net/ipv4/igmp.c

   1 /*
   2  *      Linux NET3:     Internet Group Management Protocol  [IGMP]
   3  *
   4  *      This code implements the IGMP protocol as defined in RFC1112. There has
   5  *      been a further revision of this protocol since which is now supported.
   6  *
   7  *      If you have trouble with this module be careful what gcc you have used,
   8  *      the older version didn't come out right using gcc 2.5.8, the newer one
   9  *      seems to fall out with gcc 2.6.2.
  10  *
  11  *      Authors:
  12  *              Alan Cox <alan@lxorguk.ukuu.org.uk>
  13  *
  14  *      This program is free software; you can redistribute it and/or
  15  *      modify it under the terms of the GNU General Public License
  16  *      as published by the Free Software Foundation; either version
  17  *      2 of the License, or (at your option) any later version.
  18  *
  19  *      Fixes:
  20  *
  21  *              Alan Cox        :       Added lots of __inline__ to optimise
  22  *                                      the memory usage of all the tiny little
  23  *                                      functions.
  24  *              Alan Cox        :       Dumped the header building experiment.
  25  *              Alan Cox        :       Minor tweaks ready for multicast routing
  26  *                                      and extended IGMP protocol.
  27  *              Alan Cox        :       Removed a load of inline directives. Gcc 2.5.8
  28  *                                      writes utterly bogus code otherwise (sigh)
  29  *                                      fixed IGMP loopback to behave in the manner
  30  *                                      desired by mrouted, fixed the fact it has been
  31  *                                      broken since 1.3.6 and cleaned up a few minor
  32  *                                      points.
  33  *
  34  *              Chih-Jen Chang  :       Tried to revise IGMP to Version 2
  35  *              Tsu-Sheng Tsao          E-mail: chihjenc@scf.usc.edu and tsusheng@scf.usc.edu
  36  *                                      The enhancements are mainly based on Steve Deering's
  37  *                                      ipmulti-3.5 source code.
  38  *              Chih-Jen Chang  :       Added the igmp_get_mrouter_info and
  39  *              Tsu-Sheng Tsao          igmp_set_mrouter_info to keep track of
  40  *                                      the mrouted version on that device.
  41  *              Chih-Jen Chang  :       Added the max_resp_time parameter to
  42  *              Tsu-Sheng Tsao          igmp_heard_query(). Using this parameter
  43  *                                      to identify the multicast router version
  44  *                                      and do what the IGMP version 2 specified.
  45  *              Chih-Jen Chang  :       Added a timer to revert to IGMP V2 router
  46  *              Tsu-Sheng Tsao          if the specified time expired.
  47  *              Alan Cox        :       Stop IGMP from 0.0.0.0 being accepted.
  48  *              Alan Cox        :       Use GFP_ATOMIC in the right places.
  49  *              Christian Daudt :       igmp timer wasn't set for local group
  50  *                                      memberships but was being deleted,
  51  *                                      which caused a "del_timer() called
  52  *                                      from %p with timer not initialized\n"
  53  *                                      message (960131).
  54  *              Christian Daudt :       removed del_timer from
  55  *                                      igmp_timer_expire function (960205).
  56  *             Christian Daudt :       igmp_heard_report now only calls
  57  *                                     igmp_timer_expire if tm->running is
  58  *                                     true (960216).
  59  *              Malcolm Beattie :       ttl comparison wrong in igmp_rcv made
  60  *                                      igmp_heard_query never trigger. Expiry
  61  *                                      miscalculation fixed in igmp_heard_query
  62  *                                      and random() made to return unsigned to
  63  *                                      prevent negative expiry times.
  64  *              Alexey Kuznetsov:       Wrong group leaving behaviour, backport
  65  *                                      fix from pending 2.1.x patches.
  66  *              Alan Cox:               Forget to enable FDDI support earlier.
  67  *              Alexey Kuznetsov:       Fixed leaving groups on device down.
  68  *              Alexey Kuznetsov:       Accordance to igmp-v2-06 draft.
  69  *              David L Stevens:        IGMPv3 support, with help from
  70  *                                      Vinay Kulkarni
  71  */
  72
  73 #include <linux/module.h>
  74 #include <linux/slab.h>
  75 #include <linux/uaccess.h>
  76 #include <linux/types.h>
  77 #include <linux/kernel.h>
  78 #include <linux/jiffies.h>
  79 #include <linux/string.h>
  80 #include <linux/socket.h>
  81 #include <linux/sockios.h>
  82 #include <linux/in.h>
  83 #include <linux/inet.h>
  84 #include <linux/netdevice.h>
  85 #include <linux/skbuff.h>
  86 #include <linux/inetdevice.h>
  87 #include <linux/igmp.h>
  88 #include <linux/if_arp.h>
  89 #include <linux/rtnetlink.h>
  90 #include <linux/times.h>
  91 #include <linux/pkt_sched.h>
  92 #include <linux/byteorder/generic.h>
  93
  94 #include <net/net_namespace.h>
  95 #include <net/arp.h>
  96 #include <net/ip.h>
  97 #include <net/protocol.h>
  98 #include <net/route.h>
  99 #include <net/sock.h>
 100 #include <net/checksum.h>
 101 #include <net/inet_common.h>
 102 #include <linux/netfilter_ipv4.h>
 103 #ifdef CONFIG_IP_MROUTE
 104 #include <linux/mroute.h>
 105 #endif
 106 #ifdef CONFIG_PROC_FS
 107 #include <linux/proc_fs.h>
 108 #include <linux/seq_file.h>
 109 #endif
 110
 111 #ifdef CONFIG_IP_MULTICAST
 112 /* Parameter names and values are taken from igmp-v2-06 draft */
 113
 114 #define IGMP_V2_UNSOLICITED_REPORT_INTERVAL     (10*HZ)
 115 #define IGMP_V3_UNSOLICITED_REPORT_INTERVAL     (1*HZ)
 116 #define IGMP_QUERY_INTERVAL                     (125*HZ)
 117 #define IGMP_QUERY_RESPONSE_INTERVAL            (10*HZ)
 118
 119 #define IGMP_INITIAL_REPORT_DELAY               (1)
 120
 121 /* IGMP_INITIAL_REPORT_DELAY is not from IGMP specs!
 122  * IGMP specs require to report membership immediately after
 123  * joining a group, but we delay the first report by a
 124  * small interval. It seems more natural and still does not
 125  * contradict to specs provided this delay is small enough.
 126  */
 127
 128 #define IGMP_V1_SEEN(in_dev) \
 129         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
 130          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
 131          ((in_dev)->mr_v1_seen && \
 132           time_before(jiffies, (in_dev)->mr_v1_seen)))
 133 #define IGMP_V2_SEEN(in_dev) \
 134         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
 135          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
 136          ((in_dev)->mr_v2_seen && \
 137           time_before(jiffies, (in_dev)->mr_v2_seen)))
 138
 139 static int unsolicited_report_interval(struct in_device *in_dev)
 140 {
 141         int interval_ms, interval_jiffies;
 142
 143         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 144                 interval_ms = IN_DEV_CONF_GET(
 145                         in_dev,
 146                         IGMPV2_UNSOLICITED_REPORT_INTERVAL);
 147         else /* v3 */
 148                 interval_ms = IN_DEV_CONF_GET(
 149                         in_dev,
 150                         IGMPV3_UNSOLICITED_REPORT_INTERVAL);
 151
 152         interval_jiffies = msecs_to_jiffies(interval_ms);
 153
 154         /* _timer functions can't handle a delay of 0 jiffies so ensure
 155          *  we always return a positive value.
 156          */
 157         if (interval_jiffies <= 0)
 158                 interval_jiffies = 1;
 159         return interval_jiffies;
 160 }
 161
 162 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
 163                               gfp_t gfp);
 164 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 165 static void igmpv3_clear_delrec(struct in_device *in_dev);
 166 static int sf_setstate(struct ip_mc_list *pmc);
 167 static void sf_markstate(struct ip_mc_list *pmc);
 168 #endif
 169 static void ip_mc_clear_src(struct ip_mc_list *pmc);
 170 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
 171                          int sfcount, __be32 *psfsrc, int delta);
 172
 173 static void ip_ma_put(struct ip_mc_list *im)
 174 {
 175         if (refcount_dec_and_test(&im->refcnt)) {
 176                 in_dev_put(im->interface);
 177                 kfree_rcu(im, rcu);
 178         }
 179 }
 180
 181 #define for_each_pmc_rcu(in_dev, pmc)                           \
 182         for (pmc = rcu_dereference(in_dev->mc_list);            \
 183              pmc != NULL;                                       \
 184              pmc = rcu_dereference(pmc->next_rcu))
 185
 186 #define for_each_pmc_rtnl(in_dev, pmc)                          \
 187         for (pmc = rtnl_dereference(in_dev->mc_list);           \
 188              pmc != NULL;                                       \
 189              pmc = rtnl_dereference(pmc->next_rcu))
 190
 191 #ifdef CONFIG_IP_MULTICAST
 192
 193 /*
 194  *      Timer management
 195  */
 196
 197 static void igmp_stop_timer(struct ip_mc_list *im)
 198 {
 199         spin_lock_bh(&im->lock);
 200         if (del_timer(&im->timer))
 201                 refcount_dec(&im->refcnt);
 202         im->tm_running = 0;
 203         im->reporter = 0;
 204         im->unsolicit_count = 0;
 205         spin_unlock_bh(&im->lock);
 206 }
 207
 208 /* It must be called with locked im->lock */
 209 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
 210 {
 211         int tv = prandom_u32() % max_delay;
 212
 213         im->tm_running = 1;
 214         if (!mod_timer(&im->timer, jiffies+tv+2))
 215                 refcount_inc(&im->refcnt);
 216 }
 217
 218 static void igmp_gq_start_timer(struct in_device *in_dev)
 219 {
 220         int tv = prandom_u32() % in_dev->mr_maxdelay;
 221         unsigned long exp = jiffies + tv + 2;
 222
 223         if (in_dev->mr_gq_running &&
 224             time_after_eq(exp, (in_dev->mr_gq_timer).expires))
 225                 return;
 226
 227         in_dev->mr_gq_running = 1;
 228         if (!mod_timer(&in_dev->mr_gq_timer, exp))
 229                 in_dev_hold(in_dev);
 230 }
 231
 232 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 233 {
 234         int tv = prandom_u32() % delay;
 235
 236         if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
 237                 in_dev_hold(in_dev);
 238 }
 239
 240 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 241 {
 242         spin_lock_bh(&im->lock);
 243         im->unsolicit_count = 0;
 244         if (del_timer(&im->timer)) {
 245                 if ((long)(im->timer.expires-jiffies) < max_delay) {
 246                         add_timer(&im->timer);
 247                         im->tm_running = 1;
 248                         spin_unlock_bh(&im->lock);
 249                         return;
 250                 }
 251                 refcount_dec(&im->refcnt);
 252         }
 253         igmp_start_timer(im, max_delay);
 254         spin_unlock_bh(&im->lock);
 255 }
 256
 257
 258 /*
 259  *      Send an IGMP report.
 260  */
 261
 262 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
 263
 264
 265 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
 266         int gdeleted, int sdeleted)
 267 {
 268         switch (type) {
 269         case IGMPV3_MODE_IS_INCLUDE:
 270         case IGMPV3_MODE_IS_EXCLUDE:
 271                 if (gdeleted || sdeleted)
 272                         return 0;
 273                 if (!(pmc->gsquery && !psf->sf_gsresp)) {
 274                         if (pmc->sfmode == MCAST_INCLUDE)
 275                                 return 1;
 276                         /* don't include if this source is excluded
 277                          * in all filters
 278                          */
 279                         if (psf->sf_count[MCAST_INCLUDE])
 280                                 return type == IGMPV3_MODE_IS_INCLUDE;
 281                         return pmc->sfcount[MCAST_EXCLUDE] ==
 282                                 psf->sf_count[MCAST_EXCLUDE];
 283                 }
 284                 return 0;
 285         case IGMPV3_CHANGE_TO_INCLUDE:
 286                 if (gdeleted || sdeleted)
 287                         return 0;
 288                 return psf->sf_count[MCAST_INCLUDE] != 0;
 289         case IGMPV3_CHANGE_TO_EXCLUDE:
 290                 if (gdeleted || sdeleted)
 291                         return 0;
 292                 if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
 293                     psf->sf_count[MCAST_INCLUDE])
 294                         return 0;
 295                 return pmc->sfcount[MCAST_EXCLUDE] ==
 296                         psf->sf_count[MCAST_EXCLUDE];
 297         case IGMPV3_ALLOW_NEW_SOURCES:
 298                 if (gdeleted || !psf->sf_crcount)
 299                         return 0;
 300                 return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
 301         case IGMPV3_BLOCK_OLD_SOURCES:
 302                 if (pmc->sfmode == MCAST_INCLUDE)
 303                         return gdeleted || (psf->sf_crcount && sdeleted);
 304                 return psf->sf_crcount && !gdeleted && !sdeleted;
 305         }
 306         return 0;
 307 }
 308
 309 static int
 310 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
 311 {
 312         struct ip_sf_list *psf;
 313         int scount = 0;
 314
 315         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 316                 if (!is_in(pmc, psf, type, gdeleted, sdeleted))
 317                         continue;
 318                 scount++;
 319         }
 320         return scount;
 321 }
 322
 323 /* source address selection per RFC 3376 section 4.2.13 */
 324 static __be32 igmpv3_get_srcaddr(struct net_device *dev,
 325                                  const struct flowi4 *fl4)
 326 {
 327         struct in_device *in_dev = __in_dev_get_rcu(dev);
 328
 329         if (!in_dev)
 330                 return htonl(INADDR_ANY);
 331
 332         for_ifa(in_dev) {
 333                 if (fl4->saddr == ifa->ifa_local)
 334                         return fl4->saddr;
 335         } endfor_ifa(in_dev);
 336
 337         return htonl(INADDR_ANY);
 338 }
 339
 340 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
 341 {
 342         struct sk_buff *skb;
 343         struct rtable *rt;
 344         struct iphdr *pip;
 345         struct igmpv3_report *pig;
 346         struct net *net = dev_net(dev);
 347         struct flowi4 fl4;
 348         int hlen = LL_RESERVED_SPACE(dev);
 349         int tlen = dev->needed_tailroom;
 350         unsigned int size = mtu;
 351
 352         while (1) {
 353                 skb = alloc_skb(size + hlen + tlen,
 354                                 GFP_ATOMIC | __GFP_NOWARN);
 355                 if (skb)
 356                         break;
 357                 size >>= 1;
 358                 if (size < 256)
 359                         return NULL;
 360         }
 361         skb->priority = TC_PRIO_CONTROL;
 362
 363         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 364                                    0, 0,
 365                                    IPPROTO_IGMP, 0, dev->ifindex);
 366         if (IS_ERR(rt)) {
 367                 kfree_skb(skb);
 368                 return NULL;
 369         }
 370
 371         skb_dst_set(skb, &rt->dst);
 372         skb->dev = dev;
 373
 374         skb_reserve(skb, hlen);
 375         skb_tailroom_reserve(skb, mtu, tlen);
 376
 377         skb_reset_network_header(skb);
 378         pip = ip_hdr(skb);
 379         skb_put(skb, sizeof(struct iphdr) + 4);
 380
 381         pip->version  = 4;
 382         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 383         pip->tos      = 0xc0;
 384         pip->frag_off = htons(IP_DF);
 385         pip->ttl      = 1;
 386         pip->daddr    = fl4.daddr;
 387
 388         rcu_read_lock();
 389         pip->saddr    = igmpv3_get_srcaddr(dev, &fl4);
 390         rcu_read_unlock();
 391
 392         pip->protocol = IPPROTO_IGMP;
 393         pip->tot_len  = 0;      /* filled in later */
 394         ip_select_ident(net, skb, NULL);
 395         ((u8 *)&pip[1])[0] = IPOPT_RA;
 396         ((u8 *)&pip[1])[1] = 4;
 397         ((u8 *)&pip[1])[2] = 0;
 398         ((u8 *)&pip[1])[3] = 0;
 399
 400         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 401         skb_put(skb, sizeof(*pig));
 402         pig = igmpv3_report_hdr(skb);
 403         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 404         pig->resv1 = 0;
 405         pig->csum = 0;
 406         pig->resv2 = 0;
 407         pig->ngrec = 0;
 408         return skb;
 409 }
 410
 411 static int igmpv3_sendpack(struct sk_buff *skb)
 412 {
 413         struct igmphdr *pig = igmp_hdr(skb);
 414         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 415
 416         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 417
 418         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 419 }
 420
 421 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 422 {
 423         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 424 }
 425
 426 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 427         int type, struct igmpv3_grec **ppgr, unsigned int mtu)
 428 {
 429         struct net_device *dev = pmc->interface->dev;
 430         struct igmpv3_report *pih;
 431         struct igmpv3_grec *pgr;
 432
 433         if (!skb) {
 434                 skb = igmpv3_newpack(dev, mtu);
 435                 if (!skb)
 436                         return NULL;
 437         }
 438         pgr = skb_put(skb, sizeof(struct igmpv3_grec));
 439         pgr->grec_type = type;
 440         pgr->grec_auxwords = 0;
 441         pgr->grec_nsrcs = 0;
 442         pgr->grec_mca = pmc->multiaddr;
 443         pih = igmpv3_report_hdr(skb);
 444         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 445         *ppgr = pgr;
 446         return skb;
 447 }
 448
 449 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 450
 451 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 452         int type, int gdeleted, int sdeleted)
 453 {
 454         struct net_device *dev = pmc->interface->dev;
 455         struct net *net = dev_net(dev);
 456         struct igmpv3_report *pih;
 457         struct igmpv3_grec *pgr = NULL;
 458         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 459         int scount, stotal, first, isquery, truncate;
 460         unsigned int mtu;
 461
 462         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 463                 return skb;
 464         if (ipv4_is_local_multicast(pmc->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
 465                 return skb;
 466
 467         mtu = READ_ONCE(dev->mtu);
 468         if (mtu < IPV4_MIN_MTU)
 469                 return skb;
 470
 471         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 472                   type == IGMPV3_MODE_IS_EXCLUDE;
 473         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 474                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 475
 476         stotal = scount = 0;
 477
 478         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 479
 480         if (!*psf_list)
 481                 goto empty_source;
 482
 483         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 484
 485         /* EX and TO_EX get a fresh packet, if needed */
 486         if (truncate) {
 487                 if (pih && pih->ngrec &&
 488                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 489                         if (skb)
 490                                 igmpv3_sendpack(skb);
 491                         skb = igmpv3_newpack(dev, mtu);
 492                 }
 493         }
 494         first = 1;
 495         psf_prev = NULL;
 496         for (psf = *psf_list; psf; psf = psf_next) {
 497                 __be32 *psrc;
 498
 499                 psf_next = psf->sf_next;
 500
 501                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 502                         psf_prev = psf;
 503                         continue;
 504                 }
 505
 506                 /* Based on RFC3376 5.1. Should not send source-list change
 507                  * records when there is a filter mode change.
 508                  */
 509                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 510                      (!gdeleted && pmc->crcount)) &&
 511                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 512                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 513                         goto decrease_sf_crcount;
 514
 515                 /* clear marks on query responses */
 516                 if (isquery)
 517                         psf->sf_gsresp = 0;
 518
 519                 if (AVAILABLE(skb) < sizeof(__be32) +
 520                     first*sizeof(struct igmpv3_grec)) {
 521                         if (truncate && !first)
 522                                 break;   /* truncate these */
 523                         if (pgr)
 524                                 pgr->grec_nsrcs = htons(scount);
 525                         if (skb)
 526                                 igmpv3_sendpack(skb);
 527                         skb = igmpv3_newpack(dev, mtu);
 528                         first = 1;
 529                         scount = 0;
 530                 }
 531                 if (first) {
 532                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 533                         first = 0;
 534                 }
 535                 if (!skb)
 536                         return NULL;
 537                 psrc = skb_put(skb, sizeof(__be32));
 538                 *psrc = psf->sf_inaddr;
 539                 scount++; stotal++;
 540                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 541                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 542 decrease_sf_crcount:
 543                         psf->sf_crcount--;
 544                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 545                                 if (psf_prev)
 546                                         psf_prev->sf_next = psf->sf_next;
 547                                 else
 548                                         *psf_list = psf->sf_next;
 549                                 kfree(psf);
 550                                 continue;
 551                         }
 552                 }
 553                 psf_prev = psf;
 554         }
 555
 556 empty_source:
 557         if (!stotal) {
 558                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 559                     type == IGMPV3_BLOCK_OLD_SOURCES)
 560                         return skb;
 561                 if (pmc->crcount || isquery) {
 562                         /* make sure we have room for group header */
 563                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 564                                 igmpv3_sendpack(skb);
 565                                 skb = NULL; /* add_grhead will get a new one */
 566                         }
 567                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 568                 }
 569         }
 570         if (pgr)
 571                 pgr->grec_nsrcs = htons(scount);
 572
 573         if (isquery)
 574                 pmc->gsquery = 0;       /* clear query state on report */
 575         return skb;
 576 }
 577
 578 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 579 {
 580         struct sk_buff *skb = NULL;
 581         struct net *net = dev_net(in_dev->dev);
 582         int type;
 583
 584         if (!pmc) {
 585                 rcu_read_lock();
 586                 for_each_pmc_rcu(in_dev, pmc) {
 587                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 588                                 continue;
 589                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 590                              !net->ipv4.sysctl_igmp_llm_reports)
 591                                 continue;
 592                         spin_lock_bh(&pmc->lock);
 593                         if (pmc->sfcount[MCAST_EXCLUDE])
 594                                 type = IGMPV3_MODE_IS_EXCLUDE;
 595                         else
 596                                 type = IGMPV3_MODE_IS_INCLUDE;
 597                         skb = add_grec(skb, pmc, type, 0, 0);
 598                         spin_unlock_bh(&pmc->lock);
 599                 }
 600                 rcu_read_unlock();
 601         } else {
 602                 spin_lock_bh(&pmc->lock);
 603                 if (pmc->sfcount[MCAST_EXCLUDE])
 604                         type = IGMPV3_MODE_IS_EXCLUDE;
 605                 else
 606                         type = IGMPV3_MODE_IS_INCLUDE;
 607                 skb = add_grec(skb, pmc, type, 0, 0);
 608                 spin_unlock_bh(&pmc->lock);
 609         }
 610         if (!skb)
 611                 return 0;
 612         return igmpv3_sendpack(skb);
 613 }
 614
 615 /*
 616  * remove zero-count source records from a source filter list
 617  */
 618 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 619 {
 620         struct ip_sf_list *psf_prev, *psf_next, *psf;
 621
 622         psf_prev = NULL;
 623         for (psf = *ppsf; psf; psf = psf_next) {
 624                 psf_next = psf->sf_next;
 625                 if (psf->sf_crcount == 0) {
 626                         if (psf_prev)
 627                                 psf_prev->sf_next = psf->sf_next;
 628                         else
 629                                 *ppsf = psf->sf_next;
 630                         kfree(psf);
 631                 } else
 632                         psf_prev = psf;
 633         }
 634 }
 635
 636 static void igmpv3_send_cr(struct in_device *in_dev)
 637 {
 638         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 639         struct sk_buff *skb = NULL;
 640         int type, dtype;
 641
 642         rcu_read_lock();
 643         spin_lock_bh(&in_dev->mc_tomb_lock);
 644
 645         /* deleted MCA's */
 646         pmc_prev = NULL;
 647         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 648                 pmc_next = pmc->next;
 649                 if (pmc->sfmode == MCAST_INCLUDE) {
 650                         type = IGMPV3_BLOCK_OLD_SOURCES;
 651                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 652                         skb = add_grec(skb, pmc, type, 1, 0);
 653                         skb = add_grec(skb, pmc, dtype, 1, 1);
 654                 }
 655                 if (pmc->crcount) {
 656                         if (pmc->sfmode == MCAST_EXCLUDE) {
 657                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 658                                 skb = add_grec(skb, pmc, type, 1, 0);
 659                         }
 660                         pmc->crcount--;
 661                         if (pmc->crcount == 0) {
 662                                 igmpv3_clear_zeros(&pmc->tomb);
 663                                 igmpv3_clear_zeros(&pmc->sources);
 664                         }
 665                 }
 666                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 667                         if (pmc_prev)
 668                                 pmc_prev->next = pmc_next;
 669                         else
 670                                 in_dev->mc_tomb = pmc_next;
 671                         in_dev_put(pmc->interface);
 672                         kfree(pmc);
 673                 } else
 674                         pmc_prev = pmc;
 675         }
 676         spin_unlock_bh(&in_dev->mc_tomb_lock);
 677
 678         /* change recs */
 679         for_each_pmc_rcu(in_dev, pmc) {
 680                 spin_lock_bh(&pmc->lock);
 681                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 682                         type = IGMPV3_BLOCK_OLD_SOURCES;
 683                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 684                 } else {
 685                         type = IGMPV3_ALLOW_NEW_SOURCES;
 686                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 687                 }
 688                 skb = add_grec(skb, pmc, type, 0, 0);
 689                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 690
 691                 /* filter mode changes */
 692                 if (pmc->crcount) {
 693                         if (pmc->sfmode == MCAST_EXCLUDE)
 694                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 695                         else
 696                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 697                         skb = add_grec(skb, pmc, type, 0, 0);
 698                         pmc->crcount--;
 699                 }
 700                 spin_unlock_bh(&pmc->lock);
 701         }
 702         rcu_read_unlock();
 703
 704         if (!skb)
 705                 return;
 706         (void) igmpv3_sendpack(skb);
 707 }
 708
 709 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 710         int type)
 711 {
 712         struct sk_buff *skb;
 713         struct iphdr *iph;
 714         struct igmphdr *ih;
 715         struct rtable *rt;
 716         struct net_device *dev = in_dev->dev;
 717         struct net *net = dev_net(dev);
 718         __be32  group = pmc ? pmc->multiaddr : 0;
 719         struct flowi4 fl4;
 720         __be32  dst;
 721         int hlen, tlen;
 722
 723         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 724                 return igmpv3_send_report(in_dev, pmc);
 725
 726         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 727                 return 0;
 728
 729         if (type == IGMP_HOST_LEAVE_MESSAGE)
 730                 dst = IGMP_ALL_ROUTER;
 731         else
 732                 dst = group;
 733
 734         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 735                                    0, 0,
 736                                    IPPROTO_IGMP, 0, dev->ifindex);
 737         if (IS_ERR(rt))
 738                 return -1;
 739
 740         hlen = LL_RESERVED_SPACE(dev);
 741         tlen = dev->needed_tailroom;
 742         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 743         if (!skb) {
 744                 ip_rt_put(rt);
 745                 return -1;
 746         }
 747         skb->priority = TC_PRIO_CONTROL;
 748
 749         skb_dst_set(skb, &rt->dst);
 750
 751         skb_reserve(skb, hlen);
 752
 753         skb_reset_network_header(skb);
 754         iph = ip_hdr(skb);
 755         skb_put(skb, sizeof(struct iphdr) + 4);
 756
 757         iph->version  = 4;
 758         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 759         iph->tos      = 0xc0;
 760         iph->frag_off = htons(IP_DF);
 761         iph->ttl      = 1;
 762         iph->daddr    = dst;
 763         iph->saddr    = fl4.saddr;
 764         iph->protocol = IPPROTO_IGMP;
 765         ip_select_ident(net, skb, NULL);
 766         ((u8 *)&iph[1])[0] = IPOPT_RA;
 767         ((u8 *)&iph[1])[1] = 4;
 768         ((u8 *)&iph[1])[2] = 0;
 769         ((u8 *)&iph[1])[3] = 0;
 770
 771         ih = skb_put(skb, sizeof(struct igmphdr));
 772         ih->type = type;
 773         ih->code = 0;
 774         ih->csum = 0;
 775         ih->group = group;
 776         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 777
 778         return ip_local_out(net, skb->sk, skb);
 779 }
 780
 781 static void igmp_gq_timer_expire(struct timer_list *t)
 782 {
 783         struct in_device *in_dev = from_timer(in_dev, t, mr_gq_timer);
 784
 785         in_dev->mr_gq_running = 0;
 786         igmpv3_send_report(in_dev, NULL);
 787         in_dev_put(in_dev);
 788 }
 789
 790 static void igmp_ifc_timer_expire(struct timer_list *t)
 791 {
 792         struct in_device *in_dev = from_timer(in_dev, t, mr_ifc_timer);
 793
 794         igmpv3_send_cr(in_dev);
 795         if (in_dev->mr_ifc_count) {
 796                 in_dev->mr_ifc_count--;
 797                 igmp_ifc_start_timer(in_dev,
 798                                      unsolicited_report_interval(in_dev));
 799         }
 800         in_dev_put(in_dev);
 801 }
 802
 803 static void igmp_ifc_event(struct in_device *in_dev)
 804 {
 805         struct net *net = dev_net(in_dev->dev);
 806         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 807                 return;
 808         in_dev->mr_ifc_count = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
 809         igmp_ifc_start_timer(in_dev, 1);
 810 }
 811
 812
 813 static void igmp_timer_expire(struct timer_list *t)
 814 {
 815         struct ip_mc_list *im = from_timer(im, t, timer);
 816         struct in_device *in_dev = im->interface;
 817
 818         spin_lock(&im->lock);
 819         im->tm_running = 0;
 820
 821         if (im->unsolicit_count && --im->unsolicit_count)
 822                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 823
 824         im->reporter = 1;
 825         spin_unlock(&im->lock);
 826
 827         if (IGMP_V1_SEEN(in_dev))
 828                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 829         else if (IGMP_V2_SEEN(in_dev))
 830                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 831         else
 832                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 833
 834         ip_ma_put(im);
 835 }
 836
 837 /* mark EXCLUDE-mode sources */
 838 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 839 {
 840         struct ip_sf_list *psf;
 841         int i, scount;
 842
 843         scount = 0;
 844         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 845                 if (scount == nsrcs)
 846                         break;
 847                 for (i = 0; i < nsrcs; i++) {
 848                         /* skip inactive filters */
 849                         if (psf->sf_count[MCAST_INCLUDE] ||
 850                             pmc->sfcount[MCAST_EXCLUDE] !=
 851                             psf->sf_count[MCAST_EXCLUDE])
 852                                 break;
 853                         if (srcs[i] == psf->sf_inaddr) {
 854                                 scount++;
 855                                 break;
 856                         }
 857                 }
 858         }
 859         pmc->gsquery = 0;
 860         if (scount == nsrcs)    /* all sources excluded */
 861                 return 0;
 862         return 1;
 863 }
 864
 865 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 866 {
 867         struct ip_sf_list *psf;
 868         int i, scount;
 869
 870         if (pmc->sfmode == MCAST_EXCLUDE)
 871                 return igmp_xmarksources(pmc, nsrcs, srcs);
 872
 873         /* mark INCLUDE-mode sources */
 874         scount = 0;
 875         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 876                 if (scount == nsrcs)
 877                         break;
 878                 for (i = 0; i < nsrcs; i++)
 879                         if (srcs[i] == psf->sf_inaddr) {
 880                                 psf->sf_gsresp = 1;
 881                                 scount++;
 882                                 break;
 883                         }
 884         }
 885         if (!scount) {
 886                 pmc->gsquery = 0;
 887                 return 0;
 888         }
 889         pmc->gsquery = 1;
 890         return 1;
 891 }
 892
 893 /* return true if packet was dropped */
 894 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 895 {
 896         struct ip_mc_list *im;
 897         struct net *net = dev_net(in_dev->dev);
 898
 899         /* Timers are only set for non-local groups */
 900
 901         if (group == IGMP_ALL_HOSTS)
 902                 return false;
 903         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 904                 return false;
 905
 906         rcu_read_lock();
 907         for_each_pmc_rcu(in_dev, im) {
 908                 if (im->multiaddr == group) {
 909                         igmp_stop_timer(im);
 910                         break;
 911                 }
 912         }
 913         rcu_read_unlock();
 914         return false;
 915 }
 916
 917 /* return true if packet was dropped */
 918 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 919         int len)
 920 {
 921         struct igmphdr          *ih = igmp_hdr(skb);
 922         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 923         struct ip_mc_list       *im;
 924         __be32                  group = ih->group;
 925         int                     max_delay;
 926         int                     mark = 0;
 927         struct net              *net = dev_net(in_dev->dev);
 928
 929
 930         if (len == 8) {
 931                 if (ih->code == 0) {
 932                         /* Alas, old v1 router presents here. */
 933
 934                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 935                         in_dev->mr_v1_seen = jiffies +
 936                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 937                                 in_dev->mr_qri;
 938                         group = 0;
 939                 } else {
 940                         /* v2 router present */
 941                         max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
 942                         in_dev->mr_v2_seen = jiffies +
 943                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 944                                 in_dev->mr_qri;
 945                 }
 946                 /* cancel the interface change timer */
 947                 in_dev->mr_ifc_count = 0;
 948                 if (del_timer(&in_dev->mr_ifc_timer))
 949                         __in_dev_put(in_dev);
 950                 /* clear deleted report items */
 951                 igmpv3_clear_delrec(in_dev);
 952         } else if (len < 12) {
 953                 return true;    /* ignore bogus packet; freed by caller */
 954         } else if (IGMP_V1_SEEN(in_dev)) {
 955                 /* This is a v3 query with v1 queriers present */
 956                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 957                 group = 0;
 958         } else if (IGMP_V2_SEEN(in_dev)) {
 959                 /* this is a v3 query with v2 queriers present;
 960                  * Interpretation of the max_delay code is problematic here.
 961                  * A real v2 host would use ih_code directly, while v3 has a
 962                  * different encoding. We use the v3 encoding as more likely
 963                  * to be intended in a v3 query.
 964                  */
 965                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 966                 if (!max_delay)
 967                         max_delay = 1;  /* can't mod w/ 0 */
 968         } else { /* v3 */
 969                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 970                         return true;
 971
 972                 ih3 = igmpv3_query_hdr(skb);
 973                 if (ih3->nsrcs) {
 974                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
 975                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
 976                                 return true;
 977                         ih3 = igmpv3_query_hdr(skb);
 978                 }
 979
 980                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 981                 if (!max_delay)
 982                         max_delay = 1;  /* can't mod w/ 0 */
 983                 in_dev->mr_maxdelay = max_delay;
 984
 985                 /* RFC3376, 4.1.6. QRV and 4.1.7. QQIC, when the most recently
 986                  * received value was zero, use the default or statically
 987                  * configured value.
 988                  */
 989                 in_dev->mr_qrv = ih3->qrv ?: net->ipv4.sysctl_igmp_qrv;
 990                 in_dev->mr_qi = IGMPV3_QQIC(ih3->qqic)*HZ ?: IGMP_QUERY_INTERVAL;
 991
 992                 /* RFC3376, 8.3. Query Response Interval:
 993                  * The number of seconds represented by the [Query Response
 994                  * Interval] must be less than the [Query Interval].
 995                  */
 996                 if (in_dev->mr_qri >= in_dev->mr_qi)
 997                         in_dev->mr_qri = (in_dev->mr_qi/HZ - 1)*HZ;
 998
 999                 if (!group) { /* general query */
1000                         if (ih3->nsrcs)
1001                                 return true;    /* no sources allowed */
1002                         igmp_gq_start_timer(in_dev);
1003                         return false;
1004                 }
1005                 /* mark sources to include, if group & source-specific */
1006                 mark = ih3->nsrcs != 0;
1007         }
1008
1009         /*
1010          * - Start the timers in all of our membership records
1011          *   that the query applies to for the interface on
1012          *   which the query arrived excl. those that belong
1013          *   to a "local" group (224.0.0.X)
1014          * - For timers already running check if they need to
1015          *   be reset.
1016          * - Use the igmp->igmp_code field as the maximum
1017          *   delay possible
1018          */
1019         rcu_read_lock();
1020         for_each_pmc_rcu(in_dev, im) {
1021                 int changed;
1022
1023                 if (group && group != im->multiaddr)
1024                         continue;
1025                 if (im->multiaddr == IGMP_ALL_HOSTS)
1026                         continue;
1027                 if (ipv4_is_local_multicast(im->multiaddr) &&
1028                     !net->ipv4.sysctl_igmp_llm_reports)
1029                         continue;
1030                 spin_lock_bh(&im->lock);
1031                 if (im->tm_running)
1032                         im->gsquery = im->gsquery && mark;
1033                 else
1034                         im->gsquery = mark;
1035                 changed = !im->gsquery ||
1036                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1037                 spin_unlock_bh(&im->lock);
1038                 if (changed)
1039                         igmp_mod_timer(im, max_delay);
1040         }
1041         rcu_read_unlock();
1042         return false;
1043 }
1044
1045 /* called in rcu_read_lock() section */
1046 int igmp_rcv(struct sk_buff *skb)
1047 {
1048         /* This basically follows the spec line by line -- see RFC1112 */
1049         struct igmphdr *ih;
1050         struct net_device *dev = skb->dev;
1051         struct in_device *in_dev;
1052         int len = skb->len;
1053         bool dropped = true;
1054
1055         if (netif_is_l3_master(dev)) {
1056                 dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1057                 if (!dev)
1058                         goto drop;
1059         }
1060
1061         in_dev = __in_dev_get_rcu(dev);
1062         if (!in_dev)
1063                 goto drop;
1064
1065         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1066                 goto drop;
1067
1068         if (skb_checksum_simple_validate(skb))
1069                 goto drop;
1070
1071         ih = igmp_hdr(skb);
1072         switch (ih->type) {
1073         case IGMP_HOST_MEMBERSHIP_QUERY:
1074                 dropped = igmp_heard_query(in_dev, skb, len);
1075                 break;
1076         case IGMP_HOST_MEMBERSHIP_REPORT:
1077         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1078                 /* Is it our report looped back? */
1079                 if (rt_is_output_route(skb_rtable(skb)))
1080                         break;
1081                 /* don't rely on MC router hearing unicast reports */
1082                 if (skb->pkt_type == PACKET_MULTICAST ||
1083                     skb->pkt_type == PACKET_BROADCAST)
1084                         dropped = igmp_heard_report(in_dev, ih->group);
1085                 break;
1086         case IGMP_PIM:
1087 #ifdef CONFIG_IP_PIMSM_V1
1088                 return pim_rcv_v1(skb);
1089 #endif
1090         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1091         case IGMP_DVMRP:
1092         case IGMP_TRACE:
1093         case IGMP_HOST_LEAVE_MESSAGE:
1094         case IGMP_MTRACE:
1095         case IGMP_MTRACE_RESP:
1096                 break;
1097         default:
1098                 break;
1099         }
1100
1101 drop:
1102         if (dropped)
1103                 kfree_skb(skb);
1104         else
1105                 consume_skb(skb);
1106         return 0;
1107 }
1108
1109 #endif
1110
1111
1112 /*
1113  *      Add a filter to a device
1114  */
1115
1116 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1117 {
1118         char buf[MAX_ADDR_LEN];
1119         struct net_device *dev = in_dev->dev;
1120
1121         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1122            We will get multicast token leakage, when IFF_MULTICAST
1123            is changed. This check should be done in ndo_set_rx_mode
1124            routine. Something sort of:
1125            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1126            --ANK
1127            */
1128         if (arp_mc_map(addr, buf, dev, 0) == 0)
1129                 dev_mc_add(dev, buf);
1130 }
1131
1132 /*
1133  *      Remove a filter from a device
1134  */
1135
1136 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1137 {
1138         char buf[MAX_ADDR_LEN];
1139         struct net_device *dev = in_dev->dev;
1140
1141         if (arp_mc_map(addr, buf, dev, 0) == 0)
1142                 dev_mc_del(dev, buf);
1143 }
1144
1145 #ifdef CONFIG_IP_MULTICAST
1146 /*
1147  * deleted ip_mc_list manipulation
1148  */
1149 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
1150                               gfp_t gfp)
1151 {
1152         struct ip_mc_list *pmc;
1153         struct net *net = dev_net(in_dev->dev);
1154
1155         /* this is an "ip_mc_list" for convenience; only the fields below
1156          * are actually used. In particular, the refcnt and users are not
1157          * used for management of the delete list. Using the same structure
1158          * for deleted items allows change reports to use common code with
1159          * non-deleted or query-response MCA's.
1160          */
1161         pmc = kzalloc(sizeof(*pmc), gfp);
1162         if (!pmc)
1163                 return;
1164         spin_lock_init(&pmc->lock);
1165         spin_lock_bh(&im->lock);
1166         pmc->interface = im->interface;
1167         in_dev_hold(in_dev);
1168         pmc->multiaddr = im->multiaddr;
1169         pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1170         pmc->sfmode = im->sfmode;
1171         if (pmc->sfmode == MCAST_INCLUDE) {
1172                 struct ip_sf_list *psf;
1173
1174                 pmc->tomb = im->tomb;
1175                 pmc->sources = im->sources;
1176                 im->tomb = im->sources = NULL;
1177                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1178                         psf->sf_crcount = pmc->crcount;
1179         }
1180         spin_unlock_bh(&im->lock);
1181
1182         spin_lock_bh(&in_dev->mc_tomb_lock);
1183         pmc->next = in_dev->mc_tomb;
1184         in_dev->mc_tomb = pmc;
1185         spin_unlock_bh(&in_dev->mc_tomb_lock);
1186 }
1187
1188 /*
1189  * restore ip_mc_list deleted records
1190  */
1191 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1192 {
1193         struct ip_mc_list *pmc, *pmc_prev;
1194         struct ip_sf_list *psf;
1195         struct net *net = dev_net(in_dev->dev);
1196         __be32 multiaddr = im->multiaddr;
1197
1198         spin_lock_bh(&in_dev->mc_tomb_lock);
1199         pmc_prev = NULL;
1200         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1201                 if (pmc->multiaddr == multiaddr)
1202                         break;
1203                 pmc_prev = pmc;
1204         }
1205         if (pmc) {
1206                 if (pmc_prev)
1207                         pmc_prev->next = pmc->next;
1208                 else
1209                         in_dev->mc_tomb = pmc->next;
1210         }
1211         spin_unlock_bh(&in_dev->mc_tomb_lock);
1212
1213         spin_lock_bh(&im->lock);
1214         if (pmc) {
1215                 im->interface = pmc->interface;
1216                 if (im->sfmode == MCAST_INCLUDE) {
1217                         im->tomb = pmc->tomb;
1218                         im->sources = pmc->sources;
1219                         for (psf = im->sources; psf; psf = psf->sf_next)
1220                                 psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1221                 } else {
1222                         im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1223                 }
1224                 in_dev_put(pmc->interface);
1225                 kfree(pmc);
1226         }
1227         spin_unlock_bh(&im->lock);
1228 }
1229
1230 /*
1231  * flush ip_mc_list deleted records
1232  */
1233 static void igmpv3_clear_delrec(struct in_device *in_dev)
1234 {
1235         struct ip_mc_list *pmc, *nextpmc;
1236
1237         spin_lock_bh(&in_dev->mc_tomb_lock);
1238         pmc = in_dev->mc_tomb;
1239         in_dev->mc_tomb = NULL;
1240         spin_unlock_bh(&in_dev->mc_tomb_lock);
1241
1242         for (; pmc; pmc = nextpmc) {
1243                 nextpmc = pmc->next;
1244                 ip_mc_clear_src(pmc);
1245                 in_dev_put(pmc->interface);
1246                 kfree(pmc);
1247         }
1248         /* clear dead sources, too */
1249         rcu_read_lock();
1250         for_each_pmc_rcu(in_dev, pmc) {
1251                 struct ip_sf_list *psf, *psf_next;
1252
1253                 spin_lock_bh(&pmc->lock);
1254                 psf = pmc->tomb;
1255                 pmc->tomb = NULL;
1256                 spin_unlock_bh(&pmc->lock);
1257                 for (; psf; psf = psf_next) {
1258                         psf_next = psf->sf_next;
1259                         kfree(psf);
1260                 }
1261         }
1262         rcu_read_unlock();
1263 }
1264 #endif
1265
1266 static void __igmp_group_dropped(struct ip_mc_list *im, gfp_t gfp)
1267 {
1268         struct in_device *in_dev = im->interface;
1269 #ifdef CONFIG_IP_MULTICAST
1270         struct net *net = dev_net(in_dev->dev);
1271         int reporter;
1272 #endif
1273
1274         if (im->loaded) {
1275                 im->loaded = 0;
1276                 ip_mc_filter_del(in_dev, im->multiaddr);
1277         }
1278
1279 #ifdef CONFIG_IP_MULTICAST
1280         if (im->multiaddr == IGMP_ALL_HOSTS)
1281                 return;
1282         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1283                 return;
1284
1285         reporter = im->reporter;
1286         igmp_stop_timer(im);
1287
1288         if (!in_dev->dead) {
1289                 if (IGMP_V1_SEEN(in_dev))
1290                         return;
1291                 if (IGMP_V2_SEEN(in_dev)) {
1292                         if (reporter)
1293                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1294                         return;
1295                 }
1296                 /* IGMPv3 */
1297                 igmpv3_add_delrec(in_dev, im, gfp);
1298
1299                 igmp_ifc_event(in_dev);
1300         }
1301 #endif
1302 }
1303
1304 static void igmp_group_dropped(struct ip_mc_list *im)
1305 {
1306         __igmp_group_dropped(im, GFP_KERNEL);
1307 }
1308
1309 static void igmp_group_added(struct ip_mc_list *im)
1310 {
1311         struct in_device *in_dev = im->interface;
1312 #ifdef CONFIG_IP_MULTICAST
1313         struct net *net = dev_net(in_dev->dev);
1314 #endif
1315
1316         if (im->loaded == 0) {
1317                 im->loaded = 1;
1318                 ip_mc_filter_add(in_dev, im->multiaddr);
1319         }
1320
1321 #ifdef CONFIG_IP_MULTICAST
1322         if (im->multiaddr == IGMP_ALL_HOSTS)
1323                 return;
1324         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1325                 return;
1326
1327         if (in_dev->dead)
1328                 return;
1329
1330         im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1331         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1332                 spin_lock_bh(&im->lock);
1333                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1334                 spin_unlock_bh(&im->lock);
1335                 return;
1336         }
1337         /* else, v3 */
1338
1339         /* Based on RFC3376 5.1, for newly added INCLUDE SSM, we should
1340          * not send filter-mode change record as the mode should be from
1341          * IN() to IN(A).
1342          */
1343         if (im->sfmode == MCAST_EXCLUDE)
1344                 im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1345
1346         igmp_ifc_event(in_dev);
1347 #endif
1348 }
1349
1350
1351 /*
1352  *      Multicast list managers
1353  */
1354
1355 static u32 ip_mc_hash(const struct ip_mc_list *im)
1356 {
1357         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1358 }
1359
1360 static void ip_mc_hash_add(struct in_device *in_dev,
1361                            struct ip_mc_list *im)
1362 {
1363         struct ip_mc_list __rcu **mc_hash;
1364         u32 hash;
1365
1366         mc_hash = rtnl_dereference(in_dev->mc_hash);
1367         if (mc_hash) {
1368                 hash = ip_mc_hash(im);
1369                 im->next_hash = mc_hash[hash];
1370                 rcu_assign_pointer(mc_hash[hash], im);
1371                 return;
1372         }
1373
1374         /* do not use a hash table for small number of items */
1375         if (in_dev->mc_count < 4)
1376                 return;
1377
1378         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1379                           GFP_KERNEL);
1380         if (!mc_hash)
1381                 return;
1382
1383         for_each_pmc_rtnl(in_dev, im) {
1384                 hash = ip_mc_hash(im);
1385                 im->next_hash = mc_hash[hash];
1386                 RCU_INIT_POINTER(mc_hash[hash], im);
1387         }
1388
1389         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1390 }
1391
1392 static void ip_mc_hash_remove(struct in_device *in_dev,
1393                               struct ip_mc_list *im)
1394 {
1395         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1396         struct ip_mc_list *aux;
1397
1398         if (!mc_hash)
1399                 return;
1400         mc_hash += ip_mc_hash(im);
1401         while ((aux = rtnl_dereference(*mc_hash)) != im)
1402                 mc_hash = &aux->next_hash;
1403         *mc_hash = im->next_hash;
1404 }
1405
1406
1407 /*
1408  *      A socket has joined a multicast group on device dev.
1409  */
1410 static void ____ip_mc_inc_group(struct in_device *in_dev, __be32 addr,
1411                                 unsigned int mode, gfp_t gfp)
1412 {
1413         struct ip_mc_list *im;
1414
1415         ASSERT_RTNL();
1416
1417         for_each_pmc_rtnl(in_dev, im) {
1418                 if (im->multiaddr == addr) {
1419                         im->users++;
1420                         ip_mc_add_src(in_dev, &addr, mode, 0, NULL, 0);
1421                         goto out;
1422                 }
1423         }
1424
1425         im = kzalloc(sizeof(*im), gfp);
1426         if (!im)
1427                 goto out;
1428
1429         im->users = 1;
1430         im->interface = in_dev;
1431         in_dev_hold(in_dev);
1432         im->multiaddr = addr;
1433         /* initial mode is (EX, empty) */
1434         im->sfmode = mode;
1435         im->sfcount[mode] = 1;
1436         refcount_set(&im->refcnt, 1);
1437         spin_lock_init(&im->lock);
1438 #ifdef CONFIG_IP_MULTICAST
1439         timer_setup(&im->timer, igmp_timer_expire, 0);
1440 #endif
1441
1442         im->next_rcu = in_dev->mc_list;
1443         in_dev->mc_count++;
1444         rcu_assign_pointer(in_dev->mc_list, im);
1445
1446         ip_mc_hash_add(in_dev, im);
1447
1448 #ifdef CONFIG_IP_MULTICAST
1449         igmpv3_del_delrec(in_dev, im);
1450 #endif
1451         igmp_group_added(im);
1452         if (!in_dev->dead)
1453                 ip_rt_multicast_event(in_dev);
1454 out:
1455         return;
1456 }
1457
1458 void __ip_mc_inc_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1459 {
1460         ____ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE, gfp);
1461 }
1462 EXPORT_SYMBOL(__ip_mc_inc_group);
1463
1464 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1465 {
1466         __ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE);
1467 }
1468 EXPORT_SYMBOL(ip_mc_inc_group);
1469
1470 static int ip_mc_check_iphdr(struct sk_buff *skb)
1471 {
1472         const struct iphdr *iph;
1473         unsigned int len;
1474         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1475
1476         if (!pskb_may_pull(skb, offset))
1477                 return -EINVAL;
1478
1479         iph = ip_hdr(skb);
1480
1481         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1482                 return -EINVAL;
1483
1484         offset += ip_hdrlen(skb) - sizeof(*iph);
1485
1486         if (!pskb_may_pull(skb, offset))
1487                 return -EINVAL;
1488
1489         iph = ip_hdr(skb);
1490
1491         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1492                 return -EINVAL;
1493
1494         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1495         if (skb->len < len || len < offset)
1496                 return -EINVAL;
1497
1498         skb_set_transport_header(skb, offset);
1499
1500         return 0;
1501 }
1502
1503 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1504 {
1505         unsigned int len = skb_transport_offset(skb);
1506
1507         len += sizeof(struct igmpv3_report);
1508
1509         return ip_mc_may_pull(skb, len) ? 0 : -EINVAL;
1510 }
1511
1512 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1513 {
1514         unsigned int transport_len = ip_transport_len(skb);
1515         unsigned int len;
1516
1517         /* IGMPv{1,2}? */
1518         if (transport_len != sizeof(struct igmphdr)) {
1519                 /* or IGMPv3? */
1520                 if (transport_len < sizeof(struct igmpv3_query))
1521                         return -EINVAL;
1522
1523                 len = skb_transport_offset(skb) + sizeof(struct igmpv3_query);
1524                 if (!ip_mc_may_pull(skb, len))
1525                         return -EINVAL;
1526         }
1527
1528         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1529          * all-systems destination addresses (224.0.0.1) for general queries
1530          */
1531         if (!igmp_hdr(skb)->group &&
1532             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1533                 return -EINVAL;
1534
1535         return 0;
1536 }
1537
1538 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1539 {
1540         switch (igmp_hdr(skb)->type) {
1541         case IGMP_HOST_LEAVE_MESSAGE:
1542         case IGMP_HOST_MEMBERSHIP_REPORT:
1543         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1544                 return 0;
1545         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1546                 return ip_mc_check_igmp_reportv3(skb);
1547         case IGMP_HOST_MEMBERSHIP_QUERY:
1548                 return ip_mc_check_igmp_query(skb);
1549         default:
1550                 return -ENOMSG;
1551         }
1552 }
1553
1554 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1555 {
1556         return skb_checksum_simple_validate(skb);
1557 }
1558
1559 static int ip_mc_check_igmp_csum(struct sk_buff *skb)
1560 {
1561         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1562         unsigned int transport_len = ip_transport_len(skb);
1563         struct sk_buff *skb_chk;
1564
1565         if (!ip_mc_may_pull(skb, len))
1566                 return -EINVAL;
1567
1568         skb_chk = skb_checksum_trimmed(skb, transport_len,
1569                                        ip_mc_validate_checksum);
1570         if (!skb_chk)
1571                 return -EINVAL;
1572
1573         if (skb_chk != skb)
1574                 kfree_skb(skb_chk);
1575
1576         return 0;
1577 }
1578
1579 /**
1580  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1581  * @skb: the skb to validate
1582  *
1583  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1584  * skb transport header accordingly and returns zero.
1585  *
1586  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1587  *  standard
1588  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1589  * -ENOMEM: A memory allocation failure happened.
1590  *
1591  * Caller needs to set the skb network header and free any returned skb if it
1592  * differs from the provided skb.
1593  */
1594 int ip_mc_check_igmp(struct sk_buff *skb)
1595 {
1596         int ret = ip_mc_check_iphdr(skb);
1597
1598         if (ret < 0)
1599                 return ret;
1600
1601         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1602                 return -ENOMSG;
1603
1604         ret = ip_mc_check_igmp_csum(skb);
1605         if (ret < 0)
1606                 return ret;
1607
1608         return ip_mc_check_igmp_msg(skb);
1609 }
1610 EXPORT_SYMBOL(ip_mc_check_igmp);
1611
1612 /*
1613  *      Resend IGMP JOIN report; used by netdev notifier.
1614  */
1615 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1616 {
1617 #ifdef CONFIG_IP_MULTICAST
1618         struct ip_mc_list *im;
1619         int type;
1620         struct net *net = dev_net(in_dev->dev);
1621
1622         ASSERT_RTNL();
1623
1624         for_each_pmc_rtnl(in_dev, im) {
1625                 if (im->multiaddr == IGMP_ALL_HOSTS)
1626                         continue;
1627                 if (ipv4_is_local_multicast(im->multiaddr) &&
1628                     !net->ipv4.sysctl_igmp_llm_reports)
1629                         continue;
1630
1631                 /* a failover is happening and switches
1632                  * must be notified immediately
1633                  */
1634                 if (IGMP_V1_SEEN(in_dev))
1635                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1636                 else if (IGMP_V2_SEEN(in_dev))
1637                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1638                 else
1639                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1640                 igmp_send_report(in_dev, im, type);
1641         }
1642 #endif
1643 }
1644
1645 /*
1646  *      A socket has left a multicast group on device dev
1647  */
1648
1649 void __ip_mc_dec_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1650 {
1651         struct ip_mc_list *i;
1652         struct ip_mc_list __rcu **ip;
1653
1654         ASSERT_RTNL();
1655
1656         for (ip = &in_dev->mc_list;
1657              (i = rtnl_dereference(*ip)) != NULL;
1658              ip = &i->next_rcu) {
1659                 if (i->multiaddr == addr) {
1660                         if (--i->users == 0) {
1661                                 ip_mc_hash_remove(in_dev, i);
1662                                 *ip = i->next_rcu;
1663                                 in_dev->mc_count--;
1664                                 __igmp_group_dropped(i, gfp);
1665                                 ip_mc_clear_src(i);
1666
1667                                 if (!in_dev->dead)
1668                                         ip_rt_multicast_event(in_dev);
1669
1670                                 ip_ma_put(i);
1671                                 return;
1672                         }
1673                         break;
1674                 }
1675         }
1676 }
1677 EXPORT_SYMBOL(__ip_mc_dec_group);
1678
1679 /* Device changing type */
1680
1681 void ip_mc_unmap(struct in_device *in_dev)
1682 {
1683         struct ip_mc_list *pmc;
1684
1685         ASSERT_RTNL();
1686
1687         for_each_pmc_rtnl(in_dev, pmc)
1688                 igmp_group_dropped(pmc);
1689 }
1690
1691 void ip_mc_remap(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 #ifdef CONFIG_IP_MULTICAST
1699                 igmpv3_del_delrec(in_dev, pmc);
1700 #endif
1701                 igmp_group_added(pmc);
1702         }
1703 }
1704
1705 /* Device going down */
1706
1707 void ip_mc_down(struct in_device *in_dev)
1708 {
1709         struct ip_mc_list *pmc;
1710
1711         ASSERT_RTNL();
1712
1713         for_each_pmc_rtnl(in_dev, pmc)
1714                 igmp_group_dropped(pmc);
1715
1716 #ifdef CONFIG_IP_MULTICAST
1717         in_dev->mr_ifc_count = 0;
1718         if (del_timer(&in_dev->mr_ifc_timer))
1719                 __in_dev_put(in_dev);
1720         in_dev->mr_gq_running = 0;
1721         if (del_timer(&in_dev->mr_gq_timer))
1722                 __in_dev_put(in_dev);
1723 #endif
1724
1725         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1726 }
1727
1728 #ifdef CONFIG_IP_MULTICAST
1729 static void ip_mc_reset(struct in_device *in_dev)
1730 {
1731         struct net *net = dev_net(in_dev->dev);
1732
1733         in_dev->mr_qi = IGMP_QUERY_INTERVAL;
1734         in_dev->mr_qri = IGMP_QUERY_RESPONSE_INTERVAL;
1735         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1736 }
1737 #else
1738 static void ip_mc_reset(struct in_device *in_dev)
1739 {
1740 }
1741 #endif
1742
1743 void ip_mc_init_dev(struct in_device *in_dev)
1744 {
1745         ASSERT_RTNL();
1746
1747 #ifdef CONFIG_IP_MULTICAST
1748         timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1749         timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1750 #endif
1751         ip_mc_reset(in_dev);
1752
1753         spin_lock_init(&in_dev->mc_tomb_lock);
1754 }
1755
1756 /* Device going up */
1757
1758 void ip_mc_up(struct in_device *in_dev)
1759 {
1760         struct ip_mc_list *pmc;
1761
1762         ASSERT_RTNL();
1763
1764         ip_mc_reset(in_dev);
1765         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1766
1767         for_each_pmc_rtnl(in_dev, pmc) {
1768 #ifdef CONFIG_IP_MULTICAST
1769                 igmpv3_del_delrec(in_dev, pmc);
1770 #endif
1771                 igmp_group_added(pmc);
1772         }
1773 }
1774
1775 /*
1776  *      Device is about to be destroyed: clean up.
1777  */
1778
1779 void ip_mc_destroy_dev(struct in_device *in_dev)
1780 {
1781         struct ip_mc_list *i;
1782
1783         ASSERT_RTNL();
1784
1785         /* Deactivate timers */
1786         ip_mc_down(in_dev);
1787 #ifdef CONFIG_IP_MULTICAST
1788         igmpv3_clear_delrec(in_dev);
1789 #endif
1790
1791         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1792                 in_dev->mc_list = i->next_rcu;
1793                 in_dev->mc_count--;
1794                 ip_ma_put(i);
1795         }
1796 }
1797
1798 /* RTNL is locked */
1799 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1800 {
1801         struct net_device *dev = NULL;
1802         struct in_device *idev = NULL;
1803
1804         if (imr->imr_ifindex) {
1805                 idev = inetdev_by_index(net, imr->imr_ifindex);
1806                 return idev;
1807         }
1808         if (imr->imr_address.s_addr) {
1809                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1810                 if (!dev)
1811                         return NULL;
1812         }
1813
1814         if (!dev) {
1815                 struct rtable *rt = ip_route_output(net,
1816                                                     imr->imr_multiaddr.s_addr,
1817                                                     0, 0, 0);
1818                 if (!IS_ERR(rt)) {
1819                         dev = rt->dst.dev;
1820                         ip_rt_put(rt);
1821                 }
1822         }
1823         if (dev) {
1824                 imr->imr_ifindex = dev->ifindex;
1825                 idev = __in_dev_get_rtnl(dev);
1826         }
1827         return idev;
1828 }
1829
1830 /*
1831  *      Join a socket to a group
1832  */
1833
1834 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1835         __be32 *psfsrc)
1836 {
1837         struct ip_sf_list *psf, *psf_prev;
1838         int rv = 0;
1839
1840         psf_prev = NULL;
1841         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1842                 if (psf->sf_inaddr == *psfsrc)
1843                         break;
1844                 psf_prev = psf;
1845         }
1846         if (!psf || psf->sf_count[sfmode] == 0) {
1847                 /* source filter not found, or count wrong =>  bug */
1848                 return -ESRCH;
1849         }
1850         psf->sf_count[sfmode]--;
1851         if (psf->sf_count[sfmode] == 0) {
1852                 ip_rt_multicast_event(pmc->interface);
1853         }
1854         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1855 #ifdef CONFIG_IP_MULTICAST
1856                 struct in_device *in_dev = pmc->interface;
1857                 struct net *net = dev_net(in_dev->dev);
1858 #endif
1859
1860                 /* no more filters for this source */
1861                 if (psf_prev)
1862                         psf_prev->sf_next = psf->sf_next;
1863                 else
1864                         pmc->sources = psf->sf_next;
1865 #ifdef CONFIG_IP_MULTICAST
1866                 if (psf->sf_oldin &&
1867                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1868                         psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1869                         psf->sf_next = pmc->tomb;
1870                         pmc->tomb = psf;
1871                         rv = 1;
1872                 } else
1873 #endif
1874                         kfree(psf);
1875         }
1876         return rv;
1877 }
1878
1879 #ifndef CONFIG_IP_MULTICAST
1880 #define igmp_ifc_event(x)       do { } while (0)
1881 #endif
1882
1883 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1884                          int sfcount, __be32 *psfsrc, int delta)
1885 {
1886         struct ip_mc_list *pmc;
1887         int     changerec = 0;
1888         int     i, err;
1889
1890         if (!in_dev)
1891                 return -ENODEV;
1892         rcu_read_lock();
1893         for_each_pmc_rcu(in_dev, pmc) {
1894                 if (*pmca == pmc->multiaddr)
1895                         break;
1896         }
1897         if (!pmc) {
1898                 /* MCA not found?? bug */
1899                 rcu_read_unlock();
1900                 return -ESRCH;
1901         }
1902         spin_lock_bh(&pmc->lock);
1903         rcu_read_unlock();
1904 #ifdef CONFIG_IP_MULTICAST
1905         sf_markstate(pmc);
1906 #endif
1907         if (!delta) {
1908                 err = -EINVAL;
1909                 if (!pmc->sfcount[sfmode])
1910                         goto out_unlock;
1911                 pmc->sfcount[sfmode]--;
1912         }
1913         err = 0;
1914         for (i = 0; i < sfcount; i++) {
1915                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1916
1917                 changerec |= rv > 0;
1918                 if (!err && rv < 0)
1919                         err = rv;
1920         }
1921         if (pmc->sfmode == MCAST_EXCLUDE &&
1922             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1923             pmc->sfcount[MCAST_INCLUDE]) {
1924 #ifdef CONFIG_IP_MULTICAST
1925                 struct ip_sf_list *psf;
1926                 struct net *net = dev_net(in_dev->dev);
1927 #endif
1928
1929                 /* filter mode change */
1930                 pmc->sfmode = MCAST_INCLUDE;
1931 #ifdef CONFIG_IP_MULTICAST
1932                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1933                 in_dev->mr_ifc_count = pmc->crcount;
1934                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1935                         psf->sf_crcount = 0;
1936                 igmp_ifc_event(pmc->interface);
1937         } else if (sf_setstate(pmc) || changerec) {
1938                 igmp_ifc_event(pmc->interface);
1939 #endif
1940         }
1941 out_unlock:
1942         spin_unlock_bh(&pmc->lock);
1943         return err;
1944 }
1945
1946 /*
1947  * Add multicast single-source filter to the interface list
1948  */
1949 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1950         __be32 *psfsrc)
1951 {
1952         struct ip_sf_list *psf, *psf_prev;
1953
1954         psf_prev = NULL;
1955         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1956                 if (psf->sf_inaddr == *psfsrc)
1957                         break;
1958                 psf_prev = psf;
1959         }
1960         if (!psf) {
1961                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1962                 if (!psf)
1963                         return -ENOBUFS;
1964                 psf->sf_inaddr = *psfsrc;
1965                 if (psf_prev) {
1966                         psf_prev->sf_next = psf;
1967                 } else
1968                         pmc->sources = psf;
1969         }
1970         psf->sf_count[sfmode]++;
1971         if (psf->sf_count[sfmode] == 1) {
1972                 ip_rt_multicast_event(pmc->interface);
1973         }
1974         return 0;
1975 }
1976
1977 #ifdef CONFIG_IP_MULTICAST
1978 static void sf_markstate(struct ip_mc_list *pmc)
1979 {
1980         struct ip_sf_list *psf;
1981         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1982
1983         for (psf = pmc->sources; psf; psf = psf->sf_next)
1984                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1985                         psf->sf_oldin = mca_xcount ==
1986                                 psf->sf_count[MCAST_EXCLUDE] &&
1987                                 !psf->sf_count[MCAST_INCLUDE];
1988                 } else
1989                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1990 }
1991
1992 static int sf_setstate(struct ip_mc_list *pmc)
1993 {
1994         struct ip_sf_list *psf, *dpsf;
1995         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1996         int qrv = pmc->interface->mr_qrv;
1997         int new_in, rv;
1998
1999         rv = 0;
2000         for (psf = pmc->sources; psf; psf = psf->sf_next) {
2001                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2002                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
2003                                 !psf->sf_count[MCAST_INCLUDE];
2004                 } else
2005                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
2006                 if (new_in) {
2007                         if (!psf->sf_oldin) {
2008                                 struct ip_sf_list *prev = NULL;
2009
2010                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2011                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2012                                                 break;
2013                                         prev = dpsf;
2014                                 }
2015                                 if (dpsf) {
2016                                         if (prev)
2017                                                 prev->sf_next = dpsf->sf_next;
2018                                         else
2019                                                 pmc->tomb = dpsf->sf_next;
2020                                         kfree(dpsf);
2021                                 }
2022                                 psf->sf_crcount = qrv;
2023                                 rv++;
2024                         }
2025                 } else if (psf->sf_oldin) {
2026
2027                         psf->sf_crcount = 0;
2028                         /*
2029                          * add or update "delete" records if an active filter
2030                          * is now inactive
2031                          */
2032                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2033                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2034                                         break;
2035                         if (!dpsf) {
2036                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2037                                 if (!dpsf)
2038                                         continue;
2039                                 *dpsf = *psf;
2040                                 /* pmc->lock held by callers */
2041                                 dpsf->sf_next = pmc->tomb;
2042                                 pmc->tomb = dpsf;
2043                         }
2044                         dpsf->sf_crcount = qrv;
2045                         rv++;
2046                 }
2047         }
2048         return rv;
2049 }
2050 #endif
2051
2052 /*
2053  * Add multicast source filter list to the interface list
2054  */
2055 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2056                          int sfcount, __be32 *psfsrc, int delta)
2057 {
2058         struct ip_mc_list *pmc;
2059         int     isexclude;
2060         int     i, err;
2061
2062         if (!in_dev)
2063                 return -ENODEV;
2064         rcu_read_lock();
2065         for_each_pmc_rcu(in_dev, pmc) {
2066                 if (*pmca == pmc->multiaddr)
2067                         break;
2068         }
2069         if (!pmc) {
2070                 /* MCA not found?? bug */
2071                 rcu_read_unlock();
2072                 return -ESRCH;
2073         }
2074         spin_lock_bh(&pmc->lock);
2075         rcu_read_unlock();
2076
2077 #ifdef CONFIG_IP_MULTICAST
2078         sf_markstate(pmc);
2079 #endif
2080         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2081         if (!delta)
2082                 pmc->sfcount[sfmode]++;
2083         err = 0;
2084         for (i = 0; i < sfcount; i++) {
2085                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2086                 if (err)
2087                         break;
2088         }
2089         if (err) {
2090                 int j;
2091
2092                 if (!delta)
2093                         pmc->sfcount[sfmode]--;
2094                 for (j = 0; j < i; j++)
2095                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2096         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2097 #ifdef CONFIG_IP_MULTICAST
2098                 struct ip_sf_list *psf;
2099                 struct net *net = dev_net(pmc->interface->dev);
2100                 in_dev = pmc->interface;
2101 #endif
2102
2103                 /* filter mode change */
2104                 if (pmc->sfcount[MCAST_EXCLUDE])
2105                         pmc->sfmode = MCAST_EXCLUDE;
2106                 else if (pmc->sfcount[MCAST_INCLUDE])
2107                         pmc->sfmode = MCAST_INCLUDE;
2108 #ifdef CONFIG_IP_MULTICAST
2109                 /* else no filters; keep old mode for reports */
2110
2111                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2112                 in_dev->mr_ifc_count = pmc->crcount;
2113                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2114                         psf->sf_crcount = 0;
2115                 igmp_ifc_event(in_dev);
2116         } else if (sf_setstate(pmc)) {
2117                 igmp_ifc_event(in_dev);
2118 #endif
2119         }
2120         spin_unlock_bh(&pmc->lock);
2121         return err;
2122 }
2123
2124 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2125 {
2126         struct ip_sf_list *psf, *nextpsf, *tomb, *sources;
2127
2128         spin_lock_bh(&pmc->lock);
2129         tomb = pmc->tomb;
2130         pmc->tomb = NULL;
2131         sources = pmc->sources;
2132         pmc->sources = NULL;
2133         pmc->sfmode = MCAST_EXCLUDE;
2134         pmc->sfcount[MCAST_INCLUDE] = 0;
2135         pmc->sfcount[MCAST_EXCLUDE] = 1;
2136         spin_unlock_bh(&pmc->lock);
2137
2138         for (psf = tomb; psf; psf = nextpsf) {
2139                 nextpsf = psf->sf_next;
2140                 kfree(psf);
2141         }
2142         for (psf = sources; psf; psf = nextpsf) {
2143                 nextpsf = psf->sf_next;
2144                 kfree(psf);
2145         }
2146 }
2147
2148 /* Join a multicast group
2149  */
2150 static int __ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr,
2151                               unsigned int mode)
2152 {
2153         __be32 addr = imr->imr_multiaddr.s_addr;
2154         struct ip_mc_socklist *iml, *i;
2155         struct in_device *in_dev;
2156         struct inet_sock *inet = inet_sk(sk);
2157         struct net *net = sock_net(sk);
2158         int ifindex;
2159         int count = 0;
2160         int err;
2161
2162         ASSERT_RTNL();
2163
2164         if (!ipv4_is_multicast(addr))
2165                 return -EINVAL;
2166
2167         in_dev = ip_mc_find_dev(net, imr);
2168
2169         if (!in_dev) {
2170                 err = -ENODEV;
2171                 goto done;
2172         }
2173
2174         err = -EADDRINUSE;
2175         ifindex = imr->imr_ifindex;
2176         for_each_pmc_rtnl(inet, i) {
2177                 if (i->multi.imr_multiaddr.s_addr == addr &&
2178                     i->multi.imr_ifindex == ifindex)
2179                         goto done;
2180                 count++;
2181         }
2182         err = -ENOBUFS;
2183         if (count >= net->ipv4.sysctl_igmp_max_memberships)
2184                 goto done;
2185         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2186         if (!iml)
2187                 goto done;
2188
2189         memcpy(&iml->multi, imr, sizeof(*imr));
2190         iml->next_rcu = inet->mc_list;
2191         iml->sflist = NULL;
2192         iml->sfmode = mode;
2193         rcu_assign_pointer(inet->mc_list, iml);
2194         __ip_mc_inc_group(in_dev, addr, mode);
2195         err = 0;
2196 done:
2197         return err;
2198 }
2199
2200 /* Join ASM (Any-Source Multicast) group
2201  */
2202 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2203 {
2204         return __ip_mc_join_group(sk, imr, MCAST_EXCLUDE);
2205 }
2206 EXPORT_SYMBOL(ip_mc_join_group);
2207
2208 /* Join SSM (Source-Specific Multicast) group
2209  */
2210 int ip_mc_join_group_ssm(struct sock *sk, struct ip_mreqn *imr,
2211                          unsigned int mode)
2212 {
2213         return __ip_mc_join_group(sk, imr, mode);
2214 }
2215
2216 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2217                            struct in_device *in_dev)
2218 {
2219         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2220         int err;
2221
2222         if (!psf) {
2223                 /* any-source empty exclude case */
2224                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2225                         iml->sfmode, 0, NULL, 0);
2226         }
2227         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2228                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2229         RCU_INIT_POINTER(iml->sflist, NULL);
2230         /* decrease mem now to avoid the memleak warning */
2231         atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2232         kfree_rcu(psf, rcu);
2233         return err;
2234 }
2235
2236 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2237 {
2238         struct inet_sock *inet = inet_sk(sk);
2239         struct ip_mc_socklist *iml;
2240         struct ip_mc_socklist __rcu **imlp;
2241         struct in_device *in_dev;
2242         struct net *net = sock_net(sk);
2243         __be32 group = imr->imr_multiaddr.s_addr;
2244         u32 ifindex;
2245         int ret = -EADDRNOTAVAIL;
2246
2247         ASSERT_RTNL();
2248
2249         in_dev = ip_mc_find_dev(net, imr);
2250         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2251                 ret = -ENODEV;
2252                 goto out;
2253         }
2254         ifindex = imr->imr_ifindex;
2255         for (imlp = &inet->mc_list;
2256              (iml = rtnl_dereference(*imlp)) != NULL;
2257              imlp = &iml->next_rcu) {
2258                 if (iml->multi.imr_multiaddr.s_addr != group)
2259                         continue;
2260                 if (ifindex) {
2261                         if (iml->multi.imr_ifindex != ifindex)
2262                                 continue;
2263                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2264                                 iml->multi.imr_address.s_addr)
2265                         continue;
2266
2267                 (void) ip_mc_leave_src(sk, iml, in_dev);
2268
2269                 *imlp = iml->next_rcu;
2270
2271                 if (in_dev)
2272                         ip_mc_dec_group(in_dev, group);
2273
2274                 /* decrease mem now to avoid the memleak warning */
2275                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2276                 kfree_rcu(iml, rcu);
2277                 return 0;
2278         }
2279 out:
2280         return ret;
2281 }
2282 EXPORT_SYMBOL(ip_mc_leave_group);
2283
2284 int ip_mc_source(int add, int omode, struct sock *sk, struct
2285         ip_mreq_source *mreqs, int ifindex)
2286 {
2287         int err;
2288         struct ip_mreqn imr;
2289         __be32 addr = mreqs->imr_multiaddr;
2290         struct ip_mc_socklist *pmc;
2291         struct in_device *in_dev = NULL;
2292         struct inet_sock *inet = inet_sk(sk);
2293         struct ip_sf_socklist *psl;
2294         struct net *net = sock_net(sk);
2295         int leavegroup = 0;
2296         int i, j, rv;
2297
2298         if (!ipv4_is_multicast(addr))
2299                 return -EINVAL;
2300
2301         ASSERT_RTNL();
2302
2303         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2304         imr.imr_address.s_addr = mreqs->imr_interface;
2305         imr.imr_ifindex = ifindex;
2306         in_dev = ip_mc_find_dev(net, &imr);
2307
2308         if (!in_dev) {
2309                 err = -ENODEV;
2310                 goto done;
2311         }
2312         err = -EADDRNOTAVAIL;
2313
2314         for_each_pmc_rtnl(inet, pmc) {
2315                 if ((pmc->multi.imr_multiaddr.s_addr ==
2316                      imr.imr_multiaddr.s_addr) &&
2317                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2318                         break;
2319         }
2320         if (!pmc) {             /* must have a prior join */
2321                 err = -EINVAL;
2322                 goto done;
2323         }
2324         /* if a source filter was set, must be the same mode as before */
2325         if (pmc->sflist) {
2326                 if (pmc->sfmode != omode) {
2327                         err = -EINVAL;
2328                         goto done;
2329                 }
2330         } else if (pmc->sfmode != omode) {
2331                 /* allow mode switches for empty-set filters */
2332                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2333                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2334                         NULL, 0);
2335                 pmc->sfmode = omode;
2336         }
2337
2338         psl = rtnl_dereference(pmc->sflist);
2339         if (!add) {
2340                 if (!psl)
2341                         goto done;      /* err = -EADDRNOTAVAIL */
2342                 rv = !0;
2343                 for (i = 0; i < psl->sl_count; i++) {
2344                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2345                                 sizeof(__be32));
2346                         if (rv == 0)
2347                                 break;
2348                 }
2349                 if (rv)         /* source not found */
2350                         goto done;      /* err = -EADDRNOTAVAIL */
2351
2352                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2353                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2354                         leavegroup = 1;
2355                         goto done;
2356                 }
2357
2358                 /* update the interface filter */
2359                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2360                         &mreqs->imr_sourceaddr, 1);
2361
2362                 for (j = i+1; j < psl->sl_count; j++)
2363                         psl->sl_addr[j-1] = psl->sl_addr[j];
2364                 psl->sl_count--;
2365                 err = 0;
2366                 goto done;
2367         }
2368         /* else, add a new source to the filter */
2369
2370         if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2371                 err = -ENOBUFS;
2372                 goto done;
2373         }
2374         if (!psl || psl->sl_count == psl->sl_max) {
2375                 struct ip_sf_socklist *newpsl;
2376                 int count = IP_SFBLOCK;
2377
2378                 if (psl)
2379                         count += psl->sl_max;
2380                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2381                 if (!newpsl) {
2382                         err = -ENOBUFS;
2383                         goto done;
2384                 }
2385                 newpsl->sl_max = count;
2386                 newpsl->sl_count = count - IP_SFBLOCK;
2387                 if (psl) {
2388                         for (i = 0; i < psl->sl_count; i++)
2389                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2390                         /* decrease mem now to avoid the memleak warning */
2391                         atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2392                         kfree_rcu(psl, rcu);
2393                 }
2394                 rcu_assign_pointer(pmc->sflist, newpsl);
2395                 psl = newpsl;
2396         }
2397         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2398         for (i = 0; i < psl->sl_count; i++) {
2399                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2400                         sizeof(__be32));
2401                 if (rv == 0)
2402                         break;
2403         }
2404         if (rv == 0)            /* address already there is an error */
2405                 goto done;
2406         for (j = psl->sl_count-1; j >= i; j--)
2407                 psl->sl_addr[j+1] = psl->sl_addr[j];
2408         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2409         psl->sl_count++;
2410         err = 0;
2411         /* update the interface list */
2412         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2413                 &mreqs->imr_sourceaddr, 1);
2414 done:
2415         if (leavegroup)
2416                 err = ip_mc_leave_group(sk, &imr);
2417         return err;
2418 }
2419
2420 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2421 {
2422         int err = 0;
2423         struct ip_mreqn imr;
2424         __be32 addr = msf->imsf_multiaddr;
2425         struct ip_mc_socklist *pmc;
2426         struct in_device *in_dev;
2427         struct inet_sock *inet = inet_sk(sk);
2428         struct ip_sf_socklist *newpsl, *psl;
2429         struct net *net = sock_net(sk);
2430         int leavegroup = 0;
2431
2432         if (!ipv4_is_multicast(addr))
2433                 return -EINVAL;
2434         if (msf->imsf_fmode != MCAST_INCLUDE &&
2435             msf->imsf_fmode != MCAST_EXCLUDE)
2436                 return -EINVAL;
2437
2438         ASSERT_RTNL();
2439
2440         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2441         imr.imr_address.s_addr = msf->imsf_interface;
2442         imr.imr_ifindex = ifindex;
2443         in_dev = ip_mc_find_dev(net, &imr);
2444
2445         if (!in_dev) {
2446                 err = -ENODEV;
2447                 goto done;
2448         }
2449
2450         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2451         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2452                 leavegroup = 1;
2453                 goto done;
2454         }
2455
2456         for_each_pmc_rtnl(inet, pmc) {
2457                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2458                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2459                         break;
2460         }
2461         if (!pmc) {             /* must have a prior join */
2462                 err = -EINVAL;
2463                 goto done;
2464         }
2465         if (msf->imsf_numsrc) {
2466                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2467                                                            GFP_KERNEL);
2468                 if (!newpsl) {
2469                         err = -ENOBUFS;
2470                         goto done;
2471                 }
2472                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2473                 memcpy(newpsl->sl_addr, msf->imsf_slist,
2474                         msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2475                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2476                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2477                 if (err) {
2478                         sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2479                         goto done;
2480                 }
2481         } else {
2482                 newpsl = NULL;
2483                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2484                                      msf->imsf_fmode, 0, NULL, 0);
2485         }
2486         psl = rtnl_dereference(pmc->sflist);
2487         if (psl) {
2488                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2489                         psl->sl_count, psl->sl_addr, 0);
2490                 /* decrease mem now to avoid the memleak warning */
2491                 atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2492                 kfree_rcu(psl, rcu);
2493         } else
2494                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2495                         0, NULL, 0);
2496         rcu_assign_pointer(pmc->sflist, newpsl);
2497         pmc->sfmode = msf->imsf_fmode;
2498         err = 0;
2499 done:
2500         if (leavegroup)
2501                 err = ip_mc_leave_group(sk, &imr);
2502         return err;
2503 }
2504
2505 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2506         struct ip_msfilter __user *optval, int __user *optlen)
2507 {
2508         int err, len, count, copycount;
2509         struct ip_mreqn imr;
2510         __be32 addr = msf->imsf_multiaddr;
2511         struct ip_mc_socklist *pmc;
2512         struct in_device *in_dev;
2513         struct inet_sock *inet = inet_sk(sk);
2514         struct ip_sf_socklist *psl;
2515         struct net *net = sock_net(sk);
2516
2517         ASSERT_RTNL();
2518
2519         if (!ipv4_is_multicast(addr))
2520                 return -EINVAL;
2521
2522         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2523         imr.imr_address.s_addr = msf->imsf_interface;
2524         imr.imr_ifindex = 0;
2525         in_dev = ip_mc_find_dev(net, &imr);
2526
2527         if (!in_dev) {
2528                 err = -ENODEV;
2529                 goto done;
2530         }
2531         err = -EADDRNOTAVAIL;
2532
2533         for_each_pmc_rtnl(inet, pmc) {
2534                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2535                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2536                         break;
2537         }
2538         if (!pmc)               /* must have a prior join */
2539                 goto done;
2540         msf->imsf_fmode = pmc->sfmode;
2541         psl = rtnl_dereference(pmc->sflist);
2542         if (!psl) {
2543                 len = 0;
2544                 count = 0;
2545         } else {
2546                 count = psl->sl_count;
2547         }
2548         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2549         len = copycount * sizeof(psl->sl_addr[0]);
2550         msf->imsf_numsrc = count;
2551         if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2552             copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2553                 return -EFAULT;
2554         }
2555         if (len &&
2556             copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2557                 return -EFAULT;
2558         return 0;
2559 done:
2560         return err;
2561 }
2562
2563 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2564         struct group_filter __user *optval, int __user *optlen)
2565 {
2566         int err, i, count, copycount;
2567         struct sockaddr_in *psin;
2568         __be32 addr;
2569         struct ip_mc_socklist *pmc;
2570         struct inet_sock *inet = inet_sk(sk);
2571         struct ip_sf_socklist *psl;
2572
2573         ASSERT_RTNL();
2574
2575         psin = (struct sockaddr_in *)&gsf->gf_group;
2576         if (psin->sin_family != AF_INET)
2577                 return -EINVAL;
2578         addr = psin->sin_addr.s_addr;
2579         if (!ipv4_is_multicast(addr))
2580                 return -EINVAL;
2581
2582         err = -EADDRNOTAVAIL;
2583
2584         for_each_pmc_rtnl(inet, pmc) {
2585                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2586                     pmc->multi.imr_ifindex == gsf->gf_interface)
2587                         break;
2588         }
2589         if (!pmc)               /* must have a prior join */
2590                 goto done;
2591         gsf->gf_fmode = pmc->sfmode;
2592         psl = rtnl_dereference(pmc->sflist);
2593         count = psl ? psl->sl_count : 0;
2594         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2595         gsf->gf_numsrc = count;
2596         if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2597             copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2598                 return -EFAULT;
2599         }
2600         for (i = 0; i < copycount; i++) {
2601                 struct sockaddr_storage ss;
2602
2603                 psin = (struct sockaddr_in *)&ss;
2604                 memset(&ss, 0, sizeof(ss));
2605                 psin->sin_family = AF_INET;
2606                 psin->sin_addr.s_addr = psl->sl_addr[i];
2607                 if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2608                         return -EFAULT;
2609         }
2610         return 0;
2611 done:
2612         return err;
2613 }
2614
2615 /*
2616  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2617  */
2618 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2619                    int dif, int sdif)
2620 {
2621         struct inet_sock *inet = inet_sk(sk);
2622         struct ip_mc_socklist *pmc;
2623         struct ip_sf_socklist *psl;
2624         int i;
2625         int ret;
2626
2627         ret = 1;
2628         if (!ipv4_is_multicast(loc_addr))
2629                 goto out;
2630
2631         rcu_read_lock();
2632         for_each_pmc_rcu(inet, pmc) {
2633                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2634                     (pmc->multi.imr_ifindex == dif ||
2635                      (sdif && pmc->multi.imr_ifindex == sdif)))
2636                         break;
2637         }
2638         ret = inet->mc_all;
2639         if (!pmc)
2640                 goto unlock;
2641         psl = rcu_dereference(pmc->sflist);
2642         ret = (pmc->sfmode == MCAST_EXCLUDE);
2643         if (!psl)
2644                 goto unlock;
2645
2646         for (i = 0; i < psl->sl_count; i++) {
2647                 if (psl->sl_addr[i] == rmt_addr)
2648                         break;
2649         }
2650         ret = 0;
2651         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2652                 goto unlock;
2653         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2654                 goto unlock;
2655         ret = 1;
2656 unlock:
2657         rcu_read_unlock();
2658 out:
2659         return ret;
2660 }
2661
2662 /*
2663  *      A socket is closing.
2664  */
2665
2666 void ip_mc_drop_socket(struct sock *sk)
2667 {
2668         struct inet_sock *inet = inet_sk(sk);
2669         struct ip_mc_socklist *iml;
2670         struct net *net = sock_net(sk);
2671
2672         if (!inet->mc_list)
2673                 return;
2674
2675         rtnl_lock();
2676         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2677                 struct in_device *in_dev;
2678
2679                 inet->mc_list = iml->next_rcu;
2680                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2681                 (void) ip_mc_leave_src(sk, iml, in_dev);
2682                 if (in_dev)
2683                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2684                 /* decrease mem now to avoid the memleak warning */
2685                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2686                 kfree_rcu(iml, rcu);
2687         }
2688         rtnl_unlock();
2689 }
2690
2691 /* called with rcu_read_lock() */
2692 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2693 {
2694         struct ip_mc_list *im;
2695         struct ip_mc_list __rcu **mc_hash;
2696         struct ip_sf_list *psf;
2697         int rv = 0;
2698
2699         mc_hash = rcu_dereference(in_dev->mc_hash);
2700         if (mc_hash) {
2701                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2702
2703                 for (im = rcu_dereference(mc_hash[hash]);
2704                      im != NULL;
2705                      im = rcu_dereference(im->next_hash)) {
2706                         if (im->multiaddr == mc_addr)
2707                                 break;
2708                 }
2709         } else {
2710                 for_each_pmc_rcu(in_dev, im) {
2711                         if (im->multiaddr == mc_addr)
2712                                 break;
2713                 }
2714         }
2715         if (im && proto == IPPROTO_IGMP) {
2716                 rv = 1;
2717         } else if (im) {
2718                 if (src_addr) {
2719                         for (psf = im->sources; psf; psf = psf->sf_next) {
2720                                 if (psf->sf_inaddr == src_addr)
2721                                         break;
2722                         }
2723                         if (psf)
2724                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2725                                         psf->sf_count[MCAST_EXCLUDE] !=
2726                                         im->sfcount[MCAST_EXCLUDE];
2727                         else
2728                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2729                 } else
2730                         rv = 1; /* unspecified source; tentatively allow */
2731         }
2732         return rv;
2733 }
2734
2735 #if defined(CONFIG_PROC_FS)
2736 struct igmp_mc_iter_state {
2737         struct seq_net_private p;
2738         struct net_device *dev;
2739         struct in_device *in_dev;
2740 };
2741
2742 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2743
2744 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2745 {
2746         struct net *net = seq_file_net(seq);
2747         struct ip_mc_list *im = NULL;
2748         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2749
2750         state->in_dev = NULL;
2751         for_each_netdev_rcu(net, state->dev) {
2752                 struct in_device *in_dev;
2753
2754                 in_dev = __in_dev_get_rcu(state->dev);
2755                 if (!in_dev)
2756                         continue;
2757                 im = rcu_dereference(in_dev->mc_list);
2758                 if (im) {
2759                         state->in_dev = in_dev;
2760                         break;
2761                 }
2762         }
2763         return im;
2764 }
2765
2766 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2767 {
2768         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2769
2770         im = rcu_dereference(im->next_rcu);
2771         while (!im) {
2772                 state->dev = next_net_device_rcu(state->dev);
2773                 if (!state->dev) {
2774                         state->in_dev = NULL;
2775                         break;
2776                 }
2777                 state->in_dev = __in_dev_get_rcu(state->dev);
2778                 if (!state->in_dev)
2779                         continue;
2780                 im = rcu_dereference(state->in_dev->mc_list);
2781         }
2782         return im;
2783 }
2784
2785 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2786 {
2787         struct ip_mc_list *im = igmp_mc_get_first(seq);
2788         if (im)
2789                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2790                         --pos;
2791         return pos ? NULL : im;
2792 }
2793
2794 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2795         __acquires(rcu)
2796 {
2797         rcu_read_lock();
2798         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2799 }
2800
2801 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2802 {
2803         struct ip_mc_list *im;
2804         if (v == SEQ_START_TOKEN)
2805                 im = igmp_mc_get_first(seq);
2806         else
2807                 im = igmp_mc_get_next(seq, v);
2808         ++*pos;
2809         return im;
2810 }
2811
2812 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2813         __releases(rcu)
2814 {
2815         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2816
2817         state->in_dev = NULL;
2818         state->dev = NULL;
2819         rcu_read_unlock();
2820 }
2821
2822 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2823 {
2824         if (v == SEQ_START_TOKEN)
2825                 seq_puts(seq,
2826                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2827         else {
2828                 struct ip_mc_list *im = (struct ip_mc_list *)v;
2829                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2830                 char   *querier;
2831                 long delta;
2832
2833 #ifdef CONFIG_IP_MULTICAST
2834                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2835                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2836                           "V3";
2837 #else
2838                 querier = "NONE";
2839 #endif
2840
2841                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2842                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2843                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2844                 }
2845
2846                 delta = im->timer.expires - jiffies;
2847                 seq_printf(seq,
2848                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2849                            im->multiaddr, im->users,
2850                            im->tm_running,
2851                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2852                            im->reporter);
2853         }
2854         return 0;
2855 }
2856
2857 static const struct seq_operations igmp_mc_seq_ops = {
2858         .start  =       igmp_mc_seq_start,
2859         .next   =       igmp_mc_seq_next,
2860         .stop   =       igmp_mc_seq_stop,
2861         .show   =       igmp_mc_seq_show,
2862 };
2863
2864 struct igmp_mcf_iter_state {
2865         struct seq_net_private p;
2866         struct net_device *dev;
2867         struct in_device *idev;
2868         struct ip_mc_list *im;
2869 };
2870
2871 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2872
2873 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2874 {
2875         struct net *net = seq_file_net(seq);
2876         struct ip_sf_list *psf = NULL;
2877         struct ip_mc_list *im = NULL;
2878         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2879
2880         state->idev = NULL;
2881         state->im = NULL;
2882         for_each_netdev_rcu(net, state->dev) {
2883                 struct in_device *idev;
2884                 idev = __in_dev_get_rcu(state->dev);
2885                 if (unlikely(!idev))
2886                         continue;
2887                 im = rcu_dereference(idev->mc_list);
2888                 if (likely(im)) {
2889                         spin_lock_bh(&im->lock);
2890                         psf = im->sources;
2891                         if (likely(psf)) {
2892                                 state->im = im;
2893                                 state->idev = idev;
2894                                 break;
2895                         }
2896                         spin_unlock_bh(&im->lock);
2897                 }
2898         }
2899         return psf;
2900 }
2901
2902 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2903 {
2904         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2905
2906         psf = psf->sf_next;
2907         while (!psf) {
2908                 spin_unlock_bh(&state->im->lock);
2909                 state->im = state->im->next;
2910                 while (!state->im) {
2911                         state->dev = next_net_device_rcu(state->dev);
2912                         if (!state->dev) {
2913                                 state->idev = NULL;
2914                                 goto out;
2915                         }
2916                         state->idev = __in_dev_get_rcu(state->dev);
2917                         if (!state->idev)
2918                                 continue;
2919                         state->im = rcu_dereference(state->idev->mc_list);
2920                 }
2921                 if (!state->im)
2922                         break;
2923                 spin_lock_bh(&state->im->lock);
2924                 psf = state->im->sources;
2925         }
2926 out:
2927         return psf;
2928 }
2929
2930 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2931 {
2932         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2933         if (psf)
2934                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2935                         --pos;
2936         return pos ? NULL : psf;
2937 }
2938
2939 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2940         __acquires(rcu)
2941 {
2942         rcu_read_lock();
2943         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2944 }
2945
2946 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2947 {
2948         struct ip_sf_list *psf;
2949         if (v == SEQ_START_TOKEN)
2950                 psf = igmp_mcf_get_first(seq);
2951         else
2952                 psf = igmp_mcf_get_next(seq, v);
2953         ++*pos;
2954         return psf;
2955 }
2956
2957 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2958         __releases(rcu)
2959 {
2960         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2961         if (likely(state->im)) {
2962                 spin_unlock_bh(&state->im->lock);
2963                 state->im = NULL;
2964         }
2965         state->idev = NULL;
2966         state->dev = NULL;
2967         rcu_read_unlock();
2968 }
2969
2970 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2971 {
2972         struct ip_sf_list *psf = (struct ip_sf_list *)v;
2973         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2974
2975         if (v == SEQ_START_TOKEN) {
2976                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2977         } else {
2978                 seq_printf(seq,
2979                            "%3d %6.6s 0x%08x "
2980                            "0x%08x %6lu %6lu\n",
2981                            state->dev->ifindex, state->dev->name,
2982                            ntohl(state->im->multiaddr),
2983                            ntohl(psf->sf_inaddr),
2984                            psf->sf_count[MCAST_INCLUDE],
2985                            psf->sf_count[MCAST_EXCLUDE]);
2986         }
2987         return 0;
2988 }
2989
2990 static const struct seq_operations igmp_mcf_seq_ops = {
2991         .start  =       igmp_mcf_seq_start,
2992         .next   =       igmp_mcf_seq_next,
2993         .stop   =       igmp_mcf_seq_stop,
2994         .show   =       igmp_mcf_seq_show,
2995 };
2996
2997 static int __net_init igmp_net_init(struct net *net)
2998 {
2999         struct proc_dir_entry *pde;
3000         int err;
3001
3002         pde = proc_create_net("igmp", 0444, net->proc_net, &igmp_mc_seq_ops,
3003                         sizeof(struct igmp_mc_iter_state));
3004         if (!pde)
3005                 goto out_igmp;
3006         pde = proc_create_net("mcfilter", 0444, net->proc_net,
3007                         &igmp_mcf_seq_ops, sizeof(struct igmp_mcf_iter_state));
3008         if (!pde)
3009                 goto out_mcfilter;
3010         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3011                                    SOCK_DGRAM, 0, net);
3012         if (err < 0) {
3013                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3014                        err);
3015                 goto out_sock;
3016         }
3017
3018         return 0;
3019
3020 out_sock:
3021         remove_proc_entry("mcfilter", net->proc_net);
3022 out_mcfilter:
3023         remove_proc_entry("igmp", net->proc_net);
3024 out_igmp:
3025         return -ENOMEM;
3026 }
3027
3028 static void __net_exit igmp_net_exit(struct net *net)
3029 {
3030         remove_proc_entry("mcfilter", net->proc_net);
3031         remove_proc_entry("igmp", net->proc_net);
3032         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3033 }
3034
3035 static struct pernet_operations igmp_net_ops = {
3036         .init = igmp_net_init,
3037         .exit = igmp_net_exit,
3038 };
3039 #endif
3040
3041 static int igmp_netdev_event(struct notifier_block *this,
3042                              unsigned long event, void *ptr)
3043 {
3044         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3045         struct in_device *in_dev;
3046
3047         switch (event) {
3048         case NETDEV_RESEND_IGMP:
3049                 in_dev = __in_dev_get_rtnl(dev);
3050                 if (in_dev)
3051                         ip_mc_rejoin_groups(in_dev);
3052                 break;
3053         default:
3054                 break;
3055         }
3056         return NOTIFY_DONE;
3057 }
3058
3059 static struct notifier_block igmp_notifier = {
3060         .notifier_call = igmp_netdev_event,
3061 };
3062
3063 int __init igmp_mc_init(void)
3064 {
3065 #if defined(CONFIG_PROC_FS)
3066         int err;
3067
3068         err = register_pernet_subsys(&igmp_net_ops);
3069         if (err)
3070                 return err;
3071         err = register_netdevice_notifier(&igmp_notifier);
3072         if (err)
3073                 goto reg_notif_fail;
3074         return 0;
3075
3076 reg_notif_fail:
3077         unregister_pernet_subsys(&igmp_net_ops);
3078         return err;
3079 #else
3080         return register_netdevice_notifier(&igmp_notifier);
3081 #endif
3082 }