net/ipv4/igmp.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux NET3:     Internet Group Management Protocol  [IGMP]
   4  *
   5  *      This code implements the IGMP protocol as defined in RFC1112. There has
   6  *      been a further revision of this protocol since which is now supported.
   7  *
   8  *      If you have trouble with this module be careful what gcc you have used,
   9  *      the older version didn't come out right using gcc 2.5.8, the newer one
  10  *      seems to fall out with gcc 2.6.2.
  11  *
  12  *      Authors:
  13  *              Alan Cox <alan@lxorguk.ukuu.org.uk>
  14  *
  15  *      Fixes:
  16  *
  17  *              Alan Cox        :       Added lots of __inline__ to optimise
  18  *                                      the memory usage of all the tiny little
  19  *                                      functions.
  20  *              Alan Cox        :       Dumped the header building experiment.
  21  *              Alan Cox        :       Minor tweaks ready for multicast routing
  22  *                                      and extended IGMP protocol.
  23  *              Alan Cox        :       Removed a load of inline directives. Gcc 2.5.8
  24  *                                      writes utterly bogus code otherwise (sigh)
  25  *                                      fixed IGMP loopback to behave in the manner
  26  *                                      desired by mrouted, fixed the fact it has been
  27  *                                      broken since 1.3.6 and cleaned up a few minor
  28  *                                      points.
  29  *
  30  *              Chih-Jen Chang  :       Tried to revise IGMP to Version 2
  31  *              Tsu-Sheng Tsao          E-mail: chihjenc@scf.usc.edu and tsusheng@scf.usc.edu
  32  *                                      The enhancements are mainly based on Steve Deering's
  33  *                                      ipmulti-3.5 source code.
  34  *              Chih-Jen Chang  :       Added the igmp_get_mrouter_info and
  35  *              Tsu-Sheng Tsao          igmp_set_mrouter_info to keep track of
  36  *                                      the mrouted version on that device.
  37  *              Chih-Jen Chang  :       Added the max_resp_time parameter to
  38  *              Tsu-Sheng Tsao          igmp_heard_query(). Using this parameter
  39  *                                      to identify the multicast router version
  40  *                                      and do what the IGMP version 2 specified.
  41  *              Chih-Jen Chang  :       Added a timer to revert to IGMP V2 router
  42  *              Tsu-Sheng Tsao          if the specified time expired.
  43  *              Alan Cox        :       Stop IGMP from 0.0.0.0 being accepted.
  44  *              Alan Cox        :       Use GFP_ATOMIC in the right places.
  45  *              Christian Daudt :       igmp timer wasn't set for local group
  46  *                                      memberships but was being deleted,
  47  *                                      which caused a "del_timer() called
  48  *                                      from %p with timer not initialized\n"
  49  *                                      message (960131).
  50  *              Christian Daudt :       removed del_timer from
  51  *                                      igmp_timer_expire function (960205).
  52  *             Christian Daudt :       igmp_heard_report now only calls
  53  *                                     igmp_timer_expire if tm->running is
  54  *                                     true (960216).
  55  *              Malcolm Beattie :       ttl comparison wrong in igmp_rcv made
  56  *                                      igmp_heard_query never trigger. Expiry
  57  *                                      miscalculation fixed in igmp_heard_query
  58  *                                      and random() made to return unsigned to
  59  *                                      prevent negative expiry times.
  60  *              Alexey Kuznetsov:       Wrong group leaving behaviour, backport
  61  *                                      fix from pending 2.1.x patches.
  62  *              Alan Cox:               Forget to enable FDDI support earlier.
  63  *              Alexey Kuznetsov:       Fixed leaving groups on device down.
  64  *              Alexey Kuznetsov:       Accordance to igmp-v2-06 draft.
  65  *              David L Stevens:        IGMPv3 support, with help from
  66  *                                      Vinay Kulkarni
  67  */
  68
  69 #include <linux/module.h>
  70 #include <linux/slab.h>
  71 #include <linux/uaccess.h>
  72 #include <linux/types.h>
  73 #include <linux/kernel.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/string.h>
  76 #include <linux/socket.h>
  77 #include <linux/sockios.h>
  78 #include <linux/in.h>
  79 #include <linux/inet.h>
  80 #include <linux/netdevice.h>
  81 #include <linux/skbuff.h>
  82 #include <linux/inetdevice.h>
  83 #include <linux/igmp.h>
  84 #include <linux/if_arp.h>
  85 #include <linux/rtnetlink.h>
  86 #include <linux/times.h>
  87 #include <linux/pkt_sched.h>
  88 #include <linux/byteorder/generic.h>
  89
  90 #include <net/net_namespace.h>
  91 #include <net/arp.h>
  92 #include <net/ip.h>
  93 #include <net/protocol.h>
  94 #include <net/route.h>
  95 #include <net/sock.h>
  96 #include <net/checksum.h>
  97 #include <net/inet_common.h>
  98 #include <linux/netfilter_ipv4.h>
  99 #ifdef CONFIG_IP_MROUTE
 100 #include <linux/mroute.h>
 101 #endif
 102 #ifdef CONFIG_PROC_FS
 103 #include <linux/proc_fs.h>
 104 #include <linux/seq_file.h>
 105 #endif
 106
 107 #ifdef CONFIG_IP_MULTICAST
 108 /* Parameter names and values are taken from igmp-v2-06 draft */
 109
 110 #define IGMP_V2_UNSOLICITED_REPORT_INTERVAL     (10*HZ)
 111 #define IGMP_V3_UNSOLICITED_REPORT_INTERVAL     (1*HZ)
 112 #define IGMP_QUERY_INTERVAL                     (125*HZ)
 113 #define IGMP_QUERY_RESPONSE_INTERVAL            (10*HZ)
 114
 115 #define IGMP_INITIAL_REPORT_DELAY               (1)
 116
 117 /* IGMP_INITIAL_REPORT_DELAY is not from IGMP specs!
 118  * IGMP specs require to report membership immediately after
 119  * joining a group, but we delay the first report by a
 120  * small interval. It seems more natural and still does not
 121  * contradict to specs provided this delay is small enough.
 122  */
 123
 124 #define IGMP_V1_SEEN(in_dev) \
 125         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
 126          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
 127          ((in_dev)->mr_v1_seen && \
 128           time_before(jiffies, (in_dev)->mr_v1_seen)))
 129 #define IGMP_V2_SEEN(in_dev) \
 130         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
 131          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
 132          ((in_dev)->mr_v2_seen && \
 133           time_before(jiffies, (in_dev)->mr_v2_seen)))
 134
 135 static int unsolicited_report_interval(struct in_device *in_dev)
 136 {
 137         int interval_ms, interval_jiffies;
 138
 139         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 140                 interval_ms = IN_DEV_CONF_GET(
 141                         in_dev,
 142                         IGMPV2_UNSOLICITED_REPORT_INTERVAL);
 143         else /* v3 */
 144                 interval_ms = IN_DEV_CONF_GET(
 145                         in_dev,
 146                         IGMPV3_UNSOLICITED_REPORT_INTERVAL);
 147
 148         interval_jiffies = msecs_to_jiffies(interval_ms);
 149
 150         /* _timer functions can't handle a delay of 0 jiffies so ensure
 151          *  we always return a positive value.
 152          */
 153         if (interval_jiffies <= 0)
 154                 interval_jiffies = 1;
 155         return interval_jiffies;
 156 }
 157
 158 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
 159                               gfp_t gfp);
 160 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 161 static void igmpv3_clear_delrec(struct in_device *in_dev);
 162 static int sf_setstate(struct ip_mc_list *pmc);
 163 static void sf_markstate(struct ip_mc_list *pmc);
 164 #endif
 165 static void ip_mc_clear_src(struct ip_mc_list *pmc);
 166 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
 167                          int sfcount, __be32 *psfsrc, int delta);
 168
 169 static void ip_ma_put(struct ip_mc_list *im)
 170 {
 171         if (refcount_dec_and_test(&im->refcnt)) {
 172                 in_dev_put(im->interface);
 173                 kfree_rcu(im, rcu);
 174         }
 175 }
 176
 177 #define for_each_pmc_rcu(in_dev, pmc)                           \
 178         for (pmc = rcu_dereference(in_dev->mc_list);            \
 179              pmc != NULL;                                       \
 180              pmc = rcu_dereference(pmc->next_rcu))
 181
 182 #define for_each_pmc_rtnl(in_dev, pmc)                          \
 183         for (pmc = rtnl_dereference(in_dev->mc_list);           \
 184              pmc != NULL;                                       \
 185              pmc = rtnl_dereference(pmc->next_rcu))
 186
 187 static void ip_sf_list_clear_all(struct ip_sf_list *psf)
 188 {
 189         struct ip_sf_list *next;
 190
 191         while (psf) {
 192                 next = psf->sf_next;
 193                 kfree(psf);
 194                 psf = next;
 195         }
 196 }
 197
 198 #ifdef CONFIG_IP_MULTICAST
 199
 200 /*
 201  *      Timer management
 202  */
 203
 204 static void igmp_stop_timer(struct ip_mc_list *im)
 205 {
 206         spin_lock_bh(&im->lock);
 207         if (del_timer(&im->timer))
 208                 refcount_dec(&im->refcnt);
 209         im->tm_running = 0;
 210         im->reporter = 0;
 211         im->unsolicit_count = 0;
 212         spin_unlock_bh(&im->lock);
 213 }
 214
 215 /* It must be called with locked im->lock */
 216 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
 217 {
 218         int tv = prandom_u32() % max_delay;
 219
 220         im->tm_running = 1;
 221         if (!mod_timer(&im->timer, jiffies+tv+2))
 222                 refcount_inc(&im->refcnt);
 223 }
 224
 225 static void igmp_gq_start_timer(struct in_device *in_dev)
 226 {
 227         int tv = prandom_u32() % in_dev->mr_maxdelay;
 228         unsigned long exp = jiffies + tv + 2;
 229
 230         if (in_dev->mr_gq_running &&
 231             time_after_eq(exp, (in_dev->mr_gq_timer).expires))
 232                 return;
 233
 234         in_dev->mr_gq_running = 1;
 235         if (!mod_timer(&in_dev->mr_gq_timer, exp))
 236                 in_dev_hold(in_dev);
 237 }
 238
 239 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 240 {
 241         int tv = prandom_u32() % delay;
 242
 243         if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
 244                 in_dev_hold(in_dev);
 245 }
 246
 247 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 248 {
 249         spin_lock_bh(&im->lock);
 250         im->unsolicit_count = 0;
 251         if (del_timer(&im->timer)) {
 252                 if ((long)(im->timer.expires-jiffies) < max_delay) {
 253                         add_timer(&im->timer);
 254                         im->tm_running = 1;
 255                         spin_unlock_bh(&im->lock);
 256                         return;
 257                 }
 258                 refcount_dec(&im->refcnt);
 259         }
 260         igmp_start_timer(im, max_delay);
 261         spin_unlock_bh(&im->lock);
 262 }
 263
 264
 265 /*
 266  *      Send an IGMP report.
 267  */
 268
 269 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
 270
 271
 272 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
 273         int gdeleted, int sdeleted)
 274 {
 275         switch (type) {
 276         case IGMPV3_MODE_IS_INCLUDE:
 277         case IGMPV3_MODE_IS_EXCLUDE:
 278                 if (gdeleted || sdeleted)
 279                         return 0;
 280                 if (!(pmc->gsquery && !psf->sf_gsresp)) {
 281                         if (pmc->sfmode == MCAST_INCLUDE)
 282                                 return 1;
 283                         /* don't include if this source is excluded
 284                          * in all filters
 285                          */
 286                         if (psf->sf_count[MCAST_INCLUDE])
 287                                 return type == IGMPV3_MODE_IS_INCLUDE;
 288                         return pmc->sfcount[MCAST_EXCLUDE] ==
 289                                 psf->sf_count[MCAST_EXCLUDE];
 290                 }
 291                 return 0;
 292         case IGMPV3_CHANGE_TO_INCLUDE:
 293                 if (gdeleted || sdeleted)
 294                         return 0;
 295                 return psf->sf_count[MCAST_INCLUDE] != 0;
 296         case IGMPV3_CHANGE_TO_EXCLUDE:
 297                 if (gdeleted || sdeleted)
 298                         return 0;
 299                 if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
 300                     psf->sf_count[MCAST_INCLUDE])
 301                         return 0;
 302                 return pmc->sfcount[MCAST_EXCLUDE] ==
 303                         psf->sf_count[MCAST_EXCLUDE];
 304         case IGMPV3_ALLOW_NEW_SOURCES:
 305                 if (gdeleted || !psf->sf_crcount)
 306                         return 0;
 307                 return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
 308         case IGMPV3_BLOCK_OLD_SOURCES:
 309                 if (pmc->sfmode == MCAST_INCLUDE)
 310                         return gdeleted || (psf->sf_crcount && sdeleted);
 311                 return psf->sf_crcount && !gdeleted && !sdeleted;
 312         }
 313         return 0;
 314 }
 315
 316 static int
 317 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
 318 {
 319         struct ip_sf_list *psf;
 320         int scount = 0;
 321
 322         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 323                 if (!is_in(pmc, psf, type, gdeleted, sdeleted))
 324                         continue;
 325                 scount++;
 326         }
 327         return scount;
 328 }
 329
 330 /* source address selection per RFC 3376 section 4.2.13 */
 331 static __be32 igmpv3_get_srcaddr(struct net_device *dev,
 332                                  const struct flowi4 *fl4)
 333 {
 334         struct in_device *in_dev = __in_dev_get_rcu(dev);
 335
 336         if (!in_dev)
 337                 return htonl(INADDR_ANY);
 338
 339         for_ifa(in_dev) {
 340                 if (fl4->saddr == ifa->ifa_local)
 341                         return fl4->saddr;
 342         } endfor_ifa(in_dev);
 343
 344         return htonl(INADDR_ANY);
 345 }
 346
 347 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
 348 {
 349         struct sk_buff *skb;
 350         struct rtable *rt;
 351         struct iphdr *pip;
 352         struct igmpv3_report *pig;
 353         struct net *net = dev_net(dev);
 354         struct flowi4 fl4;
 355         int hlen = LL_RESERVED_SPACE(dev);
 356         int tlen = dev->needed_tailroom;
 357         unsigned int size = mtu;
 358
 359         while (1) {
 360                 skb = alloc_skb(size + hlen + tlen,
 361                                 GFP_ATOMIC | __GFP_NOWARN);
 362                 if (skb)
 363                         break;
 364                 size >>= 1;
 365                 if (size < 256)
 366                         return NULL;
 367         }
 368         skb->priority = TC_PRIO_CONTROL;
 369
 370         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 371                                    0, 0,
 372                                    IPPROTO_IGMP, 0, dev->ifindex);
 373         if (IS_ERR(rt)) {
 374                 kfree_skb(skb);
 375                 return NULL;
 376         }
 377
 378         skb_dst_set(skb, &rt->dst);
 379         skb->dev = dev;
 380
 381         skb_reserve(skb, hlen);
 382         skb_tailroom_reserve(skb, mtu, tlen);
 383
 384         skb_reset_network_header(skb);
 385         pip = ip_hdr(skb);
 386         skb_put(skb, sizeof(struct iphdr) + 4);
 387
 388         pip->version  = 4;
 389         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 390         pip->tos      = 0xc0;
 391         pip->frag_off = htons(IP_DF);
 392         pip->ttl      = 1;
 393         pip->daddr    = fl4.daddr;
 394
 395         rcu_read_lock();
 396         pip->saddr    = igmpv3_get_srcaddr(dev, &fl4);
 397         rcu_read_unlock();
 398
 399         pip->protocol = IPPROTO_IGMP;
 400         pip->tot_len  = 0;      /* filled in later */
 401         ip_select_ident(net, skb, NULL);
 402         ((u8 *)&pip[1])[0] = IPOPT_RA;
 403         ((u8 *)&pip[1])[1] = 4;
 404         ((u8 *)&pip[1])[2] = 0;
 405         ((u8 *)&pip[1])[3] = 0;
 406
 407         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 408         skb_put(skb, sizeof(*pig));
 409         pig = igmpv3_report_hdr(skb);
 410         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 411         pig->resv1 = 0;
 412         pig->csum = 0;
 413         pig->resv2 = 0;
 414         pig->ngrec = 0;
 415         return skb;
 416 }
 417
 418 static int igmpv3_sendpack(struct sk_buff *skb)
 419 {
 420         struct igmphdr *pig = igmp_hdr(skb);
 421         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 422
 423         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 424
 425         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 426 }
 427
 428 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 429 {
 430         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 431 }
 432
 433 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 434         int type, struct igmpv3_grec **ppgr, unsigned int mtu)
 435 {
 436         struct net_device *dev = pmc->interface->dev;
 437         struct igmpv3_report *pih;
 438         struct igmpv3_grec *pgr;
 439
 440         if (!skb) {
 441                 skb = igmpv3_newpack(dev, mtu);
 442                 if (!skb)
 443                         return NULL;
 444         }
 445         pgr = skb_put(skb, sizeof(struct igmpv3_grec));
 446         pgr->grec_type = type;
 447         pgr->grec_auxwords = 0;
 448         pgr->grec_nsrcs = 0;
 449         pgr->grec_mca = pmc->multiaddr;
 450         pih = igmpv3_report_hdr(skb);
 451         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 452         *ppgr = pgr;
 453         return skb;
 454 }
 455
 456 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 457
 458 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 459         int type, int gdeleted, int sdeleted)
 460 {
 461         struct net_device *dev = pmc->interface->dev;
 462         struct net *net = dev_net(dev);
 463         struct igmpv3_report *pih;
 464         struct igmpv3_grec *pgr = NULL;
 465         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 466         int scount, stotal, first, isquery, truncate;
 467         unsigned int mtu;
 468
 469         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 470                 return skb;
 471         if (ipv4_is_local_multicast(pmc->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
 472                 return skb;
 473
 474         mtu = READ_ONCE(dev->mtu);
 475         if (mtu < IPV4_MIN_MTU)
 476                 return skb;
 477
 478         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 479                   type == IGMPV3_MODE_IS_EXCLUDE;
 480         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 481                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 482
 483         stotal = scount = 0;
 484
 485         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 486
 487         if (!*psf_list)
 488                 goto empty_source;
 489
 490         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 491
 492         /* EX and TO_EX get a fresh packet, if needed */
 493         if (truncate) {
 494                 if (pih && pih->ngrec &&
 495                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 496                         if (skb)
 497                                 igmpv3_sendpack(skb);
 498                         skb = igmpv3_newpack(dev, mtu);
 499                 }
 500         }
 501         first = 1;
 502         psf_prev = NULL;
 503         for (psf = *psf_list; psf; psf = psf_next) {
 504                 __be32 *psrc;
 505
 506                 psf_next = psf->sf_next;
 507
 508                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 509                         psf_prev = psf;
 510                         continue;
 511                 }
 512
 513                 /* Based on RFC3376 5.1. Should not send source-list change
 514                  * records when there is a filter mode change.
 515                  */
 516                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 517                      (!gdeleted && pmc->crcount)) &&
 518                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 519                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 520                         goto decrease_sf_crcount;
 521
 522                 /* clear marks on query responses */
 523                 if (isquery)
 524                         psf->sf_gsresp = 0;
 525
 526                 if (AVAILABLE(skb) < sizeof(__be32) +
 527                     first*sizeof(struct igmpv3_grec)) {
 528                         if (truncate && !first)
 529                                 break;   /* truncate these */
 530                         if (pgr)
 531                                 pgr->grec_nsrcs = htons(scount);
 532                         if (skb)
 533                                 igmpv3_sendpack(skb);
 534                         skb = igmpv3_newpack(dev, mtu);
 535                         first = 1;
 536                         scount = 0;
 537                 }
 538                 if (first) {
 539                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 540                         first = 0;
 541                 }
 542                 if (!skb)
 543                         return NULL;
 544                 psrc = skb_put(skb, sizeof(__be32));
 545                 *psrc = psf->sf_inaddr;
 546                 scount++; stotal++;
 547                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 548                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 549 decrease_sf_crcount:
 550                         psf->sf_crcount--;
 551                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 552                                 if (psf_prev)
 553                                         psf_prev->sf_next = psf->sf_next;
 554                                 else
 555                                         *psf_list = psf->sf_next;
 556                                 kfree(psf);
 557                                 continue;
 558                         }
 559                 }
 560                 psf_prev = psf;
 561         }
 562
 563 empty_source:
 564         if (!stotal) {
 565                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 566                     type == IGMPV3_BLOCK_OLD_SOURCES)
 567                         return skb;
 568                 if (pmc->crcount || isquery) {
 569                         /* make sure we have room for group header */
 570                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 571                                 igmpv3_sendpack(skb);
 572                                 skb = NULL; /* add_grhead will get a new one */
 573                         }
 574                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 575                 }
 576         }
 577         if (pgr)
 578                 pgr->grec_nsrcs = htons(scount);
 579
 580         if (isquery)
 581                 pmc->gsquery = 0;       /* clear query state on report */
 582         return skb;
 583 }
 584
 585 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 586 {
 587         struct sk_buff *skb = NULL;
 588         struct net *net = dev_net(in_dev->dev);
 589         int type;
 590
 591         if (!pmc) {
 592                 rcu_read_lock();
 593                 for_each_pmc_rcu(in_dev, pmc) {
 594                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 595                                 continue;
 596                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 597                              !net->ipv4.sysctl_igmp_llm_reports)
 598                                 continue;
 599                         spin_lock_bh(&pmc->lock);
 600                         if (pmc->sfcount[MCAST_EXCLUDE])
 601                                 type = IGMPV3_MODE_IS_EXCLUDE;
 602                         else
 603                                 type = IGMPV3_MODE_IS_INCLUDE;
 604                         skb = add_grec(skb, pmc, type, 0, 0);
 605                         spin_unlock_bh(&pmc->lock);
 606                 }
 607                 rcu_read_unlock();
 608         } else {
 609                 spin_lock_bh(&pmc->lock);
 610                 if (pmc->sfcount[MCAST_EXCLUDE])
 611                         type = IGMPV3_MODE_IS_EXCLUDE;
 612                 else
 613                         type = IGMPV3_MODE_IS_INCLUDE;
 614                 skb = add_grec(skb, pmc, type, 0, 0);
 615                 spin_unlock_bh(&pmc->lock);
 616         }
 617         if (!skb)
 618                 return 0;
 619         return igmpv3_sendpack(skb);
 620 }
 621
 622 /*
 623  * remove zero-count source records from a source filter list
 624  */
 625 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 626 {
 627         struct ip_sf_list *psf_prev, *psf_next, *psf;
 628
 629         psf_prev = NULL;
 630         for (psf = *ppsf; psf; psf = psf_next) {
 631                 psf_next = psf->sf_next;
 632                 if (psf->sf_crcount == 0) {
 633                         if (psf_prev)
 634                                 psf_prev->sf_next = psf->sf_next;
 635                         else
 636                                 *ppsf = psf->sf_next;
 637                         kfree(psf);
 638                 } else
 639                         psf_prev = psf;
 640         }
 641 }
 642
 643 static void kfree_pmc(struct ip_mc_list *pmc)
 644 {
 645         ip_sf_list_clear_all(pmc->sources);
 646         ip_sf_list_clear_all(pmc->tomb);
 647         kfree(pmc);
 648 }
 649
 650 static void igmpv3_send_cr(struct in_device *in_dev)
 651 {
 652         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 653         struct sk_buff *skb = NULL;
 654         int type, dtype;
 655
 656         rcu_read_lock();
 657         spin_lock_bh(&in_dev->mc_tomb_lock);
 658
 659         /* deleted MCA's */
 660         pmc_prev = NULL;
 661         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 662                 pmc_next = pmc->next;
 663                 if (pmc->sfmode == MCAST_INCLUDE) {
 664                         type = IGMPV3_BLOCK_OLD_SOURCES;
 665                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 666                         skb = add_grec(skb, pmc, type, 1, 0);
 667                         skb = add_grec(skb, pmc, dtype, 1, 1);
 668                 }
 669                 if (pmc->crcount) {
 670                         if (pmc->sfmode == MCAST_EXCLUDE) {
 671                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 672                                 skb = add_grec(skb, pmc, type, 1, 0);
 673                         }
 674                         pmc->crcount--;
 675                         if (pmc->crcount == 0) {
 676                                 igmpv3_clear_zeros(&pmc->tomb);
 677                                 igmpv3_clear_zeros(&pmc->sources);
 678                         }
 679                 }
 680                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 681                         if (pmc_prev)
 682                                 pmc_prev->next = pmc_next;
 683                         else
 684                                 in_dev->mc_tomb = pmc_next;
 685                         in_dev_put(pmc->interface);
 686                         kfree_pmc(pmc);
 687                 } else
 688                         pmc_prev = pmc;
 689         }
 690         spin_unlock_bh(&in_dev->mc_tomb_lock);
 691
 692         /* change recs */
 693         for_each_pmc_rcu(in_dev, pmc) {
 694                 spin_lock_bh(&pmc->lock);
 695                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 696                         type = IGMPV3_BLOCK_OLD_SOURCES;
 697                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 698                 } else {
 699                         type = IGMPV3_ALLOW_NEW_SOURCES;
 700                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 701                 }
 702                 skb = add_grec(skb, pmc, type, 0, 0);
 703                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 704
 705                 /* filter mode changes */
 706                 if (pmc->crcount) {
 707                         if (pmc->sfmode == MCAST_EXCLUDE)
 708                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 709                         else
 710                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 711                         skb = add_grec(skb, pmc, type, 0, 0);
 712                         pmc->crcount--;
 713                 }
 714                 spin_unlock_bh(&pmc->lock);
 715         }
 716         rcu_read_unlock();
 717
 718         if (!skb)
 719                 return;
 720         (void) igmpv3_sendpack(skb);
 721 }
 722
 723 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 724         int type)
 725 {
 726         struct sk_buff *skb;
 727         struct iphdr *iph;
 728         struct igmphdr *ih;
 729         struct rtable *rt;
 730         struct net_device *dev = in_dev->dev;
 731         struct net *net = dev_net(dev);
 732         __be32  group = pmc ? pmc->multiaddr : 0;
 733         struct flowi4 fl4;
 734         __be32  dst;
 735         int hlen, tlen;
 736
 737         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 738                 return igmpv3_send_report(in_dev, pmc);
 739
 740         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 741                 return 0;
 742
 743         if (type == IGMP_HOST_LEAVE_MESSAGE)
 744                 dst = IGMP_ALL_ROUTER;
 745         else
 746                 dst = group;
 747
 748         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 749                                    0, 0,
 750                                    IPPROTO_IGMP, 0, dev->ifindex);
 751         if (IS_ERR(rt))
 752                 return -1;
 753
 754         hlen = LL_RESERVED_SPACE(dev);
 755         tlen = dev->needed_tailroom;
 756         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 757         if (!skb) {
 758                 ip_rt_put(rt);
 759                 return -1;
 760         }
 761         skb->priority = TC_PRIO_CONTROL;
 762
 763         skb_dst_set(skb, &rt->dst);
 764
 765         skb_reserve(skb, hlen);
 766
 767         skb_reset_network_header(skb);
 768         iph = ip_hdr(skb);
 769         skb_put(skb, sizeof(struct iphdr) + 4);
 770
 771         iph->version  = 4;
 772         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 773         iph->tos      = 0xc0;
 774         iph->frag_off = htons(IP_DF);
 775         iph->ttl      = 1;
 776         iph->daddr    = dst;
 777         iph->saddr    = fl4.saddr;
 778         iph->protocol = IPPROTO_IGMP;
 779         ip_select_ident(net, skb, NULL);
 780         ((u8 *)&iph[1])[0] = IPOPT_RA;
 781         ((u8 *)&iph[1])[1] = 4;
 782         ((u8 *)&iph[1])[2] = 0;
 783         ((u8 *)&iph[1])[3] = 0;
 784
 785         ih = skb_put(skb, sizeof(struct igmphdr));
 786         ih->type = type;
 787         ih->code = 0;
 788         ih->csum = 0;
 789         ih->group = group;
 790         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 791
 792         return ip_local_out(net, skb->sk, skb);
 793 }
 794
 795 static void igmp_gq_timer_expire(struct timer_list *t)
 796 {
 797         struct in_device *in_dev = from_timer(in_dev, t, mr_gq_timer);
 798
 799         in_dev->mr_gq_running = 0;
 800         igmpv3_send_report(in_dev, NULL);
 801         in_dev_put(in_dev);
 802 }
 803
 804 static void igmp_ifc_timer_expire(struct timer_list *t)
 805 {
 806         struct in_device *in_dev = from_timer(in_dev, t, mr_ifc_timer);
 807
 808         igmpv3_send_cr(in_dev);
 809         if (in_dev->mr_ifc_count) {
 810                 in_dev->mr_ifc_count--;
 811                 igmp_ifc_start_timer(in_dev,
 812                                      unsolicited_report_interval(in_dev));
 813         }
 814         in_dev_put(in_dev);
 815 }
 816
 817 static void igmp_ifc_event(struct in_device *in_dev)
 818 {
 819         struct net *net = dev_net(in_dev->dev);
 820         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 821                 return;
 822         in_dev->mr_ifc_count = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
 823         igmp_ifc_start_timer(in_dev, 1);
 824 }
 825
 826
 827 static void igmp_timer_expire(struct timer_list *t)
 828 {
 829         struct ip_mc_list *im = from_timer(im, t, timer);
 830         struct in_device *in_dev = im->interface;
 831
 832         spin_lock(&im->lock);
 833         im->tm_running = 0;
 834
 835         if (im->unsolicit_count && --im->unsolicit_count)
 836                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 837
 838         im->reporter = 1;
 839         spin_unlock(&im->lock);
 840
 841         if (IGMP_V1_SEEN(in_dev))
 842                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 843         else if (IGMP_V2_SEEN(in_dev))
 844                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 845         else
 846                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 847
 848         ip_ma_put(im);
 849 }
 850
 851 /* mark EXCLUDE-mode sources */
 852 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 853 {
 854         struct ip_sf_list *psf;
 855         int i, scount;
 856
 857         scount = 0;
 858         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 859                 if (scount == nsrcs)
 860                         break;
 861                 for (i = 0; i < nsrcs; i++) {
 862                         /* skip inactive filters */
 863                         if (psf->sf_count[MCAST_INCLUDE] ||
 864                             pmc->sfcount[MCAST_EXCLUDE] !=
 865                             psf->sf_count[MCAST_EXCLUDE])
 866                                 break;
 867                         if (srcs[i] == psf->sf_inaddr) {
 868                                 scount++;
 869                                 break;
 870                         }
 871                 }
 872         }
 873         pmc->gsquery = 0;
 874         if (scount == nsrcs)    /* all sources excluded */
 875                 return 0;
 876         return 1;
 877 }
 878
 879 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 880 {
 881         struct ip_sf_list *psf;
 882         int i, scount;
 883
 884         if (pmc->sfmode == MCAST_EXCLUDE)
 885                 return igmp_xmarksources(pmc, nsrcs, srcs);
 886
 887         /* mark INCLUDE-mode sources */
 888         scount = 0;
 889         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 890                 if (scount == nsrcs)
 891                         break;
 892                 for (i = 0; i < nsrcs; i++)
 893                         if (srcs[i] == psf->sf_inaddr) {
 894                                 psf->sf_gsresp = 1;
 895                                 scount++;
 896                                 break;
 897                         }
 898         }
 899         if (!scount) {
 900                 pmc->gsquery = 0;
 901                 return 0;
 902         }
 903         pmc->gsquery = 1;
 904         return 1;
 905 }
 906
 907 /* return true if packet was dropped */
 908 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 909 {
 910         struct ip_mc_list *im;
 911         struct net *net = dev_net(in_dev->dev);
 912
 913         /* Timers are only set for non-local groups */
 914
 915         if (group == IGMP_ALL_HOSTS)
 916                 return false;
 917         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 918                 return false;
 919
 920         rcu_read_lock();
 921         for_each_pmc_rcu(in_dev, im) {
 922                 if (im->multiaddr == group) {
 923                         igmp_stop_timer(im);
 924                         break;
 925                 }
 926         }
 927         rcu_read_unlock();
 928         return false;
 929 }
 930
 931 /* return true if packet was dropped */
 932 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 933         int len)
 934 {
 935         struct igmphdr          *ih = igmp_hdr(skb);
 936         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 937         struct ip_mc_list       *im;
 938         __be32                  group = ih->group;
 939         int                     max_delay;
 940         int                     mark = 0;
 941         struct net              *net = dev_net(in_dev->dev);
 942
 943
 944         if (len == 8) {
 945                 if (ih->code == 0) {
 946                         /* Alas, old v1 router presents here. */
 947
 948                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 949                         in_dev->mr_v1_seen = jiffies +
 950                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 951                                 in_dev->mr_qri;
 952                         group = 0;
 953                 } else {
 954                         /* v2 router present */
 955                         max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
 956                         in_dev->mr_v2_seen = jiffies +
 957                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 958                                 in_dev->mr_qri;
 959                 }
 960                 /* cancel the interface change timer */
 961                 in_dev->mr_ifc_count = 0;
 962                 if (del_timer(&in_dev->mr_ifc_timer))
 963                         __in_dev_put(in_dev);
 964                 /* clear deleted report items */
 965                 igmpv3_clear_delrec(in_dev);
 966         } else if (len < 12) {
 967                 return true;    /* ignore bogus packet; freed by caller */
 968         } else if (IGMP_V1_SEEN(in_dev)) {
 969                 /* This is a v3 query with v1 queriers present */
 970                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 971                 group = 0;
 972         } else if (IGMP_V2_SEEN(in_dev)) {
 973                 /* this is a v3 query with v2 queriers present;
 974                  * Interpretation of the max_delay code is problematic here.
 975                  * A real v2 host would use ih_code directly, while v3 has a
 976                  * different encoding. We use the v3 encoding as more likely
 977                  * to be intended in a v3 query.
 978                  */
 979                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 980                 if (!max_delay)
 981                         max_delay = 1;  /* can't mod w/ 0 */
 982         } else { /* v3 */
 983                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 984                         return true;
 985
 986                 ih3 = igmpv3_query_hdr(skb);
 987                 if (ih3->nsrcs) {
 988                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
 989                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
 990                                 return true;
 991                         ih3 = igmpv3_query_hdr(skb);
 992                 }
 993
 994                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 995                 if (!max_delay)
 996                         max_delay = 1;  /* can't mod w/ 0 */
 997                 in_dev->mr_maxdelay = max_delay;
 998
 999                 /* RFC3376, 4.1.6. QRV and 4.1.7. QQIC, when the most recently
1000                  * received value was zero, use the default or statically
1001                  * configured value.
1002                  */
1003                 in_dev->mr_qrv = ih3->qrv ?: net->ipv4.sysctl_igmp_qrv;
1004                 in_dev->mr_qi = IGMPV3_QQIC(ih3->qqic)*HZ ?: IGMP_QUERY_INTERVAL;
1005
1006                 /* RFC3376, 8.3. Query Response Interval:
1007                  * The number of seconds represented by the [Query Response
1008                  * Interval] must be less than the [Query Interval].
1009                  */
1010                 if (in_dev->mr_qri >= in_dev->mr_qi)
1011                         in_dev->mr_qri = (in_dev->mr_qi/HZ - 1)*HZ;
1012
1013                 if (!group) { /* general query */
1014                         if (ih3->nsrcs)
1015                                 return true;    /* no sources allowed */
1016                         igmp_gq_start_timer(in_dev);
1017                         return false;
1018                 }
1019                 /* mark sources to include, if group & source-specific */
1020                 mark = ih3->nsrcs != 0;
1021         }
1022
1023         /*
1024          * - Start the timers in all of our membership records
1025          *   that the query applies to for the interface on
1026          *   which the query arrived excl. those that belong
1027          *   to a "local" group (224.0.0.X)
1028          * - For timers already running check if they need to
1029          *   be reset.
1030          * - Use the igmp->igmp_code field as the maximum
1031          *   delay possible
1032          */
1033         rcu_read_lock();
1034         for_each_pmc_rcu(in_dev, im) {
1035                 int changed;
1036
1037                 if (group && group != im->multiaddr)
1038                         continue;
1039                 if (im->multiaddr == IGMP_ALL_HOSTS)
1040                         continue;
1041                 if (ipv4_is_local_multicast(im->multiaddr) &&
1042                     !net->ipv4.sysctl_igmp_llm_reports)
1043                         continue;
1044                 spin_lock_bh(&im->lock);
1045                 if (im->tm_running)
1046                         im->gsquery = im->gsquery && mark;
1047                 else
1048                         im->gsquery = mark;
1049                 changed = !im->gsquery ||
1050                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1051                 spin_unlock_bh(&im->lock);
1052                 if (changed)
1053                         igmp_mod_timer(im, max_delay);
1054         }
1055         rcu_read_unlock();
1056         return false;
1057 }
1058
1059 /* called in rcu_read_lock() section */
1060 int igmp_rcv(struct sk_buff *skb)
1061 {
1062         /* This basically follows the spec line by line -- see RFC1112 */
1063         struct igmphdr *ih;
1064         struct net_device *dev = skb->dev;
1065         struct in_device *in_dev;
1066         int len = skb->len;
1067         bool dropped = true;
1068
1069         if (netif_is_l3_master(dev)) {
1070                 dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1071                 if (!dev)
1072                         goto drop;
1073         }
1074
1075         in_dev = __in_dev_get_rcu(dev);
1076         if (!in_dev)
1077                 goto drop;
1078
1079         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1080                 goto drop;
1081
1082         if (skb_checksum_simple_validate(skb))
1083                 goto drop;
1084
1085         ih = igmp_hdr(skb);
1086         switch (ih->type) {
1087         case IGMP_HOST_MEMBERSHIP_QUERY:
1088                 dropped = igmp_heard_query(in_dev, skb, len);
1089                 break;
1090         case IGMP_HOST_MEMBERSHIP_REPORT:
1091         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1092                 /* Is it our report looped back? */
1093                 if (rt_is_output_route(skb_rtable(skb)))
1094                         break;
1095                 /* don't rely on MC router hearing unicast reports */
1096                 if (skb->pkt_type == PACKET_MULTICAST ||
1097                     skb->pkt_type == PACKET_BROADCAST)
1098                         dropped = igmp_heard_report(in_dev, ih->group);
1099                 break;
1100         case IGMP_PIM:
1101 #ifdef CONFIG_IP_PIMSM_V1
1102                 return pim_rcv_v1(skb);
1103 #endif
1104         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1105         case IGMP_DVMRP:
1106         case IGMP_TRACE:
1107         case IGMP_HOST_LEAVE_MESSAGE:
1108         case IGMP_MTRACE:
1109         case IGMP_MTRACE_RESP:
1110                 break;
1111         default:
1112                 break;
1113         }
1114
1115 drop:
1116         if (dropped)
1117                 kfree_skb(skb);
1118         else
1119                 consume_skb(skb);
1120         return 0;
1121 }
1122
1123 #endif
1124
1125
1126 /*
1127  *      Add a filter to a device
1128  */
1129
1130 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1131 {
1132         char buf[MAX_ADDR_LEN];
1133         struct net_device *dev = in_dev->dev;
1134
1135         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1136            We will get multicast token leakage, when IFF_MULTICAST
1137            is changed. This check should be done in ndo_set_rx_mode
1138            routine. Something sort of:
1139            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1140            --ANK
1141            */
1142         if (arp_mc_map(addr, buf, dev, 0) == 0)
1143                 dev_mc_add(dev, buf);
1144 }
1145
1146 /*
1147  *      Remove a filter from a device
1148  */
1149
1150 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1151 {
1152         char buf[MAX_ADDR_LEN];
1153         struct net_device *dev = in_dev->dev;
1154
1155         if (arp_mc_map(addr, buf, dev, 0) == 0)
1156                 dev_mc_del(dev, buf);
1157 }
1158
1159 #ifdef CONFIG_IP_MULTICAST
1160 /*
1161  * deleted ip_mc_list manipulation
1162  */
1163 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
1164                               gfp_t gfp)
1165 {
1166         struct ip_mc_list *pmc;
1167         struct net *net = dev_net(in_dev->dev);
1168
1169         /* this is an "ip_mc_list" for convenience; only the fields below
1170          * are actually used. In particular, the refcnt and users are not
1171          * used for management of the delete list. Using the same structure
1172          * for deleted items allows change reports to use common code with
1173          * non-deleted or query-response MCA's.
1174          */
1175         pmc = kzalloc(sizeof(*pmc), gfp);
1176         if (!pmc)
1177                 return;
1178         spin_lock_init(&pmc->lock);
1179         spin_lock_bh(&im->lock);
1180         pmc->interface = im->interface;
1181         in_dev_hold(in_dev);
1182         pmc->multiaddr = im->multiaddr;
1183         pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1184         pmc->sfmode = im->sfmode;
1185         if (pmc->sfmode == MCAST_INCLUDE) {
1186                 struct ip_sf_list *psf;
1187
1188                 pmc->tomb = im->tomb;
1189                 pmc->sources = im->sources;
1190                 im->tomb = im->sources = NULL;
1191                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1192                         psf->sf_crcount = pmc->crcount;
1193         }
1194         spin_unlock_bh(&im->lock);
1195
1196         spin_lock_bh(&in_dev->mc_tomb_lock);
1197         pmc->next = in_dev->mc_tomb;
1198         in_dev->mc_tomb = pmc;
1199         spin_unlock_bh(&in_dev->mc_tomb_lock);
1200 }
1201
1202 /*
1203  * restore ip_mc_list deleted records
1204  */
1205 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1206 {
1207         struct ip_mc_list *pmc, *pmc_prev;
1208         struct ip_sf_list *psf;
1209         struct net *net = dev_net(in_dev->dev);
1210         __be32 multiaddr = im->multiaddr;
1211
1212         spin_lock_bh(&in_dev->mc_tomb_lock);
1213         pmc_prev = NULL;
1214         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1215                 if (pmc->multiaddr == multiaddr)
1216                         break;
1217                 pmc_prev = pmc;
1218         }
1219         if (pmc) {
1220                 if (pmc_prev)
1221                         pmc_prev->next = pmc->next;
1222                 else
1223                         in_dev->mc_tomb = pmc->next;
1224         }
1225         spin_unlock_bh(&in_dev->mc_tomb_lock);
1226
1227         spin_lock_bh(&im->lock);
1228         if (pmc) {
1229                 im->interface = pmc->interface;
1230                 if (im->sfmode == MCAST_INCLUDE) {
1231                         im->tomb = pmc->tomb;
1232                         pmc->tomb = NULL;
1233
1234                         im->sources = pmc->sources;
1235                         pmc->sources = NULL;
1236
1237                         for (psf = im->sources; psf; psf = psf->sf_next)
1238                                 psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1239                 } else {
1240                         im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1241                 }
1242                 in_dev_put(pmc->interface);
1243                 kfree_pmc(pmc);
1244         }
1245         spin_unlock_bh(&im->lock);
1246 }
1247
1248 /*
1249  * flush ip_mc_list deleted records
1250  */
1251 static void igmpv3_clear_delrec(struct in_device *in_dev)
1252 {
1253         struct ip_mc_list *pmc, *nextpmc;
1254
1255         spin_lock_bh(&in_dev->mc_tomb_lock);
1256         pmc = in_dev->mc_tomb;
1257         in_dev->mc_tomb = NULL;
1258         spin_unlock_bh(&in_dev->mc_tomb_lock);
1259
1260         for (; pmc; pmc = nextpmc) {
1261                 nextpmc = pmc->next;
1262                 ip_mc_clear_src(pmc);
1263                 in_dev_put(pmc->interface);
1264                 kfree_pmc(pmc);
1265         }
1266         /* clear dead sources, too */
1267         rcu_read_lock();
1268         for_each_pmc_rcu(in_dev, pmc) {
1269                 struct ip_sf_list *psf;
1270
1271                 spin_lock_bh(&pmc->lock);
1272                 psf = pmc->tomb;
1273                 pmc->tomb = NULL;
1274                 spin_unlock_bh(&pmc->lock);
1275                 ip_sf_list_clear_all(psf);
1276         }
1277         rcu_read_unlock();
1278 }
1279 #endif
1280
1281 static void __igmp_group_dropped(struct ip_mc_list *im, gfp_t gfp)
1282 {
1283         struct in_device *in_dev = im->interface;
1284 #ifdef CONFIG_IP_MULTICAST
1285         struct net *net = dev_net(in_dev->dev);
1286         int reporter;
1287 #endif
1288
1289         if (im->loaded) {
1290                 im->loaded = 0;
1291                 ip_mc_filter_del(in_dev, im->multiaddr);
1292         }
1293
1294 #ifdef CONFIG_IP_MULTICAST
1295         if (im->multiaddr == IGMP_ALL_HOSTS)
1296                 return;
1297         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1298                 return;
1299
1300         reporter = im->reporter;
1301         igmp_stop_timer(im);
1302
1303         if (!in_dev->dead) {
1304                 if (IGMP_V1_SEEN(in_dev))
1305                         return;
1306                 if (IGMP_V2_SEEN(in_dev)) {
1307                         if (reporter)
1308                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1309                         return;
1310                 }
1311                 /* IGMPv3 */
1312                 igmpv3_add_delrec(in_dev, im, gfp);
1313
1314                 igmp_ifc_event(in_dev);
1315         }
1316 #endif
1317 }
1318
1319 static void igmp_group_dropped(struct ip_mc_list *im)
1320 {
1321         __igmp_group_dropped(im, GFP_KERNEL);
1322 }
1323
1324 static void igmp_group_added(struct ip_mc_list *im)
1325 {
1326         struct in_device *in_dev = im->interface;
1327 #ifdef CONFIG_IP_MULTICAST
1328         struct net *net = dev_net(in_dev->dev);
1329 #endif
1330
1331         if (im->loaded == 0) {
1332                 im->loaded = 1;
1333                 ip_mc_filter_add(in_dev, im->multiaddr);
1334         }
1335
1336 #ifdef CONFIG_IP_MULTICAST
1337         if (im->multiaddr == IGMP_ALL_HOSTS)
1338                 return;
1339         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1340                 return;
1341
1342         if (in_dev->dead)
1343                 return;
1344
1345         im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1346         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1347                 spin_lock_bh(&im->lock);
1348                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1349                 spin_unlock_bh(&im->lock);
1350                 return;
1351         }
1352         /* else, v3 */
1353
1354         /* Based on RFC3376 5.1, for newly added INCLUDE SSM, we should
1355          * not send filter-mode change record as the mode should be from
1356          * IN() to IN(A).
1357          */
1358         if (im->sfmode == MCAST_EXCLUDE)
1359                 im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1360
1361         igmp_ifc_event(in_dev);
1362 #endif
1363 }
1364
1365
1366 /*
1367  *      Multicast list managers
1368  */
1369
1370 static u32 ip_mc_hash(const struct ip_mc_list *im)
1371 {
1372         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1373 }
1374
1375 static void ip_mc_hash_add(struct in_device *in_dev,
1376                            struct ip_mc_list *im)
1377 {
1378         struct ip_mc_list __rcu **mc_hash;
1379         u32 hash;
1380
1381         mc_hash = rtnl_dereference(in_dev->mc_hash);
1382         if (mc_hash) {
1383                 hash = ip_mc_hash(im);
1384                 im->next_hash = mc_hash[hash];
1385                 rcu_assign_pointer(mc_hash[hash], im);
1386                 return;
1387         }
1388
1389         /* do not use a hash table for small number of items */
1390         if (in_dev->mc_count < 4)
1391                 return;
1392
1393         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1394                           GFP_KERNEL);
1395         if (!mc_hash)
1396                 return;
1397
1398         for_each_pmc_rtnl(in_dev, im) {
1399                 hash = ip_mc_hash(im);
1400                 im->next_hash = mc_hash[hash];
1401                 RCU_INIT_POINTER(mc_hash[hash], im);
1402         }
1403
1404         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1405 }
1406
1407 static void ip_mc_hash_remove(struct in_device *in_dev,
1408                               struct ip_mc_list *im)
1409 {
1410         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1411         struct ip_mc_list *aux;
1412
1413         if (!mc_hash)
1414                 return;
1415         mc_hash += ip_mc_hash(im);
1416         while ((aux = rtnl_dereference(*mc_hash)) != im)
1417                 mc_hash = &aux->next_hash;
1418         *mc_hash = im->next_hash;
1419 }
1420
1421
1422 /*
1423  *      A socket has joined a multicast group on device dev.
1424  */
1425 static void ____ip_mc_inc_group(struct in_device *in_dev, __be32 addr,
1426                                 unsigned int mode, gfp_t gfp)
1427 {
1428         struct ip_mc_list *im;
1429
1430         ASSERT_RTNL();
1431
1432         for_each_pmc_rtnl(in_dev, im) {
1433                 if (im->multiaddr == addr) {
1434                         im->users++;
1435                         ip_mc_add_src(in_dev, &addr, mode, 0, NULL, 0);
1436                         goto out;
1437                 }
1438         }
1439
1440         im = kzalloc(sizeof(*im), gfp);
1441         if (!im)
1442                 goto out;
1443
1444         im->users = 1;
1445         im->interface = in_dev;
1446         in_dev_hold(in_dev);
1447         im->multiaddr = addr;
1448         /* initial mode is (EX, empty) */
1449         im->sfmode = mode;
1450         im->sfcount[mode] = 1;
1451         refcount_set(&im->refcnt, 1);
1452         spin_lock_init(&im->lock);
1453 #ifdef CONFIG_IP_MULTICAST
1454         timer_setup(&im->timer, igmp_timer_expire, 0);
1455 #endif
1456
1457         im->next_rcu = in_dev->mc_list;
1458         in_dev->mc_count++;
1459         rcu_assign_pointer(in_dev->mc_list, im);
1460
1461         ip_mc_hash_add(in_dev, im);
1462
1463 #ifdef CONFIG_IP_MULTICAST
1464         igmpv3_del_delrec(in_dev, im);
1465 #endif
1466         igmp_group_added(im);
1467         if (!in_dev->dead)
1468                 ip_rt_multicast_event(in_dev);
1469 out:
1470         return;
1471 }
1472
1473 void __ip_mc_inc_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1474 {
1475         ____ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE, gfp);
1476 }
1477 EXPORT_SYMBOL(__ip_mc_inc_group);
1478
1479 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1480 {
1481         __ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE);
1482 }
1483 EXPORT_SYMBOL(ip_mc_inc_group);
1484
1485 static int ip_mc_check_iphdr(struct sk_buff *skb)
1486 {
1487         const struct iphdr *iph;
1488         unsigned int len;
1489         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1490
1491         if (!pskb_may_pull(skb, offset))
1492                 return -EINVAL;
1493
1494         iph = ip_hdr(skb);
1495
1496         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1497                 return -EINVAL;
1498
1499         offset += ip_hdrlen(skb) - sizeof(*iph);
1500
1501         if (!pskb_may_pull(skb, offset))
1502                 return -EINVAL;
1503
1504         iph = ip_hdr(skb);
1505
1506         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1507                 return -EINVAL;
1508
1509         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1510         if (skb->len < len || len < offset)
1511                 return -EINVAL;
1512
1513         skb_set_transport_header(skb, offset);
1514
1515         return 0;
1516 }
1517
1518 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1519 {
1520         unsigned int len = skb_transport_offset(skb);
1521
1522         len += sizeof(struct igmpv3_report);
1523
1524         return ip_mc_may_pull(skb, len) ? 0 : -EINVAL;
1525 }
1526
1527 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1528 {
1529         unsigned int transport_len = ip_transport_len(skb);
1530         unsigned int len;
1531
1532         /* IGMPv{1,2}? */
1533         if (transport_len != sizeof(struct igmphdr)) {
1534                 /* or IGMPv3? */
1535                 if (transport_len < sizeof(struct igmpv3_query))
1536                         return -EINVAL;
1537
1538                 len = skb_transport_offset(skb) + sizeof(struct igmpv3_query);
1539                 if (!ip_mc_may_pull(skb, len))
1540                         return -EINVAL;
1541         }
1542
1543         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1544          * all-systems destination addresses (224.0.0.1) for general queries
1545          */
1546         if (!igmp_hdr(skb)->group &&
1547             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1548                 return -EINVAL;
1549
1550         return 0;
1551 }
1552
1553 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1554 {
1555         switch (igmp_hdr(skb)->type) {
1556         case IGMP_HOST_LEAVE_MESSAGE:
1557         case IGMP_HOST_MEMBERSHIP_REPORT:
1558         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1559                 return 0;
1560         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1561                 return ip_mc_check_igmp_reportv3(skb);
1562         case IGMP_HOST_MEMBERSHIP_QUERY:
1563                 return ip_mc_check_igmp_query(skb);
1564         default:
1565                 return -ENOMSG;
1566         }
1567 }
1568
1569 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1570 {
1571         return skb_checksum_simple_validate(skb);
1572 }
1573
1574 static int ip_mc_check_igmp_csum(struct sk_buff *skb)
1575 {
1576         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1577         unsigned int transport_len = ip_transport_len(skb);
1578         struct sk_buff *skb_chk;
1579
1580         if (!ip_mc_may_pull(skb, len))
1581                 return -EINVAL;
1582
1583         skb_chk = skb_checksum_trimmed(skb, transport_len,
1584                                        ip_mc_validate_checksum);
1585         if (!skb_chk)
1586                 return -EINVAL;
1587
1588         if (skb_chk != skb)
1589                 kfree_skb(skb_chk);
1590
1591         return 0;
1592 }
1593
1594 /**
1595  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1596  * @skb: the skb to validate
1597  *
1598  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1599  * skb transport header accordingly and returns zero.
1600  *
1601  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1602  *  standard
1603  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1604  * -ENOMEM: A memory allocation failure happened.
1605  *
1606  * Caller needs to set the skb network header and free any returned skb if it
1607  * differs from the provided skb.
1608  */
1609 int ip_mc_check_igmp(struct sk_buff *skb)
1610 {
1611         int ret = ip_mc_check_iphdr(skb);
1612
1613         if (ret < 0)
1614                 return ret;
1615
1616         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1617                 return -ENOMSG;
1618
1619         ret = ip_mc_check_igmp_csum(skb);
1620         if (ret < 0)
1621                 return ret;
1622
1623         return ip_mc_check_igmp_msg(skb);
1624 }
1625 EXPORT_SYMBOL(ip_mc_check_igmp);
1626
1627 /*
1628  *      Resend IGMP JOIN report; used by netdev notifier.
1629  */
1630 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1631 {
1632 #ifdef CONFIG_IP_MULTICAST
1633         struct ip_mc_list *im;
1634         int type;
1635         struct net *net = dev_net(in_dev->dev);
1636
1637         ASSERT_RTNL();
1638
1639         for_each_pmc_rtnl(in_dev, im) {
1640                 if (im->multiaddr == IGMP_ALL_HOSTS)
1641                         continue;
1642                 if (ipv4_is_local_multicast(im->multiaddr) &&
1643                     !net->ipv4.sysctl_igmp_llm_reports)
1644                         continue;
1645
1646                 /* a failover is happening and switches
1647                  * must be notified immediately
1648                  */
1649                 if (IGMP_V1_SEEN(in_dev))
1650                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1651                 else if (IGMP_V2_SEEN(in_dev))
1652                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1653                 else
1654                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1655                 igmp_send_report(in_dev, im, type);
1656         }
1657 #endif
1658 }
1659
1660 /*
1661  *      A socket has left a multicast group on device dev
1662  */
1663
1664 void __ip_mc_dec_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1665 {
1666         struct ip_mc_list *i;
1667         struct ip_mc_list __rcu **ip;
1668
1669         ASSERT_RTNL();
1670
1671         for (ip = &in_dev->mc_list;
1672              (i = rtnl_dereference(*ip)) != NULL;
1673              ip = &i->next_rcu) {
1674                 if (i->multiaddr == addr) {
1675                         if (--i->users == 0) {
1676                                 ip_mc_hash_remove(in_dev, i);
1677                                 *ip = i->next_rcu;
1678                                 in_dev->mc_count--;
1679                                 __igmp_group_dropped(i, gfp);
1680                                 ip_mc_clear_src(i);
1681
1682                                 if (!in_dev->dead)
1683                                         ip_rt_multicast_event(in_dev);
1684
1685                                 ip_ma_put(i);
1686                                 return;
1687                         }
1688                         break;
1689                 }
1690         }
1691 }
1692 EXPORT_SYMBOL(__ip_mc_dec_group);
1693
1694 /* Device changing type */
1695
1696 void ip_mc_unmap(struct in_device *in_dev)
1697 {
1698         struct ip_mc_list *pmc;
1699
1700         ASSERT_RTNL();
1701
1702         for_each_pmc_rtnl(in_dev, pmc)
1703                 igmp_group_dropped(pmc);
1704 }
1705
1706 void ip_mc_remap(struct in_device *in_dev)
1707 {
1708         struct ip_mc_list *pmc;
1709
1710         ASSERT_RTNL();
1711
1712         for_each_pmc_rtnl(in_dev, pmc) {
1713 #ifdef CONFIG_IP_MULTICAST
1714                 igmpv3_del_delrec(in_dev, pmc);
1715 #endif
1716                 igmp_group_added(pmc);
1717         }
1718 }
1719
1720 /* Device going down */
1721
1722 void ip_mc_down(struct in_device *in_dev)
1723 {
1724         struct ip_mc_list *pmc;
1725
1726         ASSERT_RTNL();
1727
1728         for_each_pmc_rtnl(in_dev, pmc)
1729                 igmp_group_dropped(pmc);
1730
1731 #ifdef CONFIG_IP_MULTICAST
1732         in_dev->mr_ifc_count = 0;
1733         if (del_timer(&in_dev->mr_ifc_timer))
1734                 __in_dev_put(in_dev);
1735         in_dev->mr_gq_running = 0;
1736         if (del_timer(&in_dev->mr_gq_timer))
1737                 __in_dev_put(in_dev);
1738 #endif
1739
1740         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1741 }
1742
1743 #ifdef CONFIG_IP_MULTICAST
1744 static void ip_mc_reset(struct in_device *in_dev)
1745 {
1746         struct net *net = dev_net(in_dev->dev);
1747
1748         in_dev->mr_qi = IGMP_QUERY_INTERVAL;
1749         in_dev->mr_qri = IGMP_QUERY_RESPONSE_INTERVAL;
1750         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1751 }
1752 #else
1753 static void ip_mc_reset(struct in_device *in_dev)
1754 {
1755 }
1756 #endif
1757
1758 void ip_mc_init_dev(struct in_device *in_dev)
1759 {
1760         ASSERT_RTNL();
1761
1762 #ifdef CONFIG_IP_MULTICAST
1763         timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1764         timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1765 #endif
1766         ip_mc_reset(in_dev);
1767
1768         spin_lock_init(&in_dev->mc_tomb_lock);
1769 }
1770
1771 /* Device going up */
1772
1773 void ip_mc_up(struct in_device *in_dev)
1774 {
1775         struct ip_mc_list *pmc;
1776
1777         ASSERT_RTNL();
1778
1779         ip_mc_reset(in_dev);
1780         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1781
1782         for_each_pmc_rtnl(in_dev, pmc) {
1783 #ifdef CONFIG_IP_MULTICAST
1784                 igmpv3_del_delrec(in_dev, pmc);
1785 #endif
1786                 igmp_group_added(pmc);
1787         }
1788 }
1789
1790 /*
1791  *      Device is about to be destroyed: clean up.
1792  */
1793
1794 void ip_mc_destroy_dev(struct in_device *in_dev)
1795 {
1796         struct ip_mc_list *i;
1797
1798         ASSERT_RTNL();
1799
1800         /* Deactivate timers */
1801         ip_mc_down(in_dev);
1802 #ifdef CONFIG_IP_MULTICAST
1803         igmpv3_clear_delrec(in_dev);
1804 #endif
1805
1806         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1807                 in_dev->mc_list = i->next_rcu;
1808                 in_dev->mc_count--;
1809                 ip_ma_put(i);
1810         }
1811 }
1812
1813 /* RTNL is locked */
1814 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1815 {
1816         struct net_device *dev = NULL;
1817         struct in_device *idev = NULL;
1818
1819         if (imr->imr_ifindex) {
1820                 idev = inetdev_by_index(net, imr->imr_ifindex);
1821                 return idev;
1822         }
1823         if (imr->imr_address.s_addr) {
1824                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1825                 if (!dev)
1826                         return NULL;
1827         }
1828
1829         if (!dev) {
1830                 struct rtable *rt = ip_route_output(net,
1831                                                     imr->imr_multiaddr.s_addr,
1832                                                     0, 0, 0);
1833                 if (!IS_ERR(rt)) {
1834                         dev = rt->dst.dev;
1835                         ip_rt_put(rt);
1836                 }
1837         }
1838         if (dev) {
1839                 imr->imr_ifindex = dev->ifindex;
1840                 idev = __in_dev_get_rtnl(dev);
1841         }
1842         return idev;
1843 }
1844
1845 /*
1846  *      Join a socket to a group
1847  */
1848
1849 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1850         __be32 *psfsrc)
1851 {
1852         struct ip_sf_list *psf, *psf_prev;
1853         int rv = 0;
1854
1855         psf_prev = NULL;
1856         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1857                 if (psf->sf_inaddr == *psfsrc)
1858                         break;
1859                 psf_prev = psf;
1860         }
1861         if (!psf || psf->sf_count[sfmode] == 0) {
1862                 /* source filter not found, or count wrong =>  bug */
1863                 return -ESRCH;
1864         }
1865         psf->sf_count[sfmode]--;
1866         if (psf->sf_count[sfmode] == 0) {
1867                 ip_rt_multicast_event(pmc->interface);
1868         }
1869         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1870 #ifdef CONFIG_IP_MULTICAST
1871                 struct in_device *in_dev = pmc->interface;
1872                 struct net *net = dev_net(in_dev->dev);
1873 #endif
1874
1875                 /* no more filters for this source */
1876                 if (psf_prev)
1877                         psf_prev->sf_next = psf->sf_next;
1878                 else
1879                         pmc->sources = psf->sf_next;
1880 #ifdef CONFIG_IP_MULTICAST
1881                 if (psf->sf_oldin &&
1882                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1883                         psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1884                         psf->sf_next = pmc->tomb;
1885                         pmc->tomb = psf;
1886                         rv = 1;
1887                 } else
1888 #endif
1889                         kfree(psf);
1890         }
1891         return rv;
1892 }
1893
1894 #ifndef CONFIG_IP_MULTICAST
1895 #define igmp_ifc_event(x)       do { } while (0)
1896 #endif
1897
1898 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1899                          int sfcount, __be32 *psfsrc, int delta)
1900 {
1901         struct ip_mc_list *pmc;
1902         int     changerec = 0;
1903         int     i, err;
1904
1905         if (!in_dev)
1906                 return -ENODEV;
1907         rcu_read_lock();
1908         for_each_pmc_rcu(in_dev, pmc) {
1909                 if (*pmca == pmc->multiaddr)
1910                         break;
1911         }
1912         if (!pmc) {
1913                 /* MCA not found?? bug */
1914                 rcu_read_unlock();
1915                 return -ESRCH;
1916         }
1917         spin_lock_bh(&pmc->lock);
1918         rcu_read_unlock();
1919 #ifdef CONFIG_IP_MULTICAST
1920         sf_markstate(pmc);
1921 #endif
1922         if (!delta) {
1923                 err = -EINVAL;
1924                 if (!pmc->sfcount[sfmode])
1925                         goto out_unlock;
1926                 pmc->sfcount[sfmode]--;
1927         }
1928         err = 0;
1929         for (i = 0; i < sfcount; i++) {
1930                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1931
1932                 changerec |= rv > 0;
1933                 if (!err && rv < 0)
1934                         err = rv;
1935         }
1936         if (pmc->sfmode == MCAST_EXCLUDE &&
1937             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1938             pmc->sfcount[MCAST_INCLUDE]) {
1939 #ifdef CONFIG_IP_MULTICAST
1940                 struct ip_sf_list *psf;
1941                 struct net *net = dev_net(in_dev->dev);
1942 #endif
1943
1944                 /* filter mode change */
1945                 pmc->sfmode = MCAST_INCLUDE;
1946 #ifdef CONFIG_IP_MULTICAST
1947                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1948                 in_dev->mr_ifc_count = pmc->crcount;
1949                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1950                         psf->sf_crcount = 0;
1951                 igmp_ifc_event(pmc->interface);
1952         } else if (sf_setstate(pmc) || changerec) {
1953                 igmp_ifc_event(pmc->interface);
1954 #endif
1955         }
1956 out_unlock:
1957         spin_unlock_bh(&pmc->lock);
1958         return err;
1959 }
1960
1961 /*
1962  * Add multicast single-source filter to the interface list
1963  */
1964 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1965         __be32 *psfsrc)
1966 {
1967         struct ip_sf_list *psf, *psf_prev;
1968
1969         psf_prev = NULL;
1970         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1971                 if (psf->sf_inaddr == *psfsrc)
1972                         break;
1973                 psf_prev = psf;
1974         }
1975         if (!psf) {
1976                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1977                 if (!psf)
1978                         return -ENOBUFS;
1979                 psf->sf_inaddr = *psfsrc;
1980                 if (psf_prev) {
1981                         psf_prev->sf_next = psf;
1982                 } else
1983                         pmc->sources = psf;
1984         }
1985         psf->sf_count[sfmode]++;
1986         if (psf->sf_count[sfmode] == 1) {
1987                 ip_rt_multicast_event(pmc->interface);
1988         }
1989         return 0;
1990 }
1991
1992 #ifdef CONFIG_IP_MULTICAST
1993 static void sf_markstate(struct ip_mc_list *pmc)
1994 {
1995         struct ip_sf_list *psf;
1996         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1997
1998         for (psf = pmc->sources; psf; psf = psf->sf_next)
1999                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2000                         psf->sf_oldin = mca_xcount ==
2001                                 psf->sf_count[MCAST_EXCLUDE] &&
2002                                 !psf->sf_count[MCAST_INCLUDE];
2003                 } else
2004                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
2005 }
2006
2007 static int sf_setstate(struct ip_mc_list *pmc)
2008 {
2009         struct ip_sf_list *psf, *dpsf;
2010         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2011         int qrv = pmc->interface->mr_qrv;
2012         int new_in, rv;
2013
2014         rv = 0;
2015         for (psf = pmc->sources; psf; psf = psf->sf_next) {
2016                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2017                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
2018                                 !psf->sf_count[MCAST_INCLUDE];
2019                 } else
2020                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
2021                 if (new_in) {
2022                         if (!psf->sf_oldin) {
2023                                 struct ip_sf_list *prev = NULL;
2024
2025                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2026                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2027                                                 break;
2028                                         prev = dpsf;
2029                                 }
2030                                 if (dpsf) {
2031                                         if (prev)
2032                                                 prev->sf_next = dpsf->sf_next;
2033                                         else
2034                                                 pmc->tomb = dpsf->sf_next;
2035                                         kfree(dpsf);
2036                                 }
2037                                 psf->sf_crcount = qrv;
2038                                 rv++;
2039                         }
2040                 } else if (psf->sf_oldin) {
2041
2042                         psf->sf_crcount = 0;
2043                         /*
2044                          * add or update "delete" records if an active filter
2045                          * is now inactive
2046                          */
2047                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2048                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2049                                         break;
2050                         if (!dpsf) {
2051                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2052                                 if (!dpsf)
2053                                         continue;
2054                                 *dpsf = *psf;
2055                                 /* pmc->lock held by callers */
2056                                 dpsf->sf_next = pmc->tomb;
2057                                 pmc->tomb = dpsf;
2058                         }
2059                         dpsf->sf_crcount = qrv;
2060                         rv++;
2061                 }
2062         }
2063         return rv;
2064 }
2065 #endif
2066
2067 /*
2068  * Add multicast source filter list to the interface list
2069  */
2070 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2071                          int sfcount, __be32 *psfsrc, int delta)
2072 {
2073         struct ip_mc_list *pmc;
2074         int     isexclude;
2075         int     i, err;
2076
2077         if (!in_dev)
2078                 return -ENODEV;
2079         rcu_read_lock();
2080         for_each_pmc_rcu(in_dev, pmc) {
2081                 if (*pmca == pmc->multiaddr)
2082                         break;
2083         }
2084         if (!pmc) {
2085                 /* MCA not found?? bug */
2086                 rcu_read_unlock();
2087                 return -ESRCH;
2088         }
2089         spin_lock_bh(&pmc->lock);
2090         rcu_read_unlock();
2091
2092 #ifdef CONFIG_IP_MULTICAST
2093         sf_markstate(pmc);
2094 #endif
2095         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2096         if (!delta)
2097                 pmc->sfcount[sfmode]++;
2098         err = 0;
2099         for (i = 0; i < sfcount; i++) {
2100                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2101                 if (err)
2102                         break;
2103         }
2104         if (err) {
2105                 int j;
2106
2107                 if (!delta)
2108                         pmc->sfcount[sfmode]--;
2109                 for (j = 0; j < i; j++)
2110                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2111         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2112 #ifdef CONFIG_IP_MULTICAST
2113                 struct ip_sf_list *psf;
2114                 struct net *net = dev_net(pmc->interface->dev);
2115                 in_dev = pmc->interface;
2116 #endif
2117
2118                 /* filter mode change */
2119                 if (pmc->sfcount[MCAST_EXCLUDE])
2120                         pmc->sfmode = MCAST_EXCLUDE;
2121                 else if (pmc->sfcount[MCAST_INCLUDE])
2122                         pmc->sfmode = MCAST_INCLUDE;
2123 #ifdef CONFIG_IP_MULTICAST
2124                 /* else no filters; keep old mode for reports */
2125
2126                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2127                 in_dev->mr_ifc_count = pmc->crcount;
2128                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2129                         psf->sf_crcount = 0;
2130                 igmp_ifc_event(in_dev);
2131         } else if (sf_setstate(pmc)) {
2132                 igmp_ifc_event(in_dev);
2133 #endif
2134         }
2135         spin_unlock_bh(&pmc->lock);
2136         return err;
2137 }
2138
2139 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2140 {
2141         struct ip_sf_list *tomb, *sources;
2142
2143         spin_lock_bh(&pmc->lock);
2144         tomb = pmc->tomb;
2145         pmc->tomb = NULL;
2146         sources = pmc->sources;
2147         pmc->sources = NULL;
2148         pmc->sfmode = MCAST_EXCLUDE;
2149         pmc->sfcount[MCAST_INCLUDE] = 0;
2150         pmc->sfcount[MCAST_EXCLUDE] = 1;
2151         spin_unlock_bh(&pmc->lock);
2152
2153         ip_sf_list_clear_all(tomb);
2154         ip_sf_list_clear_all(sources);
2155 }
2156
2157 /* Join a multicast group
2158  */
2159 static int __ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr,
2160                               unsigned int mode)
2161 {
2162         __be32 addr = imr->imr_multiaddr.s_addr;
2163         struct ip_mc_socklist *iml, *i;
2164         struct in_device *in_dev;
2165         struct inet_sock *inet = inet_sk(sk);
2166         struct net *net = sock_net(sk);
2167         int ifindex;
2168         int count = 0;
2169         int err;
2170
2171         ASSERT_RTNL();
2172
2173         if (!ipv4_is_multicast(addr))
2174                 return -EINVAL;
2175
2176         in_dev = ip_mc_find_dev(net, imr);
2177
2178         if (!in_dev) {
2179                 err = -ENODEV;
2180                 goto done;
2181         }
2182
2183         err = -EADDRINUSE;
2184         ifindex = imr->imr_ifindex;
2185         for_each_pmc_rtnl(inet, i) {
2186                 if (i->multi.imr_multiaddr.s_addr == addr &&
2187                     i->multi.imr_ifindex == ifindex)
2188                         goto done;
2189                 count++;
2190         }
2191         err = -ENOBUFS;
2192         if (count >= net->ipv4.sysctl_igmp_max_memberships)
2193                 goto done;
2194         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2195         if (!iml)
2196                 goto done;
2197
2198         memcpy(&iml->multi, imr, sizeof(*imr));
2199         iml->next_rcu = inet->mc_list;
2200         iml->sflist = NULL;
2201         iml->sfmode = mode;
2202         rcu_assign_pointer(inet->mc_list, iml);
2203         __ip_mc_inc_group(in_dev, addr, mode);
2204         err = 0;
2205 done:
2206         return err;
2207 }
2208
2209 /* Join ASM (Any-Source Multicast) group
2210  */
2211 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2212 {
2213         return __ip_mc_join_group(sk, imr, MCAST_EXCLUDE);
2214 }
2215 EXPORT_SYMBOL(ip_mc_join_group);
2216
2217 /* Join SSM (Source-Specific Multicast) group
2218  */
2219 int ip_mc_join_group_ssm(struct sock *sk, struct ip_mreqn *imr,
2220                          unsigned int mode)
2221 {
2222         return __ip_mc_join_group(sk, imr, mode);
2223 }
2224
2225 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2226                            struct in_device *in_dev)
2227 {
2228         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2229         int err;
2230
2231         if (!psf) {
2232                 /* any-source empty exclude case */
2233                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2234                         iml->sfmode, 0, NULL, 0);
2235         }
2236         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2237                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2238         RCU_INIT_POINTER(iml->sflist, NULL);
2239         /* decrease mem now to avoid the memleak warning */
2240         atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2241         kfree_rcu(psf, rcu);
2242         return err;
2243 }
2244
2245 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2246 {
2247         struct inet_sock *inet = inet_sk(sk);
2248         struct ip_mc_socklist *iml;
2249         struct ip_mc_socklist __rcu **imlp;
2250         struct in_device *in_dev;
2251         struct net *net = sock_net(sk);
2252         __be32 group = imr->imr_multiaddr.s_addr;
2253         u32 ifindex;
2254         int ret = -EADDRNOTAVAIL;
2255
2256         ASSERT_RTNL();
2257
2258         in_dev = ip_mc_find_dev(net, imr);
2259         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2260                 ret = -ENODEV;
2261                 goto out;
2262         }
2263         ifindex = imr->imr_ifindex;
2264         for (imlp = &inet->mc_list;
2265              (iml = rtnl_dereference(*imlp)) != NULL;
2266              imlp = &iml->next_rcu) {
2267                 if (iml->multi.imr_multiaddr.s_addr != group)
2268                         continue;
2269                 if (ifindex) {
2270                         if (iml->multi.imr_ifindex != ifindex)
2271                                 continue;
2272                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2273                                 iml->multi.imr_address.s_addr)
2274                         continue;
2275
2276                 (void) ip_mc_leave_src(sk, iml, in_dev);
2277
2278                 *imlp = iml->next_rcu;
2279
2280                 if (in_dev)
2281                         ip_mc_dec_group(in_dev, group);
2282
2283                 /* decrease mem now to avoid the memleak warning */
2284                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2285                 kfree_rcu(iml, rcu);
2286                 return 0;
2287         }
2288 out:
2289         return ret;
2290 }
2291 EXPORT_SYMBOL(ip_mc_leave_group);
2292
2293 int ip_mc_source(int add, int omode, struct sock *sk, struct
2294         ip_mreq_source *mreqs, int ifindex)
2295 {
2296         int err;
2297         struct ip_mreqn imr;
2298         __be32 addr = mreqs->imr_multiaddr;
2299         struct ip_mc_socklist *pmc;
2300         struct in_device *in_dev = NULL;
2301         struct inet_sock *inet = inet_sk(sk);
2302         struct ip_sf_socklist *psl;
2303         struct net *net = sock_net(sk);
2304         int leavegroup = 0;
2305         int i, j, rv;
2306
2307         if (!ipv4_is_multicast(addr))
2308                 return -EINVAL;
2309
2310         ASSERT_RTNL();
2311
2312         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2313         imr.imr_address.s_addr = mreqs->imr_interface;
2314         imr.imr_ifindex = ifindex;
2315         in_dev = ip_mc_find_dev(net, &imr);
2316
2317         if (!in_dev) {
2318                 err = -ENODEV;
2319                 goto done;
2320         }
2321         err = -EADDRNOTAVAIL;
2322
2323         for_each_pmc_rtnl(inet, pmc) {
2324                 if ((pmc->multi.imr_multiaddr.s_addr ==
2325                      imr.imr_multiaddr.s_addr) &&
2326                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2327                         break;
2328         }
2329         if (!pmc) {             /* must have a prior join */
2330                 err = -EINVAL;
2331                 goto done;
2332         }
2333         /* if a source filter was set, must be the same mode as before */
2334         if (pmc->sflist) {
2335                 if (pmc->sfmode != omode) {
2336                         err = -EINVAL;
2337                         goto done;
2338                 }
2339         } else if (pmc->sfmode != omode) {
2340                 /* allow mode switches for empty-set filters */
2341                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2342                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2343                         NULL, 0);
2344                 pmc->sfmode = omode;
2345         }
2346
2347         psl = rtnl_dereference(pmc->sflist);
2348         if (!add) {
2349                 if (!psl)
2350                         goto done;      /* err = -EADDRNOTAVAIL */
2351                 rv = !0;
2352                 for (i = 0; i < psl->sl_count; i++) {
2353                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2354                                 sizeof(__be32));
2355                         if (rv == 0)
2356                                 break;
2357                 }
2358                 if (rv)         /* source not found */
2359                         goto done;      /* err = -EADDRNOTAVAIL */
2360
2361                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2362                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2363                         leavegroup = 1;
2364                         goto done;
2365                 }
2366
2367                 /* update the interface filter */
2368                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2369                         &mreqs->imr_sourceaddr, 1);
2370
2371                 for (j = i+1; j < psl->sl_count; j++)
2372                         psl->sl_addr[j-1] = psl->sl_addr[j];
2373                 psl->sl_count--;
2374                 err = 0;
2375                 goto done;
2376         }
2377         /* else, add a new source to the filter */
2378
2379         if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2380                 err = -ENOBUFS;
2381                 goto done;
2382         }
2383         if (!psl || psl->sl_count == psl->sl_max) {
2384                 struct ip_sf_socklist *newpsl;
2385                 int count = IP_SFBLOCK;
2386
2387                 if (psl)
2388                         count += psl->sl_max;
2389                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2390                 if (!newpsl) {
2391                         err = -ENOBUFS;
2392                         goto done;
2393                 }
2394                 newpsl->sl_max = count;
2395                 newpsl->sl_count = count - IP_SFBLOCK;
2396                 if (psl) {
2397                         for (i = 0; i < psl->sl_count; i++)
2398                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2399                         /* decrease mem now to avoid the memleak warning */
2400                         atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2401                         kfree_rcu(psl, rcu);
2402                 }
2403                 rcu_assign_pointer(pmc->sflist, newpsl);
2404                 psl = newpsl;
2405         }
2406         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2407         for (i = 0; i < psl->sl_count; i++) {
2408                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2409                         sizeof(__be32));
2410                 if (rv == 0)
2411                         break;
2412         }
2413         if (rv == 0)            /* address already there is an error */
2414                 goto done;
2415         for (j = psl->sl_count-1; j >= i; j--)
2416                 psl->sl_addr[j+1] = psl->sl_addr[j];
2417         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2418         psl->sl_count++;
2419         err = 0;
2420         /* update the interface list */
2421         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2422                 &mreqs->imr_sourceaddr, 1);
2423 done:
2424         if (leavegroup)
2425                 err = ip_mc_leave_group(sk, &imr);
2426         return err;
2427 }
2428
2429 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2430 {
2431         int err = 0;
2432         struct ip_mreqn imr;
2433         __be32 addr = msf->imsf_multiaddr;
2434         struct ip_mc_socklist *pmc;
2435         struct in_device *in_dev;
2436         struct inet_sock *inet = inet_sk(sk);
2437         struct ip_sf_socklist *newpsl, *psl;
2438         struct net *net = sock_net(sk);
2439         int leavegroup = 0;
2440
2441         if (!ipv4_is_multicast(addr))
2442                 return -EINVAL;
2443         if (msf->imsf_fmode != MCAST_INCLUDE &&
2444             msf->imsf_fmode != MCAST_EXCLUDE)
2445                 return -EINVAL;
2446
2447         ASSERT_RTNL();
2448
2449         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2450         imr.imr_address.s_addr = msf->imsf_interface;
2451         imr.imr_ifindex = ifindex;
2452         in_dev = ip_mc_find_dev(net, &imr);
2453
2454         if (!in_dev) {
2455                 err = -ENODEV;
2456                 goto done;
2457         }
2458
2459         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2460         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2461                 leavegroup = 1;
2462                 goto done;
2463         }
2464
2465         for_each_pmc_rtnl(inet, pmc) {
2466                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2467                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2468                         break;
2469         }
2470         if (!pmc) {             /* must have a prior join */
2471                 err = -EINVAL;
2472                 goto done;
2473         }
2474         if (msf->imsf_numsrc) {
2475                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2476                                                            GFP_KERNEL);
2477                 if (!newpsl) {
2478                         err = -ENOBUFS;
2479                         goto done;
2480                 }
2481                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2482                 memcpy(newpsl->sl_addr, msf->imsf_slist,
2483                         msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2484                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2485                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2486                 if (err) {
2487                         sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2488                         goto done;
2489                 }
2490         } else {
2491                 newpsl = NULL;
2492                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2493                                      msf->imsf_fmode, 0, NULL, 0);
2494         }
2495         psl = rtnl_dereference(pmc->sflist);
2496         if (psl) {
2497                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2498                         psl->sl_count, psl->sl_addr, 0);
2499                 /* decrease mem now to avoid the memleak warning */
2500                 atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2501                 kfree_rcu(psl, rcu);
2502         } else
2503                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2504                         0, NULL, 0);
2505         rcu_assign_pointer(pmc->sflist, newpsl);
2506         pmc->sfmode = msf->imsf_fmode;
2507         err = 0;
2508 done:
2509         if (leavegroup)
2510                 err = ip_mc_leave_group(sk, &imr);
2511         return err;
2512 }
2513
2514 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2515         struct ip_msfilter __user *optval, int __user *optlen)
2516 {
2517         int err, len, count, copycount;
2518         struct ip_mreqn imr;
2519         __be32 addr = msf->imsf_multiaddr;
2520         struct ip_mc_socklist *pmc;
2521         struct in_device *in_dev;
2522         struct inet_sock *inet = inet_sk(sk);
2523         struct ip_sf_socklist *psl;
2524         struct net *net = sock_net(sk);
2525
2526         ASSERT_RTNL();
2527
2528         if (!ipv4_is_multicast(addr))
2529                 return -EINVAL;
2530
2531         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2532         imr.imr_address.s_addr = msf->imsf_interface;
2533         imr.imr_ifindex = 0;
2534         in_dev = ip_mc_find_dev(net, &imr);
2535
2536         if (!in_dev) {
2537                 err = -ENODEV;
2538                 goto done;
2539         }
2540         err = -EADDRNOTAVAIL;
2541
2542         for_each_pmc_rtnl(inet, pmc) {
2543                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2544                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2545                         break;
2546         }
2547         if (!pmc)               /* must have a prior join */
2548                 goto done;
2549         msf->imsf_fmode = pmc->sfmode;
2550         psl = rtnl_dereference(pmc->sflist);
2551         if (!psl) {
2552                 len = 0;
2553                 count = 0;
2554         } else {
2555                 count = psl->sl_count;
2556         }
2557         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2558         len = copycount * sizeof(psl->sl_addr[0]);
2559         msf->imsf_numsrc = count;
2560         if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2561             copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2562                 return -EFAULT;
2563         }
2564         if (len &&
2565             copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2566                 return -EFAULT;
2567         return 0;
2568 done:
2569         return err;
2570 }
2571
2572 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2573         struct group_filter __user *optval, int __user *optlen)
2574 {
2575         int err, i, count, copycount;
2576         struct sockaddr_in *psin;
2577         __be32 addr;
2578         struct ip_mc_socklist *pmc;
2579         struct inet_sock *inet = inet_sk(sk);
2580         struct ip_sf_socklist *psl;
2581
2582         ASSERT_RTNL();
2583
2584         psin = (struct sockaddr_in *)&gsf->gf_group;
2585         if (psin->sin_family != AF_INET)
2586                 return -EINVAL;
2587         addr = psin->sin_addr.s_addr;
2588         if (!ipv4_is_multicast(addr))
2589                 return -EINVAL;
2590
2591         err = -EADDRNOTAVAIL;
2592
2593         for_each_pmc_rtnl(inet, pmc) {
2594                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2595                     pmc->multi.imr_ifindex == gsf->gf_interface)
2596                         break;
2597         }
2598         if (!pmc)               /* must have a prior join */
2599                 goto done;
2600         gsf->gf_fmode = pmc->sfmode;
2601         psl = rtnl_dereference(pmc->sflist);
2602         count = psl ? psl->sl_count : 0;
2603         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2604         gsf->gf_numsrc = count;
2605         if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2606             copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2607                 return -EFAULT;
2608         }
2609         for (i = 0; i < copycount; i++) {
2610                 struct sockaddr_storage ss;
2611
2612                 psin = (struct sockaddr_in *)&ss;
2613                 memset(&ss, 0, sizeof(ss));
2614                 psin->sin_family = AF_INET;
2615                 psin->sin_addr.s_addr = psl->sl_addr[i];
2616                 if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2617                         return -EFAULT;
2618         }
2619         return 0;
2620 done:
2621         return err;
2622 }
2623
2624 /*
2625  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2626  */
2627 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2628                    int dif, int sdif)
2629 {
2630         struct inet_sock *inet = inet_sk(sk);
2631         struct ip_mc_socklist *pmc;
2632         struct ip_sf_socklist *psl;
2633         int i;
2634         int ret;
2635
2636         ret = 1;
2637         if (!ipv4_is_multicast(loc_addr))
2638                 goto out;
2639
2640         rcu_read_lock();
2641         for_each_pmc_rcu(inet, pmc) {
2642                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2643                     (pmc->multi.imr_ifindex == dif ||
2644                      (sdif && pmc->multi.imr_ifindex == sdif)))
2645                         break;
2646         }
2647         ret = inet->mc_all;
2648         if (!pmc)
2649                 goto unlock;
2650         psl = rcu_dereference(pmc->sflist);
2651         ret = (pmc->sfmode == MCAST_EXCLUDE);
2652         if (!psl)
2653                 goto unlock;
2654
2655         for (i = 0; i < psl->sl_count; i++) {
2656                 if (psl->sl_addr[i] == rmt_addr)
2657                         break;
2658         }
2659         ret = 0;
2660         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2661                 goto unlock;
2662         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2663                 goto unlock;
2664         ret = 1;
2665 unlock:
2666         rcu_read_unlock();
2667 out:
2668         return ret;
2669 }
2670
2671 /*
2672  *      A socket is closing.
2673  */
2674
2675 void ip_mc_drop_socket(struct sock *sk)
2676 {
2677         struct inet_sock *inet = inet_sk(sk);
2678         struct ip_mc_socklist *iml;
2679         struct net *net = sock_net(sk);
2680
2681         if (!inet->mc_list)
2682                 return;
2683
2684         rtnl_lock();
2685         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2686                 struct in_device *in_dev;
2687
2688                 inet->mc_list = iml->next_rcu;
2689                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2690                 (void) ip_mc_leave_src(sk, iml, in_dev);
2691                 if (in_dev)
2692                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2693                 /* decrease mem now to avoid the memleak warning */
2694                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2695                 kfree_rcu(iml, rcu);
2696         }
2697         rtnl_unlock();
2698 }
2699
2700 /* called with rcu_read_lock() */
2701 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2702 {
2703         struct ip_mc_list *im;
2704         struct ip_mc_list __rcu **mc_hash;
2705         struct ip_sf_list *psf;
2706         int rv = 0;
2707
2708         mc_hash = rcu_dereference(in_dev->mc_hash);
2709         if (mc_hash) {
2710                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2711
2712                 for (im = rcu_dereference(mc_hash[hash]);
2713                      im != NULL;
2714                      im = rcu_dereference(im->next_hash)) {
2715                         if (im->multiaddr == mc_addr)
2716                                 break;
2717                 }
2718         } else {
2719                 for_each_pmc_rcu(in_dev, im) {
2720                         if (im->multiaddr == mc_addr)
2721                                 break;
2722                 }
2723         }
2724         if (im && proto == IPPROTO_IGMP) {
2725                 rv = 1;
2726         } else if (im) {
2727                 if (src_addr) {
2728                         for (psf = im->sources; psf; psf = psf->sf_next) {
2729                                 if (psf->sf_inaddr == src_addr)
2730                                         break;
2731                         }
2732                         if (psf)
2733                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2734                                         psf->sf_count[MCAST_EXCLUDE] !=
2735                                         im->sfcount[MCAST_EXCLUDE];
2736                         else
2737                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2738                 } else
2739                         rv = 1; /* unspecified source; tentatively allow */
2740         }
2741         return rv;
2742 }
2743
2744 #if defined(CONFIG_PROC_FS)
2745 struct igmp_mc_iter_state {
2746         struct seq_net_private p;
2747         struct net_device *dev;
2748         struct in_device *in_dev;
2749 };
2750
2751 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2752
2753 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2754 {
2755         struct net *net = seq_file_net(seq);
2756         struct ip_mc_list *im = NULL;
2757         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2758
2759         state->in_dev = NULL;
2760         for_each_netdev_rcu(net, state->dev) {
2761                 struct in_device *in_dev;
2762
2763                 in_dev = __in_dev_get_rcu(state->dev);
2764                 if (!in_dev)
2765                         continue;
2766                 im = rcu_dereference(in_dev->mc_list);
2767                 if (im) {
2768                         state->in_dev = in_dev;
2769                         break;
2770                 }
2771         }
2772         return im;
2773 }
2774
2775 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2776 {
2777         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2778
2779         im = rcu_dereference(im->next_rcu);
2780         while (!im) {
2781                 state->dev = next_net_device_rcu(state->dev);
2782                 if (!state->dev) {
2783                         state->in_dev = NULL;
2784                         break;
2785                 }
2786                 state->in_dev = __in_dev_get_rcu(state->dev);
2787                 if (!state->in_dev)
2788                         continue;
2789                 im = rcu_dereference(state->in_dev->mc_list);
2790         }
2791         return im;
2792 }
2793
2794 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2795 {
2796         struct ip_mc_list *im = igmp_mc_get_first(seq);
2797         if (im)
2798                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2799                         --pos;
2800         return pos ? NULL : im;
2801 }
2802
2803 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2804         __acquires(rcu)
2805 {
2806         rcu_read_lock();
2807         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2808 }
2809
2810 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2811 {
2812         struct ip_mc_list *im;
2813         if (v == SEQ_START_TOKEN)
2814                 im = igmp_mc_get_first(seq);
2815         else
2816                 im = igmp_mc_get_next(seq, v);
2817         ++*pos;
2818         return im;
2819 }
2820
2821 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2822         __releases(rcu)
2823 {
2824         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2825
2826         state->in_dev = NULL;
2827         state->dev = NULL;
2828         rcu_read_unlock();
2829 }
2830
2831 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2832 {
2833         if (v == SEQ_START_TOKEN)
2834                 seq_puts(seq,
2835                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2836         else {
2837                 struct ip_mc_list *im = (struct ip_mc_list *)v;
2838                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2839                 char   *querier;
2840                 long delta;
2841
2842 #ifdef CONFIG_IP_MULTICAST
2843                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2844                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2845                           "V3";
2846 #else
2847                 querier = "NONE";
2848 #endif
2849
2850                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2851                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2852                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2853                 }
2854
2855                 delta = im->timer.expires - jiffies;
2856                 seq_printf(seq,
2857                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2858                            im->multiaddr, im->users,
2859                            im->tm_running,
2860                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2861                            im->reporter);
2862         }
2863         return 0;
2864 }
2865
2866 static const struct seq_operations igmp_mc_seq_ops = {
2867         .start  =       igmp_mc_seq_start,
2868         .next   =       igmp_mc_seq_next,
2869         .stop   =       igmp_mc_seq_stop,
2870         .show   =       igmp_mc_seq_show,
2871 };
2872
2873 struct igmp_mcf_iter_state {
2874         struct seq_net_private p;
2875         struct net_device *dev;
2876         struct in_device *idev;
2877         struct ip_mc_list *im;
2878 };
2879
2880 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2881
2882 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2883 {
2884         struct net *net = seq_file_net(seq);
2885         struct ip_sf_list *psf = NULL;
2886         struct ip_mc_list *im = NULL;
2887         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2888
2889         state->idev = NULL;
2890         state->im = NULL;
2891         for_each_netdev_rcu(net, state->dev) {
2892                 struct in_device *idev;
2893                 idev = __in_dev_get_rcu(state->dev);
2894                 if (unlikely(!idev))
2895                         continue;
2896                 im = rcu_dereference(idev->mc_list);
2897                 if (likely(im)) {
2898                         spin_lock_bh(&im->lock);
2899                         psf = im->sources;
2900                         if (likely(psf)) {
2901                                 state->im = im;
2902                                 state->idev = idev;
2903                                 break;
2904                         }
2905                         spin_unlock_bh(&im->lock);
2906                 }
2907         }
2908         return psf;
2909 }
2910
2911 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2912 {
2913         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2914
2915         psf = psf->sf_next;
2916         while (!psf) {
2917                 spin_unlock_bh(&state->im->lock);
2918                 state->im = state->im->next;
2919                 while (!state->im) {
2920                         state->dev = next_net_device_rcu(state->dev);
2921                         if (!state->dev) {
2922                                 state->idev = NULL;
2923                                 goto out;
2924                         }
2925                         state->idev = __in_dev_get_rcu(state->dev);
2926                         if (!state->idev)
2927                                 continue;
2928                         state->im = rcu_dereference(state->idev->mc_list);
2929                 }
2930                 if (!state->im)
2931                         break;
2932                 spin_lock_bh(&state->im->lock);
2933                 psf = state->im->sources;
2934         }
2935 out:
2936         return psf;
2937 }
2938
2939 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2940 {
2941         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2942         if (psf)
2943                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2944                         --pos;
2945         return pos ? NULL : psf;
2946 }
2947
2948 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2949         __acquires(rcu)
2950 {
2951         rcu_read_lock();
2952         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2953 }
2954
2955 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2956 {
2957         struct ip_sf_list *psf;
2958         if (v == SEQ_START_TOKEN)
2959                 psf = igmp_mcf_get_first(seq);
2960         else
2961                 psf = igmp_mcf_get_next(seq, v);
2962         ++*pos;
2963         return psf;
2964 }
2965
2966 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2967         __releases(rcu)
2968 {
2969         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2970         if (likely(state->im)) {
2971                 spin_unlock_bh(&state->im->lock);
2972                 state->im = NULL;
2973         }
2974         state->idev = NULL;
2975         state->dev = NULL;
2976         rcu_read_unlock();
2977 }
2978
2979 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2980 {
2981         struct ip_sf_list *psf = (struct ip_sf_list *)v;
2982         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2983
2984         if (v == SEQ_START_TOKEN) {
2985                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2986         } else {
2987                 seq_printf(seq,
2988                            "%3d %6.6s 0x%08x "
2989                            "0x%08x %6lu %6lu\n",
2990                            state->dev->ifindex, state->dev->name,
2991                            ntohl(state->im->multiaddr),
2992                            ntohl(psf->sf_inaddr),
2993                            psf->sf_count[MCAST_INCLUDE],
2994                            psf->sf_count[MCAST_EXCLUDE]);
2995         }
2996         return 0;
2997 }
2998
2999 static const struct seq_operations igmp_mcf_seq_ops = {
3000         .start  =       igmp_mcf_seq_start,
3001         .next   =       igmp_mcf_seq_next,
3002         .stop   =       igmp_mcf_seq_stop,
3003         .show   =       igmp_mcf_seq_show,
3004 };
3005
3006 static int __net_init igmp_net_init(struct net *net)
3007 {
3008         struct proc_dir_entry *pde;
3009         int err;
3010
3011         pde = proc_create_net("igmp", 0444, net->proc_net, &igmp_mc_seq_ops,
3012                         sizeof(struct igmp_mc_iter_state));
3013         if (!pde)
3014                 goto out_igmp;
3015         pde = proc_create_net("mcfilter", 0444, net->proc_net,
3016                         &igmp_mcf_seq_ops, sizeof(struct igmp_mcf_iter_state));
3017         if (!pde)
3018                 goto out_mcfilter;
3019         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3020                                    SOCK_DGRAM, 0, net);
3021         if (err < 0) {
3022                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3023                        err);
3024                 goto out_sock;
3025         }
3026
3027         return 0;
3028
3029 out_sock:
3030         remove_proc_entry("mcfilter", net->proc_net);
3031 out_mcfilter:
3032         remove_proc_entry("igmp", net->proc_net);
3033 out_igmp:
3034         return -ENOMEM;
3035 }
3036
3037 static void __net_exit igmp_net_exit(struct net *net)
3038 {
3039         remove_proc_entry("mcfilter", net->proc_net);
3040         remove_proc_entry("igmp", net->proc_net);
3041         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3042 }
3043
3044 static struct pernet_operations igmp_net_ops = {
3045         .init = igmp_net_init,
3046         .exit = igmp_net_exit,
3047 };
3048 #endif
3049
3050 static int igmp_netdev_event(struct notifier_block *this,
3051                              unsigned long event, void *ptr)
3052 {
3053         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3054         struct in_device *in_dev;
3055
3056         switch (event) {
3057         case NETDEV_RESEND_IGMP:
3058                 in_dev = __in_dev_get_rtnl(dev);
3059                 if (in_dev)
3060                         ip_mc_rejoin_groups(in_dev);
3061                 break;
3062         default:
3063                 break;
3064         }
3065         return NOTIFY_DONE;
3066 }
3067
3068 static struct notifier_block igmp_notifier = {
3069         .notifier_call = igmp_netdev_event,
3070 };
3071
3072 int __init igmp_mc_init(void)
3073 {
3074 #if defined(CONFIG_PROC_FS)
3075         int err;
3076
3077         err = register_pernet_subsys(&igmp_net_ops);
3078         if (err)
3079                 return err;
3080         err = register_netdevice_notifier(&igmp_notifier);
3081         if (err)
3082                 goto reg_notif_fail;
3083         return 0;
3084
3085 reg_notif_fail:
3086         unregister_pernet_subsys(&igmp_net_ops);
3087         return err;
3088 #else
3089         return register_netdevice_notifier(&igmp_notifier);
3090 #endif
3091 }