xref: /freebsd/sys/netinet/ip_mroute.c (revision f856af0466c076beef4ea9b15d088e1119a945b8)
1 /*-
2  * Copyright (c) 1989 Stephen Deering
3  * Copyright (c) 1992, 1993
4  *      The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * Stephen Deering of Stanford University.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *      @(#)ip_mroute.c 8.2 (Berkeley) 11/15/93
34  */
35 
36 /*
37  * IP multicast forwarding procedures
38  *
39  * Written by David Waitzman, BBN Labs, August 1988.
40  * Modified by Steve Deering, Stanford, February 1989.
41  * Modified by Mark J. Steiglitz, Stanford, May, 1991
42  * Modified by Van Jacobson, LBL, January 1993
43  * Modified by Ajit Thyagarajan, PARC, August 1993
44  * Modified by Bill Fenner, PARC, April 1995
45  * Modified by Ahmed Helmy, SGI, June 1996
46  * Modified by George Edmond Eddy (Rusty), ISI, February 1998
47  * Modified by Pavlin Radoslavov, USC/ISI, May 1998, August 1999, October 2000
48  * Modified by Hitoshi Asaeda, WIDE, August 2000
49  * Modified by Pavlin Radoslavov, ICSI, October 2002
50  *
51  * MROUTING Revision: 3.5
52  * and PIM-SMv2 and PIM-DM support, advanced API support,
53  * bandwidth metering and signaling
54  *
55  * $FreeBSD$
56  */
57 
58 #include "opt_mac.h"
59 #include "opt_mrouting.h"
60 
61 #ifdef PIM
62 #define _PIM_VT 1
63 #endif
64 
65 #include <sys/param.h>
66 #include <sys/kernel.h>
67 #include <sys/lock.h>
68 #include <sys/malloc.h>
69 #include <sys/mbuf.h>
70 #include <sys/module.h>
71 #include <sys/priv.h>
72 #include <sys/protosw.h>
73 #include <sys/signalvar.h>
74 #include <sys/socket.h>
75 #include <sys/socketvar.h>
76 #include <sys/sockio.h>
77 #include <sys/sx.h>
78 #include <sys/sysctl.h>
79 #include <sys/syslog.h>
80 #include <sys/systm.h>
81 #include <sys/time.h>
82 #include <net/if.h>
83 #include <net/netisr.h>
84 #include <net/route.h>
85 #include <netinet/in.h>
86 #include <netinet/igmp.h>
87 #include <netinet/in_systm.h>
88 #include <netinet/in_var.h>
89 #include <netinet/ip.h>
90 #include <netinet/ip_encap.h>
91 #include <netinet/ip_mroute.h>
92 #include <netinet/ip_var.h>
93 #include <netinet/ip_options.h>
94 #ifdef PIM
95 #include <netinet/pim.h>
96 #include <netinet/pim_var.h>
97 #endif
98 #include <netinet/udp.h>
99 #include <machine/in_cksum.h>
100 
101 #include <security/mac/mac_framework.h>
102 
103 /*
104  * Control debugging code for rsvp and multicast routing code.
105  * Can only set them with the debugger.
106  */
107 static u_int    rsvpdebug;		/* non-zero enables debugging	*/
108 
109 static u_int	mrtdebug;		/* any set of the flags below	*/
110 #define		DEBUG_MFC	0x02
111 #define		DEBUG_FORWARD	0x04
112 #define		DEBUG_EXPIRE	0x08
113 #define		DEBUG_XMIT	0x10
114 #define		DEBUG_PIM	0x20
115 
116 #define		VIFI_INVALID	((vifi_t) -1)
117 
118 #define M_HASCL(m)	((m)->m_flags & M_EXT)
119 
120 static MALLOC_DEFINE(M_MRTABLE, "mroutetbl", "multicast routing tables");
121 
122 /*
123  * Locking.  We use two locks: one for the virtual interface table and
124  * one for the forwarding table.  These locks may be nested in which case
125  * the VIF lock must always be taken first.  Note that each lock is used
126  * to cover not only the specific data structure but also related data
127  * structures.  It may be better to add more fine-grained locking later;
128  * it's not clear how performance-critical this code is.
129  *
130  * XXX: This module could particularly benefit from being cleaned
131  *      up to use the <sys/queue.h> macros.
132  *
133  */
134 
135 static struct mrtstat	mrtstat;
136 SYSCTL_STRUCT(_net_inet_ip, OID_AUTO, mrtstat, CTLFLAG_RW,
137     &mrtstat, mrtstat,
138     "Multicast Routing Statistics (struct mrtstat, netinet/ip_mroute.h)");
139 
140 static struct mfc	*mfctable[MFCTBLSIZ];
141 SYSCTL_OPAQUE(_net_inet_ip, OID_AUTO, mfctable, CTLFLAG_RD,
142     &mfctable, sizeof(mfctable), "S,*mfc[MFCTBLSIZ]",
143     "Multicast Forwarding Table (struct *mfc[MFCTBLSIZ], netinet/ip_mroute.h)");
144 
145 static struct mtx mfc_mtx;
146 #define	MFC_LOCK()	mtx_lock(&mfc_mtx)
147 #define	MFC_UNLOCK()	mtx_unlock(&mfc_mtx)
148 #define	MFC_LOCK_ASSERT()	do {					\
149 	mtx_assert(&mfc_mtx, MA_OWNED);					\
150 	NET_ASSERT_GIANT();						\
151 } while (0)
152 #define	MFC_LOCK_INIT()	mtx_init(&mfc_mtx, "mroute mfc table", NULL, MTX_DEF)
153 #define	MFC_LOCK_DESTROY()	mtx_destroy(&mfc_mtx)
154 
155 static struct vif	viftable[MAXVIFS];
156 SYSCTL_OPAQUE(_net_inet_ip, OID_AUTO, viftable, CTLFLAG_RD,
157     &viftable, sizeof(viftable), "S,vif[MAXVIFS]",
158     "Multicast Virtual Interfaces (struct vif[MAXVIFS], netinet/ip_mroute.h)");
159 
160 static struct mtx vif_mtx;
161 #define	VIF_LOCK()	mtx_lock(&vif_mtx)
162 #define	VIF_UNLOCK()	mtx_unlock(&vif_mtx)
163 #define	VIF_LOCK_ASSERT()	mtx_assert(&vif_mtx, MA_OWNED)
164 #define	VIF_LOCK_INIT()	mtx_init(&vif_mtx, "mroute vif table", NULL, MTX_DEF)
165 #define	VIF_LOCK_DESTROY()	mtx_destroy(&vif_mtx)
166 
167 static u_char		nexpire[MFCTBLSIZ];
168 
169 static eventhandler_tag if_detach_event_tag = NULL;
170 
171 static struct callout expire_upcalls_ch;
172 
173 #define		EXPIRE_TIMEOUT	(hz / 4)	/* 4x / second		*/
174 #define		UPCALL_EXPIRE	6		/* number of timeouts	*/
175 
176 /*
177  * Define the token bucket filter structures
178  * tbftable -> each vif has one of these for storing info
179  */
180 
181 static struct tbf tbftable[MAXVIFS];
182 #define		TBF_REPROCESS	(hz / 100)	/* 100x / second */
183 
184 /*
185  * 'Interfaces' associated with decapsulator (so we can tell
186  * packets that went through it from ones that get reflected
187  * by a broken gateway).  These interfaces are never linked into
188  * the system ifnet list & no routes point to them.  I.e., packets
189  * can't be sent this way.  They only exist as a placeholder for
190  * multicast source verification.
191  */
192 static struct ifnet multicast_decap_if[MAXVIFS];
193 
194 #define ENCAP_TTL 64
195 #define ENCAP_PROTO IPPROTO_IPIP	/* 4 */
196 
197 /* prototype IP hdr for encapsulated packets */
198 static struct ip multicast_encap_iphdr = {
199 #if BYTE_ORDER == LITTLE_ENDIAN
200 	sizeof(struct ip) >> 2, IPVERSION,
201 #else
202 	IPVERSION, sizeof(struct ip) >> 2,
203 #endif
204 	0,				/* tos */
205 	sizeof(struct ip),		/* total length */
206 	0,				/* id */
207 	0,				/* frag offset */
208 	ENCAP_TTL, ENCAP_PROTO,
209 	0,				/* checksum */
210 };
211 
212 /*
213  * Bandwidth meter variables and constants
214  */
215 static MALLOC_DEFINE(M_BWMETER, "bwmeter", "multicast upcall bw meters");
216 /*
217  * Pending timeouts are stored in a hash table, the key being the
218  * expiration time. Periodically, the entries are analysed and processed.
219  */
220 #define BW_METER_BUCKETS	1024
221 static struct bw_meter *bw_meter_timers[BW_METER_BUCKETS];
222 static struct callout bw_meter_ch;
223 #define BW_METER_PERIOD (hz)		/* periodical handling of bw meters */
224 
225 /*
226  * Pending upcalls are stored in a vector which is flushed when
227  * full, or periodically
228  */
229 static struct bw_upcall	bw_upcalls[BW_UPCALLS_MAX];
230 static u_int	bw_upcalls_n; /* # of pending upcalls */
231 static struct callout bw_upcalls_ch;
232 #define BW_UPCALLS_PERIOD (hz)		/* periodical flush of bw upcalls */
233 
234 #ifdef PIM
235 static struct pimstat pimstat;
236 SYSCTL_STRUCT(_net_inet_pim, PIMCTL_STATS, stats, CTLFLAG_RD,
237     &pimstat, pimstat,
238     "PIM Statistics (struct pimstat, netinet/pim_var.h)");
239 
240 /*
241  * Note: the PIM Register encapsulation adds the following in front of a
242  * data packet:
243  *
244  * struct pim_encap_hdr {
245  *    struct ip ip;
246  *    struct pim_encap_pimhdr  pim;
247  * }
248  *
249  */
250 
251 struct pim_encap_pimhdr {
252 	struct pim pim;
253 	uint32_t   flags;
254 };
255 
256 static struct ip pim_encap_iphdr = {
257 #if BYTE_ORDER == LITTLE_ENDIAN
258 	sizeof(struct ip) >> 2,
259 	IPVERSION,
260 #else
261 	IPVERSION,
262 	sizeof(struct ip) >> 2,
263 #endif
264 	0,			/* tos */
265 	sizeof(struct ip),	/* total length */
266 	0,			/* id */
267 	0,			/* frag offset */
268 	ENCAP_TTL,
269 	IPPROTO_PIM,
270 	0,			/* checksum */
271 };
272 
273 static struct pim_encap_pimhdr pim_encap_pimhdr = {
274     {
275 	PIM_MAKE_VT(PIM_VERSION, PIM_REGISTER), /* PIM vers and message type */
276 	0,			/* reserved */
277 	0,			/* checksum */
278     },
279     0				/* flags */
280 };
281 
282 static struct ifnet multicast_register_if;
283 static vifi_t reg_vif_num = VIFI_INVALID;
284 #endif /* PIM */
285 
286 /*
287  * Private variables.
288  */
289 static vifi_t	   numvifs;
290 static const struct encaptab *encap_cookie;
291 
292 /*
293  * one-back cache used by mroute_encapcheck to locate a tunnel's vif
294  * given a datagram's src ip address.
295  */
296 static u_long last_encap_src;
297 static struct vif *last_encap_vif;
298 
299 /*
300  * Callout for queue processing.
301  */
302 static struct callout tbf_reprocess_ch;
303 
304 static u_long	X_ip_mcast_src(int vifi);
305 static int	X_ip_mforward(struct ip *ip, struct ifnet *ifp,
306 			struct mbuf *m, struct ip_moptions *imo);
307 static int	X_ip_mrouter_done(void);
308 static int	X_ip_mrouter_get(struct socket *so, struct sockopt *m);
309 static int	X_ip_mrouter_set(struct socket *so, struct sockopt *m);
310 static int	X_legal_vif_num(int vif);
311 static int	X_mrt_ioctl(int cmd, caddr_t data);
312 
313 static int get_sg_cnt(struct sioc_sg_req *);
314 static int get_vif_cnt(struct sioc_vif_req *);
315 static void if_detached_event(void *arg __unused, struct ifnet *);
316 static int ip_mrouter_init(struct socket *, int);
317 static int add_vif(struct vifctl *);
318 static int del_vif_locked(vifi_t);
319 static int del_vif(vifi_t);
320 static int add_mfc(struct mfcctl2 *);
321 static int del_mfc(struct mfcctl2 *);
322 static int set_api_config(uint32_t *); /* chose API capabilities */
323 static int socket_send(struct socket *, struct mbuf *, struct sockaddr_in *);
324 static int set_assert(int);
325 static void expire_upcalls(void *);
326 static int ip_mdq(struct mbuf *, struct ifnet *, struct mfc *, vifi_t);
327 static void phyint_send(struct ip *, struct vif *, struct mbuf *);
328 static void encap_send(struct ip *, struct vif *, struct mbuf *);
329 static void tbf_control(struct vif *, struct mbuf *, struct ip *, u_long);
330 static void tbf_queue(struct vif *, struct mbuf *);
331 static void tbf_process_q(struct vif *);
332 static void tbf_reprocess_q(void *);
333 static int tbf_dq_sel(struct vif *, struct ip *);
334 static void tbf_send_packet(struct vif *, struct mbuf *);
335 static void tbf_update_tokens(struct vif *);
336 static int priority(struct vif *, struct ip *);
337 
338 /*
339  * Bandwidth monitoring
340  */
341 static void free_bw_list(struct bw_meter *list);
342 static int add_bw_upcall(struct bw_upcall *);
343 static int del_bw_upcall(struct bw_upcall *);
344 static void bw_meter_receive_packet(struct bw_meter *x, int plen,
345 		struct timeval *nowp);
346 static void bw_meter_prepare_upcall(struct bw_meter *x, struct timeval *nowp);
347 static void bw_upcalls_send(void);
348 static void schedule_bw_meter(struct bw_meter *x, struct timeval *nowp);
349 static void unschedule_bw_meter(struct bw_meter *x);
350 static void bw_meter_process(void);
351 static void expire_bw_upcalls_send(void *);
352 static void expire_bw_meter_process(void *);
353 
354 #ifdef PIM
355 static int pim_register_send(struct ip *, struct vif *,
356 		struct mbuf *, struct mfc *);
357 static int pim_register_send_rp(struct ip *, struct vif *,
358 		struct mbuf *, struct mfc *);
359 static int pim_register_send_upcall(struct ip *, struct vif *,
360 		struct mbuf *, struct mfc *);
361 static struct mbuf *pim_register_prepare(struct ip *, struct mbuf *);
362 #endif
363 
364 /*
365  * whether or not special PIM assert processing is enabled.
366  */
367 static int pim_assert;
368 /*
369  * Rate limit for assert notification messages, in usec
370  */
371 #define ASSERT_MSG_TIME		3000000
372 
373 /*
374  * Kernel multicast routing API capabilities and setup.
375  * If more API capabilities are added to the kernel, they should be
376  * recorded in `mrt_api_support'.
377  */
378 static const uint32_t mrt_api_support = (MRT_MFC_FLAGS_DISABLE_WRONGVIF |
379 					 MRT_MFC_FLAGS_BORDER_VIF |
380 					 MRT_MFC_RP |
381 					 MRT_MFC_BW_UPCALL);
382 static uint32_t mrt_api_config = 0;
383 
384 /*
385  * Hash function for a source, group entry
386  */
387 #define MFCHASH(a, g) MFCHASHMOD(((a) >> 20) ^ ((a) >> 10) ^ (a) ^ \
388 			((g) >> 20) ^ ((g) >> 10) ^ (g))
389 
390 /*
391  * Find a route for a given origin IP address and Multicast group address
392  * Type of service parameter to be added in the future!!!
393  * Statistics are updated by the caller if needed
394  * (mrtstat.mrts_mfc_lookups and mrtstat.mrts_mfc_misses)
395  */
396 static struct mfc *
397 mfc_find(in_addr_t o, in_addr_t g)
398 {
399     struct mfc *rt;
400 
401     MFC_LOCK_ASSERT();
402 
403     for (rt = mfctable[MFCHASH(o,g)]; rt; rt = rt->mfc_next)
404 	if ((rt->mfc_origin.s_addr == o) &&
405 		(rt->mfc_mcastgrp.s_addr == g) && (rt->mfc_stall == NULL))
406 	    break;
407     return rt;
408 }
409 
410 /*
411  * Macros to compute elapsed time efficiently
412  * Borrowed from Van Jacobson's scheduling code
413  */
414 #define TV_DELTA(a, b, delta) {					\
415 	int xxs;						\
416 	delta = (a).tv_usec - (b).tv_usec;			\
417 	if ((xxs = (a).tv_sec - (b).tv_sec)) {			\
418 		switch (xxs) {					\
419 		case 2:						\
420 		      delta += 1000000;				\
421 		      /* FALLTHROUGH */				\
422 		case 1:						\
423 		      delta += 1000000;				\
424 		      break;					\
425 		default:					\
426 		      delta += (1000000 * xxs);			\
427 		}						\
428 	}							\
429 }
430 
431 #define TV_LT(a, b) (((a).tv_usec < (b).tv_usec && \
432 	      (a).tv_sec <= (b).tv_sec) || (a).tv_sec < (b).tv_sec)
433 
434 /*
435  * Handle MRT setsockopt commands to modify the multicast routing tables.
436  */
437 static int
438 X_ip_mrouter_set(struct socket *so, struct sockopt *sopt)
439 {
440     int	error, optval;
441     vifi_t	vifi;
442     struct	vifctl vifc;
443     struct	mfcctl2 mfc;
444     struct	bw_upcall bw_upcall;
445     uint32_t	i;
446 
447     if (so != ip_mrouter && sopt->sopt_name != MRT_INIT)
448 	return EPERM;
449 
450     error = 0;
451     switch (sopt->sopt_name) {
452     case MRT_INIT:
453 	error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval);
454 	if (error)
455 	    break;
456 	error = ip_mrouter_init(so, optval);
457 	break;
458 
459     case MRT_DONE:
460 	error = ip_mrouter_done();
461 	break;
462 
463     case MRT_ADD_VIF:
464 	error = sooptcopyin(sopt, &vifc, sizeof vifc, sizeof vifc);
465 	if (error)
466 	    break;
467 	error = add_vif(&vifc);
468 	break;
469 
470     case MRT_DEL_VIF:
471 	error = sooptcopyin(sopt, &vifi, sizeof vifi, sizeof vifi);
472 	if (error)
473 	    break;
474 	error = del_vif(vifi);
475 	break;
476 
477     case MRT_ADD_MFC:
478     case MRT_DEL_MFC:
479 	/*
480 	 * select data size depending on API version.
481 	 */
482 	if (sopt->sopt_name == MRT_ADD_MFC &&
483 		mrt_api_config & MRT_API_FLAGS_ALL) {
484 	    error = sooptcopyin(sopt, &mfc, sizeof(struct mfcctl2),
485 				sizeof(struct mfcctl2));
486 	} else {
487 	    error = sooptcopyin(sopt, &mfc, sizeof(struct mfcctl),
488 				sizeof(struct mfcctl));
489 	    bzero((caddr_t)&mfc + sizeof(struct mfcctl),
490 			sizeof(mfc) - sizeof(struct mfcctl));
491 	}
492 	if (error)
493 	    break;
494 	if (sopt->sopt_name == MRT_ADD_MFC)
495 	    error = add_mfc(&mfc);
496 	else
497 	    error = del_mfc(&mfc);
498 	break;
499 
500     case MRT_ASSERT:
501 	error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval);
502 	if (error)
503 	    break;
504 	set_assert(optval);
505 	break;
506 
507     case MRT_API_CONFIG:
508 	error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
509 	if (!error)
510 	    error = set_api_config(&i);
511 	if (!error)
512 	    error = sooptcopyout(sopt, &i, sizeof i);
513 	break;
514 
515     case MRT_ADD_BW_UPCALL:
516     case MRT_DEL_BW_UPCALL:
517 	error = sooptcopyin(sopt, &bw_upcall, sizeof bw_upcall,
518 				sizeof bw_upcall);
519 	if (error)
520 	    break;
521 	if (sopt->sopt_name == MRT_ADD_BW_UPCALL)
522 	    error = add_bw_upcall(&bw_upcall);
523 	else
524 	    error = del_bw_upcall(&bw_upcall);
525 	break;
526 
527     default:
528 	error = EOPNOTSUPP;
529 	break;
530     }
531     return error;
532 }
533 
534 /*
535  * Handle MRT getsockopt commands
536  */
537 static int
538 X_ip_mrouter_get(struct socket *so, struct sockopt *sopt)
539 {
540     int error;
541     static int version = 0x0305; /* !!! why is this here? XXX */
542 
543     switch (sopt->sopt_name) {
544     case MRT_VERSION:
545 	error = sooptcopyout(sopt, &version, sizeof version);
546 	break;
547 
548     case MRT_ASSERT:
549 	error = sooptcopyout(sopt, &pim_assert, sizeof pim_assert);
550 	break;
551 
552     case MRT_API_SUPPORT:
553 	error = sooptcopyout(sopt, &mrt_api_support, sizeof mrt_api_support);
554 	break;
555 
556     case MRT_API_CONFIG:
557 	error = sooptcopyout(sopt, &mrt_api_config, sizeof mrt_api_config);
558 	break;
559 
560     default:
561 	error = EOPNOTSUPP;
562 	break;
563     }
564     return error;
565 }
566 
567 /*
568  * Handle ioctl commands to obtain information from the cache
569  */
570 static int
571 X_mrt_ioctl(int cmd, caddr_t data)
572 {
573     int error = 0;
574 
575     /*
576      * Currently the only function calling this ioctl routine is rtioctl().
577      * Typically, only root can create the raw socket in order to execute
578      * this ioctl method, however the request might be coming from a prison
579      */
580     error = priv_check(curthread, PRIV_NETINET_MROUTE);
581     if (error)
582 	return (error);
583     switch (cmd) {
584     case (SIOCGETVIFCNT):
585 	error = get_vif_cnt((struct sioc_vif_req *)data);
586 	break;
587 
588     case (SIOCGETSGCNT):
589 	error = get_sg_cnt((struct sioc_sg_req *)data);
590 	break;
591 
592     default:
593 	error = EINVAL;
594 	break;
595     }
596     return error;
597 }
598 
599 /*
600  * returns the packet, byte, rpf-failure count for the source group provided
601  */
602 static int
603 get_sg_cnt(struct sioc_sg_req *req)
604 {
605     struct mfc *rt;
606 
607     MFC_LOCK();
608     rt = mfc_find(req->src.s_addr, req->grp.s_addr);
609     if (rt == NULL) {
610 	MFC_UNLOCK();
611 	req->pktcnt = req->bytecnt = req->wrong_if = 0xffffffff;
612 	return EADDRNOTAVAIL;
613     }
614     req->pktcnt = rt->mfc_pkt_cnt;
615     req->bytecnt = rt->mfc_byte_cnt;
616     req->wrong_if = rt->mfc_wrong_if;
617     MFC_UNLOCK();
618     return 0;
619 }
620 
621 /*
622  * returns the input and output packet and byte counts on the vif provided
623  */
624 static int
625 get_vif_cnt(struct sioc_vif_req *req)
626 {
627     vifi_t vifi = req->vifi;
628 
629     VIF_LOCK();
630     if (vifi >= numvifs) {
631 	VIF_UNLOCK();
632 	return EINVAL;
633     }
634 
635     req->icount = viftable[vifi].v_pkt_in;
636     req->ocount = viftable[vifi].v_pkt_out;
637     req->ibytes = viftable[vifi].v_bytes_in;
638     req->obytes = viftable[vifi].v_bytes_out;
639     VIF_UNLOCK();
640 
641     return 0;
642 }
643 
644 static void
645 ip_mrouter_reset(void)
646 {
647     bzero((caddr_t)mfctable, sizeof(mfctable));
648     bzero((caddr_t)nexpire, sizeof(nexpire));
649 
650     pim_assert = 0;
651     mrt_api_config = 0;
652 
653     callout_init(&expire_upcalls_ch, NET_CALLOUT_MPSAFE);
654 
655     bw_upcalls_n = 0;
656     bzero((caddr_t)bw_meter_timers, sizeof(bw_meter_timers));
657     callout_init(&bw_upcalls_ch, NET_CALLOUT_MPSAFE);
658     callout_init(&bw_meter_ch, NET_CALLOUT_MPSAFE);
659 
660     callout_init(&tbf_reprocess_ch, NET_CALLOUT_MPSAFE);
661 }
662 
663 static struct mtx mrouter_mtx;		/* used to synch init/done work */
664 
665 static void
666 if_detached_event(void *arg __unused, struct ifnet *ifp)
667 {
668     vifi_t vifi;
669     int i;
670     struct mfc *mfc;
671     struct mfc *nmfc;
672     struct mfc **ppmfc;	/* Pointer to previous node's next-pointer */
673     struct rtdetq *pq;
674     struct rtdetq *npq;
675 
676     mtx_lock(&mrouter_mtx);
677     if (ip_mrouter == NULL) {
678 	mtx_unlock(&mrouter_mtx);
679     }
680 
681     /*
682      * Tear down multicast forwarder state associated with this ifnet.
683      * 1. Walk the vif list, matching vifs against this ifnet.
684      * 2. Walk the multicast forwarding cache (mfc) looking for
685      *    inner matches with this vif's index.
686      * 3. Free any pending mbufs for this mfc.
687      * 4. Free the associated mfc entry and state associated with this vif.
688      *    Be very careful about unlinking from a singly-linked list whose
689      *    "head node" is a pointer in a simple array.
690      * 5. Free vif state. This should disable ALLMULTI on the interface.
691      */
692     VIF_LOCK();
693     MFC_LOCK();
694     for (vifi = 0; vifi < numvifs; vifi++) {
695 	if (viftable[vifi].v_ifp != ifp)
696 		continue;
697 	for (i = 0; i < MFCTBLSIZ; i++) {
698 	    ppmfc = &mfctable[i];
699 	    for (mfc = mfctable[i]; mfc != NULL; ) {
700 		nmfc = mfc->mfc_next;
701 		if (mfc->mfc_parent == vifi) {
702 		    for (pq = mfc->mfc_stall; pq != NULL; ) {
703 			npq = pq->next;
704 			m_freem(pq->m);
705 			free(pq, M_MRTABLE);
706 			pq = npq;
707 		    }
708 		    free_bw_list(mfc->mfc_bw_meter);
709 		    free(mfc, M_MRTABLE);
710 		    *ppmfc = nmfc;
711 		} else {
712 		    ppmfc = &mfc->mfc_next;
713 		}
714 		mfc = nmfc;
715 	    }
716 	}
717 	del_vif_locked(vifi);
718     }
719     MFC_UNLOCK();
720     VIF_UNLOCK();
721 
722     mtx_unlock(&mrouter_mtx);
723 }
724 
725 /*
726  * Enable multicast routing
727  */
728 static int
729 ip_mrouter_init(struct socket *so, int version)
730 {
731     if (mrtdebug)
732 	log(LOG_DEBUG, "ip_mrouter_init: so_type = %d, pr_protocol = %d\n",
733 	    so->so_type, so->so_proto->pr_protocol);
734 
735     if (so->so_type != SOCK_RAW || so->so_proto->pr_protocol != IPPROTO_IGMP)
736 	return EOPNOTSUPP;
737 
738     if (version != 1)
739 	return ENOPROTOOPT;
740 
741     mtx_lock(&mrouter_mtx);
742 
743     if (ip_mrouter != NULL) {
744 	mtx_unlock(&mrouter_mtx);
745 	return EADDRINUSE;
746     }
747 
748     if_detach_event_tag = EVENTHANDLER_REGISTER(ifnet_departure_event,
749         if_detached_event, NULL, EVENTHANDLER_PRI_ANY);
750     if (if_detach_event_tag == NULL)
751 	return (ENOMEM);
752 
753     callout_reset(&expire_upcalls_ch, EXPIRE_TIMEOUT, expire_upcalls, NULL);
754 
755     callout_reset(&bw_upcalls_ch, BW_UPCALLS_PERIOD,
756 	expire_bw_upcalls_send, NULL);
757     callout_reset(&bw_meter_ch, BW_METER_PERIOD, expire_bw_meter_process, NULL);
758 
759     ip_mrouter = so;
760 
761     mtx_unlock(&mrouter_mtx);
762 
763     if (mrtdebug)
764 	log(LOG_DEBUG, "ip_mrouter_init\n");
765 
766     return 0;
767 }
768 
769 /*
770  * Disable multicast routing
771  */
772 static int
773 X_ip_mrouter_done(void)
774 {
775     vifi_t vifi;
776     int i;
777     struct ifnet *ifp;
778     struct ifreq ifr;
779     struct mfc *rt;
780     struct rtdetq *rte;
781 
782     mtx_lock(&mrouter_mtx);
783 
784     if (ip_mrouter == NULL) {
785 	mtx_unlock(&mrouter_mtx);
786 	return EINVAL;
787     }
788 
789     /*
790      * Detach/disable hooks to the reset of the system.
791      */
792     ip_mrouter = NULL;
793     mrt_api_config = 0;
794 
795     VIF_LOCK();
796     if (encap_cookie) {
797 	const struct encaptab *c = encap_cookie;
798 	encap_cookie = NULL;
799 	encap_detach(c);
800     }
801     VIF_UNLOCK();
802 
803     callout_stop(&tbf_reprocess_ch);
804 
805     VIF_LOCK();
806     /*
807      * For each phyint in use, disable promiscuous reception of all IP
808      * multicasts.
809      */
810     for (vifi = 0; vifi < numvifs; vifi++) {
811 	if (viftable[vifi].v_lcl_addr.s_addr != 0 &&
812 		!(viftable[vifi].v_flags & (VIFF_TUNNEL | VIFF_REGISTER))) {
813 	    struct sockaddr_in *so = (struct sockaddr_in *)&(ifr.ifr_addr);
814 
815 	    so->sin_len = sizeof(struct sockaddr_in);
816 	    so->sin_family = AF_INET;
817 	    so->sin_addr.s_addr = INADDR_ANY;
818 	    ifp = viftable[vifi].v_ifp;
819 	    if_allmulti(ifp, 0);
820 	}
821     }
822     bzero((caddr_t)tbftable, sizeof(tbftable));
823     bzero((caddr_t)viftable, sizeof(viftable));
824     numvifs = 0;
825     pim_assert = 0;
826     VIF_UNLOCK();
827     EVENTHANDLER_DEREGISTER(ifnet_departure_event, if_detach_event_tag);
828 
829     /*
830      * Free all multicast forwarding cache entries.
831      */
832     callout_stop(&expire_upcalls_ch);
833     callout_stop(&bw_upcalls_ch);
834     callout_stop(&bw_meter_ch);
835 
836     MFC_LOCK();
837     for (i = 0; i < MFCTBLSIZ; i++) {
838 	for (rt = mfctable[i]; rt != NULL; ) {
839 	    struct mfc *nr = rt->mfc_next;
840 
841 	    for (rte = rt->mfc_stall; rte != NULL; ) {
842 		struct rtdetq *n = rte->next;
843 
844 		m_freem(rte->m);
845 		free(rte, M_MRTABLE);
846 		rte = n;
847 	    }
848 	    free_bw_list(rt->mfc_bw_meter);
849 	    free(rt, M_MRTABLE);
850 	    rt = nr;
851 	}
852     }
853     bzero((caddr_t)mfctable, sizeof(mfctable));
854     bzero((caddr_t)nexpire, sizeof(nexpire));
855     bw_upcalls_n = 0;
856     bzero(bw_meter_timers, sizeof(bw_meter_timers));
857     MFC_UNLOCK();
858 
859     /*
860      * Reset de-encapsulation cache
861      */
862     last_encap_src = INADDR_ANY;
863     last_encap_vif = NULL;
864 #ifdef PIM
865     reg_vif_num = VIFI_INVALID;
866 #endif
867 
868     mtx_unlock(&mrouter_mtx);
869 
870     if (mrtdebug)
871 	log(LOG_DEBUG, "ip_mrouter_done\n");
872 
873     return 0;
874 }
875 
876 /*
877  * Set PIM assert processing global
878  */
879 static int
880 set_assert(int i)
881 {
882     if ((i != 1) && (i != 0))
883 	return EINVAL;
884 
885     pim_assert = i;
886 
887     return 0;
888 }
889 
890 /*
891  * Configure API capabilities
892  */
893 int
894 set_api_config(uint32_t *apival)
895 {
896     int i;
897 
898     /*
899      * We can set the API capabilities only if it is the first operation
900      * after MRT_INIT. I.e.:
901      *  - there are no vifs installed
902      *  - pim_assert is not enabled
903      *  - the MFC table is empty
904      */
905     if (numvifs > 0) {
906 	*apival = 0;
907 	return EPERM;
908     }
909     if (pim_assert) {
910 	*apival = 0;
911 	return EPERM;
912     }
913     for (i = 0; i < MFCTBLSIZ; i++) {
914 	if (mfctable[i] != NULL) {
915 	    *apival = 0;
916 	    return EPERM;
917 	}
918     }
919 
920     mrt_api_config = *apival & mrt_api_support;
921     *apival = mrt_api_config;
922 
923     return 0;
924 }
925 
926 /*
927  * Decide if a packet is from a tunnelled peer.
928  * Return 0 if not, 64 if so.  XXX yuck.. 64 ???
929  */
930 static int
931 mroute_encapcheck(const struct mbuf *m, int off, int proto, void *arg)
932 {
933     struct ip *ip = mtod(m, struct ip *);
934     int hlen = ip->ip_hl << 2;
935 
936     /*
937      * don't claim the packet if it's not to a multicast destination or if
938      * we don't have an encapsulating tunnel with the source.
939      * Note:  This code assumes that the remote site IP address
940      * uniquely identifies the tunnel (i.e., that this site has
941      * at most one tunnel with the remote site).
942      */
943     if (!IN_MULTICAST(ntohl(((struct ip *)((char *)ip+hlen))->ip_dst.s_addr)))
944 	return 0;
945     if (ip->ip_src.s_addr != last_encap_src) {
946 	struct vif *vifp = viftable;
947 	struct vif *vife = vifp + numvifs;
948 
949 	last_encap_src = ip->ip_src.s_addr;
950 	last_encap_vif = NULL;
951 	for ( ; vifp < vife; ++vifp)
952 	    if (vifp->v_rmt_addr.s_addr == ip->ip_src.s_addr) {
953 		if ((vifp->v_flags & (VIFF_TUNNEL|VIFF_SRCRT)) == VIFF_TUNNEL)
954 		    last_encap_vif = vifp;
955 		break;
956 	    }
957     }
958     if (last_encap_vif == NULL) {
959 	last_encap_src = INADDR_ANY;
960 	return 0;
961     }
962     return 64;
963 }
964 
965 /*
966  * De-encapsulate a packet and feed it back through ip input (this
967  * routine is called whenever IP gets a packet that mroute_encap_func()
968  * claimed).
969  */
970 static void
971 mroute_encap_input(struct mbuf *m, int off)
972 {
973     struct ip *ip = mtod(m, struct ip *);
974     int hlen = ip->ip_hl << 2;
975 
976     if (hlen > sizeof(struct ip))
977 	ip_stripoptions(m, (struct mbuf *) 0);
978     m->m_data += sizeof(struct ip);
979     m->m_len -= sizeof(struct ip);
980     m->m_pkthdr.len -= sizeof(struct ip);
981 
982     m->m_pkthdr.rcvif = last_encap_vif->v_ifp;
983 
984     netisr_queue(NETISR_IP, m);		/* mbuf is free'd on failure. */
985     /*
986      * normally we would need a "schednetisr(NETISR_IP)"
987      * here but we were called by ip_input and it is going
988      * to loop back & try to dequeue the packet we just
989      * queued as soon as we return so we avoid the
990      * unnecessary software interrrupt.
991      *
992      * XXX
993      * This no longer holds - we may have direct-dispatched the packet,
994      * or there may be a queue processing limit.
995      */
996 }
997 
998 extern struct domain inetdomain;
999 static struct protosw mroute_encap_protosw =
1000 {
1001 	.pr_type =		SOCK_RAW,
1002 	.pr_domain =		&inetdomain,
1003 	.pr_protocol =		IPPROTO_IPV4,
1004 	.pr_flags =		PR_ATOMIC|PR_ADDR,
1005 	.pr_input =		mroute_encap_input,
1006 	.pr_ctloutput =		rip_ctloutput,
1007 	.pr_usrreqs =		&rip_usrreqs
1008 };
1009 
1010 /*
1011  * Add a vif to the vif table
1012  */
1013 static int
1014 add_vif(struct vifctl *vifcp)
1015 {
1016     struct vif *vifp = viftable + vifcp->vifc_vifi;
1017     struct sockaddr_in sin = {sizeof sin, AF_INET};
1018     struct ifaddr *ifa;
1019     struct ifnet *ifp;
1020     int error;
1021     struct tbf *v_tbf = tbftable + vifcp->vifc_vifi;
1022 
1023     VIF_LOCK();
1024     if (vifcp->vifc_vifi >= MAXVIFS) {
1025 	VIF_UNLOCK();
1026 	return EINVAL;
1027     }
1028     if (vifp->v_lcl_addr.s_addr != INADDR_ANY) {
1029 	VIF_UNLOCK();
1030 	return EADDRINUSE;
1031     }
1032     if (vifcp->vifc_lcl_addr.s_addr == INADDR_ANY) {
1033 	VIF_UNLOCK();
1034 	return EADDRNOTAVAIL;
1035     }
1036 
1037     /* Find the interface with an address in AF_INET family */
1038 #ifdef PIM
1039     if (vifcp->vifc_flags & VIFF_REGISTER) {
1040 	/*
1041 	 * XXX: Because VIFF_REGISTER does not really need a valid
1042 	 * local interface (e.g. it could be 127.0.0.2), we don't
1043 	 * check its address.
1044 	 */
1045 	ifp = NULL;
1046     } else
1047 #endif
1048     {
1049 	sin.sin_addr = vifcp->vifc_lcl_addr;
1050 	ifa = ifa_ifwithaddr((struct sockaddr *)&sin);
1051 	if (ifa == NULL) {
1052 	    VIF_UNLOCK();
1053 	    return EADDRNOTAVAIL;
1054 	}
1055 	ifp = ifa->ifa_ifp;
1056     }
1057 
1058     if (vifcp->vifc_flags & VIFF_TUNNEL) {
1059 	if ((vifcp->vifc_flags & VIFF_SRCRT) == 0) {
1060 	    /*
1061 	     * An encapsulating tunnel is wanted.  Tell
1062 	     * mroute_encap_input() to start paying attention
1063 	     * to encapsulated packets.
1064 	     */
1065 	    if (encap_cookie == NULL) {
1066 		int i;
1067 
1068 		encap_cookie = encap_attach_func(AF_INET, IPPROTO_IPV4,
1069 				mroute_encapcheck,
1070 				(struct protosw *)&mroute_encap_protosw, NULL);
1071 
1072 		if (encap_cookie == NULL) {
1073 		    printf("ip_mroute: unable to attach encap\n");
1074 		    VIF_UNLOCK();
1075 		    return EIO;	/* XXX */
1076 		}
1077 		for (i = 0; i < MAXVIFS; ++i) {
1078 		    if_initname(&multicast_decap_if[i], "mdecap", i);
1079 		}
1080 	    }
1081 	    /*
1082 	     * Set interface to fake encapsulator interface
1083 	     */
1084 	    ifp = &multicast_decap_if[vifcp->vifc_vifi];
1085 	    /*
1086 	     * Prepare cached route entry
1087 	     */
1088 	    bzero(&vifp->v_route, sizeof(vifp->v_route));
1089 	} else {
1090 	    log(LOG_ERR, "source routed tunnels not supported\n");
1091 	    VIF_UNLOCK();
1092 	    return EOPNOTSUPP;
1093 	}
1094 #ifdef PIM
1095     } else if (vifcp->vifc_flags & VIFF_REGISTER) {
1096 	ifp = &multicast_register_if;
1097 	if (mrtdebug)
1098 	    log(LOG_DEBUG, "Adding a register vif, ifp: %p\n",
1099 		    (void *)&multicast_register_if);
1100 	if (reg_vif_num == VIFI_INVALID) {
1101 	    if_initname(&multicast_register_if, "register_vif", 0);
1102 	    multicast_register_if.if_flags = IFF_LOOPBACK;
1103 	    bzero(&vifp->v_route, sizeof(vifp->v_route));
1104 	    reg_vif_num = vifcp->vifc_vifi;
1105 	}
1106 #endif
1107     } else {		/* Make sure the interface supports multicast */
1108 	if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1109 	    VIF_UNLOCK();
1110 	    return EOPNOTSUPP;
1111 	}
1112 
1113 	/* Enable promiscuous reception of all IP multicasts from the if */
1114 	error = if_allmulti(ifp, 1);
1115 	if (error) {
1116 	    VIF_UNLOCK();
1117 	    return error;
1118 	}
1119     }
1120 
1121     /* define parameters for the tbf structure */
1122     vifp->v_tbf = v_tbf;
1123     GET_TIME(vifp->v_tbf->tbf_last_pkt_t);
1124     vifp->v_tbf->tbf_n_tok = 0;
1125     vifp->v_tbf->tbf_q_len = 0;
1126     vifp->v_tbf->tbf_max_q_len = MAXQSIZE;
1127     vifp->v_tbf->tbf_q = vifp->v_tbf->tbf_t = NULL;
1128 
1129     vifp->v_flags     = vifcp->vifc_flags;
1130     vifp->v_threshold = vifcp->vifc_threshold;
1131     vifp->v_lcl_addr  = vifcp->vifc_lcl_addr;
1132     vifp->v_rmt_addr  = vifcp->vifc_rmt_addr;
1133     vifp->v_ifp       = ifp;
1134     /* scaling up here allows division by 1024 in critical code */
1135     vifp->v_rate_limit= vifcp->vifc_rate_limit * 1024 / 1000;
1136     vifp->v_rsvp_on   = 0;
1137     vifp->v_rsvpd     = NULL;
1138     /* initialize per vif pkt counters */
1139     vifp->v_pkt_in    = 0;
1140     vifp->v_pkt_out   = 0;
1141     vifp->v_bytes_in  = 0;
1142     vifp->v_bytes_out = 0;
1143 
1144     /* Adjust numvifs up if the vifi is higher than numvifs */
1145     if (numvifs <= vifcp->vifc_vifi) numvifs = vifcp->vifc_vifi + 1;
1146 
1147     VIF_UNLOCK();
1148 
1149     if (mrtdebug)
1150 	log(LOG_DEBUG, "add_vif #%d, lcladdr %lx, %s %lx, thresh %x, rate %d\n",
1151 	    vifcp->vifc_vifi,
1152 	    (u_long)ntohl(vifcp->vifc_lcl_addr.s_addr),
1153 	    (vifcp->vifc_flags & VIFF_TUNNEL) ? "rmtaddr" : "mask",
1154 	    (u_long)ntohl(vifcp->vifc_rmt_addr.s_addr),
1155 	    vifcp->vifc_threshold,
1156 	    vifcp->vifc_rate_limit);
1157 
1158     return 0;
1159 }
1160 
1161 /*
1162  * Delete a vif from the vif table
1163  */
1164 static int
1165 del_vif_locked(vifi_t vifi)
1166 {
1167     struct vif *vifp;
1168 
1169     VIF_LOCK_ASSERT();
1170 
1171     if (vifi >= numvifs) {
1172 	return EINVAL;
1173     }
1174     vifp = &viftable[vifi];
1175     if (vifp->v_lcl_addr.s_addr == INADDR_ANY) {
1176 	return EADDRNOTAVAIL;
1177     }
1178 
1179     if (!(vifp->v_flags & (VIFF_TUNNEL | VIFF_REGISTER)))
1180 	if_allmulti(vifp->v_ifp, 0);
1181 
1182     if (vifp == last_encap_vif) {
1183 	last_encap_vif = NULL;
1184 	last_encap_src = INADDR_ANY;
1185     }
1186 
1187     /*
1188      * Free packets queued at the interface
1189      */
1190     while (vifp->v_tbf->tbf_q) {
1191 	struct mbuf *m = vifp->v_tbf->tbf_q;
1192 
1193 	vifp->v_tbf->tbf_q = m->m_act;
1194 	m_freem(m);
1195     }
1196 
1197 #ifdef PIM
1198     if (vifp->v_flags & VIFF_REGISTER)
1199 	reg_vif_num = VIFI_INVALID;
1200 #endif
1201 
1202     bzero((caddr_t)vifp->v_tbf, sizeof(*(vifp->v_tbf)));
1203     bzero((caddr_t)vifp, sizeof (*vifp));
1204 
1205     if (mrtdebug)
1206 	log(LOG_DEBUG, "del_vif %d, numvifs %d\n", vifi, numvifs);
1207 
1208     /* Adjust numvifs down */
1209     for (vifi = numvifs; vifi > 0; vifi--)
1210 	if (viftable[vifi-1].v_lcl_addr.s_addr != INADDR_ANY)
1211 	    break;
1212     numvifs = vifi;
1213 
1214     return 0;
1215 }
1216 
1217 static int
1218 del_vif(vifi_t vifi)
1219 {
1220     int cc;
1221 
1222     VIF_LOCK();
1223     cc = del_vif_locked(vifi);
1224     VIF_UNLOCK();
1225 
1226     return cc;
1227 }
1228 
1229 /*
1230  * update an mfc entry without resetting counters and S,G addresses.
1231  */
1232 static void
1233 update_mfc_params(struct mfc *rt, struct mfcctl2 *mfccp)
1234 {
1235     int i;
1236 
1237     rt->mfc_parent = mfccp->mfcc_parent;
1238     for (i = 0; i < numvifs; i++) {
1239 	rt->mfc_ttls[i] = mfccp->mfcc_ttls[i];
1240 	rt->mfc_flags[i] = mfccp->mfcc_flags[i] & mrt_api_config &
1241 	    MRT_MFC_FLAGS_ALL;
1242     }
1243     /* set the RP address */
1244     if (mrt_api_config & MRT_MFC_RP)
1245 	rt->mfc_rp = mfccp->mfcc_rp;
1246     else
1247 	rt->mfc_rp.s_addr = INADDR_ANY;
1248 }
1249 
1250 /*
1251  * fully initialize an mfc entry from the parameter.
1252  */
1253 static void
1254 init_mfc_params(struct mfc *rt, struct mfcctl2 *mfccp)
1255 {
1256     rt->mfc_origin     = mfccp->mfcc_origin;
1257     rt->mfc_mcastgrp   = mfccp->mfcc_mcastgrp;
1258 
1259     update_mfc_params(rt, mfccp);
1260 
1261     /* initialize pkt counters per src-grp */
1262     rt->mfc_pkt_cnt    = 0;
1263     rt->mfc_byte_cnt   = 0;
1264     rt->mfc_wrong_if   = 0;
1265     rt->mfc_last_assert.tv_sec = rt->mfc_last_assert.tv_usec = 0;
1266 }
1267 
1268 
1269 /*
1270  * Add an mfc entry
1271  */
1272 static int
1273 add_mfc(struct mfcctl2 *mfccp)
1274 {
1275     struct mfc *rt;
1276     u_long hash;
1277     struct rtdetq *rte;
1278     u_short nstl;
1279 
1280     VIF_LOCK();
1281     MFC_LOCK();
1282 
1283     rt = mfc_find(mfccp->mfcc_origin.s_addr, mfccp->mfcc_mcastgrp.s_addr);
1284 
1285     /* If an entry already exists, just update the fields */
1286     if (rt) {
1287 	if (mrtdebug & DEBUG_MFC)
1288 	    log(LOG_DEBUG,"add_mfc update o %lx g %lx p %x\n",
1289 		(u_long)ntohl(mfccp->mfcc_origin.s_addr),
1290 		(u_long)ntohl(mfccp->mfcc_mcastgrp.s_addr),
1291 		mfccp->mfcc_parent);
1292 
1293 	update_mfc_params(rt, mfccp);
1294 	MFC_UNLOCK();
1295 	VIF_UNLOCK();
1296 	return 0;
1297     }
1298 
1299     /*
1300      * Find the entry for which the upcall was made and update
1301      */
1302     hash = MFCHASH(mfccp->mfcc_origin.s_addr, mfccp->mfcc_mcastgrp.s_addr);
1303     for (rt = mfctable[hash], nstl = 0; rt; rt = rt->mfc_next) {
1304 
1305 	if ((rt->mfc_origin.s_addr == mfccp->mfcc_origin.s_addr) &&
1306 		(rt->mfc_mcastgrp.s_addr == mfccp->mfcc_mcastgrp.s_addr) &&
1307 		(rt->mfc_stall != NULL)) {
1308 
1309 	    if (nstl++)
1310 		log(LOG_ERR, "add_mfc %s o %lx g %lx p %x dbx %p\n",
1311 		    "multiple kernel entries",
1312 		    (u_long)ntohl(mfccp->mfcc_origin.s_addr),
1313 		    (u_long)ntohl(mfccp->mfcc_mcastgrp.s_addr),
1314 		    mfccp->mfcc_parent, (void *)rt->mfc_stall);
1315 
1316 	    if (mrtdebug & DEBUG_MFC)
1317 		log(LOG_DEBUG,"add_mfc o %lx g %lx p %x dbg %p\n",
1318 		    (u_long)ntohl(mfccp->mfcc_origin.s_addr),
1319 		    (u_long)ntohl(mfccp->mfcc_mcastgrp.s_addr),
1320 		    mfccp->mfcc_parent, (void *)rt->mfc_stall);
1321 
1322 	    init_mfc_params(rt, mfccp);
1323 
1324 	    rt->mfc_expire = 0;	/* Don't clean this guy up */
1325 	    nexpire[hash]--;
1326 
1327 	    /* free packets Qed at the end of this entry */
1328 	    for (rte = rt->mfc_stall; rte != NULL; ) {
1329 		struct rtdetq *n = rte->next;
1330 
1331 		ip_mdq(rte->m, rte->ifp, rt, -1);
1332 		m_freem(rte->m);
1333 		free(rte, M_MRTABLE);
1334 		rte = n;
1335 	    }
1336 	    rt->mfc_stall = NULL;
1337 	}
1338     }
1339 
1340     /*
1341      * It is possible that an entry is being inserted without an upcall
1342      */
1343     if (nstl == 0) {
1344 	if (mrtdebug & DEBUG_MFC)
1345 	    log(LOG_DEBUG,"add_mfc no upcall h %lu o %lx g %lx p %x\n",
1346 		hash, (u_long)ntohl(mfccp->mfcc_origin.s_addr),
1347 		(u_long)ntohl(mfccp->mfcc_mcastgrp.s_addr),
1348 		mfccp->mfcc_parent);
1349 
1350 	for (rt = mfctable[hash]; rt != NULL; rt = rt->mfc_next) {
1351 	    if ((rt->mfc_origin.s_addr == mfccp->mfcc_origin.s_addr) &&
1352 		    (rt->mfc_mcastgrp.s_addr == mfccp->mfcc_mcastgrp.s_addr)) {
1353 		init_mfc_params(rt, mfccp);
1354 		if (rt->mfc_expire)
1355 		    nexpire[hash]--;
1356 		rt->mfc_expire = 0;
1357 		break; /* XXX */
1358 	    }
1359 	}
1360 	if (rt == NULL) {		/* no upcall, so make a new entry */
1361 	    rt = (struct mfc *)malloc(sizeof(*rt), M_MRTABLE, M_NOWAIT);
1362 	    if (rt == NULL) {
1363 		MFC_UNLOCK();
1364 		VIF_UNLOCK();
1365 		return ENOBUFS;
1366 	    }
1367 
1368 	    init_mfc_params(rt, mfccp);
1369 	    rt->mfc_expire     = 0;
1370 	    rt->mfc_stall      = NULL;
1371 
1372 	    rt->mfc_bw_meter = NULL;
1373 	    /* insert new entry at head of hash chain */
1374 	    rt->mfc_next = mfctable[hash];
1375 	    mfctable[hash] = rt;
1376 	}
1377     }
1378     MFC_UNLOCK();
1379     VIF_UNLOCK();
1380     return 0;
1381 }
1382 
1383 /*
1384  * Delete an mfc entry
1385  */
1386 static int
1387 del_mfc(struct mfcctl2 *mfccp)
1388 {
1389     struct in_addr	origin;
1390     struct in_addr	mcastgrp;
1391     struct mfc		*rt;
1392     struct mfc		**nptr;
1393     u_long		hash;
1394     struct bw_meter	*list;
1395 
1396     origin = mfccp->mfcc_origin;
1397     mcastgrp = mfccp->mfcc_mcastgrp;
1398 
1399     if (mrtdebug & DEBUG_MFC)
1400 	log(LOG_DEBUG,"del_mfc orig %lx mcastgrp %lx\n",
1401 	    (u_long)ntohl(origin.s_addr), (u_long)ntohl(mcastgrp.s_addr));
1402 
1403     MFC_LOCK();
1404 
1405     hash = MFCHASH(origin.s_addr, mcastgrp.s_addr);
1406     for (nptr = &mfctable[hash]; (rt = *nptr) != NULL; nptr = &rt->mfc_next)
1407 	if (origin.s_addr == rt->mfc_origin.s_addr &&
1408 		mcastgrp.s_addr == rt->mfc_mcastgrp.s_addr &&
1409 		rt->mfc_stall == NULL)
1410 	    break;
1411     if (rt == NULL) {
1412 	MFC_UNLOCK();
1413 	return EADDRNOTAVAIL;
1414     }
1415 
1416     *nptr = rt->mfc_next;
1417 
1418     /*
1419      * free the bw_meter entries
1420      */
1421     list = rt->mfc_bw_meter;
1422     rt->mfc_bw_meter = NULL;
1423 
1424     free(rt, M_MRTABLE);
1425 
1426     free_bw_list(list);
1427 
1428     MFC_UNLOCK();
1429 
1430     return 0;
1431 }
1432 
1433 /*
1434  * Send a message to the routing daemon on the multicast routing socket
1435  */
1436 static int
1437 socket_send(struct socket *s, struct mbuf *mm, struct sockaddr_in *src)
1438 {
1439     if (s) {
1440 	SOCKBUF_LOCK(&s->so_rcv);
1441 	if (sbappendaddr_locked(&s->so_rcv, (struct sockaddr *)src, mm,
1442 	    NULL) != 0) {
1443 	    sorwakeup_locked(s);
1444 	    return 0;
1445 	}
1446 	SOCKBUF_UNLOCK(&s->so_rcv);
1447     }
1448     m_freem(mm);
1449     return -1;
1450 }
1451 
1452 /*
1453  * IP multicast forwarding function. This function assumes that the packet
1454  * pointed to by "ip" has arrived on (or is about to be sent to) the interface
1455  * pointed to by "ifp", and the packet is to be relayed to other networks
1456  * that have members of the packet's destination IP multicast group.
1457  *
1458  * The packet is returned unscathed to the caller, unless it is
1459  * erroneous, in which case a non-zero return value tells the caller to
1460  * discard it.
1461  */
1462 
1463 #define TUNNEL_LEN  12  /* # bytes of IP option for tunnel encapsulation  */
1464 
1465 static int
1466 X_ip_mforward(struct ip *ip, struct ifnet *ifp, struct mbuf *m,
1467     struct ip_moptions *imo)
1468 {
1469     struct mfc *rt;
1470     int error;
1471     vifi_t vifi;
1472 
1473     if (mrtdebug & DEBUG_FORWARD)
1474 	log(LOG_DEBUG, "ip_mforward: src %lx, dst %lx, ifp %p\n",
1475 	    (u_long)ntohl(ip->ip_src.s_addr), (u_long)ntohl(ip->ip_dst.s_addr),
1476 	    (void *)ifp);
1477 
1478     if (ip->ip_hl < (sizeof(struct ip) + TUNNEL_LEN) >> 2 ||
1479 		((u_char *)(ip + 1))[1] != IPOPT_LSRR ) {
1480 	/*
1481 	 * Packet arrived via a physical interface or
1482 	 * an encapsulated tunnel or a register_vif.
1483 	 */
1484     } else {
1485 	/*
1486 	 * Packet arrived through a source-route tunnel.
1487 	 * Source-route tunnels are no longer supported.
1488 	 */
1489 	static int last_log;
1490 	if (last_log != time_uptime) {
1491 	    last_log = time_uptime;
1492 	    log(LOG_ERR,
1493 		"ip_mforward: received source-routed packet from %lx\n",
1494 		(u_long)ntohl(ip->ip_src.s_addr));
1495 	}
1496 	return 1;
1497     }
1498 
1499     VIF_LOCK();
1500     MFC_LOCK();
1501     if (imo && ((vifi = imo->imo_multicast_vif) < numvifs)) {
1502 	if (ip->ip_ttl < 255)
1503 	    ip->ip_ttl++;	/* compensate for -1 in *_send routines */
1504 	if (rsvpdebug && ip->ip_p == IPPROTO_RSVP) {
1505 	    struct vif *vifp = viftable + vifi;
1506 
1507 	    printf("Sending IPPROTO_RSVP from %lx to %lx on vif %d (%s%s)\n",
1508 		(long)ntohl(ip->ip_src.s_addr), (long)ntohl(ip->ip_dst.s_addr),
1509 		vifi,
1510 		(vifp->v_flags & VIFF_TUNNEL) ? "tunnel on " : "",
1511 		vifp->v_ifp->if_xname);
1512 	}
1513 	error = ip_mdq(m, ifp, NULL, vifi);
1514 	MFC_UNLOCK();
1515 	VIF_UNLOCK();
1516 	return error;
1517     }
1518     if (rsvpdebug && ip->ip_p == IPPROTO_RSVP) {
1519 	printf("Warning: IPPROTO_RSVP from %lx to %lx without vif option\n",
1520 	    (long)ntohl(ip->ip_src.s_addr), (long)ntohl(ip->ip_dst.s_addr));
1521 	if (!imo)
1522 	    printf("In fact, no options were specified at all\n");
1523     }
1524 
1525     /*
1526      * Don't forward a packet with time-to-live of zero or one,
1527      * or a packet destined to a local-only group.
1528      */
1529     if (ip->ip_ttl <= 1 || ntohl(ip->ip_dst.s_addr) <= INADDR_MAX_LOCAL_GROUP) {
1530 	MFC_UNLOCK();
1531 	VIF_UNLOCK();
1532 	return 0;
1533     }
1534 
1535     /*
1536      * Determine forwarding vifs from the forwarding cache table
1537      */
1538     ++mrtstat.mrts_mfc_lookups;
1539     rt = mfc_find(ip->ip_src.s_addr, ip->ip_dst.s_addr);
1540 
1541     /* Entry exists, so forward if necessary */
1542     if (rt != NULL) {
1543 	error = ip_mdq(m, ifp, rt, -1);
1544 	MFC_UNLOCK();
1545 	VIF_UNLOCK();
1546 	return error;
1547     } else {
1548 	/*
1549 	 * If we don't have a route for packet's origin,
1550 	 * Make a copy of the packet & send message to routing daemon
1551 	 */
1552 
1553 	struct mbuf *mb0;
1554 	struct rtdetq *rte;
1555 	u_long hash;
1556 	int hlen = ip->ip_hl << 2;
1557 
1558 	++mrtstat.mrts_mfc_misses;
1559 
1560 	mrtstat.mrts_no_route++;
1561 	if (mrtdebug & (DEBUG_FORWARD | DEBUG_MFC))
1562 	    log(LOG_DEBUG, "ip_mforward: no rte s %lx g %lx\n",
1563 		(u_long)ntohl(ip->ip_src.s_addr),
1564 		(u_long)ntohl(ip->ip_dst.s_addr));
1565 
1566 	/*
1567 	 * Allocate mbufs early so that we don't do extra work if we are
1568 	 * just going to fail anyway.  Make sure to pullup the header so
1569 	 * that other people can't step on it.
1570 	 */
1571 	rte = (struct rtdetq *)malloc((sizeof *rte), M_MRTABLE, M_NOWAIT);
1572 	if (rte == NULL) {
1573 	    MFC_UNLOCK();
1574 	    VIF_UNLOCK();
1575 	    return ENOBUFS;
1576 	}
1577 	mb0 = m_copypacket(m, M_DONTWAIT);
1578 	if (mb0 && (M_HASCL(mb0) || mb0->m_len < hlen))
1579 	    mb0 = m_pullup(mb0, hlen);
1580 	if (mb0 == NULL) {
1581 	    free(rte, M_MRTABLE);
1582 	    MFC_UNLOCK();
1583 	    VIF_UNLOCK();
1584 	    return ENOBUFS;
1585 	}
1586 
1587 	/* is there an upcall waiting for this flow ? */
1588 	hash = MFCHASH(ip->ip_src.s_addr, ip->ip_dst.s_addr);
1589 	for (rt = mfctable[hash]; rt; rt = rt->mfc_next) {
1590 	    if ((ip->ip_src.s_addr == rt->mfc_origin.s_addr) &&
1591 		    (ip->ip_dst.s_addr == rt->mfc_mcastgrp.s_addr) &&
1592 		    (rt->mfc_stall != NULL))
1593 		break;
1594 	}
1595 
1596 	if (rt == NULL) {
1597 	    int i;
1598 	    struct igmpmsg *im;
1599 	    struct sockaddr_in k_igmpsrc = { sizeof k_igmpsrc, AF_INET };
1600 	    struct mbuf *mm;
1601 
1602 	    /*
1603 	     * Locate the vifi for the incoming interface for this packet.
1604 	     * If none found, drop packet.
1605 	     */
1606 	    for (vifi=0; vifi < numvifs && viftable[vifi].v_ifp != ifp; vifi++)
1607 		;
1608 	    if (vifi >= numvifs)	/* vif not found, drop packet */
1609 		goto non_fatal;
1610 
1611 	    /* no upcall, so make a new entry */
1612 	    rt = (struct mfc *)malloc(sizeof(*rt), M_MRTABLE, M_NOWAIT);
1613 	    if (rt == NULL)
1614 		goto fail;
1615 	    /* Make a copy of the header to send to the user level process */
1616 	    mm = m_copy(mb0, 0, hlen);
1617 	    if (mm == NULL)
1618 		goto fail1;
1619 
1620 	    /*
1621 	     * Send message to routing daemon to install
1622 	     * a route into the kernel table
1623 	     */
1624 
1625 	    im = mtod(mm, struct igmpmsg *);
1626 	    im->im_msgtype = IGMPMSG_NOCACHE;
1627 	    im->im_mbz = 0;
1628 	    im->im_vif = vifi;
1629 
1630 	    mrtstat.mrts_upcalls++;
1631 
1632 	    k_igmpsrc.sin_addr = ip->ip_src;
1633 	    if (socket_send(ip_mrouter, mm, &k_igmpsrc) < 0) {
1634 		log(LOG_WARNING, "ip_mforward: ip_mrouter socket queue full\n");
1635 		++mrtstat.mrts_upq_sockfull;
1636 fail1:
1637 		free(rt, M_MRTABLE);
1638 fail:
1639 		free(rte, M_MRTABLE);
1640 		m_freem(mb0);
1641 		MFC_UNLOCK();
1642 		VIF_UNLOCK();
1643 		return ENOBUFS;
1644 	    }
1645 
1646 	    /* insert new entry at head of hash chain */
1647 	    rt->mfc_origin.s_addr     = ip->ip_src.s_addr;
1648 	    rt->mfc_mcastgrp.s_addr   = ip->ip_dst.s_addr;
1649 	    rt->mfc_expire	      = UPCALL_EXPIRE;
1650 	    nexpire[hash]++;
1651 	    for (i = 0; i < numvifs; i++) {
1652 		rt->mfc_ttls[i] = 0;
1653 		rt->mfc_flags[i] = 0;
1654 	    }
1655 	    rt->mfc_parent = -1;
1656 
1657 	    rt->mfc_rp.s_addr = INADDR_ANY; /* clear the RP address */
1658 
1659 	    rt->mfc_bw_meter = NULL;
1660 
1661 	    /* link into table */
1662 	    rt->mfc_next   = mfctable[hash];
1663 	    mfctable[hash] = rt;
1664 	    rt->mfc_stall = rte;
1665 
1666 	} else {
1667 	    /* determine if q has overflowed */
1668 	    int npkts = 0;
1669 	    struct rtdetq **p;
1670 
1671 	    /*
1672 	     * XXX ouch! we need to append to the list, but we
1673 	     * only have a pointer to the front, so we have to
1674 	     * scan the entire list every time.
1675 	     */
1676 	    for (p = &rt->mfc_stall; *p != NULL; p = &(*p)->next)
1677 		npkts++;
1678 
1679 	    if (npkts > MAX_UPQ) {
1680 		mrtstat.mrts_upq_ovflw++;
1681 non_fatal:
1682 		free(rte, M_MRTABLE);
1683 		m_freem(mb0);
1684 		MFC_UNLOCK();
1685 		VIF_UNLOCK();
1686 		return 0;
1687 	    }
1688 
1689 	    /* Add this entry to the end of the queue */
1690 	    *p = rte;
1691 	}
1692 
1693 	rte->m			= mb0;
1694 	rte->ifp		= ifp;
1695 	rte->next		= NULL;
1696 
1697 	MFC_UNLOCK();
1698 	VIF_UNLOCK();
1699 
1700 	return 0;
1701     }
1702 }
1703 
1704 /*
1705  * Clean up the cache entry if upcall is not serviced
1706  */
1707 static void
1708 expire_upcalls(void *unused)
1709 {
1710     struct rtdetq *rte;
1711     struct mfc *mfc, **nptr;
1712     int i;
1713 
1714     MFC_LOCK();
1715     for (i = 0; i < MFCTBLSIZ; i++) {
1716 	if (nexpire[i] == 0)
1717 	    continue;
1718 	nptr = &mfctable[i];
1719 	for (mfc = *nptr; mfc != NULL; mfc = *nptr) {
1720 	    /*
1721 	     * Skip real cache entries
1722 	     * Make sure it wasn't marked to not expire (shouldn't happen)
1723 	     * If it expires now
1724 	     */
1725 	    if (mfc->mfc_stall != NULL && mfc->mfc_expire != 0 &&
1726 		    --mfc->mfc_expire == 0) {
1727 		if (mrtdebug & DEBUG_EXPIRE)
1728 		    log(LOG_DEBUG, "expire_upcalls: expiring (%lx %lx)\n",
1729 			(u_long)ntohl(mfc->mfc_origin.s_addr),
1730 			(u_long)ntohl(mfc->mfc_mcastgrp.s_addr));
1731 		/*
1732 		 * drop all the packets
1733 		 * free the mbuf with the pkt, if, timing info
1734 		 */
1735 		for (rte = mfc->mfc_stall; rte; ) {
1736 		    struct rtdetq *n = rte->next;
1737 
1738 		    m_freem(rte->m);
1739 		    free(rte, M_MRTABLE);
1740 		    rte = n;
1741 		}
1742 		++mrtstat.mrts_cache_cleanups;
1743 		nexpire[i]--;
1744 
1745 		/*
1746 		 * free the bw_meter entries
1747 		 */
1748 		while (mfc->mfc_bw_meter != NULL) {
1749 		    struct bw_meter *x = mfc->mfc_bw_meter;
1750 
1751 		    mfc->mfc_bw_meter = x->bm_mfc_next;
1752 		    free(x, M_BWMETER);
1753 		}
1754 
1755 		*nptr = mfc->mfc_next;
1756 		free(mfc, M_MRTABLE);
1757 	    } else {
1758 		nptr = &mfc->mfc_next;
1759 	    }
1760 	}
1761     }
1762     MFC_UNLOCK();
1763 
1764     callout_reset(&expire_upcalls_ch, EXPIRE_TIMEOUT, expire_upcalls, NULL);
1765 }
1766 
1767 /*
1768  * Packet forwarding routine once entry in the cache is made
1769  */
1770 static int
1771 ip_mdq(struct mbuf *m, struct ifnet *ifp, struct mfc *rt, vifi_t xmt_vif)
1772 {
1773     struct ip  *ip = mtod(m, struct ip *);
1774     vifi_t vifi;
1775     int plen = ip->ip_len;
1776 
1777     VIF_LOCK_ASSERT();
1778 /*
1779  * Macro to send packet on vif.  Since RSVP packets don't get counted on
1780  * input, they shouldn't get counted on output, so statistics keeping is
1781  * separate.
1782  */
1783 #define MC_SEND(ip,vifp,m) {				\
1784 		if ((vifp)->v_flags & VIFF_TUNNEL)	\
1785 		    encap_send((ip), (vifp), (m));	\
1786 		else					\
1787 		    phyint_send((ip), (vifp), (m));	\
1788 }
1789 
1790     /*
1791      * If xmt_vif is not -1, send on only the requested vif.
1792      *
1793      * (since vifi_t is u_short, -1 becomes MAXUSHORT, which > numvifs.)
1794      */
1795     if (xmt_vif < numvifs) {
1796 #ifdef PIM
1797 	if (viftable[xmt_vif].v_flags & VIFF_REGISTER)
1798 	    pim_register_send(ip, viftable + xmt_vif, m, rt);
1799 	else
1800 #endif
1801 	MC_SEND(ip, viftable + xmt_vif, m);
1802 	return 1;
1803     }
1804 
1805     /*
1806      * Don't forward if it didn't arrive from the parent vif for its origin.
1807      */
1808     vifi = rt->mfc_parent;
1809     if ((vifi >= numvifs) || (viftable[vifi].v_ifp != ifp)) {
1810 	/* came in the wrong interface */
1811 	if (mrtdebug & DEBUG_FORWARD)
1812 	    log(LOG_DEBUG, "wrong if: ifp %p vifi %d vififp %p\n",
1813 		(void *)ifp, vifi, (void *)viftable[vifi].v_ifp);
1814 	++mrtstat.mrts_wrong_if;
1815 	++rt->mfc_wrong_if;
1816 	/*
1817 	 * If we are doing PIM assert processing, send a message
1818 	 * to the routing daemon.
1819 	 *
1820 	 * XXX: A PIM-SM router needs the WRONGVIF detection so it
1821 	 * can complete the SPT switch, regardless of the type
1822 	 * of the iif (broadcast media, GRE tunnel, etc).
1823 	 */
1824 	if (pim_assert && (vifi < numvifs) && viftable[vifi].v_ifp) {
1825 	    struct timeval now;
1826 	    u_long delta;
1827 
1828 #ifdef PIM
1829 	    if (ifp == &multicast_register_if)
1830 		pimstat.pims_rcv_registers_wrongiif++;
1831 #endif
1832 
1833 	    /* Get vifi for the incoming packet */
1834 	    for (vifi=0; vifi < numvifs && viftable[vifi].v_ifp != ifp; vifi++)
1835 		;
1836 	    if (vifi >= numvifs)
1837 		return 0;	/* The iif is not found: ignore the packet. */
1838 
1839 	    if (rt->mfc_flags[vifi] & MRT_MFC_FLAGS_DISABLE_WRONGVIF)
1840 		return 0;	/* WRONGVIF disabled: ignore the packet */
1841 
1842 	    GET_TIME(now);
1843 
1844 	    TV_DELTA(now, rt->mfc_last_assert, delta);
1845 
1846 	    if (delta > ASSERT_MSG_TIME) {
1847 		struct sockaddr_in k_igmpsrc = { sizeof k_igmpsrc, AF_INET };
1848 		struct igmpmsg *im;
1849 		int hlen = ip->ip_hl << 2;
1850 		struct mbuf *mm = m_copy(m, 0, hlen);
1851 
1852 		if (mm && (M_HASCL(mm) || mm->m_len < hlen))
1853 		    mm = m_pullup(mm, hlen);
1854 		if (mm == NULL)
1855 		    return ENOBUFS;
1856 
1857 		rt->mfc_last_assert = now;
1858 
1859 		im = mtod(mm, struct igmpmsg *);
1860 		im->im_msgtype	= IGMPMSG_WRONGVIF;
1861 		im->im_mbz		= 0;
1862 		im->im_vif		= vifi;
1863 
1864 		mrtstat.mrts_upcalls++;
1865 
1866 		k_igmpsrc.sin_addr = im->im_src;
1867 		if (socket_send(ip_mrouter, mm, &k_igmpsrc) < 0) {
1868 		    log(LOG_WARNING,
1869 			"ip_mforward: ip_mrouter socket queue full\n");
1870 		    ++mrtstat.mrts_upq_sockfull;
1871 		    return ENOBUFS;
1872 		}
1873 	    }
1874 	}
1875 	return 0;
1876     }
1877 
1878     /* If I sourced this packet, it counts as output, else it was input. */
1879     if (ip->ip_src.s_addr == viftable[vifi].v_lcl_addr.s_addr) {
1880 	viftable[vifi].v_pkt_out++;
1881 	viftable[vifi].v_bytes_out += plen;
1882     } else {
1883 	viftable[vifi].v_pkt_in++;
1884 	viftable[vifi].v_bytes_in += plen;
1885     }
1886     rt->mfc_pkt_cnt++;
1887     rt->mfc_byte_cnt += plen;
1888 
1889     /*
1890      * For each vif, decide if a copy of the packet should be forwarded.
1891      * Forward if:
1892      *		- the ttl exceeds the vif's threshold
1893      *		- there are group members downstream on interface
1894      */
1895     for (vifi = 0; vifi < numvifs; vifi++)
1896 	if ((rt->mfc_ttls[vifi] > 0) && (ip->ip_ttl > rt->mfc_ttls[vifi])) {
1897 	    viftable[vifi].v_pkt_out++;
1898 	    viftable[vifi].v_bytes_out += plen;
1899 #ifdef PIM
1900 	    if (viftable[vifi].v_flags & VIFF_REGISTER)
1901 		pim_register_send(ip, viftable + vifi, m, rt);
1902 	    else
1903 #endif
1904 	    MC_SEND(ip, viftable+vifi, m);
1905 	}
1906 
1907     /*
1908      * Perform upcall-related bw measuring.
1909      */
1910     if (rt->mfc_bw_meter != NULL) {
1911 	struct bw_meter *x;
1912 	struct timeval now;
1913 
1914 	GET_TIME(now);
1915 	MFC_LOCK_ASSERT();
1916 	for (x = rt->mfc_bw_meter; x != NULL; x = x->bm_mfc_next)
1917 	    bw_meter_receive_packet(x, plen, &now);
1918     }
1919 
1920     return 0;
1921 }
1922 
1923 /*
1924  * check if a vif number is legal/ok. This is used by ip_output.
1925  */
1926 static int
1927 X_legal_vif_num(int vif)
1928 {
1929     /* XXX unlocked, matter? */
1930     return (vif >= 0 && vif < numvifs);
1931 }
1932 
1933 /*
1934  * Return the local address used by this vif
1935  */
1936 static u_long
1937 X_ip_mcast_src(int vifi)
1938 {
1939     /* XXX unlocked, matter? */
1940     if (vifi >= 0 && vifi < numvifs)
1941 	return viftable[vifi].v_lcl_addr.s_addr;
1942     else
1943 	return INADDR_ANY;
1944 }
1945 
1946 static void
1947 phyint_send(struct ip *ip, struct vif *vifp, struct mbuf *m)
1948 {
1949     struct mbuf *mb_copy;
1950     int hlen = ip->ip_hl << 2;
1951 
1952     VIF_LOCK_ASSERT();
1953 
1954     /*
1955      * Make a new reference to the packet; make sure that
1956      * the IP header is actually copied, not just referenced,
1957      * so that ip_output() only scribbles on the copy.
1958      */
1959     mb_copy = m_copypacket(m, M_DONTWAIT);
1960     if (mb_copy && (M_HASCL(mb_copy) || mb_copy->m_len < hlen))
1961 	mb_copy = m_pullup(mb_copy, hlen);
1962     if (mb_copy == NULL)
1963 	return;
1964 
1965     if (vifp->v_rate_limit == 0)
1966 	tbf_send_packet(vifp, mb_copy);
1967     else
1968 	tbf_control(vifp, mb_copy, mtod(mb_copy, struct ip *), ip->ip_len);
1969 }
1970 
1971 static void
1972 encap_send(struct ip *ip, struct vif *vifp, struct mbuf *m)
1973 {
1974     struct mbuf *mb_copy;
1975     struct ip *ip_copy;
1976     int i, len = ip->ip_len;
1977 
1978     VIF_LOCK_ASSERT();
1979 
1980     /* Take care of delayed checksums */
1981     if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
1982 	in_delayed_cksum(m);
1983 	m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1984     }
1985 
1986     /*
1987      * copy the old packet & pullup its IP header into the
1988      * new mbuf so we can modify it.  Try to fill the new
1989      * mbuf since if we don't the ethernet driver will.
1990      */
1991     MGETHDR(mb_copy, M_DONTWAIT, MT_DATA);
1992     if (mb_copy == NULL)
1993 	return;
1994 #ifdef MAC
1995     mac_create_mbuf_multicast_encap(m, vifp->v_ifp, mb_copy);
1996 #endif
1997     mb_copy->m_data += max_linkhdr;
1998     mb_copy->m_len = sizeof(multicast_encap_iphdr);
1999 
2000     if ((mb_copy->m_next = m_copypacket(m, M_DONTWAIT)) == NULL) {
2001 	m_freem(mb_copy);
2002 	return;
2003     }
2004     i = MHLEN - M_LEADINGSPACE(mb_copy);
2005     if (i > len)
2006 	i = len;
2007     mb_copy = m_pullup(mb_copy, i);
2008     if (mb_copy == NULL)
2009 	return;
2010     mb_copy->m_pkthdr.len = len + sizeof(multicast_encap_iphdr);
2011 
2012     /*
2013      * fill in the encapsulating IP header.
2014      */
2015     ip_copy = mtod(mb_copy, struct ip *);
2016     *ip_copy = multicast_encap_iphdr;
2017     ip_copy->ip_id = ip_newid();
2018     ip_copy->ip_len += len;
2019     ip_copy->ip_src = vifp->v_lcl_addr;
2020     ip_copy->ip_dst = vifp->v_rmt_addr;
2021 
2022     /*
2023      * turn the encapsulated IP header back into a valid one.
2024      */
2025     ip = (struct ip *)((caddr_t)ip_copy + sizeof(multicast_encap_iphdr));
2026     --ip->ip_ttl;
2027     ip->ip_len = htons(ip->ip_len);
2028     ip->ip_off = htons(ip->ip_off);
2029     ip->ip_sum = 0;
2030     mb_copy->m_data += sizeof(multicast_encap_iphdr);
2031     ip->ip_sum = in_cksum(mb_copy, ip->ip_hl << 2);
2032     mb_copy->m_data -= sizeof(multicast_encap_iphdr);
2033 
2034     if (vifp->v_rate_limit == 0)
2035 	tbf_send_packet(vifp, mb_copy);
2036     else
2037 	tbf_control(vifp, mb_copy, ip, ip_copy->ip_len);
2038 }
2039 
2040 /*
2041  * Token bucket filter module
2042  */
2043 
2044 static void
2045 tbf_control(struct vif *vifp, struct mbuf *m, struct ip *ip, u_long p_len)
2046 {
2047     struct tbf *t = vifp->v_tbf;
2048 
2049     VIF_LOCK_ASSERT();
2050 
2051     if (p_len > MAX_BKT_SIZE) {		/* drop if packet is too large */
2052 	mrtstat.mrts_pkt2large++;
2053 	m_freem(m);
2054 	return;
2055     }
2056 
2057     tbf_update_tokens(vifp);
2058 
2059     if (t->tbf_q_len == 0) {		/* queue empty...		*/
2060 	if (p_len <= t->tbf_n_tok) {	/* send packet if enough tokens */
2061 	    t->tbf_n_tok -= p_len;
2062 	    tbf_send_packet(vifp, m);
2063 	} else {			/* no, queue packet and try later */
2064 	    tbf_queue(vifp, m);
2065 	    callout_reset(&tbf_reprocess_ch, TBF_REPROCESS,
2066 		tbf_reprocess_q, vifp);
2067 	}
2068     } else if (t->tbf_q_len < t->tbf_max_q_len) {
2069 	/* finite queue length, so queue pkts and process queue */
2070 	tbf_queue(vifp, m);
2071 	tbf_process_q(vifp);
2072     } else {
2073 	/* queue full, try to dq and queue and process */
2074 	if (!tbf_dq_sel(vifp, ip)) {
2075 	    mrtstat.mrts_q_overflow++;
2076 	    m_freem(m);
2077 	} else {
2078 	    tbf_queue(vifp, m);
2079 	    tbf_process_q(vifp);
2080 	}
2081     }
2082 }
2083 
2084 /*
2085  * adds a packet to the queue at the interface
2086  */
2087 static void
2088 tbf_queue(struct vif *vifp, struct mbuf *m)
2089 {
2090     struct tbf *t = vifp->v_tbf;
2091 
2092     VIF_LOCK_ASSERT();
2093 
2094     if (t->tbf_t == NULL)	/* Queue was empty */
2095 	t->tbf_q = m;
2096     else			/* Insert at tail */
2097 	t->tbf_t->m_act = m;
2098 
2099     t->tbf_t = m;		/* Set new tail pointer */
2100 
2101 #ifdef DIAGNOSTIC
2102     /* Make sure we didn't get fed a bogus mbuf */
2103     if (m->m_act)
2104 	panic("tbf_queue: m_act");
2105 #endif
2106     m->m_act = NULL;
2107 
2108     t->tbf_q_len++;
2109 }
2110 
2111 /*
2112  * processes the queue at the interface
2113  */
2114 static void
2115 tbf_process_q(struct vif *vifp)
2116 {
2117     struct tbf *t = vifp->v_tbf;
2118 
2119     VIF_LOCK_ASSERT();
2120 
2121     /* loop through the queue at the interface and send as many packets
2122      * as possible
2123      */
2124     while (t->tbf_q_len > 0) {
2125 	struct mbuf *m = t->tbf_q;
2126 	int len = mtod(m, struct ip *)->ip_len;
2127 
2128 	/* determine if the packet can be sent */
2129 	if (len > t->tbf_n_tok)	/* not enough tokens, we are done */
2130 	    break;
2131 	/* ok, reduce no of tokens, dequeue and send the packet. */
2132 	t->tbf_n_tok -= len;
2133 
2134 	t->tbf_q = m->m_act;
2135 	if (--t->tbf_q_len == 0)
2136 	    t->tbf_t = NULL;
2137 
2138 	m->m_act = NULL;
2139 	tbf_send_packet(vifp, m);
2140     }
2141 }
2142 
2143 static void
2144 tbf_reprocess_q(void *xvifp)
2145 {
2146     struct vif *vifp = xvifp;
2147 
2148     if (ip_mrouter == NULL)
2149 	return;
2150     VIF_LOCK();
2151     tbf_update_tokens(vifp);
2152     tbf_process_q(vifp);
2153     if (vifp->v_tbf->tbf_q_len)
2154 	callout_reset(&tbf_reprocess_ch, TBF_REPROCESS, tbf_reprocess_q, vifp);
2155     VIF_UNLOCK();
2156 }
2157 
2158 /* function that will selectively discard a member of the queue
2159  * based on the precedence value and the priority
2160  */
2161 static int
2162 tbf_dq_sel(struct vif *vifp, struct ip *ip)
2163 {
2164     u_int p;
2165     struct mbuf *m, *last;
2166     struct mbuf **np;
2167     struct tbf *t = vifp->v_tbf;
2168 
2169     VIF_LOCK_ASSERT();
2170 
2171     p = priority(vifp, ip);
2172 
2173     np = &t->tbf_q;
2174     last = NULL;
2175     while ((m = *np) != NULL) {
2176 	if (p > priority(vifp, mtod(m, struct ip *))) {
2177 	    *np = m->m_act;
2178 	    /* If we're removing the last packet, fix the tail pointer */
2179 	    if (m == t->tbf_t)
2180 		t->tbf_t = last;
2181 	    m_freem(m);
2182 	    /* It's impossible for the queue to be empty, but check anyways. */
2183 	    if (--t->tbf_q_len == 0)
2184 		t->tbf_t = NULL;
2185 	    mrtstat.mrts_drop_sel++;
2186 	    return 1;
2187 	}
2188 	np = &m->m_act;
2189 	last = m;
2190     }
2191     return 0;
2192 }
2193 
2194 static void
2195 tbf_send_packet(struct vif *vifp, struct mbuf *m)
2196 {
2197     VIF_LOCK_ASSERT();
2198 
2199     if (vifp->v_flags & VIFF_TUNNEL)	/* If tunnel options */
2200 	ip_output(m, NULL, &vifp->v_route, IP_FORWARDING, NULL, NULL);
2201     else {
2202 	struct ip_moptions imo;
2203 	struct in_multi *imm[2];
2204 	int error;
2205 	static struct route ro; /* XXX check this */
2206 
2207 	imo.imo_multicast_ifp  = vifp->v_ifp;
2208 	imo.imo_multicast_ttl  = mtod(m, struct ip *)->ip_ttl - 1;
2209 	imo.imo_multicast_loop = 1;
2210 	imo.imo_multicast_vif  = -1;
2211 	imo.imo_num_memberships = 0;
2212 	imo.imo_max_memberships = 2;
2213 	imo.imo_membership  = &imm[0];
2214 
2215 	/*
2216 	 * Re-entrancy should not be a problem here, because
2217 	 * the packets that we send out and are looped back at us
2218 	 * should get rejected because they appear to come from
2219 	 * the loopback interface, thus preventing looping.
2220 	 */
2221 	error = ip_output(m, NULL, &ro, IP_FORWARDING, &imo, NULL);
2222 
2223 	if (mrtdebug & DEBUG_XMIT)
2224 	    log(LOG_DEBUG, "phyint_send on vif %td err %d\n",
2225 		vifp - viftable, error);
2226     }
2227 }
2228 
2229 /* determine the current time and then
2230  * the elapsed time (between the last time and time now)
2231  * in milliseconds & update the no. of tokens in the bucket
2232  */
2233 static void
2234 tbf_update_tokens(struct vif *vifp)
2235 {
2236     struct timeval tp;
2237     u_long tm;
2238     struct tbf *t = vifp->v_tbf;
2239 
2240     VIF_LOCK_ASSERT();
2241 
2242     GET_TIME(tp);
2243 
2244     TV_DELTA(tp, t->tbf_last_pkt_t, tm);
2245 
2246     /*
2247      * This formula is actually
2248      * "time in seconds" * "bytes/second".
2249      *
2250      * (tm / 1000000) * (v_rate_limit * 1000 * (1000/1024) / 8)
2251      *
2252      * The (1000/1024) was introduced in add_vif to optimize
2253      * this divide into a shift.
2254      */
2255     t->tbf_n_tok += tm * vifp->v_rate_limit / 1024 / 8;
2256     t->tbf_last_pkt_t = tp;
2257 
2258     if (t->tbf_n_tok > MAX_BKT_SIZE)
2259 	t->tbf_n_tok = MAX_BKT_SIZE;
2260 }
2261 
2262 static int
2263 priority(struct vif *vifp, struct ip *ip)
2264 {
2265     int prio = 50; /* the lowest priority -- default case */
2266 
2267     /* temporary hack; may add general packet classifier some day */
2268 
2269     /*
2270      * The UDP port space is divided up into four priority ranges:
2271      * [0, 16384)     : unclassified - lowest priority
2272      * [16384, 32768) : audio - highest priority
2273      * [32768, 49152) : whiteboard - medium priority
2274      * [49152, 65536) : video - low priority
2275      *
2276      * Everything else gets lowest priority.
2277      */
2278     if (ip->ip_p == IPPROTO_UDP) {
2279 	struct udphdr *udp = (struct udphdr *)(((char *)ip) + (ip->ip_hl << 2));
2280 	switch (ntohs(udp->uh_dport) & 0xc000) {
2281 	case 0x4000:
2282 	    prio = 70;
2283 	    break;
2284 	case 0x8000:
2285 	    prio = 60;
2286 	    break;
2287 	case 0xc000:
2288 	    prio = 55;
2289 	    break;
2290 	}
2291     }
2292     return prio;
2293 }
2294 
2295 /*
2296  * End of token bucket filter modifications
2297  */
2298 
2299 static int
2300 X_ip_rsvp_vif(struct socket *so, struct sockopt *sopt)
2301 {
2302     int error, vifi;
2303 
2304     if (so->so_type != SOCK_RAW || so->so_proto->pr_protocol != IPPROTO_RSVP)
2305 	return EOPNOTSUPP;
2306 
2307     error = sooptcopyin(sopt, &vifi, sizeof vifi, sizeof vifi);
2308     if (error)
2309 	return error;
2310 
2311     VIF_LOCK();
2312 
2313     if (vifi < 0 || vifi >= numvifs) {	/* Error if vif is invalid */
2314 	VIF_UNLOCK();
2315 	return EADDRNOTAVAIL;
2316     }
2317 
2318     if (sopt->sopt_name == IP_RSVP_VIF_ON) {
2319 	/* Check if socket is available. */
2320 	if (viftable[vifi].v_rsvpd != NULL) {
2321 	    VIF_UNLOCK();
2322 	    return EADDRINUSE;
2323 	}
2324 
2325 	viftable[vifi].v_rsvpd = so;
2326 	/* This may seem silly, but we need to be sure we don't over-increment
2327 	 * the RSVP counter, in case something slips up.
2328 	 */
2329 	if (!viftable[vifi].v_rsvp_on) {
2330 	    viftable[vifi].v_rsvp_on = 1;
2331 	    rsvp_on++;
2332 	}
2333     } else { /* must be VIF_OFF */
2334 	/*
2335 	 * XXX as an additional consistency check, one could make sure
2336 	 * that viftable[vifi].v_rsvpd == so, otherwise passing so as
2337 	 * first parameter is pretty useless.
2338 	 */
2339 	viftable[vifi].v_rsvpd = NULL;
2340 	/*
2341 	 * This may seem silly, but we need to be sure we don't over-decrement
2342 	 * the RSVP counter, in case something slips up.
2343 	 */
2344 	if (viftable[vifi].v_rsvp_on) {
2345 	    viftable[vifi].v_rsvp_on = 0;
2346 	    rsvp_on--;
2347 	}
2348     }
2349     VIF_UNLOCK();
2350     return 0;
2351 }
2352 
2353 static void
2354 X_ip_rsvp_force_done(struct socket *so)
2355 {
2356     int vifi;
2357 
2358     /* Don't bother if it is not the right type of socket. */
2359     if (so->so_type != SOCK_RAW || so->so_proto->pr_protocol != IPPROTO_RSVP)
2360 	return;
2361 
2362     VIF_LOCK();
2363 
2364     /* The socket may be attached to more than one vif...this
2365      * is perfectly legal.
2366      */
2367     for (vifi = 0; vifi < numvifs; vifi++) {
2368 	if (viftable[vifi].v_rsvpd == so) {
2369 	    viftable[vifi].v_rsvpd = NULL;
2370 	    /* This may seem silly, but we need to be sure we don't
2371 	     * over-decrement the RSVP counter, in case something slips up.
2372 	     */
2373 	    if (viftable[vifi].v_rsvp_on) {
2374 		viftable[vifi].v_rsvp_on = 0;
2375 		rsvp_on--;
2376 	    }
2377 	}
2378     }
2379 
2380     VIF_UNLOCK();
2381 }
2382 
2383 static void
2384 X_rsvp_input(struct mbuf *m, int off)
2385 {
2386     int vifi;
2387     struct ip *ip = mtod(m, struct ip *);
2388     struct sockaddr_in rsvp_src = { sizeof rsvp_src, AF_INET };
2389     struct ifnet *ifp;
2390 
2391     if (rsvpdebug)
2392 	printf("rsvp_input: rsvp_on %d\n",rsvp_on);
2393 
2394     /* Can still get packets with rsvp_on = 0 if there is a local member
2395      * of the group to which the RSVP packet is addressed.  But in this
2396      * case we want to throw the packet away.
2397      */
2398     if (!rsvp_on) {
2399 	m_freem(m);
2400 	return;
2401     }
2402 
2403     if (rsvpdebug)
2404 	printf("rsvp_input: check vifs\n");
2405 
2406 #ifdef DIAGNOSTIC
2407     M_ASSERTPKTHDR(m);
2408 #endif
2409 
2410     ifp = m->m_pkthdr.rcvif;
2411 
2412     VIF_LOCK();
2413     /* Find which vif the packet arrived on. */
2414     for (vifi = 0; vifi < numvifs; vifi++)
2415 	if (viftable[vifi].v_ifp == ifp)
2416 	    break;
2417 
2418     if (vifi == numvifs || viftable[vifi].v_rsvpd == NULL) {
2419 	/*
2420 	 * Drop the lock here to avoid holding it across rip_input.
2421 	 * This could make rsvpdebug printfs wrong.  If you care,
2422 	 * record the state of stuff before dropping the lock.
2423 	 */
2424 	VIF_UNLOCK();
2425 	/*
2426 	 * If the old-style non-vif-associated socket is set,
2427 	 * then use it.  Otherwise, drop packet since there
2428 	 * is no specific socket for this vif.
2429 	 */
2430 	if (ip_rsvpd != NULL) {
2431 	    if (rsvpdebug)
2432 		printf("rsvp_input: Sending packet up old-style socket\n");
2433 	    rip_input(m, off);  /* xxx */
2434 	} else {
2435 	    if (rsvpdebug && vifi == numvifs)
2436 		printf("rsvp_input: Can't find vif for packet.\n");
2437 	    else if (rsvpdebug && viftable[vifi].v_rsvpd == NULL)
2438 		printf("rsvp_input: No socket defined for vif %d\n",vifi);
2439 	    m_freem(m);
2440 	}
2441 	return;
2442     }
2443     rsvp_src.sin_addr = ip->ip_src;
2444 
2445     if (rsvpdebug && m)
2446 	printf("rsvp_input: m->m_len = %d, sbspace() = %ld\n",
2447 	       m->m_len,sbspace(&(viftable[vifi].v_rsvpd->so_rcv)));
2448 
2449     if (socket_send(viftable[vifi].v_rsvpd, m, &rsvp_src) < 0) {
2450 	if (rsvpdebug)
2451 	    printf("rsvp_input: Failed to append to socket\n");
2452     } else {
2453 	if (rsvpdebug)
2454 	    printf("rsvp_input: send packet up\n");
2455     }
2456     VIF_UNLOCK();
2457 }
2458 
2459 /*
2460  * Code for bandwidth monitors
2461  */
2462 
2463 /*
2464  * Define common interface for timeval-related methods
2465  */
2466 #define	BW_TIMEVALCMP(tvp, uvp, cmp) timevalcmp((tvp), (uvp), cmp)
2467 #define	BW_TIMEVALDECR(vvp, uvp) timevalsub((vvp), (uvp))
2468 #define	BW_TIMEVALADD(vvp, uvp) timevaladd((vvp), (uvp))
2469 
2470 static uint32_t
2471 compute_bw_meter_flags(struct bw_upcall *req)
2472 {
2473     uint32_t flags = 0;
2474 
2475     if (req->bu_flags & BW_UPCALL_UNIT_PACKETS)
2476 	flags |= BW_METER_UNIT_PACKETS;
2477     if (req->bu_flags & BW_UPCALL_UNIT_BYTES)
2478 	flags |= BW_METER_UNIT_BYTES;
2479     if (req->bu_flags & BW_UPCALL_GEQ)
2480 	flags |= BW_METER_GEQ;
2481     if (req->bu_flags & BW_UPCALL_LEQ)
2482 	flags |= BW_METER_LEQ;
2483 
2484     return flags;
2485 }
2486 
2487 /*
2488  * Add a bw_meter entry
2489  */
2490 static int
2491 add_bw_upcall(struct bw_upcall *req)
2492 {
2493     struct mfc *mfc;
2494     struct timeval delta = { BW_UPCALL_THRESHOLD_INTERVAL_MIN_SEC,
2495 		BW_UPCALL_THRESHOLD_INTERVAL_MIN_USEC };
2496     struct timeval now;
2497     struct bw_meter *x;
2498     uint32_t flags;
2499 
2500     if (!(mrt_api_config & MRT_MFC_BW_UPCALL))
2501 	return EOPNOTSUPP;
2502 
2503     /* Test if the flags are valid */
2504     if (!(req->bu_flags & (BW_UPCALL_UNIT_PACKETS | BW_UPCALL_UNIT_BYTES)))
2505 	return EINVAL;
2506     if (!(req->bu_flags & (BW_UPCALL_GEQ | BW_UPCALL_LEQ)))
2507 	return EINVAL;
2508     if ((req->bu_flags & (BW_UPCALL_GEQ | BW_UPCALL_LEQ))
2509 	    == (BW_UPCALL_GEQ | BW_UPCALL_LEQ))
2510 	return EINVAL;
2511 
2512     /* Test if the threshold time interval is valid */
2513     if (BW_TIMEVALCMP(&req->bu_threshold.b_time, &delta, <))
2514 	return EINVAL;
2515 
2516     flags = compute_bw_meter_flags(req);
2517 
2518     /*
2519      * Find if we have already same bw_meter entry
2520      */
2521     MFC_LOCK();
2522     mfc = mfc_find(req->bu_src.s_addr, req->bu_dst.s_addr);
2523     if (mfc == NULL) {
2524 	MFC_UNLOCK();
2525 	return EADDRNOTAVAIL;
2526     }
2527     for (x = mfc->mfc_bw_meter; x != NULL; x = x->bm_mfc_next) {
2528 	if ((BW_TIMEVALCMP(&x->bm_threshold.b_time,
2529 			   &req->bu_threshold.b_time, ==)) &&
2530 	    (x->bm_threshold.b_packets == req->bu_threshold.b_packets) &&
2531 	    (x->bm_threshold.b_bytes == req->bu_threshold.b_bytes) &&
2532 	    (x->bm_flags & BW_METER_USER_FLAGS) == flags)  {
2533 	    MFC_UNLOCK();
2534 	    return 0;		/* XXX Already installed */
2535 	}
2536     }
2537 
2538     /* Allocate the new bw_meter entry */
2539     x = (struct bw_meter *)malloc(sizeof(*x), M_BWMETER, M_NOWAIT);
2540     if (x == NULL) {
2541 	MFC_UNLOCK();
2542 	return ENOBUFS;
2543     }
2544 
2545     /* Set the new bw_meter entry */
2546     x->bm_threshold.b_time = req->bu_threshold.b_time;
2547     GET_TIME(now);
2548     x->bm_start_time = now;
2549     x->bm_threshold.b_packets = req->bu_threshold.b_packets;
2550     x->bm_threshold.b_bytes = req->bu_threshold.b_bytes;
2551     x->bm_measured.b_packets = 0;
2552     x->bm_measured.b_bytes = 0;
2553     x->bm_flags = flags;
2554     x->bm_time_next = NULL;
2555     x->bm_time_hash = BW_METER_BUCKETS;
2556 
2557     /* Add the new bw_meter entry to the front of entries for this MFC */
2558     x->bm_mfc = mfc;
2559     x->bm_mfc_next = mfc->mfc_bw_meter;
2560     mfc->mfc_bw_meter = x;
2561     schedule_bw_meter(x, &now);
2562     MFC_UNLOCK();
2563 
2564     return 0;
2565 }
2566 
2567 static void
2568 free_bw_list(struct bw_meter *list)
2569 {
2570     while (list != NULL) {
2571 	struct bw_meter *x = list;
2572 
2573 	list = list->bm_mfc_next;
2574 	unschedule_bw_meter(x);
2575 	free(x, M_BWMETER);
2576     }
2577 }
2578 
2579 /*
2580  * Delete one or multiple bw_meter entries
2581  */
2582 static int
2583 del_bw_upcall(struct bw_upcall *req)
2584 {
2585     struct mfc *mfc;
2586     struct bw_meter *x;
2587 
2588     if (!(mrt_api_config & MRT_MFC_BW_UPCALL))
2589 	return EOPNOTSUPP;
2590 
2591     MFC_LOCK();
2592     /* Find the corresponding MFC entry */
2593     mfc = mfc_find(req->bu_src.s_addr, req->bu_dst.s_addr);
2594     if (mfc == NULL) {
2595 	MFC_UNLOCK();
2596 	return EADDRNOTAVAIL;
2597     } else if (req->bu_flags & BW_UPCALL_DELETE_ALL) {
2598 	/*
2599 	 * Delete all bw_meter entries for this mfc
2600 	 */
2601 	struct bw_meter *list;
2602 
2603 	list = mfc->mfc_bw_meter;
2604 	mfc->mfc_bw_meter = NULL;
2605 	free_bw_list(list);
2606 	MFC_UNLOCK();
2607 	return 0;
2608     } else {			/* Delete a single bw_meter entry */
2609 	struct bw_meter *prev;
2610 	uint32_t flags = 0;
2611 
2612 	flags = compute_bw_meter_flags(req);
2613 
2614 	/* Find the bw_meter entry to delete */
2615 	for (prev = NULL, x = mfc->mfc_bw_meter; x != NULL;
2616 	     prev = x, x = x->bm_mfc_next) {
2617 	    if ((BW_TIMEVALCMP(&x->bm_threshold.b_time,
2618 			       &req->bu_threshold.b_time, ==)) &&
2619 		(x->bm_threshold.b_packets == req->bu_threshold.b_packets) &&
2620 		(x->bm_threshold.b_bytes == req->bu_threshold.b_bytes) &&
2621 		(x->bm_flags & BW_METER_USER_FLAGS) == flags)
2622 		break;
2623 	}
2624 	if (x != NULL) { /* Delete entry from the list for this MFC */
2625 	    if (prev != NULL)
2626 		prev->bm_mfc_next = x->bm_mfc_next;	/* remove from middle*/
2627 	    else
2628 		x->bm_mfc->mfc_bw_meter = x->bm_mfc_next;/* new head of list */
2629 
2630 	    unschedule_bw_meter(x);
2631 	    MFC_UNLOCK();
2632 	    /* Free the bw_meter entry */
2633 	    free(x, M_BWMETER);
2634 	    return 0;
2635 	} else {
2636 	    MFC_UNLOCK();
2637 	    return EINVAL;
2638 	}
2639     }
2640     /* NOTREACHED */
2641 }
2642 
2643 /*
2644  * Perform bandwidth measurement processing that may result in an upcall
2645  */
2646 static void
2647 bw_meter_receive_packet(struct bw_meter *x, int plen, struct timeval *nowp)
2648 {
2649     struct timeval delta;
2650 
2651     MFC_LOCK_ASSERT();
2652 
2653     delta = *nowp;
2654     BW_TIMEVALDECR(&delta, &x->bm_start_time);
2655 
2656     if (x->bm_flags & BW_METER_GEQ) {
2657 	/*
2658 	 * Processing for ">=" type of bw_meter entry
2659 	 */
2660 	if (BW_TIMEVALCMP(&delta, &x->bm_threshold.b_time, >)) {
2661 	    /* Reset the bw_meter entry */
2662 	    x->bm_start_time = *nowp;
2663 	    x->bm_measured.b_packets = 0;
2664 	    x->bm_measured.b_bytes = 0;
2665 	    x->bm_flags &= ~BW_METER_UPCALL_DELIVERED;
2666 	}
2667 
2668 	/* Record that a packet is received */
2669 	x->bm_measured.b_packets++;
2670 	x->bm_measured.b_bytes += plen;
2671 
2672 	/*
2673 	 * Test if we should deliver an upcall
2674 	 */
2675 	if (!(x->bm_flags & BW_METER_UPCALL_DELIVERED)) {
2676 	    if (((x->bm_flags & BW_METER_UNIT_PACKETS) &&
2677 		 (x->bm_measured.b_packets >= x->bm_threshold.b_packets)) ||
2678 		((x->bm_flags & BW_METER_UNIT_BYTES) &&
2679 		 (x->bm_measured.b_bytes >= x->bm_threshold.b_bytes))) {
2680 		/* Prepare an upcall for delivery */
2681 		bw_meter_prepare_upcall(x, nowp);
2682 		x->bm_flags |= BW_METER_UPCALL_DELIVERED;
2683 	    }
2684 	}
2685     } else if (x->bm_flags & BW_METER_LEQ) {
2686 	/*
2687 	 * Processing for "<=" type of bw_meter entry
2688 	 */
2689 	if (BW_TIMEVALCMP(&delta, &x->bm_threshold.b_time, >)) {
2690 	    /*
2691 	     * We are behind time with the multicast forwarding table
2692 	     * scanning for "<=" type of bw_meter entries, so test now
2693 	     * if we should deliver an upcall.
2694 	     */
2695 	    if (((x->bm_flags & BW_METER_UNIT_PACKETS) &&
2696 		 (x->bm_measured.b_packets <= x->bm_threshold.b_packets)) ||
2697 		((x->bm_flags & BW_METER_UNIT_BYTES) &&
2698 		 (x->bm_measured.b_bytes <= x->bm_threshold.b_bytes))) {
2699 		/* Prepare an upcall for delivery */
2700 		bw_meter_prepare_upcall(x, nowp);
2701 	    }
2702 	    /* Reschedule the bw_meter entry */
2703 	    unschedule_bw_meter(x);
2704 	    schedule_bw_meter(x, nowp);
2705 	}
2706 
2707 	/* Record that a packet is received */
2708 	x->bm_measured.b_packets++;
2709 	x->bm_measured.b_bytes += plen;
2710 
2711 	/*
2712 	 * Test if we should restart the measuring interval
2713 	 */
2714 	if ((x->bm_flags & BW_METER_UNIT_PACKETS &&
2715 	     x->bm_measured.b_packets <= x->bm_threshold.b_packets) ||
2716 	    (x->bm_flags & BW_METER_UNIT_BYTES &&
2717 	     x->bm_measured.b_bytes <= x->bm_threshold.b_bytes)) {
2718 	    /* Don't restart the measuring interval */
2719 	} else {
2720 	    /* Do restart the measuring interval */
2721 	    /*
2722 	     * XXX: note that we don't unschedule and schedule, because this
2723 	     * might be too much overhead per packet. Instead, when we process
2724 	     * all entries for a given timer hash bin, we check whether it is
2725 	     * really a timeout. If not, we reschedule at that time.
2726 	     */
2727 	    x->bm_start_time = *nowp;
2728 	    x->bm_measured.b_packets = 0;
2729 	    x->bm_measured.b_bytes = 0;
2730 	    x->bm_flags &= ~BW_METER_UPCALL_DELIVERED;
2731 	}
2732     }
2733 }
2734 
2735 /*
2736  * Prepare a bandwidth-related upcall
2737  */
2738 static void
2739 bw_meter_prepare_upcall(struct bw_meter *x, struct timeval *nowp)
2740 {
2741     struct timeval delta;
2742     struct bw_upcall *u;
2743 
2744     MFC_LOCK_ASSERT();
2745 
2746     /*
2747      * Compute the measured time interval
2748      */
2749     delta = *nowp;
2750     BW_TIMEVALDECR(&delta, &x->bm_start_time);
2751 
2752     /*
2753      * If there are too many pending upcalls, deliver them now
2754      */
2755     if (bw_upcalls_n >= BW_UPCALLS_MAX)
2756 	bw_upcalls_send();
2757 
2758     /*
2759      * Set the bw_upcall entry
2760      */
2761     u = &bw_upcalls[bw_upcalls_n++];
2762     u->bu_src = x->bm_mfc->mfc_origin;
2763     u->bu_dst = x->bm_mfc->mfc_mcastgrp;
2764     u->bu_threshold.b_time = x->bm_threshold.b_time;
2765     u->bu_threshold.b_packets = x->bm_threshold.b_packets;
2766     u->bu_threshold.b_bytes = x->bm_threshold.b_bytes;
2767     u->bu_measured.b_time = delta;
2768     u->bu_measured.b_packets = x->bm_measured.b_packets;
2769     u->bu_measured.b_bytes = x->bm_measured.b_bytes;
2770     u->bu_flags = 0;
2771     if (x->bm_flags & BW_METER_UNIT_PACKETS)
2772 	u->bu_flags |= BW_UPCALL_UNIT_PACKETS;
2773     if (x->bm_flags & BW_METER_UNIT_BYTES)
2774 	u->bu_flags |= BW_UPCALL_UNIT_BYTES;
2775     if (x->bm_flags & BW_METER_GEQ)
2776 	u->bu_flags |= BW_UPCALL_GEQ;
2777     if (x->bm_flags & BW_METER_LEQ)
2778 	u->bu_flags |= BW_UPCALL_LEQ;
2779 }
2780 
2781 /*
2782  * Send the pending bandwidth-related upcalls
2783  */
2784 static void
2785 bw_upcalls_send(void)
2786 {
2787     struct mbuf *m;
2788     int len = bw_upcalls_n * sizeof(bw_upcalls[0]);
2789     struct sockaddr_in k_igmpsrc = { sizeof k_igmpsrc, AF_INET };
2790     static struct igmpmsg igmpmsg = { 0,		/* unused1 */
2791 				      0,		/* unused2 */
2792 				      IGMPMSG_BW_UPCALL,/* im_msgtype */
2793 				      0,		/* im_mbz  */
2794 				      0,		/* im_vif  */
2795 				      0,		/* unused3 */
2796 				      { 0 },		/* im_src  */
2797 				      { 0 } };		/* im_dst  */
2798 
2799     MFC_LOCK_ASSERT();
2800 
2801     if (bw_upcalls_n == 0)
2802 	return;			/* No pending upcalls */
2803 
2804     bw_upcalls_n = 0;
2805 
2806     /*
2807      * Allocate a new mbuf, initialize it with the header and
2808      * the payload for the pending calls.
2809      */
2810     MGETHDR(m, M_DONTWAIT, MT_DATA);
2811     if (m == NULL) {
2812 	log(LOG_WARNING, "bw_upcalls_send: cannot allocate mbuf\n");
2813 	return;
2814     }
2815 
2816     m->m_len = m->m_pkthdr.len = 0;
2817     m_copyback(m, 0, sizeof(struct igmpmsg), (caddr_t)&igmpmsg);
2818     m_copyback(m, sizeof(struct igmpmsg), len, (caddr_t)&bw_upcalls[0]);
2819 
2820     /*
2821      * Send the upcalls
2822      * XXX do we need to set the address in k_igmpsrc ?
2823      */
2824     mrtstat.mrts_upcalls++;
2825     if (socket_send(ip_mrouter, m, &k_igmpsrc) < 0) {
2826 	log(LOG_WARNING, "bw_upcalls_send: ip_mrouter socket queue full\n");
2827 	++mrtstat.mrts_upq_sockfull;
2828     }
2829 }
2830 
2831 /*
2832  * Compute the timeout hash value for the bw_meter entries
2833  */
2834 #define	BW_METER_TIMEHASH(bw_meter, hash)				\
2835     do {								\
2836 	struct timeval next_timeval = (bw_meter)->bm_start_time;	\
2837 									\
2838 	BW_TIMEVALADD(&next_timeval, &(bw_meter)->bm_threshold.b_time); \
2839 	(hash) = next_timeval.tv_sec;					\
2840 	if (next_timeval.tv_usec)					\
2841 	    (hash)++; /* XXX: make sure we don't timeout early */	\
2842 	(hash) %= BW_METER_BUCKETS;					\
2843     } while (0)
2844 
2845 /*
2846  * Schedule a timer to process periodically bw_meter entry of type "<="
2847  * by linking the entry in the proper hash bucket.
2848  */
2849 static void
2850 schedule_bw_meter(struct bw_meter *x, struct timeval *nowp)
2851 {
2852     int time_hash;
2853 
2854     MFC_LOCK_ASSERT();
2855 
2856     if (!(x->bm_flags & BW_METER_LEQ))
2857 	return;		/* XXX: we schedule timers only for "<=" entries */
2858 
2859     /*
2860      * Reset the bw_meter entry
2861      */
2862     x->bm_start_time = *nowp;
2863     x->bm_measured.b_packets = 0;
2864     x->bm_measured.b_bytes = 0;
2865     x->bm_flags &= ~BW_METER_UPCALL_DELIVERED;
2866 
2867     /*
2868      * Compute the timeout hash value and insert the entry
2869      */
2870     BW_METER_TIMEHASH(x, time_hash);
2871     x->bm_time_next = bw_meter_timers[time_hash];
2872     bw_meter_timers[time_hash] = x;
2873     x->bm_time_hash = time_hash;
2874 }
2875 
2876 /*
2877  * Unschedule the periodic timer that processes bw_meter entry of type "<="
2878  * by removing the entry from the proper hash bucket.
2879  */
2880 static void
2881 unschedule_bw_meter(struct bw_meter *x)
2882 {
2883     int time_hash;
2884     struct bw_meter *prev, *tmp;
2885 
2886     MFC_LOCK_ASSERT();
2887 
2888     if (!(x->bm_flags & BW_METER_LEQ))
2889 	return;		/* XXX: we schedule timers only for "<=" entries */
2890 
2891     /*
2892      * Compute the timeout hash value and delete the entry
2893      */
2894     time_hash = x->bm_time_hash;
2895     if (time_hash >= BW_METER_BUCKETS)
2896 	return;		/* Entry was not scheduled */
2897 
2898     for (prev = NULL, tmp = bw_meter_timers[time_hash];
2899 	     tmp != NULL; prev = tmp, tmp = tmp->bm_time_next)
2900 	if (tmp == x)
2901 	    break;
2902 
2903     if (tmp == NULL)
2904 	panic("unschedule_bw_meter: bw_meter entry not found");
2905 
2906     if (prev != NULL)
2907 	prev->bm_time_next = x->bm_time_next;
2908     else
2909 	bw_meter_timers[time_hash] = x->bm_time_next;
2910 
2911     x->bm_time_next = NULL;
2912     x->bm_time_hash = BW_METER_BUCKETS;
2913 }
2914 
2915 
2916 /*
2917  * Process all "<=" type of bw_meter that should be processed now,
2918  * and for each entry prepare an upcall if necessary. Each processed
2919  * entry is rescheduled again for the (periodic) processing.
2920  *
2921  * This is run periodically (once per second normally). On each round,
2922  * all the potentially matching entries are in the hash slot that we are
2923  * looking at.
2924  */
2925 static void
2926 bw_meter_process()
2927 {
2928     static uint32_t last_tv_sec;	/* last time we processed this */
2929 
2930     uint32_t loops;
2931     int i;
2932     struct timeval now, process_endtime;
2933 
2934     GET_TIME(now);
2935     if (last_tv_sec == now.tv_sec)
2936 	return;		/* nothing to do */
2937 
2938     loops = now.tv_sec - last_tv_sec;
2939     last_tv_sec = now.tv_sec;
2940     if (loops > BW_METER_BUCKETS)
2941 	loops = BW_METER_BUCKETS;
2942 
2943     MFC_LOCK();
2944     /*
2945      * Process all bins of bw_meter entries from the one after the last
2946      * processed to the current one. On entry, i points to the last bucket
2947      * visited, so we need to increment i at the beginning of the loop.
2948      */
2949     for (i = (now.tv_sec - loops) % BW_METER_BUCKETS; loops > 0; loops--) {
2950 	struct bw_meter *x, *tmp_list;
2951 
2952 	if (++i >= BW_METER_BUCKETS)
2953 	    i = 0;
2954 
2955 	/* Disconnect the list of bw_meter entries from the bin */
2956 	tmp_list = bw_meter_timers[i];
2957 	bw_meter_timers[i] = NULL;
2958 
2959 	/* Process the list of bw_meter entries */
2960 	while (tmp_list != NULL) {
2961 	    x = tmp_list;
2962 	    tmp_list = tmp_list->bm_time_next;
2963 
2964 	    /* Test if the time interval is over */
2965 	    process_endtime = x->bm_start_time;
2966 	    BW_TIMEVALADD(&process_endtime, &x->bm_threshold.b_time);
2967 	    if (BW_TIMEVALCMP(&process_endtime, &now, >)) {
2968 		/* Not yet: reschedule, but don't reset */
2969 		int time_hash;
2970 
2971 		BW_METER_TIMEHASH(x, time_hash);
2972 		if (time_hash == i && process_endtime.tv_sec == now.tv_sec) {
2973 		    /*
2974 		     * XXX: somehow the bin processing is a bit ahead of time.
2975 		     * Put the entry in the next bin.
2976 		     */
2977 		    if (++time_hash >= BW_METER_BUCKETS)
2978 			time_hash = 0;
2979 		}
2980 		x->bm_time_next = bw_meter_timers[time_hash];
2981 		bw_meter_timers[time_hash] = x;
2982 		x->bm_time_hash = time_hash;
2983 
2984 		continue;
2985 	    }
2986 
2987 	    /*
2988 	     * Test if we should deliver an upcall
2989 	     */
2990 	    if (((x->bm_flags & BW_METER_UNIT_PACKETS) &&
2991 		 (x->bm_measured.b_packets <= x->bm_threshold.b_packets)) ||
2992 		((x->bm_flags & BW_METER_UNIT_BYTES) &&
2993 		 (x->bm_measured.b_bytes <= x->bm_threshold.b_bytes))) {
2994 		/* Prepare an upcall for delivery */
2995 		bw_meter_prepare_upcall(x, &now);
2996 	    }
2997 
2998 	    /*
2999 	     * Reschedule for next processing
3000 	     */
3001 	    schedule_bw_meter(x, &now);
3002 	}
3003     }
3004 
3005     /* Send all upcalls that are pending delivery */
3006     bw_upcalls_send();
3007 
3008     MFC_UNLOCK();
3009 }
3010 
3011 /*
3012  * A periodic function for sending all upcalls that are pending delivery
3013  */
3014 static void
3015 expire_bw_upcalls_send(void *unused)
3016 {
3017     MFC_LOCK();
3018     bw_upcalls_send();
3019     MFC_UNLOCK();
3020 
3021     callout_reset(&bw_upcalls_ch, BW_UPCALLS_PERIOD,
3022 	expire_bw_upcalls_send, NULL);
3023 }
3024 
3025 /*
3026  * A periodic function for periodic scanning of the multicast forwarding
3027  * table for processing all "<=" bw_meter entries.
3028  */
3029 static void
3030 expire_bw_meter_process(void *unused)
3031 {
3032     if (mrt_api_config & MRT_MFC_BW_UPCALL)
3033 	bw_meter_process();
3034 
3035     callout_reset(&bw_meter_ch, BW_METER_PERIOD, expire_bw_meter_process, NULL);
3036 }
3037 
3038 /*
3039  * End of bandwidth monitoring code
3040  */
3041 
3042 #ifdef PIM
3043 /*
3044  * Send the packet up to the user daemon, or eventually do kernel encapsulation
3045  *
3046  */
3047 static int
3048 pim_register_send(struct ip *ip, struct vif *vifp,
3049 	struct mbuf *m, struct mfc *rt)
3050 {
3051     struct mbuf *mb_copy, *mm;
3052 
3053     if (mrtdebug & DEBUG_PIM)
3054 	log(LOG_DEBUG, "pim_register_send: ");
3055 
3056     mb_copy = pim_register_prepare(ip, m);
3057     if (mb_copy == NULL)
3058 	return ENOBUFS;
3059 
3060     /*
3061      * Send all the fragments. Note that the mbuf for each fragment
3062      * is freed by the sending machinery.
3063      */
3064     for (mm = mb_copy; mm; mm = mb_copy) {
3065 	mb_copy = mm->m_nextpkt;
3066 	mm->m_nextpkt = 0;
3067 	mm = m_pullup(mm, sizeof(struct ip));
3068 	if (mm != NULL) {
3069 	    ip = mtod(mm, struct ip *);
3070 	    if ((mrt_api_config & MRT_MFC_RP) &&
3071 		(rt->mfc_rp.s_addr != INADDR_ANY)) {
3072 		pim_register_send_rp(ip, vifp, mm, rt);
3073 	    } else {
3074 		pim_register_send_upcall(ip, vifp, mm, rt);
3075 	    }
3076 	}
3077     }
3078 
3079     return 0;
3080 }
3081 
3082 /*
3083  * Return a copy of the data packet that is ready for PIM Register
3084  * encapsulation.
3085  * XXX: Note that in the returned copy the IP header is a valid one.
3086  */
3087 static struct mbuf *
3088 pim_register_prepare(struct ip *ip, struct mbuf *m)
3089 {
3090     struct mbuf *mb_copy = NULL;
3091     int mtu;
3092 
3093     /* Take care of delayed checksums */
3094     if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
3095 	in_delayed_cksum(m);
3096 	m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
3097     }
3098 
3099     /*
3100      * Copy the old packet & pullup its IP header into the
3101      * new mbuf so we can modify it.
3102      */
3103     mb_copy = m_copypacket(m, M_DONTWAIT);
3104     if (mb_copy == NULL)
3105 	return NULL;
3106     mb_copy = m_pullup(mb_copy, ip->ip_hl << 2);
3107     if (mb_copy == NULL)
3108 	return NULL;
3109 
3110     /* take care of the TTL */
3111     ip = mtod(mb_copy, struct ip *);
3112     --ip->ip_ttl;
3113 
3114     /* Compute the MTU after the PIM Register encapsulation */
3115     mtu = 0xffff - sizeof(pim_encap_iphdr) - sizeof(pim_encap_pimhdr);
3116 
3117     if (ip->ip_len <= mtu) {
3118 	/* Turn the IP header into a valid one */
3119 	ip->ip_len = htons(ip->ip_len);
3120 	ip->ip_off = htons(ip->ip_off);
3121 	ip->ip_sum = 0;
3122 	ip->ip_sum = in_cksum(mb_copy, ip->ip_hl << 2);
3123     } else {
3124 	/* Fragment the packet */
3125 	if (ip_fragment(ip, &mb_copy, mtu, 0, CSUM_DELAY_IP) != 0) {
3126 	    m_freem(mb_copy);
3127 	    return NULL;
3128 	}
3129     }
3130     return mb_copy;
3131 }
3132 
3133 /*
3134  * Send an upcall with the data packet to the user-level process.
3135  */
3136 static int
3137 pim_register_send_upcall(struct ip *ip, struct vif *vifp,
3138 	struct mbuf *mb_copy, struct mfc *rt)
3139 {
3140     struct mbuf *mb_first;
3141     int len = ntohs(ip->ip_len);
3142     struct igmpmsg *im;
3143     struct sockaddr_in k_igmpsrc = { sizeof k_igmpsrc, AF_INET };
3144 
3145     VIF_LOCK_ASSERT();
3146 
3147     /*
3148      * Add a new mbuf with an upcall header
3149      */
3150     MGETHDR(mb_first, M_DONTWAIT, MT_DATA);
3151     if (mb_first == NULL) {
3152 	m_freem(mb_copy);
3153 	return ENOBUFS;
3154     }
3155     mb_first->m_data += max_linkhdr;
3156     mb_first->m_pkthdr.len = len + sizeof(struct igmpmsg);
3157     mb_first->m_len = sizeof(struct igmpmsg);
3158     mb_first->m_next = mb_copy;
3159 
3160     /* Send message to routing daemon */
3161     im = mtod(mb_first, struct igmpmsg *);
3162     im->im_msgtype	= IGMPMSG_WHOLEPKT;
3163     im->im_mbz		= 0;
3164     im->im_vif		= vifp - viftable;
3165     im->im_src		= ip->ip_src;
3166     im->im_dst		= ip->ip_dst;
3167 
3168     k_igmpsrc.sin_addr	= ip->ip_src;
3169 
3170     mrtstat.mrts_upcalls++;
3171 
3172     if (socket_send(ip_mrouter, mb_first, &k_igmpsrc) < 0) {
3173 	if (mrtdebug & DEBUG_PIM)
3174 	    log(LOG_WARNING,
3175 		"mcast: pim_register_send_upcall: ip_mrouter socket queue full");
3176 	++mrtstat.mrts_upq_sockfull;
3177 	return ENOBUFS;
3178     }
3179 
3180     /* Keep statistics */
3181     pimstat.pims_snd_registers_msgs++;
3182     pimstat.pims_snd_registers_bytes += len;
3183 
3184     return 0;
3185 }
3186 
3187 /*
3188  * Encapsulate the data packet in PIM Register message and send it to the RP.
3189  */
3190 static int
3191 pim_register_send_rp(struct ip *ip, struct vif *vifp,
3192 	struct mbuf *mb_copy, struct mfc *rt)
3193 {
3194     struct mbuf *mb_first;
3195     struct ip *ip_outer;
3196     struct pim_encap_pimhdr *pimhdr;
3197     int len = ntohs(ip->ip_len);
3198     vifi_t vifi = rt->mfc_parent;
3199 
3200     VIF_LOCK_ASSERT();
3201 
3202     if ((vifi >= numvifs) || (viftable[vifi].v_lcl_addr.s_addr == 0)) {
3203 	m_freem(mb_copy);
3204 	return EADDRNOTAVAIL;		/* The iif vif is invalid */
3205     }
3206 
3207     /*
3208      * Add a new mbuf with the encapsulating header
3209      */
3210     MGETHDR(mb_first, M_DONTWAIT, MT_DATA);
3211     if (mb_first == NULL) {
3212 	m_freem(mb_copy);
3213 	return ENOBUFS;
3214     }
3215     mb_first->m_data += max_linkhdr;
3216     mb_first->m_len = sizeof(pim_encap_iphdr) + sizeof(pim_encap_pimhdr);
3217     mb_first->m_next = mb_copy;
3218 
3219     mb_first->m_pkthdr.len = len + mb_first->m_len;
3220 
3221     /*
3222      * Fill in the encapsulating IP and PIM header
3223      */
3224     ip_outer = mtod(mb_first, struct ip *);
3225     *ip_outer = pim_encap_iphdr;
3226     ip_outer->ip_id = ip_newid();
3227     ip_outer->ip_len = len + sizeof(pim_encap_iphdr) + sizeof(pim_encap_pimhdr);
3228     ip_outer->ip_src = viftable[vifi].v_lcl_addr;
3229     ip_outer->ip_dst = rt->mfc_rp;
3230     /*
3231      * Copy the inner header TOS to the outer header, and take care of the
3232      * IP_DF bit.
3233      */
3234     ip_outer->ip_tos = ip->ip_tos;
3235     if (ntohs(ip->ip_off) & IP_DF)
3236 	ip_outer->ip_off |= IP_DF;
3237     pimhdr = (struct pim_encap_pimhdr *)((caddr_t)ip_outer
3238 					 + sizeof(pim_encap_iphdr));
3239     *pimhdr = pim_encap_pimhdr;
3240     /* If the iif crosses a border, set the Border-bit */
3241     if (rt->mfc_flags[vifi] & MRT_MFC_FLAGS_BORDER_VIF & mrt_api_config)
3242 	pimhdr->flags |= htonl(PIM_BORDER_REGISTER);
3243 
3244     mb_first->m_data += sizeof(pim_encap_iphdr);
3245     pimhdr->pim.pim_cksum = in_cksum(mb_first, sizeof(pim_encap_pimhdr));
3246     mb_first->m_data -= sizeof(pim_encap_iphdr);
3247 
3248     if (vifp->v_rate_limit == 0)
3249 	tbf_send_packet(vifp, mb_first);
3250     else
3251 	tbf_control(vifp, mb_first, ip, ip_outer->ip_len);
3252 
3253     /* Keep statistics */
3254     pimstat.pims_snd_registers_msgs++;
3255     pimstat.pims_snd_registers_bytes += len;
3256 
3257     return 0;
3258 }
3259 
3260 /*
3261  * PIM-SMv2 and PIM-DM messages processing.
3262  * Receives and verifies the PIM control messages, and passes them
3263  * up to the listening socket, using rip_input().
3264  * The only message with special processing is the PIM_REGISTER message
3265  * (used by PIM-SM): the PIM header is stripped off, and the inner packet
3266  * is passed to if_simloop().
3267  */
3268 void
3269 pim_input(struct mbuf *m, int off)
3270 {
3271     struct ip *ip = mtod(m, struct ip *);
3272     struct pim *pim;
3273     int minlen;
3274     int datalen = ip->ip_len;
3275     int ip_tos;
3276     int iphlen = off;
3277 
3278     /* Keep statistics */
3279     pimstat.pims_rcv_total_msgs++;
3280     pimstat.pims_rcv_total_bytes += datalen;
3281 
3282     /*
3283      * Validate lengths
3284      */
3285     if (datalen < PIM_MINLEN) {
3286 	pimstat.pims_rcv_tooshort++;
3287 	log(LOG_ERR, "pim_input: packet size too small %d from %lx\n",
3288 	    datalen, (u_long)ip->ip_src.s_addr);
3289 	m_freem(m);
3290 	return;
3291     }
3292 
3293     /*
3294      * If the packet is at least as big as a REGISTER, go agead
3295      * and grab the PIM REGISTER header size, to avoid another
3296      * possible m_pullup() later.
3297      *
3298      * PIM_MINLEN       == pimhdr + u_int32_t == 4 + 4 = 8
3299      * PIM_REG_MINLEN   == pimhdr + reghdr + encap_iphdr == 4 + 4 + 20 = 28
3300      */
3301     minlen = iphlen + (datalen >= PIM_REG_MINLEN ? PIM_REG_MINLEN : PIM_MINLEN);
3302     /*
3303      * Get the IP and PIM headers in contiguous memory, and
3304      * possibly the PIM REGISTER header.
3305      */
3306     if ((m->m_flags & M_EXT || m->m_len < minlen) &&
3307 	(m = m_pullup(m, minlen)) == 0) {
3308 	log(LOG_ERR, "pim_input: m_pullup failure\n");
3309 	return;
3310     }
3311     /* m_pullup() may have given us a new mbuf so reset ip. */
3312     ip = mtod(m, struct ip *);
3313     ip_tos = ip->ip_tos;
3314 
3315     /* adjust mbuf to point to the PIM header */
3316     m->m_data += iphlen;
3317     m->m_len  -= iphlen;
3318     pim = mtod(m, struct pim *);
3319 
3320     /*
3321      * Validate checksum. If PIM REGISTER, exclude the data packet.
3322      *
3323      * XXX: some older PIMv2 implementations don't make this distinction,
3324      * so for compatibility reason perform the checksum over part of the
3325      * message, and if error, then over the whole message.
3326      */
3327     if (PIM_VT_T(pim->pim_vt) == PIM_REGISTER && in_cksum(m, PIM_MINLEN) == 0) {
3328 	/* do nothing, checksum okay */
3329     } else if (in_cksum(m, datalen)) {
3330 	pimstat.pims_rcv_badsum++;
3331 	if (mrtdebug & DEBUG_PIM)
3332 	    log(LOG_DEBUG, "pim_input: invalid checksum");
3333 	m_freem(m);
3334 	return;
3335     }
3336 
3337     /* PIM version check */
3338     if (PIM_VT_V(pim->pim_vt) < PIM_VERSION) {
3339 	pimstat.pims_rcv_badversion++;
3340 	log(LOG_ERR, "pim_input: incorrect version %d, expecting %d\n",
3341 	    PIM_VT_V(pim->pim_vt), PIM_VERSION);
3342 	m_freem(m);
3343 	return;
3344     }
3345 
3346     /* restore mbuf back to the outer IP */
3347     m->m_data -= iphlen;
3348     m->m_len  += iphlen;
3349 
3350     if (PIM_VT_T(pim->pim_vt) == PIM_REGISTER) {
3351 	/*
3352 	 * Since this is a REGISTER, we'll make a copy of the register
3353 	 * headers ip + pim + u_int32 + encap_ip, to be passed up to the
3354 	 * routing daemon.
3355 	 */
3356 	struct sockaddr_in dst = { sizeof(dst), AF_INET };
3357 	struct mbuf *mcp;
3358 	struct ip *encap_ip;
3359 	u_int32_t *reghdr;
3360 	struct ifnet *vifp;
3361 
3362 	VIF_LOCK();
3363 	if ((reg_vif_num >= numvifs) || (reg_vif_num == VIFI_INVALID)) {
3364 	    VIF_UNLOCK();
3365 	    if (mrtdebug & DEBUG_PIM)
3366 		log(LOG_DEBUG,
3367 		    "pim_input: register vif not set: %d\n", reg_vif_num);
3368 	    m_freem(m);
3369 	    return;
3370 	}
3371 	/* XXX need refcnt? */
3372 	vifp = viftable[reg_vif_num].v_ifp;
3373 	VIF_UNLOCK();
3374 
3375 	/*
3376 	 * Validate length
3377 	 */
3378 	if (datalen < PIM_REG_MINLEN) {
3379 	    pimstat.pims_rcv_tooshort++;
3380 	    pimstat.pims_rcv_badregisters++;
3381 	    log(LOG_ERR,
3382 		"pim_input: register packet size too small %d from %lx\n",
3383 		datalen, (u_long)ip->ip_src.s_addr);
3384 	    m_freem(m);
3385 	    return;
3386 	}
3387 
3388 	reghdr = (u_int32_t *)(pim + 1);
3389 	encap_ip = (struct ip *)(reghdr + 1);
3390 
3391 	if (mrtdebug & DEBUG_PIM) {
3392 	    log(LOG_DEBUG,
3393 		"pim_input[register], encap_ip: %lx -> %lx, encap_ip len %d\n",
3394 		(u_long)ntohl(encap_ip->ip_src.s_addr),
3395 		(u_long)ntohl(encap_ip->ip_dst.s_addr),
3396 		ntohs(encap_ip->ip_len));
3397 	}
3398 
3399 	/* verify the version number of the inner packet */
3400 	if (encap_ip->ip_v != IPVERSION) {
3401 	    pimstat.pims_rcv_badregisters++;
3402 	    if (mrtdebug & DEBUG_PIM) {
3403 		log(LOG_DEBUG, "pim_input: invalid IP version (%d) "
3404 		    "of the inner packet\n", encap_ip->ip_v);
3405 	    }
3406 	    m_freem(m);
3407 	    return;
3408 	}
3409 
3410 	/* verify the inner packet is destined to a mcast group */
3411 	if (!IN_MULTICAST(ntohl(encap_ip->ip_dst.s_addr))) {
3412 	    pimstat.pims_rcv_badregisters++;
3413 	    if (mrtdebug & DEBUG_PIM)
3414 		log(LOG_DEBUG,
3415 		    "pim_input: inner packet of register is not "
3416 		    "multicast %lx\n",
3417 		    (u_long)ntohl(encap_ip->ip_dst.s_addr));
3418 	    m_freem(m);
3419 	    return;
3420 	}
3421 
3422 	/* If a NULL_REGISTER, pass it to the daemon */
3423 	if ((ntohl(*reghdr) & PIM_NULL_REGISTER))
3424 	    goto pim_input_to_daemon;
3425 
3426 	/*
3427 	 * Copy the TOS from the outer IP header to the inner IP header.
3428 	 */
3429 	if (encap_ip->ip_tos != ip_tos) {
3430 	    /* Outer TOS -> inner TOS */
3431 	    encap_ip->ip_tos = ip_tos;
3432 	    /* Recompute the inner header checksum. Sigh... */
3433 
3434 	    /* adjust mbuf to point to the inner IP header */
3435 	    m->m_data += (iphlen + PIM_MINLEN);
3436 	    m->m_len  -= (iphlen + PIM_MINLEN);
3437 
3438 	    encap_ip->ip_sum = 0;
3439 	    encap_ip->ip_sum = in_cksum(m, encap_ip->ip_hl << 2);
3440 
3441 	    /* restore mbuf to point back to the outer IP header */
3442 	    m->m_data -= (iphlen + PIM_MINLEN);
3443 	    m->m_len  += (iphlen + PIM_MINLEN);
3444 	}
3445 
3446 	/*
3447 	 * Decapsulate the inner IP packet and loopback to forward it
3448 	 * as a normal multicast packet. Also, make a copy of the
3449 	 *     outer_iphdr + pimhdr + reghdr + encap_iphdr
3450 	 * to pass to the daemon later, so it can take the appropriate
3451 	 * actions (e.g., send back PIM_REGISTER_STOP).
3452 	 * XXX: here m->m_data points to the outer IP header.
3453 	 */
3454 	mcp = m_copy(m, 0, iphlen + PIM_REG_MINLEN);
3455 	if (mcp == NULL) {
3456 	    log(LOG_ERR,
3457 		"pim_input: pim register: could not copy register head\n");
3458 	    m_freem(m);
3459 	    return;
3460 	}
3461 
3462 	/* Keep statistics */
3463 	/* XXX: registers_bytes include only the encap. mcast pkt */
3464 	pimstat.pims_rcv_registers_msgs++;
3465 	pimstat.pims_rcv_registers_bytes += ntohs(encap_ip->ip_len);
3466 
3467 	/*
3468 	 * forward the inner ip packet; point m_data at the inner ip.
3469 	 */
3470 	m_adj(m, iphlen + PIM_MINLEN);
3471 
3472 	if (mrtdebug & DEBUG_PIM) {
3473 	    log(LOG_DEBUG,
3474 		"pim_input: forwarding decapsulated register: "
3475 		"src %lx, dst %lx, vif %d\n",
3476 		(u_long)ntohl(encap_ip->ip_src.s_addr),
3477 		(u_long)ntohl(encap_ip->ip_dst.s_addr),
3478 		reg_vif_num);
3479 	}
3480 	/* NB: vifp was collected above; can it change on us? */
3481 	if_simloop(vifp, m, dst.sin_family, 0);
3482 
3483 	/* prepare the register head to send to the mrouting daemon */
3484 	m = mcp;
3485     }
3486 
3487 pim_input_to_daemon:
3488     /*
3489      * Pass the PIM message up to the daemon; if it is a Register message,
3490      * pass the 'head' only up to the daemon. This includes the
3491      * outer IP header, PIM header, PIM-Register header and the
3492      * inner IP header.
3493      * XXX: the outer IP header pkt size of a Register is not adjust to
3494      * reflect the fact that the inner multicast data is truncated.
3495      */
3496     rip_input(m, iphlen);
3497 
3498     return;
3499 }
3500 #endif /* PIM */
3501 
3502 static int
3503 ip_mroute_modevent(module_t mod, int type, void *unused)
3504 {
3505     switch (type) {
3506     case MOD_LOAD:
3507 	mtx_init(&mrouter_mtx, "mrouter initialization", NULL, MTX_DEF);
3508 	MFC_LOCK_INIT();
3509 	VIF_LOCK_INIT();
3510 	ip_mrouter_reset();
3511 	ip_mcast_src = X_ip_mcast_src;
3512 	ip_mforward = X_ip_mforward;
3513 	ip_mrouter_done = X_ip_mrouter_done;
3514 	ip_mrouter_get = X_ip_mrouter_get;
3515 	ip_mrouter_set = X_ip_mrouter_set;
3516 	ip_rsvp_force_done = X_ip_rsvp_force_done;
3517 	ip_rsvp_vif = X_ip_rsvp_vif;
3518 	legal_vif_num = X_legal_vif_num;
3519 	mrt_ioctl = X_mrt_ioctl;
3520 	rsvp_input_p = X_rsvp_input;
3521 	break;
3522 
3523     case MOD_UNLOAD:
3524 	/*
3525 	 * Typically module unload happens after the user-level
3526 	 * process has shutdown the kernel services (the check
3527 	 * below insures someone can't just yank the module out
3528 	 * from under a running process).  But if the module is
3529 	 * just loaded and then unloaded w/o starting up a user
3530 	 * process we still need to cleanup.
3531 	 */
3532 	if (ip_mrouter)
3533 	    return EINVAL;
3534 
3535 	X_ip_mrouter_done();
3536 	ip_mcast_src = NULL;
3537 	ip_mforward = NULL;
3538 	ip_mrouter_done = NULL;
3539 	ip_mrouter_get = NULL;
3540 	ip_mrouter_set = NULL;
3541 	ip_rsvp_force_done = NULL;
3542 	ip_rsvp_vif = NULL;
3543 	legal_vif_num = NULL;
3544 	mrt_ioctl = NULL;
3545 	rsvp_input_p = NULL;
3546 	VIF_LOCK_DESTROY();
3547 	MFC_LOCK_DESTROY();
3548 	mtx_destroy(&mrouter_mtx);
3549 	break;
3550     default:
3551 	return EOPNOTSUPP;
3552     }
3553     return 0;
3554 }
3555 
3556 static moduledata_t ip_mroutemod = {
3557     "ip_mroute",
3558     ip_mroute_modevent,
3559     0
3560 };
3561 DECLARE_MODULE(ip_mroute, ip_mroutemod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3562