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 = mtu;
 357
 358         while (1) {
 359                 skb = alloc_skb(size + hlen + tlen,
 360                                 GFP_ATOMIC | __GFP_NOWARN);
 361                 if (skb)
 362                         break;
 363                 size >>= 1;
 364                 if (size < 256)
 365                         return NULL;
 366         }
 367         skb->priority = TC_PRIO_CONTROL;
 368
 369         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 370                                    0, 0,
 371                                    IPPROTO_IGMP, 0, dev->ifindex);
 372         if (IS_ERR(rt)) {
 373                 kfree_skb(skb);
 374                 return NULL;
 375         }
 376
 377         skb_dst_set(skb, &rt->dst);
 378         skb->dev = dev;
 379
 380         skb_reserve(skb, hlen);
 381         skb_tailroom_reserve(skb, mtu, tlen);
 382
 383         skb_reset_network_header(skb);
 384         pip = ip_hdr(skb);
 385         skb_put(skb, sizeof(struct iphdr) + 4);
 386
 387         pip->version  = 4;
 388         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 389         pip->tos      = 0xc0;
 390         pip->frag_off = htons(IP_DF);
 391         pip->ttl      = 1;
 392         pip->daddr    = fl4.daddr;
 393
 394         rcu_read_lock();
 395         pip->saddr    = igmpv3_get_srcaddr(dev, &fl4);
 396         rcu_read_unlock();
 397
 398         pip->protocol = IPPROTO_IGMP;
 399         pip->tot_len  = 0;      /* filled in later */
 400         ip_select_ident(net, skb, NULL);
 401         ((u8 *)&pip[1])[0] = IPOPT_RA;
 402         ((u8 *)&pip[1])[1] = 4;
 403         ((u8 *)&pip[1])[2] = 0;
 404         ((u8 *)&pip[1])[3] = 0;
 405
 406         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 407         skb_put(skb, sizeof(*pig));
 408         pig = igmpv3_report_hdr(skb);
 409         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 410         pig->resv1 = 0;
 411         pig->csum = 0;
 412         pig->resv2 = 0;
 413         pig->ngrec = 0;
 414         return skb;
 415 }
 416
 417 static int igmpv3_sendpack(struct sk_buff *skb)
 418 {
 419         struct igmphdr *pig = igmp_hdr(skb);
 420         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 421
 422         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 423
 424         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 425 }
 426
 427 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 428 {
 429         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 430 }
 431
 432 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 433         int type, struct igmpv3_grec **ppgr, unsigned int mtu)
 434 {
 435         struct net_device *dev = pmc->interface->dev;
 436         struct igmpv3_report *pih;
 437         struct igmpv3_grec *pgr;
 438
 439         if (!skb) {
 440                 skb = igmpv3_newpack(dev, mtu);
 441                 if (!skb)
 442                         return NULL;
 443         }
 444         pgr = skb_put(skb, sizeof(struct igmpv3_grec));
 445         pgr->grec_type = type;
 446         pgr->grec_auxwords = 0;
 447         pgr->grec_nsrcs = 0;
 448         pgr->grec_mca = pmc->multiaddr;
 449         pih = igmpv3_report_hdr(skb);
 450         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 451         *ppgr = pgr;
 452         return skb;
 453 }
 454
 455 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 456
 457 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 458         int type, int gdeleted, int sdeleted)
 459 {
 460         struct net_device *dev = pmc->interface->dev;
 461         struct net *net = dev_net(dev);
 462         struct igmpv3_report *pih;
 463         struct igmpv3_grec *pgr = NULL;
 464         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 465         int scount, stotal, first, isquery, truncate;
 466         unsigned int mtu;
 467
 468         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 469                 return skb;
 470         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 471             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 472                 return skb;
 473
 474         mtu = READ_ONCE(dev->mtu);
 475         if (mtu < IPV4_MIN_MTU)
 476                 return skb;
 477
 478         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 479                   type == IGMPV3_MODE_IS_EXCLUDE;
 480         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 481                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 482
 483         stotal = scount = 0;
 484
 485         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 486
 487         if (!*psf_list)
 488                 goto empty_source;
 489
 490         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 491
 492         /* EX and TO_EX get a fresh packet, if needed */
 493         if (truncate) {
 494                 if (pih && pih->ngrec &&
 495                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 496                         if (skb)
 497                                 igmpv3_sendpack(skb);
 498                         skb = igmpv3_newpack(dev, mtu);
 499                 }
 500         }
 501         first = 1;
 502         psf_prev = NULL;
 503         for (psf = *psf_list; psf; psf = psf_next) {
 504                 __be32 *psrc;
 505
 506                 psf_next = psf->sf_next;
 507
 508                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 509                         psf_prev = psf;
 510                         continue;
 511                 }
 512
 513                 /* Based on RFC3376 5.1. Should not send source-list change
 514                  * records when there is a filter mode change.
 515                  */
 516                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 517                      (!gdeleted && pmc->crcount)) &&
 518                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 519                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 520                         goto decrease_sf_crcount;
 521
 522                 /* clear marks on query responses */
 523                 if (isquery)
 524                         psf->sf_gsresp = 0;
 525
 526                 if (AVAILABLE(skb) < sizeof(__be32) +
 527                     first*sizeof(struct igmpv3_grec)) {
 528                         if (truncate && !first)
 529                                 break;   /* truncate these */
 530                         if (pgr)
 531                                 pgr->grec_nsrcs = htons(scount);
 532                         if (skb)
 533                                 igmpv3_sendpack(skb);
 534                         skb = igmpv3_newpack(dev, mtu);
 535                         first = 1;
 536                         scount = 0;
 537                 }
 538                 if (first) {
 539                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 540                         first = 0;
 541                 }
 542                 if (!skb)
 543                         return NULL;
 544                 psrc = skb_put(skb, sizeof(__be32));
 545                 *psrc = psf->sf_inaddr;
 546                 scount++; stotal++;
 547                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 548                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 549 decrease_sf_crcount:
 550                         psf->sf_crcount--;
 551                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 552                                 if (psf_prev)
 553                                         psf_prev->sf_next = psf->sf_next;
 554                                 else
 555                                         *psf_list = psf->sf_next;
 556                                 kfree(psf);
 557                                 continue;
 558                         }
 559                 }
 560                 psf_prev = psf;
 561         }
 562
 563 empty_source:
 564         if (!stotal) {
 565                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 566                     type == IGMPV3_BLOCK_OLD_SOURCES)
 567                         return skb;
 568                 if (pmc->crcount || isquery) {
 569                         /* make sure we have room for group header */
 570                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 571                                 igmpv3_sendpack(skb);
 572                                 skb = NULL; /* add_grhead will get a new one */
 573                         }
 574                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 575                 }
 576         }
 577         if (pgr)
 578                 pgr->grec_nsrcs = htons(scount);
 579
 580         if (isquery)
 581                 pmc->gsquery = 0;       /* clear query state on report */
 582         return skb;
 583 }
 584
 585 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 586 {
 587         struct sk_buff *skb = NULL;
 588         struct net *net = dev_net(in_dev->dev);
 589         int type;
 590
 591         if (!pmc) {
 592                 rcu_read_lock();
 593                 for_each_pmc_rcu(in_dev, pmc) {
 594                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 595                                 continue;
 596                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 597                             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 598                                 continue;
 599                         spin_lock_bh(&pmc->lock);
 600                         if (pmc->sfcount[MCAST_EXCLUDE])
 601                                 type = IGMPV3_MODE_IS_EXCLUDE;
 602                         else
 603                                 type = IGMPV3_MODE_IS_INCLUDE;
 604                         skb = add_grec(skb, pmc, type, 0, 0);
 605                         spin_unlock_bh(&pmc->lock);
 606                 }
 607                 rcu_read_unlock();
 608         } else {
 609                 spin_lock_bh(&pmc->lock);
 610                 if (pmc->sfcount[MCAST_EXCLUDE])
 611                         type = IGMPV3_MODE_IS_EXCLUDE;
 612                 else
 613                         type = IGMPV3_MODE_IS_INCLUDE;
 614                 skb = add_grec(skb, pmc, type, 0, 0);
 615                 spin_unlock_bh(&pmc->lock);
 616         }
 617         if (!skb)
 618                 return 0;
 619         return igmpv3_sendpack(skb);
 620 }
 621
 622 /*
 623  * remove zero-count source records from a source filter list
 624  */
 625 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 626 {
 627         struct ip_sf_list *psf_prev, *psf_next, *psf;
 628
 629         psf_prev = NULL;
 630         for (psf = *ppsf; psf; psf = psf_next) {
 631                 psf_next = psf->sf_next;
 632                 if (psf->sf_crcount == 0) {
 633                         if (psf_prev)
 634                                 psf_prev->sf_next = psf->sf_next;
 635                         else
 636                                 *ppsf = psf->sf_next;
 637                         kfree(psf);
 638                 } else
 639                         psf_prev = psf;
 640         }
 641 }
 642
 643 static void kfree_pmc(struct ip_mc_list *pmc)
 644 {
 645         ip_sf_list_clear_all(pmc->sources);
 646         ip_sf_list_clear_all(pmc->tomb);
 647         kfree(pmc);
 648 }
 649
 650 static void igmpv3_send_cr(struct in_device *in_dev)
 651 {
 652         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 653         struct sk_buff *skb = NULL;
 654         int type, dtype;
 655
 656         rcu_read_lock();
 657         spin_lock_bh(&in_dev->mc_tomb_lock);
 658
 659         /* deleted MCA's */
 660         pmc_prev = NULL;
 661         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 662                 pmc_next = pmc->next;
 663                 if (pmc->sfmode == MCAST_INCLUDE) {
 664                         type = IGMPV3_BLOCK_OLD_SOURCES;
 665                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 666                         skb = add_grec(skb, pmc, type, 1, 0);
 667                         skb = add_grec(skb, pmc, dtype, 1, 1);
 668                 }
 669                 if (pmc->crcount) {
 670                         if (pmc->sfmode == MCAST_EXCLUDE) {
 671                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 672                                 skb = add_grec(skb, pmc, type, 1, 0);
 673                         }
 674                         pmc->crcount--;
 675                         if (pmc->crcount == 0) {
 676                                 igmpv3_clear_zeros(&pmc->tomb);
 677                                 igmpv3_clear_zeros(&pmc->sources);
 678                         }
 679                 }
 680                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 681                         if (pmc_prev)
 682                                 pmc_prev->next = pmc_next;
 683                         else
 684                                 in_dev->mc_tomb = pmc_next;
 685                         in_dev_put(pmc->interface);
 686                         kfree_pmc(pmc);
 687                 } else
 688                         pmc_prev = pmc;
 689         }
 690         spin_unlock_bh(&in_dev->mc_tomb_lock);
 691
 692         /* change recs */
 693         for_each_pmc_rcu(in_dev, pmc) {
 694                 spin_lock_bh(&pmc->lock);
 695                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 696                         type = IGMPV3_BLOCK_OLD_SOURCES;
 697                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 698                 } else {
 699                         type = IGMPV3_ALLOW_NEW_SOURCES;
 700                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 701                 }
 702                 skb = add_grec(skb, pmc, type, 0, 0);
 703                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 704
 705                 /* filter mode changes */
 706                 if (pmc->crcount) {
 707                         if (pmc->sfmode == MCAST_EXCLUDE)
 708                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 709                         else
 710                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 711                         skb = add_grec(skb, pmc, type, 0, 0);
 712                         pmc->crcount--;
 713                 }
 714                 spin_unlock_bh(&pmc->lock);
 715         }
 716         rcu_read_unlock();
 717
 718         if (!skb)
 719                 return;
 720         (void) igmpv3_sendpack(skb);
 721 }
 722
 723 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 724         int type)
 725 {
 726         struct sk_buff *skb;
 727         struct iphdr *iph;
 728         struct igmphdr *ih;
 729         struct rtable *rt;
 730         struct net_device *dev = in_dev->dev;
 731         struct net *net = dev_net(dev);
 732         __be32  group = pmc ? pmc->multiaddr : 0;
 733         struct flowi4 fl4;
 734         __be32  dst;
 735         int hlen, tlen;
 736
 737         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 738                 return igmpv3_send_report(in_dev, pmc);
 739
 740         if (ipv4_is_local_multicast(group) &&
 741             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 742                 return 0;
 743
 744         if (type == IGMP_HOST_LEAVE_MESSAGE)
 745                 dst = IGMP_ALL_ROUTER;
 746         else
 747                 dst = group;
 748
 749         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 750                                    0, 0,
 751                                    IPPROTO_IGMP, 0, dev->ifindex);
 752         if (IS_ERR(rt))
 753                 return -1;
 754
 755         hlen = LL_RESERVED_SPACE(dev);
 756         tlen = dev->needed_tailroom;
 757         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 758         if (!skb) {
 759                 ip_rt_put(rt);
 760                 return -1;
 761         }
 762         skb->priority = TC_PRIO_CONTROL;
 763
 764         skb_dst_set(skb, &rt->dst);
 765
 766         skb_reserve(skb, hlen);
 767
 768         skb_reset_network_header(skb);
 769         iph = ip_hdr(skb);
 770         skb_put(skb, sizeof(struct iphdr) + 4);
 771
 772         iph->version  = 4;
 773         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 774         iph->tos      = 0xc0;
 775         iph->frag_off = htons(IP_DF);
 776         iph->ttl      = 1;
 777         iph->daddr    = dst;
 778         iph->saddr    = fl4.saddr;
 779         iph->protocol = IPPROTO_IGMP;
 780         ip_select_ident(net, skb, NULL);
 781         ((u8 *)&iph[1])[0] = IPOPT_RA;
 782         ((u8 *)&iph[1])[1] = 4;
 783         ((u8 *)&iph[1])[2] = 0;
 784         ((u8 *)&iph[1])[3] = 0;
 785
 786         ih = skb_put(skb, sizeof(struct igmphdr));
 787         ih->type = type;
 788         ih->code = 0;
 789         ih->csum = 0;
 790         ih->group = group;
 791         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 792
 793         return ip_local_out(net, skb->sk, skb);
 794 }
 795
 796 static void igmp_gq_timer_expire(struct timer_list *t)
 797 {
 798         struct in_device *in_dev = from_timer(in_dev, t, mr_gq_timer);
 799
 800         in_dev->mr_gq_running = 0;
 801         igmpv3_send_report(in_dev, NULL);
 802         in_dev_put(in_dev);
 803 }
 804
 805 static void igmp_ifc_timer_expire(struct timer_list *t)
 806 {
 807         struct in_device *in_dev = from_timer(in_dev, t, mr_ifc_timer);
 808         u32 mr_ifc_count;
 809
 810         igmpv3_send_cr(in_dev);
 811 restart:
 812         mr_ifc_count = READ_ONCE(in_dev->mr_ifc_count);
 813
 814         if (mr_ifc_count) {
 815                 if (cmpxchg(&in_dev->mr_ifc_count,
 816                             mr_ifc_count,
 817                             mr_ifc_count - 1) != mr_ifc_count)
 818                         goto restart;
 819                 igmp_ifc_start_timer(in_dev,
 820                                      unsolicited_report_interval(in_dev));
 821         }
 822         in_dev_put(in_dev);
 823 }
 824
 825 static void igmp_ifc_event(struct in_device *in_dev)
 826 {
 827         struct net *net = dev_net(in_dev->dev);
 828         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 829                 return;
 830         WRITE_ONCE(in_dev->mr_ifc_count, in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv));
 831         igmp_ifc_start_timer(in_dev, 1);
 832 }
 833
 834
 835 static void igmp_timer_expire(struct timer_list *t)
 836 {
 837         struct ip_mc_list *im = from_timer(im, t, timer);
 838         struct in_device *in_dev = im->interface;
 839
 840         spin_lock(&im->lock);
 841         im->tm_running = 0;
 842
 843         if (im->unsolicit_count && --im->unsolicit_count)
 844                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 845
 846         im->reporter = 1;
 847         spin_unlock(&im->lock);
 848
 849         if (IGMP_V1_SEEN(in_dev))
 850                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 851         else if (IGMP_V2_SEEN(in_dev))
 852                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 853         else
 854                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 855
 856         ip_ma_put(im);
 857 }
 858
 859 /* mark EXCLUDE-mode sources */
 860 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 861 {
 862         struct ip_sf_list *psf;
 863         int i, scount;
 864
 865         scount = 0;
 866         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 867                 if (scount == nsrcs)
 868                         break;
 869                 for (i = 0; i < nsrcs; i++) {
 870                         /* skip inactive filters */
 871                         if (psf->sf_count[MCAST_INCLUDE] ||
 872                             pmc->sfcount[MCAST_EXCLUDE] !=
 873                             psf->sf_count[MCAST_EXCLUDE])
 874                                 break;
 875                         if (srcs[i] == psf->sf_inaddr) {
 876                                 scount++;
 877                                 break;
 878                         }
 879                 }
 880         }
 881         pmc->gsquery = 0;
 882         if (scount == nsrcs)    /* all sources excluded */
 883                 return 0;
 884         return 1;
 885 }
 886
 887 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 888 {
 889         struct ip_sf_list *psf;
 890         int i, scount;
 891
 892         if (pmc->sfmode == MCAST_EXCLUDE)
 893                 return igmp_xmarksources(pmc, nsrcs, srcs);
 894
 895         /* mark INCLUDE-mode sources */
 896         scount = 0;
 897         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 898                 if (scount == nsrcs)
 899                         break;
 900                 for (i = 0; i < nsrcs; i++)
 901                         if (srcs[i] == psf->sf_inaddr) {
 902                                 psf->sf_gsresp = 1;
 903                                 scount++;
 904                                 break;
 905                         }
 906         }
 907         if (!scount) {
 908                 pmc->gsquery = 0;
 909                 return 0;
 910         }
 911         pmc->gsquery = 1;
 912         return 1;
 913 }
 914
 915 /* return true if packet was dropped */
 916 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 917 {
 918         struct ip_mc_list *im;
 919         struct net *net = dev_net(in_dev->dev);
 920
 921         /* Timers are only set for non-local groups */
 922
 923         if (group == IGMP_ALL_HOSTS)
 924                 return false;
 925         if (ipv4_is_local_multicast(group) &&
 926             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 927                 return false;
 928
 929         rcu_read_lock();
 930         for_each_pmc_rcu(in_dev, im) {
 931                 if (im->multiaddr == group) {
 932                         igmp_stop_timer(im);
 933                         break;
 934                 }
 935         }
 936         rcu_read_unlock();
 937         return false;
 938 }
 939
 940 /* return true if packet was dropped */
 941 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 942         int len)
 943 {
 944         struct igmphdr          *ih = igmp_hdr(skb);
 945         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 946         struct ip_mc_list       *im;
 947         __be32                  group = ih->group;
 948         int                     max_delay;
 949         int                     mark = 0;
 950         struct net              *net = dev_net(in_dev->dev);
 951
 952
 953         if (len == 8) {
 954                 if (ih->code == 0) {
 955                         /* Alas, old v1 router presents here. */
 956
 957                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 958                         in_dev->mr_v1_seen = jiffies +
 959                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 960                                 in_dev->mr_qri;
 961                         group = 0;
 962                 } else {
 963                         /* v2 router present */
 964                         max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
 965                         in_dev->mr_v2_seen = jiffies +
 966                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 967                                 in_dev->mr_qri;
 968                 }
 969                 /* cancel the interface change timer */
 970                 WRITE_ONCE(in_dev->mr_ifc_count, 0);
 971                 if (del_timer(&in_dev->mr_ifc_timer))
 972                         __in_dev_put(in_dev);
 973                 /* clear deleted report items */
 974                 igmpv3_clear_delrec(in_dev);
 975         } else if (len < 12) {
 976                 return true;    /* ignore bogus packet; freed by caller */
 977         } else if (IGMP_V1_SEEN(in_dev)) {
 978                 /* This is a v3 query with v1 queriers present */
 979                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 980                 group = 0;
 981         } else if (IGMP_V2_SEEN(in_dev)) {
 982                 /* this is a v3 query with v2 queriers present;
 983                  * Interpretation of the max_delay code is problematic here.
 984                  * A real v2 host would use ih_code directly, while v3 has a
 985                  * different encoding. We use the v3 encoding as more likely
 986                  * to be intended in a v3 query.
 987                  */
 988                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 989                 if (!max_delay)
 990                         max_delay = 1;  /* can't mod w/ 0 */
 991         } else { /* v3 */
 992                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 993                         return true;
 994
 995                 ih3 = igmpv3_query_hdr(skb);
 996                 if (ih3->nsrcs) {
 997                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
 998                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
 999                                 return true;
1000                         ih3 = igmpv3_query_hdr(skb);
1001                 }
1002
1003                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
1004                 if (!max_delay)
1005                         max_delay = 1;  /* can't mod w/ 0 */
1006                 in_dev->mr_maxdelay = max_delay;
1007
1008                 /* RFC3376, 4.1.6. QRV and 4.1.7. QQIC, when the most recently
1009                  * received value was zero, use the default or statically
1010                  * configured value.
1011                  */
1012                 in_dev->mr_qrv = ih3->qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1013                 in_dev->mr_qi = IGMPV3_QQIC(ih3->qqic)*HZ ?: IGMP_QUERY_INTERVAL;
1014
1015                 /* RFC3376, 8.3. Query Response Interval:
1016                  * The number of seconds represented by the [Query Response
1017                  * Interval] must be less than the [Query Interval].
1018                  */
1019                 if (in_dev->mr_qri >= in_dev->mr_qi)
1020                         in_dev->mr_qri = (in_dev->mr_qi/HZ - 1)*HZ;
1021
1022                 if (!group) { /* general query */
1023                         if (ih3->nsrcs)
1024                                 return true;    /* no sources allowed */
1025                         igmp_gq_start_timer(in_dev);
1026                         return false;
1027                 }
1028                 /* mark sources to include, if group & source-specific */
1029                 mark = ih3->nsrcs != 0;
1030         }
1031
1032         /*
1033          * - Start the timers in all of our membership records
1034          *   that the query applies to for the interface on
1035          *   which the query arrived excl. those that belong
1036          *   to a "local" group (224.0.0.X)
1037          * - For timers already running check if they need to
1038          *   be reset.
1039          * - Use the igmp->igmp_code field as the maximum
1040          *   delay possible
1041          */
1042         rcu_read_lock();
1043         for_each_pmc_rcu(in_dev, im) {
1044                 int changed;
1045
1046                 if (group && group != im->multiaddr)
1047                         continue;
1048                 if (im->multiaddr == IGMP_ALL_HOSTS)
1049                         continue;
1050                 if (ipv4_is_local_multicast(im->multiaddr) &&
1051                     !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1052                         continue;
1053                 spin_lock_bh(&im->lock);
1054                 if (im->tm_running)
1055                         im->gsquery = im->gsquery && mark;
1056                 else
1057                         im->gsquery = mark;
1058                 changed = !im->gsquery ||
1059                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1060                 spin_unlock_bh(&im->lock);
1061                 if (changed)
1062                         igmp_mod_timer(im, max_delay);
1063         }
1064         rcu_read_unlock();
1065         return false;
1066 }
1067
1068 /* called in rcu_read_lock() section */
1069 int igmp_rcv(struct sk_buff *skb)
1070 {
1071         /* This basically follows the spec line by line -- see RFC1112 */
1072         struct igmphdr *ih;
1073         struct net_device *dev = skb->dev;
1074         struct in_device *in_dev;
1075         int len = skb->len;
1076         bool dropped = true;
1077
1078         if (netif_is_l3_master(dev)) {
1079                 dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1080                 if (!dev)
1081                         goto drop;
1082         }
1083
1084         in_dev = __in_dev_get_rcu(dev);
1085         if (!in_dev)
1086                 goto drop;
1087
1088         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1089                 goto drop;
1090
1091         if (skb_checksum_simple_validate(skb))
1092                 goto drop;
1093
1094         ih = igmp_hdr(skb);
1095         switch (ih->type) {
1096         case IGMP_HOST_MEMBERSHIP_QUERY:
1097                 dropped = igmp_heard_query(in_dev, skb, len);
1098                 break;
1099         case IGMP_HOST_MEMBERSHIP_REPORT:
1100         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1101                 /* Is it our report looped back? */
1102                 if (rt_is_output_route(skb_rtable(skb)))
1103                         break;
1104                 /* don't rely on MC router hearing unicast reports */
1105                 if (skb->pkt_type == PACKET_MULTICAST ||
1106                     skb->pkt_type == PACKET_BROADCAST)
1107                         dropped = igmp_heard_report(in_dev, ih->group);
1108                 break;
1109         case IGMP_PIM:
1110 #ifdef CONFIG_IP_PIMSM_V1
1111                 return pim_rcv_v1(skb);
1112 #endif
1113         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1114         case IGMP_DVMRP:
1115         case IGMP_TRACE:
1116         case IGMP_HOST_LEAVE_MESSAGE:
1117         case IGMP_MTRACE:
1118         case IGMP_MTRACE_RESP:
1119                 break;
1120         default:
1121                 break;
1122         }
1123
1124 drop:
1125         if (dropped)
1126                 kfree_skb(skb);
1127         else
1128                 consume_skb(skb);
1129         return 0;
1130 }
1131
1132 #endif
1133
1134
1135 /*
1136  *      Add a filter to a device
1137  */
1138
1139 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1140 {
1141         char buf[MAX_ADDR_LEN];
1142         struct net_device *dev = in_dev->dev;
1143
1144         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1145            We will get multicast token leakage, when IFF_MULTICAST
1146            is changed. This check should be done in ndo_set_rx_mode
1147            routine. Something sort of:
1148            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1149            --ANK
1150            */
1151         if (arp_mc_map(addr, buf, dev, 0) == 0)
1152                 dev_mc_add(dev, buf);
1153 }
1154
1155 /*
1156  *      Remove a filter from a device
1157  */
1158
1159 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1160 {
1161         char buf[MAX_ADDR_LEN];
1162         struct net_device *dev = in_dev->dev;
1163
1164         if (arp_mc_map(addr, buf, dev, 0) == 0)
1165                 dev_mc_del(dev, buf);
1166 }
1167
1168 #ifdef CONFIG_IP_MULTICAST
1169 /*
1170  * deleted ip_mc_list manipulation
1171  */
1172 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
1173                               gfp_t gfp)
1174 {
1175         struct ip_mc_list *pmc;
1176         struct net *net = dev_net(in_dev->dev);
1177
1178         /* this is an "ip_mc_list" for convenience; only the fields below
1179          * are actually used. In particular, the refcnt and users are not
1180          * used for management of the delete list. Using the same structure
1181          * for deleted items allows change reports to use common code with
1182          * non-deleted or query-response MCA's.
1183          */
1184         pmc = kzalloc(sizeof(*pmc), gfp);
1185         if (!pmc)
1186                 return;
1187         spin_lock_init(&pmc->lock);
1188         spin_lock_bh(&im->lock);
1189         pmc->interface = im->interface;
1190         in_dev_hold(in_dev);
1191         pmc->multiaddr = im->multiaddr;
1192         pmc->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1193         pmc->sfmode = im->sfmode;
1194         if (pmc->sfmode == MCAST_INCLUDE) {
1195                 struct ip_sf_list *psf;
1196
1197                 pmc->tomb = im->tomb;
1198                 pmc->sources = im->sources;
1199                 im->tomb = im->sources = NULL;
1200                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1201                         psf->sf_crcount = pmc->crcount;
1202         }
1203         spin_unlock_bh(&im->lock);
1204
1205         spin_lock_bh(&in_dev->mc_tomb_lock);
1206         pmc->next = in_dev->mc_tomb;
1207         in_dev->mc_tomb = pmc;
1208         spin_unlock_bh(&in_dev->mc_tomb_lock);
1209 }
1210
1211 /*
1212  * restore ip_mc_list deleted records
1213  */
1214 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1215 {
1216         struct ip_mc_list *pmc, *pmc_prev;
1217         struct ip_sf_list *psf;
1218         struct net *net = dev_net(in_dev->dev);
1219         __be32 multiaddr = im->multiaddr;
1220
1221         spin_lock_bh(&in_dev->mc_tomb_lock);
1222         pmc_prev = NULL;
1223         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1224                 if (pmc->multiaddr == multiaddr)
1225                         break;
1226                 pmc_prev = pmc;
1227         }
1228         if (pmc) {
1229                 if (pmc_prev)
1230                         pmc_prev->next = pmc->next;
1231                 else
1232                         in_dev->mc_tomb = pmc->next;
1233         }
1234         spin_unlock_bh(&in_dev->mc_tomb_lock);
1235
1236         spin_lock_bh(&im->lock);
1237         if (pmc) {
1238                 im->interface = pmc->interface;
1239                 if (im->sfmode == MCAST_INCLUDE) {
1240                         swap(im->tomb, pmc->tomb);
1241                         swap(im->sources, pmc->sources);
1242                         for (psf = im->sources; psf; psf = psf->sf_next)
1243                                 psf->sf_crcount = in_dev->mr_qrv ?:
1244                                         READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1245                 } else {
1246                         im->crcount = in_dev->mr_qrv ?:
1247                                 READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1248                 }
1249                 in_dev_put(pmc->interface);
1250                 kfree_pmc(pmc);
1251         }
1252         spin_unlock_bh(&im->lock);
1253 }
1254
1255 /*
1256  * flush ip_mc_list deleted records
1257  */
1258 static void igmpv3_clear_delrec(struct in_device *in_dev)
1259 {
1260         struct ip_mc_list *pmc, *nextpmc;
1261
1262         spin_lock_bh(&in_dev->mc_tomb_lock);
1263         pmc = in_dev->mc_tomb;
1264         in_dev->mc_tomb = NULL;
1265         spin_unlock_bh(&in_dev->mc_tomb_lock);
1266
1267         for (; pmc; pmc = nextpmc) {
1268                 nextpmc = pmc->next;
1269                 ip_mc_clear_src(pmc);
1270                 in_dev_put(pmc->interface);
1271                 kfree_pmc(pmc);
1272         }
1273         /* clear dead sources, too */
1274         rcu_read_lock();
1275         for_each_pmc_rcu(in_dev, pmc) {
1276                 struct ip_sf_list *psf;
1277
1278                 spin_lock_bh(&pmc->lock);
1279                 psf = pmc->tomb;
1280                 pmc->tomb = NULL;
1281                 spin_unlock_bh(&pmc->lock);
1282                 ip_sf_list_clear_all(psf);
1283         }
1284         rcu_read_unlock();
1285 }
1286 #endif
1287
1288 static void __igmp_group_dropped(struct ip_mc_list *im, gfp_t gfp)
1289 {
1290         struct in_device *in_dev = im->interface;
1291 #ifdef CONFIG_IP_MULTICAST
1292         struct net *net = dev_net(in_dev->dev);
1293         int reporter;
1294 #endif
1295
1296         if (im->loaded) {
1297                 im->loaded = 0;
1298                 ip_mc_filter_del(in_dev, im->multiaddr);
1299         }
1300
1301 #ifdef CONFIG_IP_MULTICAST
1302         if (im->multiaddr == IGMP_ALL_HOSTS)
1303                 return;
1304         if (ipv4_is_local_multicast(im->multiaddr) &&
1305             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1306                 return;
1307
1308         reporter = im->reporter;
1309         igmp_stop_timer(im);
1310
1311         if (!in_dev->dead) {
1312                 if (IGMP_V1_SEEN(in_dev))
1313                         return;
1314                 if (IGMP_V2_SEEN(in_dev)) {
1315                         if (reporter)
1316                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1317                         return;
1318                 }
1319                 /* IGMPv3 */
1320                 igmpv3_add_delrec(in_dev, im, gfp);
1321
1322                 igmp_ifc_event(in_dev);
1323         }
1324 #endif
1325 }
1326
1327 static void igmp_group_dropped(struct ip_mc_list *im)
1328 {
1329         __igmp_group_dropped(im, GFP_KERNEL);
1330 }
1331
1332 static void igmp_group_added(struct ip_mc_list *im)
1333 {
1334         struct in_device *in_dev = im->interface;
1335 #ifdef CONFIG_IP_MULTICAST
1336         struct net *net = dev_net(in_dev->dev);
1337 #endif
1338
1339         if (im->loaded == 0) {
1340                 im->loaded = 1;
1341                 ip_mc_filter_add(in_dev, im->multiaddr);
1342         }
1343
1344 #ifdef CONFIG_IP_MULTICAST
1345         if (im->multiaddr == IGMP_ALL_HOSTS)
1346                 return;
1347         if (ipv4_is_local_multicast(im->multiaddr) &&
1348             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1349                 return;
1350
1351         if (in_dev->dead)
1352                 return;
1353
1354         im->unsolicit_count = READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1355         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1356                 spin_lock_bh(&im->lock);
1357                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1358                 spin_unlock_bh(&im->lock);
1359                 return;
1360         }
1361         /* else, v3 */
1362
1363         /* Based on RFC3376 5.1, for newly added INCLUDE SSM, we should
1364          * not send filter-mode change record as the mode should be from
1365          * IN() to IN(A).
1366          */
1367         if (im->sfmode == MCAST_EXCLUDE)
1368                 im->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1369
1370         igmp_ifc_event(in_dev);
1371 #endif
1372 }
1373
1374
1375 /*
1376  *      Multicast list managers
1377  */
1378
1379 static u32 ip_mc_hash(const struct ip_mc_list *im)
1380 {
1381         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1382 }
1383
1384 static void ip_mc_hash_add(struct in_device *in_dev,
1385                            struct ip_mc_list *im)
1386 {
1387         struct ip_mc_list __rcu **mc_hash;
1388         u32 hash;
1389
1390         mc_hash = rtnl_dereference(in_dev->mc_hash);
1391         if (mc_hash) {
1392                 hash = ip_mc_hash(im);
1393                 im->next_hash = mc_hash[hash];
1394                 rcu_assign_pointer(mc_hash[hash], im);
1395                 return;
1396         }
1397
1398         /* do not use a hash table for small number of items */
1399         if (in_dev->mc_count < 4)
1400                 return;
1401
1402         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1403                           GFP_KERNEL);
1404         if (!mc_hash)
1405                 return;
1406
1407         for_each_pmc_rtnl(in_dev, im) {
1408                 hash = ip_mc_hash(im);
1409                 im->next_hash = mc_hash[hash];
1410                 RCU_INIT_POINTER(mc_hash[hash], im);
1411         }
1412
1413         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1414 }
1415
1416 static void ip_mc_hash_remove(struct in_device *in_dev,
1417                               struct ip_mc_list *im)
1418 {
1419         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1420         struct ip_mc_list *aux;
1421
1422         if (!mc_hash)
1423                 return;
1424         mc_hash += ip_mc_hash(im);
1425         while ((aux = rtnl_dereference(*mc_hash)) != im)
1426                 mc_hash = &aux->next_hash;
1427         *mc_hash = im->next_hash;
1428 }
1429
1430
1431 /*
1432  *      A socket has joined a multicast group on device dev.
1433  */
1434 static void ____ip_mc_inc_group(struct in_device *in_dev, __be32 addr,
1435                                 unsigned int mode, gfp_t gfp)
1436 {
1437         struct ip_mc_list *im;
1438
1439         ASSERT_RTNL();
1440
1441         for_each_pmc_rtnl(in_dev, im) {
1442                 if (im->multiaddr == addr) {
1443                         im->users++;
1444                         ip_mc_add_src(in_dev, &addr, mode, 0, NULL, 0);
1445                         goto out;
1446                 }
1447         }
1448
1449         im = kzalloc(sizeof(*im), gfp);
1450         if (!im)
1451                 goto out;
1452
1453         im->users = 1;
1454         im->interface = in_dev;
1455         in_dev_hold(in_dev);
1456         im->multiaddr = addr;
1457         /* initial mode is (EX, empty) */
1458         im->sfmode = mode;
1459         im->sfcount[mode] = 1;
1460         refcount_set(&im->refcnt, 1);
1461         spin_lock_init(&im->lock);
1462 #ifdef CONFIG_IP_MULTICAST
1463         timer_setup(&im->timer, igmp_timer_expire, 0);
1464 #endif
1465
1466         im->next_rcu = in_dev->mc_list;
1467         in_dev->mc_count++;
1468         rcu_assign_pointer(in_dev->mc_list, im);
1469
1470         ip_mc_hash_add(in_dev, im);
1471
1472 #ifdef CONFIG_IP_MULTICAST
1473         igmpv3_del_delrec(in_dev, im);
1474 #endif
1475         igmp_group_added(im);
1476         if (!in_dev->dead)
1477                 ip_rt_multicast_event(in_dev);
1478 out:
1479         return;
1480 }
1481
1482 void __ip_mc_inc_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1483 {
1484         ____ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE, gfp);
1485 }
1486 EXPORT_SYMBOL(__ip_mc_inc_group);
1487
1488 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1489 {
1490         __ip_mc_inc_group(in_dev, addr, GFP_KERNEL);
1491 }
1492 EXPORT_SYMBOL(ip_mc_inc_group);
1493
1494 static int ip_mc_check_iphdr(struct sk_buff *skb)
1495 {
1496         const struct iphdr *iph;
1497         unsigned int len;
1498         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1499
1500         if (!pskb_may_pull(skb, offset))
1501                 return -EINVAL;
1502
1503         iph = ip_hdr(skb);
1504
1505         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1506                 return -EINVAL;
1507
1508         offset += ip_hdrlen(skb) - sizeof(*iph);
1509
1510         if (!pskb_may_pull(skb, offset))
1511                 return -EINVAL;
1512
1513         iph = ip_hdr(skb);
1514
1515         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1516                 return -EINVAL;
1517
1518         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1519         if (skb->len < len || len < offset)
1520                 return -EINVAL;
1521
1522         skb_set_transport_header(skb, offset);
1523
1524         return 0;
1525 }
1526
1527 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1528 {
1529         unsigned int len = skb_transport_offset(skb);
1530
1531         len += sizeof(struct igmpv3_report);
1532
1533         return ip_mc_may_pull(skb, len) ? 0 : -EINVAL;
1534 }
1535
1536 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1537 {
1538         unsigned int transport_len = ip_transport_len(skb);
1539         unsigned int len;
1540
1541         /* IGMPv{1,2}? */
1542         if (transport_len != sizeof(struct igmphdr)) {
1543                 /* or IGMPv3? */
1544                 if (transport_len < sizeof(struct igmpv3_query))
1545                         return -EINVAL;
1546
1547                 len = skb_transport_offset(skb) + sizeof(struct igmpv3_query);
1548                 if (!ip_mc_may_pull(skb, len))
1549                         return -EINVAL;
1550         }
1551
1552         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1553          * all-systems destination addresses (224.0.0.1) for general queries
1554          */
1555         if (!igmp_hdr(skb)->group &&
1556             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1557                 return -EINVAL;
1558
1559         return 0;
1560 }
1561
1562 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1563 {
1564         switch (igmp_hdr(skb)->type) {
1565         case IGMP_HOST_LEAVE_MESSAGE:
1566         case IGMP_HOST_MEMBERSHIP_REPORT:
1567         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1568                 return 0;
1569         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1570                 return ip_mc_check_igmp_reportv3(skb);
1571         case IGMP_HOST_MEMBERSHIP_QUERY:
1572                 return ip_mc_check_igmp_query(skb);
1573         default:
1574                 return -ENOMSG;
1575         }
1576 }
1577
1578 static __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1579 {
1580         return skb_checksum_simple_validate(skb);
1581 }
1582
1583 static int ip_mc_check_igmp_csum(struct sk_buff *skb)
1584 {
1585         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1586         unsigned int transport_len = ip_transport_len(skb);
1587         struct sk_buff *skb_chk;
1588
1589         if (!ip_mc_may_pull(skb, len))
1590                 return -EINVAL;
1591
1592         skb_chk = skb_checksum_trimmed(skb, transport_len,
1593                                        ip_mc_validate_checksum);
1594         if (!skb_chk)
1595                 return -EINVAL;
1596
1597         if (skb_chk != skb)
1598                 kfree_skb(skb_chk);
1599
1600         return 0;
1601 }
1602
1603 /**
1604  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1605  * @skb: the skb to validate
1606  *
1607  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1608  * skb transport header accordingly and returns zero.
1609  *
1610  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1611  *  standard
1612  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1613  * -ENOMEM: A memory allocation failure happened.
1614  *
1615  * Caller needs to set the skb network header and free any returned skb if it
1616  * differs from the provided skb.
1617  */
1618 int ip_mc_check_igmp(struct sk_buff *skb)
1619 {
1620         int ret = ip_mc_check_iphdr(skb);
1621
1622         if (ret < 0)
1623                 return ret;
1624
1625         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1626                 return -ENOMSG;
1627
1628         ret = ip_mc_check_igmp_csum(skb);
1629         if (ret < 0)
1630                 return ret;
1631
1632         return ip_mc_check_igmp_msg(skb);
1633 }
1634 EXPORT_SYMBOL(ip_mc_check_igmp);
1635
1636 /*
1637  *      Resend IGMP JOIN report; used by netdev notifier.
1638  */
1639 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1640 {
1641 #ifdef CONFIG_IP_MULTICAST
1642         struct ip_mc_list *im;
1643         int type;
1644         struct net *net = dev_net(in_dev->dev);
1645
1646         ASSERT_RTNL();
1647
1648         for_each_pmc_rtnl(in_dev, im) {
1649                 if (im->multiaddr == IGMP_ALL_HOSTS)
1650                         continue;
1651                 if (ipv4_is_local_multicast(im->multiaddr) &&
1652                     !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1653                         continue;
1654
1655                 /* a failover is happening and switches
1656                  * must be notified immediately
1657                  */
1658                 if (IGMP_V1_SEEN(in_dev))
1659                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1660                 else if (IGMP_V2_SEEN(in_dev))
1661                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1662                 else
1663                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1664                 igmp_send_report(in_dev, im, type);
1665         }
1666 #endif
1667 }
1668
1669 /*
1670  *      A socket has left a multicast group on device dev
1671  */
1672
1673 void __ip_mc_dec_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1674 {
1675         struct ip_mc_list *i;
1676         struct ip_mc_list __rcu **ip;
1677
1678         ASSERT_RTNL();
1679
1680         for (ip = &in_dev->mc_list;
1681              (i = rtnl_dereference(*ip)) != NULL;
1682              ip = &i->next_rcu) {
1683                 if (i->multiaddr == addr) {
1684                         if (--i->users == 0) {
1685                                 ip_mc_hash_remove(in_dev, i);
1686                                 *ip = i->next_rcu;
1687                                 in_dev->mc_count--;
1688                                 __igmp_group_dropped(i, gfp);
1689                                 ip_mc_clear_src(i);
1690
1691                                 if (!in_dev->dead)
1692                                         ip_rt_multicast_event(in_dev);
1693
1694                                 ip_ma_put(i);
1695                                 return;
1696                         }
1697                         break;
1698                 }
1699         }
1700 }
1701 EXPORT_SYMBOL(__ip_mc_dec_group);
1702
1703 /* Device changing type */
1704
1705 void ip_mc_unmap(struct in_device *in_dev)
1706 {
1707         struct ip_mc_list *pmc;
1708
1709         ASSERT_RTNL();
1710
1711         for_each_pmc_rtnl(in_dev, pmc)
1712                 igmp_group_dropped(pmc);
1713 }
1714
1715 void ip_mc_remap(struct in_device *in_dev)
1716 {
1717         struct ip_mc_list *pmc;
1718
1719         ASSERT_RTNL();
1720
1721         for_each_pmc_rtnl(in_dev, pmc) {
1722 #ifdef CONFIG_IP_MULTICAST
1723                 igmpv3_del_delrec(in_dev, pmc);
1724 #endif
1725                 igmp_group_added(pmc);
1726         }
1727 }
1728
1729 /* Device going down */
1730
1731 void ip_mc_down(struct in_device *in_dev)
1732 {
1733         struct ip_mc_list *pmc;
1734
1735         ASSERT_RTNL();
1736
1737         for_each_pmc_rtnl(in_dev, pmc)
1738                 igmp_group_dropped(pmc);
1739
1740 #ifdef CONFIG_IP_MULTICAST
1741         WRITE_ONCE(in_dev->mr_ifc_count, 0);
1742         if (del_timer(&in_dev->mr_ifc_timer))
1743                 __in_dev_put(in_dev);
1744         in_dev->mr_gq_running = 0;
1745         if (del_timer(&in_dev->mr_gq_timer))
1746                 __in_dev_put(in_dev);
1747 #endif
1748
1749         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1750 }
1751
1752 #ifdef CONFIG_IP_MULTICAST
1753 static void ip_mc_reset(struct in_device *in_dev)
1754 {
1755         struct net *net = dev_net(in_dev->dev);
1756
1757         in_dev->mr_qi = IGMP_QUERY_INTERVAL;
1758         in_dev->mr_qri = IGMP_QUERY_RESPONSE_INTERVAL;
1759         in_dev->mr_qrv = READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1760 }
1761 #else
1762 static void ip_mc_reset(struct in_device *in_dev)
1763 {
1764 }
1765 #endif
1766
1767 void ip_mc_init_dev(struct in_device *in_dev)
1768 {
1769         ASSERT_RTNL();
1770
1771 #ifdef CONFIG_IP_MULTICAST
1772         timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1773         timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1774 #endif
1775         ip_mc_reset(in_dev);
1776
1777         spin_lock_init(&in_dev->mc_tomb_lock);
1778 }
1779
1780 /* Device going up */
1781
1782 void ip_mc_up(struct in_device *in_dev)
1783 {
1784         struct ip_mc_list *pmc;
1785
1786         ASSERT_RTNL();
1787
1788         ip_mc_reset(in_dev);
1789         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1790
1791         for_each_pmc_rtnl(in_dev, pmc) {
1792 #ifdef CONFIG_IP_MULTICAST
1793                 igmpv3_del_delrec(in_dev, pmc);
1794 #endif
1795                 igmp_group_added(pmc);
1796         }
1797 }
1798
1799 /*
1800  *      Device is about to be destroyed: clean up.
1801  */
1802
1803 void ip_mc_destroy_dev(struct in_device *in_dev)
1804 {
1805         struct ip_mc_list *i;
1806
1807         ASSERT_RTNL();
1808
1809         /* Deactivate timers */
1810         ip_mc_down(in_dev);
1811 #ifdef CONFIG_IP_MULTICAST
1812         igmpv3_clear_delrec(in_dev);
1813 #endif
1814
1815         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1816                 in_dev->mc_list = i->next_rcu;
1817                 in_dev->mc_count--;
1818                 ip_mc_clear_src(i);
1819                 ip_ma_put(i);
1820         }
1821 }
1822
1823 /* RTNL is locked */
1824 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1825 {
1826         struct net_device *dev = NULL;
1827         struct in_device *idev = NULL;
1828
1829         if (imr->imr_ifindex) {
1830                 idev = inetdev_by_index(net, imr->imr_ifindex);
1831                 return idev;
1832         }
1833         if (imr->imr_address.s_addr) {
1834                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1835                 if (!dev)
1836                         return NULL;
1837         }
1838
1839         if (!dev) {
1840                 struct rtable *rt = ip_route_output(net,
1841                                                     imr->imr_multiaddr.s_addr,
1842                                                     0, 0, 0);
1843                 if (!IS_ERR(rt)) {
1844                         dev = rt->dst.dev;
1845                         ip_rt_put(rt);
1846                 }
1847         }
1848         if (dev) {
1849                 imr->imr_ifindex = dev->ifindex;
1850                 idev = __in_dev_get_rtnl(dev);
1851         }
1852         return idev;
1853 }
1854
1855 /*
1856  *      Join a socket to a group
1857  */
1858
1859 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1860         __be32 *psfsrc)
1861 {
1862         struct ip_sf_list *psf, *psf_prev;
1863         int rv = 0;
1864
1865         psf_prev = NULL;
1866         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1867                 if (psf->sf_inaddr == *psfsrc)
1868                         break;
1869                 psf_prev = psf;
1870         }
1871         if (!psf || psf->sf_count[sfmode] == 0) {
1872                 /* source filter not found, or count wrong =>  bug */
1873                 return -ESRCH;
1874         }
1875         psf->sf_count[sfmode]--;
1876         if (psf->sf_count[sfmode] == 0) {
1877                 ip_rt_multicast_event(pmc->interface);
1878         }
1879         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1880 #ifdef CONFIG_IP_MULTICAST
1881                 struct in_device *in_dev = pmc->interface;
1882                 struct net *net = dev_net(in_dev->dev);
1883 #endif
1884
1885                 /* no more filters for this source */
1886                 if (psf_prev)
1887                         psf_prev->sf_next = psf->sf_next;
1888                 else
1889                         pmc->sources = psf->sf_next;
1890 #ifdef CONFIG_IP_MULTICAST
1891                 if (psf->sf_oldin &&
1892                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1893                         psf->sf_crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1894                         psf->sf_next = pmc->tomb;
1895                         pmc->tomb = psf;
1896                         rv = 1;
1897                 } else
1898 #endif
1899                         kfree(psf);
1900         }
1901         return rv;
1902 }
1903
1904 #ifndef CONFIG_IP_MULTICAST
1905 #define igmp_ifc_event(x)       do { } while (0)
1906 #endif
1907
1908 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1909                          int sfcount, __be32 *psfsrc, int delta)
1910 {
1911         struct ip_mc_list *pmc;
1912         int     changerec = 0;
1913         int     i, err;
1914
1915         if (!in_dev)
1916                 return -ENODEV;
1917         rcu_read_lock();
1918         for_each_pmc_rcu(in_dev, pmc) {
1919                 if (*pmca == pmc->multiaddr)
1920                         break;
1921         }
1922         if (!pmc) {
1923                 /* MCA not found?? bug */
1924                 rcu_read_unlock();
1925                 return -ESRCH;
1926         }
1927         spin_lock_bh(&pmc->lock);
1928         rcu_read_unlock();
1929 #ifdef CONFIG_IP_MULTICAST
1930         sf_markstate(pmc);
1931 #endif
1932         if (!delta) {
1933                 err = -EINVAL;
1934                 if (!pmc->sfcount[sfmode])
1935                         goto out_unlock;
1936                 pmc->sfcount[sfmode]--;
1937         }
1938         err = 0;
1939         for (i = 0; i < sfcount; i++) {
1940                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1941
1942                 changerec |= rv > 0;
1943                 if (!err && rv < 0)
1944                         err = rv;
1945         }
1946         if (pmc->sfmode == MCAST_EXCLUDE &&
1947             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1948             pmc->sfcount[MCAST_INCLUDE]) {
1949 #ifdef CONFIG_IP_MULTICAST
1950                 struct ip_sf_list *psf;
1951                 struct net *net = dev_net(in_dev->dev);
1952 #endif
1953
1954                 /* filter mode change */
1955                 pmc->sfmode = MCAST_INCLUDE;
1956 #ifdef CONFIG_IP_MULTICAST
1957                 pmc->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1958                 WRITE_ONCE(in_dev->mr_ifc_count, pmc->crcount);
1959                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1960                         psf->sf_crcount = 0;
1961                 igmp_ifc_event(pmc->interface);
1962         } else if (sf_setstate(pmc) || changerec) {
1963                 igmp_ifc_event(pmc->interface);
1964 #endif
1965         }
1966 out_unlock:
1967         spin_unlock_bh(&pmc->lock);
1968         return err;
1969 }
1970
1971 /*
1972  * Add multicast single-source filter to the interface list
1973  */
1974 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1975         __be32 *psfsrc)
1976 {
1977         struct ip_sf_list *psf, *psf_prev;
1978
1979         psf_prev = NULL;
1980         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1981                 if (psf->sf_inaddr == *psfsrc)
1982                         break;
1983                 psf_prev = psf;
1984         }
1985         if (!psf) {
1986                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1987                 if (!psf)
1988                         return -ENOBUFS;
1989                 psf->sf_inaddr = *psfsrc;
1990                 if (psf_prev) {
1991                         psf_prev->sf_next = psf;
1992                 } else
1993                         pmc->sources = psf;
1994         }
1995         psf->sf_count[sfmode]++;
1996         if (psf->sf_count[sfmode] == 1) {
1997                 ip_rt_multicast_event(pmc->interface);
1998         }
1999         return 0;
2000 }
2001
2002 #ifdef CONFIG_IP_MULTICAST
2003 static void sf_markstate(struct ip_mc_list *pmc)
2004 {
2005         struct ip_sf_list *psf;
2006         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2007
2008         for (psf = pmc->sources; psf; psf = psf->sf_next)
2009                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2010                         psf->sf_oldin = mca_xcount ==
2011                                 psf->sf_count[MCAST_EXCLUDE] &&
2012                                 !psf->sf_count[MCAST_INCLUDE];
2013                 } else
2014                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
2015 }
2016
2017 static int sf_setstate(struct ip_mc_list *pmc)
2018 {
2019         struct ip_sf_list *psf, *dpsf;
2020         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2021         int qrv = pmc->interface->mr_qrv;
2022         int new_in, rv;
2023
2024         rv = 0;
2025         for (psf = pmc->sources; psf; psf = psf->sf_next) {
2026                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2027                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
2028                                 !psf->sf_count[MCAST_INCLUDE];
2029                 } else
2030                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
2031                 if (new_in) {
2032                         if (!psf->sf_oldin) {
2033                                 struct ip_sf_list *prev = NULL;
2034
2035                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2036                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2037                                                 break;
2038                                         prev = dpsf;
2039                                 }
2040                                 if (dpsf) {
2041                                         if (prev)
2042                                                 prev->sf_next = dpsf->sf_next;
2043                                         else
2044                                                 pmc->tomb = dpsf->sf_next;
2045                                         kfree(dpsf);
2046                                 }
2047                                 psf->sf_crcount = qrv;
2048                                 rv++;
2049                         }
2050                 } else if (psf->sf_oldin) {
2051
2052                         psf->sf_crcount = 0;
2053                         /*
2054                          * add or update "delete" records if an active filter
2055                          * is now inactive
2056                          */
2057                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2058                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2059                                         break;
2060                         if (!dpsf) {
2061                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2062                                 if (!dpsf)
2063                                         continue;
2064                                 *dpsf = *psf;
2065                                 /* pmc->lock held by callers */
2066                                 dpsf->sf_next = pmc->tomb;
2067                                 pmc->tomb = dpsf;
2068                         }
2069                         dpsf->sf_crcount = qrv;
2070                         rv++;
2071                 }
2072         }
2073         return rv;
2074 }
2075 #endif
2076
2077 /*
2078  * Add multicast source filter list to the interface list
2079  */
2080 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2081                          int sfcount, __be32 *psfsrc, int delta)
2082 {
2083         struct ip_mc_list *pmc;
2084         int     isexclude;
2085         int     i, err;
2086
2087         if (!in_dev)
2088                 return -ENODEV;
2089         rcu_read_lock();
2090         for_each_pmc_rcu(in_dev, pmc) {
2091                 if (*pmca == pmc->multiaddr)
2092                         break;
2093         }
2094         if (!pmc) {
2095                 /* MCA not found?? bug */
2096                 rcu_read_unlock();
2097                 return -ESRCH;
2098         }
2099         spin_lock_bh(&pmc->lock);
2100         rcu_read_unlock();
2101
2102 #ifdef CONFIG_IP_MULTICAST
2103         sf_markstate(pmc);
2104 #endif
2105         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2106         if (!delta)
2107                 pmc->sfcount[sfmode]++;
2108         err = 0;
2109         for (i = 0; i < sfcount; i++) {
2110                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2111                 if (err)
2112                         break;
2113         }
2114         if (err) {
2115                 int j;
2116
2117                 if (!delta)
2118                         pmc->sfcount[sfmode]--;
2119                 for (j = 0; j < i; j++)
2120                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2121         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2122 #ifdef CONFIG_IP_MULTICAST
2123                 struct ip_sf_list *psf;
2124                 struct net *net = dev_net(pmc->interface->dev);
2125                 in_dev = pmc->interface;
2126 #endif
2127
2128                 /* filter mode change */
2129                 if (pmc->sfcount[MCAST_EXCLUDE])
2130                         pmc->sfmode = MCAST_EXCLUDE;
2131                 else if (pmc->sfcount[MCAST_INCLUDE])
2132                         pmc->sfmode = MCAST_INCLUDE;
2133 #ifdef CONFIG_IP_MULTICAST
2134                 /* else no filters; keep old mode for reports */
2135
2136                 pmc->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
2137                 WRITE_ONCE(in_dev->mr_ifc_count, pmc->crcount);
2138                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2139                         psf->sf_crcount = 0;
2140                 igmp_ifc_event(in_dev);
2141         } else if (sf_setstate(pmc)) {
2142                 igmp_ifc_event(in_dev);
2143 #endif
2144         }
2145         spin_unlock_bh(&pmc->lock);
2146         return err;
2147 }
2148
2149 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2150 {
2151         struct ip_sf_list *tomb, *sources;
2152
2153         spin_lock_bh(&pmc->lock);
2154         tomb = pmc->tomb;
2155         pmc->tomb = NULL;
2156         sources = pmc->sources;
2157         pmc->sources = NULL;
2158         pmc->sfmode = MCAST_EXCLUDE;
2159         pmc->sfcount[MCAST_INCLUDE] = 0;
2160         pmc->sfcount[MCAST_EXCLUDE] = 1;
2161         spin_unlock_bh(&pmc->lock);
2162
2163         ip_sf_list_clear_all(tomb);
2164         ip_sf_list_clear_all(sources);
2165 }
2166
2167 /* Join a multicast group
2168  */
2169 static int __ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr,
2170                               unsigned int mode)
2171 {
2172         __be32 addr = imr->imr_multiaddr.s_addr;
2173         struct ip_mc_socklist *iml, *i;
2174         struct in_device *in_dev;
2175         struct inet_sock *inet = inet_sk(sk);
2176         struct net *net = sock_net(sk);
2177         int ifindex;
2178         int count = 0;
2179         int err;
2180
2181         ASSERT_RTNL();
2182
2183         if (!ipv4_is_multicast(addr))
2184                 return -EINVAL;
2185
2186         in_dev = ip_mc_find_dev(net, imr);
2187
2188         if (!in_dev) {
2189                 err = -ENODEV;
2190                 goto done;
2191         }
2192
2193         err = -EADDRINUSE;
2194         ifindex = imr->imr_ifindex;
2195         for_each_pmc_rtnl(inet, i) {
2196                 if (i->multi.imr_multiaddr.s_addr == addr &&
2197                     i->multi.imr_ifindex == ifindex)
2198                         goto done;
2199                 count++;
2200         }
2201         err = -ENOBUFS;
2202         if (count >= READ_ONCE(net->ipv4.sysctl_igmp_max_memberships))
2203                 goto done;
2204         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2205         if (!iml)
2206                 goto done;
2207
2208         memcpy(&iml->multi, imr, sizeof(*imr));
2209         iml->next_rcu = inet->mc_list;
2210         iml->sflist = NULL;
2211         iml->sfmode = mode;
2212         rcu_assign_pointer(inet->mc_list, iml);
2213         ____ip_mc_inc_group(in_dev, addr, mode, GFP_KERNEL);
2214         err = 0;
2215 done:
2216         return err;
2217 }
2218
2219 /* Join ASM (Any-Source Multicast) group
2220  */
2221 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2222 {
2223         return __ip_mc_join_group(sk, imr, MCAST_EXCLUDE);
2224 }
2225 EXPORT_SYMBOL(ip_mc_join_group);
2226
2227 /* Join SSM (Source-Specific Multicast) group
2228  */
2229 int ip_mc_join_group_ssm(struct sock *sk, struct ip_mreqn *imr,
2230                          unsigned int mode)
2231 {
2232         return __ip_mc_join_group(sk, imr, mode);
2233 }
2234
2235 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2236                            struct in_device *in_dev)
2237 {
2238         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2239         int err;
2240
2241         if (!psf) {
2242                 /* any-source empty exclude case */
2243                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2244                         iml->sfmode, 0, NULL, 0);
2245         }
2246         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2247                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2248         RCU_INIT_POINTER(iml->sflist, NULL);
2249         /* decrease mem now to avoid the memleak warning */
2250         atomic_sub(struct_size(psf, sl_addr, psf->sl_max), &sk->sk_omem_alloc);
2251         kfree_rcu(psf, rcu);
2252         return err;
2253 }
2254
2255 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2256 {
2257         struct inet_sock *inet = inet_sk(sk);
2258         struct ip_mc_socklist *iml;
2259         struct ip_mc_socklist __rcu **imlp;
2260         struct in_device *in_dev;
2261         struct net *net = sock_net(sk);
2262         __be32 group = imr->imr_multiaddr.s_addr;
2263         u32 ifindex;
2264         int ret = -EADDRNOTAVAIL;
2265
2266         ASSERT_RTNL();
2267
2268         in_dev = ip_mc_find_dev(net, imr);
2269         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2270                 ret = -ENODEV;
2271                 goto out;
2272         }
2273         ifindex = imr->imr_ifindex;
2274         for (imlp = &inet->mc_list;
2275              (iml = rtnl_dereference(*imlp)) != NULL;
2276              imlp = &iml->next_rcu) {
2277                 if (iml->multi.imr_multiaddr.s_addr != group)
2278                         continue;
2279                 if (ifindex) {
2280                         if (iml->multi.imr_ifindex != ifindex)
2281                                 continue;
2282                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2283                                 iml->multi.imr_address.s_addr)
2284                         continue;
2285
2286                 (void) ip_mc_leave_src(sk, iml, in_dev);
2287
2288                 *imlp = iml->next_rcu;
2289
2290                 if (in_dev)
2291                         ip_mc_dec_group(in_dev, group);
2292
2293                 /* decrease mem now to avoid the memleak warning */
2294                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2295                 kfree_rcu(iml, rcu);
2296                 return 0;
2297         }
2298 out:
2299         return ret;
2300 }
2301 EXPORT_SYMBOL(ip_mc_leave_group);
2302
2303 int ip_mc_source(int add, int omode, struct sock *sk, struct
2304         ip_mreq_source *mreqs, int ifindex)
2305 {
2306         int err;
2307         struct ip_mreqn imr;
2308         __be32 addr = mreqs->imr_multiaddr;
2309         struct ip_mc_socklist *pmc;
2310         struct in_device *in_dev = NULL;
2311         struct inet_sock *inet = inet_sk(sk);
2312         struct ip_sf_socklist *psl;
2313         struct net *net = sock_net(sk);
2314         int leavegroup = 0;
2315         int i, j, rv;
2316
2317         if (!ipv4_is_multicast(addr))
2318                 return -EINVAL;
2319
2320         ASSERT_RTNL();
2321
2322         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2323         imr.imr_address.s_addr = mreqs->imr_interface;
2324         imr.imr_ifindex = ifindex;
2325         in_dev = ip_mc_find_dev(net, &imr);
2326
2327         if (!in_dev) {
2328                 err = -ENODEV;
2329                 goto done;
2330         }
2331         err = -EADDRNOTAVAIL;
2332
2333         for_each_pmc_rtnl(inet, pmc) {
2334                 if ((pmc->multi.imr_multiaddr.s_addr ==
2335                      imr.imr_multiaddr.s_addr) &&
2336                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2337                         break;
2338         }
2339         if (!pmc) {             /* must have a prior join */
2340                 err = -EINVAL;
2341                 goto done;
2342         }
2343         /* if a source filter was set, must be the same mode as before */
2344         if (pmc->sflist) {
2345                 if (pmc->sfmode != omode) {
2346                         err = -EINVAL;
2347                         goto done;
2348                 }
2349         } else if (pmc->sfmode != omode) {
2350                 /* allow mode switches for empty-set filters */
2351                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2352                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2353                         NULL, 0);
2354                 pmc->sfmode = omode;
2355         }
2356
2357         psl = rtnl_dereference(pmc->sflist);
2358         if (!add) {
2359                 if (!psl)
2360                         goto done;      /* err = -EADDRNOTAVAIL */
2361                 rv = !0;
2362                 for (i = 0; i < psl->sl_count; i++) {
2363                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2364                                 sizeof(__be32));
2365                         if (rv == 0)
2366                                 break;
2367                 }
2368                 if (rv)         /* source not found */
2369                         goto done;      /* err = -EADDRNOTAVAIL */
2370
2371                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2372                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2373                         leavegroup = 1;
2374                         goto done;
2375                 }
2376
2377                 /* update the interface filter */
2378                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2379                         &mreqs->imr_sourceaddr, 1);
2380
2381                 for (j = i+1; j < psl->sl_count; j++)
2382                         psl->sl_addr[j-1] = psl->sl_addr[j];
2383                 psl->sl_count--;
2384                 err = 0;
2385                 goto done;
2386         }
2387         /* else, add a new source to the filter */
2388
2389         if (psl && psl->sl_count >= READ_ONCE(net->ipv4.sysctl_igmp_max_msf)) {
2390                 err = -ENOBUFS;
2391                 goto done;
2392         }
2393         if (!psl || psl->sl_count == psl->sl_max) {
2394                 struct ip_sf_socklist *newpsl;
2395                 int count = IP_SFBLOCK;
2396
2397                 if (psl)
2398                         count += psl->sl_max;
2399                 newpsl = sock_kmalloc(sk, struct_size(newpsl, sl_addr, count),
2400                                       GFP_KERNEL);
2401                 if (!newpsl) {
2402                         err = -ENOBUFS;
2403                         goto done;
2404                 }
2405                 newpsl->sl_max = count;
2406                 newpsl->sl_count = count - IP_SFBLOCK;
2407                 if (psl) {
2408                         for (i = 0; i < psl->sl_count; i++)
2409                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2410                         /* decrease mem now to avoid the memleak warning */
2411                         atomic_sub(struct_size(psl, sl_addr, psl->sl_max),
2412                                    &sk->sk_omem_alloc);
2413                 }
2414                 rcu_assign_pointer(pmc->sflist, newpsl);
2415                 if (psl)
2416                         kfree_rcu(psl, rcu);
2417                 psl = newpsl;
2418         }
2419         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2420         for (i = 0; i < psl->sl_count; i++) {
2421                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2422                         sizeof(__be32));
2423                 if (rv == 0)
2424                         break;
2425         }
2426         if (rv == 0)            /* address already there is an error */
2427                 goto done;
2428         for (j = psl->sl_count-1; j >= i; j--)
2429                 psl->sl_addr[j+1] = psl->sl_addr[j];
2430         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2431         psl->sl_count++;
2432         err = 0;
2433         /* update the interface list */
2434         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2435                 &mreqs->imr_sourceaddr, 1);
2436 done:
2437         if (leavegroup)
2438                 err = ip_mc_leave_group(sk, &imr);
2439         return err;
2440 }
2441
2442 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2443 {
2444         int err = 0;
2445         struct ip_mreqn imr;
2446         __be32 addr = msf->imsf_multiaddr;
2447         struct ip_mc_socklist *pmc;
2448         struct in_device *in_dev;
2449         struct inet_sock *inet = inet_sk(sk);
2450         struct ip_sf_socklist *newpsl, *psl;
2451         struct net *net = sock_net(sk);
2452         int leavegroup = 0;
2453
2454         if (!ipv4_is_multicast(addr))
2455                 return -EINVAL;
2456         if (msf->imsf_fmode != MCAST_INCLUDE &&
2457             msf->imsf_fmode != MCAST_EXCLUDE)
2458                 return -EINVAL;
2459
2460         ASSERT_RTNL();
2461
2462         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2463         imr.imr_address.s_addr = msf->imsf_interface;
2464         imr.imr_ifindex = ifindex;
2465         in_dev = ip_mc_find_dev(net, &imr);
2466
2467         if (!in_dev) {
2468                 err = -ENODEV;
2469                 goto done;
2470         }
2471
2472         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2473         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2474                 leavegroup = 1;
2475                 goto done;
2476         }
2477
2478         for_each_pmc_rtnl(inet, pmc) {
2479                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2480                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2481                         break;
2482         }
2483         if (!pmc) {             /* must have a prior join */
2484                 err = -EINVAL;
2485                 goto done;
2486         }
2487         if (msf->imsf_numsrc) {
2488                 newpsl = sock_kmalloc(sk, struct_size(newpsl, sl_addr,
2489                                                       msf->imsf_numsrc),
2490                                       GFP_KERNEL);
2491                 if (!newpsl) {
2492                         err = -ENOBUFS;
2493                         goto done;
2494                 }
2495                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2496                 memcpy(newpsl->sl_addr, msf->imsf_slist_flex,
2497                        flex_array_size(msf, imsf_slist_flex, msf->imsf_numsrc));
2498                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2499                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2500                 if (err) {
2501                         sock_kfree_s(sk, newpsl,
2502                                      struct_size(newpsl, sl_addr,
2503                                                  newpsl->sl_max));
2504                         goto done;
2505                 }
2506         } else {
2507                 newpsl = NULL;
2508                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2509                                      msf->imsf_fmode, 0, NULL, 0);
2510         }
2511         psl = rtnl_dereference(pmc->sflist);
2512         if (psl) {
2513                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2514                         psl->sl_count, psl->sl_addr, 0);
2515                 /* decrease mem now to avoid the memleak warning */
2516                 atomic_sub(struct_size(psl, sl_addr, psl->sl_max),
2517                            &sk->sk_omem_alloc);
2518         } else {
2519                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2520                         0, NULL, 0);
2521         }
2522         rcu_assign_pointer(pmc->sflist, newpsl);
2523         if (psl)
2524                 kfree_rcu(psl, rcu);
2525         pmc->sfmode = msf->imsf_fmode;
2526         err = 0;
2527 done:
2528         if (leavegroup)
2529                 err = ip_mc_leave_group(sk, &imr);
2530         return err;
2531 }
2532 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2533                  sockptr_t optval, sockptr_t optlen)
2534 {
2535         int err, len, count, copycount, msf_size;
2536         struct ip_mreqn imr;
2537         __be32 addr = msf->imsf_multiaddr;
2538         struct ip_mc_socklist *pmc;
2539         struct in_device *in_dev;
2540         struct inet_sock *inet = inet_sk(sk);
2541         struct ip_sf_socklist *psl;
2542         struct net *net = sock_net(sk);
2543
2544         ASSERT_RTNL();
2545
2546         if (!ipv4_is_multicast(addr))
2547                 return -EINVAL;
2548
2549         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2550         imr.imr_address.s_addr = msf->imsf_interface;
2551         imr.imr_ifindex = 0;
2552         in_dev = ip_mc_find_dev(net, &imr);
2553
2554         if (!in_dev) {
2555                 err = -ENODEV;
2556                 goto done;
2557         }
2558         err = -EADDRNOTAVAIL;
2559
2560         for_each_pmc_rtnl(inet, pmc) {
2561                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2562                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2563                         break;
2564         }
2565         if (!pmc)               /* must have a prior join */
2566                 goto done;
2567         msf->imsf_fmode = pmc->sfmode;
2568         psl = rtnl_dereference(pmc->sflist);
2569         if (!psl) {
2570                 count = 0;
2571         } else {
2572                 count = psl->sl_count;
2573         }
2574         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2575         len = flex_array_size(psl, sl_addr, copycount);
2576         msf->imsf_numsrc = count;
2577         msf_size = IP_MSFILTER_SIZE(copycount);
2578         if (copy_to_sockptr(optlen, &msf_size, sizeof(int)) ||
2579             copy_to_sockptr(optval, msf, IP_MSFILTER_SIZE(0))) {
2580                 return -EFAULT;
2581         }
2582         if (len &&
2583             copy_to_sockptr_offset(optval,
2584                                    offsetof(struct ip_msfilter, imsf_slist_flex),
2585                                    psl->sl_addr, len))
2586                 return -EFAULT;
2587         return 0;
2588 done:
2589         return err;
2590 }
2591
2592 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2593                  sockptr_t optval, size_t ss_offset)
2594 {
2595         int i, count, copycount;
2596         struct sockaddr_in *psin;
2597         __be32 addr;
2598         struct ip_mc_socklist *pmc;
2599         struct inet_sock *inet = inet_sk(sk);
2600         struct ip_sf_socklist *psl;
2601
2602         ASSERT_RTNL();
2603
2604         psin = (struct sockaddr_in *)&gsf->gf_group;
2605         if (psin->sin_family != AF_INET)
2606                 return -EINVAL;
2607         addr = psin->sin_addr.s_addr;
2608         if (!ipv4_is_multicast(addr))
2609                 return -EINVAL;
2610
2611         for_each_pmc_rtnl(inet, pmc) {
2612                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2613                     pmc->multi.imr_ifindex == gsf->gf_interface)
2614                         break;
2615         }
2616         if (!pmc)               /* must have a prior join */
2617                 return -EADDRNOTAVAIL;
2618         gsf->gf_fmode = pmc->sfmode;
2619         psl = rtnl_dereference(pmc->sflist);
2620         count = psl ? psl->sl_count : 0;
2621         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2622         gsf->gf_numsrc = count;
2623         for (i = 0; i < copycount; i++) {
2624                 struct sockaddr_storage ss;
2625
2626                 psin = (struct sockaddr_in *)&ss;
2627                 memset(&ss, 0, sizeof(ss));
2628                 psin->sin_family = AF_INET;
2629                 psin->sin_addr.s_addr = psl->sl_addr[i];
2630                 if (copy_to_sockptr_offset(optval, ss_offset,
2631                                            &ss, sizeof(ss)))
2632                         return -EFAULT;
2633                 ss_offset += sizeof(ss);
2634         }
2635         return 0;
2636 }
2637
2638 /*
2639  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2640  */
2641 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2642                    int dif, int sdif)
2643 {
2644         struct inet_sock *inet = inet_sk(sk);
2645         struct ip_mc_socklist *pmc;
2646         struct ip_sf_socklist *psl;
2647         int i;
2648         int ret;
2649
2650         ret = 1;
2651         if (!ipv4_is_multicast(loc_addr))
2652                 goto out;
2653
2654         rcu_read_lock();
2655         for_each_pmc_rcu(inet, pmc) {
2656                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2657                     (pmc->multi.imr_ifindex == dif ||
2658                      (sdif && pmc->multi.imr_ifindex == sdif)))
2659                         break;
2660         }
2661         ret = inet->mc_all;
2662         if (!pmc)
2663                 goto unlock;
2664         psl = rcu_dereference(pmc->sflist);
2665         ret = (pmc->sfmode == MCAST_EXCLUDE);
2666         if (!psl)
2667                 goto unlock;
2668
2669         for (i = 0; i < psl->sl_count; i++) {
2670                 if (psl->sl_addr[i] == rmt_addr)
2671                         break;
2672         }
2673         ret = 0;
2674         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2675                 goto unlock;
2676         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2677                 goto unlock;
2678         ret = 1;
2679 unlock:
2680         rcu_read_unlock();
2681 out:
2682         return ret;
2683 }
2684
2685 /*
2686  *      A socket is closing.
2687  */
2688
2689 void ip_mc_drop_socket(struct sock *sk)
2690 {
2691         struct inet_sock *inet = inet_sk(sk);
2692         struct ip_mc_socklist *iml;
2693         struct net *net = sock_net(sk);
2694
2695         if (!inet->mc_list)
2696                 return;
2697
2698         rtnl_lock();
2699         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2700                 struct in_device *in_dev;
2701
2702                 inet->mc_list = iml->next_rcu;
2703                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2704                 (void) ip_mc_leave_src(sk, iml, in_dev);
2705                 if (in_dev)
2706                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2707                 /* decrease mem now to avoid the memleak warning */
2708                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2709                 kfree_rcu(iml, rcu);
2710         }
2711         rtnl_unlock();
2712 }
2713
2714 /* called with rcu_read_lock() */
2715 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2716 {
2717         struct ip_mc_list *im;
2718         struct ip_mc_list __rcu **mc_hash;
2719         struct ip_sf_list *psf;
2720         int rv = 0;
2721
2722         mc_hash = rcu_dereference(in_dev->mc_hash);
2723         if (mc_hash) {
2724                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2725
2726                 for (im = rcu_dereference(mc_hash[hash]);
2727                      im != NULL;
2728                      im = rcu_dereference(im->next_hash)) {
2729                         if (im->multiaddr == mc_addr)
2730                                 break;
2731                 }
2732         } else {
2733                 for_each_pmc_rcu(in_dev, im) {
2734                         if (im->multiaddr == mc_addr)
2735                                 break;
2736                 }
2737         }
2738         if (im && proto == IPPROTO_IGMP) {
2739                 rv = 1;
2740         } else if (im) {
2741                 if (src_addr) {
2742                         spin_lock_bh(&im->lock);
2743                         for (psf = im->sources; psf; psf = psf->sf_next) {
2744                                 if (psf->sf_inaddr == src_addr)
2745                                         break;
2746                         }
2747                         if (psf)
2748                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2749                                         psf->sf_count[MCAST_EXCLUDE] !=
2750                                         im->sfcount[MCAST_EXCLUDE];
2751                         else
2752                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2753                         spin_unlock_bh(&im->lock);
2754                 } else
2755                         rv = 1; /* unspecified source; tentatively allow */
2756         }
2757         return rv;
2758 }
2759
2760 #if defined(CONFIG_PROC_FS)
2761 struct igmp_mc_iter_state {
2762         struct seq_net_private p;
2763         struct net_device *dev;
2764         struct in_device *in_dev;
2765 };
2766
2767 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2768
2769 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2770 {
2771         struct net *net = seq_file_net(seq);
2772         struct ip_mc_list *im = NULL;
2773         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2774
2775         state->in_dev = NULL;
2776         for_each_netdev_rcu(net, state->dev) {
2777                 struct in_device *in_dev;
2778
2779                 in_dev = __in_dev_get_rcu(state->dev);
2780                 if (!in_dev)
2781                         continue;
2782                 im = rcu_dereference(in_dev->mc_list);
2783                 if (im) {
2784                         state->in_dev = in_dev;
2785                         break;
2786                 }
2787         }
2788         return im;
2789 }
2790
2791 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2792 {
2793         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2794
2795         im = rcu_dereference(im->next_rcu);
2796         while (!im) {
2797                 state->dev = next_net_device_rcu(state->dev);
2798                 if (!state->dev) {
2799                         state->in_dev = NULL;
2800                         break;
2801                 }
2802                 state->in_dev = __in_dev_get_rcu(state->dev);
2803                 if (!state->in_dev)
2804                         continue;
2805                 im = rcu_dereference(state->in_dev->mc_list);
2806         }
2807         return im;
2808 }
2809
2810 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2811 {
2812         struct ip_mc_list *im = igmp_mc_get_first(seq);
2813         if (im)
2814                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2815                         --pos;
2816         return pos ? NULL : im;
2817 }
2818
2819 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2820         __acquires(rcu)
2821 {
2822         rcu_read_lock();
2823         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2824 }
2825
2826 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2827 {
2828         struct ip_mc_list *im;
2829         if (v == SEQ_START_TOKEN)
2830                 im = igmp_mc_get_first(seq);
2831         else
2832                 im = igmp_mc_get_next(seq, v);
2833         ++*pos;
2834         return im;
2835 }
2836
2837 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2838         __releases(rcu)
2839 {
2840         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2841
2842         state->in_dev = NULL;
2843         state->dev = NULL;
2844         rcu_read_unlock();
2845 }
2846
2847 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2848 {
2849         if (v == SEQ_START_TOKEN)
2850                 seq_puts(seq,
2851                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2852         else {
2853                 struct ip_mc_list *im = v;
2854                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2855                 char   *querier;
2856                 long delta;
2857
2858 #ifdef CONFIG_IP_MULTICAST
2859                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2860                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2861                           "V3";
2862 #else
2863                 querier = "NONE";
2864 #endif
2865
2866                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2867                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2868                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2869                 }
2870
2871                 delta = im->timer.expires - jiffies;
2872                 seq_printf(seq,
2873                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2874                            im->multiaddr, im->users,
2875                            im->tm_running,
2876                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2877                            im->reporter);
2878         }
2879         return 0;
2880 }
2881
2882 static const struct seq_operations igmp_mc_seq_ops = {
2883         .start  =       igmp_mc_seq_start,
2884         .next   =       igmp_mc_seq_next,
2885         .stop   =       igmp_mc_seq_stop,
2886         .show   =       igmp_mc_seq_show,
2887 };
2888
2889 struct igmp_mcf_iter_state {
2890         struct seq_net_private p;
2891         struct net_device *dev;
2892         struct in_device *idev;
2893         struct ip_mc_list *im;
2894 };
2895
2896 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2897
2898 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2899 {
2900         struct net *net = seq_file_net(seq);
2901         struct ip_sf_list *psf = NULL;
2902         struct ip_mc_list *im = NULL;
2903         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2904
2905         state->idev = NULL;
2906         state->im = NULL;
2907         for_each_netdev_rcu(net, state->dev) {
2908                 struct in_device *idev;
2909                 idev = __in_dev_get_rcu(state->dev);
2910                 if (unlikely(!idev))
2911                         continue;
2912                 im = rcu_dereference(idev->mc_list);
2913                 if (likely(im)) {
2914                         spin_lock_bh(&im->lock);
2915                         psf = im->sources;
2916                         if (likely(psf)) {
2917                                 state->im = im;
2918                                 state->idev = idev;
2919                                 break;
2920                         }
2921                         spin_unlock_bh(&im->lock);
2922                 }
2923         }
2924         return psf;
2925 }
2926
2927 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2928 {
2929         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2930
2931         psf = psf->sf_next;
2932         while (!psf) {
2933                 spin_unlock_bh(&state->im->lock);
2934                 state->im = state->im->next;
2935                 while (!state->im) {
2936                         state->dev = next_net_device_rcu(state->dev);
2937                         if (!state->dev) {
2938                                 state->idev = NULL;
2939                                 goto out;
2940                         }
2941                         state->idev = __in_dev_get_rcu(state->dev);
2942                         if (!state->idev)
2943                                 continue;
2944                         state->im = rcu_dereference(state->idev->mc_list);
2945                 }
2946                 if (!state->im)
2947                         break;
2948                 spin_lock_bh(&state->im->lock);
2949                 psf = state->im->sources;
2950         }
2951 out:
2952         return psf;
2953 }
2954
2955 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2956 {
2957         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2958         if (psf)
2959                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2960                         --pos;
2961         return pos ? NULL : psf;
2962 }
2963
2964 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2965         __acquires(rcu)
2966 {
2967         rcu_read_lock();
2968         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2969 }
2970
2971 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2972 {
2973         struct ip_sf_list *psf;
2974         if (v == SEQ_START_TOKEN)
2975                 psf = igmp_mcf_get_first(seq);
2976         else
2977                 psf = igmp_mcf_get_next(seq, v);
2978         ++*pos;
2979         return psf;
2980 }
2981
2982 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2983         __releases(rcu)
2984 {
2985         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2986         if (likely(state->im)) {
2987                 spin_unlock_bh(&state->im->lock);
2988                 state->im = NULL;
2989         }
2990         state->idev = NULL;
2991         state->dev = NULL;
2992         rcu_read_unlock();
2993 }
2994
2995 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2996 {
2997         struct ip_sf_list *psf = v;
2998         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2999
3000         if (v == SEQ_START_TOKEN) {
3001                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
3002         } else {
3003                 seq_printf(seq,
3004                            "%3d %6.6s 0x%08x "
3005                            "0x%08x %6lu %6lu\n",
3006                            state->dev->ifindex, state->dev->name,
3007                            ntohl(state->im->multiaddr),
3008                            ntohl(psf->sf_inaddr),
3009                            psf->sf_count[MCAST_INCLUDE],
3010                            psf->sf_count[MCAST_EXCLUDE]);
3011         }
3012         return 0;
3013 }
3014
3015 static const struct seq_operations igmp_mcf_seq_ops = {
3016         .start  =       igmp_mcf_seq_start,
3017         .next   =       igmp_mcf_seq_next,
3018         .stop   =       igmp_mcf_seq_stop,
3019         .show   =       igmp_mcf_seq_show,
3020 };
3021
3022 static int __net_init igmp_net_init(struct net *net)
3023 {
3024         struct proc_dir_entry *pde;
3025         int err;
3026
3027         pde = proc_create_net("igmp", 0444, net->proc_net, &igmp_mc_seq_ops,
3028                         sizeof(struct igmp_mc_iter_state));
3029         if (!pde)
3030                 goto out_igmp;
3031         pde = proc_create_net("mcfilter", 0444, net->proc_net,
3032                         &igmp_mcf_seq_ops, sizeof(struct igmp_mcf_iter_state));
3033         if (!pde)
3034                 goto out_mcfilter;
3035         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3036                                    SOCK_DGRAM, 0, net);
3037         if (err < 0) {
3038                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3039                        err);
3040                 goto out_sock;
3041         }
3042
3043         return 0;
3044
3045 out_sock:
3046         remove_proc_entry("mcfilter", net->proc_net);
3047 out_mcfilter:
3048         remove_proc_entry("igmp", net->proc_net);
3049 out_igmp:
3050         return -ENOMEM;
3051 }
3052
3053 static void __net_exit igmp_net_exit(struct net *net)
3054 {
3055         remove_proc_entry("mcfilter", net->proc_net);
3056         remove_proc_entry("igmp", net->proc_net);
3057         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3058 }
3059
3060 static struct pernet_operations igmp_net_ops = {
3061         .init = igmp_net_init,
3062         .exit = igmp_net_exit,
3063 };
3064 #endif
3065
3066 static int igmp_netdev_event(struct notifier_block *this,
3067                              unsigned long event, void *ptr)
3068 {
3069         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3070         struct in_device *in_dev;
3071
3072         switch (event) {
3073         case NETDEV_RESEND_IGMP:
3074                 in_dev = __in_dev_get_rtnl(dev);
3075                 if (in_dev)
3076                         ip_mc_rejoin_groups(in_dev);
3077                 break;
3078         default:
3079                 break;
3080         }
3081         return NOTIFY_DONE;
3082 }
3083
3084 static struct notifier_block igmp_notifier = {
3085         .notifier_call = igmp_netdev_event,
3086 };
3087
3088 int __init igmp_mc_init(void)
3089 {
3090 #if defined(CONFIG_PROC_FS)
3091         int err;
3092
3093         err = register_pernet_subsys(&igmp_net_ops);
3094         if (err)
3095                 return err;
3096         err = register_netdevice_notifier(&igmp_notifier);
3097         if (err)
3098                 goto reg_notif_fail;
3099         return 0;
3100
3101 reg_notif_fail:
3102         unregister_pernet_subsys(&igmp_net_ops);
3103         return err;
3104 #else
3105         return register_netdevice_notifier(&igmp_notifier);
3106 #endif
3107 }