_q(struct vif *);
202 static void tbf_queue(struct vif *, mblk_t *);
203 static void tbf_reprocess_q(void *);
204 static void tbf_send_packet(struct vif *, mblk_t *);
205 static void tbf_update_tokens(struct vif *);
206 static void release_mfc(struct mfcb *);
207 
208 static boolean_t is_mrouter_off(ip_stack_t *);
209 /*
210  * Encapsulation packets
211  */
212 
213 #define	ENCAP_TTL	64
214 
215 /* prototype IP hdr for encapsulated packets */
216 static ipha_t multicast_encap_iphdr = {
217 	IP_SIMPLE_HDR_VERSION,
218 	0,				/* tos */
219 	sizeof (ipha_t),		/* total length */
220 	0,				/* id */
221 	0,				/* frag offset */
222 	ENCAP_TTL, IPPROTO_ENCAP,
223 	0,				/* checksum */
224 };
225 
226 /*
227  * Rate limit for assert notification messages, in nsec.
228  */
229 #define	ASSERT_MSG_TIME		3000000000
230 
231 
232 #define	VIF_REFHOLD(vifp) {			\
233 	mutex_enter(&(vifp)->v_lock);		\
234 	(vifp)->v_refcnt++;			\
235 	mutex_exit(&(vifp)->v_lock);		\
236 }
237 
238 #define	VIF_REFRELE_LOCKED(vifp) {				\
239 	(vifp)->v_refcnt--;					\
240 	if ((vifp)->v_refcnt == 0 &&				\
241 		((vifp)->v_marks & VIF_MARK_CONDEMNED)) {	\
242 			del_vifp(vifp);				\
243 	} else {						\
244 		mutex_exit(&(vifp)->v_lock);			\
245 	}							\
246 }
247 
248 #define	VIF_REFRELE(vifp) {					\
249 	mutex_enter(&(vifp)->v_lock);				\
250 	(vifp)->v_refcnt--;					\
251 	if ((vifp)->v_refcnt == 0 &&				\
252 		((vifp)->v_marks & VIF_MARK_CONDEMNED)) {	\
253 			del_vifp(vifp);				\
254 	} else {						\
255 		mutex_exit(&(vifp)->v_lock);			\
256 	}							\
257 }
258 
259 #define	MFCB_REFHOLD(mfcb) {				\
260 	mutex_enter(&(mfcb)->mfcb_lock);		\
261 	(mfcb)->mfcb_refcnt++;				\
262 	ASSERT((mfcb)->mfcb_refcnt != 0);		\
263 	mutex_exit(&(mfcb)->mfcb_lock);			\
264 }
265 
266 #define	MFCB_REFRELE(mfcb) {					\
267 	mutex_enter(&(mfcb)->mfcb_lock);			\
268 	ASSERT((mfcb)->mfcb_refcnt != 0);			\
269 	if (--(mfcb)->mfcb_refcnt == 0 &&			\
270 		((mfcb)->mfcb_marks & MFCB_MARK_CONDEMNED)) {	\
271 			release_mfc(mfcb);			\
272 	}							\
273 	mutex_exit(&(mfcb)->mfcb_lock);				\
274 }
275 
276 /*
277  * MFCFIND:
278  * Find a route for a given origin IP address and multicast group address.
279  * Skip entries with pending upcalls.
280  * Type of service parameter to be added in the future!
281  */
282 #define	MFCFIND(mfcbp, o, g, rt) { \
283 	struct mfc *_mb_rt = NULL; \
284 	rt = NULL; \
285 	_mb_rt = mfcbp->mfcb_mfc; \
286 	while (_mb_rt) { \
287 		if ((_mb_rt->mfc_origin.s_addr == o) && \
288 		    (_mb_rt->mfc_mcastgrp.s_addr == g) && \
289 		    (_mb_rt->mfc_rte == NULL) && \
290 		    (!(_mb_rt->mfc_marks & MFCB_MARK_CONDEMNED))) {        \
291 		    rt = _mb_rt; \
292 		    break; \
293 		} \
294 	_mb_rt = _mb_rt->mfc_next; \
295 	} \
296 }
297 
298 /*
299  * BSD uses timeval with sec and usec. In SunOS 5.x uniqtime() and gethrtime()
300  * are inefficient. We use gethrestime() which returns a timespec_t with
301  * sec and nsec, the resolution is machine dependent.
302  * The following 2 macros have been changed to use nsec instead of usec.
303  */
304 /*
305  * Macros to compute elapsed time efficiently.
306  * Borrowed from Van Jacobson's scheduling code.
307  * Delta should be a hrtime_t.
308  */
309 #define	TV_DELTA(a, b, delta) { \
310 	int xxs; \
311  \
312 	delta = (a).tv_nsec - (b).tv_nsec; \
313 	if ((xxs = (a).tv_sec - (b).tv_sec) != 0) { \
314 		switch (xxs) { \
315 		case 2: \
316 		    delta += 1000000000; \
317 		    /*FALLTHROUGH*/ \
318 		case 1: \
319 		    delta += 1000000000; \
320 		    break; \
321 		default: \
322 		    delta += (1000000000 * xxs); \
323 		} \
324 	} \
325 }
326 
327 #define	TV_LT(a, b) (((a).tv_nsec < (b).tv_nsec && \
328 	(a).tv_sec <= (b).tv_sec) || (a).tv_sec < (b).tv_sec)
329 
330 /*
331  * Handle MRT setsockopt commands to modify the multicast routing tables.
332  */
333 int
334 ip_mrouter_set(int cmd, queue_t *q, int checkonly, uchar_t *data,
335     int datalen, mblk_t *first_mp)
336 {
337 	conn_t		*connp = Q_TO_CONN(q);
338 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
339 
340 	mutex_enter(&ipst->ips_ip_g_mrouter_mutex);
341 	if (cmd != MRT_INIT && connp != ipst->ips_ip_g_mrouter) {
342 		mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
343 		return (EACCES);
344 	}
345 	mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
346 
347 	if (checkonly) {
348 		/*
349 		 * do not do operation, just pretend to - new T_CHECK
350 		 * Note: Even routines further on can probably fail but
351 		 * this T_CHECK stuff is only to please XTI so it not
352 		 * necessary to be perfect.
353 		 */
354 		switch (cmd) {
355 		case MRT_INIT:
356 		case MRT_DONE:
357 		case MRT_ADD_VIF:
358 		case MRT_DEL_VIF:
359 		case MRT_ADD_MFC:
360 		case MRT_DEL_MFC:
361 		case MRT_ASSERT:
362 			return (0);
363 		default:
364 			return (EOPNOTSUPP);
365 		}
366 	}
367 
368 	/*
369 	 * make sure no command is issued after multicast routing has been
370 	 * turned off.
371 	 */
372 	if (cmd != MRT_INIT && cmd != MRT_DONE) {
373 		if (is_mrouter_off(ipst))
374 			return (EINVAL);
375 	}
376 
377 	switch (cmd) {
378 	case MRT_INIT:	return (ip_mrouter_init(connp, data, datalen, ipst));
379 	case MRT_DONE:	return (ip_mrouter_done(first_mp, ipst));
380 	case MRT_ADD_VIF:  return (add_vif((struct vifctl *)data, connp,
381 			    first_mp, ipst));
382 	case MRT_DEL_VIF:  return (del_vif((vifi_t *)data, connp, first_mp,
383 			    ipst));
384 	case MRT_ADD_MFC:  return (add_mfc((struct mfcctl *)data, ipst));
385 	case MRT_DEL_MFC:  return (del_mfc((struct mfcctl *)data, ipst));
386 	case MRT_ASSERT:   return (set_assert((int *)data, ipst));
387 	default:	   return (EOPNOTSUPP);
388 	}
389 }
390 
391 /*
392  * Handle MRT getsockopt commands
393  */
394 int
395 ip_mrouter_get(int cmd, queue_t *q, uchar_t *data)
396 {
397 	conn_t		*connp = Q_TO_CONN(q);
398 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
399 
400 	if (connp != ipst->ips_ip_g_mrouter)
401 		return (EACCES);
402 
403 	switch (cmd) {
404 	case MRT_VERSION:	return (get_version((uchar_t *)data));
405 	case MRT_ASSERT:	return (get_assert((uchar_t *)data, ipst));
406 	default:		return (EOPNOTSUPP);
407 	}
408 }
409 
410 /*
411  * Handle ioctl commands to obtain information from the cache.
412  * Called with shared access to IP. These are read_only ioctls.
413  */
414 /* ARGSUSED */
415 int
416 mrt_ioctl(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
417     ip_ioctl_cmd_t *ipip, void *if_req)
418 {
419 	mblk_t	*mp1;
420 	struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
421 	conn_t		*connp = Q_TO_CONN(q);
422 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
423 
424 	/* Existence verified in ip_wput_nondata */
425 	mp1 = mp->b_cont->b_cont;
426 
427 	switch (iocp->ioc_cmd) {
428 	case (SIOCGETVIFCNT):
429 		return (get_vif_cnt((struct sioc_vif_req *)mp1->b_rptr, ipst));
430 	case (SIOCGETSGCNT):
431 		return (get_sg_cnt((struct sioc_sg_req *)mp1->b_rptr, ipst));
432 	case (SIOCGETLSGCNT):
433 		return (get_lsg_cnt((struct sioc_lsg_req *)mp1->b_rptr, ipst));
434 	default:
435 		return (EINVAL);
436 	}
437 }
438 
439 /*
440  * Returns the packet, byte, rpf-failure count for the source, group provided.
441  */
442 static int
443 get_sg_cnt(struct sioc_sg_req *req, ip_stack_t *ipst)
444 {
445 	struct mfc *rt;
446 	struct mfcb *mfcbp;
447 
448 	mfcbp = &ipst->ips_mfcs[MFCHASH(req->src.s_addr, req->grp.s_addr)];
449 	MFCB_REFHOLD(mfcbp);
450 	MFCFIND(mfcbp, req->src.s_addr, req->grp.s_addr, rt);
451 
452 	if (rt != NULL) {
453 		mutex_enter(&rt->mfc_mutex);
454 		req->pktcnt   = rt->mfc_pkt_cnt;
455 		req->bytecnt  = rt->mfc_byte_cnt;
456 		req->wrong_if = rt->mfc_wrong_if;
457 		mutex_exit(&rt->mfc_mutex);
458 	} else
459 		req->pktcnt = req->bytecnt = req->wrong_if = 0xffffffffU;
460 
461 	MFCB_REFRELE(mfcbp);
462 	return (0);
463 }
464 
465 /*
466  * Returns the packet, byte, rpf-failure count for the source, group provided.
467  * Uses larger counters and IPv6 addresses.
468  */
469 /* ARGSUSED XXX until implemented */
470 static int
471 get_lsg_cnt(struct sioc_lsg_req *req, ip_stack_t *ipst)
472 {
473 	/* XXX TODO SIOCGETLSGCNT */
474 	return (ENXIO);
475 }
476 
477 /*
478  * Returns the input and output packet and byte counts on the vif provided.
479  */
480 static int
481 get_vif_cnt(struct sioc_vif_req *req, ip_stack_t *ipst)
482 {
483 	vifi_t vifi = req->vifi;
484 
485 	if (vifi >= ipst->ips_numvifs)
486 		return (EINVAL);
487 
488 	/*
489 	 * No locks here, an approximation is fine.
490 	 */
491 	req->icount = ipst->ips_vifs[vifi].v_pkt_in;
492 	req->ocount = ipst->ips_vifs[vifi].v_pkt_out;
493 	req->ibytes = ipst->ips_vifs[vifi].v_bytes_in;
494 	req->obytes = ipst->ips_vifs[vifi].v_bytes_out;
495 
496 	return (0);
497 }
498 
499 static int
500 get_version(uchar_t *data)
501 {
502 	int *v = (int *)data;
503 
504 	*v = 0x0305;	/* XXX !!!! */
505 
506 	return (0);
507 }
508 
509 /*
510  * Set PIM assert processing global.
511  */
512 static int
513 set_assert(int *i, ip_stack_t *ipst)
514 {
515 	if ((*i != 1) && (*i != 0))
516 		return (EINVAL);
517 
518 	ipst->ips_pim_assert = *i;
519 
520 	return (0);
521 }
522 
523 /*
524  * Get PIM assert processing global.
525  */
526 static int
527 get_assert(uchar_t *data, ip_stack_t *ipst)
528 {
529 	int *i = (int *)data;
530 
531 	*i = ipst->ips_pim_assert;
532 
533 	return (0);
534 }
535 
536 /*
537  * Enable multicast routing.
538  */
539 static int
540 ip_mrouter_init(conn_t *connp, uchar_t *data, int datalen, ip_stack_t *ipst)
541 {
542 	int	*v;
543 
544 	if (data == NULL || (datalen != sizeof (int)))
545 		return (ENOPROTOOPT);
546 
547 	v = (int *)data;
548 	if (*v != 1)
549 		return (ENOPROTOOPT);
550 
551 	mutex_enter(&ipst->ips_ip_g_mrouter_mutex);
552 	if (ipst->ips_ip_g_mrouter != NULL) {
553 		mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
554 		return (EADDRINUSE);
555 	}
556 
557 	/*
558 	 * MRT_INIT should only be allowed for RAW sockets, but we double
559 	 * check.
560 	 */
561 	if (!IPCL_IS_RAWIP(connp)) {
562 		mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
563 		return (EINVAL);
564 	}
565 
566 	ipst->ips_ip_g_mrouter = connp;
567 	connp->conn_multi_router = 1;
568 	/* In order for tunnels to work we have to turn ip_g_forward on */
569 	if (!WE_ARE_FORWARDING(ipst)) {
570 		if (ipst->ips_ip_mrtdebug > 1) {
571 			(void) mi_strlog(connp->conn_rq, 1, SL_TRACE,
572 			    "ip_mrouter_init: turning on forwarding");
573 		}
574 		ipst->ips_saved_ip_g_forward = ipst->ips_ip_g_forward;
575 		ipst->ips_ip_g_forward = IP_FORWARD_ALWAYS;
576 	}
577 
578 	mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
579 	return (0);
580 }
581 
582 void
583 ip_mrouter_stack_init(ip_stack_t *ipst)
584 {
585 	mutex_init(&ipst->ips_ip_g_mrouter_mutex, NULL, MUTEX_DEFAULT, NULL);
586 
587 	ipst->ips_vifs = kmem_zalloc(sizeof (struct vif) * (MAXVIFS+1),
588 	    KM_SLEEP);
589 	ipst->ips_mrtstat = kmem_zalloc(sizeof (struct mrtstat), KM_SLEEP);
590 	/*
591 	 * mfctable:
592 	 * Includes all mfcs, including waiting upcalls.
593 	 * Multiple mfcs per bucket.
594 	 */
595 	ipst->ips_mfcs = kmem_zalloc(sizeof (struct mfcb) * MFCTBLSIZ,
596 	    KM_SLEEP);
597 	/*
598 	 * Define the token bucket filter structures.
599 	 * tbftable -> each vif has one of these for storing info.
600 	 */
601 	ipst->ips_tbfs = kmem_zalloc(sizeof (struct tbf) * MAXVIFS, KM_SLEEP);
602 
603 	mutex_init(&ipst->ips_last_encap_lock, NULL, MUTEX_DEFAULT, NULL);
604 
605 	ipst->ips_mrtstat->mrts_vifctlSize = sizeof (struct vifctl);
606 	ipst->ips_mrtstat->mrts_mfcctlSize = sizeof (struct mfcctl);
607 }
608 
609 /*
610  * Disable multicast routing.
611  * Didn't use global timeout_val (BSD version), instead check the mfctable.
612  */
613 int
614 ip_mrouter_done(mblk_t *mp, ip_stack_t *ipst)
615 {
616 	conn_t		*mrouter;
617 	vifi_t 		vifi;
618 	struct mfc	*mfc_rt;
619 	int		i;
620 
621 	mutex_enter(&ipst->ips_ip_g_mrouter_mutex);
622 	if (ipst->ips_ip_g_mrouter == NULL) {
623 		mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
624 		return (EINVAL);
625 	}
626 
627 	mrouter = ipst->ips_ip_g_mrouter;
628 
629 	if (ipst->ips_saved_ip_g_forward != -1) {
630 		if (ipst->ips_ip_mrtdebug > 1) {
631 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
632 			    "ip_mrouter_done: turning off forwarding");
633 		}
634 		ipst->ips_ip_g_forward = ipst->ips_saved_ip_g_forward;
635 		ipst->ips_saved_ip_g_forward = -1;
636 	}
637 
638 	/*
639 	 * Always clear cache when vifs change.
640 	 * No need to get ipst->ips_last_encap_lock since we are running as
641 	 * a writer.
642 	 */
643 	mutex_enter(&ipst->ips_last_encap_lock);
644 	ipst->ips_last_encap_src = 0;
645 	ipst->ips_last_encap_vif = NULL;
646 	mutex_exit(&ipst->ips_last_encap_lock);
647 	mrouter->conn_multi_router = 0;
648 
649 	mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
650 
651 	/*
652 	 * For each phyint in use,
653 	 * disable promiscuous reception of all IP multicasts.
654 	 */
655 	for (vifi = 0; vifi < MAXVIFS; vifi++) {
656 		struct vif *vifp = ipst->ips_vifs + vifi;
657 
658 		mutex_enter(&vifp->v_lock);
659 		/*
660 		 * if the vif is active mark it condemned.
661 		 */
662 		if (vifp->v_marks & VIF_MARK_GOOD) {
663 			ASSERT(vifp->v_ipif != NULL);
664 			ipif_refhold(vifp->v_ipif);
665 			/* Phyint only */
666 			if (!(vifp->v_flags & (VIFF_TUNNEL | VIFF_REGISTER))) {
667 				ipif_t *ipif = vifp->v_ipif;
668 				ipsq_t  *ipsq;
669 				boolean_t suc;
670 				ill_t *ill;
671 
672 				ill = ipif->ipif_ill;
673 				suc = B_FALSE;
674 				if (mp == NULL) {
675 					/*
676 					 * being called from ip_close,
677 					 * lets do it synchronously.
678 					 * Clear VIF_MARK_GOOD and
679 					 * set VIF_MARK_CONDEMNED.
680 					 */
681 					vifp->v_marks &= ~VIF_MARK_GOOD;
682 					vifp->v_marks |= VIF_MARK_CONDEMNED;
683 					mutex_exit(&(vifp)->v_lock);
684 					suc = ipsq_enter(ill, B_FALSE, NEW_OP);
685 					ipsq = ill->ill_phyint->phyint_ipsq;
686 				} else {
687 					ipsq = ipsq_try_enter(ipif, NULL,
688 					    mrouter->conn_wq, mp,
689 					    ip_restart_optmgmt, NEW_OP, B_TRUE);
690 					if (ipsq == NULL) {
691 						mutex_exit(&(vifp)->v_lock);
692 						ipif_refrele(ipif);
693 						return (EINPROGRESS);
694 					}
695 					/*
696 					 * Clear VIF_MARK_GOOD and
697 					 * set VIF_MARK_CONDEMNED.
698 					 */
699 					vifp->v_marks &= ~VIF_MARK_GOOD;
700 					vifp->v_marks |= VIF_MARK_CONDEMNED;
701 					mutex_exit(&(vifp)->v_lock);
702 					suc = B_TRUE;
703 				}
704 
705 				if (suc) {
706 					(void) ip_delmulti(INADDR_ANY, ipif,
707 					    B_TRUE, B_TRUE);
708 					ipsq_exit(ipsq);
709 				}
710 				mutex_enter(&vifp->v_lock);
711 			}
712 			ipif_refrele(vifp->v_ipif);
713 			/*
714 			 * decreases the refcnt added in add_vif.
715 			 * and release v_lock.
716 			 */
717 			VIF_REFRELE_LOCKED(vifp);
718 		} else {
719 			mutex_exit(&vifp->v_lock);
720 			continue;
721 		}
722 	}
723 
724 	mutex_enter(&ipst->ips_numvifs_mutex);
725 	ipst->ips_numvifs = 0;
726 	ipst->ips_pim_assert = 0;
727 	ipst->ips_reg_vif_num = ALL_VIFS;
728 	mutex_exit(&ipst->ips_numvifs_mutex);
729 
730 	/*
731 	 * Free upcall msgs.
732 	 * Go through mfctable and stop any outstanding upcall
733 	 * timeouts remaining on mfcs.
734 	 */
735 	for (i = 0; i < MFCTBLSIZ; i++) {
736 		mutex_enter(&ipst->ips_mfcs[i].mfcb_lock);
737 		ipst->ips_mfcs[i].mfcb_refcnt++;
738 		ipst->ips_mfcs[i].mfcb_marks |= MFCB_MARK_CONDEMNED;
739 		mutex_exit(&ipst->ips_mfcs[i].mfcb_lock);
740 		mfc_rt = ipst->ips_mfcs[i].mfcb_mfc;
741 		while (mfc_rt) {
742 			/* Free upcalls */
743 			mutex_enter(&mfc_rt->mfc_mutex);
744 			if (mfc_rt->mfc_rte != NULL) {
745 				if (mfc_rt->mfc_timeout_id != 0) {
746 					/*
747 					 * OK to drop the lock as we have
748 					 * a refcnt on the bucket. timeout
749 					 * can fire but it will see that
750 					 * mfc_timeout_id == 0 and not do
751 					 * anything. see expire_upcalls().
752 					 */
753 					mfc_rt->mfc_timeout_id = 0;
754 					mutex_exit(&mfc_rt->mfc_mutex);
755 					(void) untimeout(
756 					    mfc_rt->mfc_timeout_id);
757 						mfc_rt->mfc_timeout_id = 0;
758 					mutex_enter(&mfc_rt->mfc_mutex);
759 
760 					/*
761 					 * all queued upcall packets
762 					 * and mblk will be freed in
763 					 * release_mfc().
764 					 */
765 				}
766 			}
767 
768 			mfc_rt->mfc_marks |= MFCB_MARK_CONDEMNED;
769 
770 			mutex_exit(&mfc_rt->mfc_mutex);
771 			mfc_rt = mfc_rt->mfc_next;
772 		}
773 		MFCB_REFRELE(&ipst->ips_mfcs[i]);
774 	}
775 
776 	mutex_enter(&ipst->ips_ip_g_mrouter_mutex);
777 	ipst->ips_ip_g_mrouter = NULL;
778 	mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
779 	return (0);
780 }
781 
782 void
783 ip_mrouter_stack_destroy(ip_stack_t *ipst)
784 {
785 	struct mfcb *mfcbp;
786 	struct mfc  *rt;
787 	int i;
788 
789 	for (i = 0; i < MFCTBLSIZ; i++) {
790 		mfcbp = &ipst->ips_mfcs[i];
791 
792 		while ((rt = mfcbp->mfcb_mfc) != NULL) {
793 			(void) printf("ip_mrouter_stack_destroy: free for %d\n",
794 			    i);
795 
796 			mfcbp->mfcb_mfc = rt->mfc_next;
797 			free_queue(rt);
798 			mi_free(rt);
799 		}
800 	}
801 	kmem_free(ipst->ips_vifs, sizeof (struct vif) * (MAXVIFS+1));
802 	ipst->ips_vifs = NULL;
803 	kmem_free(ipst->ips_mrtstat, sizeof (struct mrtstat));
804 	ipst->ips_mrtstat = NULL;
805 	kmem_free(ipst->ips_mfcs, sizeof (struct mfcb) * MFCTBLSIZ);
806 	ipst->ips_mfcs = NULL;
807 	kmem_free(ipst->ips_tbfs, sizeof (struct tbf) * MAXVIFS);
808 	ipst->ips_tbfs = NULL;
809 
810 	mutex_destroy(&ipst->ips_last_encap_lock);
811 	mutex_destroy(&ipst->ips_ip_g_mrouter_mutex);
812 }
813 
814 static boolean_t
815 is_mrouter_off(ip_stack_t *ipst)
816 {
817 	conn_t	*mrouter;
818 
819 	mutex_enter(&ipst->ips_ip_g_mrouter_mutex);
820 	if (ipst->ips_ip_g_mrouter == NULL) {
821 		mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
822 		return (B_TRUE);
823 	}
824 
825 	mrouter = ipst->ips_ip_g_mrouter;
826 	if (mrouter->conn_multi_router == 0) {
827 		mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
828 		return (B_TRUE);
829 	}
830 	mutex_exit(&ipst->ips_ip_g_mrouter_mutex);
831 	return (B_FALSE);
832 }
833 
834 static void
835 unlock_good_vif(struct vif *vifp)
836 {
837 	ASSERT(vifp->v_ipif != NULL);
838 	ipif_refrele(vifp->v_ipif);
839 	VIF_REFRELE(vifp);
840 }
841 
842 static boolean_t
843 lock_good_vif(struct vif *vifp)
844 {
845 	mutex_enter(&vifp->v_lock);
846 	if (!(vifp->v_marks & VIF_MARK_GOOD)) {
847 		mutex_exit(&vifp->v_lock);
848 		return (B_FALSE);
849 	}
850 
851 	ASSERT(vifp->v_ipif != NULL);
852 	mutex_enter(&vifp->v_ipif->ipif_ill->ill_lock);
853 	if (!IPIF_CAN_LOOKUP(vifp->v_ipif)) {
854 		mutex_exit(&vifp->v_ipif->ipif_ill->ill_lock);
855 		mutex_exit(&vifp->v_lock);
856 		return (B_FALSE);
857 	}
858 	ipif_refhold_locked(vifp->v_ipif);
859 	mutex_exit(&vifp->v_ipif->ipif_ill->ill_lock);
860 	vifp->v_refcnt++;
861 	mutex_exit(&vifp->v_lock);
862 	return (B_TRUE);
863 }
864 
865 /*
866  * Add a vif to the vif table.
867  */
868 static int
869 add_vif(struct vifctl *vifcp, conn_t *connp, mblk_t *first_mp, ip_stack_t *ipst)
870 {
871 	struct vif	*vifp = ipst->ips_vifs + vifcp->vifc_vifi;
872 	ipif_t		*ipif;
873 	int		error;
874 	struct tbf	*v_tbf = ipst->ips_tbfs + vifcp->vifc_vifi;
875 	ipsq_t  	*ipsq;
876 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
877 
878 	ASSERT(connp != NULL);
879 
880 	if (vifcp->vifc_vifi >= MAXVIFS)
881 		return (EINVAL);
882 
883 	if (is_mrouter_off(ipst))
884 		return (EINVAL);
885 
886 	mutex_enter(&vifp->v_lock);
887 	/*
888 	 * Viftable entry should be 0.
889 	 * if v_marks == 0 but v_refcnt != 0 means struct is being
890 	 * initialized.
891 	 *
892 	 * Also note that it is very unlikely that we will get a MRT_ADD_VIF
893 	 * request while the delete is in progress, mrouted only sends add
894 	 * requests when a new interface is added and the new interface cannot
895 	 * have the same vifi as an existing interface. We make sure that
896 	 * ill_delete will block till the vif is deleted by adding a refcnt
897 	 * to ipif in del_vif().
898 	 */
899 	if (vifp->v_lcl_addr.s_addr != 0 ||
900 	    vifp->v_marks != 0 ||
901 	    vifp->v_refcnt != 0) {
902 		mutex_exit(&vifp->v_lock);
903 		return (EADDRINUSE);
904 	}
905 
906 	/* Incoming vif should not be 0 */
907 	if (vifcp->vifc_lcl_addr.s_addr == 0) {
908 		mutex_exit(&vifp->v_lock);
909 		return (EINVAL);
910 	}
911 
912 	vifp->v_refcnt++;
913 	mutex_exit(&vifp->v_lock);
914 	/* Find the interface with the local address */
915 	ipif = ipif_lookup_addr((ipaddr_t)vifcp->vifc_lcl_addr.s_addr, NULL,
916 	    connp->conn_zoneid, CONNP_TO_WQ(connp), first_mp,
917 	    ip_restart_optmgmt, &error, ipst);
918 	if (ipif == NULL) {
919 		VIF_REFRELE(vifp);
920 		if (error == EINPROGRESS)
921 			return (error);
922 		return (EADDRNOTAVAIL);
923 	}
924 
925 	/*
926 	 * We have to be exclusive as we have to call ip_addmulti()
927 	 * This is the best position to try to be exclusive in case
928 	 * we have to wait.
929 	 */
930 	ipsq = ipsq_try_enter(ipif, NULL, CONNP_TO_WQ(connp), first_mp,
931 	    ip_restart_optmgmt, NEW_OP, B_TRUE);
932 	if ((ipsq) == NULL) {
933 		VIF_REFRELE(vifp);
934 		ipif_refrele(ipif);
935 		return (EINPROGRESS);
936 	}
937 
938 	if (ipst->ips_ip_mrtdebug > 1) {
939 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
940 		    "add_vif: src 0x%x enter",
941 		    vifcp->vifc_lcl_addr.s_addr);
942 	}
943 
944 	mutex_enter(&vifp->v_lock);
945 	/*
946 	 * Always clear cache when vifs change.
947 	 * Needed to ensure that src isn't left over from before vif was added.
948 	 * No need to get last_encap_lock, since we are running as a writer.
949 	 */
950 
951 	mutex_enter(&ipst->ips_last_encap_lock);
952 	ipst->ips_last_encap_src = 0;
953 	ipst->ips_last_encap_vif = NULL;
954 	mutex_exit(&ipst->ips_last_encap_lock);
955 
956 	if (vifcp->vifc_flags & VIFF_TUNNEL) {
957 		if ((vifcp->vifc_flags & VIFF_SRCRT) != 0) {
958 			cmn_err(CE_WARN,
959 			    "add_vif: source route tunnels not supported\n");
960 			VIF_REFRELE_LOCKED(vifp);
961 			ipif_refrele(ipif);
962 			ipsq_exit(ipsq);
963 			return (EOPNOTSUPP);
964 		}
965 		vifp->v_rmt_addr  = vifcp->vifc_rmt_addr;
966 
967 	} else {
968 		/* Phyint or Register vif */
969 		if (vifcp->vifc_flags & VIFF_REGISTER) {
970 			/*
971 			 * Note: Since all IPPROTO_IP level options (including
972 			 * MRT_ADD_VIF) are done exclusively via
973 			 * ip_optmgmt_writer(), a lock is not necessary to
974 			 * protect reg_vif_num.
975 			 */
976 			mutex_enter(&ipst->ips_numvifs_mutex);
977 			if (ipst->ips_reg_vif_num == ALL_VIFS) {
978 				ipst->ips_reg_vif_num = vifcp->vifc_vifi;
979 				mutex_exit(&ipst->ips_numvifs_mutex);
980 			} else {
981 				mutex_exit(&ipst->ips_numvifs_mutex);
982 				VIF_REFRELE_LOCKED(vifp);
983 				ipif_refrele(ipif);
984 				ipsq_exit(ipsq);
985 				return (EADDRINUSE);
986 			}
987 		}
988 
989 		/* Make sure the interface supports multicast */
990 		if ((ipif->ipif_ill->ill_flags & ILLF_MULTICAST) == 0) {
991 			VIF_REFRELE_LOCKED(vifp);
992 			ipif_refrele(ipif);
993 			if (vifcp->vifc_flags & VIFF_REGISTER) {
994 				mutex_enter(&ipst->ips_numvifs_mutex);
995 				ipst->ips_reg_vif_num = ALL_VIFS;
996 				mutex_exit(&ipst->ips_numvifs_mutex);
997 			}
998 			ipsq_exit(ipsq);
999 			return (EOPNOTSUPP);
1000 		}
1001 		/* Enable promiscuous reception of all IP mcasts from the if */
1002 		mutex_exit(&vifp->v_lock);
1003 		error = ip_addmulti(INADDR_ANY, ipif, ILGSTAT_NONE,
1004 		    MODE_IS_EXCLUDE, NULL);
1005 		mutex_enter(&vifp->v_lock);
1006 		/*
1007 		 * since we released the lock lets make sure that
1008 		 * ip_mrouter_done() has not been called.
1009 		 */
1010 		if (error != 0 || is_mrouter_off(ipst)) {
1011 			if (error == 0)
1012 				(void) ip_delmulti(INADDR_ANY, ipif, B_TRUE,
1013 				    B_TRUE);
1014 			if (vifcp->vifc_flags & VIFF_REGISTER) {
1015 				mutex_enter(&ipst->ips_numvifs_mutex);
1016 				ipst->ips_reg_vif_num = ALL_VIFS;
1017 				mutex_exit(&ipst->ips_numvifs_mutex);
1018 			}
1019 			VIF_REFRELE_LOCKED(vifp);
1020 			ipif_refrele(ipif);
1021 			ipsq_exit(ipsq);
1022 			return (error?error:EINVAL);
1023 		}
1024 	}
1025 	/* Define parameters for the tbf structure */
1026 	vifp->v_tbf = v_tbf;
1027 	gethrestime(&vifp->v_tbf->tbf_last_pkt_t);
1028 	vifp->v_tbf->tbf_n_tok = 0;
1029 	vifp->v_tbf->tbf_q_len = 0;
1030 	vifp->v_tbf->tbf_max_q_len = MAXQSIZE;
1031 	vifp->v_tbf->tbf_q = vifp->v_tbf->tbf_t = NULL;
1032 
1033 	vifp->v_flags = vifcp->vifc_flags;
1034 	vifp->v_threshold = vifcp->vifc_threshold;
1035 	vifp->v_lcl_addr = vifcp->vifc_lcl_addr;
1036 	vifp->v_ipif = ipif;
1037 	ipif_refrele(ipif);
1038 	/* Scaling up here, allows division by 1024 in critical code.	*/
1039 	vifp->v_rate_limit = vifcp->vifc_rate_limit * (1024/1000);
1040 	vifp->v_timeout_id = 0;
1041 	/* initialize per vif pkt counters */
1042 	vifp->v_pkt_in = 0;
1043 	vifp->v_pkt_out = 0;
1044 	vifp->v_bytes_in = 0;
1045 	vifp->v_bytes_out = 0;
1046 	mutex_init(&vifp->v_tbf->tbf_lock, NULL, MUTEX_DEFAULT, NULL);
1047 
1048 	/* Adjust numvifs up, if the vifi is higher than numvifs */
1049 	mutex_enter(&ipst->ips_numvifs_mutex);
1050 	if (ipst->ips_numvifs <= vifcp->vifc_vifi)
1051 		ipst->ips_numvifs = vifcp->vifc_vifi + 1;
1052 	mutex_exit(&ipst->ips_numvifs_mutex);
1053 
1054 	if (ipst->ips_ip_mrtdebug > 1) {
1055 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1056 		    "add_vif: #%d, lcladdr %x, %s %x, thresh %x, rate %d",
1057 		    vifcp->vifc_vifi,
1058 		    ntohl(vifcp->vifc_lcl_addr.s_addr),
1059 		    (vifcp->vifc_flags & VIFF_TUNNEL) ? "rmtaddr" : "mask",
1060 		    ntohl(vifcp->vifc_rmt_addr.s_addr),
1061 		    vifcp->vifc_threshold, vifcp->vifc_rate_limit);
1062 	}
1063 
1064 	vifp->v_marks = VIF_MARK_GOOD;
1065 	mutex_exit(&vifp->v_lock);
1066 	ipsq_exit(ipsq);
1067 	return (0);
1068 }
1069 
1070 
1071 /* Delete a vif from the vif table. */
1072 static void
1073 del_vifp(struct vif *vifp)
1074 {
1075 	struct tbf	*t = vifp->v_tbf;
1076 	mblk_t  *mp0;
1077 	vifi_t  vifi;
1078 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
1079 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
1080 
1081 	ASSERT(vifp->v_marks & VIF_MARK_CONDEMNED);
1082 	ASSERT(t != NULL);
1083 
1084 	if (ipst->ips_ip_mrtdebug > 1) {
1085 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1086 		    "del_vif: src 0x%x\n", vifp->v_lcl_addr.s_addr);
1087 	}
1088 
1089 	if (vifp->v_timeout_id != 0) {
1090 		(void) untimeout(vifp->v_timeout_id);
1091 		vifp->v_timeout_id = 0;
1092 	}
1093 
1094 	/*
1095 	 * Free packets queued at the interface.
1096 	 * Mrouted takes care of cleaning up mfcs - makes calls to del_mfc.
1097 	 */
1098 	mutex_enter(&t->tbf_lock);
1099 	while (t->tbf_q != NULL) {
1100 		mp0 = t->tbf_q;
1101 		t->tbf_q = t->tbf_q->b_next;
1102 		mp0->b_prev = mp0->b_next = NULL;
1103 		freemsg(mp0);
1104 	}
1105 	mutex_exit(&t->tbf_lock);
1106 
1107 	/*
1108 	 * Always clear cache when vifs change.
1109 	 * No need to get last_encap_lock since we are running as a writer.
1110 	 */
1111 	mutex_enter(&ipst->ips_last_encap_lock);
1112 	if (vifp == ipst->ips_last_encap_vif) {
1113 		ipst->ips_last_encap_vif = NULL;
1114 		ipst->ips_last_encap_src = 0;
1115 	}
1116 	mutex_exit(&ipst->ips_last_encap_lock);
1117 
1118 	mutex_destroy(&t->tbf_lock);
1119 
1120 	bzero(vifp->v_tbf, sizeof (*(vifp->v_tbf)));
1121 
1122 	/* Adjust numvifs down */
1123 	mutex_enter(&ipst->ips_numvifs_mutex);
1124 	for (vifi = ipst->ips_numvifs; vifi != 0; vifi--) /* vifi is unsigned */
1125 		if (ipst->ips_vifs[vifi - 1].v_lcl_addr.s_addr != 0)
1126 			break;
1127 	ipst->ips_numvifs = vifi;
1128 	mutex_exit(&ipst->ips_numvifs_mutex);
1129 
1130 	bzero(vifp, sizeof (*vifp));
1131 }
1132 
1133 static int
1134 del_vif(vifi_t *vifip, conn_t *connp, mblk_t *first_mp, ip_stack_t *ipst)
1135 {
1136 	struct vif	*vifp = ipst->ips_vifs + *vifip;
1137 	ipsq_t  	*ipsq;
1138 
1139 	if (*vifip >= ipst->ips_numvifs)
1140 		return (EINVAL);
1141 
1142 	mutex_enter(&vifp->v_lock);
1143 	/*
1144 	 * Not initialized
1145 	 * Here we are not looking at the vif that is being initialized
1146 	 * i.e vifp->v_marks == 0 and refcnt > 0.
1147 	 */
1148 	if (vifp->v_lcl_addr.s_addr == 0 ||
1149 	    !(vifp->v_marks & VIF_MARK_GOOD)) {
1150 		mutex_exit(&vifp->v_lock);
1151 		return (EADDRNOTAVAIL);
1152 	}
1153 
1154 	/*
1155 	 * This is an optimization, if first_mp == NULL
1156 	 * than we are being called from reset_mrt_vif_ipif()
1157 	 * so we already have exclusive access to the ipsq.
1158 	 * the ASSERT below is a check for this condition.
1159 	 */
1160 	if (first_mp != NULL &&
1161 	    !(vifp->v_flags & (VIFF_TUNNEL | VIFF_REGISTER))) {
1162 		ASSERT(connp != NULL);
1163 		/*
1164 		 * We have to be exclusive as we have to call ip_delmulti()
1165 		 * This is the best position to try to be exclusive in case
1166 		 * we have to wait.
1167 		 */
1168 		ipsq = ipsq_try_enter(vifp->v_ipif, NULL, CONNP_TO_WQ(connp),
1169 		    first_mp, ip_restart_optmgmt, NEW_OP, B_TRUE);
1170 		if ((ipsq) == NULL) {
1171 			mutex_exit(&vifp->v_lock);
1172 			return (EINPROGRESS);
1173 		}
1174 		/* recheck after being exclusive */
1175 		if (vifp->v_lcl_addr.s_addr == 0 ||
1176 		    !vifp->v_marks & VIF_MARK_GOOD) {
1177 			/*
1178 			 * someone beat us.
1179 			 */
1180 			mutex_exit(&vifp->v_lock);
1181 			ipsq_exit(ipsq);
1182 			return (EADDRNOTAVAIL);
1183 		}
1184 	}
1185 
1186 
1187 	ASSERT(IAM_WRITER_IPIF(vifp->v_ipif));
1188 
1189 	/* Clear VIF_MARK_GOOD and set VIF_MARK_CONDEMNED. */
1190 	vifp->v_marks &= ~VIF_MARK_GOOD;
1191 	vifp->v_marks |= VIF_MARK_CONDEMNED;
1192 
1193 	/* Phyint only */
1194 	if (!(vifp->v_flags & (VIFF_TUNNEL | VIFF_REGISTER))) {
1195 		ipif_t *ipif = vifp->v_ipif;
1196 		ASSERT(ipif != NULL);
1197 		/*
1198 		 * should be OK to drop the lock as we
1199 		 * have marked this as CONDEMNED.
1200 		 */
1201 		mutex_exit(&(vifp)->v_lock);
1202 		(void) ip_delmulti(INADDR_ANY, ipif, B_TRUE, B_TRUE);
1203 		if (first_mp != NULL)
1204 			ipsq_exit(ipsq);
1205 		mutex_enter(&(vifp)->v_lock);
1206 	}
1207 
1208 	/*
1209 	 * decreases the refcnt added in add_vif.
1210 	 */
1211 	VIF_REFRELE_LOCKED(vifp);
1212 	return (0);
1213 }
1214 
1215 /*
1216  * Add an mfc entry.
1217  */
1218 static int
1219 add_mfc(struct mfcctl *mfccp, ip_stack_t *ipst)
1220 {
1221 	struct mfc *rt;
1222 	struct rtdetq *rte;
1223 	ushort_t nstl;
1224 	int i;
1225 	struct mfcb *mfcbp;
1226 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
1227 
1228 	/*
1229 	 * The value of vifi is NO_VIF (==MAXVIFS) if Mrouted
1230 	 * did not have a real route for pkt.
1231 	 * We want this pkt without rt installed in the mfctable to prevent
1232 	 * multiiple tries, so go ahead and put it in mfctable, it will
1233 	 * be discarded later in ip_mdq() because the child is NULL.
1234 	 */
1235 
1236 	/* Error checking, out of bounds? */
1237 	if (mfccp->mfcc_parent > MAXVIFS) {
1238 		ip0dbg(("ADD_MFC: mfcc_parent out of range %d",
1239 		    (int)mfccp->mfcc_parent));
1240 		return (EINVAL);
1241 	}
1242 
1243 	if ((mfccp->mfcc_parent != NO_VIF) &&
1244 	    (ipst->ips_vifs[mfccp->mfcc_parent].v_ipif == NULL)) {
1245 		ip0dbg(("ADD_MFC: NULL ipif for parent vif %d\n",
1246 		    (int)mfccp->mfcc_parent));
1247 		return (EINVAL);
1248 	}
1249 
1250 	if (is_mrouter_off(ipst)) {
1251 		return (EINVAL);
1252 	}
1253 
1254 	mfcbp = &ipst->ips_mfcs[MFCHASH(mfccp->mfcc_origin.s_addr,
1255 	    mfccp->mfcc_mcastgrp.s_addr)];
1256 	MFCB_REFHOLD(mfcbp);
1257 	MFCFIND(mfcbp, mfccp->mfcc_origin.s_addr,
1258 	    mfccp->mfcc_mcastgrp.s_addr, rt);
1259 
1260 	/* If an entry already exists, just update the fields */
1261 	if (rt) {
1262 		if (ipst->ips_ip_mrtdebug > 1) {
1263 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1264 			    "add_mfc: update o %x grp %x parent %x",
1265 			    ntohl(mfccp->mfcc_origin.s_addr),
1266 			    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1267 			    mfccp->mfcc_parent);
1268 		}
1269 		mutex_enter(&rt->mfc_mutex);
1270 		rt->mfc_parent = mfccp->mfcc_parent;
1271 
1272 		mutex_enter(&ipst->ips_numvifs_mutex);
1273 		for (i = 0; i < (int)ipst->ips_numvifs; i++)
1274 			rt->mfc_ttls[i] = mfccp->mfcc_ttls[i];
1275 		mutex_exit(&ipst->ips_numvifs_mutex);
1276 		mutex_exit(&rt->mfc_mutex);
1277 
1278 		MFCB_REFRELE(mfcbp);
1279 		return (0);
1280 	}
1281 
1282 	/*
1283 	 * Find the entry for which the upcall was made and update.
1284 	 */
1285 	for (rt = mfcbp->mfcb_mfc, nstl = 0; rt; rt = rt->mfc_next) {
1286 		mutex_enter(&rt->mfc_mutex);
1287 		if ((rt->mfc_origin.s_addr == mfccp->mfcc_origin.s_addr) &&
1288 		    (rt->mfc_mcastgrp.s_addr == mfccp->mfcc_mcastgrp.s_addr) &&
1289 		    (rt->mfc_rte != NULL) &&
1290 		    !(rt->mfc_marks & MFCB_MARK_CONDEMNED)) {
1291 			if (nstl++ != 0)
1292 				cmn_err(CE_WARN,
1293 				    "add_mfc: %s o %x g %x p %x",
1294 				    "multiple kernel entries",
1295 				    ntohl(mfccp->mfcc_origin.s_addr),
1296 				    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1297 				    mfccp->mfcc_parent);
1298 
1299 			if (ipst->ips_ip_mrtdebug > 1) {
1300 				(void) mi_strlog(mrouter->conn_rq, 1,
1301 				    SL_TRACE,
1302 				    "add_mfc: o %x g %x p %x",
1303 				    ntohl(mfccp->mfcc_origin.s_addr),
1304 				    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1305 				    mfccp->mfcc_parent);
1306 			}
1307 			fill_route(rt, mfccp, ipst);
1308 
1309 			/*
1310 			 * Prevent cleanup of cache entry.
1311 			 * Timer starts in ip_mforward.
1312 			 */
1313 			if (rt->mfc_timeout_id != 0) {
1314 				timeout_id_t id;
1315 				id = rt->mfc_timeout_id;
1316 				/*
1317 				 * setting id to zero will avoid this
1318 				 * entry from being cleaned up in
1319 				 * expire_up_calls().
1320 				 */
1321 				rt->mfc_timeout_id = 0;
1322 				/*
1323 				 * dropping the lock is fine as we
1324 				 * have a refhold on the bucket.
1325 				 * so mfc cannot be freed.
1326 				 * The timeout can fire but it will see
1327 				 * that mfc_timeout_id == 0 and not cleanup.
1328 				 */
1329 				mutex_exit(&rt->mfc_mutex);
1330 				(void) untimeout(id);
1331 				mutex_enter(&rt->mfc_mutex);
1332 			}
1333 
1334 			/*
1335 			 * Send all pkts that are queued waiting for the upcall.
1336 			 * ip_mdq param tun set to 0 -
1337 			 * the return value of ip_mdq() isn't used here,
1338 			 * so value we send doesn't matter.
1339 			 */
1340 			while (rt->mfc_rte != NULL) {
1341 				rte = rt->mfc_rte;
1342 				rt->mfc_rte = rte->rte_next;
1343 				mutex_exit(&rt->mfc_mutex);
1344 				(void) ip_mdq(rte->mp, (ipha_t *)
1345 				    rte->mp->b_rptr, rte->ill, 0, rt);
1346 				freemsg(rte->mp);
1347 				mi_free((char *)rte);
1348 				mutex_enter(&rt->mfc_mutex);
1349 			}
1350 		}
1351 		mutex_exit(&rt->mfc_mutex);
1352 	}
1353 
1354 
1355 	/*
1356 	 * It is possible that an entry is being inserted without an upcall
1357 	 */
1358 	if (nstl == 0) {
1359 		mutex_enter(&(mfcbp->mfcb_lock));
1360 		if (ipst->ips_ip_mrtdebug > 1) {
1361 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1362 			    "add_mfc: no upcall o %x g %x p %x",
1363 			    ntohl(mfccp->mfcc_origin.s_addr),
1364 			    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1365 			    mfccp->mfcc_parent);
1366 		}
1367 		if (is_mrouter_off(ipst)) {
1368 			mutex_exit(&mfcbp->mfcb_lock);
1369 			MFCB_REFRELE(mfcbp);
1370 			return (EINVAL);
1371 		}
1372 
1373 		for (rt = mfcbp->mfcb_mfc; rt; rt = rt->mfc_next) {
1374 
1375 			mutex_enter(&rt->mfc_mutex);
1376 			if ((rt->mfc_origin.s_addr ==
1377 			    mfccp->mfcc_origin.s_addr) &&
1378 			    (rt->mfc_mcastgrp.s_addr ==
1379 			    mfccp->mfcc_mcastgrp.s_addr) &&
1380 			    (!(rt->mfc_marks & MFCB_MARK_CONDEMNED))) {
1381 				fill_route(rt, mfccp, ipst);
1382 				mutex_exit(&rt->mfc_mutex);
1383 				break;
1384 			}
1385 			mutex_exit(&rt->mfc_mutex);
1386 		}
1387 
1388 		/* No upcall, so make a new entry into mfctable */
1389 		if (rt == NULL) {
1390 			rt = (struct mfc *)mi_zalloc(sizeof (struct mfc));
1391 			if (rt == NULL) {
1392 				ip1dbg(("add_mfc: out of memory\n"));
1393 				mutex_exit(&mfcbp->mfcb_lock);
1394 				MFCB_REFRELE(mfcbp);
1395 				return (ENOBUFS);
1396 			}
1397 
1398 			/* Insert new entry at head of hash chain */
1399 			mutex_enter(&rt->mfc_mutex);
1400 			fill_route(rt, mfccp, ipst);
1401 
1402 			/* Link into table */
1403 			rt->mfc_next   = mfcbp->mfcb_mfc;
1404 			mfcbp->mfcb_mfc = rt;
1405 			mutex_exit(&rt->mfc_mutex);
1406 		}
1407 		mutex_exit(&mfcbp->mfcb_lock);
1408 	}
1409 
1410 	MFCB_REFRELE(mfcbp);
1411 	return (0);
1412 }
1413 
1414 /*
1415  * Fills in mfc structure from mrouted mfcctl.
1416  */
1417 static void
1418 fill_route(struct mfc *rt, struct mfcctl *mfccp, ip_stack_t *ipst)
1419 {
1420 	int i;
1421 
1422 	rt->mfc_origin		= mfccp->mfcc_origin;
1423 	rt->mfc_mcastgrp	= mfccp->mfcc_mcastgrp;
1424 	rt->mfc_parent		= mfccp->mfcc_parent;
1425 	mutex_enter(&ipst->ips_numvifs_mutex);
1426 	for (i = 0; i < (int)ipst->ips_numvifs; i++) {
1427 		rt->mfc_ttls[i] = mfccp->mfcc_ttls[i];
1428 	}
1429 	mutex_exit(&ipst->ips_numvifs_mutex);
1430 	/* Initialize pkt counters per src-grp */
1431 	rt->mfc_pkt_cnt	= 0;
1432 	rt->mfc_byte_cnt	= 0;
1433 	rt->mfc_wrong_if	= 0;
1434 	rt->mfc_last_assert.tv_sec = rt->mfc_last_assert.tv_nsec = 0;
1435 
1436 }
1437 
1438 static void
1439 free_queue(struct mfc *mfcp)
1440 {
1441 	struct rtdetq *rte0;
1442 
1443 	/*
1444 	 * Drop all queued upcall packets.
1445 	 * Free the mbuf with the pkt.
1446 	 */
1447 	while ((rte0 = mfcp->mfc_rte) != NULL) {
1448 		mfcp->mfc_rte = rte0->rte_next;
1449 		freemsg(rte0->mp);
1450 		mi_free((char *)rte0);
1451 	}
1452 }
1453 /*
1454  * go thorugh the hash bucket and free all the entries marked condemned.
1455  */
1456 void
1457 release_mfc(struct mfcb *mfcbp)
1458 {
1459 	struct mfc *current_mfcp;
1460 	struct mfc *prev_mfcp;
1461 
1462 	prev_mfcp = current_mfcp = mfcbp->mfcb_mfc;
1463 
1464 	while (current_mfcp != NULL) {
1465 		if (current_mfcp->mfc_marks & MFCB_MARK_CONDEMNED) {
1466 			if (current_mfcp == mfcbp->mfcb_mfc) {
1467 				mfcbp->mfcb_mfc = current_mfcp->mfc_next;
1468 				free_queue(current_mfcp);
1469 				mi_free(current_mfcp);
1470 				prev_mfcp = current_mfcp = mfcbp->mfcb_mfc;
1471 				continue;
1472 			}
1473 			ASSERT(prev_mfcp != NULL);
1474 			prev_mfcp->mfc_next = current_mfcp->mfc_next;
1475 			free_queue(current_mfcp);
1476 			mi_free(current_mfcp);
1477 			current_mfcp = NULL;
1478 		} else {
1479 			prev_mfcp = current_mfcp;
1480 		}
1481 
1482 		current_mfcp = prev_mfcp->mfc_next;
1483 
1484 	}
1485 	mfcbp->mfcb_marks &= ~MFCB_MARK_CONDEMNED;
1486 	ASSERT(mfcbp->mfcb_mfc != NULL || mfcbp->mfcb_marks == 0);
1487 }
1488 
1489 /*
1490  * Delete an mfc entry.
1491  */
1492 static int
1493 del_mfc(struct mfcctl *mfccp, ip_stack_t *ipst)
1494 {
1495 	struct in_addr	origin;
1496 	struct in_addr	mcastgrp;
1497 	struct mfc 	*rt;
1498 	uint_t		hash;
1499 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
1500 
1501 	origin = mfccp->mfcc_origin;
1502 	mcastgrp = mfccp->mfcc_mcastgrp;
1503 	hash = MFCHASH(origin.s_addr, mcastgrp.s_addr);
1504 
1505 	if (ipst->ips_ip_mrtdebug > 1) {
1506 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1507 		    "del_mfc: o %x g %x",
1508 		    ntohl(origin.s_addr),
1509 		    ntohl(mcastgrp.s_addr));
1510 	}
1511 
1512 	MFCB_REFHOLD(&ipst->ips_mfcs[hash]);
1513 
1514 	/* Find mfc in mfctable, finds only entries without upcalls */
1515 	for (rt = ipst->ips_mfcs[hash].mfcb_mfc; rt; rt = rt->mfc_next) {
1516 		mutex_enter(&rt->mfc_mutex);
1517 		if (origin.s_addr == rt->mfc_origin.s_addr &&
1518 		    mcastgrp.s_addr == rt->mfc_mcastgrp.s_addr &&
1519 		    rt->mfc_rte == NULL &&
1520 		    !(rt->mfc_marks & MFCB_MARK_CONDEMNED))
1521 			break;
1522 		mutex_exit(&rt->mfc_mutex);
1523 	}
1524 
1525 	/*
1526 	 * Return if there was an upcall (mfc_rte != NULL,
1527 	 * or rt not in mfctable.
1528 	 */
1529 	if (rt == NULL) {
1530 		MFCB_REFRELE(&ipst->ips_mfcs[hash]);
1531 		return (EADDRNOTAVAIL);
1532 	}
1533 
1534 
1535 	/*
1536 	 * no need to hold lock as we have a reference.
1537 	 */
1538 	ipst->ips_mfcs[hash].mfcb_marks |= MFCB_MARK_CONDEMNED;
1539 	/* error checking */
1540 	if (rt->mfc_timeout_id != 0) {
1541 		ip0dbg(("del_mfc: TIMEOUT NOT 0, rte not null"));
1542 		/*
1543 		 * Its ok to drop the lock,  the struct cannot be freed
1544 		 * since we have a ref on the hash bucket.
1545 		 */
1546 		rt->mfc_timeout_id = 0;
1547 		mutex_exit(&rt->mfc_mutex);
1548 		(void) untimeout(rt->mfc_timeout_id);
1549 		mutex_enter(&rt->mfc_mutex);
1550 	}
1551 
1552 	ASSERT(rt->mfc_rte == NULL);
1553 
1554 
1555 	/*
1556 	 * Delete the entry from the cache
1557 	 */
1558 	rt->mfc_marks |= MFCB_MARK_CONDEMNED;
1559 	mutex_exit(&rt->mfc_mutex);
1560 
1561 	MFCB_REFRELE(&ipst->ips_mfcs[hash]);
1562 
1563 	return (0);
1564 }
1565 
1566 #define	TUNNEL_LEN  12  /* # bytes of IP option for tunnel encapsulation  */
1567 
1568 /*
1569  * IP multicast forwarding function. This function assumes that the packet
1570  * pointed to by ipha has arrived on (or is about to be sent to) the interface
1571  * pointed to by "ill", and the packet is to be relayed to other networks
1572  * that have members of the packet's destination IP multicast group.
1573  *
1574  * The packet is returned unscathed to the caller, unless it is
1575  * erroneous, in which case a -1 value tells the caller (IP)
1576  * to discard it.
1577  *
1578  * Unlike BSD, SunOS 5.x needs to return to IP info about
1579  * whether pkt came in thru a tunnel, so it can be discarded, unless
1580  * it's IGMP. In BSD, the ifp is bogus for tunnels, so pkt won't try
1581  * to be delivered.
1582  * Return values are 0 - pkt is okay and phyint
1583  *		    -1 - pkt is malformed and to be tossed
1584  *                   1 - pkt came in on tunnel
1585  */
1586 int
1587 ip_mforward(ill_t *ill, ipha_t *ipha, mblk_t *mp)
1588 {
1589 	struct mfc 	*rt;
1590 	ipaddr_t	src, dst, tunnel_src = 0;
1591 	static int	srctun = 0;
1592 	vifi_t		vifi;
1593 	boolean_t	pim_reg_packet = B_FALSE;
1594 	struct mfcb *mfcbp;
1595 	ip_stack_t	*ipst = ill->ill_ipst;
1596 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
1597 
1598 	if (ipst->ips_ip_mrtdebug > 1) {
1599 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1600 		    "ip_mforward: RECV ipha_src %x, ipha_dst %x, ill %s",
1601 		    ntohl(ipha->ipha_src), ntohl(ipha->ipha_dst),
1602 		    ill->ill_name);
1603 	}
1604 
1605 	dst = ipha->ipha_dst;
1606 	if ((uint32_t)(uintptr_t)mp->b_prev == PIM_REGISTER_MARKER)
1607 		pim_reg_packet = B_TRUE;
1608 	else
1609 		tunnel_src = (ipaddr_t)(uintptr_t)mp->b_prev;
1610 
1611 	/*
1612 	 * Don't forward a packet with time-to-live of zero or one,
1613 	 * or a packet destined to a local-only group.
1614 	 */
1615 	if (CLASSD(dst) && (ipha->ipha_ttl <= 1 ||
1616 	    (ipaddr_t)ntohl(dst) <= INADDR_MAX_LOCAL_GROUP)) {
1617 		if (ipst->ips_ip_mrtdebug > 1) {
1618 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1619 			    "ip_mforward: not forwarded ttl %d,"
1620 			    " dst 0x%x ill %s",
1621 			    ipha->ipha_ttl, ntohl(dst), ill->ill_name);
1622 		}
1623 		mp->b_prev = NULL;
1624 		if (tunnel_src != 0)
1625 			return (1);
1626 		else
1627 			return (0);
1628 	}
1629 
1630 	if ((tunnel_src != 0) || pim_reg_packet) {
1631 		/*
1632 		 * Packet arrived over an encapsulated tunnel or via a PIM
1633 		 * register message. Both ip_mroute_decap() and pim_input()
1634 		 * encode information in mp->b_prev.
1635 		 */
1636 		mp->b_prev = NULL;
1637 		if (ipst->ips_ip_mrtdebug > 1) {
1638 			if (tunnel_src != 0) {
1639 				(void) mi_strlog(mrouter->conn_rq, 1,
1640 				    SL_TRACE,
1641 				    "ip_mforward: ill %s arrived via ENCAP TUN",
1642 				    ill->ill_name);
1643 			} else if (pim_reg_packet) {
1644 				(void) mi_strlog(mrouter->conn_rq, 1,
1645 				    SL_TRACE,
1646 				    "ip_mforward: ill %s arrived via"
1647 				    "  REGISTER VIF",
1648 				    ill->ill_name);
1649 			}
1650 		}
1651 	} else if ((ipha->ipha_version_and_hdr_length & 0xf) <
1652 	    (uint_t)(IP_SIMPLE_HDR_LENGTH + TUNNEL_LEN) >> 2 ||
1653 	    ((uchar_t *)(ipha + 1))[1] != IPOPT_LSRR) {
1654 		/* Packet arrived via a physical interface. */
1655 		if (ipst->ips_ip_mrtdebug > 1) {
1656 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1657 			    "ip_mforward: ill %s arrived via PHYINT",
1658 			    ill->ill_name);
1659 		}
1660 
1661 	} else {
1662 		/*
1663 		 * Packet arrived through a SRCRT tunnel.
1664 		 * Source-route tunnels are no longer supported.
1665 		 * Error message printed every 1000 times.
1666 		 */
1667 		if ((srctun++ % 1000) == 0) {
1668 			cmn_err(CE_WARN,
1669 			    "ip_mforward: received source-routed pkt from %x",
1670 			    ntohl(ipha->ipha_src));
1671 		}
1672 		return (-1);
1673 	}
1674 
1675 	ipst->ips_mrtstat->mrts_fwd_in++;
1676 	src = ipha->ipha_src;
1677 
1678 	/* Find route in cache, return NULL if not there or upcalls q'ed. */
1679 
1680 	/*
1681 	 * Lock the mfctable against changes made by ip_mforward.
1682 	 * Note that only add_mfc and del_mfc can remove entries and
1683 	 * they run with exclusive access to IP. So we do not need to
1684 	 * guard against the rt being deleted, so release lock after reading.
1685 	 */
1686 
1687 	if (is_mrouter_off(ipst))
1688 		return (-1);
1689 
1690 	mfcbp = &ipst->ips_mfcs[MFCHASH(src, dst)];
1691 	MFCB_REFHOLD(mfcbp);
1692 	MFCFIND(mfcbp, src, dst, rt);
1693 
1694 	/* Entry exists, so forward if necessary */
1695 	if (rt != NULL) {
1696 		int ret = 0;
1697 		ipst->ips_mrtstat->mrts_mfc_hits++;
1698 		if (pim_reg_packet) {
1699 			ASSERT(ipst->ips_reg_vif_num != ALL_VIFS);
1700 			ret = ip_mdq(mp, ipha,
1701 			    ipst->ips_vifs[ipst->ips_reg_vif_num].
1702 			    v_ipif->ipif_ill,
1703 			    0, rt);
1704 		} else {
1705 			ret = ip_mdq(mp, ipha, ill, tunnel_src, rt);
1706 		}
1707 
1708 		MFCB_REFRELE(mfcbp);
1709 		return (ret);
1710 
1711 		/*
1712 		 * Don't forward if we don't have a cache entry.  Mrouted will
1713 		 * always provide a cache entry in response to an upcall.
1714 		 */
1715 	} else {
1716 		/*
1717 		 * If we don't have a route for packet's origin, make a copy
1718 		 * of the packet and send message to routing daemon.
1719 		 */
1720 		struct mfc	*mfc_rt	 = NULL;
1721 		mblk_t		*mp0	 = NULL;
1722 		mblk_t		*mp_copy = NULL;
1723 		struct rtdetq	*rte	 = NULL;
1724 		struct rtdetq	*rte_m, *rte1, *prev_rte;
1725 		uint_t		hash;
1726 		int		npkts;
1727 		boolean_t	new_mfc = B_FALSE;
1728 		ipst->ips_mrtstat->mrts_mfc_misses++;
1729 		/* BSD uses mrts_no_route++ */
1730 		if (ipst->ips_ip_mrtdebug > 1) {
1731 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1732 			    "ip_mforward: no rte ill %s src %x g %x misses %d",
1733 			    ill->ill_name, ntohl(src), ntohl(dst),
1734 			    (int)ipst->ips_mrtstat->mrts_mfc_misses);
1735 		}
1736 		/*
1737 		 * The order of the following code differs from the BSD code.
1738 		 * Pre-mc3.5, the BSD code was incorrect and SunOS 5.x
1739 		 * code works, so SunOS 5.x wasn't changed to conform to the
1740 		 * BSD version.
1741 		 */
1742 
1743 		/* Lock mfctable. */
1744 		hash = MFCHASH(src, dst);
1745 		mutex_enter(&(ipst->ips_mfcs[hash].mfcb_lock));
1746 
1747 		/*
1748 		 * If we are turning off mrouted return an error
1749 		 */
1750 		if (is_mrouter_off(ipst)) {
1751 			mutex_exit(&mfcbp->mfcb_lock);
1752 			MFCB_REFRELE(mfcbp);
1753 			return (-1);
1754 		}
1755 
1756 		/* Is there an upcall waiting for this packet? */
1757 		for (mfc_rt = ipst->ips_mfcs[hash].mfcb_mfc; mfc_rt;
1758 		    mfc_rt = mfc_rt->mfc_next) {
1759 			mutex_enter(&mfc_rt->mfc_mutex);
1760 			if (ipst->ips_ip_mrtdebug > 1) {
1761 				(void) mi_strlog(mrouter->conn_rq, 1,
1762 				    SL_TRACE,
1763 				    "ip_mforward: MFCTAB hash %d o 0x%x"
1764 				    " g 0x%x\n",
1765 				    hash, ntohl(mfc_rt->mfc_origin.s_addr),
1766 				    ntohl(mfc_rt->mfc_mcastgrp.s_addr));
1767 			}
1768 			/* There is an upcall */
1769 			if ((src == mfc_rt->mfc_origin.s_addr) &&
1770 			    (dst == mfc_rt->mfc_mcastgrp.s_addr) &&
1771 			    (mfc_rt->mfc_rte != NULL) &&
1772 			    !(mfc_rt->mfc_marks & MFCB_MARK_CONDEMNED)) {
1773 				break;
1774 			}
1775 			mutex_exit(&mfc_rt->mfc_mutex);
1776 		}
1777 		/* No upcall, so make a new entry into mfctable */
1778 		if (mfc_rt == NULL) {
1779 			mfc_rt = (struct mfc *)mi_zalloc(sizeof (struct mfc));
1780 			if (mfc_rt == NULL) {
1781 				ipst->ips_mrtstat->mrts_fwd_drop++;
1782 				ip1dbg(("ip_mforward: out of memory "
1783 				    "for mfc, mfc_rt\n"));
1784 				goto error_return;
1785 			} else
1786 				new_mfc = B_TRUE;
1787 			/* Get resources */
1788 			/* TODO could copy header and dup rest */
1789 			mp_copy = copymsg(mp);
1790 			if (mp_copy == NULL) {
1791 				ipst->ips_mrtstat->mrts_fwd_drop++;
1792 				ip1dbg(("ip_mforward: out of memory for "
1793 				    "mblk, mp_copy\n"));
1794 				goto error_return;
1795 			}
1796 			mutex_enter(&mfc_rt->mfc_mutex);
1797 		}
1798 		/* Get resources for rte, whether first rte or not first. */
1799 		/* Add this packet into rtdetq */
1800 		rte = (struct rtdetq *)mi_zalloc(sizeof (struct rtdetq));
1801 		if (rte == NULL) {
1802 			ipst->ips_mrtstat->mrts_fwd_drop++;
1803 			mutex_exit(&mfc_rt->mfc_mutex);
1804 			ip1dbg(("ip_mforward: out of memory for"
1805 			    " rtdetq, rte\n"));
1806 			goto error_return;
1807 		}
1808 
1809 		mp0 = copymsg(mp);
1810 		if (mp0 == NULL) {
1811 			ipst->ips_mrtstat->mrts_fwd_drop++;
1812 			ip1dbg(("ip_mforward: out of memory for mblk, mp0\n"));
1813 			mutex_exit(&mfc_rt->mfc_mutex);
1814 			goto error_return;
1815 		}
1816 		rte->mp		= mp0;
1817 		if (pim_reg_packet) {
1818 			ASSERT(ipst->ips_reg_vif_num != ALL_VIFS);
1819 			rte->ill =
1820 			    ipst->ips_vifs[ipst->ips_reg_vif_num].
1821 			    v_ipif->ipif_ill;
1822 		} else {
1823 			rte->ill = ill;
1824 		}
1825 		rte->rte_next	= NULL;
1826 
1827 		/*
1828 		 * Determine if upcall q (rtdetq) has overflowed.
1829 		 * mfc_rt->mfc_rte is null by mi_zalloc
1830 		 * if it is the first message.
1831 		 */
1832 		for (rte_m = mfc_rt->mfc_rte, npkts = 0; rte_m;
1833 		    rte_m = rte_m->rte_next)
1834 			npkts++;
1835 		if (ipst->ips_ip_mrtdebug > 1) {
1836 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1837 			    "ip_mforward: upcalls %d\n", npkts);
1838 		}
1839 		if (npkts > MAX_UPQ) {
1840 			ipst->ips_mrtstat->mrts_upq_ovflw++;
1841 			mutex_exit(&mfc_rt->mfc_mutex);
1842 			goto error_return;
1843 		}
1844 
1845 		if (npkts == 0) {	/* first upcall */
1846 			int i = 0;
1847 			/*
1848 			 * Now finish installing the new mfc! Now that we have
1849 			 * resources!  Insert new entry at head of hash chain.
1850 			 * Use src and dst which are ipaddr_t's.
1851 			 */
1852 			mfc_rt->mfc_origin.s_addr = src;
1853 			mfc_rt->mfc_mcastgrp.s_addr = dst;
1854 
1855 			mutex_enter(&ipst->ips_numvifs_mutex);
1856 			for (i = 0; i < (int)ipst->ips_numvifs; i++)
1857 				mfc_rt->mfc_ttls[i] = 0;
1858 			mutex_exit(&ipst->ips_numvifs_mutex);
1859 			mfc_rt->mfc_parent = ALL_VIFS;
1860 
1861 			/* Link into table */
1862 			if (ipst->ips_ip_mrtdebug > 1) {
1863 				(void) mi_strlog(mrouter->conn_rq, 1,
1864 				    SL_TRACE,
1865 				    "ip_mforward: NEW MFCTAB hash %d o 0x%x "
1866 				    "g 0x%x\n", hash,
1867 				    ntohl(mfc_rt->mfc_origin.s_addr),
1868 				    ntohl(mfc_rt->mfc_mcastgrp.s_addr));
1869 			}
1870 			mfc_rt->mfc_next = ipst->ips_mfcs[hash].mfcb_mfc;
1871 			ipst->ips_mfcs[hash].mfcb_mfc = mfc_rt;
1872 			mfc_rt->mfc_rte = NULL;
1873 		}
1874 
1875 		/* Link in the upcall */
1876 		/* First upcall */
1877 		if (mfc_rt->mfc_rte == NULL)
1878 			mfc_rt->mfc_rte = rte;
1879 		else {
1880 			/* not the first upcall */
1881 			prev_rte = mfc_rt->mfc_rte;
1882 			for (rte1 = mfc_rt->mfc_rte->rte_next; rte1;
1883 			    prev_rte = rte1, rte1 = rte1->rte_next)
1884 				;
1885 			prev_rte->rte_next = rte;
1886 		}
1887 
1888 		/*
1889 		 * No upcalls waiting, this is first one, so send a message to
1890 		 * routing daemon to install a route into kernel table.
1891 		 */
1892 		if (npkts == 0) {
1893 			struct igmpmsg	*im;
1894 			/* ipha_protocol is 0, for upcall */
1895 			ASSERT(mp_copy != NULL);
1896 			im = (struct igmpmsg *)mp_copy->b_rptr;
1897 			im->im_msgtype	= IGMPMSG_NOCACHE;
1898 			im->im_mbz = 0;
1899 			mutex_enter(&ipst->ips_numvifs_mutex);
1900 			if (pim_reg_packet) {
1901 				im->im_vif = (uchar_t)ipst->ips_reg_vif_num;
1902 				mutex_exit(&ipst->ips_numvifs_mutex);
1903 			} else {
1904 				/*
1905 				 * XXX do we need to hold locks here ?
1906 				 */
1907 				for (vifi = 0;
1908 				    vifi < ipst->ips_numvifs;
1909 				    vifi++) {
1910 					if (ipst->ips_vifs[vifi].v_ipif == NULL)
1911 						continue;
1912 					if (ipst->ips_vifs[vifi].
1913 					    v_ipif->ipif_ill == ill) {
1914 						im->im_vif = (uchar_t)vifi;
1915 						break;
1916 					}
1917 				}
1918 				mutex_exit(&ipst->ips_numvifs_mutex);
1919 				ASSERT(vifi < ipst->ips_numvifs);
1920 			}
1921 
1922 			ipst->ips_mrtstat->mrts_upcalls++;
1923 			/* Timer to discard upcalls if mrouted is too slow */
1924 			mfc_rt->mfc_timeout_id = timeout(expire_upcalls,
1925 			    mfc_rt, EXPIRE_TIMEOUT * UPCALL_EXPIRE);
1926 			mutex_exit(&mfc_rt->mfc_mutex);
1927 			mutex_exit(&(ipst->ips_mfcs[hash].mfcb_lock));
1928 			/* Pass to RAWIP */
1929 			(mrouter->conn_recv)(mrouter, mp_copy, NULL);
1930 		} else {
1931 			mutex_exit(&mfc_rt->mfc_mutex);
1932 			mutex_exit(&(ipst->ips_mfcs[hash].mfcb_lock));
1933 			freemsg(mp_copy);
1934 		}
1935 
1936 		MFCB_REFRELE(mfcbp);
1937 		if (tunnel_src != 0)
1938 			return (1);
1939 		else
1940 			return (0);
1941 	error_return:
1942 		mutex_exit(&(ipst->ips_mfcs[hash].mfcb_lock));
1943 		MFCB_REFRELE(mfcbp);
1944 		if (mfc_rt != NULL && (new_mfc == B_TRUE))
1945 			mi_free((char *)mfc_rt);
1946 		if (rte != NULL)
1947 			mi_free((char *)rte);
1948 		if (mp_copy != NULL)
1949 			freemsg(mp_copy);
1950 		if (mp0 != NULL)
1951 			freemsg(mp0);
1952 		return (-1);
1953 	}
1954 }
1955 
1956 /*
1957  * Clean up the mfctable cache entry if upcall is not serviced.
1958  * SunOS 5.x has timeout per mfc, unlike BSD which has one timer.
1959  */
1960 static void
1961 expire_upcalls(void *arg)
1962 {
1963 	struct mfc *mfc_rt = arg;
1964 	uint_t hash;
1965 	struct mfc *prev_mfc, *mfc0;
1966 	ip_stack_t	*ipst;
1967 	conn_t		*mrouter;
1968 
1969 	if (mfc_rt->mfc_rte == NULL || mfc_rt->mfc_rte->ill != NULL) {
1970 		cmn_err(CE_WARN, "expire_upcalls: no ILL\n");
1971 		return;
1972 	}
1973 	ipst = mfc_rt->mfc_rte->ill->ill_ipst;
1974 	mrouter = ipst->ips_ip_g_mrouter;
1975 
1976 	hash = MFCHASH(mfc_rt->mfc_origin.s_addr, mfc_rt->mfc_mcastgrp.s_addr);
1977 	if (ipst->ips_ip_mrtdebug > 1) {
1978 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
1979 		    "expire_upcalls: hash %d s %x g %x",
1980 		    hash, ntohl(mfc_rt->mfc_origin.s_addr),
1981 		    ntohl(mfc_rt->mfc_mcastgrp.s_addr));
1982 	}
1983 	MFCB_REFHOLD(&ipst->ips_mfcs[hash]);
1984 	mutex_enter(&mfc_rt->mfc_mutex);
1985 	/*
1986 	 * if timeout has been set to zero, than the
1987 	 * entry has been filled, no need to delete it.
1988 	 */
1989 	if (mfc_rt->mfc_timeout_id == 0)
1990 		goto done;
1991 	ipst->ips_mrtstat->mrts_cache_cleanups++;
1992 	mfc_rt->mfc_timeout_id = 0;
1993 
1994 	/* Determine entry to be cleaned up in cache table. */
1995 	for (prev_mfc = mfc0 = ipst->ips_mfcs[hash].mfcb_mfc; mfc0;
1996 	    prev_mfc = mfc0, mfc0 = mfc0->mfc_next)
1997 		if (mfc0 == mfc_rt)
1998 			break;
1999 
2000 	/* del_mfc takes care of gone mfcs */
2001 	ASSERT(prev_mfc != NULL);
2002 	ASSERT(mfc0 != NULL);
2003 
2004 	/*
2005 	 * Delete the entry from the cache
2006 	 */
2007 	ipst->ips_mfcs[hash].mfcb_marks |= MFCB_MARK_CONDEMNED;
2008 	mfc_rt->mfc_marks |= MFCB_MARK_CONDEMNED;
2009 
2010 	/*
2011 	 * release_mfc will drop all queued upcall packets.
2012 	 * and will free the mbuf with the pkt, if, timing info.
2013 	 */
2014 done:
2015 	mutex_exit(&mfc_rt->mfc_mutex);
2016 	MFCB_REFRELE(&ipst->ips_mfcs[hash]);
2017 }
2018 
2019 /*
2020  * Packet forwarding routine once entry in the cache is made.
2021  */
2022 static int
2023 ip_mdq(mblk_t *mp, ipha_t *ipha, ill_t *ill, ipaddr_t tunnel_src,
2024     struct mfc *rt)
2025 {
2026 	ill_t *vill;
2027 	vifi_t vifi;
2028 	struct vif *vifp;
2029 	ipaddr_t dst = ipha->ipha_dst;
2030 	size_t  plen = msgdsize(mp);
2031 	vifi_t num_of_vifs;
2032 	ip_stack_t	*ipst = ill->ill_ipst;
2033 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2034 
2035 	if (ipst->ips_ip_mrtdebug > 1) {
2036 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2037 		    "ip_mdq: SEND src %x, ipha_dst %x, ill %s",
2038 		    ntohl(ipha->ipha_src), ntohl(ipha->ipha_dst),
2039 		    ill->ill_name);
2040 	}
2041 
2042 	/* Macro to send packet on vif */
2043 #define	MC_SEND(ipha, mp, vifp, dst) { \
2044 	if ((vifp)->v_flags & VIFF_TUNNEL) \
2045 		encap_send((ipha), (mp), (vifp), (dst)); \
2046 	else if ((vifp)->v_flags & VIFF_REGISTER) \
2047 		register_send((ipha), (mp), (vifp), (dst)); \
2048 	else \
2049 		phyint_send((ipha), (mp), (vifp), (dst)); \
2050 }
2051 
2052 	vifi = rt->mfc_parent;
2053 
2054 	/*
2055 	 * The value of vifi is MAXVIFS if the pkt had no parent, i.e.,
2056 	 * Mrouted had no route.
2057 	 * We wanted the route installed in the mfctable to prevent multiple
2058 	 * tries, so it passed add_mfc(), but is discarded here. The v_ipif is
2059 	 * NULL so we don't want to check the ill. Still needed as of Mrouted
2060 	 * 3.6.
2061 	 */
2062 	if (vifi == NO_VIF) {
2063 		ip1dbg(("ip_mdq: no route for origin ill %s, vifi is NO_VIF\n",
2064 		    ill->ill_name));
2065 		if (ipst->ips_ip_mrtdebug > 1) {
2066 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2067 			    "ip_mdq: vifi is NO_VIF ill = %s", ill->ill_name);
2068 		}
2069 		return (-1);	/* drop pkt */
2070 	}
2071 
2072 	if (!lock_good_vif(&ipst->ips_vifs[vifi]))
2073 		return (-1);
2074 	/*
2075 	 * The MFC entries are not cleaned up when an ipif goes
2076 	 * away thus this code has to guard against an MFC referencing
2077 	 * an ipif that has been closed. Note: reset_mrt_vif_ipif
2078 	 * sets the v_ipif to NULL when the ipif disappears.
2079 	 */
2080 	ASSERT(ipst->ips_vifs[vifi].v_ipif != NULL);
2081 
2082 	if (vifi >= ipst->ips_numvifs) {
2083 		cmn_err(CE_WARN, "ip_mdq: illegal vifi %d numvifs "
2084 		    "%d ill %s viftable ill %s\n",
2085 		    (int)vifi, (int)ipst->ips_numvifs, ill->ill_name,
2086 		    ipst->ips_vifs[vifi].v_ipif->ipif_ill->ill_name);
2087 		unlock_good_vif(&ipst->ips_vifs[vifi]);
2088 		return (-1);
2089 	}
2090 	/*
2091 	 * Don't forward if it didn't arrive from the parent vif for its
2092 	 * origin.
2093 	 */
2094 	vill = ipst->ips_vifs[vifi].v_ipif->ipif_ill;
2095 	if ((vill != ill && !IS_IN_SAME_ILLGRP(vill, ill)) ||
2096 	    (ipst->ips_vifs[vifi].v_rmt_addr.s_addr != tunnel_src)) {
2097 		/* Came in the wrong interface */
2098 		ip1dbg(("ip_mdq: arrived wrong if, vifi %d "
2099 			"numvifs %d ill %s viftable ill %s\n",
2100 			(int)vifi, (int)ipst->ips_numvifs, ill->ill_name,
2101 			vill->ill_name));
2102 		if (ipst->ips_ip_mrtdebug > 1) {
2103 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2104 			    "ip_mdq: arrived wrong if, vifi %d ill "
2105 			    "%s viftable ill %s\n",
2106 			    (int)vifi, ill->ill_name, vill->ill_name);
2107 		}
2108 		ipst->ips_mrtstat->mrts_wrong_if++;
2109 		rt->mfc_wrong_if++;
2110 
2111 		/*
2112 		 * If we are doing PIM assert processing and we are forwarding
2113 		 * packets on this interface, and it is a broadcast medium
2114 		 * interface (and not a tunnel), send a message to the routing.
2115 		 *
2116 		 * We use the first ipif on the list, since it's all we have.
2117 		 * Chances are the ipif_flags are the same for ipifs on the ill.
2118 		 */
2119 		if (ipst->ips_pim_assert && rt->mfc_ttls[vifi] > 0 &&
2120 		    (ill->ill_ipif->ipif_flags & IPIF_BROADCAST) &&
2121 		    !(ipst->ips_vifs[vifi].v_flags & VIFF_TUNNEL)) {
2122 			mblk_t		*mp_copy;
2123 			struct igmpmsg	*im;
2124 
2125 			/* TODO could copy header and dup rest */
2126 			mp_copy = copymsg(mp);
2127 			if (mp_copy == NULL) {
2128 				ipst->ips_mrtstat->mrts_fwd_drop++;
2129 				ip1dbg(("ip_mdq: out of memory "
2130 				    "for mblk, mp_copy\n"));
2131 				unlock_good_vif(&ipst->ips_vifs[vifi]);
2132 				return (-1);
2133 			}
2134 
2135 			im = (struct igmpmsg *)mp_copy->b_rptr;
2136 			im->im_msgtype = IGMPMSG_WRONGVIF;
2137 			im->im_mbz = 0;
2138 			im->im_vif = (ushort_t)vifi;
2139 			/* Pass to RAWIP */
2140 			(mrouter->conn_recv)(mrouter, mp_copy, NULL);
2141 		}
2142 		unlock_good_vif(&ipst->ips_vifs[vifi]);
2143 		if (tunnel_src != 0)
2144 			return (1);
2145 		else
2146 			return (0);
2147 	}
2148 	/*
2149 	 * If I sourced this packet, it counts as output, else it was input.
2150 	 */
2151 	if (ipha->ipha_src == ipst->ips_vifs[vifi].v_lcl_addr.s_addr) {
2152 		ipst->ips_vifs[vifi].v_pkt_out++;
2153 		ipst->ips_vifs[vifi].v_bytes_out += plen;
2154 	} else {
2155 		ipst->ips_vifs[vifi].v_pkt_in++;
2156 		ipst->ips_vifs[vifi].v_bytes_in += plen;
2157 	}
2158 	mutex_enter(&rt->mfc_mutex);
2159 	rt->mfc_pkt_cnt++;
2160 	rt->mfc_byte_cnt += plen;
2161 	mutex_exit(&rt->mfc_mutex);
2162 	unlock_good_vif(&ipst->ips_vifs[vifi]);
2163 	/*
2164 	 * For each vif, decide if a copy of the packet should be forwarded.
2165 	 * Forward if:
2166 	 *		- the vif threshold ttl is non-zero AND
2167 	 *		- the pkt ttl exceeds the vif's threshold
2168 	 * A non-zero mfc_ttl indicates that the vif is part of
2169 	 * the output set for the mfc entry.
2170 	 */
2171 	mutex_enter(&ipst->ips_numvifs_mutex);
2172 	num_of_vifs = ipst->ips_numvifs;
2173 	mutex_exit(&ipst->ips_numvifs_mutex);
2174 	for (vifp = ipst->ips_vifs, vifi = 0;
2175 	    vifi < num_of_vifs;
2176 	    vifp++, vifi++) {
2177 		if (!lock_good_vif(vifp))
2178 			continue;
2179 		if ((rt->mfc_ttls[vifi] > 0) &&
2180 		    (ipha->ipha_ttl > rt->mfc_ttls[vifi])) {
2181 			/*
2182 			 * lock_good_vif should not have succedded if
2183 			 * v_ipif is null.
2184 			 */
2185 			ASSERT(vifp->v_ipif != NULL);
2186 			vifp->v_pkt_out++;
2187 			vifp->v_bytes_out += plen;
2188 			MC_SEND(ipha, mp, vifp, dst);
2189 			ipst->ips_mrtstat->mrts_fwd_out++;
2190 		}
2191 		unlock_good_vif(vifp);
2192 	}
2193 	if (tunnel_src != 0)
2194 		return (1);
2195 	else
2196 		return (0);
2197 }
2198 
2199 /*
2200  * Send the packet on physical interface.
2201  * Caller assumes can continue to use mp on return.
2202  */
2203 /* ARGSUSED */
2204 static void
2205 phyint_send(ipha_t *ipha, mblk_t *mp, struct vif *vifp, ipaddr_t dst)
2206 {
2207 	mblk_t 	*mp_copy;
2208 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
2209 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2210 
2211 	/* Make a new reference to the packet */
2212 	mp_copy = copymsg(mp);	/* TODO could copy header and dup rest */
2213 	if (mp_copy == NULL) {
2214 		ipst->ips_mrtstat->mrts_fwd_drop++;
2215 		ip1dbg(("phyint_send: out of memory for mblk, mp_copy\n"));
2216 		return;
2217 	}
2218 	if (vifp->v_rate_limit <= 0)
2219 		tbf_send_packet(vifp, mp_copy);
2220 	else  {
2221 		if (ipst->ips_ip_mrtdebug > 1) {
2222 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2223 			    "phyint_send: tbf_contr rate %d "
2224 			    "vifp 0x%p mp 0x%p dst 0x%x",
2225 			    vifp->v_rate_limit, (void *)vifp, (void *)mp, dst);
2226 		}
2227 		tbf_control(vifp, mp_copy, (ipha_t *)mp_copy->b_rptr);
2228 	}
2229 }
2230 
2231 /*
2232  * Send the whole packet for REGISTER encapsulation to PIM daemon
2233  * Caller assumes it can continue to use mp on return.
2234  */
2235 /* ARGSUSED */
2236 static void
2237 register_send(ipha_t *ipha, mblk_t *mp, struct vif *vifp, ipaddr_t dst)
2238 {
2239 	struct igmpmsg	*im;
2240 	mblk_t		*mp_copy;
2241 	ipha_t		*ipha_copy;
2242 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
2243 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2244 
2245 	if (ipst->ips_ip_mrtdebug > 1) {
2246 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2247 		    "register_send: src %x, dst %x\n",
2248 		    ntohl(ipha->ipha_src), ntohl(ipha->ipha_dst));
2249 	}
2250 
2251 	/*
2252 	 * Copy the old packet & pullup its IP header into the new mblk_t so we
2253 	 * can modify it.  Try to fill the new mblk_t since if we don't the
2254 	 * ethernet driver will.
2255 	 */
2256 	mp_copy = allocb(sizeof (struct igmpmsg) + sizeof (ipha_t), BPRI_MED);
2257 	if (mp_copy == NULL) {
2258 		++ipst->ips_mrtstat->mrts_pim_nomemory;
2259 		if (ipst->ips_ip_mrtdebug > 3) {
2260 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2261 			    "register_send: allocb failure.");
2262 		}
2263 		return;
2264 	}
2265 
2266 	/*
2267 	 * Bump write pointer to account for igmpmsg being added.
2268 	 */
2269 	mp_copy->b_wptr = mp_copy->b_rptr + sizeof (struct igmpmsg);
2270 
2271 	/*
2272 	 * Chain packet to new mblk_t.
2273 	 */
2274 	if ((mp_copy->b_cont = copymsg(mp)) == NULL) {
2275 		++ipst->ips_mrtstat->mrts_pim_nomemory;
2276 		if (ipst->ips_ip_mrtdebug > 3) {
2277 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2278 			    "register_send: copymsg failure.");
2279 		}
2280 		freeb(mp_copy);
2281 		return;
2282 	}
2283 
2284 	/*
2285 	 * icmp_input() asserts that IP version field is set to an
2286 	 * appropriate version. Hence, the struct igmpmsg that this really
2287 	 * becomes, needs to have the correct IP version field.
2288 	 */
2289 	ipha_copy = (ipha_t *)mp_copy->b_rptr;
2290 	*ipha_copy = multicast_encap_iphdr;
2291 
2292 	/*
2293 	 * The kernel uses the struct igmpmsg header to encode the messages to
2294 	 * the multicast routing daemon. Fill in the fields in the header
2295 	 * starting with the message type which is IGMPMSG_WHOLEPKT
2296 	 */
2297 	im = (struct igmpmsg *)mp_copy->b_rptr;
2298 	im->im_msgtype = IGMPMSG_WHOLEPKT;
2299 	im->im_src.s_addr = ipha->ipha_src;
2300 	im->im_dst.s_addr = ipha->ipha_dst;
2301 
2302 	/*
2303 	 * Must Be Zero. This is because the struct igmpmsg is really an IP
2304 	 * header with renamed fields and the multicast routing daemon uses
2305 	 * an ipha_protocol (aka im_mbz) of 0 to distinguish these messages.
2306 	 */
2307 	im->im_mbz = 0;
2308 
2309 	++ipst->ips_mrtstat->mrts_upcalls;
2310 	if (!canputnext(mrouter->conn_rq)) {
2311 		++ipst->ips_mrtstat->mrts_pim_regsend_drops;
2312 		if (ipst->ips_ip_mrtdebug > 3) {
2313 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2314 			    "register_send: register upcall failure.");
2315 		}
2316 		freemsg(mp_copy);
2317 	} else {
2318 		/* Pass to RAWIP */
2319 		(mrouter->conn_recv)(mrouter, mp_copy, NULL);
2320 	}
2321 }
2322 
2323 /*
2324  * pim_validate_cksum handles verification of the checksum in the
2325  * pim header.  For PIM Register packets, the checksum is calculated
2326  * across the PIM header only.  For all other packets, the checksum
2327  * is for the PIM header and remainder of the packet.
2328  *
2329  * returns: B_TRUE, if checksum is okay.
2330  *          B_FALSE, if checksum is not valid.
2331  */
2332 static boolean_t
2333 pim_validate_cksum(mblk_t *mp, ipha_t *ip, struct pim *pimp)
2334 {
2335 	mblk_t *mp_dup;
2336 
2337 	if ((mp_dup = dupmsg(mp)) == NULL)
2338 		return (B_FALSE);
2339 
2340 	mp_dup->b_rptr += IPH_HDR_LENGTH(ip);
2341 	if (pimp->pim_type == PIM_REGISTER)
2342 		mp_dup->b_wptr = mp_dup->b_rptr + PIM_MINLEN;
2343 	if (IP_CSUM(mp_dup, 0, 0)) {
2344 		freemsg(mp_dup);
2345 		return (B_FALSE);
2346 	}
2347 	freemsg(mp_dup);
2348 	return (B_TRUE);
2349 }
2350 
2351 /*
2352  * int
2353  * pim_input(queue_t *, mblk_t *, ill_t *ill) - Process PIM protocol packets.
2354  *	IP Protocol 103. Register messages are decapsulated and sent
2355  *	onto multicast forwarding.
2356  */
2357 int
2358 pim_input(queue_t *q, mblk_t *mp, ill_t *ill)
2359 {
2360 	ipha_t		*eip, *ip;
2361 	int		iplen, pimlen, iphlen;
2362 	struct pim	*pimp;	/* pointer to a pim struct */
2363 	uint32_t	*reghdr;
2364 	ip_stack_t	*ipst = ill->ill_ipst;
2365 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2366 
2367 	/*
2368 	 * Pullup the msg for PIM protocol processing.
2369 	 */
2370 	if (pullupmsg(mp, -1) == 0) {
2371 		++ipst->ips_mrtstat->mrts_pim_nomemory;
2372 		freemsg(mp);
2373 		return (-1);
2374 	}
2375 
2376 	ip = (ipha_t *)mp->b_rptr;
2377 	iplen = ip->ipha_length;
2378 	iphlen = IPH_HDR_LENGTH(ip);
2379 	pimlen = ntohs(iplen) - iphlen;
2380 
2381 	/*
2382 	 * Validate lengths
2383 	 */
2384 	if (pimlen < PIM_MINLEN) {
2385 		++ipst->ips_mrtstat->mrts_pim_malformed;
2386 		if (ipst->ips_ip_mrtdebug > 1) {
2387 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2388 			    "pim_input: length not at least minlen");
2389 		}
2390 		freemsg(mp);
2391 		return (-1);
2392 	}
2393 
2394 	/*
2395 	 * Point to the PIM header.
2396 	 */
2397 	pimp = (struct pim *)((caddr_t)ip + iphlen);
2398 
2399 	/*
2400 	 * Check the version number.
2401 	 */
2402 	if (pimp->pim_vers != PIM_VERSION) {
2403 		++ipst->ips_mrtstat->mrts_pim_badversion;
2404 		if (ipst->ips_ip_mrtdebug > 1) {
2405 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2406 			    "pim_input: unknown version of PIM");
2407 		}
2408 		freemsg(mp);
2409 		return (-1);
2410 	}
2411 
2412 	/*
2413 	 * Validate the checksum
2414 	 */
2415 	if (!pim_validate_cksum(mp, ip, pimp)) {
2416 		++ipst->ips_mrtstat->mrts_pim_rcv_badcsum;
2417 		if (ipst->ips_ip_mrtdebug > 1) {
2418 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2419 			    "pim_input: invalid checksum");
2420 		}
2421 		freemsg(mp);
2422 		return (-1);
2423 	}
2424 
2425 	if (pimp->pim_type != PIM_REGISTER)
2426 		return (0);
2427 
2428 	reghdr = (uint32_t *)(pimp + 1);
2429 	eip = (ipha_t *)(reghdr + 1);
2430 
2431 	/*
2432 	 * check if the inner packet is destined to mcast group
2433 	 */
2434 	if (!CLASSD(eip->ipha_dst)) {
2435 		++ipst->ips_mrtstat->mrts_pim_badregisters;
2436 		if (ipst->ips_ip_mrtdebug > 1) {
2437 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2438 			    "pim_input: Inner pkt not mcast .. !");
2439 		}
2440 		freemsg(mp);
2441 		return (-1);
2442 	}
2443 	if (ipst->ips_ip_mrtdebug > 1) {
2444 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2445 		    "register from %x, to %x, len %d",
2446 		    ntohl(eip->ipha_src),
2447 		    ntohl(eip->ipha_dst),
2448 		    ntohs(eip->ipha_length));
2449 	}
2450 	/*
2451 	 * If the null register bit is not set, decapsulate
2452 	 * the packet before forwarding it.
2453 	 */
2454 	if (!(ntohl(*reghdr) & PIM_NULL_REGISTER)) {
2455 		mblk_t *mp_copy;
2456 
2457 		/* Copy the message */
2458 		if ((mp_copy = copymsg(mp)) == NULL) {
2459 			++ipst->ips_mrtstat->mrts_pim_nomemory;
2460 			freemsg(mp);
2461 			return (-1);
2462 		}
2463 
2464 		/*
2465 		 * Decapsulate the packet and give it to
2466 		 * register_mforward.
2467 		 */
2468 		mp_copy->b_rptr += iphlen + sizeof (pim_t) +
2469 		    sizeof (*reghdr);
2470 		if (register_mforward(q, mp_copy, ill) != 0) {
2471 			freemsg(mp);
2472 			return (-1);
2473 		}
2474 	}
2475 
2476 	/*
2477 	 * Pass all valid PIM packets up to any process(es) listening on a raw
2478 	 * PIM socket. For Solaris it is done right after pim_input() is
2479 	 * called.
2480 	 */
2481 	return (0);
2482 }
2483 
2484 /*
2485  * PIM sparse mode hook.  Called by pim_input after decapsulating
2486  * the packet. Loop back the packet, as if we have received it.
2487  * In pim_input() we have to check if the destination is a multicast address.
2488  */
2489 /* ARGSUSED */
2490 static int
2491 register_mforward(queue_t *q, mblk_t *mp, ill_t *ill)
2492 {
2493 	ip_stack_t	*ipst = ill->ill_ipst;
2494 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2495 
2496 	ASSERT(ipst->ips_reg_vif_num <= ipst->ips_numvifs);
2497 
2498 	if (ipst->ips_ip_mrtdebug > 3) {
2499 		ipha_t *ipha;
2500 
2501 		ipha = (ipha_t *)mp->b_rptr;
2502 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2503 		    "register_mforward: src %x, dst %x\n",
2504 		    ntohl(ipha->ipha_src), ntohl(ipha->ipha_dst));
2505 	}
2506 	/*
2507 	 * Need to pass in to ip_mforward() the information that the
2508 	 * packet has arrived on the register_vif. We use the solution that
2509 	 * ip_mroute_decap() employs: use mp->b_prev to pass some information
2510 	 * to ip_mforward(). Nonzero value means the packet has arrived on a
2511 	 * tunnel (ip_mroute_decap() puts the address of the other side of the
2512 	 * tunnel there.) This is safe since ip_rput() either frees the packet
2513 	 * or passes it to ip_mforward(). We use
2514 	 * PIM_REGISTER_MARKER = 0xffffffff to indicate the has arrived on the
2515 	 * register vif. If in the future we have more than one register vifs,
2516 	 * then this will need re-examination.
2517 	 */
2518 	mp->b_prev = (mblk_t *)PIM_REGISTER_MARKER;
2519 	++ipst->ips_mrtstat->mrts_pim_regforwards;
2520 	ip_rput(q, mp);
2521 	return (0);
2522 }
2523 
2524 /*
2525  * Send an encapsulated packet.
2526  * Caller assumes can continue to use mp when routine returns.
2527  */
2528 /* ARGSUSED */
2529 static void
2530 encap_send(ipha_t *ipha, mblk_t *mp, struct vif *vifp, ipaddr_t dst)
2531 {
2532 	mblk_t 	*mp_copy;
2533 	ipha_t 	*ipha_copy;
2534 	size_t	len;
2535 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
2536 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2537 
2538 	if (ipst->ips_ip_mrtdebug > 1) {
2539 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2540 		    "encap_send: vif %ld enter",
2541 		    (ptrdiff_t)(vifp - ipst->ips_vifs));
2542 	}
2543 	len = ntohs(ipha->ipha_length);
2544 
2545 	/*
2546 	 * Copy the old packet & pullup it's IP header into the
2547 	 * new mbuf so we can modify it.  Try to fill the new
2548 	 * mbuf since if we don't the ethernet driver will.
2549 	 */
2550 	mp_copy = allocb(32 + sizeof (multicast_encap_iphdr), BPRI_MED);
2551 	if (mp_copy == NULL)
2552 		return;
2553 	mp_copy->b_rptr += 32;
2554 	mp_copy->b_wptr = mp_copy->b_rptr + sizeof (multicast_encap_iphdr);
2555 	if ((mp_copy->b_cont = copymsg(mp)) == NULL) {
2556 		freeb(mp_copy);
2557 		return;
2558 	}
2559 
2560 	/*
2561 	 * Fill in the encapsulating IP header.
2562 	 * Remote tunnel dst in rmt_addr, from add_vif().
2563 	 */
2564 	ipha_copy = (ipha_t *)mp_copy->b_rptr;
2565 	*ipha_copy = multicast_encap_iphdr;
2566 	ASSERT((len + sizeof (ipha_t)) <= IP_MAXPACKET);
2567 	ipha_copy->ipha_length = htons(len + sizeof (ipha_t));
2568 	ipha_copy->ipha_src = vifp->v_lcl_addr.s_addr;
2569 	ipha_copy->ipha_dst = vifp->v_rmt_addr.s_addr;
2570 	ASSERT(ipha_copy->ipha_ident == 0);
2571 
2572 	/* Turn the encapsulated IP header back into a valid one. */
2573 	ipha = (ipha_t *)mp_copy->b_cont->b_rptr;
2574 	ipha->ipha_ttl--;
2575 	ipha->ipha_hdr_checksum = 0;
2576 	ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
2577 
2578 	if (ipst->ips_ip_mrtdebug > 1) {
2579 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2580 		    "encap_send: group 0x%x", ntohl(ipha->ipha_dst));
2581 	}
2582 	if (vifp->v_rate_limit <= 0)
2583 		tbf_send_packet(vifp, mp_copy);
2584 	else
2585 		/* ipha is from the original header */
2586 		tbf_control(vifp, mp_copy, ipha);
2587 }
2588 
2589 /*
2590  * De-encapsulate a packet and feed it back through IP input.
2591  * This routine is called whenever IP gets a packet with prototype
2592  * IPPROTO_ENCAP and a local destination address.
2593  */
2594 void
2595 ip_mroute_decap(queue_t *q, mblk_t *mp, ill_t *ill)
2596 {
2597 	ipha_t		*ipha = (ipha_t *)mp->b_rptr;
2598 	ipha_t		*ipha_encap;
2599 	int		hlen = IPH_HDR_LENGTH(ipha);
2600 	ipaddr_t	src;
2601 	struct vif	*vifp;
2602 	ip_stack_t	*ipst = ill->ill_ipst;
2603 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2604 
2605 	/*
2606 	 * Dump the packet if it's not to a multicast destination or if
2607 	 * we don't have an encapsulating tunnel with the source.
2608 	 * Note:  This code assumes that the remote site IP address
2609 	 * uniquely identifies the tunnel (i.e., that this site has
2610 	 * at most one tunnel with the remote site).
2611 	 */
2612 	ipha_encap = (ipha_t *)((char *)ipha + hlen);
2613 	if (!CLASSD(ipha_encap->ipha_dst)) {
2614 		ipst->ips_mrtstat->mrts_bad_tunnel++;
2615 		ip1dbg(("ip_mroute_decap: bad tunnel\n"));
2616 		freemsg(mp);
2617 		return;
2618 	}
2619 	src = (ipaddr_t)ipha->ipha_src;
2620 	mutex_enter(&ipst->ips_last_encap_lock);
2621 	if (src != ipst->ips_last_encap_src) {
2622 		struct vif *vife;
2623 
2624 		vifp = ipst->ips_vifs;
2625 		vife = vifp + ipst->ips_numvifs;
2626 		ipst->ips_last_encap_src = src;
2627 		ipst->ips_last_encap_vif = 0;
2628 		for (; vifp < vife; ++vifp) {
2629 			if (!lock_good_vif(vifp))
2630 				continue;
2631 			if (vifp->v_rmt_addr.s_addr == src) {
2632 				if (vifp->v_flags & VIFF_TUNNEL)
2633 					ipst->ips_last_encap_vif = vifp;
2634 				if (ipst->ips_ip_mrtdebug > 1) {
2635 					(void) mi_strlog(mrouter->conn_rq,
2636 					    1, SL_TRACE,
2637 					    "ip_mroute_decap: good tun "
2638 					    "vif %ld with %x",
2639 					    (ptrdiff_t)(vifp - ipst->ips_vifs),
2640 					    ntohl(src));
2641 				}
2642 				unlock_good_vif(vifp);
2643 				break;
2644 			}
2645 			unlock_good_vif(vifp);
2646 		}
2647 	}
2648 	if ((vifp = ipst->ips_last_encap_vif) == 0) {
2649 		mutex_exit(&ipst->ips_last_encap_lock);
2650 		ipst->ips_mrtstat->mrts_bad_tunnel++;
2651 		freemsg(mp);
2652 		ip1dbg(("ip_mroute_decap: vif %ld no tunnel with %x\n",
2653 		    (ptrdiff_t)(vifp - ipst->ips_vifs), ntohl(src)));
2654 		return;
2655 	}
2656 	mutex_exit(&ipst->ips_last_encap_lock);
2657 
2658 	/*
2659 	 * Need to pass in the tunnel source to ip_mforward (so that it can
2660 	 * verify that the packet arrived over the correct vif.)  We use b_prev
2661 	 * to pass this information. This is safe since the ip_rput either
2662 	 * frees the packet or passes it to ip_mforward.
2663 	 */
2664 	mp->b_prev = (mblk_t *)(uintptr_t)src;
2665 	mp->b_rptr += hlen;
2666 	/* Feed back into ip_rput as an M_DATA. */
2667 	ip_rput(q, mp);
2668 }
2669 
2670 /*
2671  * Remove all records with v_ipif == ipif.  Called when an interface goes away
2672  * (stream closed).  Called as writer.
2673  */
2674 void
2675 reset_mrt_vif_ipif(ipif_t *ipif)
2676 {
2677 	vifi_t vifi, tmp_vifi;
2678 	vifi_t num_of_vifs;
2679 	ip_stack_t	*ipst = ipif->ipif_ill->ill_ipst;
2680 
2681 	/* Can't check vifi >= 0 since vifi_t is unsigned! */
2682 
2683 	mutex_enter(&ipst->ips_numvifs_mutex);
2684 	num_of_vifs = ipst->ips_numvifs;
2685 	mutex_exit(&ipst->ips_numvifs_mutex);
2686 
2687 	for (vifi = num_of_vifs; vifi != 0; vifi--) {
2688 		tmp_vifi = vifi - 1;
2689 		if (ipst->ips_vifs[tmp_vifi].v_ipif == ipif) {
2690 			(void) del_vif(&tmp_vifi, NULL, NULL, ipst);
2691 		}
2692 	}
2693 }
2694 
2695 /* Remove pending upcall msgs when ill goes away.  Called by ill_delete.  */
2696 void
2697 reset_mrt_ill(ill_t *ill)
2698 {
2699 	struct mfc		*rt;
2700 	struct rtdetq	*rte;
2701 	int			i;
2702 	ip_stack_t	*ipst = ill->ill_ipst;
2703 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2704 
2705 	for (i = 0; i < MFCTBLSIZ; i++) {
2706 		MFCB_REFHOLD(&ipst->ips_mfcs[i]);
2707 		if ((rt = ipst->ips_mfcs[i].mfcb_mfc) != NULL) {
2708 			if (ipst->ips_ip_mrtdebug > 1) {
2709 				(void) mi_strlog(mrouter->conn_rq, 1,
2710 				    SL_TRACE,
2711 				    "reset_mrt_ill: mfctable [%d]", i);
2712 			}
2713 			while (rt != NULL) {
2714 				mutex_enter(&rt->mfc_mutex);
2715 				while ((rte = rt->mfc_rte) != NULL) {
2716 					if (rte->ill == ill) {
2717 						if (ipst->ips_ip_mrtdebug > 1) {
2718 						(void) mi_strlog(
2719 						    mrouter->conn_rq,
2720 						    1, SL_TRACE,
2721 						    "reset_mrt_ill: "
2722 						    "ill 0x%p", (void *)ill);
2723 						}
2724 						rt->mfc_rte = rte->rte_next;
2725 						freemsg(rte->mp);
2726 						mi_free((char *)rte);
2727 					}
2728 				}
2729 				mutex_exit(&rt->mfc_mutex);
2730 				rt = rt->mfc_next;
2731 			}
2732 		}
2733 		MFCB_REFRELE(&ipst->ips_mfcs[i]);
2734 	}
2735 }
2736 
2737 /*
2738  * Token bucket filter module.
2739  * The ipha is for mcastgrp destination for phyint and encap.
2740  */
2741 static void
2742 tbf_control(struct vif *vifp, mblk_t *mp, ipha_t *ipha)
2743 {
2744 	size_t 	p_len =  msgdsize(mp);
2745 	struct tbf	*t    = vifp->v_tbf;
2746 	timeout_id_t id = 0;
2747 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
2748 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2749 
2750 	/* Drop if packet is too large */
2751 	if (p_len > MAX_BKT_SIZE) {
2752 		ipst->ips_mrtstat->mrts_pkt2large++;
2753 		freemsg(mp);
2754 		return;
2755 	}
2756 	if (ipst->ips_ip_mrtdebug > 1) {
2757 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2758 		    "tbf_ctrl: SEND vif %ld, qlen %d, ipha_dst 0x%x",
2759 		    (ptrdiff_t)(vifp - ipst->ips_vifs), t->tbf_q_len,
2760 		    ntohl(ipha->ipha_dst));
2761 	}
2762 
2763 	mutex_enter(&t->tbf_lock);
2764 
2765 	tbf_update_tokens(vifp);
2766 
2767 	/*
2768 	 * If there are enough tokens,
2769 	 * and the queue is empty, send this packet out.
2770 	 */
2771 	if (ipst->ips_ip_mrtdebug > 1) {
2772 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2773 		    "tbf_control: vif %ld, TOKENS  %d, pkt len  %lu, qlen  %d",
2774 		    (ptrdiff_t)(vifp - ipst->ips_vifs), t->tbf_n_tok, p_len,
2775 		    t->tbf_q_len);
2776 	}
2777 	/* No packets are queued */
2778 	if (t->tbf_q_len == 0) {
2779 		/* queue empty, send packet if enough tokens */
2780 		if (p_len <= t->tbf_n_tok) {
2781 			t->tbf_n_tok -= p_len;
2782 			mutex_exit(&t->tbf_lock);
2783 			tbf_send_packet(vifp, mp);
2784 			return;
2785 		} else {
2786 			/* Queue packet and timeout till later */
2787 			tbf_queue(vifp, mp);
2788 			ASSERT(vifp->v_timeout_id == 0);
2789 			vifp->v_timeout_id = timeout(tbf_reprocess_q, vifp,
2790 			    TBF_REPROCESS);
2791 		}
2792 	} else if (t->tbf_q_len < t->tbf_max_q_len) {
2793 		/* Finite queue length, so queue pkts and process queue */
2794 		tbf_queue(vifp, mp);
2795 		tbf_process_q(vifp);
2796 	} else {
2797 		/* Check that we have UDP header with IP header */
2798 		size_t hdr_length = IPH_HDR_LENGTH(ipha) +
2799 		    sizeof (struct udphdr);
2800 
2801 		if ((mp->b_wptr - mp->b_rptr) < hdr_length) {
2802 			if (!pullupmsg(mp, hdr_length)) {
2803 				freemsg(mp);
2804 				ip1dbg(("tbf_ctl: couldn't pullup udp hdr, "
2805 				    "vif %ld src 0x%x dst 0x%x\n",
2806 				    (ptrdiff_t)(vifp - ipst->ips_vifs),
2807 				    ntohl(ipha->ipha_src),
2808 				    ntohl(ipha->ipha_dst)));
2809 				mutex_exit(&vifp->v_tbf->tbf_lock);
2810 				return;
2811 			} else
2812 				/* Have to reassign ipha after pullupmsg */
2813 				ipha = (ipha_t *)mp->b_rptr;
2814 		}
2815 		/*
2816 		 * Queue length too much,
2817 		 * try to selectively dq, or queue and process
2818 		 */
2819 		if (!tbf_dq_sel(vifp, ipha)) {
2820 			ipst->ips_mrtstat->mrts_q_overflow++;
2821 			freemsg(mp);
2822 		} else {
2823 			tbf_queue(vifp, mp);
2824 			tbf_process_q(vifp);
2825 		}
2826 	}
2827 	if (t->tbf_q_len == 0) {
2828 		id = vifp->v_timeout_id;
2829 		vifp->v_timeout_id = 0;
2830 	}
2831 	mutex_exit(&vifp->v_tbf->tbf_lock);
2832 	if (id != 0)
2833 		(void) untimeout(id);
2834 }
2835 
2836 /*
2837  * Adds a packet to the tbf queue at the interface.
2838  * The ipha is for mcastgrp destination for phyint and encap.
2839  */
2840 static void
2841 tbf_queue(struct vif *vifp, mblk_t *mp)
2842 {
2843 	struct tbf	*t = vifp->v_tbf;
2844 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
2845 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2846 
2847 	if (ipst->ips_ip_mrtdebug > 1) {
2848 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2849 		    "tbf_queue: vif %ld", (ptrdiff_t)(vifp - ipst->ips_vifs));
2850 	}
2851 	ASSERT(MUTEX_HELD(&t->tbf_lock));
2852 
2853 	if (t->tbf_t == NULL) {
2854 		/* Queue was empty */
2855 		t->tbf_q = mp;
2856 	} else {
2857 		/* Insert at tail */
2858 		t->tbf_t->b_next = mp;
2859 	}
2860 	/* set new tail pointer */
2861 	t->tbf_t = mp;
2862 
2863 	mp->b_next = mp->b_prev = NULL;
2864 
2865 	t->tbf_q_len++;
2866 }
2867 
2868 /*
2869  * Process the queue at the vif interface.
2870  * Drops the tbf_lock when sending packets.
2871  *
2872  * NOTE : The caller should quntimeout if the queue length is 0.
2873  */
2874 static void
2875 tbf_process_q(struct vif *vifp)
2876 {
2877 	mblk_t	*mp;
2878 	struct tbf	*t = vifp->v_tbf;
2879 	size_t	len;
2880 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
2881 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2882 
2883 	if (ipst->ips_ip_mrtdebug > 1) {
2884 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2885 		    "tbf_process_q 1: vif %ld qlen = %d",
2886 		    (ptrdiff_t)(vifp - ipst->ips_vifs), t->tbf_q_len);
2887 	}
2888 
2889 	/*
2890 	 * Loop through the queue at the interface and send
2891 	 * as many packets as possible.
2892 	 */
2893 	ASSERT(MUTEX_HELD(&t->tbf_lock));
2894 
2895 	while (t->tbf_q_len > 0) {
2896 		mp = t->tbf_q;
2897 		len = (size_t)msgdsize(mp); /* length of ip pkt */
2898 
2899 		/* Determine if the packet can be sent */
2900 		if (len <= t->tbf_n_tok) {
2901 			/*
2902 			 * If so, reduce no. of tokens, dequeue the packet,
2903 			 * send the packet.
2904 			 */
2905 			t->tbf_n_tok -= len;
2906 
2907 			t->tbf_q = mp->b_next;
2908 			if (--t->tbf_q_len == 0) {
2909 				t->tbf_t = NULL;
2910 			}
2911 			mp->b_next = NULL;
2912 			/* Exit mutex before sending packet, then re-enter */
2913 			mutex_exit(&t->tbf_lock);
2914 			tbf_send_packet(vifp, mp);
2915 			mutex_enter(&t->tbf_lock);
2916 		} else
2917 			break;
2918 	}
2919 }
2920 
2921 /* Called at tbf timeout to update tokens, process q and reset timer.  */
2922 static void
2923 tbf_reprocess_q(void *arg)
2924 {
2925 	struct vif *vifp = arg;
2926 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
2927 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2928 
2929 	mutex_enter(&vifp->v_tbf->tbf_lock);
2930 	vifp->v_timeout_id = 0;
2931 	tbf_update_tokens(vifp);
2932 
2933 	tbf_process_q(vifp);
2934 
2935 	if (vifp->v_tbf->tbf_q_len > 0) {
2936 		vifp->v_timeout_id = timeout(tbf_reprocess_q, vifp,
2937 		    TBF_REPROCESS);
2938 	}
2939 	mutex_exit(&vifp->v_tbf->tbf_lock);
2940 
2941 	if (ipst->ips_ip_mrtdebug > 1) {
2942 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2943 		    "tbf_reprcess_q: vif %ld timeout id = %p",
2944 		    (ptrdiff_t)(vifp - ipst->ips_vifs), vifp->v_timeout_id);
2945 	}
2946 }
2947 
2948 /*
2949  * Function that will selectively discard a member of the tbf queue,
2950  * based on the precedence value and the priority.
2951  *
2952  * NOTE : The caller should quntimeout if the queue length is 0.
2953  */
2954 static int
2955 tbf_dq_sel(struct vif *vifp, ipha_t *ipha)
2956 {
2957 	uint_t		p;
2958 	struct tbf		*t = vifp->v_tbf;
2959 	mblk_t		**np;
2960 	mblk_t		*last, *mp;
2961 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
2962 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
2963 
2964 	if (ipst->ips_ip_mrtdebug > 1) {
2965 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
2966 		    "dq_sel: vif %ld dst 0x%x",
2967 		    (ptrdiff_t)(vifp - ipst->ips_vifs), ntohl(ipha->ipha_dst));
2968 	}
2969 
2970 	ASSERT(MUTEX_HELD(&t->tbf_lock));
2971 	p = priority(vifp, ipha);
2972 
2973 	np = &t->tbf_q;
2974 	last = NULL;
2975 	while ((mp = *np) != NULL) {
2976 		if (p > (priority(vifp, (ipha_t *)mp->b_rptr))) {
2977 			*np = mp->b_next;
2978 			/* If removing the last packet, fix the tail pointer */
2979 			if (mp == t->tbf_t)
2980 				t->tbf_t = last;
2981 			mp->b_prev = mp->b_next = NULL;
2982 			freemsg(mp);
2983 			/*
2984 			 * It's impossible for the queue to be empty, but
2985 			 * we check anyway.
2986 			 */
2987 			if (--t->tbf_q_len == 0) {
2988 				t->tbf_t = NULL;
2989 			}
2990 			ipst->ips_mrtstat->mrts_drop_sel++;
2991 			return (1);
2992 		}
2993 		np = &mp->b_next;
2994 		last = mp;
2995 	}
2996 	return (0);
2997 }
2998 
2999 /* Sends packet, 2 cases - encap tunnel, phyint.  */
3000 static void
3001 tbf_send_packet(struct vif *vifp, mblk_t *mp)
3002 {
3003 	ipif_t  *ipif;
3004 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
3005 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
3006 
3007 	/* If encap tunnel options */
3008 	if (vifp->v_flags & VIFF_TUNNEL)  {
3009 		if (ipst->ips_ip_mrtdebug > 1) {
3010 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
3011 			    "tbf_send_pkt: ENCAP tunnel vif %ld",
3012 			    (ptrdiff_t)(vifp - ipst->ips_vifs));
3013 		}
3014 
3015 		/*
3016 		 * Feed into ip_wput which will set the ident field and
3017 		 * checksum the encapsulating header.
3018 		 * BSD gets the cached route vifp->v_route from ip_output()
3019 		 * to speed up route table lookups. Not necessary in SunOS 5.x.
3020 		 */
3021 		put(vifp->v_ipif->ipif_wq, mp);
3022 		return;
3023 
3024 		/* phyint */
3025 	} else {
3026 		/* Need to loop back to members on the outgoing interface. */
3027 		ipha_t  *ipha;
3028 		ipaddr_t    dst;
3029 		ipha  = (ipha_t *)mp->b_rptr;
3030 		dst  = ipha->ipha_dst;
3031 		ipif = vifp->v_ipif;
3032 
3033 		if (ilm_lookup_ipif(ipif, dst) != NULL) {
3034 			/*
3035 			 * The packet is not yet reassembled, thus we need to
3036 			 * pass it to ip_rput_local for checksum verification
3037 			 * and reassembly (and fanout the user stream).
3038 			 */
3039 			mblk_t 	*mp_loop;
3040 			ire_t	*ire;
3041 
3042 			if (ipst->ips_ip_mrtdebug > 1) {
3043 				(void) mi_strlog(mrouter->conn_rq, 1,
3044 				    SL_TRACE,
3045 				    "tbf_send_pkt: loopback vif %ld",
3046 				    (ptrdiff_t)(vifp - ipst->ips_vifs));
3047 			}
3048 			mp_loop = copymsg(mp);
3049 			ire = ire_ctable_lookup(~0, 0, IRE_BROADCAST, NULL,
3050 			    ALL_ZONES, NULL, MATCH_IRE_TYPE, ipst);
3051 
3052 			if (mp_loop != NULL && ire != NULL) {
3053 				IP_RPUT_LOCAL(ipif->ipif_rq, mp_loop,
3054 				    ((ipha_t *)mp_loop->b_rptr),
3055 				    ire, (ill_t *)ipif->ipif_rq->q_ptr);
3056 			} else {
3057 				/* Either copymsg failed or no ire */
3058 				(void) mi_strlog(mrouter->conn_rq, 1,
3059 				    SL_TRACE,
3060 				    "tbf_send_pkt: mp_loop 0x%p, ire 0x%p "
3061 				    "vif %ld\n", (void *)mp_loop, (void *)ire,
3062 				    (ptrdiff_t)(vifp - ipst->ips_vifs));
3063 			}
3064 			if (ire != NULL)
3065 				ire_refrele(ire);
3066 		}
3067 		if (ipst->ips_ip_mrtdebug > 1) {
3068 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
3069 			    "tbf_send_pkt: phyint forward  vif %ld dst = 0x%x",
3070 			    (ptrdiff_t)(vifp - ipst->ips_vifs), ntohl(dst));
3071 		}
3072 		ip_rput_forward_multicast(dst, mp, ipif);
3073 	}
3074 }
3075 
3076 /*
3077  * Determine the current time and then the elapsed time (between the last time
3078  * and time now).  Update the no. of tokens in the bucket.
3079  */
3080 static void
3081 tbf_update_tokens(struct vif *vifp)
3082 {
3083 	timespec_t	tp;
3084 	hrtime_t	tm;
3085 	struct tbf	*t = vifp->v_tbf;
3086 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
3087 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
3088 
3089 	ASSERT(MUTEX_HELD(&t->tbf_lock));
3090 
3091 	/* Time in secs and nsecs, rate limit in kbits/sec */
3092 	gethrestime(&tp);
3093 
3094 	/*LINTED*/
3095 	TV_DELTA(tp, t->tbf_last_pkt_t, tm);
3096 
3097 	/*
3098 	 * This formula is actually
3099 	 * "time in seconds" * "bytes/second".  Scaled for nsec.
3100 	 * (tm/1000000000) * (v_rate_limit * 1000 * (1000/1024) /8)
3101 	 *
3102 	 * The (1000/1024) was introduced in add_vif to optimize
3103 	 * this divide into a shift.
3104 	 */
3105 	t->tbf_n_tok += (tm/1000) * vifp->v_rate_limit / 1024 / 8;
3106 	t->tbf_last_pkt_t = tp;
3107 
3108 	if (t->tbf_n_tok > MAX_BKT_SIZE)
3109 		t->tbf_n_tok = MAX_BKT_SIZE;
3110 	if (ipst->ips_ip_mrtdebug > 1) {
3111 		(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
3112 		    "tbf_update_tok: tm %lld tok %d vif %ld",
3113 		    tm, t->tbf_n_tok, (ptrdiff_t)(vifp - ipst->ips_vifs));
3114 	}
3115 }
3116 
3117 /*
3118  * Priority currently is based on port nos.
3119  * Different forwarding mechanisms have different ways
3120  * of obtaining the port no. Hence, the vif must be
3121  * given along with the packet itself.
3122  *
3123  */
3124 static int
3125 priority(struct vif *vifp, ipha_t *ipha)
3126 {
3127 	int prio;
3128 	ip_stack_t	*ipst = vifp->v_ipif->ipif_ill->ill_ipst;
3129 	conn_t		*mrouter = ipst->ips_ip_g_mrouter;
3130 
3131 	/* Temporary hack; may add general packet classifier some day */
3132 
3133 	ASSERT(MUTEX_HELD(&vifp->v_tbf->tbf_lock));
3134 
3135 	/*
3136 	 * The UDP port space is divided up into four priority ranges:
3137 	 * [0, 16384)	: unclassified - lowest priority
3138 	 * [16384, 32768)	: audio - highest priority
3139 	 * [32768, 49152)	: whiteboard - medium priority
3140 	 * [49152, 65536)	: video - low priority
3141 	 */
3142 
3143 	if (ipha->ipha_protocol == IPPROTO_UDP) {
3144 		struct udphdr *udp =
3145 		    (struct udphdr *)((char *)ipha + IPH_HDR_LENGTH(ipha));
3146 		switch (ntohs(udp->uh_dport) & 0xc000) {
3147 		case 0x4000:
3148 			prio = 70;
3149 			break;
3150 		case 0x8000:
3151 			prio = 60;
3152 			break;
3153 		case 0xc000:
3154 			prio = 55;
3155 			break;
3156 		default:
3157 			prio = 50;
3158 			break;
3159 		}
3160 		if (ipst->ips_ip_mrtdebug > 1) {
3161 			(void) mi_strlog(mrouter->conn_rq, 1, SL_TRACE,
3162 			    "priority: port %x prio %d\n",
3163 			    ntohs(udp->uh_dport), prio);
3164 		}
3165 	} else
3166 		prio = 50;  /* default priority */
3167 	return (prio);
3168 }
3169 
3170 /*
3171  * End of token bucket filter modifications
3172  */
3173 
3174 
3175 
3176 /*
3177  * Produces data for netstat -M.
3178  */
3179 int
3180 ip_mroute_stats(mblk_t *mp, ip_stack_t *ipst)
3181 {
3182 	ipst->ips_mrtstat->mrts_vifctlSize = sizeof (struct vifctl);
3183 	ipst->ips_mrtstat->mrts_mfcctlSize = sizeof (struct mfcctl);
3184 	if (!snmp_append_data(mp, (char *)ipst->ips_mrtstat,
3185 		sizeof (struct mrtstat))) {
3186 		ip0dbg(("ip_mroute_stats: failed %ld bytes\n",
3187 		    (size_t)sizeof (struct mrtstat)));
3188 		return (0);
3189 	}
3190 	return (1);
3191 }
3192 
3193 /*
3194  * Sends info for SNMP's MIB.
3195  */
3196 int
3197 ip_mroute_vif(mblk_t *mp, ip_stack_t *ipst)
3198 {
3199 	struct vifctl 	vi;
3200 	vifi_t		vifi;
3201 
3202 	mutex_enter(&ipst->ips_numvifs_mutex);
3203 	for (vifi = 0; vifi < ipst->ips_numvifs; vifi++) {
3204 		if (ipst->ips_vifs[vifi].v_lcl_addr.s_addr == 0)
3205 			continue;
3206 		/*
3207 		 * No locks here, an approximation is fine.
3208 		 */
3209 		vi.vifc_vifi = vifi;
3210 		vi.vifc_flags = ipst->ips_vifs[vifi].v_flags;
3211 		vi.vifc_threshold = ipst->ips_vifs[vifi].v_threshold;
3212 		vi.vifc_rate_limit	= ipst->ips_vifs[vifi].v_rate_limit;
3213 		vi.vifc_lcl_addr	= ipst->ips_vifs[vifi].v_lcl_addr;
3214 		vi.vifc_rmt_addr	= ipst->ips_vifs[vifi].v_rmt_addr;
3215 		vi.vifc_pkt_in		= ipst->ips_vifs[vifi].v_pkt_in;
3216 		vi.vifc_pkt_out		= ipst->ips_vifs[vifi].v_pkt_out;
3217 
3218 		if (!snmp_append_data(mp, (char *)&vi, sizeof (vi))) {
3219 			ip0dbg(("ip_mroute_vif: failed %ld bytes\n",
3220 			    (size_t)sizeof (vi)));
3221 			mutex_exit(&ipst->ips_numvifs_mutex);
3222 			return (0);
3223 		}
3224 	}
3225 	mutex_exit(&ipst->ips_numvifs_mutex);
3226 	return (1);
3227 }
3228 
3229 /*
3230  * Called by ip_snmp_get to send up multicast routing table.
3231  */
3232 int
3233 ip_mroute_mrt(mblk_t *mp, ip_stack_t *ipst)
3234 {
3235 	int			i, j;
3236 	struct mfc		*rt;
3237 	struct mfcctl	mfcc;
3238 
3239 	/*
3240 	 * Make sure multicast has not been turned off.
3241 	 */
3242 	if (is_mrouter_off(ipst))
3243 		return (1);
3244 
3245 	/* Loop over all hash buckets and their chains */
3246 	for (i = 0; i < MFCTBLSIZ; i++) {
3247 		MFCB_REFHOLD(&ipst->ips_mfcs[i]);
3248 		for (rt = ipst->ips_mfcs[i].mfcb_mfc; rt; rt = rt->mfc_next) {
3249 			mutex_enter(&rt->mfc_mutex);
3250 			if (rt->mfc_rte != NULL ||
3251 			    (rt->mfc_marks & MFCB_MARK_CONDEMNED)) {
3252 				mutex_exit(&rt->mfc_mutex);
3253 				continue;
3254 			}
3255 			mfcc.mfcc_origin = rt->mfc_origin;
3256 			mfcc.mfcc_mcastgrp = rt->mfc_mcastgrp;
3257 			mfcc.mfcc_parent = rt->mfc_parent;
3258 			mfcc.mfcc_pkt_cnt = rt->mfc_pkt_cnt;
3259 			mutex_enter(&ipst->ips_numvifs_mutex);
3260 			for (j = 0; j < (int)ipst->ips_numvifs; j++)
3261 				mfcc.mfcc_ttls[j] = rt->mfc_ttls[j];
3262 			for (j = (int)ipst->ips_numvifs; j < MAXVIFS; j++)
3263 				mfcc.mfcc_ttls[j] = 0;
3264 			mutex_exit(&ipst->ips_numvifs_mutex);
3265 
3266 			mutex_exit(&rt->mfc_mutex);
3267 			if (!snmp_append_data(mp, (char *)&mfcc,
3268 			    sizeof (mfcc))) {
3269 				MFCB_REFRELE(&ipst->ips_mfcs[i]);
3270 				ip0dbg(("ip_mroute_mrt: failed %ld bytes\n",
3271 				    (size_t)sizeof (mfcc)));
3272 				return (0);
3273 			}
3274 		}
3275 		MFCB_REFRELE(&ipst->ips_mfcs[i]);
3276 	}
3277 	return (1);
3278 }
3279