xref: /openbmc/linux/net/ipv4/igmp.c (revision e6dec923)
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 <linux/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, struct ip_mc_list *im);
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 (refcount_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 		refcount_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 		refcount_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 	unsigned long exp = jiffies + tv + 2;
223 
224 	if (in_dev->mr_gq_running &&
225 	    time_after_eq(exp, (in_dev->mr_gq_timer).expires))
226 		return;
227 
228 	in_dev->mr_gq_running = 1;
229 	if (!mod_timer(&in_dev->mr_gq_timer, exp))
230 		in_dev_hold(in_dev);
231 }
232 
233 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
234 {
235 	int tv = prandom_u32() % delay;
236 
237 	if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
238 		in_dev_hold(in_dev);
239 }
240 
241 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
242 {
243 	spin_lock_bh(&im->lock);
244 	im->unsolicit_count = 0;
245 	if (del_timer(&im->timer)) {
246 		if ((long)(im->timer.expires-jiffies) < max_delay) {
247 			add_timer(&im->timer);
248 			im->tm_running = 1;
249 			spin_unlock_bh(&im->lock);
250 			return;
251 		}
252 		refcount_dec(&im->refcnt);
253 	}
254 	igmp_start_timer(im, max_delay);
255 	spin_unlock_bh(&im->lock);
256 }
257 
258 
259 /*
260  *	Send an IGMP report.
261  */
262 
263 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
264 
265 
266 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
267 	int gdeleted, int sdeleted)
268 {
269 	switch (type) {
270 	case IGMPV3_MODE_IS_INCLUDE:
271 	case IGMPV3_MODE_IS_EXCLUDE:
272 		if (gdeleted || sdeleted)
273 			return 0;
274 		if (!(pmc->gsquery && !psf->sf_gsresp)) {
275 			if (pmc->sfmode == MCAST_INCLUDE)
276 				return 1;
277 			/* don't include if this source is excluded
278 			 * in all filters
279 			 */
280 			if (psf->sf_count[MCAST_INCLUDE])
281 				return type == IGMPV3_MODE_IS_INCLUDE;
282 			return pmc->sfcount[MCAST_EXCLUDE] ==
283 				psf->sf_count[MCAST_EXCLUDE];
284 		}
285 		return 0;
286 	case IGMPV3_CHANGE_TO_INCLUDE:
287 		if (gdeleted || sdeleted)
288 			return 0;
289 		return psf->sf_count[MCAST_INCLUDE] != 0;
290 	case IGMPV3_CHANGE_TO_EXCLUDE:
291 		if (gdeleted || sdeleted)
292 			return 0;
293 		if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
294 		    psf->sf_count[MCAST_INCLUDE])
295 			return 0;
296 		return pmc->sfcount[MCAST_EXCLUDE] ==
297 			psf->sf_count[MCAST_EXCLUDE];
298 	case IGMPV3_ALLOW_NEW_SOURCES:
299 		if (gdeleted || !psf->sf_crcount)
300 			return 0;
301 		return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
302 	case IGMPV3_BLOCK_OLD_SOURCES:
303 		if (pmc->sfmode == MCAST_INCLUDE)
304 			return gdeleted || (psf->sf_crcount && sdeleted);
305 		return psf->sf_crcount && !gdeleted && !sdeleted;
306 	}
307 	return 0;
308 }
309 
310 static int
311 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
312 {
313 	struct ip_sf_list *psf;
314 	int scount = 0;
315 
316 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
317 		if (!is_in(pmc, psf, type, gdeleted, sdeleted))
318 			continue;
319 		scount++;
320 	}
321 	return scount;
322 }
323 
324 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
325 {
326 	struct sk_buff *skb;
327 	struct rtable *rt;
328 	struct iphdr *pip;
329 	struct igmpv3_report *pig;
330 	struct net *net = dev_net(dev);
331 	struct flowi4 fl4;
332 	int hlen = LL_RESERVED_SPACE(dev);
333 	int tlen = dev->needed_tailroom;
334 	unsigned int size = mtu;
335 
336 	while (1) {
337 		skb = alloc_skb(size + hlen + tlen,
338 				GFP_ATOMIC | __GFP_NOWARN);
339 		if (skb)
340 			break;
341 		size >>= 1;
342 		if (size < 256)
343 			return NULL;
344 	}
345 	skb->priority = TC_PRIO_CONTROL;
346 
347 	rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
348 				   0, 0,
349 				   IPPROTO_IGMP, 0, dev->ifindex);
350 	if (IS_ERR(rt)) {
351 		kfree_skb(skb);
352 		return NULL;
353 	}
354 
355 	skb_dst_set(skb, &rt->dst);
356 	skb->dev = dev;
357 
358 	skb_reserve(skb, hlen);
359 	skb_tailroom_reserve(skb, mtu, tlen);
360 
361 	skb_reset_network_header(skb);
362 	pip = ip_hdr(skb);
363 	skb_put(skb, sizeof(struct iphdr) + 4);
364 
365 	pip->version  = 4;
366 	pip->ihl      = (sizeof(struct iphdr)+4)>>2;
367 	pip->tos      = 0xc0;
368 	pip->frag_off = htons(IP_DF);
369 	pip->ttl      = 1;
370 	pip->daddr    = fl4.daddr;
371 	pip->saddr    = fl4.saddr;
372 	pip->protocol = IPPROTO_IGMP;
373 	pip->tot_len  = 0;	/* filled in later */
374 	ip_select_ident(net, skb, NULL);
375 	((u8 *)&pip[1])[0] = IPOPT_RA;
376 	((u8 *)&pip[1])[1] = 4;
377 	((u8 *)&pip[1])[2] = 0;
378 	((u8 *)&pip[1])[3] = 0;
379 
380 	skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
381 	skb_put(skb, sizeof(*pig));
382 	pig = igmpv3_report_hdr(skb);
383 	pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
384 	pig->resv1 = 0;
385 	pig->csum = 0;
386 	pig->resv2 = 0;
387 	pig->ngrec = 0;
388 	return skb;
389 }
390 
391 static int igmpv3_sendpack(struct sk_buff *skb)
392 {
393 	struct igmphdr *pig = igmp_hdr(skb);
394 	const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
395 
396 	pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
397 
398 	return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
399 }
400 
401 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
402 {
403 	return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
404 }
405 
406 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
407 	int type, struct igmpv3_grec **ppgr)
408 {
409 	struct net_device *dev = pmc->interface->dev;
410 	struct igmpv3_report *pih;
411 	struct igmpv3_grec *pgr;
412 
413 	if (!skb)
414 		skb = igmpv3_newpack(dev, dev->mtu);
415 	if (!skb)
416 		return NULL;
417 	pgr = skb_put(skb, sizeof(struct igmpv3_grec));
418 	pgr->grec_type = type;
419 	pgr->grec_auxwords = 0;
420 	pgr->grec_nsrcs = 0;
421 	pgr->grec_mca = pmc->multiaddr;
422 	pih = igmpv3_report_hdr(skb);
423 	pih->ngrec = htons(ntohs(pih->ngrec)+1);
424 	*ppgr = pgr;
425 	return skb;
426 }
427 
428 #define AVAILABLE(skb)	((skb) ? skb_availroom(skb) : 0)
429 
430 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
431 	int type, int gdeleted, int sdeleted)
432 {
433 	struct net_device *dev = pmc->interface->dev;
434 	struct net *net = dev_net(dev);
435 	struct igmpv3_report *pih;
436 	struct igmpv3_grec *pgr = NULL;
437 	struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
438 	int scount, stotal, first, isquery, truncate;
439 
440 	if (pmc->multiaddr == IGMP_ALL_HOSTS)
441 		return skb;
442 	if (ipv4_is_local_multicast(pmc->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
443 		return skb;
444 
445 	isquery = type == IGMPV3_MODE_IS_INCLUDE ||
446 		  type == IGMPV3_MODE_IS_EXCLUDE;
447 	truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
448 		    type == IGMPV3_CHANGE_TO_EXCLUDE;
449 
450 	stotal = scount = 0;
451 
452 	psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
453 
454 	if (!*psf_list)
455 		goto empty_source;
456 
457 	pih = skb ? igmpv3_report_hdr(skb) : NULL;
458 
459 	/* EX and TO_EX get a fresh packet, if needed */
460 	if (truncate) {
461 		if (pih && pih->ngrec &&
462 		    AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
463 			if (skb)
464 				igmpv3_sendpack(skb);
465 			skb = igmpv3_newpack(dev, dev->mtu);
466 		}
467 	}
468 	first = 1;
469 	psf_prev = NULL;
470 	for (psf = *psf_list; psf; psf = psf_next) {
471 		__be32 *psrc;
472 
473 		psf_next = psf->sf_next;
474 
475 		if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
476 			psf_prev = psf;
477 			continue;
478 		}
479 
480 		/* Based on RFC3376 5.1. Should not send source-list change
481 		 * records when there is a filter mode change.
482 		 */
483 		if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
484 		     (!gdeleted && pmc->crcount)) &&
485 		    (type == IGMPV3_ALLOW_NEW_SOURCES ||
486 		     type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
487 			goto decrease_sf_crcount;
488 
489 		/* clear marks on query responses */
490 		if (isquery)
491 			psf->sf_gsresp = 0;
492 
493 		if (AVAILABLE(skb) < sizeof(__be32) +
494 		    first*sizeof(struct igmpv3_grec)) {
495 			if (truncate && !first)
496 				break;	 /* truncate these */
497 			if (pgr)
498 				pgr->grec_nsrcs = htons(scount);
499 			if (skb)
500 				igmpv3_sendpack(skb);
501 			skb = igmpv3_newpack(dev, dev->mtu);
502 			first = 1;
503 			scount = 0;
504 		}
505 		if (first) {
506 			skb = add_grhead(skb, pmc, type, &pgr);
507 			first = 0;
508 		}
509 		if (!skb)
510 			return NULL;
511 		psrc = skb_put(skb, sizeof(__be32));
512 		*psrc = psf->sf_inaddr;
513 		scount++; stotal++;
514 		if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
515 		     type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
516 decrease_sf_crcount:
517 			psf->sf_crcount--;
518 			if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
519 				if (psf_prev)
520 					psf_prev->sf_next = psf->sf_next;
521 				else
522 					*psf_list = psf->sf_next;
523 				kfree(psf);
524 				continue;
525 			}
526 		}
527 		psf_prev = psf;
528 	}
529 
530 empty_source:
531 	if (!stotal) {
532 		if (type == IGMPV3_ALLOW_NEW_SOURCES ||
533 		    type == IGMPV3_BLOCK_OLD_SOURCES)
534 			return skb;
535 		if (pmc->crcount || isquery) {
536 			/* make sure we have room for group header */
537 			if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
538 				igmpv3_sendpack(skb);
539 				skb = NULL; /* add_grhead will get a new one */
540 			}
541 			skb = add_grhead(skb, pmc, type, &pgr);
542 		}
543 	}
544 	if (pgr)
545 		pgr->grec_nsrcs = htons(scount);
546 
547 	if (isquery)
548 		pmc->gsquery = 0;	/* clear query state on report */
549 	return skb;
550 }
551 
552 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
553 {
554 	struct sk_buff *skb = NULL;
555 	struct net *net = dev_net(in_dev->dev);
556 	int type;
557 
558 	if (!pmc) {
559 		rcu_read_lock();
560 		for_each_pmc_rcu(in_dev, pmc) {
561 			if (pmc->multiaddr == IGMP_ALL_HOSTS)
562 				continue;
563 			if (ipv4_is_local_multicast(pmc->multiaddr) &&
564 			     !net->ipv4.sysctl_igmp_llm_reports)
565 				continue;
566 			spin_lock_bh(&pmc->lock);
567 			if (pmc->sfcount[MCAST_EXCLUDE])
568 				type = IGMPV3_MODE_IS_EXCLUDE;
569 			else
570 				type = IGMPV3_MODE_IS_INCLUDE;
571 			skb = add_grec(skb, pmc, type, 0, 0);
572 			spin_unlock_bh(&pmc->lock);
573 		}
574 		rcu_read_unlock();
575 	} else {
576 		spin_lock_bh(&pmc->lock);
577 		if (pmc->sfcount[MCAST_EXCLUDE])
578 			type = IGMPV3_MODE_IS_EXCLUDE;
579 		else
580 			type = IGMPV3_MODE_IS_INCLUDE;
581 		skb = add_grec(skb, pmc, type, 0, 0);
582 		spin_unlock_bh(&pmc->lock);
583 	}
584 	if (!skb)
585 		return 0;
586 	return igmpv3_sendpack(skb);
587 }
588 
589 /*
590  * remove zero-count source records from a source filter list
591  */
592 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
593 {
594 	struct ip_sf_list *psf_prev, *psf_next, *psf;
595 
596 	psf_prev = NULL;
597 	for (psf = *ppsf; psf; psf = psf_next) {
598 		psf_next = psf->sf_next;
599 		if (psf->sf_crcount == 0) {
600 			if (psf_prev)
601 				psf_prev->sf_next = psf->sf_next;
602 			else
603 				*ppsf = psf->sf_next;
604 			kfree(psf);
605 		} else
606 			psf_prev = psf;
607 	}
608 }
609 
610 static void igmpv3_send_cr(struct in_device *in_dev)
611 {
612 	struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
613 	struct sk_buff *skb = NULL;
614 	int type, dtype;
615 
616 	rcu_read_lock();
617 	spin_lock_bh(&in_dev->mc_tomb_lock);
618 
619 	/* deleted MCA's */
620 	pmc_prev = NULL;
621 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
622 		pmc_next = pmc->next;
623 		if (pmc->sfmode == MCAST_INCLUDE) {
624 			type = IGMPV3_BLOCK_OLD_SOURCES;
625 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
626 			skb = add_grec(skb, pmc, type, 1, 0);
627 			skb = add_grec(skb, pmc, dtype, 1, 1);
628 		}
629 		if (pmc->crcount) {
630 			if (pmc->sfmode == MCAST_EXCLUDE) {
631 				type = IGMPV3_CHANGE_TO_INCLUDE;
632 				skb = add_grec(skb, pmc, type, 1, 0);
633 			}
634 			pmc->crcount--;
635 			if (pmc->crcount == 0) {
636 				igmpv3_clear_zeros(&pmc->tomb);
637 				igmpv3_clear_zeros(&pmc->sources);
638 			}
639 		}
640 		if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
641 			if (pmc_prev)
642 				pmc_prev->next = pmc_next;
643 			else
644 				in_dev->mc_tomb = pmc_next;
645 			in_dev_put(pmc->interface);
646 			kfree(pmc);
647 		} else
648 			pmc_prev = pmc;
649 	}
650 	spin_unlock_bh(&in_dev->mc_tomb_lock);
651 
652 	/* change recs */
653 	for_each_pmc_rcu(in_dev, pmc) {
654 		spin_lock_bh(&pmc->lock);
655 		if (pmc->sfcount[MCAST_EXCLUDE]) {
656 			type = IGMPV3_BLOCK_OLD_SOURCES;
657 			dtype = IGMPV3_ALLOW_NEW_SOURCES;
658 		} else {
659 			type = IGMPV3_ALLOW_NEW_SOURCES;
660 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
661 		}
662 		skb = add_grec(skb, pmc, type, 0, 0);
663 		skb = add_grec(skb, pmc, dtype, 0, 1);	/* deleted sources */
664 
665 		/* filter mode changes */
666 		if (pmc->crcount) {
667 			if (pmc->sfmode == MCAST_EXCLUDE)
668 				type = IGMPV3_CHANGE_TO_EXCLUDE;
669 			else
670 				type = IGMPV3_CHANGE_TO_INCLUDE;
671 			skb = add_grec(skb, pmc, type, 0, 0);
672 			pmc->crcount--;
673 		}
674 		spin_unlock_bh(&pmc->lock);
675 	}
676 	rcu_read_unlock();
677 
678 	if (!skb)
679 		return;
680 	(void) igmpv3_sendpack(skb);
681 }
682 
683 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
684 	int type)
685 {
686 	struct sk_buff *skb;
687 	struct iphdr *iph;
688 	struct igmphdr *ih;
689 	struct rtable *rt;
690 	struct net_device *dev = in_dev->dev;
691 	struct net *net = dev_net(dev);
692 	__be32	group = pmc ? pmc->multiaddr : 0;
693 	struct flowi4 fl4;
694 	__be32	dst;
695 	int hlen, tlen;
696 
697 	if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
698 		return igmpv3_send_report(in_dev, pmc);
699 
700 	if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
701 		return 0;
702 
703 	if (type == IGMP_HOST_LEAVE_MESSAGE)
704 		dst = IGMP_ALL_ROUTER;
705 	else
706 		dst = group;
707 
708 	rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
709 				   0, 0,
710 				   IPPROTO_IGMP, 0, dev->ifindex);
711 	if (IS_ERR(rt))
712 		return -1;
713 
714 	hlen = LL_RESERVED_SPACE(dev);
715 	tlen = dev->needed_tailroom;
716 	skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
717 	if (!skb) {
718 		ip_rt_put(rt);
719 		return -1;
720 	}
721 	skb->priority = TC_PRIO_CONTROL;
722 
723 	skb_dst_set(skb, &rt->dst);
724 
725 	skb_reserve(skb, hlen);
726 
727 	skb_reset_network_header(skb);
728 	iph = ip_hdr(skb);
729 	skb_put(skb, sizeof(struct iphdr) + 4);
730 
731 	iph->version  = 4;
732 	iph->ihl      = (sizeof(struct iphdr)+4)>>2;
733 	iph->tos      = 0xc0;
734 	iph->frag_off = htons(IP_DF);
735 	iph->ttl      = 1;
736 	iph->daddr    = dst;
737 	iph->saddr    = fl4.saddr;
738 	iph->protocol = IPPROTO_IGMP;
739 	ip_select_ident(net, skb, NULL);
740 	((u8 *)&iph[1])[0] = IPOPT_RA;
741 	((u8 *)&iph[1])[1] = 4;
742 	((u8 *)&iph[1])[2] = 0;
743 	((u8 *)&iph[1])[3] = 0;
744 
745 	ih = skb_put(skb, sizeof(struct igmphdr));
746 	ih->type = type;
747 	ih->code = 0;
748 	ih->csum = 0;
749 	ih->group = group;
750 	ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
751 
752 	return ip_local_out(net, skb->sk, skb);
753 }
754 
755 static void igmp_gq_timer_expire(unsigned long data)
756 {
757 	struct in_device *in_dev = (struct in_device *)data;
758 
759 	in_dev->mr_gq_running = 0;
760 	igmpv3_send_report(in_dev, NULL);
761 	in_dev_put(in_dev);
762 }
763 
764 static void igmp_ifc_timer_expire(unsigned long data)
765 {
766 	struct in_device *in_dev = (struct in_device *)data;
767 
768 	igmpv3_send_cr(in_dev);
769 	if (in_dev->mr_ifc_count) {
770 		in_dev->mr_ifc_count--;
771 		igmp_ifc_start_timer(in_dev,
772 				     unsolicited_report_interval(in_dev));
773 	}
774 	in_dev_put(in_dev);
775 }
776 
777 static void igmp_ifc_event(struct in_device *in_dev)
778 {
779 	struct net *net = dev_net(in_dev->dev);
780 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
781 		return;
782 	in_dev->mr_ifc_count = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
783 	igmp_ifc_start_timer(in_dev, 1);
784 }
785 
786 
787 static void igmp_timer_expire(unsigned long data)
788 {
789 	struct ip_mc_list *im = (struct ip_mc_list *)data;
790 	struct in_device *in_dev = im->interface;
791 
792 	spin_lock(&im->lock);
793 	im->tm_running = 0;
794 
795 	if (im->unsolicit_count) {
796 		im->unsolicit_count--;
797 		igmp_start_timer(im, unsolicited_report_interval(in_dev));
798 	}
799 	im->reporter = 1;
800 	spin_unlock(&im->lock);
801 
802 	if (IGMP_V1_SEEN(in_dev))
803 		igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
804 	else if (IGMP_V2_SEEN(in_dev))
805 		igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
806 	else
807 		igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
808 
809 	ip_ma_put(im);
810 }
811 
812 /* mark EXCLUDE-mode sources */
813 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
814 {
815 	struct ip_sf_list *psf;
816 	int i, scount;
817 
818 	scount = 0;
819 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
820 		if (scount == nsrcs)
821 			break;
822 		for (i = 0; i < nsrcs; i++) {
823 			/* skip inactive filters */
824 			if (psf->sf_count[MCAST_INCLUDE] ||
825 			    pmc->sfcount[MCAST_EXCLUDE] !=
826 			    psf->sf_count[MCAST_EXCLUDE])
827 				break;
828 			if (srcs[i] == psf->sf_inaddr) {
829 				scount++;
830 				break;
831 			}
832 		}
833 	}
834 	pmc->gsquery = 0;
835 	if (scount == nsrcs)	/* all sources excluded */
836 		return 0;
837 	return 1;
838 }
839 
840 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
841 {
842 	struct ip_sf_list *psf;
843 	int i, scount;
844 
845 	if (pmc->sfmode == MCAST_EXCLUDE)
846 		return igmp_xmarksources(pmc, nsrcs, srcs);
847 
848 	/* mark INCLUDE-mode sources */
849 	scount = 0;
850 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
851 		if (scount == nsrcs)
852 			break;
853 		for (i = 0; i < nsrcs; i++)
854 			if (srcs[i] == psf->sf_inaddr) {
855 				psf->sf_gsresp = 1;
856 				scount++;
857 				break;
858 			}
859 	}
860 	if (!scount) {
861 		pmc->gsquery = 0;
862 		return 0;
863 	}
864 	pmc->gsquery = 1;
865 	return 1;
866 }
867 
868 /* return true if packet was dropped */
869 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
870 {
871 	struct ip_mc_list *im;
872 	struct net *net = dev_net(in_dev->dev);
873 
874 	/* Timers are only set for non-local groups */
875 
876 	if (group == IGMP_ALL_HOSTS)
877 		return false;
878 	if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
879 		return false;
880 
881 	rcu_read_lock();
882 	for_each_pmc_rcu(in_dev, im) {
883 		if (im->multiaddr == group) {
884 			igmp_stop_timer(im);
885 			break;
886 		}
887 	}
888 	rcu_read_unlock();
889 	return false;
890 }
891 
892 /* return true if packet was dropped */
893 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
894 	int len)
895 {
896 	struct igmphdr 		*ih = igmp_hdr(skb);
897 	struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
898 	struct ip_mc_list	*im;
899 	__be32			group = ih->group;
900 	int			max_delay;
901 	int			mark = 0;
902 	struct net		*net = dev_net(in_dev->dev);
903 
904 
905 	if (len == 8) {
906 		if (ih->code == 0) {
907 			/* Alas, old v1 router presents here. */
908 
909 			max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
910 			in_dev->mr_v1_seen = jiffies +
911 				IGMP_V1_ROUTER_PRESENT_TIMEOUT;
912 			group = 0;
913 		} else {
914 			/* v2 router present */
915 			max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
916 			in_dev->mr_v2_seen = jiffies +
917 				IGMP_V2_ROUTER_PRESENT_TIMEOUT;
918 		}
919 		/* cancel the interface change timer */
920 		in_dev->mr_ifc_count = 0;
921 		if (del_timer(&in_dev->mr_ifc_timer))
922 			__in_dev_put(in_dev);
923 		/* clear deleted report items */
924 		igmpv3_clear_delrec(in_dev);
925 	} else if (len < 12) {
926 		return true;	/* ignore bogus packet; freed by caller */
927 	} else if (IGMP_V1_SEEN(in_dev)) {
928 		/* This is a v3 query with v1 queriers present */
929 		max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
930 		group = 0;
931 	} else if (IGMP_V2_SEEN(in_dev)) {
932 		/* this is a v3 query with v2 queriers present;
933 		 * Interpretation of the max_delay code is problematic here.
934 		 * A real v2 host would use ih_code directly, while v3 has a
935 		 * different encoding. We use the v3 encoding as more likely
936 		 * to be intended in a v3 query.
937 		 */
938 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
939 		if (!max_delay)
940 			max_delay = 1;	/* can't mod w/ 0 */
941 	} else { /* v3 */
942 		if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
943 			return true;
944 
945 		ih3 = igmpv3_query_hdr(skb);
946 		if (ih3->nsrcs) {
947 			if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
948 					   + ntohs(ih3->nsrcs)*sizeof(__be32)))
949 				return true;
950 			ih3 = igmpv3_query_hdr(skb);
951 		}
952 
953 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
954 		if (!max_delay)
955 			max_delay = 1;	/* can't mod w/ 0 */
956 		in_dev->mr_maxdelay = max_delay;
957 		if (ih3->qrv)
958 			in_dev->mr_qrv = ih3->qrv;
959 		if (!group) { /* general query */
960 			if (ih3->nsrcs)
961 				return true;	/* no sources allowed */
962 			igmp_gq_start_timer(in_dev);
963 			return false;
964 		}
965 		/* mark sources to include, if group & source-specific */
966 		mark = ih3->nsrcs != 0;
967 	}
968 
969 	/*
970 	 * - Start the timers in all of our membership records
971 	 *   that the query applies to for the interface on
972 	 *   which the query arrived excl. those that belong
973 	 *   to a "local" group (224.0.0.X)
974 	 * - For timers already running check if they need to
975 	 *   be reset.
976 	 * - Use the igmp->igmp_code field as the maximum
977 	 *   delay possible
978 	 */
979 	rcu_read_lock();
980 	for_each_pmc_rcu(in_dev, im) {
981 		int changed;
982 
983 		if (group && group != im->multiaddr)
984 			continue;
985 		if (im->multiaddr == IGMP_ALL_HOSTS)
986 			continue;
987 		if (ipv4_is_local_multicast(im->multiaddr) &&
988 		    !net->ipv4.sysctl_igmp_llm_reports)
989 			continue;
990 		spin_lock_bh(&im->lock);
991 		if (im->tm_running)
992 			im->gsquery = im->gsquery && mark;
993 		else
994 			im->gsquery = mark;
995 		changed = !im->gsquery ||
996 			igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
997 		spin_unlock_bh(&im->lock);
998 		if (changed)
999 			igmp_mod_timer(im, max_delay);
1000 	}
1001 	rcu_read_unlock();
1002 	return false;
1003 }
1004 
1005 /* called in rcu_read_lock() section */
1006 int igmp_rcv(struct sk_buff *skb)
1007 {
1008 	/* This basically follows the spec line by line -- see RFC1112 */
1009 	struct igmphdr *ih;
1010 	struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1011 	int len = skb->len;
1012 	bool dropped = true;
1013 
1014 	if (!in_dev)
1015 		goto drop;
1016 
1017 	if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1018 		goto drop;
1019 
1020 	if (skb_checksum_simple_validate(skb))
1021 		goto drop;
1022 
1023 	ih = igmp_hdr(skb);
1024 	switch (ih->type) {
1025 	case IGMP_HOST_MEMBERSHIP_QUERY:
1026 		dropped = igmp_heard_query(in_dev, skb, len);
1027 		break;
1028 	case IGMP_HOST_MEMBERSHIP_REPORT:
1029 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
1030 		/* Is it our report looped back? */
1031 		if (rt_is_output_route(skb_rtable(skb)))
1032 			break;
1033 		/* don't rely on MC router hearing unicast reports */
1034 		if (skb->pkt_type == PACKET_MULTICAST ||
1035 		    skb->pkt_type == PACKET_BROADCAST)
1036 			dropped = igmp_heard_report(in_dev, ih->group);
1037 		break;
1038 	case IGMP_PIM:
1039 #ifdef CONFIG_IP_PIMSM_V1
1040 		return pim_rcv_v1(skb);
1041 #endif
1042 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
1043 	case IGMP_DVMRP:
1044 	case IGMP_TRACE:
1045 	case IGMP_HOST_LEAVE_MESSAGE:
1046 	case IGMP_MTRACE:
1047 	case IGMP_MTRACE_RESP:
1048 		break;
1049 	default:
1050 		break;
1051 	}
1052 
1053 drop:
1054 	if (dropped)
1055 		kfree_skb(skb);
1056 	else
1057 		consume_skb(skb);
1058 	return 0;
1059 }
1060 
1061 #endif
1062 
1063 
1064 /*
1065  *	Add a filter to a device
1066  */
1067 
1068 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1069 {
1070 	char buf[MAX_ADDR_LEN];
1071 	struct net_device *dev = in_dev->dev;
1072 
1073 	/* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1074 	   We will get multicast token leakage, when IFF_MULTICAST
1075 	   is changed. This check should be done in ndo_set_rx_mode
1076 	   routine. Something sort of:
1077 	   if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1078 	   --ANK
1079 	   */
1080 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1081 		dev_mc_add(dev, buf);
1082 }
1083 
1084 /*
1085  *	Remove a filter from a device
1086  */
1087 
1088 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1089 {
1090 	char buf[MAX_ADDR_LEN];
1091 	struct net_device *dev = in_dev->dev;
1092 
1093 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1094 		dev_mc_del(dev, buf);
1095 }
1096 
1097 #ifdef CONFIG_IP_MULTICAST
1098 /*
1099  * deleted ip_mc_list manipulation
1100  */
1101 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1102 {
1103 	struct ip_mc_list *pmc;
1104 	struct net *net = dev_net(in_dev->dev);
1105 
1106 	/* this is an "ip_mc_list" for convenience; only the fields below
1107 	 * are actually used. In particular, the refcnt and users are not
1108 	 * used for management of the delete list. Using the same structure
1109 	 * for deleted items allows change reports to use common code with
1110 	 * non-deleted or query-response MCA's.
1111 	 */
1112 	pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1113 	if (!pmc)
1114 		return;
1115 	spin_lock_init(&pmc->lock);
1116 	spin_lock_bh(&im->lock);
1117 	pmc->interface = im->interface;
1118 	in_dev_hold(in_dev);
1119 	pmc->multiaddr = im->multiaddr;
1120 	pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1121 	pmc->sfmode = im->sfmode;
1122 	if (pmc->sfmode == MCAST_INCLUDE) {
1123 		struct ip_sf_list *psf;
1124 
1125 		pmc->tomb = im->tomb;
1126 		pmc->sources = im->sources;
1127 		im->tomb = im->sources = NULL;
1128 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1129 			psf->sf_crcount = pmc->crcount;
1130 	}
1131 	spin_unlock_bh(&im->lock);
1132 
1133 	spin_lock_bh(&in_dev->mc_tomb_lock);
1134 	pmc->next = in_dev->mc_tomb;
1135 	in_dev->mc_tomb = pmc;
1136 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1137 }
1138 
1139 /*
1140  * restore ip_mc_list deleted records
1141  */
1142 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1143 {
1144 	struct ip_mc_list *pmc, *pmc_prev;
1145 	struct ip_sf_list *psf;
1146 	struct net *net = dev_net(in_dev->dev);
1147 	__be32 multiaddr = im->multiaddr;
1148 
1149 	spin_lock_bh(&in_dev->mc_tomb_lock);
1150 	pmc_prev = NULL;
1151 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1152 		if (pmc->multiaddr == multiaddr)
1153 			break;
1154 		pmc_prev = pmc;
1155 	}
1156 	if (pmc) {
1157 		if (pmc_prev)
1158 			pmc_prev->next = pmc->next;
1159 		else
1160 			in_dev->mc_tomb = pmc->next;
1161 	}
1162 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1163 
1164 	spin_lock_bh(&im->lock);
1165 	if (pmc) {
1166 		im->interface = pmc->interface;
1167 		im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1168 		im->sfmode = pmc->sfmode;
1169 		if (pmc->sfmode == MCAST_INCLUDE) {
1170 			im->tomb = pmc->tomb;
1171 			im->sources = pmc->sources;
1172 			for (psf = im->sources; psf; psf = psf->sf_next)
1173 				psf->sf_crcount = im->crcount;
1174 		}
1175 		in_dev_put(pmc->interface);
1176 		kfree(pmc);
1177 	}
1178 	spin_unlock_bh(&im->lock);
1179 }
1180 
1181 /*
1182  * flush ip_mc_list deleted records
1183  */
1184 static void igmpv3_clear_delrec(struct in_device *in_dev)
1185 {
1186 	struct ip_mc_list *pmc, *nextpmc;
1187 
1188 	spin_lock_bh(&in_dev->mc_tomb_lock);
1189 	pmc = in_dev->mc_tomb;
1190 	in_dev->mc_tomb = NULL;
1191 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1192 
1193 	for (; pmc; pmc = nextpmc) {
1194 		nextpmc = pmc->next;
1195 		ip_mc_clear_src(pmc);
1196 		in_dev_put(pmc->interface);
1197 		kfree(pmc);
1198 	}
1199 	/* clear dead sources, too */
1200 	rcu_read_lock();
1201 	for_each_pmc_rcu(in_dev, pmc) {
1202 		struct ip_sf_list *psf, *psf_next;
1203 
1204 		spin_lock_bh(&pmc->lock);
1205 		psf = pmc->tomb;
1206 		pmc->tomb = NULL;
1207 		spin_unlock_bh(&pmc->lock);
1208 		for (; psf; psf = psf_next) {
1209 			psf_next = psf->sf_next;
1210 			kfree(psf);
1211 		}
1212 	}
1213 	rcu_read_unlock();
1214 }
1215 #endif
1216 
1217 static void igmp_group_dropped(struct ip_mc_list *im)
1218 {
1219 	struct in_device *in_dev = im->interface;
1220 #ifdef CONFIG_IP_MULTICAST
1221 	struct net *net = dev_net(in_dev->dev);
1222 	int reporter;
1223 #endif
1224 
1225 	if (im->loaded) {
1226 		im->loaded = 0;
1227 		ip_mc_filter_del(in_dev, im->multiaddr);
1228 	}
1229 
1230 #ifdef CONFIG_IP_MULTICAST
1231 	if (im->multiaddr == IGMP_ALL_HOSTS)
1232 		return;
1233 	if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1234 		return;
1235 
1236 	reporter = im->reporter;
1237 	igmp_stop_timer(im);
1238 
1239 	if (!in_dev->dead) {
1240 		if (IGMP_V1_SEEN(in_dev))
1241 			return;
1242 		if (IGMP_V2_SEEN(in_dev)) {
1243 			if (reporter)
1244 				igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1245 			return;
1246 		}
1247 		/* IGMPv3 */
1248 		igmpv3_add_delrec(in_dev, im);
1249 
1250 		igmp_ifc_event(in_dev);
1251 	}
1252 #endif
1253 }
1254 
1255 static void igmp_group_added(struct ip_mc_list *im)
1256 {
1257 	struct in_device *in_dev = im->interface;
1258 #ifdef CONFIG_IP_MULTICAST
1259 	struct net *net = dev_net(in_dev->dev);
1260 #endif
1261 
1262 	if (im->loaded == 0) {
1263 		im->loaded = 1;
1264 		ip_mc_filter_add(in_dev, im->multiaddr);
1265 	}
1266 
1267 #ifdef CONFIG_IP_MULTICAST
1268 	if (im->multiaddr == IGMP_ALL_HOSTS)
1269 		return;
1270 	if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1271 		return;
1272 
1273 	if (in_dev->dead)
1274 		return;
1275 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1276 		spin_lock_bh(&im->lock);
1277 		igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1278 		spin_unlock_bh(&im->lock);
1279 		return;
1280 	}
1281 	/* else, v3 */
1282 
1283 	im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1284 	igmp_ifc_event(in_dev);
1285 #endif
1286 }
1287 
1288 
1289 /*
1290  *	Multicast list managers
1291  */
1292 
1293 static u32 ip_mc_hash(const struct ip_mc_list *im)
1294 {
1295 	return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1296 }
1297 
1298 static void ip_mc_hash_add(struct in_device *in_dev,
1299 			   struct ip_mc_list *im)
1300 {
1301 	struct ip_mc_list __rcu **mc_hash;
1302 	u32 hash;
1303 
1304 	mc_hash = rtnl_dereference(in_dev->mc_hash);
1305 	if (mc_hash) {
1306 		hash = ip_mc_hash(im);
1307 		im->next_hash = mc_hash[hash];
1308 		rcu_assign_pointer(mc_hash[hash], im);
1309 		return;
1310 	}
1311 
1312 	/* do not use a hash table for small number of items */
1313 	if (in_dev->mc_count < 4)
1314 		return;
1315 
1316 	mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1317 			  GFP_KERNEL);
1318 	if (!mc_hash)
1319 		return;
1320 
1321 	for_each_pmc_rtnl(in_dev, im) {
1322 		hash = ip_mc_hash(im);
1323 		im->next_hash = mc_hash[hash];
1324 		RCU_INIT_POINTER(mc_hash[hash], im);
1325 	}
1326 
1327 	rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1328 }
1329 
1330 static void ip_mc_hash_remove(struct in_device *in_dev,
1331 			      struct ip_mc_list *im)
1332 {
1333 	struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1334 	struct ip_mc_list *aux;
1335 
1336 	if (!mc_hash)
1337 		return;
1338 	mc_hash += ip_mc_hash(im);
1339 	while ((aux = rtnl_dereference(*mc_hash)) != im)
1340 		mc_hash = &aux->next_hash;
1341 	*mc_hash = im->next_hash;
1342 }
1343 
1344 
1345 /*
1346  *	A socket has joined a multicast group on device dev.
1347  */
1348 
1349 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1350 {
1351 	struct ip_mc_list *im;
1352 #ifdef CONFIG_IP_MULTICAST
1353 	struct net *net = dev_net(in_dev->dev);
1354 #endif
1355 
1356 	ASSERT_RTNL();
1357 
1358 	for_each_pmc_rtnl(in_dev, im) {
1359 		if (im->multiaddr == addr) {
1360 			im->users++;
1361 			ip_mc_add_src(in_dev, &addr, MCAST_EXCLUDE, 0, NULL, 0);
1362 			goto out;
1363 		}
1364 	}
1365 
1366 	im = kzalloc(sizeof(*im), GFP_KERNEL);
1367 	if (!im)
1368 		goto out;
1369 
1370 	im->users = 1;
1371 	im->interface = in_dev;
1372 	in_dev_hold(in_dev);
1373 	im->multiaddr = addr;
1374 	/* initial mode is (EX, empty) */
1375 	im->sfmode = MCAST_EXCLUDE;
1376 	im->sfcount[MCAST_EXCLUDE] = 1;
1377 	refcount_set(&im->refcnt, 1);
1378 	spin_lock_init(&im->lock);
1379 #ifdef CONFIG_IP_MULTICAST
1380 	setup_timer(&im->timer, igmp_timer_expire, (unsigned long)im);
1381 	im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1382 #endif
1383 
1384 	im->next_rcu = in_dev->mc_list;
1385 	in_dev->mc_count++;
1386 	rcu_assign_pointer(in_dev->mc_list, im);
1387 
1388 	ip_mc_hash_add(in_dev, im);
1389 
1390 #ifdef CONFIG_IP_MULTICAST
1391 	igmpv3_del_delrec(in_dev, im);
1392 #endif
1393 	igmp_group_added(im);
1394 	if (!in_dev->dead)
1395 		ip_rt_multicast_event(in_dev);
1396 out:
1397 	return;
1398 }
1399 EXPORT_SYMBOL(ip_mc_inc_group);
1400 
1401 static int ip_mc_check_iphdr(struct sk_buff *skb)
1402 {
1403 	const struct iphdr *iph;
1404 	unsigned int len;
1405 	unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1406 
1407 	if (!pskb_may_pull(skb, offset))
1408 		return -EINVAL;
1409 
1410 	iph = ip_hdr(skb);
1411 
1412 	if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1413 		return -EINVAL;
1414 
1415 	offset += ip_hdrlen(skb) - sizeof(*iph);
1416 
1417 	if (!pskb_may_pull(skb, offset))
1418 		return -EINVAL;
1419 
1420 	iph = ip_hdr(skb);
1421 
1422 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1423 		return -EINVAL;
1424 
1425 	len = skb_network_offset(skb) + ntohs(iph->tot_len);
1426 	if (skb->len < len || len < offset)
1427 		return -EINVAL;
1428 
1429 	skb_set_transport_header(skb, offset);
1430 
1431 	return 0;
1432 }
1433 
1434 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1435 {
1436 	unsigned int len = skb_transport_offset(skb);
1437 
1438 	len += sizeof(struct igmpv3_report);
1439 
1440 	return pskb_may_pull(skb, len) ? 0 : -EINVAL;
1441 }
1442 
1443 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1444 {
1445 	unsigned int len = skb_transport_offset(skb);
1446 
1447 	len += sizeof(struct igmphdr);
1448 	if (skb->len < len)
1449 		return -EINVAL;
1450 
1451 	/* IGMPv{1,2}? */
1452 	if (skb->len != len) {
1453 		/* or IGMPv3? */
1454 		len += sizeof(struct igmpv3_query) - sizeof(struct igmphdr);
1455 		if (skb->len < len || !pskb_may_pull(skb, len))
1456 			return -EINVAL;
1457 	}
1458 
1459 	/* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1460 	 * all-systems destination addresses (224.0.0.1) for general queries
1461 	 */
1462 	if (!igmp_hdr(skb)->group &&
1463 	    ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1464 		return -EINVAL;
1465 
1466 	return 0;
1467 }
1468 
1469 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1470 {
1471 	switch (igmp_hdr(skb)->type) {
1472 	case IGMP_HOST_LEAVE_MESSAGE:
1473 	case IGMP_HOST_MEMBERSHIP_REPORT:
1474 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
1475 		/* fall through */
1476 		return 0;
1477 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
1478 		return ip_mc_check_igmp_reportv3(skb);
1479 	case IGMP_HOST_MEMBERSHIP_QUERY:
1480 		return ip_mc_check_igmp_query(skb);
1481 	default:
1482 		return -ENOMSG;
1483 	}
1484 }
1485 
1486 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1487 {
1488 	return skb_checksum_simple_validate(skb);
1489 }
1490 
1491 static int __ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1492 
1493 {
1494 	struct sk_buff *skb_chk;
1495 	unsigned int transport_len;
1496 	unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1497 	int ret = -EINVAL;
1498 
1499 	transport_len = ntohs(ip_hdr(skb)->tot_len) - ip_hdrlen(skb);
1500 
1501 	skb_chk = skb_checksum_trimmed(skb, transport_len,
1502 				       ip_mc_validate_checksum);
1503 	if (!skb_chk)
1504 		goto err;
1505 
1506 	if (!pskb_may_pull(skb_chk, len))
1507 		goto err;
1508 
1509 	ret = ip_mc_check_igmp_msg(skb_chk);
1510 	if (ret)
1511 		goto err;
1512 
1513 	if (skb_trimmed)
1514 		*skb_trimmed = skb_chk;
1515 	/* free now unneeded clone */
1516 	else if (skb_chk != skb)
1517 		kfree_skb(skb_chk);
1518 
1519 	ret = 0;
1520 
1521 err:
1522 	if (ret && skb_chk && skb_chk != skb)
1523 		kfree_skb(skb_chk);
1524 
1525 	return ret;
1526 }
1527 
1528 /**
1529  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1530  * @skb: the skb to validate
1531  * @skb_trimmed: to store an skb pointer trimmed to IPv4 packet tail (optional)
1532  *
1533  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1534  * skb transport header accordingly and returns zero.
1535  *
1536  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1537  *  standard
1538  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1539  * -ENOMEM: A memory allocation failure happened.
1540  *
1541  * Optionally, an skb pointer might be provided via skb_trimmed (or set it
1542  * to NULL): After parsing an IGMP packet successfully it will point to
1543  * an skb which has its tail aligned to the IP packet end. This might
1544  * either be the originally provided skb or a trimmed, cloned version if
1545  * the skb frame had data beyond the IP packet. A cloned skb allows us
1546  * to leave the original skb and its full frame unchanged (which might be
1547  * desirable for layer 2 frame jugglers).
1548  *
1549  * Caller needs to set the skb network header and free any returned skb if it
1550  * differs from the provided skb.
1551  */
1552 int ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1553 {
1554 	int ret = ip_mc_check_iphdr(skb);
1555 
1556 	if (ret < 0)
1557 		return ret;
1558 
1559 	if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1560 		return -ENOMSG;
1561 
1562 	return __ip_mc_check_igmp(skb, skb_trimmed);
1563 }
1564 EXPORT_SYMBOL(ip_mc_check_igmp);
1565 
1566 /*
1567  *	Resend IGMP JOIN report; used by netdev notifier.
1568  */
1569 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1570 {
1571 #ifdef CONFIG_IP_MULTICAST
1572 	struct ip_mc_list *im;
1573 	int type;
1574 	struct net *net = dev_net(in_dev->dev);
1575 
1576 	ASSERT_RTNL();
1577 
1578 	for_each_pmc_rtnl(in_dev, im) {
1579 		if (im->multiaddr == IGMP_ALL_HOSTS)
1580 			continue;
1581 		if (ipv4_is_local_multicast(im->multiaddr) &&
1582 		    !net->ipv4.sysctl_igmp_llm_reports)
1583 			continue;
1584 
1585 		/* a failover is happening and switches
1586 		 * must be notified immediately
1587 		 */
1588 		if (IGMP_V1_SEEN(in_dev))
1589 			type = IGMP_HOST_MEMBERSHIP_REPORT;
1590 		else if (IGMP_V2_SEEN(in_dev))
1591 			type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1592 		else
1593 			type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1594 		igmp_send_report(in_dev, im, type);
1595 	}
1596 #endif
1597 }
1598 
1599 /*
1600  *	A socket has left a multicast group on device dev
1601  */
1602 
1603 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1604 {
1605 	struct ip_mc_list *i;
1606 	struct ip_mc_list __rcu **ip;
1607 
1608 	ASSERT_RTNL();
1609 
1610 	for (ip = &in_dev->mc_list;
1611 	     (i = rtnl_dereference(*ip)) != NULL;
1612 	     ip = &i->next_rcu) {
1613 		if (i->multiaddr == addr) {
1614 			if (--i->users == 0) {
1615 				ip_mc_hash_remove(in_dev, i);
1616 				*ip = i->next_rcu;
1617 				in_dev->mc_count--;
1618 				igmp_group_dropped(i);
1619 				ip_mc_clear_src(i);
1620 
1621 				if (!in_dev->dead)
1622 					ip_rt_multicast_event(in_dev);
1623 
1624 				ip_ma_put(i);
1625 				return;
1626 			}
1627 			break;
1628 		}
1629 	}
1630 }
1631 EXPORT_SYMBOL(ip_mc_dec_group);
1632 
1633 /* Device changing type */
1634 
1635 void ip_mc_unmap(struct in_device *in_dev)
1636 {
1637 	struct ip_mc_list *pmc;
1638 
1639 	ASSERT_RTNL();
1640 
1641 	for_each_pmc_rtnl(in_dev, pmc)
1642 		igmp_group_dropped(pmc);
1643 }
1644 
1645 void ip_mc_remap(struct in_device *in_dev)
1646 {
1647 	struct ip_mc_list *pmc;
1648 
1649 	ASSERT_RTNL();
1650 
1651 	for_each_pmc_rtnl(in_dev, pmc) {
1652 #ifdef CONFIG_IP_MULTICAST
1653 		igmpv3_del_delrec(in_dev, pmc);
1654 #endif
1655 		igmp_group_added(pmc);
1656 	}
1657 }
1658 
1659 /* Device going down */
1660 
1661 void ip_mc_down(struct in_device *in_dev)
1662 {
1663 	struct ip_mc_list *pmc;
1664 
1665 	ASSERT_RTNL();
1666 
1667 	for_each_pmc_rtnl(in_dev, pmc)
1668 		igmp_group_dropped(pmc);
1669 
1670 #ifdef CONFIG_IP_MULTICAST
1671 	in_dev->mr_ifc_count = 0;
1672 	if (del_timer(&in_dev->mr_ifc_timer))
1673 		__in_dev_put(in_dev);
1674 	in_dev->mr_gq_running = 0;
1675 	if (del_timer(&in_dev->mr_gq_timer))
1676 		__in_dev_put(in_dev);
1677 #endif
1678 
1679 	ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1680 }
1681 
1682 void ip_mc_init_dev(struct in_device *in_dev)
1683 {
1684 #ifdef CONFIG_IP_MULTICAST
1685 	struct net *net = dev_net(in_dev->dev);
1686 #endif
1687 	ASSERT_RTNL();
1688 
1689 #ifdef CONFIG_IP_MULTICAST
1690 	setup_timer(&in_dev->mr_gq_timer, igmp_gq_timer_expire,
1691 			(unsigned long)in_dev);
1692 	setup_timer(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire,
1693 			(unsigned long)in_dev);
1694 	in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1695 #endif
1696 
1697 	spin_lock_init(&in_dev->mc_tomb_lock);
1698 }
1699 
1700 /* Device going up */
1701 
1702 void ip_mc_up(struct in_device *in_dev)
1703 {
1704 	struct ip_mc_list *pmc;
1705 #ifdef CONFIG_IP_MULTICAST
1706 	struct net *net = dev_net(in_dev->dev);
1707 #endif
1708 
1709 	ASSERT_RTNL();
1710 
1711 #ifdef CONFIG_IP_MULTICAST
1712 	in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1713 #endif
1714 	ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1715 
1716 	for_each_pmc_rtnl(in_dev, pmc) {
1717 #ifdef CONFIG_IP_MULTICAST
1718 		igmpv3_del_delrec(in_dev, pmc);
1719 #endif
1720 		igmp_group_added(pmc);
1721 	}
1722 }
1723 
1724 /*
1725  *	Device is about to be destroyed: clean up.
1726  */
1727 
1728 void ip_mc_destroy_dev(struct in_device *in_dev)
1729 {
1730 	struct ip_mc_list *i;
1731 
1732 	ASSERT_RTNL();
1733 
1734 	/* Deactivate timers */
1735 	ip_mc_down(in_dev);
1736 #ifdef CONFIG_IP_MULTICAST
1737 	igmpv3_clear_delrec(in_dev);
1738 #endif
1739 
1740 	while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1741 		in_dev->mc_list = i->next_rcu;
1742 		in_dev->mc_count--;
1743 		ip_ma_put(i);
1744 	}
1745 }
1746 
1747 /* RTNL is locked */
1748 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1749 {
1750 	struct net_device *dev = NULL;
1751 	struct in_device *idev = NULL;
1752 
1753 	if (imr->imr_ifindex) {
1754 		idev = inetdev_by_index(net, imr->imr_ifindex);
1755 		return idev;
1756 	}
1757 	if (imr->imr_address.s_addr) {
1758 		dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1759 		if (!dev)
1760 			return NULL;
1761 	}
1762 
1763 	if (!dev) {
1764 		struct rtable *rt = ip_route_output(net,
1765 						    imr->imr_multiaddr.s_addr,
1766 						    0, 0, 0);
1767 		if (!IS_ERR(rt)) {
1768 			dev = rt->dst.dev;
1769 			ip_rt_put(rt);
1770 		}
1771 	}
1772 	if (dev) {
1773 		imr->imr_ifindex = dev->ifindex;
1774 		idev = __in_dev_get_rtnl(dev);
1775 	}
1776 	return idev;
1777 }
1778 
1779 /*
1780  *	Join a socket to a group
1781  */
1782 
1783 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1784 	__be32 *psfsrc)
1785 {
1786 	struct ip_sf_list *psf, *psf_prev;
1787 	int rv = 0;
1788 
1789 	psf_prev = NULL;
1790 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1791 		if (psf->sf_inaddr == *psfsrc)
1792 			break;
1793 		psf_prev = psf;
1794 	}
1795 	if (!psf || psf->sf_count[sfmode] == 0) {
1796 		/* source filter not found, or count wrong =>  bug */
1797 		return -ESRCH;
1798 	}
1799 	psf->sf_count[sfmode]--;
1800 	if (psf->sf_count[sfmode] == 0) {
1801 		ip_rt_multicast_event(pmc->interface);
1802 	}
1803 	if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1804 #ifdef CONFIG_IP_MULTICAST
1805 		struct in_device *in_dev = pmc->interface;
1806 		struct net *net = dev_net(in_dev->dev);
1807 #endif
1808 
1809 		/* no more filters for this source */
1810 		if (psf_prev)
1811 			psf_prev->sf_next = psf->sf_next;
1812 		else
1813 			pmc->sources = psf->sf_next;
1814 #ifdef CONFIG_IP_MULTICAST
1815 		if (psf->sf_oldin &&
1816 		    !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1817 			psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1818 			psf->sf_next = pmc->tomb;
1819 			pmc->tomb = psf;
1820 			rv = 1;
1821 		} else
1822 #endif
1823 			kfree(psf);
1824 	}
1825 	return rv;
1826 }
1827 
1828 #ifndef CONFIG_IP_MULTICAST
1829 #define igmp_ifc_event(x)	do { } while (0)
1830 #endif
1831 
1832 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1833 			 int sfcount, __be32 *psfsrc, int delta)
1834 {
1835 	struct ip_mc_list *pmc;
1836 	int	changerec = 0;
1837 	int	i, err;
1838 
1839 	if (!in_dev)
1840 		return -ENODEV;
1841 	rcu_read_lock();
1842 	for_each_pmc_rcu(in_dev, pmc) {
1843 		if (*pmca == pmc->multiaddr)
1844 			break;
1845 	}
1846 	if (!pmc) {
1847 		/* MCA not found?? bug */
1848 		rcu_read_unlock();
1849 		return -ESRCH;
1850 	}
1851 	spin_lock_bh(&pmc->lock);
1852 	rcu_read_unlock();
1853 #ifdef CONFIG_IP_MULTICAST
1854 	sf_markstate(pmc);
1855 #endif
1856 	if (!delta) {
1857 		err = -EINVAL;
1858 		if (!pmc->sfcount[sfmode])
1859 			goto out_unlock;
1860 		pmc->sfcount[sfmode]--;
1861 	}
1862 	err = 0;
1863 	for (i = 0; i < sfcount; i++) {
1864 		int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1865 
1866 		changerec |= rv > 0;
1867 		if (!err && rv < 0)
1868 			err = rv;
1869 	}
1870 	if (pmc->sfmode == MCAST_EXCLUDE &&
1871 	    pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1872 	    pmc->sfcount[MCAST_INCLUDE]) {
1873 #ifdef CONFIG_IP_MULTICAST
1874 		struct ip_sf_list *psf;
1875 		struct net *net = dev_net(in_dev->dev);
1876 #endif
1877 
1878 		/* filter mode change */
1879 		pmc->sfmode = MCAST_INCLUDE;
1880 #ifdef CONFIG_IP_MULTICAST
1881 		pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1882 		in_dev->mr_ifc_count = pmc->crcount;
1883 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1884 			psf->sf_crcount = 0;
1885 		igmp_ifc_event(pmc->interface);
1886 	} else if (sf_setstate(pmc) || changerec) {
1887 		igmp_ifc_event(pmc->interface);
1888 #endif
1889 	}
1890 out_unlock:
1891 	spin_unlock_bh(&pmc->lock);
1892 	return err;
1893 }
1894 
1895 /*
1896  * Add multicast single-source filter to the interface list
1897  */
1898 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1899 	__be32 *psfsrc)
1900 {
1901 	struct ip_sf_list *psf, *psf_prev;
1902 
1903 	psf_prev = NULL;
1904 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1905 		if (psf->sf_inaddr == *psfsrc)
1906 			break;
1907 		psf_prev = psf;
1908 	}
1909 	if (!psf) {
1910 		psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1911 		if (!psf)
1912 			return -ENOBUFS;
1913 		psf->sf_inaddr = *psfsrc;
1914 		if (psf_prev) {
1915 			psf_prev->sf_next = psf;
1916 		} else
1917 			pmc->sources = psf;
1918 	}
1919 	psf->sf_count[sfmode]++;
1920 	if (psf->sf_count[sfmode] == 1) {
1921 		ip_rt_multicast_event(pmc->interface);
1922 	}
1923 	return 0;
1924 }
1925 
1926 #ifdef CONFIG_IP_MULTICAST
1927 static void sf_markstate(struct ip_mc_list *pmc)
1928 {
1929 	struct ip_sf_list *psf;
1930 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1931 
1932 	for (psf = pmc->sources; psf; psf = psf->sf_next)
1933 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1934 			psf->sf_oldin = mca_xcount ==
1935 				psf->sf_count[MCAST_EXCLUDE] &&
1936 				!psf->sf_count[MCAST_INCLUDE];
1937 		} else
1938 			psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1939 }
1940 
1941 static int sf_setstate(struct ip_mc_list *pmc)
1942 {
1943 	struct ip_sf_list *psf, *dpsf;
1944 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1945 	int qrv = pmc->interface->mr_qrv;
1946 	int new_in, rv;
1947 
1948 	rv = 0;
1949 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1950 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1951 			new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1952 				!psf->sf_count[MCAST_INCLUDE];
1953 		} else
1954 			new_in = psf->sf_count[MCAST_INCLUDE] != 0;
1955 		if (new_in) {
1956 			if (!psf->sf_oldin) {
1957 				struct ip_sf_list *prev = NULL;
1958 
1959 				for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
1960 					if (dpsf->sf_inaddr == psf->sf_inaddr)
1961 						break;
1962 					prev = dpsf;
1963 				}
1964 				if (dpsf) {
1965 					if (prev)
1966 						prev->sf_next = dpsf->sf_next;
1967 					else
1968 						pmc->tomb = dpsf->sf_next;
1969 					kfree(dpsf);
1970 				}
1971 				psf->sf_crcount = qrv;
1972 				rv++;
1973 			}
1974 		} else if (psf->sf_oldin) {
1975 
1976 			psf->sf_crcount = 0;
1977 			/*
1978 			 * add or update "delete" records if an active filter
1979 			 * is now inactive
1980 			 */
1981 			for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
1982 				if (dpsf->sf_inaddr == psf->sf_inaddr)
1983 					break;
1984 			if (!dpsf) {
1985 				dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
1986 				if (!dpsf)
1987 					continue;
1988 				*dpsf = *psf;
1989 				/* pmc->lock held by callers */
1990 				dpsf->sf_next = pmc->tomb;
1991 				pmc->tomb = dpsf;
1992 			}
1993 			dpsf->sf_crcount = qrv;
1994 			rv++;
1995 		}
1996 	}
1997 	return rv;
1998 }
1999 #endif
2000 
2001 /*
2002  * Add multicast source filter list to the interface list
2003  */
2004 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2005 			 int sfcount, __be32 *psfsrc, int delta)
2006 {
2007 	struct ip_mc_list *pmc;
2008 	int	isexclude;
2009 	int	i, err;
2010 
2011 	if (!in_dev)
2012 		return -ENODEV;
2013 	rcu_read_lock();
2014 	for_each_pmc_rcu(in_dev, pmc) {
2015 		if (*pmca == pmc->multiaddr)
2016 			break;
2017 	}
2018 	if (!pmc) {
2019 		/* MCA not found?? bug */
2020 		rcu_read_unlock();
2021 		return -ESRCH;
2022 	}
2023 	spin_lock_bh(&pmc->lock);
2024 	rcu_read_unlock();
2025 
2026 #ifdef CONFIG_IP_MULTICAST
2027 	sf_markstate(pmc);
2028 #endif
2029 	isexclude = pmc->sfmode == MCAST_EXCLUDE;
2030 	if (!delta)
2031 		pmc->sfcount[sfmode]++;
2032 	err = 0;
2033 	for (i = 0; i < sfcount; i++) {
2034 		err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2035 		if (err)
2036 			break;
2037 	}
2038 	if (err) {
2039 		int j;
2040 
2041 		if (!delta)
2042 			pmc->sfcount[sfmode]--;
2043 		for (j = 0; j < i; j++)
2044 			(void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2045 	} else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2046 #ifdef CONFIG_IP_MULTICAST
2047 		struct ip_sf_list *psf;
2048 		struct net *net = dev_net(pmc->interface->dev);
2049 		in_dev = pmc->interface;
2050 #endif
2051 
2052 		/* filter mode change */
2053 		if (pmc->sfcount[MCAST_EXCLUDE])
2054 			pmc->sfmode = MCAST_EXCLUDE;
2055 		else if (pmc->sfcount[MCAST_INCLUDE])
2056 			pmc->sfmode = MCAST_INCLUDE;
2057 #ifdef CONFIG_IP_MULTICAST
2058 		/* else no filters; keep old mode for reports */
2059 
2060 		pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2061 		in_dev->mr_ifc_count = pmc->crcount;
2062 		for (psf = pmc->sources; psf; psf = psf->sf_next)
2063 			psf->sf_crcount = 0;
2064 		igmp_ifc_event(in_dev);
2065 	} else if (sf_setstate(pmc)) {
2066 		igmp_ifc_event(in_dev);
2067 #endif
2068 	}
2069 	spin_unlock_bh(&pmc->lock);
2070 	return err;
2071 }
2072 
2073 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2074 {
2075 	struct ip_sf_list *psf, *nextpsf, *tomb, *sources;
2076 
2077 	spin_lock_bh(&pmc->lock);
2078 	tomb = pmc->tomb;
2079 	pmc->tomb = NULL;
2080 	sources = pmc->sources;
2081 	pmc->sources = NULL;
2082 	pmc->sfmode = MCAST_EXCLUDE;
2083 	pmc->sfcount[MCAST_INCLUDE] = 0;
2084 	pmc->sfcount[MCAST_EXCLUDE] = 1;
2085 	spin_unlock_bh(&pmc->lock);
2086 
2087 	for (psf = tomb; psf; psf = nextpsf) {
2088 		nextpsf = psf->sf_next;
2089 		kfree(psf);
2090 	}
2091 	for (psf = sources; psf; psf = nextpsf) {
2092 		nextpsf = psf->sf_next;
2093 		kfree(psf);
2094 	}
2095 }
2096 
2097 /* Join a multicast group
2098  */
2099 
2100 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2101 {
2102 	__be32 addr = imr->imr_multiaddr.s_addr;
2103 	struct ip_mc_socklist *iml, *i;
2104 	struct in_device *in_dev;
2105 	struct inet_sock *inet = inet_sk(sk);
2106 	struct net *net = sock_net(sk);
2107 	int ifindex;
2108 	int count = 0;
2109 	int err;
2110 
2111 	ASSERT_RTNL();
2112 
2113 	if (!ipv4_is_multicast(addr))
2114 		return -EINVAL;
2115 
2116 	in_dev = ip_mc_find_dev(net, imr);
2117 
2118 	if (!in_dev) {
2119 		err = -ENODEV;
2120 		goto done;
2121 	}
2122 
2123 	err = -EADDRINUSE;
2124 	ifindex = imr->imr_ifindex;
2125 	for_each_pmc_rtnl(inet, i) {
2126 		if (i->multi.imr_multiaddr.s_addr == addr &&
2127 		    i->multi.imr_ifindex == ifindex)
2128 			goto done;
2129 		count++;
2130 	}
2131 	err = -ENOBUFS;
2132 	if (count >= net->ipv4.sysctl_igmp_max_memberships)
2133 		goto done;
2134 	iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2135 	if (!iml)
2136 		goto done;
2137 
2138 	memcpy(&iml->multi, imr, sizeof(*imr));
2139 	iml->next_rcu = inet->mc_list;
2140 	iml->sflist = NULL;
2141 	iml->sfmode = MCAST_EXCLUDE;
2142 	rcu_assign_pointer(inet->mc_list, iml);
2143 	ip_mc_inc_group(in_dev, addr);
2144 	err = 0;
2145 done:
2146 	return err;
2147 }
2148 EXPORT_SYMBOL(ip_mc_join_group);
2149 
2150 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2151 			   struct in_device *in_dev)
2152 {
2153 	struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2154 	int err;
2155 
2156 	if (!psf) {
2157 		/* any-source empty exclude case */
2158 		return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2159 			iml->sfmode, 0, NULL, 0);
2160 	}
2161 	err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2162 			iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2163 	RCU_INIT_POINTER(iml->sflist, NULL);
2164 	/* decrease mem now to avoid the memleak warning */
2165 	atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2166 	kfree_rcu(psf, rcu);
2167 	return err;
2168 }
2169 
2170 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2171 {
2172 	struct inet_sock *inet = inet_sk(sk);
2173 	struct ip_mc_socklist *iml;
2174 	struct ip_mc_socklist __rcu **imlp;
2175 	struct in_device *in_dev;
2176 	struct net *net = sock_net(sk);
2177 	__be32 group = imr->imr_multiaddr.s_addr;
2178 	u32 ifindex;
2179 	int ret = -EADDRNOTAVAIL;
2180 
2181 	ASSERT_RTNL();
2182 
2183 	in_dev = ip_mc_find_dev(net, imr);
2184 	if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2185 		ret = -ENODEV;
2186 		goto out;
2187 	}
2188 	ifindex = imr->imr_ifindex;
2189 	for (imlp = &inet->mc_list;
2190 	     (iml = rtnl_dereference(*imlp)) != NULL;
2191 	     imlp = &iml->next_rcu) {
2192 		if (iml->multi.imr_multiaddr.s_addr != group)
2193 			continue;
2194 		if (ifindex) {
2195 			if (iml->multi.imr_ifindex != ifindex)
2196 				continue;
2197 		} else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2198 				iml->multi.imr_address.s_addr)
2199 			continue;
2200 
2201 		(void) ip_mc_leave_src(sk, iml, in_dev);
2202 
2203 		*imlp = iml->next_rcu;
2204 
2205 		if (in_dev)
2206 			ip_mc_dec_group(in_dev, group);
2207 
2208 		/* decrease mem now to avoid the memleak warning */
2209 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2210 		kfree_rcu(iml, rcu);
2211 		return 0;
2212 	}
2213 out:
2214 	return ret;
2215 }
2216 EXPORT_SYMBOL(ip_mc_leave_group);
2217 
2218 int ip_mc_source(int add, int omode, struct sock *sk, struct
2219 	ip_mreq_source *mreqs, int ifindex)
2220 {
2221 	int err;
2222 	struct ip_mreqn imr;
2223 	__be32 addr = mreqs->imr_multiaddr;
2224 	struct ip_mc_socklist *pmc;
2225 	struct in_device *in_dev = NULL;
2226 	struct inet_sock *inet = inet_sk(sk);
2227 	struct ip_sf_socklist *psl;
2228 	struct net *net = sock_net(sk);
2229 	int leavegroup = 0;
2230 	int i, j, rv;
2231 
2232 	if (!ipv4_is_multicast(addr))
2233 		return -EINVAL;
2234 
2235 	ASSERT_RTNL();
2236 
2237 	imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2238 	imr.imr_address.s_addr = mreqs->imr_interface;
2239 	imr.imr_ifindex = ifindex;
2240 	in_dev = ip_mc_find_dev(net, &imr);
2241 
2242 	if (!in_dev) {
2243 		err = -ENODEV;
2244 		goto done;
2245 	}
2246 	err = -EADDRNOTAVAIL;
2247 
2248 	for_each_pmc_rtnl(inet, pmc) {
2249 		if ((pmc->multi.imr_multiaddr.s_addr ==
2250 		     imr.imr_multiaddr.s_addr) &&
2251 		    (pmc->multi.imr_ifindex == imr.imr_ifindex))
2252 			break;
2253 	}
2254 	if (!pmc) {		/* must have a prior join */
2255 		err = -EINVAL;
2256 		goto done;
2257 	}
2258 	/* if a source filter was set, must be the same mode as before */
2259 	if (pmc->sflist) {
2260 		if (pmc->sfmode != omode) {
2261 			err = -EINVAL;
2262 			goto done;
2263 		}
2264 	} else if (pmc->sfmode != omode) {
2265 		/* allow mode switches for empty-set filters */
2266 		ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2267 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2268 			NULL, 0);
2269 		pmc->sfmode = omode;
2270 	}
2271 
2272 	psl = rtnl_dereference(pmc->sflist);
2273 	if (!add) {
2274 		if (!psl)
2275 			goto done;	/* err = -EADDRNOTAVAIL */
2276 		rv = !0;
2277 		for (i = 0; i < psl->sl_count; i++) {
2278 			rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2279 				sizeof(__be32));
2280 			if (rv == 0)
2281 				break;
2282 		}
2283 		if (rv)		/* source not found */
2284 			goto done;	/* err = -EADDRNOTAVAIL */
2285 
2286 		/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2287 		if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2288 			leavegroup = 1;
2289 			goto done;
2290 		}
2291 
2292 		/* update the interface filter */
2293 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2294 			&mreqs->imr_sourceaddr, 1);
2295 
2296 		for (j = i+1; j < psl->sl_count; j++)
2297 			psl->sl_addr[j-1] = psl->sl_addr[j];
2298 		psl->sl_count--;
2299 		err = 0;
2300 		goto done;
2301 	}
2302 	/* else, add a new source to the filter */
2303 
2304 	if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2305 		err = -ENOBUFS;
2306 		goto done;
2307 	}
2308 	if (!psl || psl->sl_count == psl->sl_max) {
2309 		struct ip_sf_socklist *newpsl;
2310 		int count = IP_SFBLOCK;
2311 
2312 		if (psl)
2313 			count += psl->sl_max;
2314 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2315 		if (!newpsl) {
2316 			err = -ENOBUFS;
2317 			goto done;
2318 		}
2319 		newpsl->sl_max = count;
2320 		newpsl->sl_count = count - IP_SFBLOCK;
2321 		if (psl) {
2322 			for (i = 0; i < psl->sl_count; i++)
2323 				newpsl->sl_addr[i] = psl->sl_addr[i];
2324 			/* decrease mem now to avoid the memleak warning */
2325 			atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2326 			kfree_rcu(psl, rcu);
2327 		}
2328 		rcu_assign_pointer(pmc->sflist, newpsl);
2329 		psl = newpsl;
2330 	}
2331 	rv = 1;	/* > 0 for insert logic below if sl_count is 0 */
2332 	for (i = 0; i < psl->sl_count; i++) {
2333 		rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2334 			sizeof(__be32));
2335 		if (rv == 0)
2336 			break;
2337 	}
2338 	if (rv == 0)		/* address already there is an error */
2339 		goto done;
2340 	for (j = psl->sl_count-1; j >= i; j--)
2341 		psl->sl_addr[j+1] = psl->sl_addr[j];
2342 	psl->sl_addr[i] = mreqs->imr_sourceaddr;
2343 	psl->sl_count++;
2344 	err = 0;
2345 	/* update the interface list */
2346 	ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2347 		&mreqs->imr_sourceaddr, 1);
2348 done:
2349 	if (leavegroup)
2350 		err = ip_mc_leave_group(sk, &imr);
2351 	return err;
2352 }
2353 
2354 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2355 {
2356 	int err = 0;
2357 	struct ip_mreqn	imr;
2358 	__be32 addr = msf->imsf_multiaddr;
2359 	struct ip_mc_socklist *pmc;
2360 	struct in_device *in_dev;
2361 	struct inet_sock *inet = inet_sk(sk);
2362 	struct ip_sf_socklist *newpsl, *psl;
2363 	struct net *net = sock_net(sk);
2364 	int leavegroup = 0;
2365 
2366 	if (!ipv4_is_multicast(addr))
2367 		return -EINVAL;
2368 	if (msf->imsf_fmode != MCAST_INCLUDE &&
2369 	    msf->imsf_fmode != MCAST_EXCLUDE)
2370 		return -EINVAL;
2371 
2372 	ASSERT_RTNL();
2373 
2374 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2375 	imr.imr_address.s_addr = msf->imsf_interface;
2376 	imr.imr_ifindex = ifindex;
2377 	in_dev = ip_mc_find_dev(net, &imr);
2378 
2379 	if (!in_dev) {
2380 		err = -ENODEV;
2381 		goto done;
2382 	}
2383 
2384 	/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2385 	if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2386 		leavegroup = 1;
2387 		goto done;
2388 	}
2389 
2390 	for_each_pmc_rtnl(inet, pmc) {
2391 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2392 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2393 			break;
2394 	}
2395 	if (!pmc) {		/* must have a prior join */
2396 		err = -EINVAL;
2397 		goto done;
2398 	}
2399 	if (msf->imsf_numsrc) {
2400 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2401 							   GFP_KERNEL);
2402 		if (!newpsl) {
2403 			err = -ENOBUFS;
2404 			goto done;
2405 		}
2406 		newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2407 		memcpy(newpsl->sl_addr, msf->imsf_slist,
2408 			msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2409 		err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2410 			msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2411 		if (err) {
2412 			sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2413 			goto done;
2414 		}
2415 	} else {
2416 		newpsl = NULL;
2417 		(void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2418 				     msf->imsf_fmode, 0, NULL, 0);
2419 	}
2420 	psl = rtnl_dereference(pmc->sflist);
2421 	if (psl) {
2422 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2423 			psl->sl_count, psl->sl_addr, 0);
2424 		/* decrease mem now to avoid the memleak warning */
2425 		atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2426 		kfree_rcu(psl, rcu);
2427 	} else
2428 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2429 			0, NULL, 0);
2430 	rcu_assign_pointer(pmc->sflist, newpsl);
2431 	pmc->sfmode = msf->imsf_fmode;
2432 	err = 0;
2433 done:
2434 	if (leavegroup)
2435 		err = ip_mc_leave_group(sk, &imr);
2436 	return err;
2437 }
2438 
2439 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2440 	struct ip_msfilter __user *optval, int __user *optlen)
2441 {
2442 	int err, len, count, copycount;
2443 	struct ip_mreqn	imr;
2444 	__be32 addr = msf->imsf_multiaddr;
2445 	struct ip_mc_socklist *pmc;
2446 	struct in_device *in_dev;
2447 	struct inet_sock *inet = inet_sk(sk);
2448 	struct ip_sf_socklist *psl;
2449 	struct net *net = sock_net(sk);
2450 
2451 	ASSERT_RTNL();
2452 
2453 	if (!ipv4_is_multicast(addr))
2454 		return -EINVAL;
2455 
2456 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2457 	imr.imr_address.s_addr = msf->imsf_interface;
2458 	imr.imr_ifindex = 0;
2459 	in_dev = ip_mc_find_dev(net, &imr);
2460 
2461 	if (!in_dev) {
2462 		err = -ENODEV;
2463 		goto done;
2464 	}
2465 	err = -EADDRNOTAVAIL;
2466 
2467 	for_each_pmc_rtnl(inet, pmc) {
2468 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2469 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2470 			break;
2471 	}
2472 	if (!pmc)		/* must have a prior join */
2473 		goto done;
2474 	msf->imsf_fmode = pmc->sfmode;
2475 	psl = rtnl_dereference(pmc->sflist);
2476 	if (!psl) {
2477 		len = 0;
2478 		count = 0;
2479 	} else {
2480 		count = psl->sl_count;
2481 	}
2482 	copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2483 	len = copycount * sizeof(psl->sl_addr[0]);
2484 	msf->imsf_numsrc = count;
2485 	if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2486 	    copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2487 		return -EFAULT;
2488 	}
2489 	if (len &&
2490 	    copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2491 		return -EFAULT;
2492 	return 0;
2493 done:
2494 	return err;
2495 }
2496 
2497 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2498 	struct group_filter __user *optval, int __user *optlen)
2499 {
2500 	int err, i, count, copycount;
2501 	struct sockaddr_in *psin;
2502 	__be32 addr;
2503 	struct ip_mc_socklist *pmc;
2504 	struct inet_sock *inet = inet_sk(sk);
2505 	struct ip_sf_socklist *psl;
2506 
2507 	ASSERT_RTNL();
2508 
2509 	psin = (struct sockaddr_in *)&gsf->gf_group;
2510 	if (psin->sin_family != AF_INET)
2511 		return -EINVAL;
2512 	addr = psin->sin_addr.s_addr;
2513 	if (!ipv4_is_multicast(addr))
2514 		return -EINVAL;
2515 
2516 	err = -EADDRNOTAVAIL;
2517 
2518 	for_each_pmc_rtnl(inet, pmc) {
2519 		if (pmc->multi.imr_multiaddr.s_addr == addr &&
2520 		    pmc->multi.imr_ifindex == gsf->gf_interface)
2521 			break;
2522 	}
2523 	if (!pmc)		/* must have a prior join */
2524 		goto done;
2525 	gsf->gf_fmode = pmc->sfmode;
2526 	psl = rtnl_dereference(pmc->sflist);
2527 	count = psl ? psl->sl_count : 0;
2528 	copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2529 	gsf->gf_numsrc = count;
2530 	if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2531 	    copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2532 		return -EFAULT;
2533 	}
2534 	for (i = 0; i < copycount; i++) {
2535 		struct sockaddr_storage ss;
2536 
2537 		psin = (struct sockaddr_in *)&ss;
2538 		memset(&ss, 0, sizeof(ss));
2539 		psin->sin_family = AF_INET;
2540 		psin->sin_addr.s_addr = psl->sl_addr[i];
2541 		if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2542 			return -EFAULT;
2543 	}
2544 	return 0;
2545 done:
2546 	return err;
2547 }
2548 
2549 /*
2550  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2551  */
2552 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr, int dif)
2553 {
2554 	struct inet_sock *inet = inet_sk(sk);
2555 	struct ip_mc_socklist *pmc;
2556 	struct ip_sf_socklist *psl;
2557 	int i;
2558 	int ret;
2559 
2560 	ret = 1;
2561 	if (!ipv4_is_multicast(loc_addr))
2562 		goto out;
2563 
2564 	rcu_read_lock();
2565 	for_each_pmc_rcu(inet, pmc) {
2566 		if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2567 		    pmc->multi.imr_ifindex == dif)
2568 			break;
2569 	}
2570 	ret = inet->mc_all;
2571 	if (!pmc)
2572 		goto unlock;
2573 	psl = rcu_dereference(pmc->sflist);
2574 	ret = (pmc->sfmode == MCAST_EXCLUDE);
2575 	if (!psl)
2576 		goto unlock;
2577 
2578 	for (i = 0; i < psl->sl_count; i++) {
2579 		if (psl->sl_addr[i] == rmt_addr)
2580 			break;
2581 	}
2582 	ret = 0;
2583 	if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2584 		goto unlock;
2585 	if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2586 		goto unlock;
2587 	ret = 1;
2588 unlock:
2589 	rcu_read_unlock();
2590 out:
2591 	return ret;
2592 }
2593 
2594 /*
2595  *	A socket is closing.
2596  */
2597 
2598 void ip_mc_drop_socket(struct sock *sk)
2599 {
2600 	struct inet_sock *inet = inet_sk(sk);
2601 	struct ip_mc_socklist *iml;
2602 	struct net *net = sock_net(sk);
2603 
2604 	if (!inet->mc_list)
2605 		return;
2606 
2607 	rtnl_lock();
2608 	while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2609 		struct in_device *in_dev;
2610 
2611 		inet->mc_list = iml->next_rcu;
2612 		in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2613 		(void) ip_mc_leave_src(sk, iml, in_dev);
2614 		if (in_dev)
2615 			ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2616 		/* decrease mem now to avoid the memleak warning */
2617 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2618 		kfree_rcu(iml, rcu);
2619 	}
2620 	rtnl_unlock();
2621 }
2622 
2623 /* called with rcu_read_lock() */
2624 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2625 {
2626 	struct ip_mc_list *im;
2627 	struct ip_mc_list __rcu **mc_hash;
2628 	struct ip_sf_list *psf;
2629 	int rv = 0;
2630 
2631 	mc_hash = rcu_dereference(in_dev->mc_hash);
2632 	if (mc_hash) {
2633 		u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2634 
2635 		for (im = rcu_dereference(mc_hash[hash]);
2636 		     im != NULL;
2637 		     im = rcu_dereference(im->next_hash)) {
2638 			if (im->multiaddr == mc_addr)
2639 				break;
2640 		}
2641 	} else {
2642 		for_each_pmc_rcu(in_dev, im) {
2643 			if (im->multiaddr == mc_addr)
2644 				break;
2645 		}
2646 	}
2647 	if (im && proto == IPPROTO_IGMP) {
2648 		rv = 1;
2649 	} else if (im) {
2650 		if (src_addr) {
2651 			for (psf = im->sources; psf; psf = psf->sf_next) {
2652 				if (psf->sf_inaddr == src_addr)
2653 					break;
2654 			}
2655 			if (psf)
2656 				rv = psf->sf_count[MCAST_INCLUDE] ||
2657 					psf->sf_count[MCAST_EXCLUDE] !=
2658 					im->sfcount[MCAST_EXCLUDE];
2659 			else
2660 				rv = im->sfcount[MCAST_EXCLUDE] != 0;
2661 		} else
2662 			rv = 1; /* unspecified source; tentatively allow */
2663 	}
2664 	return rv;
2665 }
2666 
2667 #if defined(CONFIG_PROC_FS)
2668 struct igmp_mc_iter_state {
2669 	struct seq_net_private p;
2670 	struct net_device *dev;
2671 	struct in_device *in_dev;
2672 };
2673 
2674 #define	igmp_mc_seq_private(seq)	((struct igmp_mc_iter_state *)(seq)->private)
2675 
2676 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2677 {
2678 	struct net *net = seq_file_net(seq);
2679 	struct ip_mc_list *im = NULL;
2680 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2681 
2682 	state->in_dev = NULL;
2683 	for_each_netdev_rcu(net, state->dev) {
2684 		struct in_device *in_dev;
2685 
2686 		in_dev = __in_dev_get_rcu(state->dev);
2687 		if (!in_dev)
2688 			continue;
2689 		im = rcu_dereference(in_dev->mc_list);
2690 		if (im) {
2691 			state->in_dev = in_dev;
2692 			break;
2693 		}
2694 	}
2695 	return im;
2696 }
2697 
2698 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2699 {
2700 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2701 
2702 	im = rcu_dereference(im->next_rcu);
2703 	while (!im) {
2704 		state->dev = next_net_device_rcu(state->dev);
2705 		if (!state->dev) {
2706 			state->in_dev = NULL;
2707 			break;
2708 		}
2709 		state->in_dev = __in_dev_get_rcu(state->dev);
2710 		if (!state->in_dev)
2711 			continue;
2712 		im = rcu_dereference(state->in_dev->mc_list);
2713 	}
2714 	return im;
2715 }
2716 
2717 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2718 {
2719 	struct ip_mc_list *im = igmp_mc_get_first(seq);
2720 	if (im)
2721 		while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2722 			--pos;
2723 	return pos ? NULL : im;
2724 }
2725 
2726 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2727 	__acquires(rcu)
2728 {
2729 	rcu_read_lock();
2730 	return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2731 }
2732 
2733 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2734 {
2735 	struct ip_mc_list *im;
2736 	if (v == SEQ_START_TOKEN)
2737 		im = igmp_mc_get_first(seq);
2738 	else
2739 		im = igmp_mc_get_next(seq, v);
2740 	++*pos;
2741 	return im;
2742 }
2743 
2744 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2745 	__releases(rcu)
2746 {
2747 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2748 
2749 	state->in_dev = NULL;
2750 	state->dev = NULL;
2751 	rcu_read_unlock();
2752 }
2753 
2754 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2755 {
2756 	if (v == SEQ_START_TOKEN)
2757 		seq_puts(seq,
2758 			 "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2759 	else {
2760 		struct ip_mc_list *im = (struct ip_mc_list *)v;
2761 		struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2762 		char   *querier;
2763 		long delta;
2764 
2765 #ifdef CONFIG_IP_MULTICAST
2766 		querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2767 			  IGMP_V2_SEEN(state->in_dev) ? "V2" :
2768 			  "V3";
2769 #else
2770 		querier = "NONE";
2771 #endif
2772 
2773 		if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2774 			seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2775 				   state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2776 		}
2777 
2778 		delta = im->timer.expires - jiffies;
2779 		seq_printf(seq,
2780 			   "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2781 			   im->multiaddr, im->users,
2782 			   im->tm_running,
2783 			   im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2784 			   im->reporter);
2785 	}
2786 	return 0;
2787 }
2788 
2789 static const struct seq_operations igmp_mc_seq_ops = {
2790 	.start	=	igmp_mc_seq_start,
2791 	.next	=	igmp_mc_seq_next,
2792 	.stop	=	igmp_mc_seq_stop,
2793 	.show	=	igmp_mc_seq_show,
2794 };
2795 
2796 static int igmp_mc_seq_open(struct inode *inode, struct file *file)
2797 {
2798 	return seq_open_net(inode, file, &igmp_mc_seq_ops,
2799 			sizeof(struct igmp_mc_iter_state));
2800 }
2801 
2802 static const struct file_operations igmp_mc_seq_fops = {
2803 	.owner		=	THIS_MODULE,
2804 	.open		=	igmp_mc_seq_open,
2805 	.read		=	seq_read,
2806 	.llseek		=	seq_lseek,
2807 	.release	=	seq_release_net,
2808 };
2809 
2810 struct igmp_mcf_iter_state {
2811 	struct seq_net_private p;
2812 	struct net_device *dev;
2813 	struct in_device *idev;
2814 	struct ip_mc_list *im;
2815 };
2816 
2817 #define igmp_mcf_seq_private(seq)	((struct igmp_mcf_iter_state *)(seq)->private)
2818 
2819 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2820 {
2821 	struct net *net = seq_file_net(seq);
2822 	struct ip_sf_list *psf = NULL;
2823 	struct ip_mc_list *im = NULL;
2824 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2825 
2826 	state->idev = NULL;
2827 	state->im = NULL;
2828 	for_each_netdev_rcu(net, state->dev) {
2829 		struct in_device *idev;
2830 		idev = __in_dev_get_rcu(state->dev);
2831 		if (unlikely(!idev))
2832 			continue;
2833 		im = rcu_dereference(idev->mc_list);
2834 		if (likely(im)) {
2835 			spin_lock_bh(&im->lock);
2836 			psf = im->sources;
2837 			if (likely(psf)) {
2838 				state->im = im;
2839 				state->idev = idev;
2840 				break;
2841 			}
2842 			spin_unlock_bh(&im->lock);
2843 		}
2844 	}
2845 	return psf;
2846 }
2847 
2848 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2849 {
2850 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2851 
2852 	psf = psf->sf_next;
2853 	while (!psf) {
2854 		spin_unlock_bh(&state->im->lock);
2855 		state->im = state->im->next;
2856 		while (!state->im) {
2857 			state->dev = next_net_device_rcu(state->dev);
2858 			if (!state->dev) {
2859 				state->idev = NULL;
2860 				goto out;
2861 			}
2862 			state->idev = __in_dev_get_rcu(state->dev);
2863 			if (!state->idev)
2864 				continue;
2865 			state->im = rcu_dereference(state->idev->mc_list);
2866 		}
2867 		if (!state->im)
2868 			break;
2869 		spin_lock_bh(&state->im->lock);
2870 		psf = state->im->sources;
2871 	}
2872 out:
2873 	return psf;
2874 }
2875 
2876 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2877 {
2878 	struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2879 	if (psf)
2880 		while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2881 			--pos;
2882 	return pos ? NULL : psf;
2883 }
2884 
2885 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2886 	__acquires(rcu)
2887 {
2888 	rcu_read_lock();
2889 	return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2890 }
2891 
2892 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2893 {
2894 	struct ip_sf_list *psf;
2895 	if (v == SEQ_START_TOKEN)
2896 		psf = igmp_mcf_get_first(seq);
2897 	else
2898 		psf = igmp_mcf_get_next(seq, v);
2899 	++*pos;
2900 	return psf;
2901 }
2902 
2903 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2904 	__releases(rcu)
2905 {
2906 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2907 	if (likely(state->im)) {
2908 		spin_unlock_bh(&state->im->lock);
2909 		state->im = NULL;
2910 	}
2911 	state->idev = NULL;
2912 	state->dev = NULL;
2913 	rcu_read_unlock();
2914 }
2915 
2916 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2917 {
2918 	struct ip_sf_list *psf = (struct ip_sf_list *)v;
2919 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2920 
2921 	if (v == SEQ_START_TOKEN) {
2922 		seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2923 	} else {
2924 		seq_printf(seq,
2925 			   "%3d %6.6s 0x%08x "
2926 			   "0x%08x %6lu %6lu\n",
2927 			   state->dev->ifindex, state->dev->name,
2928 			   ntohl(state->im->multiaddr),
2929 			   ntohl(psf->sf_inaddr),
2930 			   psf->sf_count[MCAST_INCLUDE],
2931 			   psf->sf_count[MCAST_EXCLUDE]);
2932 	}
2933 	return 0;
2934 }
2935 
2936 static const struct seq_operations igmp_mcf_seq_ops = {
2937 	.start	=	igmp_mcf_seq_start,
2938 	.next	=	igmp_mcf_seq_next,
2939 	.stop	=	igmp_mcf_seq_stop,
2940 	.show	=	igmp_mcf_seq_show,
2941 };
2942 
2943 static int igmp_mcf_seq_open(struct inode *inode, struct file *file)
2944 {
2945 	return seq_open_net(inode, file, &igmp_mcf_seq_ops,
2946 			sizeof(struct igmp_mcf_iter_state));
2947 }
2948 
2949 static const struct file_operations igmp_mcf_seq_fops = {
2950 	.owner		=	THIS_MODULE,
2951 	.open		=	igmp_mcf_seq_open,
2952 	.read		=	seq_read,
2953 	.llseek		=	seq_lseek,
2954 	.release	=	seq_release_net,
2955 };
2956 
2957 static int __net_init igmp_net_init(struct net *net)
2958 {
2959 	struct proc_dir_entry *pde;
2960 	int err;
2961 
2962 	pde = proc_create("igmp", S_IRUGO, net->proc_net, &igmp_mc_seq_fops);
2963 	if (!pde)
2964 		goto out_igmp;
2965 	pde = proc_create("mcfilter", S_IRUGO, net->proc_net,
2966 			  &igmp_mcf_seq_fops);
2967 	if (!pde)
2968 		goto out_mcfilter;
2969 	err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
2970 				   SOCK_DGRAM, 0, net);
2971 	if (err < 0) {
2972 		pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
2973 		       err);
2974 		goto out_sock;
2975 	}
2976 
2977 	/* Sysctl initialization */
2978 	net->ipv4.sysctl_igmp_max_memberships = 20;
2979 	net->ipv4.sysctl_igmp_max_msf = 10;
2980 	/* IGMP reports for link-local multicast groups are enabled by default */
2981 	net->ipv4.sysctl_igmp_llm_reports = 1;
2982 	net->ipv4.sysctl_igmp_qrv = 2;
2983 	return 0;
2984 
2985 out_sock:
2986 	remove_proc_entry("mcfilter", net->proc_net);
2987 out_mcfilter:
2988 	remove_proc_entry("igmp", net->proc_net);
2989 out_igmp:
2990 	return -ENOMEM;
2991 }
2992 
2993 static void __net_exit igmp_net_exit(struct net *net)
2994 {
2995 	remove_proc_entry("mcfilter", net->proc_net);
2996 	remove_proc_entry("igmp", net->proc_net);
2997 	inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
2998 }
2999 
3000 static struct pernet_operations igmp_net_ops = {
3001 	.init = igmp_net_init,
3002 	.exit = igmp_net_exit,
3003 };
3004 #endif
3005 
3006 static int igmp_netdev_event(struct notifier_block *this,
3007 			     unsigned long event, void *ptr)
3008 {
3009 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3010 	struct in_device *in_dev;
3011 
3012 	switch (event) {
3013 	case NETDEV_RESEND_IGMP:
3014 		in_dev = __in_dev_get_rtnl(dev);
3015 		if (in_dev)
3016 			ip_mc_rejoin_groups(in_dev);
3017 		break;
3018 	default:
3019 		break;
3020 	}
3021 	return NOTIFY_DONE;
3022 }
3023 
3024 static struct notifier_block igmp_notifier = {
3025 	.notifier_call = igmp_netdev_event,
3026 };
3027 
3028 int __init igmp_mc_init(void)
3029 {
3030 #if defined(CONFIG_PROC_FS)
3031 	int err;
3032 
3033 	err = register_pernet_subsys(&igmp_net_ops);
3034 	if (err)
3035 		return err;
3036 	err = register_netdevice_notifier(&igmp_notifier);
3037 	if (err)
3038 		goto reg_notif_fail;
3039 	return 0;
3040 
3041 reg_notif_fail:
3042 	unregister_pernet_subsys(&igmp_net_ops);
3043 	return err;
3044 #else
3045 	return register_netdevice_notifier(&igmp_notifier);
3046 #endif
3047 }
3048