xref: /openbmc/linux/net/ipv4/igmp.c (revision b5476022bbada3764609368f03329ca287528dc8)
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, unsigned int mtu)
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, mtu);
415 		if (!skb)
416 			return NULL;
417 	}
418 	pgr = skb_put(skb, sizeof(struct igmpv3_grec));
419 	pgr->grec_type = type;
420 	pgr->grec_auxwords = 0;
421 	pgr->grec_nsrcs = 0;
422 	pgr->grec_mca = pmc->multiaddr;
423 	pih = igmpv3_report_hdr(skb);
424 	pih->ngrec = htons(ntohs(pih->ngrec)+1);
425 	*ppgr = pgr;
426 	return skb;
427 }
428 
429 #define AVAILABLE(skb)	((skb) ? skb_availroom(skb) : 0)
430 
431 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
432 	int type, int gdeleted, int sdeleted)
433 {
434 	struct net_device *dev = pmc->interface->dev;
435 	struct net *net = dev_net(dev);
436 	struct igmpv3_report *pih;
437 	struct igmpv3_grec *pgr = NULL;
438 	struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
439 	int scount, stotal, first, isquery, truncate;
440 	unsigned int mtu;
441 
442 	if (pmc->multiaddr == IGMP_ALL_HOSTS)
443 		return skb;
444 	if (ipv4_is_local_multicast(pmc->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
445 		return skb;
446 
447 	mtu = READ_ONCE(dev->mtu);
448 	if (mtu < IPV4_MIN_MTU)
449 		return skb;
450 
451 	isquery = type == IGMPV3_MODE_IS_INCLUDE ||
452 		  type == IGMPV3_MODE_IS_EXCLUDE;
453 	truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
454 		    type == IGMPV3_CHANGE_TO_EXCLUDE;
455 
456 	stotal = scount = 0;
457 
458 	psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
459 
460 	if (!*psf_list)
461 		goto empty_source;
462 
463 	pih = skb ? igmpv3_report_hdr(skb) : NULL;
464 
465 	/* EX and TO_EX get a fresh packet, if needed */
466 	if (truncate) {
467 		if (pih && pih->ngrec &&
468 		    AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
469 			if (skb)
470 				igmpv3_sendpack(skb);
471 			skb = igmpv3_newpack(dev, mtu);
472 		}
473 	}
474 	first = 1;
475 	psf_prev = NULL;
476 	for (psf = *psf_list; psf; psf = psf_next) {
477 		__be32 *psrc;
478 
479 		psf_next = psf->sf_next;
480 
481 		if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
482 			psf_prev = psf;
483 			continue;
484 		}
485 
486 		/* Based on RFC3376 5.1. Should not send source-list change
487 		 * records when there is a filter mode change.
488 		 */
489 		if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
490 		     (!gdeleted && pmc->crcount)) &&
491 		    (type == IGMPV3_ALLOW_NEW_SOURCES ||
492 		     type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
493 			goto decrease_sf_crcount;
494 
495 		/* clear marks on query responses */
496 		if (isquery)
497 			psf->sf_gsresp = 0;
498 
499 		if (AVAILABLE(skb) < sizeof(__be32) +
500 		    first*sizeof(struct igmpv3_grec)) {
501 			if (truncate && !first)
502 				break;	 /* truncate these */
503 			if (pgr)
504 				pgr->grec_nsrcs = htons(scount);
505 			if (skb)
506 				igmpv3_sendpack(skb);
507 			skb = igmpv3_newpack(dev, mtu);
508 			first = 1;
509 			scount = 0;
510 		}
511 		if (first) {
512 			skb = add_grhead(skb, pmc, type, &pgr, mtu);
513 			first = 0;
514 		}
515 		if (!skb)
516 			return NULL;
517 		psrc = skb_put(skb, sizeof(__be32));
518 		*psrc = psf->sf_inaddr;
519 		scount++; stotal++;
520 		if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
521 		     type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
522 decrease_sf_crcount:
523 			psf->sf_crcount--;
524 			if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
525 				if (psf_prev)
526 					psf_prev->sf_next = psf->sf_next;
527 				else
528 					*psf_list = psf->sf_next;
529 				kfree(psf);
530 				continue;
531 			}
532 		}
533 		psf_prev = psf;
534 	}
535 
536 empty_source:
537 	if (!stotal) {
538 		if (type == IGMPV3_ALLOW_NEW_SOURCES ||
539 		    type == IGMPV3_BLOCK_OLD_SOURCES)
540 			return skb;
541 		if (pmc->crcount || isquery) {
542 			/* make sure we have room for group header */
543 			if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
544 				igmpv3_sendpack(skb);
545 				skb = NULL; /* add_grhead will get a new one */
546 			}
547 			skb = add_grhead(skb, pmc, type, &pgr, mtu);
548 		}
549 	}
550 	if (pgr)
551 		pgr->grec_nsrcs = htons(scount);
552 
553 	if (isquery)
554 		pmc->gsquery = 0;	/* clear query state on report */
555 	return skb;
556 }
557 
558 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
559 {
560 	struct sk_buff *skb = NULL;
561 	struct net *net = dev_net(in_dev->dev);
562 	int type;
563 
564 	if (!pmc) {
565 		rcu_read_lock();
566 		for_each_pmc_rcu(in_dev, pmc) {
567 			if (pmc->multiaddr == IGMP_ALL_HOSTS)
568 				continue;
569 			if (ipv4_is_local_multicast(pmc->multiaddr) &&
570 			     !net->ipv4.sysctl_igmp_llm_reports)
571 				continue;
572 			spin_lock_bh(&pmc->lock);
573 			if (pmc->sfcount[MCAST_EXCLUDE])
574 				type = IGMPV3_MODE_IS_EXCLUDE;
575 			else
576 				type = IGMPV3_MODE_IS_INCLUDE;
577 			skb = add_grec(skb, pmc, type, 0, 0);
578 			spin_unlock_bh(&pmc->lock);
579 		}
580 		rcu_read_unlock();
581 	} else {
582 		spin_lock_bh(&pmc->lock);
583 		if (pmc->sfcount[MCAST_EXCLUDE])
584 			type = IGMPV3_MODE_IS_EXCLUDE;
585 		else
586 			type = IGMPV3_MODE_IS_INCLUDE;
587 		skb = add_grec(skb, pmc, type, 0, 0);
588 		spin_unlock_bh(&pmc->lock);
589 	}
590 	if (!skb)
591 		return 0;
592 	return igmpv3_sendpack(skb);
593 }
594 
595 /*
596  * remove zero-count source records from a source filter list
597  */
598 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
599 {
600 	struct ip_sf_list *psf_prev, *psf_next, *psf;
601 
602 	psf_prev = NULL;
603 	for (psf = *ppsf; psf; psf = psf_next) {
604 		psf_next = psf->sf_next;
605 		if (psf->sf_crcount == 0) {
606 			if (psf_prev)
607 				psf_prev->sf_next = psf->sf_next;
608 			else
609 				*ppsf = psf->sf_next;
610 			kfree(psf);
611 		} else
612 			psf_prev = psf;
613 	}
614 }
615 
616 static void igmpv3_send_cr(struct in_device *in_dev)
617 {
618 	struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
619 	struct sk_buff *skb = NULL;
620 	int type, dtype;
621 
622 	rcu_read_lock();
623 	spin_lock_bh(&in_dev->mc_tomb_lock);
624 
625 	/* deleted MCA's */
626 	pmc_prev = NULL;
627 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
628 		pmc_next = pmc->next;
629 		if (pmc->sfmode == MCAST_INCLUDE) {
630 			type = IGMPV3_BLOCK_OLD_SOURCES;
631 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
632 			skb = add_grec(skb, pmc, type, 1, 0);
633 			skb = add_grec(skb, pmc, dtype, 1, 1);
634 		}
635 		if (pmc->crcount) {
636 			if (pmc->sfmode == MCAST_EXCLUDE) {
637 				type = IGMPV3_CHANGE_TO_INCLUDE;
638 				skb = add_grec(skb, pmc, type, 1, 0);
639 			}
640 			pmc->crcount--;
641 			if (pmc->crcount == 0) {
642 				igmpv3_clear_zeros(&pmc->tomb);
643 				igmpv3_clear_zeros(&pmc->sources);
644 			}
645 		}
646 		if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
647 			if (pmc_prev)
648 				pmc_prev->next = pmc_next;
649 			else
650 				in_dev->mc_tomb = pmc_next;
651 			in_dev_put(pmc->interface);
652 			kfree(pmc);
653 		} else
654 			pmc_prev = pmc;
655 	}
656 	spin_unlock_bh(&in_dev->mc_tomb_lock);
657 
658 	/* change recs */
659 	for_each_pmc_rcu(in_dev, pmc) {
660 		spin_lock_bh(&pmc->lock);
661 		if (pmc->sfcount[MCAST_EXCLUDE]) {
662 			type = IGMPV3_BLOCK_OLD_SOURCES;
663 			dtype = IGMPV3_ALLOW_NEW_SOURCES;
664 		} else {
665 			type = IGMPV3_ALLOW_NEW_SOURCES;
666 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
667 		}
668 		skb = add_grec(skb, pmc, type, 0, 0);
669 		skb = add_grec(skb, pmc, dtype, 0, 1);	/* deleted sources */
670 
671 		/* filter mode changes */
672 		if (pmc->crcount) {
673 			if (pmc->sfmode == MCAST_EXCLUDE)
674 				type = IGMPV3_CHANGE_TO_EXCLUDE;
675 			else
676 				type = IGMPV3_CHANGE_TO_INCLUDE;
677 			skb = add_grec(skb, pmc, type, 0, 0);
678 			pmc->crcount--;
679 		}
680 		spin_unlock_bh(&pmc->lock);
681 	}
682 	rcu_read_unlock();
683 
684 	if (!skb)
685 		return;
686 	(void) igmpv3_sendpack(skb);
687 }
688 
689 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
690 	int type)
691 {
692 	struct sk_buff *skb;
693 	struct iphdr *iph;
694 	struct igmphdr *ih;
695 	struct rtable *rt;
696 	struct net_device *dev = in_dev->dev;
697 	struct net *net = dev_net(dev);
698 	__be32	group = pmc ? pmc->multiaddr : 0;
699 	struct flowi4 fl4;
700 	__be32	dst;
701 	int hlen, tlen;
702 
703 	if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
704 		return igmpv3_send_report(in_dev, pmc);
705 
706 	if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
707 		return 0;
708 
709 	if (type == IGMP_HOST_LEAVE_MESSAGE)
710 		dst = IGMP_ALL_ROUTER;
711 	else
712 		dst = group;
713 
714 	rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
715 				   0, 0,
716 				   IPPROTO_IGMP, 0, dev->ifindex);
717 	if (IS_ERR(rt))
718 		return -1;
719 
720 	hlen = LL_RESERVED_SPACE(dev);
721 	tlen = dev->needed_tailroom;
722 	skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
723 	if (!skb) {
724 		ip_rt_put(rt);
725 		return -1;
726 	}
727 	skb->priority = TC_PRIO_CONTROL;
728 
729 	skb_dst_set(skb, &rt->dst);
730 
731 	skb_reserve(skb, hlen);
732 
733 	skb_reset_network_header(skb);
734 	iph = ip_hdr(skb);
735 	skb_put(skb, sizeof(struct iphdr) + 4);
736 
737 	iph->version  = 4;
738 	iph->ihl      = (sizeof(struct iphdr)+4)>>2;
739 	iph->tos      = 0xc0;
740 	iph->frag_off = htons(IP_DF);
741 	iph->ttl      = 1;
742 	iph->daddr    = dst;
743 	iph->saddr    = fl4.saddr;
744 	iph->protocol = IPPROTO_IGMP;
745 	ip_select_ident(net, skb, NULL);
746 	((u8 *)&iph[1])[0] = IPOPT_RA;
747 	((u8 *)&iph[1])[1] = 4;
748 	((u8 *)&iph[1])[2] = 0;
749 	((u8 *)&iph[1])[3] = 0;
750 
751 	ih = skb_put(skb, sizeof(struct igmphdr));
752 	ih->type = type;
753 	ih->code = 0;
754 	ih->csum = 0;
755 	ih->group = group;
756 	ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
757 
758 	return ip_local_out(net, skb->sk, skb);
759 }
760 
761 static void igmp_gq_timer_expire(struct timer_list *t)
762 {
763 	struct in_device *in_dev = from_timer(in_dev, t, mr_gq_timer);
764 
765 	in_dev->mr_gq_running = 0;
766 	igmpv3_send_report(in_dev, NULL);
767 	in_dev_put(in_dev);
768 }
769 
770 static void igmp_ifc_timer_expire(struct timer_list *t)
771 {
772 	struct in_device *in_dev = from_timer(in_dev, t, mr_ifc_timer);
773 
774 	igmpv3_send_cr(in_dev);
775 	if (in_dev->mr_ifc_count) {
776 		in_dev->mr_ifc_count--;
777 		igmp_ifc_start_timer(in_dev,
778 				     unsolicited_report_interval(in_dev));
779 	}
780 	in_dev_put(in_dev);
781 }
782 
783 static void igmp_ifc_event(struct in_device *in_dev)
784 {
785 	struct net *net = dev_net(in_dev->dev);
786 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
787 		return;
788 	in_dev->mr_ifc_count = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
789 	igmp_ifc_start_timer(in_dev, 1);
790 }
791 
792 
793 static void igmp_timer_expire(struct timer_list *t)
794 {
795 	struct ip_mc_list *im = from_timer(im, t, timer);
796 	struct in_device *in_dev = im->interface;
797 
798 	spin_lock(&im->lock);
799 	im->tm_running = 0;
800 
801 	if (im->unsolicit_count) {
802 		im->unsolicit_count--;
803 		igmp_start_timer(im, unsolicited_report_interval(in_dev));
804 	}
805 	im->reporter = 1;
806 	spin_unlock(&im->lock);
807 
808 	if (IGMP_V1_SEEN(in_dev))
809 		igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
810 	else if (IGMP_V2_SEEN(in_dev))
811 		igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
812 	else
813 		igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
814 
815 	ip_ma_put(im);
816 }
817 
818 /* mark EXCLUDE-mode sources */
819 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
820 {
821 	struct ip_sf_list *psf;
822 	int i, scount;
823 
824 	scount = 0;
825 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
826 		if (scount == nsrcs)
827 			break;
828 		for (i = 0; i < nsrcs; i++) {
829 			/* skip inactive filters */
830 			if (psf->sf_count[MCAST_INCLUDE] ||
831 			    pmc->sfcount[MCAST_EXCLUDE] !=
832 			    psf->sf_count[MCAST_EXCLUDE])
833 				break;
834 			if (srcs[i] == psf->sf_inaddr) {
835 				scount++;
836 				break;
837 			}
838 		}
839 	}
840 	pmc->gsquery = 0;
841 	if (scount == nsrcs)	/* all sources excluded */
842 		return 0;
843 	return 1;
844 }
845 
846 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
847 {
848 	struct ip_sf_list *psf;
849 	int i, scount;
850 
851 	if (pmc->sfmode == MCAST_EXCLUDE)
852 		return igmp_xmarksources(pmc, nsrcs, srcs);
853 
854 	/* mark INCLUDE-mode sources */
855 	scount = 0;
856 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
857 		if (scount == nsrcs)
858 			break;
859 		for (i = 0; i < nsrcs; i++)
860 			if (srcs[i] == psf->sf_inaddr) {
861 				psf->sf_gsresp = 1;
862 				scount++;
863 				break;
864 			}
865 	}
866 	if (!scount) {
867 		pmc->gsquery = 0;
868 		return 0;
869 	}
870 	pmc->gsquery = 1;
871 	return 1;
872 }
873 
874 /* return true if packet was dropped */
875 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
876 {
877 	struct ip_mc_list *im;
878 	struct net *net = dev_net(in_dev->dev);
879 
880 	/* Timers are only set for non-local groups */
881 
882 	if (group == IGMP_ALL_HOSTS)
883 		return false;
884 	if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
885 		return false;
886 
887 	rcu_read_lock();
888 	for_each_pmc_rcu(in_dev, im) {
889 		if (im->multiaddr == group) {
890 			igmp_stop_timer(im);
891 			break;
892 		}
893 	}
894 	rcu_read_unlock();
895 	return false;
896 }
897 
898 /* return true if packet was dropped */
899 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
900 	int len)
901 {
902 	struct igmphdr 		*ih = igmp_hdr(skb);
903 	struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
904 	struct ip_mc_list	*im;
905 	__be32			group = ih->group;
906 	int			max_delay;
907 	int			mark = 0;
908 	struct net		*net = dev_net(in_dev->dev);
909 
910 
911 	if (len == 8) {
912 		if (ih->code == 0) {
913 			/* Alas, old v1 router presents here. */
914 
915 			max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
916 			in_dev->mr_v1_seen = jiffies +
917 				IGMP_V1_ROUTER_PRESENT_TIMEOUT;
918 			group = 0;
919 		} else {
920 			/* v2 router present */
921 			max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
922 			in_dev->mr_v2_seen = jiffies +
923 				IGMP_V2_ROUTER_PRESENT_TIMEOUT;
924 		}
925 		/* cancel the interface change timer */
926 		in_dev->mr_ifc_count = 0;
927 		if (del_timer(&in_dev->mr_ifc_timer))
928 			__in_dev_put(in_dev);
929 		/* clear deleted report items */
930 		igmpv3_clear_delrec(in_dev);
931 	} else if (len < 12) {
932 		return true;	/* ignore bogus packet; freed by caller */
933 	} else if (IGMP_V1_SEEN(in_dev)) {
934 		/* This is a v3 query with v1 queriers present */
935 		max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
936 		group = 0;
937 	} else if (IGMP_V2_SEEN(in_dev)) {
938 		/* this is a v3 query with v2 queriers present;
939 		 * Interpretation of the max_delay code is problematic here.
940 		 * A real v2 host would use ih_code directly, while v3 has a
941 		 * different encoding. We use the v3 encoding as more likely
942 		 * to be intended in a v3 query.
943 		 */
944 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
945 		if (!max_delay)
946 			max_delay = 1;	/* can't mod w/ 0 */
947 	} else { /* v3 */
948 		if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
949 			return true;
950 
951 		ih3 = igmpv3_query_hdr(skb);
952 		if (ih3->nsrcs) {
953 			if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
954 					   + ntohs(ih3->nsrcs)*sizeof(__be32)))
955 				return true;
956 			ih3 = igmpv3_query_hdr(skb);
957 		}
958 
959 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
960 		if (!max_delay)
961 			max_delay = 1;	/* can't mod w/ 0 */
962 		in_dev->mr_maxdelay = max_delay;
963 		if (ih3->qrv)
964 			in_dev->mr_qrv = ih3->qrv;
965 		if (!group) { /* general query */
966 			if (ih3->nsrcs)
967 				return true;	/* no sources allowed */
968 			igmp_gq_start_timer(in_dev);
969 			return false;
970 		}
971 		/* mark sources to include, if group & source-specific */
972 		mark = ih3->nsrcs != 0;
973 	}
974 
975 	/*
976 	 * - Start the timers in all of our membership records
977 	 *   that the query applies to for the interface on
978 	 *   which the query arrived excl. those that belong
979 	 *   to a "local" group (224.0.0.X)
980 	 * - For timers already running check if they need to
981 	 *   be reset.
982 	 * - Use the igmp->igmp_code field as the maximum
983 	 *   delay possible
984 	 */
985 	rcu_read_lock();
986 	for_each_pmc_rcu(in_dev, im) {
987 		int changed;
988 
989 		if (group && group != im->multiaddr)
990 			continue;
991 		if (im->multiaddr == IGMP_ALL_HOSTS)
992 			continue;
993 		if (ipv4_is_local_multicast(im->multiaddr) &&
994 		    !net->ipv4.sysctl_igmp_llm_reports)
995 			continue;
996 		spin_lock_bh(&im->lock);
997 		if (im->tm_running)
998 			im->gsquery = im->gsquery && mark;
999 		else
1000 			im->gsquery = mark;
1001 		changed = !im->gsquery ||
1002 			igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1003 		spin_unlock_bh(&im->lock);
1004 		if (changed)
1005 			igmp_mod_timer(im, max_delay);
1006 	}
1007 	rcu_read_unlock();
1008 	return false;
1009 }
1010 
1011 /* called in rcu_read_lock() section */
1012 int igmp_rcv(struct sk_buff *skb)
1013 {
1014 	/* This basically follows the spec line by line -- see RFC1112 */
1015 	struct igmphdr *ih;
1016 	struct net_device *dev = skb->dev;
1017 	struct in_device *in_dev;
1018 	int len = skb->len;
1019 	bool dropped = true;
1020 
1021 	if (netif_is_l3_master(dev)) {
1022 		dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1023 		if (!dev)
1024 			goto drop;
1025 	}
1026 
1027 	in_dev = __in_dev_get_rcu(dev);
1028 	if (!in_dev)
1029 		goto drop;
1030 
1031 	if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1032 		goto drop;
1033 
1034 	if (skb_checksum_simple_validate(skb))
1035 		goto drop;
1036 
1037 	ih = igmp_hdr(skb);
1038 	switch (ih->type) {
1039 	case IGMP_HOST_MEMBERSHIP_QUERY:
1040 		dropped = igmp_heard_query(in_dev, skb, len);
1041 		break;
1042 	case IGMP_HOST_MEMBERSHIP_REPORT:
1043 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
1044 		/* Is it our report looped back? */
1045 		if (rt_is_output_route(skb_rtable(skb)))
1046 			break;
1047 		/* don't rely on MC router hearing unicast reports */
1048 		if (skb->pkt_type == PACKET_MULTICAST ||
1049 		    skb->pkt_type == PACKET_BROADCAST)
1050 			dropped = igmp_heard_report(in_dev, ih->group);
1051 		break;
1052 	case IGMP_PIM:
1053 #ifdef CONFIG_IP_PIMSM_V1
1054 		return pim_rcv_v1(skb);
1055 #endif
1056 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
1057 	case IGMP_DVMRP:
1058 	case IGMP_TRACE:
1059 	case IGMP_HOST_LEAVE_MESSAGE:
1060 	case IGMP_MTRACE:
1061 	case IGMP_MTRACE_RESP:
1062 		break;
1063 	default:
1064 		break;
1065 	}
1066 
1067 drop:
1068 	if (dropped)
1069 		kfree_skb(skb);
1070 	else
1071 		consume_skb(skb);
1072 	return 0;
1073 }
1074 
1075 #endif
1076 
1077 
1078 /*
1079  *	Add a filter to a device
1080  */
1081 
1082 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1083 {
1084 	char buf[MAX_ADDR_LEN];
1085 	struct net_device *dev = in_dev->dev;
1086 
1087 	/* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1088 	   We will get multicast token leakage, when IFF_MULTICAST
1089 	   is changed. This check should be done in ndo_set_rx_mode
1090 	   routine. Something sort of:
1091 	   if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1092 	   --ANK
1093 	   */
1094 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1095 		dev_mc_add(dev, buf);
1096 }
1097 
1098 /*
1099  *	Remove a filter from a device
1100  */
1101 
1102 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1103 {
1104 	char buf[MAX_ADDR_LEN];
1105 	struct net_device *dev = in_dev->dev;
1106 
1107 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1108 		dev_mc_del(dev, buf);
1109 }
1110 
1111 #ifdef CONFIG_IP_MULTICAST
1112 /*
1113  * deleted ip_mc_list manipulation
1114  */
1115 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1116 {
1117 	struct ip_mc_list *pmc;
1118 	struct net *net = dev_net(in_dev->dev);
1119 
1120 	/* this is an "ip_mc_list" for convenience; only the fields below
1121 	 * are actually used. In particular, the refcnt and users are not
1122 	 * used for management of the delete list. Using the same structure
1123 	 * for deleted items allows change reports to use common code with
1124 	 * non-deleted or query-response MCA's.
1125 	 */
1126 	pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1127 	if (!pmc)
1128 		return;
1129 	spin_lock_init(&pmc->lock);
1130 	spin_lock_bh(&im->lock);
1131 	pmc->interface = im->interface;
1132 	in_dev_hold(in_dev);
1133 	pmc->multiaddr = im->multiaddr;
1134 	pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1135 	pmc->sfmode = im->sfmode;
1136 	if (pmc->sfmode == MCAST_INCLUDE) {
1137 		struct ip_sf_list *psf;
1138 
1139 		pmc->tomb = im->tomb;
1140 		pmc->sources = im->sources;
1141 		im->tomb = im->sources = NULL;
1142 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1143 			psf->sf_crcount = pmc->crcount;
1144 	}
1145 	spin_unlock_bh(&im->lock);
1146 
1147 	spin_lock_bh(&in_dev->mc_tomb_lock);
1148 	pmc->next = in_dev->mc_tomb;
1149 	in_dev->mc_tomb = pmc;
1150 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1151 }
1152 
1153 /*
1154  * restore ip_mc_list deleted records
1155  */
1156 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1157 {
1158 	struct ip_mc_list *pmc, *pmc_prev;
1159 	struct ip_sf_list *psf;
1160 	struct net *net = dev_net(in_dev->dev);
1161 	__be32 multiaddr = im->multiaddr;
1162 
1163 	spin_lock_bh(&in_dev->mc_tomb_lock);
1164 	pmc_prev = NULL;
1165 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1166 		if (pmc->multiaddr == multiaddr)
1167 			break;
1168 		pmc_prev = pmc;
1169 	}
1170 	if (pmc) {
1171 		if (pmc_prev)
1172 			pmc_prev->next = pmc->next;
1173 		else
1174 			in_dev->mc_tomb = pmc->next;
1175 	}
1176 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1177 
1178 	spin_lock_bh(&im->lock);
1179 	if (pmc) {
1180 		im->interface = pmc->interface;
1181 		im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1182 		im->sfmode = pmc->sfmode;
1183 		if (pmc->sfmode == MCAST_INCLUDE) {
1184 			im->tomb = pmc->tomb;
1185 			im->sources = pmc->sources;
1186 			for (psf = im->sources; psf; psf = psf->sf_next)
1187 				psf->sf_crcount = im->crcount;
1188 		}
1189 		in_dev_put(pmc->interface);
1190 		kfree(pmc);
1191 	}
1192 	spin_unlock_bh(&im->lock);
1193 }
1194 
1195 /*
1196  * flush ip_mc_list deleted records
1197  */
1198 static void igmpv3_clear_delrec(struct in_device *in_dev)
1199 {
1200 	struct ip_mc_list *pmc, *nextpmc;
1201 
1202 	spin_lock_bh(&in_dev->mc_tomb_lock);
1203 	pmc = in_dev->mc_tomb;
1204 	in_dev->mc_tomb = NULL;
1205 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1206 
1207 	for (; pmc; pmc = nextpmc) {
1208 		nextpmc = pmc->next;
1209 		ip_mc_clear_src(pmc);
1210 		in_dev_put(pmc->interface);
1211 		kfree(pmc);
1212 	}
1213 	/* clear dead sources, too */
1214 	rcu_read_lock();
1215 	for_each_pmc_rcu(in_dev, pmc) {
1216 		struct ip_sf_list *psf, *psf_next;
1217 
1218 		spin_lock_bh(&pmc->lock);
1219 		psf = pmc->tomb;
1220 		pmc->tomb = NULL;
1221 		spin_unlock_bh(&pmc->lock);
1222 		for (; psf; psf = psf_next) {
1223 			psf_next = psf->sf_next;
1224 			kfree(psf);
1225 		}
1226 	}
1227 	rcu_read_unlock();
1228 }
1229 #endif
1230 
1231 static void igmp_group_dropped(struct ip_mc_list *im)
1232 {
1233 	struct in_device *in_dev = im->interface;
1234 #ifdef CONFIG_IP_MULTICAST
1235 	struct net *net = dev_net(in_dev->dev);
1236 	int reporter;
1237 #endif
1238 
1239 	if (im->loaded) {
1240 		im->loaded = 0;
1241 		ip_mc_filter_del(in_dev, im->multiaddr);
1242 	}
1243 
1244 #ifdef CONFIG_IP_MULTICAST
1245 	if (im->multiaddr == IGMP_ALL_HOSTS)
1246 		return;
1247 	if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1248 		return;
1249 
1250 	reporter = im->reporter;
1251 	igmp_stop_timer(im);
1252 
1253 	if (!in_dev->dead) {
1254 		if (IGMP_V1_SEEN(in_dev))
1255 			return;
1256 		if (IGMP_V2_SEEN(in_dev)) {
1257 			if (reporter)
1258 				igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1259 			return;
1260 		}
1261 		/* IGMPv3 */
1262 		igmpv3_add_delrec(in_dev, im);
1263 
1264 		igmp_ifc_event(in_dev);
1265 	}
1266 #endif
1267 }
1268 
1269 static void igmp_group_added(struct ip_mc_list *im)
1270 {
1271 	struct in_device *in_dev = im->interface;
1272 #ifdef CONFIG_IP_MULTICAST
1273 	struct net *net = dev_net(in_dev->dev);
1274 #endif
1275 
1276 	if (im->loaded == 0) {
1277 		im->loaded = 1;
1278 		ip_mc_filter_add(in_dev, im->multiaddr);
1279 	}
1280 
1281 #ifdef CONFIG_IP_MULTICAST
1282 	if (im->multiaddr == IGMP_ALL_HOSTS)
1283 		return;
1284 	if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1285 		return;
1286 
1287 	if (in_dev->dead)
1288 		return;
1289 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1290 		spin_lock_bh(&im->lock);
1291 		igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1292 		spin_unlock_bh(&im->lock);
1293 		return;
1294 	}
1295 	/* else, v3 */
1296 
1297 	im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1298 	igmp_ifc_event(in_dev);
1299 #endif
1300 }
1301 
1302 
1303 /*
1304  *	Multicast list managers
1305  */
1306 
1307 static u32 ip_mc_hash(const struct ip_mc_list *im)
1308 {
1309 	return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1310 }
1311 
1312 static void ip_mc_hash_add(struct in_device *in_dev,
1313 			   struct ip_mc_list *im)
1314 {
1315 	struct ip_mc_list __rcu **mc_hash;
1316 	u32 hash;
1317 
1318 	mc_hash = rtnl_dereference(in_dev->mc_hash);
1319 	if (mc_hash) {
1320 		hash = ip_mc_hash(im);
1321 		im->next_hash = mc_hash[hash];
1322 		rcu_assign_pointer(mc_hash[hash], im);
1323 		return;
1324 	}
1325 
1326 	/* do not use a hash table for small number of items */
1327 	if (in_dev->mc_count < 4)
1328 		return;
1329 
1330 	mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1331 			  GFP_KERNEL);
1332 	if (!mc_hash)
1333 		return;
1334 
1335 	for_each_pmc_rtnl(in_dev, im) {
1336 		hash = ip_mc_hash(im);
1337 		im->next_hash = mc_hash[hash];
1338 		RCU_INIT_POINTER(mc_hash[hash], im);
1339 	}
1340 
1341 	rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1342 }
1343 
1344 static void ip_mc_hash_remove(struct in_device *in_dev,
1345 			      struct ip_mc_list *im)
1346 {
1347 	struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1348 	struct ip_mc_list *aux;
1349 
1350 	if (!mc_hash)
1351 		return;
1352 	mc_hash += ip_mc_hash(im);
1353 	while ((aux = rtnl_dereference(*mc_hash)) != im)
1354 		mc_hash = &aux->next_hash;
1355 	*mc_hash = im->next_hash;
1356 }
1357 
1358 
1359 /*
1360  *	A socket has joined a multicast group on device dev.
1361  */
1362 
1363 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1364 {
1365 	struct ip_mc_list *im;
1366 #ifdef CONFIG_IP_MULTICAST
1367 	struct net *net = dev_net(in_dev->dev);
1368 #endif
1369 
1370 	ASSERT_RTNL();
1371 
1372 	for_each_pmc_rtnl(in_dev, im) {
1373 		if (im->multiaddr == addr) {
1374 			im->users++;
1375 			ip_mc_add_src(in_dev, &addr, MCAST_EXCLUDE, 0, NULL, 0);
1376 			goto out;
1377 		}
1378 	}
1379 
1380 	im = kzalloc(sizeof(*im), GFP_KERNEL);
1381 	if (!im)
1382 		goto out;
1383 
1384 	im->users = 1;
1385 	im->interface = in_dev;
1386 	in_dev_hold(in_dev);
1387 	im->multiaddr = addr;
1388 	/* initial mode is (EX, empty) */
1389 	im->sfmode = MCAST_EXCLUDE;
1390 	im->sfcount[MCAST_EXCLUDE] = 1;
1391 	refcount_set(&im->refcnt, 1);
1392 	spin_lock_init(&im->lock);
1393 #ifdef CONFIG_IP_MULTICAST
1394 	timer_setup(&im->timer, igmp_timer_expire, 0);
1395 	im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1396 #endif
1397 
1398 	im->next_rcu = in_dev->mc_list;
1399 	in_dev->mc_count++;
1400 	rcu_assign_pointer(in_dev->mc_list, im);
1401 
1402 	ip_mc_hash_add(in_dev, im);
1403 
1404 #ifdef CONFIG_IP_MULTICAST
1405 	igmpv3_del_delrec(in_dev, im);
1406 #endif
1407 	igmp_group_added(im);
1408 	if (!in_dev->dead)
1409 		ip_rt_multicast_event(in_dev);
1410 out:
1411 	return;
1412 }
1413 EXPORT_SYMBOL(ip_mc_inc_group);
1414 
1415 static int ip_mc_check_iphdr(struct sk_buff *skb)
1416 {
1417 	const struct iphdr *iph;
1418 	unsigned int len;
1419 	unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1420 
1421 	if (!pskb_may_pull(skb, offset))
1422 		return -EINVAL;
1423 
1424 	iph = ip_hdr(skb);
1425 
1426 	if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1427 		return -EINVAL;
1428 
1429 	offset += ip_hdrlen(skb) - sizeof(*iph);
1430 
1431 	if (!pskb_may_pull(skb, offset))
1432 		return -EINVAL;
1433 
1434 	iph = ip_hdr(skb);
1435 
1436 	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1437 		return -EINVAL;
1438 
1439 	len = skb_network_offset(skb) + ntohs(iph->tot_len);
1440 	if (skb->len < len || len < offset)
1441 		return -EINVAL;
1442 
1443 	skb_set_transport_header(skb, offset);
1444 
1445 	return 0;
1446 }
1447 
1448 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1449 {
1450 	unsigned int len = skb_transport_offset(skb);
1451 
1452 	len += sizeof(struct igmpv3_report);
1453 
1454 	return pskb_may_pull(skb, len) ? 0 : -EINVAL;
1455 }
1456 
1457 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1458 {
1459 	unsigned int len = skb_transport_offset(skb);
1460 
1461 	len += sizeof(struct igmphdr);
1462 	if (skb->len < len)
1463 		return -EINVAL;
1464 
1465 	/* IGMPv{1,2}? */
1466 	if (skb->len != len) {
1467 		/* or IGMPv3? */
1468 		len += sizeof(struct igmpv3_query) - sizeof(struct igmphdr);
1469 		if (skb->len < len || !pskb_may_pull(skb, len))
1470 			return -EINVAL;
1471 	}
1472 
1473 	/* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1474 	 * all-systems destination addresses (224.0.0.1) for general queries
1475 	 */
1476 	if (!igmp_hdr(skb)->group &&
1477 	    ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1478 		return -EINVAL;
1479 
1480 	return 0;
1481 }
1482 
1483 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1484 {
1485 	switch (igmp_hdr(skb)->type) {
1486 	case IGMP_HOST_LEAVE_MESSAGE:
1487 	case IGMP_HOST_MEMBERSHIP_REPORT:
1488 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
1489 		/* fall through */
1490 		return 0;
1491 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
1492 		return ip_mc_check_igmp_reportv3(skb);
1493 	case IGMP_HOST_MEMBERSHIP_QUERY:
1494 		return ip_mc_check_igmp_query(skb);
1495 	default:
1496 		return -ENOMSG;
1497 	}
1498 }
1499 
1500 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1501 {
1502 	return skb_checksum_simple_validate(skb);
1503 }
1504 
1505 static int __ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1506 
1507 {
1508 	struct sk_buff *skb_chk;
1509 	unsigned int transport_len;
1510 	unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1511 	int ret = -EINVAL;
1512 
1513 	transport_len = ntohs(ip_hdr(skb)->tot_len) - ip_hdrlen(skb);
1514 
1515 	skb_chk = skb_checksum_trimmed(skb, transport_len,
1516 				       ip_mc_validate_checksum);
1517 	if (!skb_chk)
1518 		goto err;
1519 
1520 	if (!pskb_may_pull(skb_chk, len))
1521 		goto err;
1522 
1523 	ret = ip_mc_check_igmp_msg(skb_chk);
1524 	if (ret)
1525 		goto err;
1526 
1527 	if (skb_trimmed)
1528 		*skb_trimmed = skb_chk;
1529 	/* free now unneeded clone */
1530 	else if (skb_chk != skb)
1531 		kfree_skb(skb_chk);
1532 
1533 	ret = 0;
1534 
1535 err:
1536 	if (ret && skb_chk && skb_chk != skb)
1537 		kfree_skb(skb_chk);
1538 
1539 	return ret;
1540 }
1541 
1542 /**
1543  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1544  * @skb: the skb to validate
1545  * @skb_trimmed: to store an skb pointer trimmed to IPv4 packet tail (optional)
1546  *
1547  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1548  * skb transport header accordingly and returns zero.
1549  *
1550  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1551  *  standard
1552  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1553  * -ENOMEM: A memory allocation failure happened.
1554  *
1555  * Optionally, an skb pointer might be provided via skb_trimmed (or set it
1556  * to NULL): After parsing an IGMP packet successfully it will point to
1557  * an skb which has its tail aligned to the IP packet end. This might
1558  * either be the originally provided skb or a trimmed, cloned version if
1559  * the skb frame had data beyond the IP packet. A cloned skb allows us
1560  * to leave the original skb and its full frame unchanged (which might be
1561  * desirable for layer 2 frame jugglers).
1562  *
1563  * Caller needs to set the skb network header and free any returned skb if it
1564  * differs from the provided skb.
1565  */
1566 int ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1567 {
1568 	int ret = ip_mc_check_iphdr(skb);
1569 
1570 	if (ret < 0)
1571 		return ret;
1572 
1573 	if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1574 		return -ENOMSG;
1575 
1576 	return __ip_mc_check_igmp(skb, skb_trimmed);
1577 }
1578 EXPORT_SYMBOL(ip_mc_check_igmp);
1579 
1580 /*
1581  *	Resend IGMP JOIN report; used by netdev notifier.
1582  */
1583 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1584 {
1585 #ifdef CONFIG_IP_MULTICAST
1586 	struct ip_mc_list *im;
1587 	int type;
1588 	struct net *net = dev_net(in_dev->dev);
1589 
1590 	ASSERT_RTNL();
1591 
1592 	for_each_pmc_rtnl(in_dev, im) {
1593 		if (im->multiaddr == IGMP_ALL_HOSTS)
1594 			continue;
1595 		if (ipv4_is_local_multicast(im->multiaddr) &&
1596 		    !net->ipv4.sysctl_igmp_llm_reports)
1597 			continue;
1598 
1599 		/* a failover is happening and switches
1600 		 * must be notified immediately
1601 		 */
1602 		if (IGMP_V1_SEEN(in_dev))
1603 			type = IGMP_HOST_MEMBERSHIP_REPORT;
1604 		else if (IGMP_V2_SEEN(in_dev))
1605 			type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1606 		else
1607 			type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1608 		igmp_send_report(in_dev, im, type);
1609 	}
1610 #endif
1611 }
1612 
1613 /*
1614  *	A socket has left a multicast group on device dev
1615  */
1616 
1617 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1618 {
1619 	struct ip_mc_list *i;
1620 	struct ip_mc_list __rcu **ip;
1621 
1622 	ASSERT_RTNL();
1623 
1624 	for (ip = &in_dev->mc_list;
1625 	     (i = rtnl_dereference(*ip)) != NULL;
1626 	     ip = &i->next_rcu) {
1627 		if (i->multiaddr == addr) {
1628 			if (--i->users == 0) {
1629 				ip_mc_hash_remove(in_dev, i);
1630 				*ip = i->next_rcu;
1631 				in_dev->mc_count--;
1632 				igmp_group_dropped(i);
1633 				ip_mc_clear_src(i);
1634 
1635 				if (!in_dev->dead)
1636 					ip_rt_multicast_event(in_dev);
1637 
1638 				ip_ma_put(i);
1639 				return;
1640 			}
1641 			break;
1642 		}
1643 	}
1644 }
1645 EXPORT_SYMBOL(ip_mc_dec_group);
1646 
1647 /* Device changing type */
1648 
1649 void ip_mc_unmap(struct in_device *in_dev)
1650 {
1651 	struct ip_mc_list *pmc;
1652 
1653 	ASSERT_RTNL();
1654 
1655 	for_each_pmc_rtnl(in_dev, pmc)
1656 		igmp_group_dropped(pmc);
1657 }
1658 
1659 void ip_mc_remap(struct in_device *in_dev)
1660 {
1661 	struct ip_mc_list *pmc;
1662 
1663 	ASSERT_RTNL();
1664 
1665 	for_each_pmc_rtnl(in_dev, pmc) {
1666 #ifdef CONFIG_IP_MULTICAST
1667 		igmpv3_del_delrec(in_dev, pmc);
1668 #endif
1669 		igmp_group_added(pmc);
1670 	}
1671 }
1672 
1673 /* Device going down */
1674 
1675 void ip_mc_down(struct in_device *in_dev)
1676 {
1677 	struct ip_mc_list *pmc;
1678 
1679 	ASSERT_RTNL();
1680 
1681 	for_each_pmc_rtnl(in_dev, pmc)
1682 		igmp_group_dropped(pmc);
1683 
1684 #ifdef CONFIG_IP_MULTICAST
1685 	in_dev->mr_ifc_count = 0;
1686 	if (del_timer(&in_dev->mr_ifc_timer))
1687 		__in_dev_put(in_dev);
1688 	in_dev->mr_gq_running = 0;
1689 	if (del_timer(&in_dev->mr_gq_timer))
1690 		__in_dev_put(in_dev);
1691 #endif
1692 
1693 	ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1694 }
1695 
1696 void ip_mc_init_dev(struct in_device *in_dev)
1697 {
1698 #ifdef CONFIG_IP_MULTICAST
1699 	struct net *net = dev_net(in_dev->dev);
1700 #endif
1701 	ASSERT_RTNL();
1702 
1703 #ifdef CONFIG_IP_MULTICAST
1704 	timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1705 	timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1706 	in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1707 #endif
1708 
1709 	spin_lock_init(&in_dev->mc_tomb_lock);
1710 }
1711 
1712 /* Device going up */
1713 
1714 void ip_mc_up(struct in_device *in_dev)
1715 {
1716 	struct ip_mc_list *pmc;
1717 #ifdef CONFIG_IP_MULTICAST
1718 	struct net *net = dev_net(in_dev->dev);
1719 #endif
1720 
1721 	ASSERT_RTNL();
1722 
1723 #ifdef CONFIG_IP_MULTICAST
1724 	in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1725 #endif
1726 	ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1727 
1728 	for_each_pmc_rtnl(in_dev, pmc) {
1729 #ifdef CONFIG_IP_MULTICAST
1730 		igmpv3_del_delrec(in_dev, pmc);
1731 #endif
1732 		igmp_group_added(pmc);
1733 	}
1734 }
1735 
1736 /*
1737  *	Device is about to be destroyed: clean up.
1738  */
1739 
1740 void ip_mc_destroy_dev(struct in_device *in_dev)
1741 {
1742 	struct ip_mc_list *i;
1743 
1744 	ASSERT_RTNL();
1745 
1746 	/* Deactivate timers */
1747 	ip_mc_down(in_dev);
1748 #ifdef CONFIG_IP_MULTICAST
1749 	igmpv3_clear_delrec(in_dev);
1750 #endif
1751 
1752 	while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1753 		in_dev->mc_list = i->next_rcu;
1754 		in_dev->mc_count--;
1755 		ip_ma_put(i);
1756 	}
1757 }
1758 
1759 /* RTNL is locked */
1760 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1761 {
1762 	struct net_device *dev = NULL;
1763 	struct in_device *idev = NULL;
1764 
1765 	if (imr->imr_ifindex) {
1766 		idev = inetdev_by_index(net, imr->imr_ifindex);
1767 		return idev;
1768 	}
1769 	if (imr->imr_address.s_addr) {
1770 		dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1771 		if (!dev)
1772 			return NULL;
1773 	}
1774 
1775 	if (!dev) {
1776 		struct rtable *rt = ip_route_output(net,
1777 						    imr->imr_multiaddr.s_addr,
1778 						    0, 0, 0);
1779 		if (!IS_ERR(rt)) {
1780 			dev = rt->dst.dev;
1781 			ip_rt_put(rt);
1782 		}
1783 	}
1784 	if (dev) {
1785 		imr->imr_ifindex = dev->ifindex;
1786 		idev = __in_dev_get_rtnl(dev);
1787 	}
1788 	return idev;
1789 }
1790 
1791 /*
1792  *	Join a socket to a group
1793  */
1794 
1795 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1796 	__be32 *psfsrc)
1797 {
1798 	struct ip_sf_list *psf, *psf_prev;
1799 	int rv = 0;
1800 
1801 	psf_prev = NULL;
1802 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1803 		if (psf->sf_inaddr == *psfsrc)
1804 			break;
1805 		psf_prev = psf;
1806 	}
1807 	if (!psf || psf->sf_count[sfmode] == 0) {
1808 		/* source filter not found, or count wrong =>  bug */
1809 		return -ESRCH;
1810 	}
1811 	psf->sf_count[sfmode]--;
1812 	if (psf->sf_count[sfmode] == 0) {
1813 		ip_rt_multicast_event(pmc->interface);
1814 	}
1815 	if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1816 #ifdef CONFIG_IP_MULTICAST
1817 		struct in_device *in_dev = pmc->interface;
1818 		struct net *net = dev_net(in_dev->dev);
1819 #endif
1820 
1821 		/* no more filters for this source */
1822 		if (psf_prev)
1823 			psf_prev->sf_next = psf->sf_next;
1824 		else
1825 			pmc->sources = psf->sf_next;
1826 #ifdef CONFIG_IP_MULTICAST
1827 		if (psf->sf_oldin &&
1828 		    !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1829 			psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1830 			psf->sf_next = pmc->tomb;
1831 			pmc->tomb = psf;
1832 			rv = 1;
1833 		} else
1834 #endif
1835 			kfree(psf);
1836 	}
1837 	return rv;
1838 }
1839 
1840 #ifndef CONFIG_IP_MULTICAST
1841 #define igmp_ifc_event(x)	do { } while (0)
1842 #endif
1843 
1844 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1845 			 int sfcount, __be32 *psfsrc, int delta)
1846 {
1847 	struct ip_mc_list *pmc;
1848 	int	changerec = 0;
1849 	int	i, err;
1850 
1851 	if (!in_dev)
1852 		return -ENODEV;
1853 	rcu_read_lock();
1854 	for_each_pmc_rcu(in_dev, pmc) {
1855 		if (*pmca == pmc->multiaddr)
1856 			break;
1857 	}
1858 	if (!pmc) {
1859 		/* MCA not found?? bug */
1860 		rcu_read_unlock();
1861 		return -ESRCH;
1862 	}
1863 	spin_lock_bh(&pmc->lock);
1864 	rcu_read_unlock();
1865 #ifdef CONFIG_IP_MULTICAST
1866 	sf_markstate(pmc);
1867 #endif
1868 	if (!delta) {
1869 		err = -EINVAL;
1870 		if (!pmc->sfcount[sfmode])
1871 			goto out_unlock;
1872 		pmc->sfcount[sfmode]--;
1873 	}
1874 	err = 0;
1875 	for (i = 0; i < sfcount; i++) {
1876 		int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1877 
1878 		changerec |= rv > 0;
1879 		if (!err && rv < 0)
1880 			err = rv;
1881 	}
1882 	if (pmc->sfmode == MCAST_EXCLUDE &&
1883 	    pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1884 	    pmc->sfcount[MCAST_INCLUDE]) {
1885 #ifdef CONFIG_IP_MULTICAST
1886 		struct ip_sf_list *psf;
1887 		struct net *net = dev_net(in_dev->dev);
1888 #endif
1889 
1890 		/* filter mode change */
1891 		pmc->sfmode = MCAST_INCLUDE;
1892 #ifdef CONFIG_IP_MULTICAST
1893 		pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1894 		in_dev->mr_ifc_count = pmc->crcount;
1895 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1896 			psf->sf_crcount = 0;
1897 		igmp_ifc_event(pmc->interface);
1898 	} else if (sf_setstate(pmc) || changerec) {
1899 		igmp_ifc_event(pmc->interface);
1900 #endif
1901 	}
1902 out_unlock:
1903 	spin_unlock_bh(&pmc->lock);
1904 	return err;
1905 }
1906 
1907 /*
1908  * Add multicast single-source filter to the interface list
1909  */
1910 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1911 	__be32 *psfsrc)
1912 {
1913 	struct ip_sf_list *psf, *psf_prev;
1914 
1915 	psf_prev = NULL;
1916 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1917 		if (psf->sf_inaddr == *psfsrc)
1918 			break;
1919 		psf_prev = psf;
1920 	}
1921 	if (!psf) {
1922 		psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1923 		if (!psf)
1924 			return -ENOBUFS;
1925 		psf->sf_inaddr = *psfsrc;
1926 		if (psf_prev) {
1927 			psf_prev->sf_next = psf;
1928 		} else
1929 			pmc->sources = psf;
1930 	}
1931 	psf->sf_count[sfmode]++;
1932 	if (psf->sf_count[sfmode] == 1) {
1933 		ip_rt_multicast_event(pmc->interface);
1934 	}
1935 	return 0;
1936 }
1937 
1938 #ifdef CONFIG_IP_MULTICAST
1939 static void sf_markstate(struct ip_mc_list *pmc)
1940 {
1941 	struct ip_sf_list *psf;
1942 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1943 
1944 	for (psf = pmc->sources; psf; psf = psf->sf_next)
1945 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1946 			psf->sf_oldin = mca_xcount ==
1947 				psf->sf_count[MCAST_EXCLUDE] &&
1948 				!psf->sf_count[MCAST_INCLUDE];
1949 		} else
1950 			psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1951 }
1952 
1953 static int sf_setstate(struct ip_mc_list *pmc)
1954 {
1955 	struct ip_sf_list *psf, *dpsf;
1956 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1957 	int qrv = pmc->interface->mr_qrv;
1958 	int new_in, rv;
1959 
1960 	rv = 0;
1961 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1962 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1963 			new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1964 				!psf->sf_count[MCAST_INCLUDE];
1965 		} else
1966 			new_in = psf->sf_count[MCAST_INCLUDE] != 0;
1967 		if (new_in) {
1968 			if (!psf->sf_oldin) {
1969 				struct ip_sf_list *prev = NULL;
1970 
1971 				for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
1972 					if (dpsf->sf_inaddr == psf->sf_inaddr)
1973 						break;
1974 					prev = dpsf;
1975 				}
1976 				if (dpsf) {
1977 					if (prev)
1978 						prev->sf_next = dpsf->sf_next;
1979 					else
1980 						pmc->tomb = dpsf->sf_next;
1981 					kfree(dpsf);
1982 				}
1983 				psf->sf_crcount = qrv;
1984 				rv++;
1985 			}
1986 		} else if (psf->sf_oldin) {
1987 
1988 			psf->sf_crcount = 0;
1989 			/*
1990 			 * add or update "delete" records if an active filter
1991 			 * is now inactive
1992 			 */
1993 			for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
1994 				if (dpsf->sf_inaddr == psf->sf_inaddr)
1995 					break;
1996 			if (!dpsf) {
1997 				dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
1998 				if (!dpsf)
1999 					continue;
2000 				*dpsf = *psf;
2001 				/* pmc->lock held by callers */
2002 				dpsf->sf_next = pmc->tomb;
2003 				pmc->tomb = dpsf;
2004 			}
2005 			dpsf->sf_crcount = qrv;
2006 			rv++;
2007 		}
2008 	}
2009 	return rv;
2010 }
2011 #endif
2012 
2013 /*
2014  * Add multicast source filter list to the interface list
2015  */
2016 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2017 			 int sfcount, __be32 *psfsrc, int delta)
2018 {
2019 	struct ip_mc_list *pmc;
2020 	int	isexclude;
2021 	int	i, err;
2022 
2023 	if (!in_dev)
2024 		return -ENODEV;
2025 	rcu_read_lock();
2026 	for_each_pmc_rcu(in_dev, pmc) {
2027 		if (*pmca == pmc->multiaddr)
2028 			break;
2029 	}
2030 	if (!pmc) {
2031 		/* MCA not found?? bug */
2032 		rcu_read_unlock();
2033 		return -ESRCH;
2034 	}
2035 	spin_lock_bh(&pmc->lock);
2036 	rcu_read_unlock();
2037 
2038 #ifdef CONFIG_IP_MULTICAST
2039 	sf_markstate(pmc);
2040 #endif
2041 	isexclude = pmc->sfmode == MCAST_EXCLUDE;
2042 	if (!delta)
2043 		pmc->sfcount[sfmode]++;
2044 	err = 0;
2045 	for (i = 0; i < sfcount; i++) {
2046 		err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2047 		if (err)
2048 			break;
2049 	}
2050 	if (err) {
2051 		int j;
2052 
2053 		if (!delta)
2054 			pmc->sfcount[sfmode]--;
2055 		for (j = 0; j < i; j++)
2056 			(void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2057 	} else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2058 #ifdef CONFIG_IP_MULTICAST
2059 		struct ip_sf_list *psf;
2060 		struct net *net = dev_net(pmc->interface->dev);
2061 		in_dev = pmc->interface;
2062 #endif
2063 
2064 		/* filter mode change */
2065 		if (pmc->sfcount[MCAST_EXCLUDE])
2066 			pmc->sfmode = MCAST_EXCLUDE;
2067 		else if (pmc->sfcount[MCAST_INCLUDE])
2068 			pmc->sfmode = MCAST_INCLUDE;
2069 #ifdef CONFIG_IP_MULTICAST
2070 		/* else no filters; keep old mode for reports */
2071 
2072 		pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2073 		in_dev->mr_ifc_count = pmc->crcount;
2074 		for (psf = pmc->sources; psf; psf = psf->sf_next)
2075 			psf->sf_crcount = 0;
2076 		igmp_ifc_event(in_dev);
2077 	} else if (sf_setstate(pmc)) {
2078 		igmp_ifc_event(in_dev);
2079 #endif
2080 	}
2081 	spin_unlock_bh(&pmc->lock);
2082 	return err;
2083 }
2084 
2085 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2086 {
2087 	struct ip_sf_list *psf, *nextpsf, *tomb, *sources;
2088 
2089 	spin_lock_bh(&pmc->lock);
2090 	tomb = pmc->tomb;
2091 	pmc->tomb = NULL;
2092 	sources = pmc->sources;
2093 	pmc->sources = NULL;
2094 	pmc->sfmode = MCAST_EXCLUDE;
2095 	pmc->sfcount[MCAST_INCLUDE] = 0;
2096 	pmc->sfcount[MCAST_EXCLUDE] = 1;
2097 	spin_unlock_bh(&pmc->lock);
2098 
2099 	for (psf = tomb; psf; psf = nextpsf) {
2100 		nextpsf = psf->sf_next;
2101 		kfree(psf);
2102 	}
2103 	for (psf = sources; psf; psf = nextpsf) {
2104 		nextpsf = psf->sf_next;
2105 		kfree(psf);
2106 	}
2107 }
2108 
2109 /* Join a multicast group
2110  */
2111 
2112 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2113 {
2114 	__be32 addr = imr->imr_multiaddr.s_addr;
2115 	struct ip_mc_socklist *iml, *i;
2116 	struct in_device *in_dev;
2117 	struct inet_sock *inet = inet_sk(sk);
2118 	struct net *net = sock_net(sk);
2119 	int ifindex;
2120 	int count = 0;
2121 	int err;
2122 
2123 	ASSERT_RTNL();
2124 
2125 	if (!ipv4_is_multicast(addr))
2126 		return -EINVAL;
2127 
2128 	in_dev = ip_mc_find_dev(net, imr);
2129 
2130 	if (!in_dev) {
2131 		err = -ENODEV;
2132 		goto done;
2133 	}
2134 
2135 	err = -EADDRINUSE;
2136 	ifindex = imr->imr_ifindex;
2137 	for_each_pmc_rtnl(inet, i) {
2138 		if (i->multi.imr_multiaddr.s_addr == addr &&
2139 		    i->multi.imr_ifindex == ifindex)
2140 			goto done;
2141 		count++;
2142 	}
2143 	err = -ENOBUFS;
2144 	if (count >= net->ipv4.sysctl_igmp_max_memberships)
2145 		goto done;
2146 	iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2147 	if (!iml)
2148 		goto done;
2149 
2150 	memcpy(&iml->multi, imr, sizeof(*imr));
2151 	iml->next_rcu = inet->mc_list;
2152 	iml->sflist = NULL;
2153 	iml->sfmode = MCAST_EXCLUDE;
2154 	rcu_assign_pointer(inet->mc_list, iml);
2155 	ip_mc_inc_group(in_dev, addr);
2156 	err = 0;
2157 done:
2158 	return err;
2159 }
2160 EXPORT_SYMBOL(ip_mc_join_group);
2161 
2162 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2163 			   struct in_device *in_dev)
2164 {
2165 	struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2166 	int err;
2167 
2168 	if (!psf) {
2169 		/* any-source empty exclude case */
2170 		return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2171 			iml->sfmode, 0, NULL, 0);
2172 	}
2173 	err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2174 			iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2175 	RCU_INIT_POINTER(iml->sflist, NULL);
2176 	/* decrease mem now to avoid the memleak warning */
2177 	atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2178 	kfree_rcu(psf, rcu);
2179 	return err;
2180 }
2181 
2182 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2183 {
2184 	struct inet_sock *inet = inet_sk(sk);
2185 	struct ip_mc_socklist *iml;
2186 	struct ip_mc_socklist __rcu **imlp;
2187 	struct in_device *in_dev;
2188 	struct net *net = sock_net(sk);
2189 	__be32 group = imr->imr_multiaddr.s_addr;
2190 	u32 ifindex;
2191 	int ret = -EADDRNOTAVAIL;
2192 
2193 	ASSERT_RTNL();
2194 
2195 	in_dev = ip_mc_find_dev(net, imr);
2196 	if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2197 		ret = -ENODEV;
2198 		goto out;
2199 	}
2200 	ifindex = imr->imr_ifindex;
2201 	for (imlp = &inet->mc_list;
2202 	     (iml = rtnl_dereference(*imlp)) != NULL;
2203 	     imlp = &iml->next_rcu) {
2204 		if (iml->multi.imr_multiaddr.s_addr != group)
2205 			continue;
2206 		if (ifindex) {
2207 			if (iml->multi.imr_ifindex != ifindex)
2208 				continue;
2209 		} else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2210 				iml->multi.imr_address.s_addr)
2211 			continue;
2212 
2213 		(void) ip_mc_leave_src(sk, iml, in_dev);
2214 
2215 		*imlp = iml->next_rcu;
2216 
2217 		if (in_dev)
2218 			ip_mc_dec_group(in_dev, group);
2219 
2220 		/* decrease mem now to avoid the memleak warning */
2221 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2222 		kfree_rcu(iml, rcu);
2223 		return 0;
2224 	}
2225 out:
2226 	return ret;
2227 }
2228 EXPORT_SYMBOL(ip_mc_leave_group);
2229 
2230 int ip_mc_source(int add, int omode, struct sock *sk, struct
2231 	ip_mreq_source *mreqs, int ifindex)
2232 {
2233 	int err;
2234 	struct ip_mreqn imr;
2235 	__be32 addr = mreqs->imr_multiaddr;
2236 	struct ip_mc_socklist *pmc;
2237 	struct in_device *in_dev = NULL;
2238 	struct inet_sock *inet = inet_sk(sk);
2239 	struct ip_sf_socklist *psl;
2240 	struct net *net = sock_net(sk);
2241 	int leavegroup = 0;
2242 	int i, j, rv;
2243 
2244 	if (!ipv4_is_multicast(addr))
2245 		return -EINVAL;
2246 
2247 	ASSERT_RTNL();
2248 
2249 	imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2250 	imr.imr_address.s_addr = mreqs->imr_interface;
2251 	imr.imr_ifindex = ifindex;
2252 	in_dev = ip_mc_find_dev(net, &imr);
2253 
2254 	if (!in_dev) {
2255 		err = -ENODEV;
2256 		goto done;
2257 	}
2258 	err = -EADDRNOTAVAIL;
2259 
2260 	for_each_pmc_rtnl(inet, pmc) {
2261 		if ((pmc->multi.imr_multiaddr.s_addr ==
2262 		     imr.imr_multiaddr.s_addr) &&
2263 		    (pmc->multi.imr_ifindex == imr.imr_ifindex))
2264 			break;
2265 	}
2266 	if (!pmc) {		/* must have a prior join */
2267 		err = -EINVAL;
2268 		goto done;
2269 	}
2270 	/* if a source filter was set, must be the same mode as before */
2271 	if (pmc->sflist) {
2272 		if (pmc->sfmode != omode) {
2273 			err = -EINVAL;
2274 			goto done;
2275 		}
2276 	} else if (pmc->sfmode != omode) {
2277 		/* allow mode switches for empty-set filters */
2278 		ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2279 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2280 			NULL, 0);
2281 		pmc->sfmode = omode;
2282 	}
2283 
2284 	psl = rtnl_dereference(pmc->sflist);
2285 	if (!add) {
2286 		if (!psl)
2287 			goto done;	/* err = -EADDRNOTAVAIL */
2288 		rv = !0;
2289 		for (i = 0; i < psl->sl_count; i++) {
2290 			rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2291 				sizeof(__be32));
2292 			if (rv == 0)
2293 				break;
2294 		}
2295 		if (rv)		/* source not found */
2296 			goto done;	/* err = -EADDRNOTAVAIL */
2297 
2298 		/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2299 		if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2300 			leavegroup = 1;
2301 			goto done;
2302 		}
2303 
2304 		/* update the interface filter */
2305 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2306 			&mreqs->imr_sourceaddr, 1);
2307 
2308 		for (j = i+1; j < psl->sl_count; j++)
2309 			psl->sl_addr[j-1] = psl->sl_addr[j];
2310 		psl->sl_count--;
2311 		err = 0;
2312 		goto done;
2313 	}
2314 	/* else, add a new source to the filter */
2315 
2316 	if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2317 		err = -ENOBUFS;
2318 		goto done;
2319 	}
2320 	if (!psl || psl->sl_count == psl->sl_max) {
2321 		struct ip_sf_socklist *newpsl;
2322 		int count = IP_SFBLOCK;
2323 
2324 		if (psl)
2325 			count += psl->sl_max;
2326 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2327 		if (!newpsl) {
2328 			err = -ENOBUFS;
2329 			goto done;
2330 		}
2331 		newpsl->sl_max = count;
2332 		newpsl->sl_count = count - IP_SFBLOCK;
2333 		if (psl) {
2334 			for (i = 0; i < psl->sl_count; i++)
2335 				newpsl->sl_addr[i] = psl->sl_addr[i];
2336 			/* decrease mem now to avoid the memleak warning */
2337 			atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2338 			kfree_rcu(psl, rcu);
2339 		}
2340 		rcu_assign_pointer(pmc->sflist, newpsl);
2341 		psl = newpsl;
2342 	}
2343 	rv = 1;	/* > 0 for insert logic below if sl_count is 0 */
2344 	for (i = 0; i < psl->sl_count; i++) {
2345 		rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2346 			sizeof(__be32));
2347 		if (rv == 0)
2348 			break;
2349 	}
2350 	if (rv == 0)		/* address already there is an error */
2351 		goto done;
2352 	for (j = psl->sl_count-1; j >= i; j--)
2353 		psl->sl_addr[j+1] = psl->sl_addr[j];
2354 	psl->sl_addr[i] = mreqs->imr_sourceaddr;
2355 	psl->sl_count++;
2356 	err = 0;
2357 	/* update the interface list */
2358 	ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2359 		&mreqs->imr_sourceaddr, 1);
2360 done:
2361 	if (leavegroup)
2362 		err = ip_mc_leave_group(sk, &imr);
2363 	return err;
2364 }
2365 
2366 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2367 {
2368 	int err = 0;
2369 	struct ip_mreqn	imr;
2370 	__be32 addr = msf->imsf_multiaddr;
2371 	struct ip_mc_socklist *pmc;
2372 	struct in_device *in_dev;
2373 	struct inet_sock *inet = inet_sk(sk);
2374 	struct ip_sf_socklist *newpsl, *psl;
2375 	struct net *net = sock_net(sk);
2376 	int leavegroup = 0;
2377 
2378 	if (!ipv4_is_multicast(addr))
2379 		return -EINVAL;
2380 	if (msf->imsf_fmode != MCAST_INCLUDE &&
2381 	    msf->imsf_fmode != MCAST_EXCLUDE)
2382 		return -EINVAL;
2383 
2384 	ASSERT_RTNL();
2385 
2386 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2387 	imr.imr_address.s_addr = msf->imsf_interface;
2388 	imr.imr_ifindex = ifindex;
2389 	in_dev = ip_mc_find_dev(net, &imr);
2390 
2391 	if (!in_dev) {
2392 		err = -ENODEV;
2393 		goto done;
2394 	}
2395 
2396 	/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2397 	if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2398 		leavegroup = 1;
2399 		goto done;
2400 	}
2401 
2402 	for_each_pmc_rtnl(inet, pmc) {
2403 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2404 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2405 			break;
2406 	}
2407 	if (!pmc) {		/* must have a prior join */
2408 		err = -EINVAL;
2409 		goto done;
2410 	}
2411 	if (msf->imsf_numsrc) {
2412 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2413 							   GFP_KERNEL);
2414 		if (!newpsl) {
2415 			err = -ENOBUFS;
2416 			goto done;
2417 		}
2418 		newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2419 		memcpy(newpsl->sl_addr, msf->imsf_slist,
2420 			msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2421 		err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2422 			msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2423 		if (err) {
2424 			sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2425 			goto done;
2426 		}
2427 	} else {
2428 		newpsl = NULL;
2429 		(void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2430 				     msf->imsf_fmode, 0, NULL, 0);
2431 	}
2432 	psl = rtnl_dereference(pmc->sflist);
2433 	if (psl) {
2434 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2435 			psl->sl_count, psl->sl_addr, 0);
2436 		/* decrease mem now to avoid the memleak warning */
2437 		atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2438 		kfree_rcu(psl, rcu);
2439 	} else
2440 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2441 			0, NULL, 0);
2442 	rcu_assign_pointer(pmc->sflist, newpsl);
2443 	pmc->sfmode = msf->imsf_fmode;
2444 	err = 0;
2445 done:
2446 	if (leavegroup)
2447 		err = ip_mc_leave_group(sk, &imr);
2448 	return err;
2449 }
2450 
2451 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2452 	struct ip_msfilter __user *optval, int __user *optlen)
2453 {
2454 	int err, len, count, copycount;
2455 	struct ip_mreqn	imr;
2456 	__be32 addr = msf->imsf_multiaddr;
2457 	struct ip_mc_socklist *pmc;
2458 	struct in_device *in_dev;
2459 	struct inet_sock *inet = inet_sk(sk);
2460 	struct ip_sf_socklist *psl;
2461 	struct net *net = sock_net(sk);
2462 
2463 	ASSERT_RTNL();
2464 
2465 	if (!ipv4_is_multicast(addr))
2466 		return -EINVAL;
2467 
2468 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2469 	imr.imr_address.s_addr = msf->imsf_interface;
2470 	imr.imr_ifindex = 0;
2471 	in_dev = ip_mc_find_dev(net, &imr);
2472 
2473 	if (!in_dev) {
2474 		err = -ENODEV;
2475 		goto done;
2476 	}
2477 	err = -EADDRNOTAVAIL;
2478 
2479 	for_each_pmc_rtnl(inet, pmc) {
2480 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2481 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2482 			break;
2483 	}
2484 	if (!pmc)		/* must have a prior join */
2485 		goto done;
2486 	msf->imsf_fmode = pmc->sfmode;
2487 	psl = rtnl_dereference(pmc->sflist);
2488 	if (!psl) {
2489 		len = 0;
2490 		count = 0;
2491 	} else {
2492 		count = psl->sl_count;
2493 	}
2494 	copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2495 	len = copycount * sizeof(psl->sl_addr[0]);
2496 	msf->imsf_numsrc = count;
2497 	if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2498 	    copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2499 		return -EFAULT;
2500 	}
2501 	if (len &&
2502 	    copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2503 		return -EFAULT;
2504 	return 0;
2505 done:
2506 	return err;
2507 }
2508 
2509 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2510 	struct group_filter __user *optval, int __user *optlen)
2511 {
2512 	int err, i, count, copycount;
2513 	struct sockaddr_in *psin;
2514 	__be32 addr;
2515 	struct ip_mc_socklist *pmc;
2516 	struct inet_sock *inet = inet_sk(sk);
2517 	struct ip_sf_socklist *psl;
2518 
2519 	ASSERT_RTNL();
2520 
2521 	psin = (struct sockaddr_in *)&gsf->gf_group;
2522 	if (psin->sin_family != AF_INET)
2523 		return -EINVAL;
2524 	addr = psin->sin_addr.s_addr;
2525 	if (!ipv4_is_multicast(addr))
2526 		return -EINVAL;
2527 
2528 	err = -EADDRNOTAVAIL;
2529 
2530 	for_each_pmc_rtnl(inet, pmc) {
2531 		if (pmc->multi.imr_multiaddr.s_addr == addr &&
2532 		    pmc->multi.imr_ifindex == gsf->gf_interface)
2533 			break;
2534 	}
2535 	if (!pmc)		/* must have a prior join */
2536 		goto done;
2537 	gsf->gf_fmode = pmc->sfmode;
2538 	psl = rtnl_dereference(pmc->sflist);
2539 	count = psl ? psl->sl_count : 0;
2540 	copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2541 	gsf->gf_numsrc = count;
2542 	if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2543 	    copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2544 		return -EFAULT;
2545 	}
2546 	for (i = 0; i < copycount; i++) {
2547 		struct sockaddr_storage ss;
2548 
2549 		psin = (struct sockaddr_in *)&ss;
2550 		memset(&ss, 0, sizeof(ss));
2551 		psin->sin_family = AF_INET;
2552 		psin->sin_addr.s_addr = psl->sl_addr[i];
2553 		if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2554 			return -EFAULT;
2555 	}
2556 	return 0;
2557 done:
2558 	return err;
2559 }
2560 
2561 /*
2562  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2563  */
2564 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2565 		   int dif, int sdif)
2566 {
2567 	struct inet_sock *inet = inet_sk(sk);
2568 	struct ip_mc_socklist *pmc;
2569 	struct ip_sf_socklist *psl;
2570 	int i;
2571 	int ret;
2572 
2573 	ret = 1;
2574 	if (!ipv4_is_multicast(loc_addr))
2575 		goto out;
2576 
2577 	rcu_read_lock();
2578 	for_each_pmc_rcu(inet, pmc) {
2579 		if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2580 		    (pmc->multi.imr_ifindex == dif ||
2581 		     (sdif && pmc->multi.imr_ifindex == sdif)))
2582 			break;
2583 	}
2584 	ret = inet->mc_all;
2585 	if (!pmc)
2586 		goto unlock;
2587 	psl = rcu_dereference(pmc->sflist);
2588 	ret = (pmc->sfmode == MCAST_EXCLUDE);
2589 	if (!psl)
2590 		goto unlock;
2591 
2592 	for (i = 0; i < psl->sl_count; i++) {
2593 		if (psl->sl_addr[i] == rmt_addr)
2594 			break;
2595 	}
2596 	ret = 0;
2597 	if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2598 		goto unlock;
2599 	if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2600 		goto unlock;
2601 	ret = 1;
2602 unlock:
2603 	rcu_read_unlock();
2604 out:
2605 	return ret;
2606 }
2607 
2608 /*
2609  *	A socket is closing.
2610  */
2611 
2612 void ip_mc_drop_socket(struct sock *sk)
2613 {
2614 	struct inet_sock *inet = inet_sk(sk);
2615 	struct ip_mc_socklist *iml;
2616 	struct net *net = sock_net(sk);
2617 
2618 	if (!inet->mc_list)
2619 		return;
2620 
2621 	rtnl_lock();
2622 	while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2623 		struct in_device *in_dev;
2624 
2625 		inet->mc_list = iml->next_rcu;
2626 		in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2627 		(void) ip_mc_leave_src(sk, iml, in_dev);
2628 		if (in_dev)
2629 			ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2630 		/* decrease mem now to avoid the memleak warning */
2631 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2632 		kfree_rcu(iml, rcu);
2633 	}
2634 	rtnl_unlock();
2635 }
2636 
2637 /* called with rcu_read_lock() */
2638 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2639 {
2640 	struct ip_mc_list *im;
2641 	struct ip_mc_list __rcu **mc_hash;
2642 	struct ip_sf_list *psf;
2643 	int rv = 0;
2644 
2645 	mc_hash = rcu_dereference(in_dev->mc_hash);
2646 	if (mc_hash) {
2647 		u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2648 
2649 		for (im = rcu_dereference(mc_hash[hash]);
2650 		     im != NULL;
2651 		     im = rcu_dereference(im->next_hash)) {
2652 			if (im->multiaddr == mc_addr)
2653 				break;
2654 		}
2655 	} else {
2656 		for_each_pmc_rcu(in_dev, im) {
2657 			if (im->multiaddr == mc_addr)
2658 				break;
2659 		}
2660 	}
2661 	if (im && proto == IPPROTO_IGMP) {
2662 		rv = 1;
2663 	} else if (im) {
2664 		if (src_addr) {
2665 			for (psf = im->sources; psf; psf = psf->sf_next) {
2666 				if (psf->sf_inaddr == src_addr)
2667 					break;
2668 			}
2669 			if (psf)
2670 				rv = psf->sf_count[MCAST_INCLUDE] ||
2671 					psf->sf_count[MCAST_EXCLUDE] !=
2672 					im->sfcount[MCAST_EXCLUDE];
2673 			else
2674 				rv = im->sfcount[MCAST_EXCLUDE] != 0;
2675 		} else
2676 			rv = 1; /* unspecified source; tentatively allow */
2677 	}
2678 	return rv;
2679 }
2680 
2681 #if defined(CONFIG_PROC_FS)
2682 struct igmp_mc_iter_state {
2683 	struct seq_net_private p;
2684 	struct net_device *dev;
2685 	struct in_device *in_dev;
2686 };
2687 
2688 #define	igmp_mc_seq_private(seq)	((struct igmp_mc_iter_state *)(seq)->private)
2689 
2690 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2691 {
2692 	struct net *net = seq_file_net(seq);
2693 	struct ip_mc_list *im = NULL;
2694 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2695 
2696 	state->in_dev = NULL;
2697 	for_each_netdev_rcu(net, state->dev) {
2698 		struct in_device *in_dev;
2699 
2700 		in_dev = __in_dev_get_rcu(state->dev);
2701 		if (!in_dev)
2702 			continue;
2703 		im = rcu_dereference(in_dev->mc_list);
2704 		if (im) {
2705 			state->in_dev = in_dev;
2706 			break;
2707 		}
2708 	}
2709 	return im;
2710 }
2711 
2712 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2713 {
2714 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2715 
2716 	im = rcu_dereference(im->next_rcu);
2717 	while (!im) {
2718 		state->dev = next_net_device_rcu(state->dev);
2719 		if (!state->dev) {
2720 			state->in_dev = NULL;
2721 			break;
2722 		}
2723 		state->in_dev = __in_dev_get_rcu(state->dev);
2724 		if (!state->in_dev)
2725 			continue;
2726 		im = rcu_dereference(state->in_dev->mc_list);
2727 	}
2728 	return im;
2729 }
2730 
2731 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2732 {
2733 	struct ip_mc_list *im = igmp_mc_get_first(seq);
2734 	if (im)
2735 		while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2736 			--pos;
2737 	return pos ? NULL : im;
2738 }
2739 
2740 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2741 	__acquires(rcu)
2742 {
2743 	rcu_read_lock();
2744 	return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2745 }
2746 
2747 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2748 {
2749 	struct ip_mc_list *im;
2750 	if (v == SEQ_START_TOKEN)
2751 		im = igmp_mc_get_first(seq);
2752 	else
2753 		im = igmp_mc_get_next(seq, v);
2754 	++*pos;
2755 	return im;
2756 }
2757 
2758 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2759 	__releases(rcu)
2760 {
2761 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2762 
2763 	state->in_dev = NULL;
2764 	state->dev = NULL;
2765 	rcu_read_unlock();
2766 }
2767 
2768 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2769 {
2770 	if (v == SEQ_START_TOKEN)
2771 		seq_puts(seq,
2772 			 "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2773 	else {
2774 		struct ip_mc_list *im = (struct ip_mc_list *)v;
2775 		struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2776 		char   *querier;
2777 		long delta;
2778 
2779 #ifdef CONFIG_IP_MULTICAST
2780 		querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2781 			  IGMP_V2_SEEN(state->in_dev) ? "V2" :
2782 			  "V3";
2783 #else
2784 		querier = "NONE";
2785 #endif
2786 
2787 		if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2788 			seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2789 				   state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2790 		}
2791 
2792 		delta = im->timer.expires - jiffies;
2793 		seq_printf(seq,
2794 			   "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2795 			   im->multiaddr, im->users,
2796 			   im->tm_running,
2797 			   im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2798 			   im->reporter);
2799 	}
2800 	return 0;
2801 }
2802 
2803 static const struct seq_operations igmp_mc_seq_ops = {
2804 	.start	=	igmp_mc_seq_start,
2805 	.next	=	igmp_mc_seq_next,
2806 	.stop	=	igmp_mc_seq_stop,
2807 	.show	=	igmp_mc_seq_show,
2808 };
2809 
2810 static int igmp_mc_seq_open(struct inode *inode, struct file *file)
2811 {
2812 	return seq_open_net(inode, file, &igmp_mc_seq_ops,
2813 			sizeof(struct igmp_mc_iter_state));
2814 }
2815 
2816 static const struct file_operations igmp_mc_seq_fops = {
2817 	.owner		=	THIS_MODULE,
2818 	.open		=	igmp_mc_seq_open,
2819 	.read		=	seq_read,
2820 	.llseek		=	seq_lseek,
2821 	.release	=	seq_release_net,
2822 };
2823 
2824 struct igmp_mcf_iter_state {
2825 	struct seq_net_private p;
2826 	struct net_device *dev;
2827 	struct in_device *idev;
2828 	struct ip_mc_list *im;
2829 };
2830 
2831 #define igmp_mcf_seq_private(seq)	((struct igmp_mcf_iter_state *)(seq)->private)
2832 
2833 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2834 {
2835 	struct net *net = seq_file_net(seq);
2836 	struct ip_sf_list *psf = NULL;
2837 	struct ip_mc_list *im = NULL;
2838 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2839 
2840 	state->idev = NULL;
2841 	state->im = NULL;
2842 	for_each_netdev_rcu(net, state->dev) {
2843 		struct in_device *idev;
2844 		idev = __in_dev_get_rcu(state->dev);
2845 		if (unlikely(!idev))
2846 			continue;
2847 		im = rcu_dereference(idev->mc_list);
2848 		if (likely(im)) {
2849 			spin_lock_bh(&im->lock);
2850 			psf = im->sources;
2851 			if (likely(psf)) {
2852 				state->im = im;
2853 				state->idev = idev;
2854 				break;
2855 			}
2856 			spin_unlock_bh(&im->lock);
2857 		}
2858 	}
2859 	return psf;
2860 }
2861 
2862 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2863 {
2864 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2865 
2866 	psf = psf->sf_next;
2867 	while (!psf) {
2868 		spin_unlock_bh(&state->im->lock);
2869 		state->im = state->im->next;
2870 		while (!state->im) {
2871 			state->dev = next_net_device_rcu(state->dev);
2872 			if (!state->dev) {
2873 				state->idev = NULL;
2874 				goto out;
2875 			}
2876 			state->idev = __in_dev_get_rcu(state->dev);
2877 			if (!state->idev)
2878 				continue;
2879 			state->im = rcu_dereference(state->idev->mc_list);
2880 		}
2881 		if (!state->im)
2882 			break;
2883 		spin_lock_bh(&state->im->lock);
2884 		psf = state->im->sources;
2885 	}
2886 out:
2887 	return psf;
2888 }
2889 
2890 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2891 {
2892 	struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2893 	if (psf)
2894 		while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2895 			--pos;
2896 	return pos ? NULL : psf;
2897 }
2898 
2899 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2900 	__acquires(rcu)
2901 {
2902 	rcu_read_lock();
2903 	return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2904 }
2905 
2906 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2907 {
2908 	struct ip_sf_list *psf;
2909 	if (v == SEQ_START_TOKEN)
2910 		psf = igmp_mcf_get_first(seq);
2911 	else
2912 		psf = igmp_mcf_get_next(seq, v);
2913 	++*pos;
2914 	return psf;
2915 }
2916 
2917 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2918 	__releases(rcu)
2919 {
2920 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2921 	if (likely(state->im)) {
2922 		spin_unlock_bh(&state->im->lock);
2923 		state->im = NULL;
2924 	}
2925 	state->idev = NULL;
2926 	state->dev = NULL;
2927 	rcu_read_unlock();
2928 }
2929 
2930 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2931 {
2932 	struct ip_sf_list *psf = (struct ip_sf_list *)v;
2933 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2934 
2935 	if (v == SEQ_START_TOKEN) {
2936 		seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2937 	} else {
2938 		seq_printf(seq,
2939 			   "%3d %6.6s 0x%08x "
2940 			   "0x%08x %6lu %6lu\n",
2941 			   state->dev->ifindex, state->dev->name,
2942 			   ntohl(state->im->multiaddr),
2943 			   ntohl(psf->sf_inaddr),
2944 			   psf->sf_count[MCAST_INCLUDE],
2945 			   psf->sf_count[MCAST_EXCLUDE]);
2946 	}
2947 	return 0;
2948 }
2949 
2950 static const struct seq_operations igmp_mcf_seq_ops = {
2951 	.start	=	igmp_mcf_seq_start,
2952 	.next	=	igmp_mcf_seq_next,
2953 	.stop	=	igmp_mcf_seq_stop,
2954 	.show	=	igmp_mcf_seq_show,
2955 };
2956 
2957 static int igmp_mcf_seq_open(struct inode *inode, struct file *file)
2958 {
2959 	return seq_open_net(inode, file, &igmp_mcf_seq_ops,
2960 			sizeof(struct igmp_mcf_iter_state));
2961 }
2962 
2963 static const struct file_operations igmp_mcf_seq_fops = {
2964 	.owner		=	THIS_MODULE,
2965 	.open		=	igmp_mcf_seq_open,
2966 	.read		=	seq_read,
2967 	.llseek		=	seq_lseek,
2968 	.release	=	seq_release_net,
2969 };
2970 
2971 static int __net_init igmp_net_init(struct net *net)
2972 {
2973 	struct proc_dir_entry *pde;
2974 	int err;
2975 
2976 	pde = proc_create("igmp", S_IRUGO, net->proc_net, &igmp_mc_seq_fops);
2977 	if (!pde)
2978 		goto out_igmp;
2979 	pde = proc_create("mcfilter", S_IRUGO, net->proc_net,
2980 			  &igmp_mcf_seq_fops);
2981 	if (!pde)
2982 		goto out_mcfilter;
2983 	err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
2984 				   SOCK_DGRAM, 0, net);
2985 	if (err < 0) {
2986 		pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
2987 		       err);
2988 		goto out_sock;
2989 	}
2990 
2991 	return 0;
2992 
2993 out_sock:
2994 	remove_proc_entry("mcfilter", net->proc_net);
2995 out_mcfilter:
2996 	remove_proc_entry("igmp", net->proc_net);
2997 out_igmp:
2998 	return -ENOMEM;
2999 }
3000 
3001 static void __net_exit igmp_net_exit(struct net *net)
3002 {
3003 	remove_proc_entry("mcfilter", net->proc_net);
3004 	remove_proc_entry("igmp", net->proc_net);
3005 	inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3006 }
3007 
3008 static struct pernet_operations igmp_net_ops = {
3009 	.init = igmp_net_init,
3010 	.exit = igmp_net_exit,
3011 };
3012 #endif
3013 
3014 static int igmp_netdev_event(struct notifier_block *this,
3015 			     unsigned long event, void *ptr)
3016 {
3017 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3018 	struct in_device *in_dev;
3019 
3020 	switch (event) {
3021 	case NETDEV_RESEND_IGMP:
3022 		in_dev = __in_dev_get_rtnl(dev);
3023 		if (in_dev)
3024 			ip_mc_rejoin_groups(in_dev);
3025 		break;
3026 	default:
3027 		break;
3028 	}
3029 	return NOTIFY_DONE;
3030 }
3031 
3032 static struct notifier_block igmp_notifier = {
3033 	.notifier_call = igmp_netdev_event,
3034 };
3035 
3036 int __init igmp_mc_init(void)
3037 {
3038 #if defined(CONFIG_PROC_FS)
3039 	int err;
3040 
3041 	err = register_pernet_subsys(&igmp_net_ops);
3042 	if (err)
3043 		return err;
3044 	err = register_netdevice_notifier(&igmp_notifier);
3045 	if (err)
3046 		goto reg_notif_fail;
3047 	return 0;
3048 
3049 reg_notif_fail:
3050 	unregister_pernet_subsys(&igmp_net_ops);
3051 	return err;
3052 #else
3053 	return register_netdevice_notifier(&igmp_notifier);
3054 #endif
3055 }
3056