xref: /openbmc/linux/net/ipv4/igmp.c (revision 4a44a19b)
1 /*
2  *	Linux NET3:	Internet Group Management Protocol  [IGMP]
3  *
4  *	This code implements the IGMP protocol as defined in RFC1112. There has
5  *	been a further revision of this protocol since which is now supported.
6  *
7  *	If you have trouble with this module be careful what gcc you have used,
8  *	the older version didn't come out right using gcc 2.5.8, the newer one
9  *	seems to fall out with gcc 2.6.2.
10  *
11  *	Authors:
12  *		Alan Cox <alan@lxorguk.ukuu.org.uk>
13  *
14  *	This program is free software; you can redistribute it and/or
15  *	modify it under the terms of the GNU General Public License
16  *	as published by the Free Software Foundation; either version
17  *	2 of the License, or (at your option) any later version.
18  *
19  *	Fixes:
20  *
21  *		Alan Cox	:	Added lots of __inline__ to optimise
22  *					the memory usage of all the tiny little
23  *					functions.
24  *		Alan Cox	:	Dumped the header building experiment.
25  *		Alan Cox	:	Minor tweaks ready for multicast routing
26  *					and extended IGMP protocol.
27  *		Alan Cox	:	Removed a load of inline directives. Gcc 2.5.8
28  *					writes utterly bogus code otherwise (sigh)
29  *					fixed IGMP loopback to behave in the manner
30  *					desired by mrouted, fixed the fact it has been
31  *					broken since 1.3.6 and cleaned up a few minor
32  *					points.
33  *
34  *		Chih-Jen Chang	:	Tried to revise IGMP to Version 2
35  *		Tsu-Sheng Tsao		E-mail: chihjenc@scf.usc.edu and tsusheng@scf.usc.edu
36  *					The enhancements are mainly based on Steve Deering's
37  * 					ipmulti-3.5 source code.
38  *		Chih-Jen Chang	:	Added the igmp_get_mrouter_info and
39  *		Tsu-Sheng Tsao		igmp_set_mrouter_info to keep track of
40  *					the mrouted version on that device.
41  *		Chih-Jen Chang	:	Added the max_resp_time parameter to
42  *		Tsu-Sheng Tsao		igmp_heard_query(). Using this parameter
43  *					to identify the multicast router version
44  *					and do what the IGMP version 2 specified.
45  *		Chih-Jen Chang	:	Added a timer to revert to IGMP V2 router
46  *		Tsu-Sheng Tsao		if the specified time expired.
47  *		Alan Cox	:	Stop IGMP from 0.0.0.0 being accepted.
48  *		Alan Cox	:	Use GFP_ATOMIC in the right places.
49  *		Christian Daudt :	igmp timer wasn't set for local group
50  *					memberships but was being deleted,
51  *					which caused a "del_timer() called
52  *					from %p with timer not initialized\n"
53  *					message (960131).
54  *		Christian Daudt :	removed del_timer from
55  *					igmp_timer_expire function (960205).
56  *             Christian Daudt :       igmp_heard_report now only calls
57  *                                     igmp_timer_expire if tm->running is
58  *                                     true (960216).
59  *		Malcolm Beattie :	ttl comparison wrong in igmp_rcv made
60  *					igmp_heard_query never trigger. Expiry
61  *					miscalculation fixed in igmp_heard_query
62  *					and random() made to return unsigned to
63  *					prevent negative expiry times.
64  *		Alexey Kuznetsov:	Wrong group leaving behaviour, backport
65  *					fix from pending 2.1.x patches.
66  *		Alan Cox:		Forget to enable FDDI support earlier.
67  *		Alexey Kuznetsov:	Fixed leaving groups on device down.
68  *		Alexey Kuznetsov:	Accordance to igmp-v2-06 draft.
69  *		David L Stevens:	IGMPv3 support, with help from
70  *					Vinay Kulkarni
71  */
72 
73 #include <linux/module.h>
74 #include <linux/slab.h>
75 #include <asm/uaccess.h>
76 #include <linux/types.h>
77 #include <linux/kernel.h>
78 #include <linux/jiffies.h>
79 #include <linux/string.h>
80 #include <linux/socket.h>
81 #include <linux/sockios.h>
82 #include <linux/in.h>
83 #include <linux/inet.h>
84 #include <linux/netdevice.h>
85 #include <linux/skbuff.h>
86 #include <linux/inetdevice.h>
87 #include <linux/igmp.h>
88 #include <linux/if_arp.h>
89 #include <linux/rtnetlink.h>
90 #include <linux/times.h>
91 #include <linux/pkt_sched.h>
92 
93 #include <net/net_namespace.h>
94 #include <net/arp.h>
95 #include <net/ip.h>
96 #include <net/protocol.h>
97 #include <net/route.h>
98 #include <net/sock.h>
99 #include <net/checksum.h>
100 #include <linux/netfilter_ipv4.h>
101 #ifdef CONFIG_IP_MROUTE
102 #include <linux/mroute.h>
103 #endif
104 #ifdef CONFIG_PROC_FS
105 #include <linux/proc_fs.h>
106 #include <linux/seq_file.h>
107 #endif
108 
109 #define IP_MAX_MEMBERSHIPS	20
110 #define IP_MAX_MSF		10
111 
112 #ifdef CONFIG_IP_MULTICAST
113 /* Parameter names and values are taken from igmp-v2-06 draft */
114 
115 #define IGMP_V1_Router_Present_Timeout		(400*HZ)
116 #define IGMP_V2_Router_Present_Timeout		(400*HZ)
117 #define IGMP_V2_Unsolicited_Report_Interval	(10*HZ)
118 #define IGMP_V3_Unsolicited_Report_Interval	(1*HZ)
119 #define IGMP_Query_Response_Interval		(10*HZ)
120 #define IGMP_Query_Robustness_Variable		2
121 
122 
123 #define IGMP_Initial_Report_Delay		(1)
124 
125 /* IGMP_Initial_Report_Delay is not from IGMP specs!
126  * IGMP specs require to report membership immediately after
127  * joining a group, but we delay the first report by a
128  * small interval. It seems more natural and still does not
129  * contradict to specs provided this delay is small enough.
130  */
131 
132 #define IGMP_V1_SEEN(in_dev) \
133 	(IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
134 	 IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
135 	 ((in_dev)->mr_v1_seen && \
136 	  time_before(jiffies, (in_dev)->mr_v1_seen)))
137 #define IGMP_V2_SEEN(in_dev) \
138 	(IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
139 	 IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
140 	 ((in_dev)->mr_v2_seen && \
141 	  time_before(jiffies, (in_dev)->mr_v2_seen)))
142 
143 static int unsolicited_report_interval(struct in_device *in_dev)
144 {
145 	int interval_ms, interval_jiffies;
146 
147 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
148 		interval_ms = IN_DEV_CONF_GET(
149 			in_dev,
150 			IGMPV2_UNSOLICITED_REPORT_INTERVAL);
151 	else /* v3 */
152 		interval_ms = IN_DEV_CONF_GET(
153 			in_dev,
154 			IGMPV3_UNSOLICITED_REPORT_INTERVAL);
155 
156 	interval_jiffies = msecs_to_jiffies(interval_ms);
157 
158 	/* _timer functions can't handle a delay of 0 jiffies so ensure
159 	 *  we always return a positive value.
160 	 */
161 	if (interval_jiffies <= 0)
162 		interval_jiffies = 1;
163 	return interval_jiffies;
164 }
165 
166 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im);
167 static void igmpv3_del_delrec(struct in_device *in_dev, __be32 multiaddr);
168 static void igmpv3_clear_delrec(struct in_device *in_dev);
169 static int sf_setstate(struct ip_mc_list *pmc);
170 static void sf_markstate(struct ip_mc_list *pmc);
171 #endif
172 static void ip_mc_clear_src(struct ip_mc_list *pmc);
173 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
174 			 int sfcount, __be32 *psfsrc, int delta);
175 
176 static void ip_ma_put(struct ip_mc_list *im)
177 {
178 	if (atomic_dec_and_test(&im->refcnt)) {
179 		in_dev_put(im->interface);
180 		kfree_rcu(im, rcu);
181 	}
182 }
183 
184 #define for_each_pmc_rcu(in_dev, pmc)				\
185 	for (pmc = rcu_dereference(in_dev->mc_list);		\
186 	     pmc != NULL;					\
187 	     pmc = rcu_dereference(pmc->next_rcu))
188 
189 #define for_each_pmc_rtnl(in_dev, pmc)				\
190 	for (pmc = rtnl_dereference(in_dev->mc_list);		\
191 	     pmc != NULL;					\
192 	     pmc = rtnl_dereference(pmc->next_rcu))
193 
194 #ifdef CONFIG_IP_MULTICAST
195 
196 /*
197  *	Timer management
198  */
199 
200 static void igmp_stop_timer(struct ip_mc_list *im)
201 {
202 	spin_lock_bh(&im->lock);
203 	if (del_timer(&im->timer))
204 		atomic_dec(&im->refcnt);
205 	im->tm_running = 0;
206 	im->reporter = 0;
207 	im->unsolicit_count = 0;
208 	spin_unlock_bh(&im->lock);
209 }
210 
211 /* It must be called with locked im->lock */
212 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
213 {
214 	int tv = prandom_u32() % max_delay;
215 
216 	im->tm_running = 1;
217 	if (!mod_timer(&im->timer, jiffies+tv+2))
218 		atomic_inc(&im->refcnt);
219 }
220 
221 static void igmp_gq_start_timer(struct in_device *in_dev)
222 {
223 	int tv = prandom_u32() % in_dev->mr_maxdelay;
224 
225 	in_dev->mr_gq_running = 1;
226 	if (!mod_timer(&in_dev->mr_gq_timer, jiffies+tv+2))
227 		in_dev_hold(in_dev);
228 }
229 
230 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
231 {
232 	int tv = prandom_u32() % delay;
233 
234 	if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
235 		in_dev_hold(in_dev);
236 }
237 
238 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
239 {
240 	spin_lock_bh(&im->lock);
241 	im->unsolicit_count = 0;
242 	if (del_timer(&im->timer)) {
243 		if ((long)(im->timer.expires-jiffies) < max_delay) {
244 			add_timer(&im->timer);
245 			im->tm_running = 1;
246 			spin_unlock_bh(&im->lock);
247 			return;
248 		}
249 		atomic_dec(&im->refcnt);
250 	}
251 	igmp_start_timer(im, max_delay);
252 	spin_unlock_bh(&im->lock);
253 }
254 
255 
256 /*
257  *	Send an IGMP report.
258  */
259 
260 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
261 
262 
263 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
264 	int gdeleted, int sdeleted)
265 {
266 	switch (type) {
267 	case IGMPV3_MODE_IS_INCLUDE:
268 	case IGMPV3_MODE_IS_EXCLUDE:
269 		if (gdeleted || sdeleted)
270 			return 0;
271 		if (!(pmc->gsquery && !psf->sf_gsresp)) {
272 			if (pmc->sfmode == MCAST_INCLUDE)
273 				return 1;
274 			/* don't include if this source is excluded
275 			 * in all filters
276 			 */
277 			if (psf->sf_count[MCAST_INCLUDE])
278 				return type == IGMPV3_MODE_IS_INCLUDE;
279 			return pmc->sfcount[MCAST_EXCLUDE] ==
280 				psf->sf_count[MCAST_EXCLUDE];
281 		}
282 		return 0;
283 	case IGMPV3_CHANGE_TO_INCLUDE:
284 		if (gdeleted || sdeleted)
285 			return 0;
286 		return psf->sf_count[MCAST_INCLUDE] != 0;
287 	case IGMPV3_CHANGE_TO_EXCLUDE:
288 		if (gdeleted || sdeleted)
289 			return 0;
290 		if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
291 		    psf->sf_count[MCAST_INCLUDE])
292 			return 0;
293 		return pmc->sfcount[MCAST_EXCLUDE] ==
294 			psf->sf_count[MCAST_EXCLUDE];
295 	case IGMPV3_ALLOW_NEW_SOURCES:
296 		if (gdeleted || !psf->sf_crcount)
297 			return 0;
298 		return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
299 	case IGMPV3_BLOCK_OLD_SOURCES:
300 		if (pmc->sfmode == MCAST_INCLUDE)
301 			return gdeleted || (psf->sf_crcount && sdeleted);
302 		return psf->sf_crcount && !gdeleted && !sdeleted;
303 	}
304 	return 0;
305 }
306 
307 static int
308 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
309 {
310 	struct ip_sf_list *psf;
311 	int scount = 0;
312 
313 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
314 		if (!is_in(pmc, psf, type, gdeleted, sdeleted))
315 			continue;
316 		scount++;
317 	}
318 	return scount;
319 }
320 
321 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
322 {
323 	struct sk_buff *skb;
324 	struct rtable *rt;
325 	struct iphdr *pip;
326 	struct igmpv3_report *pig;
327 	struct net *net = dev_net(dev);
328 	struct flowi4 fl4;
329 	int hlen = LL_RESERVED_SPACE(dev);
330 	int tlen = dev->needed_tailroom;
331 	unsigned int size = mtu;
332 
333 	while (1) {
334 		skb = alloc_skb(size + hlen + tlen,
335 				GFP_ATOMIC | __GFP_NOWARN);
336 		if (skb)
337 			break;
338 		size >>= 1;
339 		if (size < 256)
340 			return NULL;
341 	}
342 	skb->priority = TC_PRIO_CONTROL;
343 
344 	rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
345 				   0, 0,
346 				   IPPROTO_IGMP, 0, dev->ifindex);
347 	if (IS_ERR(rt)) {
348 		kfree_skb(skb);
349 		return NULL;
350 	}
351 
352 	skb_dst_set(skb, &rt->dst);
353 	skb->dev = dev;
354 
355 	skb->reserved_tailroom = skb_end_offset(skb) -
356 				 min(mtu, skb_end_offset(skb));
357 	skb_reserve(skb, hlen);
358 
359 	skb_reset_network_header(skb);
360 	pip = ip_hdr(skb);
361 	skb_put(skb, sizeof(struct iphdr) + 4);
362 
363 	pip->version  = 4;
364 	pip->ihl      = (sizeof(struct iphdr)+4)>>2;
365 	pip->tos      = 0xc0;
366 	pip->frag_off = htons(IP_DF);
367 	pip->ttl      = 1;
368 	pip->daddr    = fl4.daddr;
369 	pip->saddr    = fl4.saddr;
370 	pip->protocol = IPPROTO_IGMP;
371 	pip->tot_len  = 0;	/* filled in later */
372 	ip_select_ident(skb, NULL);
373 	((u8 *)&pip[1])[0] = IPOPT_RA;
374 	((u8 *)&pip[1])[1] = 4;
375 	((u8 *)&pip[1])[2] = 0;
376 	((u8 *)&pip[1])[3] = 0;
377 
378 	skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
379 	skb_put(skb, sizeof(*pig));
380 	pig = igmpv3_report_hdr(skb);
381 	pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
382 	pig->resv1 = 0;
383 	pig->csum = 0;
384 	pig->resv2 = 0;
385 	pig->ngrec = 0;
386 	return skb;
387 }
388 
389 static int igmpv3_sendpack(struct sk_buff *skb)
390 {
391 	struct igmphdr *pig = igmp_hdr(skb);
392 	const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
393 
394 	pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
395 
396 	return ip_local_out(skb);
397 }
398 
399 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
400 {
401 	return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
402 }
403 
404 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
405 	int type, struct igmpv3_grec **ppgr)
406 {
407 	struct net_device *dev = pmc->interface->dev;
408 	struct igmpv3_report *pih;
409 	struct igmpv3_grec *pgr;
410 
411 	if (!skb)
412 		skb = igmpv3_newpack(dev, dev->mtu);
413 	if (!skb)
414 		return NULL;
415 	pgr = (struct igmpv3_grec *)skb_put(skb, sizeof(struct igmpv3_grec));
416 	pgr->grec_type = type;
417 	pgr->grec_auxwords = 0;
418 	pgr->grec_nsrcs = 0;
419 	pgr->grec_mca = pmc->multiaddr;
420 	pih = igmpv3_report_hdr(skb);
421 	pih->ngrec = htons(ntohs(pih->ngrec)+1);
422 	*ppgr = pgr;
423 	return skb;
424 }
425 
426 #define AVAILABLE(skb)	((skb) ? skb_availroom(skb) : 0)
427 
428 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
429 	int type, int gdeleted, int sdeleted)
430 {
431 	struct net_device *dev = pmc->interface->dev;
432 	struct igmpv3_report *pih;
433 	struct igmpv3_grec *pgr = NULL;
434 	struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
435 	int scount, stotal, first, isquery, truncate;
436 
437 	if (pmc->multiaddr == IGMP_ALL_HOSTS)
438 		return skb;
439 
440 	isquery = type == IGMPV3_MODE_IS_INCLUDE ||
441 		  type == IGMPV3_MODE_IS_EXCLUDE;
442 	truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
443 		    type == IGMPV3_CHANGE_TO_EXCLUDE;
444 
445 	stotal = scount = 0;
446 
447 	psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
448 
449 	if (!*psf_list)
450 		goto empty_source;
451 
452 	pih = skb ? igmpv3_report_hdr(skb) : NULL;
453 
454 	/* EX and TO_EX get a fresh packet, if needed */
455 	if (truncate) {
456 		if (pih && pih->ngrec &&
457 		    AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
458 			if (skb)
459 				igmpv3_sendpack(skb);
460 			skb = igmpv3_newpack(dev, dev->mtu);
461 		}
462 	}
463 	first = 1;
464 	psf_prev = NULL;
465 	for (psf = *psf_list; psf; psf = psf_next) {
466 		__be32 *psrc;
467 
468 		psf_next = psf->sf_next;
469 
470 		if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
471 			psf_prev = psf;
472 			continue;
473 		}
474 
475 		/* clear marks on query responses */
476 		if (isquery)
477 			psf->sf_gsresp = 0;
478 
479 		if (AVAILABLE(skb) < sizeof(__be32) +
480 		    first*sizeof(struct igmpv3_grec)) {
481 			if (truncate && !first)
482 				break;	 /* truncate these */
483 			if (pgr)
484 				pgr->grec_nsrcs = htons(scount);
485 			if (skb)
486 				igmpv3_sendpack(skb);
487 			skb = igmpv3_newpack(dev, dev->mtu);
488 			first = 1;
489 			scount = 0;
490 		}
491 		if (first) {
492 			skb = add_grhead(skb, pmc, type, &pgr);
493 			first = 0;
494 		}
495 		if (!skb)
496 			return NULL;
497 		psrc = (__be32 *)skb_put(skb, sizeof(__be32));
498 		*psrc = psf->sf_inaddr;
499 		scount++; stotal++;
500 		if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
501 		     type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
502 			psf->sf_crcount--;
503 			if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
504 				if (psf_prev)
505 					psf_prev->sf_next = psf->sf_next;
506 				else
507 					*psf_list = psf->sf_next;
508 				kfree(psf);
509 				continue;
510 			}
511 		}
512 		psf_prev = psf;
513 	}
514 
515 empty_source:
516 	if (!stotal) {
517 		if (type == IGMPV3_ALLOW_NEW_SOURCES ||
518 		    type == IGMPV3_BLOCK_OLD_SOURCES)
519 			return skb;
520 		if (pmc->crcount || isquery) {
521 			/* make sure we have room for group header */
522 			if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
523 				igmpv3_sendpack(skb);
524 				skb = NULL; /* add_grhead will get a new one */
525 			}
526 			skb = add_grhead(skb, pmc, type, &pgr);
527 		}
528 	}
529 	if (pgr)
530 		pgr->grec_nsrcs = htons(scount);
531 
532 	if (isquery)
533 		pmc->gsquery = 0;	/* clear query state on report */
534 	return skb;
535 }
536 
537 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
538 {
539 	struct sk_buff *skb = NULL;
540 	int type;
541 
542 	if (!pmc) {
543 		rcu_read_lock();
544 		for_each_pmc_rcu(in_dev, pmc) {
545 			if (pmc->multiaddr == IGMP_ALL_HOSTS)
546 				continue;
547 			spin_lock_bh(&pmc->lock);
548 			if (pmc->sfcount[MCAST_EXCLUDE])
549 				type = IGMPV3_MODE_IS_EXCLUDE;
550 			else
551 				type = IGMPV3_MODE_IS_INCLUDE;
552 			skb = add_grec(skb, pmc, type, 0, 0);
553 			spin_unlock_bh(&pmc->lock);
554 		}
555 		rcu_read_unlock();
556 	} else {
557 		spin_lock_bh(&pmc->lock);
558 		if (pmc->sfcount[MCAST_EXCLUDE])
559 			type = IGMPV3_MODE_IS_EXCLUDE;
560 		else
561 			type = IGMPV3_MODE_IS_INCLUDE;
562 		skb = add_grec(skb, pmc, type, 0, 0);
563 		spin_unlock_bh(&pmc->lock);
564 	}
565 	if (!skb)
566 		return 0;
567 	return igmpv3_sendpack(skb);
568 }
569 
570 /*
571  * remove zero-count source records from a source filter list
572  */
573 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
574 {
575 	struct ip_sf_list *psf_prev, *psf_next, *psf;
576 
577 	psf_prev = NULL;
578 	for (psf = *ppsf; psf; psf = psf_next) {
579 		psf_next = psf->sf_next;
580 		if (psf->sf_crcount == 0) {
581 			if (psf_prev)
582 				psf_prev->sf_next = psf->sf_next;
583 			else
584 				*ppsf = psf->sf_next;
585 			kfree(psf);
586 		} else
587 			psf_prev = psf;
588 	}
589 }
590 
591 static void igmpv3_send_cr(struct in_device *in_dev)
592 {
593 	struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
594 	struct sk_buff *skb = NULL;
595 	int type, dtype;
596 
597 	rcu_read_lock();
598 	spin_lock_bh(&in_dev->mc_tomb_lock);
599 
600 	/* deleted MCA's */
601 	pmc_prev = NULL;
602 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
603 		pmc_next = pmc->next;
604 		if (pmc->sfmode == MCAST_INCLUDE) {
605 			type = IGMPV3_BLOCK_OLD_SOURCES;
606 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
607 			skb = add_grec(skb, pmc, type, 1, 0);
608 			skb = add_grec(skb, pmc, dtype, 1, 1);
609 		}
610 		if (pmc->crcount) {
611 			if (pmc->sfmode == MCAST_EXCLUDE) {
612 				type = IGMPV3_CHANGE_TO_INCLUDE;
613 				skb = add_grec(skb, pmc, type, 1, 0);
614 			}
615 			pmc->crcount--;
616 			if (pmc->crcount == 0) {
617 				igmpv3_clear_zeros(&pmc->tomb);
618 				igmpv3_clear_zeros(&pmc->sources);
619 			}
620 		}
621 		if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
622 			if (pmc_prev)
623 				pmc_prev->next = pmc_next;
624 			else
625 				in_dev->mc_tomb = pmc_next;
626 			in_dev_put(pmc->interface);
627 			kfree(pmc);
628 		} else
629 			pmc_prev = pmc;
630 	}
631 	spin_unlock_bh(&in_dev->mc_tomb_lock);
632 
633 	/* change recs */
634 	for_each_pmc_rcu(in_dev, pmc) {
635 		spin_lock_bh(&pmc->lock);
636 		if (pmc->sfcount[MCAST_EXCLUDE]) {
637 			type = IGMPV3_BLOCK_OLD_SOURCES;
638 			dtype = IGMPV3_ALLOW_NEW_SOURCES;
639 		} else {
640 			type = IGMPV3_ALLOW_NEW_SOURCES;
641 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
642 		}
643 		skb = add_grec(skb, pmc, type, 0, 0);
644 		skb = add_grec(skb, pmc, dtype, 0, 1);	/* deleted sources */
645 
646 		/* filter mode changes */
647 		if (pmc->crcount) {
648 			if (pmc->sfmode == MCAST_EXCLUDE)
649 				type = IGMPV3_CHANGE_TO_EXCLUDE;
650 			else
651 				type = IGMPV3_CHANGE_TO_INCLUDE;
652 			skb = add_grec(skb, pmc, type, 0, 0);
653 			pmc->crcount--;
654 		}
655 		spin_unlock_bh(&pmc->lock);
656 	}
657 	rcu_read_unlock();
658 
659 	if (!skb)
660 		return;
661 	(void) igmpv3_sendpack(skb);
662 }
663 
664 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
665 	int type)
666 {
667 	struct sk_buff *skb;
668 	struct iphdr *iph;
669 	struct igmphdr *ih;
670 	struct rtable *rt;
671 	struct net_device *dev = in_dev->dev;
672 	struct net *net = dev_net(dev);
673 	__be32	group = pmc ? pmc->multiaddr : 0;
674 	struct flowi4 fl4;
675 	__be32	dst;
676 	int hlen, tlen;
677 
678 	if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
679 		return igmpv3_send_report(in_dev, pmc);
680 	else if (type == IGMP_HOST_LEAVE_MESSAGE)
681 		dst = IGMP_ALL_ROUTER;
682 	else
683 		dst = group;
684 
685 	rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
686 				   0, 0,
687 				   IPPROTO_IGMP, 0, dev->ifindex);
688 	if (IS_ERR(rt))
689 		return -1;
690 
691 	hlen = LL_RESERVED_SPACE(dev);
692 	tlen = dev->needed_tailroom;
693 	skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
694 	if (skb == NULL) {
695 		ip_rt_put(rt);
696 		return -1;
697 	}
698 	skb->priority = TC_PRIO_CONTROL;
699 
700 	skb_dst_set(skb, &rt->dst);
701 
702 	skb_reserve(skb, hlen);
703 
704 	skb_reset_network_header(skb);
705 	iph = ip_hdr(skb);
706 	skb_put(skb, sizeof(struct iphdr) + 4);
707 
708 	iph->version  = 4;
709 	iph->ihl      = (sizeof(struct iphdr)+4)>>2;
710 	iph->tos      = 0xc0;
711 	iph->frag_off = htons(IP_DF);
712 	iph->ttl      = 1;
713 	iph->daddr    = dst;
714 	iph->saddr    = fl4.saddr;
715 	iph->protocol = IPPROTO_IGMP;
716 	ip_select_ident(skb, NULL);
717 	((u8 *)&iph[1])[0] = IPOPT_RA;
718 	((u8 *)&iph[1])[1] = 4;
719 	((u8 *)&iph[1])[2] = 0;
720 	((u8 *)&iph[1])[3] = 0;
721 
722 	ih = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
723 	ih->type = type;
724 	ih->code = 0;
725 	ih->csum = 0;
726 	ih->group = group;
727 	ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
728 
729 	return ip_local_out(skb);
730 }
731 
732 static void igmp_gq_timer_expire(unsigned long data)
733 {
734 	struct in_device *in_dev = (struct in_device *)data;
735 
736 	in_dev->mr_gq_running = 0;
737 	igmpv3_send_report(in_dev, NULL);
738 	in_dev_put(in_dev);
739 }
740 
741 static void igmp_ifc_timer_expire(unsigned long data)
742 {
743 	struct in_device *in_dev = (struct in_device *)data;
744 
745 	igmpv3_send_cr(in_dev);
746 	if (in_dev->mr_ifc_count) {
747 		in_dev->mr_ifc_count--;
748 		igmp_ifc_start_timer(in_dev,
749 				     unsolicited_report_interval(in_dev));
750 	}
751 	in_dev_put(in_dev);
752 }
753 
754 static void igmp_ifc_event(struct in_device *in_dev)
755 {
756 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
757 		return;
758 	in_dev->mr_ifc_count = in_dev->mr_qrv ?: sysctl_igmp_qrv;
759 	igmp_ifc_start_timer(in_dev, 1);
760 }
761 
762 
763 static void igmp_timer_expire(unsigned long data)
764 {
765 	struct ip_mc_list *im = (struct ip_mc_list *)data;
766 	struct in_device *in_dev = im->interface;
767 
768 	spin_lock(&im->lock);
769 	im->tm_running = 0;
770 
771 	if (im->unsolicit_count) {
772 		im->unsolicit_count--;
773 		igmp_start_timer(im, unsolicited_report_interval(in_dev));
774 	}
775 	im->reporter = 1;
776 	spin_unlock(&im->lock);
777 
778 	if (IGMP_V1_SEEN(in_dev))
779 		igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
780 	else if (IGMP_V2_SEEN(in_dev))
781 		igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
782 	else
783 		igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
784 
785 	ip_ma_put(im);
786 }
787 
788 /* mark EXCLUDE-mode sources */
789 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
790 {
791 	struct ip_sf_list *psf;
792 	int i, scount;
793 
794 	scount = 0;
795 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
796 		if (scount == nsrcs)
797 			break;
798 		for (i = 0; i < nsrcs; i++) {
799 			/* skip inactive filters */
800 			if (psf->sf_count[MCAST_INCLUDE] ||
801 			    pmc->sfcount[MCAST_EXCLUDE] !=
802 			    psf->sf_count[MCAST_EXCLUDE])
803 				break;
804 			if (srcs[i] == psf->sf_inaddr) {
805 				scount++;
806 				break;
807 			}
808 		}
809 	}
810 	pmc->gsquery = 0;
811 	if (scount == nsrcs)	/* all sources excluded */
812 		return 0;
813 	return 1;
814 }
815 
816 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
817 {
818 	struct ip_sf_list *psf;
819 	int i, scount;
820 
821 	if (pmc->sfmode == MCAST_EXCLUDE)
822 		return igmp_xmarksources(pmc, nsrcs, srcs);
823 
824 	/* mark INCLUDE-mode sources */
825 	scount = 0;
826 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
827 		if (scount == nsrcs)
828 			break;
829 		for (i = 0; i < nsrcs; i++)
830 			if (srcs[i] == psf->sf_inaddr) {
831 				psf->sf_gsresp = 1;
832 				scount++;
833 				break;
834 			}
835 	}
836 	if (!scount) {
837 		pmc->gsquery = 0;
838 		return 0;
839 	}
840 	pmc->gsquery = 1;
841 	return 1;
842 }
843 
844 /* return true if packet was dropped */
845 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
846 {
847 	struct ip_mc_list *im;
848 
849 	/* Timers are only set for non-local groups */
850 
851 	if (group == IGMP_ALL_HOSTS)
852 		return false;
853 
854 	rcu_read_lock();
855 	for_each_pmc_rcu(in_dev, im) {
856 		if (im->multiaddr == group) {
857 			igmp_stop_timer(im);
858 			break;
859 		}
860 	}
861 	rcu_read_unlock();
862 	return false;
863 }
864 
865 /* return true if packet was dropped */
866 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
867 	int len)
868 {
869 	struct igmphdr 		*ih = igmp_hdr(skb);
870 	struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
871 	struct ip_mc_list	*im;
872 	__be32			group = ih->group;
873 	int			max_delay;
874 	int			mark = 0;
875 
876 
877 	if (len == 8) {
878 		if (ih->code == 0) {
879 			/* Alas, old v1 router presents here. */
880 
881 			max_delay = IGMP_Query_Response_Interval;
882 			in_dev->mr_v1_seen = jiffies +
883 				IGMP_V1_Router_Present_Timeout;
884 			group = 0;
885 		} else {
886 			/* v2 router present */
887 			max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
888 			in_dev->mr_v2_seen = jiffies +
889 				IGMP_V2_Router_Present_Timeout;
890 		}
891 		/* cancel the interface change timer */
892 		in_dev->mr_ifc_count = 0;
893 		if (del_timer(&in_dev->mr_ifc_timer))
894 			__in_dev_put(in_dev);
895 		/* clear deleted report items */
896 		igmpv3_clear_delrec(in_dev);
897 	} else if (len < 12) {
898 		return true;	/* ignore bogus packet; freed by caller */
899 	} else if (IGMP_V1_SEEN(in_dev)) {
900 		/* This is a v3 query with v1 queriers present */
901 		max_delay = IGMP_Query_Response_Interval;
902 		group = 0;
903 	} else if (IGMP_V2_SEEN(in_dev)) {
904 		/* this is a v3 query with v2 queriers present;
905 		 * Interpretation of the max_delay code is problematic here.
906 		 * A real v2 host would use ih_code directly, while v3 has a
907 		 * different encoding. We use the v3 encoding as more likely
908 		 * to be intended in a v3 query.
909 		 */
910 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
911 		if (!max_delay)
912 			max_delay = 1;	/* can't mod w/ 0 */
913 	} else { /* v3 */
914 		if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
915 			return true;
916 
917 		ih3 = igmpv3_query_hdr(skb);
918 		if (ih3->nsrcs) {
919 			if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
920 					   + ntohs(ih3->nsrcs)*sizeof(__be32)))
921 				return true;
922 			ih3 = igmpv3_query_hdr(skb);
923 		}
924 
925 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
926 		if (!max_delay)
927 			max_delay = 1;	/* can't mod w/ 0 */
928 		in_dev->mr_maxdelay = max_delay;
929 		if (ih3->qrv)
930 			in_dev->mr_qrv = ih3->qrv;
931 		if (!group) { /* general query */
932 			if (ih3->nsrcs)
933 				return true;	/* no sources allowed */
934 			igmp_gq_start_timer(in_dev);
935 			return false;
936 		}
937 		/* mark sources to include, if group & source-specific */
938 		mark = ih3->nsrcs != 0;
939 	}
940 
941 	/*
942 	 * - Start the timers in all of our membership records
943 	 *   that the query applies to for the interface on
944 	 *   which the query arrived excl. those that belong
945 	 *   to a "local" group (224.0.0.X)
946 	 * - For timers already running check if they need to
947 	 *   be reset.
948 	 * - Use the igmp->igmp_code field as the maximum
949 	 *   delay possible
950 	 */
951 	rcu_read_lock();
952 	for_each_pmc_rcu(in_dev, im) {
953 		int changed;
954 
955 		if (group && group != im->multiaddr)
956 			continue;
957 		if (im->multiaddr == IGMP_ALL_HOSTS)
958 			continue;
959 		spin_lock_bh(&im->lock);
960 		if (im->tm_running)
961 			im->gsquery = im->gsquery && mark;
962 		else
963 			im->gsquery = mark;
964 		changed = !im->gsquery ||
965 			igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
966 		spin_unlock_bh(&im->lock);
967 		if (changed)
968 			igmp_mod_timer(im, max_delay);
969 	}
970 	rcu_read_unlock();
971 	return false;
972 }
973 
974 /* called in rcu_read_lock() section */
975 int igmp_rcv(struct sk_buff *skb)
976 {
977 	/* This basically follows the spec line by line -- see RFC1112 */
978 	struct igmphdr *ih;
979 	struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
980 	int len = skb->len;
981 	bool dropped = true;
982 
983 	if (in_dev == NULL)
984 		goto drop;
985 
986 	if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
987 		goto drop;
988 
989 	if (skb_checksum_simple_validate(skb))
990 		goto drop;
991 
992 	ih = igmp_hdr(skb);
993 	switch (ih->type) {
994 	case IGMP_HOST_MEMBERSHIP_QUERY:
995 		dropped = igmp_heard_query(in_dev, skb, len);
996 		break;
997 	case IGMP_HOST_MEMBERSHIP_REPORT:
998 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
999 		/* Is it our report looped back? */
1000 		if (rt_is_output_route(skb_rtable(skb)))
1001 			break;
1002 		/* don't rely on MC router hearing unicast reports */
1003 		if (skb->pkt_type == PACKET_MULTICAST ||
1004 		    skb->pkt_type == PACKET_BROADCAST)
1005 			dropped = igmp_heard_report(in_dev, ih->group);
1006 		break;
1007 	case IGMP_PIM:
1008 #ifdef CONFIG_IP_PIMSM_V1
1009 		return pim_rcv_v1(skb);
1010 #endif
1011 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
1012 	case IGMP_DVMRP:
1013 	case IGMP_TRACE:
1014 	case IGMP_HOST_LEAVE_MESSAGE:
1015 	case IGMP_MTRACE:
1016 	case IGMP_MTRACE_RESP:
1017 		break;
1018 	default:
1019 		break;
1020 	}
1021 
1022 drop:
1023 	if (dropped)
1024 		kfree_skb(skb);
1025 	else
1026 		consume_skb(skb);
1027 	return 0;
1028 }
1029 
1030 #endif
1031 
1032 
1033 /*
1034  *	Add a filter to a device
1035  */
1036 
1037 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1038 {
1039 	char buf[MAX_ADDR_LEN];
1040 	struct net_device *dev = in_dev->dev;
1041 
1042 	/* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1043 	   We will get multicast token leakage, when IFF_MULTICAST
1044 	   is changed. This check should be done in ndo_set_rx_mode
1045 	   routine. Something sort of:
1046 	   if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1047 	   --ANK
1048 	   */
1049 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1050 		dev_mc_add(dev, buf);
1051 }
1052 
1053 /*
1054  *	Remove a filter from a device
1055  */
1056 
1057 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1058 {
1059 	char buf[MAX_ADDR_LEN];
1060 	struct net_device *dev = in_dev->dev;
1061 
1062 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1063 		dev_mc_del(dev, buf);
1064 }
1065 
1066 #ifdef CONFIG_IP_MULTICAST
1067 /*
1068  * deleted ip_mc_list manipulation
1069  */
1070 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1071 {
1072 	struct ip_mc_list *pmc;
1073 
1074 	/* this is an "ip_mc_list" for convenience; only the fields below
1075 	 * are actually used. In particular, the refcnt and users are not
1076 	 * used for management of the delete list. Using the same structure
1077 	 * for deleted items allows change reports to use common code with
1078 	 * non-deleted or query-response MCA's.
1079 	 */
1080 	pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1081 	if (!pmc)
1082 		return;
1083 	spin_lock_bh(&im->lock);
1084 	pmc->interface = im->interface;
1085 	in_dev_hold(in_dev);
1086 	pmc->multiaddr = im->multiaddr;
1087 	pmc->crcount = in_dev->mr_qrv ?: sysctl_igmp_qrv;
1088 	pmc->sfmode = im->sfmode;
1089 	if (pmc->sfmode == MCAST_INCLUDE) {
1090 		struct ip_sf_list *psf;
1091 
1092 		pmc->tomb = im->tomb;
1093 		pmc->sources = im->sources;
1094 		im->tomb = im->sources = NULL;
1095 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1096 			psf->sf_crcount = pmc->crcount;
1097 	}
1098 	spin_unlock_bh(&im->lock);
1099 
1100 	spin_lock_bh(&in_dev->mc_tomb_lock);
1101 	pmc->next = in_dev->mc_tomb;
1102 	in_dev->mc_tomb = pmc;
1103 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1104 }
1105 
1106 static void igmpv3_del_delrec(struct in_device *in_dev, __be32 multiaddr)
1107 {
1108 	struct ip_mc_list *pmc, *pmc_prev;
1109 	struct ip_sf_list *psf, *psf_next;
1110 
1111 	spin_lock_bh(&in_dev->mc_tomb_lock);
1112 	pmc_prev = NULL;
1113 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1114 		if (pmc->multiaddr == multiaddr)
1115 			break;
1116 		pmc_prev = pmc;
1117 	}
1118 	if (pmc) {
1119 		if (pmc_prev)
1120 			pmc_prev->next = pmc->next;
1121 		else
1122 			in_dev->mc_tomb = pmc->next;
1123 	}
1124 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1125 	if (pmc) {
1126 		for (psf = pmc->tomb; psf; psf = psf_next) {
1127 			psf_next = psf->sf_next;
1128 			kfree(psf);
1129 		}
1130 		in_dev_put(pmc->interface);
1131 		kfree(pmc);
1132 	}
1133 }
1134 
1135 static void igmpv3_clear_delrec(struct in_device *in_dev)
1136 {
1137 	struct ip_mc_list *pmc, *nextpmc;
1138 
1139 	spin_lock_bh(&in_dev->mc_tomb_lock);
1140 	pmc = in_dev->mc_tomb;
1141 	in_dev->mc_tomb = NULL;
1142 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1143 
1144 	for (; pmc; pmc = nextpmc) {
1145 		nextpmc = pmc->next;
1146 		ip_mc_clear_src(pmc);
1147 		in_dev_put(pmc->interface);
1148 		kfree(pmc);
1149 	}
1150 	/* clear dead sources, too */
1151 	rcu_read_lock();
1152 	for_each_pmc_rcu(in_dev, pmc) {
1153 		struct ip_sf_list *psf, *psf_next;
1154 
1155 		spin_lock_bh(&pmc->lock);
1156 		psf = pmc->tomb;
1157 		pmc->tomb = NULL;
1158 		spin_unlock_bh(&pmc->lock);
1159 		for (; psf; psf = psf_next) {
1160 			psf_next = psf->sf_next;
1161 			kfree(psf);
1162 		}
1163 	}
1164 	rcu_read_unlock();
1165 }
1166 #endif
1167 
1168 static void igmp_group_dropped(struct ip_mc_list *im)
1169 {
1170 	struct in_device *in_dev = im->interface;
1171 #ifdef CONFIG_IP_MULTICAST
1172 	int reporter;
1173 #endif
1174 
1175 	if (im->loaded) {
1176 		im->loaded = 0;
1177 		ip_mc_filter_del(in_dev, im->multiaddr);
1178 	}
1179 
1180 #ifdef CONFIG_IP_MULTICAST
1181 	if (im->multiaddr == IGMP_ALL_HOSTS)
1182 		return;
1183 
1184 	reporter = im->reporter;
1185 	igmp_stop_timer(im);
1186 
1187 	if (!in_dev->dead) {
1188 		if (IGMP_V1_SEEN(in_dev))
1189 			return;
1190 		if (IGMP_V2_SEEN(in_dev)) {
1191 			if (reporter)
1192 				igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1193 			return;
1194 		}
1195 		/* IGMPv3 */
1196 		igmpv3_add_delrec(in_dev, im);
1197 
1198 		igmp_ifc_event(in_dev);
1199 	}
1200 #endif
1201 }
1202 
1203 static void igmp_group_added(struct ip_mc_list *im)
1204 {
1205 	struct in_device *in_dev = im->interface;
1206 
1207 	if (im->loaded == 0) {
1208 		im->loaded = 1;
1209 		ip_mc_filter_add(in_dev, im->multiaddr);
1210 	}
1211 
1212 #ifdef CONFIG_IP_MULTICAST
1213 	if (im->multiaddr == IGMP_ALL_HOSTS)
1214 		return;
1215 
1216 	if (in_dev->dead)
1217 		return;
1218 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1219 		spin_lock_bh(&im->lock);
1220 		igmp_start_timer(im, IGMP_Initial_Report_Delay);
1221 		spin_unlock_bh(&im->lock);
1222 		return;
1223 	}
1224 	/* else, v3 */
1225 
1226 	im->crcount = in_dev->mr_qrv ?: sysctl_igmp_qrv;
1227 	igmp_ifc_event(in_dev);
1228 #endif
1229 }
1230 
1231 
1232 /*
1233  *	Multicast list managers
1234  */
1235 
1236 static u32 ip_mc_hash(const struct ip_mc_list *im)
1237 {
1238 	return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1239 }
1240 
1241 static void ip_mc_hash_add(struct in_device *in_dev,
1242 			   struct ip_mc_list *im)
1243 {
1244 	struct ip_mc_list __rcu **mc_hash;
1245 	u32 hash;
1246 
1247 	mc_hash = rtnl_dereference(in_dev->mc_hash);
1248 	if (mc_hash) {
1249 		hash = ip_mc_hash(im);
1250 		im->next_hash = mc_hash[hash];
1251 		rcu_assign_pointer(mc_hash[hash], im);
1252 		return;
1253 	}
1254 
1255 	/* do not use a hash table for small number of items */
1256 	if (in_dev->mc_count < 4)
1257 		return;
1258 
1259 	mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1260 			  GFP_KERNEL);
1261 	if (!mc_hash)
1262 		return;
1263 
1264 	for_each_pmc_rtnl(in_dev, im) {
1265 		hash = ip_mc_hash(im);
1266 		im->next_hash = mc_hash[hash];
1267 		RCU_INIT_POINTER(mc_hash[hash], im);
1268 	}
1269 
1270 	rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1271 }
1272 
1273 static void ip_mc_hash_remove(struct in_device *in_dev,
1274 			      struct ip_mc_list *im)
1275 {
1276 	struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1277 	struct ip_mc_list *aux;
1278 
1279 	if (!mc_hash)
1280 		return;
1281 	mc_hash += ip_mc_hash(im);
1282 	while ((aux = rtnl_dereference(*mc_hash)) != im)
1283 		mc_hash = &aux->next_hash;
1284 	*mc_hash = im->next_hash;
1285 }
1286 
1287 
1288 /*
1289  *	A socket has joined a multicast group on device dev.
1290  */
1291 
1292 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1293 {
1294 	struct ip_mc_list *im;
1295 
1296 	ASSERT_RTNL();
1297 
1298 	for_each_pmc_rtnl(in_dev, im) {
1299 		if (im->multiaddr == addr) {
1300 			im->users++;
1301 			ip_mc_add_src(in_dev, &addr, MCAST_EXCLUDE, 0, NULL, 0);
1302 			goto out;
1303 		}
1304 	}
1305 
1306 	im = kzalloc(sizeof(*im), GFP_KERNEL);
1307 	if (!im)
1308 		goto out;
1309 
1310 	im->users = 1;
1311 	im->interface = in_dev;
1312 	in_dev_hold(in_dev);
1313 	im->multiaddr = addr;
1314 	/* initial mode is (EX, empty) */
1315 	im->sfmode = MCAST_EXCLUDE;
1316 	im->sfcount[MCAST_EXCLUDE] = 1;
1317 	atomic_set(&im->refcnt, 1);
1318 	spin_lock_init(&im->lock);
1319 #ifdef CONFIG_IP_MULTICAST
1320 	setup_timer(&im->timer, igmp_timer_expire, (unsigned long)im);
1321 	im->unsolicit_count = sysctl_igmp_qrv;
1322 #endif
1323 
1324 	im->next_rcu = in_dev->mc_list;
1325 	in_dev->mc_count++;
1326 	rcu_assign_pointer(in_dev->mc_list, im);
1327 
1328 	ip_mc_hash_add(in_dev, im);
1329 
1330 #ifdef CONFIG_IP_MULTICAST
1331 	igmpv3_del_delrec(in_dev, im->multiaddr);
1332 #endif
1333 	igmp_group_added(im);
1334 	if (!in_dev->dead)
1335 		ip_rt_multicast_event(in_dev);
1336 out:
1337 	return;
1338 }
1339 EXPORT_SYMBOL(ip_mc_inc_group);
1340 
1341 /*
1342  *	Resend IGMP JOIN report; used by netdev notifier.
1343  */
1344 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1345 {
1346 #ifdef CONFIG_IP_MULTICAST
1347 	struct ip_mc_list *im;
1348 	int type;
1349 
1350 	ASSERT_RTNL();
1351 
1352 	for_each_pmc_rtnl(in_dev, im) {
1353 		if (im->multiaddr == IGMP_ALL_HOSTS)
1354 			continue;
1355 
1356 		/* a failover is happening and switches
1357 		 * must be notified immediately
1358 		 */
1359 		if (IGMP_V1_SEEN(in_dev))
1360 			type = IGMP_HOST_MEMBERSHIP_REPORT;
1361 		else if (IGMP_V2_SEEN(in_dev))
1362 			type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1363 		else
1364 			type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1365 		igmp_send_report(in_dev, im, type);
1366 	}
1367 #endif
1368 }
1369 
1370 /*
1371  *	A socket has left a multicast group on device dev
1372  */
1373 
1374 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1375 {
1376 	struct ip_mc_list *i;
1377 	struct ip_mc_list __rcu **ip;
1378 
1379 	ASSERT_RTNL();
1380 
1381 	for (ip = &in_dev->mc_list;
1382 	     (i = rtnl_dereference(*ip)) != NULL;
1383 	     ip = &i->next_rcu) {
1384 		if (i->multiaddr == addr) {
1385 			if (--i->users == 0) {
1386 				ip_mc_hash_remove(in_dev, i);
1387 				*ip = i->next_rcu;
1388 				in_dev->mc_count--;
1389 				igmp_group_dropped(i);
1390 				ip_mc_clear_src(i);
1391 
1392 				if (!in_dev->dead)
1393 					ip_rt_multicast_event(in_dev);
1394 
1395 				ip_ma_put(i);
1396 				return;
1397 			}
1398 			break;
1399 		}
1400 	}
1401 }
1402 EXPORT_SYMBOL(ip_mc_dec_group);
1403 
1404 /* Device changing type */
1405 
1406 void ip_mc_unmap(struct in_device *in_dev)
1407 {
1408 	struct ip_mc_list *pmc;
1409 
1410 	ASSERT_RTNL();
1411 
1412 	for_each_pmc_rtnl(in_dev, pmc)
1413 		igmp_group_dropped(pmc);
1414 }
1415 
1416 void ip_mc_remap(struct in_device *in_dev)
1417 {
1418 	struct ip_mc_list *pmc;
1419 
1420 	ASSERT_RTNL();
1421 
1422 	for_each_pmc_rtnl(in_dev, pmc)
1423 		igmp_group_added(pmc);
1424 }
1425 
1426 /* Device going down */
1427 
1428 void ip_mc_down(struct in_device *in_dev)
1429 {
1430 	struct ip_mc_list *pmc;
1431 
1432 	ASSERT_RTNL();
1433 
1434 	for_each_pmc_rtnl(in_dev, pmc)
1435 		igmp_group_dropped(pmc);
1436 
1437 #ifdef CONFIG_IP_MULTICAST
1438 	in_dev->mr_ifc_count = 0;
1439 	if (del_timer(&in_dev->mr_ifc_timer))
1440 		__in_dev_put(in_dev);
1441 	in_dev->mr_gq_running = 0;
1442 	if (del_timer(&in_dev->mr_gq_timer))
1443 		__in_dev_put(in_dev);
1444 	igmpv3_clear_delrec(in_dev);
1445 #endif
1446 
1447 	ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1448 }
1449 
1450 void ip_mc_init_dev(struct in_device *in_dev)
1451 {
1452 	ASSERT_RTNL();
1453 
1454 #ifdef CONFIG_IP_MULTICAST
1455 	setup_timer(&in_dev->mr_gq_timer, igmp_gq_timer_expire,
1456 			(unsigned long)in_dev);
1457 	setup_timer(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire,
1458 			(unsigned long)in_dev);
1459 	in_dev->mr_qrv = sysctl_igmp_qrv;
1460 #endif
1461 
1462 	spin_lock_init(&in_dev->mc_tomb_lock);
1463 }
1464 
1465 /* Device going up */
1466 
1467 void ip_mc_up(struct in_device *in_dev)
1468 {
1469 	struct ip_mc_list *pmc;
1470 
1471 	ASSERT_RTNL();
1472 
1473 #ifdef CONFIG_IP_MULTICAST
1474 	in_dev->mr_qrv = sysctl_igmp_qrv;
1475 #endif
1476 	ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1477 
1478 	for_each_pmc_rtnl(in_dev, pmc)
1479 		igmp_group_added(pmc);
1480 }
1481 
1482 /*
1483  *	Device is about to be destroyed: clean up.
1484  */
1485 
1486 void ip_mc_destroy_dev(struct in_device *in_dev)
1487 {
1488 	struct ip_mc_list *i;
1489 
1490 	ASSERT_RTNL();
1491 
1492 	/* Deactivate timers */
1493 	ip_mc_down(in_dev);
1494 
1495 	while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1496 		in_dev->mc_list = i->next_rcu;
1497 		in_dev->mc_count--;
1498 
1499 		/* We've dropped the groups in ip_mc_down already */
1500 		ip_mc_clear_src(i);
1501 		ip_ma_put(i);
1502 	}
1503 }
1504 
1505 /* RTNL is locked */
1506 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1507 {
1508 	struct net_device *dev = NULL;
1509 	struct in_device *idev = NULL;
1510 
1511 	if (imr->imr_ifindex) {
1512 		idev = inetdev_by_index(net, imr->imr_ifindex);
1513 		return idev;
1514 	}
1515 	if (imr->imr_address.s_addr) {
1516 		dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1517 		if (!dev)
1518 			return NULL;
1519 	}
1520 
1521 	if (!dev) {
1522 		struct rtable *rt = ip_route_output(net,
1523 						    imr->imr_multiaddr.s_addr,
1524 						    0, 0, 0);
1525 		if (!IS_ERR(rt)) {
1526 			dev = rt->dst.dev;
1527 			ip_rt_put(rt);
1528 		}
1529 	}
1530 	if (dev) {
1531 		imr->imr_ifindex = dev->ifindex;
1532 		idev = __in_dev_get_rtnl(dev);
1533 	}
1534 	return idev;
1535 }
1536 
1537 /*
1538  *	Join a socket to a group
1539  */
1540 int sysctl_igmp_max_memberships __read_mostly = IP_MAX_MEMBERSHIPS;
1541 int sysctl_igmp_max_msf __read_mostly = IP_MAX_MSF;
1542 #ifdef CONFIG_IP_MULTICAST
1543 int sysctl_igmp_qrv __read_mostly = IGMP_Query_Robustness_Variable;
1544 #endif
1545 
1546 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1547 	__be32 *psfsrc)
1548 {
1549 	struct ip_sf_list *psf, *psf_prev;
1550 	int rv = 0;
1551 
1552 	psf_prev = NULL;
1553 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1554 		if (psf->sf_inaddr == *psfsrc)
1555 			break;
1556 		psf_prev = psf;
1557 	}
1558 	if (!psf || psf->sf_count[sfmode] == 0) {
1559 		/* source filter not found, or count wrong =>  bug */
1560 		return -ESRCH;
1561 	}
1562 	psf->sf_count[sfmode]--;
1563 	if (psf->sf_count[sfmode] == 0) {
1564 		ip_rt_multicast_event(pmc->interface);
1565 	}
1566 	if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1567 #ifdef CONFIG_IP_MULTICAST
1568 		struct in_device *in_dev = pmc->interface;
1569 #endif
1570 
1571 		/* no more filters for this source */
1572 		if (psf_prev)
1573 			psf_prev->sf_next = psf->sf_next;
1574 		else
1575 			pmc->sources = psf->sf_next;
1576 #ifdef CONFIG_IP_MULTICAST
1577 		if (psf->sf_oldin &&
1578 		    !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1579 			psf->sf_crcount = in_dev->mr_qrv ?: sysctl_igmp_qrv;
1580 			psf->sf_next = pmc->tomb;
1581 			pmc->tomb = psf;
1582 			rv = 1;
1583 		} else
1584 #endif
1585 			kfree(psf);
1586 	}
1587 	return rv;
1588 }
1589 
1590 #ifndef CONFIG_IP_MULTICAST
1591 #define igmp_ifc_event(x)	do { } while (0)
1592 #endif
1593 
1594 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1595 			 int sfcount, __be32 *psfsrc, int delta)
1596 {
1597 	struct ip_mc_list *pmc;
1598 	int	changerec = 0;
1599 	int	i, err;
1600 
1601 	if (!in_dev)
1602 		return -ENODEV;
1603 	rcu_read_lock();
1604 	for_each_pmc_rcu(in_dev, pmc) {
1605 		if (*pmca == pmc->multiaddr)
1606 			break;
1607 	}
1608 	if (!pmc) {
1609 		/* MCA not found?? bug */
1610 		rcu_read_unlock();
1611 		return -ESRCH;
1612 	}
1613 	spin_lock_bh(&pmc->lock);
1614 	rcu_read_unlock();
1615 #ifdef CONFIG_IP_MULTICAST
1616 	sf_markstate(pmc);
1617 #endif
1618 	if (!delta) {
1619 		err = -EINVAL;
1620 		if (!pmc->sfcount[sfmode])
1621 			goto out_unlock;
1622 		pmc->sfcount[sfmode]--;
1623 	}
1624 	err = 0;
1625 	for (i = 0; i < sfcount; i++) {
1626 		int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1627 
1628 		changerec |= rv > 0;
1629 		if (!err && rv < 0)
1630 			err = rv;
1631 	}
1632 	if (pmc->sfmode == MCAST_EXCLUDE &&
1633 	    pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1634 	    pmc->sfcount[MCAST_INCLUDE]) {
1635 #ifdef CONFIG_IP_MULTICAST
1636 		struct ip_sf_list *psf;
1637 #endif
1638 
1639 		/* filter mode change */
1640 		pmc->sfmode = MCAST_INCLUDE;
1641 #ifdef CONFIG_IP_MULTICAST
1642 		pmc->crcount = in_dev->mr_qrv ?: sysctl_igmp_qrv;
1643 		in_dev->mr_ifc_count = pmc->crcount;
1644 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1645 			psf->sf_crcount = 0;
1646 		igmp_ifc_event(pmc->interface);
1647 	} else if (sf_setstate(pmc) || changerec) {
1648 		igmp_ifc_event(pmc->interface);
1649 #endif
1650 	}
1651 out_unlock:
1652 	spin_unlock_bh(&pmc->lock);
1653 	return err;
1654 }
1655 
1656 /*
1657  * Add multicast single-source filter to the interface list
1658  */
1659 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1660 	__be32 *psfsrc)
1661 {
1662 	struct ip_sf_list *psf, *psf_prev;
1663 
1664 	psf_prev = NULL;
1665 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1666 		if (psf->sf_inaddr == *psfsrc)
1667 			break;
1668 		psf_prev = psf;
1669 	}
1670 	if (!psf) {
1671 		psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1672 		if (!psf)
1673 			return -ENOBUFS;
1674 		psf->sf_inaddr = *psfsrc;
1675 		if (psf_prev) {
1676 			psf_prev->sf_next = psf;
1677 		} else
1678 			pmc->sources = psf;
1679 	}
1680 	psf->sf_count[sfmode]++;
1681 	if (psf->sf_count[sfmode] == 1) {
1682 		ip_rt_multicast_event(pmc->interface);
1683 	}
1684 	return 0;
1685 }
1686 
1687 #ifdef CONFIG_IP_MULTICAST
1688 static void sf_markstate(struct ip_mc_list *pmc)
1689 {
1690 	struct ip_sf_list *psf;
1691 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1692 
1693 	for (psf = pmc->sources; psf; psf = psf->sf_next)
1694 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1695 			psf->sf_oldin = mca_xcount ==
1696 				psf->sf_count[MCAST_EXCLUDE] &&
1697 				!psf->sf_count[MCAST_INCLUDE];
1698 		} else
1699 			psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1700 }
1701 
1702 static int sf_setstate(struct ip_mc_list *pmc)
1703 {
1704 	struct ip_sf_list *psf, *dpsf;
1705 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1706 	int qrv = pmc->interface->mr_qrv;
1707 	int new_in, rv;
1708 
1709 	rv = 0;
1710 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1711 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1712 			new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1713 				!psf->sf_count[MCAST_INCLUDE];
1714 		} else
1715 			new_in = psf->sf_count[MCAST_INCLUDE] != 0;
1716 		if (new_in) {
1717 			if (!psf->sf_oldin) {
1718 				struct ip_sf_list *prev = NULL;
1719 
1720 				for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
1721 					if (dpsf->sf_inaddr == psf->sf_inaddr)
1722 						break;
1723 					prev = dpsf;
1724 				}
1725 				if (dpsf) {
1726 					if (prev)
1727 						prev->sf_next = dpsf->sf_next;
1728 					else
1729 						pmc->tomb = dpsf->sf_next;
1730 					kfree(dpsf);
1731 				}
1732 				psf->sf_crcount = qrv;
1733 				rv++;
1734 			}
1735 		} else if (psf->sf_oldin) {
1736 
1737 			psf->sf_crcount = 0;
1738 			/*
1739 			 * add or update "delete" records if an active filter
1740 			 * is now inactive
1741 			 */
1742 			for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
1743 				if (dpsf->sf_inaddr == psf->sf_inaddr)
1744 					break;
1745 			if (!dpsf) {
1746 				dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
1747 				if (!dpsf)
1748 					continue;
1749 				*dpsf = *psf;
1750 				/* pmc->lock held by callers */
1751 				dpsf->sf_next = pmc->tomb;
1752 				pmc->tomb = dpsf;
1753 			}
1754 			dpsf->sf_crcount = qrv;
1755 			rv++;
1756 		}
1757 	}
1758 	return rv;
1759 }
1760 #endif
1761 
1762 /*
1763  * Add multicast source filter list to the interface list
1764  */
1765 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1766 			 int sfcount, __be32 *psfsrc, int delta)
1767 {
1768 	struct ip_mc_list *pmc;
1769 	int	isexclude;
1770 	int	i, err;
1771 
1772 	if (!in_dev)
1773 		return -ENODEV;
1774 	rcu_read_lock();
1775 	for_each_pmc_rcu(in_dev, pmc) {
1776 		if (*pmca == pmc->multiaddr)
1777 			break;
1778 	}
1779 	if (!pmc) {
1780 		/* MCA not found?? bug */
1781 		rcu_read_unlock();
1782 		return -ESRCH;
1783 	}
1784 	spin_lock_bh(&pmc->lock);
1785 	rcu_read_unlock();
1786 
1787 #ifdef CONFIG_IP_MULTICAST
1788 	sf_markstate(pmc);
1789 #endif
1790 	isexclude = pmc->sfmode == MCAST_EXCLUDE;
1791 	if (!delta)
1792 		pmc->sfcount[sfmode]++;
1793 	err = 0;
1794 	for (i = 0; i < sfcount; i++) {
1795 		err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
1796 		if (err)
1797 			break;
1798 	}
1799 	if (err) {
1800 		int j;
1801 
1802 		if (!delta)
1803 			pmc->sfcount[sfmode]--;
1804 		for (j = 0; j < i; j++)
1805 			(void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
1806 	} else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
1807 #ifdef CONFIG_IP_MULTICAST
1808 		struct ip_sf_list *psf;
1809 		in_dev = pmc->interface;
1810 #endif
1811 
1812 		/* filter mode change */
1813 		if (pmc->sfcount[MCAST_EXCLUDE])
1814 			pmc->sfmode = MCAST_EXCLUDE;
1815 		else if (pmc->sfcount[MCAST_INCLUDE])
1816 			pmc->sfmode = MCAST_INCLUDE;
1817 #ifdef CONFIG_IP_MULTICAST
1818 		/* else no filters; keep old mode for reports */
1819 
1820 		pmc->crcount = in_dev->mr_qrv ?: sysctl_igmp_qrv;
1821 		in_dev->mr_ifc_count = pmc->crcount;
1822 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1823 			psf->sf_crcount = 0;
1824 		igmp_ifc_event(in_dev);
1825 	} else if (sf_setstate(pmc)) {
1826 		igmp_ifc_event(in_dev);
1827 #endif
1828 	}
1829 	spin_unlock_bh(&pmc->lock);
1830 	return err;
1831 }
1832 
1833 static void ip_mc_clear_src(struct ip_mc_list *pmc)
1834 {
1835 	struct ip_sf_list *psf, *nextpsf;
1836 
1837 	for (psf = pmc->tomb; psf; psf = nextpsf) {
1838 		nextpsf = psf->sf_next;
1839 		kfree(psf);
1840 	}
1841 	pmc->tomb = NULL;
1842 	for (psf = pmc->sources; psf; psf = nextpsf) {
1843 		nextpsf = psf->sf_next;
1844 		kfree(psf);
1845 	}
1846 	pmc->sources = NULL;
1847 	pmc->sfmode = MCAST_EXCLUDE;
1848 	pmc->sfcount[MCAST_INCLUDE] = 0;
1849 	pmc->sfcount[MCAST_EXCLUDE] = 1;
1850 }
1851 
1852 
1853 /*
1854  * Join a multicast group
1855  */
1856 int ip_mc_join_group(struct sock *sk , struct ip_mreqn *imr)
1857 {
1858 	int err;
1859 	__be32 addr = imr->imr_multiaddr.s_addr;
1860 	struct ip_mc_socklist *iml = NULL, *i;
1861 	struct in_device *in_dev;
1862 	struct inet_sock *inet = inet_sk(sk);
1863 	struct net *net = sock_net(sk);
1864 	int ifindex;
1865 	int count = 0;
1866 
1867 	if (!ipv4_is_multicast(addr))
1868 		return -EINVAL;
1869 
1870 	rtnl_lock();
1871 
1872 	in_dev = ip_mc_find_dev(net, imr);
1873 
1874 	if (!in_dev) {
1875 		iml = NULL;
1876 		err = -ENODEV;
1877 		goto done;
1878 	}
1879 
1880 	err = -EADDRINUSE;
1881 	ifindex = imr->imr_ifindex;
1882 	for_each_pmc_rtnl(inet, i) {
1883 		if (i->multi.imr_multiaddr.s_addr == addr &&
1884 		    i->multi.imr_ifindex == ifindex)
1885 			goto done;
1886 		count++;
1887 	}
1888 	err = -ENOBUFS;
1889 	if (count >= sysctl_igmp_max_memberships)
1890 		goto done;
1891 	iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
1892 	if (iml == NULL)
1893 		goto done;
1894 
1895 	memcpy(&iml->multi, imr, sizeof(*imr));
1896 	iml->next_rcu = inet->mc_list;
1897 	iml->sflist = NULL;
1898 	iml->sfmode = MCAST_EXCLUDE;
1899 	rcu_assign_pointer(inet->mc_list, iml);
1900 	ip_mc_inc_group(in_dev, addr);
1901 	err = 0;
1902 done:
1903 	rtnl_unlock();
1904 	return err;
1905 }
1906 EXPORT_SYMBOL(ip_mc_join_group);
1907 
1908 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
1909 			   struct in_device *in_dev)
1910 {
1911 	struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
1912 	int err;
1913 
1914 	if (psf == NULL) {
1915 		/* any-source empty exclude case */
1916 		return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
1917 			iml->sfmode, 0, NULL, 0);
1918 	}
1919 	err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
1920 			iml->sfmode, psf->sl_count, psf->sl_addr, 0);
1921 	RCU_INIT_POINTER(iml->sflist, NULL);
1922 	/* decrease mem now to avoid the memleak warning */
1923 	atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
1924 	kfree_rcu(psf, rcu);
1925 	return err;
1926 }
1927 
1928 /*
1929  *	Ask a socket to leave a group.
1930  */
1931 
1932 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
1933 {
1934 	struct inet_sock *inet = inet_sk(sk);
1935 	struct ip_mc_socklist *iml;
1936 	struct ip_mc_socklist __rcu **imlp;
1937 	struct in_device *in_dev;
1938 	struct net *net = sock_net(sk);
1939 	__be32 group = imr->imr_multiaddr.s_addr;
1940 	u32 ifindex;
1941 	int ret = -EADDRNOTAVAIL;
1942 
1943 	rtnl_lock();
1944 	in_dev = ip_mc_find_dev(net, imr);
1945 	if (!in_dev) {
1946 		ret = -ENODEV;
1947 		goto out;
1948 	}
1949 	ifindex = imr->imr_ifindex;
1950 	for (imlp = &inet->mc_list;
1951 	     (iml = rtnl_dereference(*imlp)) != NULL;
1952 	     imlp = &iml->next_rcu) {
1953 		if (iml->multi.imr_multiaddr.s_addr != group)
1954 			continue;
1955 		if (ifindex) {
1956 			if (iml->multi.imr_ifindex != ifindex)
1957 				continue;
1958 		} else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
1959 				iml->multi.imr_address.s_addr)
1960 			continue;
1961 
1962 		(void) ip_mc_leave_src(sk, iml, in_dev);
1963 
1964 		*imlp = iml->next_rcu;
1965 
1966 		ip_mc_dec_group(in_dev, group);
1967 		rtnl_unlock();
1968 		/* decrease mem now to avoid the memleak warning */
1969 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
1970 		kfree_rcu(iml, rcu);
1971 		return 0;
1972 	}
1973 out:
1974 	rtnl_unlock();
1975 	return ret;
1976 }
1977 EXPORT_SYMBOL(ip_mc_leave_group);
1978 
1979 int ip_mc_source(int add, int omode, struct sock *sk, struct
1980 	ip_mreq_source *mreqs, int ifindex)
1981 {
1982 	int err;
1983 	struct ip_mreqn imr;
1984 	__be32 addr = mreqs->imr_multiaddr;
1985 	struct ip_mc_socklist *pmc;
1986 	struct in_device *in_dev = NULL;
1987 	struct inet_sock *inet = inet_sk(sk);
1988 	struct ip_sf_socklist *psl;
1989 	struct net *net = sock_net(sk);
1990 	int leavegroup = 0;
1991 	int i, j, rv;
1992 
1993 	if (!ipv4_is_multicast(addr))
1994 		return -EINVAL;
1995 
1996 	rtnl_lock();
1997 
1998 	imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
1999 	imr.imr_address.s_addr = mreqs->imr_interface;
2000 	imr.imr_ifindex = ifindex;
2001 	in_dev = ip_mc_find_dev(net, &imr);
2002 
2003 	if (!in_dev) {
2004 		err = -ENODEV;
2005 		goto done;
2006 	}
2007 	err = -EADDRNOTAVAIL;
2008 
2009 	for_each_pmc_rtnl(inet, pmc) {
2010 		if ((pmc->multi.imr_multiaddr.s_addr ==
2011 		     imr.imr_multiaddr.s_addr) &&
2012 		    (pmc->multi.imr_ifindex == imr.imr_ifindex))
2013 			break;
2014 	}
2015 	if (!pmc) {		/* must have a prior join */
2016 		err = -EINVAL;
2017 		goto done;
2018 	}
2019 	/* if a source filter was set, must be the same mode as before */
2020 	if (pmc->sflist) {
2021 		if (pmc->sfmode != omode) {
2022 			err = -EINVAL;
2023 			goto done;
2024 		}
2025 	} else if (pmc->sfmode != omode) {
2026 		/* allow mode switches for empty-set filters */
2027 		ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2028 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2029 			NULL, 0);
2030 		pmc->sfmode = omode;
2031 	}
2032 
2033 	psl = rtnl_dereference(pmc->sflist);
2034 	if (!add) {
2035 		if (!psl)
2036 			goto done;	/* err = -EADDRNOTAVAIL */
2037 		rv = !0;
2038 		for (i = 0; i < psl->sl_count; i++) {
2039 			rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2040 				sizeof(__be32));
2041 			if (rv == 0)
2042 				break;
2043 		}
2044 		if (rv)		/* source not found */
2045 			goto done;	/* err = -EADDRNOTAVAIL */
2046 
2047 		/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2048 		if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2049 			leavegroup = 1;
2050 			goto done;
2051 		}
2052 
2053 		/* update the interface filter */
2054 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2055 			&mreqs->imr_sourceaddr, 1);
2056 
2057 		for (j = i+1; j < psl->sl_count; j++)
2058 			psl->sl_addr[j-1] = psl->sl_addr[j];
2059 		psl->sl_count--;
2060 		err = 0;
2061 		goto done;
2062 	}
2063 	/* else, add a new source to the filter */
2064 
2065 	if (psl && psl->sl_count >= sysctl_igmp_max_msf) {
2066 		err = -ENOBUFS;
2067 		goto done;
2068 	}
2069 	if (!psl || psl->sl_count == psl->sl_max) {
2070 		struct ip_sf_socklist *newpsl;
2071 		int count = IP_SFBLOCK;
2072 
2073 		if (psl)
2074 			count += psl->sl_max;
2075 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2076 		if (!newpsl) {
2077 			err = -ENOBUFS;
2078 			goto done;
2079 		}
2080 		newpsl->sl_max = count;
2081 		newpsl->sl_count = count - IP_SFBLOCK;
2082 		if (psl) {
2083 			for (i = 0; i < psl->sl_count; i++)
2084 				newpsl->sl_addr[i] = psl->sl_addr[i];
2085 			/* decrease mem now to avoid the memleak warning */
2086 			atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2087 			kfree_rcu(psl, rcu);
2088 		}
2089 		rcu_assign_pointer(pmc->sflist, newpsl);
2090 		psl = newpsl;
2091 	}
2092 	rv = 1;	/* > 0 for insert logic below if sl_count is 0 */
2093 	for (i = 0; i < psl->sl_count; i++) {
2094 		rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2095 			sizeof(__be32));
2096 		if (rv == 0)
2097 			break;
2098 	}
2099 	if (rv == 0)		/* address already there is an error */
2100 		goto done;
2101 	for (j = psl->sl_count-1; j >= i; j--)
2102 		psl->sl_addr[j+1] = psl->sl_addr[j];
2103 	psl->sl_addr[i] = mreqs->imr_sourceaddr;
2104 	psl->sl_count++;
2105 	err = 0;
2106 	/* update the interface list */
2107 	ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2108 		&mreqs->imr_sourceaddr, 1);
2109 done:
2110 	rtnl_unlock();
2111 	if (leavegroup)
2112 		return ip_mc_leave_group(sk, &imr);
2113 	return err;
2114 }
2115 
2116 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2117 {
2118 	int err = 0;
2119 	struct ip_mreqn	imr;
2120 	__be32 addr = msf->imsf_multiaddr;
2121 	struct ip_mc_socklist *pmc;
2122 	struct in_device *in_dev;
2123 	struct inet_sock *inet = inet_sk(sk);
2124 	struct ip_sf_socklist *newpsl, *psl;
2125 	struct net *net = sock_net(sk);
2126 	int leavegroup = 0;
2127 
2128 	if (!ipv4_is_multicast(addr))
2129 		return -EINVAL;
2130 	if (msf->imsf_fmode != MCAST_INCLUDE &&
2131 	    msf->imsf_fmode != MCAST_EXCLUDE)
2132 		return -EINVAL;
2133 
2134 	rtnl_lock();
2135 
2136 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2137 	imr.imr_address.s_addr = msf->imsf_interface;
2138 	imr.imr_ifindex = ifindex;
2139 	in_dev = ip_mc_find_dev(net, &imr);
2140 
2141 	if (!in_dev) {
2142 		err = -ENODEV;
2143 		goto done;
2144 	}
2145 
2146 	/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2147 	if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2148 		leavegroup = 1;
2149 		goto done;
2150 	}
2151 
2152 	for_each_pmc_rtnl(inet, pmc) {
2153 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2154 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2155 			break;
2156 	}
2157 	if (!pmc) {		/* must have a prior join */
2158 		err = -EINVAL;
2159 		goto done;
2160 	}
2161 	if (msf->imsf_numsrc) {
2162 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2163 							   GFP_KERNEL);
2164 		if (!newpsl) {
2165 			err = -ENOBUFS;
2166 			goto done;
2167 		}
2168 		newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2169 		memcpy(newpsl->sl_addr, msf->imsf_slist,
2170 			msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2171 		err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2172 			msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2173 		if (err) {
2174 			sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2175 			goto done;
2176 		}
2177 	} else {
2178 		newpsl = NULL;
2179 		(void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2180 				     msf->imsf_fmode, 0, NULL, 0);
2181 	}
2182 	psl = rtnl_dereference(pmc->sflist);
2183 	if (psl) {
2184 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2185 			psl->sl_count, psl->sl_addr, 0);
2186 		/* decrease mem now to avoid the memleak warning */
2187 		atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2188 		kfree_rcu(psl, rcu);
2189 	} else
2190 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2191 			0, NULL, 0);
2192 	rcu_assign_pointer(pmc->sflist, newpsl);
2193 	pmc->sfmode = msf->imsf_fmode;
2194 	err = 0;
2195 done:
2196 	rtnl_unlock();
2197 	if (leavegroup)
2198 		err = ip_mc_leave_group(sk, &imr);
2199 	return err;
2200 }
2201 
2202 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2203 	struct ip_msfilter __user *optval, int __user *optlen)
2204 {
2205 	int err, len, count, copycount;
2206 	struct ip_mreqn	imr;
2207 	__be32 addr = msf->imsf_multiaddr;
2208 	struct ip_mc_socklist *pmc;
2209 	struct in_device *in_dev;
2210 	struct inet_sock *inet = inet_sk(sk);
2211 	struct ip_sf_socklist *psl;
2212 	struct net *net = sock_net(sk);
2213 
2214 	if (!ipv4_is_multicast(addr))
2215 		return -EINVAL;
2216 
2217 	rtnl_lock();
2218 
2219 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2220 	imr.imr_address.s_addr = msf->imsf_interface;
2221 	imr.imr_ifindex = 0;
2222 	in_dev = ip_mc_find_dev(net, &imr);
2223 
2224 	if (!in_dev) {
2225 		err = -ENODEV;
2226 		goto done;
2227 	}
2228 	err = -EADDRNOTAVAIL;
2229 
2230 	for_each_pmc_rtnl(inet, pmc) {
2231 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2232 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2233 			break;
2234 	}
2235 	if (!pmc)		/* must have a prior join */
2236 		goto done;
2237 	msf->imsf_fmode = pmc->sfmode;
2238 	psl = rtnl_dereference(pmc->sflist);
2239 	rtnl_unlock();
2240 	if (!psl) {
2241 		len = 0;
2242 		count = 0;
2243 	} else {
2244 		count = psl->sl_count;
2245 	}
2246 	copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2247 	len = copycount * sizeof(psl->sl_addr[0]);
2248 	msf->imsf_numsrc = count;
2249 	if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2250 	    copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2251 		return -EFAULT;
2252 	}
2253 	if (len &&
2254 	    copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2255 		return -EFAULT;
2256 	return 0;
2257 done:
2258 	rtnl_unlock();
2259 	return err;
2260 }
2261 
2262 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2263 	struct group_filter __user *optval, int __user *optlen)
2264 {
2265 	int err, i, count, copycount;
2266 	struct sockaddr_in *psin;
2267 	__be32 addr;
2268 	struct ip_mc_socklist *pmc;
2269 	struct inet_sock *inet = inet_sk(sk);
2270 	struct ip_sf_socklist *psl;
2271 
2272 	psin = (struct sockaddr_in *)&gsf->gf_group;
2273 	if (psin->sin_family != AF_INET)
2274 		return -EINVAL;
2275 	addr = psin->sin_addr.s_addr;
2276 	if (!ipv4_is_multicast(addr))
2277 		return -EINVAL;
2278 
2279 	rtnl_lock();
2280 
2281 	err = -EADDRNOTAVAIL;
2282 
2283 	for_each_pmc_rtnl(inet, pmc) {
2284 		if (pmc->multi.imr_multiaddr.s_addr == addr &&
2285 		    pmc->multi.imr_ifindex == gsf->gf_interface)
2286 			break;
2287 	}
2288 	if (!pmc)		/* must have a prior join */
2289 		goto done;
2290 	gsf->gf_fmode = pmc->sfmode;
2291 	psl = rtnl_dereference(pmc->sflist);
2292 	rtnl_unlock();
2293 	count = psl ? psl->sl_count : 0;
2294 	copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2295 	gsf->gf_numsrc = count;
2296 	if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2297 	    copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2298 		return -EFAULT;
2299 	}
2300 	for (i = 0; i < copycount; i++) {
2301 		struct sockaddr_storage ss;
2302 
2303 		psin = (struct sockaddr_in *)&ss;
2304 		memset(&ss, 0, sizeof(ss));
2305 		psin->sin_family = AF_INET;
2306 		psin->sin_addr.s_addr = psl->sl_addr[i];
2307 		if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2308 			return -EFAULT;
2309 	}
2310 	return 0;
2311 done:
2312 	rtnl_unlock();
2313 	return err;
2314 }
2315 
2316 /*
2317  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2318  */
2319 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr, int dif)
2320 {
2321 	struct inet_sock *inet = inet_sk(sk);
2322 	struct ip_mc_socklist *pmc;
2323 	struct ip_sf_socklist *psl;
2324 	int i;
2325 	int ret;
2326 
2327 	ret = 1;
2328 	if (!ipv4_is_multicast(loc_addr))
2329 		goto out;
2330 
2331 	rcu_read_lock();
2332 	for_each_pmc_rcu(inet, pmc) {
2333 		if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2334 		    pmc->multi.imr_ifindex == dif)
2335 			break;
2336 	}
2337 	ret = inet->mc_all;
2338 	if (!pmc)
2339 		goto unlock;
2340 	psl = rcu_dereference(pmc->sflist);
2341 	ret = (pmc->sfmode == MCAST_EXCLUDE);
2342 	if (!psl)
2343 		goto unlock;
2344 
2345 	for (i = 0; i < psl->sl_count; i++) {
2346 		if (psl->sl_addr[i] == rmt_addr)
2347 			break;
2348 	}
2349 	ret = 0;
2350 	if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2351 		goto unlock;
2352 	if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2353 		goto unlock;
2354 	ret = 1;
2355 unlock:
2356 	rcu_read_unlock();
2357 out:
2358 	return ret;
2359 }
2360 
2361 /*
2362  *	A socket is closing.
2363  */
2364 
2365 void ip_mc_drop_socket(struct sock *sk)
2366 {
2367 	struct inet_sock *inet = inet_sk(sk);
2368 	struct ip_mc_socklist *iml;
2369 	struct net *net = sock_net(sk);
2370 
2371 	if (inet->mc_list == NULL)
2372 		return;
2373 
2374 	rtnl_lock();
2375 	while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2376 		struct in_device *in_dev;
2377 
2378 		inet->mc_list = iml->next_rcu;
2379 		in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2380 		(void) ip_mc_leave_src(sk, iml, in_dev);
2381 		if (in_dev != NULL)
2382 			ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2383 		/* decrease mem now to avoid the memleak warning */
2384 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2385 		kfree_rcu(iml, rcu);
2386 	}
2387 	rtnl_unlock();
2388 }
2389 
2390 /* called with rcu_read_lock() */
2391 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u16 proto)
2392 {
2393 	struct ip_mc_list *im;
2394 	struct ip_mc_list __rcu **mc_hash;
2395 	struct ip_sf_list *psf;
2396 	int rv = 0;
2397 
2398 	mc_hash = rcu_dereference(in_dev->mc_hash);
2399 	if (mc_hash) {
2400 		u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2401 
2402 		for (im = rcu_dereference(mc_hash[hash]);
2403 		     im != NULL;
2404 		     im = rcu_dereference(im->next_hash)) {
2405 			if (im->multiaddr == mc_addr)
2406 				break;
2407 		}
2408 	} else {
2409 		for_each_pmc_rcu(in_dev, im) {
2410 			if (im->multiaddr == mc_addr)
2411 				break;
2412 		}
2413 	}
2414 	if (im && proto == IPPROTO_IGMP) {
2415 		rv = 1;
2416 	} else if (im) {
2417 		if (src_addr) {
2418 			for (psf = im->sources; psf; psf = psf->sf_next) {
2419 				if (psf->sf_inaddr == src_addr)
2420 					break;
2421 			}
2422 			if (psf)
2423 				rv = psf->sf_count[MCAST_INCLUDE] ||
2424 					psf->sf_count[MCAST_EXCLUDE] !=
2425 					im->sfcount[MCAST_EXCLUDE];
2426 			else
2427 				rv = im->sfcount[MCAST_EXCLUDE] != 0;
2428 		} else
2429 			rv = 1; /* unspecified source; tentatively allow */
2430 	}
2431 	return rv;
2432 }
2433 
2434 #if defined(CONFIG_PROC_FS)
2435 struct igmp_mc_iter_state {
2436 	struct seq_net_private p;
2437 	struct net_device *dev;
2438 	struct in_device *in_dev;
2439 };
2440 
2441 #define	igmp_mc_seq_private(seq)	((struct igmp_mc_iter_state *)(seq)->private)
2442 
2443 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2444 {
2445 	struct net *net = seq_file_net(seq);
2446 	struct ip_mc_list *im = NULL;
2447 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2448 
2449 	state->in_dev = NULL;
2450 	for_each_netdev_rcu(net, state->dev) {
2451 		struct in_device *in_dev;
2452 
2453 		in_dev = __in_dev_get_rcu(state->dev);
2454 		if (!in_dev)
2455 			continue;
2456 		im = rcu_dereference(in_dev->mc_list);
2457 		if (im) {
2458 			state->in_dev = in_dev;
2459 			break;
2460 		}
2461 	}
2462 	return im;
2463 }
2464 
2465 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2466 {
2467 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2468 
2469 	im = rcu_dereference(im->next_rcu);
2470 	while (!im) {
2471 		state->dev = next_net_device_rcu(state->dev);
2472 		if (!state->dev) {
2473 			state->in_dev = NULL;
2474 			break;
2475 		}
2476 		state->in_dev = __in_dev_get_rcu(state->dev);
2477 		if (!state->in_dev)
2478 			continue;
2479 		im = rcu_dereference(state->in_dev->mc_list);
2480 	}
2481 	return im;
2482 }
2483 
2484 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2485 {
2486 	struct ip_mc_list *im = igmp_mc_get_first(seq);
2487 	if (im)
2488 		while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2489 			--pos;
2490 	return pos ? NULL : im;
2491 }
2492 
2493 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2494 	__acquires(rcu)
2495 {
2496 	rcu_read_lock();
2497 	return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2498 }
2499 
2500 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2501 {
2502 	struct ip_mc_list *im;
2503 	if (v == SEQ_START_TOKEN)
2504 		im = igmp_mc_get_first(seq);
2505 	else
2506 		im = igmp_mc_get_next(seq, v);
2507 	++*pos;
2508 	return im;
2509 }
2510 
2511 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2512 	__releases(rcu)
2513 {
2514 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2515 
2516 	state->in_dev = NULL;
2517 	state->dev = NULL;
2518 	rcu_read_unlock();
2519 }
2520 
2521 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2522 {
2523 	if (v == SEQ_START_TOKEN)
2524 		seq_puts(seq,
2525 			 "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2526 	else {
2527 		struct ip_mc_list *im = (struct ip_mc_list *)v;
2528 		struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2529 		char   *querier;
2530 		long delta;
2531 
2532 #ifdef CONFIG_IP_MULTICAST
2533 		querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2534 			  IGMP_V2_SEEN(state->in_dev) ? "V2" :
2535 			  "V3";
2536 #else
2537 		querier = "NONE";
2538 #endif
2539 
2540 		if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2541 			seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2542 				   state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2543 		}
2544 
2545 		delta = im->timer.expires - jiffies;
2546 		seq_printf(seq,
2547 			   "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2548 			   im->multiaddr, im->users,
2549 			   im->tm_running,
2550 			   im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2551 			   im->reporter);
2552 	}
2553 	return 0;
2554 }
2555 
2556 static const struct seq_operations igmp_mc_seq_ops = {
2557 	.start	=	igmp_mc_seq_start,
2558 	.next	=	igmp_mc_seq_next,
2559 	.stop	=	igmp_mc_seq_stop,
2560 	.show	=	igmp_mc_seq_show,
2561 };
2562 
2563 static int igmp_mc_seq_open(struct inode *inode, struct file *file)
2564 {
2565 	return seq_open_net(inode, file, &igmp_mc_seq_ops,
2566 			sizeof(struct igmp_mc_iter_state));
2567 }
2568 
2569 static const struct file_operations igmp_mc_seq_fops = {
2570 	.owner		=	THIS_MODULE,
2571 	.open		=	igmp_mc_seq_open,
2572 	.read		=	seq_read,
2573 	.llseek		=	seq_lseek,
2574 	.release	=	seq_release_net,
2575 };
2576 
2577 struct igmp_mcf_iter_state {
2578 	struct seq_net_private p;
2579 	struct net_device *dev;
2580 	struct in_device *idev;
2581 	struct ip_mc_list *im;
2582 };
2583 
2584 #define igmp_mcf_seq_private(seq)	((struct igmp_mcf_iter_state *)(seq)->private)
2585 
2586 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2587 {
2588 	struct net *net = seq_file_net(seq);
2589 	struct ip_sf_list *psf = NULL;
2590 	struct ip_mc_list *im = NULL;
2591 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2592 
2593 	state->idev = NULL;
2594 	state->im = NULL;
2595 	for_each_netdev_rcu(net, state->dev) {
2596 		struct in_device *idev;
2597 		idev = __in_dev_get_rcu(state->dev);
2598 		if (unlikely(idev == NULL))
2599 			continue;
2600 		im = rcu_dereference(idev->mc_list);
2601 		if (likely(im != NULL)) {
2602 			spin_lock_bh(&im->lock);
2603 			psf = im->sources;
2604 			if (likely(psf != NULL)) {
2605 				state->im = im;
2606 				state->idev = idev;
2607 				break;
2608 			}
2609 			spin_unlock_bh(&im->lock);
2610 		}
2611 	}
2612 	return psf;
2613 }
2614 
2615 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2616 {
2617 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2618 
2619 	psf = psf->sf_next;
2620 	while (!psf) {
2621 		spin_unlock_bh(&state->im->lock);
2622 		state->im = state->im->next;
2623 		while (!state->im) {
2624 			state->dev = next_net_device_rcu(state->dev);
2625 			if (!state->dev) {
2626 				state->idev = NULL;
2627 				goto out;
2628 			}
2629 			state->idev = __in_dev_get_rcu(state->dev);
2630 			if (!state->idev)
2631 				continue;
2632 			state->im = rcu_dereference(state->idev->mc_list);
2633 		}
2634 		if (!state->im)
2635 			break;
2636 		spin_lock_bh(&state->im->lock);
2637 		psf = state->im->sources;
2638 	}
2639 out:
2640 	return psf;
2641 }
2642 
2643 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2644 {
2645 	struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2646 	if (psf)
2647 		while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2648 			--pos;
2649 	return pos ? NULL : psf;
2650 }
2651 
2652 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2653 	__acquires(rcu)
2654 {
2655 	rcu_read_lock();
2656 	return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2657 }
2658 
2659 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2660 {
2661 	struct ip_sf_list *psf;
2662 	if (v == SEQ_START_TOKEN)
2663 		psf = igmp_mcf_get_first(seq);
2664 	else
2665 		psf = igmp_mcf_get_next(seq, v);
2666 	++*pos;
2667 	return psf;
2668 }
2669 
2670 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2671 	__releases(rcu)
2672 {
2673 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2674 	if (likely(state->im != NULL)) {
2675 		spin_unlock_bh(&state->im->lock);
2676 		state->im = NULL;
2677 	}
2678 	state->idev = NULL;
2679 	state->dev = NULL;
2680 	rcu_read_unlock();
2681 }
2682 
2683 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2684 {
2685 	struct ip_sf_list *psf = (struct ip_sf_list *)v;
2686 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2687 
2688 	if (v == SEQ_START_TOKEN) {
2689 		seq_printf(seq,
2690 			   "%3s %6s "
2691 			   "%10s %10s %6s %6s\n", "Idx",
2692 			   "Device", "MCA",
2693 			   "SRC", "INC", "EXC");
2694 	} else {
2695 		seq_printf(seq,
2696 			   "%3d %6.6s 0x%08x "
2697 			   "0x%08x %6lu %6lu\n",
2698 			   state->dev->ifindex, state->dev->name,
2699 			   ntohl(state->im->multiaddr),
2700 			   ntohl(psf->sf_inaddr),
2701 			   psf->sf_count[MCAST_INCLUDE],
2702 			   psf->sf_count[MCAST_EXCLUDE]);
2703 	}
2704 	return 0;
2705 }
2706 
2707 static const struct seq_operations igmp_mcf_seq_ops = {
2708 	.start	=	igmp_mcf_seq_start,
2709 	.next	=	igmp_mcf_seq_next,
2710 	.stop	=	igmp_mcf_seq_stop,
2711 	.show	=	igmp_mcf_seq_show,
2712 };
2713 
2714 static int igmp_mcf_seq_open(struct inode *inode, struct file *file)
2715 {
2716 	return seq_open_net(inode, file, &igmp_mcf_seq_ops,
2717 			sizeof(struct igmp_mcf_iter_state));
2718 }
2719 
2720 static const struct file_operations igmp_mcf_seq_fops = {
2721 	.owner		=	THIS_MODULE,
2722 	.open		=	igmp_mcf_seq_open,
2723 	.read		=	seq_read,
2724 	.llseek		=	seq_lseek,
2725 	.release	=	seq_release_net,
2726 };
2727 
2728 static int __net_init igmp_net_init(struct net *net)
2729 {
2730 	struct proc_dir_entry *pde;
2731 
2732 	pde = proc_create("igmp", S_IRUGO, net->proc_net, &igmp_mc_seq_fops);
2733 	if (!pde)
2734 		goto out_igmp;
2735 	pde = proc_create("mcfilter", S_IRUGO, net->proc_net,
2736 			  &igmp_mcf_seq_fops);
2737 	if (!pde)
2738 		goto out_mcfilter;
2739 	return 0;
2740 
2741 out_mcfilter:
2742 	remove_proc_entry("igmp", net->proc_net);
2743 out_igmp:
2744 	return -ENOMEM;
2745 }
2746 
2747 static void __net_exit igmp_net_exit(struct net *net)
2748 {
2749 	remove_proc_entry("mcfilter", net->proc_net);
2750 	remove_proc_entry("igmp", net->proc_net);
2751 }
2752 
2753 static struct pernet_operations igmp_net_ops = {
2754 	.init = igmp_net_init,
2755 	.exit = igmp_net_exit,
2756 };
2757 #endif
2758 
2759 static int igmp_netdev_event(struct notifier_block *this,
2760 			     unsigned long event, void *ptr)
2761 {
2762 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2763 	struct in_device *in_dev;
2764 
2765 	switch (event) {
2766 	case NETDEV_RESEND_IGMP:
2767 		in_dev = __in_dev_get_rtnl(dev);
2768 		if (in_dev)
2769 			ip_mc_rejoin_groups(in_dev);
2770 		break;
2771 	default:
2772 		break;
2773 	}
2774 	return NOTIFY_DONE;
2775 }
2776 
2777 static struct notifier_block igmp_notifier = {
2778 	.notifier_call = igmp_netdev_event,
2779 };
2780 
2781 int __init igmp_mc_init(void)
2782 {
2783 #if defined(CONFIG_PROC_FS)
2784 	int err;
2785 
2786 	err = register_pernet_subsys(&igmp_net_ops);
2787 	if (err)
2788 		return err;
2789 	err = register_netdevice_notifier(&igmp_notifier);
2790 	if (err)
2791 		goto reg_notif_fail;
2792 	return 0;
2793 
2794 reg_notif_fail:
2795 	unregister_pernet_subsys(&igmp_net_ops);
2796 	return err;
2797 #else
2798 	return register_netdevice_notifier(&igmp_notifier);
2799 #endif
2800 }
2801