xref: /netbsd-src/sys/netinet/ip_mroute.c (revision b757af438b42b93f8c6571f026d8b8ef3eaf5fc9)
1 /*	$NetBSD: ip_mroute.c,v 1.123 2012/03/22 20:34:39 drochner 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 1.123 2012/03/22 20:34:39 drochner 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/ioctl.h>
117 #include <sys/syslog.h>
118 
119 #include <net/if.h>
120 #include <net/route.h>
121 #include <net/raw_cb.h>
122 
123 #include <netinet/in.h>
124 #include <netinet/in_var.h>
125 #include <netinet/in_systm.h>
126 #include <netinet/ip.h>
127 #include <netinet/ip_var.h>
128 #include <netinet/in_pcb.h>
129 #include <netinet/udp.h>
130 #include <netinet/igmp.h>
131 #include <netinet/igmp_var.h>
132 #include <netinet/ip_mroute.h>
133 #ifdef PIM
134 #include <netinet/pim.h>
135 #include <netinet/pim_var.h>
136 #endif
137 #include <netinet/ip_encap.h>
138 
139 #ifdef FAST_IPSEC
140 #include <netipsec/ipsec.h>
141 #include <netipsec/key.h>
142 #endif
143 
144 #define IP_MULTICASTOPTS 0
145 #define	M_PULLUP(m, len)						 \
146 	do {								 \
147 		if ((m) && ((m)->m_flags & M_EXT || (m)->m_len < (len))) \
148 			(m) = m_pullup((m), (len));			 \
149 	} while (/*CONSTCOND*/ 0)
150 
151 /*
152  * Globals.  All but ip_mrouter and ip_mrtproto could be static,
153  * except for netstat or debugging purposes.
154  */
155 struct socket  *ip_mrouter  = NULL;
156 int		ip_mrtproto = IGMP_DVMRP;    /* for netstat only */
157 
158 #define NO_RTE_FOUND 	0x1
159 #define RTE_FOUND	0x2
160 
161 #define	MFCHASH(a, g)							\
162 	((((a).s_addr >> 20) ^ ((a).s_addr >> 10) ^ (a).s_addr ^	\
163 	  ((g).s_addr >> 20) ^ ((g).s_addr >> 10) ^ (g).s_addr) & mfchash)
164 LIST_HEAD(mfchashhdr, mfc) *mfchashtbl;
165 u_long	mfchash;
166 
167 u_char		nexpire[MFCTBLSIZ];
168 struct vif	viftable[MAXVIFS];
169 struct mrtstat	mrtstat;
170 u_int		mrtdebug = 0;	  /* debug level 	*/
171 #define		DEBUG_MFC	0x02
172 #define		DEBUG_FORWARD	0x04
173 #define		DEBUG_EXPIRE	0x08
174 #define		DEBUG_XMIT	0x10
175 #define		DEBUG_PIM	0x20
176 
177 #define		VIFI_INVALID	((vifi_t) -1)
178 
179 u_int       	tbfdebug = 0;     /* tbf debug level 	*/
180 #ifdef RSVP_ISI
181 u_int		rsvpdebug = 0;	  /* rsvp debug level   */
182 extern struct socket *ip_rsvpd;
183 extern int rsvp_on;
184 #endif /* RSVP_ISI */
185 
186 /* vif attachment using sys/netinet/ip_encap.c */
187 static void vif_input(struct mbuf *, ...);
188 static int vif_encapcheck(struct mbuf *, int, int, void *);
189 
190 static const struct protosw vif_protosw =
191 { SOCK_RAW,	&inetdomain,	IPPROTO_IPV4,	PR_ATOMIC|PR_ADDR,
192   vif_input,	rip_output,	0,		rip_ctloutput,
193   rip_usrreq,
194   0,            0,              0,              0,
195 };
196 
197 #define		EXPIRE_TIMEOUT	(hz / 4)	/* 4x / second */
198 #define		UPCALL_EXPIRE	6		/* number of timeouts */
199 
200 /*
201  * Define the token bucket filter structures
202  */
203 
204 #define		TBF_REPROCESS	(hz / 100)	/* 100x / second */
205 
206 static int get_sg_cnt(struct sioc_sg_req *);
207 static int get_vif_cnt(struct sioc_vif_req *);
208 static int ip_mrouter_init(struct socket *, int);
209 static int set_assert(int);
210 static int add_vif(struct vifctl *);
211 static int del_vif(vifi_t *);
212 static void update_mfc_params(struct mfc *, struct mfcctl2 *);
213 static void init_mfc_params(struct mfc *, struct mfcctl2 *);
214 static void expire_mfc(struct mfc *);
215 static int add_mfc(struct sockopt *);
216 #ifdef UPCALL_TIMING
217 static void collate(struct timeval *);
218 #endif
219 static int del_mfc(struct sockopt *);
220 static int set_api_config(struct sockopt *); /* chose API capabilities */
221 static int socket_send(struct socket *, struct mbuf *, struct sockaddr_in *);
222 static void expire_upcalls(void *);
223 #ifdef RSVP_ISI
224 static int ip_mdq(struct mbuf *, struct ifnet *, struct mfc *, vifi_t);
225 #else
226 static int ip_mdq(struct mbuf *, struct ifnet *, struct mfc *);
227 #endif
228 static void phyint_send(struct ip *, struct vif *, struct mbuf *);
229 static void encap_send(struct ip *, struct vif *, struct mbuf *);
230 static void tbf_control(struct vif *, struct mbuf *, struct ip *, u_int32_t);
231 static void tbf_queue(struct vif *, struct mbuf *);
232 static void tbf_process_q(struct vif *);
233 static void tbf_reprocess_q(void *);
234 static int tbf_dq_sel(struct vif *, struct ip *);
235 static void tbf_send_packet(struct vif *, struct mbuf *);
236 static void tbf_update_tokens(struct vif *);
237 static int priority(struct vif *, struct ip *);
238 
239 /*
240  * Bandwidth monitoring
241  */
242 static void free_bw_list(struct bw_meter *);
243 static int add_bw_upcall(struct bw_upcall *);
244 static int del_bw_upcall(struct bw_upcall *);
245 static void bw_meter_receive_packet(struct bw_meter *, int , struct timeval *);
246 static void bw_meter_prepare_upcall(struct bw_meter *, struct timeval *);
247 static void bw_upcalls_send(void);
248 static void schedule_bw_meter(struct bw_meter *, struct timeval *);
249 static void unschedule_bw_meter(struct bw_meter *);
250 static void bw_meter_process(void);
251 static void expire_bw_upcalls_send(void *);
252 static void expire_bw_meter_process(void *);
253 
254 #ifdef PIM
255 static int pim_register_send(struct ip *, struct vif *,
256 		struct mbuf *, struct mfc *);
257 static int pim_register_send_rp(struct ip *, struct vif *,
258 		struct mbuf *, struct mfc *);
259 static int pim_register_send_upcall(struct ip *, struct vif *,
260 		struct mbuf *, struct mfc *);
261 static struct mbuf *pim_register_prepare(struct ip *, struct mbuf *);
262 #endif
263 
264 /*
265  * 'Interfaces' associated with decapsulator (so we can tell
266  * packets that went through it from ones that get reflected
267  * by a broken gateway).  These interfaces are never linked into
268  * the system ifnet list & no routes point to them.  I.e., packets
269  * can't be sent this way.  They only exist as a placeholder for
270  * multicast source verification.
271  */
272 #if 0
273 struct ifnet multicast_decap_if[MAXVIFS];
274 #endif
275 
276 #define	ENCAP_TTL	64
277 #define	ENCAP_PROTO	IPPROTO_IPIP	/* 4 */
278 
279 /* prototype IP hdr for encapsulated packets */
280 struct ip multicast_encap_iphdr = {
281 	.ip_hl = sizeof(struct ip) >> 2,
282 	.ip_v = IPVERSION,
283 	.ip_len = sizeof(struct ip),
284 	.ip_ttl = ENCAP_TTL,
285 	.ip_p = ENCAP_PROTO,
286 };
287 
288 /*
289  * Bandwidth meter variables and constants
290  */
291 
292 /*
293  * Pending timeouts are stored in a hash table, the key being the
294  * expiration time. Periodically, the entries are analysed and processed.
295  */
296 #define BW_METER_BUCKETS	1024
297 static struct bw_meter *bw_meter_timers[BW_METER_BUCKETS];
298 struct callout bw_meter_ch;
299 #define BW_METER_PERIOD (hz)		/* periodical handling of bw meters */
300 
301 /*
302  * Pending upcalls are stored in a vector which is flushed when
303  * full, or periodically
304  */
305 static struct bw_upcall	bw_upcalls[BW_UPCALLS_MAX];
306 static u_int	bw_upcalls_n; /* # of pending upcalls */
307 struct callout	bw_upcalls_ch;
308 #define BW_UPCALLS_PERIOD (hz)		/* periodical flush of bw upcalls */
309 
310 #ifdef PIM
311 struct pimstat pimstat;
312 
313 /*
314  * Note: the PIM Register encapsulation adds the following in front of a
315  * data packet:
316  *
317  * struct pim_encap_hdr {
318  *    struct ip ip;
319  *    struct pim_encap_pimhdr  pim;
320  * }
321  *
322  */
323 
324 struct pim_encap_pimhdr {
325 	struct pim pim;
326 	uint32_t   flags;
327 };
328 
329 static struct ip pim_encap_iphdr = {
330 	.ip_v = IPVERSION,
331 	.ip_hl = sizeof(struct ip) >> 2,
332 	.ip_len = sizeof(struct ip),
333 	.ip_ttl = ENCAP_TTL,
334 	.ip_p = IPPROTO_PIM,
335 };
336 
337 static struct pim_encap_pimhdr pim_encap_pimhdr = {
338     {
339 	PIM_MAKE_VT(PIM_VERSION, PIM_REGISTER), /* PIM vers and message type */
340 	0,			/* reserved */
341 	0,			/* checksum */
342     },
343     0				/* flags */
344 };
345 
346 static struct ifnet multicast_register_if;
347 static vifi_t reg_vif_num = VIFI_INVALID;
348 #endif /* PIM */
349 
350 
351 /*
352  * Private variables.
353  */
354 static vifi_t	   numvifs = 0;
355 
356 static struct callout expire_upcalls_ch;
357 
358 /*
359  * whether or not special PIM assert processing is enabled.
360  */
361 static int pim_assert;
362 /*
363  * Rate limit for assert notification messages, in usec
364  */
365 #define ASSERT_MSG_TIME		3000000
366 
367 /*
368  * Kernel multicast routing API capabilities and setup.
369  * If more API capabilities are added to the kernel, they should be
370  * recorded in `mrt_api_support'.
371  */
372 static const u_int32_t mrt_api_support = (MRT_MFC_FLAGS_DISABLE_WRONGVIF |
373 					  MRT_MFC_FLAGS_BORDER_VIF |
374 					  MRT_MFC_RP |
375 					  MRT_MFC_BW_UPCALL);
376 static u_int32_t mrt_api_config = 0;
377 
378 /*
379  * Find a route for a given origin IP address and Multicast group address
380  * Type of service parameter to be added in the future!!!
381  * Statistics are updated by the caller if needed
382  * (mrtstat.mrts_mfc_lookups and mrtstat.mrts_mfc_misses)
383  */
384 static struct mfc *
385 mfc_find(struct in_addr *o, struct in_addr *g)
386 {
387 	struct mfc *rt;
388 
389 	LIST_FOREACH(rt, &mfchashtbl[MFCHASH(*o, *g)], mfc_hash) {
390 		if (in_hosteq(rt->mfc_origin, *o) &&
391 		    in_hosteq(rt->mfc_mcastgrp, *g) &&
392 		    (rt->mfc_stall == NULL))
393 			break;
394 	}
395 
396 	return (rt);
397 }
398 
399 /*
400  * Macros to compute elapsed time efficiently
401  * Borrowed from Van Jacobson's scheduling code
402  */
403 #define TV_DELTA(a, b, delta) do {					\
404 	int xxs;							\
405 	delta = (a).tv_usec - (b).tv_usec;				\
406 	xxs = (a).tv_sec - (b).tv_sec;					\
407 	switch (xxs) {							\
408 	case 2:								\
409 		delta += 1000000;					\
410 		/* fall through */					\
411 	case 1:								\
412 		delta += 1000000;					\
413 		/* fall through */					\
414 	case 0:								\
415 		break;							\
416 	default:							\
417 		delta += (1000000 * xxs);				\
418 		break;							\
419 	}								\
420 } while (/*CONSTCOND*/ 0)
421 
422 #ifdef UPCALL_TIMING
423 u_int32_t upcall_data[51];
424 #endif /* UPCALL_TIMING */
425 
426 /*
427  * Handle MRT setsockopt commands to modify the multicast routing tables.
428  */
429 int
430 ip_mrouter_set(struct socket *so, struct sockopt *sopt)
431 {
432 	int error;
433 	int optval;
434 	struct vifctl vifc;
435 	vifi_t vifi;
436 	struct bw_upcall bwuc;
437 
438 	if (sopt->sopt_name != MRT_INIT && so != ip_mrouter)
439 		error = ENOPROTOOPT;
440 	else {
441 		switch (sopt->sopt_name) {
442 		case MRT_INIT:
443 			error = sockopt_getint(sopt, &optval);
444 			if (error)
445 				break;
446 
447 			error = ip_mrouter_init(so, optval);
448 			break;
449 		case MRT_DONE:
450 			error = ip_mrouter_done();
451 			break;
452 		case MRT_ADD_VIF:
453 			error = sockopt_get(sopt, &vifc, sizeof(vifc));
454 			if (error)
455 				break;
456 			error = add_vif(&vifc);
457 			break;
458 		case MRT_DEL_VIF:
459 			error = sockopt_get(sopt, &vifi, sizeof(vifi));
460 			if (error)
461 				break;
462 			error = del_vif(&vifi);
463 			break;
464 		case MRT_ADD_MFC:
465 			error = add_mfc(sopt);
466 			break;
467 		case MRT_DEL_MFC:
468 			error = del_mfc(sopt);
469 			break;
470 		case MRT_ASSERT:
471 			error = sockopt_getint(sopt, &optval);
472 			if (error)
473 				break;
474 			error = set_assert(optval);
475 			break;
476 		case MRT_API_CONFIG:
477 			error = set_api_config(sopt);
478 			break;
479 		case MRT_ADD_BW_UPCALL:
480 			error = sockopt_get(sopt, &bwuc, sizeof(bwuc));
481 			if (error)
482 				break;
483 			error = add_bw_upcall(&bwuc);
484 			break;
485 		case MRT_DEL_BW_UPCALL:
486 			error = sockopt_get(sopt, &bwuc, sizeof(bwuc));
487 			if (error)
488 				break;
489 			error = del_bw_upcall(&bwuc);
490 			break;
491 		default:
492 			error = ENOPROTOOPT;
493 			break;
494 		}
495 	}
496 	return (error);
497 }
498 
499 /*
500  * Handle MRT getsockopt commands
501  */
502 int
503 ip_mrouter_get(struct socket *so, struct sockopt *sopt)
504 {
505 	int error;
506 
507 	if (so != ip_mrouter)
508 		error = ENOPROTOOPT;
509 	else {
510 		switch (sopt->sopt_name) {
511 		case MRT_VERSION:
512 			error = sockopt_setint(sopt, 0x0305); /* XXX !!!! */
513 			break;
514 		case MRT_ASSERT:
515 			error = sockopt_setint(sopt, pim_assert);
516 			break;
517 		case MRT_API_SUPPORT:
518 			error = sockopt_set(sopt, &mrt_api_support,
519 			    sizeof(mrt_api_support));
520 			break;
521 		case MRT_API_CONFIG:
522 			error = sockopt_set(sopt, &mrt_api_config,
523 			    sizeof(mrt_api_config));
524 			break;
525 		default:
526 			error = ENOPROTOOPT;
527 			break;
528 		}
529 	}
530 	return (error);
531 }
532 
533 /*
534  * Handle ioctl commands to obtain information from the cache
535  */
536 int
537 mrt_ioctl(struct socket *so, u_long cmd, void *data)
538 {
539 	int error;
540 
541 	if (so != ip_mrouter)
542 		error = EINVAL;
543 	else
544 		switch (cmd) {
545 		case SIOCGETVIFCNT:
546 			error = get_vif_cnt((struct sioc_vif_req *)data);
547 			break;
548 		case SIOCGETSGCNT:
549 			error = get_sg_cnt((struct sioc_sg_req *)data);
550 			break;
551 		default:
552 			error = EINVAL;
553 			break;
554 		}
555 
556 	return (error);
557 }
558 
559 /*
560  * returns the packet, byte, rpf-failure count for the source group provided
561  */
562 static int
563 get_sg_cnt(struct sioc_sg_req *req)
564 {
565 	int s;
566 	struct mfc *rt;
567 
568 	s = splsoftnet();
569 	rt = mfc_find(&req->src, &req->grp);
570 	if (rt == NULL) {
571 		splx(s);
572 		req->pktcnt = req->bytecnt = req->wrong_if = 0xffffffff;
573 		return (EADDRNOTAVAIL);
574 	}
575 	req->pktcnt = rt->mfc_pkt_cnt;
576 	req->bytecnt = rt->mfc_byte_cnt;
577 	req->wrong_if = rt->mfc_wrong_if;
578 	splx(s);
579 
580 	return (0);
581 }
582 
583 /*
584  * returns the input and output packet and byte counts on the vif provided
585  */
586 static int
587 get_vif_cnt(struct sioc_vif_req *req)
588 {
589 	vifi_t vifi = req->vifi;
590 
591 	if (vifi >= numvifs)
592 		return (EINVAL);
593 
594 	req->icount = viftable[vifi].v_pkt_in;
595 	req->ocount = viftable[vifi].v_pkt_out;
596 	req->ibytes = viftable[vifi].v_bytes_in;
597 	req->obytes = viftable[vifi].v_bytes_out;
598 
599 	return (0);
600 }
601 
602 /*
603  * Enable multicast routing
604  */
605 static int
606 ip_mrouter_init(struct socket *so, int v)
607 {
608 	if (mrtdebug)
609 		log(LOG_DEBUG,
610 		    "ip_mrouter_init: so_type = %d, pr_protocol = %d\n",
611 		    so->so_type, so->so_proto->pr_protocol);
612 
613 	if (so->so_type != SOCK_RAW ||
614 	    so->so_proto->pr_protocol != IPPROTO_IGMP)
615 		return (EOPNOTSUPP);
616 
617 	if (v != 1)
618 		return (EINVAL);
619 
620 	if (ip_mrouter != NULL)
621 		return (EADDRINUSE);
622 
623 	ip_mrouter = so;
624 
625 	mfchashtbl = hashinit(MFCTBLSIZ, HASH_LIST, true, &mfchash);
626 	memset((void *)nexpire, 0, sizeof(nexpire));
627 
628 	pim_assert = 0;
629 
630 	callout_init(&expire_upcalls_ch, 0);
631 	callout_reset(&expire_upcalls_ch, EXPIRE_TIMEOUT,
632 		      expire_upcalls, NULL);
633 
634 	callout_init(&bw_upcalls_ch, 0);
635 	callout_reset(&bw_upcalls_ch, BW_UPCALLS_PERIOD,
636 		      expire_bw_upcalls_send, NULL);
637 
638 	callout_init(&bw_meter_ch, 0);
639 	callout_reset(&bw_meter_ch, BW_METER_PERIOD,
640 		      expire_bw_meter_process, NULL);
641 
642 	if (mrtdebug)
643 		log(LOG_DEBUG, "ip_mrouter_init\n");
644 
645 	return (0);
646 }
647 
648 /*
649  * Disable multicast routing
650  */
651 int
652 ip_mrouter_done(void)
653 {
654 	vifi_t vifi;
655 	struct vif *vifp;
656 	int i;
657 	int s;
658 
659 	s = splsoftnet();
660 
661 	/* Clear out all the vifs currently in use. */
662 	for (vifi = 0; vifi < numvifs; vifi++) {
663 		vifp = &viftable[vifi];
664 		if (!in_nullhost(vifp->v_lcl_addr))
665 			reset_vif(vifp);
666 	}
667 
668 	numvifs = 0;
669 	pim_assert = 0;
670 	mrt_api_config = 0;
671 
672 	callout_stop(&expire_upcalls_ch);
673 	callout_stop(&bw_upcalls_ch);
674 	callout_stop(&bw_meter_ch);
675 
676 	/*
677 	 * Free all multicast forwarding cache entries.
678 	 */
679 	for (i = 0; i < MFCTBLSIZ; i++) {
680 		struct mfc *rt, *nrt;
681 
682 		for (rt = LIST_FIRST(&mfchashtbl[i]); rt; rt = nrt) {
683 			nrt = LIST_NEXT(rt, mfc_hash);
684 
685 			expire_mfc(rt);
686 		}
687 	}
688 
689 	memset((void *)nexpire, 0, sizeof(nexpire));
690 	hashdone(mfchashtbl, HASH_LIST, mfchash);
691 	mfchashtbl = NULL;
692 
693 	bw_upcalls_n = 0;
694 	memset(bw_meter_timers, 0, sizeof(bw_meter_timers));
695 
696 	/* Reset de-encapsulation cache. */
697 
698 	ip_mrouter = NULL;
699 
700 	splx(s);
701 
702 	if (mrtdebug)
703 		log(LOG_DEBUG, "ip_mrouter_done\n");
704 
705 	return (0);
706 }
707 
708 void
709 ip_mrouter_detach(struct ifnet *ifp)
710 {
711 	int vifi, i;
712 	struct vif *vifp;
713 	struct mfc *rt;
714 	struct rtdetq *rte;
715 
716 	/* XXX not sure about side effect to userland routing daemon */
717 	for (vifi = 0; vifi < numvifs; vifi++) {
718 		vifp = &viftable[vifi];
719 		if (vifp->v_ifp == ifp)
720 			reset_vif(vifp);
721 	}
722 	for (i = 0; i < MFCTBLSIZ; i++) {
723 		if (nexpire[i] == 0)
724 			continue;
725 		LIST_FOREACH(rt, &mfchashtbl[i], mfc_hash) {
726 			for (rte = rt->mfc_stall; rte; rte = rte->next) {
727 				if (rte->ifp == ifp)
728 					rte->ifp = NULL;
729 			}
730 		}
731 	}
732 }
733 
734 /*
735  * Set PIM assert processing global
736  */
737 static int
738 set_assert(int i)
739 {
740 	pim_assert = !!i;
741 	return (0);
742 }
743 
744 /*
745  * Configure API capabilities
746  */
747 static int
748 set_api_config(struct sockopt *sopt)
749 {
750 	u_int32_t apival;
751 	int i, error;
752 
753 	/*
754 	 * We can set the API capabilities only if it is the first operation
755 	 * after MRT_INIT. I.e.:
756 	 *  - there are no vifs installed
757 	 *  - pim_assert is not enabled
758 	 *  - the MFC table is empty
759 	 */
760 	error = sockopt_get(sopt, &apival, sizeof(apival));
761 	if (error)
762 		return (error);
763 	if (numvifs > 0)
764 		return (EPERM);
765 	if (pim_assert)
766 		return (EPERM);
767 	for (i = 0; i < MFCTBLSIZ; i++) {
768 		if (LIST_FIRST(&mfchashtbl[i]) != NULL)
769 			return (EPERM);
770 	}
771 
772 	mrt_api_config = apival & mrt_api_support;
773 	return (0);
774 }
775 
776 /*
777  * Add a vif to the vif table
778  */
779 static int
780 add_vif(struct vifctl *vifcp)
781 {
782 	struct vif *vifp;
783 	struct ifaddr *ifa;
784 	struct ifnet *ifp;
785 	int error, s;
786 	struct sockaddr_in sin;
787 
788 	if (vifcp->vifc_vifi >= MAXVIFS)
789 		return (EINVAL);
790 	if (in_nullhost(vifcp->vifc_lcl_addr))
791 		return (EADDRNOTAVAIL);
792 
793 	vifp = &viftable[vifcp->vifc_vifi];
794 	if (!in_nullhost(vifp->v_lcl_addr))
795 		return (EADDRINUSE);
796 
797 	/* Find the interface with an address in AF_INET family. */
798 #ifdef PIM
799 	if (vifcp->vifc_flags & VIFF_REGISTER) {
800 		/*
801 		 * XXX: Because VIFF_REGISTER does not really need a valid
802 		 * local interface (e.g. it could be 127.0.0.2), we don't
803 		 * check its address.
804 		 */
805 	    ifp = NULL;
806 	} else
807 #endif
808 	{
809 		sockaddr_in_init(&sin, &vifcp->vifc_lcl_addr, 0);
810 		ifa = ifa_ifwithaddr(sintosa(&sin));
811 		if (ifa == NULL)
812 			return (EADDRNOTAVAIL);
813 		ifp = ifa->ifa_ifp;
814 	}
815 
816 	if (vifcp->vifc_flags & VIFF_TUNNEL) {
817 		if (vifcp->vifc_flags & VIFF_SRCRT) {
818 			log(LOG_ERR, "source routed tunnels not supported\n");
819 			return (EOPNOTSUPP);
820 		}
821 
822 		/* attach this vif to decapsulator dispatch table */
823 		/*
824 		 * XXX Use addresses in registration so that matching
825 		 * can be done with radix tree in decapsulator.  But,
826 		 * we need to check inner header for multicast, so
827 		 * this requires both radix tree lookup and then a
828 		 * function to check, and this is not supported yet.
829 		 */
830 		vifp->v_encap_cookie = encap_attach_func(AF_INET, IPPROTO_IPV4,
831 		    vif_encapcheck, &vif_protosw, vifp);
832 		if (!vifp->v_encap_cookie)
833 			return (EINVAL);
834 
835 		/* Create a fake encapsulation interface. */
836 		ifp = malloc(sizeof(*ifp), M_MRTABLE, M_WAITOK|M_ZERO);
837 		snprintf(ifp->if_xname, sizeof(ifp->if_xname),
838 			 "mdecap%d", vifcp->vifc_vifi);
839 
840 		/* Prepare cached route entry. */
841 		memset(&vifp->v_route, 0, sizeof(vifp->v_route));
842 #ifdef PIM
843 	} else if (vifcp->vifc_flags & VIFF_REGISTER) {
844 		ifp = &multicast_register_if;
845 		if (mrtdebug)
846 			log(LOG_DEBUG, "Adding a register vif, ifp: %p\n",
847 			    (void *)ifp);
848 		if (reg_vif_num == VIFI_INVALID) {
849 			memset(ifp, 0, sizeof(*ifp));
850 			snprintf(ifp->if_xname, sizeof(ifp->if_xname),
851 				 "register_vif");
852 			ifp->if_flags = IFF_LOOPBACK;
853 			memset(&vifp->v_route, 0, sizeof(vifp->v_route));
854 			reg_vif_num = vifcp->vifc_vifi;
855 		}
856 #endif
857 	} else {
858 		/* Make sure the interface supports multicast. */
859 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
860 			return (EOPNOTSUPP);
861 
862 		/* Enable promiscuous reception of all IP multicasts. */
863 		sockaddr_in_init(&sin, &zeroin_addr, 0);
864 		error = if_mcast_op(ifp, SIOCADDMULTI, sintosa(&sin));
865 		if (error)
866 			return (error);
867 	}
868 
869 	s = splsoftnet();
870 
871 	/* Define parameters for the tbf structure. */
872 	vifp->tbf_q = NULL;
873 	vifp->tbf_t = &vifp->tbf_q;
874 	microtime(&vifp->tbf_last_pkt_t);
875 	vifp->tbf_n_tok = 0;
876 	vifp->tbf_q_len = 0;
877 	vifp->tbf_max_q_len = MAXQSIZE;
878 
879 	vifp->v_flags = vifcp->vifc_flags;
880 	vifp->v_threshold = vifcp->vifc_threshold;
881 	/* scaling up here allows division by 1024 in critical code */
882 	vifp->v_rate_limit = vifcp->vifc_rate_limit * 1024 / 1000;
883 	vifp->v_lcl_addr = vifcp->vifc_lcl_addr;
884 	vifp->v_rmt_addr = vifcp->vifc_rmt_addr;
885 	vifp->v_ifp = ifp;
886 	/* Initialize per vif pkt counters. */
887 	vifp->v_pkt_in = 0;
888 	vifp->v_pkt_out = 0;
889 	vifp->v_bytes_in = 0;
890 	vifp->v_bytes_out = 0;
891 
892 	callout_init(&vifp->v_repq_ch, 0);
893 
894 #ifdef RSVP_ISI
895 	vifp->v_rsvp_on = 0;
896 	vifp->v_rsvpd = NULL;
897 #endif /* RSVP_ISI */
898 
899 	splx(s);
900 
901 	/* Adjust numvifs up if the vifi is higher than numvifs. */
902 	if (numvifs <= vifcp->vifc_vifi)
903 		numvifs = vifcp->vifc_vifi + 1;
904 
905 	if (mrtdebug)
906 		log(LOG_DEBUG, "add_vif #%d, lcladdr %x, %s %x, thresh %x, rate %d\n",
907 		    vifcp->vifc_vifi,
908 		    ntohl(vifcp->vifc_lcl_addr.s_addr),
909 		    (vifcp->vifc_flags & VIFF_TUNNEL) ? "rmtaddr" : "mask",
910 		    ntohl(vifcp->vifc_rmt_addr.s_addr),
911 		    vifcp->vifc_threshold,
912 		    vifcp->vifc_rate_limit);
913 
914 	return (0);
915 }
916 
917 void
918 reset_vif(struct vif *vifp)
919 {
920 	struct mbuf *m, *n;
921 	struct ifnet *ifp;
922 	struct sockaddr_in sin;
923 
924 	callout_stop(&vifp->v_repq_ch);
925 
926 	/* detach this vif from decapsulator dispatch table */
927 	encap_detach(vifp->v_encap_cookie);
928 	vifp->v_encap_cookie = NULL;
929 
930 	/*
931 	 * Free packets queued at the interface
932 	 */
933 	for (m = vifp->tbf_q; m != NULL; m = n) {
934 		n = m->m_nextpkt;
935 		m_freem(m);
936 	}
937 
938 	if (vifp->v_flags & VIFF_TUNNEL)
939 		free(vifp->v_ifp, M_MRTABLE);
940 	else if (vifp->v_flags & VIFF_REGISTER) {
941 #ifdef PIM
942 		reg_vif_num = VIFI_INVALID;
943 #endif
944 	} else {
945 		sockaddr_in_init(&sin, &zeroin_addr, 0);
946 		ifp = vifp->v_ifp;
947 		if_mcast_op(ifp, SIOCDELMULTI, sintosa(&sin));
948 	}
949 	memset((void *)vifp, 0, sizeof(*vifp));
950 }
951 
952 /*
953  * Delete a vif from the vif table
954  */
955 static int
956 del_vif(vifi_t *vifip)
957 {
958 	struct vif *vifp;
959 	vifi_t vifi;
960 	int s;
961 
962 	if (*vifip >= numvifs)
963 		return (EINVAL);
964 
965 	vifp = &viftable[*vifip];
966 	if (in_nullhost(vifp->v_lcl_addr))
967 		return (EADDRNOTAVAIL);
968 
969 	s = splsoftnet();
970 
971 	reset_vif(vifp);
972 
973 	/* Adjust numvifs down */
974 	for (vifi = numvifs; vifi > 0; vifi--)
975 		if (!in_nullhost(viftable[vifi - 1].v_lcl_addr))
976 			break;
977 	numvifs = vifi;
978 
979 	splx(s);
980 
981 	if (mrtdebug)
982 		log(LOG_DEBUG, "del_vif %d, numvifs %d\n", *vifip, numvifs);
983 
984 	return (0);
985 }
986 
987 /*
988  * update an mfc entry without resetting counters and S,G addresses.
989  */
990 static void
991 update_mfc_params(struct mfc *rt, struct mfcctl2 *mfccp)
992 {
993 	int i;
994 
995 	rt->mfc_parent = mfccp->mfcc_parent;
996 	for (i = 0; i < numvifs; i++) {
997 		rt->mfc_ttls[i] = mfccp->mfcc_ttls[i];
998 		rt->mfc_flags[i] = mfccp->mfcc_flags[i] & mrt_api_config &
999 			MRT_MFC_FLAGS_ALL;
1000 	}
1001 	/* set the RP address */
1002 	if (mrt_api_config & MRT_MFC_RP)
1003 		rt->mfc_rp = mfccp->mfcc_rp;
1004 	else
1005 		rt->mfc_rp = zeroin_addr;
1006 }
1007 
1008 /*
1009  * fully initialize an mfc entry from the parameter.
1010  */
1011 static void
1012 init_mfc_params(struct mfc *rt, struct mfcctl2 *mfccp)
1013 {
1014 	rt->mfc_origin     = mfccp->mfcc_origin;
1015 	rt->mfc_mcastgrp   = mfccp->mfcc_mcastgrp;
1016 
1017 	update_mfc_params(rt, mfccp);
1018 
1019 	/* initialize pkt counters per src-grp */
1020 	rt->mfc_pkt_cnt    = 0;
1021 	rt->mfc_byte_cnt   = 0;
1022 	rt->mfc_wrong_if   = 0;
1023 	timerclear(&rt->mfc_last_assert);
1024 }
1025 
1026 static void
1027 expire_mfc(struct mfc *rt)
1028 {
1029 	struct rtdetq *rte, *nrte;
1030 
1031 	free_bw_list(rt->mfc_bw_meter);
1032 
1033 	for (rte = rt->mfc_stall; rte != NULL; rte = nrte) {
1034 		nrte = rte->next;
1035 		m_freem(rte->m);
1036 		free(rte, M_MRTABLE);
1037 	}
1038 
1039 	LIST_REMOVE(rt, mfc_hash);
1040 	free(rt, M_MRTABLE);
1041 }
1042 
1043 /*
1044  * Add an mfc entry
1045  */
1046 static int
1047 add_mfc(struct sockopt *sopt)
1048 {
1049 	struct mfcctl2 mfcctl2;
1050 	struct mfcctl2 *mfccp;
1051 	struct mfc *rt;
1052 	u_int32_t hash = 0;
1053 	struct rtdetq *rte, *nrte;
1054 	u_short nstl;
1055 	int s;
1056 	int mfcctl_size = sizeof(struct mfcctl);
1057 	int error;
1058 
1059 	if (mrt_api_config & MRT_API_FLAGS_ALL)
1060 		mfcctl_size = sizeof(struct mfcctl2);
1061 
1062 	/*
1063 	 * select data size depending on API version.
1064 	 */
1065 	mfccp = &mfcctl2;
1066 	memset(&mfcctl2, 0, sizeof(mfcctl2));
1067 
1068 	if (mrt_api_config & MRT_API_FLAGS_ALL)
1069 		error = sockopt_get(sopt, mfccp, sizeof(struct mfcctl2));
1070 	else
1071 		error = sockopt_get(sopt, mfccp, sizeof(struct mfcctl));
1072 
1073 	if (error)
1074 		return (error);
1075 
1076 	s = splsoftnet();
1077 	rt = mfc_find(&mfccp->mfcc_origin, &mfccp->mfcc_mcastgrp);
1078 
1079 	/* If an entry already exists, just update the fields */
1080 	if (rt) {
1081 		if (mrtdebug & DEBUG_MFC)
1082 			log(LOG_DEBUG, "add_mfc update o %x g %x p %x\n",
1083 			    ntohl(mfccp->mfcc_origin.s_addr),
1084 			    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1085 			    mfccp->mfcc_parent);
1086 
1087 		update_mfc_params(rt, mfccp);
1088 
1089 		splx(s);
1090 		return (0);
1091 	}
1092 
1093 	/*
1094 	 * Find the entry for which the upcall was made and update
1095 	 */
1096 	nstl = 0;
1097 	hash = MFCHASH(mfccp->mfcc_origin, mfccp->mfcc_mcastgrp);
1098 	LIST_FOREACH(rt, &mfchashtbl[hash], mfc_hash) {
1099 		if (in_hosteq(rt->mfc_origin, mfccp->mfcc_origin) &&
1100 		    in_hosteq(rt->mfc_mcastgrp, mfccp->mfcc_mcastgrp) &&
1101 		    rt->mfc_stall != NULL) {
1102 			if (nstl++)
1103 				log(LOG_ERR, "add_mfc %s o %x g %x p %x dbx %p\n",
1104 				    "multiple kernel entries",
1105 				    ntohl(mfccp->mfcc_origin.s_addr),
1106 				    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1107 				    mfccp->mfcc_parent, rt->mfc_stall);
1108 
1109 			if (mrtdebug & DEBUG_MFC)
1110 				log(LOG_DEBUG, "add_mfc o %x g %x p %x dbg %p\n",
1111 				    ntohl(mfccp->mfcc_origin.s_addr),
1112 				    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1113 				    mfccp->mfcc_parent, rt->mfc_stall);
1114 
1115 			rte = rt->mfc_stall;
1116 			init_mfc_params(rt, mfccp);
1117 			rt->mfc_stall = NULL;
1118 
1119 			rt->mfc_expire = 0; /* Don't clean this guy up */
1120 			nexpire[hash]--;
1121 
1122 			/* free packets Qed at the end of this entry */
1123 			for (; rte != NULL; rte = nrte) {
1124 				nrte = rte->next;
1125 				if (rte->ifp) {
1126 #ifdef RSVP_ISI
1127 					ip_mdq(rte->m, rte->ifp, rt, -1);
1128 #else
1129 					ip_mdq(rte->m, rte->ifp, rt);
1130 #endif /* RSVP_ISI */
1131 				}
1132 				m_freem(rte->m);
1133 #ifdef UPCALL_TIMING
1134 				collate(&rte->t);
1135 #endif /* UPCALL_TIMING */
1136 				free(rte, M_MRTABLE);
1137 			}
1138 		}
1139 	}
1140 
1141 	/*
1142 	 * It is possible that an entry is being inserted without an upcall
1143 	 */
1144 	if (nstl == 0) {
1145 		/*
1146 		 * No mfc; make a new one
1147 		 */
1148 		if (mrtdebug & DEBUG_MFC)
1149 			log(LOG_DEBUG, "add_mfc no upcall o %x g %x p %x\n",
1150 			    ntohl(mfccp->mfcc_origin.s_addr),
1151 			    ntohl(mfccp->mfcc_mcastgrp.s_addr),
1152 			    mfccp->mfcc_parent);
1153 
1154 		LIST_FOREACH(rt, &mfchashtbl[hash], mfc_hash) {
1155 			if (in_hosteq(rt->mfc_origin, mfccp->mfcc_origin) &&
1156 			    in_hosteq(rt->mfc_mcastgrp, mfccp->mfcc_mcastgrp)) {
1157 				init_mfc_params(rt, mfccp);
1158 				if (rt->mfc_expire)
1159 					nexpire[hash]--;
1160 				rt->mfc_expire = 0;
1161 				break; /* XXX */
1162 			}
1163 		}
1164 		if (rt == NULL) {	/* no upcall, so make a new entry */
1165 			rt = (struct mfc *)malloc(sizeof(*rt), M_MRTABLE,
1166 						  M_NOWAIT);
1167 			if (rt == NULL) {
1168 				splx(s);
1169 				return (ENOBUFS);
1170 			}
1171 
1172 			init_mfc_params(rt, mfccp);
1173 			rt->mfc_expire	= 0;
1174 			rt->mfc_stall	= NULL;
1175 			rt->mfc_bw_meter = NULL;
1176 
1177 			/* insert new entry at head of hash chain */
1178 			LIST_INSERT_HEAD(&mfchashtbl[hash], rt, mfc_hash);
1179 		}
1180 	}
1181 
1182 	splx(s);
1183 	return (0);
1184 }
1185 
1186 #ifdef UPCALL_TIMING
1187 /*
1188  * collect delay statistics on the upcalls
1189  */
1190 static void
1191 collate(struct timeval *t)
1192 {
1193 	u_int32_t d;
1194 	struct timeval tp;
1195 	u_int32_t delta;
1196 
1197 	microtime(&tp);
1198 
1199 	if (timercmp(t, &tp, <)) {
1200 		TV_DELTA(tp, *t, delta);
1201 
1202 		d = delta >> 10;
1203 		if (d > 50)
1204 			d = 50;
1205 
1206 		++upcall_data[d];
1207 	}
1208 }
1209 #endif /* UPCALL_TIMING */
1210 
1211 /*
1212  * Delete an mfc entry
1213  */
1214 static int
1215 del_mfc(struct sockopt *sopt)
1216 {
1217 	struct mfcctl2 mfcctl2;
1218 	struct mfcctl2 *mfccp;
1219 	struct mfc *rt;
1220 	int s;
1221 	int error;
1222 
1223 	/*
1224 	 * XXX: for deleting MFC entries the information in entries
1225 	 * of size "struct mfcctl" is sufficient.
1226 	 */
1227 
1228 	mfccp = &mfcctl2;
1229 	memset(&mfcctl2, 0, sizeof(mfcctl2));
1230 
1231 	error = sockopt_get(sopt, mfccp, sizeof(struct mfcctl));
1232 	if (error) {
1233 		/* Try with the size of mfcctl2. */
1234 		error = sockopt_get(sopt, mfccp, sizeof(struct mfcctl2));
1235 		if (error)
1236 			return (error);
1237 	}
1238 
1239 	if (mrtdebug & DEBUG_MFC)
1240 		log(LOG_DEBUG, "del_mfc origin %x mcastgrp %x\n",
1241 		    ntohl(mfccp->mfcc_origin.s_addr),
1242 		    ntohl(mfccp->mfcc_mcastgrp.s_addr));
1243 
1244 	s = splsoftnet();
1245 
1246 	rt = mfc_find(&mfccp->mfcc_origin, &mfccp->mfcc_mcastgrp);
1247 	if (rt == NULL) {
1248 		splx(s);
1249 		return (EADDRNOTAVAIL);
1250 	}
1251 
1252 	/*
1253 	 * free the bw_meter entries
1254 	 */
1255 	free_bw_list(rt->mfc_bw_meter);
1256 	rt->mfc_bw_meter = NULL;
1257 
1258 	LIST_REMOVE(rt, mfc_hash);
1259 	free(rt, M_MRTABLE);
1260 
1261 	splx(s);
1262 	return (0);
1263 }
1264 
1265 static int
1266 socket_send(struct socket *s, struct mbuf *mm, struct sockaddr_in *src)
1267 {
1268 	if (s) {
1269 		if (sbappendaddr(&s->so_rcv, sintosa(src), mm, NULL) != 0) {
1270 			sorwakeup(s);
1271 			return (0);
1272 		}
1273 	}
1274 	m_freem(mm);
1275 	return (-1);
1276 }
1277 
1278 /*
1279  * IP multicast forwarding function. This function assumes that the packet
1280  * pointed to by "ip" has arrived on (or is about to be sent to) the interface
1281  * pointed to by "ifp", and the packet is to be relayed to other networks
1282  * that have members of the packet's destination IP multicast group.
1283  *
1284  * The packet is returned unscathed to the caller, unless it is
1285  * erroneous, in which case a non-zero return value tells the caller to
1286  * discard it.
1287  */
1288 
1289 #define IP_HDR_LEN  20	/* # bytes of fixed IP header (excluding options) */
1290 #define TUNNEL_LEN  12  /* # bytes of IP option for tunnel encapsulation  */
1291 
1292 int
1293 #ifdef RSVP_ISI
1294 ip_mforward(struct mbuf *m, struct ifnet *ifp, struct ip_moptions *imo)
1295 #else
1296 ip_mforward(struct mbuf *m, struct ifnet *ifp)
1297 #endif /* RSVP_ISI */
1298 {
1299 	struct ip *ip = mtod(m, struct ip *);
1300 	struct mfc *rt;
1301 	static int srctun = 0;
1302 	struct mbuf *mm;
1303 	struct sockaddr_in sin;
1304 	int s;
1305 	vifi_t vifi;
1306 
1307 	if (mrtdebug & DEBUG_FORWARD)
1308 		log(LOG_DEBUG, "ip_mforward: src %x, dst %x, ifp %p\n",
1309 		    ntohl(ip->ip_src.s_addr), ntohl(ip->ip_dst.s_addr), ifp);
1310 
1311 	if (ip->ip_hl < (IP_HDR_LEN + TUNNEL_LEN) >> 2 ||
1312 	    ((u_char *)(ip + 1))[1] != IPOPT_LSRR) {
1313 		/*
1314 		 * Packet arrived via a physical interface or
1315 		 * an encapsulated tunnel or a register_vif.
1316 		 */
1317 	} else {
1318 		/*
1319 		 * Packet arrived through a source-route tunnel.
1320 		 * Source-route tunnels are no longer supported.
1321 		 */
1322 		if ((srctun++ % 1000) == 0)
1323 			log(LOG_ERR,
1324 			    "ip_mforward: received source-routed packet from %x\n",
1325 			    ntohl(ip->ip_src.s_addr));
1326 
1327 		return (1);
1328 	}
1329 
1330 	/*
1331 	 * Clear any in-bound checksum flags for this packet.
1332 	 */
1333 	m->m_pkthdr.csum_flags = 0;
1334 
1335 #ifdef RSVP_ISI
1336 	if (imo && ((vifi = imo->imo_multicast_vif) < numvifs)) {
1337 		if (ip->ip_ttl < MAXTTL)
1338 			ip->ip_ttl++;	/* compensate for -1 in *_send routines */
1339 		if (rsvpdebug && ip->ip_p == IPPROTO_RSVP) {
1340 			struct vif *vifp = viftable + vifi;
1341 			printf("Sending IPPROTO_RSVP from %x to %x on vif %d (%s%s)\n",
1342 			    ntohl(ip->ip_src), ntohl(ip->ip_dst), vifi,
1343 			    (vifp->v_flags & VIFF_TUNNEL) ? "tunnel on " : "",
1344 			    vifp->v_ifp->if_xname);
1345 		}
1346 		return (ip_mdq(m, ifp, NULL, vifi));
1347 	}
1348 	if (rsvpdebug && ip->ip_p == IPPROTO_RSVP) {
1349 		printf("Warning: IPPROTO_RSVP from %x to %x without vif option\n",
1350 		    ntohl(ip->ip_src), ntohl(ip->ip_dst));
1351 	}
1352 #endif /* RSVP_ISI */
1353 
1354 	/*
1355 	 * Don't forward a packet with time-to-live of zero or one,
1356 	 * or a packet destined to a local-only group.
1357 	 */
1358 	if (ip->ip_ttl <= 1 || IN_LOCAL_GROUP(ip->ip_dst.s_addr))
1359 		return (0);
1360 
1361 	/*
1362 	 * Determine forwarding vifs from the forwarding cache table
1363 	 */
1364 	s = splsoftnet();
1365 	++mrtstat.mrts_mfc_lookups;
1366 	rt = mfc_find(&ip->ip_src, &ip->ip_dst);
1367 
1368 	/* Entry exists, so forward if necessary */
1369 	if (rt != NULL) {
1370 		splx(s);
1371 #ifdef RSVP_ISI
1372 		return (ip_mdq(m, ifp, rt, -1));
1373 #else
1374 		return (ip_mdq(m, ifp, rt));
1375 #endif /* RSVP_ISI */
1376 	} else {
1377 		/*
1378 		 * If we don't have a route for packet's origin,
1379 		 * Make a copy of the packet & send message to routing daemon
1380 		 */
1381 
1382 		struct mbuf *mb0;
1383 		struct rtdetq *rte;
1384 		u_int32_t hash;
1385 		int hlen = ip->ip_hl << 2;
1386 #ifdef UPCALL_TIMING
1387 		struct timeval tp;
1388 
1389 		microtime(&tp);
1390 #endif /* UPCALL_TIMING */
1391 
1392 		++mrtstat.mrts_mfc_misses;
1393 
1394 		mrtstat.mrts_no_route++;
1395 		if (mrtdebug & (DEBUG_FORWARD | DEBUG_MFC))
1396 			log(LOG_DEBUG, "ip_mforward: no rte s %x g %x\n",
1397 			    ntohl(ip->ip_src.s_addr),
1398 			    ntohl(ip->ip_dst.s_addr));
1399 
1400 		/*
1401 		 * Allocate mbufs early so that we don't do extra work if we are
1402 		 * just going to fail anyway.  Make sure to pullup the header so
1403 		 * that other people can't step on it.
1404 		 */
1405 		rte = (struct rtdetq *)malloc(sizeof(*rte), M_MRTABLE,
1406 					      M_NOWAIT);
1407 		if (rte == NULL) {
1408 			splx(s);
1409 			return (ENOBUFS);
1410 		}
1411 		mb0 = m_copypacket(m, M_DONTWAIT);
1412 		M_PULLUP(mb0, hlen);
1413 		if (mb0 == NULL) {
1414 			free(rte, M_MRTABLE);
1415 			splx(s);
1416 			return (ENOBUFS);
1417 		}
1418 
1419 		/* is there an upcall waiting for this flow? */
1420 		hash = MFCHASH(ip->ip_src, ip->ip_dst);
1421 		LIST_FOREACH(rt, &mfchashtbl[hash], mfc_hash) {
1422 			if (in_hosteq(ip->ip_src, rt->mfc_origin) &&
1423 			    in_hosteq(ip->ip_dst, rt->mfc_mcastgrp) &&
1424 			    rt->mfc_stall != NULL)
1425 				break;
1426 		}
1427 
1428 		if (rt == NULL) {
1429 			int i;
1430 			struct igmpmsg *im;
1431 
1432 			/*
1433 			 * Locate the vifi for the incoming interface for
1434 			 * this packet.
1435 			 * If none found, drop packet.
1436 			 */
1437 			for (vifi = 0; vifi < numvifs &&
1438 				 viftable[vifi].v_ifp != ifp; vifi++)
1439 				;
1440 			if (vifi >= numvifs) /* vif not found, drop packet */
1441 				goto non_fatal;
1442 
1443 			/* no upcall, so make a new entry */
1444 			rt = (struct mfc *)malloc(sizeof(*rt), M_MRTABLE,
1445 						  M_NOWAIT);
1446 			if (rt == NULL)
1447 				goto fail;
1448 
1449 			/*
1450 			 * Make a copy of the header to send to the user level
1451 			 * process
1452 			 */
1453 			mm = m_copym(m, 0, hlen, M_DONTWAIT);
1454 			M_PULLUP(mm, hlen);
1455 			if (mm == NULL)
1456 				goto fail1;
1457 
1458 			/*
1459 			 * Send message to routing daemon to install
1460 			 * a route into the kernel table
1461 			 */
1462 
1463 			im = mtod(mm, struct igmpmsg *);
1464 			im->im_msgtype = IGMPMSG_NOCACHE;
1465 			im->im_mbz = 0;
1466 			im->im_vif = vifi;
1467 
1468 			mrtstat.mrts_upcalls++;
1469 
1470 			sockaddr_in_init(&sin, &ip->ip_src, 0);
1471 			if (socket_send(ip_mrouter, mm, &sin) < 0) {
1472 				log(LOG_WARNING,
1473 				    "ip_mforward: ip_mrouter socket queue full\n");
1474 				++mrtstat.mrts_upq_sockfull;
1475 			fail1:
1476 				free(rt, M_MRTABLE);
1477 			fail:
1478 				free(rte, M_MRTABLE);
1479 				m_freem(mb0);
1480 				splx(s);
1481 				return (ENOBUFS);
1482 			}
1483 
1484 			/* insert new entry at head of hash chain */
1485 			rt->mfc_origin = ip->ip_src;
1486 			rt->mfc_mcastgrp = ip->ip_dst;
1487 			rt->mfc_pkt_cnt = 0;
1488 			rt->mfc_byte_cnt = 0;
1489 			rt->mfc_wrong_if = 0;
1490 			rt->mfc_expire = UPCALL_EXPIRE;
1491 			nexpire[hash]++;
1492 			for (i = 0; i < numvifs; i++) {
1493 				rt->mfc_ttls[i] = 0;
1494 				rt->mfc_flags[i] = 0;
1495 			}
1496 			rt->mfc_parent = -1;
1497 
1498 			/* clear the RP address */
1499 			rt->mfc_rp = zeroin_addr;
1500 
1501 			rt->mfc_bw_meter = NULL;
1502 
1503 			/* link into table */
1504 			LIST_INSERT_HEAD(&mfchashtbl[hash], rt, mfc_hash);
1505 			/* Add this entry to the end of the queue */
1506 			rt->mfc_stall = rte;
1507 		} else {
1508 			/* determine if q has overflowed */
1509 			struct rtdetq **p;
1510 			int npkts = 0;
1511 
1512 			/*
1513 			 * XXX ouch! we need to append to the list, but we
1514 			 * only have a pointer to the front, so we have to
1515 			 * scan the entire list every time.
1516 			 */
1517 			for (p = &rt->mfc_stall; *p != NULL; p = &(*p)->next)
1518 				if (++npkts > MAX_UPQ) {
1519 					mrtstat.mrts_upq_ovflw++;
1520 				non_fatal:
1521 					free(rte, M_MRTABLE);
1522 					m_freem(mb0);
1523 					splx(s);
1524 					return (0);
1525 				}
1526 
1527 			/* Add this entry to the end of the queue */
1528 			*p = rte;
1529 		}
1530 
1531 		rte->next = NULL;
1532 		rte->m = mb0;
1533 		rte->ifp = ifp;
1534 #ifdef UPCALL_TIMING
1535 		rte->t = tp;
1536 #endif /* UPCALL_TIMING */
1537 
1538 		splx(s);
1539 
1540 		return (0);
1541 	}
1542 }
1543 
1544 
1545 /*ARGSUSED*/
1546 static void
1547 expire_upcalls(void *v)
1548 {
1549 	int i;
1550 	int s;
1551 
1552 	s = splsoftnet();
1553 
1554 	for (i = 0; i < MFCTBLSIZ; i++) {
1555 		struct mfc *rt, *nrt;
1556 
1557 		if (nexpire[i] == 0)
1558 			continue;
1559 
1560 		for (rt = LIST_FIRST(&mfchashtbl[i]); rt; rt = nrt) {
1561 			nrt = LIST_NEXT(rt, mfc_hash);
1562 
1563 			if (rt->mfc_expire == 0 || --rt->mfc_expire > 0)
1564 				continue;
1565 			nexpire[i]--;
1566 
1567 			/*
1568 			 * free the bw_meter entries
1569 			 */
1570 			while (rt->mfc_bw_meter != NULL) {
1571 				struct bw_meter *x = rt->mfc_bw_meter;
1572 
1573 				rt->mfc_bw_meter = x->bm_mfc_next;
1574 				free(x, M_BWMETER);
1575 			}
1576 
1577 			++mrtstat.mrts_cache_cleanups;
1578 			if (mrtdebug & DEBUG_EXPIRE)
1579 				log(LOG_DEBUG,
1580 				    "expire_upcalls: expiring (%x %x)\n",
1581 				    ntohl(rt->mfc_origin.s_addr),
1582 				    ntohl(rt->mfc_mcastgrp.s_addr));
1583 
1584 			expire_mfc(rt);
1585 		}
1586 	}
1587 
1588 	splx(s);
1589 	callout_reset(&expire_upcalls_ch, EXPIRE_TIMEOUT,
1590 	    expire_upcalls, NULL);
1591 }
1592 
1593 /*
1594  * Packet forwarding routine once entry in the cache is made
1595  */
1596 static int
1597 #ifdef RSVP_ISI
1598 ip_mdq(struct mbuf *m, struct ifnet *ifp, struct mfc *rt, vifi_t xmt_vif)
1599 #else
1600 ip_mdq(struct mbuf *m, struct ifnet *ifp, struct mfc *rt)
1601 #endif /* RSVP_ISI */
1602 {
1603 	struct ip  *ip = mtod(m, struct ip *);
1604 	vifi_t vifi;
1605 	struct vif *vifp;
1606 	struct sockaddr_in sin;
1607 	int plen = ntohs(ip->ip_len) - (ip->ip_hl << 2);
1608 
1609 /*
1610  * Macro to send packet on vif.  Since RSVP packets don't get counted on
1611  * input, they shouldn't get counted on output, so statistics keeping is
1612  * separate.
1613  */
1614 #define MC_SEND(ip, vifp, m) do {					\
1615 	if ((vifp)->v_flags & VIFF_TUNNEL)				\
1616 		encap_send((ip), (vifp), (m));				\
1617 	else								\
1618 		phyint_send((ip), (vifp), (m));				\
1619 } while (/*CONSTCOND*/ 0)
1620 
1621 #ifdef RSVP_ISI
1622 	/*
1623 	 * If xmt_vif is not -1, send on only the requested vif.
1624 	 *
1625 	 * (since vifi_t is u_short, -1 becomes MAXUSHORT, which > numvifs.
1626 	 */
1627 	if (xmt_vif < numvifs) {
1628 #ifdef PIM
1629 		if (viftable[xmt_vif].v_flags & VIFF_REGISTER)
1630 			pim_register_send(ip, viftable + xmt_vif, m, rt);
1631 		else
1632 #endif
1633 		MC_SEND(ip, viftable + xmt_vif, m);
1634 		return (1);
1635 	}
1636 #endif /* RSVP_ISI */
1637 
1638 	/*
1639 	 * Don't forward if it didn't arrive from the parent vif for its origin.
1640 	 */
1641 	vifi = rt->mfc_parent;
1642 	if ((vifi >= numvifs) || (viftable[vifi].v_ifp != ifp)) {
1643 		/* came in the wrong interface */
1644 		if (mrtdebug & DEBUG_FORWARD)
1645 			log(LOG_DEBUG, "wrong if: ifp %p vifi %d vififp %p\n",
1646 			    ifp, vifi,
1647 			    vifi >= numvifs ? 0 : viftable[vifi].v_ifp);
1648 		++mrtstat.mrts_wrong_if;
1649 		++rt->mfc_wrong_if;
1650 		/*
1651 		 * If we are doing PIM assert processing, send a message
1652 		 * to the routing daemon.
1653 		 *
1654 		 * XXX: A PIM-SM router needs the WRONGVIF detection so it
1655 		 * can complete the SPT switch, regardless of the type
1656 		 * of the iif (broadcast media, GRE tunnel, etc).
1657 		 */
1658 		if (pim_assert && (vifi < numvifs) && viftable[vifi].v_ifp) {
1659 			struct timeval now;
1660 			u_int32_t delta;
1661 
1662 #ifdef PIM
1663 			if (ifp == &multicast_register_if)
1664 				pimstat.pims_rcv_registers_wrongiif++;
1665 #endif
1666 
1667 			/* Get vifi for the incoming packet */
1668 			for (vifi = 0;
1669 			     vifi < numvifs && viftable[vifi].v_ifp != ifp;
1670 			     vifi++)
1671 			    ;
1672 			if (vifi >= numvifs) {
1673 				/* The iif is not found: ignore the packet. */
1674 				return (0);
1675 			}
1676 
1677 			if (rt->mfc_flags[vifi] &
1678 			    MRT_MFC_FLAGS_DISABLE_WRONGVIF) {
1679 				/* WRONGVIF disabled: ignore the packet */
1680 				return (0);
1681 			}
1682 
1683 			microtime(&now);
1684 
1685 			TV_DELTA(rt->mfc_last_assert, now, delta);
1686 
1687 			if (delta > ASSERT_MSG_TIME) {
1688 				struct igmpmsg *im;
1689 				int hlen = ip->ip_hl << 2;
1690 				struct mbuf *mm =
1691 				    m_copym(m, 0, hlen, M_DONTWAIT);
1692 
1693 				M_PULLUP(mm, hlen);
1694 				if (mm == NULL)
1695 					return (ENOBUFS);
1696 
1697 				rt->mfc_last_assert = now;
1698 
1699 				im = mtod(mm, struct igmpmsg *);
1700 				im->im_msgtype	= IGMPMSG_WRONGVIF;
1701 				im->im_mbz	= 0;
1702 				im->im_vif	= vifi;
1703 
1704 				mrtstat.mrts_upcalls++;
1705 
1706 				sockaddr_in_init(&sin, &im->im_src, 0);
1707 				if (socket_send(ip_mrouter, mm, &sin) < 0) {
1708 					log(LOG_WARNING,
1709 					    "ip_mforward: ip_mrouter socket queue full\n");
1710 					++mrtstat.mrts_upq_sockfull;
1711 					return (ENOBUFS);
1712 				}
1713 			}
1714 		}
1715 		return (0);
1716 	}
1717 
1718 	/* If I sourced this packet, it counts as output, else it was input. */
1719 	if (in_hosteq(ip->ip_src, viftable[vifi].v_lcl_addr)) {
1720 		viftable[vifi].v_pkt_out++;
1721 		viftable[vifi].v_bytes_out += plen;
1722 	} else {
1723 		viftable[vifi].v_pkt_in++;
1724 		viftable[vifi].v_bytes_in += plen;
1725 	}
1726 	rt->mfc_pkt_cnt++;
1727 	rt->mfc_byte_cnt += plen;
1728 
1729 	/*
1730 	 * For each vif, decide if a copy of the packet should be forwarded.
1731 	 * Forward if:
1732 	 *		- the ttl exceeds the vif's threshold
1733 	 *		- there are group members downstream on interface
1734 	 */
1735 	for (vifp = viftable, vifi = 0; vifi < numvifs; vifp++, vifi++)
1736 		if ((rt->mfc_ttls[vifi] > 0) &&
1737 			(ip->ip_ttl > rt->mfc_ttls[vifi])) {
1738 			vifp->v_pkt_out++;
1739 			vifp->v_bytes_out += plen;
1740 #ifdef PIM
1741 			if (vifp->v_flags & VIFF_REGISTER)
1742 				pim_register_send(ip, vifp, m, rt);
1743 			else
1744 #endif
1745 			MC_SEND(ip, vifp, m);
1746 		}
1747 
1748 	/*
1749 	 * Perform upcall-related bw measuring.
1750 	 */
1751 	if (rt->mfc_bw_meter != NULL) {
1752 		struct bw_meter *x;
1753 		struct timeval now;
1754 
1755 		microtime(&now);
1756 		for (x = rt->mfc_bw_meter; x != NULL; x = x->bm_mfc_next)
1757 			bw_meter_receive_packet(x, plen, &now);
1758 	}
1759 
1760 	return (0);
1761 }
1762 
1763 #ifdef RSVP_ISI
1764 /*
1765  * check if a vif number is legal/ok. This is used by ip_output.
1766  */
1767 int
1768 legal_vif_num(int vif)
1769 {
1770 	if (vif >= 0 && vif < numvifs)
1771 		return (1);
1772 	else
1773 		return (0);
1774 }
1775 #endif /* RSVP_ISI */
1776 
1777 static void
1778 phyint_send(struct ip *ip, struct vif *vifp, struct mbuf *m)
1779 {
1780 	struct mbuf *mb_copy;
1781 	int hlen = ip->ip_hl << 2;
1782 
1783 	/*
1784 	 * Make a new reference to the packet; make sure that
1785 	 * the IP header is actually copied, not just referenced,
1786 	 * so that ip_output() only scribbles on the copy.
1787 	 */
1788 	mb_copy = m_copypacket(m, M_DONTWAIT);
1789 	M_PULLUP(mb_copy, hlen);
1790 	if (mb_copy == NULL)
1791 		return;
1792 
1793 	if (vifp->v_rate_limit <= 0)
1794 		tbf_send_packet(vifp, mb_copy);
1795 	else
1796 		tbf_control(vifp, mb_copy, mtod(mb_copy, struct ip *),
1797 		    ntohs(ip->ip_len));
1798 }
1799 
1800 static void
1801 encap_send(struct ip *ip, struct vif *vifp, struct mbuf *m)
1802 {
1803 	struct mbuf *mb_copy;
1804 	struct ip *ip_copy;
1805 	int i, len = ntohs(ip->ip_len) + sizeof(multicast_encap_iphdr);
1806 
1807 	/* Take care of delayed checksums */
1808 	if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
1809 		in_delayed_cksum(m);
1810 		m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
1811 	}
1812 
1813 	/*
1814 	 * copy the old packet & pullup it's IP header into the
1815 	 * new mbuf so we can modify it.  Try to fill the new
1816 	 * mbuf since if we don't the ethernet driver will.
1817 	 */
1818 	MGETHDR(mb_copy, M_DONTWAIT, MT_DATA);
1819 	if (mb_copy == NULL)
1820 		return;
1821 	mb_copy->m_data += max_linkhdr;
1822 	mb_copy->m_pkthdr.len = len;
1823 	mb_copy->m_len = sizeof(multicast_encap_iphdr);
1824 
1825 	if ((mb_copy->m_next = m_copypacket(m, M_DONTWAIT)) == NULL) {
1826 		m_freem(mb_copy);
1827 		return;
1828 	}
1829 	i = MHLEN - max_linkhdr;
1830 	if (i > len)
1831 		i = len;
1832 	mb_copy = m_pullup(mb_copy, i);
1833 	if (mb_copy == NULL)
1834 		return;
1835 
1836 	/*
1837 	 * fill in the encapsulating IP header.
1838 	 */
1839 	ip_copy = mtod(mb_copy, struct ip *);
1840 	*ip_copy = multicast_encap_iphdr;
1841 	if (len < IP_MINFRAGSIZE)
1842 		ip_copy->ip_id = 0;
1843 	else
1844 		ip_copy->ip_id = ip_newid(NULL);
1845 	ip_copy->ip_len = htons(len);
1846 	ip_copy->ip_src = vifp->v_lcl_addr;
1847 	ip_copy->ip_dst = vifp->v_rmt_addr;
1848 
1849 	/*
1850 	 * turn the encapsulated IP header back into a valid one.
1851 	 */
1852 	ip = (struct ip *)((char *)ip_copy + sizeof(multicast_encap_iphdr));
1853 	--ip->ip_ttl;
1854 	ip->ip_sum = 0;
1855 	mb_copy->m_data += sizeof(multicast_encap_iphdr);
1856 	ip->ip_sum = in_cksum(mb_copy, ip->ip_hl << 2);
1857 	mb_copy->m_data -= sizeof(multicast_encap_iphdr);
1858 
1859 	if (vifp->v_rate_limit <= 0)
1860 		tbf_send_packet(vifp, mb_copy);
1861 	else
1862 		tbf_control(vifp, mb_copy, ip, ntohs(ip_copy->ip_len));
1863 }
1864 
1865 /*
1866  * De-encapsulate a packet and feed it back through ip input.
1867  */
1868 static void
1869 vif_input(struct mbuf *m, ...)
1870 {
1871 	int off, proto;
1872 	va_list ap;
1873 	struct vif *vifp;
1874 	int s;
1875 	struct ifqueue *ifq;
1876 
1877 	va_start(ap, m);
1878 	off = va_arg(ap, int);
1879 	proto = va_arg(ap, int);
1880 	va_end(ap);
1881 
1882 	vifp = (struct vif *)encap_getarg(m);
1883 	if (!vifp || proto != ENCAP_PROTO) {
1884 		m_freem(m);
1885 		mrtstat.mrts_bad_tunnel++;
1886 		return;
1887 	}
1888 
1889 	m_adj(m, off);
1890 	m->m_pkthdr.rcvif = vifp->v_ifp;
1891 	ifq = &ipintrq;
1892 	s = splnet();
1893 	if (IF_QFULL(ifq)) {
1894 		IF_DROP(ifq);
1895 		m_freem(m);
1896 	} else {
1897 		IF_ENQUEUE(ifq, m);
1898 		/*
1899 		 * normally we would need a "schednetisr(NETISR_IP)"
1900 		 * here but we were called by ip_input and it is going
1901 		 * to loop back & try to dequeue the packet we just
1902 		 * queued as soon as we return so we avoid the
1903 		 * unnecessary software interrrupt.
1904 		 */
1905 	}
1906 	splx(s);
1907 }
1908 
1909 /*
1910  * Check if the packet should be received on the vif denoted by arg.
1911  * (The encap selection code will call this once per vif since each is
1912  * registered separately.)
1913  */
1914 static int
1915 vif_encapcheck(struct mbuf *m, int off, int proto, void *arg)
1916 {
1917 	struct vif *vifp;
1918 	struct ip ip;
1919 
1920 #ifdef DIAGNOSTIC
1921 	if (!arg || proto != IPPROTO_IPV4)
1922 		panic("unexpected arg in vif_encapcheck");
1923 #endif
1924 
1925 	/*
1926 	 * Accept the packet only if the inner heaader is multicast
1927 	 * and the outer header matches a tunnel-mode vif.  Order
1928 	 * checks in the hope that common non-matching packets will be
1929 	 * rejected quickly.  Assume that unicast IPv4 traffic in a
1930 	 * parallel tunnel (e.g. gif(4)) is unlikely.
1931 	 */
1932 
1933 	/* Obtain the outer IP header and the vif pointer. */
1934 	m_copydata((struct mbuf *)m, 0, sizeof(ip), (void *)&ip);
1935 	vifp = (struct vif *)arg;
1936 
1937 	/*
1938 	 * The outer source must match the vif's remote peer address.
1939 	 * For a multicast router with several tunnels, this is the
1940 	 * only check that will fail on packets in other tunnels,
1941 	 * assuming the local address is the same.
1942 	 */
1943 	if (!in_hosteq(vifp->v_rmt_addr, ip.ip_src))
1944 		return 0;
1945 
1946 	/* The outer destination must match the vif's local address. */
1947 	if (!in_hosteq(vifp->v_lcl_addr, ip.ip_dst))
1948 		return 0;
1949 
1950 	/* The vif must be of tunnel type. */
1951 	if ((vifp->v_flags & VIFF_TUNNEL) == 0)
1952 		return 0;
1953 
1954 	/* Check that the inner destination is multicast. */
1955 	m_copydata((struct mbuf *)m, off, sizeof(ip), (void *)&ip);
1956 	if (!IN_MULTICAST(ip.ip_dst.s_addr))
1957 		return 0;
1958 
1959 	/*
1960 	 * We have checked that both the outer src and dst addresses
1961 	 * match the vif, and that the inner destination is multicast
1962 	 * (224/5).  By claiming more than 64, we intend to
1963 	 * preferentially take packets that also match a parallel
1964 	 * gif(4).
1965 	 */
1966 	return 32 + 32 + 5;
1967 }
1968 
1969 /*
1970  * Token bucket filter module
1971  */
1972 static void
1973 tbf_control(struct vif *vifp, struct mbuf *m, struct ip *ip, u_int32_t len)
1974 {
1975 
1976 	if (len > MAX_BKT_SIZE) {
1977 		/* drop if packet is too large */
1978 		mrtstat.mrts_pkt2large++;
1979 		m_freem(m);
1980 		return;
1981 	}
1982 
1983 	tbf_update_tokens(vifp);
1984 
1985 	/*
1986 	 * If there are enough tokens, and the queue is empty, send this packet
1987 	 * out immediately.  Otherwise, try to insert it on this vif's queue.
1988 	 */
1989 	if (vifp->tbf_q_len == 0) {
1990 		if (len <= vifp->tbf_n_tok) {
1991 			vifp->tbf_n_tok -= len;
1992 			tbf_send_packet(vifp, m);
1993 		} else {
1994 			/* queue packet and timeout till later */
1995 			tbf_queue(vifp, m);
1996 			callout_reset(&vifp->v_repq_ch, TBF_REPROCESS,
1997 			    tbf_reprocess_q, vifp);
1998 		}
1999 	} else {
2000 		if (vifp->tbf_q_len >= vifp->tbf_max_q_len &&
2001 		    !tbf_dq_sel(vifp, ip)) {
2002 			/* queue full, and couldn't make room */
2003 			mrtstat.mrts_q_overflow++;
2004 			m_freem(m);
2005 		} else {
2006 			/* queue length low enough, or made room */
2007 			tbf_queue(vifp, m);
2008 			tbf_process_q(vifp);
2009 		}
2010 	}
2011 }
2012 
2013 /*
2014  * adds a packet to the queue at the interface
2015  */
2016 static void
2017 tbf_queue(struct vif *vifp, struct mbuf *m)
2018 {
2019 	int s = splsoftnet();
2020 
2021 	/* insert at tail */
2022 	*vifp->tbf_t = m;
2023 	vifp->tbf_t = &m->m_nextpkt;
2024 	vifp->tbf_q_len++;
2025 
2026 	splx(s);
2027 }
2028 
2029 
2030 /*
2031  * processes the queue at the interface
2032  */
2033 static void
2034 tbf_process_q(struct vif *vifp)
2035 {
2036 	struct mbuf *m;
2037 	int len;
2038 	int s = splsoftnet();
2039 
2040 	/*
2041 	 * Loop through the queue at the interface and send as many packets
2042 	 * as possible.
2043 	 */
2044 	for (m = vifp->tbf_q; m != NULL; m = vifp->tbf_q) {
2045 		len = ntohs(mtod(m, struct ip *)->ip_len);
2046 
2047 		/* determine if the packet can be sent */
2048 		if (len <= vifp->tbf_n_tok) {
2049 			/* if so,
2050 			 * reduce no of tokens, dequeue the packet,
2051 			 * send the packet.
2052 			 */
2053 			if ((vifp->tbf_q = m->m_nextpkt) == NULL)
2054 				vifp->tbf_t = &vifp->tbf_q;
2055 			--vifp->tbf_q_len;
2056 
2057 			m->m_nextpkt = NULL;
2058 			vifp->tbf_n_tok -= len;
2059 			tbf_send_packet(vifp, m);
2060 		} else
2061 			break;
2062 	}
2063 	splx(s);
2064 }
2065 
2066 static void
2067 tbf_reprocess_q(void *arg)
2068 {
2069 	struct vif *vifp = arg;
2070 
2071 	if (ip_mrouter == NULL)
2072 		return;
2073 
2074 	tbf_update_tokens(vifp);
2075 	tbf_process_q(vifp);
2076 
2077 	if (vifp->tbf_q_len != 0)
2078 		callout_reset(&vifp->v_repq_ch, TBF_REPROCESS,
2079 		    tbf_reprocess_q, vifp);
2080 }
2081 
2082 /* function that will selectively discard a member of the queue
2083  * based on the precedence value and the priority
2084  */
2085 static int
2086 tbf_dq_sel(struct vif *vifp, struct ip *ip)
2087 {
2088 	u_int p;
2089 	struct mbuf **mp, *m;
2090 	int s = splsoftnet();
2091 
2092 	p = priority(vifp, ip);
2093 
2094 	for (mp = &vifp->tbf_q, m = *mp;
2095 	    m != NULL;
2096 	    mp = &m->m_nextpkt, m = *mp) {
2097 		if (p > priority(vifp, mtod(m, struct ip *))) {
2098 			if ((*mp = m->m_nextpkt) == NULL)
2099 				vifp->tbf_t = mp;
2100 			--vifp->tbf_q_len;
2101 
2102 			m_freem(m);
2103 			mrtstat.mrts_drop_sel++;
2104 			splx(s);
2105 			return (1);
2106 		}
2107 	}
2108 	splx(s);
2109 	return (0);
2110 }
2111 
2112 static void
2113 tbf_send_packet(struct vif *vifp, struct mbuf *m)
2114 {
2115 	int error;
2116 	int s = splsoftnet();
2117 
2118 	if (vifp->v_flags & VIFF_TUNNEL) {
2119 		/* If tunnel options */
2120 		ip_output(m, NULL, &vifp->v_route, IP_FORWARDING, NULL, NULL);
2121 	} else {
2122 		/* if physical interface option, extract the options and then send */
2123 		struct ip_moptions imo;
2124 
2125 		imo.imo_multicast_ifp = vifp->v_ifp;
2126 		imo.imo_multicast_ttl = mtod(m, struct ip *)->ip_ttl - 1;
2127 		imo.imo_multicast_loop = 1;
2128 #ifdef RSVP_ISI
2129 		imo.imo_multicast_vif = -1;
2130 #endif
2131 
2132 		error = ip_output(m, NULL, NULL, IP_FORWARDING|IP_MULTICASTOPTS,
2133 		    &imo, NULL);
2134 
2135 		if (mrtdebug & DEBUG_XMIT)
2136 			log(LOG_DEBUG, "phyint_send on vif %ld err %d\n",
2137 			    (long)(vifp - viftable), error);
2138 	}
2139 	splx(s);
2140 }
2141 
2142 /* determine the current time and then
2143  * the elapsed time (between the last time and time now)
2144  * in milliseconds & update the no. of tokens in the bucket
2145  */
2146 static void
2147 tbf_update_tokens(struct vif *vifp)
2148 {
2149 	struct timeval tp;
2150 	u_int32_t tm;
2151 	int s = splsoftnet();
2152 
2153 	microtime(&tp);
2154 
2155 	TV_DELTA(tp, vifp->tbf_last_pkt_t, tm);
2156 
2157 	/*
2158 	 * This formula is actually
2159 	 * "time in seconds" * "bytes/second".
2160 	 *
2161 	 * (tm / 1000000) * (v_rate_limit * 1000 * (1000/1024) / 8)
2162 	 *
2163 	 * The (1000/1024) was introduced in add_vif to optimize
2164 	 * this divide into a shift.
2165 	 */
2166 	vifp->tbf_n_tok += tm * vifp->v_rate_limit / 8192;
2167 	vifp->tbf_last_pkt_t = tp;
2168 
2169 	if (vifp->tbf_n_tok > MAX_BKT_SIZE)
2170 		vifp->tbf_n_tok = MAX_BKT_SIZE;
2171 
2172 	splx(s);
2173 }
2174 
2175 static int
2176 priority(struct vif *vifp, struct ip *ip)
2177 {
2178 	int prio = 50;	/* the lowest priority -- default case */
2179 
2180 	/* temporary hack; may add general packet classifier some day */
2181 
2182 	/*
2183 	 * The UDP port space is divided up into four priority ranges:
2184 	 * [0, 16384)     : unclassified - lowest priority
2185 	 * [16384, 32768) : audio - highest priority
2186 	 * [32768, 49152) : whiteboard - medium priority
2187 	 * [49152, 65536) : video - low priority
2188 	 */
2189 	if (ip->ip_p == IPPROTO_UDP) {
2190 		struct udphdr *udp = (struct udphdr *)(((char *)ip) + (ip->ip_hl << 2));
2191 
2192 		switch (ntohs(udp->uh_dport) & 0xc000) {
2193 		case 0x4000:
2194 			prio = 70;
2195 			break;
2196 		case 0x8000:
2197 			prio = 60;
2198 			break;
2199 		case 0xc000:
2200 			prio = 55;
2201 			break;
2202 		}
2203 
2204 		if (tbfdebug > 1)
2205 			log(LOG_DEBUG, "port %x prio %d\n",
2206 			    ntohs(udp->uh_dport), prio);
2207 	}
2208 
2209 	return (prio);
2210 }
2211 
2212 /*
2213  * End of token bucket filter modifications
2214  */
2215 #ifdef RSVP_ISI
2216 int
2217 ip_rsvp_vif_init(struct socket *so, struct mbuf *m)
2218 {
2219 	int vifi, s;
2220 
2221 	if (rsvpdebug)
2222 		printf("ip_rsvp_vif_init: so_type = %d, pr_protocol = %d\n",
2223 		    so->so_type, so->so_proto->pr_protocol);
2224 
2225 	if (so->so_type != SOCK_RAW ||
2226 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
2227 		return (EOPNOTSUPP);
2228 
2229 	/* Check mbuf. */
2230 	if (m == NULL || m->m_len != sizeof(int)) {
2231 		return (EINVAL);
2232 	}
2233 	vifi = *(mtod(m, int *));
2234 
2235 	if (rsvpdebug)
2236 		printf("ip_rsvp_vif_init: vif = %d rsvp_on = %d\n",
2237 		       vifi, rsvp_on);
2238 
2239 	s = splsoftnet();
2240 
2241 	/* Check vif. */
2242 	if (!legal_vif_num(vifi)) {
2243 		splx(s);
2244 		return (EADDRNOTAVAIL);
2245 	}
2246 
2247 	/* Check if socket is available. */
2248 	if (viftable[vifi].v_rsvpd != NULL) {
2249 		splx(s);
2250 		return (EADDRINUSE);
2251 	}
2252 
2253 	viftable[vifi].v_rsvpd = so;
2254 	/*
2255 	 * This may seem silly, but we need to be sure we don't over-increment
2256 	 * the RSVP counter, in case something slips up.
2257 	 */
2258 	if (!viftable[vifi].v_rsvp_on) {
2259 		viftable[vifi].v_rsvp_on = 1;
2260 		rsvp_on++;
2261 	}
2262 
2263 	splx(s);
2264 	return (0);
2265 }
2266 
2267 int
2268 ip_rsvp_vif_done(struct socket *so, struct mbuf *m)
2269 {
2270 	int vifi, s;
2271 
2272 	if (rsvpdebug)
2273 		printf("ip_rsvp_vif_done: so_type = %d, pr_protocol = %d\n",
2274 		    so->so_type, so->so_proto->pr_protocol);
2275 
2276 	if (so->so_type != SOCK_RAW ||
2277 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
2278 		return (EOPNOTSUPP);
2279 
2280 	/* Check mbuf. */
2281 	if (m == NULL || m->m_len != sizeof(int)) {
2282 		return (EINVAL);
2283 	}
2284 	vifi = *(mtod(m, int *));
2285 
2286 	s = splsoftnet();
2287 
2288 	/* Check vif. */
2289 	if (!legal_vif_num(vifi)) {
2290 		splx(s);
2291 		return (EADDRNOTAVAIL);
2292 	}
2293 
2294 	if (rsvpdebug)
2295 		printf("ip_rsvp_vif_done: v_rsvpd = %x so = %x\n",
2296 		    viftable[vifi].v_rsvpd, so);
2297 
2298 	viftable[vifi].v_rsvpd = NULL;
2299 	/*
2300 	 * This may seem silly, but we need to be sure we don't over-decrement
2301 	 * the RSVP counter, in case something slips up.
2302 	 */
2303 	if (viftable[vifi].v_rsvp_on) {
2304 		viftable[vifi].v_rsvp_on = 0;
2305 		rsvp_on--;
2306 	}
2307 
2308 	splx(s);
2309 	return (0);
2310 }
2311 
2312 void
2313 ip_rsvp_force_done(struct socket *so)
2314 {
2315 	int vifi, s;
2316 
2317 	/* Don't bother if it is not the right type of socket. */
2318 	if (so->so_type != SOCK_RAW ||
2319 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
2320 		return;
2321 
2322 	s = splsoftnet();
2323 
2324 	/*
2325 	 * The socket may be attached to more than one vif...this
2326 	 * is perfectly legal.
2327 	 */
2328 	for (vifi = 0; vifi < numvifs; vifi++) {
2329 		if (viftable[vifi].v_rsvpd == so) {
2330 			viftable[vifi].v_rsvpd = NULL;
2331 			/*
2332 			 * This may seem silly, but we need to be sure we don't
2333 			 * over-decrement the RSVP counter, in case something
2334 			 * slips up.
2335 			 */
2336 			if (viftable[vifi].v_rsvp_on) {
2337 				viftable[vifi].v_rsvp_on = 0;
2338 				rsvp_on--;
2339 			}
2340 		}
2341 	}
2342 
2343 	splx(s);
2344 	return;
2345 }
2346 
2347 void
2348 rsvp_input(struct mbuf *m, struct ifnet *ifp)
2349 {
2350 	int vifi, s;
2351 	struct ip *ip = mtod(m, struct ip *);
2352 	struct sockaddr_in rsvp_src;
2353 
2354 	if (rsvpdebug)
2355 		printf("rsvp_input: rsvp_on %d\n", rsvp_on);
2356 
2357 	/*
2358 	 * Can still get packets with rsvp_on = 0 if there is a local member
2359 	 * of the group to which the RSVP packet is addressed.  But in this
2360 	 * case we want to throw the packet away.
2361 	 */
2362 	if (!rsvp_on) {
2363 		m_freem(m);
2364 		return;
2365 	}
2366 
2367 	/*
2368 	 * If the old-style non-vif-associated socket is set, then use
2369 	 * it and ignore the new ones.
2370 	 */
2371 	if (ip_rsvpd != NULL) {
2372 		if (rsvpdebug)
2373 			printf("rsvp_input: "
2374 			    "Sending packet up old-style socket\n");
2375 		rip_input(m);	/*XXX*/
2376 		return;
2377 	}
2378 
2379 	s = splsoftnet();
2380 
2381 	if (rsvpdebug)
2382 		printf("rsvp_input: check vifs\n");
2383 
2384 	/* Find which vif the packet arrived on. */
2385 	for (vifi = 0; vifi < numvifs; vifi++) {
2386 		if (viftable[vifi].v_ifp == ifp)
2387 			break;
2388 	}
2389 
2390 	if (vifi == numvifs) {
2391 		/* Can't find vif packet arrived on. Drop packet. */
2392 		if (rsvpdebug)
2393 			printf("rsvp_input: "
2394 			    "Can't find vif for packet...dropping it.\n");
2395 		m_freem(m);
2396 		splx(s);
2397 		return;
2398 	}
2399 
2400 	if (rsvpdebug)
2401 		printf("rsvp_input: check socket\n");
2402 
2403 	if (viftable[vifi].v_rsvpd == NULL) {
2404 		/*
2405 		 * drop packet, since there is no specific socket for this
2406 		 * interface
2407 		 */
2408 		if (rsvpdebug)
2409 			printf("rsvp_input: No socket defined for vif %d\n",
2410 			    vifi);
2411 		m_freem(m);
2412 		splx(s);
2413 		return;
2414 	}
2415 
2416 	sockaddr_in_init(&rsvp_src, &ip->ip_src, 0);
2417 
2418 	if (rsvpdebug && m)
2419 		printf("rsvp_input: m->m_len = %d, sbspace() = %d\n",
2420 		    m->m_len, sbspace(&viftable[vifi].v_rsvpd->so_rcv));
2421 
2422 	if (socket_send(viftable[vifi].v_rsvpd, m, &rsvp_src) < 0)
2423 		if (rsvpdebug)
2424 			printf("rsvp_input: Failed to append to socket\n");
2425 	else
2426 		if (rsvpdebug)
2427 			printf("rsvp_input: send packet up\n");
2428 
2429 	splx(s);
2430 }
2431 #endif /* RSVP_ISI */
2432 
2433 /*
2434  * Code for bandwidth monitors
2435  */
2436 
2437 /*
2438  * Define common interface for timeval-related methods
2439  */
2440 #define	BW_TIMEVALCMP(tvp, uvp, cmp) timercmp((tvp), (uvp), cmp)
2441 #define	BW_TIMEVALDECR(vvp, uvp) timersub((vvp), (uvp), (vvp))
2442 #define	BW_TIMEVALADD(vvp, uvp) timeradd((vvp), (uvp), (vvp))
2443 
2444 static uint32_t
2445 compute_bw_meter_flags(struct bw_upcall *req)
2446 {
2447     uint32_t flags = 0;
2448 
2449     if (req->bu_flags & BW_UPCALL_UNIT_PACKETS)
2450 	flags |= BW_METER_UNIT_PACKETS;
2451     if (req->bu_flags & BW_UPCALL_UNIT_BYTES)
2452 	flags |= BW_METER_UNIT_BYTES;
2453     if (req->bu_flags & BW_UPCALL_GEQ)
2454 	flags |= BW_METER_GEQ;
2455     if (req->bu_flags & BW_UPCALL_LEQ)
2456 	flags |= BW_METER_LEQ;
2457 
2458     return flags;
2459 }
2460 
2461 /*
2462  * Add a bw_meter entry
2463  */
2464 static int
2465 add_bw_upcall(struct bw_upcall *req)
2466 {
2467     int s;
2468     struct mfc *mfc;
2469     struct timeval delta = { BW_UPCALL_THRESHOLD_INTERVAL_MIN_SEC,
2470 		BW_UPCALL_THRESHOLD_INTERVAL_MIN_USEC };
2471     struct timeval now;
2472     struct bw_meter *x;
2473     uint32_t flags;
2474 
2475     if (!(mrt_api_config & MRT_MFC_BW_UPCALL))
2476 	return EOPNOTSUPP;
2477 
2478     /* Test if the flags are valid */
2479     if (!(req->bu_flags & (BW_UPCALL_UNIT_PACKETS | BW_UPCALL_UNIT_BYTES)))
2480 	return EINVAL;
2481     if (!(req->bu_flags & (BW_UPCALL_GEQ | BW_UPCALL_LEQ)))
2482 	return EINVAL;
2483     if ((req->bu_flags & (BW_UPCALL_GEQ | BW_UPCALL_LEQ))
2484 	    == (BW_UPCALL_GEQ | BW_UPCALL_LEQ))
2485 	return EINVAL;
2486 
2487     /* Test if the threshold time interval is valid */
2488     if (BW_TIMEVALCMP(&req->bu_threshold.b_time, &delta, <))
2489 	return EINVAL;
2490 
2491     flags = compute_bw_meter_flags(req);
2492 
2493     /*
2494      * Find if we have already same bw_meter entry
2495      */
2496     s = splsoftnet();
2497     mfc = mfc_find(&req->bu_src, &req->bu_dst);
2498     if (mfc == NULL) {
2499 	splx(s);
2500 	return EADDRNOTAVAIL;
2501     }
2502     for (x = mfc->mfc_bw_meter; x != NULL; x = x->bm_mfc_next) {
2503 	if ((BW_TIMEVALCMP(&x->bm_threshold.b_time,
2504 			   &req->bu_threshold.b_time, ==)) &&
2505 	    (x->bm_threshold.b_packets == req->bu_threshold.b_packets) &&
2506 	    (x->bm_threshold.b_bytes == req->bu_threshold.b_bytes) &&
2507 	    (x->bm_flags & BW_METER_USER_FLAGS) == flags)  {
2508 	    splx(s);
2509 	    return 0;		/* XXX Already installed */
2510 	}
2511     }
2512 
2513     /* Allocate the new bw_meter entry */
2514     x = (struct bw_meter *)malloc(sizeof(*x), M_BWMETER, M_NOWAIT);
2515     if (x == NULL) {
2516 	splx(s);
2517 	return ENOBUFS;
2518     }
2519 
2520     /* Set the new bw_meter entry */
2521     x->bm_threshold.b_time = req->bu_threshold.b_time;
2522     microtime(&now);
2523     x->bm_start_time = now;
2524     x->bm_threshold.b_packets = req->bu_threshold.b_packets;
2525     x->bm_threshold.b_bytes = req->bu_threshold.b_bytes;
2526     x->bm_measured.b_packets = 0;
2527     x->bm_measured.b_bytes = 0;
2528     x->bm_flags = flags;
2529     x->bm_time_next = NULL;
2530     x->bm_time_hash = BW_METER_BUCKETS;
2531 
2532     /* Add the new bw_meter entry to the front of entries for this MFC */
2533     x->bm_mfc = mfc;
2534     x->bm_mfc_next = mfc->mfc_bw_meter;
2535     mfc->mfc_bw_meter = x;
2536     schedule_bw_meter(x, &now);
2537     splx(s);
2538 
2539     return 0;
2540 }
2541 
2542 static void
2543 free_bw_list(struct bw_meter *list)
2544 {
2545     while (list != NULL) {
2546 	struct bw_meter *x = list;
2547 
2548 	list = list->bm_mfc_next;
2549 	unschedule_bw_meter(x);
2550 	free(x, M_BWMETER);
2551     }
2552 }
2553 
2554 /*
2555  * Delete one or multiple bw_meter entries
2556  */
2557 static int
2558 del_bw_upcall(struct bw_upcall *req)
2559 {
2560     int s;
2561     struct mfc *mfc;
2562     struct bw_meter *x;
2563 
2564     if (!(mrt_api_config & MRT_MFC_BW_UPCALL))
2565 	return EOPNOTSUPP;
2566 
2567     s = splsoftnet();
2568     /* Find the corresponding MFC entry */
2569     mfc = mfc_find(&req->bu_src, &req->bu_dst);
2570     if (mfc == NULL) {
2571 	splx(s);
2572 	return EADDRNOTAVAIL;
2573     } else if (req->bu_flags & BW_UPCALL_DELETE_ALL) {
2574 	/*
2575 	 * Delete all bw_meter entries for this mfc
2576 	 */
2577 	struct bw_meter *list;
2578 
2579 	list = mfc->mfc_bw_meter;
2580 	mfc->mfc_bw_meter = NULL;
2581 	free_bw_list(list);
2582 	splx(s);
2583 	return 0;
2584     } else {			/* Delete a single bw_meter entry */
2585 	struct bw_meter *prev;
2586 	uint32_t flags = 0;
2587 
2588 	flags = compute_bw_meter_flags(req);
2589 
2590 	/* Find the bw_meter entry to delete */
2591 	for (prev = NULL, x = mfc->mfc_bw_meter; x != NULL;
2592 	     prev = x, x = x->bm_mfc_next) {
2593 	    if ((BW_TIMEVALCMP(&x->bm_threshold.b_time,
2594 			       &req->bu_threshold.b_time, ==)) &&
2595 		(x->bm_threshold.b_packets == req->bu_threshold.b_packets) &&
2596 		(x->bm_threshold.b_bytes == req->bu_threshold.b_bytes) &&
2597 		(x->bm_flags & BW_METER_USER_FLAGS) == flags)
2598 		break;
2599 	}
2600 	if (x != NULL) { /* Delete entry from the list for this MFC */
2601 	    if (prev != NULL)
2602 		prev->bm_mfc_next = x->bm_mfc_next;	/* remove from middle*/
2603 	    else
2604 		x->bm_mfc->mfc_bw_meter = x->bm_mfc_next;/* new head of list */
2605 
2606 	    unschedule_bw_meter(x);
2607 	    splx(s);
2608 	    /* Free the bw_meter entry */
2609 	    free(x, M_BWMETER);
2610 	    return 0;
2611 	} else {
2612 	    splx(s);
2613 	    return EINVAL;
2614 	}
2615     }
2616     /* NOTREACHED */
2617 }
2618 
2619 /*
2620  * Perform bandwidth measurement processing that may result in an upcall
2621  */
2622 static void
2623 bw_meter_receive_packet(struct bw_meter *x, int plen, struct timeval *nowp)
2624 {
2625     struct timeval delta;
2626 
2627     delta = *nowp;
2628     BW_TIMEVALDECR(&delta, &x->bm_start_time);
2629 
2630     if (x->bm_flags & BW_METER_GEQ) {
2631 	/*
2632 	 * Processing for ">=" type of bw_meter entry
2633 	 */
2634 	if (BW_TIMEVALCMP(&delta, &x->bm_threshold.b_time, >)) {
2635 	    /* Reset the bw_meter entry */
2636 	    x->bm_start_time = *nowp;
2637 	    x->bm_measured.b_packets = 0;
2638 	    x->bm_measured.b_bytes = 0;
2639 	    x->bm_flags &= ~BW_METER_UPCALL_DELIVERED;
2640 	}
2641 
2642 	/* Record that a packet is received */
2643 	x->bm_measured.b_packets++;
2644 	x->bm_measured.b_bytes += plen;
2645 
2646 	/*
2647 	 * Test if we should deliver an upcall
2648 	 */
2649 	if (!(x->bm_flags & BW_METER_UPCALL_DELIVERED)) {
2650 	    if (((x->bm_flags & BW_METER_UNIT_PACKETS) &&
2651 		 (x->bm_measured.b_packets >= x->bm_threshold.b_packets)) ||
2652 		((x->bm_flags & BW_METER_UNIT_BYTES) &&
2653 		 (x->bm_measured.b_bytes >= x->bm_threshold.b_bytes))) {
2654 		/* Prepare an upcall for delivery */
2655 		bw_meter_prepare_upcall(x, nowp);
2656 		x->bm_flags |= BW_METER_UPCALL_DELIVERED;
2657 	    }
2658 	}
2659     } else if (x->bm_flags & BW_METER_LEQ) {
2660 	/*
2661 	 * Processing for "<=" type of bw_meter entry
2662 	 */
2663 	if (BW_TIMEVALCMP(&delta, &x->bm_threshold.b_time, >)) {
2664 	    /*
2665 	     * We are behind time with the multicast forwarding table
2666 	     * scanning for "<=" type of bw_meter entries, so test now
2667 	     * if we should deliver an upcall.
2668 	     */
2669 	    if (((x->bm_flags & BW_METER_UNIT_PACKETS) &&
2670 		 (x->bm_measured.b_packets <= x->bm_threshold.b_packets)) ||
2671 		((x->bm_flags & BW_METER_UNIT_BYTES) &&
2672 		 (x->bm_measured.b_bytes <= x->bm_threshold.b_bytes))) {
2673 		/* Prepare an upcall for delivery */
2674 		bw_meter_prepare_upcall(x, nowp);
2675 	    }
2676 	    /* Reschedule the bw_meter entry */
2677 	    unschedule_bw_meter(x);
2678 	    schedule_bw_meter(x, nowp);
2679 	}
2680 
2681 	/* Record that a packet is received */
2682 	x->bm_measured.b_packets++;
2683 	x->bm_measured.b_bytes += plen;
2684 
2685 	/*
2686 	 * Test if we should restart the measuring interval
2687 	 */
2688 	if ((x->bm_flags & BW_METER_UNIT_PACKETS &&
2689 	     x->bm_measured.b_packets <= x->bm_threshold.b_packets) ||
2690 	    (x->bm_flags & BW_METER_UNIT_BYTES &&
2691 	     x->bm_measured.b_bytes <= x->bm_threshold.b_bytes)) {
2692 	    /* Don't restart the measuring interval */
2693 	} else {
2694 	    /* Do restart the measuring interval */
2695 	    /*
2696 	     * XXX: note that we don't unschedule and schedule, because this
2697 	     * might be too much overhead per packet. Instead, when we process
2698 	     * all entries for a given timer hash bin, we check whether it is
2699 	     * really a timeout. If not, we reschedule at that time.
2700 	     */
2701 	    x->bm_start_time = *nowp;
2702 	    x->bm_measured.b_packets = 0;
2703 	    x->bm_measured.b_bytes = 0;
2704 	    x->bm_flags &= ~BW_METER_UPCALL_DELIVERED;
2705 	}
2706     }
2707 }
2708 
2709 /*
2710  * Prepare a bandwidth-related upcall
2711  */
2712 static void
2713 bw_meter_prepare_upcall(struct bw_meter *x, struct timeval *nowp)
2714 {
2715     struct timeval delta;
2716     struct bw_upcall *u;
2717 
2718     /*
2719      * Compute the measured time interval
2720      */
2721     delta = *nowp;
2722     BW_TIMEVALDECR(&delta, &x->bm_start_time);
2723 
2724     /*
2725      * If there are too many pending upcalls, deliver them now
2726      */
2727     if (bw_upcalls_n >= BW_UPCALLS_MAX)
2728 	bw_upcalls_send();
2729 
2730     /*
2731      * Set the bw_upcall entry
2732      */
2733     u = &bw_upcalls[bw_upcalls_n++];
2734     u->bu_src = x->bm_mfc->mfc_origin;
2735     u->bu_dst = x->bm_mfc->mfc_mcastgrp;
2736     u->bu_threshold.b_time = x->bm_threshold.b_time;
2737     u->bu_threshold.b_packets = x->bm_threshold.b_packets;
2738     u->bu_threshold.b_bytes = x->bm_threshold.b_bytes;
2739     u->bu_measured.b_time = delta;
2740     u->bu_measured.b_packets = x->bm_measured.b_packets;
2741     u->bu_measured.b_bytes = x->bm_measured.b_bytes;
2742     u->bu_flags = 0;
2743     if (x->bm_flags & BW_METER_UNIT_PACKETS)
2744 	u->bu_flags |= BW_UPCALL_UNIT_PACKETS;
2745     if (x->bm_flags & BW_METER_UNIT_BYTES)
2746 	u->bu_flags |= BW_UPCALL_UNIT_BYTES;
2747     if (x->bm_flags & BW_METER_GEQ)
2748 	u->bu_flags |= BW_UPCALL_GEQ;
2749     if (x->bm_flags & BW_METER_LEQ)
2750 	u->bu_flags |= BW_UPCALL_LEQ;
2751 }
2752 
2753 /*
2754  * Send the pending bandwidth-related upcalls
2755  */
2756 static void
2757 bw_upcalls_send(void)
2758 {
2759     struct mbuf *m;
2760     int len = bw_upcalls_n * sizeof(bw_upcalls[0]);
2761     struct sockaddr_in k_igmpsrc = {
2762 	    .sin_len = sizeof(k_igmpsrc),
2763 	    .sin_family = AF_INET,
2764     };
2765     static struct igmpmsg igmpmsg = { 0,		/* unused1 */
2766 				      0,		/* unused2 */
2767 				      IGMPMSG_BW_UPCALL,/* im_msgtype */
2768 				      0,		/* im_mbz  */
2769 				      0,		/* im_vif  */
2770 				      0,		/* unused3 */
2771 				      { 0 },		/* im_src  */
2772 				      { 0 } };		/* im_dst  */
2773 
2774     if (bw_upcalls_n == 0)
2775 	return;			/* No pending upcalls */
2776 
2777     bw_upcalls_n = 0;
2778 
2779     /*
2780      * Allocate a new mbuf, initialize it with the header and
2781      * the payload for the pending calls.
2782      */
2783     MGETHDR(m, M_DONTWAIT, MT_HEADER);
2784     if (m == NULL) {
2785 	log(LOG_WARNING, "bw_upcalls_send: cannot allocate mbuf\n");
2786 	return;
2787     }
2788 
2789     m->m_len = m->m_pkthdr.len = 0;
2790     m_copyback(m, 0, sizeof(struct igmpmsg), (void *)&igmpmsg);
2791     m_copyback(m, sizeof(struct igmpmsg), len, (void *)&bw_upcalls[0]);
2792 
2793     /*
2794      * Send the upcalls
2795      * XXX do we need to set the address in k_igmpsrc ?
2796      */
2797     mrtstat.mrts_upcalls++;
2798     if (socket_send(ip_mrouter, m, &k_igmpsrc) < 0) {
2799 	log(LOG_WARNING, "bw_upcalls_send: ip_mrouter socket queue full\n");
2800 	++mrtstat.mrts_upq_sockfull;
2801     }
2802 }
2803 
2804 /*
2805  * Compute the timeout hash value for the bw_meter entries
2806  */
2807 #define	BW_METER_TIMEHASH(bw_meter, hash)				\
2808     do {								\
2809 	struct timeval next_timeval = (bw_meter)->bm_start_time;	\
2810 									\
2811 	BW_TIMEVALADD(&next_timeval, &(bw_meter)->bm_threshold.b_time); \
2812 	(hash) = next_timeval.tv_sec;					\
2813 	if (next_timeval.tv_usec)					\
2814 	    (hash)++; /* XXX: make sure we don't timeout early */	\
2815 	(hash) %= BW_METER_BUCKETS;					\
2816     } while (/*CONSTCOND*/ 0)
2817 
2818 /*
2819  * Schedule a timer to process periodically bw_meter entry of type "<="
2820  * by linking the entry in the proper hash bucket.
2821  */
2822 static void
2823 schedule_bw_meter(struct bw_meter *x, struct timeval *nowp)
2824 {
2825     int time_hash;
2826 
2827     if (!(x->bm_flags & BW_METER_LEQ))
2828 	return;		/* XXX: we schedule timers only for "<=" entries */
2829 
2830     /*
2831      * Reset the bw_meter entry
2832      */
2833     x->bm_start_time = *nowp;
2834     x->bm_measured.b_packets = 0;
2835     x->bm_measured.b_bytes = 0;
2836     x->bm_flags &= ~BW_METER_UPCALL_DELIVERED;
2837 
2838     /*
2839      * Compute the timeout hash value and insert the entry
2840      */
2841     BW_METER_TIMEHASH(x, time_hash);
2842     x->bm_time_next = bw_meter_timers[time_hash];
2843     bw_meter_timers[time_hash] = x;
2844     x->bm_time_hash = time_hash;
2845 }
2846 
2847 /*
2848  * Unschedule the periodic timer that processes bw_meter entry of type "<="
2849  * by removing the entry from the proper hash bucket.
2850  */
2851 static void
2852 unschedule_bw_meter(struct bw_meter *x)
2853 {
2854     int time_hash;
2855     struct bw_meter *prev, *tmp;
2856 
2857     if (!(x->bm_flags & BW_METER_LEQ))
2858 	return;		/* XXX: we schedule timers only for "<=" entries */
2859 
2860     /*
2861      * Compute the timeout hash value and delete the entry
2862      */
2863     time_hash = x->bm_time_hash;
2864     if (time_hash >= BW_METER_BUCKETS)
2865 	return;		/* Entry was not scheduled */
2866 
2867     for (prev = NULL, tmp = bw_meter_timers[time_hash];
2868 	     tmp != NULL; prev = tmp, tmp = tmp->bm_time_next)
2869 	if (tmp == x)
2870 	    break;
2871 
2872     if (tmp == NULL)
2873 	panic("unschedule_bw_meter: bw_meter entry not found");
2874 
2875     if (prev != NULL)
2876 	prev->bm_time_next = x->bm_time_next;
2877     else
2878 	bw_meter_timers[time_hash] = x->bm_time_next;
2879 
2880     x->bm_time_next = NULL;
2881     x->bm_time_hash = BW_METER_BUCKETS;
2882 }
2883 
2884 /*
2885  * Process all "<=" type of bw_meter that should be processed now,
2886  * and for each entry prepare an upcall if necessary. Each processed
2887  * entry is rescheduled again for the (periodic) processing.
2888  *
2889  * This is run periodically (once per second normally). On each round,
2890  * all the potentially matching entries are in the hash slot that we are
2891  * looking at.
2892  */
2893 static void
2894 bw_meter_process(void)
2895 {
2896     int s;
2897     static uint32_t last_tv_sec;	/* last time we processed this */
2898 
2899     uint32_t loops;
2900     int i;
2901     struct timeval now, process_endtime;
2902 
2903     microtime(&now);
2904     if (last_tv_sec == now.tv_sec)
2905 	return;		/* nothing to do */
2906 
2907     loops = now.tv_sec - last_tv_sec;
2908     last_tv_sec = now.tv_sec;
2909     if (loops > BW_METER_BUCKETS)
2910 	loops = BW_METER_BUCKETS;
2911 
2912     s = splsoftnet();
2913     /*
2914      * Process all bins of bw_meter entries from the one after the last
2915      * processed to the current one. On entry, i points to the last bucket
2916      * visited, so we need to increment i at the beginning of the loop.
2917      */
2918     for (i = (now.tv_sec - loops) % BW_METER_BUCKETS; loops > 0; loops--) {
2919 	struct bw_meter *x, *tmp_list;
2920 
2921 	if (++i >= BW_METER_BUCKETS)
2922 	    i = 0;
2923 
2924 	/* Disconnect the list of bw_meter entries from the bin */
2925 	tmp_list = bw_meter_timers[i];
2926 	bw_meter_timers[i] = NULL;
2927 
2928 	/* Process the list of bw_meter entries */
2929 	while (tmp_list != NULL) {
2930 	    x = tmp_list;
2931 	    tmp_list = tmp_list->bm_time_next;
2932 
2933 	    /* Test if the time interval is over */
2934 	    process_endtime = x->bm_start_time;
2935 	    BW_TIMEVALADD(&process_endtime, &x->bm_threshold.b_time);
2936 	    if (BW_TIMEVALCMP(&process_endtime, &now, >)) {
2937 		/* Not yet: reschedule, but don't reset */
2938 		int time_hash;
2939 
2940 		BW_METER_TIMEHASH(x, time_hash);
2941 		if (time_hash == i && process_endtime.tv_sec == now.tv_sec) {
2942 		    /*
2943 		     * XXX: somehow the bin processing is a bit ahead of time.
2944 		     * Put the entry in the next bin.
2945 		     */
2946 		    if (++time_hash >= BW_METER_BUCKETS)
2947 			time_hash = 0;
2948 		}
2949 		x->bm_time_next = bw_meter_timers[time_hash];
2950 		bw_meter_timers[time_hash] = x;
2951 		x->bm_time_hash = time_hash;
2952 
2953 		continue;
2954 	    }
2955 
2956 	    /*
2957 	     * Test if we should deliver an upcall
2958 	     */
2959 	    if (((x->bm_flags & BW_METER_UNIT_PACKETS) &&
2960 		 (x->bm_measured.b_packets <= x->bm_threshold.b_packets)) ||
2961 		((x->bm_flags & BW_METER_UNIT_BYTES) &&
2962 		 (x->bm_measured.b_bytes <= x->bm_threshold.b_bytes))) {
2963 		/* Prepare an upcall for delivery */
2964 		bw_meter_prepare_upcall(x, &now);
2965 	    }
2966 
2967 	    /*
2968 	     * Reschedule for next processing
2969 	     */
2970 	    schedule_bw_meter(x, &now);
2971 	}
2972     }
2973 
2974     /* Send all upcalls that are pending delivery */
2975     bw_upcalls_send();
2976 
2977     splx(s);
2978 }
2979 
2980 /*
2981  * A periodic function for sending all upcalls that are pending delivery
2982  */
2983 static void
2984 expire_bw_upcalls_send(void *unused)
2985 {
2986     int s;
2987 
2988     s = splsoftnet();
2989     bw_upcalls_send();
2990     splx(s);
2991 
2992     callout_reset(&bw_upcalls_ch, BW_UPCALLS_PERIOD,
2993 		  expire_bw_upcalls_send, NULL);
2994 }
2995 
2996 /*
2997  * A periodic function for periodic scanning of the multicast forwarding
2998  * table for processing all "<=" bw_meter entries.
2999  */
3000 static void
3001 expire_bw_meter_process(void *unused)
3002 {
3003     if (mrt_api_config & MRT_MFC_BW_UPCALL)
3004 	bw_meter_process();
3005 
3006     callout_reset(&bw_meter_ch, BW_METER_PERIOD,
3007 		  expire_bw_meter_process, NULL);
3008 }
3009 
3010 /*
3011  * End of bandwidth monitoring code
3012  */
3013 
3014 #ifdef PIM
3015 /*
3016  * Send the packet up to the user daemon, or eventually do kernel encapsulation
3017  */
3018 static int
3019 pim_register_send(struct ip *ip, struct vif *vifp,
3020 	struct mbuf *m, struct mfc *rt)
3021 {
3022     struct mbuf *mb_copy, *mm;
3023 
3024     if (mrtdebug & DEBUG_PIM)
3025         log(LOG_DEBUG, "pim_register_send: ");
3026 
3027     mb_copy = pim_register_prepare(ip, m);
3028     if (mb_copy == NULL)
3029 	return ENOBUFS;
3030 
3031     /*
3032      * Send all the fragments. Note that the mbuf for each fragment
3033      * is freed by the sending machinery.
3034      */
3035     for (mm = mb_copy; mm; mm = mb_copy) {
3036 	mb_copy = mm->m_nextpkt;
3037 	mm->m_nextpkt = NULL;
3038 	mm = m_pullup(mm, sizeof(struct ip));
3039 	if (mm != NULL) {
3040 	    ip = mtod(mm, struct ip *);
3041 	    if ((mrt_api_config & MRT_MFC_RP) &&
3042 		!in_nullhost(rt->mfc_rp)) {
3043 		pim_register_send_rp(ip, vifp, mm, rt);
3044 	    } else {
3045 		pim_register_send_upcall(ip, vifp, mm, rt);
3046 	    }
3047 	}
3048     }
3049 
3050     return 0;
3051 }
3052 
3053 /*
3054  * Return a copy of the data packet that is ready for PIM Register
3055  * encapsulation.
3056  * XXX: Note that in the returned copy the IP header is a valid one.
3057  */
3058 static struct mbuf *
3059 pim_register_prepare(struct ip *ip, struct mbuf *m)
3060 {
3061     struct mbuf *mb_copy = NULL;
3062     int mtu;
3063 
3064     /* Take care of delayed checksums */
3065     if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
3066 	in_delayed_cksum(m);
3067 	m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4);
3068     }
3069 
3070     /*
3071      * Copy the old packet & pullup its IP header into the
3072      * new mbuf so we can modify it.
3073      */
3074     mb_copy = m_copypacket(m, M_DONTWAIT);
3075     if (mb_copy == NULL)
3076 	return NULL;
3077     mb_copy = m_pullup(mb_copy, ip->ip_hl << 2);
3078     if (mb_copy == NULL)
3079 	return NULL;
3080 
3081     /* take care of the TTL */
3082     ip = mtod(mb_copy, struct ip *);
3083     --ip->ip_ttl;
3084 
3085     /* Compute the MTU after the PIM Register encapsulation */
3086     mtu = 0xffff - sizeof(pim_encap_iphdr) - sizeof(pim_encap_pimhdr);
3087 
3088     if (ntohs(ip->ip_len) <= mtu) {
3089 	/* Turn the IP header into a valid one */
3090 	ip->ip_sum = 0;
3091 	ip->ip_sum = in_cksum(mb_copy, ip->ip_hl << 2);
3092     } else {
3093 	/* Fragment the packet */
3094 	if (ip_fragment(mb_copy, NULL, mtu) != 0) {
3095 	    /* XXX: mb_copy was freed by ip_fragment() */
3096 	    return NULL;
3097 	}
3098     }
3099     return mb_copy;
3100 }
3101 
3102 /*
3103  * Send an upcall with the data packet to the user-level process.
3104  */
3105 static int
3106 pim_register_send_upcall(struct ip *ip, struct vif *vifp,
3107     struct mbuf *mb_copy, struct mfc *rt)
3108 {
3109     struct mbuf *mb_first;
3110     int len = ntohs(ip->ip_len);
3111     struct igmpmsg *im;
3112     struct sockaddr_in k_igmpsrc = {
3113 	    .sin_len = sizeof(k_igmpsrc),
3114 	    .sin_family = AF_INET,
3115     };
3116 
3117     /*
3118      * Add a new mbuf with an upcall header
3119      */
3120     MGETHDR(mb_first, M_DONTWAIT, MT_HEADER);
3121     if (mb_first == NULL) {
3122 	m_freem(mb_copy);
3123 	return ENOBUFS;
3124     }
3125     mb_first->m_data += max_linkhdr;
3126     mb_first->m_pkthdr.len = len + sizeof(struct igmpmsg);
3127     mb_first->m_len = sizeof(struct igmpmsg);
3128     mb_first->m_next = mb_copy;
3129 
3130     /* Send message to routing daemon */
3131     im = mtod(mb_first, struct igmpmsg *);
3132     im->im_msgtype	= IGMPMSG_WHOLEPKT;
3133     im->im_mbz		= 0;
3134     im->im_vif		= vifp - viftable;
3135     im->im_src		= ip->ip_src;
3136     im->im_dst		= ip->ip_dst;
3137 
3138     k_igmpsrc.sin_addr	= ip->ip_src;
3139 
3140     mrtstat.mrts_upcalls++;
3141 
3142     if (socket_send(ip_mrouter, mb_first, &k_igmpsrc) < 0) {
3143 	if (mrtdebug & DEBUG_PIM)
3144 	    log(LOG_WARNING,
3145 		"mcast: pim_register_send_upcall: ip_mrouter socket queue full");
3146 	++mrtstat.mrts_upq_sockfull;
3147 	return ENOBUFS;
3148     }
3149 
3150     /* Keep statistics */
3151     pimstat.pims_snd_registers_msgs++;
3152     pimstat.pims_snd_registers_bytes += len;
3153 
3154     return 0;
3155 }
3156 
3157 /*
3158  * Encapsulate the data packet in PIM Register message and send it to the RP.
3159  */
3160 static int
3161 pim_register_send_rp(struct ip *ip, struct vif *vifp,
3162 	struct mbuf *mb_copy, struct mfc *rt)
3163 {
3164     struct mbuf *mb_first;
3165     struct ip *ip_outer;
3166     struct pim_encap_pimhdr *pimhdr;
3167     int len = ntohs(ip->ip_len);
3168     vifi_t vifi = rt->mfc_parent;
3169 
3170     if ((vifi >= numvifs) || in_nullhost(viftable[vifi].v_lcl_addr)) {
3171 	m_freem(mb_copy);
3172 	return EADDRNOTAVAIL;		/* The iif vif is invalid */
3173     }
3174 
3175     /*
3176      * Add a new mbuf with the encapsulating header
3177      */
3178     MGETHDR(mb_first, M_DONTWAIT, MT_HEADER);
3179     if (mb_first == NULL) {
3180 	m_freem(mb_copy);
3181 	return ENOBUFS;
3182     }
3183     mb_first->m_data += max_linkhdr;
3184     mb_first->m_len = sizeof(pim_encap_iphdr) + sizeof(pim_encap_pimhdr);
3185     mb_first->m_next = mb_copy;
3186 
3187     mb_first->m_pkthdr.len = len + mb_first->m_len;
3188 
3189     /*
3190      * Fill in the encapsulating IP and PIM header
3191      */
3192     ip_outer = mtod(mb_first, struct ip *);
3193     *ip_outer = pim_encap_iphdr;
3194      if (mb_first->m_pkthdr.len < IP_MINFRAGSIZE)
3195 	ip_outer->ip_id = 0;
3196     else
3197 	ip_outer->ip_id = ip_newid(NULL);
3198     ip_outer->ip_len = htons(len + sizeof(pim_encap_iphdr) +
3199 			     sizeof(pim_encap_pimhdr));
3200     ip_outer->ip_src = viftable[vifi].v_lcl_addr;
3201     ip_outer->ip_dst = rt->mfc_rp;
3202     /*
3203      * Copy the inner header TOS to the outer header, and take care of the
3204      * IP_DF bit.
3205      */
3206     ip_outer->ip_tos = ip->ip_tos;
3207     if (ntohs(ip->ip_off) & IP_DF)
3208 	ip_outer->ip_off |= htons(IP_DF);
3209     pimhdr = (struct pim_encap_pimhdr *)((char *)ip_outer
3210 					 + sizeof(pim_encap_iphdr));
3211     *pimhdr = pim_encap_pimhdr;
3212     /* If the iif crosses a border, set the Border-bit */
3213     if (rt->mfc_flags[vifi] & MRT_MFC_FLAGS_BORDER_VIF & mrt_api_config)
3214 	pimhdr->flags |= htonl(PIM_BORDER_REGISTER);
3215 
3216     mb_first->m_data += sizeof(pim_encap_iphdr);
3217     pimhdr->pim.pim_cksum = in_cksum(mb_first, sizeof(pim_encap_pimhdr));
3218     mb_first->m_data -= sizeof(pim_encap_iphdr);
3219 
3220     if (vifp->v_rate_limit == 0)
3221 	tbf_send_packet(vifp, mb_first);
3222     else
3223 	tbf_control(vifp, mb_first, ip, ntohs(ip_outer->ip_len));
3224 
3225     /* Keep statistics */
3226     pimstat.pims_snd_registers_msgs++;
3227     pimstat.pims_snd_registers_bytes += len;
3228 
3229     return 0;
3230 }
3231 
3232 /*
3233  * PIM-SMv2 and PIM-DM messages processing.
3234  * Receives and verifies the PIM control messages, and passes them
3235  * up to the listening socket, using rip_input().
3236  * The only message with special processing is the PIM_REGISTER message
3237  * (used by PIM-SM): the PIM header is stripped off, and the inner packet
3238  * is passed to if_simloop().
3239  */
3240 void
3241 pim_input(struct mbuf *m, ...)
3242 {
3243     struct ip *ip = mtod(m, struct ip *);
3244     struct pim *pim;
3245     int minlen;
3246     int datalen;
3247     int ip_tos;
3248     int proto;
3249     int iphlen;
3250     va_list ap;
3251 
3252     va_start(ap, m);
3253     iphlen = va_arg(ap, int);
3254     proto = va_arg(ap, int);
3255     va_end(ap);
3256 
3257     datalen = ntohs(ip->ip_len) - iphlen;
3258 
3259     /* Keep statistics */
3260     pimstat.pims_rcv_total_msgs++;
3261     pimstat.pims_rcv_total_bytes += datalen;
3262 
3263     /*
3264      * Validate lengths
3265      */
3266     if (datalen < PIM_MINLEN) {
3267 	pimstat.pims_rcv_tooshort++;
3268 	log(LOG_ERR, "pim_input: packet size too small %d from %lx\n",
3269 	    datalen, (u_long)ip->ip_src.s_addr);
3270 	m_freem(m);
3271 	return;
3272     }
3273 
3274     /*
3275      * If the packet is at least as big as a REGISTER, go agead
3276      * and grab the PIM REGISTER header size, to avoid another
3277      * possible m_pullup() later.
3278      *
3279      * PIM_MINLEN       == pimhdr + u_int32_t == 4 + 4 = 8
3280      * PIM_REG_MINLEN   == pimhdr + reghdr + encap_iphdr == 4 + 4 + 20 = 28
3281      */
3282     minlen = iphlen + (datalen >= PIM_REG_MINLEN ? PIM_REG_MINLEN : PIM_MINLEN);
3283     /*
3284      * Get the IP and PIM headers in contiguous memory, and
3285      * possibly the PIM REGISTER header.
3286      */
3287     if ((m->m_flags & M_EXT || m->m_len < minlen) &&
3288 	(m = m_pullup(m, minlen)) == NULL) {
3289 	log(LOG_ERR, "pim_input: m_pullup failure\n");
3290 	return;
3291     }
3292     /* m_pullup() may have given us a new mbuf so reset ip. */
3293     ip = mtod(m, struct ip *);
3294     ip_tos = ip->ip_tos;
3295 
3296     /* adjust mbuf to point to the PIM header */
3297     m->m_data += iphlen;
3298     m->m_len  -= iphlen;
3299     pim = mtod(m, struct pim *);
3300 
3301     /*
3302      * Validate checksum. If PIM REGISTER, exclude the data packet.
3303      *
3304      * XXX: some older PIMv2 implementations don't make this distinction,
3305      * so for compatibility reason perform the checksum over part of the
3306      * message, and if error, then over the whole message.
3307      */
3308     if (PIM_VT_T(pim->pim_vt) == PIM_REGISTER && in_cksum(m, PIM_MINLEN) == 0) {
3309 	/* do nothing, checksum okay */
3310     } else if (in_cksum(m, datalen)) {
3311 	pimstat.pims_rcv_badsum++;
3312 	if (mrtdebug & DEBUG_PIM)
3313 	    log(LOG_DEBUG, "pim_input: invalid checksum");
3314 	m_freem(m);
3315 	return;
3316     }
3317 
3318     /* PIM version check */
3319     if (PIM_VT_V(pim->pim_vt) < PIM_VERSION) {
3320 	pimstat.pims_rcv_badversion++;
3321 	log(LOG_ERR, "pim_input: incorrect version %d, expecting %d\n",
3322 	    PIM_VT_V(pim->pim_vt), PIM_VERSION);
3323 	m_freem(m);
3324 	return;
3325     }
3326 
3327     /* restore mbuf back to the outer IP */
3328     m->m_data -= iphlen;
3329     m->m_len  += iphlen;
3330 
3331     if (PIM_VT_T(pim->pim_vt) == PIM_REGISTER) {
3332 	/*
3333 	 * Since this is a REGISTER, we'll make a copy of the register
3334 	 * headers ip + pim + u_int32 + encap_ip, to be passed up to the
3335 	 * routing daemon.
3336 	 */
3337 	int s;
3338 	struct sockaddr_in dst = {
3339 		.sin_len = sizeof(dst),
3340 		.sin_family = AF_INET,
3341 	};
3342 	struct mbuf *mcp;
3343 	struct ip *encap_ip;
3344 	u_int32_t *reghdr;
3345 	struct ifnet *vifp;
3346 
3347 	s = splsoftnet();
3348 	if ((reg_vif_num >= numvifs) || (reg_vif_num == VIFI_INVALID)) {
3349 	    splx(s);
3350 	    if (mrtdebug & DEBUG_PIM)
3351 		log(LOG_DEBUG,
3352 		    "pim_input: register vif not set: %d\n", reg_vif_num);
3353 	    m_freem(m);
3354 	    return;
3355 	}
3356 	/* XXX need refcnt? */
3357 	vifp = viftable[reg_vif_num].v_ifp;
3358 	splx(s);
3359 
3360 	/*
3361 	 * Validate length
3362 	 */
3363 	if (datalen < PIM_REG_MINLEN) {
3364 	    pimstat.pims_rcv_tooshort++;
3365 	    pimstat.pims_rcv_badregisters++;
3366 	    log(LOG_ERR,
3367 		"pim_input: register packet size too small %d from %lx\n",
3368 		datalen, (u_long)ip->ip_src.s_addr);
3369 	    m_freem(m);
3370 	    return;
3371 	}
3372 
3373 	reghdr = (u_int32_t *)(pim + 1);
3374 	encap_ip = (struct ip *)(reghdr + 1);
3375 
3376 	if (mrtdebug & DEBUG_PIM) {
3377 	    log(LOG_DEBUG,
3378 		"pim_input[register], encap_ip: %lx -> %lx, encap_ip len %d\n",
3379 		(u_long)ntohl(encap_ip->ip_src.s_addr),
3380 		(u_long)ntohl(encap_ip->ip_dst.s_addr),
3381 		ntohs(encap_ip->ip_len));
3382 	}
3383 
3384 	/* verify the version number of the inner packet */
3385 	if (encap_ip->ip_v != IPVERSION) {
3386 	    pimstat.pims_rcv_badregisters++;
3387 	    if (mrtdebug & DEBUG_PIM) {
3388 		log(LOG_DEBUG, "pim_input: invalid IP version (%d) "
3389 		    "of the inner packet\n", encap_ip->ip_v);
3390 	    }
3391 	    m_freem(m);
3392 	    return;
3393 	}
3394 
3395 	/* verify the inner packet is destined to a mcast group */
3396 	if (!IN_MULTICAST(encap_ip->ip_dst.s_addr)) {
3397 	    pimstat.pims_rcv_badregisters++;
3398 	    if (mrtdebug & DEBUG_PIM)
3399 		log(LOG_DEBUG,
3400 		    "pim_input: inner packet of register is not "
3401 		    "multicast %lx\n",
3402 		    (u_long)ntohl(encap_ip->ip_dst.s_addr));
3403 	    m_freem(m);
3404 	    return;
3405 	}
3406 
3407 	/* If a NULL_REGISTER, pass it to the daemon */
3408 	if ((ntohl(*reghdr) & PIM_NULL_REGISTER))
3409 	    goto pim_input_to_daemon;
3410 
3411 	/*
3412 	 * Copy the TOS from the outer IP header to the inner IP header.
3413 	 */
3414 	if (encap_ip->ip_tos != ip_tos) {
3415 	    /* Outer TOS -> inner TOS */
3416 	    encap_ip->ip_tos = ip_tos;
3417 	    /* Recompute the inner header checksum. Sigh... */
3418 
3419 	    /* adjust mbuf to point to the inner IP header */
3420 	    m->m_data += (iphlen + PIM_MINLEN);
3421 	    m->m_len  -= (iphlen + PIM_MINLEN);
3422 
3423 	    encap_ip->ip_sum = 0;
3424 	    encap_ip->ip_sum = in_cksum(m, encap_ip->ip_hl << 2);
3425 
3426 	    /* restore mbuf to point back to the outer IP header */
3427 	    m->m_data -= (iphlen + PIM_MINLEN);
3428 	    m->m_len  += (iphlen + PIM_MINLEN);
3429 	}
3430 
3431 	/*
3432 	 * Decapsulate the inner IP packet and loopback to forward it
3433 	 * as a normal multicast packet. Also, make a copy of the
3434 	 *     outer_iphdr + pimhdr + reghdr + encap_iphdr
3435 	 * to pass to the daemon later, so it can take the appropriate
3436 	 * actions (e.g., send back PIM_REGISTER_STOP).
3437 	 * XXX: here m->m_data points to the outer IP header.
3438 	 */
3439 	mcp = m_copym(m, 0, iphlen + PIM_REG_MINLEN, M_DONTWAIT);
3440 	if (mcp == NULL) {
3441 	    log(LOG_ERR,
3442 		"pim_input: pim register: could not copy register head\n");
3443 	    m_freem(m);
3444 	    return;
3445 	}
3446 
3447 	/* Keep statistics */
3448 	/* XXX: registers_bytes include only the encap. mcast pkt */
3449 	pimstat.pims_rcv_registers_msgs++;
3450 	pimstat.pims_rcv_registers_bytes += ntohs(encap_ip->ip_len);
3451 
3452 	/*
3453 	 * forward the inner ip packet; point m_data at the inner ip.
3454 	 */
3455 	m_adj(m, iphlen + PIM_MINLEN);
3456 
3457 	if (mrtdebug & DEBUG_PIM) {
3458 	    log(LOG_DEBUG,
3459 		"pim_input: forwarding decapsulated register: "
3460 		"src %lx, dst %lx, vif %d\n",
3461 		(u_long)ntohl(encap_ip->ip_src.s_addr),
3462 		(u_long)ntohl(encap_ip->ip_dst.s_addr),
3463 		reg_vif_num);
3464 	}
3465 	/* NB: vifp was collected above; can it change on us? */
3466 	looutput(vifp, m, (struct sockaddr *)&dst, NULL);
3467 
3468 	/* prepare the register head to send to the mrouting daemon */
3469 	m = mcp;
3470     }
3471 
3472 pim_input_to_daemon:
3473     /*
3474      * Pass the PIM message up to the daemon; if it is a Register message,
3475      * pass the 'head' only up to the daemon. This includes the
3476      * outer IP header, PIM header, PIM-Register header and the
3477      * inner IP header.
3478      * XXX: the outer IP header pkt size of a Register is not adjust to
3479      * reflect the fact that the inner multicast data is truncated.
3480      */
3481     rip_input(m, iphlen, proto);
3482 
3483     return;
3484 }
3485 #endif /* PIM */
3486