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 <alan@lxorguk.ukuu.org.uk>
  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: chihjenc@scf.usc.edu and tsusheng@scf.usc.edu
  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_RO(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_RO(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 (refcount_inc_not_zero(&im->refcnt)) {
 220                 if (mod_timer(&im->timer, jiffies + tv + 2))
 221                         ip_ma_put(im);
 222         }
 223 }
 224
 225 static void igmp_gq_start_timer(struct in_device *in_dev)
 226 {
 227         int tv = get_random_u32_below(in_dev->mr_maxdelay);
 228         unsigned long exp = jiffies + tv + 2;
 229
 230         if (in_dev->mr_gq_running &&
 231             time_after_eq(exp, (in_dev->mr_gq_timer).expires))
 232                 return;
 233
 234         in_dev->mr_gq_running = 1;
 235         if (!mod_timer(&in_dev->mr_gq_timer, exp))
 236                 in_dev_hold(in_dev);
 237 }
 238
 239 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 240 {
 241         int tv = get_random_u32_below(delay);
 242
 243         if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
 244                 in_dev_hold(in_dev);
 245 }
 246
 247 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 248 {
 249         spin_lock_bh(&im->lock);
 250         im->unsolicit_count = 0;
 251         if (del_timer(&im->timer)) {
 252                 if ((long)(im->timer.expires-jiffies) < max_delay) {
 253                         add_timer(&im->timer);
 254                         im->tm_running = 1;
 255                         spin_unlock_bh(&im->lock);
 256                         return;
 257                 }
 258                 refcount_dec(&im->refcnt);
 259         }
 260         igmp_start_timer(im, max_delay);
 261         spin_unlock_bh(&im->lock);
 262 }
 263
 264
 265 /*
 266  *      Send an IGMP report.
 267  */
 268
 269 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
 270
 271
 272 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
 273         int gdeleted, int sdeleted)
 274 {
 275         switch (type) {
 276         case IGMPV3_MODE_IS_INCLUDE:
 277         case IGMPV3_MODE_IS_EXCLUDE:
 278                 if (gdeleted || sdeleted)
 279                         return 0;
 280                 if (!(pmc->gsquery && !psf->sf_gsresp)) {
 281                         if (pmc->sfmode == MCAST_INCLUDE)
 282                                 return 1;
 283                         /* don't include if this source is excluded
 284                          * in all filters
 285                          */
 286                         if (psf->sf_count[MCAST_INCLUDE])
 287                                 return type == IGMPV3_MODE_IS_INCLUDE;
 288                         return pmc->sfcount[MCAST_EXCLUDE] ==
 289                                 psf->sf_count[MCAST_EXCLUDE];
 290                 }
 291                 return 0;
 292         case IGMPV3_CHANGE_TO_INCLUDE:
 293                 if (gdeleted || sdeleted)
 294                         return 0;
 295                 return psf->sf_count[MCAST_INCLUDE] != 0;
 296         case IGMPV3_CHANGE_TO_EXCLUDE:
 297                 if (gdeleted || sdeleted)
 298                         return 0;
 299                 if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
 300                     psf->sf_count[MCAST_INCLUDE])
 301                         return 0;
 302                 return pmc->sfcount[MCAST_EXCLUDE] ==
 303                         psf->sf_count[MCAST_EXCLUDE];
 304         case IGMPV3_ALLOW_NEW_SOURCES:
 305                 if (gdeleted || !psf->sf_crcount)
 306                         return 0;
 307                 return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
 308         case IGMPV3_BLOCK_OLD_SOURCES:
 309                 if (pmc->sfmode == MCAST_INCLUDE)
 310                         return gdeleted || (psf->sf_crcount && sdeleted);
 311                 return psf->sf_crcount && !gdeleted && !sdeleted;
 312         }
 313         return 0;
 314 }
 315
 316 static int
 317 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
 318 {
 319         struct ip_sf_list *psf;
 320         int scount = 0;
 321
 322         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 323                 if (!is_in(pmc, psf, type, gdeleted, sdeleted))
 324                         continue;
 325                 scount++;
 326         }
 327         return scount;
 328 }
 329
 330 /* source address selection per RFC 3376 section 4.2.13 */
 331 static __be32 igmpv3_get_srcaddr(struct net_device *dev,
 332                                  const struct flowi4 *fl4)
 333 {
 334         struct in_device *in_dev = __in_dev_get_rcu(dev);
 335         const struct in_ifaddr *ifa;
 336
 337         if (!in_dev)
 338                 return htonl(INADDR_ANY);
 339
 340         in_dev_for_each_ifa_rcu(ifa, in_dev) {
 341                 if (fl4->saddr == ifa->ifa_local)
 342                         return fl4->saddr;
 343         }
 344
 345         return htonl(INADDR_ANY);
 346 }
 347
 348 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
 349 {
 350         struct sk_buff *skb;
 351         struct rtable *rt;
 352         struct iphdr *pip;
 353         struct igmpv3_report *pig;
 354         struct net *net = dev_net(dev);
 355         struct flowi4 fl4;
 356         int hlen = LL_RESERVED_SPACE(dev);
 357         int tlen = dev->needed_tailroom;
 358         unsigned int size;
 359
 360         size = min(mtu, IP_MAX_MTU);
 361         while (1) {
 362                 skb = alloc_skb(size + hlen + tlen,
 363                                 GFP_ATOMIC | __GFP_NOWARN);
 364                 if (skb)
 365                         break;
 366                 size >>= 1;
 367                 if (size < 256)
 368                         return NULL;
 369         }
 370         skb->priority = TC_PRIO_CONTROL;
 371
 372         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 373                                    0, 0,
 374                                    IPPROTO_IGMP, 0, dev->ifindex);
 375         if (IS_ERR(rt)) {
 376                 kfree_skb(skb);
 377                 return NULL;
 378         }
 379
 380         skb_dst_set(skb, &rt->dst);
 381         skb->dev = dev;
 382
 383         skb_reserve(skb, hlen);
 384         skb_tailroom_reserve(skb, mtu, tlen);
 385
 386         skb_reset_network_header(skb);
 387         pip = ip_hdr(skb);
 388         skb_put(skb, sizeof(struct iphdr) + 4);
 389
 390         pip->version  = 4;
 391         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 392         pip->tos      = 0xc0;
 393         pip->frag_off = htons(IP_DF);
 394         pip->ttl      = 1;
 395         pip->daddr    = fl4.daddr;
 396
 397         rcu_read_lock();
 398         pip->saddr    = igmpv3_get_srcaddr(dev, &fl4);
 399         rcu_read_unlock();
 400
 401         pip->protocol = IPPROTO_IGMP;
 402         pip->tot_len  = 0;      /* filled in later */
 403         ip_select_ident(net, skb, NULL);
 404         ((u8 *)&pip[1])[0] = IPOPT_RA;
 405         ((u8 *)&pip[1])[1] = 4;
 406         ((u8 *)&pip[1])[2] = 0;
 407         ((u8 *)&pip[1])[3] = 0;
 408
 409         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 410         skb_put(skb, sizeof(*pig));
 411         pig = igmpv3_report_hdr(skb);
 412         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 413         pig->resv1 = 0;
 414         pig->csum = 0;
 415         pig->resv2 = 0;
 416         pig->ngrec = 0;
 417         return skb;
 418 }
 419
 420 static int igmpv3_sendpack(struct sk_buff *skb)
 421 {
 422         struct igmphdr *pig = igmp_hdr(skb);
 423         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 424
 425         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 426
 427         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 428 }
 429
 430 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 431 {
 432         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 433 }
 434
 435 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 436         int type, struct igmpv3_grec **ppgr, unsigned int mtu)
 437 {
 438         struct net_device *dev = pmc->interface->dev;
 439         struct igmpv3_report *pih;
 440         struct igmpv3_grec *pgr;
 441
 442         if (!skb) {
 443                 skb = igmpv3_newpack(dev, mtu);
 444                 if (!skb)
 445                         return NULL;
 446         }
 447         pgr = skb_put(skb, sizeof(struct igmpv3_grec));
 448         pgr->grec_type = type;
 449         pgr->grec_auxwords = 0;
 450         pgr->grec_nsrcs = 0;
 451         pgr->grec_mca = pmc->multiaddr;
 452         pih = igmpv3_report_hdr(skb);
 453         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 454         *ppgr = pgr;
 455         return skb;
 456 }
 457
 458 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 459
 460 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 461         int type, int gdeleted, int sdeleted)
 462 {
 463         struct net_device *dev = pmc->interface->dev;
 464         struct net *net = dev_net(dev);
 465         struct igmpv3_report *pih;
 466         struct igmpv3_grec *pgr = NULL;
 467         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 468         int scount, stotal, first, isquery, truncate;
 469         unsigned int mtu;
 470
 471         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 472                 return skb;
 473         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 474             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 475                 return skb;
 476
 477         mtu = READ_ONCE(dev->mtu);
 478         if (mtu < IPV4_MIN_MTU)
 479                 return skb;
 480
 481         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 482                   type == IGMPV3_MODE_IS_EXCLUDE;
 483         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 484                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 485
 486         stotal = scount = 0;
 487
 488         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 489
 490         if (!*psf_list)
 491                 goto empty_source;
 492
 493         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 494
 495         /* EX and TO_EX get a fresh packet, if needed */
 496         if (truncate) {
 497                 if (pih && pih->ngrec &&
 498                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 499                         if (skb)
 500                                 igmpv3_sendpack(skb);
 501                         skb = igmpv3_newpack(dev, mtu);
 502                 }
 503         }
 504         first = 1;
 505         psf_prev = NULL;
 506         for (psf = *psf_list; psf; psf = psf_next) {
 507                 __be32 *psrc;
 508
 509                 psf_next = psf->sf_next;
 510
 511                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 512                         psf_prev = psf;
 513                         continue;
 514                 }
 515
 516                 /* Based on RFC3376 5.1. Should not send source-list change
 517                  * records when there is a filter mode change.
 518                  */
 519                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 520                      (!gdeleted && pmc->crcount)) &&
 521                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 522                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 523                         goto decrease_sf_crcount;
 524
 525                 /* clear marks on query responses */
 526                 if (isquery)
 527                         psf->sf_gsresp = 0;
 528
 529                 if (AVAILABLE(skb) < sizeof(__be32) +
 530                     first*sizeof(struct igmpv3_grec)) {
 531                         if (truncate && !first)
 532                                 break;   /* truncate these */
 533                         if (pgr)
 534                                 pgr->grec_nsrcs = htons(scount);
 535                         if (skb)
 536                                 igmpv3_sendpack(skb);
 537                         skb = igmpv3_newpack(dev, mtu);
 538                         first = 1;
 539                         scount = 0;
 540                 }
 541                 if (first) {
 542                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 543                         first = 0;
 544                 }
 545                 if (!skb)
 546                         return NULL;
 547                 psrc = skb_put(skb, sizeof(__be32));
 548                 *psrc = psf->sf_inaddr;
 549                 scount++; stotal++;
 550                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 551                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 552 decrease_sf_crcount:
 553                         psf->sf_crcount--;
 554                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 555                                 if (psf_prev)
 556                                         psf_prev->sf_next = psf->sf_next;
 557                                 else
 558                                         *psf_list = psf->sf_next;
 559                                 kfree(psf);
 560                                 continue;
 561                         }
 562                 }
 563                 psf_prev = psf;
 564         }
 565
 566 empty_source:
 567         if (!stotal) {
 568                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 569                     type == IGMPV3_BLOCK_OLD_SOURCES)
 570                         return skb;
 571                 if (pmc->crcount || isquery) {
 572                         /* make sure we have room for group header */
 573                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 574                                 igmpv3_sendpack(skb);
 575                                 skb = NULL; /* add_grhead will get a new one */
 576                         }
 577                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 578                 }
 579         }
 580         if (pgr)
 581                 pgr->grec_nsrcs = htons(scount);
 582
 583         if (isquery)
 584                 pmc->gsquery = 0;       /* clear query state on report */
 585         return skb;
 586 }
 587
 588 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 589 {
 590         struct sk_buff *skb = NULL;
 591         struct net *net = dev_net(in_dev->dev);
 592         int type;
 593
 594         if (!pmc) {
 595                 rcu_read_lock();
 596                 for_each_pmc_rcu(in_dev, pmc) {
 597                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 598                                 continue;
 599                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 600                             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 601                                 continue;
 602                         spin_lock_bh(&pmc->lock);
 603                         if (pmc->sfcount[MCAST_EXCLUDE])
 604                                 type = IGMPV3_MODE_IS_EXCLUDE;
 605                         else
 606                                 type = IGMPV3_MODE_IS_INCLUDE;
 607                         skb = add_grec(skb, pmc, type, 0, 0);
 608                         spin_unlock_bh(&pmc->lock);
 609                 }
 610                 rcu_read_unlock();
 611         } else {
 612                 spin_lock_bh(&pmc->lock);
 613                 if (pmc->sfcount[MCAST_EXCLUDE])
 614                         type = IGMPV3_MODE_IS_EXCLUDE;
 615                 else
 616                         type = IGMPV3_MODE_IS_INCLUDE;
 617                 skb = add_grec(skb, pmc, type, 0, 0);
 618                 spin_unlock_bh(&pmc->lock);
 619         }
 620         if (!skb)
 621                 return 0;
 622         return igmpv3_sendpack(skb);
 623 }
 624
 625 /*
 626  * remove zero-count source records from a source filter list
 627  */
 628 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 629 {
 630         struct ip_sf_list *psf_prev, *psf_next, *psf;
 631
 632         psf_prev = NULL;
 633         for (psf = *ppsf; psf; psf = psf_next) {
 634                 psf_next = psf->sf_next;
 635                 if (psf->sf_crcount == 0) {
 636                         if (psf_prev)
 637                                 psf_prev->sf_next = psf->sf_next;
 638                         else
 639                                 *ppsf = psf->sf_next;
 640                         kfree(psf);
 641                 } else
 642                         psf_prev = psf;
 643         }
 644 }
 645
 646 static void kfree_pmc(struct ip_mc_list *pmc)
 647 {
 648         ip_sf_list_clear_all(pmc->sources);
 649         ip_sf_list_clear_all(pmc->tomb);
 650         kfree(pmc);
 651 }
 652
 653 static void igmpv3_send_cr(struct in_device *in_dev)
 654 {
 655         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 656         struct sk_buff *skb = NULL;
 657         int type, dtype;
 658
 659         rcu_read_lock();
 660         spin_lock_bh(&in_dev->mc_tomb_lock);
 661
 662         /* deleted MCA's */
 663         pmc_prev = NULL;
 664         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 665                 pmc_next = pmc->next;
 666                 if (pmc->sfmode == MCAST_INCLUDE) {
 667                         type = IGMPV3_BLOCK_OLD_SOURCES;
 668                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 669                         skb = add_grec(skb, pmc, type, 1, 0);
 670                         skb = add_grec(skb, pmc, dtype, 1, 1);
 671                 }
 672                 if (pmc->crcount) {
 673                         if (pmc->sfmode == MCAST_EXCLUDE) {
 674                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 675                                 skb = add_grec(skb, pmc, type, 1, 0);
 676                         }
 677                         pmc->crcount--;
 678                         if (pmc->crcount == 0) {
 679                                 igmpv3_clear_zeros(&pmc->tomb);
 680                                 igmpv3_clear_zeros(&pmc->sources);
 681                         }
 682                 }
 683                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 684                         if (pmc_prev)
 685                                 pmc_prev->next = pmc_next;
 686                         else
 687                                 in_dev->mc_tomb = pmc_next;
 688                         in_dev_put(pmc->interface);
 689                         kfree_pmc(pmc);
 690                 } else
 691                         pmc_prev = pmc;
 692         }
 693         spin_unlock_bh(&in_dev->mc_tomb_lock);
 694
 695         /* change recs */
 696         for_each_pmc_rcu(in_dev, pmc) {
 697                 spin_lock_bh(&pmc->lock);
 698                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 699                         type = IGMPV3_BLOCK_OLD_SOURCES;
 700                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 701                 } else {
 702                         type = IGMPV3_ALLOW_NEW_SOURCES;
 703                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 704                 }
 705                 skb = add_grec(skb, pmc, type, 0, 0);
 706                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 707
 708                 /* filter mode changes */
 709                 if (pmc->crcount) {
 710                         if (pmc->sfmode == MCAST_EXCLUDE)
 711                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 712                         else
 713                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 714                         skb = add_grec(skb, pmc, type, 0, 0);
 715                         pmc->crcount--;
 716                 }
 717                 spin_unlock_bh(&pmc->lock);
 718         }
 719         rcu_read_unlock();
 720
 721         if (!skb)
 722                 return;
 723         (void) igmpv3_sendpack(skb);
 724 }
 725
 726 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 727         int type)
 728 {
 729         struct sk_buff *skb;
 730         struct iphdr *iph;
 731         struct igmphdr *ih;
 732         struct rtable *rt;
 733         struct net_device *dev = in_dev->dev;
 734         struct net *net = dev_net(dev);
 735         __be32  group = pmc ? pmc->multiaddr : 0;
 736         struct flowi4 fl4;
 737         __be32  dst;
 738         int hlen, tlen;
 739
 740         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 741                 return igmpv3_send_report(in_dev, pmc);
 742
 743         if (ipv4_is_local_multicast(group) &&
 744             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 745                 return 0;
 746
 747         if (type == IGMP_HOST_LEAVE_MESSAGE)
 748                 dst = IGMP_ALL_ROUTER;
 749         else
 750                 dst = group;
 751
 752         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 753                                    0, 0,
 754                                    IPPROTO_IGMP, 0, dev->ifindex);
 755         if (IS_ERR(rt))
 756                 return -1;
 757
 758         hlen = LL_RESERVED_SPACE(dev);
 759         tlen = dev->needed_tailroom;
 760         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 761         if (!skb) {
 762                 ip_rt_put(rt);
 763                 return -1;
 764         }
 765         skb->priority = TC_PRIO_CONTROL;
 766
 767         skb_dst_set(skb, &rt->dst);
 768
 769         skb_reserve(skb, hlen);
 770
 771         skb_reset_network_header(skb);
 772         iph = ip_hdr(skb);
 773         skb_put(skb, sizeof(struct iphdr) + 4);
 774
 775         iph->version  = 4;
 776         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 777         iph->tos      = 0xc0;
 778         iph->frag_off = htons(IP_DF);
 779         iph->ttl      = 1;
 780         iph->daddr    = dst;
 781         iph->saddr    = fl4.saddr;
 782         iph->protocol = IPPROTO_IGMP;
 783         ip_select_ident(net, skb, NULL);
 784         ((u8 *)&iph[1])[0] = IPOPT_RA;
 785         ((u8 *)&iph[1])[1] = 4;
 786         ((u8 *)&iph[1])[2] = 0;
 787         ((u8 *)&iph[1])[3] = 0;
 788
 789         ih = skb_put(skb, sizeof(struct igmphdr));
 790         ih->type = type;
 791         ih->code = 0;
 792         ih->csum = 0;
 793         ih->group = group;
 794         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 795
 796         return ip_local_out(net, skb->sk, skb);
 797 }
 798
 799 static void igmp_gq_timer_expire(struct timer_list *t)
 800 {
 801         struct in_device *in_dev = from_timer(in_dev, t, mr_gq_timer);
 802
 803         in_dev->mr_gq_running = 0;
 804         igmpv3_send_report(in_dev, NULL);
 805         in_dev_put(in_dev);
 806 }
 807
 808 static void igmp_ifc_timer_expire(struct timer_list *t)
 809 {
 810         struct in_device *in_dev = from_timer(in_dev, t, mr_ifc_timer);
 811         u32 mr_ifc_count;
 812
 813         igmpv3_send_cr(in_dev);
 814 restart:
 815         mr_ifc_count = READ_ONCE(in_dev->mr_ifc_count);
 816
 817         if (mr_ifc_count) {
 818                 if (cmpxchg(&in_dev->mr_ifc_count,
 819                             mr_ifc_count,
 820                             mr_ifc_count - 1) != mr_ifc_count)
 821                         goto restart;
 822                 igmp_ifc_start_timer(in_dev,
 823                                      unsolicited_report_interval(in_dev));
 824         }
 825         in_dev_put(in_dev);
 826 }
 827
 828 static void igmp_ifc_event(struct in_device *in_dev)
 829 {
 830         struct net *net = dev_net(in_dev->dev);
 831         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 832                 return;
 833         WRITE_ONCE(in_dev->mr_ifc_count, in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv));
 834         igmp_ifc_start_timer(in_dev, 1);
 835 }
 836
 837
 838 static void igmp_timer_expire(struct timer_list *t)
 839 {
 840         struct ip_mc_list *im = from_timer(im, t, timer);
 841         struct in_device *in_dev = im->interface;
 842
 843         spin_lock(&im->lock);
 844         im->tm_running = 0;
 845
 846         if (im->unsolicit_count && --im->unsolicit_count)
 847                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 848
 849         im->reporter = 1;
 850         spin_unlock(&im->lock);
 851
 852         if (IGMP_V1_SEEN(in_dev))
 853                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 854         else if (IGMP_V2_SEEN(in_dev))
 855                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 856         else
 857                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 858
 859         ip_ma_put(im);
 860 }
 861
 862 /* mark EXCLUDE-mode sources */
 863 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 864 {
 865         struct ip_sf_list *psf;
 866         int i, scount;
 867
 868         scount = 0;
 869         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 870                 if (scount == nsrcs)
 871                         break;
 872                 for (i = 0; i < nsrcs; i++) {
 873                         /* skip inactive filters */
 874                         if (psf->sf_count[MCAST_INCLUDE] ||
 875                             pmc->sfcount[MCAST_EXCLUDE] !=
 876                             psf->sf_count[MCAST_EXCLUDE])
 877                                 break;
 878                         if (srcs[i] == psf->sf_inaddr) {
 879                                 scount++;
 880                                 break;
 881                         }
 882                 }
 883         }
 884         pmc->gsquery = 0;
 885         if (scount == nsrcs)    /* all sources excluded */
 886                 return 0;
 887         return 1;
 888 }
 889
 890 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 891 {
 892         struct ip_sf_list *psf;
 893         int i, scount;
 894
 895         if (pmc->sfmode == MCAST_EXCLUDE)
 896                 return igmp_xmarksources(pmc, nsrcs, srcs);
 897
 898         /* mark INCLUDE-mode sources */
 899         scount = 0;
 900         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 901                 if (scount == nsrcs)
 902                         break;
 903                 for (i = 0; i < nsrcs; i++)
 904                         if (srcs[i] == psf->sf_inaddr) {
 905                                 psf->sf_gsresp = 1;
 906                                 scount++;
 907                                 break;
 908                         }
 909         }
 910         if (!scount) {
 911                 pmc->gsquery = 0;
 912                 return 0;
 913         }
 914         pmc->gsquery = 1;
 915         return 1;
 916 }
 917
 918 /* return true if packet was dropped */
 919 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 920 {
 921         struct ip_mc_list *im;
 922         struct net *net = dev_net(in_dev->dev);
 923
 924         /* Timers are only set for non-local groups */
 925
 926         if (group == IGMP_ALL_HOSTS)
 927                 return false;
 928         if (ipv4_is_local_multicast(group) &&
 929             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 930                 return false;
 931
 932         rcu_read_lock();
 933         for_each_pmc_rcu(in_dev, im) {
 934                 if (im->multiaddr == group) {
 935                         igmp_stop_timer(im);
 936                         break;
 937                 }
 938         }
 939         rcu_read_unlock();
 940         return false;
 941 }
 942
 943 /* return true if packet was dropped */
 944 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 945         int len)
 946 {
 947         struct igmphdr          *ih = igmp_hdr(skb);
 948         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 949         struct ip_mc_list       *im;
 950         __be32                  group = ih->group;
 951         int                     max_delay;
 952         int                     mark = 0;
 953         struct net              *net = dev_net(in_dev->dev);
 954
 955
 956         if (len == 8) {
 957                 if (ih->code == 0) {
 958                         /* Alas, old v1 router presents here. */
 959
 960                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 961                         in_dev->mr_v1_seen = jiffies +
 962                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 963                                 in_dev->mr_qri;
 964                         group = 0;
 965                 } else {
 966                         /* v2 router present */
 967                         max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
 968                         in_dev->mr_v2_seen = jiffies +
 969                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 970                                 in_dev->mr_qri;
 971                 }
 972                 /* cancel the interface change timer */
 973                 WRITE_ONCE(in_dev->mr_ifc_count, 0);
 974                 if (del_timer(&in_dev->mr_ifc_timer))
 975                         __in_dev_put(in_dev);
 976                 /* clear deleted report items */
 977                 igmpv3_clear_delrec(in_dev);
 978         } else if (len < 12) {
 979                 return true;    /* ignore bogus packet; freed by caller */
 980         } else if (IGMP_V1_SEEN(in_dev)) {
 981                 /* This is a v3 query with v1 queriers present */
 982                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 983                 group = 0;
 984         } else if (IGMP_V2_SEEN(in_dev)) {
 985                 /* this is a v3 query with v2 queriers present;
 986                  * Interpretation of the max_delay code is problematic here.
 987                  * A real v2 host would use ih_code directly, while v3 has a
 988                  * different encoding. We use the v3 encoding as more likely
 989                  * to be intended in a v3 query.
 990                  */
 991                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 992                 if (!max_delay)
 993                         max_delay = 1;  /* can't mod w/ 0 */
 994         } else { /* v3 */
 995                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 996                         return true;
 997
 998                 ih3 = igmpv3_query_hdr(skb);
 999                 if (ih3->nsrcs) {
1000                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
1001                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
1002                                 return true;
1003                         ih3 = igmpv3_query_hdr(skb);
1004                 }
1005
1006                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
1007                 if (!max_delay)
1008                         max_delay = 1;  /* can't mod w/ 0 */
1009                 in_dev->mr_maxdelay = max_delay;
1010
1011                 /* RFC3376, 4.1.6. QRV and 4.1.7. QQIC, when the most recently
1012                  * received value was zero, use the default or statically
1013                  * configured value.
1014                  */
1015                 in_dev->mr_qrv = ih3->qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1016                 in_dev->mr_qi = IGMPV3_QQIC(ih3->qqic)*HZ ?: IGMP_QUERY_INTERVAL;
1017
1018                 /* RFC3376, 8.3. Query Response Interval:
1019                  * The number of seconds represented by the [Query Response
1020                  * Interval] must be less than the [Query Interval].
1021                  */
1022                 if (in_dev->mr_qri >= in_dev->mr_qi)
1023                         in_dev->mr_qri = (in_dev->mr_qi/HZ - 1)*HZ;
1024
1025                 if (!group) { /* general query */
1026                         if (ih3->nsrcs)
1027                                 return true;    /* no sources allowed */
1028                         igmp_gq_start_timer(in_dev);
1029                         return false;
1030                 }
1031                 /* mark sources to include, if group & source-specific */
1032                 mark = ih3->nsrcs != 0;
1033         }
1034
1035         /*
1036          * - Start the timers in all of our membership records
1037          *   that the query applies to for the interface on
1038          *   which the query arrived excl. those that belong
1039          *   to a "local" group (224.0.0.X)
1040          * - For timers already running check if they need to
1041          *   be reset.
1042          * - Use the igmp->igmp_code field as the maximum
1043          *   delay possible
1044          */
1045         rcu_read_lock();
1046         for_each_pmc_rcu(in_dev, im) {
1047                 int changed;
1048
1049                 if (group && group != im->multiaddr)
1050                         continue;
1051                 if (im->multiaddr == IGMP_ALL_HOSTS)
1052                         continue;
1053                 if (ipv4_is_local_multicast(im->multiaddr) &&
1054                     !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1055                         continue;
1056                 spin_lock_bh(&im->lock);
1057                 if (im->tm_running)
1058                         im->gsquery = im->gsquery && mark;
1059                 else
1060                         im->gsquery = mark;
1061                 changed = !im->gsquery ||
1062                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1063                 spin_unlock_bh(&im->lock);
1064                 if (changed)
1065                         igmp_mod_timer(im, max_delay);
1066         }
1067         rcu_read_unlock();
1068         return false;
1069 }
1070
1071 /* called in rcu_read_lock() section */
1072 int igmp_rcv(struct sk_buff *skb)
1073 {
1074         /* This basically follows the spec line by line -- see RFC1112 */
1075         struct igmphdr *ih;
1076         struct net_device *dev = skb->dev;
1077         struct in_device *in_dev;
1078         int len = skb->len;
1079         bool dropped = true;
1080
1081         if (netif_is_l3_master(dev)) {
1082                 dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1083                 if (!dev)
1084                         goto drop;
1085         }
1086
1087         in_dev = __in_dev_get_rcu(dev);
1088         if (!in_dev)
1089                 goto drop;
1090
1091         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1092                 goto drop;
1093
1094         if (skb_checksum_simple_validate(skb))
1095                 goto drop;
1096
1097         ih = igmp_hdr(skb);
1098         switch (ih->type) {
1099         case IGMP_HOST_MEMBERSHIP_QUERY:
1100                 dropped = igmp_heard_query(in_dev, skb, len);
1101                 break;
1102         case IGMP_HOST_MEMBERSHIP_REPORT:
1103         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1104                 /* Is it our report looped back? */
1105                 if (rt_is_output_route(skb_rtable(skb)))
1106                         break;
1107                 /* don't rely on MC router hearing unicast reports */
1108                 if (skb->pkt_type == PACKET_MULTICAST ||
1109                     skb->pkt_type == PACKET_BROADCAST)
1110                         dropped = igmp_heard_report(in_dev, ih->group);
1111                 break;
1112         case IGMP_PIM:
1113 #ifdef CONFIG_IP_PIMSM_V1
1114                 return pim_rcv_v1(skb);
1115 #endif
1116         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1117         case IGMP_DVMRP:
1118         case IGMP_TRACE:
1119         case IGMP_HOST_LEAVE_MESSAGE:
1120         case IGMP_MTRACE:
1121         case IGMP_MTRACE_RESP:
1122                 break;
1123         default:
1124                 break;
1125         }
1126
1127 drop:
1128         if (dropped)
1129                 kfree_skb(skb);
1130         else
1131                 consume_skb(skb);
1132         return 0;
1133 }
1134
1135 #endif
1136
1137
1138 /*
1139  *      Add a filter to a device
1140  */
1141
1142 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1143 {
1144         char buf[MAX_ADDR_LEN];
1145         struct net_device *dev = in_dev->dev;
1146
1147         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1148            We will get multicast token leakage, when IFF_MULTICAST
1149            is changed. This check should be done in ndo_set_rx_mode
1150            routine. Something sort of:
1151            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1152            --ANK
1153            */
1154         if (arp_mc_map(addr, buf, dev, 0) == 0)
1155                 dev_mc_add(dev, buf);
1156 }
1157
1158 /*
1159  *      Remove a filter from a device
1160  */
1161
1162 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1163 {
1164         char buf[MAX_ADDR_LEN];
1165         struct net_device *dev = in_dev->dev;
1166
1167         if (arp_mc_map(addr, buf, dev, 0) == 0)
1168                 dev_mc_del(dev, buf);
1169 }
1170
1171 #ifdef CONFIG_IP_MULTICAST
1172 /*
1173  * deleted ip_mc_list manipulation
1174  */
1175 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
1176                               gfp_t gfp)
1177 {
1178         struct ip_mc_list *pmc;
1179         struct net *net = dev_net(in_dev->dev);
1180
1181         /* this is an "ip_mc_list" for convenience; only the fields below
1182          * are actually used. In particular, the refcnt and users are not
1183          * used for management of the delete list. Using the same structure
1184          * for deleted items allows change reports to use common code with
1185          * non-deleted or query-response MCA's.
1186          */
1187         pmc = kzalloc(sizeof(*pmc), gfp);
1188         if (!pmc)
1189                 return;
1190         spin_lock_init(&pmc->lock);
1191         spin_lock_bh(&im->lock);
1192         pmc->interface = im->interface;
1193         in_dev_hold(in_dev);
1194         pmc->multiaddr = im->multiaddr;
1195         pmc->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1196         pmc->sfmode = im->sfmode;
1197         if (pmc->sfmode == MCAST_INCLUDE) {
1198                 struct ip_sf_list *psf;
1199
1200                 pmc->tomb = im->tomb;
1201                 pmc->sources = im->sources;
1202                 im->tomb = im->sources = NULL;
1203                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1204                         psf->sf_crcount = pmc->crcount;
1205         }
1206         spin_unlock_bh(&im->lock);
1207
1208         spin_lock_bh(&in_dev->mc_tomb_lock);
1209         pmc->next = in_dev->mc_tomb;
1210         in_dev->mc_tomb = pmc;
1211         spin_unlock_bh(&in_dev->mc_tomb_lock);
1212 }
1213
1214 /*
1215  * restore ip_mc_list deleted records
1216  */
1217 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1218 {
1219         struct ip_mc_list *pmc, *pmc_prev;
1220         struct ip_sf_list *psf;
1221         struct net *net = dev_net(in_dev->dev);
1222         __be32 multiaddr = im->multiaddr;
1223
1224         spin_lock_bh(&in_dev->mc_tomb_lock);
1225         pmc_prev = NULL;
1226         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1227                 if (pmc->multiaddr == multiaddr)
1228                         break;
1229                 pmc_prev = pmc;
1230         }
1231         if (pmc) {
1232                 if (pmc_prev)
1233                         pmc_prev->next = pmc->next;
1234                 else
1235                         in_dev->mc_tomb = pmc->next;
1236         }
1237         spin_unlock_bh(&in_dev->mc_tomb_lock);
1238
1239         spin_lock_bh(&im->lock);
1240         if (pmc) {
1241                 im->interface = pmc->interface;
1242                 if (im->sfmode == MCAST_INCLUDE) {
1243                         swap(im->tomb, pmc->tomb);
1244                         swap(im->sources, pmc->sources);
1245                         for (psf = im->sources; psf; psf = psf->sf_next)
1246                                 psf->sf_crcount = in_dev->mr_qrv ?:
1247                                         READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1248                 } else {
1249                         im->crcount = in_dev->mr_qrv ?:
1250                                 READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1251                 }
1252                 in_dev_put(pmc->interface);
1253                 kfree_pmc(pmc);
1254         }
1255         spin_unlock_bh(&im->lock);
1256 }
1257
1258 /*
1259  * flush ip_mc_list deleted records
1260  */
1261 static void igmpv3_clear_delrec(struct in_device *in_dev)
1262 {
1263         struct ip_mc_list *pmc, *nextpmc;
1264
1265         spin_lock_bh(&in_dev->mc_tomb_lock);
1266         pmc = in_dev->mc_tomb;
1267         in_dev->mc_tomb = NULL;
1268         spin_unlock_bh(&in_dev->mc_tomb_lock);
1269
1270         for (; pmc; pmc = nextpmc) {
1271                 nextpmc = pmc->next;
1272                 ip_mc_clear_src(pmc);
1273                 in_dev_put(pmc->interface);
1274                 kfree_pmc(pmc);
1275         }
1276         /* clear dead sources, too */
1277         rcu_read_lock();
1278         for_each_pmc_rcu(in_dev, pmc) {
1279                 struct ip_sf_list *psf;
1280
1281                 spin_lock_bh(&pmc->lock);
1282                 psf = pmc->tomb;
1283                 pmc->tomb = NULL;
1284                 spin_unlock_bh(&pmc->lock);
1285                 ip_sf_list_clear_all(psf);
1286         }
1287         rcu_read_unlock();
1288 }
1289 #endif
1290
1291 static void __igmp_group_dropped(struct ip_mc_list *im, gfp_t gfp)
1292 {
1293         struct in_device *in_dev = im->interface;
1294 #ifdef CONFIG_IP_MULTICAST
1295         struct net *net = dev_net(in_dev->dev);
1296         int reporter;
1297 #endif
1298
1299         if (im->loaded) {
1300                 im->loaded = 0;
1301                 ip_mc_filter_del(in_dev, im->multiaddr);
1302         }
1303
1304 #ifdef CONFIG_IP_MULTICAST
1305         if (im->multiaddr == IGMP_ALL_HOSTS)
1306                 return;
1307         if (ipv4_is_local_multicast(im->multiaddr) &&
1308             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1309                 return;
1310
1311         reporter = im->reporter;
1312         igmp_stop_timer(im);
1313
1314         if (!in_dev->dead) {
1315                 if (IGMP_V1_SEEN(in_dev))
1316                         return;
1317                 if (IGMP_V2_SEEN(in_dev)) {
1318                         if (reporter)
1319                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1320                         return;
1321                 }
1322                 /* IGMPv3 */
1323                 igmpv3_add_delrec(in_dev, im, gfp);
1324
1325                 igmp_ifc_event(in_dev);
1326         }
1327 #endif
1328 }
1329
1330 static void igmp_group_dropped(struct ip_mc_list *im)
1331 {
1332         __igmp_group_dropped(im, GFP_KERNEL);
1333 }
1334
1335 static void igmp_group_added(struct ip_mc_list *im)
1336 {
1337         struct in_device *in_dev = im->interface;
1338 #ifdef CONFIG_IP_MULTICAST
1339         struct net *net = dev_net(in_dev->dev);
1340 #endif
1341
1342         if (im->loaded == 0) {
1343                 im->loaded = 1;
1344                 ip_mc_filter_add(in_dev, im->multiaddr);
1345         }
1346
1347 #ifdef CONFIG_IP_MULTICAST
1348         if (im->multiaddr == IGMP_ALL_HOSTS)
1349                 return;
1350         if (ipv4_is_local_multicast(im->multiaddr) &&
1351             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1352                 return;
1353
1354         if (in_dev->dead)
1355                 return;
1356
1357         im->unsolicit_count = READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1358         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1359                 spin_lock_bh(&im->lock);
1360                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1361                 spin_unlock_bh(&im->lock);
1362                 return;
1363         }
1364         /* else, v3 */
1365
1366         /* Based on RFC3376 5.1, for newly added INCLUDE SSM, we should
1367          * not send filter-mode change record as the mode should be from
1368          * IN() to IN(A).
1369          */
1370         if (im->sfmode == MCAST_EXCLUDE)
1371                 im->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1372
1373         igmp_ifc_event(in_dev);
1374 #endif
1375 }
1376
1377
1378 /*
1379  *      Multicast list managers
1380  */
1381
1382 static u32 ip_mc_hash(const struct ip_mc_list *im)
1383 {
1384         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1385 }
1386
1387 static void ip_mc_hash_add(struct in_device *in_dev,
1388                            struct ip_mc_list *im)
1389 {
1390         struct ip_mc_list __rcu **mc_hash;
1391         u32 hash;
1392
1393         mc_hash = rtnl_dereference(in_dev->mc_hash);
1394         if (mc_hash) {
1395                 hash = ip_mc_hash(im);
1396                 im->next_hash = mc_hash[hash];
1397                 rcu_assign_pointer(mc_hash[hash], im);
1398                 return;
1399         }
1400
1401         /* do not use a hash table for small number of items */
1402         if (in_dev->mc_count < 4)
1403                 return;
1404
1405         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1406                           GFP_KERNEL);
1407         if (!mc_hash)
1408                 return;
1409
1410         for_each_pmc_rtnl(in_dev, im) {
1411                 hash = ip_mc_hash(im);
1412                 im->next_hash = mc_hash[hash];
1413                 RCU_INIT_POINTER(mc_hash[hash], im);
1414         }
1415
1416         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1417 }
1418
1419 static void ip_mc_hash_remove(struct in_device *in_dev,
1420                               struct ip_mc_list *im)
1421 {
1422         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1423         struct ip_mc_list *aux;
1424
1425         if (!mc_hash)
1426                 return;
1427         mc_hash += ip_mc_hash(im);
1428         while ((aux = rtnl_dereference(*mc_hash)) != im)
1429                 mc_hash = &aux->next_hash;
1430         *mc_hash = im->next_hash;
1431 }
1432
1433
1434 /*
1435  *      A socket has joined a multicast group on device dev.
1436  */
1437 static void ____ip_mc_inc_group(struct in_device *in_dev, __be32 addr,
1438                                 unsigned int mode, gfp_t gfp)
1439 {
1440         struct ip_mc_list *im;
1441
1442         ASSERT_RTNL();
1443
1444         for_each_pmc_rtnl(in_dev, im) {
1445                 if (im->multiaddr == addr) {
1446                         im->users++;
1447                         ip_mc_add_src(in_dev, &addr, mode, 0, NULL, 0);
1448                         goto out;
1449                 }
1450         }
1451
1452         im = kzalloc(sizeof(*im), gfp);
1453         if (!im)
1454                 goto out;
1455
1456         im->users = 1;
1457         im->interface = in_dev;
1458         in_dev_hold(in_dev);
1459         im->multiaddr = addr;
1460         /* initial mode is (EX, empty) */
1461         im->sfmode = mode;
1462         im->sfcount[mode] = 1;
1463         refcount_set(&im->refcnt, 1);
1464         spin_lock_init(&im->lock);
1465 #ifdef CONFIG_IP_MULTICAST
1466         timer_setup(&im->timer, igmp_timer_expire, 0);
1467 #endif
1468
1469         im->next_rcu = in_dev->mc_list;
1470         in_dev->mc_count++;
1471         rcu_assign_pointer(in_dev->mc_list, im);
1472
1473         ip_mc_hash_add(in_dev, im);
1474
1475 #ifdef CONFIG_IP_MULTICAST
1476         igmpv3_del_delrec(in_dev, im);
1477 #endif
1478         igmp_group_added(im);
1479         if (!in_dev->dead)
1480                 ip_rt_multicast_event(in_dev);
1481 out:
1482         return;
1483 }
1484
1485 void __ip_mc_inc_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1486 {
1487         ____ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE, gfp);
1488 }
1489 EXPORT_SYMBOL(__ip_mc_inc_group);
1490
1491 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1492 {
1493         __ip_mc_inc_group(in_dev, addr, GFP_KERNEL);
1494 }
1495 EXPORT_SYMBOL(ip_mc_inc_group);
1496
1497 static int ip_mc_check_iphdr(struct sk_buff *skb)
1498 {
1499         const struct iphdr *iph;
1500         unsigned int len;
1501         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1502
1503         if (!pskb_may_pull(skb, offset))
1504                 return -EINVAL;
1505
1506         iph = ip_hdr(skb);
1507
1508         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1509                 return -EINVAL;
1510
1511         offset += ip_hdrlen(skb) - sizeof(*iph);
1512
1513         if (!pskb_may_pull(skb, offset))
1514                 return -EINVAL;
1515
1516         iph = ip_hdr(skb);
1517
1518         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1519                 return -EINVAL;
1520
1521         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1522         if (skb->len < len || len < offset)
1523                 return -EINVAL;
1524
1525         skb_set_transport_header(skb, offset);
1526
1527         return 0;
1528 }
1529
1530 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1531 {
1532         unsigned int len = skb_transport_offset(skb);
1533
1534         len += sizeof(struct igmpv3_report);
1535
1536         return ip_mc_may_pull(skb, len) ? 0 : -EINVAL;
1537 }
1538
1539 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1540 {
1541         unsigned int transport_len = ip_transport_len(skb);
1542         unsigned int len;
1543
1544         /* IGMPv{1,2}? */
1545         if (transport_len != sizeof(struct igmphdr)) {
1546                 /* or IGMPv3? */
1547                 if (transport_len < sizeof(struct igmpv3_query))
1548                         return -EINVAL;
1549
1550                 len = skb_transport_offset(skb) + sizeof(struct igmpv3_query);
1551                 if (!ip_mc_may_pull(skb, len))
1552                         return -EINVAL;
1553         }
1554
1555         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1556          * all-systems destination addresses (224.0.0.1) for general queries
1557          */
1558         if (!igmp_hdr(skb)->group &&
1559             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1560                 return -EINVAL;
1561
1562         return 0;
1563 }
1564
1565 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1566 {
1567         switch (igmp_hdr(skb)->type) {
1568         case IGMP_HOST_LEAVE_MESSAGE:
1569         case IGMP_HOST_MEMBERSHIP_REPORT:
1570         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1571                 return 0;
1572         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1573                 return ip_mc_check_igmp_reportv3(skb);
1574         case IGMP_HOST_MEMBERSHIP_QUERY:
1575                 return ip_mc_check_igmp_query(skb);
1576         default:
1577                 return -ENOMSG;
1578         }
1579 }
1580
1581 static __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1582 {
1583         return skb_checksum_simple_validate(skb);
1584 }
1585
1586 static int ip_mc_check_igmp_csum(struct sk_buff *skb)
1587 {
1588         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1589         unsigned int transport_len = ip_transport_len(skb);
1590         struct sk_buff *skb_chk;
1591
1592         if (!ip_mc_may_pull(skb, len))
1593                 return -EINVAL;
1594
1595         skb_chk = skb_checksum_trimmed(skb, transport_len,
1596                                        ip_mc_validate_checksum);
1597         if (!skb_chk)
1598                 return -EINVAL;
1599
1600         if (skb_chk != skb)
1601                 kfree_skb(skb_chk);
1602
1603         return 0;
1604 }
1605
1606 /**
1607  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1608  * @skb: the skb to validate
1609  *
1610  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1611  * skb transport header accordingly and returns zero.
1612  *
1613  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1614  *  standard
1615  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1616  * -ENOMEM: A memory allocation failure happened.
1617  *
1618  * Caller needs to set the skb network header and free any returned skb if it
1619  * differs from the provided skb.
1620  */
1621 int ip_mc_check_igmp(struct sk_buff *skb)
1622 {
1623         int ret = ip_mc_check_iphdr(skb);
1624
1625         if (ret < 0)
1626                 return ret;
1627
1628         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1629                 return -ENOMSG;
1630
1631         ret = ip_mc_check_igmp_csum(skb);
1632         if (ret < 0)
1633                 return ret;
1634
1635         return ip_mc_check_igmp_msg(skb);
1636 }
1637 EXPORT_SYMBOL(ip_mc_check_igmp);
1638
1639 /*
1640  *      Resend IGMP JOIN report; used by netdev notifier.
1641  */
1642 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1643 {
1644 #ifdef CONFIG_IP_MULTICAST
1645         struct ip_mc_list *im;
1646         int type;
1647         struct net *net = dev_net(in_dev->dev);
1648
1649         ASSERT_RTNL();
1650
1651         for_each_pmc_rtnl(in_dev, im) {
1652                 if (im->multiaddr == IGMP_ALL_HOSTS)
1653                         continue;
1654                 if (ipv4_is_local_multicast(im->multiaddr) &&
1655                     !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1656                         continue;
1657
1658                 /* a failover is happening and switches
1659                  * must be notified immediately
1660                  */
1661                 if (IGMP_V1_SEEN(in_dev))
1662                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1663                 else if (IGMP_V2_SEEN(in_dev))
1664                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1665                 else
1666                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1667                 igmp_send_report(in_dev, im, type);
1668         }
1669 #endif
1670 }
1671
1672 /*
1673  *      A socket has left a multicast group on device dev
1674  */
1675
1676 void __ip_mc_dec_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1677 {
1678         struct ip_mc_list *i;
1679         struct ip_mc_list __rcu **ip;
1680
1681         ASSERT_RTNL();
1682
1683         for (ip = &in_dev->mc_list;
1684              (i = rtnl_dereference(*ip)) != NULL;
1685              ip = &i->next_rcu) {
1686                 if (i->multiaddr == addr) {
1687                         if (--i->users == 0) {
1688                                 ip_mc_hash_remove(in_dev, i);
1689                                 *ip = i->next_rcu;
1690                                 in_dev->mc_count--;
1691                                 __igmp_group_dropped(i, gfp);
1692                                 ip_mc_clear_src(i);
1693
1694                                 if (!in_dev->dead)
1695                                         ip_rt_multicast_event(in_dev);
1696
1697                                 ip_ma_put(i);
1698                                 return;
1699                         }
1700                         break;
1701                 }
1702         }
1703 }
1704 EXPORT_SYMBOL(__ip_mc_dec_group);
1705
1706 /* Device changing type */
1707
1708 void ip_mc_unmap(struct in_device *in_dev)
1709 {
1710         struct ip_mc_list *pmc;
1711
1712         ASSERT_RTNL();
1713
1714         for_each_pmc_rtnl(in_dev, pmc)
1715                 igmp_group_dropped(pmc);
1716 }
1717
1718 void ip_mc_remap(struct in_device *in_dev)
1719 {
1720         struct ip_mc_list *pmc;
1721
1722         ASSERT_RTNL();
1723
1724         for_each_pmc_rtnl(in_dev, pmc) {
1725 #ifdef CONFIG_IP_MULTICAST
1726                 igmpv3_del_delrec(in_dev, pmc);
1727 #endif
1728                 igmp_group_added(pmc);
1729         }
1730 }
1731
1732 /* Device going down */
1733
1734 void ip_mc_down(struct in_device *in_dev)
1735 {
1736         struct ip_mc_list *pmc;
1737
1738         ASSERT_RTNL();
1739
1740         for_each_pmc_rtnl(in_dev, pmc)
1741                 igmp_group_dropped(pmc);
1742
1743 #ifdef CONFIG_IP_MULTICAST
1744         WRITE_ONCE(in_dev->mr_ifc_count, 0);
1745         if (del_timer(&in_dev->mr_ifc_timer))
1746                 __in_dev_put(in_dev);
1747         in_dev->mr_gq_running = 0;
1748         if (del_timer(&in_dev->mr_gq_timer))
1749                 __in_dev_put(in_dev);
1750 #endif
1751
1752         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1753 }
1754
1755 #ifdef CONFIG_IP_MULTICAST
1756 static void ip_mc_reset(struct in_device *in_dev)
1757 {
1758         struct net *net = dev_net(in_dev->dev);
1759
1760         in_dev->mr_qi = IGMP_QUERY_INTERVAL;
1761         in_dev->mr_qri = IGMP_QUERY_RESPONSE_INTERVAL;
1762         in_dev->mr_qrv = READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1763 }
1764 #else
1765 static void ip_mc_reset(struct in_device *in_dev)
1766 {
1767 }
1768 #endif
1769
1770 void ip_mc_init_dev(struct in_device *in_dev)
1771 {
1772         ASSERT_RTNL();
1773
1774 #ifdef CONFIG_IP_MULTICAST
1775         timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1776         timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1777 #endif
1778         ip_mc_reset(in_dev);
1779
1780         spin_lock_init(&in_dev->mc_tomb_lock);
1781 }
1782
1783 /* Device going up */
1784
1785 void ip_mc_up(struct in_device *in_dev)
1786 {
1787         struct ip_mc_list *pmc;
1788
1789         ASSERT_RTNL();
1790
1791         ip_mc_reset(in_dev);
1792         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1793
1794         for_each_pmc_rtnl(in_dev, pmc) {
1795 #ifdef CONFIG_IP_MULTICAST
1796                 igmpv3_del_delrec(in_dev, pmc);
1797 #endif
1798                 igmp_group_added(pmc);
1799         }
1800 }
1801
1802 /*
1803  *      Device is about to be destroyed: clean up.
1804  */
1805
1806 void ip_mc_destroy_dev(struct in_device *in_dev)
1807 {
1808         struct ip_mc_list *i;
1809
1810         ASSERT_RTNL();
1811
1812         /* Deactivate timers */
1813         ip_mc_down(in_dev);
1814 #ifdef CONFIG_IP_MULTICAST
1815         igmpv3_clear_delrec(in_dev);
1816 #endif
1817
1818         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1819                 in_dev->mc_list = i->next_rcu;
1820                 in_dev->mc_count--;
1821                 ip_mc_clear_src(i);
1822                 ip_ma_put(i);
1823         }
1824 }
1825
1826 /* RTNL is locked */
1827 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1828 {
1829         struct net_device *dev = NULL;
1830         struct in_device *idev = NULL;
1831
1832         if (imr->imr_ifindex) {
1833                 idev = inetdev_by_index(net, imr->imr_ifindex);
1834                 return idev;
1835         }
1836         if (imr->imr_address.s_addr) {
1837                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1838                 if (!dev)
1839                         return NULL;
1840         }
1841
1842         if (!dev) {
1843                 struct rtable *rt = ip_route_output(net,
1844                                                     imr->imr_multiaddr.s_addr,
1845                                                     0, 0, 0,
1846                                                     RT_SCOPE_UNIVERSE);
1847                 if (!IS_ERR(rt)) {
1848                         dev = rt->dst.dev;
1849                         ip_rt_put(rt);
1850                 }
1851         }
1852         if (dev) {
1853                 imr->imr_ifindex = dev->ifindex;
1854                 idev = __in_dev_get_rtnl(dev);
1855         }
1856         return idev;
1857 }
1858
1859 /*
1860  *      Join a socket to a group
1861  */
1862
1863 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1864         __be32 *psfsrc)
1865 {
1866         struct ip_sf_list *psf, *psf_prev;
1867         int rv = 0;
1868
1869         psf_prev = NULL;
1870         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1871                 if (psf->sf_inaddr == *psfsrc)
1872                         break;
1873                 psf_prev = psf;
1874         }
1875         if (!psf || psf->sf_count[sfmode] == 0) {
1876                 /* source filter not found, or count wrong =>  bug */
1877                 return -ESRCH;
1878         }
1879         psf->sf_count[sfmode]--;
1880         if (psf->sf_count[sfmode] == 0) {
1881                 ip_rt_multicast_event(pmc->interface);
1882         }
1883         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1884 #ifdef CONFIG_IP_MULTICAST
1885                 struct in_device *in_dev = pmc->interface;
1886                 struct net *net = dev_net(in_dev->dev);
1887 #endif
1888
1889                 /* no more filters for this source */
1890                 if (psf_prev)
1891                         psf_prev->sf_next = psf->sf_next;
1892                 else
1893                         pmc->sources = psf->sf_next;
1894 #ifdef CONFIG_IP_MULTICAST
1895                 if (psf->sf_oldin &&
1896                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1897                         psf->sf_crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1898                         psf->sf_next = pmc->tomb;
1899                         pmc->tomb = psf;
1900                         rv = 1;
1901                 } else
1902 #endif
1903                         kfree(psf);
1904         }
1905         return rv;
1906 }
1907
1908 #ifndef CONFIG_IP_MULTICAST
1909 #define igmp_ifc_event(x)       do { } while (0)
1910 #endif
1911
1912 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1913                          int sfcount, __be32 *psfsrc, int delta)
1914 {
1915         struct ip_mc_list *pmc;
1916         int     changerec = 0;
1917         int     i, err;
1918
1919         if (!in_dev)
1920                 return -ENODEV;
1921         rcu_read_lock();
1922         for_each_pmc_rcu(in_dev, pmc) {
1923                 if (*pmca == pmc->multiaddr)
1924                         break;
1925         }
1926         if (!pmc) {
1927                 /* MCA not found?? bug */
1928                 rcu_read_unlock();
1929                 return -ESRCH;
1930         }
1931         spin_lock_bh(&pmc->lock);
1932         rcu_read_unlock();
1933 #ifdef CONFIG_IP_MULTICAST
1934         sf_markstate(pmc);
1935 #endif
1936         if (!delta) {
1937                 err = -EINVAL;
1938                 if (!pmc->sfcount[sfmode])
1939                         goto out_unlock;
1940                 pmc->sfcount[sfmode]--;
1941         }
1942         err = 0;
1943         for (i = 0; i < sfcount; i++) {
1944                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1945
1946                 changerec |= rv > 0;
1947                 if (!err && rv < 0)
1948                         err = rv;
1949         }
1950         if (pmc->sfmode == MCAST_EXCLUDE &&
1951             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1952             pmc->sfcount[MCAST_INCLUDE]) {
1953 #ifdef CONFIG_IP_MULTICAST
1954                 struct ip_sf_list *psf;
1955                 struct net *net = dev_net(in_dev->dev);
1956 #endif
1957
1958                 /* filter mode change */
1959                 pmc->sfmode = MCAST_INCLUDE;
1960 #ifdef CONFIG_IP_MULTICAST
1961                 pmc->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
1962                 WRITE_ONCE(in_dev->mr_ifc_count, pmc->crcount);
1963                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1964                         psf->sf_crcount = 0;
1965                 igmp_ifc_event(pmc->interface);
1966         } else if (sf_setstate(pmc) || changerec) {
1967                 igmp_ifc_event(pmc->interface);
1968 #endif
1969         }
1970 out_unlock:
1971         spin_unlock_bh(&pmc->lock);
1972         return err;
1973 }
1974
1975 /*
1976  * Add multicast single-source filter to the interface list
1977  */
1978 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1979         __be32 *psfsrc)
1980 {
1981         struct ip_sf_list *psf, *psf_prev;
1982
1983         psf_prev = NULL;
1984         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1985                 if (psf->sf_inaddr == *psfsrc)
1986                         break;
1987                 psf_prev = psf;
1988         }
1989         if (!psf) {
1990                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1991                 if (!psf)
1992                         return -ENOBUFS;
1993                 psf->sf_inaddr = *psfsrc;
1994                 if (psf_prev) {
1995                         psf_prev->sf_next = psf;
1996                 } else
1997                         pmc->sources = psf;
1998         }
1999         psf->sf_count[sfmode]++;
2000         if (psf->sf_count[sfmode] == 1) {
2001                 ip_rt_multicast_event(pmc->interface);
2002         }
2003         return 0;
2004 }
2005
2006 #ifdef CONFIG_IP_MULTICAST
2007 static void sf_markstate(struct ip_mc_list *pmc)
2008 {
2009         struct ip_sf_list *psf;
2010         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2011
2012         for (psf = pmc->sources; psf; psf = psf->sf_next)
2013                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2014                         psf->sf_oldin = mca_xcount ==
2015                                 psf->sf_count[MCAST_EXCLUDE] &&
2016                                 !psf->sf_count[MCAST_INCLUDE];
2017                 } else
2018                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
2019 }
2020
2021 static int sf_setstate(struct ip_mc_list *pmc)
2022 {
2023         struct ip_sf_list *psf, *dpsf;
2024         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2025         int qrv = pmc->interface->mr_qrv;
2026         int new_in, rv;
2027
2028         rv = 0;
2029         for (psf = pmc->sources; psf; psf = psf->sf_next) {
2030                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2031                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
2032                                 !psf->sf_count[MCAST_INCLUDE];
2033                 } else
2034                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
2035                 if (new_in) {
2036                         if (!psf->sf_oldin) {
2037                                 struct ip_sf_list *prev = NULL;
2038
2039                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2040                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2041                                                 break;
2042                                         prev = dpsf;
2043                                 }
2044                                 if (dpsf) {
2045                                         if (prev)
2046                                                 prev->sf_next = dpsf->sf_next;
2047                                         else
2048                                                 pmc->tomb = dpsf->sf_next;
2049                                         kfree(dpsf);
2050                                 }
2051                                 psf->sf_crcount = qrv;
2052                                 rv++;
2053                         }
2054                 } else if (psf->sf_oldin) {
2055
2056                         psf->sf_crcount = 0;
2057                         /*
2058                          * add or update "delete" records if an active filter
2059                          * is now inactive
2060                          */
2061                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2062                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2063                                         break;
2064                         if (!dpsf) {
2065                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2066                                 if (!dpsf)
2067                                         continue;
2068                                 *dpsf = *psf;
2069                                 /* pmc->lock held by callers */
2070                                 dpsf->sf_next = pmc->tomb;
2071                                 pmc->tomb = dpsf;
2072                         }
2073                         dpsf->sf_crcount = qrv;
2074                         rv++;
2075                 }
2076         }
2077         return rv;
2078 }
2079 #endif
2080
2081 /*
2082  * Add multicast source filter list to the interface list
2083  */
2084 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2085                          int sfcount, __be32 *psfsrc, int delta)
2086 {
2087         struct ip_mc_list *pmc;
2088         int     isexclude;
2089         int     i, err;
2090
2091         if (!in_dev)
2092                 return -ENODEV;
2093         rcu_read_lock();
2094         for_each_pmc_rcu(in_dev, pmc) {
2095                 if (*pmca == pmc->multiaddr)
2096                         break;
2097         }
2098         if (!pmc) {
2099                 /* MCA not found?? bug */
2100                 rcu_read_unlock();
2101                 return -ESRCH;
2102         }
2103         spin_lock_bh(&pmc->lock);
2104         rcu_read_unlock();
2105
2106 #ifdef CONFIG_IP_MULTICAST
2107         sf_markstate(pmc);
2108 #endif
2109         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2110         if (!delta)
2111                 pmc->sfcount[sfmode]++;
2112         err = 0;
2113         for (i = 0; i < sfcount; i++) {
2114                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2115                 if (err)
2116                         break;
2117         }
2118         if (err) {
2119                 int j;
2120
2121                 if (!delta)
2122                         pmc->sfcount[sfmode]--;
2123                 for (j = 0; j < i; j++)
2124                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2125         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2126 #ifdef CONFIG_IP_MULTICAST
2127                 struct ip_sf_list *psf;
2128                 struct net *net = dev_net(pmc->interface->dev);
2129                 in_dev = pmc->interface;
2130 #endif
2131
2132                 /* filter mode change */
2133                 if (pmc->sfcount[MCAST_EXCLUDE])
2134                         pmc->sfmode = MCAST_EXCLUDE;
2135                 else if (pmc->sfcount[MCAST_INCLUDE])
2136                         pmc->sfmode = MCAST_INCLUDE;
2137 #ifdef CONFIG_IP_MULTICAST
2138                 /* else no filters; keep old mode for reports */
2139
2140                 pmc->crcount = in_dev->mr_qrv ?: READ_ONCE(net->ipv4.sysctl_igmp_qrv);
2141                 WRITE_ONCE(in_dev->mr_ifc_count, pmc->crcount);
2142                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2143                         psf->sf_crcount = 0;
2144                 igmp_ifc_event(in_dev);
2145         } else if (sf_setstate(pmc)) {
2146                 igmp_ifc_event(in_dev);
2147 #endif
2148         }
2149         spin_unlock_bh(&pmc->lock);
2150         return err;
2151 }
2152
2153 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2154 {
2155         struct ip_sf_list *tomb, *sources;
2156
2157         spin_lock_bh(&pmc->lock);
2158         tomb = pmc->tomb;
2159         pmc->tomb = NULL;
2160         sources = pmc->sources;
2161         pmc->sources = NULL;
2162         pmc->sfmode = MCAST_EXCLUDE;
2163         pmc->sfcount[MCAST_INCLUDE] = 0;
2164         pmc->sfcount[MCAST_EXCLUDE] = 1;
2165         spin_unlock_bh(&pmc->lock);
2166
2167         ip_sf_list_clear_all(tomb);
2168         ip_sf_list_clear_all(sources);
2169 }
2170
2171 /* Join a multicast group
2172  */
2173 static int __ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr,
2174                               unsigned int mode)
2175 {
2176         __be32 addr = imr->imr_multiaddr.s_addr;
2177         struct ip_mc_socklist *iml, *i;
2178         struct in_device *in_dev;
2179         struct inet_sock *inet = inet_sk(sk);
2180         struct net *net = sock_net(sk);
2181         int ifindex;
2182         int count = 0;
2183         int err;
2184
2185         ASSERT_RTNL();
2186
2187         if (!ipv4_is_multicast(addr))
2188                 return -EINVAL;
2189
2190         in_dev = ip_mc_find_dev(net, imr);
2191
2192         if (!in_dev) {
2193                 err = -ENODEV;
2194                 goto done;
2195         }
2196
2197         err = -EADDRINUSE;
2198         ifindex = imr->imr_ifindex;
2199         for_each_pmc_rtnl(inet, i) {
2200                 if (i->multi.imr_multiaddr.s_addr == addr &&
2201                     i->multi.imr_ifindex == ifindex)
2202                         goto done;
2203                 count++;
2204         }
2205         err = -ENOBUFS;
2206         if (count >= READ_ONCE(net->ipv4.sysctl_igmp_max_memberships))
2207                 goto done;
2208         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2209         if (!iml)
2210                 goto done;
2211
2212         memcpy(&iml->multi, imr, sizeof(*imr));
2213         iml->next_rcu = inet->mc_list;
2214         iml->sflist = NULL;
2215         iml->sfmode = mode;
2216         rcu_assign_pointer(inet->mc_list, iml);
2217         ____ip_mc_inc_group(in_dev, addr, mode, GFP_KERNEL);
2218         err = 0;
2219 done:
2220         return err;
2221 }
2222
2223 /* Join ASM (Any-Source Multicast) group
2224  */
2225 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2226 {
2227         return __ip_mc_join_group(sk, imr, MCAST_EXCLUDE);
2228 }
2229 EXPORT_SYMBOL(ip_mc_join_group);
2230
2231 /* Join SSM (Source-Specific Multicast) group
2232  */
2233 int ip_mc_join_group_ssm(struct sock *sk, struct ip_mreqn *imr,
2234                          unsigned int mode)
2235 {
2236         return __ip_mc_join_group(sk, imr, mode);
2237 }
2238
2239 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2240                            struct in_device *in_dev)
2241 {
2242         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2243         int err;
2244
2245         if (!psf) {
2246                 /* any-source empty exclude case */
2247                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2248                         iml->sfmode, 0, NULL, 0);
2249         }
2250         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2251                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2252         RCU_INIT_POINTER(iml->sflist, NULL);
2253         /* decrease mem now to avoid the memleak warning */
2254         atomic_sub(struct_size(psf, sl_addr, psf->sl_max), &sk->sk_omem_alloc);
2255         kfree_rcu(psf, rcu);
2256         return err;
2257 }
2258
2259 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2260 {
2261         struct inet_sock *inet = inet_sk(sk);
2262         struct ip_mc_socklist *iml;
2263         struct ip_mc_socklist __rcu **imlp;
2264         struct in_device *in_dev;
2265         struct net *net = sock_net(sk);
2266         __be32 group = imr->imr_multiaddr.s_addr;
2267         u32 ifindex;
2268         int ret = -EADDRNOTAVAIL;
2269
2270         ASSERT_RTNL();
2271
2272         in_dev = ip_mc_find_dev(net, imr);
2273         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2274                 ret = -ENODEV;
2275                 goto out;
2276         }
2277         ifindex = imr->imr_ifindex;
2278         for (imlp = &inet->mc_list;
2279              (iml = rtnl_dereference(*imlp)) != NULL;
2280              imlp = &iml->next_rcu) {
2281                 if (iml->multi.imr_multiaddr.s_addr != group)
2282                         continue;
2283                 if (ifindex) {
2284                         if (iml->multi.imr_ifindex != ifindex)
2285                                 continue;
2286                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2287                                 iml->multi.imr_address.s_addr)
2288                         continue;
2289
2290                 (void) ip_mc_leave_src(sk, iml, in_dev);
2291
2292                 *imlp = iml->next_rcu;
2293
2294                 if (in_dev)
2295                         ip_mc_dec_group(in_dev, group);
2296
2297                 /* decrease mem now to avoid the memleak warning */
2298                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2299                 kfree_rcu(iml, rcu);
2300                 return 0;
2301         }
2302 out:
2303         return ret;
2304 }
2305 EXPORT_SYMBOL(ip_mc_leave_group);
2306
2307 int ip_mc_source(int add, int omode, struct sock *sk, struct
2308         ip_mreq_source *mreqs, int ifindex)
2309 {
2310         int err;
2311         struct ip_mreqn imr;
2312         __be32 addr = mreqs->imr_multiaddr;
2313         struct ip_mc_socklist *pmc;
2314         struct in_device *in_dev = NULL;
2315         struct inet_sock *inet = inet_sk(sk);
2316         struct ip_sf_socklist *psl;
2317         struct net *net = sock_net(sk);
2318         int leavegroup = 0;
2319         int i, j, rv;
2320
2321         if (!ipv4_is_multicast(addr))
2322                 return -EINVAL;
2323
2324         ASSERT_RTNL();
2325
2326         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2327         imr.imr_address.s_addr = mreqs->imr_interface;
2328         imr.imr_ifindex = ifindex;
2329         in_dev = ip_mc_find_dev(net, &imr);
2330
2331         if (!in_dev) {
2332                 err = -ENODEV;
2333                 goto done;
2334         }
2335         err = -EADDRNOTAVAIL;
2336
2337         for_each_pmc_rtnl(inet, pmc) {
2338                 if ((pmc->multi.imr_multiaddr.s_addr ==
2339                      imr.imr_multiaddr.s_addr) &&
2340                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2341                         break;
2342         }
2343         if (!pmc) {             /* must have a prior join */
2344                 err = -EINVAL;
2345                 goto done;
2346         }
2347         /* if a source filter was set, must be the same mode as before */
2348         if (pmc->sflist) {
2349                 if (pmc->sfmode != omode) {
2350                         err = -EINVAL;
2351                         goto done;
2352                 }
2353         } else if (pmc->sfmode != omode) {
2354                 /* allow mode switches for empty-set filters */
2355                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2356                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2357                         NULL, 0);
2358                 pmc->sfmode = omode;
2359         }
2360
2361         psl = rtnl_dereference(pmc->sflist);
2362         if (!add) {
2363                 if (!psl)
2364                         goto done;      /* err = -EADDRNOTAVAIL */
2365                 rv = !0;
2366                 for (i = 0; i < psl->sl_count; i++) {
2367                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2368                                 sizeof(__be32));
2369                         if (rv == 0)
2370                                 break;
2371                 }
2372                 if (rv)         /* source not found */
2373                         goto done;      /* err = -EADDRNOTAVAIL */
2374
2375                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2376                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2377                         leavegroup = 1;
2378                         goto done;
2379                 }
2380
2381                 /* update the interface filter */
2382                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2383                         &mreqs->imr_sourceaddr, 1);
2384
2385                 for (j = i+1; j < psl->sl_count; j++)
2386                         psl->sl_addr[j-1] = psl->sl_addr[j];
2387                 psl->sl_count--;
2388                 err = 0;
2389                 goto done;
2390         }
2391         /* else, add a new source to the filter */
2392
2393         if (psl && psl->sl_count >= READ_ONCE(net->ipv4.sysctl_igmp_max_msf)) {
2394                 err = -ENOBUFS;
2395                 goto done;
2396         }
2397         if (!psl || psl->sl_count == psl->sl_max) {
2398                 struct ip_sf_socklist *newpsl;
2399                 int count = IP_SFBLOCK;
2400
2401                 if (psl)
2402                         count += psl->sl_max;
2403                 newpsl = sock_kmalloc(sk, struct_size(newpsl, sl_addr, count),
2404                                       GFP_KERNEL);
2405                 if (!newpsl) {
2406                         err = -ENOBUFS;
2407                         goto done;
2408                 }
2409                 newpsl->sl_max = count;
2410                 newpsl->sl_count = count - IP_SFBLOCK;
2411                 if (psl) {
2412                         for (i = 0; i < psl->sl_count; i++)
2413                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2414                         /* decrease mem now to avoid the memleak warning */
2415                         atomic_sub(struct_size(psl, sl_addr, psl->sl_max),
2416                                    &sk->sk_omem_alloc);
2417                 }
2418                 rcu_assign_pointer(pmc->sflist, newpsl);
2419                 if (psl)
2420                         kfree_rcu(psl, rcu);
2421                 psl = newpsl;
2422         }
2423         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2424         for (i = 0; i < psl->sl_count; i++) {
2425                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2426                         sizeof(__be32));
2427                 if (rv == 0)
2428                         break;
2429         }
2430         if (rv == 0)            /* address already there is an error */
2431                 goto done;
2432         for (j = psl->sl_count-1; j >= i; j--)
2433                 psl->sl_addr[j+1] = psl->sl_addr[j];
2434         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2435         psl->sl_count++;
2436         err = 0;
2437         /* update the interface list */
2438         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2439                 &mreqs->imr_sourceaddr, 1);
2440 done:
2441         if (leavegroup)
2442                 err = ip_mc_leave_group(sk, &imr);
2443         return err;
2444 }
2445
2446 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2447 {
2448         int err = 0;
2449         struct ip_mreqn imr;
2450         __be32 addr = msf->imsf_multiaddr;
2451         struct ip_mc_socklist *pmc;
2452         struct in_device *in_dev;
2453         struct inet_sock *inet = inet_sk(sk);
2454         struct ip_sf_socklist *newpsl, *psl;
2455         struct net *net = sock_net(sk);
2456         int leavegroup = 0;
2457
2458         if (!ipv4_is_multicast(addr))
2459                 return -EINVAL;
2460         if (msf->imsf_fmode != MCAST_INCLUDE &&
2461             msf->imsf_fmode != MCAST_EXCLUDE)
2462                 return -EINVAL;
2463
2464         ASSERT_RTNL();
2465
2466         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2467         imr.imr_address.s_addr = msf->imsf_interface;
2468         imr.imr_ifindex = ifindex;
2469         in_dev = ip_mc_find_dev(net, &imr);
2470
2471         if (!in_dev) {
2472                 err = -ENODEV;
2473                 goto done;
2474         }
2475
2476         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2477         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2478                 leavegroup = 1;
2479                 goto done;
2480         }
2481
2482         for_each_pmc_rtnl(inet, pmc) {
2483                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2484                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2485                         break;
2486         }
2487         if (!pmc) {             /* must have a prior join */
2488                 err = -EINVAL;
2489                 goto done;
2490         }
2491         if (msf->imsf_numsrc) {
2492                 newpsl = sock_kmalloc(sk, struct_size(newpsl, sl_addr,
2493                                                       msf->imsf_numsrc),
2494                                       GFP_KERNEL);
2495                 if (!newpsl) {
2496                         err = -ENOBUFS;
2497                         goto done;
2498                 }
2499                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2500                 memcpy(newpsl->sl_addr, msf->imsf_slist_flex,
2501                        flex_array_size(msf, imsf_slist_flex, msf->imsf_numsrc));
2502                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2503                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2504                 if (err) {
2505                         sock_kfree_s(sk, newpsl,
2506                                      struct_size(newpsl, sl_addr,
2507                                                  newpsl->sl_max));
2508                         goto done;
2509                 }
2510         } else {
2511                 newpsl = NULL;
2512                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2513                                      msf->imsf_fmode, 0, NULL, 0);
2514         }
2515         psl = rtnl_dereference(pmc->sflist);
2516         if (psl) {
2517                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2518                         psl->sl_count, psl->sl_addr, 0);
2519                 /* decrease mem now to avoid the memleak warning */
2520                 atomic_sub(struct_size(psl, sl_addr, psl->sl_max),
2521                            &sk->sk_omem_alloc);
2522         } else {
2523                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2524                         0, NULL, 0);
2525         }
2526         rcu_assign_pointer(pmc->sflist, newpsl);
2527         if (psl)
2528                 kfree_rcu(psl, rcu);
2529         pmc->sfmode = msf->imsf_fmode;
2530         err = 0;
2531 done:
2532         if (leavegroup)
2533                 err = ip_mc_leave_group(sk, &imr);
2534         return err;
2535 }
2536 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2537                  sockptr_t optval, sockptr_t optlen)
2538 {
2539         int err, len, count, copycount, msf_size;
2540         struct ip_mreqn imr;
2541         __be32 addr = msf->imsf_multiaddr;
2542         struct ip_mc_socklist *pmc;
2543         struct in_device *in_dev;
2544         struct inet_sock *inet = inet_sk(sk);
2545         struct ip_sf_socklist *psl;
2546         struct net *net = sock_net(sk);
2547
2548         ASSERT_RTNL();
2549
2550         if (!ipv4_is_multicast(addr))
2551                 return -EINVAL;
2552
2553         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2554         imr.imr_address.s_addr = msf->imsf_interface;
2555         imr.imr_ifindex = 0;
2556         in_dev = ip_mc_find_dev(net, &imr);
2557
2558         if (!in_dev) {
2559                 err = -ENODEV;
2560                 goto done;
2561         }
2562         err = -EADDRNOTAVAIL;
2563
2564         for_each_pmc_rtnl(inet, pmc) {
2565                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2566                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2567                         break;
2568         }
2569         if (!pmc)               /* must have a prior join */
2570                 goto done;
2571         msf->imsf_fmode = pmc->sfmode;
2572         psl = rtnl_dereference(pmc->sflist);
2573         if (!psl) {
2574                 count = 0;
2575         } else {
2576                 count = psl->sl_count;
2577         }
2578         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2579         len = flex_array_size(psl, sl_addr, copycount);
2580         msf->imsf_numsrc = count;
2581         msf_size = IP_MSFILTER_SIZE(copycount);
2582         if (copy_to_sockptr(optlen, &msf_size, sizeof(int)) ||
2583             copy_to_sockptr(optval, msf, IP_MSFILTER_SIZE(0))) {
2584                 return -EFAULT;
2585         }
2586         if (len &&
2587             copy_to_sockptr_offset(optval,
2588                                    offsetof(struct ip_msfilter, imsf_slist_flex),
2589                                    psl->sl_addr, len))
2590                 return -EFAULT;
2591         return 0;
2592 done:
2593         return err;
2594 }
2595
2596 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2597                  sockptr_t optval, size_t ss_offset)
2598 {
2599         int i, count, copycount;
2600         struct sockaddr_in *psin;
2601         __be32 addr;
2602         struct ip_mc_socklist *pmc;
2603         struct inet_sock *inet = inet_sk(sk);
2604         struct ip_sf_socklist *psl;
2605
2606         ASSERT_RTNL();
2607
2608         psin = (struct sockaddr_in *)&gsf->gf_group;
2609         if (psin->sin_family != AF_INET)
2610                 return -EINVAL;
2611         addr = psin->sin_addr.s_addr;
2612         if (!ipv4_is_multicast(addr))
2613                 return -EINVAL;
2614
2615         for_each_pmc_rtnl(inet, pmc) {
2616                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2617                     pmc->multi.imr_ifindex == gsf->gf_interface)
2618                         break;
2619         }
2620         if (!pmc)               /* must have a prior join */
2621                 return -EADDRNOTAVAIL;
2622         gsf->gf_fmode = pmc->sfmode;
2623         psl = rtnl_dereference(pmc->sflist);
2624         count = psl ? psl->sl_count : 0;
2625         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2626         gsf->gf_numsrc = count;
2627         for (i = 0; i < copycount; i++) {
2628                 struct sockaddr_storage ss;
2629
2630                 psin = (struct sockaddr_in *)&ss;
2631                 memset(&ss, 0, sizeof(ss));
2632                 psin->sin_family = AF_INET;
2633                 psin->sin_addr.s_addr = psl->sl_addr[i];
2634                 if (copy_to_sockptr_offset(optval, ss_offset,
2635                                            &ss, sizeof(ss)))
2636                         return -EFAULT;
2637                 ss_offset += sizeof(ss);
2638         }
2639         return 0;
2640 }
2641
2642 /*
2643  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2644  */
2645 int ip_mc_sf_allow(const struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2646                    int dif, int sdif)
2647 {
2648         const struct inet_sock *inet = inet_sk(sk);
2649         struct ip_mc_socklist *pmc;
2650         struct ip_sf_socklist *psl;
2651         int i;
2652         int ret;
2653
2654         ret = 1;
2655         if (!ipv4_is_multicast(loc_addr))
2656                 goto out;
2657
2658         rcu_read_lock();
2659         for_each_pmc_rcu(inet, pmc) {
2660                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2661                     (pmc->multi.imr_ifindex == dif ||
2662                      (sdif && pmc->multi.imr_ifindex == sdif)))
2663                         break;
2664         }
2665         ret = inet_test_bit(MC_ALL, sk);
2666         if (!pmc)
2667                 goto unlock;
2668         psl = rcu_dereference(pmc->sflist);
2669         ret = (pmc->sfmode == MCAST_EXCLUDE);
2670         if (!psl)
2671                 goto unlock;
2672
2673         for (i = 0; i < psl->sl_count; i++) {
2674                 if (psl->sl_addr[i] == rmt_addr)
2675                         break;
2676         }
2677         ret = 0;
2678         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2679                 goto unlock;
2680         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2681                 goto unlock;
2682         ret = 1;
2683 unlock:
2684         rcu_read_unlock();
2685 out:
2686         return ret;
2687 }
2688
2689 /*
2690  *      A socket is closing.
2691  */
2692
2693 void ip_mc_drop_socket(struct sock *sk)
2694 {
2695         struct inet_sock *inet = inet_sk(sk);
2696         struct ip_mc_socklist *iml;
2697         struct net *net = sock_net(sk);
2698
2699         if (!inet->mc_list)
2700                 return;
2701
2702         rtnl_lock();
2703         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2704                 struct in_device *in_dev;
2705
2706                 inet->mc_list = iml->next_rcu;
2707                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2708                 (void) ip_mc_leave_src(sk, iml, in_dev);
2709                 if (in_dev)
2710                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2711                 /* decrease mem now to avoid the memleak warning */
2712                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2713                 kfree_rcu(iml, rcu);
2714         }
2715         rtnl_unlock();
2716 }
2717
2718 /* called with rcu_read_lock() */
2719 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2720 {
2721         struct ip_mc_list *im;
2722         struct ip_mc_list __rcu **mc_hash;
2723         struct ip_sf_list *psf;
2724         int rv = 0;
2725
2726         mc_hash = rcu_dereference(in_dev->mc_hash);
2727         if (mc_hash) {
2728                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2729
2730                 for (im = rcu_dereference(mc_hash[hash]);
2731                      im != NULL;
2732                      im = rcu_dereference(im->next_hash)) {
2733                         if (im->multiaddr == mc_addr)
2734                                 break;
2735                 }
2736         } else {
2737                 for_each_pmc_rcu(in_dev, im) {
2738                         if (im->multiaddr == mc_addr)
2739                                 break;
2740                 }
2741         }
2742         if (im && proto == IPPROTO_IGMP) {
2743                 rv = 1;
2744         } else if (im) {
2745                 if (src_addr) {
2746                         spin_lock_bh(&im->lock);
2747                         for (psf = im->sources; psf; psf = psf->sf_next) {
2748                                 if (psf->sf_inaddr == src_addr)
2749                                         break;
2750                         }
2751                         if (psf)
2752                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2753                                         psf->sf_count[MCAST_EXCLUDE] !=
2754                                         im->sfcount[MCAST_EXCLUDE];
2755                         else
2756                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2757                         spin_unlock_bh(&im->lock);
2758                 } else
2759                         rv = 1; /* unspecified source; tentatively allow */
2760         }
2761         return rv;
2762 }
2763
2764 #if defined(CONFIG_PROC_FS)
2765 struct igmp_mc_iter_state {
2766         struct seq_net_private p;
2767         struct net_device *dev;
2768         struct in_device *in_dev;
2769 };
2770
2771 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2772
2773 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2774 {
2775         struct net *net = seq_file_net(seq);
2776         struct ip_mc_list *im = NULL;
2777         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2778
2779         state->in_dev = NULL;
2780         for_each_netdev_rcu(net, state->dev) {
2781                 struct in_device *in_dev;
2782
2783                 in_dev = __in_dev_get_rcu(state->dev);
2784                 if (!in_dev)
2785                         continue;
2786                 im = rcu_dereference(in_dev->mc_list);
2787                 if (im) {
2788                         state->in_dev = in_dev;
2789                         break;
2790                 }
2791         }
2792         return im;
2793 }
2794
2795 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2796 {
2797         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2798
2799         im = rcu_dereference(im->next_rcu);
2800         while (!im) {
2801                 state->dev = next_net_device_rcu(state->dev);
2802                 if (!state->dev) {
2803                         state->in_dev = NULL;
2804                         break;
2805                 }
2806                 state->in_dev = __in_dev_get_rcu(state->dev);
2807                 if (!state->in_dev)
2808                         continue;
2809                 im = rcu_dereference(state->in_dev->mc_list);
2810         }
2811         return im;
2812 }
2813
2814 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2815 {
2816         struct ip_mc_list *im = igmp_mc_get_first(seq);
2817         if (im)
2818                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2819                         --pos;
2820         return pos ? NULL : im;
2821 }
2822
2823 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2824         __acquires(rcu)
2825 {
2826         rcu_read_lock();
2827         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2828 }
2829
2830 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2831 {
2832         struct ip_mc_list *im;
2833         if (v == SEQ_START_TOKEN)
2834                 im = igmp_mc_get_first(seq);
2835         else
2836                 im = igmp_mc_get_next(seq, v);
2837         ++*pos;
2838         return im;
2839 }
2840
2841 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2842         __releases(rcu)
2843 {
2844         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2845
2846         state->in_dev = NULL;
2847         state->dev = NULL;
2848         rcu_read_unlock();
2849 }
2850
2851 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2852 {
2853         if (v == SEQ_START_TOKEN)
2854                 seq_puts(seq,
2855                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2856         else {
2857                 struct ip_mc_list *im = v;
2858                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2859                 char   *querier;
2860                 long delta;
2861
2862 #ifdef CONFIG_IP_MULTICAST
2863                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2864                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2865                           "V3";
2866 #else
2867                 querier = "NONE";
2868 #endif
2869
2870                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2871                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2872                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2873                 }
2874
2875                 delta = im->timer.expires - jiffies;
2876                 seq_printf(seq,
2877                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2878                            im->multiaddr, im->users,
2879                            im->tm_running,
2880                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2881                            im->reporter);
2882         }
2883         return 0;
2884 }
2885
2886 static const struct seq_operations igmp_mc_seq_ops = {
2887         .start  =       igmp_mc_seq_start,
2888         .next   =       igmp_mc_seq_next,
2889         .stop   =       igmp_mc_seq_stop,
2890         .show   =       igmp_mc_seq_show,
2891 };
2892
2893 struct igmp_mcf_iter_state {
2894         struct seq_net_private p;
2895         struct net_device *dev;
2896         struct in_device *idev;
2897         struct ip_mc_list *im;
2898 };
2899
2900 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2901
2902 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2903 {
2904         struct net *net = seq_file_net(seq);
2905         struct ip_sf_list *psf = NULL;
2906         struct ip_mc_list *im = NULL;
2907         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2908
2909         state->idev = NULL;
2910         state->im = NULL;
2911         for_each_netdev_rcu(net, state->dev) {
2912                 struct in_device *idev;
2913                 idev = __in_dev_get_rcu(state->dev);
2914                 if (unlikely(!idev))
2915                         continue;
2916                 im = rcu_dereference(idev->mc_list);
2917                 if (likely(im)) {
2918                         spin_lock_bh(&im->lock);
2919                         psf = im->sources;
2920                         if (likely(psf)) {
2921                                 state->im = im;
2922                                 state->idev = idev;
2923                                 break;
2924                         }
2925                         spin_unlock_bh(&im->lock);
2926                 }
2927         }
2928         return psf;
2929 }
2930
2931 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2932 {
2933         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2934
2935         psf = psf->sf_next;
2936         while (!psf) {
2937                 spin_unlock_bh(&state->im->lock);
2938                 state->im = state->im->next;
2939                 while (!state->im) {
2940                         state->dev = next_net_device_rcu(state->dev);
2941                         if (!state->dev) {
2942                                 state->idev = NULL;
2943                                 goto out;
2944                         }
2945                         state->idev = __in_dev_get_rcu(state->dev);
2946                         if (!state->idev)
2947                                 continue;
2948                         state->im = rcu_dereference(state->idev->mc_list);
2949                 }
2950                 spin_lock_bh(&state->im->lock);
2951                 psf = state->im->sources;
2952         }
2953 out:
2954         return psf;
2955 }
2956
2957 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2958 {
2959         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2960         if (psf)
2961                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2962                         --pos;
2963         return pos ? NULL : psf;
2964 }
2965
2966 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2967         __acquires(rcu)
2968 {
2969         rcu_read_lock();
2970         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2971 }
2972
2973 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2974 {
2975         struct ip_sf_list *psf;
2976         if (v == SEQ_START_TOKEN)
2977                 psf = igmp_mcf_get_first(seq);
2978         else
2979                 psf = igmp_mcf_get_next(seq, v);
2980         ++*pos;
2981         return psf;
2982 }
2983
2984 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2985         __releases(rcu)
2986 {
2987         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2988         if (likely(state->im)) {
2989                 spin_unlock_bh(&state->im->lock);
2990                 state->im = NULL;
2991         }
2992         state->idev = NULL;
2993         state->dev = NULL;
2994         rcu_read_unlock();
2995 }
2996
2997 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2998 {
2999         struct ip_sf_list *psf = v;
3000         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
3001
3002         if (v == SEQ_START_TOKEN) {
3003                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
3004         } else {
3005                 seq_printf(seq,
3006                            "%3d %6.6s 0x%08x "
3007                            "0x%08x %6lu %6lu\n",
3008                            state->dev->ifindex, state->dev->name,
3009                            ntohl(state->im->multiaddr),
3010                            ntohl(psf->sf_inaddr),
3011                            psf->sf_count[MCAST_INCLUDE],
3012                            psf->sf_count[MCAST_EXCLUDE]);
3013         }
3014         return 0;
3015 }
3016
3017 static const struct seq_operations igmp_mcf_seq_ops = {
3018         .start  =       igmp_mcf_seq_start,
3019         .next   =       igmp_mcf_seq_next,
3020         .stop   =       igmp_mcf_seq_stop,
3021         .show   =       igmp_mcf_seq_show,
3022 };
3023
3024 static int __net_init igmp_net_init(struct net *net)
3025 {
3026         struct proc_dir_entry *pde;
3027         int err;
3028
3029         pde = proc_create_net("igmp", 0444, net->proc_net, &igmp_mc_seq_ops,
3030                         sizeof(struct igmp_mc_iter_state));
3031         if (!pde)
3032                 goto out_igmp;
3033         pde = proc_create_net("mcfilter", 0444, net->proc_net,
3034                         &igmp_mcf_seq_ops, sizeof(struct igmp_mcf_iter_state));
3035         if (!pde)
3036                 goto out_mcfilter;
3037         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3038                                    SOCK_DGRAM, 0, net);
3039         if (err < 0) {
3040                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3041                        err);
3042                 goto out_sock;
3043         }
3044
3045         return 0;
3046
3047 out_sock:
3048         remove_proc_entry("mcfilter", net->proc_net);
3049 out_mcfilter:
3050         remove_proc_entry("igmp", net->proc_net);
3051 out_igmp:
3052         return -ENOMEM;
3053 }
3054
3055 static void __net_exit igmp_net_exit(struct net *net)
3056 {
3057         remove_proc_entry("mcfilter", net->proc_net);
3058         remove_proc_entry("igmp", net->proc_net);
3059         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3060 }
3061
3062 static struct pernet_operations igmp_net_ops = {
3063         .init = igmp_net_init,
3064         .exit = igmp_net_exit,
3065 };
3066 #endif
3067
3068 static int igmp_netdev_event(struct notifier_block *this,
3069                              unsigned long event, void *ptr)
3070 {
3071         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3072         struct in_device *in_dev;
3073
3074         switch (event) {
3075         case NETDEV_RESEND_IGMP:
3076                 in_dev = __in_dev_get_rtnl(dev);
3077                 if (in_dev)
3078                         ip_mc_rejoin_groups(in_dev);
3079                 break;
3080         default:
3081                 break;
3082         }
3083         return NOTIFY_DONE;
3084 }
3085
3086 static struct notifier_block igmp_notifier = {
3087         .notifier_call = igmp_netdev_event,
3088 };
3089
3090 int __init igmp_mc_init(void)
3091 {
3092 #if defined(CONFIG_PROC_FS)
3093         int err;
3094
3095         err = register_pernet_subsys(&igmp_net_ops);
3096         if (err)
3097                 return err;
3098         err = register_netdevice_notifier(&igmp_notifier);
3099         if (err)
3100                 goto reg_notif_fail;
3101         return 0;
3102
3103 reg_notif_fail:
3104         unregister_pernet_subsys(&igmp_net_ops);
3105         return err;
3106 #else
3107         return register_netdevice_notifier(&igmp_notifier);
3108 #endif
3109 }