net/ipv4/igmp.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux NET3:     Internet Group Management Protocol  [IGMP]
   4  *
   5  *      This code implements the IGMP protocol as defined in RFC1112. There has
   6  *      been a further revision of this protocol since which is now supported.
   7  *
   8  *      If you have trouble with this module be careful what gcc you have used,
   9  *      the older version didn't come out right using gcc 2.5.8, the newer one
  10  *      seems to fall out with gcc 2.6.2.
  11  *
  12  *      Authors:
  13  *              Alan Cox <[email protected]>
  14  *
  15  *      Fixes:
  16  *
  17  *              Alan Cox        :       Added lots of __inline__ to optimise
  18  *                                      the memory usage of all the tiny little
  19  *                                      functions.
  20  *              Alan Cox        :       Dumped the header building experiment.
  21  *              Alan Cox        :       Minor tweaks ready for multicast routing
  22  *                                      and extended IGMP protocol.
  23  *              Alan Cox        :       Removed a load of inline directives. Gcc 2.5.8
  24  *                                      writes utterly bogus code otherwise (sigh)
  25  *                                      fixed IGMP loopback to behave in the manner
  26  *                                      desired by mrouted, fixed the fact it has been
  27  *                                      broken since 1.3.6 and cleaned up a few minor
  28  *                                      points.
  29  *
  30  *              Chih-Jen Chang  :       Tried to revise IGMP to Version 2
  31  *              Tsu-Sheng Tsao          E-mail: [email protected] and [email protected]
  32  *                                      The enhancements are mainly based on Steve Deering's
  33  *                                      ipmulti-3.5 source code.
  34  *              Chih-Jen Chang  :       Added the igmp_get_mrouter_info and
  35  *              Tsu-Sheng Tsao          igmp_set_mrouter_info to keep track of
  36  *                                      the mrouted version on that device.
  37  *              Chih-Jen Chang  :       Added the max_resp_time parameter to
  38  *              Tsu-Sheng Tsao          igmp_heard_query(). Using this parameter
  39  *                                      to identify the multicast router version
  40  *                                      and do what the IGMP version 2 specified.
  41  *              Chih-Jen Chang  :       Added a timer to revert to IGMP V2 router
  42  *              Tsu-Sheng Tsao          if the specified time expired.
  43  *              Alan Cox        :       Stop IGMP from 0.0.0.0 being accepted.
  44  *              Alan Cox        :       Use GFP_ATOMIC in the right places.
  45  *              Christian Daudt :       igmp timer wasn't set for local group
  46  *                                      memberships but was being deleted,
  47  *                                      which caused a "del_timer() called
  48  *                                      from %p with timer not initialized\n"
  49  *                                      message (960131).
  50  *              Christian Daudt :       removed del_timer from
  51  *                                      igmp_timer_expire function (960205).
  52  *             Christian Daudt :       igmp_heard_report now only calls
  53  *                                     igmp_timer_expire if tm->running is
  54  *                                     true (960216).
  55  *              Malcolm Beattie :       ttl comparison wrong in igmp_rcv made
  56  *                                      igmp_heard_query never trigger. Expiry
  57  *                                      miscalculation fixed in igmp_heard_query
  58  *                                      and random() made to return unsigned to
  59  *                                      prevent negative expiry times.
  60  *              Alexey Kuznetsov:       Wrong group leaving behaviour, backport
  61  *                                      fix from pending 2.1.x patches.
  62  *              Alan Cox:               Forget to enable FDDI support earlier.
  63  *              Alexey Kuznetsov:       Fixed leaving groups on device down.
  64  *              Alexey Kuznetsov:       Accordance to igmp-v2-06 draft.
  65  *              David L Stevens:        IGMPv3 support, with help from
  66  *                                      Vinay Kulkarni
  67  */
  68
  69 #include <linux/module.h>
  70 #include <linux/slab.h>
  71 #include <linux/uaccess.h>
  72 #include <linux/types.h>
  73 #include <linux/kernel.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/string.h>
  76 #include <linux/socket.h>
  77 #include <linux/sockios.h>
  78 #include <linux/in.h>
  79 #include <linux/inet.h>
  80 #include <linux/netdevice.h>
  81 #include <linux/skbuff.h>
  82 #include <linux/inetdevice.h>
  83 #include <linux/igmp.h>
  84 #include <linux/if_arp.h>
  85 #include <linux/rtnetlink.h>
  86 #include <linux/times.h>
  87 #include <linux/pkt_sched.h>
  88 #include <linux/byteorder/generic.h>
  89
  90 #include <net/net_namespace.h>
  91 #include <net/arp.h>
  92 #include <net/ip.h>
  93 #include <net/protocol.h>
  94 #include <net/route.h>
  95 #include <net/sock.h>
  96 #include <net/checksum.h>
  97 #include <net/inet_common.h>
  98 #include <linux/netfilter_ipv4.h>
  99 #ifdef CONFIG_IP_MROUTE
 100 #include <linux/mroute.h>
 101 #endif
 102 #ifdef CONFIG_PROC_FS
 103 #include <linux/proc_fs.h>
 104 #include <linux/seq_file.h>
 105 #endif
 106
 107 #ifdef CONFIG_IP_MULTICAST
 108 /* Parameter names and values are taken from igmp-v2-06 draft */
 109
 110 #define IGMP_QUERY_INTERVAL                     (125*HZ)
 111 #define IGMP_QUERY_RESPONSE_INTERVAL            (10*HZ)
 112
 113 #define IGMP_INITIAL_REPORT_DELAY               (1)
 114
 115 /* IGMP_INITIAL_REPORT_DELAY is not from IGMP specs!
 116  * IGMP specs require to report membership immediately after
 117  * joining a group, but we delay the first report by a
 118  * small interval. It seems more natural and still does not
 119  * contradict to specs provided this delay is small enough.
 120  */
 121
 122 #define IGMP_V1_SEEN(in_dev) \
 123         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
 124          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
 125          ((in_dev)->mr_v1_seen && \
 126           time_before(jiffies, (in_dev)->mr_v1_seen)))
 127 #define IGMP_V2_SEEN(in_dev) \
 128         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
 129          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
 130          ((in_dev)->mr_v2_seen && \
 131           time_before(jiffies, (in_dev)->mr_v2_seen)))
 132
 133 static int unsolicited_report_interval(struct in_device *in_dev)
 134 {
 135         int interval_ms, interval_jiffies;
 136
 137         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 138                 interval_ms = IN_DEV_CONF_GET(
 139                         in_dev,
 140                         IGMPV2_UNSOLICITED_REPORT_INTERVAL);
 141         else /* v3 */
 142                 interval_ms = IN_DEV_CONF_GET(
 143                         in_dev,
 144                         IGMPV3_UNSOLICITED_REPORT_INTERVAL);
 145
 146         interval_jiffies = msecs_to_jiffies(interval_ms);
 147
 148         /* _timer functions can't handle a delay of 0 jiffies so ensure
 149          *  we always return a positive value.
 150          */
 151         if (interval_jiffies <= 0)
 152                 interval_jiffies = 1;
 153         return interval_jiffies;
 154 }
 155
 156 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
 157                               gfp_t gfp);
 158 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 159 static void igmpv3_clear_delrec(struct in_device *in_dev);
 160 static int sf_setstate(struct ip_mc_list *pmc);
 161 static void sf_markstate(struct ip_mc_list *pmc);
 162 #endif
 163 static void ip_mc_clear_src(struct ip_mc_list *pmc);
 164 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
 165                          int sfcount, __be32 *psfsrc, int delta);
 166
 167 static void ip_ma_put(struct ip_mc_list *im)
 168 {
 169         if (refcount_dec_and_test(&im->refcnt)) {
 170                 in_dev_put(im->interface);
 171                 kfree_rcu(im, rcu);
 172         }
 173 }
 174
 175 #define for_each_pmc_rcu(in_dev, pmc)                           \
 176         for (pmc = rcu_dereference(in_dev->mc_list);            \
 177              pmc != NULL;                                       \
 178              pmc = rcu_dereference(pmc->next_rcu))
 179
 180 #define for_each_pmc_rtnl(in_dev, pmc)                          \
 181         for (pmc = rtnl_dereference(in_dev->mc_list);           \
 182              pmc != NULL;                                       \
 183              pmc = rtnl_dereference(pmc->next_rcu))
 184
 185 static void ip_sf_list_clear_all(struct ip_sf_list *psf)
 186 {
 187         struct ip_sf_list *next;
 188
 189         while (psf) {
 190                 next = psf->sf_next;
 191                 kfree(psf);
 192                 psf = next;
 193         }
 194 }
 195
 196 #ifdef CONFIG_IP_MULTICAST
 197
 198 /*
 199  *      Timer management
 200  */
 201
 202 static void igmp_stop_timer(struct ip_mc_list *im)
 203 {
 204         spin_lock_bh(&im->lock);
 205         if (del_timer(&im->timer))
 206                 refcount_dec(&im->refcnt);
 207         im->tm_running = 0;
 208         im->reporter = 0;
 209         im->unsolicit_count = 0;
 210         spin_unlock_bh(&im->lock);
 211 }
 212
 213 /* It must be called with locked im->lock */
 214 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
 215 {
 216         int tv = prandom_u32() % max_delay;
 217
 218         im->tm_running = 1;
 219         if (!mod_timer(&im->timer, jiffies+tv+2))
 220                 refcount_inc(&im->refcnt);
 221 }
 222
 223 static void igmp_gq_start_timer(struct in_device *in_dev)
 224 {
 225         int tv = prandom_u32() % in_dev->mr_maxdelay;
 226         unsigned long exp = jiffies + tv + 2;
 227
 228         if (in_dev->mr_gq_running &&
 229             time_after_eq(exp, (in_dev->mr_gq_timer).expires))
 230                 return;
 231
 232         in_dev->mr_gq_running = 1;
 233         if (!mod_timer(&in_dev->mr_gq_timer, exp))
 234                 in_dev_hold(in_dev);
 235 }
 236
 237 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 238 {
 239         int tv = prandom_u32() % delay;
 240
 241         if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
 242                 in_dev_hold(in_dev);
 243 }
 244
 245 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 246 {
 247         spin_lock_bh(&im->lock);
 248         im->unsolicit_count = 0;
 249         if (del_timer(&im->timer)) {
 250                 if ((long)(im->timer.expires-jiffies) < max_delay) {
 251                         add_timer(&im->timer);
 252                         im->tm_running = 1;
 253                         spin_unlock_bh(&im->lock);
 254                         return;
 255                 }
 256                 refcount_dec(&im->refcnt);
 257         }
 258         igmp_start_timer(im, max_delay);
 259         spin_unlock_bh(&im->lock);
 260 }
 261
 262
 263 /*
 264  *      Send an IGMP report.
 265  */
 266
 267 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
 268
 269
 270 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
 271         int gdeleted, int sdeleted)
 272 {
 273         switch (type) {
 274         case IGMPV3_MODE_IS_INCLUDE:
 275         case IGMPV3_MODE_IS_EXCLUDE:
 276                 if (gdeleted || sdeleted)
 277                         return 0;
 278                 if (!(pmc->gsquery && !psf->sf_gsresp)) {
 279                         if (pmc->sfmode == MCAST_INCLUDE)
 280                                 return 1;
 281                         /* don't include if this source is excluded
 282                          * in all filters
 283                          */
 284                         if (psf->sf_count[MCAST_INCLUDE])
 285                                 return type == IGMPV3_MODE_IS_INCLUDE;
 286                         return pmc->sfcount[MCAST_EXCLUDE] ==
 287                                 psf->sf_count[MCAST_EXCLUDE];
 288                 }
 289                 return 0;
 290         case IGMPV3_CHANGE_TO_INCLUDE:
 291                 if (gdeleted || sdeleted)
 292                         return 0;
 293                 return psf->sf_count[MCAST_INCLUDE] != 0;
 294         case IGMPV3_CHANGE_TO_EXCLUDE:
 295                 if (gdeleted || sdeleted)
 296                         return 0;
 297                 if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
 298                     psf->sf_count[MCAST_INCLUDE])
 299                         return 0;
 300                 return pmc->sfcount[MCAST_EXCLUDE] ==
 301                         psf->sf_count[MCAST_EXCLUDE];
 302         case IGMPV3_ALLOW_NEW_SOURCES:
 303                 if (gdeleted || !psf->sf_crcount)
 304                         return 0;
 305                 return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
 306         case IGMPV3_BLOCK_OLD_SOURCES:
 307                 if (pmc->sfmode == MCAST_INCLUDE)
 308                         return gdeleted || (psf->sf_crcount && sdeleted);
 309                 return psf->sf_crcount && !gdeleted && !sdeleted;
 310         }
 311         return 0;
 312 }
 313
 314 static int
 315 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
 316 {
 317         struct ip_sf_list *psf;
 318         int scount = 0;
 319
 320         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 321                 if (!is_in(pmc, psf, type, gdeleted, sdeleted))
 322                         continue;
 323                 scount++;
 324         }
 325         return scount;
 326 }
 327
 328 /* source address selection per RFC 3376 section 4.2.13 */
 329 static __be32 igmpv3_get_srcaddr(struct net_device *dev,
 330                                  const struct flowi4 *fl4)
 331 {
 332         struct in_device *in_dev = __in_dev_get_rcu(dev);
 333         const struct in_ifaddr *ifa;
 334
 335         if (!in_dev)
 336                 return htonl(INADDR_ANY);
 337
 338         in_dev_for_each_ifa_rcu(ifa, in_dev) {
 339                 if (fl4->saddr == ifa->ifa_local)
 340                         return fl4->saddr;
 341         }
 342
 343         return htonl(INADDR_ANY);
 344 }
 345
 346 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
 347 {
 348         struct sk_buff *skb;
 349         struct rtable *rt;
 350         struct iphdr *pip;
 351         struct igmpv3_report *pig;
 352         struct net *net = dev_net(dev);
 353         struct flowi4 fl4;
 354         int hlen = LL_RESERVED_SPACE(dev);
 355         int tlen = dev->needed_tailroom;
 356         unsigned int size = mtu;
 357
 358         while (1) {
 359                 skb = alloc_skb(size + hlen + tlen,
 360                                 GFP_ATOMIC | __GFP_NOWARN);
 361                 if (skb)
 362                         break;
 363                 size >>= 1;
 364                 if (size < 256)
 365                         return NULL;
 366         }
 367         skb->priority = TC_PRIO_CONTROL;
 368
 369         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 370                                    0, 0,
 371                                    IPPROTO_IGMP, 0, dev->ifindex);
 372         if (IS_ERR(rt)) {
 373                 kfree_skb(skb);
 374                 return NULL;
 375         }
 376
 377         skb_dst_set(skb, &rt->dst);
 378         skb->dev = dev;
 379
 380         skb_reserve(skb, hlen);
 381         skb_tailroom_reserve(skb, mtu, tlen);
 382
 383         skb_reset_network_header(skb);
 384         pip = ip_hdr(skb);
 385         skb_put(skb, sizeof(struct iphdr) + 4);
 386
 387         pip->version  = 4;
 388         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 389         pip->tos      = 0xc0;
 390         pip->frag_off = htons(IP_DF);
 391         pip->ttl      = 1;
 392         pip->daddr    = fl4.daddr;
 393
 394         rcu_read_lock();
 395         pip->saddr    = igmpv3_get_srcaddr(dev, &fl4);
 396         rcu_read_unlock();
 397
 398         pip->protocol = IPPROTO_IGMP;
 399         pip->tot_len  = 0;      /* filled in later */
 400         ip_select_ident(net, skb, NULL);
 401         ((u8 *)&pip[1])[0] = IPOPT_RA;
 402         ((u8 *)&pip[1])[1] = 4;
 403         ((u8 *)&pip[1])[2] = 0;
 404         ((u8 *)&pip[1])[3] = 0;
 405
 406         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 407         skb_put(skb, sizeof(*pig));
 408         pig = igmpv3_report_hdr(skb);
 409         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 410         pig->resv1 = 0;
 411         pig->csum = 0;
 412         pig->resv2 = 0;
 413         pig->ngrec = 0;
 414         return skb;
 415 }
 416
 417 static int igmpv3_sendpack(struct sk_buff *skb)
 418 {
 419         struct igmphdr *pig = igmp_hdr(skb);
 420         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 421
 422         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 423
 424         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 425 }
 426
 427 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 428 {
 429         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 430 }
 431
 432 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 433         int type, struct igmpv3_grec **ppgr, unsigned int mtu)
 434 {
 435         struct net_device *dev = pmc->interface->dev;
 436         struct igmpv3_report *pih;
 437         struct igmpv3_grec *pgr;
 438
 439         if (!skb) {
 440                 skb = igmpv3_newpack(dev, mtu);
 441                 if (!skb)
 442                         return NULL;
 443         }
 444         pgr = skb_put(skb, sizeof(struct igmpv3_grec));
 445         pgr->grec_type = type;
 446         pgr->grec_auxwords = 0;
 447         pgr->grec_nsrcs = 0;
 448         pgr->grec_mca = pmc->multiaddr;
 449         pih = igmpv3_report_hdr(skb);
 450         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 451         *ppgr = pgr;
 452         return skb;
 453 }
 454
 455 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 456
 457 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 458         int type, int gdeleted, int sdeleted)
 459 {
 460         struct net_device *dev = pmc->interface->dev;
 461         struct net *net = dev_net(dev);
 462         struct igmpv3_report *pih;
 463         struct igmpv3_grec *pgr = NULL;
 464         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 465         int scount, stotal, first, isquery, truncate;
 466         unsigned int mtu;
 467
 468         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 469                 return skb;
 470         if (ipv4_is_local_multicast(pmc->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
 471                 return skb;
 472
 473         mtu = READ_ONCE(dev->mtu);
 474         if (mtu < IPV4_MIN_MTU)
 475                 return skb;
 476
 477         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 478                   type == IGMPV3_MODE_IS_EXCLUDE;
 479         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 480                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 481
 482         stotal = scount = 0;
 483
 484         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 485
 486         if (!*psf_list)
 487                 goto empty_source;
 488
 489         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 490
 491         /* EX and TO_EX get a fresh packet, if needed */
 492         if (truncate) {
 493                 if (pih && pih->ngrec &&
 494                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 495                         if (skb)
 496                                 igmpv3_sendpack(skb);
 497                         skb = igmpv3_newpack(dev, mtu);
 498                 }
 499         }
 500         first = 1;
 501         psf_prev = NULL;
 502         for (psf = *psf_list; psf; psf = psf_next) {
 503                 __be32 *psrc;
 504
 505                 psf_next = psf->sf_next;
 506
 507                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 508                         psf_prev = psf;
 509                         continue;
 510                 }
 511
 512                 /* Based on RFC3376 5.1. Should not send source-list change
 513                  * records when there is a filter mode change.
 514                  */
 515                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 516                      (!gdeleted && pmc->crcount)) &&
 517                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 518                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 519                         goto decrease_sf_crcount;
 520
 521                 /* clear marks on query responses */
 522                 if (isquery)
 523                         psf->sf_gsresp = 0;
 524
 525                 if (AVAILABLE(skb) < sizeof(__be32) +
 526                     first*sizeof(struct igmpv3_grec)) {
 527                         if (truncate && !first)
 528                                 break;   /* truncate these */
 529                         if (pgr)
 530                                 pgr->grec_nsrcs = htons(scount);
 531                         if (skb)
 532                                 igmpv3_sendpack(skb);
 533                         skb = igmpv3_newpack(dev, mtu);
 534                         first = 1;
 535                         scount = 0;
 536                 }
 537                 if (first) {
 538                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 539                         first = 0;
 540                 }
 541                 if (!skb)
 542                         return NULL;
 543                 psrc = skb_put(skb, sizeof(__be32));
 544                 *psrc = psf->sf_inaddr;
 545                 scount++; stotal++;
 546                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 547                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 548 decrease_sf_crcount:
 549                         psf->sf_crcount--;
 550                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 551                                 if (psf_prev)
 552                                         psf_prev->sf_next = psf->sf_next;
 553                                 else
 554                                         *psf_list = psf->sf_next;
 555                                 kfree(psf);
 556                                 continue;
 557                         }
 558                 }
 559                 psf_prev = psf;
 560         }
 561
 562 empty_source:
 563         if (!stotal) {
 564                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 565                     type == IGMPV3_BLOCK_OLD_SOURCES)
 566                         return skb;
 567                 if (pmc->crcount || isquery) {
 568                         /* make sure we have room for group header */
 569                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 570                                 igmpv3_sendpack(skb);
 571                                 skb = NULL; /* add_grhead will get a new one */
 572                         }
 573                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 574                 }
 575         }
 576         if (pgr)
 577                 pgr->grec_nsrcs = htons(scount);
 578
 579         if (isquery)
 580                 pmc->gsquery = 0;       /* clear query state on report */
 581         return skb;
 582 }
 583
 584 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 585 {
 586         struct sk_buff *skb = NULL;
 587         struct net *net = dev_net(in_dev->dev);
 588         int type;
 589
 590         if (!pmc) {
 591                 rcu_read_lock();
 592                 for_each_pmc_rcu(in_dev, pmc) {
 593                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 594                                 continue;
 595                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 596                              !net->ipv4.sysctl_igmp_llm_reports)
 597                                 continue;
 598                         spin_lock_bh(&pmc->lock);
 599                         if (pmc->sfcount[MCAST_EXCLUDE])
 600                                 type = IGMPV3_MODE_IS_EXCLUDE;
 601                         else
 602                                 type = IGMPV3_MODE_IS_INCLUDE;
 603                         skb = add_grec(skb, pmc, type, 0, 0);
 604                         spin_unlock_bh(&pmc->lock);
 605                 }
 606                 rcu_read_unlock();
 607         } else {
 608                 spin_lock_bh(&pmc->lock);
 609                 if (pmc->sfcount[MCAST_EXCLUDE])
 610                         type = IGMPV3_MODE_IS_EXCLUDE;
 611                 else
 612                         type = IGMPV3_MODE_IS_INCLUDE;
 613                 skb = add_grec(skb, pmc, type, 0, 0);
 614                 spin_unlock_bh(&pmc->lock);
 615         }
 616         if (!skb)
 617                 return 0;
 618         return igmpv3_sendpack(skb);
 619 }
 620
 621 /*
 622  * remove zero-count source records from a source filter list
 623  */
 624 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 625 {
 626         struct ip_sf_list *psf_prev, *psf_next, *psf;
 627
 628         psf_prev = NULL;
 629         for (psf = *ppsf; psf; psf = psf_next) {
 630                 psf_next = psf->sf_next;
 631                 if (psf->sf_crcount == 0) {
 632                         if (psf_prev)
 633                                 psf_prev->sf_next = psf->sf_next;
 634                         else
 635                                 *ppsf = psf->sf_next;
 636                         kfree(psf);
 637                 } else
 638                         psf_prev = psf;
 639         }
 640 }
 641
 642 static void kfree_pmc(struct ip_mc_list *pmc)
 643 {
 644         ip_sf_list_clear_all(pmc->sources);
 645         ip_sf_list_clear_all(pmc->tomb);
 646         kfree(pmc);
 647 }
 648
 649 static void igmpv3_send_cr(struct in_device *in_dev)
 650 {
 651         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 652         struct sk_buff *skb = NULL;
 653         int type, dtype;
 654
 655         rcu_read_lock();
 656         spin_lock_bh(&in_dev->mc_tomb_lock);
 657
 658         /* deleted MCA's */
 659         pmc_prev = NULL;
 660         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 661                 pmc_next = pmc->next;
 662                 if (pmc->sfmode == MCAST_INCLUDE) {
 663                         type = IGMPV3_BLOCK_OLD_SOURCES;
 664                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 665                         skb = add_grec(skb, pmc, type, 1, 0);
 666                         skb = add_grec(skb, pmc, dtype, 1, 1);
 667                 }
 668                 if (pmc->crcount) {
 669                         if (pmc->sfmode == MCAST_EXCLUDE) {
 670                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 671                                 skb = add_grec(skb, pmc, type, 1, 0);
 672                         }
 673                         pmc->crcount--;
 674                         if (pmc->crcount == 0) {
 675                                 igmpv3_clear_zeros(&pmc->tomb);
 676                                 igmpv3_clear_zeros(&pmc->sources);
 677                         }
 678                 }
 679                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 680                         if (pmc_prev)
 681                                 pmc_prev->next = pmc_next;
 682                         else
 683                                 in_dev->mc_tomb = pmc_next;
 684                         in_dev_put(pmc->interface);
 685                         kfree_pmc(pmc);
 686                 } else
 687                         pmc_prev = pmc;
 688         }
 689         spin_unlock_bh(&in_dev->mc_tomb_lock);
 690
 691         /* change recs */
 692         for_each_pmc_rcu(in_dev, pmc) {
 693                 spin_lock_bh(&pmc->lock);
 694                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 695                         type = IGMPV3_BLOCK_OLD_SOURCES;
 696                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 697                 } else {
 698                         type = IGMPV3_ALLOW_NEW_SOURCES;
 699                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 700                 }
 701                 skb = add_grec(skb, pmc, type, 0, 0);
 702                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 703
 704                 /* filter mode changes */
 705                 if (pmc->crcount) {
 706                         if (pmc->sfmode == MCAST_EXCLUDE)
 707                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 708                         else
 709                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 710                         skb = add_grec(skb, pmc, type, 0, 0);
 711                         pmc->crcount--;
 712                 }
 713                 spin_unlock_bh(&pmc->lock);
 714         }
 715         rcu_read_unlock();
 716
 717         if (!skb)
 718                 return;
 719         (void) igmpv3_sendpack(skb);
 720 }
 721
 722 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 723         int type)
 724 {
 725         struct sk_buff *skb;
 726         struct iphdr *iph;
 727         struct igmphdr *ih;
 728         struct rtable *rt;
 729         struct net_device *dev = in_dev->dev;
 730         struct net *net = dev_net(dev);
 731         __be32  group = pmc ? pmc->multiaddr : 0;
 732         struct flowi4 fl4;
 733         __be32  dst;
 734         int hlen, tlen;
 735
 736         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 737                 return igmpv3_send_report(in_dev, pmc);
 738
 739         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 740                 return 0;
 741
 742         if (type == IGMP_HOST_LEAVE_MESSAGE)
 743                 dst = IGMP_ALL_ROUTER;
 744         else
 745                 dst = group;
 746
 747         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 748                                    0, 0,
 749                                    IPPROTO_IGMP, 0, dev->ifindex);
 750         if (IS_ERR(rt))
 751                 return -1;
 752
 753         hlen = LL_RESERVED_SPACE(dev);
 754         tlen = dev->needed_tailroom;
 755         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 756         if (!skb) {
 757                 ip_rt_put(rt);
 758                 return -1;
 759         }
 760         skb->priority = TC_PRIO_CONTROL;
 761
 762         skb_dst_set(skb, &rt->dst);
 763
 764         skb_reserve(skb, hlen);
 765
 766         skb_reset_network_header(skb);
 767         iph = ip_hdr(skb);
 768         skb_put(skb, sizeof(struct iphdr) + 4);
 769
 770         iph->version  = 4;
 771         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 772         iph->tos      = 0xc0;
 773         iph->frag_off = htons(IP_DF);
 774         iph->ttl      = 1;
 775         iph->daddr    = dst;
 776         iph->saddr    = fl4.saddr;
 777         iph->protocol = IPPROTO_IGMP;
 778         ip_select_ident(net, skb, NULL);
 779         ((u8 *)&iph[1])[0] = IPOPT_RA;
 780         ((u8 *)&iph[1])[1] = 4;
 781         ((u8 *)&iph[1])[2] = 0;
 782         ((u8 *)&iph[1])[3] = 0;
 783
 784         ih = skb_put(skb, sizeof(struct igmphdr));
 785         ih->type = type;
 786         ih->code = 0;
 787         ih->csum = 0;
 788         ih->group = group;
 789         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 790
 791         return ip_local_out(net, skb->sk, skb);
 792 }
 793
 794 static void igmp_gq_timer_expire(struct timer_list *t)
 795 {
 796         struct in_device *in_dev = from_timer(in_dev, t, mr_gq_timer);
 797
 798         in_dev->mr_gq_running = 0;
 799         igmpv3_send_report(in_dev, NULL);
 800         in_dev_put(in_dev);
 801 }
 802
 803 static void igmp_ifc_timer_expire(struct timer_list *t)
 804 {
 805         struct in_device *in_dev = from_timer(in_dev, t, mr_ifc_timer);
 806
 807         igmpv3_send_cr(in_dev);
 808         if (in_dev->mr_ifc_count) {
 809                 in_dev->mr_ifc_count--;
 810                 igmp_ifc_start_timer(in_dev,
 811                                      unsolicited_report_interval(in_dev));
 812         }
 813         in_dev_put(in_dev);
 814 }
 815
 816 static void igmp_ifc_event(struct in_device *in_dev)
 817 {
 818         struct net *net = dev_net(in_dev->dev);
 819         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 820                 return;
 821         in_dev->mr_ifc_count = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
 822         igmp_ifc_start_timer(in_dev, 1);
 823 }
 824
 825
 826 static void igmp_timer_expire(struct timer_list *t)
 827 {
 828         struct ip_mc_list *im = from_timer(im, t, timer);
 829         struct in_device *in_dev = im->interface;
 830
 831         spin_lock(&im->lock);
 832         im->tm_running = 0;
 833
 834         if (im->unsolicit_count && --im->unsolicit_count)
 835                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 836
 837         im->reporter = 1;
 838         spin_unlock(&im->lock);
 839
 840         if (IGMP_V1_SEEN(in_dev))
 841                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 842         else if (IGMP_V2_SEEN(in_dev))
 843                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 844         else
 845                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 846
 847         ip_ma_put(im);
 848 }
 849
 850 /* mark EXCLUDE-mode sources */
 851 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 852 {
 853         struct ip_sf_list *psf;
 854         int i, scount;
 855
 856         scount = 0;
 857         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 858                 if (scount == nsrcs)
 859                         break;
 860                 for (i = 0; i < nsrcs; i++) {
 861                         /* skip inactive filters */
 862                         if (psf->sf_count[MCAST_INCLUDE] ||
 863                             pmc->sfcount[MCAST_EXCLUDE] !=
 864                             psf->sf_count[MCAST_EXCLUDE])
 865                                 break;
 866                         if (srcs[i] == psf->sf_inaddr) {
 867                                 scount++;
 868                                 break;
 869                         }
 870                 }
 871         }
 872         pmc->gsquery = 0;
 873         if (scount == nsrcs)    /* all sources excluded */
 874                 return 0;
 875         return 1;
 876 }
 877
 878 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 879 {
 880         struct ip_sf_list *psf;
 881         int i, scount;
 882
 883         if (pmc->sfmode == MCAST_EXCLUDE)
 884                 return igmp_xmarksources(pmc, nsrcs, srcs);
 885
 886         /* mark INCLUDE-mode sources */
 887         scount = 0;
 888         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 889                 if (scount == nsrcs)
 890                         break;
 891                 for (i = 0; i < nsrcs; i++)
 892                         if (srcs[i] == psf->sf_inaddr) {
 893                                 psf->sf_gsresp = 1;
 894                                 scount++;
 895                                 break;
 896                         }
 897         }
 898         if (!scount) {
 899                 pmc->gsquery = 0;
 900                 return 0;
 901         }
 902         pmc->gsquery = 1;
 903         return 1;
 904 }
 905
 906 /* return true if packet was dropped */
 907 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 908 {
 909         struct ip_mc_list *im;
 910         struct net *net = dev_net(in_dev->dev);
 911
 912         /* Timers are only set for non-local groups */
 913
 914         if (group == IGMP_ALL_HOSTS)
 915                 return false;
 916         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 917                 return false;
 918
 919         rcu_read_lock();
 920         for_each_pmc_rcu(in_dev, im) {
 921                 if (im->multiaddr == group) {
 922                         igmp_stop_timer(im);
 923                         break;
 924                 }
 925         }
 926         rcu_read_unlock();
 927         return false;
 928 }
 929
 930 /* return true if packet was dropped */
 931 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 932         int len)
 933 {
 934         struct igmphdr          *ih = igmp_hdr(skb);
 935         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 936         struct ip_mc_list       *im;
 937         __be32                  group = ih->group;
 938         int                     max_delay;
 939         int                     mark = 0;
 940         struct net              *net = dev_net(in_dev->dev);
 941
 942
 943         if (len == 8) {
 944                 if (ih->code == 0) {
 945                         /* Alas, old v1 router presents here. */
 946
 947                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 948                         in_dev->mr_v1_seen = jiffies +
 949                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 950                                 in_dev->mr_qri;
 951                         group = 0;
 952                 } else {
 953                         /* v2 router present */
 954                         max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
 955                         in_dev->mr_v2_seen = jiffies +
 956                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 957                                 in_dev->mr_qri;
 958                 }
 959                 /* cancel the interface change timer */
 960                 in_dev->mr_ifc_count = 0;
 961                 if (del_timer(&in_dev->mr_ifc_timer))
 962                         __in_dev_put(in_dev);
 963                 /* clear deleted report items */
 964                 igmpv3_clear_delrec(in_dev);
 965         } else if (len < 12) {
 966                 return true;    /* ignore bogus packet; freed by caller */
 967         } else if (IGMP_V1_SEEN(in_dev)) {
 968                 /* This is a v3 query with v1 queriers present */
 969                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 970                 group = 0;
 971         } else if (IGMP_V2_SEEN(in_dev)) {
 972                 /* this is a v3 query with v2 queriers present;
 973                  * Interpretation of the max_delay code is problematic here.
 974                  * A real v2 host would use ih_code directly, while v3 has a
 975                  * different encoding. We use the v3 encoding as more likely
 976                  * to be intended in a v3 query.
 977                  */
 978                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 979                 if (!max_delay)
 980                         max_delay = 1;  /* can't mod w/ 0 */
 981         } else { /* v3 */
 982                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 983                         return true;
 984
 985                 ih3 = igmpv3_query_hdr(skb);
 986                 if (ih3->nsrcs) {
 987                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
 988                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
 989                                 return true;
 990                         ih3 = igmpv3_query_hdr(skb);
 991                 }
 992
 993                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 994                 if (!max_delay)
 995                         max_delay = 1;  /* can't mod w/ 0 */
 996                 in_dev->mr_maxdelay = max_delay;
 997
 998                 /* RFC3376, 4.1.6. QRV and 4.1.7. QQIC, when the most recently
 999                  * received value was zero, use the default or statically
1000                  * configured value.
1001                  */
1002                 in_dev->mr_qrv = ih3->qrv ?: net->ipv4.sysctl_igmp_qrv;
1003                 in_dev->mr_qi = IGMPV3_QQIC(ih3->qqic)*HZ ?: IGMP_QUERY_INTERVAL;
1004
1005                 /* RFC3376, 8.3. Query Response Interval:
1006                  * The number of seconds represented by the [Query Response
1007                  * Interval] must be less than the [Query Interval].
1008                  */
1009                 if (in_dev->mr_qri >= in_dev->mr_qi)
1010                         in_dev->mr_qri = (in_dev->mr_qi/HZ - 1)*HZ;
1011
1012                 if (!group) { /* general query */
1013                         if (ih3->nsrcs)
1014                                 return true;    /* no sources allowed */
1015                         igmp_gq_start_timer(in_dev);
1016                         return false;
1017                 }
1018                 /* mark sources to include, if group & source-specific */
1019                 mark = ih3->nsrcs != 0;
1020         }
1021
1022         /*
1023          * - Start the timers in all of our membership records
1024          *   that the query applies to for the interface on
1025          *   which the query arrived excl. those that belong
1026          *   to a "local" group (224.0.0.X)
1027          * - For timers already running check if they need to
1028          *   be reset.
1029          * - Use the igmp->igmp_code field as the maximum
1030          *   delay possible
1031          */
1032         rcu_read_lock();
1033         for_each_pmc_rcu(in_dev, im) {
1034                 int changed;
1035
1036                 if (group && group != im->multiaddr)
1037                         continue;
1038                 if (im->multiaddr == IGMP_ALL_HOSTS)
1039                         continue;
1040                 if (ipv4_is_local_multicast(im->multiaddr) &&
1041                     !net->ipv4.sysctl_igmp_llm_reports)
1042                         continue;
1043                 spin_lock_bh(&im->lock);
1044                 if (im->tm_running)
1045                         im->gsquery = im->gsquery && mark;
1046                 else
1047                         im->gsquery = mark;
1048                 changed = !im->gsquery ||
1049                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1050                 spin_unlock_bh(&im->lock);
1051                 if (changed)
1052                         igmp_mod_timer(im, max_delay);
1053         }
1054         rcu_read_unlock();
1055         return false;
1056 }
1057
1058 /* called in rcu_read_lock() section */
1059 int igmp_rcv(struct sk_buff *skb)
1060 {
1061         /* This basically follows the spec line by line -- see RFC1112 */
1062         struct igmphdr *ih;
1063         struct net_device *dev = skb->dev;
1064         struct in_device *in_dev;
1065         int len = skb->len;
1066         bool dropped = true;
1067
1068         if (netif_is_l3_master(dev)) {
1069                 dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1070                 if (!dev)
1071                         goto drop;
1072         }
1073
1074         in_dev = __in_dev_get_rcu(dev);
1075         if (!in_dev)
1076                 goto drop;
1077
1078         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1079                 goto drop;
1080
1081         if (skb_checksum_simple_validate(skb))
1082                 goto drop;
1083
1084         ih = igmp_hdr(skb);
1085         switch (ih->type) {
1086         case IGMP_HOST_MEMBERSHIP_QUERY:
1087                 dropped = igmp_heard_query(in_dev, skb, len);
1088                 break;
1089         case IGMP_HOST_MEMBERSHIP_REPORT:
1090         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1091                 /* Is it our report looped back? */
1092                 if (rt_is_output_route(skb_rtable(skb)))
1093                         break;
1094                 /* don't rely on MC router hearing unicast reports */
1095                 if (skb->pkt_type == PACKET_MULTICAST ||
1096                     skb->pkt_type == PACKET_BROADCAST)
1097                         dropped = igmp_heard_report(in_dev, ih->group);
1098                 break;
1099         case IGMP_PIM:
1100 #ifdef CONFIG_IP_PIMSM_V1
1101                 return pim_rcv_v1(skb);
1102 #endif
1103         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1104         case IGMP_DVMRP:
1105         case IGMP_TRACE:
1106         case IGMP_HOST_LEAVE_MESSAGE:
1107         case IGMP_MTRACE:
1108         case IGMP_MTRACE_RESP:
1109                 break;
1110         default:
1111                 break;
1112         }
1113
1114 drop:
1115         if (dropped)
1116                 kfree_skb(skb);
1117         else
1118                 consume_skb(skb);
1119         return 0;
1120 }
1121
1122 #endif
1123
1124
1125 /*
1126  *      Add a filter to a device
1127  */
1128
1129 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1130 {
1131         char buf[MAX_ADDR_LEN];
1132         struct net_device *dev = in_dev->dev;
1133
1134         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1135            We will get multicast token leakage, when IFF_MULTICAST
1136            is changed. This check should be done in ndo_set_rx_mode
1137            routine. Something sort of:
1138            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1139            --ANK
1140            */
1141         if (arp_mc_map(addr, buf, dev, 0) == 0)
1142                 dev_mc_add(dev, buf);
1143 }
1144
1145 /*
1146  *      Remove a filter from a device
1147  */
1148
1149 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1150 {
1151         char buf[MAX_ADDR_LEN];
1152         struct net_device *dev = in_dev->dev;
1153
1154         if (arp_mc_map(addr, buf, dev, 0) == 0)
1155                 dev_mc_del(dev, buf);
1156 }
1157
1158 #ifdef CONFIG_IP_MULTICAST
1159 /*
1160  * deleted ip_mc_list manipulation
1161  */
1162 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
1163                               gfp_t gfp)
1164 {
1165         struct ip_mc_list *pmc;
1166         struct net *net = dev_net(in_dev->dev);
1167
1168         /* this is an "ip_mc_list" for convenience; only the fields below
1169          * are actually used. In particular, the refcnt and users are not
1170          * used for management of the delete list. Using the same structure
1171          * for deleted items allows change reports to use common code with
1172          * non-deleted or query-response MCA's.
1173          */
1174         pmc = kzalloc(sizeof(*pmc), gfp);
1175         if (!pmc)
1176                 return;
1177         spin_lock_init(&pmc->lock);
1178         spin_lock_bh(&im->lock);
1179         pmc->interface = im->interface;
1180         in_dev_hold(in_dev);
1181         pmc->multiaddr = im->multiaddr;
1182         pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1183         pmc->sfmode = im->sfmode;
1184         if (pmc->sfmode == MCAST_INCLUDE) {
1185                 struct ip_sf_list *psf;
1186
1187                 pmc->tomb = im->tomb;
1188                 pmc->sources = im->sources;
1189                 im->tomb = im->sources = NULL;
1190                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1191                         psf->sf_crcount = pmc->crcount;
1192         }
1193         spin_unlock_bh(&im->lock);
1194
1195         spin_lock_bh(&in_dev->mc_tomb_lock);
1196         pmc->next = in_dev->mc_tomb;
1197         in_dev->mc_tomb = pmc;
1198         spin_unlock_bh(&in_dev->mc_tomb_lock);
1199 }
1200
1201 /*
1202  * restore ip_mc_list deleted records
1203  */
1204 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1205 {
1206         struct ip_mc_list *pmc, *pmc_prev;
1207         struct ip_sf_list *psf;
1208         struct net *net = dev_net(in_dev->dev);
1209         __be32 multiaddr = im->multiaddr;
1210
1211         spin_lock_bh(&in_dev->mc_tomb_lock);
1212         pmc_prev = NULL;
1213         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1214                 if (pmc->multiaddr == multiaddr)
1215                         break;
1216                 pmc_prev = pmc;
1217         }
1218         if (pmc) {
1219                 if (pmc_prev)
1220                         pmc_prev->next = pmc->next;
1221                 else
1222                         in_dev->mc_tomb = pmc->next;
1223         }
1224         spin_unlock_bh(&in_dev->mc_tomb_lock);
1225
1226         spin_lock_bh(&im->lock);
1227         if (pmc) {
1228                 im->interface = pmc->interface;
1229                 if (im->sfmode == MCAST_INCLUDE) {
1230                         swap(im->tomb, pmc->tomb);
1231                         swap(im->sources, pmc->sources);
1232                         for (psf = im->sources; psf; psf = psf->sf_next)
1233                                 psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1234                 } else {
1235                         im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1236                 }
1237                 in_dev_put(pmc->interface);
1238                 kfree_pmc(pmc);
1239         }
1240         spin_unlock_bh(&im->lock);
1241 }
1242
1243 /*
1244  * flush ip_mc_list deleted records
1245  */
1246 static void igmpv3_clear_delrec(struct in_device *in_dev)
1247 {
1248         struct ip_mc_list *pmc, *nextpmc;
1249
1250         spin_lock_bh(&in_dev->mc_tomb_lock);
1251         pmc = in_dev->mc_tomb;
1252         in_dev->mc_tomb = NULL;
1253         spin_unlock_bh(&in_dev->mc_tomb_lock);
1254
1255         for (; pmc; pmc = nextpmc) {
1256                 nextpmc = pmc->next;
1257                 ip_mc_clear_src(pmc);
1258                 in_dev_put(pmc->interface);
1259                 kfree_pmc(pmc);
1260         }
1261         /* clear dead sources, too */
1262         rcu_read_lock();
1263         for_each_pmc_rcu(in_dev, pmc) {
1264                 struct ip_sf_list *psf;
1265
1266                 spin_lock_bh(&pmc->lock);
1267                 psf = pmc->tomb;
1268                 pmc->tomb = NULL;
1269                 spin_unlock_bh(&pmc->lock);
1270                 ip_sf_list_clear_all(psf);
1271         }
1272         rcu_read_unlock();
1273 }
1274 #endif
1275
1276 static void __igmp_group_dropped(struct ip_mc_list *im, gfp_t gfp)
1277 {
1278         struct in_device *in_dev = im->interface;
1279 #ifdef CONFIG_IP_MULTICAST
1280         struct net *net = dev_net(in_dev->dev);
1281         int reporter;
1282 #endif
1283
1284         if (im->loaded) {
1285                 im->loaded = 0;
1286                 ip_mc_filter_del(in_dev, im->multiaddr);
1287         }
1288
1289 #ifdef CONFIG_IP_MULTICAST
1290         if (im->multiaddr == IGMP_ALL_HOSTS)
1291                 return;
1292         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1293                 return;
1294
1295         reporter = im->reporter;
1296         igmp_stop_timer(im);
1297
1298         if (!in_dev->dead) {
1299                 if (IGMP_V1_SEEN(in_dev))
1300                         return;
1301                 if (IGMP_V2_SEEN(in_dev)) {
1302                         if (reporter)
1303                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1304                         return;
1305                 }
1306                 /* IGMPv3 */
1307                 igmpv3_add_delrec(in_dev, im, gfp);
1308
1309                 igmp_ifc_event(in_dev);
1310         }
1311 #endif
1312 }
1313
1314 static void igmp_group_dropped(struct ip_mc_list *im)
1315 {
1316         __igmp_group_dropped(im, GFP_KERNEL);
1317 }
1318
1319 static void igmp_group_added(struct ip_mc_list *im)
1320 {
1321         struct in_device *in_dev = im->interface;
1322 #ifdef CONFIG_IP_MULTICAST
1323         struct net *net = dev_net(in_dev->dev);
1324 #endif
1325
1326         if (im->loaded == 0) {
1327                 im->loaded = 1;
1328                 ip_mc_filter_add(in_dev, im->multiaddr);
1329         }
1330
1331 #ifdef CONFIG_IP_MULTICAST
1332         if (im->multiaddr == IGMP_ALL_HOSTS)
1333                 return;
1334         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1335                 return;
1336
1337         if (in_dev->dead)
1338                 return;
1339
1340         im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1341         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1342                 spin_lock_bh(&im->lock);
1343                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1344                 spin_unlock_bh(&im->lock);
1345                 return;
1346         }
1347         /* else, v3 */
1348
1349         /* Based on RFC3376 5.1, for newly added INCLUDE SSM, we should
1350          * not send filter-mode change record as the mode should be from
1351          * IN() to IN(A).
1352          */
1353         if (im->sfmode == MCAST_EXCLUDE)
1354                 im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1355
1356         igmp_ifc_event(in_dev);
1357 #endif
1358 }
1359
1360
1361 /*
1362  *      Multicast list managers
1363  */
1364
1365 static u32 ip_mc_hash(const struct ip_mc_list *im)
1366 {
1367         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1368 }
1369
1370 static void ip_mc_hash_add(struct in_device *in_dev,
1371                            struct ip_mc_list *im)
1372 {
1373         struct ip_mc_list __rcu **mc_hash;
1374         u32 hash;
1375
1376         mc_hash = rtnl_dereference(in_dev->mc_hash);
1377         if (mc_hash) {
1378                 hash = ip_mc_hash(im);
1379                 im->next_hash = mc_hash[hash];
1380                 rcu_assign_pointer(mc_hash[hash], im);
1381                 return;
1382         }
1383
1384         /* do not use a hash table for small number of items */
1385         if (in_dev->mc_count < 4)
1386                 return;
1387
1388         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1389                           GFP_KERNEL);
1390         if (!mc_hash)
1391                 return;
1392
1393         for_each_pmc_rtnl(in_dev, im) {
1394                 hash = ip_mc_hash(im);
1395                 im->next_hash = mc_hash[hash];
1396                 RCU_INIT_POINTER(mc_hash[hash], im);
1397         }
1398
1399         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1400 }
1401
1402 static void ip_mc_hash_remove(struct in_device *in_dev,
1403                               struct ip_mc_list *im)
1404 {
1405         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1406         struct ip_mc_list *aux;
1407
1408         if (!mc_hash)
1409                 return;
1410         mc_hash += ip_mc_hash(im);
1411         while ((aux = rtnl_dereference(*mc_hash)) != im)
1412                 mc_hash = &aux->next_hash;
1413         *mc_hash = im->next_hash;
1414 }
1415
1416
1417 /*
1418  *      A socket has joined a multicast group on device dev.
1419  */
1420 static void ____ip_mc_inc_group(struct in_device *in_dev, __be32 addr,
1421                                 unsigned int mode, gfp_t gfp)
1422 {
1423         struct ip_mc_list *im;
1424
1425         ASSERT_RTNL();
1426
1427         for_each_pmc_rtnl(in_dev, im) {
1428                 if (im->multiaddr == addr) {
1429                         im->users++;
1430                         ip_mc_add_src(in_dev, &addr, mode, 0, NULL, 0);
1431                         goto out;
1432                 }
1433         }
1434
1435         im = kzalloc(sizeof(*im), gfp);
1436         if (!im)
1437                 goto out;
1438
1439         im->users = 1;
1440         im->interface = in_dev;
1441         in_dev_hold(in_dev);
1442         im->multiaddr = addr;
1443         /* initial mode is (EX, empty) */
1444         im->sfmode = mode;
1445         im->sfcount[mode] = 1;
1446         refcount_set(&im->refcnt, 1);
1447         spin_lock_init(&im->lock);
1448 #ifdef CONFIG_IP_MULTICAST
1449         timer_setup(&im->timer, igmp_timer_expire, 0);
1450 #endif
1451
1452         im->next_rcu = in_dev->mc_list;
1453         in_dev->mc_count++;
1454         rcu_assign_pointer(in_dev->mc_list, im);
1455
1456         ip_mc_hash_add(in_dev, im);
1457
1458 #ifdef CONFIG_IP_MULTICAST
1459         igmpv3_del_delrec(in_dev, im);
1460 #endif
1461         igmp_group_added(im);
1462         if (!in_dev->dead)
1463                 ip_rt_multicast_event(in_dev);
1464 out:
1465         return;
1466 }
1467
1468 void __ip_mc_inc_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1469 {
1470         ____ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE, gfp);
1471 }
1472 EXPORT_SYMBOL(__ip_mc_inc_group);
1473
1474 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1475 {
1476         __ip_mc_inc_group(in_dev, addr, GFP_KERNEL);
1477 }
1478 EXPORT_SYMBOL(ip_mc_inc_group);
1479
1480 static int ip_mc_check_iphdr(struct sk_buff *skb)
1481 {
1482         const struct iphdr *iph;
1483         unsigned int len;
1484         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1485
1486         if (!pskb_may_pull(skb, offset))
1487                 return -EINVAL;
1488
1489         iph = ip_hdr(skb);
1490
1491         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1492                 return -EINVAL;
1493
1494         offset += ip_hdrlen(skb) - sizeof(*iph);
1495
1496         if (!pskb_may_pull(skb, offset))
1497                 return -EINVAL;
1498
1499         iph = ip_hdr(skb);
1500
1501         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1502                 return -EINVAL;
1503
1504         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1505         if (skb->len < len || len < offset)
1506                 return -EINVAL;
1507
1508         skb_set_transport_header(skb, offset);
1509
1510         return 0;
1511 }
1512
1513 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1514 {
1515         unsigned int len = skb_transport_offset(skb);
1516
1517         len += sizeof(struct igmpv3_report);
1518
1519         return ip_mc_may_pull(skb, len) ? 0 : -EINVAL;
1520 }
1521
1522 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1523 {
1524         unsigned int transport_len = ip_transport_len(skb);
1525         unsigned int len;
1526
1527         /* IGMPv{1,2}? */
1528         if (transport_len != sizeof(struct igmphdr)) {
1529                 /* or IGMPv3? */
1530                 if (transport_len < sizeof(struct igmpv3_query))
1531                         return -EINVAL;
1532
1533                 len = skb_transport_offset(skb) + sizeof(struct igmpv3_query);
1534                 if (!ip_mc_may_pull(skb, len))
1535                         return -EINVAL;
1536         }
1537
1538         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1539          * all-systems destination addresses (224.0.0.1) for general queries
1540          */
1541         if (!igmp_hdr(skb)->group &&
1542             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1543                 return -EINVAL;
1544
1545         return 0;
1546 }
1547
1548 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1549 {
1550         switch (igmp_hdr(skb)->type) {
1551         case IGMP_HOST_LEAVE_MESSAGE:
1552         case IGMP_HOST_MEMBERSHIP_REPORT:
1553         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1554                 return 0;
1555         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1556                 return ip_mc_check_igmp_reportv3(skb);
1557         case IGMP_HOST_MEMBERSHIP_QUERY:
1558                 return ip_mc_check_igmp_query(skb);
1559         default:
1560                 return -ENOMSG;
1561         }
1562 }
1563
1564 static __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1565 {
1566         return skb_checksum_simple_validate(skb);
1567 }
1568
1569 static int ip_mc_check_igmp_csum(struct sk_buff *skb)
1570 {
1571         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1572         unsigned int transport_len = ip_transport_len(skb);
1573         struct sk_buff *skb_chk;
1574
1575         if (!ip_mc_may_pull(skb, len))
1576                 return -EINVAL;
1577
1578         skb_chk = skb_checksum_trimmed(skb, transport_len,
1579                                        ip_mc_validate_checksum);
1580         if (!skb_chk)
1581                 return -EINVAL;
1582
1583         if (skb_chk != skb)
1584                 kfree_skb(skb_chk);
1585
1586         return 0;
1587 }
1588
1589 /**
1590  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1591  * @skb: the skb to validate
1592  *
1593  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1594  * skb transport header accordingly and returns zero.
1595  *
1596  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1597  *  standard
1598  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1599  * -ENOMEM: A memory allocation failure happened.
1600  *
1601  * Caller needs to set the skb network header and free any returned skb if it
1602  * differs from the provided skb.
1603  */
1604 int ip_mc_check_igmp(struct sk_buff *skb)
1605 {
1606         int ret = ip_mc_check_iphdr(skb);
1607
1608         if (ret < 0)
1609                 return ret;
1610
1611         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1612                 return -ENOMSG;
1613
1614         ret = ip_mc_check_igmp_csum(skb);
1615         if (ret < 0)
1616                 return ret;
1617
1618         return ip_mc_check_igmp_msg(skb);
1619 }
1620 EXPORT_SYMBOL(ip_mc_check_igmp);
1621
1622 /*
1623  *      Resend IGMP JOIN report; used by netdev notifier.
1624  */
1625 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1626 {
1627 #ifdef CONFIG_IP_MULTICAST
1628         struct ip_mc_list *im;
1629         int type;
1630         struct net *net = dev_net(in_dev->dev);
1631
1632         ASSERT_RTNL();
1633
1634         for_each_pmc_rtnl(in_dev, im) {
1635                 if (im->multiaddr == IGMP_ALL_HOSTS)
1636                         continue;
1637                 if (ipv4_is_local_multicast(im->multiaddr) &&
1638                     !net->ipv4.sysctl_igmp_llm_reports)
1639                         continue;
1640
1641                 /* a failover is happening and switches
1642                  * must be notified immediately
1643                  */
1644                 if (IGMP_V1_SEEN(in_dev))
1645                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1646                 else if (IGMP_V2_SEEN(in_dev))
1647                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1648                 else
1649                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1650                 igmp_send_report(in_dev, im, type);
1651         }
1652 #endif
1653 }
1654
1655 /*
1656  *      A socket has left a multicast group on device dev
1657  */
1658
1659 void __ip_mc_dec_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1660 {
1661         struct ip_mc_list *i;
1662         struct ip_mc_list __rcu **ip;
1663
1664         ASSERT_RTNL();
1665
1666         for (ip = &in_dev->mc_list;
1667              (i = rtnl_dereference(*ip)) != NULL;
1668              ip = &i->next_rcu) {
1669                 if (i->multiaddr == addr) {
1670                         if (--i->users == 0) {
1671                                 ip_mc_hash_remove(in_dev, i);
1672                                 *ip = i->next_rcu;
1673                                 in_dev->mc_count--;
1674                                 __igmp_group_dropped(i, gfp);
1675                                 ip_mc_clear_src(i);
1676
1677                                 if (!in_dev->dead)
1678                                         ip_rt_multicast_event(in_dev);
1679
1680                                 ip_ma_put(i);
1681                                 return;
1682                         }
1683                         break;
1684                 }
1685         }
1686 }
1687 EXPORT_SYMBOL(__ip_mc_dec_group);
1688
1689 /* Device changing type */
1690
1691 void ip_mc_unmap(struct in_device *in_dev)
1692 {
1693         struct ip_mc_list *pmc;
1694
1695         ASSERT_RTNL();
1696
1697         for_each_pmc_rtnl(in_dev, pmc)
1698                 igmp_group_dropped(pmc);
1699 }
1700
1701 void ip_mc_remap(struct in_device *in_dev)
1702 {
1703         struct ip_mc_list *pmc;
1704
1705         ASSERT_RTNL();
1706
1707         for_each_pmc_rtnl(in_dev, pmc) {
1708 #ifdef CONFIG_IP_MULTICAST
1709                 igmpv3_del_delrec(in_dev, pmc);
1710 #endif
1711                 igmp_group_added(pmc);
1712         }
1713 }
1714
1715 /* Device going down */
1716
1717 void ip_mc_down(struct in_device *in_dev)
1718 {
1719         struct ip_mc_list *pmc;
1720
1721         ASSERT_RTNL();
1722
1723         for_each_pmc_rtnl(in_dev, pmc)
1724                 igmp_group_dropped(pmc);
1725
1726 #ifdef CONFIG_IP_MULTICAST
1727         in_dev->mr_ifc_count = 0;
1728         if (del_timer(&in_dev->mr_ifc_timer))
1729                 __in_dev_put(in_dev);
1730         in_dev->mr_gq_running = 0;
1731         if (del_timer(&in_dev->mr_gq_timer))
1732                 __in_dev_put(in_dev);
1733 #endif
1734
1735         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1736 }
1737
1738 #ifdef CONFIG_IP_MULTICAST
1739 static void ip_mc_reset(struct in_device *in_dev)
1740 {
1741         struct net *net = dev_net(in_dev->dev);
1742
1743         in_dev->mr_qi = IGMP_QUERY_INTERVAL;
1744         in_dev->mr_qri = IGMP_QUERY_RESPONSE_INTERVAL;
1745         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1746 }
1747 #else
1748 static void ip_mc_reset(struct in_device *in_dev)
1749 {
1750 }
1751 #endif
1752
1753 void ip_mc_init_dev(struct in_device *in_dev)
1754 {
1755         ASSERT_RTNL();
1756
1757 #ifdef CONFIG_IP_MULTICAST
1758         timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1759         timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1760 #endif
1761         ip_mc_reset(in_dev);
1762
1763         spin_lock_init(&in_dev->mc_tomb_lock);
1764 }
1765
1766 /* Device going up */
1767
1768 void ip_mc_up(struct in_device *in_dev)
1769 {
1770         struct ip_mc_list *pmc;
1771
1772         ASSERT_RTNL();
1773
1774         ip_mc_reset(in_dev);
1775         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1776
1777         for_each_pmc_rtnl(in_dev, pmc) {
1778 #ifdef CONFIG_IP_MULTICAST
1779                 igmpv3_del_delrec(in_dev, pmc);
1780 #endif
1781                 igmp_group_added(pmc);
1782         }
1783 }
1784
1785 /*
1786  *      Device is about to be destroyed: clean up.
1787  */
1788
1789 void ip_mc_destroy_dev(struct in_device *in_dev)
1790 {
1791         struct ip_mc_list *i;
1792
1793         ASSERT_RTNL();
1794
1795         /* Deactivate timers */
1796         ip_mc_down(in_dev);
1797 #ifdef CONFIG_IP_MULTICAST
1798         igmpv3_clear_delrec(in_dev);
1799 #endif
1800
1801         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1802                 in_dev->mc_list = i->next_rcu;
1803                 in_dev->mc_count--;
1804                 ip_mc_clear_src(i);
1805                 ip_ma_put(i);
1806         }
1807 }
1808
1809 /* RTNL is locked */
1810 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1811 {
1812         struct net_device *dev = NULL;
1813         struct in_device *idev = NULL;
1814
1815         if (imr->imr_ifindex) {
1816                 idev = inetdev_by_index(net, imr->imr_ifindex);
1817                 return idev;
1818         }
1819         if (imr->imr_address.s_addr) {
1820                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1821                 if (!dev)
1822                         return NULL;
1823         }
1824
1825         if (!dev) {
1826                 struct rtable *rt = ip_route_output(net,
1827                                                     imr->imr_multiaddr.s_addr,
1828                                                     0, 0, 0);
1829                 if (!IS_ERR(rt)) {
1830                         dev = rt->dst.dev;
1831                         ip_rt_put(rt);
1832                 }
1833         }
1834         if (dev) {
1835                 imr->imr_ifindex = dev->ifindex;
1836                 idev = __in_dev_get_rtnl(dev);
1837         }
1838         return idev;
1839 }
1840
1841 /*
1842  *      Join a socket to a group
1843  */
1844
1845 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1846         __be32 *psfsrc)
1847 {
1848         struct ip_sf_list *psf, *psf_prev;
1849         int rv = 0;
1850
1851         psf_prev = NULL;
1852         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1853                 if (psf->sf_inaddr == *psfsrc)
1854                         break;
1855                 psf_prev = psf;
1856         }
1857         if (!psf || psf->sf_count[sfmode] == 0) {
1858                 /* source filter not found, or count wrong =>  bug */
1859                 return -ESRCH;
1860         }
1861         psf->sf_count[sfmode]--;
1862         if (psf->sf_count[sfmode] == 0) {
1863                 ip_rt_multicast_event(pmc->interface);
1864         }
1865         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1866 #ifdef CONFIG_IP_MULTICAST
1867                 struct in_device *in_dev = pmc->interface;
1868                 struct net *net = dev_net(in_dev->dev);
1869 #endif
1870
1871                 /* no more filters for this source */
1872                 if (psf_prev)
1873                         psf_prev->sf_next = psf->sf_next;
1874                 else
1875                         pmc->sources = psf->sf_next;
1876 #ifdef CONFIG_IP_MULTICAST
1877                 if (psf->sf_oldin &&
1878                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1879                         psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1880                         psf->sf_next = pmc->tomb;
1881                         pmc->tomb = psf;
1882                         rv = 1;
1883                 } else
1884 #endif
1885                         kfree(psf);
1886         }
1887         return rv;
1888 }
1889
1890 #ifndef CONFIG_IP_MULTICAST
1891 #define igmp_ifc_event(x)       do { } while (0)
1892 #endif
1893
1894 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1895                          int sfcount, __be32 *psfsrc, int delta)
1896 {
1897         struct ip_mc_list *pmc;
1898         int     changerec = 0;
1899         int     i, err;
1900
1901         if (!in_dev)
1902                 return -ENODEV;
1903         rcu_read_lock();
1904         for_each_pmc_rcu(in_dev, pmc) {
1905                 if (*pmca == pmc->multiaddr)
1906                         break;
1907         }
1908         if (!pmc) {
1909                 /* MCA not found?? bug */
1910                 rcu_read_unlock();
1911                 return -ESRCH;
1912         }
1913         spin_lock_bh(&pmc->lock);
1914         rcu_read_unlock();
1915 #ifdef CONFIG_IP_MULTICAST
1916         sf_markstate(pmc);
1917 #endif
1918         if (!delta) {
1919                 err = -EINVAL;
1920                 if (!pmc->sfcount[sfmode])
1921                         goto out_unlock;
1922                 pmc->sfcount[sfmode]--;
1923         }
1924         err = 0;
1925         for (i = 0; i < sfcount; i++) {
1926                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1927
1928                 changerec |= rv > 0;
1929                 if (!err && rv < 0)
1930                         err = rv;
1931         }
1932         if (pmc->sfmode == MCAST_EXCLUDE &&
1933             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1934             pmc->sfcount[MCAST_INCLUDE]) {
1935 #ifdef CONFIG_IP_MULTICAST
1936                 struct ip_sf_list *psf;
1937                 struct net *net = dev_net(in_dev->dev);
1938 #endif
1939
1940                 /* filter mode change */
1941                 pmc->sfmode = MCAST_INCLUDE;
1942 #ifdef CONFIG_IP_MULTICAST
1943                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1944                 in_dev->mr_ifc_count = pmc->crcount;
1945                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1946                         psf->sf_crcount = 0;
1947                 igmp_ifc_event(pmc->interface);
1948         } else if (sf_setstate(pmc) || changerec) {
1949                 igmp_ifc_event(pmc->interface);
1950 #endif
1951         }
1952 out_unlock:
1953         spin_unlock_bh(&pmc->lock);
1954         return err;
1955 }
1956
1957 /*
1958  * Add multicast single-source filter to the interface list
1959  */
1960 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1961         __be32 *psfsrc)
1962 {
1963         struct ip_sf_list *psf, *psf_prev;
1964
1965         psf_prev = NULL;
1966         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1967                 if (psf->sf_inaddr == *psfsrc)
1968                         break;
1969                 psf_prev = psf;
1970         }
1971         if (!psf) {
1972                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1973                 if (!psf)
1974                         return -ENOBUFS;
1975                 psf->sf_inaddr = *psfsrc;
1976                 if (psf_prev) {
1977                         psf_prev->sf_next = psf;
1978                 } else
1979                         pmc->sources = psf;
1980         }
1981         psf->sf_count[sfmode]++;
1982         if (psf->sf_count[sfmode] == 1) {
1983                 ip_rt_multicast_event(pmc->interface);
1984         }
1985         return 0;
1986 }
1987
1988 #ifdef CONFIG_IP_MULTICAST
1989 static void sf_markstate(struct ip_mc_list *pmc)
1990 {
1991         struct ip_sf_list *psf;
1992         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1993
1994         for (psf = pmc->sources; psf; psf = psf->sf_next)
1995                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1996                         psf->sf_oldin = mca_xcount ==
1997                                 psf->sf_count[MCAST_EXCLUDE] &&
1998                                 !psf->sf_count[MCAST_INCLUDE];
1999                 } else
2000                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
2001 }
2002
2003 static int sf_setstate(struct ip_mc_list *pmc)
2004 {
2005         struct ip_sf_list *psf, *dpsf;
2006         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2007         int qrv = pmc->interface->mr_qrv;
2008         int new_in, rv;
2009
2010         rv = 0;
2011         for (psf = pmc->sources; psf; psf = psf->sf_next) {
2012                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2013                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
2014                                 !psf->sf_count[MCAST_INCLUDE];
2015                 } else
2016                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
2017                 if (new_in) {
2018                         if (!psf->sf_oldin) {
2019                                 struct ip_sf_list *prev = NULL;
2020
2021                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2022                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2023                                                 break;
2024                                         prev = dpsf;
2025                                 }
2026                                 if (dpsf) {
2027                                         if (prev)
2028                                                 prev->sf_next = dpsf->sf_next;
2029                                         else
2030                                                 pmc->tomb = dpsf->sf_next;
2031                                         kfree(dpsf);
2032                                 }
2033                                 psf->sf_crcount = qrv;
2034                                 rv++;
2035                         }
2036                 } else if (psf->sf_oldin) {
2037
2038                         psf->sf_crcount = 0;
2039                         /*
2040                          * add or update "delete" records if an active filter
2041                          * is now inactive
2042                          */
2043                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2044                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2045                                         break;
2046                         if (!dpsf) {
2047                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2048                                 if (!dpsf)
2049                                         continue;
2050                                 *dpsf = *psf;
2051                                 /* pmc->lock held by callers */
2052                                 dpsf->sf_next = pmc->tomb;
2053                                 pmc->tomb = dpsf;
2054                         }
2055                         dpsf->sf_crcount = qrv;
2056                         rv++;
2057                 }
2058         }
2059         return rv;
2060 }
2061 #endif
2062
2063 /*
2064  * Add multicast source filter list to the interface list
2065  */
2066 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2067                          int sfcount, __be32 *psfsrc, int delta)
2068 {
2069         struct ip_mc_list *pmc;
2070         int     isexclude;
2071         int     i, err;
2072
2073         if (!in_dev)
2074                 return -ENODEV;
2075         rcu_read_lock();
2076         for_each_pmc_rcu(in_dev, pmc) {
2077                 if (*pmca == pmc->multiaddr)
2078                         break;
2079         }
2080         if (!pmc) {
2081                 /* MCA not found?? bug */
2082                 rcu_read_unlock();
2083                 return -ESRCH;
2084         }
2085         spin_lock_bh(&pmc->lock);
2086         rcu_read_unlock();
2087
2088 #ifdef CONFIG_IP_MULTICAST
2089         sf_markstate(pmc);
2090 #endif
2091         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2092         if (!delta)
2093                 pmc->sfcount[sfmode]++;
2094         err = 0;
2095         for (i = 0; i < sfcount; i++) {
2096                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2097                 if (err)
2098                         break;
2099         }
2100         if (err) {
2101                 int j;
2102
2103                 if (!delta)
2104                         pmc->sfcount[sfmode]--;
2105                 for (j = 0; j < i; j++)
2106                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2107         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2108 #ifdef CONFIG_IP_MULTICAST
2109                 struct ip_sf_list *psf;
2110                 struct net *net = dev_net(pmc->interface->dev);
2111                 in_dev = pmc->interface;
2112 #endif
2113
2114                 /* filter mode change */
2115                 if (pmc->sfcount[MCAST_EXCLUDE])
2116                         pmc->sfmode = MCAST_EXCLUDE;
2117                 else if (pmc->sfcount[MCAST_INCLUDE])
2118                         pmc->sfmode = MCAST_INCLUDE;
2119 #ifdef CONFIG_IP_MULTICAST
2120                 /* else no filters; keep old mode for reports */
2121
2122                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2123                 in_dev->mr_ifc_count = pmc->crcount;
2124                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2125                         psf->sf_crcount = 0;
2126                 igmp_ifc_event(in_dev);
2127         } else if (sf_setstate(pmc)) {
2128                 igmp_ifc_event(in_dev);
2129 #endif
2130         }
2131         spin_unlock_bh(&pmc->lock);
2132         return err;
2133 }
2134
2135 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2136 {
2137         struct ip_sf_list *tomb, *sources;
2138
2139         spin_lock_bh(&pmc->lock);
2140         tomb = pmc->tomb;
2141         pmc->tomb = NULL;
2142         sources = pmc->sources;
2143         pmc->sources = NULL;
2144         pmc->sfmode = MCAST_EXCLUDE;
2145         pmc->sfcount[MCAST_INCLUDE] = 0;
2146         pmc->sfcount[MCAST_EXCLUDE] = 1;
2147         spin_unlock_bh(&pmc->lock);
2148
2149         ip_sf_list_clear_all(tomb);
2150         ip_sf_list_clear_all(sources);
2151 }
2152
2153 /* Join a multicast group
2154  */
2155 static int __ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr,
2156                               unsigned int mode)
2157 {
2158         __be32 addr = imr->imr_multiaddr.s_addr;
2159         struct ip_mc_socklist *iml, *i;
2160         struct in_device *in_dev;
2161         struct inet_sock *inet = inet_sk(sk);
2162         struct net *net = sock_net(sk);
2163         int ifindex;
2164         int count = 0;
2165         int err;
2166
2167         ASSERT_RTNL();
2168
2169         if (!ipv4_is_multicast(addr))
2170                 return -EINVAL;
2171
2172         in_dev = ip_mc_find_dev(net, imr);
2173
2174         if (!in_dev) {
2175                 err = -ENODEV;
2176                 goto done;
2177         }
2178
2179         err = -EADDRINUSE;
2180         ifindex = imr->imr_ifindex;
2181         for_each_pmc_rtnl(inet, i) {
2182                 if (i->multi.imr_multiaddr.s_addr == addr &&
2183                     i->multi.imr_ifindex == ifindex)
2184                         goto done;
2185                 count++;
2186         }
2187         err = -ENOBUFS;
2188         if (count >= net->ipv4.sysctl_igmp_max_memberships)
2189                 goto done;
2190         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2191         if (!iml)
2192                 goto done;
2193
2194         memcpy(&iml->multi, imr, sizeof(*imr));
2195         iml->next_rcu = inet->mc_list;
2196         iml->sflist = NULL;
2197         iml->sfmode = mode;
2198         rcu_assign_pointer(inet->mc_list, iml);
2199         ____ip_mc_inc_group(in_dev, addr, mode, GFP_KERNEL);
2200         err = 0;
2201 done:
2202         return err;
2203 }
2204
2205 /* Join ASM (Any-Source Multicast) group
2206  */
2207 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2208 {
2209         return __ip_mc_join_group(sk, imr, MCAST_EXCLUDE);
2210 }
2211 EXPORT_SYMBOL(ip_mc_join_group);
2212
2213 /* Join SSM (Source-Specific Multicast) group
2214  */
2215 int ip_mc_join_group_ssm(struct sock *sk, struct ip_mreqn *imr,
2216                          unsigned int mode)
2217 {
2218         return __ip_mc_join_group(sk, imr, mode);
2219 }
2220
2221 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2222                            struct in_device *in_dev)
2223 {
2224         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2225         int err;
2226
2227         if (!psf) {
2228                 /* any-source empty exclude case */
2229                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2230                         iml->sfmode, 0, NULL, 0);
2231         }
2232         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2233                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2234         RCU_INIT_POINTER(iml->sflist, NULL);
2235         /* decrease mem now to avoid the memleak warning */
2236         atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2237         kfree_rcu(psf, rcu);
2238         return err;
2239 }
2240
2241 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2242 {
2243         struct inet_sock *inet = inet_sk(sk);
2244         struct ip_mc_socklist *iml;
2245         struct ip_mc_socklist __rcu **imlp;
2246         struct in_device *in_dev;
2247         struct net *net = sock_net(sk);
2248         __be32 group = imr->imr_multiaddr.s_addr;
2249         u32 ifindex;
2250         int ret = -EADDRNOTAVAIL;
2251
2252         ASSERT_RTNL();
2253
2254         in_dev = ip_mc_find_dev(net, imr);
2255         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2256                 ret = -ENODEV;
2257                 goto out;
2258         }
2259         ifindex = imr->imr_ifindex;
2260         for (imlp = &inet->mc_list;
2261              (iml = rtnl_dereference(*imlp)) != NULL;
2262              imlp = &iml->next_rcu) {
2263                 if (iml->multi.imr_multiaddr.s_addr != group)
2264                         continue;
2265                 if (ifindex) {
2266                         if (iml->multi.imr_ifindex != ifindex)
2267                                 continue;
2268                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2269                                 iml->multi.imr_address.s_addr)
2270                         continue;
2271
2272                 (void) ip_mc_leave_src(sk, iml, in_dev);
2273
2274                 *imlp = iml->next_rcu;
2275
2276                 if (in_dev)
2277                         ip_mc_dec_group(in_dev, group);
2278
2279                 /* decrease mem now to avoid the memleak warning */
2280                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2281                 kfree_rcu(iml, rcu);
2282                 return 0;
2283         }
2284 out:
2285         return ret;
2286 }
2287 EXPORT_SYMBOL(ip_mc_leave_group);
2288
2289 int ip_mc_source(int add, int omode, struct sock *sk, struct
2290         ip_mreq_source *mreqs, int ifindex)
2291 {
2292         int err;
2293         struct ip_mreqn imr;
2294         __be32 addr = mreqs->imr_multiaddr;
2295         struct ip_mc_socklist *pmc;
2296         struct in_device *in_dev = NULL;
2297         struct inet_sock *inet = inet_sk(sk);
2298         struct ip_sf_socklist *psl;
2299         struct net *net = sock_net(sk);
2300         int leavegroup = 0;
2301         int i, j, rv;
2302
2303         if (!ipv4_is_multicast(addr))
2304                 return -EINVAL;
2305
2306         ASSERT_RTNL();
2307
2308         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2309         imr.imr_address.s_addr = mreqs->imr_interface;
2310         imr.imr_ifindex = ifindex;
2311         in_dev = ip_mc_find_dev(net, &imr);
2312
2313         if (!in_dev) {
2314                 err = -ENODEV;
2315                 goto done;
2316         }
2317         err = -EADDRNOTAVAIL;
2318
2319         for_each_pmc_rtnl(inet, pmc) {
2320                 if ((pmc->multi.imr_multiaddr.s_addr ==
2321                      imr.imr_multiaddr.s_addr) &&
2322                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2323                         break;
2324         }
2325         if (!pmc) {             /* must have a prior join */
2326                 err = -EINVAL;
2327                 goto done;
2328         }
2329         /* if a source filter was set, must be the same mode as before */
2330         if (pmc->sflist) {
2331                 if (pmc->sfmode != omode) {
2332                         err = -EINVAL;
2333                         goto done;
2334                 }
2335         } else if (pmc->sfmode != omode) {
2336                 /* allow mode switches for empty-set filters */
2337                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2338                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2339                         NULL, 0);
2340                 pmc->sfmode = omode;
2341         }
2342
2343         psl = rtnl_dereference(pmc->sflist);
2344         if (!add) {
2345                 if (!psl)
2346                         goto done;      /* err = -EADDRNOTAVAIL */
2347                 rv = !0;
2348                 for (i = 0; i < psl->sl_count; i++) {
2349                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2350                                 sizeof(__be32));
2351                         if (rv == 0)
2352                                 break;
2353                 }
2354                 if (rv)         /* source not found */
2355                         goto done;      /* err = -EADDRNOTAVAIL */
2356
2357                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2358                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2359                         leavegroup = 1;
2360                         goto done;
2361                 }
2362
2363                 /* update the interface filter */
2364                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2365                         &mreqs->imr_sourceaddr, 1);
2366
2367                 for (j = i+1; j < psl->sl_count; j++)
2368                         psl->sl_addr[j-1] = psl->sl_addr[j];
2369                 psl->sl_count--;
2370                 err = 0;
2371                 goto done;
2372         }
2373         /* else, add a new source to the filter */
2374
2375         if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2376                 err = -ENOBUFS;
2377                 goto done;
2378         }
2379         if (!psl || psl->sl_count == psl->sl_max) {
2380                 struct ip_sf_socklist *newpsl;
2381                 int count = IP_SFBLOCK;
2382
2383                 if (psl)
2384                         count += psl->sl_max;
2385                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2386                 if (!newpsl) {
2387                         err = -ENOBUFS;
2388                         goto done;
2389                 }
2390                 newpsl->sl_max = count;
2391                 newpsl->sl_count = count - IP_SFBLOCK;
2392                 if (psl) {
2393                         for (i = 0; i < psl->sl_count; i++)
2394                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2395                         /* decrease mem now to avoid the memleak warning */
2396                         atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2397                         kfree_rcu(psl, rcu);
2398                 }
2399                 rcu_assign_pointer(pmc->sflist, newpsl);
2400                 psl = newpsl;
2401         }
2402         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2403         for (i = 0; i < psl->sl_count; i++) {
2404                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2405                         sizeof(__be32));
2406                 if (rv == 0)
2407                         break;
2408         }
2409         if (rv == 0)            /* address already there is an error */
2410                 goto done;
2411         for (j = psl->sl_count-1; j >= i; j--)
2412                 psl->sl_addr[j+1] = psl->sl_addr[j];
2413         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2414         psl->sl_count++;
2415         err = 0;
2416         /* update the interface list */
2417         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2418                 &mreqs->imr_sourceaddr, 1);
2419 done:
2420         if (leavegroup)
2421                 err = ip_mc_leave_group(sk, &imr);
2422         return err;
2423 }
2424
2425 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2426 {
2427         int err = 0;
2428         struct ip_mreqn imr;
2429         __be32 addr = msf->imsf_multiaddr;
2430         struct ip_mc_socklist *pmc;
2431         struct in_device *in_dev;
2432         struct inet_sock *inet = inet_sk(sk);
2433         struct ip_sf_socklist *newpsl, *psl;
2434         struct net *net = sock_net(sk);
2435         int leavegroup = 0;
2436
2437         if (!ipv4_is_multicast(addr))
2438                 return -EINVAL;
2439         if (msf->imsf_fmode != MCAST_INCLUDE &&
2440             msf->imsf_fmode != MCAST_EXCLUDE)
2441                 return -EINVAL;
2442
2443         ASSERT_RTNL();
2444
2445         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2446         imr.imr_address.s_addr = msf->imsf_interface;
2447         imr.imr_ifindex = ifindex;
2448         in_dev = ip_mc_find_dev(net, &imr);
2449
2450         if (!in_dev) {
2451                 err = -ENODEV;
2452                 goto done;
2453         }
2454
2455         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2456         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2457                 leavegroup = 1;
2458                 goto done;
2459         }
2460
2461         for_each_pmc_rtnl(inet, pmc) {
2462                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2463                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2464                         break;
2465         }
2466         if (!pmc) {             /* must have a prior join */
2467                 err = -EINVAL;
2468                 goto done;
2469         }
2470         if (msf->imsf_numsrc) {
2471                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2472                                                            GFP_KERNEL);
2473                 if (!newpsl) {
2474                         err = -ENOBUFS;
2475                         goto done;
2476                 }
2477                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2478                 memcpy(newpsl->sl_addr, msf->imsf_slist,
2479                         msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2480                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2481                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2482                 if (err) {
2483                         sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2484                         goto done;
2485                 }
2486         } else {
2487                 newpsl = NULL;
2488                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2489                                      msf->imsf_fmode, 0, NULL, 0);
2490         }
2491         psl = rtnl_dereference(pmc->sflist);
2492         if (psl) {
2493                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2494                         psl->sl_count, psl->sl_addr, 0);
2495                 /* decrease mem now to avoid the memleak warning */
2496                 atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2497                 kfree_rcu(psl, rcu);
2498         } else
2499                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2500                         0, NULL, 0);
2501         rcu_assign_pointer(pmc->sflist, newpsl);
2502         pmc->sfmode = msf->imsf_fmode;
2503         err = 0;
2504 done:
2505         if (leavegroup)
2506                 err = ip_mc_leave_group(sk, &imr);
2507         return err;
2508 }
2509
2510 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2511         struct ip_msfilter __user *optval, int __user *optlen)
2512 {
2513         int err, len, count, copycount;
2514         struct ip_mreqn imr;
2515         __be32 addr = msf->imsf_multiaddr;
2516         struct ip_mc_socklist *pmc;
2517         struct in_device *in_dev;
2518         struct inet_sock *inet = inet_sk(sk);
2519         struct ip_sf_socklist *psl;
2520         struct net *net = sock_net(sk);
2521
2522         ASSERT_RTNL();
2523
2524         if (!ipv4_is_multicast(addr))
2525                 return -EINVAL;
2526
2527         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2528         imr.imr_address.s_addr = msf->imsf_interface;
2529         imr.imr_ifindex = 0;
2530         in_dev = ip_mc_find_dev(net, &imr);
2531
2532         if (!in_dev) {
2533                 err = -ENODEV;
2534                 goto done;
2535         }
2536         err = -EADDRNOTAVAIL;
2537
2538         for_each_pmc_rtnl(inet, pmc) {
2539                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2540                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2541                         break;
2542         }
2543         if (!pmc)               /* must have a prior join */
2544                 goto done;
2545         msf->imsf_fmode = pmc->sfmode;
2546         psl = rtnl_dereference(pmc->sflist);
2547         if (!psl) {
2548                 len = 0;
2549                 count = 0;
2550         } else {
2551                 count = psl->sl_count;
2552         }
2553         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2554         len = copycount * sizeof(psl->sl_addr[0]);
2555         msf->imsf_numsrc = count;
2556         if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2557             copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2558                 return -EFAULT;
2559         }
2560         if (len &&
2561             copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2562                 return -EFAULT;
2563         return 0;
2564 done:
2565         return err;
2566 }
2567
2568 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2569         struct sockaddr_storage __user *p)
2570 {
2571         int i, count, copycount;
2572         struct sockaddr_in *psin;
2573         __be32 addr;
2574         struct ip_mc_socklist *pmc;
2575         struct inet_sock *inet = inet_sk(sk);
2576         struct ip_sf_socklist *psl;
2577
2578         ASSERT_RTNL();
2579
2580         psin = (struct sockaddr_in *)&gsf->gf_group;
2581         if (psin->sin_family != AF_INET)
2582                 return -EINVAL;
2583         addr = psin->sin_addr.s_addr;
2584         if (!ipv4_is_multicast(addr))
2585                 return -EINVAL;
2586
2587         for_each_pmc_rtnl(inet, pmc) {
2588                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2589                     pmc->multi.imr_ifindex == gsf->gf_interface)
2590                         break;
2591         }
2592         if (!pmc)               /* must have a prior join */
2593                 return -EADDRNOTAVAIL;
2594         gsf->gf_fmode = pmc->sfmode;
2595         psl = rtnl_dereference(pmc->sflist);
2596         count = psl ? psl->sl_count : 0;
2597         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2598         gsf->gf_numsrc = count;
2599         for (i = 0; i < copycount; i++, p++) {
2600                 struct sockaddr_storage ss;
2601
2602                 psin = (struct sockaddr_in *)&ss;
2603                 memset(&ss, 0, sizeof(ss));
2604                 psin->sin_family = AF_INET;
2605                 psin->sin_addr.s_addr = psl->sl_addr[i];
2606                 if (copy_to_user(p, &ss, sizeof(ss)))
2607                         return -EFAULT;
2608         }
2609         return 0;
2610 }
2611
2612 /*
2613  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2614  */
2615 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2616                    int dif, int sdif)
2617 {
2618         struct inet_sock *inet = inet_sk(sk);
2619         struct ip_mc_socklist *pmc;
2620         struct ip_sf_socklist *psl;
2621         int i;
2622         int ret;
2623
2624         ret = 1;
2625         if (!ipv4_is_multicast(loc_addr))
2626                 goto out;
2627
2628         rcu_read_lock();
2629         for_each_pmc_rcu(inet, pmc) {
2630                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2631                     (pmc->multi.imr_ifindex == dif ||
2632                      (sdif && pmc->multi.imr_ifindex == sdif)))
2633                         break;
2634         }
2635         ret = inet->mc_all;
2636         if (!pmc)
2637                 goto unlock;
2638         psl = rcu_dereference(pmc->sflist);
2639         ret = (pmc->sfmode == MCAST_EXCLUDE);
2640         if (!psl)
2641                 goto unlock;
2642
2643         for (i = 0; i < psl->sl_count; i++) {
2644                 if (psl->sl_addr[i] == rmt_addr)
2645                         break;
2646         }
2647         ret = 0;
2648         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2649                 goto unlock;
2650         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2651                 goto unlock;
2652         ret = 1;
2653 unlock:
2654         rcu_read_unlock();
2655 out:
2656         return ret;
2657 }
2658
2659 /*
2660  *      A socket is closing.
2661  */
2662
2663 void ip_mc_drop_socket(struct sock *sk)
2664 {
2665         struct inet_sock *inet = inet_sk(sk);
2666         struct ip_mc_socklist *iml;
2667         struct net *net = sock_net(sk);
2668
2669         if (!inet->mc_list)
2670                 return;
2671
2672         rtnl_lock();
2673         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2674                 struct in_device *in_dev;
2675
2676                 inet->mc_list = iml->next_rcu;
2677                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2678                 (void) ip_mc_leave_src(sk, iml, in_dev);
2679                 if (in_dev)
2680                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2681                 /* decrease mem now to avoid the memleak warning */
2682                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2683                 kfree_rcu(iml, rcu);
2684         }
2685         rtnl_unlock();
2686 }
2687
2688 /* called with rcu_read_lock() */
2689 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2690 {
2691         struct ip_mc_list *im;
2692         struct ip_mc_list __rcu **mc_hash;
2693         struct ip_sf_list *psf;
2694         int rv = 0;
2695
2696         mc_hash = rcu_dereference(in_dev->mc_hash);
2697         if (mc_hash) {
2698                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2699
2700                 for (im = rcu_dereference(mc_hash[hash]);
2701                      im != NULL;
2702                      im = rcu_dereference(im->next_hash)) {
2703                         if (im->multiaddr == mc_addr)
2704                                 break;
2705                 }
2706         } else {
2707                 for_each_pmc_rcu(in_dev, im) {
2708                         if (im->multiaddr == mc_addr)
2709                                 break;
2710                 }
2711         }
2712         if (im && proto == IPPROTO_IGMP) {
2713                 rv = 1;
2714         } else if (im) {
2715                 if (src_addr) {
2716                         for (psf = im->sources; psf; psf = psf->sf_next) {
2717                                 if (psf->sf_inaddr == src_addr)
2718                                         break;
2719                         }
2720                         if (psf)
2721                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2722                                         psf->sf_count[MCAST_EXCLUDE] !=
2723                                         im->sfcount[MCAST_EXCLUDE];
2724                         else
2725                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2726                 } else
2727                         rv = 1; /* unspecified source; tentatively allow */
2728         }
2729         return rv;
2730 }
2731
2732 #if defined(CONFIG_PROC_FS)
2733 struct igmp_mc_iter_state {
2734         struct seq_net_private p;
2735         struct net_device *dev;
2736         struct in_device *in_dev;
2737 };
2738
2739 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2740
2741 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2742 {
2743         struct net *net = seq_file_net(seq);
2744         struct ip_mc_list *im = NULL;
2745         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2746
2747         state->in_dev = NULL;
2748         for_each_netdev_rcu(net, state->dev) {
2749                 struct in_device *in_dev;
2750
2751                 in_dev = __in_dev_get_rcu(state->dev);
2752                 if (!in_dev)
2753                         continue;
2754                 im = rcu_dereference(in_dev->mc_list);
2755                 if (im) {
2756                         state->in_dev = in_dev;
2757                         break;
2758                 }
2759         }
2760         return im;
2761 }
2762
2763 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2764 {
2765         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2766
2767         im = rcu_dereference(im->next_rcu);
2768         while (!im) {
2769                 state->dev = next_net_device_rcu(state->dev);
2770                 if (!state->dev) {
2771                         state->in_dev = NULL;
2772                         break;
2773                 }
2774                 state->in_dev = __in_dev_get_rcu(state->dev);
2775                 if (!state->in_dev)
2776                         continue;
2777                 im = rcu_dereference(state->in_dev->mc_list);
2778         }
2779         return im;
2780 }
2781
2782 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2783 {
2784         struct ip_mc_list *im = igmp_mc_get_first(seq);
2785         if (im)
2786                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2787                         --pos;
2788         return pos ? NULL : im;
2789 }
2790
2791 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2792         __acquires(rcu)
2793 {
2794         rcu_read_lock();
2795         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2796 }
2797
2798 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2799 {
2800         struct ip_mc_list *im;
2801         if (v == SEQ_START_TOKEN)
2802                 im = igmp_mc_get_first(seq);
2803         else
2804                 im = igmp_mc_get_next(seq, v);
2805         ++*pos;
2806         return im;
2807 }
2808
2809 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2810         __releases(rcu)
2811 {
2812         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2813
2814         state->in_dev = NULL;
2815         state->dev = NULL;
2816         rcu_read_unlock();
2817 }
2818
2819 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2820 {
2821         if (v == SEQ_START_TOKEN)
2822                 seq_puts(seq,
2823                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2824         else {
2825                 struct ip_mc_list *im = (struct ip_mc_list *)v;
2826                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2827                 char   *querier;
2828                 long delta;
2829
2830 #ifdef CONFIG_IP_MULTICAST
2831                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2832                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2833                           "V3";
2834 #else
2835                 querier = "NONE";
2836 #endif
2837
2838                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2839                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2840                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2841                 }
2842
2843                 delta = im->timer.expires - jiffies;
2844                 seq_printf(seq,
2845                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2846                            im->multiaddr, im->users,
2847                            im->tm_running,
2848                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2849                            im->reporter);
2850         }
2851         return 0;
2852 }
2853
2854 static const struct seq_operations igmp_mc_seq_ops = {
2855         .start  =       igmp_mc_seq_start,
2856         .next   =       igmp_mc_seq_next,
2857         .stop   =       igmp_mc_seq_stop,
2858         .show   =       igmp_mc_seq_show,
2859 };
2860
2861 struct igmp_mcf_iter_state {
2862         struct seq_net_private p;
2863         struct net_device *dev;
2864         struct in_device *idev;
2865         struct ip_mc_list *im;
2866 };
2867
2868 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2869
2870 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2871 {
2872         struct net *net = seq_file_net(seq);
2873         struct ip_sf_list *psf = NULL;
2874         struct ip_mc_list *im = NULL;
2875         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2876
2877         state->idev = NULL;
2878         state->im = NULL;
2879         for_each_netdev_rcu(net, state->dev) {
2880                 struct in_device *idev;
2881                 idev = __in_dev_get_rcu(state->dev);
2882                 if (unlikely(!idev))
2883                         continue;
2884                 im = rcu_dereference(idev->mc_list);
2885                 if (likely(im)) {
2886                         spin_lock_bh(&im->lock);
2887                         psf = im->sources;
2888                         if (likely(psf)) {
2889                                 state->im = im;
2890                                 state->idev = idev;
2891                                 break;
2892                         }
2893                         spin_unlock_bh(&im->lock);
2894                 }
2895         }
2896         return psf;
2897 }
2898
2899 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2900 {
2901         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2902
2903         psf = psf->sf_next;
2904         while (!psf) {
2905                 spin_unlock_bh(&state->im->lock);
2906                 state->im = state->im->next;
2907                 while (!state->im) {
2908                         state->dev = next_net_device_rcu(state->dev);
2909                         if (!state->dev) {
2910                                 state->idev = NULL;
2911                                 goto out;
2912                         }
2913                         state->idev = __in_dev_get_rcu(state->dev);
2914                         if (!state->idev)
2915                                 continue;
2916                         state->im = rcu_dereference(state->idev->mc_list);
2917                 }
2918                 if (!state->im)
2919                         break;
2920                 spin_lock_bh(&state->im->lock);
2921                 psf = state->im->sources;
2922         }
2923 out:
2924         return psf;
2925 }
2926
2927 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2928 {
2929         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2930         if (psf)
2931                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2932                         --pos;
2933         return pos ? NULL : psf;
2934 }
2935
2936 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2937         __acquires(rcu)
2938 {
2939         rcu_read_lock();
2940         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2941 }
2942
2943 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2944 {
2945         struct ip_sf_list *psf;
2946         if (v == SEQ_START_TOKEN)
2947                 psf = igmp_mcf_get_first(seq);
2948         else
2949                 psf = igmp_mcf_get_next(seq, v);
2950         ++*pos;
2951         return psf;
2952 }
2953
2954 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2955         __releases(rcu)
2956 {
2957         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2958         if (likely(state->im)) {
2959                 spin_unlock_bh(&state->im->lock);
2960                 state->im = NULL;
2961         }
2962         state->idev = NULL;
2963         state->dev = NULL;
2964         rcu_read_unlock();
2965 }
2966
2967 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2968 {
2969         struct ip_sf_list *psf = (struct ip_sf_list *)v;
2970         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2971
2972         if (v == SEQ_START_TOKEN) {
2973                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2974         } else {
2975                 seq_printf(seq,
2976                            "%3d %6.6s 0x%08x "
2977                            "0x%08x %6lu %6lu\n",
2978                            state->dev->ifindex, state->dev->name,
2979                            ntohl(state->im->multiaddr),
2980                            ntohl(psf->sf_inaddr),
2981                            psf->sf_count[MCAST_INCLUDE],
2982                            psf->sf_count[MCAST_EXCLUDE]);
2983         }
2984         return 0;
2985 }
2986
2987 static const struct seq_operations igmp_mcf_seq_ops = {
2988         .start  =       igmp_mcf_seq_start,
2989         .next   =       igmp_mcf_seq_next,
2990         .stop   =       igmp_mcf_seq_stop,
2991         .show   =       igmp_mcf_seq_show,
2992 };
2993
2994 static int __net_init igmp_net_init(struct net *net)
2995 {
2996         struct proc_dir_entry *pde;
2997         int err;
2998
2999         pde = proc_create_net("igmp", 0444, net->proc_net, &igmp_mc_seq_ops,
3000                         sizeof(struct igmp_mc_iter_state));
3001         if (!pde)
3002                 goto out_igmp;
3003         pde = proc_create_net("mcfilter", 0444, net->proc_net,
3004                         &igmp_mcf_seq_ops, sizeof(struct igmp_mcf_iter_state));
3005         if (!pde)
3006                 goto out_mcfilter;
3007         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3008                                    SOCK_DGRAM, 0, net);
3009         if (err < 0) {
3010                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3011                        err);
3012                 goto out_sock;
3013         }
3014
3015         return 0;
3016
3017 out_sock:
3018         remove_proc_entry("mcfilter", net->proc_net);
3019 out_mcfilter:
3020         remove_proc_entry("igmp", net->proc_net);
3021 out_igmp:
3022         return -ENOMEM;
3023 }
3024
3025 static void __net_exit igmp_net_exit(struct net *net)
3026 {
3027         remove_proc_entry("mcfilter", net->proc_net);
3028         remove_proc_entry("igmp", net->proc_net);
3029         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3030 }
3031
3032 static struct pernet_operations igmp_net_ops = {
3033         .init = igmp_net_init,
3034         .exit = igmp_net_exit,
3035 };
3036 #endif
3037
3038 static int igmp_netdev_event(struct notifier_block *this,
3039                              unsigned long event, void *ptr)
3040 {
3041         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3042         struct in_device *in_dev;
3043
3044         switch (event) {
3045         case NETDEV_RESEND_IGMP:
3046                 in_dev = __in_dev_get_rtnl(dev);
3047                 if (in_dev)
3048                         ip_mc_rejoin_groups(in_dev);
3049                 break;
3050         default:
3051                 break;
3052         }
3053         return NOTIFY_DONE;
3054 }
3055
3056 static struct notifier_block igmp_notifier = {
3057         .notifier_call = igmp_netdev_event,
3058 };
3059
3060 int __init igmp_mc_init(void)
3061 {
3062 #if defined(CONFIG_PROC_FS)
3063         int err;
3064
3065         err = register_pernet_subsys(&igmp_net_ops);
3066         if (err)
3067                 return err;
3068         err = register_netdevice_notifier(&igmp_notifier);
3069         if (err)
3070                 goto reg_notif_fail;
3071         return 0;
3072
3073 reg_notif_fail:
3074         unregister_pernet_subsys(&igmp_net_ops);
3075         return err;
3076 #else
3077         return register_netdevice_notifier(&igmp_notifier);
3078 #endif
3079 }