xref: /netbsd-src/sys/netinet/ip_mroute.c (revision 2c6fc41c810f5088457889d00eba558e8bc74d9e)
1 /*	$NetBSD: ip_mroute.c,v 1.129 2014/05/18 14:46:16 rmind Exp $	*/
2 
3 /*
4  * Copyright (c) 1992, 1993
5  *      The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Stephen Deering of Stanford University.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *      @(#)ip_mroute.c 8.2 (Berkeley) 11/15/93
35  */
36 
37 /*
38  * Copyright (c) 1989 Stephen Deering
39  *
40  * This code is derived from software contributed to Berkeley by
41  * Stephen Deering of Stanford University.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *      This product includes software developed by the University of
54  *      California, Berkeley and its contributors.
55  * 4. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  *
71  *      @(#)ip_mroute.c 8.2 (Berkeley) 11/15/93
72  */
73 
74 /*
75  * IP multicast forwarding procedures
76  *
77  * Written by David Waitzman, BBN Labs, August 1988.
78  * Modified by Steve Deering, Stanford, February 1989.
79  * Modified by Mark J. Steiglitz, Stanford, May, 1991
80  * Modified by Van Jacobson, LBL, January 1993
81  * Modified by Ajit Thyagarajan, PARC, August 1993
82  * Modified by Bill Fenner, PARC, April 1994
83  * Modified by Charles M. Hannum, NetBSD, May 1995.
84  * Modified by Ahmed Helmy, SGI, June 1996
85  * Modified by George Edmond Eddy (Rusty), ISI, February 1998
86  * Modified by Pavlin Radoslavov, USC/ISI, May 1998, August 1999, October 2000
87  * Modified by Hitoshi Asaeda, WIDE, August 2000
88  * Modified by Pavlin Radoslavov, ICSI, October 2002
89  *
90  * MROUTING Revision: 1.2
91  * and PIM-SMv2 and PIM-DM support, advanced API support,
92  * bandwidth metering and signaling
93  */
94 
95 #include <sys/cdefs.h>
96 __KERNEL_RCSID(0, "$NetBSD: ip_mroute.c,v 1.129 2014/05/18 14:46:16 rmind Exp $");
97 
98 #include "opt_inet.h"
99 #include "opt_ipsec.h"
100 #include "opt_pim.h"
101 
102 #ifdef PIM
103 #define _PIM_VT 1
104 #endif
105 
106 #include <sys/param.h>
107 #include <sys/systm.h>
108 #include <sys/callout.h>
109 #include <sys/mbuf.h>
110 #include <sys/socket.h>
111 #include <sys/socketvar.h>
112 #include <sys/protosw.h>
113 #include <sys/errno.h>
114 #include <sys/time.h>
115 #include <sys/kernel.h>
116 #include <sys/kmem.h>
117 #include <sys/ioctl.h>
118 #include <sys/syslog.h>
119 
120 #include <net/if.h>
121 #include <net/route.h>
122 #include <net/raw_cb.h>
123 
124 #include <netinet/in.h>
125 #include <netinet/in_var.h>
126 #include <netinet/in_systm.h>
127 #include <netinet/ip.h>
128 #include <netinet/ip_var.h>
129 #include <netinet/in_pcb.h>
130 #include <netinet/udp.h>
131 #include <netinet/igmp.h>
132 #include <netinet/igmp_var.h>
133 #include <netinet/ip_mroute.h>
134 #ifdef PIM
135 #include <netinet/pim.h>
136 #include <netinet/pim_var.h>
137 #endif
138 #include <netinet/ip_encap.h>
139 
140 #ifdef IPSEC
141 #include <netipsec/ipsec.h>
142 #include <netipsec/key.h>
143 #endif
144 
145 #define IP_MULTICASTOPTS 0
146 #define	M_PULLUP(m, len)						 \
147 	do {								 \
148 		if ((m) && ((m)->m_flags & M_EXT || (m)->m_len < (len))) \
149 			(m) = m_pullup((m), (len));			 \
150 	} while (/*CONSTCOND*/ 0)
151 
152 /*
153  * Globals.  All but ip_mrouter and ip_mrtproto could be static,
154  * except for netstat or debugging purposes.
155  */
156 struct socket  *ip_mrouter  = NULL;
157 int		ip_mrtproto = IGMP_DVMRP;    /* for netstat only */
158 
159 #define NO_RTE_FOUND 	0x1
160 #define RTE_FOUND	0x2
161 
162 #define	MFCHASH(a, g)							\
163 	((((a).s_addr >> 20) ^ ((a).s_addr >> 10) ^ (a).s_addr ^	\
164 	  ((g).s_addr >> 20) ^ ((g).s_addr >> 10) ^ (g).s_addr) & mfchash)
165 LIST_HEAD(mfchashhdr, mfc) *mfchashtbl;
166 u_long	mfchash;
167 
168 u_char		nexpire[MFCTBLSIZ];
169 struct vif	viftable[MAXVIFS];
170 struct mrtstat	mrtstat;
171 u_int		mrtdebug = 0;	  /* debug level 	*/
172 #define		DEBUG_MFC	0x02
173 #define		DEBUG_FORWARD	0x04
174 #define		DEBUG_EXPIRE	0x08
175 #define		DEBUG_XMIT	0x10
176 #define		DEBUG_PIM	0x20
177 
178 #define		VIFI_INVALID	((vifi_t) -1)
179 
180 u_int       	tbfdebug = 0;     /* tbf debug level 	*/
181 #ifdef RSVP_ISI
182 u_int		rsvpdebug = 0;	  /* rsvp debug level   */
183 extern struct socket *ip_rsvpd;
184 extern int rsvp_on;
185 #endif /* RSVP_ISI */
186 
187 /* vif attachment using sys/netinet/ip_encap.c */
188 static void vif_input(struct mbuf *, ...);
189 static int vif_encapcheck(struct mbuf *, int, int, void *);
190 
191 static const struct protosw vif_protosw = {
192 	.pr_type	= SOCK_RAW,
193 	.pr_domain	= &inetdomain,
194 	.pr_protocol	= IPPROTO_IPV4,
195 	.pr_flags	= PR_ATOMIC|PR_ADDR,
196 	.pr_input	= vif_input,
197 	.pr_output	= rip_output,
198 	.pr_ctloutput	= rip_ctloutput,
199 	.pr_usrreqs	= &rip_usrreqs,
200 };
201 
202 #define		EXPIRE_TIMEOUT	(hz / 4)	/* 4x / second */
203 #define		UPCALL_EXPIRE	6		/* number of timeouts */
204 
205 /*
206  * Define the token bucket filter structures
207  */
208 
209 #define		TBF_REPROCESS	(hz / 100)	/* 100x / second */
210 
211 static int get_sg_cnt(struct sioc_sg_req *);
212 static int get_vif_cnt(struct sioc_vif_req *);
213 static int ip_mrouter_init(struct socket *, int);
214 static int set_assert(int);
215 static int add_vif(struct vifctl *);
216 static int del_vif(vifi_t *);
217 static void update_mfc_params(struct mfc *, struct mfcctl2 *);
218 static void init_mfc_params(struct mfc *, struct mfcctl2 *);
219 static void expire_mfc(struct mfc *);
220 static int add_mfc(struct sockopt *);
221 #ifdef UPCALL_TIMING
222 static void collate(struct timeval *);
223 #endif
224 static int del_mfc(struct sockopt *);
225 static int set_api_config(struct sockopt *); /* chose API capabilities */
226 static int socket_send(struct socket *, struct mbuf *, struct sockaddr_in *);
227 static void expire_upcalls(void *);
228 #ifdef RSVP_ISI
229 static int ip_mdq(struct mbuf *, struct ifnet *, struct mfc *, vifi_t);
230 #else
231 static int ip_mdq(struct mbuf *, struct ifnet *, struct mfc *);
232 #endif
233 static void phyint_send(struct ip *, struct vif *, struct mbuf *);
234 static void encap_send(struct ip *, struct vif *, struct mbuf *);
235 static void tbf_control(struct vif *, struct mbuf *, struct ip *, u_int32_t);
236 static void tbf_queue(struct vif *, struct mbuf *);
237 static void tbf_process_q(struct vif *);
238 static void tbf_reprocess_q(void *);
239 static int tbf_dq_sel(struct vif *, struct ip *);
240 static void tbf_send_packet(struct vif *, struct mbuf *);
241 static void tbf_update_tokens(struct vif *);
242 static int priority(struct vif *, struct ip *);
243 
244 /*
245  * Bandwidth monitoring
246  */
247 static void free_bw_list(struct bw_meter *);
248 static int add_bw_upcall(struct bw_upcall *);
249 static int del_bw_upcall(struct bw_upcall *);
250 static void bw_meter_receive_packet(struct bw_meter *, int , struct timeval *);
251 static void bw_meter_prepare_upcall(struct bw_meter *, struct timeval *);
252 static void bw_upcalls_send(void);
253 static void schedule_bw_meter(struct bw_meter *, struct timeval *);
254 static void unschedule_bw_meter(struct bw_meter *);
255 static void bw_meter_process(void);
256 static void expire_bw_upcalls_send(void *);
257 static void expire_bw_meter_process(void *);
258 
259 #ifdef PIM
260 static int pim_register_send(struct ip *, struct vif *,
261 		struct mbuf *, struct mfc *);
262 static int pim_register_send_rp(struct ip *, struct vif *,
263 		struct mbuf *, struct mfc *);
264 static int pim_register_send_upcall(struct ip *, struct vif *,
265 		struct mbuf *, struct mfc *);
266 static struct mbuf *pim_register_prepare(struct ip *, struct mbuf *);
267 #endif
268 
269 /*
270  * 'Interfaces' associated with decapsulator (so we can tell
271  * packets that went through it from ones that get reflected
272  * by a broken gateway).  These interfaces are never linked into
273  * the system ifnet list & no routes point to them.  I.e., packets
274  * can't be sent this way.  They only exist as a placeholder for
275  * multicast source verification.
276  */
277 #if 0
278 struct ifnet multicast_decap_if[MAXVIFS];
279 #endif
280 
281 #define	ENCAP_TTL	64
282 #define	ENCAP_PROTO	IPPROTO_IPIP	/* 4 */
283 
284 /* prototype IP hdr for encapsulated packets */
285 struct ip multicast_encap_iphdr = {
286 	.ip_hl = sizeof(struct ip) >> 2,
287 	.ip_v = IPVERSION,
288 	.ip_len = sizeof(struct ip),
289 	.ip_ttl = ENCAP_TTL,
290 	.ip_p = ENCAP_PROTO,
291 };
292 
293 /*
294  * Bandwidth meter variables and constants
295  */
296 
297 /*
298  * Pending timeouts are stored in a hash table, the key being the
299  * expiration time. Periodically, the entries are analysed and processed.
300  */
301 #define BW_METER_BUCKETS	1024
302 static struct bw_meter *bw_meter_timers[BW_METER_BUCKETS];
303 struct callout bw_meter_ch;
304 #define BW_METER_PERIOD (hz)		/* periodical handling of bw meters */
305 
306 /*
307  * Pending upcalls are stored in a vector which is flushed when
308  * full, or periodically
309  */
310 static struct bw_upcall	bw_upcalls[BW_UPCALLS_MAX];
311 static u_int	bw_upcalls_n; /* # of pending upcalls */
312 struct callout	bw_upcalls_ch;
313 #define BW_UPCALLS_PERIOD (hz)		/* periodical flush of bw upcalls */
314 
315 #ifdef PIM
316 struct pimstat pimstat;
317 
318 /*
319  * Note: the PIM Register encapsulation adds the following in front of a
320  * data packet:
321  *
322  * struct pim_encap_hdr {
323  *    struct ip ip;
324  *    struct pim_encap_pimhdr  pim;
325  * }
326  *
327  */
328 
329 struct pim_encap_pimhdr {
330 	struct pim pim;
331 	uint32_t   flags;
332 };
333 
334 static struct ip pim_encap_iphdr = {
335 	.ip_v = IPVERSION,
336 	.ip_hl = sizeof(struct ip) >> 2,
337 	.ip_len = sizeof(struct ip),
338 	.ip_ttl = ENCAP_TTL,
339 	.ip_p = IPPROTO_PIM,
340 };
341 
342 static struct pim_encap_pimhdr pim_encap_pimhdr = {
343     {
344 	PIM_MAKE_VT(PIM_VERSION, PIM_REGISTER), /* PIM vers and message type */
345 	0,			/* reserved */
346 	0,			/* checksum */
347     },
348     0				/* flags */
349 };
350 
351 static struct ifnet multicast_register_if;
352 static vifi_t reg_vif_num = VIFI_INVALID;
353 #endif /* PIM */
354 
355 
356 /*
357  * Private variables.
358  */
359 static vifi_t	   numvifs = 0;
360 
361 static struct callout expire_upcalls_ch;
362 
363 /*
364  * whether or not special PIM assert processing is enabled.
365  */
366 static int pim_assert;
367 /*
368  * Rate limit for assert notification messages, in usec
369  */
370 #define ASSERT_MSG_TIME		3000000
371 
372 /*
373  * Kernel multicast routing API capabilities and setup.
374  * If more API capabilities are added to the kernel, they should be
375  * recorded in `mrt_api_support'.
376  */
377 static const u_int32_t mrt_api_support = (MRT_MFC_FLAGS_DISABLE_WRONGVIF |
378 					  MRT_MFC_FLAGS_BORDER_VIF |
379 					  MRT_MFC_RP |
380 					  MRT_MFC_BW_UPCALL);
381 static u_int32_t mrt_api_config = 0;
382 
383 /*
384  * Find a route for a given origin IP address and Multicast group address
385  * Type of service parameter to be added in the future!!!
386  * Statistics are updated by the caller if needed
387  * (mrtstat.mrts_mfc_lookups and mrtstat.mrts_mfc_misses)
388  */
389 static struct mfc *
390 mfc_find(struct in_addr *o, struct in_addr *g)
391 {
392 	struct mfc *rt;
393 
394 	LIST_FOREACH(rt, &mfchashtbl[MFCHASH(*o, *g)], mfc_hash) {
395 		if (in_hosteq(rt->mfc_origin, *o) &&
396 		    in_hosteq(rt->mfc_mcastgrp, *g) &&
397 		    (rt->mfc_stall == NULL))
398 			break;
399 	}
400 
401 	return (rt);
402 }
403 
404 /*
405  * Macros to compute elapsed time efficiently
406  * Borrowed from Van Jacobson's scheduling code
407  */
408 #define TV_DELTA(a, b, delta) do {					\
409 	int xxs;							\
410 	delta = (a).tv_usec - (b).tv_usec;				\
411 	xxs = (a).tv_sec - (b).tv_sec;					\
412 	switch (xxs) {							\
413 	case 2:								\
414 		delta += 1000000;					\
415 		/* fall through */					\
416 	case 1:								\
417 		delta += 1000000;					\
418 		/* fall through */					\
419 	case 0:								\
420 		break;							\
421 	default:							\
422 		delta += (1000000 * xxs);				\
423 		break;							\
424 	}								\
425 } while (/*CONSTCOND*/ 0)
426 
427 #ifdef UPCALL_TIMING
428 u_int32_t upcall_data[51];
429 #endif /* UPCALL_TIMING */
430 
431 /*
432  * Handle MRT setsockopt commands to modify the multicast routing tables.
433  */
434 int
435 ip_mrouter_set(struct socket *so, struct sockopt *sopt)
436 {
437 	int error;
438 	int optval;
439 	struct vifctl vifc;
440 	vifi_t vifi;
441 	struct bw_upcall bwuc;
442 
443 	if (sopt->sopt_name != MRT_INIT && so != ip_mrouter)
444 		error = ENOPROTOOPT;
445 	else {
446 		switch (sopt->sopt_name) {
447 		case MRT_INIT:
448 			error = sockopt_getint(sopt, &optval);
449 			if (error)
450 				break;
451 
452 			error = ip_mrouter_init(so, optval);
453 			break;
454 		case MRT_DONE:
455 			error = ip_mrouter_done();
456 			break;
457 		case MRT_ADD_VIF:
458 			error = sockopt_get(sopt, &vifc, sizeof(vifc));
459 			if (error)
460 				break;
461 			error = add_vif(&vifc);
462 			break;
463 		case MRT_DEL_VIF:
464 			error = sockopt_get(sopt, &vifi, sizeof(vifi));
465 			if (error)
466 				break;
467 			error = del_vif(&vifi);
468 			break;
469 		case MRT_ADD_MFC:
470 			error = add_mfc(sopt);
471 			break;
472 		case MRT_DEL_MFC:
473 			error = del_mfc(sopt);
474 			break;
475 		case MRT_ASSERT:
476 			error = sockopt_getint(sopt, &optval);
477 			if (error)
478 				break;
479 			error = set_assert(optval);
480 			break;
481 		case MRT_API_CONFIG:
482 			error = set_api_config(sopt);
483 			break;
484 		case MRT_ADD_BW_UPCALL:
485 			error = sockopt_get(sopt, &bwuc, sizeof(bwuc));
486 			if (error)
487 				break;
488 			error = add_bw_upcall(&bwuc);
489 			break;
490 		case MRT_DEL_BW_UPCALL:
491 			error = sockopt_get(sopt, &bwuc, sizeof(bwuc));
492 			if (error)
493 				break;
494 			error = del_bw_upcall(&bwuc);
495 			break;
496 		default:
497 			error = ENOPROTOOPT;
498 			break;
499 		}
500 	}
501 	return (error);
502 }
503 
504 /*
505  * Handle MRT getsockopt commands
506  */
507 int
508 ip_mrouter_get(struct socket *so, struct sockopt *sopt)
509 {
510 	int error;
511 
512 	if (so != ip_mrouter)
513 		error = ENOPROTOOPT;
514 	else {
515 		switch (sopt->sopt_name) {
516 		case MRT_VERSION:
517 			error = sockopt_setint(sopt, 0x0305); /* XXX !!!! */
518 			break;
519 		case MRT_ASSERT:
520 			error = sockopt_setint(sopt, pim_assert);
521 			break;
522 		case MRT_API_SUPPORT:
523 			error = sockopt_set(sopt, &mrt_api_support,
524 			    sizeof(mrt_api_support));
525 			break;
526 		case MRT_API_CONFIG:
527 			error = sockopt_set(sopt, &mrt_api_config,
528 			    sizeof(mrt_api_config));
529 			break;
530 		default:
531 			error = ENOPROTOOPT;
532 			break;
533 		}
534 	}
535 	return (error);
536 }
537 
538 /*
539  * Handle ioctl commands to obtain information from the cache
540  */
541 int
542 mrt_ioctl(struct socket *so, u_long cmd, void *data)
543 {
544 	int error;
545 
546 	if (so != ip_mrouter)
547 		error = EINVAL;
548 	else
549 		switch (cmd) {
550 		case SIOCGETVIFCNT:
551 			error = get_vif_cnt((struct sioc_vif_req *)data);
552 			break;
553 		case SIOCGETSGCNT:
554 			error = get_sg_cnt((struct sioc_sg_req *)data);
555 			break;
556 		default:
557 			error = EINVAL;
558 			break;
559 		}
560 
561 	return (error);
562 }
563 
564 /*
565  * returns the packet, byte, rpf-failure count for the source group provided
566  */
567 static int
568 get_sg_cnt(struct sioc_sg_req *req)
569 {
570 	int s;
571 	struct mfc *rt;
572 
573 	s = splsoftnet();
574 	rt = mfc_find(&req->src, &req->grp);
575 	if (rt == NULL) {
576 		splx(s);
577 		req->pktcnt = req->bytecnt = req->wrong_if = 0xffffffff;
578 		return (EADDRNOTAVAIL);
579 	}
580 	req->pktcnt = rt->mfc_pkt_cnt;
581 	req->bytecnt = rt->mfc_byte_cnt;
582 	req->wrong_if = rt->mfc_wrong_if;
583 	splx(s);
584 
585 	return (0);
586 }
587 
588 /*
589  * returns the input and output packet and byte counts on the vif provided
590  */
591 static int
592 get_vif_cnt(struct sioc_vif_req *req)
593 {
594 	vifi_t vifi = req->vifi;
595 
596 	if (vifi >= numvifs)
597 		return (EINVAL);
598 
599 	req->icount = viftable[vifi].v_pkt_in;
600 	req->ocount = viftable[vifi].v_pkt_out;
601 	req->ibytes = viftable[vifi].v_bytes_in;
602 	req->obytes = viftable[vifi].v_bytes_out;
603 
604 	return (0);
605 }
606 
607 /*
608  * Enable multicast routing
609  */
610 static int
611 ip_mrouter_init(struct socket *so, int v)
612 {
613 	if (mrtdebug)
614 		log(LOG_DEBUG,
615 		    "ip_mrouter_init: so_type = %d, pr_protocol = %d\n",
616 		    so->so_type, so->so_proto->pr_protocol);
617 
618 	if (so->so_type != SOCK_RAW ||
619 	    so->so_proto->pr_protocol != IPPROTO_IGMP)
620 		return (EOPNOTSUPP);
621 
622 	if (v != 1)
623 		return (EINVAL);
624 
625 	if (ip_mrouter != NULL)
626 		return (EADDRINUSE);
627 
628 	ip_mrouter = so;
629 
630 	mfchashtbl = hashinit(MFCTBLSIZ, HASH_LIST, true, &mfchash);
631 	memset((void *)nexpire, 0, sizeof(nexpire));
632 
633 	pim_assert = 0;
634 
635 	callout_init(&expire_upcalls_ch, 0);
636 	callout_reset(&expire_upcalls_ch, EXPIRE_TIMEOUT,
637 		      expire_upcalls, NULL);
638 
639 	callout_init(&bw_upcalls_ch, 0);
640 	callout_reset(&bw_upcalls_ch, BW_UPCALLS_PERIOD,
641 		      expire_bw_upcalls_send, NULL);
642 
643 	callout_init(&bw_meter_ch, 0);
644 	callout_reset(&bw_meter_ch, BW_METER_PERIOD,
645 		      expire_bw_meter_process, NULL);
646 
647 	if (mrtdebug)
648 		log(LOG_DEBUG, "ip_mrouter_init\n");
649 
650 	return (0);
651 }
652 
653 /*
654  * Disable multicast routing
655  */
656 int
657 ip_mrouter_done(void)
658 {
659 	vifi_t vifi;
660 	struct vif *vifp;
661 	int i;
662 	int s;
663 
664 	s = splsoftnet();
665 
666 	/* Clear out all the vifs currently in use. */
667 	for (vifi = 0; vifi < numvifs; vifi++) {
668 		vifp = &viftable[vifi];
669 		if (!in_nullhost(vifp->v_lcl_addr))
670 			reset_vif(vifp);
671 	}
672 
673 	numvifs = 0;
674 	pim_assert = 0;
675 	mrt_api_config = 0;
676 
677 	callout_stop(&expire_upcalls_ch);
678 	callout_stop(&bw_upcalls_ch);
679 	callout_stop(&bw_meter_ch);
680 
681 	/*
682 	 * Free all multicast forwarding cache entries.
683 	 */
684 	for (i = 0; i < MFCTBLSIZ; i++) {
685 		struct mfc *rt, *nrt;
686 
687 		for (rt = LIST_FIRST(&mfchashtbl[i]); rt; rt = nrt) {
688 			nrt = LIST_NEXT(rt, mfc_hash);
689 
690 			expire_mfc(rt);
691 		}
692 	}
693 
694 	memset((void *)nexpire, 0, sizeof(nexpire));
695 	hashdone(mfchashtbl, HASH_LIST, mfchash);
696 	mfchashtbl = NULL;
697 
698 	bw_upcalls_n = 0;
699 	memset(bw_meter_timers, 0, sizeof(bw_meter_timers));
700 
701 	/* Reset de-encapsulation cache. */
702 
703 	ip_mrouter = NULL;
704 
705 	splx(s);
706 
707 	if (mrtdebug)
708 		log(LOG_DEBUG, "ip_mrouter_done\n");
709 
710 	return (0);
711 }
712 
713 void
714 ip_mrouter_detach(struct ifnet *ifp)
715 {
716 	int vifi, i;
717 	struct vif *vifp;
718 	struct mfc *rt;
719 	struct rtdetq *rte;
720 
721 	/* XXX not sure about side effect to userland routing daemon */
722 	for (vifi = 0; vifi < numvifs; vifi++) {
723 		vifp = &viftable[vifi];
724 		if (vifp->v_ifp == ifp)
725 			reset_vif(vifp);
726 	}
727 	for (i = 0; i < MFCTBLSIZ; i++) {
728 		if (nexpire[i] == 0)
729 			continue;
730 		LIST_FOREACH(rt, &mfchashtbl[i], mfc_hash) {
731 			for (rte = rt->mfc_stall; rte; rte = rte->next) {
732 				if (rte->ifp == ifp)
733 					rte->ifp = NULL;
734 			}
735 		}
736 	}
737 }
738 
739 /*
740  * Set PIM assert processing global
741  */
742 static int
743 set_assert(int i)
744 {
745 	pim_assert = !!i;
746 	return (0);
747 }
748 
749 /*
750  * Configure API capabilities
751  */
752 static int
753 set_api_config(struct sockopt *sopt)
754 {
755 	u_int32_t apival;
756 	int i, error;
757 
758 	/*
759 	 * We can set the API capabilities only if it is the first operation
760 	 * after MRT_INIT. I.e.:
761 	 *  - there are no vifs installed
762 	 *  - pim_assert is not enabled
763 	 *  - the MFC table is empty
764 	 */
765 	error = sockopt_get(sopt, &apival, sizeof(apival));
766 	if (error)
767 		return (error);
768 	if (numvifs > 0)
769 		return (EPERM);
770 	if (pim_assert)
771 		return (EPERM);
772 	for (i = 0; i < MFCTBLSIZ; i++) {
773 		if (LIST_FIRST(&mfchashtbl[i]) != NULL)
774 			return (EPERM);
775 	}
776 
777 	mrt_api_config = apival & mrt_api_support;
778 	return (0);
779 }
780 
781 /*
782  * Add a vif to the vif table
783  */
784 static int
785 add_vif(struct vifctl *vifcp)
786 {
787 	struct vif *vifp;
788 	struct ifaddr *ifa;
789 	struct ifnet *ifp;
790 	int error, s;
791 	struct sockaddr_in sin;
792 
793 	if (vifcp->vifc_vifi >= MAXVIFS)
794 		return (EINVAL);
795 	if (in_nullhost(vifcp->vifc_lcl_addr))
796 		return (EADDRNOTAVAIL);
797 
798 	vifp = &viftable[vifcp->vifc_vifi];
799 	if (!in_nullhost(vifp->v_lcl_addr))
800 		return (EADDRINUSE);
801 
802 	/* Find the interface with an address in AF_INET family. */
803 #ifdef PIM
804 	if (vifcp->vifc_flags & VIFF_REGISTER) {
805 		/*
806 		 * XXX: Because VIFF_REGISTER does not really need a valid
807 		 * local interface (e.g. it could be 127.0.0.2), we don't
808 		 * check its address.
809 		 */
810 	    ifp = NULL;
811 	} else
812 #endif
813 	{
814 		sockaddr_in_init(&sin, &vifcp->vifc_lcl_addr, 0);
815 		ifa = ifa_ifwithaddr(sintosa(&sin));
816 		if (ifa == NULL)
817 			return (EADDRNOTAVAIL);
818 		ifp = ifa->ifa_ifp;
819 	}
820 
821 	if (vifcp->vifc_flags & VIFF_TUNNEL) {
822 		if (vifcp->vifc_flags & VIFF_SRCRT) {
823 			log(LOG_ERR, "source routed tunnels not supported\n");
824 			return (EOPNOTSUPP);
825 		}
826 
827 		/* attach this vif to decapsulator dispatch table */
828 		/*
829 		 * XXX Use addresses in registration so that matching
830 		 * can be done with radix tree in decapsulator.  But,
831 		 * we need to check inner header for multicast, so
832 		 * this requires both radix tree lookup and then a
833 		 * function to check, and this is not supported yet.
834 		 */
835 		vifp->v_encap_cookie = encap_attach_func(AF_INET, IPPROTO_IPV4,
836 		    vif_encapcheck, &vif_protosw, vifp);
837 		if (!vifp->v_encap_cookie)
838 			return (EINVAL);
839 
840 		/* Create a fake encapsulation interface. */
841 		ifp = malloc(sizeof(*ifp), M_MRTABLE, M_WAITOK|M_ZERO);
842 		snprintf(ifp->if_xname, sizeof(ifp->if_xname),
843 			 "mdecap%d", vifcp->vifc_vifi);
844 
845 		/* Prepare cached route entry. */
846 		memset(&vifp->v_route, 0, sizeof(vifp->v_route));
847 #ifdef PIM
848 	} else if (vifcp->vifc_flags & VIFF_REGISTER) {
849 		ifp = &multicast_register_if;
850 		if (mrtdebug)
851 			log(LOG_DEBUG, "Adding a register vif, ifp: %p\n",
852 			    (void *)ifp);
853 		if (reg_vif_num == VIFI_INVALID) {
854 			memset(ifp, 0, sizeof(*ifp));
855 			snprintf(ifp->if_xname, sizeof(ifp->if_xname),
856 				 "register_vif");
857 			ifp->if_flags = IFF_LOOPBACK;
858 			memset(&vifp->v_route, 0, sizeof(vifp->v_route));
859 			reg_vif_num = vifcp->vifc_vifi;
860 		}
861 #endif
862 	} else {
863 		/* Make sure the interface supports multicast. */
864 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
865 			return (EOPNOTSUPP);
866 
867 		/* Enable promiscuous reception of all IP multicasts. */
868 		sockaddr_in_init(&sin, &zeroin_addr, 0);
869 		error = if_mcast_op(ifp, SIOCADDMULTI, sintosa(&sin));
870 		if (error)
871 			return (error);
872 	}
873 
874 	s = splsoftnet();
875 
876 	/* Define parameters for the tbf structure. */
877 	vifp->tbf_q = NULL;
878 	vifp->tbf_t = &vifp->tbf_q;
879 	microtime(&vifp->tbf_last_pkt_t);
880 	vifp->tbf_n_tok = 0;
881 	vifp->tbf_q_len = 0;
882 	vifp->tbf_max_q_len = MAXQSIZE;
883 
884 	vifp->v_flags = vifcp->vifc_flags;
885 	vifp->v_threshold = vifcp->vifc_threshold;
886 	/* scaling up here allows division by 1024 in critical code */
887 	vifp->v_rate_limit = vifcp->vifc_rate_limit * 1024 / 1000;
888 	vifp->v_lcl_addr = vifcp->vifc_lcl_addr;
889 	vifp->v_rmt_addr = vifcp->vifc_rmt_addr;
890 	vifp->v_ifp = ifp;
891 	/* Initialize per vif pkt counters. */
892 	vifp->v_pkt_in = 0;
893 	vifp->v_pkt_out = 0;
894 	vifp->v_bytes_in = 0;
895 	vifp->v_bytes_out = 0;
896 
897 	callout_init(&vifp->v_repq_ch, 0);
898 
899 #ifdef RSVP_ISI
900 	vifp->v_rsvp_on = 0;
901 	vifp->v_rsvpd = NULL;
902 #endif /* RSVP_ISI */
903 
904 	splx(s);
905 
906 	/* Adjust numvifs up if the vifi is higher than numvifs. */
907 	if (numvifs <= vifcp->vifc_vifi)
908 		numvifs = vifcp->vifc_vifi + 1;
909 
910 	if (mrtdebug)
911 		log(LOG_DEBUG, "add_vif #%d, lcladdr %x, %s %x, thresh %x, rate %d\n",
912 		    vifcp->vifc_vifi,
913 		    ntohl(vifcp->vifc_lcl_addr.s_addr),
914 		    (vifcp->vifc_flags & VIFF_TUNNEL) ? "rmtaddr" : "mask",
915 		    ntohl(vifcp->vifc_rmt_addr.s_addr),
916 		    vifcp->vifc_threshold,
917 		    vifcp->vifc_rate_limit);
918 
919 	return (0);
920 }
921 
922 void
923 reset_vif(struct vif *vifp)
924 {
925 	struct mbuf *m, *n;
926 	struct ifnet *ifp;
927 	struct sockaddr_in sin;
928 
929 	callout_stop(&vifp->v_repq_ch);
930 
931 	/* detach this vif from decapsulator dispatch table */
932 	encap_detach(vifp->v_encap_cookie);
933 	vifp->v_encap_cookie = NULL;
934 
935 	/*
936 	 * Free packets queued at the interface
937 	 */
938 	for (m = vifp->tbf_q; m != NULL; m = n) {
939 		n = m->m_nextpkt;
940 		m_freem(m);
941 	}
942 
943 	if (vifp->v_flags & VIFF_TUNNEL)
944 		free(vifp->v_ifp, M_MRTABLE);
945 	else if (vifp->v_flags & VIFF_REGISTER) {
946 #ifdef PIM
947 		reg_vif_num = VIFI_INVALID;
948 #endif
949 	} else {
950 		sockaddr_in_init(&sin, &zeroin_addr, 0);
951 		ifp = vifp->v_ifp;
952 		if_mcast_op(ifp, SIOCDELMULTI, sintosa(&sin));
953 	}
954 	memset((void *)vifp, 0, sizeof(*vifp));
955 }
956 
957 /*
958  * Delete a vif from the vif table
959  */
960 static int
961 del_vif(vifi_t *vifip)
962 {
963 	struct vif *vifp;
964 	vifi_t vifi;
965 	int s;
966 
967 	if (*vifip >= numvifs)
968 		return (EINVAL);
969 
970 	vifp = &viftable[*vifip];
971 	if (in_nullhost(vifp->v_lcl_addr))
972 		return (EADDRNOTAVAIL);
973 
974 	s = splsoftnet();
975 
976 	reset_vif(vifp);
977 
978 	/* Adjust numvifs down */
979 	for (vifi = numvifs; vifi > 0; vifi--)
980 		if (!in_nullhost(viftable[vifi - 1].v_lcl_addr))
981 			break;
982 	numvifs = vifi;
983 
984 	splx(s);
985 
986 	if (mrtdebug)
987 		log(LOG_DEBUG, "del_vif %d, numvifs %d\n", *vifip, numvifs);
988 
989 	return (0);
990 }
991 
992 /*
993  * update an mfc entry without resetting counters and S,G addresses.
994  */
995 static void
996 update_mfc_params(struct mfc *rt, struct mfcctl2 *mfccp)
997 {
998 	int i;
999 
1000 	rt->mfc_parent = mfccp->mfcc_parent;
1001 	for (i = 0; i < numvifs; i++) {
1002 		rt->mfc_ttls[i] = mfccp->mfcc_ttls[i];
1003 		rt->mfc_flags[i] = mfccp->mfcc_flags[i] & mrt_api_config &
1004 			MRT_MFC_FLAGS_ALL;
1005 	}
1006 	/* set the RP address */
1007 	if (mrt_api_config & MRT_MFC_RP)
1008 		rt->mfc_rp = mfccp->mfcc_rp;
1009 	else
1010 		rt->mfc_rp = zeroin_addr;
1011 }
1012 
1013 /*
1014  * fully initialize an mfc entry from the parameter.
1015  */
1016 static void
1017 init_mfc_params(struct mfc *rt, struct mfcctl2 *mfccp)
1018 {
1019 	rt->mfc_origin     = mfccp->mfcc_origin;
1020 	rt->mfc_mcastgrp   = mfccp->mfcc_mcastgrp;
1021 
1022 	update_mfc_params(rt, mfccp);
1023 
1024 	/* initialize pkt counters per src-grp */
1025 	rt->mfc_pkt_cnt    = 0;
1026 	rt->mfc_byte_cnt   = 0;
1027 	rt->mfc_wrong_if   = 0;
1028 	timerclear(&rt->mfc_last_assert);
1029 }
1030 
1031 static void
1032 expire_mfc(struct mfc *rt)
1033 {
1034 	struct rtdetq *rte, *nrte;
1035 
1036 	free_bw_list(rt->mfc_bw_meter);
1037 
1038 	for (rte = rt->mfc_stall; rte != NULL; rte = nrte) {
1039 		nrte = rte->next;
1040 		m_freem(rte->m);
1041 		free(rte, M_MRTABLE);
1042 	}
1043 
1044 	LIST_REMOVE(rt, mfc_hash);
1045 	free(rt, M_MRTABLE);
1046 }
1047 
1048 /*
1049  * Add an mfc entry
1050  */
1051 static int
1052 add_mfc(struct sockopt *sopt)
1053 {
1054 	struct mfcctl2 mfcctl2;
1055 	struct mfcctl2 *mfccp;
1056 	struct mfc *rt;
1057 	u_int32_t hash = 0;
1058 	struct rtdetq *rte, *nrte;
1059 	u_short nstl;
1060 	int s;
1061 	int error;
1062 
1063 	/*
1064 	 * select data size depending on API version.
1065 	 */
1066 	mfccp = &mfcctl2;
1067 	memset(&mfcctl2, 0, sizeof(mfcctl2));
1068 
1069 	if (mrt_api_config & MRT_API_FLAGS_ALL)
1070 		error = sockopt_get(sopt, mfccp, sizeof(struct mfcctl2));
1071 	else
1072 		error = sockopt_get(sopt, mfccp, sizeof(struct mfcctl));
1073 
1074 	if (error)
1075 		return (error);
1076 
1077 	s = splsoftnet();
1078 	rt = mfc_find(&mfccp->mfcc_origin, &mfccp->mfcc_mcastgrp);
1079 
1080 	/* If an entry already exists, just update the fields */
1081 	if (rt) {
1082 		if (mrtdebug & DEBUG_MFC)
1083 			log(LOG_DEBUG, "add_mfc update o %x g %x p %x\n",
1084 			    ntohl(mfccp->mfcc_origin.s_addr),
1085 			    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1086 			    mfccp->mfcc_parent);
1087 
1088 		update_mfc_params(rt, mfccp);
1089 
1090 		splx(s);
1091 		return (0);
1092 	}
1093 
1094 	/*
1095 	 * Find the entry for which the upcall was made and update
1096 	 */
1097 	nstl = 0;
1098 	hash = MFCHASH(mfccp->mfcc_origin, mfccp->mfcc_mcastgrp);
1099 	LIST_FOREACH(rt, &mfchashtbl[hash], mfc_hash) {
1100 		if (in_hosteq(rt->mfc_origin, mfccp->mfcc_origin) &&
1101 		    in_hosteq(rt->mfc_mcastgrp, mfccp->mfcc_mcastgrp) &&
1102 		    rt->mfc_stall != NULL) {
1103 			if (nstl++)
1104 				log(LOG_ERR, "add_mfc %s o %x g %x p %x dbx %p\n",
1105 				    "multiple kernel entries",
1106 				    ntohl(mfccp->mfcc_origin.s_addr),
1107 				    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1108 				    mfccp->mfcc_parent, rt->mfc_stall);
1109 
1110 			if (mrtdebug & DEBUG_MFC)
1111 				log(LOG_DEBUG, "add_mfc o %x g %x p %x dbg %p\n",
1112 				    ntohl(mfccp->mfcc_origin.s_addr),
1113 				    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1114 				    mfccp->mfcc_parent, rt->mfc_stall);
1115 
1116 			rte = rt->mfc_stall;
1117 			init_mfc_params(rt, mfccp);
1118 			rt->mfc_stall = NULL;
1119 
1120 			rt->mfc_expire = 0; /* Don't clean this guy up */
1121 			nexpire[hash]--;
1122 
1123 			/* free packets Qed at the end of this entry */
1124 			for (; rte != NULL; rte = nrte) {
1125 				nrte = rte->next;
1126 				if (rte->ifp) {
1127 #ifdef RSVP_ISI
1128 					ip_mdq(rte->m, rte->ifp, rt, -1);
1129 #else
1130 					ip_mdq(rte->m, rte->ifp, rt);
1131 #endif /* RSVP_ISI */
1132 				}
1133 				m_freem(rte->m);
1134 #ifdef UPCALL_TIMING
1135 				collate(&rte->t);
1136 #endif /* UPCALL_TIMING */
1137 				free(rte, M_MRTABLE);
1138 			}
1139 		}
1140 	}
1141 
1142 	/*
1143 	 * It is possible that an entry is being inserted without an upcall
1144 	 */
1145 	if (nstl == 0) {
1146 		/*
1147 		 * No mfc; make a new one
1148 		 */
1149 		if (mrtdebug & DEBUG_MFC)
1150 			log(LOG_DEBUG, "add_mfc no upcall o %x g %x p %x\n",
1151 			    ntohl(mfccp->mfcc_origin.s_addr),
1152 			    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1153 			    mfccp->mfcc_parent);
1154 
1155 		LIST_FOREACH(rt, &mfchashtbl[hash], mfc_hash) {
1156 			if (in_hosteq(rt->mfc_origin, mfccp->mfcc_origin) &&
1157 			    in_hosteq(rt->mfc_mcastgrp, mfccp->mfcc_mcastgrp)) {
1158 				init_mfc_params(rt, mfccp);
1159 				if (rt->mfc_expire)
1160 					nexpire[hash]--;
1161 				rt->mfc_expire = 0;
1162 				break; /* XXX */
1163 			}
1164 		}
1165 		if (rt == NULL) {	/* no upcall, so make a new entry */
1166 			rt = (struct mfc *)malloc(sizeof(*rt), M_MRTABLE,
1167 						  M_NOWAIT);
1168 			if (rt == NULL) {
1169 				splx(s);
1170 				return (ENOBUFS);
1171 			}
1172 
1173 			init_mfc_params(rt, mfccp);
1174 			rt->mfc_expire	= 0;
1175 			rt->mfc_stall	= NULL;
1176 			rt->mfc_bw_meter = NULL;
1177 
1178 			/* insert new entry at head of hash chain */
1179 			LIST_INSERT_HEAD(&mfchashtbl[hash], rt, mfc_hash);
1180 		}
1181 	}
1182 
1183 	splx(s);
1184 	return (0);
1185 }
1186 
1187 #ifdef UPCALL_TIMING
1188 /*
1189  * collect delay statistics on the upcalls
1190  */
1191 static void
1192 collate(struct timeval *t)
1193 {
1194 	u_int32_t d;
1195 	struct timeval tp;
1196 	u_int32_t delta;
1197 
1198 	microtime(&tp);
1199 
1200 	if (timercmp(t, &tp, <)) {
1201 		TV_DELTA(tp, *t, delta);
1202 
1203 		d = delta >> 10;
1204 		if (d > 50)
1205 			d = 50;
1206 
1207 		++upcall_data[d];
1208 	}
1209 }
1210 #endif /* UPCALL_TIMING */
1211 
1212 /*
1213  * Delete an mfc entry
1214  */
1215 static int
1216 del_mfc(struct sockopt *sopt)
1217 {
1218 	struct mfcctl2 mfcctl2;
1219 	struct mfcctl2 *mfccp;
1220 	struct mfc *rt;
1221 	int s;
1222 	int error;
1223 
1224 	/*
1225 	 * XXX: for deleting MFC entries the information in entries
1226 	 * of size "struct mfcctl" is sufficient.
1227 	 */
1228 
1229 	mfccp = &mfcctl2;
1230 	memset(&mfcctl2, 0, sizeof(mfcctl2));
1231 
1232 	error = sockopt_get(sopt, mfccp, sizeof(struct mfcctl));
1233 	if (error) {
1234 		/* Try with the size of mfcctl2. */
1235 		error = sockopt_get(sopt, mfccp, sizeof(struct mfcctl2));
1236 		if (error)
1237 			return (error);
1238 	}
1239 
1240 	if (mrtdebug & DEBUG_MFC)
1241 		log(LOG_DEBUG, "del_mfc origin %x mcastgrp %x\n",
1242 		    ntohl(mfccp->mfcc_origin.s_addr),
1243 		    ntohl(mfccp->mfcc_mcastgrp.s_addr));
1244 
1245 	s = splsoftnet();
1246 
1247 	rt = mfc_find(&mfccp->mfcc_origin, &mfccp->mfcc_mcastgrp);
1248 	if (rt == NULL) {
1249 		splx(s);
1250 		return (EADDRNOTAVAIL);
1251 	}
1252 
1253 	/*
1254 	 * free the bw_meter entries
1255 	 */
1256 	free_bw_list(rt->mfc_bw_meter);
1257 	rt->mfc_bw_meter = NULL;
1258 
1259 	LIST_REMOVE(rt, mfc_hash);
1260 	free(rt, M_MRTABLE);
1261 
1262 	splx(s);
1263 	return (0);
1264 }
1265 
1266 static int
1267 socket_send(struct socket *s, struct mbuf *mm, struct sockaddr_in *src)
1268 {
1269 	if (s) {
1270 		if (sbappendaddr(&s->so_rcv, sintosa(src), mm, NULL) != 0) {
1271 			sorwakeup(s);
1272 			return (0);
1273 		}
1274 	}
1275 	m_freem(mm);
1276 	return (-1);
1277 }
1278 
1279 /*
1280  * IP multicast forwarding function. This function assumes that the packet
1281  * pointed to by "ip" has arrived on (or is about to be sent to) the interface
1282  * pointed to by "ifp", and the packet is to be relayed to other networks
1283  * that have members of the packet's destination IP multicast group.
1284  *
1285  * The packet is returned unscathed to the caller, unless it is
1286  * erroneous, in which case a non-zero return value tells the caller to
1287  * discard it.
1288  */
1289 
1290 #define IP_HDR_LEN  20	/* # bytes of fixed IP header (excluding options) */
1291 #define TUNNEL_LEN  12  /* # bytes of IP option for tunnel encapsulation  */
1292 
1293 int
1294 #ifdef RSVP_ISI
1295 ip_mforward(struct mbuf *m, struct ifnet *ifp, struct ip_moptions *imo)
1296 #else
1297 ip_mforward(struct mbuf *m, struct ifnet *ifp)
1298 #endif /* RSVP_ISI */
1299 {
1300 	struct ip *ip = mtod(m, struct ip *);
1301 	struct mfc *rt;
1302 	static int srctun = 0;
1303 	struct mbuf *mm;
1304 	struct sockaddr_in sin;
1305 	int s;
1306 	vifi_t vifi;
1307 
1308 	if (mrtdebug & DEBUG_FORWARD)
1309 		log(LOG_DEBUG, "ip_mforward: src %x, dst %x, ifp %p\n",
1310 		    ntohl(ip->ip_src.s_addr), ntohl(ip->ip_dst.s_addr), ifp);
1311 
1312 	if (ip->ip_hl < (IP_HDR_LEN + TUNNEL_LEN) >> 2 ||
1313 	    ((u_char *)(ip + 1))[1] != IPOPT_LSRR) {
1314 		/*
1315 		 * Packet arrived via a physical interface or
1316 		 * an encapsulated tunnel or a register_vif.
1317 		 */
1318 	} else {
1319 		/*
1320 		 * Packet arrived through a source-route tunnel.
1321 		 * Source-route tunnels are no longer supported.
1322 		 */
1323 		if ((srctun++ % 1000) == 0)
1324 			log(LOG_ERR,
1325 			    "ip_mforward: received source-routed packet from %x\n",
1326 			    ntohl(ip->ip_src.s_addr));
1327 
1328 		return (1);
1329 	}
1330 
1331 	/*
1332 	 * Clear any in-bound checksum flags for this packet.
1333 	 */
1334 	m->m_pkthdr.csum_flags = 0;
1335 
1336 #ifdef RSVP_ISI
1337 	if (imo && ((vifi = imo->imo_multicast_vif) < numvifs)) {
1338 		if (ip->ip_ttl < MAXTTL)
1339 			ip->ip_ttl++;	/* compensate for -1 in *_send routines */
1340 		if (rsvpdebug && ip->ip_p == IPPROTO_RSVP) {
1341 			struct vif *vifp = viftable + vifi;
1342 			printf("Sending IPPROTO_RSVP from %x to %x on vif %d (%s%s)\n",
1343 			    ntohl(ip->ip_src), ntohl(ip->ip_dst), vifi,
1344 			    (vifp->v_flags & VIFF_TUNNEL) ? "tunnel on " : "",
1345 			    vifp->v_ifp->if_xname);
1346 		}
1347 		return (ip_mdq(m, ifp, NULL, vifi));
1348 	}
1349 	if (rsvpdebug && ip->ip_p == IPPROTO_RSVP) {
1350 		printf("Warning: IPPROTO_RSVP from %x to %x without vif option\n",
1351 		    ntohl(ip->ip_src), ntohl(ip->ip_dst));
1352 	}
1353 #endif /* RSVP_ISI */
1354 
1355 	/*
1356 	 * Don't forward a packet with time-to-live of zero or one,
1357 	 * or a packet destined to a local-only group.
1358 	 */
1359 	if (ip->ip_ttl <= 1 || IN_LOCAL_GROUP(ip->ip_dst.s_addr))
1360 		return (0);
1361 
1362 	/*
1363 	 * Determine forwarding vifs from the forwarding cache table
1364 	 */
1365 	s = splsoftnet();
1366 	++mrtstat.mrts_mfc_lookups;
1367 	rt = mfc_find(&ip->ip_src, &ip->ip_dst);
1368 
1369 	/* Entry exists, so forward if necessary */
1370 	if (rt != NULL) {
1371 		splx(s);
1372 #ifdef RSVP_ISI
1373 		return (ip_mdq(m, ifp, rt, -1));
1374 #else
1375 		return (ip_mdq(m, ifp, rt));
1376 #endif /* RSVP_ISI */
1377 	} else {
1378 		/*
1379 		 * If we don't have a route for packet's origin,
1380 		 * Make a copy of the packet & send message to routing daemon
1381 		 */
1382 
1383 		struct mbuf *mb0;
1384 		struct rtdetq *rte;
1385 		u_int32_t hash;
1386 		int hlen = ip->ip_hl << 2;
1387 #ifdef UPCALL_TIMING
1388 		struct timeval tp;
1389 
1390 		microtime(&tp);
1391 #endif /* UPCALL_TIMING */
1392 
1393 		++mrtstat.mrts_mfc_misses;
1394 
1395 		mrtstat.mrts_no_route++;
1396 		if (mrtdebug & (DEBUG_FORWARD | DEBUG_MFC))
1397 			log(LOG_DEBUG, "ip_mforward: no rte s %x g %x\n",
1398 			    ntohl(ip->ip_src.s_addr),
1399 			    ntohl(ip->ip_dst.s_addr));
1400 
1401 		/*
1402 		 * Allocate mbufs early so that we don't do extra work if we are
1403 		 * just going to fail anyway.  Make sure to pullup the header so
1404 		 * that other people can't step on it.
1405 		 */
1406 		rte = (struct rtdetq *)malloc(sizeof(*rte), M_MRTABLE,
1407 					      M_NOWAIT);
1408 		if (rte == NULL) {
1409 			splx(s);
1410 			return (ENOBUFS);
1411 		}
1412 		mb0 = m_copypacket(m, M_DONTWAIT);
1413 		M_PULLUP(mb0, hlen);
1414 		if (mb0 == NULL) {
1415 			free(rte, M_MRTABLE);
1416 			splx(s);
1417 			return (ENOBUFS);
1418 		}
1419 
1420 		/* is there an upcall waiting for this flow? */
1421 		hash = MFCHASH(ip->ip_src, ip->ip_dst);
1422 		LIST_FOREACH(rt, &mfchashtbl[hash], mfc_hash) {
1423 			if (in_hosteq(ip->ip_src, rt->mfc_origin) &&
1424 			    in_hosteq(ip->ip_dst, rt->mfc_mcastgrp) &&
1425 			    rt->mfc_stall != NULL)
1426 				break;
1427 		}
1428 
1429 		if (rt == NULL) {
1430 			int i;
1431 			struct igmpmsg *im;
1432 
1433 			/*
1434 			 * Locate the vifi for the incoming interface for
1435 			 * this packet.
1436 			 * If none found, drop packet.
1437 			 */
1438 			for (vifi = 0; vifi < numvifs &&
1439 				 viftable[vifi].v_ifp != ifp; vifi++)
1440 				;
1441 			if (vifi >= numvifs) /* vif not found, drop packet */
1442 				goto non_fatal;
1443 
1444 			/* no upcall, so make a new entry */
1445 			rt = (struct mfc *)malloc(sizeof(*rt), M_MRTABLE,
1446 						  M_NOWAIT);
1447 			if (rt == NULL)
1448 				goto fail;
1449 
1450 			/*
1451 			 * Make a copy of the header to send to the user level
1452 			 * process
1453 			 */
1454 			mm = m_copym(m, 0, hlen, M_DONTWAIT);
1455 			M_PULLUP(mm, hlen);
1456 			if (mm == NULL)
1457 				goto fail1;
1458 
1459 			/*
1460 			 * Send message to routing daemon to install
1461 			 * a route into the kernel table
1462 			 */
1463 
1464 			im = mtod(mm, struct igmpmsg *);
1465 			im->im_msgtype = IGMPMSG_NOCACHE;
1466 			im->im_mbz = 0;
1467 			im->im_vif = vifi;
1468 
1469 			mrtstat.mrts_upcalls++;
1470 
1471 			sockaddr_in_init(&sin, &ip->ip_src, 0);
1472 			if (socket_send(ip_mrouter, mm, &sin) < 0) {
1473 				log(LOG_WARNING,
1474 				    "ip_mforward: ip_mrouter socket queue full\n");
1475 				++mrtstat.mrts_upq_sockfull;
1476 			fail1:
1477 				free(rt, M_MRTABLE);
1478 			fail:
1479 				free(rte, M_MRTABLE);
1480 				m_freem(mb0);
1481 				splx(s);
1482 				return (ENOBUFS);
1483 			}
1484 
1485 			/* insert new entry at head of hash chain */
1486 			rt->mfc_origin = ip->ip_src;
1487 			rt->mfc_mcastgrp = ip->ip_dst;
1488 			rt->mfc_pkt_cnt = 0;
1489 			rt->mfc_byte_cnt = 0;
1490 			rt->mfc_wrong_if = 0;
1491 			rt->mfc_expire = UPCALL_EXPIRE;
1492 			nexpire[hash]++;
1493 			for (i = 0; i < numvifs; i++) {
1494 				rt->mfc_ttls[i] = 0;
1495 				rt->mfc_flags[i] = 0;
1496 			}
1497 			rt->mfc_parent = -1;
1498 
1499 			/* clear the RP address */
1500 			rt->mfc_rp = zeroin_addr;
1501 
1502 			rt->mfc_bw_meter = NULL;
1503 
1504 			/* link into table */
1505 			LIST_INSERT_HEAD(&mfchashtbl[hash], rt, mfc_hash);
1506 			/* Add this entry to the end of the queue */
1507 			rt->mfc_stall = rte;
1508 		} else {
1509 			/* determine if q has overflowed */
1510 			struct rtdetq **p;
1511 			int npkts = 0;
1512 
1513 			/*
1514 			 * XXX ouch! we need to append to the list, but we
1515 			 * only have a pointer to the front, so we have to
1516 			 * scan the entire list every time.
1517 			 */
1518 			for (p = &rt->mfc_stall; *p != NULL; p = &(*p)->next)
1519 				if (++npkts > MAX_UPQ) {
1520 					mrtstat.mrts_upq_ovflw++;
1521 				non_fatal:
1522 					free(rte, M_MRTABLE);
1523 					m_freem(mb0);
1524 					splx(s);
1525 					return (0);
1526 				}
1527 
1528 			/* Add this entry to the end of the queue */
1529 			*p = rte;
1530 		}
1531 
1532 		rte->next = NULL;
1533 		rte->m = mb0;
1534 		rte->ifp = ifp;
1535 #ifdef UPCALL_TIMING
1536 		rte->t = tp;
1537 #endif /* UPCALL_TIMING */
1538 
1539 		splx(s);
1540 
1541 		return (0);
1542 	}
1543 }
1544 
1545 
1546 /*ARGSUSED*/
1547 static void
1548 expire_upcalls(void *v)
1549 {
1550 	int i;
1551 	int s;
1552 
1553 	s = splsoftnet();
1554 
1555 	for (i = 0; i < MFCTBLSIZ; i++) {
1556 		struct mfc *rt, *nrt;
1557 
1558 		if (nexpire[i] == 0)
1559 			continue;
1560 
1561 		for (rt = LIST_FIRST(&mfchashtbl[i]); rt; rt = nrt) {
1562 			nrt = LIST_NEXT(rt, mfc_hash);
1563 
1564 			if (rt->mfc_expire == 0 || --rt->mfc_expire > 0)
1565 				continue;
1566 			nexpire[i]--;
1567 
1568 			/*
1569 			 * free the bw_meter entries
1570 			 */
1571 			while (rt->mfc_bw_meter != NULL) {
1572 				struct bw_meter *x = rt->mfc_bw_meter;
1573 
1574 				rt->mfc_bw_meter = x->bm_mfc_next;
1575 				kmem_free(x, sizeof(*x));
1576 			}
1577 
1578 			++mrtstat.mrts_cache_cleanups;
1579 			if (mrtdebug & DEBUG_EXPIRE)
1580 				log(LOG_DEBUG,
1581 				    "expire_upcalls: expiring (%x %x)\n",
1582 				    ntohl(rt->mfc_origin.s_addr),
1583 				    ntohl(rt->mfc_mcastgrp.s_addr));
1584 
1585 			expire_mfc(rt);
1586 		}
1587 	}
1588 
1589 	splx(s);
1590 	callout_reset(&expire_upcalls_ch, EXPIRE_TIMEOUT,
1591 	    expire_upcalls, NULL);
1592 }
1593 
1594 /*
1595  * Packet forwarding routine once entry in the cache is made
1596  */
1597 static int
1598 #ifdef RSVP_ISI
1599 ip_mdq(struct mbuf *m, struct ifnet *ifp, struct mfc *rt, vifi_t xmt_vif)
1600 #else
1601 ip_mdq(struct mbuf *m, struct ifnet *ifp, struct mfc *rt)
1602 #endif /* RSVP_ISI */
1603 {
1604 	struct ip  *ip = mtod(m, struct ip *);
1605 	vifi_t vifi;
1606 	struct vif *vifp;
1607 	struct sockaddr_in sin;
1608 	int plen = ntohs(ip->ip_len) - (ip->ip_hl << 2);
1609 
1610 /*
1611  * Macro to send packet on vif.  Since RSVP packets don't get counted on
1612  * input, they shouldn't get counted on output, so statistics keeping is
1613  * separate.
1614  */
1615 #define MC_SEND(ip, vifp, m) do {					\
1616 	if ((vifp)->v_flags & VIFF_TUNNEL)				\
1617 		encap_send((ip), (vifp), (m));				\
1618 	else								\
1619 		phyint_send((ip), (vifp), (m));				\
1620 } while (/*CONSTCOND*/ 0)
1621 
1622 #ifdef RSVP_ISI
1623 	/*
1624 	 * If xmt_vif is not -1, send on only the requested vif.
1625 	 *
1626 	 * (since vifi_t is u_short, -1 becomes MAXUSHORT, which > numvifs.
1627 	 */
1628 	if (xmt_vif < numvifs) {
1629 #ifdef PIM
1630 		if (viftable[xmt_vif].v_flags & VIFF_REGISTER)
1631 			pim_register_send(ip, viftable + xmt_vif, m, rt);
1632 		else
1633 #endif
1634 		MC_SEND(ip, viftable + xmt_vif, m);
1635 		return (1);
1636 	}
1637 #endif /* RSVP_ISI */
1638 
1639 	/*
1640 	 * Don't forward if it didn't arrive from the parent vif for its origin.
1641 	 */
1642 	vifi = rt->mfc_parent;
1643 	if ((vifi >= numvifs) || (viftable[vifi].v_ifp != ifp)) {
1644 		/* came in the wrong interface */
1645 		if (mrtdebug & DEBUG_FORWARD)
1646 			log(LOG_DEBUG, "wrong if: ifp %p vifi %d vififp %p\n",
1647 			    ifp, vifi,
1648 			    vifi >= numvifs ? 0 : viftable[vifi].v_ifp);
1649 		++mrtstat.mrts_wrong_if;
1650 		++rt->mfc_wrong_if;
1651 		/*
1652 		 * If we are doing PIM assert processing, send a message
1653 		 * to the routing daemon.
1654 		 *
1655 		 * XXX: A PIM-SM router needs the WRONGVIF detection so it
1656 		 * can complete the SPT switch, regardless of the type
1657 		 * of the iif (broadcast media, GRE tunnel, etc).
1658 		 */
1659 		if (pim_assert && (vifi < numvifs) && viftable[vifi].v_ifp) {
1660 			struct timeval now;
1661 			u_int32_t delta;
1662 
1663 #ifdef PIM
1664 			if (ifp == &multicast_register_if)
1665 				pimstat.pims_rcv_registers_wrongiif++;
1666 #endif
1667 
1668 			/* Get vifi for the incoming packet */
1669 			for (vifi = 0;
1670 			     vifi < numvifs && viftable[vifi].v_ifp != ifp;
1671 			     vifi++)
1672 			    ;
1673 			if (vifi >= numvifs) {
1674 				/* The iif is not found: ignore the packet. */
1675 				return (0);
1676 			}
1677 
1678 			if (rt->mfc_flags[vifi] &
1679 			    MRT_MFC_FLAGS_DISABLE_WRONGVIF) {
1680 				/* WRONGVIF disabled: ignore the packet */
1681 				return (0);
1682 			}
1683 
1684 			microtime(&now);
1685 
1686 			TV_DELTA(rt->mfc_last_assert, now, delta);
1687 
1688 			if (delta > ASSERT_MSG_TIME) {
1689 				struct igmpmsg *im;
1690 				int hlen = ip->ip_hl << 2;
1691 				struct mbuf *mm =
1692 				    m_copym(m, 0, hlen, M_DONTWAIT);
1693 
1694 				M_PULLUP(mm, hlen);
1695 				if (mm == NULL)
1696 					return (ENOBUFS);
1697 
1698 				rt->mfc_last_assert = now;
1699 
1700 				im = mtod(mm, struct igmpmsg *);
1701 				im->im_msgtype	= IGMPMSG_WRONGVIF;
1702 				im->im_mbz	= 0;
1703 				im->im_vif	= vifi;
1704 
1705 				mrtstat.mrts_upcalls++;
1706 
1707 				sockaddr_in_init(&sin, &im->im_src, 0);
1708 				if (socket_send(ip_mrouter, mm, &sin) < 0) {
1709 					log(LOG_WARNING,
1710 					    "ip_mforward: ip_mrouter socket queue full\n");
1711 					++mrtstat.mrts_upq_sockfull;
1712 					return (ENOBUFS);
1713 				}
1714 			}
1715 		}
1716 		return (0);
1717 	}
1718 
1719 	/* If I sourced this packet, it counts as output, else it was input. */
1720 	if (in_hosteq(ip->ip_src, viftable[vifi].v_lcl_addr)) {
1721 		viftable[vifi].v_pkt_out++;
1722 		viftable[vifi].v_bytes_out += plen;
1723 	} else {
1724 		viftable[vifi].v_pkt_in++;
1725 		viftable[vifi].v_bytes_in += plen;
1726 	}
1727 	rt->mfc_pkt_cnt++;
1728 	rt->mfc_byte_cnt += plen;
1729 
1730 	/*
1731 	 * For each vif, decide if a copy of the packet should be forwarded.
1732 	 * Forward if:
1733 	 *		- the ttl exceeds the vif's threshold
1734 	 *		- there are group members downstream on interface
1735 	 */
1736 	for (vifp = viftable, vifi = 0; vifi < numvifs; vifp++, vifi++)
1737 		if ((rt->mfc_ttls[vifi] > 0) &&
1738 			(ip->ip_ttl > rt->mfc_ttls[vifi])) {
1739 			vifp->v_pkt_out++;
1740 			vifp->v_bytes_out += plen;
1741 #ifdef PIM
1742 			if (vifp->v_flags & VIFF_REGISTER)
1743 				pim_register_send(ip, vifp, m, rt);
1744 			else
1745 #endif
1746 			MC_SEND(ip, vifp, m);
1747 		}
1748 
1749 	/*
1750 	 * Perform upcall-related bw measuring.
1751 	 */
1752 	if (rt->mfc_bw_meter != NULL) {
1753 		struct bw_meter *x;
1754 		struct timeval now;
1755 
1756 		microtime(&now);
1757 		for (x = rt->mfc_bw_meter; x != NULL; x = x->bm_mfc_next)
1758 			bw_meter_receive_packet(x, plen, &now);
1759 	}
1760 
1761 	return (0);
1762 }
1763 
1764 #ifdef RSVP_ISI
1765 /*
1766  * check if a vif number is legal/ok. This is used by ip_output.
1767  */
1768 int
1769 legal_vif_num(int vif)
1770 {
1771 	if (vif >= 0 && vif < numvifs)
1772 		return (1);
1773 	else
1774 		return (0);
1775 }
1776 #endif /* RSVP_ISI */
1777 
1778 static void
1779 phyint_send(struct ip *ip, struct vif *vifp, struct mbuf *m)
1780 {
1781 	struct mbuf *mb_copy;
1782 	int hlen = ip->ip_hl << 2;
1783 
1784 	/*
1785 	 * Make a new reference to the packet; make sure that
1786 	 * the IP header is actually copied, not just referenced,
1787 	 * so that ip_output() only scribbles on the copy.
1788 	 */
1789 	mb_copy = m_copypacket(m, M_DONTWAIT);
1790 	M_PULLUP(mb_copy, hlen);
1791 	if (mb_copy == NULL)
1792 		return;
1793 
1794 	if (vifp->v_rate_limit <= 0)
1795 		tbf_send_packet(vifp, mb_copy);
1796 	else
1797 		tbf_control(vifp, mb_copy, mtod(mb_copy, struct ip *),
1798 		    ntohs(ip->ip_len));
1799 }
1800 
1801 static void
1802 encap_send(struct ip *ip, struct vif *vifp, struct mbuf *m)
1803 {
1804 	struct mbuf *mb_copy;
1805 	struct ip *ip_copy;
1806 	int i, len = ntohs(ip->ip_len) + sizeof(multicast_encap_iphdr);
1807 
1808 	/* Take care of delayed checksums */
1809 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1810 		in_delayed_cksum(m);
1811 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
1812 	}
1813 
1814 	/*
1815 	 * copy the old packet & pullup it's IP header into the
1816 	 * new mbuf so we can modify it.  Try to fill the new
1817 	 * mbuf since if we don't the ethernet driver will.
1818 	 */
1819 	MGETHDR(mb_copy, M_DONTWAIT, MT_DATA);
1820 	if (mb_copy == NULL)
1821 		return;
1822 	mb_copy->m_data += max_linkhdr;
1823 	mb_copy->m_pkthdr.len = len;
1824 	mb_copy->m_len = sizeof(multicast_encap_iphdr);
1825 
1826 	if ((mb_copy->m_next = m_copypacket(m, M_DONTWAIT)) == NULL) {
1827 		m_freem(mb_copy);
1828 		return;
1829 	}
1830 	i = MHLEN - max_linkhdr;
1831 	if (i > len)
1832 		i = len;
1833 	mb_copy = m_pullup(mb_copy, i);
1834 	if (mb_copy == NULL)
1835 		return;
1836 
1837 	/*
1838 	 * fill in the encapsulating IP header.
1839 	 */
1840 	ip_copy = mtod(mb_copy, struct ip *);
1841 	*ip_copy = multicast_encap_iphdr;
1842 	if (len < IP_MINFRAGSIZE)
1843 		ip_copy->ip_id = 0;
1844 	else
1845 		ip_copy->ip_id = ip_newid(NULL);
1846 	ip_copy->ip_len = htons(len);
1847 	ip_copy->ip_src = vifp->v_lcl_addr;
1848 	ip_copy->ip_dst = vifp->v_rmt_addr;
1849 
1850 	/*
1851 	 * turn the encapsulated IP header back into a valid one.
1852 	 */
1853 	ip = (struct ip *)((char *)ip_copy + sizeof(multicast_encap_iphdr));
1854 	--ip->ip_ttl;
1855 	ip->ip_sum = 0;
1856 	mb_copy->m_data += sizeof(multicast_encap_iphdr);
1857 	ip->ip_sum = in_cksum(mb_copy, ip->ip_hl << 2);
1858 	mb_copy->m_data -= sizeof(multicast_encap_iphdr);
1859 
1860 	if (vifp->v_rate_limit <= 0)
1861 		tbf_send_packet(vifp, mb_copy);
1862 	else
1863 		tbf_control(vifp, mb_copy, ip, ntohs(ip_copy->ip_len));
1864 }
1865 
1866 /*
1867  * De-encapsulate a packet and feed it back through ip input.
1868  */
1869 static void
1870 vif_input(struct mbuf *m, ...)
1871 {
1872 	int off, proto;
1873 	va_list ap;
1874 	struct vif *vifp;
1875 	int s;
1876 	struct ifqueue *ifq;
1877 
1878 	va_start(ap, m);
1879 	off = va_arg(ap, int);
1880 	proto = va_arg(ap, int);
1881 	va_end(ap);
1882 
1883 	vifp = (struct vif *)encap_getarg(m);
1884 	if (!vifp || proto != ENCAP_PROTO) {
1885 		m_freem(m);
1886 		mrtstat.mrts_bad_tunnel++;
1887 		return;
1888 	}
1889 
1890 	m_adj(m, off);
1891 	m->m_pkthdr.rcvif = vifp->v_ifp;
1892 	ifq = &ipintrq;
1893 	s = splnet();
1894 	if (IF_QFULL(ifq)) {
1895 		IF_DROP(ifq);
1896 		m_freem(m);
1897 	} else {
1898 		IF_ENQUEUE(ifq, m);
1899 		/*
1900 		 * normally we would need a "schednetisr(NETISR_IP)"
1901 		 * here but we were called by ip_input and it is going
1902 		 * to loop back & try to dequeue the packet we just
1903 		 * queued as soon as we return so we avoid the
1904 		 * unnecessary software interrrupt.
1905 		 */
1906 	}
1907 	splx(s);
1908 }
1909 
1910 /*
1911  * Check if the packet should be received on the vif denoted by arg.
1912  * (The encap selection code will call this once per vif since each is
1913  * registered separately.)
1914  */
1915 static int
1916 vif_encapcheck(struct mbuf *m, int off, int proto, void *arg)
1917 {
1918 	struct vif *vifp;
1919 	struct ip ip;
1920 
1921 #ifdef DIAGNOSTIC
1922 	if (!arg || proto != IPPROTO_IPV4)
1923 		panic("unexpected arg in vif_encapcheck");
1924 #endif
1925 
1926 	/*
1927 	 * Accept the packet only if the inner heaader is multicast
1928 	 * and the outer header matches a tunnel-mode vif.  Order
1929 	 * checks in the hope that common non-matching packets will be
1930 	 * rejected quickly.  Assume that unicast IPv4 traffic in a
1931 	 * parallel tunnel (e.g. gif(4)) is unlikely.
1932 	 */
1933 
1934 	/* Obtain the outer IP header and the vif pointer. */
1935 	m_copydata((struct mbuf *)m, 0, sizeof(ip), (void *)&ip);
1936 	vifp = (struct vif *)arg;
1937 
1938 	/*
1939 	 * The outer source must match the vif's remote peer address.
1940 	 * For a multicast router with several tunnels, this is the
1941 	 * only check that will fail on packets in other tunnels,
1942 	 * assuming the local address is the same.
1943 	 */
1944 	if (!in_hosteq(vifp->v_rmt_addr, ip.ip_src))
1945 		return 0;
1946 
1947 	/* The outer destination must match the vif's local address. */
1948 	if (!in_hosteq(vifp->v_lcl_addr, ip.ip_dst))
1949 		return 0;
1950 
1951 	/* The vif must be of tunnel type. */
1952 	if ((vifp->v_flags & VIFF_TUNNEL) == 0)
1953 		return 0;
1954 
1955 	/* Check that the inner destination is multicast. */
1956 	m_copydata((struct mbuf *)m, off, sizeof(ip), (void *)&ip);
1957 	if (!IN_MULTICAST(ip.ip_dst.s_addr))
1958 		return 0;
1959 
1960 	/*
1961 	 * We have checked that both the outer src and dst addresses
1962 	 * match the vif, and that the inner destination is multicast
1963 	 * (224/5).  By claiming more than 64, we intend to
1964 	 * preferentially take packets that also match a parallel
1965 	 * gif(4).
1966 	 */
1967 	return 32 + 32 + 5;
1968 }
1969 
1970 /*
1971  * Token bucket filter module
1972  */
1973 static void
1974 tbf_control(struct vif *vifp, struct mbuf *m, struct ip *ip, u_int32_t len)
1975 {
1976 
1977 	if (len > MAX_BKT_SIZE) {
1978 		/* drop if packet is too large */
1979 		mrtstat.mrts_pkt2large++;
1980 		m_freem(m);
1981 		return;
1982 	}
1983 
1984 	tbf_update_tokens(vifp);
1985 
1986 	/*
1987 	 * If there are enough tokens, and the queue is empty, send this packet
1988 	 * out immediately.  Otherwise, try to insert it on this vif's queue.
1989 	 */
1990 	if (vifp->tbf_q_len == 0) {
1991 		if (len <= vifp->tbf_n_tok) {
1992 			vifp->tbf_n_tok -= len;
1993 			tbf_send_packet(vifp, m);
1994 		} else {
1995 			/* queue packet and timeout till later */
1996 			tbf_queue(vifp, m);
1997 			callout_reset(&vifp->v_repq_ch, TBF_REPROCESS,
1998 			    tbf_reprocess_q, vifp);
1999 		}
2000 	} else {
2001 		if (vifp->tbf_q_len >= vifp->tbf_max_q_len &&
2002 		    !tbf_dq_sel(vifp, ip)) {
2003 			/* queue full, and couldn't make room */
2004 			mrtstat.mrts_q_overflow++;
2005 			m_freem(m);
2006 		} else {
2007 			/* queue length low enough, or made room */
2008 			tbf_queue(vifp, m);
2009 			tbf_process_q(vifp);
2010 		}
2011 	}
2012 }
2013 
2014 /*
2015  * adds a packet to the queue at the interface
2016  */
2017 static void
2018 tbf_queue(struct vif *vifp, struct mbuf *m)
2019 {
2020 	int s = splsoftnet();
2021 
2022 	/* insert at tail */
2023 	*vifp->tbf_t = m;
2024 	vifp->tbf_t = &m->m_nextpkt;
2025 	vifp->tbf_q_len++;
2026 
2027 	splx(s);
2028 }
2029 
2030 
2031 /*
2032  * processes the queue at the interface
2033  */
2034 static void
2035 tbf_process_q(struct vif *vifp)
2036 {
2037 	struct mbuf *m;
2038 	int len;
2039 	int s = splsoftnet();
2040 
2041 	/*
2042 	 * Loop through the queue at the interface and send as many packets
2043 	 * as possible.
2044 	 */
2045 	for (m = vifp->tbf_q; m != NULL; m = vifp->tbf_q) {
2046 		len = ntohs(mtod(m, struct ip *)->ip_len);
2047 
2048 		/* determine if the packet can be sent */
2049 		if (len <= vifp->tbf_n_tok) {
2050 			/* if so,
2051 			 * reduce no of tokens, dequeue the packet,
2052 			 * send the packet.
2053 			 */
2054 			if ((vifp->tbf_q = m->m_nextpkt) == NULL)
2055 				vifp->tbf_t = &vifp->tbf_q;
2056 			--vifp->tbf_q_len;
2057 
2058 			m->m_nextpkt = NULL;
2059 			vifp->tbf_n_tok -= len;
2060 			tbf_send_packet(vifp, m);
2061 		} else
2062 			break;
2063 	}
2064 	splx(s);
2065 }
2066 
2067 static void
2068 tbf_reprocess_q(void *arg)
2069 {
2070 	struct vif *vifp = arg;
2071 
2072 	if (ip_mrouter == NULL)
2073 		return;
2074 
2075 	tbf_update_tokens(vifp);
2076 	tbf_process_q(vifp);
2077 
2078 	if (vifp->tbf_q_len != 0)
2079 		callout_reset(&vifp->v_repq_ch, TBF_REPROCESS,
2080 		    tbf_reprocess_q, vifp);
2081 }
2082 
2083 /* function that will selectively discard a member of the queue
2084  * based on the precedence value and the priority
2085  */
2086 static int
2087 tbf_dq_sel(struct vif *vifp, struct ip *ip)
2088 {
2089 	u_int p;
2090 	struct mbuf **mp, *m;
2091 	int s = splsoftnet();
2092 
2093 	p = priority(vifp, ip);
2094 
2095 	for (mp = &vifp->tbf_q, m = *mp;
2096 	    m != NULL;
2097 	    mp = &m->m_nextpkt, m = *mp) {
2098 		if (p > priority(vifp, mtod(m, struct ip *))) {
2099 			if ((*mp = m->m_nextpkt) == NULL)
2100 				vifp->tbf_t = mp;
2101 			--vifp->tbf_q_len;
2102 
2103 			m_freem(m);
2104 			mrtstat.mrts_drop_sel++;
2105 			splx(s);
2106 			return (1);
2107 		}
2108 	}
2109 	splx(s);
2110 	return (0);
2111 }
2112 
2113 static void
2114 tbf_send_packet(struct vif *vifp, struct mbuf *m)
2115 {
2116 	int error;
2117 	int s = splsoftnet();
2118 
2119 	if (vifp->v_flags & VIFF_TUNNEL) {
2120 		/* If tunnel options */
2121 		ip_output(m, NULL, &vifp->v_route, IP_FORWARDING, NULL, NULL);
2122 	} else {
2123 		/* if physical interface option, extract the options and then send */
2124 		struct ip_moptions imo;
2125 
2126 		imo.imo_multicast_ifp = vifp->v_ifp;
2127 		imo.imo_multicast_ttl = mtod(m, struct ip *)->ip_ttl - 1;
2128 		imo.imo_multicast_loop = 1;
2129 #ifdef RSVP_ISI
2130 		imo.imo_multicast_vif = -1;
2131 #endif
2132 
2133 		error = ip_output(m, NULL, NULL, IP_FORWARDING|IP_MULTICASTOPTS,
2134 		    &imo, NULL);
2135 
2136 		if (mrtdebug & DEBUG_XMIT)
2137 			log(LOG_DEBUG, "phyint_send on vif %ld err %d\n",
2138 			    (long)(vifp - viftable), error);
2139 	}
2140 	splx(s);
2141 }
2142 
2143 /* determine the current time and then
2144  * the elapsed time (between the last time and time now)
2145  * in milliseconds & update the no. of tokens in the bucket
2146  */
2147 static void
2148 tbf_update_tokens(struct vif *vifp)
2149 {
2150 	struct timeval tp;
2151 	u_int32_t tm;
2152 	int s = splsoftnet();
2153 
2154 	microtime(&tp);
2155 
2156 	TV_DELTA(tp, vifp->tbf_last_pkt_t, tm);
2157 
2158 	/*
2159 	 * This formula is actually
2160 	 * "time in seconds" * "bytes/second".
2161 	 *
2162 	 * (tm / 1000000) * (v_rate_limit * 1000 * (1000/1024) / 8)
2163 	 *
2164 	 * The (1000/1024) was introduced in add_vif to optimize
2165 	 * this divide into a shift.
2166 	 */
2167 	vifp->tbf_n_tok += tm * vifp->v_rate_limit / 8192;
2168 	vifp->tbf_last_pkt_t = tp;
2169 
2170 	if (vifp->tbf_n_tok > MAX_BKT_SIZE)
2171 		vifp->tbf_n_tok = MAX_BKT_SIZE;
2172 
2173 	splx(s);
2174 }
2175 
2176 static int
2177 priority(struct vif *vifp, struct ip *ip)
2178 {
2179 	int prio = 50;	/* the lowest priority -- default case */
2180 
2181 	/* temporary hack; may add general packet classifier some day */
2182 
2183 	/*
2184 	 * The UDP port space is divided up into four priority ranges:
2185 	 * [0, 16384)     : unclassified - lowest priority
2186 	 * [16384, 32768) : audio - highest priority
2187 	 * [32768, 49152) : whiteboard - medium priority
2188 	 * [49152, 65536) : video - low priority
2189 	 */
2190 	if (ip->ip_p == IPPROTO_UDP) {
2191 		struct udphdr *udp = (struct udphdr *)(((char *)ip) + (ip->ip_hl << 2));
2192 
2193 		switch (ntohs(udp->uh_dport) & 0xc000) {
2194 		case 0x4000:
2195 			prio = 70;
2196 			break;
2197 		case 0x8000:
2198 			prio = 60;
2199 			break;
2200 		case 0xc000:
2201 			prio = 55;
2202 			break;
2203 		}
2204 
2205 		if (tbfdebug > 1)
2206 			log(LOG_DEBUG, "port %x prio %d\n",
2207 			    ntohs(udp->uh_dport), prio);
2208 	}
2209 
2210 	return (prio);
2211 }
2212 
2213 /*
2214  * End of token bucket filter modifications
2215  */
2216 #ifdef RSVP_ISI
2217 int
2218 ip_rsvp_vif_init(struct socket *so, struct mbuf *m)
2219 {
2220 	int vifi, s;
2221 
2222 	if (rsvpdebug)
2223 		printf("ip_rsvp_vif_init: so_type = %d, pr_protocol = %d\n",
2224 		    so->so_type, so->so_proto->pr_protocol);
2225 
2226 	if (so->so_type != SOCK_RAW ||
2227 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
2228 		return (EOPNOTSUPP);
2229 
2230 	/* Check mbuf. */
2231 	if (m == NULL || m->m_len != sizeof(int)) {
2232 		return (EINVAL);
2233 	}
2234 	vifi = *(mtod(m, int *));
2235 
2236 	if (rsvpdebug)
2237 		printf("ip_rsvp_vif_init: vif = %d rsvp_on = %d\n",
2238 		       vifi, rsvp_on);
2239 
2240 	s = splsoftnet();
2241 
2242 	/* Check vif. */
2243 	if (!legal_vif_num(vifi)) {
2244 		splx(s);
2245 		return (EADDRNOTAVAIL);
2246 	}
2247 
2248 	/* Check if socket is available. */
2249 	if (viftable[vifi].v_rsvpd != NULL) {
2250 		splx(s);
2251 		return (EADDRINUSE);
2252 	}
2253 
2254 	viftable[vifi].v_rsvpd = so;
2255 	/*
2256 	 * This may seem silly, but we need to be sure we don't over-increment
2257 	 * the RSVP counter, in case something slips up.
2258 	 */
2259 	if (!viftable[vifi].v_rsvp_on) {
2260 		viftable[vifi].v_rsvp_on = 1;
2261 		rsvp_on++;
2262 	}
2263 
2264 	splx(s);
2265 	return (0);
2266 }
2267 
2268 int
2269 ip_rsvp_vif_done(struct socket *so, struct mbuf *m)
2270 {
2271 	int vifi, s;
2272 
2273 	if (rsvpdebug)
2274 		printf("ip_rsvp_vif_done: so_type = %d, pr_protocol = %d\n",
2275 		    so->so_type, so->so_proto->pr_protocol);
2276 
2277 	if (so->so_type != SOCK_RAW ||
2278 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
2279 		return (EOPNOTSUPP);
2280 
2281 	/* Check mbuf. */
2282 	if (m == NULL || m->m_len != sizeof(int)) {
2283 		return (EINVAL);
2284 	}
2285 	vifi = *(mtod(m, int *));
2286 
2287 	s = splsoftnet();
2288 
2289 	/* Check vif. */
2290 	if (!legal_vif_num(vifi)) {
2291 		splx(s);
2292 		return (EADDRNOTAVAIL);
2293 	}
2294 
2295 	if (rsvpdebug)
2296 		printf("ip_rsvp_vif_done: v_rsvpd = %x so = %x\n",
2297 		    viftable[vifi].v_rsvpd, so);
2298 
2299 	viftable[vifi].v_rsvpd = NULL;
2300 	/*
2301 	 * This may seem silly, but we need to be sure we don't over-decrement
2302 	 * the RSVP counter, in case something slips up.
2303 	 */
2304 	if (viftable[vifi].v_rsvp_on) {
2305 		viftable[vifi].v_rsvp_on = 0;
2306 		rsvp_on--;
2307 	}
2308 
2309 	splx(s);
2310 	return (0);
2311 }
2312 
2313 void
2314 ip_rsvp_force_done(struct socket *so)
2315 {
2316 	int vifi, s;
2317 
2318 	/* Don't bother if it is not the right type of socket. */
2319 	if (so->so_type != SOCK_RAW ||
2320 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
2321 		return;
2322 
2323 	s = splsoftnet();
2324 
2325 	/*
2326 	 * The socket may be attached to more than one vif...this
2327 	 * is perfectly legal.
2328 	 */
2329 	for (vifi = 0; vifi < numvifs; vifi++) {
2330 		if (viftable[vifi].v_rsvpd == so) {
2331 			viftable[vifi].v_rsvpd = NULL;
2332 			/*
2333 			 * This may seem silly, but we need to be sure we don't
2334 			 * over-decrement the RSVP counter, in case something
2335 			 * slips up.
2336 			 */
2337 			if (viftable[vifi].v_rsvp_on) {
2338 				viftable[vifi].v_rsvp_on = 0;
2339 				rsvp_on--;
2340 			}
2341 		}
2342 	}
2343 
2344 	splx(s);
2345 	return;
2346 }
2347 
2348 void
2349 rsvp_input(struct mbuf *m, struct ifnet *ifp)
2350 {
2351 	int vifi, s;
2352 	struct ip *ip = mtod(m, struct ip *);
2353 	struct sockaddr_in rsvp_src;
2354 
2355 	if (rsvpdebug)
2356 		printf("rsvp_input: rsvp_on %d\n", rsvp_on);
2357 
2358 	/*
2359 	 * Can still get packets with rsvp_on = 0 if there is a local member
2360 	 * of the group to which the RSVP packet is addressed.  But in this
2361 	 * case we want to throw the packet away.
2362 	 */
2363 	if (!rsvp_on) {
2364 		m_freem(m);
2365 		return;
2366 	}
2367 
2368 	/*
2369 	 * If the old-style non-vif-associated socket is set, then use
2370 	 * it and ignore the new ones.
2371 	 */
2372 	if (ip_rsvpd != NULL) {
2373 		if (rsvpdebug)
2374 			printf("rsvp_input: "
2375 			    "Sending packet up old-style socket\n");
2376 		rip_input(m);	/*XXX*/
2377 		return;
2378 	}
2379 
2380 	s = splsoftnet();
2381 
2382 	if (rsvpdebug)
2383 		printf("rsvp_input: check vifs\n");
2384 
2385 	/* Find which vif the packet arrived on. */
2386 	for (vifi = 0; vifi < numvifs; vifi++) {
2387 		if (viftable[vifi].v_ifp == ifp)
2388 			break;
2389 	}
2390 
2391 	if (vifi == numvifs) {
2392 		/* Can't find vif packet arrived on. Drop packet. */
2393 		if (rsvpdebug)
2394 			printf("rsvp_input: "
2395 			    "Can't find vif for packet...dropping it.\n");
2396 		m_freem(m);
2397 		splx(s);
2398 		return;
2399 	}
2400 
2401 	if (rsvpdebug)
2402 		printf("rsvp_input: check socket\n");
2403 
2404 	if (viftable[vifi].v_rsvpd == NULL) {
2405 		/*
2406 		 * drop packet, since there is no specific socket for this
2407 		 * interface
2408 		 */
2409 		if (rsvpdebug)
2410 			printf("rsvp_input: No socket defined for vif %d\n",
2411 			    vifi);
2412 		m_freem(m);
2413 		splx(s);
2414 		return;
2415 	}
2416 
2417 	sockaddr_in_init(&rsvp_src, &ip->ip_src, 0);
2418 
2419 	if (rsvpdebug && m)
2420 		printf("rsvp_input: m->m_len = %d, sbspace() = %d\n",
2421 		    m->m_len, sbspace(&viftable[vifi].v_rsvpd->so_rcv));
2422 
2423 	if (socket_send(viftable[vifi].v_rsvpd, m, &rsvp_src) < 0)
2424 		if (rsvpdebug)
2425 			printf("rsvp_input: Failed to append to socket\n");
2426 	else
2427 		if (rsvpdebug)
2428 			printf("rsvp_input: send packet up\n");
2429 
2430 	splx(s);
2431 }
2432 #endif /* RSVP_ISI */
2433 
2434 /*
2435  * Code for bandwidth monitors
2436  */
2437 
2438 /*
2439  * Define common interface for timeval-related methods
2440  */
2441 #define	BW_TIMEVALCMP(tvp, uvp, cmp) timercmp((tvp), (uvp), cmp)
2442 #define	BW_TIMEVALDECR(vvp, uvp) timersub((vvp), (uvp), (vvp))
2443 #define	BW_TIMEVALADD(vvp, uvp) timeradd((vvp), (uvp), (vvp))
2444 
2445 static uint32_t
2446 compute_bw_meter_flags(struct bw_upcall *req)
2447 {
2448     uint32_t flags = 0;
2449 
2450     if (req->bu_flags & BW_UPCALL_UNIT_PACKETS)
2451 	flags |= BW_METER_UNIT_PACKETS;
2452     if (req->bu_flags & BW_UPCALL_UNIT_BYTES)
2453 	flags |= BW_METER_UNIT_BYTES;
2454     if (req->bu_flags & BW_UPCALL_GEQ)
2455 	flags |= BW_METER_GEQ;
2456     if (req->bu_flags & BW_UPCALL_LEQ)
2457 	flags |= BW_METER_LEQ;
2458 
2459     return flags;
2460 }
2461 
2462 /*
2463  * Add a bw_meter entry
2464  */
2465 static int
2466 add_bw_upcall(struct bw_upcall *req)
2467 {
2468     int s;
2469     struct mfc *mfc;
2470     struct timeval delta = { BW_UPCALL_THRESHOLD_INTERVAL_MIN_SEC,
2471 		BW_UPCALL_THRESHOLD_INTERVAL_MIN_USEC };
2472     struct timeval now;
2473     struct bw_meter *x;
2474     uint32_t flags;
2475 
2476     if (!(mrt_api_config & MRT_MFC_BW_UPCALL))
2477 	return EOPNOTSUPP;
2478 
2479     /* Test if the flags are valid */
2480     if (!(req->bu_flags & (BW_UPCALL_UNIT_PACKETS | BW_UPCALL_UNIT_BYTES)))
2481 	return EINVAL;
2482     if (!(req->bu_flags & (BW_UPCALL_GEQ | BW_UPCALL_LEQ)))
2483 	return EINVAL;
2484     if ((req->bu_flags & (BW_UPCALL_GEQ | BW_UPCALL_LEQ))
2485 	    == (BW_UPCALL_GEQ | BW_UPCALL_LEQ))
2486 	return EINVAL;
2487 
2488     /* Test if the threshold time interval is valid */
2489     if (BW_TIMEVALCMP(&req->bu_threshold.b_time, &delta, <))
2490 	return EINVAL;
2491 
2492     flags = compute_bw_meter_flags(req);
2493 
2494     /*
2495      * Find if we have already same bw_meter entry
2496      */
2497     s = splsoftnet();
2498     mfc = mfc_find(&req->bu_src, &req->bu_dst);
2499     if (mfc == NULL) {
2500 	splx(s);
2501 	return EADDRNOTAVAIL;
2502     }
2503     for (x = mfc->mfc_bw_meter; x != NULL; x = x->bm_mfc_next) {
2504 	if ((BW_TIMEVALCMP(&x->bm_threshold.b_time,
2505 			   &req->bu_threshold.b_time, ==)) &&
2506 	    (x->bm_threshold.b_packets == req->bu_threshold.b_packets) &&
2507 	    (x->bm_threshold.b_bytes == req->bu_threshold.b_bytes) &&
2508 	    (x->bm_flags & BW_METER_USER_FLAGS) == flags)  {
2509 	    splx(s);
2510 	    return 0;		/* XXX Already installed */
2511 	}
2512     }
2513 
2514     /* Allocate the new bw_meter entry */
2515     x = kmem_intr_alloc(sizeof(*x), KM_NOSLEEP);
2516     if (x == NULL) {
2517 	splx(s);
2518 	return ENOBUFS;
2519     }
2520 
2521     /* Set the new bw_meter entry */
2522     x->bm_threshold.b_time = req->bu_threshold.b_time;
2523     microtime(&now);
2524     x->bm_start_time = now;
2525     x->bm_threshold.b_packets = req->bu_threshold.b_packets;
2526     x->bm_threshold.b_bytes = req->bu_threshold.b_bytes;
2527     x->bm_measured.b_packets = 0;
2528     x->bm_measured.b_bytes = 0;
2529     x->bm_flags = flags;
2530     x->bm_time_next = NULL;
2531     x->bm_time_hash = BW_METER_BUCKETS;
2532 
2533     /* Add the new bw_meter entry to the front of entries for this MFC */
2534     x->bm_mfc = mfc;
2535     x->bm_mfc_next = mfc->mfc_bw_meter;
2536     mfc->mfc_bw_meter = x;
2537     schedule_bw_meter(x, &now);
2538     splx(s);
2539 
2540     return 0;
2541 }
2542 
2543 static void
2544 free_bw_list(struct bw_meter *list)
2545 {
2546     while (list != NULL) {
2547 	struct bw_meter *x = list;
2548 
2549 	list = list->bm_mfc_next;
2550 	unschedule_bw_meter(x);
2551 	kmem_free(x, sizeof(*x));
2552     }
2553 }
2554 
2555 /*
2556  * Delete one or multiple bw_meter entries
2557  */
2558 static int
2559 del_bw_upcall(struct bw_upcall *req)
2560 {
2561     int s;
2562     struct mfc *mfc;
2563     struct bw_meter *x;
2564 
2565     if (!(mrt_api_config & MRT_MFC_BW_UPCALL))
2566 	return EOPNOTSUPP;
2567 
2568     s = splsoftnet();
2569     /* Find the corresponding MFC entry */
2570     mfc = mfc_find(&req->bu_src, &req->bu_dst);
2571     if (mfc == NULL) {
2572 	splx(s);
2573 	return EADDRNOTAVAIL;
2574     } else if (req->bu_flags & BW_UPCALL_DELETE_ALL) {
2575 	/*
2576 	 * Delete all bw_meter entries for this mfc
2577 	 */
2578 	struct bw_meter *list;
2579 
2580 	list = mfc->mfc_bw_meter;
2581 	mfc->mfc_bw_meter = NULL;
2582 	free_bw_list(list);
2583 	splx(s);
2584 	return 0;
2585     } else {			/* Delete a single bw_meter entry */
2586 	struct bw_meter *prev;
2587 	uint32_t flags = 0;
2588 
2589 	flags = compute_bw_meter_flags(req);
2590 
2591 	/* Find the bw_meter entry to delete */
2592 	for (prev = NULL, x = mfc->mfc_bw_meter; x != NULL;
2593 	     prev = x, x = x->bm_mfc_next) {
2594 	    if ((BW_TIMEVALCMP(&x->bm_threshold.b_time,
2595 			       &req->bu_threshold.b_time, ==)) &&
2596 		(x->bm_threshold.b_packets == req->bu_threshold.b_packets) &&
2597 		(x->bm_threshold.b_bytes == req->bu_threshold.b_bytes) &&
2598 		(x->bm_flags & BW_METER_USER_FLAGS) == flags)
2599 		break;
2600 	}
2601 	if (x != NULL) { /* Delete entry from the list for this MFC */
2602 	    if (prev != NULL)
2603 		prev->bm_mfc_next = x->bm_mfc_next;	/* remove from middle*/
2604 	    else
2605 		x->bm_mfc->mfc_bw_meter = x->bm_mfc_next;/* new head of list */
2606 
2607 	    unschedule_bw_meter(x);
2608 	    splx(s);
2609 	    /* Free the bw_meter entry */
2610 	    kmem_free(x, sizeof(*x));
2611 	    return 0;
2612 	} else {
2613 	    splx(s);
2614 	    return EINVAL;
2615 	}
2616     }
2617     /* NOTREACHED */
2618 }
2619 
2620 /*
2621  * Perform bandwidth measurement processing that may result in an upcall
2622  */
2623 static void
2624 bw_meter_receive_packet(struct bw_meter *x, int plen, struct timeval *nowp)
2625 {
2626     struct timeval delta;
2627 
2628     delta = *nowp;
2629     BW_TIMEVALDECR(&delta, &x->bm_start_time);
2630 
2631     if (x->bm_flags & BW_METER_GEQ) {
2632 	/*
2633 	 * Processing for ">=" type of bw_meter entry
2634 	 */
2635 	if (BW_TIMEVALCMP(&delta, &x->bm_threshold.b_time, >)) {
2636 	    /* Reset the bw_meter entry */
2637 	    x->bm_start_time = *nowp;
2638 	    x->bm_measured.b_packets = 0;
2639 	    x->bm_measured.b_bytes = 0;
2640 	    x->bm_flags &= ~BW_METER_UPCALL_DELIVERED;
2641 	}
2642 
2643 	/* Record that a packet is received */
2644 	x->bm_measured.b_packets++;
2645 	x->bm_measured.b_bytes += plen;
2646 
2647 	/*
2648 	 * Test if we should deliver an upcall
2649 	 */
2650 	if (!(x->bm_flags & BW_METER_UPCALL_DELIVERED)) {
2651 	    if (((x->bm_flags & BW_METER_UNIT_PACKETS) &&
2652 		 (x->bm_measured.b_packets >= x->bm_threshold.b_packets)) ||
2653 		((x->bm_flags & BW_METER_UNIT_BYTES) &&
2654 		 (x->bm_measured.b_bytes >= x->bm_threshold.b_bytes))) {
2655 		/* Prepare an upcall for delivery */
2656 		bw_meter_prepare_upcall(x, nowp);
2657 		x->bm_flags |= BW_METER_UPCALL_DELIVERED;
2658 	    }
2659 	}
2660     } else if (x->bm_flags & BW_METER_LEQ) {
2661 	/*
2662 	 * Processing for "<=" type of bw_meter entry
2663 	 */
2664 	if (BW_TIMEVALCMP(&delta, &x->bm_threshold.b_time, >)) {
2665 	    /*
2666 	     * We are behind time with the multicast forwarding table
2667 	     * scanning for "<=" type of bw_meter entries, so test now
2668 	     * if we should deliver an upcall.
2669 	     */
2670 	    if (((x->bm_flags & BW_METER_UNIT_PACKETS) &&
2671 		 (x->bm_measured.b_packets <= x->bm_threshold.b_packets)) ||
2672 		((x->bm_flags & BW_METER_UNIT_BYTES) &&
2673 		 (x->bm_measured.b_bytes <= x->bm_threshold.b_bytes))) {
2674 		/* Prepare an upcall for delivery */
2675 		bw_meter_prepare_upcall(x, nowp);
2676 	    }
2677 	    /* Reschedule the bw_meter entry */
2678 	    unschedule_bw_meter(x);
2679 	    schedule_bw_meter(x, nowp);
2680 	}
2681 
2682 	/* Record that a packet is received */
2683 	x->bm_measured.b_packets++;
2684 	x->bm_measured.b_bytes += plen;
2685 
2686 	/*
2687 	 * Test if we should restart the measuring interval
2688 	 */
2689 	if ((x->bm_flags & BW_METER_UNIT_PACKETS &&
2690 	     x->bm_measured.b_packets <= x->bm_threshold.b_packets) ||
2691 	    (x->bm_flags & BW_METER_UNIT_BYTES &&
2692 	     x->bm_measured.b_bytes <= x->bm_threshold.b_bytes)) {
2693 	    /* Don't restart the measuring interval */
2694 	} else {
2695 	    /* Do restart the measuring interval */
2696 	    /*
2697 	     * XXX: note that we don't unschedule and schedule, because this
2698 	     * might be too much overhead per packet. Instead, when we process
2699 	     * all entries for a given timer hash bin, we check whether it is
2700 	     * really a timeout. If not, we reschedule at that time.
2701 	     */
2702 	    x->bm_start_time = *nowp;
2703 	    x->bm_measured.b_packets = 0;
2704 	    x->bm_measured.b_bytes = 0;
2705 	    x->bm_flags &= ~BW_METER_UPCALL_DELIVERED;
2706 	}
2707     }
2708 }
2709 
2710 /*
2711  * Prepare a bandwidth-related upcall
2712  */
2713 static void
2714 bw_meter_prepare_upcall(struct bw_meter *x, struct timeval *nowp)
2715 {
2716     struct timeval delta;
2717     struct bw_upcall *u;
2718 
2719     /*
2720      * Compute the measured time interval
2721      */
2722     delta = *nowp;
2723     BW_TIMEVALDECR(&delta, &x->bm_start_time);
2724 
2725     /*
2726      * If there are too many pending upcalls, deliver them now
2727      */
2728     if (bw_upcalls_n >= BW_UPCALLS_MAX)
2729 	bw_upcalls_send();
2730 
2731     /*
2732      * Set the bw_upcall entry
2733      */
2734     u = &bw_upcalls[bw_upcalls_n++];
2735     u->bu_src = x->bm_mfc->mfc_origin;
2736     u->bu_dst = x->bm_mfc->mfc_mcastgrp;
2737     u->bu_threshold.b_time = x->bm_threshold.b_time;
2738     u->bu_threshold.b_packets = x->bm_threshold.b_packets;
2739     u->bu_threshold.b_bytes = x->bm_threshold.b_bytes;
2740     u->bu_measured.b_time = delta;
2741     u->bu_measured.b_packets = x->bm_measured.b_packets;
2742     u->bu_measured.b_bytes = x->bm_measured.b_bytes;
2743     u->bu_flags = 0;
2744     if (x->bm_flags & BW_METER_UNIT_PACKETS)
2745 	u->bu_flags |= BW_UPCALL_UNIT_PACKETS;
2746     if (x->bm_flags & BW_METER_UNIT_BYTES)
2747 	u->bu_flags |= BW_UPCALL_UNIT_BYTES;
2748     if (x->bm_flags & BW_METER_GEQ)
2749 	u->bu_flags |= BW_UPCALL_GEQ;
2750     if (x->bm_flags & BW_METER_LEQ)
2751 	u->bu_flags |= BW_UPCALL_LEQ;
2752 }
2753 
2754 /*
2755  * Send the pending bandwidth-related upcalls
2756  */
2757 static void
2758 bw_upcalls_send(void)
2759 {
2760     struct mbuf *m;
2761     int len = bw_upcalls_n * sizeof(bw_upcalls[0]);
2762     struct sockaddr_in k_igmpsrc = {
2763 	    .sin_len = sizeof(k_igmpsrc),
2764 	    .sin_family = AF_INET,
2765     };
2766     static struct igmpmsg igmpmsg = { 0,		/* unused1 */
2767 				      0,		/* unused2 */
2768 				      IGMPMSG_BW_UPCALL,/* im_msgtype */
2769 				      0,		/* im_mbz  */
2770 				      0,		/* im_vif  */
2771 				      0,		/* unused3 */
2772 				      { 0 },		/* im_src  */
2773 				      { 0 } };		/* im_dst  */
2774 
2775     if (bw_upcalls_n == 0)
2776 	return;			/* No pending upcalls */
2777 
2778     bw_upcalls_n = 0;
2779 
2780     /*
2781      * Allocate a new mbuf, initialize it with the header and
2782      * the payload for the pending calls.
2783      */
2784     MGETHDR(m, M_DONTWAIT, MT_HEADER);
2785     if (m == NULL) {
2786 	log(LOG_WARNING, "bw_upcalls_send: cannot allocate mbuf\n");
2787 	return;
2788     }
2789 
2790     m->m_len = m->m_pkthdr.len = 0;
2791     m_copyback(m, 0, sizeof(struct igmpmsg), (void *)&igmpmsg);
2792     m_copyback(m, sizeof(struct igmpmsg), len, (void *)&bw_upcalls[0]);
2793 
2794     /*
2795      * Send the upcalls
2796      * XXX do we need to set the address in k_igmpsrc ?
2797      */
2798     mrtstat.mrts_upcalls++;
2799     if (socket_send(ip_mrouter, m, &k_igmpsrc) < 0) {
2800 	log(LOG_WARNING, "bw_upcalls_send: ip_mrouter socket queue full\n");
2801 	++mrtstat.mrts_upq_sockfull;
2802     }
2803 }
2804 
2805 /*
2806  * Compute the timeout hash value for the bw_meter entries
2807  */
2808 #define	BW_METER_TIMEHASH(bw_meter, hash)				\
2809     do {								\
2810 	struct timeval next_timeval = (bw_meter)->bm_start_time;	\
2811 									\
2812 	BW_TIMEVALADD(&next_timeval, &(bw_meter)->bm_threshold.b_time); \
2813 	(hash) = next_timeval.tv_sec;					\
2814 	if (next_timeval.tv_usec)					\
2815 	    (hash)++; /* XXX: make sure we don't timeout early */	\
2816 	(hash) %= BW_METER_BUCKETS;					\
2817     } while (/*CONSTCOND*/ 0)
2818 
2819 /*
2820  * Schedule a timer to process periodically bw_meter entry of type "<="
2821  * by linking the entry in the proper hash bucket.
2822  */
2823 static void
2824 schedule_bw_meter(struct bw_meter *x, struct timeval *nowp)
2825 {
2826     int time_hash;
2827 
2828     if (!(x->bm_flags & BW_METER_LEQ))
2829 	return;		/* XXX: we schedule timers only for "<=" entries */
2830 
2831     /*
2832      * Reset the bw_meter entry
2833      */
2834     x->bm_start_time = *nowp;
2835     x->bm_measured.b_packets = 0;
2836     x->bm_measured.b_bytes = 0;
2837     x->bm_flags &= ~BW_METER_UPCALL_DELIVERED;
2838 
2839     /*
2840      * Compute the timeout hash value and insert the entry
2841      */
2842     BW_METER_TIMEHASH(x, time_hash);
2843     x->bm_time_next = bw_meter_timers[time_hash];
2844     bw_meter_timers[time_hash] = x;
2845     x->bm_time_hash = time_hash;
2846 }
2847 
2848 /*
2849  * Unschedule the periodic timer that processes bw_meter entry of type "<="
2850  * by removing the entry from the proper hash bucket.
2851  */
2852 static void
2853 unschedule_bw_meter(struct bw_meter *x)
2854 {
2855     int time_hash;
2856     struct bw_meter *prev, *tmp;
2857 
2858     if (!(x->bm_flags & BW_METER_LEQ))
2859 	return;		/* XXX: we schedule timers only for "<=" entries */
2860 
2861     /*
2862      * Compute the timeout hash value and delete the entry
2863      */
2864     time_hash = x->bm_time_hash;
2865     if (time_hash >= BW_METER_BUCKETS)
2866 	return;		/* Entry was not scheduled */
2867 
2868     for (prev = NULL, tmp = bw_meter_timers[time_hash];
2869 	     tmp != NULL; prev = tmp, tmp = tmp->bm_time_next)
2870 	if (tmp == x)
2871 	    break;
2872 
2873     if (tmp == NULL)
2874 	panic("unschedule_bw_meter: bw_meter entry not found");
2875 
2876     if (prev != NULL)
2877 	prev->bm_time_next = x->bm_time_next;
2878     else
2879 	bw_meter_timers[time_hash] = x->bm_time_next;
2880 
2881     x->bm_time_next = NULL;
2882     x->bm_time_hash = BW_METER_BUCKETS;
2883 }
2884 
2885 /*
2886  * Process all "<=" type of bw_meter that should be processed now,
2887  * and for each entry prepare an upcall if necessary. Each processed
2888  * entry is rescheduled again for the (periodic) processing.
2889  *
2890  * This is run periodically (once per second normally). On each round,
2891  * all the potentially matching entries are in the hash slot that we are
2892  * looking at.
2893  */
2894 static void
2895 bw_meter_process(void)
2896 {
2897     int s;
2898     static uint32_t last_tv_sec;	/* last time we processed this */
2899 
2900     uint32_t loops;
2901     int i;
2902     struct timeval now, process_endtime;
2903 
2904     microtime(&now);
2905     if (last_tv_sec == now.tv_sec)
2906 	return;		/* nothing to do */
2907 
2908     loops = now.tv_sec - last_tv_sec;
2909     last_tv_sec = now.tv_sec;
2910     if (loops > BW_METER_BUCKETS)
2911 	loops = BW_METER_BUCKETS;
2912 
2913     s = splsoftnet();
2914     /*
2915      * Process all bins of bw_meter entries from the one after the last
2916      * processed to the current one. On entry, i points to the last bucket
2917      * visited, so we need to increment i at the beginning of the loop.
2918      */
2919     for (i = (now.tv_sec - loops) % BW_METER_BUCKETS; loops > 0; loops--) {
2920 	struct bw_meter *x, *tmp_list;
2921 
2922 	if (++i >= BW_METER_BUCKETS)
2923 	    i = 0;
2924 
2925 	/* Disconnect the list of bw_meter entries from the bin */
2926 	tmp_list = bw_meter_timers[i];
2927 	bw_meter_timers[i] = NULL;
2928 
2929 	/* Process the list of bw_meter entries */
2930 	while (tmp_list != NULL) {
2931 	    x = tmp_list;
2932 	    tmp_list = tmp_list->bm_time_next;
2933 
2934 	    /* Test if the time interval is over */
2935 	    process_endtime = x->bm_start_time;
2936 	    BW_TIMEVALADD(&process_endtime, &x->bm_threshold.b_time);
2937 	    if (BW_TIMEVALCMP(&process_endtime, &now, >)) {
2938 		/* Not yet: reschedule, but don't reset */
2939 		int time_hash;
2940 
2941 		BW_METER_TIMEHASH(x, time_hash);
2942 		if (time_hash == i && process_endtime.tv_sec == now.tv_sec) {
2943 		    /*
2944 		     * XXX: somehow the bin processing is a bit ahead of time.
2945 		     * Put the entry in the next bin.
2946 		     */
2947 		    if (++time_hash >= BW_METER_BUCKETS)
2948 			time_hash = 0;
2949 		}
2950 		x->bm_time_next = bw_meter_timers[time_hash];
2951 		bw_meter_timers[time_hash] = x;
2952 		x->bm_time_hash = time_hash;
2953 
2954 		continue;
2955 	    }
2956 
2957 	    /*
2958 	     * Test if we should deliver an upcall
2959 	     */
2960 	    if (((x->bm_flags & BW_METER_UNIT_PACKETS) &&
2961 		 (x->bm_measured.b_packets <= x->bm_threshold.b_packets)) ||
2962 		((x->bm_flags & BW_METER_UNIT_BYTES) &&
2963 		 (x->bm_measured.b_bytes <= x->bm_threshold.b_bytes))) {
2964 		/* Prepare an upcall for delivery */
2965 		bw_meter_prepare_upcall(x, &now);
2966 	    }
2967 
2968 	    /*
2969 	     * Reschedule for next processing
2970 	     */
2971 	    schedule_bw_meter(x, &now);
2972 	}
2973     }
2974 
2975     /* Send all upcalls that are pending delivery */
2976     bw_upcalls_send();
2977 
2978     splx(s);
2979 }
2980 
2981 /*
2982  * A periodic function for sending all upcalls that are pending delivery
2983  */
2984 static void
2985 expire_bw_upcalls_send(void *unused)
2986 {
2987     int s;
2988 
2989     s = splsoftnet();
2990     bw_upcalls_send();
2991     splx(s);
2992 
2993     callout_reset(&bw_upcalls_ch, BW_UPCALLS_PERIOD,
2994 		  expire_bw_upcalls_send, NULL);
2995 }
2996 
2997 /*
2998  * A periodic function for periodic scanning of the multicast forwarding
2999  * table for processing all "<=" bw_meter entries.
3000  */
3001 static void
3002 expire_bw_meter_process(void *unused)
3003 {
3004     if (mrt_api_config & MRT_MFC_BW_UPCALL)
3005 	bw_meter_process();
3006 
3007     callout_reset(&bw_meter_ch, BW_METER_PERIOD,
3008 		  expire_bw_meter_process, NULL);
3009 }
3010 
3011 /*
3012  * End of bandwidth monitoring code
3013  */
3014 
3015 #ifdef PIM
3016 /*
3017  * Send the packet up to the user daemon, or eventually do kernel encapsulation
3018  */
3019 static int
3020 pim_register_send(struct ip *ip, struct vif *vifp,
3021 	struct mbuf *m, struct mfc *rt)
3022 {
3023     struct mbuf *mb_copy, *mm;
3024 
3025     if (mrtdebug & DEBUG_PIM)
3026         log(LOG_DEBUG, "pim_register_send: \n");
3027 
3028     mb_copy = pim_register_prepare(ip, m);
3029     if (mb_copy == NULL)
3030 	return ENOBUFS;
3031 
3032     /*
3033      * Send all the fragments. Note that the mbuf for each fragment
3034      * is freed by the sending machinery.
3035      */
3036     for (mm = mb_copy; mm; mm = mb_copy) {
3037 	mb_copy = mm->m_nextpkt;
3038 	mm->m_nextpkt = NULL;
3039 	mm = m_pullup(mm, sizeof(struct ip));
3040 	if (mm != NULL) {
3041 	    ip = mtod(mm, struct ip *);
3042 	    if ((mrt_api_config & MRT_MFC_RP) &&
3043 		!in_nullhost(rt->mfc_rp)) {
3044 		pim_register_send_rp(ip, vifp, mm, rt);
3045 	    } else {
3046 		pim_register_send_upcall(ip, vifp, mm, rt);
3047 	    }
3048 	}
3049     }
3050 
3051     return 0;
3052 }
3053 
3054 /*
3055  * Return a copy of the data packet that is ready for PIM Register
3056  * encapsulation.
3057  * XXX: Note that in the returned copy the IP header is a valid one.
3058  */
3059 static struct mbuf *
3060 pim_register_prepare(struct ip *ip, struct mbuf *m)
3061 {
3062     struct mbuf *mb_copy = NULL;
3063     int mtu;
3064 
3065     /* Take care of delayed checksums */
3066     if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
3067 	in_delayed_cksum(m);
3068 	m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
3069     }
3070 
3071     /*
3072      * Copy the old packet & pullup its IP header into the
3073      * new mbuf so we can modify it.
3074      */
3075     mb_copy = m_copypacket(m, M_DONTWAIT);
3076     if (mb_copy == NULL)
3077 	return NULL;
3078     mb_copy = m_pullup(mb_copy, ip->ip_hl << 2);
3079     if (mb_copy == NULL)
3080 	return NULL;
3081 
3082     /* take care of the TTL */
3083     ip = mtod(mb_copy, struct ip *);
3084     --ip->ip_ttl;
3085 
3086     /* Compute the MTU after the PIM Register encapsulation */
3087     mtu = 0xffff - sizeof(pim_encap_iphdr) - sizeof(pim_encap_pimhdr);
3088 
3089     if (ntohs(ip->ip_len) <= mtu) {
3090 	/* Turn the IP header into a valid one */
3091 	ip->ip_sum = 0;
3092 	ip->ip_sum = in_cksum(mb_copy, ip->ip_hl << 2);
3093     } else {
3094 	/* Fragment the packet */
3095 	if (ip_fragment(mb_copy, NULL, mtu) != 0) {
3096 	    /* XXX: mb_copy was freed by ip_fragment() */
3097 	    return NULL;
3098 	}
3099     }
3100     return mb_copy;
3101 }
3102 
3103 /*
3104  * Send an upcall with the data packet to the user-level process.
3105  */
3106 static int
3107 pim_register_send_upcall(struct ip *ip, struct vif *vifp,
3108     struct mbuf *mb_copy, struct mfc *rt)
3109 {
3110     struct mbuf *mb_first;
3111     int len = ntohs(ip->ip_len);
3112     struct igmpmsg *im;
3113     struct sockaddr_in k_igmpsrc = {
3114 	    .sin_len = sizeof(k_igmpsrc),
3115 	    .sin_family = AF_INET,
3116     };
3117 
3118     /*
3119      * Add a new mbuf with an upcall header
3120      */
3121     MGETHDR(mb_first, M_DONTWAIT, MT_HEADER);
3122     if (mb_first == NULL) {
3123 	m_freem(mb_copy);
3124 	return ENOBUFS;
3125     }
3126     mb_first->m_data += max_linkhdr;
3127     mb_first->m_pkthdr.len = len + sizeof(struct igmpmsg);
3128     mb_first->m_len = sizeof(struct igmpmsg);
3129     mb_first->m_next = mb_copy;
3130 
3131     /* Send message to routing daemon */
3132     im = mtod(mb_first, struct igmpmsg *);
3133     im->im_msgtype	= IGMPMSG_WHOLEPKT;
3134     im->im_mbz		= 0;
3135     im->im_vif		= vifp - viftable;
3136     im->im_src		= ip->ip_src;
3137     im->im_dst		= ip->ip_dst;
3138 
3139     k_igmpsrc.sin_addr	= ip->ip_src;
3140 
3141     mrtstat.mrts_upcalls++;
3142 
3143     if (socket_send(ip_mrouter, mb_first, &k_igmpsrc) < 0) {
3144 	if (mrtdebug & DEBUG_PIM)
3145 	    log(LOG_WARNING,
3146 		"mcast: pim_register_send_upcall: ip_mrouter socket queue full\n");
3147 	++mrtstat.mrts_upq_sockfull;
3148 	return ENOBUFS;
3149     }
3150 
3151     /* Keep statistics */
3152     pimstat.pims_snd_registers_msgs++;
3153     pimstat.pims_snd_registers_bytes += len;
3154 
3155     return 0;
3156 }
3157 
3158 /*
3159  * Encapsulate the data packet in PIM Register message and send it to the RP.
3160  */
3161 static int
3162 pim_register_send_rp(struct ip *ip, struct vif *vifp,
3163 	struct mbuf *mb_copy, struct mfc *rt)
3164 {
3165     struct mbuf *mb_first;
3166     struct ip *ip_outer;
3167     struct pim_encap_pimhdr *pimhdr;
3168     int len = ntohs(ip->ip_len);
3169     vifi_t vifi = rt->mfc_parent;
3170 
3171     if ((vifi >= numvifs) || in_nullhost(viftable[vifi].v_lcl_addr)) {
3172 	m_freem(mb_copy);
3173 	return EADDRNOTAVAIL;		/* The iif vif is invalid */
3174     }
3175 
3176     /*
3177      * Add a new mbuf with the encapsulating header
3178      */
3179     MGETHDR(mb_first, M_DONTWAIT, MT_HEADER);
3180     if (mb_first == NULL) {
3181 	m_freem(mb_copy);
3182 	return ENOBUFS;
3183     }
3184     mb_first->m_data += max_linkhdr;
3185     mb_first->m_len = sizeof(pim_encap_iphdr) + sizeof(pim_encap_pimhdr);
3186     mb_first->m_next = mb_copy;
3187 
3188     mb_first->m_pkthdr.len = len + mb_first->m_len;
3189 
3190     /*
3191      * Fill in the encapsulating IP and PIM header
3192      */
3193     ip_outer = mtod(mb_first, struct ip *);
3194     *ip_outer = pim_encap_iphdr;
3195      if (mb_first->m_pkthdr.len < IP_MINFRAGSIZE)
3196 	ip_outer->ip_id = 0;
3197     else
3198 	ip_outer->ip_id = ip_newid(NULL);
3199     ip_outer->ip_len = htons(len + sizeof(pim_encap_iphdr) +
3200 			     sizeof(pim_encap_pimhdr));
3201     ip_outer->ip_src = viftable[vifi].v_lcl_addr;
3202     ip_outer->ip_dst = rt->mfc_rp;
3203     /*
3204      * Copy the inner header TOS to the outer header, and take care of the
3205      * IP_DF bit.
3206      */
3207     ip_outer->ip_tos = ip->ip_tos;
3208     if (ntohs(ip->ip_off) & IP_DF)
3209 	ip_outer->ip_off |= htons(IP_DF);
3210     pimhdr = (struct pim_encap_pimhdr *)((char *)ip_outer
3211 					 + sizeof(pim_encap_iphdr));
3212     *pimhdr = pim_encap_pimhdr;
3213     /* If the iif crosses a border, set the Border-bit */
3214     if (rt->mfc_flags[vifi] & MRT_MFC_FLAGS_BORDER_VIF & mrt_api_config)
3215 	pimhdr->flags |= htonl(PIM_BORDER_REGISTER);
3216 
3217     mb_first->m_data += sizeof(pim_encap_iphdr);
3218     pimhdr->pim.pim_cksum = in_cksum(mb_first, sizeof(pim_encap_pimhdr));
3219     mb_first->m_data -= sizeof(pim_encap_iphdr);
3220 
3221     if (vifp->v_rate_limit == 0)
3222 	tbf_send_packet(vifp, mb_first);
3223     else
3224 	tbf_control(vifp, mb_first, ip, ntohs(ip_outer->ip_len));
3225 
3226     /* Keep statistics */
3227     pimstat.pims_snd_registers_msgs++;
3228     pimstat.pims_snd_registers_bytes += len;
3229 
3230     return 0;
3231 }
3232 
3233 /*
3234  * PIM-SMv2 and PIM-DM messages processing.
3235  * Receives and verifies the PIM control messages, and passes them
3236  * up to the listening socket, using rip_input().
3237  * The only message with special processing is the PIM_REGISTER message
3238  * (used by PIM-SM): the PIM header is stripped off, and the inner packet
3239  * is passed to if_simloop().
3240  */
3241 void
3242 pim_input(struct mbuf *m, ...)
3243 {
3244     struct ip *ip = mtod(m, struct ip *);
3245     struct pim *pim;
3246     int minlen;
3247     int datalen;
3248     int ip_tos;
3249     int proto;
3250     int iphlen;
3251     va_list ap;
3252 
3253     va_start(ap, m);
3254     iphlen = va_arg(ap, int);
3255     proto = va_arg(ap, int);
3256     va_end(ap);
3257 
3258     datalen = ntohs(ip->ip_len) - iphlen;
3259 
3260     /* Keep statistics */
3261     pimstat.pims_rcv_total_msgs++;
3262     pimstat.pims_rcv_total_bytes += datalen;
3263 
3264     /*
3265      * Validate lengths
3266      */
3267     if (datalen < PIM_MINLEN) {
3268 	pimstat.pims_rcv_tooshort++;
3269 	log(LOG_ERR, "pim_input: packet size too small %d from %lx\n",
3270 	    datalen, (u_long)ip->ip_src.s_addr);
3271 	m_freem(m);
3272 	return;
3273     }
3274 
3275     /*
3276      * If the packet is at least as big as a REGISTER, go agead
3277      * and grab the PIM REGISTER header size, to avoid another
3278      * possible m_pullup() later.
3279      *
3280      * PIM_MINLEN       == pimhdr + u_int32_t == 4 + 4 = 8
3281      * PIM_REG_MINLEN   == pimhdr + reghdr + encap_iphdr == 4 + 4 + 20 = 28
3282      */
3283     minlen = iphlen + (datalen >= PIM_REG_MINLEN ? PIM_REG_MINLEN : PIM_MINLEN);
3284     /*
3285      * Get the IP and PIM headers in contiguous memory, and
3286      * possibly the PIM REGISTER header.
3287      */
3288     if ((m->m_flags & M_EXT || m->m_len < minlen) &&
3289 	(m = m_pullup(m, minlen)) == NULL) {
3290 	log(LOG_ERR, "pim_input: m_pullup failure\n");
3291 	return;
3292     }
3293     /* m_pullup() may have given us a new mbuf so reset ip. */
3294     ip = mtod(m, struct ip *);
3295     ip_tos = ip->ip_tos;
3296 
3297     /* adjust mbuf to point to the PIM header */
3298     m->m_data += iphlen;
3299     m->m_len  -= iphlen;
3300     pim = mtod(m, struct pim *);
3301 
3302     /*
3303      * Validate checksum. If PIM REGISTER, exclude the data packet.
3304      *
3305      * XXX: some older PIMv2 implementations don't make this distinction,
3306      * so for compatibility reason perform the checksum over part of the
3307      * message, and if error, then over the whole message.
3308      */
3309     if (PIM_VT_T(pim->pim_vt) == PIM_REGISTER && in_cksum(m, PIM_MINLEN) == 0) {
3310 	/* do nothing, checksum okay */
3311     } else if (in_cksum(m, datalen)) {
3312 	pimstat.pims_rcv_badsum++;
3313 	if (mrtdebug & DEBUG_PIM)
3314 	    log(LOG_DEBUG, "pim_input: invalid checksum\n");
3315 	m_freem(m);
3316 	return;
3317     }
3318 
3319     /* PIM version check */
3320     if (PIM_VT_V(pim->pim_vt) < PIM_VERSION) {
3321 	pimstat.pims_rcv_badversion++;
3322 	log(LOG_ERR, "pim_input: incorrect version %d, expecting %d\n",
3323 	    PIM_VT_V(pim->pim_vt), PIM_VERSION);
3324 	m_freem(m);
3325 	return;
3326     }
3327 
3328     /* restore mbuf back to the outer IP */
3329     m->m_data -= iphlen;
3330     m->m_len  += iphlen;
3331 
3332     if (PIM_VT_T(pim->pim_vt) == PIM_REGISTER) {
3333 	/*
3334 	 * Since this is a REGISTER, we'll make a copy of the register
3335 	 * headers ip + pim + u_int32 + encap_ip, to be passed up to the
3336 	 * routing daemon.
3337 	 */
3338 	int s;
3339 	struct sockaddr_in dst = {
3340 		.sin_len = sizeof(dst),
3341 		.sin_family = AF_INET,
3342 	};
3343 	struct mbuf *mcp;
3344 	struct ip *encap_ip;
3345 	u_int32_t *reghdr;
3346 	struct ifnet *vifp;
3347 
3348 	s = splsoftnet();
3349 	if ((reg_vif_num >= numvifs) || (reg_vif_num == VIFI_INVALID)) {
3350 	    splx(s);
3351 	    if (mrtdebug & DEBUG_PIM)
3352 		log(LOG_DEBUG,
3353 		    "pim_input: register vif not set: %d\n", reg_vif_num);
3354 	    m_freem(m);
3355 	    return;
3356 	}
3357 	/* XXX need refcnt? */
3358 	vifp = viftable[reg_vif_num].v_ifp;
3359 	splx(s);
3360 
3361 	/*
3362 	 * Validate length
3363 	 */
3364 	if (datalen < PIM_REG_MINLEN) {
3365 	    pimstat.pims_rcv_tooshort++;
3366 	    pimstat.pims_rcv_badregisters++;
3367 	    log(LOG_ERR,
3368 		"pim_input: register packet size too small %d from %lx\n",
3369 		datalen, (u_long)ip->ip_src.s_addr);
3370 	    m_freem(m);
3371 	    return;
3372 	}
3373 
3374 	reghdr = (u_int32_t *)(pim + 1);
3375 	encap_ip = (struct ip *)(reghdr + 1);
3376 
3377 	if (mrtdebug & DEBUG_PIM) {
3378 	    log(LOG_DEBUG,
3379 		"pim_input[register], encap_ip: %lx -> %lx, encap_ip len %d\n",
3380 		(u_long)ntohl(encap_ip->ip_src.s_addr),
3381 		(u_long)ntohl(encap_ip->ip_dst.s_addr),
3382 		ntohs(encap_ip->ip_len));
3383 	}
3384 
3385 	/* verify the version number of the inner packet */
3386 	if (encap_ip->ip_v != IPVERSION) {
3387 	    pimstat.pims_rcv_badregisters++;
3388 	    if (mrtdebug & DEBUG_PIM) {
3389 		log(LOG_DEBUG, "pim_input: invalid IP version (%d) "
3390 		    "of the inner packet\n", encap_ip->ip_v);
3391 	    }
3392 	    m_freem(m);
3393 	    return;
3394 	}
3395 
3396 	/* verify the inner packet is destined to a mcast group */
3397 	if (!IN_MULTICAST(encap_ip->ip_dst.s_addr)) {
3398 	    pimstat.pims_rcv_badregisters++;
3399 	    if (mrtdebug & DEBUG_PIM)
3400 		log(LOG_DEBUG,
3401 		    "pim_input: inner packet of register is not "
3402 		    "multicast %lx\n",
3403 		    (u_long)ntohl(encap_ip->ip_dst.s_addr));
3404 	    m_freem(m);
3405 	    return;
3406 	}
3407 
3408 	/* If a NULL_REGISTER, pass it to the daemon */
3409 	if ((ntohl(*reghdr) & PIM_NULL_REGISTER))
3410 	    goto pim_input_to_daemon;
3411 
3412 	/*
3413 	 * Copy the TOS from the outer IP header to the inner IP header.
3414 	 */
3415 	if (encap_ip->ip_tos != ip_tos) {
3416 	    /* Outer TOS -> inner TOS */
3417 	    encap_ip->ip_tos = ip_tos;
3418 	    /* Recompute the inner header checksum. Sigh... */
3419 
3420 	    /* adjust mbuf to point to the inner IP header */
3421 	    m->m_data += (iphlen + PIM_MINLEN);
3422 	    m->m_len  -= (iphlen + PIM_MINLEN);
3423 
3424 	    encap_ip->ip_sum = 0;
3425 	    encap_ip->ip_sum = in_cksum(m, encap_ip->ip_hl << 2);
3426 
3427 	    /* restore mbuf to point back to the outer IP header */
3428 	    m->m_data -= (iphlen + PIM_MINLEN);
3429 	    m->m_len  += (iphlen + PIM_MINLEN);
3430 	}
3431 
3432 	/*
3433 	 * Decapsulate the inner IP packet and loopback to forward it
3434 	 * as a normal multicast packet. Also, make a copy of the
3435 	 *     outer_iphdr + pimhdr + reghdr + encap_iphdr
3436 	 * to pass to the daemon later, so it can take the appropriate
3437 	 * actions (e.g., send back PIM_REGISTER_STOP).
3438 	 * XXX: here m->m_data points to the outer IP header.
3439 	 */
3440 	mcp = m_copym(m, 0, iphlen + PIM_REG_MINLEN, M_DONTWAIT);
3441 	if (mcp == NULL) {
3442 	    log(LOG_ERR,
3443 		"pim_input: pim register: could not copy register head\n");
3444 	    m_freem(m);
3445 	    return;
3446 	}
3447 
3448 	/* Keep statistics */
3449 	/* XXX: registers_bytes include only the encap. mcast pkt */
3450 	pimstat.pims_rcv_registers_msgs++;
3451 	pimstat.pims_rcv_registers_bytes += ntohs(encap_ip->ip_len);
3452 
3453 	/*
3454 	 * forward the inner ip packet; point m_data at the inner ip.
3455 	 */
3456 	m_adj(m, iphlen + PIM_MINLEN);
3457 
3458 	if (mrtdebug & DEBUG_PIM) {
3459 	    log(LOG_DEBUG,
3460 		"pim_input: forwarding decapsulated register: "
3461 		"src %lx, dst %lx, vif %d\n",
3462 		(u_long)ntohl(encap_ip->ip_src.s_addr),
3463 		(u_long)ntohl(encap_ip->ip_dst.s_addr),
3464 		reg_vif_num);
3465 	}
3466 	/* NB: vifp was collected above; can it change on us? */
3467 	looutput(vifp, m, (struct sockaddr *)&dst, NULL);
3468 
3469 	/* prepare the register head to send to the mrouting daemon */
3470 	m = mcp;
3471     }
3472 
3473 pim_input_to_daemon:
3474     /*
3475      * Pass the PIM message up to the daemon; if it is a Register message,
3476      * pass the 'head' only up to the daemon. This includes the
3477      * outer IP header, PIM header, PIM-Register header and the
3478      * inner IP header.
3479      * XXX: the outer IP header pkt size of a Register is not adjust to
3480      * reflect the fact that the inner multicast data is truncated.
3481      */
3482     rip_input(m, iphlen, proto);
3483 
3484     return;
3485 }
3486 #endif /* PIM */
3487