xref: /linux/net/ipv4/igmp.c (revision 3f2fb9a834cb1fcddbae22deca7fde136944dc89)
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 #include <linux/pkt_sched.h>
92 
93 #include <net/net_namespace.h>
94 #include <net/arp.h>
95 #include <net/ip.h>
96 #include <net/protocol.h>
97 #include <net/route.h>
98 #include <net/sock.h>
99 #include <net/checksum.h>
100 #include <net/inet_common.h>
101 #include <linux/netfilter_ipv4.h>
102 #ifdef CONFIG_IP_MROUTE
103 #include <linux/mroute.h>
104 #endif
105 #ifdef CONFIG_PROC_FS
106 #include <linux/proc_fs.h>
107 #include <linux/seq_file.h>
108 #endif
109 
110 #ifdef CONFIG_IP_MULTICAST
111 /* Parameter names and values are taken from igmp-v2-06 draft */
112 
113 #define IGMP_V1_ROUTER_PRESENT_TIMEOUT		(400*HZ)
114 #define IGMP_V2_ROUTER_PRESENT_TIMEOUT		(400*HZ)
115 #define IGMP_V2_UNSOLICITED_REPORT_INTERVAL	(10*HZ)
116 #define IGMP_V3_UNSOLICITED_REPORT_INTERVAL	(1*HZ)
117 #define IGMP_QUERY_RESPONSE_INTERVAL		(10*HZ)
118 #define IGMP_QUERY_ROBUSTNESS_VARIABLE		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 int unsolicited_report_interval(struct in_device *in_dev)
142 {
143 	int interval_ms, interval_jiffies;
144 
145 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
146 		interval_ms = IN_DEV_CONF_GET(
147 			in_dev,
148 			IGMPV2_UNSOLICITED_REPORT_INTERVAL);
149 	else /* v3 */
150 		interval_ms = IN_DEV_CONF_GET(
151 			in_dev,
152 			IGMPV3_UNSOLICITED_REPORT_INTERVAL);
153 
154 	interval_jiffies = msecs_to_jiffies(interval_ms);
155 
156 	/* _timer functions can't handle a delay of 0 jiffies so ensure
157 	 *  we always return a positive value.
158 	 */
159 	if (interval_jiffies <= 0)
160 		interval_jiffies = 1;
161 	return interval_jiffies;
162 }
163 
164 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im);
165 static void igmpv3_del_delrec(struct in_device *in_dev, __be32 multiaddr);
166 static void igmpv3_clear_delrec(struct in_device *in_dev);
167 static int sf_setstate(struct ip_mc_list *pmc);
168 static void sf_markstate(struct ip_mc_list *pmc);
169 #endif
170 static void ip_mc_clear_src(struct ip_mc_list *pmc);
171 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
172 			 int sfcount, __be32 *psfsrc, int delta);
173 
174 static void ip_ma_put(struct ip_mc_list *im)
175 {
176 	if (atomic_dec_and_test(&im->refcnt)) {
177 		in_dev_put(im->interface);
178 		kfree_rcu(im, rcu);
179 	}
180 }
181 
182 #define for_each_pmc_rcu(in_dev, pmc)				\
183 	for (pmc = rcu_dereference(in_dev->mc_list);		\
184 	     pmc != NULL;					\
185 	     pmc = rcu_dereference(pmc->next_rcu))
186 
187 #define for_each_pmc_rtnl(in_dev, pmc)				\
188 	for (pmc = rtnl_dereference(in_dev->mc_list);		\
189 	     pmc != NULL;					\
190 	     pmc = rtnl_dereference(pmc->next_rcu))
191 
192 #ifdef CONFIG_IP_MULTICAST
193 
194 /*
195  *	Timer management
196  */
197 
198 static void igmp_stop_timer(struct ip_mc_list *im)
199 {
200 	spin_lock_bh(&im->lock);
201 	if (del_timer(&im->timer))
202 		atomic_dec(&im->refcnt);
203 	im->tm_running = 0;
204 	im->reporter = 0;
205 	im->unsolicit_count = 0;
206 	spin_unlock_bh(&im->lock);
207 }
208 
209 /* It must be called with locked im->lock */
210 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
211 {
212 	int tv = prandom_u32() % max_delay;
213 
214 	im->tm_running = 1;
215 	if (!mod_timer(&im->timer, jiffies+tv+2))
216 		atomic_inc(&im->refcnt);
217 }
218 
219 static void igmp_gq_start_timer(struct in_device *in_dev)
220 {
221 	int tv = prandom_u32() % in_dev->mr_maxdelay;
222 
223 	in_dev->mr_gq_running = 1;
224 	if (!mod_timer(&in_dev->mr_gq_timer, jiffies+tv+2))
225 		in_dev_hold(in_dev);
226 }
227 
228 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
229 {
230 	int tv = prandom_u32() % delay;
231 
232 	if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
233 		in_dev_hold(in_dev);
234 }
235 
236 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
237 {
238 	spin_lock_bh(&im->lock);
239 	im->unsolicit_count = 0;
240 	if (del_timer(&im->timer)) {
241 		if ((long)(im->timer.expires-jiffies) < max_delay) {
242 			add_timer(&im->timer);
243 			im->tm_running = 1;
244 			spin_unlock_bh(&im->lock);
245 			return;
246 		}
247 		atomic_dec(&im->refcnt);
248 	}
249 	igmp_start_timer(im, max_delay);
250 	spin_unlock_bh(&im->lock);
251 }
252 
253 
254 /*
255  *	Send an IGMP report.
256  */
257 
258 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
259 
260 
261 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
262 	int gdeleted, int sdeleted)
263 {
264 	switch (type) {
265 	case IGMPV3_MODE_IS_INCLUDE:
266 	case IGMPV3_MODE_IS_EXCLUDE:
267 		if (gdeleted || sdeleted)
268 			return 0;
269 		if (!(pmc->gsquery && !psf->sf_gsresp)) {
270 			if (pmc->sfmode == MCAST_INCLUDE)
271 				return 1;
272 			/* don't include if this source is excluded
273 			 * in all filters
274 			 */
275 			if (psf->sf_count[MCAST_INCLUDE])
276 				return type == IGMPV3_MODE_IS_INCLUDE;
277 			return pmc->sfcount[MCAST_EXCLUDE] ==
278 				psf->sf_count[MCAST_EXCLUDE];
279 		}
280 		return 0;
281 	case IGMPV3_CHANGE_TO_INCLUDE:
282 		if (gdeleted || sdeleted)
283 			return 0;
284 		return psf->sf_count[MCAST_INCLUDE] != 0;
285 	case IGMPV3_CHANGE_TO_EXCLUDE:
286 		if (gdeleted || sdeleted)
287 			return 0;
288 		if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
289 		    psf->sf_count[MCAST_INCLUDE])
290 			return 0;
291 		return pmc->sfcount[MCAST_EXCLUDE] ==
292 			psf->sf_count[MCAST_EXCLUDE];
293 	case IGMPV3_ALLOW_NEW_SOURCES:
294 		if (gdeleted || !psf->sf_crcount)
295 			return 0;
296 		return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
297 	case IGMPV3_BLOCK_OLD_SOURCES:
298 		if (pmc->sfmode == MCAST_INCLUDE)
299 			return gdeleted || (psf->sf_crcount && sdeleted);
300 		return psf->sf_crcount && !gdeleted && !sdeleted;
301 	}
302 	return 0;
303 }
304 
305 static int
306 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
307 {
308 	struct ip_sf_list *psf;
309 	int scount = 0;
310 
311 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
312 		if (!is_in(pmc, psf, type, gdeleted, sdeleted))
313 			continue;
314 		scount++;
315 	}
316 	return scount;
317 }
318 
319 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
320 {
321 	struct sk_buff *skb;
322 	struct rtable *rt;
323 	struct iphdr *pip;
324 	struct igmpv3_report *pig;
325 	struct net *net = dev_net(dev);
326 	struct flowi4 fl4;
327 	int hlen = LL_RESERVED_SPACE(dev);
328 	int tlen = dev->needed_tailroom;
329 	unsigned int size = mtu;
330 
331 	while (1) {
332 		skb = alloc_skb(size + hlen + tlen,
333 				GFP_ATOMIC | __GFP_NOWARN);
334 		if (skb)
335 			break;
336 		size >>= 1;
337 		if (size < 256)
338 			return NULL;
339 	}
340 	skb->priority = TC_PRIO_CONTROL;
341 
342 	rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
343 				   0, 0,
344 				   IPPROTO_IGMP, 0, dev->ifindex);
345 	if (IS_ERR(rt)) {
346 		kfree_skb(skb);
347 		return NULL;
348 	}
349 
350 	skb_dst_set(skb, &rt->dst);
351 	skb->dev = dev;
352 
353 	skb->reserved_tailroom = skb_end_offset(skb) -
354 				 min(mtu, skb_end_offset(skb));
355 	skb_reserve(skb, hlen);
356 
357 	skb_reset_network_header(skb);
358 	pip = ip_hdr(skb);
359 	skb_put(skb, sizeof(struct iphdr) + 4);
360 
361 	pip->version  = 4;
362 	pip->ihl      = (sizeof(struct iphdr)+4)>>2;
363 	pip->tos      = 0xc0;
364 	pip->frag_off = htons(IP_DF);
365 	pip->ttl      = 1;
366 	pip->daddr    = fl4.daddr;
367 	pip->saddr    = fl4.saddr;
368 	pip->protocol = IPPROTO_IGMP;
369 	pip->tot_len  = 0;	/* filled in later */
370 	ip_select_ident(net, skb, NULL);
371 	((u8 *)&pip[1])[0] = IPOPT_RA;
372 	((u8 *)&pip[1])[1] = 4;
373 	((u8 *)&pip[1])[2] = 0;
374 	((u8 *)&pip[1])[3] = 0;
375 
376 	skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
377 	skb_put(skb, sizeof(*pig));
378 	pig = igmpv3_report_hdr(skb);
379 	pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
380 	pig->resv1 = 0;
381 	pig->csum = 0;
382 	pig->resv2 = 0;
383 	pig->ngrec = 0;
384 	return skb;
385 }
386 
387 static int igmpv3_sendpack(struct sk_buff *skb)
388 {
389 	struct igmphdr *pig = igmp_hdr(skb);
390 	const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
391 
392 	pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
393 
394 	return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
395 }
396 
397 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
398 {
399 	return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
400 }
401 
402 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
403 	int type, struct igmpv3_grec **ppgr)
404 {
405 	struct net_device *dev = pmc->interface->dev;
406 	struct igmpv3_report *pih;
407 	struct igmpv3_grec *pgr;
408 
409 	if (!skb)
410 		skb = igmpv3_newpack(dev, dev->mtu);
411 	if (!skb)
412 		return NULL;
413 	pgr = (struct igmpv3_grec *)skb_put(skb, sizeof(struct igmpv3_grec));
414 	pgr->grec_type = type;
415 	pgr->grec_auxwords = 0;
416 	pgr->grec_nsrcs = 0;
417 	pgr->grec_mca = pmc->multiaddr;
418 	pih = igmpv3_report_hdr(skb);
419 	pih->ngrec = htons(ntohs(pih->ngrec)+1);
420 	*ppgr = pgr;
421 	return skb;
422 }
423 
424 #define AVAILABLE(skb)	((skb) ? skb_availroom(skb) : 0)
425 
426 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
427 	int type, int gdeleted, int sdeleted)
428 {
429 	struct net_device *dev = pmc->interface->dev;
430 	struct net *net = dev_net(dev);
431 	struct igmpv3_report *pih;
432 	struct igmpv3_grec *pgr = NULL;
433 	struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
434 	int scount, stotal, first, isquery, truncate;
435 
436 	if (pmc->multiaddr == IGMP_ALL_HOSTS)
437 		return skb;
438 	if (ipv4_is_local_multicast(pmc->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
439 		return skb;
440 
441 	isquery = type == IGMPV3_MODE_IS_INCLUDE ||
442 		  type == IGMPV3_MODE_IS_EXCLUDE;
443 	truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
444 		    type == IGMPV3_CHANGE_TO_EXCLUDE;
445 
446 	stotal = scount = 0;
447 
448 	psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
449 
450 	if (!*psf_list)
451 		goto empty_source;
452 
453 	pih = skb ? igmpv3_report_hdr(skb) : NULL;
454 
455 	/* EX and TO_EX get a fresh packet, if needed */
456 	if (truncate) {
457 		if (pih && pih->ngrec &&
458 		    AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
459 			if (skb)
460 				igmpv3_sendpack(skb);
461 			skb = igmpv3_newpack(dev, dev->mtu);
462 		}
463 	}
464 	first = 1;
465 	psf_prev = NULL;
466 	for (psf = *psf_list; psf; psf = psf_next) {
467 		__be32 *psrc;
468 
469 		psf_next = psf->sf_next;
470 
471 		if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
472 			psf_prev = psf;
473 			continue;
474 		}
475 
476 		/* clear marks on query responses */
477 		if (isquery)
478 			psf->sf_gsresp = 0;
479 
480 		if (AVAILABLE(skb) < sizeof(__be32) +
481 		    first*sizeof(struct igmpv3_grec)) {
482 			if (truncate && !first)
483 				break;	 /* truncate these */
484 			if (pgr)
485 				pgr->grec_nsrcs = htons(scount);
486 			if (skb)
487 				igmpv3_sendpack(skb);
488 			skb = igmpv3_newpack(dev, dev->mtu);
489 			first = 1;
490 			scount = 0;
491 		}
492 		if (first) {
493 			skb = add_grhead(skb, pmc, type, &pgr);
494 			first = 0;
495 		}
496 		if (!skb)
497 			return NULL;
498 		psrc = (__be32 *)skb_put(skb, sizeof(__be32));
499 		*psrc = psf->sf_inaddr;
500 		scount++; stotal++;
501 		if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
502 		     type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
503 			psf->sf_crcount--;
504 			if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
505 				if (psf_prev)
506 					psf_prev->sf_next = psf->sf_next;
507 				else
508 					*psf_list = psf->sf_next;
509 				kfree(psf);
510 				continue;
511 			}
512 		}
513 		psf_prev = psf;
514 	}
515 
516 empty_source:
517 	if (!stotal) {
518 		if (type == IGMPV3_ALLOW_NEW_SOURCES ||
519 		    type == IGMPV3_BLOCK_OLD_SOURCES)
520 			return skb;
521 		if (pmc->crcount || isquery) {
522 			/* make sure we have room for group header */
523 			if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
524 				igmpv3_sendpack(skb);
525 				skb = NULL; /* add_grhead will get a new one */
526 			}
527 			skb = add_grhead(skb, pmc, type, &pgr);
528 		}
529 	}
530 	if (pgr)
531 		pgr->grec_nsrcs = htons(scount);
532 
533 	if (isquery)
534 		pmc->gsquery = 0;	/* clear query state on report */
535 	return skb;
536 }
537 
538 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
539 {
540 	struct sk_buff *skb = NULL;
541 	struct net *net = dev_net(in_dev->dev);
542 	int type;
543 
544 	if (!pmc) {
545 		rcu_read_lock();
546 		for_each_pmc_rcu(in_dev, pmc) {
547 			if (pmc->multiaddr == IGMP_ALL_HOSTS)
548 				continue;
549 			if (ipv4_is_local_multicast(pmc->multiaddr) &&
550 			     !net->ipv4.sysctl_igmp_llm_reports)
551 				continue;
552 			spin_lock_bh(&pmc->lock);
553 			if (pmc->sfcount[MCAST_EXCLUDE])
554 				type = IGMPV3_MODE_IS_EXCLUDE;
555 			else
556 				type = IGMPV3_MODE_IS_INCLUDE;
557 			skb = add_grec(skb, pmc, type, 0, 0);
558 			spin_unlock_bh(&pmc->lock);
559 		}
560 		rcu_read_unlock();
561 	} else {
562 		spin_lock_bh(&pmc->lock);
563 		if (pmc->sfcount[MCAST_EXCLUDE])
564 			type = IGMPV3_MODE_IS_EXCLUDE;
565 		else
566 			type = IGMPV3_MODE_IS_INCLUDE;
567 		skb = add_grec(skb, pmc, type, 0, 0);
568 		spin_unlock_bh(&pmc->lock);
569 	}
570 	if (!skb)
571 		return 0;
572 	return igmpv3_sendpack(skb);
573 }
574 
575 /*
576  * remove zero-count source records from a source filter list
577  */
578 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
579 {
580 	struct ip_sf_list *psf_prev, *psf_next, *psf;
581 
582 	psf_prev = NULL;
583 	for (psf = *ppsf; psf; psf = psf_next) {
584 		psf_next = psf->sf_next;
585 		if (psf->sf_crcount == 0) {
586 			if (psf_prev)
587 				psf_prev->sf_next = psf->sf_next;
588 			else
589 				*ppsf = psf->sf_next;
590 			kfree(psf);
591 		} else
592 			psf_prev = psf;
593 	}
594 }
595 
596 static void igmpv3_send_cr(struct in_device *in_dev)
597 {
598 	struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
599 	struct sk_buff *skb = NULL;
600 	int type, dtype;
601 
602 	rcu_read_lock();
603 	spin_lock_bh(&in_dev->mc_tomb_lock);
604 
605 	/* deleted MCA's */
606 	pmc_prev = NULL;
607 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
608 		pmc_next = pmc->next;
609 		if (pmc->sfmode == MCAST_INCLUDE) {
610 			type = IGMPV3_BLOCK_OLD_SOURCES;
611 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
612 			skb = add_grec(skb, pmc, type, 1, 0);
613 			skb = add_grec(skb, pmc, dtype, 1, 1);
614 		}
615 		if (pmc->crcount) {
616 			if (pmc->sfmode == MCAST_EXCLUDE) {
617 				type = IGMPV3_CHANGE_TO_INCLUDE;
618 				skb = add_grec(skb, pmc, type, 1, 0);
619 			}
620 			pmc->crcount--;
621 			if (pmc->crcount == 0) {
622 				igmpv3_clear_zeros(&pmc->tomb);
623 				igmpv3_clear_zeros(&pmc->sources);
624 			}
625 		}
626 		if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
627 			if (pmc_prev)
628 				pmc_prev->next = pmc_next;
629 			else
630 				in_dev->mc_tomb = pmc_next;
631 			in_dev_put(pmc->interface);
632 			kfree(pmc);
633 		} else
634 			pmc_prev = pmc;
635 	}
636 	spin_unlock_bh(&in_dev->mc_tomb_lock);
637 
638 	/* change recs */
639 	for_each_pmc_rcu(in_dev, pmc) {
640 		spin_lock_bh(&pmc->lock);
641 		if (pmc->sfcount[MCAST_EXCLUDE]) {
642 			type = IGMPV3_BLOCK_OLD_SOURCES;
643 			dtype = IGMPV3_ALLOW_NEW_SOURCES;
644 		} else {
645 			type = IGMPV3_ALLOW_NEW_SOURCES;
646 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
647 		}
648 		skb = add_grec(skb, pmc, type, 0, 0);
649 		skb = add_grec(skb, pmc, dtype, 0, 1);	/* deleted sources */
650 
651 		/* filter mode changes */
652 		if (pmc->crcount) {
653 			if (pmc->sfmode == MCAST_EXCLUDE)
654 				type = IGMPV3_CHANGE_TO_EXCLUDE;
655 			else
656 				type = IGMPV3_CHANGE_TO_INCLUDE;
657 			skb = add_grec(skb, pmc, type, 0, 0);
658 			pmc->crcount--;
659 		}
660 		spin_unlock_bh(&pmc->lock);
661 	}
662 	rcu_read_unlock();
663 
664 	if (!skb)
665 		return;
666 	(void) igmpv3_sendpack(skb);
667 }
668 
669 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
670 	int type)
671 {
672 	struct sk_buff *skb;
673 	struct iphdr *iph;
674 	struct igmphdr *ih;
675 	struct rtable *rt;
676 	struct net_device *dev = in_dev->dev;
677 	struct net *net = dev_net(dev);
678 	__be32	group = pmc ? pmc->multiaddr : 0;
679 	struct flowi4 fl4;
680 	__be32	dst;
681 	int hlen, tlen;
682 
683 	if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
684 		return igmpv3_send_report(in_dev, pmc);
685 
686 	if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
687 		return 0;
688 
689 	if (type == IGMP_HOST_LEAVE_MESSAGE)
690 		dst = IGMP_ALL_ROUTER;
691 	else
692 		dst = group;
693 
694 	rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
695 				   0, 0,
696 				   IPPROTO_IGMP, 0, dev->ifindex);
697 	if (IS_ERR(rt))
698 		return -1;
699 
700 	hlen = LL_RESERVED_SPACE(dev);
701 	tlen = dev->needed_tailroom;
702 	skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
703 	if (!skb) {
704 		ip_rt_put(rt);
705 		return -1;
706 	}
707 	skb->priority = TC_PRIO_CONTROL;
708 
709 	skb_dst_set(skb, &rt->dst);
710 
711 	skb_reserve(skb, hlen);
712 
713 	skb_reset_network_header(skb);
714 	iph = ip_hdr(skb);
715 	skb_put(skb, sizeof(struct iphdr) + 4);
716 
717 	iph->version  = 4;
718 	iph->ihl      = (sizeof(struct iphdr)+4)>>2;
719 	iph->tos      = 0xc0;
720 	iph->frag_off = htons(IP_DF);
721 	iph->ttl      = 1;
722 	iph->daddr    = dst;
723 	iph->saddr    = fl4.saddr;
724 	iph->protocol = IPPROTO_IGMP;
725 	ip_select_ident(net, skb, NULL);
726 	((u8 *)&iph[1])[0] = IPOPT_RA;
727 	((u8 *)&iph[1])[1] = 4;
728 	((u8 *)&iph[1])[2] = 0;
729 	((u8 *)&iph[1])[3] = 0;
730 
731 	ih = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
732 	ih->type = type;
733 	ih->code = 0;
734 	ih->csum = 0;
735 	ih->group = group;
736 	ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
737 
738 	return ip_local_out(net, skb->sk, skb);
739 }
740 
741 static void igmp_gq_timer_expire(unsigned long data)
742 {
743 	struct in_device *in_dev = (struct in_device *)data;
744 
745 	in_dev->mr_gq_running = 0;
746 	igmpv3_send_report(in_dev, NULL);
747 	in_dev_put(in_dev);
748 }
749 
750 static void igmp_ifc_timer_expire(unsigned long data)
751 {
752 	struct in_device *in_dev = (struct in_device *)data;
753 
754 	igmpv3_send_cr(in_dev);
755 	if (in_dev->mr_ifc_count) {
756 		in_dev->mr_ifc_count--;
757 		igmp_ifc_start_timer(in_dev,
758 				     unsolicited_report_interval(in_dev));
759 	}
760 	in_dev_put(in_dev);
761 }
762 
763 static void igmp_ifc_event(struct in_device *in_dev)
764 {
765 	struct net *net = dev_net(in_dev->dev);
766 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
767 		return;
768 	in_dev->mr_ifc_count = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
769 	igmp_ifc_start_timer(in_dev, 1);
770 }
771 
772 
773 static void igmp_timer_expire(unsigned long data)
774 {
775 	struct ip_mc_list *im = (struct ip_mc_list *)data;
776 	struct in_device *in_dev = im->interface;
777 
778 	spin_lock(&im->lock);
779 	im->tm_running = 0;
780 
781 	if (im->unsolicit_count) {
782 		im->unsolicit_count--;
783 		igmp_start_timer(im, unsolicited_report_interval(in_dev));
784 	}
785 	im->reporter = 1;
786 	spin_unlock(&im->lock);
787 
788 	if (IGMP_V1_SEEN(in_dev))
789 		igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
790 	else if (IGMP_V2_SEEN(in_dev))
791 		igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
792 	else
793 		igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
794 
795 	ip_ma_put(im);
796 }
797 
798 /* mark EXCLUDE-mode sources */
799 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
800 {
801 	struct ip_sf_list *psf;
802 	int i, scount;
803 
804 	scount = 0;
805 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
806 		if (scount == nsrcs)
807 			break;
808 		for (i = 0; i < nsrcs; i++) {
809 			/* skip inactive filters */
810 			if (psf->sf_count[MCAST_INCLUDE] ||
811 			    pmc->sfcount[MCAST_EXCLUDE] !=
812 			    psf->sf_count[MCAST_EXCLUDE])
813 				break;
814 			if (srcs[i] == psf->sf_inaddr) {
815 				scount++;
816 				break;
817 			}
818 		}
819 	}
820 	pmc->gsquery = 0;
821 	if (scount == nsrcs)	/* all sources excluded */
822 		return 0;
823 	return 1;
824 }
825 
826 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
827 {
828 	struct ip_sf_list *psf;
829 	int i, scount;
830 
831 	if (pmc->sfmode == MCAST_EXCLUDE)
832 		return igmp_xmarksources(pmc, nsrcs, srcs);
833 
834 	/* mark INCLUDE-mode sources */
835 	scount = 0;
836 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
837 		if (scount == nsrcs)
838 			break;
839 		for (i = 0; i < nsrcs; i++)
840 			if (srcs[i] == psf->sf_inaddr) {
841 				psf->sf_gsresp = 1;
842 				scount++;
843 				break;
844 			}
845 	}
846 	if (!scount) {
847 		pmc->gsquery = 0;
848 		return 0;
849 	}
850 	pmc->gsquery = 1;
851 	return 1;
852 }
853 
854 /* return true if packet was dropped */
855 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
856 {
857 	struct ip_mc_list *im;
858 	struct net *net = dev_net(in_dev->dev);
859 
860 	/* Timers are only set for non-local groups */
861 
862 	if (group == IGMP_ALL_HOSTS)
863 		return false;
864 	if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
865 		return false;
866 
867 	rcu_read_lock();
868 	for_each_pmc_rcu(in_dev, im) {
869 		if (im->multiaddr == group) {
870 			igmp_stop_timer(im);
871 			break;
872 		}
873 	}
874 	rcu_read_unlock();
875 	return false;
876 }
877 
878 /* return true if packet was dropped */
879 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
880 	int len)
881 {
882 	struct igmphdr 		*ih = igmp_hdr(skb);
883 	struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
884 	struct ip_mc_list	*im;
885 	__be32			group = ih->group;
886 	int			max_delay;
887 	int			mark = 0;
888 	struct net		*net = dev_net(in_dev->dev);
889 
890 
891 	if (len == 8) {
892 		if (ih->code == 0) {
893 			/* Alas, old v1 router presents here. */
894 
895 			max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
896 			in_dev->mr_v1_seen = jiffies +
897 				IGMP_V1_ROUTER_PRESENT_TIMEOUT;
898 			group = 0;
899 		} else {
900 			/* v2 router present */
901 			max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
902 			in_dev->mr_v2_seen = jiffies +
903 				IGMP_V2_ROUTER_PRESENT_TIMEOUT;
904 		}
905 		/* cancel the interface change timer */
906 		in_dev->mr_ifc_count = 0;
907 		if (del_timer(&in_dev->mr_ifc_timer))
908 			__in_dev_put(in_dev);
909 		/* clear deleted report items */
910 		igmpv3_clear_delrec(in_dev);
911 	} else if (len < 12) {
912 		return true;	/* ignore bogus packet; freed by caller */
913 	} else if (IGMP_V1_SEEN(in_dev)) {
914 		/* This is a v3 query with v1 queriers present */
915 		max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
916 		group = 0;
917 	} else if (IGMP_V2_SEEN(in_dev)) {
918 		/* this is a v3 query with v2 queriers present;
919 		 * Interpretation of the max_delay code is problematic here.
920 		 * A real v2 host would use ih_code directly, while v3 has a
921 		 * different encoding. We use the v3 encoding as more likely
922 		 * to be intended in a v3 query.
923 		 */
924 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
925 		if (!max_delay)
926 			max_delay = 1;	/* can't mod w/ 0 */
927 	} else { /* v3 */
928 		if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
929 			return true;
930 
931 		ih3 = igmpv3_query_hdr(skb);
932 		if (ih3->nsrcs) {
933 			if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
934 					   + ntohs(ih3->nsrcs)*sizeof(__be32)))
935 				return true;
936 			ih3 = igmpv3_query_hdr(skb);
937 		}
938 
939 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
940 		if (!max_delay)
941 			max_delay = 1;	/* can't mod w/ 0 */
942 		in_dev->mr_maxdelay = max_delay;
943 		if (ih3->qrv)
944 			in_dev->mr_qrv = ih3->qrv;
945 		if (!group) { /* general query */
946 			if (ih3->nsrcs)
947 				return true;	/* no sources allowed */
948 			igmp_gq_start_timer(in_dev);
949 			return false;
950 		}
951 		/* mark sources to include, if group & source-specific */
952 		mark = ih3->nsrcs != 0;
953 	}
954 
955 	/*
956 	 * - Start the timers in all of our membership records
957 	 *   that the query applies to for the interface on
958 	 *   which the query arrived excl. those that belong
959 	 *   to a "local" group (224.0.0.X)
960 	 * - For timers already running check if they need to
961 	 *   be reset.
962 	 * - Use the igmp->igmp_code field as the maximum
963 	 *   delay possible
964 	 */
965 	rcu_read_lock();
966 	for_each_pmc_rcu(in_dev, im) {
967 		int changed;
968 
969 		if (group && group != im->multiaddr)
970 			continue;
971 		if (im->multiaddr == IGMP_ALL_HOSTS)
972 			continue;
973 		if (ipv4_is_local_multicast(im->multiaddr) &&
974 		    !net->ipv4.sysctl_igmp_llm_reports)
975 			continue;
976 		spin_lock_bh(&im->lock);
977 		if (im->tm_running)
978 			im->gsquery = im->gsquery && mark;
979 		else
980 			im->gsquery = mark;
981 		changed = !im->gsquery ||
982 			igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
983 		spin_unlock_bh(&im->lock);
984 		if (changed)
985 			igmp_mod_timer(im, max_delay);
986 	}
987 	rcu_read_unlock();
988 	return false;
989 }
990 
991 /* called in rcu_read_lock() section */
992 int igmp_rcv(struct sk_buff *skb)
993 {
994 	/* This basically follows the spec line by line -- see RFC1112 */
995 	struct igmphdr *ih;
996 	struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
997 	int len = skb->len;
998 	bool dropped = true;
999 
1000 	if (!in_dev)
1001 		goto drop;
1002 
1003 	if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1004 		goto drop;
1005 
1006 	if (skb_checksum_simple_validate(skb))
1007 		goto drop;
1008 
1009 	ih = igmp_hdr(skb);
1010 	switch (ih->type) {
1011 	case IGMP_HOST_MEMBERSHIP_QUERY:
1012 		dropped = igmp_heard_query(in_dev, skb, len);
1013 		break;
1014 	case IGMP_HOST_MEMBERSHIP_REPORT:
1015 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
1016 		/* Is it our report looped back? */
1017 		if (rt_is_output_route(skb_rtable(skb)))
1018 			break;
1019 		/* don't rely on MC router hearing unicast reports */
1020 		if (skb->pkt_type == PACKET_MULTICAST ||
1021 		    skb->pkt_type == PACKET_BROADCAST)
1022 			dropped = igmp_heard_report(in_dev, ih->group);
1023 		break;
1024 	case IGMP_PIM:
1025 #ifdef CONFIG_IP_PIMSM_V1
1026 		return pim_rcv_v1(skb);
1027 #endif
1028 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
1029 	case IGMP_DVMRP:
1030 	case IGMP_TRACE:
1031 	case IGMP_HOST_LEAVE_MESSAGE:
1032 	case IGMP_MTRACE:
1033 	case IGMP_MTRACE_RESP:
1034 		break;
1035 	default:
1036 		break;
1037 	}
1038 
1039 drop:
1040 	if (dropped)
1041 		kfree_skb(skb);
1042 	else
1043 		consume_skb(skb);
1044 	return 0;
1045 }
1046 
1047 #endif
1048 
1049 
1050 /*
1051  *	Add a filter to a device
1052  */
1053 
1054 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1055 {
1056 	char buf[MAX_ADDR_LEN];
1057 	struct net_device *dev = in_dev->dev;
1058 
1059 	/* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1060 	   We will get multicast token leakage, when IFF_MULTICAST
1061 	   is changed. This check should be done in ndo_set_rx_mode
1062 	   routine. Something sort of:
1063 	   if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1064 	   --ANK
1065 	   */
1066 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1067 		dev_mc_add(dev, buf);
1068 }
1069 
1070 /*
1071  *	Remove a filter from a device
1072  */
1073 
1074 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1075 {
1076 	char buf[MAX_ADDR_LEN];
1077 	struct net_device *dev = in_dev->dev;
1078 
1079 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1080 		dev_mc_del(dev, buf);
1081 }
1082 
1083 #ifdef CONFIG_IP_MULTICAST
1084 /*
1085  * deleted ip_mc_list manipulation
1086  */
1087 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1088 {
1089 	struct ip_mc_list *pmc;
1090 	struct net *net = dev_net(in_dev->dev);
1091 
1092 	/* this is an "ip_mc_list" for convenience; only the fields below
1093 	 * are actually used. In particular, the refcnt and users are not
1094 	 * used for management of the delete list. Using the same structure
1095 	 * for deleted items allows change reports to use common code with
1096 	 * non-deleted or query-response MCA's.
1097 	 */
1098 	pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1099 	if (!pmc)
1100 		return;
1101 	spin_lock_bh(&im->lock);
1102 	pmc->interface = im->interface;
1103 	in_dev_hold(in_dev);
1104 	pmc->multiaddr = im->multiaddr;
1105 	pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1106 	pmc->sfmode = im->sfmode;
1107 	if (pmc->sfmode == MCAST_INCLUDE) {
1108 		struct ip_sf_list *psf;
1109 
1110 		pmc->tomb = im->tomb;
1111 		pmc->sources = im->sources;
1112 		im->tomb = im->sources = NULL;
1113 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1114 			psf->sf_crcount = pmc->crcount;
1115 	}
1116 	spin_unlock_bh(&im->lock);
1117 
1118 	spin_lock_bh(&in_dev->mc_tomb_lock);
1119 	pmc->next = in_dev->mc_tomb;
1120 	in_dev->mc_tomb = pmc;
1121 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1122 }
1123 
1124 static void igmpv3_del_delrec(struct in_device *in_dev, __be32 multiaddr)
1125 {
1126 	struct ip_mc_list *pmc, *pmc_prev;
1127 	struct ip_sf_list *psf, *psf_next;
1128 
1129 	spin_lock_bh(&in_dev->mc_tomb_lock);
1130 	pmc_prev = NULL;
1131 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1132 		if (pmc->multiaddr == multiaddr)
1133 			break;
1134 		pmc_prev = pmc;
1135 	}
1136 	if (pmc) {
1137 		if (pmc_prev)
1138 			pmc_prev->next = pmc->next;
1139 		else
1140 			in_dev->mc_tomb = pmc->next;
1141 	}
1142 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1143 	if (pmc) {
1144 		for (psf = pmc->tomb; psf; psf = psf_next) {
1145 			psf_next = psf->sf_next;
1146 			kfree(psf);
1147 		}
1148 		in_dev_put(pmc->interface);
1149 		kfree(pmc);
1150 	}
1151 }
1152 
1153 static void igmpv3_clear_delrec(struct in_device *in_dev)
1154 {
1155 	struct ip_mc_list *pmc, *nextpmc;
1156 
1157 	spin_lock_bh(&in_dev->mc_tomb_lock);
1158 	pmc = in_dev->mc_tomb;
1159 	in_dev->mc_tomb = NULL;
1160 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1161 
1162 	for (; pmc; pmc = nextpmc) {
1163 		nextpmc = pmc->next;
1164 		ip_mc_clear_src(pmc);
1165 		in_dev_put(pmc->interface);
1166 		kfree(pmc);
1167 	}
1168 	/* clear dead sources, too */
1169 	rcu_read_lock();
1170 	for_each_pmc_rcu(in_dev, pmc) {
1171 		struct ip_sf_list *psf, *psf_next;
1172 
1173 		spin_lock_bh(&pmc->lock);
1174 		psf = pmc->tomb;
1175 		pmc->tomb = NULL;
1176 		spin_unlock_bh(&pmc->lock);
1177 		for (; psf; psf = psf_next) {
1178 			psf_next = psf->sf_next;
1179 			kfree(psf);
1180 		}
1181 	}
1182 	rcu_read_unlock();
1183 }
1184 #endif
1185 
1186 static void igmp_group_dropped(struct ip_mc_list *im)
1187 {
1188 	struct in_device *in_dev = im->interface;
1189 #ifdef CONFIG_IP_MULTICAST
1190 	struct net *net = dev_net(in_dev->dev);
1191 	int reporter;
1192 #endif
1193 
1194 	if (im->loaded) {
1195 		im->loaded = 0;
1196 		ip_mc_filter_del(in_dev, im->multiaddr);
1197 	}
1198 
1199 #ifdef CONFIG_IP_MULTICAST
1200 	if (im->multiaddr == IGMP_ALL_HOSTS)
1201 		return;
1202 	if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1203 		return;
1204 
1205 	reporter = im->reporter;
1206 	igmp_stop_timer(im);
1207 
1208 	if (!in_dev->dead) {
1209 		if (IGMP_V1_SEEN(in_dev))
1210 			return;
1211 		if (IGMP_V2_SEEN(in_dev)) {
1212 			if (reporter)
1213 				igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1214 			return;
1215 		}
1216 		/* IGMPv3 */
1217 		igmpv3_add_delrec(in_dev, im);
1218 
1219 		igmp_ifc_event(in_dev);
1220 	}
1221 #endif
1222 }
1223 
1224 static void igmp_group_added(struct ip_mc_list *im)
1225 {
1226 	struct in_device *in_dev = im->interface;
1227 #ifdef CONFIG_IP_MULTICAST
1228 	struct net *net = dev_net(in_dev->dev);
1229 #endif
1230 
1231 	if (im->loaded == 0) {
1232 		im->loaded = 1;
1233 		ip_mc_filter_add(in_dev, im->multiaddr);
1234 	}
1235 
1236 #ifdef CONFIG_IP_MULTICAST
1237 	if (im->multiaddr == IGMP_ALL_HOSTS)
1238 		return;
1239 	if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1240 		return;
1241 
1242 	if (in_dev->dead)
1243 		return;
1244 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1245 		spin_lock_bh(&im->lock);
1246 		igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1247 		spin_unlock_bh(&im->lock);
1248 		return;
1249 	}
1250 	/* else, v3 */
1251 
1252 	im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1253 	igmp_ifc_event(in_dev);
1254 #endif
1255 }
1256 
1257 
1258 /*
1259  *	Multicast list managers
1260  */
1261 
1262 static u32 ip_mc_hash(const struct ip_mc_list *im)
1263 {
1264 	return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1265 }
1266 
1267 static void ip_mc_hash_add(struct in_device *in_dev,
1268 			   struct ip_mc_list *im)
1269 {
1270 	struct ip_mc_list __rcu **mc_hash;
1271 	u32 hash;
1272 
1273 	mc_hash = rtnl_dereference(in_dev->mc_hash);
1274 	if (mc_hash) {
1275 		hash = ip_mc_hash(im);
1276 		im->next_hash = mc_hash[hash];
1277 		rcu_assign_pointer(mc_hash[hash], im);
1278 		return;
1279 	}
1280 
1281 	/* do not use a hash table for small number of items */
1282 	if (in_dev->mc_count < 4)
1283 		return;
1284 
1285 	mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1286 			  GFP_KERNEL);
1287 	if (!mc_hash)
1288 		return;
1289 
1290 	for_each_pmc_rtnl(in_dev, im) {
1291 		hash = ip_mc_hash(im);
1292 		im->next_hash = mc_hash[hash];
1293 		RCU_INIT_POINTER(mc_hash[hash], im);
1294 	}
1295 
1296 	rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1297 }
1298 
1299 static void ip_mc_hash_remove(struct in_device *in_dev,
1300 			      struct ip_mc_list *im)
1301 {
1302 	struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1303 	struct ip_mc_list *aux;
1304 
1305 	if (!mc_hash)
1306 		return;
1307 	mc_hash += ip_mc_hash(im);
1308 	while ((aux = rtnl_dereference(*mc_hash)) != im)
1309 		mc_hash = &aux->next_hash;
1310 	*mc_hash = im->next_hash;
1311 }
1312 
1313 
1314 /*
1315  *	A socket has joined a multicast group on device dev.
1316  */
1317 
1318 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1319 {
1320 	struct ip_mc_list *im;
1321 #ifdef CONFIG_IP_MULTICAST
1322 	struct net *net = dev_net(in_dev->dev);
1323 #endif
1324 
1325 	ASSERT_RTNL();
1326 
1327 	for_each_pmc_rtnl(in_dev, im) {
1328 		if (im->multiaddr == addr) {
1329 			im->users++;
1330 			ip_mc_add_src(in_dev, &addr, MCAST_EXCLUDE, 0, NULL, 0);
1331 			goto out;
1332 		}
1333 	}
1334 
1335 	im = kzalloc(sizeof(*im), GFP_KERNEL);
1336 	if (!im)
1337 		goto out;
1338 
1339 	im->users = 1;
1340 	im->interface = in_dev;
1341 	in_dev_hold(in_dev);
1342 	im->multiaddr = addr;
1343 	/* initial mode is (EX, empty) */
1344 	im->sfmode = MCAST_EXCLUDE;
1345 	im->sfcount[MCAST_EXCLUDE] = 1;
1346 	atomic_set(&im->refcnt, 1);
1347 	spin_lock_init(&im->lock);
1348 #ifdef CONFIG_IP_MULTICAST
1349 	setup_timer(&im->timer, igmp_timer_expire, (unsigned long)im);
1350 	im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1351 #endif
1352 
1353 	im->next_rcu = in_dev->mc_list;
1354 	in_dev->mc_count++;
1355 	rcu_assign_pointer(in_dev->mc_list, im);
1356 
1357 	ip_mc_hash_add(in_dev, im);
1358 
1359 #ifdef CONFIG_IP_MULTICAST
1360 	igmpv3_del_delrec(in_dev, im->multiaddr);
1361 #endif
1362 	igmp_group_added(im);
1363 	if (!in_dev->dead)
1364 		ip_rt_multicast_event(in_dev);
1365 out:
1366 	return;
1367 }
1368 EXPORT_SYMBOL(ip_mc_inc_group);
1369 
1370 static int ip_mc_check_iphdr(struct sk_buff *skb)
1371 {
1372 	const struct iphdr *iph;
1373 	unsigned int len;
1374 	unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1375 
1376 	if (!pskb_may_pull(skb, offset))
1377 		return -EINVAL;
1378 
1379 	iph = ip_hdr(skb);
1380 
1381 	if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1382 		return -EINVAL;
1383 
1384 	offset += ip_hdrlen(skb) - sizeof(*iph);
1385 
1386 	if (!pskb_may_pull(skb, offset))
1387 		return -EINVAL;
1388 
1389 	iph = ip_hdr(skb);
1390 
1391 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1392 		return -EINVAL;
1393 
1394 	len = skb_network_offset(skb) + ntohs(iph->tot_len);
1395 	if (skb->len < len || len < offset)
1396 		return -EINVAL;
1397 
1398 	skb_set_transport_header(skb, offset);
1399 
1400 	return 0;
1401 }
1402 
1403 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1404 {
1405 	unsigned int len = skb_transport_offset(skb);
1406 
1407 	len += sizeof(struct igmpv3_report);
1408 
1409 	return pskb_may_pull(skb, len) ? 0 : -EINVAL;
1410 }
1411 
1412 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1413 {
1414 	unsigned int len = skb_transport_offset(skb);
1415 
1416 	len += sizeof(struct igmphdr);
1417 	if (skb->len < len)
1418 		return -EINVAL;
1419 
1420 	/* IGMPv{1,2}? */
1421 	if (skb->len != len) {
1422 		/* or IGMPv3? */
1423 		len += sizeof(struct igmpv3_query) - sizeof(struct igmphdr);
1424 		if (skb->len < len || !pskb_may_pull(skb, len))
1425 			return -EINVAL;
1426 	}
1427 
1428 	/* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1429 	 * all-systems destination addresses (224.0.0.1) for general queries
1430 	 */
1431 	if (!igmp_hdr(skb)->group &&
1432 	    ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1433 		return -EINVAL;
1434 
1435 	return 0;
1436 }
1437 
1438 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1439 {
1440 	switch (igmp_hdr(skb)->type) {
1441 	case IGMP_HOST_LEAVE_MESSAGE:
1442 	case IGMP_HOST_MEMBERSHIP_REPORT:
1443 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
1444 		/* fall through */
1445 		return 0;
1446 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
1447 		return ip_mc_check_igmp_reportv3(skb);
1448 	case IGMP_HOST_MEMBERSHIP_QUERY:
1449 		return ip_mc_check_igmp_query(skb);
1450 	default:
1451 		return -ENOMSG;
1452 	}
1453 }
1454 
1455 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1456 {
1457 	return skb_checksum_simple_validate(skb);
1458 }
1459 
1460 static int __ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1461 
1462 {
1463 	struct sk_buff *skb_chk;
1464 	unsigned int transport_len;
1465 	unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1466 	int ret = -EINVAL;
1467 
1468 	transport_len = ntohs(ip_hdr(skb)->tot_len) - ip_hdrlen(skb);
1469 
1470 	skb_chk = skb_checksum_trimmed(skb, transport_len,
1471 				       ip_mc_validate_checksum);
1472 	if (!skb_chk)
1473 		goto err;
1474 
1475 	if (!pskb_may_pull(skb_chk, len))
1476 		goto err;
1477 
1478 	ret = ip_mc_check_igmp_msg(skb_chk);
1479 	if (ret)
1480 		goto err;
1481 
1482 	if (skb_trimmed)
1483 		*skb_trimmed = skb_chk;
1484 	/* free now unneeded clone */
1485 	else if (skb_chk != skb)
1486 		kfree_skb(skb_chk);
1487 
1488 	ret = 0;
1489 
1490 err:
1491 	if (ret && skb_chk && skb_chk != skb)
1492 		kfree_skb(skb_chk);
1493 
1494 	return ret;
1495 }
1496 
1497 /**
1498  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1499  * @skb: the skb to validate
1500  * @skb_trimmed: to store an skb pointer trimmed to IPv4 packet tail (optional)
1501  *
1502  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1503  * skb transport header accordingly and returns zero.
1504  *
1505  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1506  *  standard
1507  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1508  * -ENOMEM: A memory allocation failure happened.
1509  *
1510  * Optionally, an skb pointer might be provided via skb_trimmed (or set it
1511  * to NULL): After parsing an IGMP packet successfully it will point to
1512  * an skb which has its tail aligned to the IP packet end. This might
1513  * either be the originally provided skb or a trimmed, cloned version if
1514  * the skb frame had data beyond the IP packet. A cloned skb allows us
1515  * to leave the original skb and its full frame unchanged (which might be
1516  * desirable for layer 2 frame jugglers).
1517  *
1518  * Caller needs to set the skb network header and free any returned skb if it
1519  * differs from the provided skb.
1520  */
1521 int ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1522 {
1523 	int ret = ip_mc_check_iphdr(skb);
1524 
1525 	if (ret < 0)
1526 		return ret;
1527 
1528 	if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1529 		return -ENOMSG;
1530 
1531 	return __ip_mc_check_igmp(skb, skb_trimmed);
1532 }
1533 EXPORT_SYMBOL(ip_mc_check_igmp);
1534 
1535 /*
1536  *	Resend IGMP JOIN report; used by netdev notifier.
1537  */
1538 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1539 {
1540 #ifdef CONFIG_IP_MULTICAST
1541 	struct ip_mc_list *im;
1542 	int type;
1543 	struct net *net = dev_net(in_dev->dev);
1544 
1545 	ASSERT_RTNL();
1546 
1547 	for_each_pmc_rtnl(in_dev, im) {
1548 		if (im->multiaddr == IGMP_ALL_HOSTS)
1549 			continue;
1550 		if (ipv4_is_local_multicast(im->multiaddr) &&
1551 		    !net->ipv4.sysctl_igmp_llm_reports)
1552 			continue;
1553 
1554 		/* a failover is happening and switches
1555 		 * must be notified immediately
1556 		 */
1557 		if (IGMP_V1_SEEN(in_dev))
1558 			type = IGMP_HOST_MEMBERSHIP_REPORT;
1559 		else if (IGMP_V2_SEEN(in_dev))
1560 			type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1561 		else
1562 			type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1563 		igmp_send_report(in_dev, im, type);
1564 	}
1565 #endif
1566 }
1567 
1568 /*
1569  *	A socket has left a multicast group on device dev
1570  */
1571 
1572 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1573 {
1574 	struct ip_mc_list *i;
1575 	struct ip_mc_list __rcu **ip;
1576 
1577 	ASSERT_RTNL();
1578 
1579 	for (ip = &in_dev->mc_list;
1580 	     (i = rtnl_dereference(*ip)) != NULL;
1581 	     ip = &i->next_rcu) {
1582 		if (i->multiaddr == addr) {
1583 			if (--i->users == 0) {
1584 				ip_mc_hash_remove(in_dev, i);
1585 				*ip = i->next_rcu;
1586 				in_dev->mc_count--;
1587 				igmp_group_dropped(i);
1588 				ip_mc_clear_src(i);
1589 
1590 				if (!in_dev->dead)
1591 					ip_rt_multicast_event(in_dev);
1592 
1593 				ip_ma_put(i);
1594 				return;
1595 			}
1596 			break;
1597 		}
1598 	}
1599 }
1600 EXPORT_SYMBOL(ip_mc_dec_group);
1601 
1602 /* Device changing type */
1603 
1604 void ip_mc_unmap(struct in_device *in_dev)
1605 {
1606 	struct ip_mc_list *pmc;
1607 
1608 	ASSERT_RTNL();
1609 
1610 	for_each_pmc_rtnl(in_dev, pmc)
1611 		igmp_group_dropped(pmc);
1612 }
1613 
1614 void ip_mc_remap(struct in_device *in_dev)
1615 {
1616 	struct ip_mc_list *pmc;
1617 
1618 	ASSERT_RTNL();
1619 
1620 	for_each_pmc_rtnl(in_dev, pmc)
1621 		igmp_group_added(pmc);
1622 }
1623 
1624 /* Device going down */
1625 
1626 void ip_mc_down(struct in_device *in_dev)
1627 {
1628 	struct ip_mc_list *pmc;
1629 
1630 	ASSERT_RTNL();
1631 
1632 	for_each_pmc_rtnl(in_dev, pmc)
1633 		igmp_group_dropped(pmc);
1634 
1635 #ifdef CONFIG_IP_MULTICAST
1636 	in_dev->mr_ifc_count = 0;
1637 	if (del_timer(&in_dev->mr_ifc_timer))
1638 		__in_dev_put(in_dev);
1639 	in_dev->mr_gq_running = 0;
1640 	if (del_timer(&in_dev->mr_gq_timer))
1641 		__in_dev_put(in_dev);
1642 	igmpv3_clear_delrec(in_dev);
1643 #endif
1644 
1645 	ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1646 }
1647 
1648 void ip_mc_init_dev(struct in_device *in_dev)
1649 {
1650 #ifdef CONFIG_IP_MULTICAST
1651 	struct net *net = dev_net(in_dev->dev);
1652 #endif
1653 	ASSERT_RTNL();
1654 
1655 #ifdef CONFIG_IP_MULTICAST
1656 	setup_timer(&in_dev->mr_gq_timer, igmp_gq_timer_expire,
1657 			(unsigned long)in_dev);
1658 	setup_timer(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire,
1659 			(unsigned long)in_dev);
1660 	in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1661 #endif
1662 
1663 	spin_lock_init(&in_dev->mc_tomb_lock);
1664 }
1665 
1666 /* Device going up */
1667 
1668 void ip_mc_up(struct in_device *in_dev)
1669 {
1670 	struct ip_mc_list *pmc;
1671 #ifdef CONFIG_IP_MULTICAST
1672 	struct net *net = dev_net(in_dev->dev);
1673 #endif
1674 
1675 	ASSERT_RTNL();
1676 
1677 #ifdef CONFIG_IP_MULTICAST
1678 	in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1679 #endif
1680 	ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1681 
1682 	for_each_pmc_rtnl(in_dev, pmc)
1683 		igmp_group_added(pmc);
1684 }
1685 
1686 /*
1687  *	Device is about to be destroyed: clean up.
1688  */
1689 
1690 void ip_mc_destroy_dev(struct in_device *in_dev)
1691 {
1692 	struct ip_mc_list *i;
1693 
1694 	ASSERT_RTNL();
1695 
1696 	/* Deactivate timers */
1697 	ip_mc_down(in_dev);
1698 
1699 	while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1700 		in_dev->mc_list = i->next_rcu;
1701 		in_dev->mc_count--;
1702 
1703 		/* We've dropped the groups in ip_mc_down already */
1704 		ip_mc_clear_src(i);
1705 		ip_ma_put(i);
1706 	}
1707 }
1708 
1709 /* RTNL is locked */
1710 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1711 {
1712 	struct net_device *dev = NULL;
1713 	struct in_device *idev = NULL;
1714 
1715 	if (imr->imr_ifindex) {
1716 		idev = inetdev_by_index(net, imr->imr_ifindex);
1717 		return idev;
1718 	}
1719 	if (imr->imr_address.s_addr) {
1720 		dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1721 		if (!dev)
1722 			return NULL;
1723 	}
1724 
1725 	if (!dev) {
1726 		struct rtable *rt = ip_route_output(net,
1727 						    imr->imr_multiaddr.s_addr,
1728 						    0, 0, 0);
1729 		if (!IS_ERR(rt)) {
1730 			dev = rt->dst.dev;
1731 			ip_rt_put(rt);
1732 		}
1733 	}
1734 	if (dev) {
1735 		imr->imr_ifindex = dev->ifindex;
1736 		idev = __in_dev_get_rtnl(dev);
1737 	}
1738 	return idev;
1739 }
1740 
1741 /*
1742  *	Join a socket to a group
1743  */
1744 
1745 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1746 	__be32 *psfsrc)
1747 {
1748 	struct ip_sf_list *psf, *psf_prev;
1749 	int rv = 0;
1750 
1751 	psf_prev = NULL;
1752 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1753 		if (psf->sf_inaddr == *psfsrc)
1754 			break;
1755 		psf_prev = psf;
1756 	}
1757 	if (!psf || psf->sf_count[sfmode] == 0) {
1758 		/* source filter not found, or count wrong =>  bug */
1759 		return -ESRCH;
1760 	}
1761 	psf->sf_count[sfmode]--;
1762 	if (psf->sf_count[sfmode] == 0) {
1763 		ip_rt_multicast_event(pmc->interface);
1764 	}
1765 	if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1766 #ifdef CONFIG_IP_MULTICAST
1767 		struct in_device *in_dev = pmc->interface;
1768 		struct net *net = dev_net(in_dev->dev);
1769 #endif
1770 
1771 		/* no more filters for this source */
1772 		if (psf_prev)
1773 			psf_prev->sf_next = psf->sf_next;
1774 		else
1775 			pmc->sources = psf->sf_next;
1776 #ifdef CONFIG_IP_MULTICAST
1777 		if (psf->sf_oldin &&
1778 		    !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1779 			psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1780 			psf->sf_next = pmc->tomb;
1781 			pmc->tomb = psf;
1782 			rv = 1;
1783 		} else
1784 #endif
1785 			kfree(psf);
1786 	}
1787 	return rv;
1788 }
1789 
1790 #ifndef CONFIG_IP_MULTICAST
1791 #define igmp_ifc_event(x)	do { } while (0)
1792 #endif
1793 
1794 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1795 			 int sfcount, __be32 *psfsrc, int delta)
1796 {
1797 	struct ip_mc_list *pmc;
1798 	int	changerec = 0;
1799 	int	i, err;
1800 
1801 	if (!in_dev)
1802 		return -ENODEV;
1803 	rcu_read_lock();
1804 	for_each_pmc_rcu(in_dev, pmc) {
1805 		if (*pmca == pmc->multiaddr)
1806 			break;
1807 	}
1808 	if (!pmc) {
1809 		/* MCA not found?? bug */
1810 		rcu_read_unlock();
1811 		return -ESRCH;
1812 	}
1813 	spin_lock_bh(&pmc->lock);
1814 	rcu_read_unlock();
1815 #ifdef CONFIG_IP_MULTICAST
1816 	sf_markstate(pmc);
1817 #endif
1818 	if (!delta) {
1819 		err = -EINVAL;
1820 		if (!pmc->sfcount[sfmode])
1821 			goto out_unlock;
1822 		pmc->sfcount[sfmode]--;
1823 	}
1824 	err = 0;
1825 	for (i = 0; i < sfcount; i++) {
1826 		int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1827 
1828 		changerec |= rv > 0;
1829 		if (!err && rv < 0)
1830 			err = rv;
1831 	}
1832 	if (pmc->sfmode == MCAST_EXCLUDE &&
1833 	    pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1834 	    pmc->sfcount[MCAST_INCLUDE]) {
1835 #ifdef CONFIG_IP_MULTICAST
1836 		struct ip_sf_list *psf;
1837 		struct net *net = dev_net(in_dev->dev);
1838 #endif
1839 
1840 		/* filter mode change */
1841 		pmc->sfmode = MCAST_INCLUDE;
1842 #ifdef CONFIG_IP_MULTICAST
1843 		pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1844 		in_dev->mr_ifc_count = pmc->crcount;
1845 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1846 			psf->sf_crcount = 0;
1847 		igmp_ifc_event(pmc->interface);
1848 	} else if (sf_setstate(pmc) || changerec) {
1849 		igmp_ifc_event(pmc->interface);
1850 #endif
1851 	}
1852 out_unlock:
1853 	spin_unlock_bh(&pmc->lock);
1854 	return err;
1855 }
1856 
1857 /*
1858  * Add multicast single-source filter to the interface list
1859  */
1860 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1861 	__be32 *psfsrc)
1862 {
1863 	struct ip_sf_list *psf, *psf_prev;
1864 
1865 	psf_prev = NULL;
1866 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1867 		if (psf->sf_inaddr == *psfsrc)
1868 			break;
1869 		psf_prev = psf;
1870 	}
1871 	if (!psf) {
1872 		psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1873 		if (!psf)
1874 			return -ENOBUFS;
1875 		psf->sf_inaddr = *psfsrc;
1876 		if (psf_prev) {
1877 			psf_prev->sf_next = psf;
1878 		} else
1879 			pmc->sources = psf;
1880 	}
1881 	psf->sf_count[sfmode]++;
1882 	if (psf->sf_count[sfmode] == 1) {
1883 		ip_rt_multicast_event(pmc->interface);
1884 	}
1885 	return 0;
1886 }
1887 
1888 #ifdef CONFIG_IP_MULTICAST
1889 static void sf_markstate(struct ip_mc_list *pmc)
1890 {
1891 	struct ip_sf_list *psf;
1892 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1893 
1894 	for (psf = pmc->sources; psf; psf = psf->sf_next)
1895 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1896 			psf->sf_oldin = mca_xcount ==
1897 				psf->sf_count[MCAST_EXCLUDE] &&
1898 				!psf->sf_count[MCAST_INCLUDE];
1899 		} else
1900 			psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1901 }
1902 
1903 static int sf_setstate(struct ip_mc_list *pmc)
1904 {
1905 	struct ip_sf_list *psf, *dpsf;
1906 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1907 	int qrv = pmc->interface->mr_qrv;
1908 	int new_in, rv;
1909 
1910 	rv = 0;
1911 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1912 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1913 			new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1914 				!psf->sf_count[MCAST_INCLUDE];
1915 		} else
1916 			new_in = psf->sf_count[MCAST_INCLUDE] != 0;
1917 		if (new_in) {
1918 			if (!psf->sf_oldin) {
1919 				struct ip_sf_list *prev = NULL;
1920 
1921 				for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
1922 					if (dpsf->sf_inaddr == psf->sf_inaddr)
1923 						break;
1924 					prev = dpsf;
1925 				}
1926 				if (dpsf) {
1927 					if (prev)
1928 						prev->sf_next = dpsf->sf_next;
1929 					else
1930 						pmc->tomb = dpsf->sf_next;
1931 					kfree(dpsf);
1932 				}
1933 				psf->sf_crcount = qrv;
1934 				rv++;
1935 			}
1936 		} else if (psf->sf_oldin) {
1937 
1938 			psf->sf_crcount = 0;
1939 			/*
1940 			 * add or update "delete" records if an active filter
1941 			 * is now inactive
1942 			 */
1943 			for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
1944 				if (dpsf->sf_inaddr == psf->sf_inaddr)
1945 					break;
1946 			if (!dpsf) {
1947 				dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
1948 				if (!dpsf)
1949 					continue;
1950 				*dpsf = *psf;
1951 				/* pmc->lock held by callers */
1952 				dpsf->sf_next = pmc->tomb;
1953 				pmc->tomb = dpsf;
1954 			}
1955 			dpsf->sf_crcount = qrv;
1956 			rv++;
1957 		}
1958 	}
1959 	return rv;
1960 }
1961 #endif
1962 
1963 /*
1964  * Add multicast source filter list to the interface list
1965  */
1966 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1967 			 int sfcount, __be32 *psfsrc, int delta)
1968 {
1969 	struct ip_mc_list *pmc;
1970 	int	isexclude;
1971 	int	i, err;
1972 
1973 	if (!in_dev)
1974 		return -ENODEV;
1975 	rcu_read_lock();
1976 	for_each_pmc_rcu(in_dev, pmc) {
1977 		if (*pmca == pmc->multiaddr)
1978 			break;
1979 	}
1980 	if (!pmc) {
1981 		/* MCA not found?? bug */
1982 		rcu_read_unlock();
1983 		return -ESRCH;
1984 	}
1985 	spin_lock_bh(&pmc->lock);
1986 	rcu_read_unlock();
1987 
1988 #ifdef CONFIG_IP_MULTICAST
1989 	sf_markstate(pmc);
1990 #endif
1991 	isexclude = pmc->sfmode == MCAST_EXCLUDE;
1992 	if (!delta)
1993 		pmc->sfcount[sfmode]++;
1994 	err = 0;
1995 	for (i = 0; i < sfcount; i++) {
1996 		err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
1997 		if (err)
1998 			break;
1999 	}
2000 	if (err) {
2001 		int j;
2002 
2003 		if (!delta)
2004 			pmc->sfcount[sfmode]--;
2005 		for (j = 0; j < i; j++)
2006 			(void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2007 	} else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2008 #ifdef CONFIG_IP_MULTICAST
2009 		struct ip_sf_list *psf;
2010 		struct net *net = dev_net(pmc->interface->dev);
2011 		in_dev = pmc->interface;
2012 #endif
2013 
2014 		/* filter mode change */
2015 		if (pmc->sfcount[MCAST_EXCLUDE])
2016 			pmc->sfmode = MCAST_EXCLUDE;
2017 		else if (pmc->sfcount[MCAST_INCLUDE])
2018 			pmc->sfmode = MCAST_INCLUDE;
2019 #ifdef CONFIG_IP_MULTICAST
2020 		/* else no filters; keep old mode for reports */
2021 
2022 		pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2023 		in_dev->mr_ifc_count = pmc->crcount;
2024 		for (psf = pmc->sources; psf; psf = psf->sf_next)
2025 			psf->sf_crcount = 0;
2026 		igmp_ifc_event(in_dev);
2027 	} else if (sf_setstate(pmc)) {
2028 		igmp_ifc_event(in_dev);
2029 #endif
2030 	}
2031 	spin_unlock_bh(&pmc->lock);
2032 	return err;
2033 }
2034 
2035 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2036 {
2037 	struct ip_sf_list *psf, *nextpsf;
2038 
2039 	for (psf = pmc->tomb; psf; psf = nextpsf) {
2040 		nextpsf = psf->sf_next;
2041 		kfree(psf);
2042 	}
2043 	pmc->tomb = NULL;
2044 	for (psf = pmc->sources; psf; psf = nextpsf) {
2045 		nextpsf = psf->sf_next;
2046 		kfree(psf);
2047 	}
2048 	pmc->sources = NULL;
2049 	pmc->sfmode = MCAST_EXCLUDE;
2050 	pmc->sfcount[MCAST_INCLUDE] = 0;
2051 	pmc->sfcount[MCAST_EXCLUDE] = 1;
2052 }
2053 
2054 /* Join a multicast group
2055  */
2056 
2057 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2058 {
2059 	__be32 addr = imr->imr_multiaddr.s_addr;
2060 	struct ip_mc_socklist *iml, *i;
2061 	struct in_device *in_dev;
2062 	struct inet_sock *inet = inet_sk(sk);
2063 	struct net *net = sock_net(sk);
2064 	int ifindex;
2065 	int count = 0;
2066 	int err;
2067 
2068 	ASSERT_RTNL();
2069 
2070 	if (!ipv4_is_multicast(addr))
2071 		return -EINVAL;
2072 
2073 	in_dev = ip_mc_find_dev(net, imr);
2074 
2075 	if (!in_dev) {
2076 		err = -ENODEV;
2077 		goto done;
2078 	}
2079 
2080 	err = -EADDRINUSE;
2081 	ifindex = imr->imr_ifindex;
2082 	for_each_pmc_rtnl(inet, i) {
2083 		if (i->multi.imr_multiaddr.s_addr == addr &&
2084 		    i->multi.imr_ifindex == ifindex)
2085 			goto done;
2086 		count++;
2087 	}
2088 	err = -ENOBUFS;
2089 	if (count >= net->ipv4.sysctl_igmp_max_memberships)
2090 		goto done;
2091 	iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2092 	if (!iml)
2093 		goto done;
2094 
2095 	memcpy(&iml->multi, imr, sizeof(*imr));
2096 	iml->next_rcu = inet->mc_list;
2097 	iml->sflist = NULL;
2098 	iml->sfmode = MCAST_EXCLUDE;
2099 	rcu_assign_pointer(inet->mc_list, iml);
2100 	ip_mc_inc_group(in_dev, addr);
2101 	err = 0;
2102 done:
2103 	return err;
2104 }
2105 EXPORT_SYMBOL(ip_mc_join_group);
2106 
2107 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2108 			   struct in_device *in_dev)
2109 {
2110 	struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2111 	int err;
2112 
2113 	if (!psf) {
2114 		/* any-source empty exclude case */
2115 		return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2116 			iml->sfmode, 0, NULL, 0);
2117 	}
2118 	err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2119 			iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2120 	RCU_INIT_POINTER(iml->sflist, NULL);
2121 	/* decrease mem now to avoid the memleak warning */
2122 	atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2123 	kfree_rcu(psf, rcu);
2124 	return err;
2125 }
2126 
2127 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2128 {
2129 	struct inet_sock *inet = inet_sk(sk);
2130 	struct ip_mc_socklist *iml;
2131 	struct ip_mc_socklist __rcu **imlp;
2132 	struct in_device *in_dev;
2133 	struct net *net = sock_net(sk);
2134 	__be32 group = imr->imr_multiaddr.s_addr;
2135 	u32 ifindex;
2136 	int ret = -EADDRNOTAVAIL;
2137 
2138 	ASSERT_RTNL();
2139 
2140 	in_dev = ip_mc_find_dev(net, imr);
2141 	if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2142 		ret = -ENODEV;
2143 		goto out;
2144 	}
2145 	ifindex = imr->imr_ifindex;
2146 	for (imlp = &inet->mc_list;
2147 	     (iml = rtnl_dereference(*imlp)) != NULL;
2148 	     imlp = &iml->next_rcu) {
2149 		if (iml->multi.imr_multiaddr.s_addr != group)
2150 			continue;
2151 		if (ifindex) {
2152 			if (iml->multi.imr_ifindex != ifindex)
2153 				continue;
2154 		} else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2155 				iml->multi.imr_address.s_addr)
2156 			continue;
2157 
2158 		(void) ip_mc_leave_src(sk, iml, in_dev);
2159 
2160 		*imlp = iml->next_rcu;
2161 
2162 		if (in_dev)
2163 			ip_mc_dec_group(in_dev, group);
2164 
2165 		/* decrease mem now to avoid the memleak warning */
2166 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2167 		kfree_rcu(iml, rcu);
2168 		return 0;
2169 	}
2170 out:
2171 	return ret;
2172 }
2173 EXPORT_SYMBOL(ip_mc_leave_group);
2174 
2175 int ip_mc_source(int add, int omode, struct sock *sk, struct
2176 	ip_mreq_source *mreqs, int ifindex)
2177 {
2178 	int err;
2179 	struct ip_mreqn imr;
2180 	__be32 addr = mreqs->imr_multiaddr;
2181 	struct ip_mc_socklist *pmc;
2182 	struct in_device *in_dev = NULL;
2183 	struct inet_sock *inet = inet_sk(sk);
2184 	struct ip_sf_socklist *psl;
2185 	struct net *net = sock_net(sk);
2186 	int leavegroup = 0;
2187 	int i, j, rv;
2188 
2189 	if (!ipv4_is_multicast(addr))
2190 		return -EINVAL;
2191 
2192 	ASSERT_RTNL();
2193 
2194 	imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2195 	imr.imr_address.s_addr = mreqs->imr_interface;
2196 	imr.imr_ifindex = ifindex;
2197 	in_dev = ip_mc_find_dev(net, &imr);
2198 
2199 	if (!in_dev) {
2200 		err = -ENODEV;
2201 		goto done;
2202 	}
2203 	err = -EADDRNOTAVAIL;
2204 
2205 	for_each_pmc_rtnl(inet, pmc) {
2206 		if ((pmc->multi.imr_multiaddr.s_addr ==
2207 		     imr.imr_multiaddr.s_addr) &&
2208 		    (pmc->multi.imr_ifindex == imr.imr_ifindex))
2209 			break;
2210 	}
2211 	if (!pmc) {		/* must have a prior join */
2212 		err = -EINVAL;
2213 		goto done;
2214 	}
2215 	/* if a source filter was set, must be the same mode as before */
2216 	if (pmc->sflist) {
2217 		if (pmc->sfmode != omode) {
2218 			err = -EINVAL;
2219 			goto done;
2220 		}
2221 	} else if (pmc->sfmode != omode) {
2222 		/* allow mode switches for empty-set filters */
2223 		ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2224 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2225 			NULL, 0);
2226 		pmc->sfmode = omode;
2227 	}
2228 
2229 	psl = rtnl_dereference(pmc->sflist);
2230 	if (!add) {
2231 		if (!psl)
2232 			goto done;	/* err = -EADDRNOTAVAIL */
2233 		rv = !0;
2234 		for (i = 0; i < psl->sl_count; i++) {
2235 			rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2236 				sizeof(__be32));
2237 			if (rv == 0)
2238 				break;
2239 		}
2240 		if (rv)		/* source not found */
2241 			goto done;	/* err = -EADDRNOTAVAIL */
2242 
2243 		/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2244 		if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2245 			leavegroup = 1;
2246 			goto done;
2247 		}
2248 
2249 		/* update the interface filter */
2250 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2251 			&mreqs->imr_sourceaddr, 1);
2252 
2253 		for (j = i+1; j < psl->sl_count; j++)
2254 			psl->sl_addr[j-1] = psl->sl_addr[j];
2255 		psl->sl_count--;
2256 		err = 0;
2257 		goto done;
2258 	}
2259 	/* else, add a new source to the filter */
2260 
2261 	if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2262 		err = -ENOBUFS;
2263 		goto done;
2264 	}
2265 	if (!psl || psl->sl_count == psl->sl_max) {
2266 		struct ip_sf_socklist *newpsl;
2267 		int count = IP_SFBLOCK;
2268 
2269 		if (psl)
2270 			count += psl->sl_max;
2271 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2272 		if (!newpsl) {
2273 			err = -ENOBUFS;
2274 			goto done;
2275 		}
2276 		newpsl->sl_max = count;
2277 		newpsl->sl_count = count - IP_SFBLOCK;
2278 		if (psl) {
2279 			for (i = 0; i < psl->sl_count; i++)
2280 				newpsl->sl_addr[i] = psl->sl_addr[i];
2281 			/* decrease mem now to avoid the memleak warning */
2282 			atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2283 			kfree_rcu(psl, rcu);
2284 		}
2285 		rcu_assign_pointer(pmc->sflist, newpsl);
2286 		psl = newpsl;
2287 	}
2288 	rv = 1;	/* > 0 for insert logic below if sl_count is 0 */
2289 	for (i = 0; i < psl->sl_count; i++) {
2290 		rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2291 			sizeof(__be32));
2292 		if (rv == 0)
2293 			break;
2294 	}
2295 	if (rv == 0)		/* address already there is an error */
2296 		goto done;
2297 	for (j = psl->sl_count-1; j >= i; j--)
2298 		psl->sl_addr[j+1] = psl->sl_addr[j];
2299 	psl->sl_addr[i] = mreqs->imr_sourceaddr;
2300 	psl->sl_count++;
2301 	err = 0;
2302 	/* update the interface list */
2303 	ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2304 		&mreqs->imr_sourceaddr, 1);
2305 done:
2306 	if (leavegroup)
2307 		err = ip_mc_leave_group(sk, &imr);
2308 	return err;
2309 }
2310 
2311 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2312 {
2313 	int err = 0;
2314 	struct ip_mreqn	imr;
2315 	__be32 addr = msf->imsf_multiaddr;
2316 	struct ip_mc_socklist *pmc;
2317 	struct in_device *in_dev;
2318 	struct inet_sock *inet = inet_sk(sk);
2319 	struct ip_sf_socklist *newpsl, *psl;
2320 	struct net *net = sock_net(sk);
2321 	int leavegroup = 0;
2322 
2323 	if (!ipv4_is_multicast(addr))
2324 		return -EINVAL;
2325 	if (msf->imsf_fmode != MCAST_INCLUDE &&
2326 	    msf->imsf_fmode != MCAST_EXCLUDE)
2327 		return -EINVAL;
2328 
2329 	ASSERT_RTNL();
2330 
2331 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2332 	imr.imr_address.s_addr = msf->imsf_interface;
2333 	imr.imr_ifindex = ifindex;
2334 	in_dev = ip_mc_find_dev(net, &imr);
2335 
2336 	if (!in_dev) {
2337 		err = -ENODEV;
2338 		goto done;
2339 	}
2340 
2341 	/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2342 	if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2343 		leavegroup = 1;
2344 		goto done;
2345 	}
2346 
2347 	for_each_pmc_rtnl(inet, pmc) {
2348 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2349 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2350 			break;
2351 	}
2352 	if (!pmc) {		/* must have a prior join */
2353 		err = -EINVAL;
2354 		goto done;
2355 	}
2356 	if (msf->imsf_numsrc) {
2357 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2358 							   GFP_KERNEL);
2359 		if (!newpsl) {
2360 			err = -ENOBUFS;
2361 			goto done;
2362 		}
2363 		newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2364 		memcpy(newpsl->sl_addr, msf->imsf_slist,
2365 			msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2366 		err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2367 			msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2368 		if (err) {
2369 			sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2370 			goto done;
2371 		}
2372 	} else {
2373 		newpsl = NULL;
2374 		(void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2375 				     msf->imsf_fmode, 0, NULL, 0);
2376 	}
2377 	psl = rtnl_dereference(pmc->sflist);
2378 	if (psl) {
2379 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2380 			psl->sl_count, psl->sl_addr, 0);
2381 		/* decrease mem now to avoid the memleak warning */
2382 		atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2383 		kfree_rcu(psl, rcu);
2384 	} else
2385 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2386 			0, NULL, 0);
2387 	rcu_assign_pointer(pmc->sflist, newpsl);
2388 	pmc->sfmode = msf->imsf_fmode;
2389 	err = 0;
2390 done:
2391 	if (leavegroup)
2392 		err = ip_mc_leave_group(sk, &imr);
2393 	return err;
2394 }
2395 
2396 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2397 	struct ip_msfilter __user *optval, int __user *optlen)
2398 {
2399 	int err, len, count, copycount;
2400 	struct ip_mreqn	imr;
2401 	__be32 addr = msf->imsf_multiaddr;
2402 	struct ip_mc_socklist *pmc;
2403 	struct in_device *in_dev;
2404 	struct inet_sock *inet = inet_sk(sk);
2405 	struct ip_sf_socklist *psl;
2406 	struct net *net = sock_net(sk);
2407 
2408 	ASSERT_RTNL();
2409 
2410 	if (!ipv4_is_multicast(addr))
2411 		return -EINVAL;
2412 
2413 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2414 	imr.imr_address.s_addr = msf->imsf_interface;
2415 	imr.imr_ifindex = 0;
2416 	in_dev = ip_mc_find_dev(net, &imr);
2417 
2418 	if (!in_dev) {
2419 		err = -ENODEV;
2420 		goto done;
2421 	}
2422 	err = -EADDRNOTAVAIL;
2423 
2424 	for_each_pmc_rtnl(inet, pmc) {
2425 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2426 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2427 			break;
2428 	}
2429 	if (!pmc)		/* must have a prior join */
2430 		goto done;
2431 	msf->imsf_fmode = pmc->sfmode;
2432 	psl = rtnl_dereference(pmc->sflist);
2433 	if (!psl) {
2434 		len = 0;
2435 		count = 0;
2436 	} else {
2437 		count = psl->sl_count;
2438 	}
2439 	copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2440 	len = copycount * sizeof(psl->sl_addr[0]);
2441 	msf->imsf_numsrc = count;
2442 	if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2443 	    copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2444 		return -EFAULT;
2445 	}
2446 	if (len &&
2447 	    copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2448 		return -EFAULT;
2449 	return 0;
2450 done:
2451 	return err;
2452 }
2453 
2454 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2455 	struct group_filter __user *optval, int __user *optlen)
2456 {
2457 	int err, i, count, copycount;
2458 	struct sockaddr_in *psin;
2459 	__be32 addr;
2460 	struct ip_mc_socklist *pmc;
2461 	struct inet_sock *inet = inet_sk(sk);
2462 	struct ip_sf_socklist *psl;
2463 
2464 	ASSERT_RTNL();
2465 
2466 	psin = (struct sockaddr_in *)&gsf->gf_group;
2467 	if (psin->sin_family != AF_INET)
2468 		return -EINVAL;
2469 	addr = psin->sin_addr.s_addr;
2470 	if (!ipv4_is_multicast(addr))
2471 		return -EINVAL;
2472 
2473 	err = -EADDRNOTAVAIL;
2474 
2475 	for_each_pmc_rtnl(inet, pmc) {
2476 		if (pmc->multi.imr_multiaddr.s_addr == addr &&
2477 		    pmc->multi.imr_ifindex == gsf->gf_interface)
2478 			break;
2479 	}
2480 	if (!pmc)		/* must have a prior join */
2481 		goto done;
2482 	gsf->gf_fmode = pmc->sfmode;
2483 	psl = rtnl_dereference(pmc->sflist);
2484 	count = psl ? psl->sl_count : 0;
2485 	copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2486 	gsf->gf_numsrc = count;
2487 	if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2488 	    copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2489 		return -EFAULT;
2490 	}
2491 	for (i = 0; i < copycount; i++) {
2492 		struct sockaddr_storage ss;
2493 
2494 		psin = (struct sockaddr_in *)&ss;
2495 		memset(&ss, 0, sizeof(ss));
2496 		psin->sin_family = AF_INET;
2497 		psin->sin_addr.s_addr = psl->sl_addr[i];
2498 		if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2499 			return -EFAULT;
2500 	}
2501 	return 0;
2502 done:
2503 	return err;
2504 }
2505 
2506 /*
2507  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2508  */
2509 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr, int dif)
2510 {
2511 	struct inet_sock *inet = inet_sk(sk);
2512 	struct ip_mc_socklist *pmc;
2513 	struct ip_sf_socklist *psl;
2514 	int i;
2515 	int ret;
2516 
2517 	ret = 1;
2518 	if (!ipv4_is_multicast(loc_addr))
2519 		goto out;
2520 
2521 	rcu_read_lock();
2522 	for_each_pmc_rcu(inet, pmc) {
2523 		if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2524 		    pmc->multi.imr_ifindex == dif)
2525 			break;
2526 	}
2527 	ret = inet->mc_all;
2528 	if (!pmc)
2529 		goto unlock;
2530 	psl = rcu_dereference(pmc->sflist);
2531 	ret = (pmc->sfmode == MCAST_EXCLUDE);
2532 	if (!psl)
2533 		goto unlock;
2534 
2535 	for (i = 0; i < psl->sl_count; i++) {
2536 		if (psl->sl_addr[i] == rmt_addr)
2537 			break;
2538 	}
2539 	ret = 0;
2540 	if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2541 		goto unlock;
2542 	if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2543 		goto unlock;
2544 	ret = 1;
2545 unlock:
2546 	rcu_read_unlock();
2547 out:
2548 	return ret;
2549 }
2550 
2551 /*
2552  *	A socket is closing.
2553  */
2554 
2555 void ip_mc_drop_socket(struct sock *sk)
2556 {
2557 	struct inet_sock *inet = inet_sk(sk);
2558 	struct ip_mc_socklist *iml;
2559 	struct net *net = sock_net(sk);
2560 
2561 	if (!inet->mc_list)
2562 		return;
2563 
2564 	rtnl_lock();
2565 	while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2566 		struct in_device *in_dev;
2567 
2568 		inet->mc_list = iml->next_rcu;
2569 		in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2570 		(void) ip_mc_leave_src(sk, iml, in_dev);
2571 		if (in_dev)
2572 			ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2573 		/* decrease mem now to avoid the memleak warning */
2574 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2575 		kfree_rcu(iml, rcu);
2576 	}
2577 	rtnl_unlock();
2578 }
2579 
2580 /* called with rcu_read_lock() */
2581 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2582 {
2583 	struct ip_mc_list *im;
2584 	struct ip_mc_list __rcu **mc_hash;
2585 	struct ip_sf_list *psf;
2586 	int rv = 0;
2587 
2588 	mc_hash = rcu_dereference(in_dev->mc_hash);
2589 	if (mc_hash) {
2590 		u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2591 
2592 		for (im = rcu_dereference(mc_hash[hash]);
2593 		     im != NULL;
2594 		     im = rcu_dereference(im->next_hash)) {
2595 			if (im->multiaddr == mc_addr)
2596 				break;
2597 		}
2598 	} else {
2599 		for_each_pmc_rcu(in_dev, im) {
2600 			if (im->multiaddr == mc_addr)
2601 				break;
2602 		}
2603 	}
2604 	if (im && proto == IPPROTO_IGMP) {
2605 		rv = 1;
2606 	} else if (im) {
2607 		if (src_addr) {
2608 			for (psf = im->sources; psf; psf = psf->sf_next) {
2609 				if (psf->sf_inaddr == src_addr)
2610 					break;
2611 			}
2612 			if (psf)
2613 				rv = psf->sf_count[MCAST_INCLUDE] ||
2614 					psf->sf_count[MCAST_EXCLUDE] !=
2615 					im->sfcount[MCAST_EXCLUDE];
2616 			else
2617 				rv = im->sfcount[MCAST_EXCLUDE] != 0;
2618 		} else
2619 			rv = 1; /* unspecified source; tentatively allow */
2620 	}
2621 	return rv;
2622 }
2623 
2624 #if defined(CONFIG_PROC_FS)
2625 struct igmp_mc_iter_state {
2626 	struct seq_net_private p;
2627 	struct net_device *dev;
2628 	struct in_device *in_dev;
2629 };
2630 
2631 #define	igmp_mc_seq_private(seq)	((struct igmp_mc_iter_state *)(seq)->private)
2632 
2633 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2634 {
2635 	struct net *net = seq_file_net(seq);
2636 	struct ip_mc_list *im = NULL;
2637 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2638 
2639 	state->in_dev = NULL;
2640 	for_each_netdev_rcu(net, state->dev) {
2641 		struct in_device *in_dev;
2642 
2643 		in_dev = __in_dev_get_rcu(state->dev);
2644 		if (!in_dev)
2645 			continue;
2646 		im = rcu_dereference(in_dev->mc_list);
2647 		if (im) {
2648 			state->in_dev = in_dev;
2649 			break;
2650 		}
2651 	}
2652 	return im;
2653 }
2654 
2655 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2656 {
2657 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2658 
2659 	im = rcu_dereference(im->next_rcu);
2660 	while (!im) {
2661 		state->dev = next_net_device_rcu(state->dev);
2662 		if (!state->dev) {
2663 			state->in_dev = NULL;
2664 			break;
2665 		}
2666 		state->in_dev = __in_dev_get_rcu(state->dev);
2667 		if (!state->in_dev)
2668 			continue;
2669 		im = rcu_dereference(state->in_dev->mc_list);
2670 	}
2671 	return im;
2672 }
2673 
2674 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2675 {
2676 	struct ip_mc_list *im = igmp_mc_get_first(seq);
2677 	if (im)
2678 		while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2679 			--pos;
2680 	return pos ? NULL : im;
2681 }
2682 
2683 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2684 	__acquires(rcu)
2685 {
2686 	rcu_read_lock();
2687 	return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2688 }
2689 
2690 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2691 {
2692 	struct ip_mc_list *im;
2693 	if (v == SEQ_START_TOKEN)
2694 		im = igmp_mc_get_first(seq);
2695 	else
2696 		im = igmp_mc_get_next(seq, v);
2697 	++*pos;
2698 	return im;
2699 }
2700 
2701 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2702 	__releases(rcu)
2703 {
2704 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2705 
2706 	state->in_dev = NULL;
2707 	state->dev = NULL;
2708 	rcu_read_unlock();
2709 }
2710 
2711 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2712 {
2713 	if (v == SEQ_START_TOKEN)
2714 		seq_puts(seq,
2715 			 "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2716 	else {
2717 		struct ip_mc_list *im = (struct ip_mc_list *)v;
2718 		struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2719 		char   *querier;
2720 		long delta;
2721 
2722 #ifdef CONFIG_IP_MULTICAST
2723 		querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2724 			  IGMP_V2_SEEN(state->in_dev) ? "V2" :
2725 			  "V3";
2726 #else
2727 		querier = "NONE";
2728 #endif
2729 
2730 		if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2731 			seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2732 				   state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2733 		}
2734 
2735 		delta = im->timer.expires - jiffies;
2736 		seq_printf(seq,
2737 			   "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2738 			   im->multiaddr, im->users,
2739 			   im->tm_running,
2740 			   im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2741 			   im->reporter);
2742 	}
2743 	return 0;
2744 }
2745 
2746 static const struct seq_operations igmp_mc_seq_ops = {
2747 	.start	=	igmp_mc_seq_start,
2748 	.next	=	igmp_mc_seq_next,
2749 	.stop	=	igmp_mc_seq_stop,
2750 	.show	=	igmp_mc_seq_show,
2751 };
2752 
2753 static int igmp_mc_seq_open(struct inode *inode, struct file *file)
2754 {
2755 	return seq_open_net(inode, file, &igmp_mc_seq_ops,
2756 			sizeof(struct igmp_mc_iter_state));
2757 }
2758 
2759 static const struct file_operations igmp_mc_seq_fops = {
2760 	.owner		=	THIS_MODULE,
2761 	.open		=	igmp_mc_seq_open,
2762 	.read		=	seq_read,
2763 	.llseek		=	seq_lseek,
2764 	.release	=	seq_release_net,
2765 };
2766 
2767 struct igmp_mcf_iter_state {
2768 	struct seq_net_private p;
2769 	struct net_device *dev;
2770 	struct in_device *idev;
2771 	struct ip_mc_list *im;
2772 };
2773 
2774 #define igmp_mcf_seq_private(seq)	((struct igmp_mcf_iter_state *)(seq)->private)
2775 
2776 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2777 {
2778 	struct net *net = seq_file_net(seq);
2779 	struct ip_sf_list *psf = NULL;
2780 	struct ip_mc_list *im = NULL;
2781 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2782 
2783 	state->idev = NULL;
2784 	state->im = NULL;
2785 	for_each_netdev_rcu(net, state->dev) {
2786 		struct in_device *idev;
2787 		idev = __in_dev_get_rcu(state->dev);
2788 		if (unlikely(!idev))
2789 			continue;
2790 		im = rcu_dereference(idev->mc_list);
2791 		if (likely(im)) {
2792 			spin_lock_bh(&im->lock);
2793 			psf = im->sources;
2794 			if (likely(psf)) {
2795 				state->im = im;
2796 				state->idev = idev;
2797 				break;
2798 			}
2799 			spin_unlock_bh(&im->lock);
2800 		}
2801 	}
2802 	return psf;
2803 }
2804 
2805 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2806 {
2807 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2808 
2809 	psf = psf->sf_next;
2810 	while (!psf) {
2811 		spin_unlock_bh(&state->im->lock);
2812 		state->im = state->im->next;
2813 		while (!state->im) {
2814 			state->dev = next_net_device_rcu(state->dev);
2815 			if (!state->dev) {
2816 				state->idev = NULL;
2817 				goto out;
2818 			}
2819 			state->idev = __in_dev_get_rcu(state->dev);
2820 			if (!state->idev)
2821 				continue;
2822 			state->im = rcu_dereference(state->idev->mc_list);
2823 		}
2824 		if (!state->im)
2825 			break;
2826 		spin_lock_bh(&state->im->lock);
2827 		psf = state->im->sources;
2828 	}
2829 out:
2830 	return psf;
2831 }
2832 
2833 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2834 {
2835 	struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2836 	if (psf)
2837 		while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2838 			--pos;
2839 	return pos ? NULL : psf;
2840 }
2841 
2842 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2843 	__acquires(rcu)
2844 {
2845 	rcu_read_lock();
2846 	return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2847 }
2848 
2849 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2850 {
2851 	struct ip_sf_list *psf;
2852 	if (v == SEQ_START_TOKEN)
2853 		psf = igmp_mcf_get_first(seq);
2854 	else
2855 		psf = igmp_mcf_get_next(seq, v);
2856 	++*pos;
2857 	return psf;
2858 }
2859 
2860 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2861 	__releases(rcu)
2862 {
2863 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2864 	if (likely(state->im)) {
2865 		spin_unlock_bh(&state->im->lock);
2866 		state->im = NULL;
2867 	}
2868 	state->idev = NULL;
2869 	state->dev = NULL;
2870 	rcu_read_unlock();
2871 }
2872 
2873 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2874 {
2875 	struct ip_sf_list *psf = (struct ip_sf_list *)v;
2876 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2877 
2878 	if (v == SEQ_START_TOKEN) {
2879 		seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2880 	} else {
2881 		seq_printf(seq,
2882 			   "%3d %6.6s 0x%08x "
2883 			   "0x%08x %6lu %6lu\n",
2884 			   state->dev->ifindex, state->dev->name,
2885 			   ntohl(state->im->multiaddr),
2886 			   ntohl(psf->sf_inaddr),
2887 			   psf->sf_count[MCAST_INCLUDE],
2888 			   psf->sf_count[MCAST_EXCLUDE]);
2889 	}
2890 	return 0;
2891 }
2892 
2893 static const struct seq_operations igmp_mcf_seq_ops = {
2894 	.start	=	igmp_mcf_seq_start,
2895 	.next	=	igmp_mcf_seq_next,
2896 	.stop	=	igmp_mcf_seq_stop,
2897 	.show	=	igmp_mcf_seq_show,
2898 };
2899 
2900 static int igmp_mcf_seq_open(struct inode *inode, struct file *file)
2901 {
2902 	return seq_open_net(inode, file, &igmp_mcf_seq_ops,
2903 			sizeof(struct igmp_mcf_iter_state));
2904 }
2905 
2906 static const struct file_operations igmp_mcf_seq_fops = {
2907 	.owner		=	THIS_MODULE,
2908 	.open		=	igmp_mcf_seq_open,
2909 	.read		=	seq_read,
2910 	.llseek		=	seq_lseek,
2911 	.release	=	seq_release_net,
2912 };
2913 
2914 static int __net_init igmp_net_init(struct net *net)
2915 {
2916 	struct proc_dir_entry *pde;
2917 	int err;
2918 
2919 	pde = proc_create("igmp", S_IRUGO, net->proc_net, &igmp_mc_seq_fops);
2920 	if (!pde)
2921 		goto out_igmp;
2922 	pde = proc_create("mcfilter", S_IRUGO, net->proc_net,
2923 			  &igmp_mcf_seq_fops);
2924 	if (!pde)
2925 		goto out_mcfilter;
2926 	err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
2927 				   SOCK_DGRAM, 0, net);
2928 	if (err < 0) {
2929 		pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
2930 		       err);
2931 		goto out_sock;
2932 	}
2933 
2934 	/* Sysctl initialization */
2935 	net->ipv4.sysctl_igmp_max_memberships = 20;
2936 	net->ipv4.sysctl_igmp_max_msf = 10;
2937 	/* IGMP reports for link-local multicast groups are enabled by default */
2938 	net->ipv4.sysctl_igmp_llm_reports = 1;
2939 	net->ipv4.sysctl_igmp_qrv = 2;
2940 	return 0;
2941 
2942 out_sock:
2943 	remove_proc_entry("mcfilter", net->proc_net);
2944 out_mcfilter:
2945 	remove_proc_entry("igmp", net->proc_net);
2946 out_igmp:
2947 	return -ENOMEM;
2948 }
2949 
2950 static void __net_exit igmp_net_exit(struct net *net)
2951 {
2952 	remove_proc_entry("mcfilter", net->proc_net);
2953 	remove_proc_entry("igmp", net->proc_net);
2954 	inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
2955 }
2956 
2957 static struct pernet_operations igmp_net_ops = {
2958 	.init = igmp_net_init,
2959 	.exit = igmp_net_exit,
2960 };
2961 #endif
2962 
2963 static int igmp_netdev_event(struct notifier_block *this,
2964 			     unsigned long event, void *ptr)
2965 {
2966 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2967 	struct in_device *in_dev;
2968 
2969 	switch (event) {
2970 	case NETDEV_RESEND_IGMP:
2971 		in_dev = __in_dev_get_rtnl(dev);
2972 		if (in_dev)
2973 			ip_mc_rejoin_groups(in_dev);
2974 		break;
2975 	default:
2976 		break;
2977 	}
2978 	return NOTIFY_DONE;
2979 }
2980 
2981 static struct notifier_block igmp_notifier = {
2982 	.notifier_call = igmp_netdev_event,
2983 };
2984 
2985 int __init igmp_mc_init(void)
2986 {
2987 #if defined(CONFIG_PROC_FS)
2988 	int err;
2989 
2990 	err = register_pernet_subsys(&igmp_net_ops);
2991 	if (err)
2992 		return err;
2993 	err = register_netdevice_notifier(&igmp_notifier);
2994 	if (err)
2995 		goto reg_notif_fail;
2996 	return 0;
2997 
2998 reg_notif_fail:
2999 	unregister_pernet_subsys(&igmp_net_ops);
3000 	return err;
3001 #else
3002 	return register_netdevice_notifier(&igmp_notifier);
3003 #endif
3004 }
3005