xref: /openbmc/linux/net/ipv4/igmp.c (revision 54cbac81)
1 /*
2  *	Linux NET3:	Internet Group Management Protocol  [IGMP]
3  *
4  *	This code implements the IGMP protocol as defined in RFC1112. There has
5  *	been a further revision of this protocol since which is now supported.
6  *
7  *	If you have trouble with this module be careful what gcc you have used,
8  *	the older version didn't come out right using gcc 2.5.8, the newer one
9  *	seems to fall out with gcc 2.6.2.
10  *
11  *	Authors:
12  *		Alan Cox <alan@lxorguk.ukuu.org.uk>
13  *
14  *	This program is free software; you can redistribute it and/or
15  *	modify it under the terms of the GNU General Public License
16  *	as published by the Free Software Foundation; either version
17  *	2 of the License, or (at your option) any later version.
18  *
19  *	Fixes:
20  *
21  *		Alan Cox	:	Added lots of __inline__ to optimise
22  *					the memory usage of all the tiny little
23  *					functions.
24  *		Alan Cox	:	Dumped the header building experiment.
25  *		Alan Cox	:	Minor tweaks ready for multicast routing
26  *					and extended IGMP protocol.
27  *		Alan Cox	:	Removed a load of inline directives. Gcc 2.5.8
28  *					writes utterly bogus code otherwise (sigh)
29  *					fixed IGMP loopback to behave in the manner
30  *					desired by mrouted, fixed the fact it has been
31  *					broken since 1.3.6 and cleaned up a few minor
32  *					points.
33  *
34  *		Chih-Jen Chang	:	Tried to revise IGMP to Version 2
35  *		Tsu-Sheng Tsao		E-mail: chihjenc@scf.usc.edu and tsusheng@scf.usc.edu
36  *					The enhancements are mainly based on Steve Deering's
37  * 					ipmulti-3.5 source code.
38  *		Chih-Jen Chang	:	Added the igmp_get_mrouter_info and
39  *		Tsu-Sheng Tsao		igmp_set_mrouter_info to keep track of
40  *					the mrouted version on that device.
41  *		Chih-Jen Chang	:	Added the max_resp_time parameter to
42  *		Tsu-Sheng Tsao		igmp_heard_query(). Using this parameter
43  *					to identify the multicast router version
44  *					and do what the IGMP version 2 specified.
45  *		Chih-Jen Chang	:	Added a timer to revert to IGMP V2 router
46  *		Tsu-Sheng Tsao		if the specified time expired.
47  *		Alan Cox	:	Stop IGMP from 0.0.0.0 being accepted.
48  *		Alan Cox	:	Use GFP_ATOMIC in the right places.
49  *		Christian Daudt :	igmp timer wasn't set for local group
50  *					memberships but was being deleted,
51  *					which caused a "del_timer() called
52  *					from %p with timer not initialized\n"
53  *					message (960131).
54  *		Christian Daudt :	removed del_timer from
55  *					igmp_timer_expire function (960205).
56  *             Christian Daudt :       igmp_heard_report now only calls
57  *                                     igmp_timer_expire if tm->running is
58  *                                     true (960216).
59  *		Malcolm Beattie :	ttl comparison wrong in igmp_rcv made
60  *					igmp_heard_query never trigger. Expiry
61  *					miscalculation fixed in igmp_heard_query
62  *					and random() made to return unsigned to
63  *					prevent negative expiry times.
64  *		Alexey Kuznetsov:	Wrong group leaving behaviour, backport
65  *					fix from pending 2.1.x patches.
66  *		Alan Cox:		Forget to enable FDDI support earlier.
67  *		Alexey Kuznetsov:	Fixed leaving groups on device down.
68  *		Alexey Kuznetsov:	Accordance to igmp-v2-06 draft.
69  *		David L Stevens:	IGMPv3 support, with help from
70  *					Vinay Kulkarni
71  */
72 
73 #include <linux/module.h>
74 #include <linux/slab.h>
75 #include <asm/uaccess.h>
76 #include <linux/types.h>
77 #include <linux/kernel.h>
78 #include <linux/jiffies.h>
79 #include <linux/string.h>
80 #include <linux/socket.h>
81 #include <linux/sockios.h>
82 #include <linux/in.h>
83 #include <linux/inet.h>
84 #include <linux/netdevice.h>
85 #include <linux/skbuff.h>
86 #include <linux/inetdevice.h>
87 #include <linux/igmp.h>
88 #include <linux/if_arp.h>
89 #include <linux/rtnetlink.h>
90 #include <linux/times.h>
91 
92 #include <net/net_namespace.h>
93 #include <net/arp.h>
94 #include <net/ip.h>
95 #include <net/protocol.h>
96 #include <net/route.h>
97 #include <net/sock.h>
98 #include <net/checksum.h>
99 #include <linux/netfilter_ipv4.h>
100 #ifdef CONFIG_IP_MROUTE
101 #include <linux/mroute.h>
102 #endif
103 #ifdef CONFIG_PROC_FS
104 #include <linux/proc_fs.h>
105 #include <linux/seq_file.h>
106 #endif
107 
108 #define IP_MAX_MEMBERSHIPS	20
109 #define IP_MAX_MSF		10
110 
111 #ifdef CONFIG_IP_MULTICAST
112 /* Parameter names and values are taken from igmp-v2-06 draft */
113 
114 #define IGMP_V1_Router_Present_Timeout		(400*HZ)
115 #define IGMP_V2_Router_Present_Timeout		(400*HZ)
116 #define IGMP_Unsolicited_Report_Interval	(10*HZ)
117 #define IGMP_Query_Response_Interval		(10*HZ)
118 #define IGMP_Unsolicited_Report_Count		2
119 
120 
121 #define IGMP_Initial_Report_Delay		(1)
122 
123 /* IGMP_Initial_Report_Delay is not from IGMP specs!
124  * IGMP specs require to report membership immediately after
125  * joining a group, but we delay the first report by a
126  * small interval. It seems more natural and still does not
127  * contradict to specs provided this delay is small enough.
128  */
129 
130 #define IGMP_V1_SEEN(in_dev) \
131 	(IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
132 	 IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
133 	 ((in_dev)->mr_v1_seen && \
134 	  time_before(jiffies, (in_dev)->mr_v1_seen)))
135 #define IGMP_V2_SEEN(in_dev) \
136 	(IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
137 	 IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
138 	 ((in_dev)->mr_v2_seen && \
139 	  time_before(jiffies, (in_dev)->mr_v2_seen)))
140 
141 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im);
142 static void igmpv3_del_delrec(struct in_device *in_dev, __be32 multiaddr);
143 static void igmpv3_clear_delrec(struct in_device *in_dev);
144 static int sf_setstate(struct ip_mc_list *pmc);
145 static void sf_markstate(struct ip_mc_list *pmc);
146 #endif
147 static void ip_mc_clear_src(struct ip_mc_list *pmc);
148 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
149 			 int sfcount, __be32 *psfsrc, int delta);
150 
151 static void ip_ma_put(struct ip_mc_list *im)
152 {
153 	if (atomic_dec_and_test(&im->refcnt)) {
154 		in_dev_put(im->interface);
155 		kfree_rcu(im, rcu);
156 	}
157 }
158 
159 #define for_each_pmc_rcu(in_dev, pmc)				\
160 	for (pmc = rcu_dereference(in_dev->mc_list);		\
161 	     pmc != NULL;					\
162 	     pmc = rcu_dereference(pmc->next_rcu))
163 
164 #define for_each_pmc_rtnl(in_dev, pmc)				\
165 	for (pmc = rtnl_dereference(in_dev->mc_list);		\
166 	     pmc != NULL;					\
167 	     pmc = rtnl_dereference(pmc->next_rcu))
168 
169 #ifdef CONFIG_IP_MULTICAST
170 
171 /*
172  *	Timer management
173  */
174 
175 static void igmp_stop_timer(struct ip_mc_list *im)
176 {
177 	spin_lock_bh(&im->lock);
178 	if (del_timer(&im->timer))
179 		atomic_dec(&im->refcnt);
180 	im->tm_running = 0;
181 	im->reporter = 0;
182 	im->unsolicit_count = 0;
183 	spin_unlock_bh(&im->lock);
184 }
185 
186 /* It must be called with locked im->lock */
187 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
188 {
189 	int tv = net_random() % max_delay;
190 
191 	im->tm_running = 1;
192 	if (!mod_timer(&im->timer, jiffies+tv+2))
193 		atomic_inc(&im->refcnt);
194 }
195 
196 static void igmp_gq_start_timer(struct in_device *in_dev)
197 {
198 	int tv = net_random() % in_dev->mr_maxdelay;
199 
200 	in_dev->mr_gq_running = 1;
201 	if (!mod_timer(&in_dev->mr_gq_timer, jiffies+tv+2))
202 		in_dev_hold(in_dev);
203 }
204 
205 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
206 {
207 	int tv = net_random() % delay;
208 
209 	if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
210 		in_dev_hold(in_dev);
211 }
212 
213 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
214 {
215 	spin_lock_bh(&im->lock);
216 	im->unsolicit_count = 0;
217 	if (del_timer(&im->timer)) {
218 		if ((long)(im->timer.expires-jiffies) < max_delay) {
219 			add_timer(&im->timer);
220 			im->tm_running = 1;
221 			spin_unlock_bh(&im->lock);
222 			return;
223 		}
224 		atomic_dec(&im->refcnt);
225 	}
226 	igmp_start_timer(im, max_delay);
227 	spin_unlock_bh(&im->lock);
228 }
229 
230 
231 /*
232  *	Send an IGMP report.
233  */
234 
235 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
236 
237 
238 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
239 	int gdeleted, int sdeleted)
240 {
241 	switch (type) {
242 	case IGMPV3_MODE_IS_INCLUDE:
243 	case IGMPV3_MODE_IS_EXCLUDE:
244 		if (gdeleted || sdeleted)
245 			return 0;
246 		if (!(pmc->gsquery && !psf->sf_gsresp)) {
247 			if (pmc->sfmode == MCAST_INCLUDE)
248 				return 1;
249 			/* don't include if this source is excluded
250 			 * in all filters
251 			 */
252 			if (psf->sf_count[MCAST_INCLUDE])
253 				return type == IGMPV3_MODE_IS_INCLUDE;
254 			return pmc->sfcount[MCAST_EXCLUDE] ==
255 				psf->sf_count[MCAST_EXCLUDE];
256 		}
257 		return 0;
258 	case IGMPV3_CHANGE_TO_INCLUDE:
259 		if (gdeleted || sdeleted)
260 			return 0;
261 		return psf->sf_count[MCAST_INCLUDE] != 0;
262 	case IGMPV3_CHANGE_TO_EXCLUDE:
263 		if (gdeleted || sdeleted)
264 			return 0;
265 		if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
266 		    psf->sf_count[MCAST_INCLUDE])
267 			return 0;
268 		return pmc->sfcount[MCAST_EXCLUDE] ==
269 			psf->sf_count[MCAST_EXCLUDE];
270 	case IGMPV3_ALLOW_NEW_SOURCES:
271 		if (gdeleted || !psf->sf_crcount)
272 			return 0;
273 		return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
274 	case IGMPV3_BLOCK_OLD_SOURCES:
275 		if (pmc->sfmode == MCAST_INCLUDE)
276 			return gdeleted || (psf->sf_crcount && sdeleted);
277 		return psf->sf_crcount && !gdeleted && !sdeleted;
278 	}
279 	return 0;
280 }
281 
282 static int
283 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
284 {
285 	struct ip_sf_list *psf;
286 	int scount = 0;
287 
288 	for (psf=pmc->sources; psf; psf=psf->sf_next) {
289 		if (!is_in(pmc, psf, type, gdeleted, sdeleted))
290 			continue;
291 		scount++;
292 	}
293 	return scount;
294 }
295 
296 #define igmp_skb_size(skb) (*(unsigned int *)((skb)->cb))
297 
298 static struct sk_buff *igmpv3_newpack(struct net_device *dev, int size)
299 {
300 	struct sk_buff *skb;
301 	struct rtable *rt;
302 	struct iphdr *pip;
303 	struct igmpv3_report *pig;
304 	struct net *net = dev_net(dev);
305 	struct flowi4 fl4;
306 	int hlen = LL_RESERVED_SPACE(dev);
307 	int tlen = dev->needed_tailroom;
308 
309 	while (1) {
310 		skb = alloc_skb(size + hlen + tlen,
311 				GFP_ATOMIC | __GFP_NOWARN);
312 		if (skb)
313 			break;
314 		size >>= 1;
315 		if (size < 256)
316 			return NULL;
317 	}
318 	igmp_skb_size(skb) = size;
319 
320 	rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
321 				   0, 0,
322 				   IPPROTO_IGMP, 0, dev->ifindex);
323 	if (IS_ERR(rt)) {
324 		kfree_skb(skb);
325 		return NULL;
326 	}
327 
328 	skb_dst_set(skb, &rt->dst);
329 	skb->dev = dev;
330 
331 	skb_reserve(skb, hlen);
332 
333 	skb_reset_network_header(skb);
334 	pip = ip_hdr(skb);
335 	skb_put(skb, sizeof(struct iphdr) + 4);
336 
337 	pip->version  = 4;
338 	pip->ihl      = (sizeof(struct iphdr)+4)>>2;
339 	pip->tos      = 0xc0;
340 	pip->frag_off = htons(IP_DF);
341 	pip->ttl      = 1;
342 	pip->daddr    = fl4.daddr;
343 	pip->saddr    = fl4.saddr;
344 	pip->protocol = IPPROTO_IGMP;
345 	pip->tot_len  = 0;	/* filled in later */
346 	ip_select_ident(pip, &rt->dst, NULL);
347 	((u8 *)&pip[1])[0] = IPOPT_RA;
348 	((u8 *)&pip[1])[1] = 4;
349 	((u8 *)&pip[1])[2] = 0;
350 	((u8 *)&pip[1])[3] = 0;
351 
352 	skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
353 	skb_put(skb, sizeof(*pig));
354 	pig = igmpv3_report_hdr(skb);
355 	pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
356 	pig->resv1 = 0;
357 	pig->csum = 0;
358 	pig->resv2 = 0;
359 	pig->ngrec = 0;
360 	return skb;
361 }
362 
363 static int igmpv3_sendpack(struct sk_buff *skb)
364 {
365 	struct igmphdr *pig = igmp_hdr(skb);
366 	const int igmplen = skb->tail - skb->transport_header;
367 
368 	pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
369 
370 	return ip_local_out(skb);
371 }
372 
373 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
374 {
375 	return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
376 }
377 
378 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
379 	int type, struct igmpv3_grec **ppgr)
380 {
381 	struct net_device *dev = pmc->interface->dev;
382 	struct igmpv3_report *pih;
383 	struct igmpv3_grec *pgr;
384 
385 	if (!skb)
386 		skb = igmpv3_newpack(dev, dev->mtu);
387 	if (!skb)
388 		return NULL;
389 	pgr = (struct igmpv3_grec *)skb_put(skb, sizeof(struct igmpv3_grec));
390 	pgr->grec_type = type;
391 	pgr->grec_auxwords = 0;
392 	pgr->grec_nsrcs = 0;
393 	pgr->grec_mca = pmc->multiaddr;
394 	pih = igmpv3_report_hdr(skb);
395 	pih->ngrec = htons(ntohs(pih->ngrec)+1);
396 	*ppgr = pgr;
397 	return skb;
398 }
399 
400 #define AVAILABLE(skb) ((skb) ? ((skb)->dev ? igmp_skb_size(skb) - (skb)->len : \
401 	skb_tailroom(skb)) : 0)
402 
403 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
404 	int type, int gdeleted, int sdeleted)
405 {
406 	struct net_device *dev = pmc->interface->dev;
407 	struct igmpv3_report *pih;
408 	struct igmpv3_grec *pgr = NULL;
409 	struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
410 	int scount, stotal, first, isquery, truncate;
411 
412 	if (pmc->multiaddr == IGMP_ALL_HOSTS)
413 		return skb;
414 
415 	isquery = type == IGMPV3_MODE_IS_INCLUDE ||
416 		  type == IGMPV3_MODE_IS_EXCLUDE;
417 	truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
418 		    type == IGMPV3_CHANGE_TO_EXCLUDE;
419 
420 	stotal = scount = 0;
421 
422 	psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
423 
424 	if (!*psf_list)
425 		goto empty_source;
426 
427 	pih = skb ? igmpv3_report_hdr(skb) : NULL;
428 
429 	/* EX and TO_EX get a fresh packet, if needed */
430 	if (truncate) {
431 		if (pih && pih->ngrec &&
432 		    AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
433 			if (skb)
434 				igmpv3_sendpack(skb);
435 			skb = igmpv3_newpack(dev, dev->mtu);
436 		}
437 	}
438 	first = 1;
439 	psf_prev = NULL;
440 	for (psf=*psf_list; psf; psf=psf_next) {
441 		__be32 *psrc;
442 
443 		psf_next = psf->sf_next;
444 
445 		if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
446 			psf_prev = psf;
447 			continue;
448 		}
449 
450 		/* clear marks on query responses */
451 		if (isquery)
452 			psf->sf_gsresp = 0;
453 
454 		if (AVAILABLE(skb) < sizeof(__be32) +
455 		    first*sizeof(struct igmpv3_grec)) {
456 			if (truncate && !first)
457 				break;	 /* truncate these */
458 			if (pgr)
459 				pgr->grec_nsrcs = htons(scount);
460 			if (skb)
461 				igmpv3_sendpack(skb);
462 			skb = igmpv3_newpack(dev, dev->mtu);
463 			first = 1;
464 			scount = 0;
465 		}
466 		if (first) {
467 			skb = add_grhead(skb, pmc, type, &pgr);
468 			first = 0;
469 		}
470 		if (!skb)
471 			return NULL;
472 		psrc = (__be32 *)skb_put(skb, sizeof(__be32));
473 		*psrc = psf->sf_inaddr;
474 		scount++; stotal++;
475 		if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
476 		     type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
477 			psf->sf_crcount--;
478 			if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
479 				if (psf_prev)
480 					psf_prev->sf_next = psf->sf_next;
481 				else
482 					*psf_list = psf->sf_next;
483 				kfree(psf);
484 				continue;
485 			}
486 		}
487 		psf_prev = psf;
488 	}
489 
490 empty_source:
491 	if (!stotal) {
492 		if (type == IGMPV3_ALLOW_NEW_SOURCES ||
493 		    type == IGMPV3_BLOCK_OLD_SOURCES)
494 			return skb;
495 		if (pmc->crcount || isquery) {
496 			/* make sure we have room for group header */
497 			if (skb && AVAILABLE(skb)<sizeof(struct igmpv3_grec)) {
498 				igmpv3_sendpack(skb);
499 				skb = NULL; /* add_grhead will get a new one */
500 			}
501 			skb = add_grhead(skb, pmc, type, &pgr);
502 		}
503 	}
504 	if (pgr)
505 		pgr->grec_nsrcs = htons(scount);
506 
507 	if (isquery)
508 		pmc->gsquery = 0;	/* clear query state on report */
509 	return skb;
510 }
511 
512 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
513 {
514 	struct sk_buff *skb = NULL;
515 	int type;
516 
517 	if (!pmc) {
518 		rcu_read_lock();
519 		for_each_pmc_rcu(in_dev, pmc) {
520 			if (pmc->multiaddr == IGMP_ALL_HOSTS)
521 				continue;
522 			spin_lock_bh(&pmc->lock);
523 			if (pmc->sfcount[MCAST_EXCLUDE])
524 				type = IGMPV3_MODE_IS_EXCLUDE;
525 			else
526 				type = IGMPV3_MODE_IS_INCLUDE;
527 			skb = add_grec(skb, pmc, type, 0, 0);
528 			spin_unlock_bh(&pmc->lock);
529 		}
530 		rcu_read_unlock();
531 	} else {
532 		spin_lock_bh(&pmc->lock);
533 		if (pmc->sfcount[MCAST_EXCLUDE])
534 			type = IGMPV3_MODE_IS_EXCLUDE;
535 		else
536 			type = IGMPV3_MODE_IS_INCLUDE;
537 		skb = add_grec(skb, pmc, type, 0, 0);
538 		spin_unlock_bh(&pmc->lock);
539 	}
540 	if (!skb)
541 		return 0;
542 	return igmpv3_sendpack(skb);
543 }
544 
545 /*
546  * remove zero-count source records from a source filter list
547  */
548 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
549 {
550 	struct ip_sf_list *psf_prev, *psf_next, *psf;
551 
552 	psf_prev = NULL;
553 	for (psf=*ppsf; psf; psf = psf_next) {
554 		psf_next = psf->sf_next;
555 		if (psf->sf_crcount == 0) {
556 			if (psf_prev)
557 				psf_prev->sf_next = psf->sf_next;
558 			else
559 				*ppsf = psf->sf_next;
560 			kfree(psf);
561 		} else
562 			psf_prev = psf;
563 	}
564 }
565 
566 static void igmpv3_send_cr(struct in_device *in_dev)
567 {
568 	struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
569 	struct sk_buff *skb = NULL;
570 	int type, dtype;
571 
572 	rcu_read_lock();
573 	spin_lock_bh(&in_dev->mc_tomb_lock);
574 
575 	/* deleted MCA's */
576 	pmc_prev = NULL;
577 	for (pmc=in_dev->mc_tomb; pmc; pmc=pmc_next) {
578 		pmc_next = pmc->next;
579 		if (pmc->sfmode == MCAST_INCLUDE) {
580 			type = IGMPV3_BLOCK_OLD_SOURCES;
581 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
582 			skb = add_grec(skb, pmc, type, 1, 0);
583 			skb = add_grec(skb, pmc, dtype, 1, 1);
584 		}
585 		if (pmc->crcount) {
586 			if (pmc->sfmode == MCAST_EXCLUDE) {
587 				type = IGMPV3_CHANGE_TO_INCLUDE;
588 				skb = add_grec(skb, pmc, type, 1, 0);
589 			}
590 			pmc->crcount--;
591 			if (pmc->crcount == 0) {
592 				igmpv3_clear_zeros(&pmc->tomb);
593 				igmpv3_clear_zeros(&pmc->sources);
594 			}
595 		}
596 		if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
597 			if (pmc_prev)
598 				pmc_prev->next = pmc_next;
599 			else
600 				in_dev->mc_tomb = pmc_next;
601 			in_dev_put(pmc->interface);
602 			kfree(pmc);
603 		} else
604 			pmc_prev = pmc;
605 	}
606 	spin_unlock_bh(&in_dev->mc_tomb_lock);
607 
608 	/* change recs */
609 	for_each_pmc_rcu(in_dev, pmc) {
610 		spin_lock_bh(&pmc->lock);
611 		if (pmc->sfcount[MCAST_EXCLUDE]) {
612 			type = IGMPV3_BLOCK_OLD_SOURCES;
613 			dtype = IGMPV3_ALLOW_NEW_SOURCES;
614 		} else {
615 			type = IGMPV3_ALLOW_NEW_SOURCES;
616 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
617 		}
618 		skb = add_grec(skb, pmc, type, 0, 0);
619 		skb = add_grec(skb, pmc, dtype, 0, 1);	/* deleted sources */
620 
621 		/* filter mode changes */
622 		if (pmc->crcount) {
623 			if (pmc->sfmode == MCAST_EXCLUDE)
624 				type = IGMPV3_CHANGE_TO_EXCLUDE;
625 			else
626 				type = IGMPV3_CHANGE_TO_INCLUDE;
627 			skb = add_grec(skb, pmc, type, 0, 0);
628 			pmc->crcount--;
629 		}
630 		spin_unlock_bh(&pmc->lock);
631 	}
632 	rcu_read_unlock();
633 
634 	if (!skb)
635 		return;
636 	(void) igmpv3_sendpack(skb);
637 }
638 
639 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
640 	int type)
641 {
642 	struct sk_buff *skb;
643 	struct iphdr *iph;
644 	struct igmphdr *ih;
645 	struct rtable *rt;
646 	struct net_device *dev = in_dev->dev;
647 	struct net *net = dev_net(dev);
648 	__be32	group = pmc ? pmc->multiaddr : 0;
649 	struct flowi4 fl4;
650 	__be32	dst;
651 	int hlen, tlen;
652 
653 	if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
654 		return igmpv3_send_report(in_dev, pmc);
655 	else if (type == IGMP_HOST_LEAVE_MESSAGE)
656 		dst = IGMP_ALL_ROUTER;
657 	else
658 		dst = group;
659 
660 	rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
661 				   0, 0,
662 				   IPPROTO_IGMP, 0, dev->ifindex);
663 	if (IS_ERR(rt))
664 		return -1;
665 
666 	hlen = LL_RESERVED_SPACE(dev);
667 	tlen = dev->needed_tailroom;
668 	skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
669 	if (skb == NULL) {
670 		ip_rt_put(rt);
671 		return -1;
672 	}
673 
674 	skb_dst_set(skb, &rt->dst);
675 
676 	skb_reserve(skb, hlen);
677 
678 	skb_reset_network_header(skb);
679 	iph = ip_hdr(skb);
680 	skb_put(skb, sizeof(struct iphdr) + 4);
681 
682 	iph->version  = 4;
683 	iph->ihl      = (sizeof(struct iphdr)+4)>>2;
684 	iph->tos      = 0xc0;
685 	iph->frag_off = htons(IP_DF);
686 	iph->ttl      = 1;
687 	iph->daddr    = dst;
688 	iph->saddr    = fl4.saddr;
689 	iph->protocol = IPPROTO_IGMP;
690 	ip_select_ident(iph, &rt->dst, NULL);
691 	((u8 *)&iph[1])[0] = IPOPT_RA;
692 	((u8 *)&iph[1])[1] = 4;
693 	((u8 *)&iph[1])[2] = 0;
694 	((u8 *)&iph[1])[3] = 0;
695 
696 	ih = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
697 	ih->type = type;
698 	ih->code = 0;
699 	ih->csum = 0;
700 	ih->group = group;
701 	ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
702 
703 	return ip_local_out(skb);
704 }
705 
706 static void igmp_gq_timer_expire(unsigned long data)
707 {
708 	struct in_device *in_dev = (struct in_device *)data;
709 
710 	in_dev->mr_gq_running = 0;
711 	igmpv3_send_report(in_dev, NULL);
712 	__in_dev_put(in_dev);
713 }
714 
715 static void igmp_ifc_timer_expire(unsigned long data)
716 {
717 	struct in_device *in_dev = (struct in_device *)data;
718 
719 	igmpv3_send_cr(in_dev);
720 	if (in_dev->mr_ifc_count) {
721 		in_dev->mr_ifc_count--;
722 		igmp_ifc_start_timer(in_dev, IGMP_Unsolicited_Report_Interval);
723 	}
724 	__in_dev_put(in_dev);
725 }
726 
727 static void igmp_ifc_event(struct in_device *in_dev)
728 {
729 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
730 		return;
731 	in_dev->mr_ifc_count = in_dev->mr_qrv ? in_dev->mr_qrv :
732 		IGMP_Unsolicited_Report_Count;
733 	igmp_ifc_start_timer(in_dev, 1);
734 }
735 
736 
737 static void igmp_timer_expire(unsigned long data)
738 {
739 	struct ip_mc_list *im=(struct ip_mc_list *)data;
740 	struct in_device *in_dev = im->interface;
741 
742 	spin_lock(&im->lock);
743 	im->tm_running = 0;
744 
745 	if (im->unsolicit_count) {
746 		im->unsolicit_count--;
747 		igmp_start_timer(im, IGMP_Unsolicited_Report_Interval);
748 	}
749 	im->reporter = 1;
750 	spin_unlock(&im->lock);
751 
752 	if (IGMP_V1_SEEN(in_dev))
753 		igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
754 	else if (IGMP_V2_SEEN(in_dev))
755 		igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
756 	else
757 		igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
758 
759 	ip_ma_put(im);
760 }
761 
762 /* mark EXCLUDE-mode sources */
763 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
764 {
765 	struct ip_sf_list *psf;
766 	int i, scount;
767 
768 	scount = 0;
769 	for (psf=pmc->sources; psf; psf=psf->sf_next) {
770 		if (scount == nsrcs)
771 			break;
772 		for (i=0; i<nsrcs; i++) {
773 			/* skip inactive filters */
774 			if (psf->sf_count[MCAST_INCLUDE] ||
775 			    pmc->sfcount[MCAST_EXCLUDE] !=
776 			    psf->sf_count[MCAST_EXCLUDE])
777 				break;
778 			if (srcs[i] == psf->sf_inaddr) {
779 				scount++;
780 				break;
781 			}
782 		}
783 	}
784 	pmc->gsquery = 0;
785 	if (scount == nsrcs)	/* all sources excluded */
786 		return 0;
787 	return 1;
788 }
789 
790 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
791 {
792 	struct ip_sf_list *psf;
793 	int i, scount;
794 
795 	if (pmc->sfmode == MCAST_EXCLUDE)
796 		return igmp_xmarksources(pmc, nsrcs, srcs);
797 
798 	/* mark INCLUDE-mode sources */
799 	scount = 0;
800 	for (psf=pmc->sources; psf; psf=psf->sf_next) {
801 		if (scount == nsrcs)
802 			break;
803 		for (i=0; i<nsrcs; i++)
804 			if (srcs[i] == psf->sf_inaddr) {
805 				psf->sf_gsresp = 1;
806 				scount++;
807 				break;
808 			}
809 	}
810 	if (!scount) {
811 		pmc->gsquery = 0;
812 		return 0;
813 	}
814 	pmc->gsquery = 1;
815 	return 1;
816 }
817 
818 /* return true if packet was dropped */
819 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
820 {
821 	struct ip_mc_list *im;
822 
823 	/* Timers are only set for non-local groups */
824 
825 	if (group == IGMP_ALL_HOSTS)
826 		return false;
827 
828 	rcu_read_lock();
829 	for_each_pmc_rcu(in_dev, im) {
830 		if (im->multiaddr == group) {
831 			igmp_stop_timer(im);
832 			break;
833 		}
834 	}
835 	rcu_read_unlock();
836 	return false;
837 }
838 
839 /* return true if packet was dropped */
840 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
841 	int len)
842 {
843 	struct igmphdr 		*ih = igmp_hdr(skb);
844 	struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
845 	struct ip_mc_list	*im;
846 	__be32			group = ih->group;
847 	int			max_delay;
848 	int			mark = 0;
849 
850 
851 	if (len == 8) {
852 		if (ih->code == 0) {
853 			/* Alas, old v1 router presents here. */
854 
855 			max_delay = IGMP_Query_Response_Interval;
856 			in_dev->mr_v1_seen = jiffies +
857 				IGMP_V1_Router_Present_Timeout;
858 			group = 0;
859 		} else {
860 			/* v2 router present */
861 			max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
862 			in_dev->mr_v2_seen = jiffies +
863 				IGMP_V2_Router_Present_Timeout;
864 		}
865 		/* cancel the interface change timer */
866 		in_dev->mr_ifc_count = 0;
867 		if (del_timer(&in_dev->mr_ifc_timer))
868 			__in_dev_put(in_dev);
869 		/* clear deleted report items */
870 		igmpv3_clear_delrec(in_dev);
871 	} else if (len < 12) {
872 		return true;	/* ignore bogus packet; freed by caller */
873 	} else if (IGMP_V1_SEEN(in_dev)) {
874 		/* This is a v3 query with v1 queriers present */
875 		max_delay = IGMP_Query_Response_Interval;
876 		group = 0;
877 	} else if (IGMP_V2_SEEN(in_dev)) {
878 		/* this is a v3 query with v2 queriers present;
879 		 * Interpretation of the max_delay code is problematic here.
880 		 * A real v2 host would use ih_code directly, while v3 has a
881 		 * different encoding. We use the v3 encoding as more likely
882 		 * to be intended in a v3 query.
883 		 */
884 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
885 		if (!max_delay)
886 			max_delay = 1;	/* can't mod w/ 0 */
887 	} else { /* v3 */
888 		if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
889 			return true;
890 
891 		ih3 = igmpv3_query_hdr(skb);
892 		if (ih3->nsrcs) {
893 			if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
894 					   + ntohs(ih3->nsrcs)*sizeof(__be32)))
895 				return true;
896 			ih3 = igmpv3_query_hdr(skb);
897 		}
898 
899 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
900 		if (!max_delay)
901 			max_delay = 1;	/* can't mod w/ 0 */
902 		in_dev->mr_maxdelay = max_delay;
903 		if (ih3->qrv)
904 			in_dev->mr_qrv = ih3->qrv;
905 		if (!group) { /* general query */
906 			if (ih3->nsrcs)
907 				return false;	/* no sources allowed */
908 			igmp_gq_start_timer(in_dev);
909 			return false;
910 		}
911 		/* mark sources to include, if group & source-specific */
912 		mark = ih3->nsrcs != 0;
913 	}
914 
915 	/*
916 	 * - Start the timers in all of our membership records
917 	 *   that the query applies to for the interface on
918 	 *   which the query arrived excl. those that belong
919 	 *   to a "local" group (224.0.0.X)
920 	 * - For timers already running check if they need to
921 	 *   be reset.
922 	 * - Use the igmp->igmp_code field as the maximum
923 	 *   delay possible
924 	 */
925 	rcu_read_lock();
926 	for_each_pmc_rcu(in_dev, im) {
927 		int changed;
928 
929 		if (group && group != im->multiaddr)
930 			continue;
931 		if (im->multiaddr == IGMP_ALL_HOSTS)
932 			continue;
933 		spin_lock_bh(&im->lock);
934 		if (im->tm_running)
935 			im->gsquery = im->gsquery && mark;
936 		else
937 			im->gsquery = mark;
938 		changed = !im->gsquery ||
939 			igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
940 		spin_unlock_bh(&im->lock);
941 		if (changed)
942 			igmp_mod_timer(im, max_delay);
943 	}
944 	rcu_read_unlock();
945 	return false;
946 }
947 
948 /* called in rcu_read_lock() section */
949 int igmp_rcv(struct sk_buff *skb)
950 {
951 	/* This basically follows the spec line by line -- see RFC1112 */
952 	struct igmphdr *ih;
953 	struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
954 	int len = skb->len;
955 	bool dropped = true;
956 
957 	if (in_dev == NULL)
958 		goto drop;
959 
960 	if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
961 		goto drop;
962 
963 	switch (skb->ip_summed) {
964 	case CHECKSUM_COMPLETE:
965 		if (!csum_fold(skb->csum))
966 			break;
967 		/* fall through */
968 	case CHECKSUM_NONE:
969 		skb->csum = 0;
970 		if (__skb_checksum_complete(skb))
971 			goto drop;
972 	}
973 
974 	ih = igmp_hdr(skb);
975 	switch (ih->type) {
976 	case IGMP_HOST_MEMBERSHIP_QUERY:
977 		dropped = igmp_heard_query(in_dev, skb, len);
978 		break;
979 	case IGMP_HOST_MEMBERSHIP_REPORT:
980 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
981 		/* Is it our report looped back? */
982 		if (rt_is_output_route(skb_rtable(skb)))
983 			break;
984 		/* don't rely on MC router hearing unicast reports */
985 		if (skb->pkt_type == PACKET_MULTICAST ||
986 		    skb->pkt_type == PACKET_BROADCAST)
987 			dropped = igmp_heard_report(in_dev, ih->group);
988 		break;
989 	case IGMP_PIM:
990 #ifdef CONFIG_IP_PIMSM_V1
991 		return pim_rcv_v1(skb);
992 #endif
993 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
994 	case IGMP_DVMRP:
995 	case IGMP_TRACE:
996 	case IGMP_HOST_LEAVE_MESSAGE:
997 	case IGMP_MTRACE:
998 	case IGMP_MTRACE_RESP:
999 		break;
1000 	default:
1001 		break;
1002 	}
1003 
1004 drop:
1005 	if (dropped)
1006 		kfree_skb(skb);
1007 	else
1008 		consume_skb(skb);
1009 	return 0;
1010 }
1011 
1012 #endif
1013 
1014 
1015 /*
1016  *	Add a filter to a device
1017  */
1018 
1019 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1020 {
1021 	char buf[MAX_ADDR_LEN];
1022 	struct net_device *dev = in_dev->dev;
1023 
1024 	/* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1025 	   We will get multicast token leakage, when IFF_MULTICAST
1026 	   is changed. This check should be done in ndo_set_rx_mode
1027 	   routine. Something sort of:
1028 	   if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1029 	   --ANK
1030 	   */
1031 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1032 		dev_mc_add(dev, buf);
1033 }
1034 
1035 /*
1036  *	Remove a filter from a device
1037  */
1038 
1039 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1040 {
1041 	char buf[MAX_ADDR_LEN];
1042 	struct net_device *dev = in_dev->dev;
1043 
1044 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1045 		dev_mc_del(dev, buf);
1046 }
1047 
1048 #ifdef CONFIG_IP_MULTICAST
1049 /*
1050  * deleted ip_mc_list manipulation
1051  */
1052 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1053 {
1054 	struct ip_mc_list *pmc;
1055 
1056 	/* this is an "ip_mc_list" for convenience; only the fields below
1057 	 * are actually used. In particular, the refcnt and users are not
1058 	 * used for management of the delete list. Using the same structure
1059 	 * for deleted items allows change reports to use common code with
1060 	 * non-deleted or query-response MCA's.
1061 	 */
1062 	pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1063 	if (!pmc)
1064 		return;
1065 	spin_lock_bh(&im->lock);
1066 	pmc->interface = im->interface;
1067 	in_dev_hold(in_dev);
1068 	pmc->multiaddr = im->multiaddr;
1069 	pmc->crcount = in_dev->mr_qrv ? in_dev->mr_qrv :
1070 		IGMP_Unsolicited_Report_Count;
1071 	pmc->sfmode = im->sfmode;
1072 	if (pmc->sfmode == MCAST_INCLUDE) {
1073 		struct ip_sf_list *psf;
1074 
1075 		pmc->tomb = im->tomb;
1076 		pmc->sources = im->sources;
1077 		im->tomb = im->sources = NULL;
1078 		for (psf=pmc->sources; psf; psf=psf->sf_next)
1079 			psf->sf_crcount = pmc->crcount;
1080 	}
1081 	spin_unlock_bh(&im->lock);
1082 
1083 	spin_lock_bh(&in_dev->mc_tomb_lock);
1084 	pmc->next = in_dev->mc_tomb;
1085 	in_dev->mc_tomb = pmc;
1086 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1087 }
1088 
1089 static void igmpv3_del_delrec(struct in_device *in_dev, __be32 multiaddr)
1090 {
1091 	struct ip_mc_list *pmc, *pmc_prev;
1092 	struct ip_sf_list *psf, *psf_next;
1093 
1094 	spin_lock_bh(&in_dev->mc_tomb_lock);
1095 	pmc_prev = NULL;
1096 	for (pmc=in_dev->mc_tomb; pmc; pmc=pmc->next) {
1097 		if (pmc->multiaddr == multiaddr)
1098 			break;
1099 		pmc_prev = pmc;
1100 	}
1101 	if (pmc) {
1102 		if (pmc_prev)
1103 			pmc_prev->next = pmc->next;
1104 		else
1105 			in_dev->mc_tomb = pmc->next;
1106 	}
1107 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1108 	if (pmc) {
1109 		for (psf=pmc->tomb; psf; psf=psf_next) {
1110 			psf_next = psf->sf_next;
1111 			kfree(psf);
1112 		}
1113 		in_dev_put(pmc->interface);
1114 		kfree(pmc);
1115 	}
1116 }
1117 
1118 static void igmpv3_clear_delrec(struct in_device *in_dev)
1119 {
1120 	struct ip_mc_list *pmc, *nextpmc;
1121 
1122 	spin_lock_bh(&in_dev->mc_tomb_lock);
1123 	pmc = in_dev->mc_tomb;
1124 	in_dev->mc_tomb = NULL;
1125 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1126 
1127 	for (; pmc; pmc = nextpmc) {
1128 		nextpmc = pmc->next;
1129 		ip_mc_clear_src(pmc);
1130 		in_dev_put(pmc->interface);
1131 		kfree(pmc);
1132 	}
1133 	/* clear dead sources, too */
1134 	rcu_read_lock();
1135 	for_each_pmc_rcu(in_dev, pmc) {
1136 		struct ip_sf_list *psf, *psf_next;
1137 
1138 		spin_lock_bh(&pmc->lock);
1139 		psf = pmc->tomb;
1140 		pmc->tomb = NULL;
1141 		spin_unlock_bh(&pmc->lock);
1142 		for (; psf; psf=psf_next) {
1143 			psf_next = psf->sf_next;
1144 			kfree(psf);
1145 		}
1146 	}
1147 	rcu_read_unlock();
1148 }
1149 #endif
1150 
1151 static void igmp_group_dropped(struct ip_mc_list *im)
1152 {
1153 	struct in_device *in_dev = im->interface;
1154 #ifdef CONFIG_IP_MULTICAST
1155 	int reporter;
1156 #endif
1157 
1158 	if (im->loaded) {
1159 		im->loaded = 0;
1160 		ip_mc_filter_del(in_dev, im->multiaddr);
1161 	}
1162 
1163 #ifdef CONFIG_IP_MULTICAST
1164 	if (im->multiaddr == IGMP_ALL_HOSTS)
1165 		return;
1166 
1167 	reporter = im->reporter;
1168 	igmp_stop_timer(im);
1169 
1170 	if (!in_dev->dead) {
1171 		if (IGMP_V1_SEEN(in_dev))
1172 			return;
1173 		if (IGMP_V2_SEEN(in_dev)) {
1174 			if (reporter)
1175 				igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1176 			return;
1177 		}
1178 		/* IGMPv3 */
1179 		igmpv3_add_delrec(in_dev, im);
1180 
1181 		igmp_ifc_event(in_dev);
1182 	}
1183 #endif
1184 }
1185 
1186 static void igmp_group_added(struct ip_mc_list *im)
1187 {
1188 	struct in_device *in_dev = im->interface;
1189 
1190 	if (im->loaded == 0) {
1191 		im->loaded = 1;
1192 		ip_mc_filter_add(in_dev, im->multiaddr);
1193 	}
1194 
1195 #ifdef CONFIG_IP_MULTICAST
1196 	if (im->multiaddr == IGMP_ALL_HOSTS)
1197 		return;
1198 
1199 	if (in_dev->dead)
1200 		return;
1201 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1202 		spin_lock_bh(&im->lock);
1203 		igmp_start_timer(im, IGMP_Initial_Report_Delay);
1204 		spin_unlock_bh(&im->lock);
1205 		return;
1206 	}
1207 	/* else, v3 */
1208 
1209 	im->crcount = in_dev->mr_qrv ? in_dev->mr_qrv :
1210 		IGMP_Unsolicited_Report_Count;
1211 	igmp_ifc_event(in_dev);
1212 #endif
1213 }
1214 
1215 
1216 /*
1217  *	Multicast list managers
1218  */
1219 
1220 
1221 /*
1222  *	A socket has joined a multicast group on device dev.
1223  */
1224 
1225 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1226 {
1227 	struct ip_mc_list *im;
1228 
1229 	ASSERT_RTNL();
1230 
1231 	for_each_pmc_rtnl(in_dev, im) {
1232 		if (im->multiaddr == addr) {
1233 			im->users++;
1234 			ip_mc_add_src(in_dev, &addr, MCAST_EXCLUDE, 0, NULL, 0);
1235 			goto out;
1236 		}
1237 	}
1238 
1239 	im = kzalloc(sizeof(*im), GFP_KERNEL);
1240 	if (!im)
1241 		goto out;
1242 
1243 	im->users = 1;
1244 	im->interface = in_dev;
1245 	in_dev_hold(in_dev);
1246 	im->multiaddr = addr;
1247 	/* initial mode is (EX, empty) */
1248 	im->sfmode = MCAST_EXCLUDE;
1249 	im->sfcount[MCAST_EXCLUDE] = 1;
1250 	atomic_set(&im->refcnt, 1);
1251 	spin_lock_init(&im->lock);
1252 #ifdef CONFIG_IP_MULTICAST
1253 	setup_timer(&im->timer, &igmp_timer_expire, (unsigned long)im);
1254 	im->unsolicit_count = IGMP_Unsolicited_Report_Count;
1255 #endif
1256 
1257 	im->next_rcu = in_dev->mc_list;
1258 	in_dev->mc_count++;
1259 	rcu_assign_pointer(in_dev->mc_list, im);
1260 
1261 #ifdef CONFIG_IP_MULTICAST
1262 	igmpv3_del_delrec(in_dev, im->multiaddr);
1263 #endif
1264 	igmp_group_added(im);
1265 	if (!in_dev->dead)
1266 		ip_rt_multicast_event(in_dev);
1267 out:
1268 	return;
1269 }
1270 EXPORT_SYMBOL(ip_mc_inc_group);
1271 
1272 /*
1273  *	Resend IGMP JOIN report; used for bonding.
1274  *	Called with rcu_read_lock()
1275  */
1276 void ip_mc_rejoin_groups(struct in_device *in_dev)
1277 {
1278 #ifdef CONFIG_IP_MULTICAST
1279 	struct ip_mc_list *im;
1280 	int type;
1281 
1282 	for_each_pmc_rcu(in_dev, im) {
1283 		if (im->multiaddr == IGMP_ALL_HOSTS)
1284 			continue;
1285 
1286 		/* a failover is happening and switches
1287 		 * must be notified immediately
1288 		 */
1289 		if (IGMP_V1_SEEN(in_dev))
1290 			type = IGMP_HOST_MEMBERSHIP_REPORT;
1291 		else if (IGMP_V2_SEEN(in_dev))
1292 			type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1293 		else
1294 			type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1295 		igmp_send_report(in_dev, im, type);
1296 	}
1297 #endif
1298 }
1299 EXPORT_SYMBOL(ip_mc_rejoin_groups);
1300 
1301 /*
1302  *	A socket has left a multicast group on device dev
1303  */
1304 
1305 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1306 {
1307 	struct ip_mc_list *i;
1308 	struct ip_mc_list __rcu **ip;
1309 
1310 	ASSERT_RTNL();
1311 
1312 	for (ip = &in_dev->mc_list;
1313 	     (i = rtnl_dereference(*ip)) != NULL;
1314 	     ip = &i->next_rcu) {
1315 		if (i->multiaddr == addr) {
1316 			if (--i->users == 0) {
1317 				*ip = i->next_rcu;
1318 				in_dev->mc_count--;
1319 				igmp_group_dropped(i);
1320 				ip_mc_clear_src(i);
1321 
1322 				if (!in_dev->dead)
1323 					ip_rt_multicast_event(in_dev);
1324 
1325 				ip_ma_put(i);
1326 				return;
1327 			}
1328 			break;
1329 		}
1330 	}
1331 }
1332 EXPORT_SYMBOL(ip_mc_dec_group);
1333 
1334 /* Device changing type */
1335 
1336 void ip_mc_unmap(struct in_device *in_dev)
1337 {
1338 	struct ip_mc_list *pmc;
1339 
1340 	ASSERT_RTNL();
1341 
1342 	for_each_pmc_rtnl(in_dev, pmc)
1343 		igmp_group_dropped(pmc);
1344 }
1345 
1346 void ip_mc_remap(struct in_device *in_dev)
1347 {
1348 	struct ip_mc_list *pmc;
1349 
1350 	ASSERT_RTNL();
1351 
1352 	for_each_pmc_rtnl(in_dev, pmc)
1353 		igmp_group_added(pmc);
1354 }
1355 
1356 /* Device going down */
1357 
1358 void ip_mc_down(struct in_device *in_dev)
1359 {
1360 	struct ip_mc_list *pmc;
1361 
1362 	ASSERT_RTNL();
1363 
1364 	for_each_pmc_rtnl(in_dev, pmc)
1365 		igmp_group_dropped(pmc);
1366 
1367 #ifdef CONFIG_IP_MULTICAST
1368 	in_dev->mr_ifc_count = 0;
1369 	if (del_timer(&in_dev->mr_ifc_timer))
1370 		__in_dev_put(in_dev);
1371 	in_dev->mr_gq_running = 0;
1372 	if (del_timer(&in_dev->mr_gq_timer))
1373 		__in_dev_put(in_dev);
1374 	igmpv3_clear_delrec(in_dev);
1375 #endif
1376 
1377 	ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1378 }
1379 
1380 void ip_mc_init_dev(struct in_device *in_dev)
1381 {
1382 	ASSERT_RTNL();
1383 
1384 	in_dev->mc_tomb = NULL;
1385 #ifdef CONFIG_IP_MULTICAST
1386 	in_dev->mr_gq_running = 0;
1387 	setup_timer(&in_dev->mr_gq_timer, igmp_gq_timer_expire,
1388 			(unsigned long)in_dev);
1389 	in_dev->mr_ifc_count = 0;
1390 	in_dev->mc_count     = 0;
1391 	setup_timer(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire,
1392 			(unsigned long)in_dev);
1393 	in_dev->mr_qrv = IGMP_Unsolicited_Report_Count;
1394 #endif
1395 
1396 	spin_lock_init(&in_dev->mc_tomb_lock);
1397 }
1398 
1399 /* Device going up */
1400 
1401 void ip_mc_up(struct in_device *in_dev)
1402 {
1403 	struct ip_mc_list *pmc;
1404 
1405 	ASSERT_RTNL();
1406 
1407 	ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1408 
1409 	for_each_pmc_rtnl(in_dev, pmc)
1410 		igmp_group_added(pmc);
1411 }
1412 
1413 /*
1414  *	Device is about to be destroyed: clean up.
1415  */
1416 
1417 void ip_mc_destroy_dev(struct in_device *in_dev)
1418 {
1419 	struct ip_mc_list *i;
1420 
1421 	ASSERT_RTNL();
1422 
1423 	/* Deactivate timers */
1424 	ip_mc_down(in_dev);
1425 
1426 	while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1427 		in_dev->mc_list = i->next_rcu;
1428 		in_dev->mc_count--;
1429 
1430 		/* We've dropped the groups in ip_mc_down already */
1431 		ip_mc_clear_src(i);
1432 		ip_ma_put(i);
1433 	}
1434 }
1435 
1436 /* RTNL is locked */
1437 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1438 {
1439 	struct net_device *dev = NULL;
1440 	struct in_device *idev = NULL;
1441 
1442 	if (imr->imr_ifindex) {
1443 		idev = inetdev_by_index(net, imr->imr_ifindex);
1444 		return idev;
1445 	}
1446 	if (imr->imr_address.s_addr) {
1447 		dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1448 		if (!dev)
1449 			return NULL;
1450 	}
1451 
1452 	if (!dev) {
1453 		struct rtable *rt = ip_route_output(net,
1454 						    imr->imr_multiaddr.s_addr,
1455 						    0, 0, 0);
1456 		if (!IS_ERR(rt)) {
1457 			dev = rt->dst.dev;
1458 			ip_rt_put(rt);
1459 		}
1460 	}
1461 	if (dev) {
1462 		imr->imr_ifindex = dev->ifindex;
1463 		idev = __in_dev_get_rtnl(dev);
1464 	}
1465 	return idev;
1466 }
1467 
1468 /*
1469  *	Join a socket to a group
1470  */
1471 int sysctl_igmp_max_memberships __read_mostly = IP_MAX_MEMBERSHIPS;
1472 int sysctl_igmp_max_msf __read_mostly = IP_MAX_MSF;
1473 
1474 
1475 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1476 	__be32 *psfsrc)
1477 {
1478 	struct ip_sf_list *psf, *psf_prev;
1479 	int rv = 0;
1480 
1481 	psf_prev = NULL;
1482 	for (psf=pmc->sources; psf; psf=psf->sf_next) {
1483 		if (psf->sf_inaddr == *psfsrc)
1484 			break;
1485 		psf_prev = psf;
1486 	}
1487 	if (!psf || psf->sf_count[sfmode] == 0) {
1488 		/* source filter not found, or count wrong =>  bug */
1489 		return -ESRCH;
1490 	}
1491 	psf->sf_count[sfmode]--;
1492 	if (psf->sf_count[sfmode] == 0) {
1493 		ip_rt_multicast_event(pmc->interface);
1494 	}
1495 	if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1496 #ifdef CONFIG_IP_MULTICAST
1497 		struct in_device *in_dev = pmc->interface;
1498 #endif
1499 
1500 		/* no more filters for this source */
1501 		if (psf_prev)
1502 			psf_prev->sf_next = psf->sf_next;
1503 		else
1504 			pmc->sources = psf->sf_next;
1505 #ifdef CONFIG_IP_MULTICAST
1506 		if (psf->sf_oldin &&
1507 		    !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1508 			psf->sf_crcount = in_dev->mr_qrv ? in_dev->mr_qrv :
1509 				IGMP_Unsolicited_Report_Count;
1510 			psf->sf_next = pmc->tomb;
1511 			pmc->tomb = psf;
1512 			rv = 1;
1513 		} else
1514 #endif
1515 			kfree(psf);
1516 	}
1517 	return rv;
1518 }
1519 
1520 #ifndef CONFIG_IP_MULTICAST
1521 #define igmp_ifc_event(x)	do { } while (0)
1522 #endif
1523 
1524 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1525 			 int sfcount, __be32 *psfsrc, int delta)
1526 {
1527 	struct ip_mc_list *pmc;
1528 	int	changerec = 0;
1529 	int	i, err;
1530 
1531 	if (!in_dev)
1532 		return -ENODEV;
1533 	rcu_read_lock();
1534 	for_each_pmc_rcu(in_dev, pmc) {
1535 		if (*pmca == pmc->multiaddr)
1536 			break;
1537 	}
1538 	if (!pmc) {
1539 		/* MCA not found?? bug */
1540 		rcu_read_unlock();
1541 		return -ESRCH;
1542 	}
1543 	spin_lock_bh(&pmc->lock);
1544 	rcu_read_unlock();
1545 #ifdef CONFIG_IP_MULTICAST
1546 	sf_markstate(pmc);
1547 #endif
1548 	if (!delta) {
1549 		err = -EINVAL;
1550 		if (!pmc->sfcount[sfmode])
1551 			goto out_unlock;
1552 		pmc->sfcount[sfmode]--;
1553 	}
1554 	err = 0;
1555 	for (i=0; i<sfcount; i++) {
1556 		int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1557 
1558 		changerec |= rv > 0;
1559 		if (!err && rv < 0)
1560 			err = rv;
1561 	}
1562 	if (pmc->sfmode == MCAST_EXCLUDE &&
1563 	    pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1564 	    pmc->sfcount[MCAST_INCLUDE]) {
1565 #ifdef CONFIG_IP_MULTICAST
1566 		struct ip_sf_list *psf;
1567 #endif
1568 
1569 		/* filter mode change */
1570 		pmc->sfmode = MCAST_INCLUDE;
1571 #ifdef CONFIG_IP_MULTICAST
1572 		pmc->crcount = in_dev->mr_qrv ? in_dev->mr_qrv :
1573 			IGMP_Unsolicited_Report_Count;
1574 		in_dev->mr_ifc_count = pmc->crcount;
1575 		for (psf=pmc->sources; psf; psf = psf->sf_next)
1576 			psf->sf_crcount = 0;
1577 		igmp_ifc_event(pmc->interface);
1578 	} else if (sf_setstate(pmc) || changerec) {
1579 		igmp_ifc_event(pmc->interface);
1580 #endif
1581 	}
1582 out_unlock:
1583 	spin_unlock_bh(&pmc->lock);
1584 	return err;
1585 }
1586 
1587 /*
1588  * Add multicast single-source filter to the interface list
1589  */
1590 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1591 	__be32 *psfsrc)
1592 {
1593 	struct ip_sf_list *psf, *psf_prev;
1594 
1595 	psf_prev = NULL;
1596 	for (psf=pmc->sources; psf; psf=psf->sf_next) {
1597 		if (psf->sf_inaddr == *psfsrc)
1598 			break;
1599 		psf_prev = psf;
1600 	}
1601 	if (!psf) {
1602 		psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1603 		if (!psf)
1604 			return -ENOBUFS;
1605 		psf->sf_inaddr = *psfsrc;
1606 		if (psf_prev) {
1607 			psf_prev->sf_next = psf;
1608 		} else
1609 			pmc->sources = psf;
1610 	}
1611 	psf->sf_count[sfmode]++;
1612 	if (psf->sf_count[sfmode] == 1) {
1613 		ip_rt_multicast_event(pmc->interface);
1614 	}
1615 	return 0;
1616 }
1617 
1618 #ifdef CONFIG_IP_MULTICAST
1619 static void sf_markstate(struct ip_mc_list *pmc)
1620 {
1621 	struct ip_sf_list *psf;
1622 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1623 
1624 	for (psf=pmc->sources; psf; psf=psf->sf_next)
1625 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1626 			psf->sf_oldin = mca_xcount ==
1627 				psf->sf_count[MCAST_EXCLUDE] &&
1628 				!psf->sf_count[MCAST_INCLUDE];
1629 		} else
1630 			psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1631 }
1632 
1633 static int sf_setstate(struct ip_mc_list *pmc)
1634 {
1635 	struct ip_sf_list *psf, *dpsf;
1636 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1637 	int qrv = pmc->interface->mr_qrv;
1638 	int new_in, rv;
1639 
1640 	rv = 0;
1641 	for (psf=pmc->sources; psf; psf=psf->sf_next) {
1642 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1643 			new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1644 				!psf->sf_count[MCAST_INCLUDE];
1645 		} else
1646 			new_in = psf->sf_count[MCAST_INCLUDE] != 0;
1647 		if (new_in) {
1648 			if (!psf->sf_oldin) {
1649 				struct ip_sf_list *prev = NULL;
1650 
1651 				for (dpsf=pmc->tomb; dpsf; dpsf=dpsf->sf_next) {
1652 					if (dpsf->sf_inaddr == psf->sf_inaddr)
1653 						break;
1654 					prev = dpsf;
1655 				}
1656 				if (dpsf) {
1657 					if (prev)
1658 						prev->sf_next = dpsf->sf_next;
1659 					else
1660 						pmc->tomb = dpsf->sf_next;
1661 					kfree(dpsf);
1662 				}
1663 				psf->sf_crcount = qrv;
1664 				rv++;
1665 			}
1666 		} else if (psf->sf_oldin) {
1667 
1668 			psf->sf_crcount = 0;
1669 			/*
1670 			 * add or update "delete" records if an active filter
1671 			 * is now inactive
1672 			 */
1673 			for (dpsf=pmc->tomb; dpsf; dpsf=dpsf->sf_next)
1674 				if (dpsf->sf_inaddr == psf->sf_inaddr)
1675 					break;
1676 			if (!dpsf) {
1677 				dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
1678 				if (!dpsf)
1679 					continue;
1680 				*dpsf = *psf;
1681 				/* pmc->lock held by callers */
1682 				dpsf->sf_next = pmc->tomb;
1683 				pmc->tomb = dpsf;
1684 			}
1685 			dpsf->sf_crcount = qrv;
1686 			rv++;
1687 		}
1688 	}
1689 	return rv;
1690 }
1691 #endif
1692 
1693 /*
1694  * Add multicast source filter list to the interface list
1695  */
1696 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1697 			 int sfcount, __be32 *psfsrc, int delta)
1698 {
1699 	struct ip_mc_list *pmc;
1700 	int	isexclude;
1701 	int	i, err;
1702 
1703 	if (!in_dev)
1704 		return -ENODEV;
1705 	rcu_read_lock();
1706 	for_each_pmc_rcu(in_dev, pmc) {
1707 		if (*pmca == pmc->multiaddr)
1708 			break;
1709 	}
1710 	if (!pmc) {
1711 		/* MCA not found?? bug */
1712 		rcu_read_unlock();
1713 		return -ESRCH;
1714 	}
1715 	spin_lock_bh(&pmc->lock);
1716 	rcu_read_unlock();
1717 
1718 #ifdef CONFIG_IP_MULTICAST
1719 	sf_markstate(pmc);
1720 #endif
1721 	isexclude = pmc->sfmode == MCAST_EXCLUDE;
1722 	if (!delta)
1723 		pmc->sfcount[sfmode]++;
1724 	err = 0;
1725 	for (i=0; i<sfcount; i++) {
1726 		err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
1727 		if (err)
1728 			break;
1729 	}
1730 	if (err) {
1731 		int j;
1732 
1733 		if (!delta)
1734 			pmc->sfcount[sfmode]--;
1735 		for (j=0; j<i; j++)
1736 			(void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
1737 	} else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
1738 #ifdef CONFIG_IP_MULTICAST
1739 		struct ip_sf_list *psf;
1740 		in_dev = pmc->interface;
1741 #endif
1742 
1743 		/* filter mode change */
1744 		if (pmc->sfcount[MCAST_EXCLUDE])
1745 			pmc->sfmode = MCAST_EXCLUDE;
1746 		else if (pmc->sfcount[MCAST_INCLUDE])
1747 			pmc->sfmode = MCAST_INCLUDE;
1748 #ifdef CONFIG_IP_MULTICAST
1749 		/* else no filters; keep old mode for reports */
1750 
1751 		pmc->crcount = in_dev->mr_qrv ? in_dev->mr_qrv :
1752 			IGMP_Unsolicited_Report_Count;
1753 		in_dev->mr_ifc_count = pmc->crcount;
1754 		for (psf=pmc->sources; psf; psf = psf->sf_next)
1755 			psf->sf_crcount = 0;
1756 		igmp_ifc_event(in_dev);
1757 	} else if (sf_setstate(pmc)) {
1758 		igmp_ifc_event(in_dev);
1759 #endif
1760 	}
1761 	spin_unlock_bh(&pmc->lock);
1762 	return err;
1763 }
1764 
1765 static void ip_mc_clear_src(struct ip_mc_list *pmc)
1766 {
1767 	struct ip_sf_list *psf, *nextpsf;
1768 
1769 	for (psf=pmc->tomb; psf; psf=nextpsf) {
1770 		nextpsf = psf->sf_next;
1771 		kfree(psf);
1772 	}
1773 	pmc->tomb = NULL;
1774 	for (psf=pmc->sources; psf; psf=nextpsf) {
1775 		nextpsf = psf->sf_next;
1776 		kfree(psf);
1777 	}
1778 	pmc->sources = NULL;
1779 	pmc->sfmode = MCAST_EXCLUDE;
1780 	pmc->sfcount[MCAST_INCLUDE] = 0;
1781 	pmc->sfcount[MCAST_EXCLUDE] = 1;
1782 }
1783 
1784 
1785 /*
1786  * Join a multicast group
1787  */
1788 int ip_mc_join_group(struct sock *sk , struct ip_mreqn *imr)
1789 {
1790 	int err;
1791 	__be32 addr = imr->imr_multiaddr.s_addr;
1792 	struct ip_mc_socklist *iml = NULL, *i;
1793 	struct in_device *in_dev;
1794 	struct inet_sock *inet = inet_sk(sk);
1795 	struct net *net = sock_net(sk);
1796 	int ifindex;
1797 	int count = 0;
1798 
1799 	if (!ipv4_is_multicast(addr))
1800 		return -EINVAL;
1801 
1802 	rtnl_lock();
1803 
1804 	in_dev = ip_mc_find_dev(net, imr);
1805 
1806 	if (!in_dev) {
1807 		iml = NULL;
1808 		err = -ENODEV;
1809 		goto done;
1810 	}
1811 
1812 	err = -EADDRINUSE;
1813 	ifindex = imr->imr_ifindex;
1814 	for_each_pmc_rtnl(inet, i) {
1815 		if (i->multi.imr_multiaddr.s_addr == addr &&
1816 		    i->multi.imr_ifindex == ifindex)
1817 			goto done;
1818 		count++;
1819 	}
1820 	err = -ENOBUFS;
1821 	if (count >= sysctl_igmp_max_memberships)
1822 		goto done;
1823 	iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
1824 	if (iml == NULL)
1825 		goto done;
1826 
1827 	memcpy(&iml->multi, imr, sizeof(*imr));
1828 	iml->next_rcu = inet->mc_list;
1829 	iml->sflist = NULL;
1830 	iml->sfmode = MCAST_EXCLUDE;
1831 	rcu_assign_pointer(inet->mc_list, iml);
1832 	ip_mc_inc_group(in_dev, addr);
1833 	err = 0;
1834 done:
1835 	rtnl_unlock();
1836 	return err;
1837 }
1838 EXPORT_SYMBOL(ip_mc_join_group);
1839 
1840 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
1841 			   struct in_device *in_dev)
1842 {
1843 	struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
1844 	int err;
1845 
1846 	if (psf == NULL) {
1847 		/* any-source empty exclude case */
1848 		return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
1849 			iml->sfmode, 0, NULL, 0);
1850 	}
1851 	err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
1852 			iml->sfmode, psf->sl_count, psf->sl_addr, 0);
1853 	RCU_INIT_POINTER(iml->sflist, NULL);
1854 	/* decrease mem now to avoid the memleak warning */
1855 	atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
1856 	kfree_rcu(psf, rcu);
1857 	return err;
1858 }
1859 
1860 /*
1861  *	Ask a socket to leave a group.
1862  */
1863 
1864 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
1865 {
1866 	struct inet_sock *inet = inet_sk(sk);
1867 	struct ip_mc_socklist *iml;
1868 	struct ip_mc_socklist __rcu **imlp;
1869 	struct in_device *in_dev;
1870 	struct net *net = sock_net(sk);
1871 	__be32 group = imr->imr_multiaddr.s_addr;
1872 	u32 ifindex;
1873 	int ret = -EADDRNOTAVAIL;
1874 
1875 	rtnl_lock();
1876 	in_dev = ip_mc_find_dev(net, imr);
1877 	ifindex = imr->imr_ifindex;
1878 	for (imlp = &inet->mc_list;
1879 	     (iml = rtnl_dereference(*imlp)) != NULL;
1880 	     imlp = &iml->next_rcu) {
1881 		if (iml->multi.imr_multiaddr.s_addr != group)
1882 			continue;
1883 		if (ifindex) {
1884 			if (iml->multi.imr_ifindex != ifindex)
1885 				continue;
1886 		} else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
1887 				iml->multi.imr_address.s_addr)
1888 			continue;
1889 
1890 		(void) ip_mc_leave_src(sk, iml, in_dev);
1891 
1892 		*imlp = iml->next_rcu;
1893 
1894 		if (in_dev)
1895 			ip_mc_dec_group(in_dev, group);
1896 		rtnl_unlock();
1897 		/* decrease mem now to avoid the memleak warning */
1898 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
1899 		kfree_rcu(iml, rcu);
1900 		return 0;
1901 	}
1902 	if (!in_dev)
1903 		ret = -ENODEV;
1904 	rtnl_unlock();
1905 	return ret;
1906 }
1907 EXPORT_SYMBOL(ip_mc_leave_group);
1908 
1909 int ip_mc_source(int add, int omode, struct sock *sk, struct
1910 	ip_mreq_source *mreqs, int ifindex)
1911 {
1912 	int err;
1913 	struct ip_mreqn imr;
1914 	__be32 addr = mreqs->imr_multiaddr;
1915 	struct ip_mc_socklist *pmc;
1916 	struct in_device *in_dev = NULL;
1917 	struct inet_sock *inet = inet_sk(sk);
1918 	struct ip_sf_socklist *psl;
1919 	struct net *net = sock_net(sk);
1920 	int leavegroup = 0;
1921 	int i, j, rv;
1922 
1923 	if (!ipv4_is_multicast(addr))
1924 		return -EINVAL;
1925 
1926 	rtnl_lock();
1927 
1928 	imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
1929 	imr.imr_address.s_addr = mreqs->imr_interface;
1930 	imr.imr_ifindex = ifindex;
1931 	in_dev = ip_mc_find_dev(net, &imr);
1932 
1933 	if (!in_dev) {
1934 		err = -ENODEV;
1935 		goto done;
1936 	}
1937 	err = -EADDRNOTAVAIL;
1938 
1939 	for_each_pmc_rtnl(inet, pmc) {
1940 		if ((pmc->multi.imr_multiaddr.s_addr ==
1941 		     imr.imr_multiaddr.s_addr) &&
1942 		    (pmc->multi.imr_ifindex == imr.imr_ifindex))
1943 			break;
1944 	}
1945 	if (!pmc) {		/* must have a prior join */
1946 		err = -EINVAL;
1947 		goto done;
1948 	}
1949 	/* if a source filter was set, must be the same mode as before */
1950 	if (pmc->sflist) {
1951 		if (pmc->sfmode != omode) {
1952 			err = -EINVAL;
1953 			goto done;
1954 		}
1955 	} else if (pmc->sfmode != omode) {
1956 		/* allow mode switches for empty-set filters */
1957 		ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
1958 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
1959 			NULL, 0);
1960 		pmc->sfmode = omode;
1961 	}
1962 
1963 	psl = rtnl_dereference(pmc->sflist);
1964 	if (!add) {
1965 		if (!psl)
1966 			goto done;	/* err = -EADDRNOTAVAIL */
1967 		rv = !0;
1968 		for (i=0; i<psl->sl_count; i++) {
1969 			rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
1970 				sizeof(__be32));
1971 			if (rv == 0)
1972 				break;
1973 		}
1974 		if (rv)		/* source not found */
1975 			goto done;	/* err = -EADDRNOTAVAIL */
1976 
1977 		/* special case - (INCLUDE, empty) == LEAVE_GROUP */
1978 		if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
1979 			leavegroup = 1;
1980 			goto done;
1981 		}
1982 
1983 		/* update the interface filter */
1984 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
1985 			&mreqs->imr_sourceaddr, 1);
1986 
1987 		for (j=i+1; j<psl->sl_count; j++)
1988 			psl->sl_addr[j-1] = psl->sl_addr[j];
1989 		psl->sl_count--;
1990 		err = 0;
1991 		goto done;
1992 	}
1993 	/* else, add a new source to the filter */
1994 
1995 	if (psl && psl->sl_count >= sysctl_igmp_max_msf) {
1996 		err = -ENOBUFS;
1997 		goto done;
1998 	}
1999 	if (!psl || psl->sl_count == psl->sl_max) {
2000 		struct ip_sf_socklist *newpsl;
2001 		int count = IP_SFBLOCK;
2002 
2003 		if (psl)
2004 			count += psl->sl_max;
2005 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2006 		if (!newpsl) {
2007 			err = -ENOBUFS;
2008 			goto done;
2009 		}
2010 		newpsl->sl_max = count;
2011 		newpsl->sl_count = count - IP_SFBLOCK;
2012 		if (psl) {
2013 			for (i=0; i<psl->sl_count; i++)
2014 				newpsl->sl_addr[i] = psl->sl_addr[i];
2015 			/* decrease mem now to avoid the memleak warning */
2016 			atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2017 			kfree_rcu(psl, rcu);
2018 		}
2019 		rcu_assign_pointer(pmc->sflist, newpsl);
2020 		psl = newpsl;
2021 	}
2022 	rv = 1;	/* > 0 for insert logic below if sl_count is 0 */
2023 	for (i=0; i<psl->sl_count; i++) {
2024 		rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2025 			sizeof(__be32));
2026 		if (rv == 0)
2027 			break;
2028 	}
2029 	if (rv == 0)		/* address already there is an error */
2030 		goto done;
2031 	for (j=psl->sl_count-1; j>=i; j--)
2032 		psl->sl_addr[j+1] = psl->sl_addr[j];
2033 	psl->sl_addr[i] = mreqs->imr_sourceaddr;
2034 	psl->sl_count++;
2035 	err = 0;
2036 	/* update the interface list */
2037 	ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2038 		&mreqs->imr_sourceaddr, 1);
2039 done:
2040 	rtnl_unlock();
2041 	if (leavegroup)
2042 		return ip_mc_leave_group(sk, &imr);
2043 	return err;
2044 }
2045 
2046 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2047 {
2048 	int err = 0;
2049 	struct ip_mreqn	imr;
2050 	__be32 addr = msf->imsf_multiaddr;
2051 	struct ip_mc_socklist *pmc;
2052 	struct in_device *in_dev;
2053 	struct inet_sock *inet = inet_sk(sk);
2054 	struct ip_sf_socklist *newpsl, *psl;
2055 	struct net *net = sock_net(sk);
2056 	int leavegroup = 0;
2057 
2058 	if (!ipv4_is_multicast(addr))
2059 		return -EINVAL;
2060 	if (msf->imsf_fmode != MCAST_INCLUDE &&
2061 	    msf->imsf_fmode != MCAST_EXCLUDE)
2062 		return -EINVAL;
2063 
2064 	rtnl_lock();
2065 
2066 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2067 	imr.imr_address.s_addr = msf->imsf_interface;
2068 	imr.imr_ifindex = ifindex;
2069 	in_dev = ip_mc_find_dev(net, &imr);
2070 
2071 	if (!in_dev) {
2072 		err = -ENODEV;
2073 		goto done;
2074 	}
2075 
2076 	/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2077 	if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2078 		leavegroup = 1;
2079 		goto done;
2080 	}
2081 
2082 	for_each_pmc_rtnl(inet, pmc) {
2083 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2084 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2085 			break;
2086 	}
2087 	if (!pmc) {		/* must have a prior join */
2088 		err = -EINVAL;
2089 		goto done;
2090 	}
2091 	if (msf->imsf_numsrc) {
2092 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2093 							   GFP_KERNEL);
2094 		if (!newpsl) {
2095 			err = -ENOBUFS;
2096 			goto done;
2097 		}
2098 		newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2099 		memcpy(newpsl->sl_addr, msf->imsf_slist,
2100 			msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2101 		err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2102 			msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2103 		if (err) {
2104 			sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2105 			goto done;
2106 		}
2107 	} else {
2108 		newpsl = NULL;
2109 		(void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2110 				     msf->imsf_fmode, 0, NULL, 0);
2111 	}
2112 	psl = rtnl_dereference(pmc->sflist);
2113 	if (psl) {
2114 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2115 			psl->sl_count, psl->sl_addr, 0);
2116 		/* decrease mem now to avoid the memleak warning */
2117 		atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2118 		kfree_rcu(psl, rcu);
2119 	} else
2120 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2121 			0, NULL, 0);
2122 	rcu_assign_pointer(pmc->sflist, newpsl);
2123 	pmc->sfmode = msf->imsf_fmode;
2124 	err = 0;
2125 done:
2126 	rtnl_unlock();
2127 	if (leavegroup)
2128 		err = ip_mc_leave_group(sk, &imr);
2129 	return err;
2130 }
2131 
2132 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2133 	struct ip_msfilter __user *optval, int __user *optlen)
2134 {
2135 	int err, len, count, copycount;
2136 	struct ip_mreqn	imr;
2137 	__be32 addr = msf->imsf_multiaddr;
2138 	struct ip_mc_socklist *pmc;
2139 	struct in_device *in_dev;
2140 	struct inet_sock *inet = inet_sk(sk);
2141 	struct ip_sf_socklist *psl;
2142 	struct net *net = sock_net(sk);
2143 
2144 	if (!ipv4_is_multicast(addr))
2145 		return -EINVAL;
2146 
2147 	rtnl_lock();
2148 
2149 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2150 	imr.imr_address.s_addr = msf->imsf_interface;
2151 	imr.imr_ifindex = 0;
2152 	in_dev = ip_mc_find_dev(net, &imr);
2153 
2154 	if (!in_dev) {
2155 		err = -ENODEV;
2156 		goto done;
2157 	}
2158 	err = -EADDRNOTAVAIL;
2159 
2160 	for_each_pmc_rtnl(inet, pmc) {
2161 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2162 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2163 			break;
2164 	}
2165 	if (!pmc)		/* must have a prior join */
2166 		goto done;
2167 	msf->imsf_fmode = pmc->sfmode;
2168 	psl = rtnl_dereference(pmc->sflist);
2169 	rtnl_unlock();
2170 	if (!psl) {
2171 		len = 0;
2172 		count = 0;
2173 	} else {
2174 		count = psl->sl_count;
2175 	}
2176 	copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2177 	len = copycount * sizeof(psl->sl_addr[0]);
2178 	msf->imsf_numsrc = count;
2179 	if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2180 	    copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2181 		return -EFAULT;
2182 	}
2183 	if (len &&
2184 	    copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2185 		return -EFAULT;
2186 	return 0;
2187 done:
2188 	rtnl_unlock();
2189 	return err;
2190 }
2191 
2192 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2193 	struct group_filter __user *optval, int __user *optlen)
2194 {
2195 	int err, i, count, copycount;
2196 	struct sockaddr_in *psin;
2197 	__be32 addr;
2198 	struct ip_mc_socklist *pmc;
2199 	struct inet_sock *inet = inet_sk(sk);
2200 	struct ip_sf_socklist *psl;
2201 
2202 	psin = (struct sockaddr_in *)&gsf->gf_group;
2203 	if (psin->sin_family != AF_INET)
2204 		return -EINVAL;
2205 	addr = psin->sin_addr.s_addr;
2206 	if (!ipv4_is_multicast(addr))
2207 		return -EINVAL;
2208 
2209 	rtnl_lock();
2210 
2211 	err = -EADDRNOTAVAIL;
2212 
2213 	for_each_pmc_rtnl(inet, pmc) {
2214 		if (pmc->multi.imr_multiaddr.s_addr == addr &&
2215 		    pmc->multi.imr_ifindex == gsf->gf_interface)
2216 			break;
2217 	}
2218 	if (!pmc)		/* must have a prior join */
2219 		goto done;
2220 	gsf->gf_fmode = pmc->sfmode;
2221 	psl = rtnl_dereference(pmc->sflist);
2222 	rtnl_unlock();
2223 	count = psl ? psl->sl_count : 0;
2224 	copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2225 	gsf->gf_numsrc = count;
2226 	if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2227 	    copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2228 		return -EFAULT;
2229 	}
2230 	for (i=0; i<copycount; i++) {
2231 		struct sockaddr_storage ss;
2232 
2233 		psin = (struct sockaddr_in *)&ss;
2234 		memset(&ss, 0, sizeof(ss));
2235 		psin->sin_family = AF_INET;
2236 		psin->sin_addr.s_addr = psl->sl_addr[i];
2237 		if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2238 			return -EFAULT;
2239 	}
2240 	return 0;
2241 done:
2242 	rtnl_unlock();
2243 	return err;
2244 }
2245 
2246 /*
2247  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2248  */
2249 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr, int dif)
2250 {
2251 	struct inet_sock *inet = inet_sk(sk);
2252 	struct ip_mc_socklist *pmc;
2253 	struct ip_sf_socklist *psl;
2254 	int i;
2255 	int ret;
2256 
2257 	ret = 1;
2258 	if (!ipv4_is_multicast(loc_addr))
2259 		goto out;
2260 
2261 	rcu_read_lock();
2262 	for_each_pmc_rcu(inet, pmc) {
2263 		if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2264 		    pmc->multi.imr_ifindex == dif)
2265 			break;
2266 	}
2267 	ret = inet->mc_all;
2268 	if (!pmc)
2269 		goto unlock;
2270 	psl = rcu_dereference(pmc->sflist);
2271 	ret = (pmc->sfmode == MCAST_EXCLUDE);
2272 	if (!psl)
2273 		goto unlock;
2274 
2275 	for (i=0; i<psl->sl_count; i++) {
2276 		if (psl->sl_addr[i] == rmt_addr)
2277 			break;
2278 	}
2279 	ret = 0;
2280 	if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2281 		goto unlock;
2282 	if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2283 		goto unlock;
2284 	ret = 1;
2285 unlock:
2286 	rcu_read_unlock();
2287 out:
2288 	return ret;
2289 }
2290 
2291 /*
2292  *	A socket is closing.
2293  */
2294 
2295 void ip_mc_drop_socket(struct sock *sk)
2296 {
2297 	struct inet_sock *inet = inet_sk(sk);
2298 	struct ip_mc_socklist *iml;
2299 	struct net *net = sock_net(sk);
2300 
2301 	if (inet->mc_list == NULL)
2302 		return;
2303 
2304 	rtnl_lock();
2305 	while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2306 		struct in_device *in_dev;
2307 
2308 		inet->mc_list = iml->next_rcu;
2309 		in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2310 		(void) ip_mc_leave_src(sk, iml, in_dev);
2311 		if (in_dev != NULL)
2312 			ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2313 		/* decrease mem now to avoid the memleak warning */
2314 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2315 		kfree_rcu(iml, rcu);
2316 	}
2317 	rtnl_unlock();
2318 }
2319 
2320 /* called with rcu_read_lock() */
2321 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u16 proto)
2322 {
2323 	struct ip_mc_list *im;
2324 	struct ip_sf_list *psf;
2325 	int rv = 0;
2326 
2327 	for_each_pmc_rcu(in_dev, im) {
2328 		if (im->multiaddr == mc_addr)
2329 			break;
2330 	}
2331 	if (im && proto == IPPROTO_IGMP) {
2332 		rv = 1;
2333 	} else if (im) {
2334 		if (src_addr) {
2335 			for (psf=im->sources; psf; psf=psf->sf_next) {
2336 				if (psf->sf_inaddr == src_addr)
2337 					break;
2338 			}
2339 			if (psf)
2340 				rv = psf->sf_count[MCAST_INCLUDE] ||
2341 					psf->sf_count[MCAST_EXCLUDE] !=
2342 					im->sfcount[MCAST_EXCLUDE];
2343 			else
2344 				rv = im->sfcount[MCAST_EXCLUDE] != 0;
2345 		} else
2346 			rv = 1; /* unspecified source; tentatively allow */
2347 	}
2348 	return rv;
2349 }
2350 
2351 #if defined(CONFIG_PROC_FS)
2352 struct igmp_mc_iter_state {
2353 	struct seq_net_private p;
2354 	struct net_device *dev;
2355 	struct in_device *in_dev;
2356 };
2357 
2358 #define	igmp_mc_seq_private(seq)	((struct igmp_mc_iter_state *)(seq)->private)
2359 
2360 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2361 {
2362 	struct net *net = seq_file_net(seq);
2363 	struct ip_mc_list *im = NULL;
2364 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2365 
2366 	state->in_dev = NULL;
2367 	for_each_netdev_rcu(net, state->dev) {
2368 		struct in_device *in_dev;
2369 
2370 		in_dev = __in_dev_get_rcu(state->dev);
2371 		if (!in_dev)
2372 			continue;
2373 		im = rcu_dereference(in_dev->mc_list);
2374 		if (im) {
2375 			state->in_dev = in_dev;
2376 			break;
2377 		}
2378 	}
2379 	return im;
2380 }
2381 
2382 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2383 {
2384 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2385 
2386 	im = rcu_dereference(im->next_rcu);
2387 	while (!im) {
2388 		state->dev = next_net_device_rcu(state->dev);
2389 		if (!state->dev) {
2390 			state->in_dev = NULL;
2391 			break;
2392 		}
2393 		state->in_dev = __in_dev_get_rcu(state->dev);
2394 		if (!state->in_dev)
2395 			continue;
2396 		im = rcu_dereference(state->in_dev->mc_list);
2397 	}
2398 	return im;
2399 }
2400 
2401 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2402 {
2403 	struct ip_mc_list *im = igmp_mc_get_first(seq);
2404 	if (im)
2405 		while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2406 			--pos;
2407 	return pos ? NULL : im;
2408 }
2409 
2410 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2411 	__acquires(rcu)
2412 {
2413 	rcu_read_lock();
2414 	return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2415 }
2416 
2417 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2418 {
2419 	struct ip_mc_list *im;
2420 	if (v == SEQ_START_TOKEN)
2421 		im = igmp_mc_get_first(seq);
2422 	else
2423 		im = igmp_mc_get_next(seq, v);
2424 	++*pos;
2425 	return im;
2426 }
2427 
2428 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2429 	__releases(rcu)
2430 {
2431 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2432 
2433 	state->in_dev = NULL;
2434 	state->dev = NULL;
2435 	rcu_read_unlock();
2436 }
2437 
2438 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2439 {
2440 	if (v == SEQ_START_TOKEN)
2441 		seq_puts(seq,
2442 			 "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2443 	else {
2444 		struct ip_mc_list *im = (struct ip_mc_list *)v;
2445 		struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2446 		char   *querier;
2447 		long delta;
2448 
2449 #ifdef CONFIG_IP_MULTICAST
2450 		querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2451 			  IGMP_V2_SEEN(state->in_dev) ? "V2" :
2452 			  "V3";
2453 #else
2454 		querier = "NONE";
2455 #endif
2456 
2457 		if (rcu_dereference(state->in_dev->mc_list) == im) {
2458 			seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2459 				   state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2460 		}
2461 
2462 		delta = im->timer.expires - jiffies;
2463 		seq_printf(seq,
2464 			   "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2465 			   im->multiaddr, im->users,
2466 			   im->tm_running,
2467 			   im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2468 			   im->reporter);
2469 	}
2470 	return 0;
2471 }
2472 
2473 static const struct seq_operations igmp_mc_seq_ops = {
2474 	.start	=	igmp_mc_seq_start,
2475 	.next	=	igmp_mc_seq_next,
2476 	.stop	=	igmp_mc_seq_stop,
2477 	.show	=	igmp_mc_seq_show,
2478 };
2479 
2480 static int igmp_mc_seq_open(struct inode *inode, struct file *file)
2481 {
2482 	return seq_open_net(inode, file, &igmp_mc_seq_ops,
2483 			sizeof(struct igmp_mc_iter_state));
2484 }
2485 
2486 static const struct file_operations igmp_mc_seq_fops = {
2487 	.owner		=	THIS_MODULE,
2488 	.open		=	igmp_mc_seq_open,
2489 	.read		=	seq_read,
2490 	.llseek		=	seq_lseek,
2491 	.release	=	seq_release_net,
2492 };
2493 
2494 struct igmp_mcf_iter_state {
2495 	struct seq_net_private p;
2496 	struct net_device *dev;
2497 	struct in_device *idev;
2498 	struct ip_mc_list *im;
2499 };
2500 
2501 #define igmp_mcf_seq_private(seq)	((struct igmp_mcf_iter_state *)(seq)->private)
2502 
2503 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2504 {
2505 	struct net *net = seq_file_net(seq);
2506 	struct ip_sf_list *psf = NULL;
2507 	struct ip_mc_list *im = NULL;
2508 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2509 
2510 	state->idev = NULL;
2511 	state->im = NULL;
2512 	for_each_netdev_rcu(net, state->dev) {
2513 		struct in_device *idev;
2514 		idev = __in_dev_get_rcu(state->dev);
2515 		if (unlikely(idev == NULL))
2516 			continue;
2517 		im = rcu_dereference(idev->mc_list);
2518 		if (likely(im != NULL)) {
2519 			spin_lock_bh(&im->lock);
2520 			psf = im->sources;
2521 			if (likely(psf != NULL)) {
2522 				state->im = im;
2523 				state->idev = idev;
2524 				break;
2525 			}
2526 			spin_unlock_bh(&im->lock);
2527 		}
2528 	}
2529 	return psf;
2530 }
2531 
2532 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2533 {
2534 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2535 
2536 	psf = psf->sf_next;
2537 	while (!psf) {
2538 		spin_unlock_bh(&state->im->lock);
2539 		state->im = state->im->next;
2540 		while (!state->im) {
2541 			state->dev = next_net_device_rcu(state->dev);
2542 			if (!state->dev) {
2543 				state->idev = NULL;
2544 				goto out;
2545 			}
2546 			state->idev = __in_dev_get_rcu(state->dev);
2547 			if (!state->idev)
2548 				continue;
2549 			state->im = rcu_dereference(state->idev->mc_list);
2550 		}
2551 		if (!state->im)
2552 			break;
2553 		spin_lock_bh(&state->im->lock);
2554 		psf = state->im->sources;
2555 	}
2556 out:
2557 	return psf;
2558 }
2559 
2560 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2561 {
2562 	struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2563 	if (psf)
2564 		while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2565 			--pos;
2566 	return pos ? NULL : psf;
2567 }
2568 
2569 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2570 	__acquires(rcu)
2571 {
2572 	rcu_read_lock();
2573 	return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2574 }
2575 
2576 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2577 {
2578 	struct ip_sf_list *psf;
2579 	if (v == SEQ_START_TOKEN)
2580 		psf = igmp_mcf_get_first(seq);
2581 	else
2582 		psf = igmp_mcf_get_next(seq, v);
2583 	++*pos;
2584 	return psf;
2585 }
2586 
2587 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2588 	__releases(rcu)
2589 {
2590 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2591 	if (likely(state->im != NULL)) {
2592 		spin_unlock_bh(&state->im->lock);
2593 		state->im = NULL;
2594 	}
2595 	state->idev = NULL;
2596 	state->dev = NULL;
2597 	rcu_read_unlock();
2598 }
2599 
2600 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2601 {
2602 	struct ip_sf_list *psf = (struct ip_sf_list *)v;
2603 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2604 
2605 	if (v == SEQ_START_TOKEN) {
2606 		seq_printf(seq,
2607 			   "%3s %6s "
2608 			   "%10s %10s %6s %6s\n", "Idx",
2609 			   "Device", "MCA",
2610 			   "SRC", "INC", "EXC");
2611 	} else {
2612 		seq_printf(seq,
2613 			   "%3d %6.6s 0x%08x "
2614 			   "0x%08x %6lu %6lu\n",
2615 			   state->dev->ifindex, state->dev->name,
2616 			   ntohl(state->im->multiaddr),
2617 			   ntohl(psf->sf_inaddr),
2618 			   psf->sf_count[MCAST_INCLUDE],
2619 			   psf->sf_count[MCAST_EXCLUDE]);
2620 	}
2621 	return 0;
2622 }
2623 
2624 static const struct seq_operations igmp_mcf_seq_ops = {
2625 	.start	=	igmp_mcf_seq_start,
2626 	.next	=	igmp_mcf_seq_next,
2627 	.stop	=	igmp_mcf_seq_stop,
2628 	.show	=	igmp_mcf_seq_show,
2629 };
2630 
2631 static int igmp_mcf_seq_open(struct inode *inode, struct file *file)
2632 {
2633 	return seq_open_net(inode, file, &igmp_mcf_seq_ops,
2634 			sizeof(struct igmp_mcf_iter_state));
2635 }
2636 
2637 static const struct file_operations igmp_mcf_seq_fops = {
2638 	.owner		=	THIS_MODULE,
2639 	.open		=	igmp_mcf_seq_open,
2640 	.read		=	seq_read,
2641 	.llseek		=	seq_lseek,
2642 	.release	=	seq_release_net,
2643 };
2644 
2645 static int __net_init igmp_net_init(struct net *net)
2646 {
2647 	struct proc_dir_entry *pde;
2648 
2649 	pde = proc_net_fops_create(net, "igmp", S_IRUGO, &igmp_mc_seq_fops);
2650 	if (!pde)
2651 		goto out_igmp;
2652 	pde = proc_net_fops_create(net, "mcfilter", S_IRUGO, &igmp_mcf_seq_fops);
2653 	if (!pde)
2654 		goto out_mcfilter;
2655 	return 0;
2656 
2657 out_mcfilter:
2658 	proc_net_remove(net, "igmp");
2659 out_igmp:
2660 	return -ENOMEM;
2661 }
2662 
2663 static void __net_exit igmp_net_exit(struct net *net)
2664 {
2665 	proc_net_remove(net, "mcfilter");
2666 	proc_net_remove(net, "igmp");
2667 }
2668 
2669 static struct pernet_operations igmp_net_ops = {
2670 	.init = igmp_net_init,
2671 	.exit = igmp_net_exit,
2672 };
2673 
2674 int __init igmp_mc_proc_init(void)
2675 {
2676 	return register_pernet_subsys(&igmp_net_ops);
2677 }
2678 #endif
2679