xref: /openbmc/linux/net/ipv4/igmp.c (revision 588b48ca)
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_Unsolicited_Report_Count		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 #define igmp_skb_size(skb) (*(unsigned int *)((skb)->cb))
322 
323 static struct sk_buff *igmpv3_newpack(struct net_device *dev, int size)
324 {
325 	struct sk_buff *skb;
326 	struct rtable *rt;
327 	struct iphdr *pip;
328 	struct igmpv3_report *pig;
329 	struct net *net = dev_net(dev);
330 	struct flowi4 fl4;
331 	int hlen = LL_RESERVED_SPACE(dev);
332 	int tlen = dev->needed_tailroom;
333 
334 	while (1) {
335 		skb = alloc_skb(size + hlen + tlen,
336 				GFP_ATOMIC | __GFP_NOWARN);
337 		if (skb)
338 			break;
339 		size >>= 1;
340 		if (size < 256)
341 			return NULL;
342 	}
343 	skb->priority = TC_PRIO_CONTROL;
344 	igmp_skb_size(skb) = size;
345 
346 	rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
347 				   0, 0,
348 				   IPPROTO_IGMP, 0, dev->ifindex);
349 	if (IS_ERR(rt)) {
350 		kfree_skb(skb);
351 		return NULL;
352 	}
353 
354 	skb_dst_set(skb, &rt->dst);
355 	skb->dev = dev;
356 
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)->dev ? igmp_skb_size(skb) - (skb)->len : \
427 	skb_tailroom(skb)) : 0)
428 
429 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
430 	int type, int gdeleted, int sdeleted)
431 {
432 	struct net_device *dev = pmc->interface->dev;
433 	struct igmpv3_report *pih;
434 	struct igmpv3_grec *pgr = NULL;
435 	struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
436 	int scount, stotal, first, isquery, truncate;
437 
438 	if (pmc->multiaddr == IGMP_ALL_HOSTS)
439 		return skb;
440 
441 	isquery = type == IGMPV3_MODE_IS_INCLUDE ||
442 		  type == IGMPV3_MODE_IS_EXCLUDE;
443 	truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
444 		    type == IGMPV3_CHANGE_TO_EXCLUDE;
445 
446 	stotal = scount = 0;
447 
448 	psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
449 
450 	if (!*psf_list)
451 		goto empty_source;
452 
453 	pih = skb ? igmpv3_report_hdr(skb) : NULL;
454 
455 	/* EX and TO_EX get a fresh packet, if needed */
456 	if (truncate) {
457 		if (pih && pih->ngrec &&
458 		    AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
459 			if (skb)
460 				igmpv3_sendpack(skb);
461 			skb = igmpv3_newpack(dev, dev->mtu);
462 		}
463 	}
464 	first = 1;
465 	psf_prev = NULL;
466 	for (psf = *psf_list; psf; psf = psf_next) {
467 		__be32 *psrc;
468 
469 		psf_next = psf->sf_next;
470 
471 		if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
472 			psf_prev = psf;
473 			continue;
474 		}
475 
476 		/* clear marks on query responses */
477 		if (isquery)
478 			psf->sf_gsresp = 0;
479 
480 		if (AVAILABLE(skb) < sizeof(__be32) +
481 		    first*sizeof(struct igmpv3_grec)) {
482 			if (truncate && !first)
483 				break;	 /* truncate these */
484 			if (pgr)
485 				pgr->grec_nsrcs = htons(scount);
486 			if (skb)
487 				igmpv3_sendpack(skb);
488 			skb = igmpv3_newpack(dev, dev->mtu);
489 			first = 1;
490 			scount = 0;
491 		}
492 		if (first) {
493 			skb = add_grhead(skb, pmc, type, &pgr);
494 			first = 0;
495 		}
496 		if (!skb)
497 			return NULL;
498 		psrc = (__be32 *)skb_put(skb, sizeof(__be32));
499 		*psrc = psf->sf_inaddr;
500 		scount++; stotal++;
501 		if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
502 		     type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
503 			psf->sf_crcount--;
504 			if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
505 				if (psf_prev)
506 					psf_prev->sf_next = psf->sf_next;
507 				else
508 					*psf_list = psf->sf_next;
509 				kfree(psf);
510 				continue;
511 			}
512 		}
513 		psf_prev = psf;
514 	}
515 
516 empty_source:
517 	if (!stotal) {
518 		if (type == IGMPV3_ALLOW_NEW_SOURCES ||
519 		    type == IGMPV3_BLOCK_OLD_SOURCES)
520 			return skb;
521 		if (pmc->crcount || isquery) {
522 			/* make sure we have room for group header */
523 			if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
524 				igmpv3_sendpack(skb);
525 				skb = NULL; /* add_grhead will get a new one */
526 			}
527 			skb = add_grhead(skb, pmc, type, &pgr);
528 		}
529 	}
530 	if (pgr)
531 		pgr->grec_nsrcs = htons(scount);
532 
533 	if (isquery)
534 		pmc->gsquery = 0;	/* clear query state on report */
535 	return skb;
536 }
537 
538 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
539 {
540 	struct sk_buff *skb = NULL;
541 	int type;
542 
543 	if (!pmc) {
544 		rcu_read_lock();
545 		for_each_pmc_rcu(in_dev, pmc) {
546 			if (pmc->multiaddr == IGMP_ALL_HOSTS)
547 				continue;
548 			spin_lock_bh(&pmc->lock);
549 			if (pmc->sfcount[MCAST_EXCLUDE])
550 				type = IGMPV3_MODE_IS_EXCLUDE;
551 			else
552 				type = IGMPV3_MODE_IS_INCLUDE;
553 			skb = add_grec(skb, pmc, type, 0, 0);
554 			spin_unlock_bh(&pmc->lock);
555 		}
556 		rcu_read_unlock();
557 	} else {
558 		spin_lock_bh(&pmc->lock);
559 		if (pmc->sfcount[MCAST_EXCLUDE])
560 			type = IGMPV3_MODE_IS_EXCLUDE;
561 		else
562 			type = IGMPV3_MODE_IS_INCLUDE;
563 		skb = add_grec(skb, pmc, type, 0, 0);
564 		spin_unlock_bh(&pmc->lock);
565 	}
566 	if (!skb)
567 		return 0;
568 	return igmpv3_sendpack(skb);
569 }
570 
571 /*
572  * remove zero-count source records from a source filter list
573  */
574 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
575 {
576 	struct ip_sf_list *psf_prev, *psf_next, *psf;
577 
578 	psf_prev = NULL;
579 	for (psf = *ppsf; psf; psf = psf_next) {
580 		psf_next = psf->sf_next;
581 		if (psf->sf_crcount == 0) {
582 			if (psf_prev)
583 				psf_prev->sf_next = psf->sf_next;
584 			else
585 				*ppsf = psf->sf_next;
586 			kfree(psf);
587 		} else
588 			psf_prev = psf;
589 	}
590 }
591 
592 static void igmpv3_send_cr(struct in_device *in_dev)
593 {
594 	struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
595 	struct sk_buff *skb = NULL;
596 	int type, dtype;
597 
598 	rcu_read_lock();
599 	spin_lock_bh(&in_dev->mc_tomb_lock);
600 
601 	/* deleted MCA's */
602 	pmc_prev = NULL;
603 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
604 		pmc_next = pmc->next;
605 		if (pmc->sfmode == MCAST_INCLUDE) {
606 			type = IGMPV3_BLOCK_OLD_SOURCES;
607 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
608 			skb = add_grec(skb, pmc, type, 1, 0);
609 			skb = add_grec(skb, pmc, dtype, 1, 1);
610 		}
611 		if (pmc->crcount) {
612 			if (pmc->sfmode == MCAST_EXCLUDE) {
613 				type = IGMPV3_CHANGE_TO_INCLUDE;
614 				skb = add_grec(skb, pmc, type, 1, 0);
615 			}
616 			pmc->crcount--;
617 			if (pmc->crcount == 0) {
618 				igmpv3_clear_zeros(&pmc->tomb);
619 				igmpv3_clear_zeros(&pmc->sources);
620 			}
621 		}
622 		if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
623 			if (pmc_prev)
624 				pmc_prev->next = pmc_next;
625 			else
626 				in_dev->mc_tomb = pmc_next;
627 			in_dev_put(pmc->interface);
628 			kfree(pmc);
629 		} else
630 			pmc_prev = pmc;
631 	}
632 	spin_unlock_bh(&in_dev->mc_tomb_lock);
633 
634 	/* change recs */
635 	for_each_pmc_rcu(in_dev, pmc) {
636 		spin_lock_bh(&pmc->lock);
637 		if (pmc->sfcount[MCAST_EXCLUDE]) {
638 			type = IGMPV3_BLOCK_OLD_SOURCES;
639 			dtype = IGMPV3_ALLOW_NEW_SOURCES;
640 		} else {
641 			type = IGMPV3_ALLOW_NEW_SOURCES;
642 			dtype = IGMPV3_BLOCK_OLD_SOURCES;
643 		}
644 		skb = add_grec(skb, pmc, type, 0, 0);
645 		skb = add_grec(skb, pmc, dtype, 0, 1);	/* deleted sources */
646 
647 		/* filter mode changes */
648 		if (pmc->crcount) {
649 			if (pmc->sfmode == MCAST_EXCLUDE)
650 				type = IGMPV3_CHANGE_TO_EXCLUDE;
651 			else
652 				type = IGMPV3_CHANGE_TO_INCLUDE;
653 			skb = add_grec(skb, pmc, type, 0, 0);
654 			pmc->crcount--;
655 		}
656 		spin_unlock_bh(&pmc->lock);
657 	}
658 	rcu_read_unlock();
659 
660 	if (!skb)
661 		return;
662 	(void) igmpv3_sendpack(skb);
663 }
664 
665 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
666 	int type)
667 {
668 	struct sk_buff *skb;
669 	struct iphdr *iph;
670 	struct igmphdr *ih;
671 	struct rtable *rt;
672 	struct net_device *dev = in_dev->dev;
673 	struct net *net = dev_net(dev);
674 	__be32	group = pmc ? pmc->multiaddr : 0;
675 	struct flowi4 fl4;
676 	__be32	dst;
677 	int hlen, tlen;
678 
679 	if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
680 		return igmpv3_send_report(in_dev, pmc);
681 	else if (type == IGMP_HOST_LEAVE_MESSAGE)
682 		dst = IGMP_ALL_ROUTER;
683 	else
684 		dst = group;
685 
686 	rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
687 				   0, 0,
688 				   IPPROTO_IGMP, 0, dev->ifindex);
689 	if (IS_ERR(rt))
690 		return -1;
691 
692 	hlen = LL_RESERVED_SPACE(dev);
693 	tlen = dev->needed_tailroom;
694 	skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
695 	if (skb == NULL) {
696 		ip_rt_put(rt);
697 		return -1;
698 	}
699 	skb->priority = TC_PRIO_CONTROL;
700 
701 	skb_dst_set(skb, &rt->dst);
702 
703 	skb_reserve(skb, hlen);
704 
705 	skb_reset_network_header(skb);
706 	iph = ip_hdr(skb);
707 	skb_put(skb, sizeof(struct iphdr) + 4);
708 
709 	iph->version  = 4;
710 	iph->ihl      = (sizeof(struct iphdr)+4)>>2;
711 	iph->tos      = 0xc0;
712 	iph->frag_off = htons(IP_DF);
713 	iph->ttl      = 1;
714 	iph->daddr    = dst;
715 	iph->saddr    = fl4.saddr;
716 	iph->protocol = IPPROTO_IGMP;
717 	ip_select_ident(skb, NULL);
718 	((u8 *)&iph[1])[0] = IPOPT_RA;
719 	((u8 *)&iph[1])[1] = 4;
720 	((u8 *)&iph[1])[2] = 0;
721 	((u8 *)&iph[1])[3] = 0;
722 
723 	ih = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
724 	ih->type = type;
725 	ih->code = 0;
726 	ih->csum = 0;
727 	ih->group = group;
728 	ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
729 
730 	return ip_local_out(skb);
731 }
732 
733 static void igmp_gq_timer_expire(unsigned long data)
734 {
735 	struct in_device *in_dev = (struct in_device *)data;
736 
737 	in_dev->mr_gq_running = 0;
738 	igmpv3_send_report(in_dev, NULL);
739 	in_dev_put(in_dev);
740 }
741 
742 static void igmp_ifc_timer_expire(unsigned long data)
743 {
744 	struct in_device *in_dev = (struct in_device *)data;
745 
746 	igmpv3_send_cr(in_dev);
747 	if (in_dev->mr_ifc_count) {
748 		in_dev->mr_ifc_count--;
749 		igmp_ifc_start_timer(in_dev,
750 				     unsolicited_report_interval(in_dev));
751 	}
752 	in_dev_put(in_dev);
753 }
754 
755 static void igmp_ifc_event(struct in_device *in_dev)
756 {
757 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
758 		return;
759 	in_dev->mr_ifc_count = in_dev->mr_qrv ? in_dev->mr_qrv :
760 		IGMP_Unsolicited_Report_Count;
761 	igmp_ifc_start_timer(in_dev, 1);
762 }
763 
764 
765 static void igmp_timer_expire(unsigned long data)
766 {
767 	struct ip_mc_list *im = (struct ip_mc_list *)data;
768 	struct in_device *in_dev = im->interface;
769 
770 	spin_lock(&im->lock);
771 	im->tm_running = 0;
772 
773 	if (im->unsolicit_count) {
774 		im->unsolicit_count--;
775 		igmp_start_timer(im, unsolicited_report_interval(in_dev));
776 	}
777 	im->reporter = 1;
778 	spin_unlock(&im->lock);
779 
780 	if (IGMP_V1_SEEN(in_dev))
781 		igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
782 	else if (IGMP_V2_SEEN(in_dev))
783 		igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
784 	else
785 		igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
786 
787 	ip_ma_put(im);
788 }
789 
790 /* mark EXCLUDE-mode sources */
791 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
792 {
793 	struct ip_sf_list *psf;
794 	int i, scount;
795 
796 	scount = 0;
797 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
798 		if (scount == nsrcs)
799 			break;
800 		for (i = 0; i < nsrcs; i++) {
801 			/* skip inactive filters */
802 			if (psf->sf_count[MCAST_INCLUDE] ||
803 			    pmc->sfcount[MCAST_EXCLUDE] !=
804 			    psf->sf_count[MCAST_EXCLUDE])
805 				break;
806 			if (srcs[i] == psf->sf_inaddr) {
807 				scount++;
808 				break;
809 			}
810 		}
811 	}
812 	pmc->gsquery = 0;
813 	if (scount == nsrcs)	/* all sources excluded */
814 		return 0;
815 	return 1;
816 }
817 
818 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
819 {
820 	struct ip_sf_list *psf;
821 	int i, scount;
822 
823 	if (pmc->sfmode == MCAST_EXCLUDE)
824 		return igmp_xmarksources(pmc, nsrcs, srcs);
825 
826 	/* mark INCLUDE-mode sources */
827 	scount = 0;
828 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
829 		if (scount == nsrcs)
830 			break;
831 		for (i = 0; i < nsrcs; i++)
832 			if (srcs[i] == psf->sf_inaddr) {
833 				psf->sf_gsresp = 1;
834 				scount++;
835 				break;
836 			}
837 	}
838 	if (!scount) {
839 		pmc->gsquery = 0;
840 		return 0;
841 	}
842 	pmc->gsquery = 1;
843 	return 1;
844 }
845 
846 /* return true if packet was dropped */
847 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
848 {
849 	struct ip_mc_list *im;
850 
851 	/* Timers are only set for non-local groups */
852 
853 	if (group == IGMP_ALL_HOSTS)
854 		return false;
855 
856 	rcu_read_lock();
857 	for_each_pmc_rcu(in_dev, im) {
858 		if (im->multiaddr == group) {
859 			igmp_stop_timer(im);
860 			break;
861 		}
862 	}
863 	rcu_read_unlock();
864 	return false;
865 }
866 
867 /* return true if packet was dropped */
868 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
869 	int len)
870 {
871 	struct igmphdr 		*ih = igmp_hdr(skb);
872 	struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
873 	struct ip_mc_list	*im;
874 	__be32			group = ih->group;
875 	int			max_delay;
876 	int			mark = 0;
877 
878 
879 	if (len == 8) {
880 		if (ih->code == 0) {
881 			/* Alas, old v1 router presents here. */
882 
883 			max_delay = IGMP_Query_Response_Interval;
884 			in_dev->mr_v1_seen = jiffies +
885 				IGMP_V1_Router_Present_Timeout;
886 			group = 0;
887 		} else {
888 			/* v2 router present */
889 			max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
890 			in_dev->mr_v2_seen = jiffies +
891 				IGMP_V2_Router_Present_Timeout;
892 		}
893 		/* cancel the interface change timer */
894 		in_dev->mr_ifc_count = 0;
895 		if (del_timer(&in_dev->mr_ifc_timer))
896 			__in_dev_put(in_dev);
897 		/* clear deleted report items */
898 		igmpv3_clear_delrec(in_dev);
899 	} else if (len < 12) {
900 		return true;	/* ignore bogus packet; freed by caller */
901 	} else if (IGMP_V1_SEEN(in_dev)) {
902 		/* This is a v3 query with v1 queriers present */
903 		max_delay = IGMP_Query_Response_Interval;
904 		group = 0;
905 	} else if (IGMP_V2_SEEN(in_dev)) {
906 		/* this is a v3 query with v2 queriers present;
907 		 * Interpretation of the max_delay code is problematic here.
908 		 * A real v2 host would use ih_code directly, while v3 has a
909 		 * different encoding. We use the v3 encoding as more likely
910 		 * to be intended in a v3 query.
911 		 */
912 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
913 		if (!max_delay)
914 			max_delay = 1;	/* can't mod w/ 0 */
915 	} else { /* v3 */
916 		if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
917 			return true;
918 
919 		ih3 = igmpv3_query_hdr(skb);
920 		if (ih3->nsrcs) {
921 			if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
922 					   + ntohs(ih3->nsrcs)*sizeof(__be32)))
923 				return true;
924 			ih3 = igmpv3_query_hdr(skb);
925 		}
926 
927 		max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
928 		if (!max_delay)
929 			max_delay = 1;	/* can't mod w/ 0 */
930 		in_dev->mr_maxdelay = max_delay;
931 		if (ih3->qrv)
932 			in_dev->mr_qrv = ih3->qrv;
933 		if (!group) { /* general query */
934 			if (ih3->nsrcs)
935 				return false;	/* no sources allowed */
936 			igmp_gq_start_timer(in_dev);
937 			return false;
938 		}
939 		/* mark sources to include, if group & source-specific */
940 		mark = ih3->nsrcs != 0;
941 	}
942 
943 	/*
944 	 * - Start the timers in all of our membership records
945 	 *   that the query applies to for the interface on
946 	 *   which the query arrived excl. those that belong
947 	 *   to a "local" group (224.0.0.X)
948 	 * - For timers already running check if they need to
949 	 *   be reset.
950 	 * - Use the igmp->igmp_code field as the maximum
951 	 *   delay possible
952 	 */
953 	rcu_read_lock();
954 	for_each_pmc_rcu(in_dev, im) {
955 		int changed;
956 
957 		if (group && group != im->multiaddr)
958 			continue;
959 		if (im->multiaddr == IGMP_ALL_HOSTS)
960 			continue;
961 		spin_lock_bh(&im->lock);
962 		if (im->tm_running)
963 			im->gsquery = im->gsquery && mark;
964 		else
965 			im->gsquery = mark;
966 		changed = !im->gsquery ||
967 			igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
968 		spin_unlock_bh(&im->lock);
969 		if (changed)
970 			igmp_mod_timer(im, max_delay);
971 	}
972 	rcu_read_unlock();
973 	return false;
974 }
975 
976 /* called in rcu_read_lock() section */
977 int igmp_rcv(struct sk_buff *skb)
978 {
979 	/* This basically follows the spec line by line -- see RFC1112 */
980 	struct igmphdr *ih;
981 	struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
982 	int len = skb->len;
983 	bool dropped = true;
984 
985 	if (in_dev == NULL)
986 		goto drop;
987 
988 	if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
989 		goto drop;
990 
991 	if (skb_checksum_simple_validate(skb))
992 		goto drop;
993 
994 	ih = igmp_hdr(skb);
995 	switch (ih->type) {
996 	case IGMP_HOST_MEMBERSHIP_QUERY:
997 		dropped = igmp_heard_query(in_dev, skb, len);
998 		break;
999 	case IGMP_HOST_MEMBERSHIP_REPORT:
1000 	case IGMPV2_HOST_MEMBERSHIP_REPORT:
1001 		/* Is it our report looped back? */
1002 		if (rt_is_output_route(skb_rtable(skb)))
1003 			break;
1004 		/* don't rely on MC router hearing unicast reports */
1005 		if (skb->pkt_type == PACKET_MULTICAST ||
1006 		    skb->pkt_type == PACKET_BROADCAST)
1007 			dropped = igmp_heard_report(in_dev, ih->group);
1008 		break;
1009 	case IGMP_PIM:
1010 #ifdef CONFIG_IP_PIMSM_V1
1011 		return pim_rcv_v1(skb);
1012 #endif
1013 	case IGMPV3_HOST_MEMBERSHIP_REPORT:
1014 	case IGMP_DVMRP:
1015 	case IGMP_TRACE:
1016 	case IGMP_HOST_LEAVE_MESSAGE:
1017 	case IGMP_MTRACE:
1018 	case IGMP_MTRACE_RESP:
1019 		break;
1020 	default:
1021 		break;
1022 	}
1023 
1024 drop:
1025 	if (dropped)
1026 		kfree_skb(skb);
1027 	else
1028 		consume_skb(skb);
1029 	return 0;
1030 }
1031 
1032 #endif
1033 
1034 
1035 /*
1036  *	Add a filter to a device
1037  */
1038 
1039 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1040 {
1041 	char buf[MAX_ADDR_LEN];
1042 	struct net_device *dev = in_dev->dev;
1043 
1044 	/* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1045 	   We will get multicast token leakage, when IFF_MULTICAST
1046 	   is changed. This check should be done in ndo_set_rx_mode
1047 	   routine. Something sort of:
1048 	   if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1049 	   --ANK
1050 	   */
1051 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1052 		dev_mc_add(dev, buf);
1053 }
1054 
1055 /*
1056  *	Remove a filter from a device
1057  */
1058 
1059 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1060 {
1061 	char buf[MAX_ADDR_LEN];
1062 	struct net_device *dev = in_dev->dev;
1063 
1064 	if (arp_mc_map(addr, buf, dev, 0) == 0)
1065 		dev_mc_del(dev, buf);
1066 }
1067 
1068 #ifdef CONFIG_IP_MULTICAST
1069 /*
1070  * deleted ip_mc_list manipulation
1071  */
1072 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1073 {
1074 	struct ip_mc_list *pmc;
1075 
1076 	/* this is an "ip_mc_list" for convenience; only the fields below
1077 	 * are actually used. In particular, the refcnt and users are not
1078 	 * used for management of the delete list. Using the same structure
1079 	 * for deleted items allows change reports to use common code with
1080 	 * non-deleted or query-response MCA's.
1081 	 */
1082 	pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1083 	if (!pmc)
1084 		return;
1085 	spin_lock_bh(&im->lock);
1086 	pmc->interface = im->interface;
1087 	in_dev_hold(in_dev);
1088 	pmc->multiaddr = im->multiaddr;
1089 	pmc->crcount = in_dev->mr_qrv ? in_dev->mr_qrv :
1090 		IGMP_Unsolicited_Report_Count;
1091 	pmc->sfmode = im->sfmode;
1092 	if (pmc->sfmode == MCAST_INCLUDE) {
1093 		struct ip_sf_list *psf;
1094 
1095 		pmc->tomb = im->tomb;
1096 		pmc->sources = im->sources;
1097 		im->tomb = im->sources = NULL;
1098 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1099 			psf->sf_crcount = pmc->crcount;
1100 	}
1101 	spin_unlock_bh(&im->lock);
1102 
1103 	spin_lock_bh(&in_dev->mc_tomb_lock);
1104 	pmc->next = in_dev->mc_tomb;
1105 	in_dev->mc_tomb = pmc;
1106 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1107 }
1108 
1109 static void igmpv3_del_delrec(struct in_device *in_dev, __be32 multiaddr)
1110 {
1111 	struct ip_mc_list *pmc, *pmc_prev;
1112 	struct ip_sf_list *psf, *psf_next;
1113 
1114 	spin_lock_bh(&in_dev->mc_tomb_lock);
1115 	pmc_prev = NULL;
1116 	for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1117 		if (pmc->multiaddr == multiaddr)
1118 			break;
1119 		pmc_prev = pmc;
1120 	}
1121 	if (pmc) {
1122 		if (pmc_prev)
1123 			pmc_prev->next = pmc->next;
1124 		else
1125 			in_dev->mc_tomb = pmc->next;
1126 	}
1127 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1128 	if (pmc) {
1129 		for (psf = pmc->tomb; psf; psf = psf_next) {
1130 			psf_next = psf->sf_next;
1131 			kfree(psf);
1132 		}
1133 		in_dev_put(pmc->interface);
1134 		kfree(pmc);
1135 	}
1136 }
1137 
1138 static void igmpv3_clear_delrec(struct in_device *in_dev)
1139 {
1140 	struct ip_mc_list *pmc, *nextpmc;
1141 
1142 	spin_lock_bh(&in_dev->mc_tomb_lock);
1143 	pmc = in_dev->mc_tomb;
1144 	in_dev->mc_tomb = NULL;
1145 	spin_unlock_bh(&in_dev->mc_tomb_lock);
1146 
1147 	for (; pmc; pmc = nextpmc) {
1148 		nextpmc = pmc->next;
1149 		ip_mc_clear_src(pmc);
1150 		in_dev_put(pmc->interface);
1151 		kfree(pmc);
1152 	}
1153 	/* clear dead sources, too */
1154 	rcu_read_lock();
1155 	for_each_pmc_rcu(in_dev, pmc) {
1156 		struct ip_sf_list *psf, *psf_next;
1157 
1158 		spin_lock_bh(&pmc->lock);
1159 		psf = pmc->tomb;
1160 		pmc->tomb = NULL;
1161 		spin_unlock_bh(&pmc->lock);
1162 		for (; psf; psf = psf_next) {
1163 			psf_next = psf->sf_next;
1164 			kfree(psf);
1165 		}
1166 	}
1167 	rcu_read_unlock();
1168 }
1169 #endif
1170 
1171 static void igmp_group_dropped(struct ip_mc_list *im)
1172 {
1173 	struct in_device *in_dev = im->interface;
1174 #ifdef CONFIG_IP_MULTICAST
1175 	int reporter;
1176 #endif
1177 
1178 	if (im->loaded) {
1179 		im->loaded = 0;
1180 		ip_mc_filter_del(in_dev, im->multiaddr);
1181 	}
1182 
1183 #ifdef CONFIG_IP_MULTICAST
1184 	if (im->multiaddr == IGMP_ALL_HOSTS)
1185 		return;
1186 
1187 	reporter = im->reporter;
1188 	igmp_stop_timer(im);
1189 
1190 	if (!in_dev->dead) {
1191 		if (IGMP_V1_SEEN(in_dev))
1192 			return;
1193 		if (IGMP_V2_SEEN(in_dev)) {
1194 			if (reporter)
1195 				igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1196 			return;
1197 		}
1198 		/* IGMPv3 */
1199 		igmpv3_add_delrec(in_dev, im);
1200 
1201 		igmp_ifc_event(in_dev);
1202 	}
1203 #endif
1204 }
1205 
1206 static void igmp_group_added(struct ip_mc_list *im)
1207 {
1208 	struct in_device *in_dev = im->interface;
1209 
1210 	if (im->loaded == 0) {
1211 		im->loaded = 1;
1212 		ip_mc_filter_add(in_dev, im->multiaddr);
1213 	}
1214 
1215 #ifdef CONFIG_IP_MULTICAST
1216 	if (im->multiaddr == IGMP_ALL_HOSTS)
1217 		return;
1218 
1219 	if (in_dev->dead)
1220 		return;
1221 	if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1222 		spin_lock_bh(&im->lock);
1223 		igmp_start_timer(im, IGMP_Initial_Report_Delay);
1224 		spin_unlock_bh(&im->lock);
1225 		return;
1226 	}
1227 	/* else, v3 */
1228 
1229 	im->crcount = in_dev->mr_qrv ? in_dev->mr_qrv :
1230 		IGMP_Unsolicited_Report_Count;
1231 	igmp_ifc_event(in_dev);
1232 #endif
1233 }
1234 
1235 
1236 /*
1237  *	Multicast list managers
1238  */
1239 
1240 static u32 ip_mc_hash(const struct ip_mc_list *im)
1241 {
1242 	return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1243 }
1244 
1245 static void ip_mc_hash_add(struct in_device *in_dev,
1246 			   struct ip_mc_list *im)
1247 {
1248 	struct ip_mc_list __rcu **mc_hash;
1249 	u32 hash;
1250 
1251 	mc_hash = rtnl_dereference(in_dev->mc_hash);
1252 	if (mc_hash) {
1253 		hash = ip_mc_hash(im);
1254 		im->next_hash = mc_hash[hash];
1255 		rcu_assign_pointer(mc_hash[hash], im);
1256 		return;
1257 	}
1258 
1259 	/* do not use a hash table for small number of items */
1260 	if (in_dev->mc_count < 4)
1261 		return;
1262 
1263 	mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1264 			  GFP_KERNEL);
1265 	if (!mc_hash)
1266 		return;
1267 
1268 	for_each_pmc_rtnl(in_dev, im) {
1269 		hash = ip_mc_hash(im);
1270 		im->next_hash = mc_hash[hash];
1271 		RCU_INIT_POINTER(mc_hash[hash], im);
1272 	}
1273 
1274 	rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1275 }
1276 
1277 static void ip_mc_hash_remove(struct in_device *in_dev,
1278 			      struct ip_mc_list *im)
1279 {
1280 	struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1281 	struct ip_mc_list *aux;
1282 
1283 	if (!mc_hash)
1284 		return;
1285 	mc_hash += ip_mc_hash(im);
1286 	while ((aux = rtnl_dereference(*mc_hash)) != im)
1287 		mc_hash = &aux->next_hash;
1288 	*mc_hash = im->next_hash;
1289 }
1290 
1291 
1292 /*
1293  *	A socket has joined a multicast group on device dev.
1294  */
1295 
1296 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1297 {
1298 	struct ip_mc_list *im;
1299 
1300 	ASSERT_RTNL();
1301 
1302 	for_each_pmc_rtnl(in_dev, im) {
1303 		if (im->multiaddr == addr) {
1304 			im->users++;
1305 			ip_mc_add_src(in_dev, &addr, MCAST_EXCLUDE, 0, NULL, 0);
1306 			goto out;
1307 		}
1308 	}
1309 
1310 	im = kzalloc(sizeof(*im), GFP_KERNEL);
1311 	if (!im)
1312 		goto out;
1313 
1314 	im->users = 1;
1315 	im->interface = in_dev;
1316 	in_dev_hold(in_dev);
1317 	im->multiaddr = addr;
1318 	/* initial mode is (EX, empty) */
1319 	im->sfmode = MCAST_EXCLUDE;
1320 	im->sfcount[MCAST_EXCLUDE] = 1;
1321 	atomic_set(&im->refcnt, 1);
1322 	spin_lock_init(&im->lock);
1323 #ifdef CONFIG_IP_MULTICAST
1324 	setup_timer(&im->timer, igmp_timer_expire, (unsigned long)im);
1325 	im->unsolicit_count = IGMP_Unsolicited_Report_Count;
1326 #endif
1327 
1328 	im->next_rcu = in_dev->mc_list;
1329 	in_dev->mc_count++;
1330 	rcu_assign_pointer(in_dev->mc_list, im);
1331 
1332 	ip_mc_hash_add(in_dev, im);
1333 
1334 #ifdef CONFIG_IP_MULTICAST
1335 	igmpv3_del_delrec(in_dev, im->multiaddr);
1336 #endif
1337 	igmp_group_added(im);
1338 	if (!in_dev->dead)
1339 		ip_rt_multicast_event(in_dev);
1340 out:
1341 	return;
1342 }
1343 EXPORT_SYMBOL(ip_mc_inc_group);
1344 
1345 /*
1346  *	Resend IGMP JOIN report; used by netdev notifier.
1347  */
1348 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1349 {
1350 #ifdef CONFIG_IP_MULTICAST
1351 	struct ip_mc_list *im;
1352 	int type;
1353 
1354 	ASSERT_RTNL();
1355 
1356 	for_each_pmc_rtnl(in_dev, im) {
1357 		if (im->multiaddr == IGMP_ALL_HOSTS)
1358 			continue;
1359 
1360 		/* a failover is happening and switches
1361 		 * must be notified immediately
1362 		 */
1363 		if (IGMP_V1_SEEN(in_dev))
1364 			type = IGMP_HOST_MEMBERSHIP_REPORT;
1365 		else if (IGMP_V2_SEEN(in_dev))
1366 			type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1367 		else
1368 			type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1369 		igmp_send_report(in_dev, im, type);
1370 	}
1371 #endif
1372 }
1373 
1374 /*
1375  *	A socket has left a multicast group on device dev
1376  */
1377 
1378 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1379 {
1380 	struct ip_mc_list *i;
1381 	struct ip_mc_list __rcu **ip;
1382 
1383 	ASSERT_RTNL();
1384 
1385 	for (ip = &in_dev->mc_list;
1386 	     (i = rtnl_dereference(*ip)) != NULL;
1387 	     ip = &i->next_rcu) {
1388 		if (i->multiaddr == addr) {
1389 			if (--i->users == 0) {
1390 				ip_mc_hash_remove(in_dev, i);
1391 				*ip = i->next_rcu;
1392 				in_dev->mc_count--;
1393 				igmp_group_dropped(i);
1394 				ip_mc_clear_src(i);
1395 
1396 				if (!in_dev->dead)
1397 					ip_rt_multicast_event(in_dev);
1398 
1399 				ip_ma_put(i);
1400 				return;
1401 			}
1402 			break;
1403 		}
1404 	}
1405 }
1406 EXPORT_SYMBOL(ip_mc_dec_group);
1407 
1408 /* Device changing type */
1409 
1410 void ip_mc_unmap(struct in_device *in_dev)
1411 {
1412 	struct ip_mc_list *pmc;
1413 
1414 	ASSERT_RTNL();
1415 
1416 	for_each_pmc_rtnl(in_dev, pmc)
1417 		igmp_group_dropped(pmc);
1418 }
1419 
1420 void ip_mc_remap(struct in_device *in_dev)
1421 {
1422 	struct ip_mc_list *pmc;
1423 
1424 	ASSERT_RTNL();
1425 
1426 	for_each_pmc_rtnl(in_dev, pmc)
1427 		igmp_group_added(pmc);
1428 }
1429 
1430 /* Device going down */
1431 
1432 void ip_mc_down(struct in_device *in_dev)
1433 {
1434 	struct ip_mc_list *pmc;
1435 
1436 	ASSERT_RTNL();
1437 
1438 	for_each_pmc_rtnl(in_dev, pmc)
1439 		igmp_group_dropped(pmc);
1440 
1441 #ifdef CONFIG_IP_MULTICAST
1442 	in_dev->mr_ifc_count = 0;
1443 	if (del_timer(&in_dev->mr_ifc_timer))
1444 		__in_dev_put(in_dev);
1445 	in_dev->mr_gq_running = 0;
1446 	if (del_timer(&in_dev->mr_gq_timer))
1447 		__in_dev_put(in_dev);
1448 	igmpv3_clear_delrec(in_dev);
1449 #endif
1450 
1451 	ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1452 }
1453 
1454 void ip_mc_init_dev(struct in_device *in_dev)
1455 {
1456 	ASSERT_RTNL();
1457 
1458 #ifdef CONFIG_IP_MULTICAST
1459 	setup_timer(&in_dev->mr_gq_timer, igmp_gq_timer_expire,
1460 			(unsigned long)in_dev);
1461 	setup_timer(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire,
1462 			(unsigned long)in_dev);
1463 	in_dev->mr_qrv = IGMP_Unsolicited_Report_Count;
1464 #endif
1465 
1466 	spin_lock_init(&in_dev->mc_tomb_lock);
1467 }
1468 
1469 /* Device going up */
1470 
1471 void ip_mc_up(struct in_device *in_dev)
1472 {
1473 	struct ip_mc_list *pmc;
1474 
1475 	ASSERT_RTNL();
1476 
1477 	ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1478 
1479 	for_each_pmc_rtnl(in_dev, pmc)
1480 		igmp_group_added(pmc);
1481 }
1482 
1483 /*
1484  *	Device is about to be destroyed: clean up.
1485  */
1486 
1487 void ip_mc_destroy_dev(struct in_device *in_dev)
1488 {
1489 	struct ip_mc_list *i;
1490 
1491 	ASSERT_RTNL();
1492 
1493 	/* Deactivate timers */
1494 	ip_mc_down(in_dev);
1495 
1496 	while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1497 		in_dev->mc_list = i->next_rcu;
1498 		in_dev->mc_count--;
1499 
1500 		/* We've dropped the groups in ip_mc_down already */
1501 		ip_mc_clear_src(i);
1502 		ip_ma_put(i);
1503 	}
1504 }
1505 
1506 /* RTNL is locked */
1507 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1508 {
1509 	struct net_device *dev = NULL;
1510 	struct in_device *idev = NULL;
1511 
1512 	if (imr->imr_ifindex) {
1513 		idev = inetdev_by_index(net, imr->imr_ifindex);
1514 		return idev;
1515 	}
1516 	if (imr->imr_address.s_addr) {
1517 		dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1518 		if (!dev)
1519 			return NULL;
1520 	}
1521 
1522 	if (!dev) {
1523 		struct rtable *rt = ip_route_output(net,
1524 						    imr->imr_multiaddr.s_addr,
1525 						    0, 0, 0);
1526 		if (!IS_ERR(rt)) {
1527 			dev = rt->dst.dev;
1528 			ip_rt_put(rt);
1529 		}
1530 	}
1531 	if (dev) {
1532 		imr->imr_ifindex = dev->ifindex;
1533 		idev = __in_dev_get_rtnl(dev);
1534 	}
1535 	return idev;
1536 }
1537 
1538 /*
1539  *	Join a socket to a group
1540  */
1541 int sysctl_igmp_max_memberships __read_mostly = IP_MAX_MEMBERSHIPS;
1542 int sysctl_igmp_max_msf __read_mostly = IP_MAX_MSF;
1543 
1544 
1545 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1546 	__be32 *psfsrc)
1547 {
1548 	struct ip_sf_list *psf, *psf_prev;
1549 	int rv = 0;
1550 
1551 	psf_prev = NULL;
1552 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1553 		if (psf->sf_inaddr == *psfsrc)
1554 			break;
1555 		psf_prev = psf;
1556 	}
1557 	if (!psf || psf->sf_count[sfmode] == 0) {
1558 		/* source filter not found, or count wrong =>  bug */
1559 		return -ESRCH;
1560 	}
1561 	psf->sf_count[sfmode]--;
1562 	if (psf->sf_count[sfmode] == 0) {
1563 		ip_rt_multicast_event(pmc->interface);
1564 	}
1565 	if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1566 #ifdef CONFIG_IP_MULTICAST
1567 		struct in_device *in_dev = pmc->interface;
1568 #endif
1569 
1570 		/* no more filters for this source */
1571 		if (psf_prev)
1572 			psf_prev->sf_next = psf->sf_next;
1573 		else
1574 			pmc->sources = psf->sf_next;
1575 #ifdef CONFIG_IP_MULTICAST
1576 		if (psf->sf_oldin &&
1577 		    !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1578 			psf->sf_crcount = in_dev->mr_qrv ? in_dev->mr_qrv :
1579 				IGMP_Unsolicited_Report_Count;
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 ? in_dev->mr_qrv :
1643 			IGMP_Unsolicited_Report_Count;
1644 		in_dev->mr_ifc_count = pmc->crcount;
1645 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1646 			psf->sf_crcount = 0;
1647 		igmp_ifc_event(pmc->interface);
1648 	} else if (sf_setstate(pmc) || changerec) {
1649 		igmp_ifc_event(pmc->interface);
1650 #endif
1651 	}
1652 out_unlock:
1653 	spin_unlock_bh(&pmc->lock);
1654 	return err;
1655 }
1656 
1657 /*
1658  * Add multicast single-source filter to the interface list
1659  */
1660 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1661 	__be32 *psfsrc)
1662 {
1663 	struct ip_sf_list *psf, *psf_prev;
1664 
1665 	psf_prev = NULL;
1666 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1667 		if (psf->sf_inaddr == *psfsrc)
1668 			break;
1669 		psf_prev = psf;
1670 	}
1671 	if (!psf) {
1672 		psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1673 		if (!psf)
1674 			return -ENOBUFS;
1675 		psf->sf_inaddr = *psfsrc;
1676 		if (psf_prev) {
1677 			psf_prev->sf_next = psf;
1678 		} else
1679 			pmc->sources = psf;
1680 	}
1681 	psf->sf_count[sfmode]++;
1682 	if (psf->sf_count[sfmode] == 1) {
1683 		ip_rt_multicast_event(pmc->interface);
1684 	}
1685 	return 0;
1686 }
1687 
1688 #ifdef CONFIG_IP_MULTICAST
1689 static void sf_markstate(struct ip_mc_list *pmc)
1690 {
1691 	struct ip_sf_list *psf;
1692 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1693 
1694 	for (psf = pmc->sources; psf; psf = psf->sf_next)
1695 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1696 			psf->sf_oldin = mca_xcount ==
1697 				psf->sf_count[MCAST_EXCLUDE] &&
1698 				!psf->sf_count[MCAST_INCLUDE];
1699 		} else
1700 			psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1701 }
1702 
1703 static int sf_setstate(struct ip_mc_list *pmc)
1704 {
1705 	struct ip_sf_list *psf, *dpsf;
1706 	int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1707 	int qrv = pmc->interface->mr_qrv;
1708 	int new_in, rv;
1709 
1710 	rv = 0;
1711 	for (psf = pmc->sources; psf; psf = psf->sf_next) {
1712 		if (pmc->sfcount[MCAST_EXCLUDE]) {
1713 			new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1714 				!psf->sf_count[MCAST_INCLUDE];
1715 		} else
1716 			new_in = psf->sf_count[MCAST_INCLUDE] != 0;
1717 		if (new_in) {
1718 			if (!psf->sf_oldin) {
1719 				struct ip_sf_list *prev = NULL;
1720 
1721 				for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
1722 					if (dpsf->sf_inaddr == psf->sf_inaddr)
1723 						break;
1724 					prev = dpsf;
1725 				}
1726 				if (dpsf) {
1727 					if (prev)
1728 						prev->sf_next = dpsf->sf_next;
1729 					else
1730 						pmc->tomb = dpsf->sf_next;
1731 					kfree(dpsf);
1732 				}
1733 				psf->sf_crcount = qrv;
1734 				rv++;
1735 			}
1736 		} else if (psf->sf_oldin) {
1737 
1738 			psf->sf_crcount = 0;
1739 			/*
1740 			 * add or update "delete" records if an active filter
1741 			 * is now inactive
1742 			 */
1743 			for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
1744 				if (dpsf->sf_inaddr == psf->sf_inaddr)
1745 					break;
1746 			if (!dpsf) {
1747 				dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
1748 				if (!dpsf)
1749 					continue;
1750 				*dpsf = *psf;
1751 				/* pmc->lock held by callers */
1752 				dpsf->sf_next = pmc->tomb;
1753 				pmc->tomb = dpsf;
1754 			}
1755 			dpsf->sf_crcount = qrv;
1756 			rv++;
1757 		}
1758 	}
1759 	return rv;
1760 }
1761 #endif
1762 
1763 /*
1764  * Add multicast source filter list to the interface list
1765  */
1766 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1767 			 int sfcount, __be32 *psfsrc, int delta)
1768 {
1769 	struct ip_mc_list *pmc;
1770 	int	isexclude;
1771 	int	i, err;
1772 
1773 	if (!in_dev)
1774 		return -ENODEV;
1775 	rcu_read_lock();
1776 	for_each_pmc_rcu(in_dev, pmc) {
1777 		if (*pmca == pmc->multiaddr)
1778 			break;
1779 	}
1780 	if (!pmc) {
1781 		/* MCA not found?? bug */
1782 		rcu_read_unlock();
1783 		return -ESRCH;
1784 	}
1785 	spin_lock_bh(&pmc->lock);
1786 	rcu_read_unlock();
1787 
1788 #ifdef CONFIG_IP_MULTICAST
1789 	sf_markstate(pmc);
1790 #endif
1791 	isexclude = pmc->sfmode == MCAST_EXCLUDE;
1792 	if (!delta)
1793 		pmc->sfcount[sfmode]++;
1794 	err = 0;
1795 	for (i = 0; i < sfcount; i++) {
1796 		err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
1797 		if (err)
1798 			break;
1799 	}
1800 	if (err) {
1801 		int j;
1802 
1803 		if (!delta)
1804 			pmc->sfcount[sfmode]--;
1805 		for (j = 0; j < i; j++)
1806 			(void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
1807 	} else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
1808 #ifdef CONFIG_IP_MULTICAST
1809 		struct ip_sf_list *psf;
1810 		in_dev = pmc->interface;
1811 #endif
1812 
1813 		/* filter mode change */
1814 		if (pmc->sfcount[MCAST_EXCLUDE])
1815 			pmc->sfmode = MCAST_EXCLUDE;
1816 		else if (pmc->sfcount[MCAST_INCLUDE])
1817 			pmc->sfmode = MCAST_INCLUDE;
1818 #ifdef CONFIG_IP_MULTICAST
1819 		/* else no filters; keep old mode for reports */
1820 
1821 		pmc->crcount = in_dev->mr_qrv ? in_dev->mr_qrv :
1822 			IGMP_Unsolicited_Report_Count;
1823 		in_dev->mr_ifc_count = pmc->crcount;
1824 		for (psf = pmc->sources; psf; psf = psf->sf_next)
1825 			psf->sf_crcount = 0;
1826 		igmp_ifc_event(in_dev);
1827 	} else if (sf_setstate(pmc)) {
1828 		igmp_ifc_event(in_dev);
1829 #endif
1830 	}
1831 	spin_unlock_bh(&pmc->lock);
1832 	return err;
1833 }
1834 
1835 static void ip_mc_clear_src(struct ip_mc_list *pmc)
1836 {
1837 	struct ip_sf_list *psf, *nextpsf;
1838 
1839 	for (psf = pmc->tomb; psf; psf = nextpsf) {
1840 		nextpsf = psf->sf_next;
1841 		kfree(psf);
1842 	}
1843 	pmc->tomb = NULL;
1844 	for (psf = pmc->sources; psf; psf = nextpsf) {
1845 		nextpsf = psf->sf_next;
1846 		kfree(psf);
1847 	}
1848 	pmc->sources = NULL;
1849 	pmc->sfmode = MCAST_EXCLUDE;
1850 	pmc->sfcount[MCAST_INCLUDE] = 0;
1851 	pmc->sfcount[MCAST_EXCLUDE] = 1;
1852 }
1853 
1854 
1855 /*
1856  * Join a multicast group
1857  */
1858 int ip_mc_join_group(struct sock *sk , struct ip_mreqn *imr)
1859 {
1860 	int err;
1861 	__be32 addr = imr->imr_multiaddr.s_addr;
1862 	struct ip_mc_socklist *iml = NULL, *i;
1863 	struct in_device *in_dev;
1864 	struct inet_sock *inet = inet_sk(sk);
1865 	struct net *net = sock_net(sk);
1866 	int ifindex;
1867 	int count = 0;
1868 
1869 	if (!ipv4_is_multicast(addr))
1870 		return -EINVAL;
1871 
1872 	rtnl_lock();
1873 
1874 	in_dev = ip_mc_find_dev(net, imr);
1875 
1876 	if (!in_dev) {
1877 		iml = NULL;
1878 		err = -ENODEV;
1879 		goto done;
1880 	}
1881 
1882 	err = -EADDRINUSE;
1883 	ifindex = imr->imr_ifindex;
1884 	for_each_pmc_rtnl(inet, i) {
1885 		if (i->multi.imr_multiaddr.s_addr == addr &&
1886 		    i->multi.imr_ifindex == ifindex)
1887 			goto done;
1888 		count++;
1889 	}
1890 	err = -ENOBUFS;
1891 	if (count >= sysctl_igmp_max_memberships)
1892 		goto done;
1893 	iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
1894 	if (iml == NULL)
1895 		goto done;
1896 
1897 	memcpy(&iml->multi, imr, sizeof(*imr));
1898 	iml->next_rcu = inet->mc_list;
1899 	iml->sflist = NULL;
1900 	iml->sfmode = MCAST_EXCLUDE;
1901 	rcu_assign_pointer(inet->mc_list, iml);
1902 	ip_mc_inc_group(in_dev, addr);
1903 	err = 0;
1904 done:
1905 	rtnl_unlock();
1906 	return err;
1907 }
1908 EXPORT_SYMBOL(ip_mc_join_group);
1909 
1910 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
1911 			   struct in_device *in_dev)
1912 {
1913 	struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
1914 	int err;
1915 
1916 	if (psf == NULL) {
1917 		/* any-source empty exclude case */
1918 		return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
1919 			iml->sfmode, 0, NULL, 0);
1920 	}
1921 	err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
1922 			iml->sfmode, psf->sl_count, psf->sl_addr, 0);
1923 	RCU_INIT_POINTER(iml->sflist, NULL);
1924 	/* decrease mem now to avoid the memleak warning */
1925 	atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
1926 	kfree_rcu(psf, rcu);
1927 	return err;
1928 }
1929 
1930 /*
1931  *	Ask a socket to leave a group.
1932  */
1933 
1934 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
1935 {
1936 	struct inet_sock *inet = inet_sk(sk);
1937 	struct ip_mc_socklist *iml;
1938 	struct ip_mc_socklist __rcu **imlp;
1939 	struct in_device *in_dev;
1940 	struct net *net = sock_net(sk);
1941 	__be32 group = imr->imr_multiaddr.s_addr;
1942 	u32 ifindex;
1943 	int ret = -EADDRNOTAVAIL;
1944 
1945 	rtnl_lock();
1946 	in_dev = ip_mc_find_dev(net, imr);
1947 	if (!in_dev) {
1948 		ret = -ENODEV;
1949 		goto out;
1950 	}
1951 	ifindex = imr->imr_ifindex;
1952 	for (imlp = &inet->mc_list;
1953 	     (iml = rtnl_dereference(*imlp)) != NULL;
1954 	     imlp = &iml->next_rcu) {
1955 		if (iml->multi.imr_multiaddr.s_addr != group)
1956 			continue;
1957 		if (ifindex) {
1958 			if (iml->multi.imr_ifindex != ifindex)
1959 				continue;
1960 		} else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
1961 				iml->multi.imr_address.s_addr)
1962 			continue;
1963 
1964 		(void) ip_mc_leave_src(sk, iml, in_dev);
1965 
1966 		*imlp = iml->next_rcu;
1967 
1968 		ip_mc_dec_group(in_dev, group);
1969 		rtnl_unlock();
1970 		/* decrease mem now to avoid the memleak warning */
1971 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
1972 		kfree_rcu(iml, rcu);
1973 		return 0;
1974 	}
1975 out:
1976 	rtnl_unlock();
1977 	return ret;
1978 }
1979 EXPORT_SYMBOL(ip_mc_leave_group);
1980 
1981 int ip_mc_source(int add, int omode, struct sock *sk, struct
1982 	ip_mreq_source *mreqs, int ifindex)
1983 {
1984 	int err;
1985 	struct ip_mreqn imr;
1986 	__be32 addr = mreqs->imr_multiaddr;
1987 	struct ip_mc_socklist *pmc;
1988 	struct in_device *in_dev = NULL;
1989 	struct inet_sock *inet = inet_sk(sk);
1990 	struct ip_sf_socklist *psl;
1991 	struct net *net = sock_net(sk);
1992 	int leavegroup = 0;
1993 	int i, j, rv;
1994 
1995 	if (!ipv4_is_multicast(addr))
1996 		return -EINVAL;
1997 
1998 	rtnl_lock();
1999 
2000 	imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2001 	imr.imr_address.s_addr = mreqs->imr_interface;
2002 	imr.imr_ifindex = ifindex;
2003 	in_dev = ip_mc_find_dev(net, &imr);
2004 
2005 	if (!in_dev) {
2006 		err = -ENODEV;
2007 		goto done;
2008 	}
2009 	err = -EADDRNOTAVAIL;
2010 
2011 	for_each_pmc_rtnl(inet, pmc) {
2012 		if ((pmc->multi.imr_multiaddr.s_addr ==
2013 		     imr.imr_multiaddr.s_addr) &&
2014 		    (pmc->multi.imr_ifindex == imr.imr_ifindex))
2015 			break;
2016 	}
2017 	if (!pmc) {		/* must have a prior join */
2018 		err = -EINVAL;
2019 		goto done;
2020 	}
2021 	/* if a source filter was set, must be the same mode as before */
2022 	if (pmc->sflist) {
2023 		if (pmc->sfmode != omode) {
2024 			err = -EINVAL;
2025 			goto done;
2026 		}
2027 	} else if (pmc->sfmode != omode) {
2028 		/* allow mode switches for empty-set filters */
2029 		ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2030 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2031 			NULL, 0);
2032 		pmc->sfmode = omode;
2033 	}
2034 
2035 	psl = rtnl_dereference(pmc->sflist);
2036 	if (!add) {
2037 		if (!psl)
2038 			goto done;	/* err = -EADDRNOTAVAIL */
2039 		rv = !0;
2040 		for (i = 0; i < psl->sl_count; i++) {
2041 			rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2042 				sizeof(__be32));
2043 			if (rv == 0)
2044 				break;
2045 		}
2046 		if (rv)		/* source not found */
2047 			goto done;	/* err = -EADDRNOTAVAIL */
2048 
2049 		/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2050 		if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2051 			leavegroup = 1;
2052 			goto done;
2053 		}
2054 
2055 		/* update the interface filter */
2056 		ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2057 			&mreqs->imr_sourceaddr, 1);
2058 
2059 		for (j = i+1; j < psl->sl_count; j++)
2060 			psl->sl_addr[j-1] = psl->sl_addr[j];
2061 		psl->sl_count--;
2062 		err = 0;
2063 		goto done;
2064 	}
2065 	/* else, add a new source to the filter */
2066 
2067 	if (psl && psl->sl_count >= sysctl_igmp_max_msf) {
2068 		err = -ENOBUFS;
2069 		goto done;
2070 	}
2071 	if (!psl || psl->sl_count == psl->sl_max) {
2072 		struct ip_sf_socklist *newpsl;
2073 		int count = IP_SFBLOCK;
2074 
2075 		if (psl)
2076 			count += psl->sl_max;
2077 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2078 		if (!newpsl) {
2079 			err = -ENOBUFS;
2080 			goto done;
2081 		}
2082 		newpsl->sl_max = count;
2083 		newpsl->sl_count = count - IP_SFBLOCK;
2084 		if (psl) {
2085 			for (i = 0; i < psl->sl_count; i++)
2086 				newpsl->sl_addr[i] = psl->sl_addr[i];
2087 			/* decrease mem now to avoid the memleak warning */
2088 			atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2089 			kfree_rcu(psl, rcu);
2090 		}
2091 		rcu_assign_pointer(pmc->sflist, newpsl);
2092 		psl = newpsl;
2093 	}
2094 	rv = 1;	/* > 0 for insert logic below if sl_count is 0 */
2095 	for (i = 0; i < psl->sl_count; i++) {
2096 		rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2097 			sizeof(__be32));
2098 		if (rv == 0)
2099 			break;
2100 	}
2101 	if (rv == 0)		/* address already there is an error */
2102 		goto done;
2103 	for (j = psl->sl_count-1; j >= i; j--)
2104 		psl->sl_addr[j+1] = psl->sl_addr[j];
2105 	psl->sl_addr[i] = mreqs->imr_sourceaddr;
2106 	psl->sl_count++;
2107 	err = 0;
2108 	/* update the interface list */
2109 	ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2110 		&mreqs->imr_sourceaddr, 1);
2111 done:
2112 	rtnl_unlock();
2113 	if (leavegroup)
2114 		return ip_mc_leave_group(sk, &imr);
2115 	return err;
2116 }
2117 
2118 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2119 {
2120 	int err = 0;
2121 	struct ip_mreqn	imr;
2122 	__be32 addr = msf->imsf_multiaddr;
2123 	struct ip_mc_socklist *pmc;
2124 	struct in_device *in_dev;
2125 	struct inet_sock *inet = inet_sk(sk);
2126 	struct ip_sf_socklist *newpsl, *psl;
2127 	struct net *net = sock_net(sk);
2128 	int leavegroup = 0;
2129 
2130 	if (!ipv4_is_multicast(addr))
2131 		return -EINVAL;
2132 	if (msf->imsf_fmode != MCAST_INCLUDE &&
2133 	    msf->imsf_fmode != MCAST_EXCLUDE)
2134 		return -EINVAL;
2135 
2136 	rtnl_lock();
2137 
2138 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2139 	imr.imr_address.s_addr = msf->imsf_interface;
2140 	imr.imr_ifindex = ifindex;
2141 	in_dev = ip_mc_find_dev(net, &imr);
2142 
2143 	if (!in_dev) {
2144 		err = -ENODEV;
2145 		goto done;
2146 	}
2147 
2148 	/* special case - (INCLUDE, empty) == LEAVE_GROUP */
2149 	if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2150 		leavegroup = 1;
2151 		goto done;
2152 	}
2153 
2154 	for_each_pmc_rtnl(inet, pmc) {
2155 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2156 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2157 			break;
2158 	}
2159 	if (!pmc) {		/* must have a prior join */
2160 		err = -EINVAL;
2161 		goto done;
2162 	}
2163 	if (msf->imsf_numsrc) {
2164 		newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2165 							   GFP_KERNEL);
2166 		if (!newpsl) {
2167 			err = -ENOBUFS;
2168 			goto done;
2169 		}
2170 		newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2171 		memcpy(newpsl->sl_addr, msf->imsf_slist,
2172 			msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2173 		err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2174 			msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2175 		if (err) {
2176 			sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2177 			goto done;
2178 		}
2179 	} else {
2180 		newpsl = NULL;
2181 		(void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2182 				     msf->imsf_fmode, 0, NULL, 0);
2183 	}
2184 	psl = rtnl_dereference(pmc->sflist);
2185 	if (psl) {
2186 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2187 			psl->sl_count, psl->sl_addr, 0);
2188 		/* decrease mem now to avoid the memleak warning */
2189 		atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2190 		kfree_rcu(psl, rcu);
2191 	} else
2192 		(void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2193 			0, NULL, 0);
2194 	rcu_assign_pointer(pmc->sflist, newpsl);
2195 	pmc->sfmode = msf->imsf_fmode;
2196 	err = 0;
2197 done:
2198 	rtnl_unlock();
2199 	if (leavegroup)
2200 		err = ip_mc_leave_group(sk, &imr);
2201 	return err;
2202 }
2203 
2204 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2205 	struct ip_msfilter __user *optval, int __user *optlen)
2206 {
2207 	int err, len, count, copycount;
2208 	struct ip_mreqn	imr;
2209 	__be32 addr = msf->imsf_multiaddr;
2210 	struct ip_mc_socklist *pmc;
2211 	struct in_device *in_dev;
2212 	struct inet_sock *inet = inet_sk(sk);
2213 	struct ip_sf_socklist *psl;
2214 	struct net *net = sock_net(sk);
2215 
2216 	if (!ipv4_is_multicast(addr))
2217 		return -EINVAL;
2218 
2219 	rtnl_lock();
2220 
2221 	imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2222 	imr.imr_address.s_addr = msf->imsf_interface;
2223 	imr.imr_ifindex = 0;
2224 	in_dev = ip_mc_find_dev(net, &imr);
2225 
2226 	if (!in_dev) {
2227 		err = -ENODEV;
2228 		goto done;
2229 	}
2230 	err = -EADDRNOTAVAIL;
2231 
2232 	for_each_pmc_rtnl(inet, pmc) {
2233 		if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2234 		    pmc->multi.imr_ifindex == imr.imr_ifindex)
2235 			break;
2236 	}
2237 	if (!pmc)		/* must have a prior join */
2238 		goto done;
2239 	msf->imsf_fmode = pmc->sfmode;
2240 	psl = rtnl_dereference(pmc->sflist);
2241 	rtnl_unlock();
2242 	if (!psl) {
2243 		len = 0;
2244 		count = 0;
2245 	} else {
2246 		count = psl->sl_count;
2247 	}
2248 	copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2249 	len = copycount * sizeof(psl->sl_addr[0]);
2250 	msf->imsf_numsrc = count;
2251 	if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2252 	    copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2253 		return -EFAULT;
2254 	}
2255 	if (len &&
2256 	    copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2257 		return -EFAULT;
2258 	return 0;
2259 done:
2260 	rtnl_unlock();
2261 	return err;
2262 }
2263 
2264 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2265 	struct group_filter __user *optval, int __user *optlen)
2266 {
2267 	int err, i, count, copycount;
2268 	struct sockaddr_in *psin;
2269 	__be32 addr;
2270 	struct ip_mc_socklist *pmc;
2271 	struct inet_sock *inet = inet_sk(sk);
2272 	struct ip_sf_socklist *psl;
2273 
2274 	psin = (struct sockaddr_in *)&gsf->gf_group;
2275 	if (psin->sin_family != AF_INET)
2276 		return -EINVAL;
2277 	addr = psin->sin_addr.s_addr;
2278 	if (!ipv4_is_multicast(addr))
2279 		return -EINVAL;
2280 
2281 	rtnl_lock();
2282 
2283 	err = -EADDRNOTAVAIL;
2284 
2285 	for_each_pmc_rtnl(inet, pmc) {
2286 		if (pmc->multi.imr_multiaddr.s_addr == addr &&
2287 		    pmc->multi.imr_ifindex == gsf->gf_interface)
2288 			break;
2289 	}
2290 	if (!pmc)		/* must have a prior join */
2291 		goto done;
2292 	gsf->gf_fmode = pmc->sfmode;
2293 	psl = rtnl_dereference(pmc->sflist);
2294 	rtnl_unlock();
2295 	count = psl ? psl->sl_count : 0;
2296 	copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2297 	gsf->gf_numsrc = count;
2298 	if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2299 	    copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2300 		return -EFAULT;
2301 	}
2302 	for (i = 0; i < copycount; i++) {
2303 		struct sockaddr_storage ss;
2304 
2305 		psin = (struct sockaddr_in *)&ss;
2306 		memset(&ss, 0, sizeof(ss));
2307 		psin->sin_family = AF_INET;
2308 		psin->sin_addr.s_addr = psl->sl_addr[i];
2309 		if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2310 			return -EFAULT;
2311 	}
2312 	return 0;
2313 done:
2314 	rtnl_unlock();
2315 	return err;
2316 }
2317 
2318 /*
2319  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2320  */
2321 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr, int dif)
2322 {
2323 	struct inet_sock *inet = inet_sk(sk);
2324 	struct ip_mc_socklist *pmc;
2325 	struct ip_sf_socklist *psl;
2326 	int i;
2327 	int ret;
2328 
2329 	ret = 1;
2330 	if (!ipv4_is_multicast(loc_addr))
2331 		goto out;
2332 
2333 	rcu_read_lock();
2334 	for_each_pmc_rcu(inet, pmc) {
2335 		if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2336 		    pmc->multi.imr_ifindex == dif)
2337 			break;
2338 	}
2339 	ret = inet->mc_all;
2340 	if (!pmc)
2341 		goto unlock;
2342 	psl = rcu_dereference(pmc->sflist);
2343 	ret = (pmc->sfmode == MCAST_EXCLUDE);
2344 	if (!psl)
2345 		goto unlock;
2346 
2347 	for (i = 0; i < psl->sl_count; i++) {
2348 		if (psl->sl_addr[i] == rmt_addr)
2349 			break;
2350 	}
2351 	ret = 0;
2352 	if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2353 		goto unlock;
2354 	if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2355 		goto unlock;
2356 	ret = 1;
2357 unlock:
2358 	rcu_read_unlock();
2359 out:
2360 	return ret;
2361 }
2362 
2363 /*
2364  *	A socket is closing.
2365  */
2366 
2367 void ip_mc_drop_socket(struct sock *sk)
2368 {
2369 	struct inet_sock *inet = inet_sk(sk);
2370 	struct ip_mc_socklist *iml;
2371 	struct net *net = sock_net(sk);
2372 
2373 	if (inet->mc_list == NULL)
2374 		return;
2375 
2376 	rtnl_lock();
2377 	while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2378 		struct in_device *in_dev;
2379 
2380 		inet->mc_list = iml->next_rcu;
2381 		in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2382 		(void) ip_mc_leave_src(sk, iml, in_dev);
2383 		if (in_dev != NULL)
2384 			ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2385 		/* decrease mem now to avoid the memleak warning */
2386 		atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2387 		kfree_rcu(iml, rcu);
2388 	}
2389 	rtnl_unlock();
2390 }
2391 
2392 /* called with rcu_read_lock() */
2393 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u16 proto)
2394 {
2395 	struct ip_mc_list *im;
2396 	struct ip_mc_list __rcu **mc_hash;
2397 	struct ip_sf_list *psf;
2398 	int rv = 0;
2399 
2400 	mc_hash = rcu_dereference(in_dev->mc_hash);
2401 	if (mc_hash) {
2402 		u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2403 
2404 		for (im = rcu_dereference(mc_hash[hash]);
2405 		     im != NULL;
2406 		     im = rcu_dereference(im->next_hash)) {
2407 			if (im->multiaddr == mc_addr)
2408 				break;
2409 		}
2410 	} else {
2411 		for_each_pmc_rcu(in_dev, im) {
2412 			if (im->multiaddr == mc_addr)
2413 				break;
2414 		}
2415 	}
2416 	if (im && proto == IPPROTO_IGMP) {
2417 		rv = 1;
2418 	} else if (im) {
2419 		if (src_addr) {
2420 			for (psf = im->sources; psf; psf = psf->sf_next) {
2421 				if (psf->sf_inaddr == src_addr)
2422 					break;
2423 			}
2424 			if (psf)
2425 				rv = psf->sf_count[MCAST_INCLUDE] ||
2426 					psf->sf_count[MCAST_EXCLUDE] !=
2427 					im->sfcount[MCAST_EXCLUDE];
2428 			else
2429 				rv = im->sfcount[MCAST_EXCLUDE] != 0;
2430 		} else
2431 			rv = 1; /* unspecified source; tentatively allow */
2432 	}
2433 	return rv;
2434 }
2435 
2436 #if defined(CONFIG_PROC_FS)
2437 struct igmp_mc_iter_state {
2438 	struct seq_net_private p;
2439 	struct net_device *dev;
2440 	struct in_device *in_dev;
2441 };
2442 
2443 #define	igmp_mc_seq_private(seq)	((struct igmp_mc_iter_state *)(seq)->private)
2444 
2445 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2446 {
2447 	struct net *net = seq_file_net(seq);
2448 	struct ip_mc_list *im = NULL;
2449 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2450 
2451 	state->in_dev = NULL;
2452 	for_each_netdev_rcu(net, state->dev) {
2453 		struct in_device *in_dev;
2454 
2455 		in_dev = __in_dev_get_rcu(state->dev);
2456 		if (!in_dev)
2457 			continue;
2458 		im = rcu_dereference(in_dev->mc_list);
2459 		if (im) {
2460 			state->in_dev = in_dev;
2461 			break;
2462 		}
2463 	}
2464 	return im;
2465 }
2466 
2467 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2468 {
2469 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2470 
2471 	im = rcu_dereference(im->next_rcu);
2472 	while (!im) {
2473 		state->dev = next_net_device_rcu(state->dev);
2474 		if (!state->dev) {
2475 			state->in_dev = NULL;
2476 			break;
2477 		}
2478 		state->in_dev = __in_dev_get_rcu(state->dev);
2479 		if (!state->in_dev)
2480 			continue;
2481 		im = rcu_dereference(state->in_dev->mc_list);
2482 	}
2483 	return im;
2484 }
2485 
2486 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2487 {
2488 	struct ip_mc_list *im = igmp_mc_get_first(seq);
2489 	if (im)
2490 		while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2491 			--pos;
2492 	return pos ? NULL : im;
2493 }
2494 
2495 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2496 	__acquires(rcu)
2497 {
2498 	rcu_read_lock();
2499 	return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2500 }
2501 
2502 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2503 {
2504 	struct ip_mc_list *im;
2505 	if (v == SEQ_START_TOKEN)
2506 		im = igmp_mc_get_first(seq);
2507 	else
2508 		im = igmp_mc_get_next(seq, v);
2509 	++*pos;
2510 	return im;
2511 }
2512 
2513 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2514 	__releases(rcu)
2515 {
2516 	struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2517 
2518 	state->in_dev = NULL;
2519 	state->dev = NULL;
2520 	rcu_read_unlock();
2521 }
2522 
2523 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2524 {
2525 	if (v == SEQ_START_TOKEN)
2526 		seq_puts(seq,
2527 			 "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2528 	else {
2529 		struct ip_mc_list *im = (struct ip_mc_list *)v;
2530 		struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2531 		char   *querier;
2532 		long delta;
2533 
2534 #ifdef CONFIG_IP_MULTICAST
2535 		querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2536 			  IGMP_V2_SEEN(state->in_dev) ? "V2" :
2537 			  "V3";
2538 #else
2539 		querier = "NONE";
2540 #endif
2541 
2542 		if (rcu_dereference(state->in_dev->mc_list) == im) {
2543 			seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2544 				   state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2545 		}
2546 
2547 		delta = im->timer.expires - jiffies;
2548 		seq_printf(seq,
2549 			   "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2550 			   im->multiaddr, im->users,
2551 			   im->tm_running,
2552 			   im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2553 			   im->reporter);
2554 	}
2555 	return 0;
2556 }
2557 
2558 static const struct seq_operations igmp_mc_seq_ops = {
2559 	.start	=	igmp_mc_seq_start,
2560 	.next	=	igmp_mc_seq_next,
2561 	.stop	=	igmp_mc_seq_stop,
2562 	.show	=	igmp_mc_seq_show,
2563 };
2564 
2565 static int igmp_mc_seq_open(struct inode *inode, struct file *file)
2566 {
2567 	return seq_open_net(inode, file, &igmp_mc_seq_ops,
2568 			sizeof(struct igmp_mc_iter_state));
2569 }
2570 
2571 static const struct file_operations igmp_mc_seq_fops = {
2572 	.owner		=	THIS_MODULE,
2573 	.open		=	igmp_mc_seq_open,
2574 	.read		=	seq_read,
2575 	.llseek		=	seq_lseek,
2576 	.release	=	seq_release_net,
2577 };
2578 
2579 struct igmp_mcf_iter_state {
2580 	struct seq_net_private p;
2581 	struct net_device *dev;
2582 	struct in_device *idev;
2583 	struct ip_mc_list *im;
2584 };
2585 
2586 #define igmp_mcf_seq_private(seq)	((struct igmp_mcf_iter_state *)(seq)->private)
2587 
2588 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2589 {
2590 	struct net *net = seq_file_net(seq);
2591 	struct ip_sf_list *psf = NULL;
2592 	struct ip_mc_list *im = NULL;
2593 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2594 
2595 	state->idev = NULL;
2596 	state->im = NULL;
2597 	for_each_netdev_rcu(net, state->dev) {
2598 		struct in_device *idev;
2599 		idev = __in_dev_get_rcu(state->dev);
2600 		if (unlikely(idev == NULL))
2601 			continue;
2602 		im = rcu_dereference(idev->mc_list);
2603 		if (likely(im != NULL)) {
2604 			spin_lock_bh(&im->lock);
2605 			psf = im->sources;
2606 			if (likely(psf != NULL)) {
2607 				state->im = im;
2608 				state->idev = idev;
2609 				break;
2610 			}
2611 			spin_unlock_bh(&im->lock);
2612 		}
2613 	}
2614 	return psf;
2615 }
2616 
2617 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2618 {
2619 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2620 
2621 	psf = psf->sf_next;
2622 	while (!psf) {
2623 		spin_unlock_bh(&state->im->lock);
2624 		state->im = state->im->next;
2625 		while (!state->im) {
2626 			state->dev = next_net_device_rcu(state->dev);
2627 			if (!state->dev) {
2628 				state->idev = NULL;
2629 				goto out;
2630 			}
2631 			state->idev = __in_dev_get_rcu(state->dev);
2632 			if (!state->idev)
2633 				continue;
2634 			state->im = rcu_dereference(state->idev->mc_list);
2635 		}
2636 		if (!state->im)
2637 			break;
2638 		spin_lock_bh(&state->im->lock);
2639 		psf = state->im->sources;
2640 	}
2641 out:
2642 	return psf;
2643 }
2644 
2645 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2646 {
2647 	struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2648 	if (psf)
2649 		while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2650 			--pos;
2651 	return pos ? NULL : psf;
2652 }
2653 
2654 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2655 	__acquires(rcu)
2656 {
2657 	rcu_read_lock();
2658 	return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2659 }
2660 
2661 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2662 {
2663 	struct ip_sf_list *psf;
2664 	if (v == SEQ_START_TOKEN)
2665 		psf = igmp_mcf_get_first(seq);
2666 	else
2667 		psf = igmp_mcf_get_next(seq, v);
2668 	++*pos;
2669 	return psf;
2670 }
2671 
2672 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2673 	__releases(rcu)
2674 {
2675 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2676 	if (likely(state->im != NULL)) {
2677 		spin_unlock_bh(&state->im->lock);
2678 		state->im = NULL;
2679 	}
2680 	state->idev = NULL;
2681 	state->dev = NULL;
2682 	rcu_read_unlock();
2683 }
2684 
2685 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2686 {
2687 	struct ip_sf_list *psf = (struct ip_sf_list *)v;
2688 	struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2689 
2690 	if (v == SEQ_START_TOKEN) {
2691 		seq_printf(seq,
2692 			   "%3s %6s "
2693 			   "%10s %10s %6s %6s\n", "Idx",
2694 			   "Device", "MCA",
2695 			   "SRC", "INC", "EXC");
2696 	} else {
2697 		seq_printf(seq,
2698 			   "%3d %6.6s 0x%08x "
2699 			   "0x%08x %6lu %6lu\n",
2700 			   state->dev->ifindex, state->dev->name,
2701 			   ntohl(state->im->multiaddr),
2702 			   ntohl(psf->sf_inaddr),
2703 			   psf->sf_count[MCAST_INCLUDE],
2704 			   psf->sf_count[MCAST_EXCLUDE]);
2705 	}
2706 	return 0;
2707 }
2708 
2709 static const struct seq_operations igmp_mcf_seq_ops = {
2710 	.start	=	igmp_mcf_seq_start,
2711 	.next	=	igmp_mcf_seq_next,
2712 	.stop	=	igmp_mcf_seq_stop,
2713 	.show	=	igmp_mcf_seq_show,
2714 };
2715 
2716 static int igmp_mcf_seq_open(struct inode *inode, struct file *file)
2717 {
2718 	return seq_open_net(inode, file, &igmp_mcf_seq_ops,
2719 			sizeof(struct igmp_mcf_iter_state));
2720 }
2721 
2722 static const struct file_operations igmp_mcf_seq_fops = {
2723 	.owner		=	THIS_MODULE,
2724 	.open		=	igmp_mcf_seq_open,
2725 	.read		=	seq_read,
2726 	.llseek		=	seq_lseek,
2727 	.release	=	seq_release_net,
2728 };
2729 
2730 static int __net_init igmp_net_init(struct net *net)
2731 {
2732 	struct proc_dir_entry *pde;
2733 
2734 	pde = proc_create("igmp", S_IRUGO, net->proc_net, &igmp_mc_seq_fops);
2735 	if (!pde)
2736 		goto out_igmp;
2737 	pde = proc_create("mcfilter", S_IRUGO, net->proc_net,
2738 			  &igmp_mcf_seq_fops);
2739 	if (!pde)
2740 		goto out_mcfilter;
2741 	return 0;
2742 
2743 out_mcfilter:
2744 	remove_proc_entry("igmp", net->proc_net);
2745 out_igmp:
2746 	return -ENOMEM;
2747 }
2748 
2749 static void __net_exit igmp_net_exit(struct net *net)
2750 {
2751 	remove_proc_entry("mcfilter", net->proc_net);
2752 	remove_proc_entry("igmp", net->proc_net);
2753 }
2754 
2755 static struct pernet_operations igmp_net_ops = {
2756 	.init = igmp_net_init,
2757 	.exit = igmp_net_exit,
2758 };
2759 #endif
2760 
2761 static int igmp_netdev_event(struct notifier_block *this,
2762 			     unsigned long event, void *ptr)
2763 {
2764 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2765 	struct in_device *in_dev;
2766 
2767 	switch (event) {
2768 	case NETDEV_RESEND_IGMP:
2769 		in_dev = __in_dev_get_rtnl(dev);
2770 		if (in_dev)
2771 			ip_mc_rejoin_groups(in_dev);
2772 		break;
2773 	default:
2774 		break;
2775 	}
2776 	return NOTIFY_DONE;
2777 }
2778 
2779 static struct notifier_block igmp_notifier = {
2780 	.notifier_call = igmp_netdev_event,
2781 };
2782 
2783 int __init igmp_mc_init(void)
2784 {
2785 #if defined(CONFIG_PROC_FS)
2786 	int err;
2787 
2788 	err = register_pernet_subsys(&igmp_net_ops);
2789 	if (err)
2790 		return err;
2791 	err = register_netdevice_notifier(&igmp_notifier);
2792 	if (err)
2793 		goto reg_notif_fail;
2794 	return 0;
2795 
2796 reg_notif_fail:
2797 	unregister_pernet_subsys(&igmp_net_ops);
2798 	return err;
2799 #else
2800 	return register_netdevice_notifier(&igmp_notifier);
2801 #endif
2802 }
2803