xref: /openbmc/linux/net/ipv4/igmp.c (revision dea54fba)
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 net_device *dev = skb->dev;
1011 	struct in_device *in_dev;
1012 	int len = skb->len;
1013 	bool dropped = true;
1014 
1015 	if (netif_is_l3_master(dev)) {
1016 		dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1017 		if (!dev)
1018 			goto drop;
1019 	}
1020 
1021 	in_dev = __in_dev_get_rcu(dev);
1022 	if (!in_dev)
1023 		goto drop;
1024 
1025 	if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1026 		goto drop;
1027 
1028 	if (skb_checksum_simple_validate(skb))
1029 		goto drop;
1030 
1031 	ih = igmp_hdr(skb);
1032 	switch (ih->type) {
1033 	case IGMP_HOST_MEMBERSHIP_QUERY:
1034 		dropped = igmp_heard_query(in_dev, skb, len);
1035 		break;
1036 	case IGMP_HOST_MEMBERSHIP_REPORT:
1037 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
1038 		/* Is it our report looped back? */
1039 		if (rt_is_output_route(skb_rtable(skb)))
1040 			break;
1041 		/* don't rely on MC router hearing unicast reports */
1042 		if (skb->pkt_type == PACKET_MULTICAST ||
1043 		    skb->pkt_type == PACKET_BROADCAST)
1044 			dropped = igmp_heard_report(in_dev, ih->group);
1045 		break;
1046 	case IGMP_PIM:
1047 #ifdef CONFIG_IP_PIMSM_V1
1048 		return pim_rcv_v1(skb);
1049 #endif
1050 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
1051 	case IGMP_DVMRP:
1052 	case IGMP_TRACE:
1053 	case IGMP_HOST_LEAVE_MESSAGE:
1054 	case IGMP_MTRACE:
1055 	case IGMP_MTRACE_RESP:
1056 		break;
1057 	default:
1058 		break;
1059 	}
1060 
1061 drop:
1062 	if (dropped)
1063 		kfree_skb(skb);
1064 	else
1065 		consume_skb(skb);
1066 	return 0;
1067 }
1068 
1069 #endif
1070 
1071 
1072 /*
1073  *	Add a filter to a device
1074  */
1075 
1076 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1077 {
1078 	char buf[MAX_ADDR_LEN];
1079 	struct net_device *dev = in_dev->dev;
1080 
1081 	/* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1082 	   We will get multicast token leakage, when IFF_MULTICAST
1083 	   is changed. This check should be done in ndo_set_rx_mode
1084 	   routine. Something sort of:
1085 	   if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1086 	   --ANK
1087 	   */
1088 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1089 		dev_mc_add(dev, buf);
1090 }
1091 
1092 /*
1093  *	Remove a filter from a device
1094  */
1095 
1096 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1097 {
1098 	char buf[MAX_ADDR_LEN];
1099 	struct net_device *dev = in_dev->dev;
1100 
1101 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1102 		dev_mc_del(dev, buf);
1103 }
1104 
1105 #ifdef CONFIG_IP_MULTICAST
1106 /*
1107  * deleted ip_mc_list manipulation
1108  */
1109 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1110 {
1111 	struct ip_mc_list *pmc;
1112 	struct net *net = dev_net(in_dev->dev);
1113 
1114 	/* this is an "ip_mc_list" for convenience; only the fields below
1115 	 * are actually used. In particular, the refcnt and users are not
1116 	 * used for management of the delete list. Using the same structure
1117 	 * for deleted items allows change reports to use common code with
1118 	 * non-deleted or query-response MCA's.
1119 	 */
1120 	pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1121 	if (!pmc)
1122 		return;
1123 	spin_lock_init(&pmc->lock);
1124 	spin_lock_bh(&im->lock);
1125 	pmc->interface = im->interface;
1126 	in_dev_hold(in_dev);
1127 	pmc->multiaddr = im->multiaddr;
1128 	pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1129 	pmc->sfmode = im->sfmode;
1130 	if (pmc->sfmode == MCAST_INCLUDE) {
1131 		struct ip_sf_list *psf;
1132 
1133 		pmc->tomb = im->tomb;
1134 		pmc->sources = im->sources;
1135 		im->tomb = im->sources = NULL;
1136 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1137 			psf->sf_crcount = pmc->crcount;
1138 	}
1139 	spin_unlock_bh(&im->lock);
1140 
1141 	spin_lock_bh(&in_dev->mc_tomb_lock);
1142 	pmc->next = in_dev->mc_tomb;
1143 	in_dev->mc_tomb = pmc;
1144 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1145 }
1146 
1147 /*
1148  * restore ip_mc_list deleted records
1149  */
1150 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1151 {
1152 	struct ip_mc_list *pmc, *pmc_prev;
1153 	struct ip_sf_list *psf;
1154 	struct net *net = dev_net(in_dev->dev);
1155 	__be32 multiaddr = im->multiaddr;
1156 
1157 	spin_lock_bh(&in_dev->mc_tomb_lock);
1158 	pmc_prev = NULL;
1159 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1160 		if (pmc->multiaddr == multiaddr)
1161 			break;
1162 		pmc_prev = pmc;
1163 	}
1164 	if (pmc) {
1165 		if (pmc_prev)
1166 			pmc_prev->next = pmc->next;
1167 		else
1168 			in_dev->mc_tomb = pmc->next;
1169 	}
1170 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1171 
1172 	spin_lock_bh(&im->lock);
1173 	if (pmc) {
1174 		im->interface = pmc->interface;
1175 		im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1176 		im->sfmode = pmc->sfmode;
1177 		if (pmc->sfmode == MCAST_INCLUDE) {
1178 			im->tomb = pmc->tomb;
1179 			im->sources = pmc->sources;
1180 			for (psf = im->sources; psf; psf = psf->sf_next)
1181 				psf->sf_crcount = im->crcount;
1182 		}
1183 		in_dev_put(pmc->interface);
1184 		kfree(pmc);
1185 	}
1186 	spin_unlock_bh(&im->lock);
1187 }
1188 
1189 /*
1190  * flush ip_mc_list deleted records
1191  */
1192 static void igmpv3_clear_delrec(struct in_device *in_dev)
1193 {
1194 	struct ip_mc_list *pmc, *nextpmc;
1195 
1196 	spin_lock_bh(&in_dev->mc_tomb_lock);
1197 	pmc = in_dev->mc_tomb;
1198 	in_dev->mc_tomb = NULL;
1199 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1200 
1201 	for (; pmc; pmc = nextpmc) {
1202 		nextpmc = pmc->next;
1203 		ip_mc_clear_src(pmc);
1204 		in_dev_put(pmc->interface);
1205 		kfree(pmc);
1206 	}
1207 	/* clear dead sources, too */
1208 	rcu_read_lock();
1209 	for_each_pmc_rcu(in_dev, pmc) {
1210 		struct ip_sf_list *psf, *psf_next;
1211 
1212 		spin_lock_bh(&pmc->lock);
1213 		psf = pmc->tomb;
1214 		pmc->tomb = NULL;
1215 		spin_unlock_bh(&pmc->lock);
1216 		for (; psf; psf = psf_next) {
1217 			psf_next = psf->sf_next;
1218 			kfree(psf);
1219 		}
1220 	}
1221 	rcu_read_unlock();
1222 }
1223 #endif
1224 
1225 static void igmp_group_dropped(struct ip_mc_list *im)
1226 {
1227 	struct in_device *in_dev = im->interface;
1228 #ifdef CONFIG_IP_MULTICAST
1229 	struct net *net = dev_net(in_dev->dev);
1230 	int reporter;
1231 #endif
1232 
1233 	if (im->loaded) {
1234 		im->loaded = 0;
1235 		ip_mc_filter_del(in_dev, im->multiaddr);
1236 	}
1237 
1238 #ifdef CONFIG_IP_MULTICAST
1239 	if (im->multiaddr == IGMP_ALL_HOSTS)
1240 		return;
1241 	if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1242 		return;
1243 
1244 	reporter = im->reporter;
1245 	igmp_stop_timer(im);
1246 
1247 	if (!in_dev->dead) {
1248 		if (IGMP_V1_SEEN(in_dev))
1249 			return;
1250 		if (IGMP_V2_SEEN(in_dev)) {
1251 			if (reporter)
1252 				igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1253 			return;
1254 		}
1255 		/* IGMPv3 */
1256 		igmpv3_add_delrec(in_dev, im);
1257 
1258 		igmp_ifc_event(in_dev);
1259 	}
1260 #endif
1261 }
1262 
1263 static void igmp_group_added(struct ip_mc_list *im)
1264 {
1265 	struct in_device *in_dev = im->interface;
1266 #ifdef CONFIG_IP_MULTICAST
1267 	struct net *net = dev_net(in_dev->dev);
1268 #endif
1269 
1270 	if (im->loaded == 0) {
1271 		im->loaded = 1;
1272 		ip_mc_filter_add(in_dev, im->multiaddr);
1273 	}
1274 
1275 #ifdef CONFIG_IP_MULTICAST
1276 	if (im->multiaddr == IGMP_ALL_HOSTS)
1277 		return;
1278 	if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1279 		return;
1280 
1281 	if (in_dev->dead)
1282 		return;
1283 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1284 		spin_lock_bh(&im->lock);
1285 		igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1286 		spin_unlock_bh(&im->lock);
1287 		return;
1288 	}
1289 	/* else, v3 */
1290 
1291 	im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1292 	igmp_ifc_event(in_dev);
1293 #endif
1294 }
1295 
1296 
1297 /*
1298  *	Multicast list managers
1299  */
1300 
1301 static u32 ip_mc_hash(const struct ip_mc_list *im)
1302 {
1303 	return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1304 }
1305 
1306 static void ip_mc_hash_add(struct in_device *in_dev,
1307 			   struct ip_mc_list *im)
1308 {
1309 	struct ip_mc_list __rcu **mc_hash;
1310 	u32 hash;
1311 
1312 	mc_hash = rtnl_dereference(in_dev->mc_hash);
1313 	if (mc_hash) {
1314 		hash = ip_mc_hash(im);
1315 		im->next_hash = mc_hash[hash];
1316 		rcu_assign_pointer(mc_hash[hash], im);
1317 		return;
1318 	}
1319 
1320 	/* do not use a hash table for small number of items */
1321 	if (in_dev->mc_count < 4)
1322 		return;
1323 
1324 	mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1325 			  GFP_KERNEL);
1326 	if (!mc_hash)
1327 		return;
1328 
1329 	for_each_pmc_rtnl(in_dev, im) {
1330 		hash = ip_mc_hash(im);
1331 		im->next_hash = mc_hash[hash];
1332 		RCU_INIT_POINTER(mc_hash[hash], im);
1333 	}
1334 
1335 	rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1336 }
1337 
1338 static void ip_mc_hash_remove(struct in_device *in_dev,
1339 			      struct ip_mc_list *im)
1340 {
1341 	struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1342 	struct ip_mc_list *aux;
1343 
1344 	if (!mc_hash)
1345 		return;
1346 	mc_hash += ip_mc_hash(im);
1347 	while ((aux = rtnl_dereference(*mc_hash)) != im)
1348 		mc_hash = &aux->next_hash;
1349 	*mc_hash = im->next_hash;
1350 }
1351 
1352 
1353 /*
1354  *	A socket has joined a multicast group on device dev.
1355  */
1356 
1357 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1358 {
1359 	struct ip_mc_list *im;
1360 #ifdef CONFIG_IP_MULTICAST
1361 	struct net *net = dev_net(in_dev->dev);
1362 #endif
1363 
1364 	ASSERT_RTNL();
1365 
1366 	for_each_pmc_rtnl(in_dev, im) {
1367 		if (im->multiaddr == addr) {
1368 			im->users++;
1369 			ip_mc_add_src(in_dev, &addr, MCAST_EXCLUDE, 0, NULL, 0);
1370 			goto out;
1371 		}
1372 	}
1373 
1374 	im = kzalloc(sizeof(*im), GFP_KERNEL);
1375 	if (!im)
1376 		goto out;
1377 
1378 	im->users = 1;
1379 	im->interface = in_dev;
1380 	in_dev_hold(in_dev);
1381 	im->multiaddr = addr;
1382 	/* initial mode is (EX, empty) */
1383 	im->sfmode = MCAST_EXCLUDE;
1384 	im->sfcount[MCAST_EXCLUDE] = 1;
1385 	refcount_set(&im->refcnt, 1);
1386 	spin_lock_init(&im->lock);
1387 #ifdef CONFIG_IP_MULTICAST
1388 	setup_timer(&im->timer, igmp_timer_expire, (unsigned long)im);
1389 	im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1390 #endif
1391 
1392 	im->next_rcu = in_dev->mc_list;
1393 	in_dev->mc_count++;
1394 	rcu_assign_pointer(in_dev->mc_list, im);
1395 
1396 	ip_mc_hash_add(in_dev, im);
1397 
1398 #ifdef CONFIG_IP_MULTICAST
1399 	igmpv3_del_delrec(in_dev, im);
1400 #endif
1401 	igmp_group_added(im);
1402 	if (!in_dev->dead)
1403 		ip_rt_multicast_event(in_dev);
1404 out:
1405 	return;
1406 }
1407 EXPORT_SYMBOL(ip_mc_inc_group);
1408 
1409 static int ip_mc_check_iphdr(struct sk_buff *skb)
1410 {
1411 	const struct iphdr *iph;
1412 	unsigned int len;
1413 	unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1414 
1415 	if (!pskb_may_pull(skb, offset))
1416 		return -EINVAL;
1417 
1418 	iph = ip_hdr(skb);
1419 
1420 	if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1421 		return -EINVAL;
1422 
1423 	offset += ip_hdrlen(skb) - sizeof(*iph);
1424 
1425 	if (!pskb_may_pull(skb, offset))
1426 		return -EINVAL;
1427 
1428 	iph = ip_hdr(skb);
1429 
1430 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1431 		return -EINVAL;
1432 
1433 	len = skb_network_offset(skb) + ntohs(iph->tot_len);
1434 	if (skb->len < len || len < offset)
1435 		return -EINVAL;
1436 
1437 	skb_set_transport_header(skb, offset);
1438 
1439 	return 0;
1440 }
1441 
1442 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1443 {
1444 	unsigned int len = skb_transport_offset(skb);
1445 
1446 	len += sizeof(struct igmpv3_report);
1447 
1448 	return pskb_may_pull(skb, len) ? 0 : -EINVAL;
1449 }
1450 
1451 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1452 {
1453 	unsigned int len = skb_transport_offset(skb);
1454 
1455 	len += sizeof(struct igmphdr);
1456 	if (skb->len < len)
1457 		return -EINVAL;
1458 
1459 	/* IGMPv{1,2}? */
1460 	if (skb->len != len) {
1461 		/* or IGMPv3? */
1462 		len += sizeof(struct igmpv3_query) - sizeof(struct igmphdr);
1463 		if (skb->len < len || !pskb_may_pull(skb, len))
1464 			return -EINVAL;
1465 	}
1466 
1467 	/* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1468 	 * all-systems destination addresses (224.0.0.1) for general queries
1469 	 */
1470 	if (!igmp_hdr(skb)->group &&
1471 	    ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1472 		return -EINVAL;
1473 
1474 	return 0;
1475 }
1476 
1477 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1478 {
1479 	switch (igmp_hdr(skb)->type) {
1480 	case IGMP_HOST_LEAVE_MESSAGE:
1481 	case IGMP_HOST_MEMBERSHIP_REPORT:
1482 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
1483 		/* fall through */
1484 		return 0;
1485 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
1486 		return ip_mc_check_igmp_reportv3(skb);
1487 	case IGMP_HOST_MEMBERSHIP_QUERY:
1488 		return ip_mc_check_igmp_query(skb);
1489 	default:
1490 		return -ENOMSG;
1491 	}
1492 }
1493 
1494 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1495 {
1496 	return skb_checksum_simple_validate(skb);
1497 }
1498 
1499 static int __ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1500 
1501 {
1502 	struct sk_buff *skb_chk;
1503 	unsigned int transport_len;
1504 	unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1505 	int ret = -EINVAL;
1506 
1507 	transport_len = ntohs(ip_hdr(skb)->tot_len) - ip_hdrlen(skb);
1508 
1509 	skb_chk = skb_checksum_trimmed(skb, transport_len,
1510 				       ip_mc_validate_checksum);
1511 	if (!skb_chk)
1512 		goto err;
1513 
1514 	if (!pskb_may_pull(skb_chk, len))
1515 		goto err;
1516 
1517 	ret = ip_mc_check_igmp_msg(skb_chk);
1518 	if (ret)
1519 		goto err;
1520 
1521 	if (skb_trimmed)
1522 		*skb_trimmed = skb_chk;
1523 	/* free now unneeded clone */
1524 	else if (skb_chk != skb)
1525 		kfree_skb(skb_chk);
1526 
1527 	ret = 0;
1528 
1529 err:
1530 	if (ret && skb_chk && skb_chk != skb)
1531 		kfree_skb(skb_chk);
1532 
1533 	return ret;
1534 }
1535 
1536 /**
1537  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1538  * @skb: the skb to validate
1539  * @skb_trimmed: to store an skb pointer trimmed to IPv4 packet tail (optional)
1540  *
1541  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1542  * skb transport header accordingly and returns zero.
1543  *
1544  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1545  *  standard
1546  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1547  * -ENOMEM: A memory allocation failure happened.
1548  *
1549  * Optionally, an skb pointer might be provided via skb_trimmed (or set it
1550  * to NULL): After parsing an IGMP packet successfully it will point to
1551  * an skb which has its tail aligned to the IP packet end. This might
1552  * either be the originally provided skb or a trimmed, cloned version if
1553  * the skb frame had data beyond the IP packet. A cloned skb allows us
1554  * to leave the original skb and its full frame unchanged (which might be
1555  * desirable for layer 2 frame jugglers).
1556  *
1557  * Caller needs to set the skb network header and free any returned skb if it
1558  * differs from the provided skb.
1559  */
1560 int ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1561 {
1562 	int ret = ip_mc_check_iphdr(skb);
1563 
1564 	if (ret < 0)
1565 		return ret;
1566 
1567 	if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1568 		return -ENOMSG;
1569 
1570 	return __ip_mc_check_igmp(skb, skb_trimmed);
1571 }
1572 EXPORT_SYMBOL(ip_mc_check_igmp);
1573 
1574 /*
1575  *	Resend IGMP JOIN report; used by netdev notifier.
1576  */
1577 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1578 {
1579 #ifdef CONFIG_IP_MULTICAST
1580 	struct ip_mc_list *im;
1581 	int type;
1582 	struct net *net = dev_net(in_dev->dev);
1583 
1584 	ASSERT_RTNL();
1585 
1586 	for_each_pmc_rtnl(in_dev, im) {
1587 		if (im->multiaddr == IGMP_ALL_HOSTS)
1588 			continue;
1589 		if (ipv4_is_local_multicast(im->multiaddr) &&
1590 		    !net->ipv4.sysctl_igmp_llm_reports)
1591 			continue;
1592 
1593 		/* a failover is happening and switches
1594 		 * must be notified immediately
1595 		 */
1596 		if (IGMP_V1_SEEN(in_dev))
1597 			type = IGMP_HOST_MEMBERSHIP_REPORT;
1598 		else if (IGMP_V2_SEEN(in_dev))
1599 			type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1600 		else
1601 			type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1602 		igmp_send_report(in_dev, im, type);
1603 	}
1604 #endif
1605 }
1606 
1607 /*
1608  *	A socket has left a multicast group on device dev
1609  */
1610 
1611 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1612 {
1613 	struct ip_mc_list *i;
1614 	struct ip_mc_list __rcu **ip;
1615 
1616 	ASSERT_RTNL();
1617 
1618 	for (ip = &in_dev->mc_list;
1619 	     (i = rtnl_dereference(*ip)) != NULL;
1620 	     ip = &i->next_rcu) {
1621 		if (i->multiaddr == addr) {
1622 			if (--i->users == 0) {
1623 				ip_mc_hash_remove(in_dev, i);
1624 				*ip = i->next_rcu;
1625 				in_dev->mc_count--;
1626 				igmp_group_dropped(i);
1627 				ip_mc_clear_src(i);
1628 
1629 				if (!in_dev->dead)
1630 					ip_rt_multicast_event(in_dev);
1631 
1632 				ip_ma_put(i);
1633 				return;
1634 			}
1635 			break;
1636 		}
1637 	}
1638 }
1639 EXPORT_SYMBOL(ip_mc_dec_group);
1640 
1641 /* Device changing type */
1642 
1643 void ip_mc_unmap(struct in_device *in_dev)
1644 {
1645 	struct ip_mc_list *pmc;
1646 
1647 	ASSERT_RTNL();
1648 
1649 	for_each_pmc_rtnl(in_dev, pmc)
1650 		igmp_group_dropped(pmc);
1651 }
1652 
1653 void ip_mc_remap(struct in_device *in_dev)
1654 {
1655 	struct ip_mc_list *pmc;
1656 
1657 	ASSERT_RTNL();
1658 
1659 	for_each_pmc_rtnl(in_dev, pmc) {
1660 #ifdef CONFIG_IP_MULTICAST
1661 		igmpv3_del_delrec(in_dev, pmc);
1662 #endif
1663 		igmp_group_added(pmc);
1664 	}
1665 }
1666 
1667 /* Device going down */
1668 
1669 void ip_mc_down(struct in_device *in_dev)
1670 {
1671 	struct ip_mc_list *pmc;
1672 
1673 	ASSERT_RTNL();
1674 
1675 	for_each_pmc_rtnl(in_dev, pmc)
1676 		igmp_group_dropped(pmc);
1677 
1678 #ifdef CONFIG_IP_MULTICAST
1679 	in_dev->mr_ifc_count = 0;
1680 	if (del_timer(&in_dev->mr_ifc_timer))
1681 		__in_dev_put(in_dev);
1682 	in_dev->mr_gq_running = 0;
1683 	if (del_timer(&in_dev->mr_gq_timer))
1684 		__in_dev_put(in_dev);
1685 #endif
1686 
1687 	ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1688 }
1689 
1690 void ip_mc_init_dev(struct in_device *in_dev)
1691 {
1692 #ifdef CONFIG_IP_MULTICAST
1693 	struct net *net = dev_net(in_dev->dev);
1694 #endif
1695 	ASSERT_RTNL();
1696 
1697 #ifdef CONFIG_IP_MULTICAST
1698 	setup_timer(&in_dev->mr_gq_timer, igmp_gq_timer_expire,
1699 			(unsigned long)in_dev);
1700 	setup_timer(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire,
1701 			(unsigned long)in_dev);
1702 	in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1703 #endif
1704 
1705 	spin_lock_init(&in_dev->mc_tomb_lock);
1706 }
1707 
1708 /* Device going up */
1709 
1710 void ip_mc_up(struct in_device *in_dev)
1711 {
1712 	struct ip_mc_list *pmc;
1713 #ifdef CONFIG_IP_MULTICAST
1714 	struct net *net = dev_net(in_dev->dev);
1715 #endif
1716 
1717 	ASSERT_RTNL();
1718 
1719 #ifdef CONFIG_IP_MULTICAST
1720 	in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1721 #endif
1722 	ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1723 
1724 	for_each_pmc_rtnl(in_dev, pmc) {
1725 #ifdef CONFIG_IP_MULTICAST
1726 		igmpv3_del_delrec(in_dev, pmc);
1727 #endif
1728 		igmp_group_added(pmc);
1729 	}
1730 }
1731 
1732 /*
1733  *	Device is about to be destroyed: clean up.
1734  */
1735 
1736 void ip_mc_destroy_dev(struct in_device *in_dev)
1737 {
1738 	struct ip_mc_list *i;
1739 
1740 	ASSERT_RTNL();
1741 
1742 	/* Deactivate timers */
1743 	ip_mc_down(in_dev);
1744 #ifdef CONFIG_IP_MULTICAST
1745 	igmpv3_clear_delrec(in_dev);
1746 #endif
1747 
1748 	while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1749 		in_dev->mc_list = i->next_rcu;
1750 		in_dev->mc_count--;
1751 		ip_ma_put(i);
1752 	}
1753 }
1754 
1755 /* RTNL is locked */
1756 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1757 {
1758 	struct net_device *dev = NULL;
1759 	struct in_device *idev = NULL;
1760 
1761 	if (imr->imr_ifindex) {
1762 		idev = inetdev_by_index(net, imr->imr_ifindex);
1763 		return idev;
1764 	}
1765 	if (imr->imr_address.s_addr) {
1766 		dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1767 		if (!dev)
1768 			return NULL;
1769 	}
1770 
1771 	if (!dev) {
1772 		struct rtable *rt = ip_route_output(net,
1773 						    imr->imr_multiaddr.s_addr,
1774 						    0, 0, 0);
1775 		if (!IS_ERR(rt)) {
1776 			dev = rt->dst.dev;
1777 			ip_rt_put(rt);
1778 		}
1779 	}
1780 	if (dev) {
1781 		imr->imr_ifindex = dev->ifindex;
1782 		idev = __in_dev_get_rtnl(dev);
1783 	}
1784 	return idev;
1785 }
1786 
1787 /*
1788  *	Join a socket to a group
1789  */
1790 
1791 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1792 	__be32 *psfsrc)
1793 {
1794 	struct ip_sf_list *psf, *psf_prev;
1795 	int rv = 0;
1796 
1797 	psf_prev = NULL;
1798 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1799 		if (psf->sf_inaddr == *psfsrc)
1800 			break;
1801 		psf_prev = psf;
1802 	}
1803 	if (!psf || psf->sf_count[sfmode] == 0) {
1804 		/* source filter not found, or count wrong =>  bug */
1805 		return -ESRCH;
1806 	}
1807 	psf->sf_count[sfmode]--;
1808 	if (psf->sf_count[sfmode] == 0) {
1809 		ip_rt_multicast_event(pmc->interface);
1810 	}
1811 	if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1812 #ifdef CONFIG_IP_MULTICAST
1813 		struct in_device *in_dev = pmc->interface;
1814 		struct net *net = dev_net(in_dev->dev);
1815 #endif
1816 
1817 		/* no more filters for this source */
1818 		if (psf_prev)
1819 			psf_prev->sf_next = psf->sf_next;
1820 		else
1821 			pmc->sources = psf->sf_next;
1822 #ifdef CONFIG_IP_MULTICAST
1823 		if (psf->sf_oldin &&
1824 		    !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1825 			psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1826 			psf->sf_next = pmc->tomb;
1827 			pmc->tomb = psf;
1828 			rv = 1;
1829 		} else
1830 #endif
1831 			kfree(psf);
1832 	}
1833 	return rv;
1834 }
1835 
1836 #ifndef CONFIG_IP_MULTICAST
1837 #define igmp_ifc_event(x)	do { } while (0)
1838 #endif
1839 
1840 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1841 			 int sfcount, __be32 *psfsrc, int delta)
1842 {
1843 	struct ip_mc_list *pmc;
1844 	int	changerec = 0;
1845 	int	i, err;
1846 
1847 	if (!in_dev)
1848 		return -ENODEV;
1849 	rcu_read_lock();
1850 	for_each_pmc_rcu(in_dev, pmc) {
1851 		if (*pmca == pmc->multiaddr)
1852 			break;
1853 	}
1854 	if (!pmc) {
1855 		/* MCA not found?? bug */
1856 		rcu_read_unlock();
1857 		return -ESRCH;
1858 	}
1859 	spin_lock_bh(&pmc->lock);
1860 	rcu_read_unlock();
1861 #ifdef CONFIG_IP_MULTICAST
1862 	sf_markstate(pmc);
1863 #endif
1864 	if (!delta) {
1865 		err = -EINVAL;
1866 		if (!pmc->sfcount[sfmode])
1867 			goto out_unlock;
1868 		pmc->sfcount[sfmode]--;
1869 	}
1870 	err = 0;
1871 	for (i = 0; i < sfcount; i++) {
1872 		int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1873 
1874 		changerec |= rv > 0;
1875 		if (!err && rv < 0)
1876 			err = rv;
1877 	}
1878 	if (pmc->sfmode == MCAST_EXCLUDE &&
1879 	    pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1880 	    pmc->sfcount[MCAST_INCLUDE]) {
1881 #ifdef CONFIG_IP_MULTICAST
1882 		struct ip_sf_list *psf;
1883 		struct net *net = dev_net(in_dev->dev);
1884 #endif
1885 
1886 		/* filter mode change */
1887 		pmc->sfmode = MCAST_INCLUDE;
1888 #ifdef CONFIG_IP_MULTICAST
1889 		pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1890 		in_dev->mr_ifc_count = pmc->crcount;
1891 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1892 			psf->sf_crcount = 0;
1893 		igmp_ifc_event(pmc->interface);
1894 	} else if (sf_setstate(pmc) || changerec) {
1895 		igmp_ifc_event(pmc->interface);
1896 #endif
1897 	}
1898 out_unlock:
1899 	spin_unlock_bh(&pmc->lock);
1900 	return err;
1901 }
1902 
1903 /*
1904  * Add multicast single-source filter to the interface list
1905  */
1906 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1907 	__be32 *psfsrc)
1908 {
1909 	struct ip_sf_list *psf, *psf_prev;
1910 
1911 	psf_prev = NULL;
1912 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1913 		if (psf->sf_inaddr == *psfsrc)
1914 			break;
1915 		psf_prev = psf;
1916 	}
1917 	if (!psf) {
1918 		psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1919 		if (!psf)
1920 			return -ENOBUFS;
1921 		psf->sf_inaddr = *psfsrc;
1922 		if (psf_prev) {
1923 			psf_prev->sf_next = psf;
1924 		} else
1925 			pmc->sources = psf;
1926 	}
1927 	psf->sf_count[sfmode]++;
1928 	if (psf->sf_count[sfmode] == 1) {
1929 		ip_rt_multicast_event(pmc->interface);
1930 	}
1931 	return 0;
1932 }
1933 
1934 #ifdef CONFIG_IP_MULTICAST
1935 static void sf_markstate(struct ip_mc_list *pmc)
1936 {
1937 	struct ip_sf_list *psf;
1938 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1939 
1940 	for (psf = pmc->sources; psf; psf = psf->sf_next)
1941 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1942 			psf->sf_oldin = mca_xcount ==
1943 				psf->sf_count[MCAST_EXCLUDE] &&
1944 				!psf->sf_count[MCAST_INCLUDE];
1945 		} else
1946 			psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1947 }
1948 
1949 static int sf_setstate(struct ip_mc_list *pmc)
1950 {
1951 	struct ip_sf_list *psf, *dpsf;
1952 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1953 	int qrv = pmc->interface->mr_qrv;
1954 	int new_in, rv;
1955 
1956 	rv = 0;
1957 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1958 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1959 			new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1960 				!psf->sf_count[MCAST_INCLUDE];
1961 		} else
1962 			new_in = psf->sf_count[MCAST_INCLUDE] != 0;
1963 		if (new_in) {
1964 			if (!psf->sf_oldin) {
1965 				struct ip_sf_list *prev = NULL;
1966 
1967 				for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
1968 					if (dpsf->sf_inaddr == psf->sf_inaddr)
1969 						break;
1970 					prev = dpsf;
1971 				}
1972 				if (dpsf) {
1973 					if (prev)
1974 						prev->sf_next = dpsf->sf_next;
1975 					else
1976 						pmc->tomb = dpsf->sf_next;
1977 					kfree(dpsf);
1978 				}
1979 				psf->sf_crcount = qrv;
1980 				rv++;
1981 			}
1982 		} else if (psf->sf_oldin) {
1983 
1984 			psf->sf_crcount = 0;
1985 			/*
1986 			 * add or update "delete" records if an active filter
1987 			 * is now inactive
1988 			 */
1989 			for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
1990 				if (dpsf->sf_inaddr == psf->sf_inaddr)
1991 					break;
1992 			if (!dpsf) {
1993 				dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
1994 				if (!dpsf)
1995 					continue;
1996 				*dpsf = *psf;
1997 				/* pmc->lock held by callers */
1998 				dpsf->sf_next = pmc->tomb;
1999 				pmc->tomb = dpsf;
2000 			}
2001 			dpsf->sf_crcount = qrv;
2002 			rv++;
2003 		}
2004 	}
2005 	return rv;
2006 }
2007 #endif
2008 
2009 /*
2010  * Add multicast source filter list to the interface list
2011  */
2012 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2013 			 int sfcount, __be32 *psfsrc, int delta)
2014 {
2015 	struct ip_mc_list *pmc;
2016 	int	isexclude;
2017 	int	i, err;
2018 
2019 	if (!in_dev)
2020 		return -ENODEV;
2021 	rcu_read_lock();
2022 	for_each_pmc_rcu(in_dev, pmc) {
2023 		if (*pmca == pmc->multiaddr)
2024 			break;
2025 	}
2026 	if (!pmc) {
2027 		/* MCA not found?? bug */
2028 		rcu_read_unlock();
2029 		return -ESRCH;
2030 	}
2031 	spin_lock_bh(&pmc->lock);
2032 	rcu_read_unlock();
2033 
2034 #ifdef CONFIG_IP_MULTICAST
2035 	sf_markstate(pmc);
2036 #endif
2037 	isexclude = pmc->sfmode == MCAST_EXCLUDE;
2038 	if (!delta)
2039 		pmc->sfcount[sfmode]++;
2040 	err = 0;
2041 	for (i = 0; i < sfcount; i++) {
2042 		err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2043 		if (err)
2044 			break;
2045 	}
2046 	if (err) {
2047 		int j;
2048 
2049 		if (!delta)
2050 			pmc->sfcount[sfmode]--;
2051 		for (j = 0; j < i; j++)
2052 			(void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2053 	} else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2054 #ifdef CONFIG_IP_MULTICAST
2055 		struct ip_sf_list *psf;
2056 		struct net *net = dev_net(pmc->interface->dev);
2057 		in_dev = pmc->interface;
2058 #endif
2059 
2060 		/* filter mode change */
2061 		if (pmc->sfcount[MCAST_EXCLUDE])
2062 			pmc->sfmode = MCAST_EXCLUDE;
2063 		else if (pmc->sfcount[MCAST_INCLUDE])
2064 			pmc->sfmode = MCAST_INCLUDE;
2065 #ifdef CONFIG_IP_MULTICAST
2066 		/* else no filters; keep old mode for reports */
2067 
2068 		pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2069 		in_dev->mr_ifc_count = pmc->crcount;
2070 		for (psf = pmc->sources; psf; psf = psf->sf_next)
2071 			psf->sf_crcount = 0;
2072 		igmp_ifc_event(in_dev);
2073 	} else if (sf_setstate(pmc)) {
2074 		igmp_ifc_event(in_dev);
2075 #endif
2076 	}
2077 	spin_unlock_bh(&pmc->lock);
2078 	return err;
2079 }
2080 
2081 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2082 {
2083 	struct ip_sf_list *psf, *nextpsf, *tomb, *sources;
2084 
2085 	spin_lock_bh(&pmc->lock);
2086 	tomb = pmc->tomb;
2087 	pmc->tomb = NULL;
2088 	sources = pmc->sources;
2089 	pmc->sources = NULL;
2090 	pmc->sfmode = MCAST_EXCLUDE;
2091 	pmc->sfcount[MCAST_INCLUDE] = 0;
2092 	pmc->sfcount[MCAST_EXCLUDE] = 1;
2093 	spin_unlock_bh(&pmc->lock);
2094 
2095 	for (psf = tomb; psf; psf = nextpsf) {
2096 		nextpsf = psf->sf_next;
2097 		kfree(psf);
2098 	}
2099 	for (psf = sources; psf; psf = nextpsf) {
2100 		nextpsf = psf->sf_next;
2101 		kfree(psf);
2102 	}
2103 }
2104 
2105 /* Join a multicast group
2106  */
2107 
2108 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2109 {
2110 	__be32 addr = imr->imr_multiaddr.s_addr;
2111 	struct ip_mc_socklist *iml, *i;
2112 	struct in_device *in_dev;
2113 	struct inet_sock *inet = inet_sk(sk);
2114 	struct net *net = sock_net(sk);
2115 	int ifindex;
2116 	int count = 0;
2117 	int err;
2118 
2119 	ASSERT_RTNL();
2120 
2121 	if (!ipv4_is_multicast(addr))
2122 		return -EINVAL;
2123 
2124 	in_dev = ip_mc_find_dev(net, imr);
2125 
2126 	if (!in_dev) {
2127 		err = -ENODEV;
2128 		goto done;
2129 	}
2130 
2131 	err = -EADDRINUSE;
2132 	ifindex = imr->imr_ifindex;
2133 	for_each_pmc_rtnl(inet, i) {
2134 		if (i->multi.imr_multiaddr.s_addr == addr &&
2135 		    i->multi.imr_ifindex == ifindex)
2136 			goto done;
2137 		count++;
2138 	}
2139 	err = -ENOBUFS;
2140 	if (count >= net->ipv4.sysctl_igmp_max_memberships)
2141 		goto done;
2142 	iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2143 	if (!iml)
2144 		goto done;
2145 
2146 	memcpy(&iml->multi, imr, sizeof(*imr));
2147 	iml->next_rcu = inet->mc_list;
2148 	iml->sflist = NULL;
2149 	iml->sfmode = MCAST_EXCLUDE;
2150 	rcu_assign_pointer(inet->mc_list, iml);
2151 	ip_mc_inc_group(in_dev, addr);
2152 	err = 0;
2153 done:
2154 	return err;
2155 }
2156 EXPORT_SYMBOL(ip_mc_join_group);
2157 
2158 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2159 			   struct in_device *in_dev)
2160 {
2161 	struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2162 	int err;
2163 
2164 	if (!psf) {
2165 		/* any-source empty exclude case */
2166 		return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2167 			iml->sfmode, 0, NULL, 0);
2168 	}
2169 	err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2170 			iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2171 	RCU_INIT_POINTER(iml->sflist, NULL);
2172 	/* decrease mem now to avoid the memleak warning */
2173 	atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2174 	kfree_rcu(psf, rcu);
2175 	return err;
2176 }
2177 
2178 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2179 {
2180 	struct inet_sock *inet = inet_sk(sk);
2181 	struct ip_mc_socklist *iml;
2182 	struct ip_mc_socklist __rcu **imlp;
2183 	struct in_device *in_dev;
2184 	struct net *net = sock_net(sk);
2185 	__be32 group = imr->imr_multiaddr.s_addr;
2186 	u32 ifindex;
2187 	int ret = -EADDRNOTAVAIL;
2188 
2189 	ASSERT_RTNL();
2190 
2191 	in_dev = ip_mc_find_dev(net, imr);
2192 	if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2193 		ret = -ENODEV;
2194 		goto out;
2195 	}
2196 	ifindex = imr->imr_ifindex;
2197 	for (imlp = &inet->mc_list;
2198 	     (iml = rtnl_dereference(*imlp)) != NULL;
2199 	     imlp = &iml->next_rcu) {
2200 		if (iml->multi.imr_multiaddr.s_addr != group)
2201 			continue;
2202 		if (ifindex) {
2203 			if (iml->multi.imr_ifindex != ifindex)
2204 				continue;
2205 		} else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2206 				iml->multi.imr_address.s_addr)
2207 			continue;
2208 
2209 		(void) ip_mc_leave_src(sk, iml, in_dev);
2210 
2211 		*imlp = iml->next_rcu;
2212 
2213 		if (in_dev)
2214 			ip_mc_dec_group(in_dev, group);
2215 
2216 		/* decrease mem now to avoid the memleak warning */
2217 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2218 		kfree_rcu(iml, rcu);
2219 		return 0;
2220 	}
2221 out:
2222 	return ret;
2223 }
2224 EXPORT_SYMBOL(ip_mc_leave_group);
2225 
2226 int ip_mc_source(int add, int omode, struct sock *sk, struct
2227 	ip_mreq_source *mreqs, int ifindex)
2228 {
2229 	int err;
2230 	struct ip_mreqn imr;
2231 	__be32 addr = mreqs->imr_multiaddr;
2232 	struct ip_mc_socklist *pmc;
2233 	struct in_device *in_dev = NULL;
2234 	struct inet_sock *inet = inet_sk(sk);
2235 	struct ip_sf_socklist *psl;
2236 	struct net *net = sock_net(sk);
2237 	int leavegroup = 0;
2238 	int i, j, rv;
2239 
2240 	if (!ipv4_is_multicast(addr))
2241 		return -EINVAL;
2242 
2243 	ASSERT_RTNL();
2244 
2245 	imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2246 	imr.imr_address.s_addr = mreqs->imr_interface;
2247 	imr.imr_ifindex = ifindex;
2248 	in_dev = ip_mc_find_dev(net, &imr);
2249 
2250 	if (!in_dev) {
2251 		err = -ENODEV;
2252 		goto done;
2253 	}
2254 	err = -EADDRNOTAVAIL;
2255 
2256 	for_each_pmc_rtnl(inet, pmc) {
2257 		if ((pmc->multi.imr_multiaddr.s_addr ==
2258 		     imr.imr_multiaddr.s_addr) &&
2259 		    (pmc->multi.imr_ifindex == imr.imr_ifindex))
2260 			break;
2261 	}
2262 	if (!pmc) {		/* must have a prior join */
2263 		err = -EINVAL;
2264 		goto done;
2265 	}
2266 	/* if a source filter was set, must be the same mode as before */
2267 	if (pmc->sflist) {
2268 		if (pmc->sfmode != omode) {
2269 			err = -EINVAL;
2270 			goto done;
2271 		}
2272 	} else if (pmc->sfmode != omode) {
2273 		/* allow mode switches for empty-set filters */
2274 		ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2275 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2276 			NULL, 0);
2277 		pmc->sfmode = omode;
2278 	}
2279 
2280 	psl = rtnl_dereference(pmc->sflist);
2281 	if (!add) {
2282 		if (!psl)
2283 			goto done;	/* err = -EADDRNOTAVAIL */
2284 		rv = !0;
2285 		for (i = 0; i < psl->sl_count; i++) {
2286 			rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2287 				sizeof(__be32));
2288 			if (rv == 0)
2289 				break;
2290 		}
2291 		if (rv)		/* source not found */
2292 			goto done;	/* err = -EADDRNOTAVAIL */
2293 
2294 		/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2295 		if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2296 			leavegroup = 1;
2297 			goto done;
2298 		}
2299 
2300 		/* update the interface filter */
2301 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2302 			&mreqs->imr_sourceaddr, 1);
2303 
2304 		for (j = i+1; j < psl->sl_count; j++)
2305 			psl->sl_addr[j-1] = psl->sl_addr[j];
2306 		psl->sl_count--;
2307 		err = 0;
2308 		goto done;
2309 	}
2310 	/* else, add a new source to the filter */
2311 
2312 	if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2313 		err = -ENOBUFS;
2314 		goto done;
2315 	}
2316 	if (!psl || psl->sl_count == psl->sl_max) {
2317 		struct ip_sf_socklist *newpsl;
2318 		int count = IP_SFBLOCK;
2319 
2320 		if (psl)
2321 			count += psl->sl_max;
2322 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2323 		if (!newpsl) {
2324 			err = -ENOBUFS;
2325 			goto done;
2326 		}
2327 		newpsl->sl_max = count;
2328 		newpsl->sl_count = count - IP_SFBLOCK;
2329 		if (psl) {
2330 			for (i = 0; i < psl->sl_count; i++)
2331 				newpsl->sl_addr[i] = psl->sl_addr[i];
2332 			/* decrease mem now to avoid the memleak warning */
2333 			atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2334 			kfree_rcu(psl, rcu);
2335 		}
2336 		rcu_assign_pointer(pmc->sflist, newpsl);
2337 		psl = newpsl;
2338 	}
2339 	rv = 1;	/* > 0 for insert logic below if sl_count is 0 */
2340 	for (i = 0; i < psl->sl_count; i++) {
2341 		rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2342 			sizeof(__be32));
2343 		if (rv == 0)
2344 			break;
2345 	}
2346 	if (rv == 0)		/* address already there is an error */
2347 		goto done;
2348 	for (j = psl->sl_count-1; j >= i; j--)
2349 		psl->sl_addr[j+1] = psl->sl_addr[j];
2350 	psl->sl_addr[i] = mreqs->imr_sourceaddr;
2351 	psl->sl_count++;
2352 	err = 0;
2353 	/* update the interface list */
2354 	ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2355 		&mreqs->imr_sourceaddr, 1);
2356 done:
2357 	if (leavegroup)
2358 		err = ip_mc_leave_group(sk, &imr);
2359 	return err;
2360 }
2361 
2362 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2363 {
2364 	int err = 0;
2365 	struct ip_mreqn	imr;
2366 	__be32 addr = msf->imsf_multiaddr;
2367 	struct ip_mc_socklist *pmc;
2368 	struct in_device *in_dev;
2369 	struct inet_sock *inet = inet_sk(sk);
2370 	struct ip_sf_socklist *newpsl, *psl;
2371 	struct net *net = sock_net(sk);
2372 	int leavegroup = 0;
2373 
2374 	if (!ipv4_is_multicast(addr))
2375 		return -EINVAL;
2376 	if (msf->imsf_fmode != MCAST_INCLUDE &&
2377 	    msf->imsf_fmode != MCAST_EXCLUDE)
2378 		return -EINVAL;
2379 
2380 	ASSERT_RTNL();
2381 
2382 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2383 	imr.imr_address.s_addr = msf->imsf_interface;
2384 	imr.imr_ifindex = ifindex;
2385 	in_dev = ip_mc_find_dev(net, &imr);
2386 
2387 	if (!in_dev) {
2388 		err = -ENODEV;
2389 		goto done;
2390 	}
2391 
2392 	/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2393 	if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2394 		leavegroup = 1;
2395 		goto done;
2396 	}
2397 
2398 	for_each_pmc_rtnl(inet, pmc) {
2399 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2400 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2401 			break;
2402 	}
2403 	if (!pmc) {		/* must have a prior join */
2404 		err = -EINVAL;
2405 		goto done;
2406 	}
2407 	if (msf->imsf_numsrc) {
2408 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2409 							   GFP_KERNEL);
2410 		if (!newpsl) {
2411 			err = -ENOBUFS;
2412 			goto done;
2413 		}
2414 		newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2415 		memcpy(newpsl->sl_addr, msf->imsf_slist,
2416 			msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2417 		err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2418 			msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2419 		if (err) {
2420 			sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2421 			goto done;
2422 		}
2423 	} else {
2424 		newpsl = NULL;
2425 		(void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2426 				     msf->imsf_fmode, 0, NULL, 0);
2427 	}
2428 	psl = rtnl_dereference(pmc->sflist);
2429 	if (psl) {
2430 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2431 			psl->sl_count, psl->sl_addr, 0);
2432 		/* decrease mem now to avoid the memleak warning */
2433 		atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2434 		kfree_rcu(psl, rcu);
2435 	} else
2436 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2437 			0, NULL, 0);
2438 	rcu_assign_pointer(pmc->sflist, newpsl);
2439 	pmc->sfmode = msf->imsf_fmode;
2440 	err = 0;
2441 done:
2442 	if (leavegroup)
2443 		err = ip_mc_leave_group(sk, &imr);
2444 	return err;
2445 }
2446 
2447 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2448 	struct ip_msfilter __user *optval, int __user *optlen)
2449 {
2450 	int err, len, count, copycount;
2451 	struct ip_mreqn	imr;
2452 	__be32 addr = msf->imsf_multiaddr;
2453 	struct ip_mc_socklist *pmc;
2454 	struct in_device *in_dev;
2455 	struct inet_sock *inet = inet_sk(sk);
2456 	struct ip_sf_socklist *psl;
2457 	struct net *net = sock_net(sk);
2458 
2459 	ASSERT_RTNL();
2460 
2461 	if (!ipv4_is_multicast(addr))
2462 		return -EINVAL;
2463 
2464 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2465 	imr.imr_address.s_addr = msf->imsf_interface;
2466 	imr.imr_ifindex = 0;
2467 	in_dev = ip_mc_find_dev(net, &imr);
2468 
2469 	if (!in_dev) {
2470 		err = -ENODEV;
2471 		goto done;
2472 	}
2473 	err = -EADDRNOTAVAIL;
2474 
2475 	for_each_pmc_rtnl(inet, pmc) {
2476 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2477 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2478 			break;
2479 	}
2480 	if (!pmc)		/* must have a prior join */
2481 		goto done;
2482 	msf->imsf_fmode = pmc->sfmode;
2483 	psl = rtnl_dereference(pmc->sflist);
2484 	if (!psl) {
2485 		len = 0;
2486 		count = 0;
2487 	} else {
2488 		count = psl->sl_count;
2489 	}
2490 	copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2491 	len = copycount * sizeof(psl->sl_addr[0]);
2492 	msf->imsf_numsrc = count;
2493 	if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2494 	    copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2495 		return -EFAULT;
2496 	}
2497 	if (len &&
2498 	    copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2499 		return -EFAULT;
2500 	return 0;
2501 done:
2502 	return err;
2503 }
2504 
2505 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2506 	struct group_filter __user *optval, int __user *optlen)
2507 {
2508 	int err, i, count, copycount;
2509 	struct sockaddr_in *psin;
2510 	__be32 addr;
2511 	struct ip_mc_socklist *pmc;
2512 	struct inet_sock *inet = inet_sk(sk);
2513 	struct ip_sf_socklist *psl;
2514 
2515 	ASSERT_RTNL();
2516 
2517 	psin = (struct sockaddr_in *)&gsf->gf_group;
2518 	if (psin->sin_family != AF_INET)
2519 		return -EINVAL;
2520 	addr = psin->sin_addr.s_addr;
2521 	if (!ipv4_is_multicast(addr))
2522 		return -EINVAL;
2523 
2524 	err = -EADDRNOTAVAIL;
2525 
2526 	for_each_pmc_rtnl(inet, pmc) {
2527 		if (pmc->multi.imr_multiaddr.s_addr == addr &&
2528 		    pmc->multi.imr_ifindex == gsf->gf_interface)
2529 			break;
2530 	}
2531 	if (!pmc)		/* must have a prior join */
2532 		goto done;
2533 	gsf->gf_fmode = pmc->sfmode;
2534 	psl = rtnl_dereference(pmc->sflist);
2535 	count = psl ? psl->sl_count : 0;
2536 	copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2537 	gsf->gf_numsrc = count;
2538 	if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2539 	    copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2540 		return -EFAULT;
2541 	}
2542 	for (i = 0; i < copycount; i++) {
2543 		struct sockaddr_storage ss;
2544 
2545 		psin = (struct sockaddr_in *)&ss;
2546 		memset(&ss, 0, sizeof(ss));
2547 		psin->sin_family = AF_INET;
2548 		psin->sin_addr.s_addr = psl->sl_addr[i];
2549 		if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2550 			return -EFAULT;
2551 	}
2552 	return 0;
2553 done:
2554 	return err;
2555 }
2556 
2557 /*
2558  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2559  */
2560 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr, int dif)
2561 {
2562 	struct inet_sock *inet = inet_sk(sk);
2563 	struct ip_mc_socklist *pmc;
2564 	struct ip_sf_socklist *psl;
2565 	int i;
2566 	int ret;
2567 
2568 	ret = 1;
2569 	if (!ipv4_is_multicast(loc_addr))
2570 		goto out;
2571 
2572 	rcu_read_lock();
2573 	for_each_pmc_rcu(inet, pmc) {
2574 		if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2575 		    pmc->multi.imr_ifindex == dif)
2576 			break;
2577 	}
2578 	ret = inet->mc_all;
2579 	if (!pmc)
2580 		goto unlock;
2581 	psl = rcu_dereference(pmc->sflist);
2582 	ret = (pmc->sfmode == MCAST_EXCLUDE);
2583 	if (!psl)
2584 		goto unlock;
2585 
2586 	for (i = 0; i < psl->sl_count; i++) {
2587 		if (psl->sl_addr[i] == rmt_addr)
2588 			break;
2589 	}
2590 	ret = 0;
2591 	if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2592 		goto unlock;
2593 	if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2594 		goto unlock;
2595 	ret = 1;
2596 unlock:
2597 	rcu_read_unlock();
2598 out:
2599 	return ret;
2600 }
2601 
2602 /*
2603  *	A socket is closing.
2604  */
2605 
2606 void ip_mc_drop_socket(struct sock *sk)
2607 {
2608 	struct inet_sock *inet = inet_sk(sk);
2609 	struct ip_mc_socklist *iml;
2610 	struct net *net = sock_net(sk);
2611 
2612 	if (!inet->mc_list)
2613 		return;
2614 
2615 	rtnl_lock();
2616 	while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2617 		struct in_device *in_dev;
2618 
2619 		inet->mc_list = iml->next_rcu;
2620 		in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2621 		(void) ip_mc_leave_src(sk, iml, in_dev);
2622 		if (in_dev)
2623 			ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2624 		/* decrease mem now to avoid the memleak warning */
2625 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2626 		kfree_rcu(iml, rcu);
2627 	}
2628 	rtnl_unlock();
2629 }
2630 
2631 /* called with rcu_read_lock() */
2632 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2633 {
2634 	struct ip_mc_list *im;
2635 	struct ip_mc_list __rcu **mc_hash;
2636 	struct ip_sf_list *psf;
2637 	int rv = 0;
2638 
2639 	mc_hash = rcu_dereference(in_dev->mc_hash);
2640 	if (mc_hash) {
2641 		u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2642 
2643 		for (im = rcu_dereference(mc_hash[hash]);
2644 		     im != NULL;
2645 		     im = rcu_dereference(im->next_hash)) {
2646 			if (im->multiaddr == mc_addr)
2647 				break;
2648 		}
2649 	} else {
2650 		for_each_pmc_rcu(in_dev, im) {
2651 			if (im->multiaddr == mc_addr)
2652 				break;
2653 		}
2654 	}
2655 	if (im && proto == IPPROTO_IGMP) {
2656 		rv = 1;
2657 	} else if (im) {
2658 		if (src_addr) {
2659 			for (psf = im->sources; psf; psf = psf->sf_next) {
2660 				if (psf->sf_inaddr == src_addr)
2661 					break;
2662 			}
2663 			if (psf)
2664 				rv = psf->sf_count[MCAST_INCLUDE] ||
2665 					psf->sf_count[MCAST_EXCLUDE] !=
2666 					im->sfcount[MCAST_EXCLUDE];
2667 			else
2668 				rv = im->sfcount[MCAST_EXCLUDE] != 0;
2669 		} else
2670 			rv = 1; /* unspecified source; tentatively allow */
2671 	}
2672 	return rv;
2673 }
2674 
2675 #if defined(CONFIG_PROC_FS)
2676 struct igmp_mc_iter_state {
2677 	struct seq_net_private p;
2678 	struct net_device *dev;
2679 	struct in_device *in_dev;
2680 };
2681 
2682 #define	igmp_mc_seq_private(seq)	((struct igmp_mc_iter_state *)(seq)->private)
2683 
2684 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2685 {
2686 	struct net *net = seq_file_net(seq);
2687 	struct ip_mc_list *im = NULL;
2688 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2689 
2690 	state->in_dev = NULL;
2691 	for_each_netdev_rcu(net, state->dev) {
2692 		struct in_device *in_dev;
2693 
2694 		in_dev = __in_dev_get_rcu(state->dev);
2695 		if (!in_dev)
2696 			continue;
2697 		im = rcu_dereference(in_dev->mc_list);
2698 		if (im) {
2699 			state->in_dev = in_dev;
2700 			break;
2701 		}
2702 	}
2703 	return im;
2704 }
2705 
2706 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2707 {
2708 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2709 
2710 	im = rcu_dereference(im->next_rcu);
2711 	while (!im) {
2712 		state->dev = next_net_device_rcu(state->dev);
2713 		if (!state->dev) {
2714 			state->in_dev = NULL;
2715 			break;
2716 		}
2717 		state->in_dev = __in_dev_get_rcu(state->dev);
2718 		if (!state->in_dev)
2719 			continue;
2720 		im = rcu_dereference(state->in_dev->mc_list);
2721 	}
2722 	return im;
2723 }
2724 
2725 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2726 {
2727 	struct ip_mc_list *im = igmp_mc_get_first(seq);
2728 	if (im)
2729 		while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2730 			--pos;
2731 	return pos ? NULL : im;
2732 }
2733 
2734 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2735 	__acquires(rcu)
2736 {
2737 	rcu_read_lock();
2738 	return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2739 }
2740 
2741 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2742 {
2743 	struct ip_mc_list *im;
2744 	if (v == SEQ_START_TOKEN)
2745 		im = igmp_mc_get_first(seq);
2746 	else
2747 		im = igmp_mc_get_next(seq, v);
2748 	++*pos;
2749 	return im;
2750 }
2751 
2752 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2753 	__releases(rcu)
2754 {
2755 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2756 
2757 	state->in_dev = NULL;
2758 	state->dev = NULL;
2759 	rcu_read_unlock();
2760 }
2761 
2762 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2763 {
2764 	if (v == SEQ_START_TOKEN)
2765 		seq_puts(seq,
2766 			 "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2767 	else {
2768 		struct ip_mc_list *im = (struct ip_mc_list *)v;
2769 		struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2770 		char   *querier;
2771 		long delta;
2772 
2773 #ifdef CONFIG_IP_MULTICAST
2774 		querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2775 			  IGMP_V2_SEEN(state->in_dev) ? "V2" :
2776 			  "V3";
2777 #else
2778 		querier = "NONE";
2779 #endif
2780 
2781 		if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2782 			seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2783 				   state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2784 		}
2785 
2786 		delta = im->timer.expires - jiffies;
2787 		seq_printf(seq,
2788 			   "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2789 			   im->multiaddr, im->users,
2790 			   im->tm_running,
2791 			   im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2792 			   im->reporter);
2793 	}
2794 	return 0;
2795 }
2796 
2797 static const struct seq_operations igmp_mc_seq_ops = {
2798 	.start	=	igmp_mc_seq_start,
2799 	.next	=	igmp_mc_seq_next,
2800 	.stop	=	igmp_mc_seq_stop,
2801 	.show	=	igmp_mc_seq_show,
2802 };
2803 
2804 static int igmp_mc_seq_open(struct inode *inode, struct file *file)
2805 {
2806 	return seq_open_net(inode, file, &igmp_mc_seq_ops,
2807 			sizeof(struct igmp_mc_iter_state));
2808 }
2809 
2810 static const struct file_operations igmp_mc_seq_fops = {
2811 	.owner		=	THIS_MODULE,
2812 	.open		=	igmp_mc_seq_open,
2813 	.read		=	seq_read,
2814 	.llseek		=	seq_lseek,
2815 	.release	=	seq_release_net,
2816 };
2817 
2818 struct igmp_mcf_iter_state {
2819 	struct seq_net_private p;
2820 	struct net_device *dev;
2821 	struct in_device *idev;
2822 	struct ip_mc_list *im;
2823 };
2824 
2825 #define igmp_mcf_seq_private(seq)	((struct igmp_mcf_iter_state *)(seq)->private)
2826 
2827 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2828 {
2829 	struct net *net = seq_file_net(seq);
2830 	struct ip_sf_list *psf = NULL;
2831 	struct ip_mc_list *im = NULL;
2832 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2833 
2834 	state->idev = NULL;
2835 	state->im = NULL;
2836 	for_each_netdev_rcu(net, state->dev) {
2837 		struct in_device *idev;
2838 		idev = __in_dev_get_rcu(state->dev);
2839 		if (unlikely(!idev))
2840 			continue;
2841 		im = rcu_dereference(idev->mc_list);
2842 		if (likely(im)) {
2843 			spin_lock_bh(&im->lock);
2844 			psf = im->sources;
2845 			if (likely(psf)) {
2846 				state->im = im;
2847 				state->idev = idev;
2848 				break;
2849 			}
2850 			spin_unlock_bh(&im->lock);
2851 		}
2852 	}
2853 	return psf;
2854 }
2855 
2856 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2857 {
2858 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2859 
2860 	psf = psf->sf_next;
2861 	while (!psf) {
2862 		spin_unlock_bh(&state->im->lock);
2863 		state->im = state->im->next;
2864 		while (!state->im) {
2865 			state->dev = next_net_device_rcu(state->dev);
2866 			if (!state->dev) {
2867 				state->idev = NULL;
2868 				goto out;
2869 			}
2870 			state->idev = __in_dev_get_rcu(state->dev);
2871 			if (!state->idev)
2872 				continue;
2873 			state->im = rcu_dereference(state->idev->mc_list);
2874 		}
2875 		if (!state->im)
2876 			break;
2877 		spin_lock_bh(&state->im->lock);
2878 		psf = state->im->sources;
2879 	}
2880 out:
2881 	return psf;
2882 }
2883 
2884 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2885 {
2886 	struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2887 	if (psf)
2888 		while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2889 			--pos;
2890 	return pos ? NULL : psf;
2891 }
2892 
2893 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2894 	__acquires(rcu)
2895 {
2896 	rcu_read_lock();
2897 	return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2898 }
2899 
2900 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2901 {
2902 	struct ip_sf_list *psf;
2903 	if (v == SEQ_START_TOKEN)
2904 		psf = igmp_mcf_get_first(seq);
2905 	else
2906 		psf = igmp_mcf_get_next(seq, v);
2907 	++*pos;
2908 	return psf;
2909 }
2910 
2911 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2912 	__releases(rcu)
2913 {
2914 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2915 	if (likely(state->im)) {
2916 		spin_unlock_bh(&state->im->lock);
2917 		state->im = NULL;
2918 	}
2919 	state->idev = NULL;
2920 	state->dev = NULL;
2921 	rcu_read_unlock();
2922 }
2923 
2924 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2925 {
2926 	struct ip_sf_list *psf = (struct ip_sf_list *)v;
2927 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2928 
2929 	if (v == SEQ_START_TOKEN) {
2930 		seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2931 	} else {
2932 		seq_printf(seq,
2933 			   "%3d %6.6s 0x%08x "
2934 			   "0x%08x %6lu %6lu\n",
2935 			   state->dev->ifindex, state->dev->name,
2936 			   ntohl(state->im->multiaddr),
2937 			   ntohl(psf->sf_inaddr),
2938 			   psf->sf_count[MCAST_INCLUDE],
2939 			   psf->sf_count[MCAST_EXCLUDE]);
2940 	}
2941 	return 0;
2942 }
2943 
2944 static const struct seq_operations igmp_mcf_seq_ops = {
2945 	.start	=	igmp_mcf_seq_start,
2946 	.next	=	igmp_mcf_seq_next,
2947 	.stop	=	igmp_mcf_seq_stop,
2948 	.show	=	igmp_mcf_seq_show,
2949 };
2950 
2951 static int igmp_mcf_seq_open(struct inode *inode, struct file *file)
2952 {
2953 	return seq_open_net(inode, file, &igmp_mcf_seq_ops,
2954 			sizeof(struct igmp_mcf_iter_state));
2955 }
2956 
2957 static const struct file_operations igmp_mcf_seq_fops = {
2958 	.owner		=	THIS_MODULE,
2959 	.open		=	igmp_mcf_seq_open,
2960 	.read		=	seq_read,
2961 	.llseek		=	seq_lseek,
2962 	.release	=	seq_release_net,
2963 };
2964 
2965 static int __net_init igmp_net_init(struct net *net)
2966 {
2967 	struct proc_dir_entry *pde;
2968 	int err;
2969 
2970 	pde = proc_create("igmp", S_IRUGO, net->proc_net, &igmp_mc_seq_fops);
2971 	if (!pde)
2972 		goto out_igmp;
2973 	pde = proc_create("mcfilter", S_IRUGO, net->proc_net,
2974 			  &igmp_mcf_seq_fops);
2975 	if (!pde)
2976 		goto out_mcfilter;
2977 	err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
2978 				   SOCK_DGRAM, 0, net);
2979 	if (err < 0) {
2980 		pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
2981 		       err);
2982 		goto out_sock;
2983 	}
2984 
2985 	return 0;
2986 
2987 out_sock:
2988 	remove_proc_entry("mcfilter", net->proc_net);
2989 out_mcfilter:
2990 	remove_proc_entry("igmp", net->proc_net);
2991 out_igmp:
2992 	return -ENOMEM;
2993 }
2994 
2995 static void __net_exit igmp_net_exit(struct net *net)
2996 {
2997 	remove_proc_entry("mcfilter", net->proc_net);
2998 	remove_proc_entry("igmp", net->proc_net);
2999 	inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3000 }
3001 
3002 static struct pernet_operations igmp_net_ops = {
3003 	.init = igmp_net_init,
3004 	.exit = igmp_net_exit,
3005 };
3006 #endif
3007 
3008 static int igmp_netdev_event(struct notifier_block *this,
3009 			     unsigned long event, void *ptr)
3010 {
3011 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3012 	struct in_device *in_dev;
3013 
3014 	switch (event) {
3015 	case NETDEV_RESEND_IGMP:
3016 		in_dev = __in_dev_get_rtnl(dev);
3017 		if (in_dev)
3018 			ip_mc_rejoin_groups(in_dev);
3019 		break;
3020 	default:
3021 		break;
3022 	}
3023 	return NOTIFY_DONE;
3024 }
3025 
3026 static struct notifier_block igmp_notifier = {
3027 	.notifier_call = igmp_netdev_event,
3028 };
3029 
3030 int __init igmp_mc_init(void)
3031 {
3032 #if defined(CONFIG_PROC_FS)
3033 	int err;
3034 
3035 	err = register_pernet_subsys(&igmp_net_ops);
3036 	if (err)
3037 		return err;
3038 	err = register_netdevice_notifier(&igmp_notifier);
3039 	if (err)
3040 		goto reg_notif_fail;
3041 	return 0;
3042 
3043 reg_notif_fail:
3044 	unregister_pernet_subsys(&igmp_net_ops);
3045 	return err;
3046 #else
3047 	return register_netdevice_notifier(&igmp_notifier);
3048 #endif
3049 }
3050