net/ipv4/igmp.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux NET3:     Internet Group Management Protocol  [IGMP]
   4  *
   5  *      This code implements the IGMP protocol as defined in RFC1112. There has
   6  *      been a further revision of this protocol since which is now supported.
   7  *
   8  *      If you have trouble with this module be careful what gcc you have used,
   9  *      the older version didn't come out right using gcc 2.5.8, the newer one
  10  *      seems to fall out with gcc 2.6.2.
  11  *
  12  *      Authors:
  13  *              Alan Cox <[email protected]>
  14  *
  15  *      Fixes:
  16  *
  17  *              Alan Cox        :       Added lots of __inline__ to optimise
  18  *                                      the memory usage of all the tiny little
  19  *                                      functions.
  20  *              Alan Cox        :       Dumped the header building experiment.
  21  *              Alan Cox        :       Minor tweaks ready for multicast routing
  22  *                                      and extended IGMP protocol.
  23  *              Alan Cox        :       Removed a load of inline directives. Gcc 2.5.8
  24  *                                      writes utterly bogus code otherwise (sigh)
  25  *                                      fixed IGMP loopback to behave in the manner
  26  *                                      desired by mrouted, fixed the fact it has been
  27  *                                      broken since 1.3.6 and cleaned up a few minor
  28  *                                      points.
  29  *
  30  *              Chih-Jen Chang  :       Tried to revise IGMP to Version 2
  31  *              Tsu-Sheng Tsao          E-mail: [email protected] and [email protected]
  32  *                                      The enhancements are mainly based on Steve Deering's
  33  *                                      ipmulti-3.5 source code.
  34  *              Chih-Jen Chang  :       Added the igmp_get_mrouter_info and
  35  *              Tsu-Sheng Tsao          igmp_set_mrouter_info to keep track of
  36  *                                      the mrouted version on that device.
  37  *              Chih-Jen Chang  :       Added the max_resp_time parameter to
  38  *              Tsu-Sheng Tsao          igmp_heard_query(). Using this parameter
  39  *                                      to identify the multicast router version
  40  *                                      and do what the IGMP version 2 specified.
  41  *              Chih-Jen Chang  :       Added a timer to revert to IGMP V2 router
  42  *              Tsu-Sheng Tsao          if the specified time expired.
  43  *              Alan Cox        :       Stop IGMP from 0.0.0.0 being accepted.
  44  *              Alan Cox        :       Use GFP_ATOMIC in the right places.
  45  *              Christian Daudt :       igmp timer wasn't set for local group
  46  *                                      memberships but was being deleted,
  47  *                                      which caused a "del_timer() called
  48  *                                      from %p with timer not initialized\n"
  49  *                                      message (960131).
  50  *              Christian Daudt :       removed del_timer from
  51  *                                      igmp_timer_expire function (960205).
  52  *             Christian Daudt :       igmp_heard_report now only calls
  53  *                                     igmp_timer_expire if tm->running is
  54  *                                     true (960216).
  55  *              Malcolm Beattie :       ttl comparison wrong in igmp_rcv made
  56  *                                      igmp_heard_query never trigger. Expiry
  57  *                                      miscalculation fixed in igmp_heard_query
  58  *                                      and random() made to return unsigned to
  59  *                                      prevent negative expiry times.
  60  *              Alexey Kuznetsov:       Wrong group leaving behaviour, backport
  61  *                                      fix from pending 2.1.x patches.
  62  *              Alan Cox:               Forget to enable FDDI support earlier.
  63  *              Alexey Kuznetsov:       Fixed leaving groups on device down.
  64  *              Alexey Kuznetsov:       Accordance to igmp-v2-06 draft.
  65  *              David L Stevens:        IGMPv3 support, with help from
  66  *                                      Vinay Kulkarni
  67  */
  68
  69 #include <linux/module.h>
  70 #include <linux/slab.h>
  71 #include <linux/uaccess.h>
  72 #include <linux/types.h>
  73 #include <linux/kernel.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/string.h>
  76 #include <linux/socket.h>
  77 #include <linux/sockios.h>
  78 #include <linux/in.h>
  79 #include <linux/inet.h>
  80 #include <linux/netdevice.h>
  81 #include <linux/skbuff.h>
  82 #include <linux/inetdevice.h>
  83 #include <linux/igmp.h>
  84 #include <linux/if_arp.h>
  85 #include <linux/rtnetlink.h>
  86 #include <linux/times.h>
  87 #include <linux/pkt_sched.h>
  88 #include <linux/byteorder/generic.h>
  89
  90 #include <net/net_namespace.h>
  91 #include <net/arp.h>
  92 #include <net/ip.h>
  93 #include <net/protocol.h>
  94 #include <net/route.h>
  95 #include <net/sock.h>
  96 #include <net/checksum.h>
  97 #include <net/inet_common.h>
  98 #include <linux/netfilter_ipv4.h>
  99 #ifdef CONFIG_IP_MROUTE
 100 #include <linux/mroute.h>
 101 #endif
 102 #ifdef CONFIG_PROC_FS
 103 #include <linux/proc_fs.h>
 104 #include <linux/seq_file.h>
 105 #endif
 106
 107 #ifdef CONFIG_IP_MULTICAST
 108 /* Parameter names and values are taken from igmp-v2-06 draft */
 109
 110 #define IGMP_QUERY_INTERVAL                     (125*HZ)
 111 #define IGMP_QUERY_RESPONSE_INTERVAL            (10*HZ)
 112
 113 #define IGMP_INITIAL_REPORT_DELAY               (1)
 114
 115 /* IGMP_INITIAL_REPORT_DELAY is not from IGMP specs!
 116  * IGMP specs require to report membership immediately after
 117  * joining a group, but we delay the first report by a
 118  * small interval. It seems more natural and still does not
 119  * contradict to specs provided this delay is small enough.
 120  */
 121
 122 #define IGMP_V1_SEEN(in_dev) \
 123         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
 124          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
 125          ((in_dev)->mr_v1_seen && \
 126           time_before(jiffies, (in_dev)->mr_v1_seen)))
 127 #define IGMP_V2_SEEN(in_dev) \
 128         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
 129          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
 130          ((in_dev)->mr_v2_seen && \
 131           time_before(jiffies, (in_dev)->mr_v2_seen)))
 132
 133 static int unsolicited_report_interval(struct in_device *in_dev)
 134 {
 135         int interval_ms, interval_jiffies;
 136
 137         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 138                 interval_ms = IN_DEV_CONF_GET(
 139                         in_dev,
 140                         IGMPV2_UNSOLICITED_REPORT_INTERVAL);
 141         else /* v3 */
 142                 interval_ms = IN_DEV_CONF_GET(
 143                         in_dev,
 144                         IGMPV3_UNSOLICITED_REPORT_INTERVAL);
 145
 146         interval_jiffies = msecs_to_jiffies(interval_ms);
 147
 148         /* _timer functions can't handle a delay of 0 jiffies so ensure
 149          *  we always return a positive value.
 150          */
 151         if (interval_jiffies <= 0)
 152                 interval_jiffies = 1;
 153         return interval_jiffies;
 154 }
 155
 156 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
 157                               gfp_t gfp);
 158 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 159 static void igmpv3_clear_delrec(struct in_device *in_dev);
 160 static int sf_setstate(struct ip_mc_list *pmc);
 161 static void sf_markstate(struct ip_mc_list *pmc);
 162 #endif
 163 static void ip_mc_clear_src(struct ip_mc_list *pmc);
 164 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
 165                          int sfcount, __be32 *psfsrc, int delta);
 166
 167 static void ip_ma_put(struct ip_mc_list *im)
 168 {
 169         if (refcount_dec_and_test(&im->refcnt)) {
 170                 in_dev_put(im->interface);
 171                 kfree_rcu(im, rcu);
 172         }
 173 }
 174
 175 #define for_each_pmc_rcu(in_dev, pmc)                           \
 176         for (pmc = rcu_dereference(in_dev->mc_list);            \
 177              pmc != NULL;                                       \
 178              pmc = rcu_dereference(pmc->next_rcu))
 179
 180 #define for_each_pmc_rtnl(in_dev, pmc)                          \
 181         for (pmc = rtnl_dereference(in_dev->mc_list);           \
 182              pmc != NULL;                                       \
 183              pmc = rtnl_dereference(pmc->next_rcu))
 184
 185 static void ip_sf_list_clear_all(struct ip_sf_list *psf)
 186 {
 187         struct ip_sf_list *next;
 188
 189         while (psf) {
 190                 next = psf->sf_next;
 191                 kfree(psf);
 192                 psf = next;
 193         }
 194 }
 195
 196 #ifdef CONFIG_IP_MULTICAST
 197
 198 /*
 199  *      Timer management
 200  */
 201
 202 static void igmp_stop_timer(struct ip_mc_list *im)
 203 {
 204         spin_lock_bh(&im->lock);
 205         if (del_timer(&im->timer))
 206                 refcount_dec(&im->refcnt);
 207         im->tm_running = 0;
 208         im->reporter = 0;
 209         im->unsolicit_count = 0;
 210         spin_unlock_bh(&im->lock);
 211 }
 212
 213 /* It must be called with locked im->lock */
 214 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
 215 {
 216         int tv = get_random_u32_below(max_delay);
 217
 218         im->tm_running = 1;
 219         if (!mod_timer(&im->timer, jiffies+tv+2))
 220                 refcount_inc(&im->refcnt);
 221 }
 222
 223 static void igmp_gq_start_timer(struct in_device *in_dev)
 224 {
 225         int tv = get_random_u32_below(in_dev->mr_maxdelay);
 226         unsigned long exp = jiffies + tv + 2;
 227
 228         if (in_dev->mr_gq_running &&
 229             time_after_eq(exp, (in_dev->mr_gq_timer).expires))
 230                 return;
 231
 232         in_dev->mr_gq_running = 1;
 233         if (!mod_timer(&in_dev->mr_gq_timer, exp))
 234                 in_dev_hold(in_dev);
 235 }
 236
 237 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 238 {
 239         int tv = get_random_u32_below(delay);
 240
 241         if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
 242                 in_dev_hold(in_dev);
 243 }
 244
 245 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 246 {
 247         spin_lock_bh(&im->lock);
 248         im->unsolicit_count = 0;
 249         if (del_timer(&im->timer)) {
 250                 if ((long)(im->timer.expires-jiffies) < max_delay) {
 251                         add_timer(&im->timer);
 252                         im->tm_running = 1;
 253                         spin_unlock_bh(&im->lock);
 254                         return;
 255                 }
 256                 refcount_dec(&im->refcnt);
 257         }
 258         igmp_start_timer(im, max_delay);
 259         spin_unlock_bh(&im->lock);
 260 }
 261
 262
 263 /*
 264  *      Send an IGMP report.
 265  */
 266
 267 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
 268
 269
 270 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
 271         int gdeleted, int sdeleted)
 272 {
 273         switch (type) {
 274         case IGMPV3_MODE_IS_INCLUDE:
 275         case IGMPV3_MODE_IS_EXCLUDE:
 276                 if (gdeleted || sdeleted)
 277                         return 0;
 278                 if (!(pmc->gsquery && !psf->sf_gsresp)) {
 279                         if (pmc->sfmode == MCAST_INCLUDE)
 280                                 return 1;
 281                         /* don't include if this source is excluded
 282                          * in all filters
 283                          */
 284                         if (psf->sf_count[MCAST_INCLUDE])
 285                                 return type == IGMPV3_MODE_IS_INCLUDE;
 286                         return pmc->sfcount[MCAST_EXCLUDE] ==
 287                                 psf->sf_count[MCAST_EXCLUDE];
 288                 }
 289                 return 0;
 290         case IGMPV3_CHANGE_TO_INCLUDE:
 291                 if (gdeleted || sdeleted)
 292                         return 0;
 293                 return psf->sf_count[MCAST_INCLUDE] != 0;
 294         case IGMPV3_CHANGE_TO_EXCLUDE:
 295                 if (gdeleted || sdeleted)
 296                         return 0;
 297                 if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
 298                     psf->sf_count[MCAST_INCLUDE])
 299                         return 0;
 300                 return pmc->sfcount[MCAST_EXCLUDE] ==
 301                         psf->sf_count[MCAST_EXCLUDE];
 302         case IGMPV3_ALLOW_NEW_SOURCES:
 303                 if (gdeleted || !psf->sf_crcount)
 304                         return 0;
 305                 return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
 306         case IGMPV3_BLOCK_OLD_SOURCES:
 307                 if (pmc->sfmode == MCAST_INCLUDE)
 308                         return gdeleted || (psf->sf_crcount && sdeleted);
 309                 return psf->sf_crcount && !gdeleted && !sdeleted;
 310         }
 311         return 0;
 312 }
 313
 314 static int
 315 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
 316 {
 317         struct ip_sf_list *psf;
 318         int scount = 0;
 319
 320         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 321                 if (!is_in(pmc, psf, type, gdeleted, sdeleted))
 322                         continue;
 323                 scount++;
 324         }
 325         return scount;
 326 }
 327
 328 /* source address selection per RFC 3376 section 4.2.13 */
 329 static __be32 igmpv3_get_srcaddr(struct net_device *dev,
 330                                  const struct flowi4 *fl4)
 331 {
 332         struct in_device *in_dev = __in_dev_get_rcu(dev);
 333         const struct in_ifaddr *ifa;
 334
 335         if (!in_dev)
 336                 return htonl(INADDR_ANY);
 337
 338         in_dev_for_each_ifa_rcu(ifa, in_dev) {
 339                 if (fl4->saddr == ifa->ifa_local)
 340                         return fl4->saddr;
 341         }
 342
 343         return htonl(INADDR_ANY);
 344 }
 345
 346 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
 347 {
 348         struct sk_buff *skb;
 349         struct rtable *rt;
 350         struct iphdr *pip;
 351         struct igmpv3_report *pig;
 352         struct net *net = dev_net(dev);
 353         struct flowi4 fl4;
 354         int hlen = LL_RESERVED_SPACE(dev);
 355         int tlen = dev->needed_tailroom;
 356         unsigned int size;
 357
 358         size = min(mtu, IP_MAX_MTU);
 359         while (1) {
 360                 skb = alloc_skb(size + hlen + tlen,
 361                                 GFP_ATOMIC | __GFP_NOWARN);
 362                 if (skb)
 363                         break;
 364                 size >>= 1;
 365                 if (size < 256)
 366                         return NULL;
 367         }
 368         skb->priority = TC_PRIO_CONTROL;
 369
 370         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 371                                    0, 0,
 372                                    IPPROTO_IGMP, 0, dev->ifindex);
 373         if (IS_ERR(rt)) {
 374                 kfree_skb(skb);
 375                 return NULL;
 376         }
 377
 378         skb_dst_set(skb, &rt->dst);
 379         skb->dev = dev;
 380
 381         skb_reserve(skb, hlen);
 382         skb_tailroom_reserve(skb, mtu, tlen);
 383
 384         skb_reset_network_header(skb);
 385         pip = ip_hdr(skb);
 386         skb_put(skb, sizeof(struct iphdr) + 4);
 387
 388         pip->version  = 4;
 389         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 390         pip->tos      = 0xc0;
 391         pip->frag_off = htons(IP_DF);
 392         pip->ttl      = 1;
 393         pip->daddr    = fl4.daddr;
 394
 395         rcu_read_lock();
 396         pip->saddr    = igmpv3_get_srcaddr(dev, &fl4);
 397         rcu_read_unlock();
 398
 399         pip->protocol = IPPROTO_IGMP;
 400         pip->tot_len  = 0;      /* filled in later */
 401         ip_select_ident(net, skb, NULL);
 402         ((u8 *)&pip[1])[0] = IPOPT_RA;
 403         ((u8 *)&pip[1])[1] = 4;
 404         ((u8 *)&pip[1])[2] = 0;
 405         ((u8 *)&pip[1])[3] = 0;
 406
 407         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 408         skb_put(skb, sizeof(*pig));
 409         pig = igmpv3_report_hdr(skb);
 410         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 411         pig->resv1 = 0;
 412         pig->csum = 0;
 413         pig->resv2 = 0;
 414         pig->ngrec = 0;
 415         return skb;
 416 }
 417
 418 static int igmpv3_sendpack(struct sk_buff *skb)
 419 {
 420         struct igmphdr *pig = igmp_hdr(skb);
 421         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 422
 423         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 424
 425         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 426 }
 427
 428 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 429 {
 430         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 431 }
 432
 433 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 434         int type, struct igmpv3_grec **ppgr, unsigned int mtu)
 435 {
 436         struct net_device *dev = pmc->interface->dev;
 437         struct igmpv3_report *pih;
 438         struct igmpv3_grec *pgr;
 439
 440         if (!skb) {
 441                 skb = igmpv3_newpack(dev, mtu);
 442                 if (!skb)
 443                         return NULL;
 444         }
 445         pgr = skb_put(skb, sizeof(struct igmpv3_grec));
 446         pgr->grec_type = type;
 447         pgr->grec_auxwords = 0;
 448         pgr->grec_nsrcs = 0;
 449         pgr->grec_mca = pmc->multiaddr;
 450         pih = igmpv3_report_hdr(skb);
 451         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 452         *ppgr = pgr;
 453         return skb;
 454 }
 455
 456 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 457
 458 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 459         int type, int gdeleted, int sdeleted)
 460 {
 461         struct net_device *dev = pmc->interface->dev;
 462         struct net *net = dev_net(dev);
 463         struct igmpv3_report *pih;
 464         struct igmpv3_grec *pgr = NULL;
 465         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 466         int scount, stotal, first, isquery, truncate;
 467         unsigned int mtu;
 468
 469         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 470                 return skb;
 471         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 472             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 473                 return skb;
 474
 475         mtu = READ_ONCE(dev->mtu);
 476         if (mtu < IPV4_MIN_MTU)
 477                 return skb;
 478
 479         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 480                   type == IGMPV3_MODE_IS_EXCLUDE;
 481         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 482                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 483
 484         stotal = scount = 0;
 485
 486         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 487
 488         if (!*psf_list)
 489                 goto empty_source;
 490
 491         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 492
 493         /* EX and TO_EX get a fresh packet, if needed */
 494         if (truncate) {
 495                 if (pih && pih->ngrec &&
 496                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 497                         if (skb)
 498                                 igmpv3_sendpack(skb);
 499                         skb = igmpv3_newpack(dev, mtu);
 500                 }
 501         }
 502         first = 1;
 503         psf_prev = NULL;
 504         for (psf = *psf_list; psf; psf = psf_next) {
 505                 __be32 *psrc;
 506
 507                 psf_next = psf->sf_next;
 508
 509                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 510                         psf_prev = psf;
 511                         continue;
 512                 }
 513
 514                 /* Based on RFC3376 5.1. Should not send source-list change
 515                  * records when there is a filter mode change.
 516                  */
 517                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 518                      (!gdeleted && pmc->crcount)) &&
 519                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 520                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 521                         goto decrease_sf_crcount;
 522
 523                 /* clear marks on query responses */
 524                 if (isquery)
 525                         psf->sf_gsresp = 0;
 526
 527                 if (AVAILABLE(skb) < sizeof(__be32) +
 528                     first*sizeof(struct igmpv3_grec)) {
 529                         if (truncate && !first)
 530                                 break;   /* truncate these */
 531                         if (pgr)
 532                                 pgr->grec_nsrcs = htons(scount);
 533                         if (skb)
 534                                 igmpv3_sendpack(skb);
 535                         skb = igmpv3_newpack(dev, mtu);
 536                         first = 1;
 537                         scount = 0;
 538                 }
 539                 if (first) {
 540                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 541                         first = 0;
 542                 }
 543                 if (!skb)
 544                         return NULL;
 545                 psrc = skb_put(skb, sizeof(__be32));
 546                 *psrc = psf->sf_inaddr;
 547                 scount++; stotal++;
 548                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 549                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 550 decrease_sf_crcount:
 551                         psf->sf_crcount--;
 552                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 553                                 if (psf_prev)
 554                                         psf_prev->sf_next = psf->sf_next;
 555                                 else
 556                                         *psf_list = psf->sf_next;
 557                                 kfree(psf);
 558                                 continue;
 559                         }
 560                 }
 561                 psf_prev = psf;
 562         }
 563
 564 empty_source:
 565         if (!stotal) {
 566                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 567                     type == IGMPV3_BLOCK_OLD_SOURCES)
 568                         return skb;
 569                 if (pmc->crcount || isquery) {
 570                         /* make sure we have room for group header */
 571                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 572                                 igmpv3_sendpack(skb);
 573                                 skb = NULL; /* add_grhead will get a new one */
 574                         }
 575                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 576                 }
 577         }
 578         if (pgr)
 579                 pgr->grec_nsrcs = htons(scount);
 580
 581         if (isquery)
 582                 pmc->gsquery = 0;       /* clear query state on report */
 583         return skb;
 584 }
 585
 586 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 587 {
 588         struct sk_buff *skb = NULL;
 589         struct net *net = dev_net(in_dev->dev);
 590         int type;
 591
 592         if (!pmc) {
 593                 rcu_read_lock();
 594                 for_each_pmc_rcu(in_dev, pmc) {
 595                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 596                                 continue;
 597                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 598                             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 599                                 continue;
 600                         spin_lock_bh(&pmc->lock);
 601                         if (pmc->sfcount[MCAST_EXCLUDE])
 602                                 type = IGMPV3_MODE_IS_EXCLUDE;
 603                         else
 604                                 type = IGMPV3_MODE_IS_INCLUDE;
 605                         skb = add_grec(skb, pmc, type, 0, 0);
 606                         spin_unlock_bh(&pmc->lock);
 607                 }
 608                 rcu_read_unlock();
 609         } else {
 610                 spin_lock_bh(&pmc->lock);
 611                 if (pmc->sfcount[MCAST_EXCLUDE])
 612                         type = IGMPV3_MODE_IS_EXCLUDE;
 613                 else
 614                         type = IGMPV3_MODE_IS_INCLUDE;
 615                 skb = add_grec(skb, pmc, type, 0, 0);
 616                 spin_unlock_bh(&pmc->lock);
 617         }
 618         if (!skb)
 619                 return 0;
 620         return igmpv3_sendpack(skb);
 621 }
 622
 623 /*
 624  * remove zero-count source records from a source filter list
 625  */
 626 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 627 {
 628         struct ip_sf_list *psf_prev, *psf_next, *psf;
 629
 630         psf_prev = NULL;
 631         for (psf = *ppsf; psf; psf = psf_next) {
 632                 psf_next = psf->sf_next;
 633                 if (psf->sf_crcount == 0) {
 634                         if (psf_prev)
 635                                 psf_prev->sf_next = psf->sf_next;
 636                         else
 637                                 *ppsf = psf->sf_next;
 638                         kfree(psf);
 639                 } else
 640                         psf_prev = psf;
 641         }
 642 }
 643
 644 static void kfree_pmc(struct ip_mc_list *pmc)
 645 {
 646         ip_sf_list_clear_all(pmc->sources);
 647         ip_sf_list_clear_all(pmc->tomb);
 648         kfree(pmc);
 649 }
 650
 651 static void igmpv3_send_cr(struct in_device *in_dev)
 652 {
 653         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 654         struct sk_buff *skb = NULL;
 655         int type, dtype;
 656
 657         rcu_read_lock();
 658         spin_lock_bh(&in_dev->mc_tomb_lock);
 659
 660         /* deleted MCA's */
 661         pmc_prev = NULL;
 662         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 663                 pmc_next = pmc->next;
 664                 if (pmc->sfmode == MCAST_INCLUDE) {
 665                         type = IGMPV3_BLOCK_OLD_SOURCES;
 666                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 667                         skb = add_grec(skb, pmc, type, 1, 0);
 668                         skb = add_grec(skb, pmc, dtype, 1, 1);
 669                 }
 670                 if (pmc->crcount) {
 671                         if (pmc->sfmode == MCAST_EXCLUDE) {
 672                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 673                                 skb = add_grec(skb, pmc, type, 1, 0);
 674                         }
 675                         pmc->crcount--;
 676                         if (pmc->crcount == 0) {
 677                                 igmpv3_clear_zeros(&pmc->tomb);
 678                                 igmpv3_clear_zeros(&pmc->sources);
 679                         }
 680                 }
 681                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 682                         if (pmc_prev)
 683                                 pmc_prev->next = pmc_next;
 684                         else
 685                                 in_dev->mc_tomb = pmc_next;
 686                         in_dev_put(pmc->interface);
 687                         kfree_pmc(pmc);
 688                 } else
 689                         pmc_prev = pmc;
 690         }
 691         spin_unlock_bh(&in_dev->mc_tomb_lock);
 692
 693         /* change recs */
 694         for_each_pmc_rcu(in_dev, pmc) {
 695                 spin_lock_bh(&pmc->lock);
 696                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 697                         type = IGMPV3_BLOCK_OLD_SOURCES;
 698                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 699                 } else {
 700                         type = IGMPV3_ALLOW_NEW_SOURCES;
 701                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 702                 }
 703                 skb = add_grec(skb, pmc, type, 0, 0);
 704                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 705
 706                 /* filter mode changes */
 707                 if (pmc->crcount) {
 708                         if (pmc->sfmode == MCAST_EXCLUDE)
 709                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 710                         else
 711                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 712                         skb = add_grec(skb, pmc, type, 0, 0);
 713                         pmc->crcount--;
 714                 }
 715                 spin_unlock_bh(&pmc->lock);
 716         }
 717         rcu_read_unlock();
 718
 719         if (!skb)
 720                 return;
 721         (void) igmpv3_sendpack(skb);
 722 }
 723
 724 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 725         int type)
 726 {
 727         struct sk_buff *skb;
 728         struct iphdr *iph;
 729         struct igmphdr *ih;
 730         struct rtable *rt;
 731         struct net_device *dev = in_dev->dev;
 732         struct net *net = dev_net(dev);
 733         __be32  group = pmc ? pmc->multiaddr : 0;
 734         struct flowi4 fl4;
 735         __be32  dst;
 736         int hlen, tlen;
 737
 738         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 739                 return igmpv3_send_report(in_dev, pmc);
 740
 741         if (ipv4_is_local_multicast(group) &&
 742             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 743                 return 0;
 744
 745         if (type == IGMP_HOST_LEAVE_MESSAGE)
 746                 dst = IGMP_ALL_ROUTER;
 747         else
 748                 dst = group;
 749
 750         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 751                                    0, 0,
 752                                    IPPROTO_IGMP, 0, dev->ifindex);
 753         if (IS_ERR(rt))
 754                 return -1;
 755
 756         hlen = LL_RESERVED_SPACE(dev);
 757         tlen = dev->needed_tailroom;
 758         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 759         if (!skb) {
 760                 ip_rt_put(rt);
 761                 return -1;
 762         }
 763         skb->priority = TC_PRIO_CONTROL;
 764
 765         skb_dst_set(skb, &rt->dst);
 766
 767         skb_reserve(skb, hlen);
 768
 769         skb_reset_network_header(skb);
 770         iph = ip_hdr(skb);
 771         skb_put(skb, sizeof(struct iphdr) + 4);
 772
 773         iph->version  = 4;
 774         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 775         iph->tos      = 0xc0;
 776         iph->frag_off = htons(IP_DF);
 777         iph->ttl      = 1;
 778         iph->daddr    = dst;
 779         iph->saddr    = fl4.saddr;
 780         iph->protocol = IPPROTO_IGMP;
 781         ip_select_ident(net, skb, NULL);
 782         ((u8 *)&iph[1])[0] = IPOPT_RA;
 783         ((u8 *)&iph[1])[1] = 4;
 784         ((u8 *)&iph[1])[2] = 0;
 785         ((u8 *)&iph[1])[3] = 0;
 786
 787         ih = skb_put(skb, sizeof(struct igmphdr));
 788         ih->type = type;
 789         ih->code = 0;
 790         ih->csum = 0;
 791         ih->group = group;
 792         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 793
 794         return ip_local_out(net, skb->sk, skb);
 795 }
 796
 797 static void igmp_gq_timer_expire(struct timer_list *t)
 798 {
 799         struct in_device *in_dev = from_timer(in_dev, t, mr_gq_timer);
 800
 801         in_dev->mr_gq_running = 0;
 802         igmpv3_send_report(in_dev, NULL);
 803         in_dev_put(in_dev);
 804 }
 805
 806 static void igmp_ifc_timer_expire(struct timer_list *t)
 807 {
 808         struct in_device *in_dev = from_timer(in_dev, t, mr_ifc_timer);
 809         u32 mr_ifc_count;
 810
 811         igmpv3_send_cr(in_dev);
 812 restart:
 813         mr_ifc_count = READ_ONCE(in_dev->mr_ifc_count);
 814
 815         if (mr_ifc_count) {
 816                 if (cmpxchg(&in_dev->mr_ifc_count,
 817                             mr_ifc_count,
 818                             mr_ifc_count - 1) != mr_ifc_count)
 819                         goto restart;
 820                 igmp_ifc_start_timer(in_dev,
 821                                      unsolicited_report_interval(in_dev));
 822         }
 823         in_dev_put(in_dev);
 824 }
 825
 826 static void igmp_ifc_event(struct in_device *in_dev)
 827 {
 828         struct net *net = dev_net(in_dev->dev);
 829         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 830                 return;
 831         WRITE_ONCE(in_dev->mr_ifc_count, in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv));
 832         igmp_ifc_start_timer(in_dev, 1);
 833 }
 834
 835
 836 static void igmp_timer_expire(struct timer_list *t)
 837 {
 838         struct ip_mc_list *im = from_timer(im, t, timer);
 839         struct in_device *in_dev = im->interface;
 840
 841         spin_lock(&im->lock);
 842         im->tm_running = 0;
 843
 844         if (im->unsolicit_count && --im->unsolicit_count)
 845                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 846
 847         im->reporter = 1;
 848         spin_unlock(&im->lock);
 849
 850         if (IGMP_V1_SEEN(in_dev))
 851                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 852         else if (IGMP_V2_SEEN(in_dev))
 853                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 854         else
 855                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 856
 857         ip_ma_put(im);
 858 }
 859
 860 /* mark EXCLUDE-mode sources */
 861 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 862 {
 863         struct ip_sf_list *psf;
 864         int i, scount;
 865
 866         scount = 0;
 867         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 868                 if (scount == nsrcs)
 869                         break;
 870                 for (i = 0; i < nsrcs; i++) {
 871                         /* skip inactive filters */
 872                         if (psf->sf_count[MCAST_INCLUDE] ||
 873                             pmc->sfcount[MCAST_EXCLUDE] !=
 874                             psf->sf_count[MCAST_EXCLUDE])
 875                                 break;
 876                         if (srcs[i] == psf->sf_inaddr) {
 877                                 scount++;
 878                                 break;
 879                         }
 880                 }
 881         }
 882         pmc->gsquery = 0;
 883         if (scount == nsrcs)    /* all sources excluded */
 884                 return 0;
 885         return 1;
 886 }
 887
 888 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 889 {
 890         struct ip_sf_list *psf;
 891         int i, scount;
 892
 893         if (pmc->sfmode == MCAST_EXCLUDE)
 894                 return igmp_xmarksources(pmc, nsrcs, srcs);
 895
 896         /* mark INCLUDE-mode sources */
 897         scount = 0;
 898         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 899                 if (scount == nsrcs)
 900                         break;
 901                 for (i = 0; i < nsrcs; i++)
 902                         if (srcs[i] == psf->sf_inaddr) {
 903                                 psf->sf_gsresp = 1;
 904                                 scount++;
 905                                 break;
 906                         }
 907         }
 908         if (!scount) {
 909                 pmc->gsquery = 0;
 910                 return 0;
 911         }
 912         pmc->gsquery = 1;
 913         return 1;
 914 }
 915
 916 /* return true if packet was dropped */
 917 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 918 {
 919         struct ip_mc_list *im;
 920         struct net *net = dev_net(in_dev->dev);
 921
 922         /* Timers are only set for non-local groups */
 923
 924         if (group == IGMP_ALL_HOSTS)
 925                 return false;
 926         if (ipv4_is_local_multicast(group) &&
 927             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 928                 return false;
 929
 930         rcu_read_lock();
 931         for_each_pmc_rcu(in_dev, im) {
 932                 if (im->multiaddr == group) {
 933                         igmp_stop_timer(im);
 934                         break;
 935                 }
 936         }
 937         rcu_read_unlock();
 938         return false;
 939 }
 940
 941 /* return true if packet was dropped */
 942 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 943         int len)
 944 {
 945         struct igmphdr          *ih = igmp_hdr(skb);
 946         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 947         struct ip_mc_list       *im;
 948         __be32                  group = ih->group;
 949         int                     max_delay;
 950         int                     mark = 0;
 951         struct net              *net = dev_net(in_dev->dev);
 952
 953
 954         if (len == 8) {
 955                 if (ih->code == 0) {
 956                         /* Alas, old v1 router presents here. */
 957
 958                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 959                         in_dev->mr_v1_seen = jiffies +
 960                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 961                                 in_dev->mr_qri;
 962                         group = 0;
 963                 } else {
 964                         /* v2 router present */
 965                         max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
 966                         in_dev->mr_v2_seen = jiffies +
 967                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 968                                 in_dev->mr_qri;
 969                 }
 970                 /* cancel the interface change timer */
 971                 WRITE_ONCE(in_dev->mr_ifc_count, 0);
 972                 if (del_timer(&in_dev->mr_ifc_timer))
 973                         __in_dev_put(in_dev);
 974                 /* clear deleted report items */
 975                 igmpv3_clear_delrec(in_dev);
 976         } else if (len < 12) {
 977                 return true;    /* ignore bogus packet; freed by caller */
 978         } else if (IGMP_V1_SEEN(in_dev)) {
 979                 /* This is a v3 query with v1 queriers present */
 980                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 981                 group = 0;
 982         } else if (IGMP_V2_SEEN(in_dev)) {
 983                 /* this is a v3 query with v2 queriers present;
 984                  * Interpretation of the max_delay code is problematic here.
 985                  * A real v2 host would use ih_code directly, while v3 has a
 986                  * different encoding. We use the v3 encoding as more likely
 987                  * to be intended in a v3 query.
 988                  */
 989                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 990                 if (!max_delay)
 991                         max_delay = 1;  /* can't mod w/ 0 */
 992         } else { /* v3 */
 993                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 994                         return true;
 995
 996                 ih3 = igmpv3_query_hdr(skb);
 997                 if (ih3->nsrcs) {
 998                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
 999                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
1000                                 return true;
1001                         ih3 = igmpv3_query_hdr(skb);
1002                 }
1003
1004                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
1005                 if (!max_delay)
1006                         max_delay = 1;  /* can't mod w/ 0 */
1007                 in_dev->mr_maxdelay = max_delay;
1008
1009                 /* RFC3376, 4.1.6. QRV and 4.1.7. QQIC, when the most recently
1010                  * received value was zero, use the default or statically
1011                  * configured value.
1012                  */
1013                 in_dev->mr_qrv = ih3->qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1014                 in_dev->mr_qi = IGMPV3_QQIC(ih3->qqic)*HZ ?: IGMP_QUERY_INTERVAL;
1015
1016                 /* RFC3376, 8.3. Query Response Interval:
1017                  * The number of seconds represented by the [Query Response
1018                  * Interval] must be less than the [Query Interval].
1019                  */
1020                 if (in_dev->mr_qri >= in_dev->mr_qi)
1021                         in_dev->mr_qri = (in_dev->mr_qi/HZ - 1)*HZ;
1022
1023                 if (!group) { /* general query */
1024                         if (ih3->nsrcs)
1025                                 return true;    /* no sources allowed */
1026                         igmp_gq_start_timer(in_dev);
1027                         return false;
1028                 }
1029                 /* mark sources to include, if group & source-specific */
1030                 mark = ih3->nsrcs != 0;
1031         }
1032
1033         /*
1034          * - Start the timers in all of our membership records
1035          *   that the query applies to for the interface on
1036          *   which the query arrived excl. those that belong
1037          *   to a "local" group (224.0.0.X)
1038          * - For timers already running check if they need to
1039          *   be reset.
1040          * - Use the igmp->igmp_code field as the maximum
1041          *   delay possible
1042          */
1043         rcu_read_lock();
1044         for_each_pmc_rcu(in_dev, im) {
1045                 int changed;
1046
1047                 if (group && group != im->multiaddr)
1048                         continue;
1049                 if (im->multiaddr == IGMP_ALL_HOSTS)
1050                         continue;
1051                 if (ipv4_is_local_multicast(im->multiaddr) &&
1052                     !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1053                         continue;
1054                 spin_lock_bh(&im->lock);
1055                 if (im->tm_running)
1056                         im->gsquery = im->gsquery && mark;
1057                 else
1058                         im->gsquery = mark;
1059                 changed = !im->gsquery ||
1060                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1061                 spin_unlock_bh(&im->lock);
1062                 if (changed)
1063                         igmp_mod_timer(im, max_delay);
1064         }
1065         rcu_read_unlock();
1066         return false;
1067 }
1068
1069 /* called in rcu_read_lock() section */
1070 int igmp_rcv(struct sk_buff *skb)
1071 {
1072         /* This basically follows the spec line by line -- see RFC1112 */
1073         struct igmphdr *ih;
1074         struct net_device *dev = skb->dev;
1075         struct in_device *in_dev;
1076         int len = skb->len;
1077         bool dropped = true;
1078
1079         if (netif_is_l3_master(dev)) {
1080                 dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1081                 if (!dev)
1082                         goto drop;
1083         }
1084
1085         in_dev = __in_dev_get_rcu(dev);
1086         if (!in_dev)
1087                 goto drop;
1088
1089         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1090                 goto drop;
1091
1092         if (skb_checksum_simple_validate(skb))
1093                 goto drop;
1094
1095         ih = igmp_hdr(skb);
1096         switch (ih->type) {
1097         case IGMP_HOST_MEMBERSHIP_QUERY:
1098                 dropped = igmp_heard_query(in_dev, skb, len);
1099                 break;
1100         case IGMP_HOST_MEMBERSHIP_REPORT:
1101         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1102                 /* Is it our report looped back? */
1103                 if (rt_is_output_route(skb_rtable(skb)))
1104                         break;
1105                 /* don't rely on MC router hearing unicast reports */
1106                 if (skb->pkt_type == PACKET_MULTICAST ||
1107                     skb->pkt_type == PACKET_BROADCAST)
1108                         dropped = igmp_heard_report(in_dev, ih->group);
1109                 break;
1110         case IGMP_PIM:
1111 #ifdef CONFIG_IP_PIMSM_V1
1112                 return pim_rcv_v1(skb);
1113 #endif
1114         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1115         case IGMP_DVMRP:
1116         case IGMP_TRACE:
1117         case IGMP_HOST_LEAVE_MESSAGE:
1118         case IGMP_MTRACE:
1119         case IGMP_MTRACE_RESP:
1120                 break;
1121         default:
1122                 break;
1123         }
1124
1125 drop:
1126         if (dropped)
1127                 kfree_skb(skb);
1128         else
1129                 consume_skb(skb);
1130         return 0;
1131 }
1132
1133 #endif
1134
1135
1136 /*
1137  *      Add a filter to a device
1138  */
1139
1140 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1141 {
1142         char buf[MAX_ADDR_LEN];
1143         struct net_device *dev = in_dev->dev;
1144
1145         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1146            We will get multicast token leakage, when IFF_MULTICAST
1147            is changed. This check should be done in ndo_set_rx_mode
1148            routine. Something sort of:
1149            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1150            --ANK
1151            */
1152         if (arp_mc_map(addr, buf, dev, 0) == 0)
1153                 dev_mc_add(dev, buf);
1154 }
1155
1156 /*
1157  *      Remove a filter from a device
1158  */
1159
1160 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1161 {
1162         char buf[MAX_ADDR_LEN];
1163         struct net_device *dev = in_dev->dev;
1164
1165         if (arp_mc_map(addr, buf, dev, 0) == 0)
1166                 dev_mc_del(dev, buf);
1167 }
1168
1169 #ifdef CONFIG_IP_MULTICAST
1170 /*
1171  * deleted ip_mc_list manipulation
1172  */
1173 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
1174                               gfp_t gfp)
1175 {
1176         struct ip_mc_list *pmc;
1177         struct net *net = dev_net(in_dev->dev);
1178
1179         /* this is an "ip_mc_list" for convenience; only the fields below
1180          * are actually used. In particular, the refcnt and users are not
1181          * used for management of the delete list. Using the same structure
1182          * for deleted items allows change reports to use common code with
1183          * non-deleted or query-response MCA's.
1184          */
1185         pmc = kzalloc(sizeof(*pmc), gfp);
1186         if (!pmc)
1187                 return;
1188         spin_lock_init(&pmc->lock);
1189         spin_lock_bh(&im->lock);
1190         pmc->interface = im->interface;
1191         in_dev_hold(in_dev);
1192         pmc->multiaddr = im->multiaddr;
1193         pmc->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1194         pmc->sfmode = im->sfmode;
1195         if (pmc->sfmode == MCAST_INCLUDE) {
1196                 struct ip_sf_list *psf;
1197
1198                 pmc->tomb = im->tomb;
1199                 pmc->sources = im->sources;
1200                 im->tomb = im->sources = NULL;
1201                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1202                         psf->sf_crcount = pmc->crcount;
1203         }
1204         spin_unlock_bh(&im->lock);
1205
1206         spin_lock_bh(&in_dev->mc_tomb_lock);
1207         pmc->next = in_dev->mc_tomb;
1208         in_dev->mc_tomb = pmc;
1209         spin_unlock_bh(&in_dev->mc_tomb_lock);
1210 }
1211
1212 /*
1213  * restore ip_mc_list deleted records
1214  */
1215 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1216 {
1217         struct ip_mc_list *pmc, *pmc_prev;
1218         struct ip_sf_list *psf;
1219         struct net *net = dev_net(in_dev->dev);
1220         __be32 multiaddr = im->multiaddr;
1221
1222         spin_lock_bh(&in_dev->mc_tomb_lock);
1223         pmc_prev = NULL;
1224         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1225                 if (pmc->multiaddr == multiaddr)
1226                         break;
1227                 pmc_prev = pmc;
1228         }
1229         if (pmc) {
1230                 if (pmc_prev)
1231                         pmc_prev->next = pmc->next;
1232                 else
1233                         in_dev->mc_tomb = pmc->next;
1234         }
1235         spin_unlock_bh(&in_dev->mc_tomb_lock);
1236
1237         spin_lock_bh(&im->lock);
1238         if (pmc) {
1239                 im->interface = pmc->interface;
1240                 if (im->sfmode == MCAST_INCLUDE) {
1241                         swap(im->tomb, pmc->tomb);
1242                         swap(im->sources, pmc->sources);
1243                         for (psf = im->sources; psf; psf = psf->sf_next)
1244                                 psf->sf_crcount = in_dev->mr_qrv ?:
1245                                         READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1246                 } else {
1247                         im->crcount = in_dev->mr_qrv ?:
1248                                 READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1249                 }
1250                 in_dev_put(pmc->interface);
1251                 kfree_pmc(pmc);
1252         }
1253         spin_unlock_bh(&im->lock);
1254 }
1255
1256 /*
1257  * flush ip_mc_list deleted records
1258  */
1259 static void igmpv3_clear_delrec(struct in_device *in_dev)
1260 {
1261         struct ip_mc_list *pmc, *nextpmc;
1262
1263         spin_lock_bh(&in_dev->mc_tomb_lock);
1264         pmc = in_dev->mc_tomb;
1265         in_dev->mc_tomb = NULL;
1266         spin_unlock_bh(&in_dev->mc_tomb_lock);
1267
1268         for (; pmc; pmc = nextpmc) {
1269                 nextpmc = pmc->next;
1270                 ip_mc_clear_src(pmc);
1271                 in_dev_put(pmc->interface);
1272                 kfree_pmc(pmc);
1273         }
1274         /* clear dead sources, too */
1275         rcu_read_lock();
1276         for_each_pmc_rcu(in_dev, pmc) {
1277                 struct ip_sf_list *psf;
1278
1279                 spin_lock_bh(&pmc->lock);
1280                 psf = pmc->tomb;
1281                 pmc->tomb = NULL;
1282                 spin_unlock_bh(&pmc->lock);
1283                 ip_sf_list_clear_all(psf);
1284         }
1285         rcu_read_unlock();
1286 }
1287 #endif
1288
1289 static void __igmp_group_dropped(struct ip_mc_list *im, gfp_t gfp)
1290 {
1291         struct in_device *in_dev = im->interface;
1292 #ifdef CONFIG_IP_MULTICAST
1293         struct net *net = dev_net(in_dev->dev);
1294         int reporter;
1295 #endif
1296
1297         if (im->loaded) {
1298                 im->loaded = 0;
1299                 ip_mc_filter_del(in_dev, im->multiaddr);
1300         }
1301
1302 #ifdef CONFIG_IP_MULTICAST
1303         if (im->multiaddr == IGMP_ALL_HOSTS)
1304                 return;
1305         if (ipv4_is_local_multicast(im->multiaddr) &&
1306             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1307                 return;
1308
1309         reporter = im->reporter;
1310         igmp_stop_timer(im);
1311
1312         if (!in_dev->dead) {
1313                 if (IGMP_V1_SEEN(in_dev))
1314                         return;
1315                 if (IGMP_V2_SEEN(in_dev)) {
1316                         if (reporter)
1317                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1318                         return;
1319                 }
1320                 /* IGMPv3 */
1321                 igmpv3_add_delrec(in_dev, im, gfp);
1322
1323                 igmp_ifc_event(in_dev);
1324         }
1325 #endif
1326 }
1327
1328 static void igmp_group_dropped(struct ip_mc_list *im)
1329 {
1330         __igmp_group_dropped(im, GFP_KERNEL);
1331 }
1332
1333 static void igmp_group_added(struct ip_mc_list *im)
1334 {
1335         struct in_device *in_dev = im->interface;
1336 #ifdef CONFIG_IP_MULTICAST
1337         struct net *net = dev_net(in_dev->dev);
1338 #endif
1339
1340         if (im->loaded == 0) {
1341                 im->loaded = 1;
1342                 ip_mc_filter_add(in_dev, im->multiaddr);
1343         }
1344
1345 #ifdef CONFIG_IP_MULTICAST
1346         if (im->multiaddr == IGMP_ALL_HOSTS)
1347                 return;
1348         if (ipv4_is_local_multicast(im->multiaddr) &&
1349             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1350                 return;
1351
1352         if (in_dev->dead)
1353                 return;
1354
1355         im->unsolicit_count = READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1356         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1357                 spin_lock_bh(&im->lock);
1358                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1359                 spin_unlock_bh(&im->lock);
1360                 return;
1361         }
1362         /* else, v3 */
1363
1364         /* Based on RFC3376 5.1, for newly added INCLUDE SSM, we should
1365          * not send filter-mode change record as the mode should be from
1366          * IN() to IN(A).
1367          */
1368         if (im->sfmode == MCAST_EXCLUDE)
1369                 im->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1370
1371         igmp_ifc_event(in_dev);
1372 #endif
1373 }
1374
1375
1376 /*
1377  *      Multicast list managers
1378  */
1379
1380 static u32 ip_mc_hash(const struct ip_mc_list *im)
1381 {
1382         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1383 }
1384
1385 static void ip_mc_hash_add(struct in_device *in_dev,
1386                            struct ip_mc_list *im)
1387 {
1388         struct ip_mc_list __rcu **mc_hash;
1389         u32 hash;
1390
1391         mc_hash = rtnl_dereference(in_dev->mc_hash);
1392         if (mc_hash) {
1393                 hash = ip_mc_hash(im);
1394                 im->next_hash = mc_hash[hash];
1395                 rcu_assign_pointer(mc_hash[hash], im);
1396                 return;
1397         }
1398
1399         /* do not use a hash table for small number of items */
1400         if (in_dev->mc_count < 4)
1401                 return;
1402
1403         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1404                           GFP_KERNEL);
1405         if (!mc_hash)
1406                 return;
1407
1408         for_each_pmc_rtnl(in_dev, im) {
1409                 hash = ip_mc_hash(im);
1410                 im->next_hash = mc_hash[hash];
1411                 RCU_INIT_POINTER(mc_hash[hash], im);
1412         }
1413
1414         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1415 }
1416
1417 static void ip_mc_hash_remove(struct in_device *in_dev,
1418                               struct ip_mc_list *im)
1419 {
1420         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1421         struct ip_mc_list *aux;
1422
1423         if (!mc_hash)
1424                 return;
1425         mc_hash += ip_mc_hash(im);
1426         while ((aux = rtnl_dereference(*mc_hash)) != im)
1427                 mc_hash = &aux->next_hash;
1428         *mc_hash = im->next_hash;
1429 }
1430
1431
1432 /*
1433  *      A socket has joined a multicast group on device dev.
1434  */
1435 static void ____ip_mc_inc_group(struct in_device *in_dev, __be32 addr,
1436                                 unsigned int mode, gfp_t gfp)
1437 {
1438         struct ip_mc_list *im;
1439
1440         ASSERT_RTNL();
1441
1442         for_each_pmc_rtnl(in_dev, im) {
1443                 if (im->multiaddr == addr) {
1444                         im->users++;
1445                         ip_mc_add_src(in_dev, &addr, mode, 0, NULL, 0);
1446                         goto out;
1447                 }
1448         }
1449
1450         im = kzalloc(sizeof(*im), gfp);
1451         if (!im)
1452                 goto out;
1453
1454         im->users = 1;
1455         im->interface = in_dev;
1456         in_dev_hold(in_dev);
1457         im->multiaddr = addr;
1458         /* initial mode is (EX, empty) */
1459         im->sfmode = mode;
1460         im->sfcount[mode] = 1;
1461         refcount_set(&im->refcnt, 1);
1462         spin_lock_init(&im->lock);
1463 #ifdef CONFIG_IP_MULTICAST
1464         timer_setup(&im->timer, igmp_timer_expire, 0);
1465 #endif
1466
1467         im->next_rcu = in_dev->mc_list;
1468         in_dev->mc_count++;
1469         rcu_assign_pointer(in_dev->mc_list, im);
1470
1471         ip_mc_hash_add(in_dev, im);
1472
1473 #ifdef CONFIG_IP_MULTICAST
1474         igmpv3_del_delrec(in_dev, im);
1475 #endif
1476         igmp_group_added(im);
1477         if (!in_dev->dead)
1478                 ip_rt_multicast_event(in_dev);
1479 out:
1480         return;
1481 }
1482
1483 void __ip_mc_inc_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1484 {
1485         ____ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE, gfp);
1486 }
1487 EXPORT_SYMBOL(__ip_mc_inc_group);
1488
1489 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1490 {
1491         __ip_mc_inc_group(in_dev, addr, GFP_KERNEL);
1492 }
1493 EXPORT_SYMBOL(ip_mc_inc_group);
1494
1495 static int ip_mc_check_iphdr(struct sk_buff *skb)
1496 {
1497         const struct iphdr *iph;
1498         unsigned int len;
1499         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1500
1501         if (!pskb_may_pull(skb, offset))
1502                 return -EINVAL;
1503
1504         iph = ip_hdr(skb);
1505
1506         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1507                 return -EINVAL;
1508
1509         offset += ip_hdrlen(skb) - sizeof(*iph);
1510
1511         if (!pskb_may_pull(skb, offset))
1512                 return -EINVAL;
1513
1514         iph = ip_hdr(skb);
1515
1516         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1517                 return -EINVAL;
1518
1519         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1520         if (skb->len < len || len < offset)
1521                 return -EINVAL;
1522
1523         skb_set_transport_header(skb, offset);
1524
1525         return 0;
1526 }
1527
1528 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1529 {
1530         unsigned int len = skb_transport_offset(skb);
1531
1532         len += sizeof(struct igmpv3_report);
1533
1534         return ip_mc_may_pull(skb, len) ? 0 : -EINVAL;
1535 }
1536
1537 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1538 {
1539         unsigned int transport_len = ip_transport_len(skb);
1540         unsigned int len;
1541
1542         /* IGMPv{1,2}? */
1543         if (transport_len != sizeof(struct igmphdr)) {
1544                 /* or IGMPv3? */
1545                 if (transport_len < sizeof(struct igmpv3_query))
1546                         return -EINVAL;
1547
1548                 len = skb_transport_offset(skb) + sizeof(struct igmpv3_query);
1549                 if (!ip_mc_may_pull(skb, len))
1550                         return -EINVAL;
1551         }
1552
1553         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1554          * all-systems destination addresses (224.0.0.1) for general queries
1555          */
1556         if (!igmp_hdr(skb)->group &&
1557             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1558                 return -EINVAL;
1559
1560         return 0;
1561 }
1562
1563 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1564 {
1565         switch (igmp_hdr(skb)->type) {
1566         case IGMP_HOST_LEAVE_MESSAGE:
1567         case IGMP_HOST_MEMBERSHIP_REPORT:
1568         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1569                 return 0;
1570         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1571                 return ip_mc_check_igmp_reportv3(skb);
1572         case IGMP_HOST_MEMBERSHIP_QUERY:
1573                 return ip_mc_check_igmp_query(skb);
1574         default:
1575                 return -ENOMSG;
1576         }
1577 }
1578
1579 static __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1580 {
1581         return skb_checksum_simple_validate(skb);
1582 }
1583
1584 static int ip_mc_check_igmp_csum(struct sk_buff *skb)
1585 {
1586         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1587         unsigned int transport_len = ip_transport_len(skb);
1588         struct sk_buff *skb_chk;
1589
1590         if (!ip_mc_may_pull(skb, len))
1591                 return -EINVAL;
1592
1593         skb_chk = skb_checksum_trimmed(skb, transport_len,
1594                                        ip_mc_validate_checksum);
1595         if (!skb_chk)
1596                 return -EINVAL;
1597
1598         if (skb_chk != skb)
1599                 kfree_skb(skb_chk);
1600
1601         return 0;
1602 }
1603
1604 /**
1605  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1606  * @skb: the skb to validate
1607  *
1608  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1609  * skb transport header accordingly and returns zero.
1610  *
1611  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1612  *  standard
1613  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1614  * -ENOMEM: A memory allocation failure happened.
1615  *
1616  * Caller needs to set the skb network header and free any returned skb if it
1617  * differs from the provided skb.
1618  */
1619 int ip_mc_check_igmp(struct sk_buff *skb)
1620 {
1621         int ret = ip_mc_check_iphdr(skb);
1622
1623         if (ret < 0)
1624                 return ret;
1625
1626         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1627                 return -ENOMSG;
1628
1629         ret = ip_mc_check_igmp_csum(skb);
1630         if (ret < 0)
1631                 return ret;
1632
1633         return ip_mc_check_igmp_msg(skb);
1634 }
1635 EXPORT_SYMBOL(ip_mc_check_igmp);
1636
1637 /*
1638  *      Resend IGMP JOIN report; used by netdev notifier.
1639  */
1640 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1641 {
1642 #ifdef CONFIG_IP_MULTICAST
1643         struct ip_mc_list *im;
1644         int type;
1645         struct net *net = dev_net(in_dev->dev);
1646
1647         ASSERT_RTNL();
1648
1649         for_each_pmc_rtnl(in_dev, im) {
1650                 if (im->multiaddr == IGMP_ALL_HOSTS)
1651                         continue;
1652                 if (ipv4_is_local_multicast(im->multiaddr) &&
1653                     !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1654                         continue;
1655
1656                 /* a failover is happening and switches
1657                  * must be notified immediately
1658                  */
1659                 if (IGMP_V1_SEEN(in_dev))
1660                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1661                 else if (IGMP_V2_SEEN(in_dev))
1662                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1663                 else
1664                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1665                 igmp_send_report(in_dev, im, type);
1666         }
1667 #endif
1668 }
1669
1670 /*
1671  *      A socket has left a multicast group on device dev
1672  */
1673
1674 void __ip_mc_dec_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1675 {
1676         struct ip_mc_list *i;
1677         struct ip_mc_list __rcu **ip;
1678
1679         ASSERT_RTNL();
1680
1681         for (ip = &in_dev->mc_list;
1682              (i = rtnl_dereference(*ip)) != NULL;
1683              ip = &i->next_rcu) {
1684                 if (i->multiaddr == addr) {
1685                         if (--i->users == 0) {
1686                                 ip_mc_hash_remove(in_dev, i);
1687                                 *ip = i->next_rcu;
1688                                 in_dev->mc_count--;
1689                                 __igmp_group_dropped(i, gfp);
1690                                 ip_mc_clear_src(i);
1691
1692                                 if (!in_dev->dead)
1693                                         ip_rt_multicast_event(in_dev);
1694
1695                                 ip_ma_put(i);
1696                                 return;
1697                         }
1698                         break;
1699                 }
1700         }
1701 }
1702 EXPORT_SYMBOL(__ip_mc_dec_group);
1703
1704 /* Device changing type */
1705
1706 void ip_mc_unmap(struct in_device *in_dev)
1707 {
1708         struct ip_mc_list *pmc;
1709
1710         ASSERT_RTNL();
1711
1712         for_each_pmc_rtnl(in_dev, pmc)
1713                 igmp_group_dropped(pmc);
1714 }
1715
1716 void ip_mc_remap(struct in_device *in_dev)
1717 {
1718         struct ip_mc_list *pmc;
1719
1720         ASSERT_RTNL();
1721
1722         for_each_pmc_rtnl(in_dev, pmc) {
1723 #ifdef CONFIG_IP_MULTICAST
1724                 igmpv3_del_delrec(in_dev, pmc);
1725 #endif
1726                 igmp_group_added(pmc);
1727         }
1728 }
1729
1730 /* Device going down */
1731
1732 void ip_mc_down(struct in_device *in_dev)
1733 {
1734         struct ip_mc_list *pmc;
1735
1736         ASSERT_RTNL();
1737
1738         for_each_pmc_rtnl(in_dev, pmc)
1739                 igmp_group_dropped(pmc);
1740
1741 #ifdef CONFIG_IP_MULTICAST
1742         WRITE_ONCE(in_dev->mr_ifc_count, 0);
1743         if (del_timer(&in_dev->mr_ifc_timer))
1744                 __in_dev_put(in_dev);
1745         in_dev->mr_gq_running = 0;
1746         if (del_timer(&in_dev->mr_gq_timer))
1747                 __in_dev_put(in_dev);
1748 #endif
1749
1750         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1751 }
1752
1753 #ifdef CONFIG_IP_MULTICAST
1754 static void ip_mc_reset(struct in_device *in_dev)
1755 {
1756         struct net *net = dev_net(in_dev->dev);
1757
1758         in_dev->mr_qi = IGMP_QUERY_INTERVAL;
1759         in_dev->mr_qri = IGMP_QUERY_RESPONSE_INTERVAL;
1760         in_dev->mr_qrv = READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1761 }
1762 #else
1763 static void ip_mc_reset(struct in_device *in_dev)
1764 {
1765 }
1766 #endif
1767
1768 void ip_mc_init_dev(struct in_device *in_dev)
1769 {
1770         ASSERT_RTNL();
1771
1772 #ifdef CONFIG_IP_MULTICAST
1773         timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1774         timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1775 #endif
1776         ip_mc_reset(in_dev);
1777
1778         spin_lock_init(&in_dev->mc_tomb_lock);
1779 }
1780
1781 /* Device going up */
1782
1783 void ip_mc_up(struct in_device *in_dev)
1784 {
1785         struct ip_mc_list *pmc;
1786
1787         ASSERT_RTNL();
1788
1789         ip_mc_reset(in_dev);
1790         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1791
1792         for_each_pmc_rtnl(in_dev, pmc) {
1793 #ifdef CONFIG_IP_MULTICAST
1794                 igmpv3_del_delrec(in_dev, pmc);
1795 #endif
1796                 igmp_group_added(pmc);
1797         }
1798 }
1799
1800 /*
1801  *      Device is about to be destroyed: clean up.
1802  */
1803
1804 void ip_mc_destroy_dev(struct in_device *in_dev)
1805 {
1806         struct ip_mc_list *i;
1807
1808         ASSERT_RTNL();
1809
1810         /* Deactivate timers */
1811         ip_mc_down(in_dev);
1812 #ifdef CONFIG_IP_MULTICAST
1813         igmpv3_clear_delrec(in_dev);
1814 #endif
1815
1816         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1817                 in_dev->mc_list = i->next_rcu;
1818                 in_dev->mc_count--;
1819                 ip_mc_clear_src(i);
1820                 ip_ma_put(i);
1821         }
1822 }
1823
1824 /* RTNL is locked */
1825 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1826 {
1827         struct net_device *dev = NULL;
1828         struct in_device *idev = NULL;
1829
1830         if (imr->imr_ifindex) {
1831                 idev = inetdev_by_index(net, imr->imr_ifindex);
1832                 return idev;
1833         }
1834         if (imr->imr_address.s_addr) {
1835                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1836                 if (!dev)
1837                         return NULL;
1838         }
1839
1840         if (!dev) {
1841                 struct rtable *rt = ip_route_output(net,
1842                                                     imr->imr_multiaddr.s_addr,
1843                                                     0, 0, 0);
1844                 if (!IS_ERR(rt)) {
1845                         dev = rt->dst.dev;
1846                         ip_rt_put(rt);
1847                 }
1848         }
1849         if (dev) {
1850                 imr->imr_ifindex = dev->ifindex;
1851                 idev = __in_dev_get_rtnl(dev);
1852         }
1853         return idev;
1854 }
1855
1856 /*
1857  *      Join a socket to a group
1858  */
1859
1860 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1861         __be32 *psfsrc)
1862 {
1863         struct ip_sf_list *psf, *psf_prev;
1864         int rv = 0;
1865
1866         psf_prev = NULL;
1867         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1868                 if (psf->sf_inaddr == *psfsrc)
1869                         break;
1870                 psf_prev = psf;
1871         }
1872         if (!psf || psf->sf_count[sfmode] == 0) {
1873                 /* source filter not found, or count wrong =>  bug */
1874                 return -ESRCH;
1875         }
1876         psf->sf_count[sfmode]--;
1877         if (psf->sf_count[sfmode] == 0) {
1878                 ip_rt_multicast_event(pmc->interface);
1879         }
1880         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1881 #ifdef CONFIG_IP_MULTICAST
1882                 struct in_device *in_dev = pmc->interface;
1883                 struct net *net = dev_net(in_dev->dev);
1884 #endif
1885
1886                 /* no more filters for this source */
1887                 if (psf_prev)
1888                         psf_prev->sf_next = psf->sf_next;
1889                 else
1890                         pmc->sources = psf->sf_next;
1891 #ifdef CONFIG_IP_MULTICAST
1892                 if (psf->sf_oldin &&
1893                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1894                         psf->sf_crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1895                         psf->sf_next = pmc->tomb;
1896                         pmc->tomb = psf;
1897                         rv = 1;
1898                 } else
1899 #endif
1900                         kfree(psf);
1901         }
1902         return rv;
1903 }
1904
1905 #ifndef CONFIG_IP_MULTICAST
1906 #define igmp_ifc_event(x)       do { } while (0)
1907 #endif
1908
1909 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1910                          int sfcount, __be32 *psfsrc, int delta)
1911 {
1912         struct ip_mc_list *pmc;
1913         int     changerec = 0;
1914         int     i, err;
1915
1916         if (!in_dev)
1917                 return -ENODEV;
1918         rcu_read_lock();
1919         for_each_pmc_rcu(in_dev, pmc) {
1920                 if (*pmca == pmc->multiaddr)
1921                         break;
1922         }
1923         if (!pmc) {
1924                 /* MCA not found?? bug */
1925                 rcu_read_unlock();
1926                 return -ESRCH;
1927         }
1928         spin_lock_bh(&pmc->lock);
1929         rcu_read_unlock();
1930 #ifdef CONFIG_IP_MULTICAST
1931         sf_markstate(pmc);
1932 #endif
1933         if (!delta) {
1934                 err = -EINVAL;
1935                 if (!pmc->sfcount[sfmode])
1936                         goto out_unlock;
1937                 pmc->sfcount[sfmode]--;
1938         }
1939         err = 0;
1940         for (i = 0; i < sfcount; i++) {
1941                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1942
1943                 changerec |= rv > 0;
1944                 if (!err && rv < 0)
1945                         err = rv;
1946         }
1947         if (pmc->sfmode == MCAST_EXCLUDE &&
1948             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1949             pmc->sfcount[MCAST_INCLUDE]) {
1950 #ifdef CONFIG_IP_MULTICAST
1951                 struct ip_sf_list *psf;
1952                 struct net *net = dev_net(in_dev->dev);
1953 #endif
1954
1955                 /* filter mode change */
1956                 pmc->sfmode = MCAST_INCLUDE;
1957 #ifdef CONFIG_IP_MULTICAST
1958                 pmc->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1959                 WRITE_ONCE(in_dev->mr_ifc_count, pmc->crcount);
1960                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1961                         psf->sf_crcount = 0;
1962                 igmp_ifc_event(pmc->interface);
1963         } else if (sf_setstate(pmc) || changerec) {
1964                 igmp_ifc_event(pmc->interface);
1965 #endif
1966         }
1967 out_unlock:
1968         spin_unlock_bh(&pmc->lock);
1969         return err;
1970 }
1971
1972 /*
1973  * Add multicast single-source filter to the interface list
1974  */
1975 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1976         __be32 *psfsrc)
1977 {
1978         struct ip_sf_list *psf, *psf_prev;
1979
1980         psf_prev = NULL;
1981         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1982                 if (psf->sf_inaddr == *psfsrc)
1983                         break;
1984                 psf_prev = psf;
1985         }
1986         if (!psf) {
1987                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1988                 if (!psf)
1989                         return -ENOBUFS;
1990                 psf->sf_inaddr = *psfsrc;
1991                 if (psf_prev) {
1992                         psf_prev->sf_next = psf;
1993                 } else
1994                         pmc->sources = psf;
1995         }
1996         psf->sf_count[sfmode]++;
1997         if (psf->sf_count[sfmode] == 1) {
1998                 ip_rt_multicast_event(pmc->interface);
1999         }
2000         return 0;
2001 }
2002
2003 #ifdef CONFIG_IP_MULTICAST
2004 static void sf_markstate(struct ip_mc_list *pmc)
2005 {
2006         struct ip_sf_list *psf;
2007         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2008
2009         for (psf = pmc->sources; psf; psf = psf->sf_next)
2010                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2011                         psf->sf_oldin = mca_xcount ==
2012                                 psf->sf_count[MCAST_EXCLUDE] &&
2013                                 !psf->sf_count[MCAST_INCLUDE];
2014                 } else
2015                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
2016 }
2017
2018 static int sf_setstate(struct ip_mc_list *pmc)
2019 {
2020         struct ip_sf_list *psf, *dpsf;
2021         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2022         int qrv = pmc->interface->mr_qrv;
2023         int new_in, rv;
2024
2025         rv = 0;
2026         for (psf = pmc->sources; psf; psf = psf->sf_next) {
2027                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2028                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
2029                                 !psf->sf_count[MCAST_INCLUDE];
2030                 } else
2031                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
2032                 if (new_in) {
2033                         if (!psf->sf_oldin) {
2034                                 struct ip_sf_list *prev = NULL;
2035
2036                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2037                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2038                                                 break;
2039                                         prev = dpsf;
2040                                 }
2041                                 if (dpsf) {
2042                                         if (prev)
2043                                                 prev->sf_next = dpsf->sf_next;
2044                                         else
2045                                                 pmc->tomb = dpsf->sf_next;
2046                                         kfree(dpsf);
2047                                 }
2048                                 psf->sf_crcount = qrv;
2049                                 rv++;
2050                         }
2051                 } else if (psf->sf_oldin) {
2052
2053                         psf->sf_crcount = 0;
2054                         /*
2055                          * add or update "delete" records if an active filter
2056                          * is now inactive
2057                          */
2058                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2059                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2060                                         break;
2061                         if (!dpsf) {
2062                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2063                                 if (!dpsf)
2064                                         continue;
2065                                 *dpsf = *psf;
2066                                 /* pmc->lock held by callers */
2067                                 dpsf->sf_next = pmc->tomb;
2068                                 pmc->tomb = dpsf;
2069                         }
2070                         dpsf->sf_crcount = qrv;
2071                         rv++;
2072                 }
2073         }
2074         return rv;
2075 }
2076 #endif
2077
2078 /*
2079  * Add multicast source filter list to the interface list
2080  */
2081 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2082                          int sfcount, __be32 *psfsrc, int delta)
2083 {
2084         struct ip_mc_list *pmc;
2085         int     isexclude;
2086         int     i, err;
2087
2088         if (!in_dev)
2089                 return -ENODEV;
2090         rcu_read_lock();
2091         for_each_pmc_rcu(in_dev, pmc) {
2092                 if (*pmca == pmc->multiaddr)
2093                         break;
2094         }
2095         if (!pmc) {
2096                 /* MCA not found?? bug */
2097                 rcu_read_unlock();
2098                 return -ESRCH;
2099         }
2100         spin_lock_bh(&pmc->lock);
2101         rcu_read_unlock();
2102
2103 #ifdef CONFIG_IP_MULTICAST
2104         sf_markstate(pmc);
2105 #endif
2106         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2107         if (!delta)
2108                 pmc->sfcount[sfmode]++;
2109         err = 0;
2110         for (i = 0; i < sfcount; i++) {
2111                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2112                 if (err)
2113                         break;
2114         }
2115         if (err) {
2116                 int j;
2117
2118                 if (!delta)
2119                         pmc->sfcount[sfmode]--;
2120                 for (j = 0; j < i; j++)
2121                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2122         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2123 #ifdef CONFIG_IP_MULTICAST
2124                 struct ip_sf_list *psf;
2125                 struct net *net = dev_net(pmc->interface->dev);
2126                 in_dev = pmc->interface;
2127 #endif
2128
2129                 /* filter mode change */
2130                 if (pmc->sfcount[MCAST_EXCLUDE])
2131                         pmc->sfmode = MCAST_EXCLUDE;
2132                 else if (pmc->sfcount[MCAST_INCLUDE])
2133                         pmc->sfmode = MCAST_INCLUDE;
2134 #ifdef CONFIG_IP_MULTICAST
2135                 /* else no filters; keep old mode for reports */
2136
2137                 pmc->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
2138                 WRITE_ONCE(in_dev->mr_ifc_count, pmc->crcount);
2139                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2140                         psf->sf_crcount = 0;
2141                 igmp_ifc_event(in_dev);
2142         } else if (sf_setstate(pmc)) {
2143                 igmp_ifc_event(in_dev);
2144 #endif
2145         }
2146         spin_unlock_bh(&pmc->lock);
2147         return err;
2148 }
2149
2150 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2151 {
2152         struct ip_sf_list *tomb, *sources;
2153
2154         spin_lock_bh(&pmc->lock);
2155         tomb = pmc->tomb;
2156         pmc->tomb = NULL;
2157         sources = pmc->sources;
2158         pmc->sources = NULL;
2159         pmc->sfmode = MCAST_EXCLUDE;
2160         pmc->sfcount[MCAST_INCLUDE] = 0;
2161         pmc->sfcount[MCAST_EXCLUDE] = 1;
2162         spin_unlock_bh(&pmc->lock);
2163
2164         ip_sf_list_clear_all(tomb);
2165         ip_sf_list_clear_all(sources);
2166 }
2167
2168 /* Join a multicast group
2169  */
2170 static int __ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr,
2171                               unsigned int mode)
2172 {
2173         __be32 addr = imr->imr_multiaddr.s_addr;
2174         struct ip_mc_socklist *iml, *i;
2175         struct in_device *in_dev;
2176         struct inet_sock *inet = inet_sk(sk);
2177         struct net *net = sock_net(sk);
2178         int ifindex;
2179         int count = 0;
2180         int err;
2181
2182         ASSERT_RTNL();
2183
2184         if (!ipv4_is_multicast(addr))
2185                 return -EINVAL;
2186
2187         in_dev = ip_mc_find_dev(net, imr);
2188
2189         if (!in_dev) {
2190                 err = -ENODEV;
2191                 goto done;
2192         }
2193
2194         err = -EADDRINUSE;
2195         ifindex = imr->imr_ifindex;
2196         for_each_pmc_rtnl(inet, i) {
2197                 if (i->multi.imr_multiaddr.s_addr == addr &&
2198                     i->multi.imr_ifindex == ifindex)
2199                         goto done;
2200                 count++;
2201         }
2202         err = -ENOBUFS;
2203         if (count >= READ_ONCE(net->ipv4.sysctl_igmp_max_memberships))
2204                 goto done;
2205         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2206         if (!iml)
2207                 goto done;
2208
2209         memcpy(&iml->multi, imr, sizeof(*imr));
2210         iml->next_rcu = inet->mc_list;
2211         iml->sflist = NULL;
2212         iml->sfmode = mode;
2213         rcu_assign_pointer(inet->mc_list, iml);
2214         ____ip_mc_inc_group(in_dev, addr, mode, GFP_KERNEL);
2215         err = 0;
2216 done:
2217         return err;
2218 }
2219
2220 /* Join ASM (Any-Source Multicast) group
2221  */
2222 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2223 {
2224         return __ip_mc_join_group(sk, imr, MCAST_EXCLUDE);
2225 }
2226 EXPORT_SYMBOL(ip_mc_join_group);
2227
2228 /* Join SSM (Source-Specific Multicast) group
2229  */
2230 int ip_mc_join_group_ssm(struct sock *sk, struct ip_mreqn *imr,
2231                          unsigned int mode)
2232 {
2233         return __ip_mc_join_group(sk, imr, mode);
2234 }
2235
2236 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2237                            struct in_device *in_dev)
2238 {
2239         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2240         int err;
2241
2242         if (!psf) {
2243                 /* any-source empty exclude case */
2244                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2245                         iml->sfmode, 0, NULL, 0);
2246         }
2247         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2248                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2249         RCU_INIT_POINTER(iml->sflist, NULL);
2250         /* decrease mem now to avoid the memleak warning */
2251         atomic_sub(struct_size(psf, sl_addr, psf->sl_max), &sk->sk_omem_alloc);
2252         kfree_rcu(psf, rcu);
2253         return err;
2254 }
2255
2256 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2257 {
2258         struct inet_sock *inet = inet_sk(sk);
2259         struct ip_mc_socklist *iml;
2260         struct ip_mc_socklist __rcu **imlp;
2261         struct in_device *in_dev;
2262         struct net *net = sock_net(sk);
2263         __be32 group = imr->imr_multiaddr.s_addr;
2264         u32 ifindex;
2265         int ret = -EADDRNOTAVAIL;
2266
2267         ASSERT_RTNL();
2268
2269         in_dev = ip_mc_find_dev(net, imr);
2270         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2271                 ret = -ENODEV;
2272                 goto out;
2273         }
2274         ifindex = imr->imr_ifindex;
2275         for (imlp = &inet->mc_list;
2276              (iml = rtnl_dereference(*imlp)) != NULL;
2277              imlp = &iml->next_rcu) {
2278                 if (iml->multi.imr_multiaddr.s_addr != group)
2279                         continue;
2280                 if (ifindex) {
2281                         if (iml->multi.imr_ifindex != ifindex)
2282                                 continue;
2283                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2284                                 iml->multi.imr_address.s_addr)
2285                         continue;
2286
2287                 (void) ip_mc_leave_src(sk, iml, in_dev);
2288
2289                 *imlp = iml->next_rcu;
2290
2291                 if (in_dev)
2292                         ip_mc_dec_group(in_dev, group);
2293
2294                 /* decrease mem now to avoid the memleak warning */
2295                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2296                 kfree_rcu(iml, rcu);
2297                 return 0;
2298         }
2299 out:
2300         return ret;
2301 }
2302 EXPORT_SYMBOL(ip_mc_leave_group);
2303
2304 int ip_mc_source(int add, int omode, struct sock *sk, struct
2305         ip_mreq_source *mreqs, int ifindex)
2306 {
2307         int err;
2308         struct ip_mreqn imr;
2309         __be32 addr = mreqs->imr_multiaddr;
2310         struct ip_mc_socklist *pmc;
2311         struct in_device *in_dev = NULL;
2312         struct inet_sock *inet = inet_sk(sk);
2313         struct ip_sf_socklist *psl;
2314         struct net *net = sock_net(sk);
2315         int leavegroup = 0;
2316         int i, j, rv;
2317
2318         if (!ipv4_is_multicast(addr))
2319                 return -EINVAL;
2320
2321         ASSERT_RTNL();
2322
2323         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2324         imr.imr_address.s_addr = mreqs->imr_interface;
2325         imr.imr_ifindex = ifindex;
2326         in_dev = ip_mc_find_dev(net, &imr);
2327
2328         if (!in_dev) {
2329                 err = -ENODEV;
2330                 goto done;
2331         }
2332         err = -EADDRNOTAVAIL;
2333
2334         for_each_pmc_rtnl(inet, pmc) {
2335                 if ((pmc->multi.imr_multiaddr.s_addr ==
2336                      imr.imr_multiaddr.s_addr) &&
2337                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2338                         break;
2339         }
2340         if (!pmc) {             /* must have a prior join */
2341                 err = -EINVAL;
2342                 goto done;
2343         }
2344         /* if a source filter was set, must be the same mode as before */
2345         if (pmc->sflist) {
2346                 if (pmc->sfmode != omode) {
2347                         err = -EINVAL;
2348                         goto done;
2349                 }
2350         } else if (pmc->sfmode != omode) {
2351                 /* allow mode switches for empty-set filters */
2352                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2353                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2354                         NULL, 0);
2355                 pmc->sfmode = omode;
2356         }
2357
2358         psl = rtnl_dereference(pmc->sflist);
2359         if (!add) {
2360                 if (!psl)
2361                         goto done;      /* err = -EADDRNOTAVAIL */
2362                 rv = !0;
2363                 for (i = 0; i < psl->sl_count; i++) {
2364                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2365                                 sizeof(__be32));
2366                         if (rv == 0)
2367                                 break;
2368                 }
2369                 if (rv)         /* source not found */
2370                         goto done;      /* err = -EADDRNOTAVAIL */
2371
2372                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2373                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2374                         leavegroup = 1;
2375                         goto done;
2376                 }
2377
2378                 /* update the interface filter */
2379                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2380                         &mreqs->imr_sourceaddr, 1);
2381
2382                 for (j = i+1; j < psl->sl_count; j++)
2383                         psl->sl_addr[j-1] = psl->sl_addr[j];
2384                 psl->sl_count--;
2385                 err = 0;
2386                 goto done;
2387         }
2388         /* else, add a new source to the filter */
2389
2390         if (psl && psl->sl_count >= READ_ONCE(net->ipv4.sysctl_igmp_max_msf)) {
2391                 err = -ENOBUFS;
2392                 goto done;
2393         }
2394         if (!psl || psl->sl_count == psl->sl_max) {
2395                 struct ip_sf_socklist *newpsl;
2396                 int count = IP_SFBLOCK;
2397
2398                 if (psl)
2399                         count += psl->sl_max;
2400                 newpsl = sock_kmalloc(sk, struct_size(newpsl, sl_addr, count),
2401                                       GFP_KERNEL);
2402                 if (!newpsl) {
2403                         err = -ENOBUFS;
2404                         goto done;
2405                 }
2406                 newpsl->sl_max = count;
2407                 newpsl->sl_count = count - IP_SFBLOCK;
2408                 if (psl) {
2409                         for (i = 0; i < psl->sl_count; i++)
2410                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2411                         /* decrease mem now to avoid the memleak warning */
2412                         atomic_sub(struct_size(psl, sl_addr, psl->sl_max),
2413                                    &sk->sk_omem_alloc);
2414                 }
2415                 rcu_assign_pointer(pmc->sflist, newpsl);
2416                 if (psl)
2417                         kfree_rcu(psl, rcu);
2418                 psl = newpsl;
2419         }
2420         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2421         for (i = 0; i < psl->sl_count; i++) {
2422                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2423                         sizeof(__be32));
2424                 if (rv == 0)
2425                         break;
2426         }
2427         if (rv == 0)            /* address already there is an error */
2428                 goto done;
2429         for (j = psl->sl_count-1; j >= i; j--)
2430                 psl->sl_addr[j+1] = psl->sl_addr[j];
2431         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2432         psl->sl_count++;
2433         err = 0;
2434         /* update the interface list */
2435         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2436                 &mreqs->imr_sourceaddr, 1);
2437 done:
2438         if (leavegroup)
2439                 err = ip_mc_leave_group(sk, &imr);
2440         return err;
2441 }
2442
2443 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2444 {
2445         int err = 0;
2446         struct ip_mreqn imr;
2447         __be32 addr = msf->imsf_multiaddr;
2448         struct ip_mc_socklist *pmc;
2449         struct in_device *in_dev;
2450         struct inet_sock *inet = inet_sk(sk);
2451         struct ip_sf_socklist *newpsl, *psl;
2452         struct net *net = sock_net(sk);
2453         int leavegroup = 0;
2454
2455         if (!ipv4_is_multicast(addr))
2456                 return -EINVAL;
2457         if (msf->imsf_fmode != MCAST_INCLUDE &&
2458             msf->imsf_fmode != MCAST_EXCLUDE)
2459                 return -EINVAL;
2460
2461         ASSERT_RTNL();
2462
2463         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2464         imr.imr_address.s_addr = msf->imsf_interface;
2465         imr.imr_ifindex = ifindex;
2466         in_dev = ip_mc_find_dev(net, &imr);
2467
2468         if (!in_dev) {
2469                 err = -ENODEV;
2470                 goto done;
2471         }
2472
2473         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2474         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2475                 leavegroup = 1;
2476                 goto done;
2477         }
2478
2479         for_each_pmc_rtnl(inet, pmc) {
2480                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2481                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2482                         break;
2483         }
2484         if (!pmc) {             /* must have a prior join */
2485                 err = -EINVAL;
2486                 goto done;
2487         }
2488         if (msf->imsf_numsrc) {
2489                 newpsl = sock_kmalloc(sk, struct_size(newpsl, sl_addr,
2490                                                       msf->imsf_numsrc),
2491                                       GFP_KERNEL);
2492                 if (!newpsl) {
2493                         err = -ENOBUFS;
2494                         goto done;
2495                 }
2496                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2497                 memcpy(newpsl->sl_addr, msf->imsf_slist_flex,
2498                        flex_array_size(msf, imsf_slist_flex, msf->imsf_numsrc));
2499                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2500                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2501                 if (err) {
2502                         sock_kfree_s(sk, newpsl,
2503                                      struct_size(newpsl, sl_addr,
2504                                                  newpsl->sl_max));
2505                         goto done;
2506                 }
2507         } else {
2508                 newpsl = NULL;
2509                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2510                                      msf->imsf_fmode, 0, NULL, 0);
2511         }
2512         psl = rtnl_dereference(pmc->sflist);
2513         if (psl) {
2514                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2515                         psl->sl_count, psl->sl_addr, 0);
2516                 /* decrease mem now to avoid the memleak warning */
2517                 atomic_sub(struct_size(psl, sl_addr, psl->sl_max),
2518                            &sk->sk_omem_alloc);
2519         } else {
2520                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2521                         0, NULL, 0);
2522         }
2523         rcu_assign_pointer(pmc->sflist, newpsl);
2524         if (psl)
2525                 kfree_rcu(psl, rcu);
2526         pmc->sfmode = msf->imsf_fmode;
2527         err = 0;
2528 done:
2529         if (leavegroup)
2530                 err = ip_mc_leave_group(sk, &imr);
2531         return err;
2532 }
2533 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2534                  sockptr_t optval, sockptr_t optlen)
2535 {
2536         int err, len, count, copycount, msf_size;
2537         struct ip_mreqn imr;
2538         __be32 addr = msf->imsf_multiaddr;
2539         struct ip_mc_socklist *pmc;
2540         struct in_device *in_dev;
2541         struct inet_sock *inet = inet_sk(sk);
2542         struct ip_sf_socklist *psl;
2543         struct net *net = sock_net(sk);
2544
2545         ASSERT_RTNL();
2546
2547         if (!ipv4_is_multicast(addr))
2548                 return -EINVAL;
2549
2550         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2551         imr.imr_address.s_addr = msf->imsf_interface;
2552         imr.imr_ifindex = 0;
2553         in_dev = ip_mc_find_dev(net, &imr);
2554
2555         if (!in_dev) {
2556                 err = -ENODEV;
2557                 goto done;
2558         }
2559         err = -EADDRNOTAVAIL;
2560
2561         for_each_pmc_rtnl(inet, pmc) {
2562                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2563                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2564                         break;
2565         }
2566         if (!pmc)               /* must have a prior join */
2567                 goto done;
2568         msf->imsf_fmode = pmc->sfmode;
2569         psl = rtnl_dereference(pmc->sflist);
2570         if (!psl) {
2571                 count = 0;
2572         } else {
2573                 count = psl->sl_count;
2574         }
2575         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2576         len = flex_array_size(psl, sl_addr, copycount);
2577         msf->imsf_numsrc = count;
2578         msf_size = IP_MSFILTER_SIZE(copycount);
2579         if (copy_to_sockptr(optlen, &msf_size, sizeof(int)) ||
2580             copy_to_sockptr(optval, msf, IP_MSFILTER_SIZE(0))) {
2581                 return -EFAULT;
2582         }
2583         if (len &&
2584             copy_to_sockptr_offset(optval,
2585                                    offsetof(struct ip_msfilter, imsf_slist_flex),
2586                                    psl->sl_addr, len))
2587                 return -EFAULT;
2588         return 0;
2589 done:
2590         return err;
2591 }
2592
2593 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2594                  sockptr_t optval, size_t ss_offset)
2595 {
2596         int i, count, copycount;
2597         struct sockaddr_in *psin;
2598         __be32 addr;
2599         struct ip_mc_socklist *pmc;
2600         struct inet_sock *inet = inet_sk(sk);
2601         struct ip_sf_socklist *psl;
2602
2603         ASSERT_RTNL();
2604
2605         psin = (struct sockaddr_in *)&gsf->gf_group;
2606         if (psin->sin_family != AF_INET)
2607                 return -EINVAL;
2608         addr = psin->sin_addr.s_addr;
2609         if (!ipv4_is_multicast(addr))
2610                 return -EINVAL;
2611
2612         for_each_pmc_rtnl(inet, pmc) {
2613                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2614                     pmc->multi.imr_ifindex == gsf->gf_interface)
2615                         break;
2616         }
2617         if (!pmc)               /* must have a prior join */
2618                 return -EADDRNOTAVAIL;
2619         gsf->gf_fmode = pmc->sfmode;
2620         psl = rtnl_dereference(pmc->sflist);
2621         count = psl ? psl->sl_count : 0;
2622         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2623         gsf->gf_numsrc = count;
2624         for (i = 0; i < copycount; i++) {
2625                 struct sockaddr_storage ss;
2626
2627                 psin = (struct sockaddr_in *)&ss;
2628                 memset(&ss, 0, sizeof(ss));
2629                 psin->sin_family = AF_INET;
2630                 psin->sin_addr.s_addr = psl->sl_addr[i];
2631                 if (copy_to_sockptr_offset(optval, ss_offset,
2632                                            &ss, sizeof(ss)))
2633                         return -EFAULT;
2634                 ss_offset += sizeof(ss);
2635         }
2636         return 0;
2637 }
2638
2639 /*
2640  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2641  */
2642 int ip_mc_sf_allow(const struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2643                    int dif, int sdif)
2644 {
2645         const struct inet_sock *inet = inet_sk(sk);
2646         struct ip_mc_socklist *pmc;
2647         struct ip_sf_socklist *psl;
2648         int i;
2649         int ret;
2650
2651         ret = 1;
2652         if (!ipv4_is_multicast(loc_addr))
2653                 goto out;
2654
2655         rcu_read_lock();
2656         for_each_pmc_rcu(inet, pmc) {
2657                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2658                     (pmc->multi.imr_ifindex == dif ||
2659                      (sdif && pmc->multi.imr_ifindex == sdif)))
2660                         break;
2661         }
2662         ret = inet_test_bit(MC_ALL, sk);
2663         if (!pmc)
2664                 goto unlock;
2665         psl = rcu_dereference(pmc->sflist);
2666         ret = (pmc->sfmode == MCAST_EXCLUDE);
2667         if (!psl)
2668                 goto unlock;
2669
2670         for (i = 0; i < psl->sl_count; i++) {
2671                 if (psl->sl_addr[i] == rmt_addr)
2672                         break;
2673         }
2674         ret = 0;
2675         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2676                 goto unlock;
2677         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2678                 goto unlock;
2679         ret = 1;
2680 unlock:
2681         rcu_read_unlock();
2682 out:
2683         return ret;
2684 }
2685
2686 /*
2687  *      A socket is closing.
2688  */
2689
2690 void ip_mc_drop_socket(struct sock *sk)
2691 {
2692         struct inet_sock *inet = inet_sk(sk);
2693         struct ip_mc_socklist *iml;
2694         struct net *net = sock_net(sk);
2695
2696         if (!inet->mc_list)
2697                 return;
2698
2699         rtnl_lock();
2700         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2701                 struct in_device *in_dev;
2702
2703                 inet->mc_list = iml->next_rcu;
2704                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2705                 (void) ip_mc_leave_src(sk, iml, in_dev);
2706                 if (in_dev)
2707                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2708                 /* decrease mem now to avoid the memleak warning */
2709                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2710                 kfree_rcu(iml, rcu);
2711         }
2712         rtnl_unlock();
2713 }
2714
2715 /* called with rcu_read_lock() */
2716 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2717 {
2718         struct ip_mc_list *im;
2719         struct ip_mc_list __rcu **mc_hash;
2720         struct ip_sf_list *psf;
2721         int rv = 0;
2722
2723         mc_hash = rcu_dereference(in_dev->mc_hash);
2724         if (mc_hash) {
2725                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2726
2727                 for (im = rcu_dereference(mc_hash[hash]);
2728                      im != NULL;
2729                      im = rcu_dereference(im->next_hash)) {
2730                         if (im->multiaddr == mc_addr)
2731                                 break;
2732                 }
2733         } else {
2734                 for_each_pmc_rcu(in_dev, im) {
2735                         if (im->multiaddr == mc_addr)
2736                                 break;
2737                 }
2738         }
2739         if (im && proto == IPPROTO_IGMP) {
2740                 rv = 1;
2741         } else if (im) {
2742                 if (src_addr) {
2743                         spin_lock_bh(&im->lock);
2744                         for (psf = im->sources; psf; psf = psf->sf_next) {
2745                                 if (psf->sf_inaddr == src_addr)
2746                                         break;
2747                         }
2748                         if (psf)
2749                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2750                                         psf->sf_count[MCAST_EXCLUDE] !=
2751                                         im->sfcount[MCAST_EXCLUDE];
2752                         else
2753                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2754                         spin_unlock_bh(&im->lock);
2755                 } else
2756                         rv = 1; /* unspecified source; tentatively allow */
2757         }
2758         return rv;
2759 }
2760
2761 #if defined(CONFIG_PROC_FS)
2762 struct igmp_mc_iter_state {
2763         struct seq_net_private p;
2764         struct net_device *dev;
2765         struct in_device *in_dev;
2766 };
2767
2768 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2769
2770 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2771 {
2772         struct net *net = seq_file_net(seq);
2773         struct ip_mc_list *im = NULL;
2774         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2775
2776         state->in_dev = NULL;
2777         for_each_netdev_rcu(net, state->dev) {
2778                 struct in_device *in_dev;
2779
2780                 in_dev = __in_dev_get_rcu(state->dev);
2781                 if (!in_dev)
2782                         continue;
2783                 im = rcu_dereference(in_dev->mc_list);
2784                 if (im) {
2785                         state->in_dev = in_dev;
2786                         break;
2787                 }
2788         }
2789         return im;
2790 }
2791
2792 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2793 {
2794         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2795
2796         im = rcu_dereference(im->next_rcu);
2797         while (!im) {
2798                 state->dev = next_net_device_rcu(state->dev);
2799                 if (!state->dev) {
2800                         state->in_dev = NULL;
2801                         break;
2802                 }
2803                 state->in_dev = __in_dev_get_rcu(state->dev);
2804                 if (!state->in_dev)
2805                         continue;
2806                 im = rcu_dereference(state->in_dev->mc_list);
2807         }
2808         return im;
2809 }
2810
2811 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2812 {
2813         struct ip_mc_list *im = igmp_mc_get_first(seq);
2814         if (im)
2815                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2816                         --pos;
2817         return pos ? NULL : im;
2818 }
2819
2820 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2821         __acquires(rcu)
2822 {
2823         rcu_read_lock();
2824         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2825 }
2826
2827 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2828 {
2829         struct ip_mc_list *im;
2830         if (v == SEQ_START_TOKEN)
2831                 im = igmp_mc_get_first(seq);
2832         else
2833                 im = igmp_mc_get_next(seq, v);
2834         ++*pos;
2835         return im;
2836 }
2837
2838 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2839         __releases(rcu)
2840 {
2841         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2842
2843         state->in_dev = NULL;
2844         state->dev = NULL;
2845         rcu_read_unlock();
2846 }
2847
2848 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2849 {
2850         if (v == SEQ_START_TOKEN)
2851                 seq_puts(seq,
2852                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2853         else {
2854                 struct ip_mc_list *im = v;
2855                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2856                 char   *querier;
2857                 long delta;
2858
2859 #ifdef CONFIG_IP_MULTICAST
2860                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2861                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2862                           "V3";
2863 #else
2864                 querier = "NONE";
2865 #endif
2866
2867                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2868                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2869                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2870                 }
2871
2872                 delta = im->timer.expires - jiffies;
2873                 seq_printf(seq,
2874                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2875                            im->multiaddr, im->users,
2876                            im->tm_running,
2877                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2878                            im->reporter);
2879         }
2880         return 0;
2881 }
2882
2883 static const struct seq_operations igmp_mc_seq_ops = {
2884         .start  =       igmp_mc_seq_start,
2885         .next   =       igmp_mc_seq_next,
2886         .stop   =       igmp_mc_seq_stop,
2887         .show   =       igmp_mc_seq_show,
2888 };
2889
2890 struct igmp_mcf_iter_state {
2891         struct seq_net_private p;
2892         struct net_device *dev;
2893         struct in_device *idev;
2894         struct ip_mc_list *im;
2895 };
2896
2897 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2898
2899 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2900 {
2901         struct net *net = seq_file_net(seq);
2902         struct ip_sf_list *psf = NULL;
2903         struct ip_mc_list *im = NULL;
2904         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2905
2906         state->idev = NULL;
2907         state->im = NULL;
2908         for_each_netdev_rcu(net, state->dev) {
2909                 struct in_device *idev;
2910                 idev = __in_dev_get_rcu(state->dev);
2911                 if (unlikely(!idev))
2912                         continue;
2913                 im = rcu_dereference(idev->mc_list);
2914                 if (likely(im)) {
2915                         spin_lock_bh(&im->lock);
2916                         psf = im->sources;
2917                         if (likely(psf)) {
2918                                 state->im = im;
2919                                 state->idev = idev;
2920                                 break;
2921                         }
2922                         spin_unlock_bh(&im->lock);
2923                 }
2924         }
2925         return psf;
2926 }
2927
2928 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2929 {
2930         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2931
2932         psf = psf->sf_next;
2933         while (!psf) {
2934                 spin_unlock_bh(&state->im->lock);
2935                 state->im = state->im->next;
2936                 while (!state->im) {
2937                         state->dev = next_net_device_rcu(state->dev);
2938                         if (!state->dev) {
2939                                 state->idev = NULL;
2940                                 goto out;
2941                         }
2942                         state->idev = __in_dev_get_rcu(state->dev);
2943                         if (!state->idev)
2944                                 continue;
2945                         state->im = rcu_dereference(state->idev->mc_list);
2946                 }
2947                 spin_lock_bh(&state->im->lock);
2948                 psf = state->im->sources;
2949         }
2950 out:
2951         return psf;
2952 }
2953
2954 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2955 {
2956         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2957         if (psf)
2958                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2959                         --pos;
2960         return pos ? NULL : psf;
2961 }
2962
2963 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2964         __acquires(rcu)
2965 {
2966         rcu_read_lock();
2967         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2968 }
2969
2970 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2971 {
2972         struct ip_sf_list *psf;
2973         if (v == SEQ_START_TOKEN)
2974                 psf = igmp_mcf_get_first(seq);
2975         else
2976                 psf = igmp_mcf_get_next(seq, v);
2977         ++*pos;
2978         return psf;
2979 }
2980
2981 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2982         __releases(rcu)
2983 {
2984         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2985         if (likely(state->im)) {
2986                 spin_unlock_bh(&state->im->lock);
2987                 state->im = NULL;
2988         }
2989         state->idev = NULL;
2990         state->dev = NULL;
2991         rcu_read_unlock();
2992 }
2993
2994 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2995 {
2996         struct ip_sf_list *psf = v;
2997         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2998
2999         if (v == SEQ_START_TOKEN) {
3000                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
3001         } else {
3002                 seq_printf(seq,
3003                            "%3d %6.6s 0x%08x "
3004                            "0x%08x %6lu %6lu\n",
3005                            state->dev->ifindex, state->dev->name,
3006                            ntohl(state->im->multiaddr),
3007                            ntohl(psf->sf_inaddr),
3008                            psf->sf_count[MCAST_INCLUDE],
3009                            psf->sf_count[MCAST_EXCLUDE]);
3010         }
3011         return 0;
3012 }
3013
3014 static const struct seq_operations igmp_mcf_seq_ops = {
3015         .start  =       igmp_mcf_seq_start,
3016         .next   =       igmp_mcf_seq_next,
3017         .stop   =       igmp_mcf_seq_stop,
3018         .show   =       igmp_mcf_seq_show,
3019 };
3020
3021 static int __net_init igmp_net_init(struct net *net)
3022 {
3023         struct proc_dir_entry *pde;
3024         int err;
3025
3026         pde = proc_create_net("igmp", 0444, net->proc_net, &igmp_mc_seq_ops,
3027                         sizeof(struct igmp_mc_iter_state));
3028         if (!pde)
3029                 goto out_igmp;
3030         pde = proc_create_net("mcfilter", 0444, net->proc_net,
3031                         &igmp_mcf_seq_ops, sizeof(struct igmp_mcf_iter_state));
3032         if (!pde)
3033                 goto out_mcfilter;
3034         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3035                                    SOCK_DGRAM, 0, net);
3036         if (err < 0) {
3037                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3038                        err);
3039                 goto out_sock;
3040         }
3041
3042         return 0;
3043
3044 out_sock:
3045         remove_proc_entry("mcfilter", net->proc_net);
3046 out_mcfilter:
3047         remove_proc_entry("igmp", net->proc_net);
3048 out_igmp:
3049         return -ENOMEM;
3050 }
3051
3052 static void __net_exit igmp_net_exit(struct net *net)
3053 {
3054         remove_proc_entry("mcfilter", net->proc_net);
3055         remove_proc_entry("igmp", net->proc_net);
3056         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3057 }
3058
3059 static struct pernet_operations igmp_net_ops = {
3060         .init = igmp_net_init,
3061         .exit = igmp_net_exit,
3062 };
3063 #endif
3064
3065 static int igmp_netdev_event(struct notifier_block *this,
3066                              unsigned long event, void *ptr)
3067 {
3068         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3069         struct in_device *in_dev;
3070
3071         switch (event) {
3072         case NETDEV_RESEND_IGMP:
3073                 in_dev = __in_dev_get_rtnl(dev);
3074                 if (in_dev)
3075                         ip_mc_rejoin_groups(in_dev);
3076                 break;
3077         default:
3078                 break;
3079         }
3080         return NOTIFY_DONE;
3081 }
3082
3083 static struct notifier_block igmp_notifier = {
3084         .notifier_call = igmp_netdev_event,
3085 };
3086
3087 int __init igmp_mc_init(void)
3088 {
3089 #if defined(CONFIG_PROC_FS)
3090         int err;
3091
3092         err = register_pernet_subsys(&igmp_net_ops);
3093         if (err)
3094                 return err;
3095         err = register_netdevice_notifier(&igmp_notifier);
3096         if (err)
3097                 goto reg_notif_fail;
3098         return 0;
3099
3100 reg_notif_fail:
3101         unregister_pernet_subsys(&igmp_net_ops);
3102         return err;
3103 #else
3104         return register_netdevice_notifier(&igmp_notifier);
3105 #endif
3106 }