net/ipv4/igmp.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux NET3:     Internet Group Management Protocol  [IGMP]
   4  *
   5  *      This code implements the IGMP protocol as defined in RFC1112. There has
   6  *      been a further revision of this protocol since which is now supported.
   7  *
   8  *      If you have trouble with this module be careful what gcc you have used,
   9  *      the older version didn't come out right using gcc 2.5.8, the newer one
  10  *      seems to fall out with gcc 2.6.2.
  11  *
  12  *      Authors:
  13  *              Alan Cox <alan@lxorguk.ukuu.org.uk>
  14  *
  15  *      Fixes:
  16  *
  17  *              Alan Cox        :       Added lots of __inline__ to optimise
  18  *                                      the memory usage of all the tiny little
  19  *                                      functions.
  20  *              Alan Cox        :       Dumped the header building experiment.
  21  *              Alan Cox        :       Minor tweaks ready for multicast routing
  22  *                                      and extended IGMP protocol.
  23  *              Alan Cox        :       Removed a load of inline directives. Gcc 2.5.8
  24  *                                      writes utterly bogus code otherwise (sigh)
  25  *                                      fixed IGMP loopback to behave in the manner
  26  *                                      desired by mrouted, fixed the fact it has been
  27  *                                      broken since 1.3.6 and cleaned up a few minor
  28  *                                      points.
  29  *
  30  *              Chih-Jen Chang  :       Tried to revise IGMP to Version 2
  31  *              Tsu-Sheng Tsao          E-mail: chihjenc@scf.usc.edu and tsusheng@scf.usc.edu
  32  *                                      The enhancements are mainly based on Steve Deering's
  33  *                                      ipmulti-3.5 source code.
  34  *              Chih-Jen Chang  :       Added the igmp_get_mrouter_info and
  35  *              Tsu-Sheng Tsao          igmp_set_mrouter_info to keep track of
  36  *                                      the mrouted version on that device.
  37  *              Chih-Jen Chang  :       Added the max_resp_time parameter to
  38  *              Tsu-Sheng Tsao          igmp_heard_query(). Using this parameter
  39  *                                      to identify the multicast router version
  40  *                                      and do what the IGMP version 2 specified.
  41  *              Chih-Jen Chang  :       Added a timer to revert to IGMP V2 router
  42  *              Tsu-Sheng Tsao          if the specified time expired.
  43  *              Alan Cox        :       Stop IGMP from 0.0.0.0 being accepted.
  44  *              Alan Cox        :       Use GFP_ATOMIC in the right places.
  45  *              Christian Daudt :       igmp timer wasn't set for local group
  46  *                                      memberships but was being deleted,
  47  *                                      which caused a "del_timer() called
  48  *                                      from %p with timer not initialized\n"
  49  *                                      message (960131).
  50  *              Christian Daudt :       removed del_timer from
  51  *                                      igmp_timer_expire function (960205).
  52  *             Christian Daudt :       igmp_heard_report now only calls
  53  *                                     igmp_timer_expire if tm->running is
  54  *                                     true (960216).
  55  *              Malcolm Beattie :       ttl comparison wrong in igmp_rcv made
  56  *                                      igmp_heard_query never trigger. Expiry
  57  *                                      miscalculation fixed in igmp_heard_query
  58  *                                      and random() made to return unsigned to
  59  *                                      prevent negative expiry times.
  60  *              Alexey Kuznetsov:       Wrong group leaving behaviour, backport
  61  *                                      fix from pending 2.1.x patches.
  62  *              Alan Cox:               Forget to enable FDDI support earlier.
  63  *              Alexey Kuznetsov:       Fixed leaving groups on device down.
  64  *              Alexey Kuznetsov:       Accordance to igmp-v2-06 draft.
  65  *              David L Stevens:        IGMPv3 support, with help from
  66  *                                      Vinay Kulkarni
  67  */
  68
  69 #include <linux/module.h>
  70 #include <linux/slab.h>
  71 #include <linux/uaccess.h>
  72 #include <linux/types.h>
  73 #include <linux/kernel.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/string.h>
  76 #include <linux/socket.h>
  77 #include <linux/sockios.h>
  78 #include <linux/in.h>
  79 #include <linux/inet.h>
  80 #include <linux/netdevice.h>
  81 #include <linux/skbuff.h>
  82 #include <linux/inetdevice.h>
  83 #include <linux/igmp.h>
  84 #include <linux/if_arp.h>
  85 #include <linux/rtnetlink.h>
  86 #include <linux/times.h>
  87 #include <linux/pkt_sched.h>
  88 #include <linux/byteorder/generic.h>
  89
  90 #include <net/net_namespace.h>
  91 #include <net/arp.h>
  92 #include <net/ip.h>
  93 #include <net/protocol.h>
  94 #include <net/route.h>
  95 #include <net/sock.h>
  96 #include <net/checksum.h>
  97 #include <net/inet_common.h>
  98 #include <linux/netfilter_ipv4.h>
  99 #ifdef CONFIG_IP_MROUTE
 100 #include <linux/mroute.h>
 101 #endif
 102 #ifdef CONFIG_PROC_FS
 103 #include <linux/proc_fs.h>
 104 #include <linux/seq_file.h>
 105 #endif
 106
 107 #ifdef CONFIG_IP_MULTICAST
 108 /* Parameter names and values are taken from igmp-v2-06 draft */
 109
 110 #define IGMP_QUERY_INTERVAL                     (125*HZ)
 111 #define IGMP_QUERY_RESPONSE_INTERVAL            (10*HZ)
 112
 113 #define IGMP_INITIAL_REPORT_DELAY               (1)
 114
 115 /* IGMP_INITIAL_REPORT_DELAY is not from IGMP specs!
 116  * IGMP specs require to report membership immediately after
 117  * joining a group, but we delay the first report by a
 118  * small interval. It seems more natural and still does not
 119  * contradict to specs provided this delay is small enough.
 120  */
 121
 122 #define IGMP_V1_SEEN(in_dev) \
 123         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
 124          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
 125          ((in_dev)->mr_v1_seen && \
 126           time_before(jiffies, (in_dev)->mr_v1_seen)))
 127 #define IGMP_V2_SEEN(in_dev) \
 128         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
 129          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
 130          ((in_dev)->mr_v2_seen && \
 131           time_before(jiffies, (in_dev)->mr_v2_seen)))
 132
 133 static int unsolicited_report_interval(struct in_device *in_dev)
 134 {
 135         int interval_ms, interval_jiffies;
 136
 137         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 138                 interval_ms = IN_DEV_CONF_GET(
 139                         in_dev,
 140                         IGMPV2_UNSOLICITED_REPORT_INTERVAL);
 141         else /* v3 */
 142                 interval_ms = IN_DEV_CONF_GET(
 143                         in_dev,
 144                         IGMPV3_UNSOLICITED_REPORT_INTERVAL);
 145
 146         interval_jiffies = msecs_to_jiffies(interval_ms);
 147
 148         /* _timer functions can't handle a delay of 0 jiffies so ensure
 149          *  we always return a positive value.
 150          */
 151         if (interval_jiffies <= 0)
 152                 interval_jiffies = 1;
 153         return interval_jiffies;
 154 }
 155
 156 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
 157                               gfp_t gfp);
 158 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 159 static void igmpv3_clear_delrec(struct in_device *in_dev);
 160 static int sf_setstate(struct ip_mc_list *pmc);
 161 static void sf_markstate(struct ip_mc_list *pmc);
 162 #endif
 163 static void ip_mc_clear_src(struct ip_mc_list *pmc);
 164 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
 165                          int sfcount, __be32 *psfsrc, int delta);
 166
 167 static void ip_ma_put(struct ip_mc_list *im)
 168 {
 169         if (refcount_dec_and_test(&im->refcnt)) {
 170                 in_dev_put(im->interface);
 171                 kfree_rcu(im, rcu);
 172         }
 173 }
 174
 175 #define for_each_pmc_rcu(in_dev, pmc)                           \
 176         for (pmc = rcu_dereference(in_dev->mc_list);            \
 177              pmc != NULL;                                       \
 178              pmc = rcu_dereference(pmc->next_rcu))
 179
 180 #define for_each_pmc_rtnl(in_dev, pmc)                          \
 181         for (pmc = rtnl_dereference(in_dev->mc_list);           \
 182              pmc != NULL;                                       \
 183              pmc = rtnl_dereference(pmc->next_rcu))
 184
 185 static void ip_sf_list_clear_all(struct ip_sf_list *psf)
 186 {
 187         struct ip_sf_list *next;
 188
 189         while (psf) {
 190                 next = psf->sf_next;
 191                 kfree(psf);
 192                 psf = next;
 193         }
 194 }
 195
 196 #ifdef CONFIG_IP_MULTICAST
 197
 198 /*
 199  *      Timer management
 200  */
 201
 202 static void igmp_stop_timer(struct ip_mc_list *im)
 203 {
 204         spin_lock_bh(&im->lock);
 205         if (del_timer(&im->timer))
 206                 refcount_dec(&im->refcnt);
 207         im->tm_running = 0;
 208         im->reporter = 0;
 209         im->unsolicit_count = 0;
 210         spin_unlock_bh(&im->lock);
 211 }
 212
 213 /* It must be called with locked im->lock */
 214 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
 215 {
 216         int tv = prandom_u32() % max_delay;
 217
 218         im->tm_running = 1;
 219         if (!mod_timer(&im->timer, jiffies+tv+2))
 220                 refcount_inc(&im->refcnt);
 221 }
 222
 223 static void igmp_gq_start_timer(struct in_device *in_dev)
 224 {
 225         int tv = prandom_u32() % in_dev->mr_maxdelay;
 226         unsigned long exp = jiffies + tv + 2;
 227
 228         if (in_dev->mr_gq_running &&
 229             time_after_eq(exp, (in_dev->mr_gq_timer).expires))
 230                 return;
 231
 232         in_dev->mr_gq_running = 1;
 233         if (!mod_timer(&in_dev->mr_gq_timer, exp))
 234                 in_dev_hold(in_dev);
 235 }
 236
 237 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 238 {
 239         int tv = prandom_u32() % delay;
 240
 241         if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
 242                 in_dev_hold(in_dev);
 243 }
 244
 245 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 246 {
 247         spin_lock_bh(&im->lock);
 248         im->unsolicit_count = 0;
 249         if (del_timer(&im->timer)) {
 250                 if ((long)(im->timer.expires-jiffies) < max_delay) {
 251                         add_timer(&im->timer);
 252                         im->tm_running = 1;
 253                         spin_unlock_bh(&im->lock);
 254                         return;
 255                 }
 256                 refcount_dec(&im->refcnt);
 257         }
 258         igmp_start_timer(im, max_delay);
 259         spin_unlock_bh(&im->lock);
 260 }
 261
 262
 263 /*
 264  *      Send an IGMP report.
 265  */
 266
 267 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
 268
 269
 270 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
 271         int gdeleted, int sdeleted)
 272 {
 273         switch (type) {
 274         case IGMPV3_MODE_IS_INCLUDE:
 275         case IGMPV3_MODE_IS_EXCLUDE:
 276                 if (gdeleted || sdeleted)
 277                         return 0;
 278                 if (!(pmc->gsquery && !psf->sf_gsresp)) {
 279                         if (pmc->sfmode == MCAST_INCLUDE)
 280                                 return 1;
 281                         /* don't include if this source is excluded
 282                          * in all filters
 283                          */
 284                         if (psf->sf_count[MCAST_INCLUDE])
 285                                 return type == IGMPV3_MODE_IS_INCLUDE;
 286                         return pmc->sfcount[MCAST_EXCLUDE] ==
 287                                 psf->sf_count[MCAST_EXCLUDE];
 288                 }
 289                 return 0;
 290         case IGMPV3_CHANGE_TO_INCLUDE:
 291                 if (gdeleted || sdeleted)
 292                         return 0;
 293                 return psf->sf_count[MCAST_INCLUDE] != 0;
 294         case IGMPV3_CHANGE_TO_EXCLUDE:
 295                 if (gdeleted || sdeleted)
 296                         return 0;
 297                 if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
 298                     psf->sf_count[MCAST_INCLUDE])
 299                         return 0;
 300                 return pmc->sfcount[MCAST_EXCLUDE] ==
 301                         psf->sf_count[MCAST_EXCLUDE];
 302         case IGMPV3_ALLOW_NEW_SOURCES:
 303                 if (gdeleted || !psf->sf_crcount)
 304                         return 0;
 305                 return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
 306         case IGMPV3_BLOCK_OLD_SOURCES:
 307                 if (pmc->sfmode == MCAST_INCLUDE)
 308                         return gdeleted || (psf->sf_crcount && sdeleted);
 309                 return psf->sf_crcount && !gdeleted && !sdeleted;
 310         }
 311         return 0;
 312 }
 313
 314 static int
 315 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
 316 {
 317         struct ip_sf_list *psf;
 318         int scount = 0;
 319
 320         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 321                 if (!is_in(pmc, psf, type, gdeleted, sdeleted))
 322                         continue;
 323                 scount++;
 324         }
 325         return scount;
 326 }
 327
 328 /* source address selection per RFC 3376 section 4.2.13 */
 329 static __be32 igmpv3_get_srcaddr(struct net_device *dev,
 330                                  const struct flowi4 *fl4)
 331 {
 332         struct in_device *in_dev = __in_dev_get_rcu(dev);
 333         const struct in_ifaddr *ifa;
 334
 335         if (!in_dev)
 336                 return htonl(INADDR_ANY);
 337
 338         in_dev_for_each_ifa_rcu(ifa, in_dev) {
 339                 if (fl4->saddr == ifa->ifa_local)
 340                         return fl4->saddr;
 341         }
 342
 343         return htonl(INADDR_ANY);
 344 }
 345
 346 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
 347 {
 348         struct sk_buff *skb;
 349         struct rtable *rt;
 350         struct iphdr *pip;
 351         struct igmpv3_report *pig;
 352         struct net *net = dev_net(dev);
 353         struct flowi4 fl4;
 354         int hlen = LL_RESERVED_SPACE(dev);
 355         int tlen = dev->needed_tailroom;
 356         unsigned int size = mtu;
 357
 358         while (1) {
 359                 skb = alloc_skb(size + hlen + tlen,
 360                                 GFP_ATOMIC | __GFP_NOWARN);
 361                 if (skb)
 362                         break;
 363                 size >>= 1;
 364                 if (size < 256)
 365                         return NULL;
 366         }
 367         skb->priority = TC_PRIO_CONTROL;
 368
 369         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 370                                    0, 0,
 371                                    IPPROTO_IGMP, 0, dev->ifindex);
 372         if (IS_ERR(rt)) {
 373                 kfree_skb(skb);
 374                 return NULL;
 375         }
 376
 377         skb_dst_set(skb, &rt->dst);
 378         skb->dev = dev;
 379
 380         skb_reserve(skb, hlen);
 381         skb_tailroom_reserve(skb, mtu, tlen);
 382
 383         skb_reset_network_header(skb);
 384         pip = ip_hdr(skb);
 385         skb_put(skb, sizeof(struct iphdr) + 4);
 386
 387         pip->version  = 4;
 388         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 389         pip->tos      = 0xc0;
 390         pip->frag_off = htons(IP_DF);
 391         pip->ttl      = 1;
 392         pip->daddr    = fl4.daddr;
 393
 394         rcu_read_lock();
 395         pip->saddr    = igmpv3_get_srcaddr(dev, &fl4);
 396         rcu_read_unlock();
 397
 398         pip->protocol = IPPROTO_IGMP;
 399         pip->tot_len  = 0;      /* filled in later */
 400         ip_select_ident(net, skb, NULL);
 401         ((u8 *)&pip[1])[0] = IPOPT_RA;
 402         ((u8 *)&pip[1])[1] = 4;
 403         ((u8 *)&pip[1])[2] = 0;
 404         ((u8 *)&pip[1])[3] = 0;
 405
 406         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 407         skb_put(skb, sizeof(*pig));
 408         pig = igmpv3_report_hdr(skb);
 409         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 410         pig->resv1 = 0;
 411         pig->csum = 0;
 412         pig->resv2 = 0;
 413         pig->ngrec = 0;
 414         return skb;
 415 }
 416
 417 static int igmpv3_sendpack(struct sk_buff *skb)
 418 {
 419         struct igmphdr *pig = igmp_hdr(skb);
 420         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 421
 422         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 423
 424         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 425 }
 426
 427 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 428 {
 429         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 430 }
 431
 432 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 433         int type, struct igmpv3_grec **ppgr, unsigned int mtu)
 434 {
 435         struct net_device *dev = pmc->interface->dev;
 436         struct igmpv3_report *pih;
 437         struct igmpv3_grec *pgr;
 438
 439         if (!skb) {
 440                 skb = igmpv3_newpack(dev, mtu);
 441                 if (!skb)
 442                         return NULL;
 443         }
 444         pgr = skb_put(skb, sizeof(struct igmpv3_grec));
 445         pgr->grec_type = type;
 446         pgr->grec_auxwords = 0;
 447         pgr->grec_nsrcs = 0;
 448         pgr->grec_mca = pmc->multiaddr;
 449         pih = igmpv3_report_hdr(skb);
 450         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 451         *ppgr = pgr;
 452         return skb;
 453 }
 454
 455 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 456
 457 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 458         int type, int gdeleted, int sdeleted)
 459 {
 460         struct net_device *dev = pmc->interface->dev;
 461         struct net *net = dev_net(dev);
 462         struct igmpv3_report *pih;
 463         struct igmpv3_grec *pgr = NULL;
 464         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 465         int scount, stotal, first, isquery, truncate;
 466         unsigned int mtu;
 467
 468         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 469                 return skb;
 470         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 471             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 472                 return skb;
 473
 474         mtu = READ_ONCE(dev->mtu);
 475         if (mtu < IPV4_MIN_MTU)
 476                 return skb;
 477
 478         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 479                   type == IGMPV3_MODE_IS_EXCLUDE;
 480         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 481                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 482
 483         stotal = scount = 0;
 484
 485         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 486
 487         if (!*psf_list)
 488                 goto empty_source;
 489
 490         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 491
 492         /* EX and TO_EX get a fresh packet, if needed */
 493         if (truncate) {
 494                 if (pih && pih->ngrec &&
 495                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 496                         if (skb)
 497                                 igmpv3_sendpack(skb);
 498                         skb = igmpv3_newpack(dev, mtu);
 499                 }
 500         }
 501         first = 1;
 502         psf_prev = NULL;
 503         for (psf = *psf_list; psf; psf = psf_next) {
 504                 __be32 *psrc;
 505
 506                 psf_next = psf->sf_next;
 507
 508                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 509                         psf_prev = psf;
 510                         continue;
 511                 }
 512
 513                 /* Based on RFC3376 5.1. Should not send source-list change
 514                  * records when there is a filter mode change.
 515                  */
 516                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 517                      (!gdeleted && pmc->crcount)) &&
 518                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 519                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 520                         goto decrease_sf_crcount;
 521
 522                 /* clear marks on query responses */
 523                 if (isquery)
 524                         psf->sf_gsresp = 0;
 525
 526                 if (AVAILABLE(skb) < sizeof(__be32) +
 527                     first*sizeof(struct igmpv3_grec)) {
 528                         if (truncate && !first)
 529                                 break;   /* truncate these */
 530                         if (pgr)
 531                                 pgr->grec_nsrcs = htons(scount);
 532                         if (skb)
 533                                 igmpv3_sendpack(skb);
 534                         skb = igmpv3_newpack(dev, mtu);
 535                         first = 1;
 536                         scount = 0;
 537                 }
 538                 if (first) {
 539                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 540                         first = 0;
 541                 }
 542                 if (!skb)
 543                         return NULL;
 544                 psrc = skb_put(skb, sizeof(__be32));
 545                 *psrc = psf->sf_inaddr;
 546                 scount++; stotal++;
 547                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 548                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 549 decrease_sf_crcount:
 550                         psf->sf_crcount--;
 551                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 552                                 if (psf_prev)
 553                                         psf_prev->sf_next = psf->sf_next;
 554                                 else
 555                                         *psf_list = psf->sf_next;
 556                                 kfree(psf);
 557                                 continue;
 558                         }
 559                 }
 560                 psf_prev = psf;
 561         }
 562
 563 empty_source:
 564         if (!stotal) {
 565                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 566                     type == IGMPV3_BLOCK_OLD_SOURCES)
 567                         return skb;
 568                 if (pmc->crcount || isquery) {
 569                         /* make sure we have room for group header */
 570                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 571                                 igmpv3_sendpack(skb);
 572                                 skb = NULL; /* add_grhead will get a new one */
 573                         }
 574                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 575                 }
 576         }
 577         if (pgr)
 578                 pgr->grec_nsrcs = htons(scount);
 579
 580         if (isquery)
 581                 pmc->gsquery = 0;       /* clear query state on report */
 582         return skb;
 583 }
 584
 585 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 586 {
 587         struct sk_buff *skb = NULL;
 588         struct net *net = dev_net(in_dev->dev);
 589         int type;
 590
 591         if (!pmc) {
 592                 rcu_read_lock();
 593                 for_each_pmc_rcu(in_dev, pmc) {
 594                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 595                                 continue;
 596                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 597                             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 598                                 continue;
 599                         spin_lock_bh(&pmc->lock);
 600                         if (pmc->sfcount[MCAST_EXCLUDE])
 601                                 type = IGMPV3_MODE_IS_EXCLUDE;
 602                         else
 603                                 type = IGMPV3_MODE_IS_INCLUDE;
 604                         skb = add_grec(skb, pmc, type, 0, 0);
 605                         spin_unlock_bh(&pmc->lock);
 606                 }
 607                 rcu_read_unlock();
 608         } else {
 609                 spin_lock_bh(&pmc->lock);
 610                 if (pmc->sfcount[MCAST_EXCLUDE])
 611                         type = IGMPV3_MODE_IS_EXCLUDE;
 612                 else
 613                         type = IGMPV3_MODE_IS_INCLUDE;
 614                 skb = add_grec(skb, pmc, type, 0, 0);
 615                 spin_unlock_bh(&pmc->lock);
 616         }
 617         if (!skb)
 618                 return 0;
 619         return igmpv3_sendpack(skb);
 620 }
 621
 622 /*
 623  * remove zero-count source records from a source filter list
 624  */
 625 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 626 {
 627         struct ip_sf_list *psf_prev, *psf_next, *psf;
 628
 629         psf_prev = NULL;
 630         for (psf = *ppsf; psf; psf = psf_next) {
 631                 psf_next = psf->sf_next;
 632                 if (psf->sf_crcount == 0) {
 633                         if (psf_prev)
 634                                 psf_prev->sf_next = psf->sf_next;
 635                         else
 636                                 *ppsf = psf->sf_next;
 637                         kfree(psf);
 638                 } else
 639                         psf_prev = psf;
 640         }
 641 }
 642
 643 static void kfree_pmc(struct ip_mc_list *pmc)
 644 {
 645         ip_sf_list_clear_all(pmc->sources);
 646         ip_sf_list_clear_all(pmc->tomb);
 647         kfree(pmc);
 648 }
 649
 650 static void igmpv3_send_cr(struct in_device *in_dev)
 651 {
 652         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 653         struct sk_buff *skb = NULL;
 654         int type, dtype;
 655
 656         rcu_read_lock();
 657         spin_lock_bh(&in_dev->mc_tomb_lock);
 658
 659         /* deleted MCA's */
 660         pmc_prev = NULL;
 661         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 662                 pmc_next = pmc->next;
 663                 if (pmc->sfmode == MCAST_INCLUDE) {
 664                         type = IGMPV3_BLOCK_OLD_SOURCES;
 665                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 666                         skb = add_grec(skb, pmc, type, 1, 0);
 667                         skb = add_grec(skb, pmc, dtype, 1, 1);
 668                 }
 669                 if (pmc->crcount) {
 670                         if (pmc->sfmode == MCAST_EXCLUDE) {
 671                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 672                                 skb = add_grec(skb, pmc, type, 1, 0);
 673                         }
 674                         pmc->crcount--;
 675                         if (pmc->crcount == 0) {
 676                                 igmpv3_clear_zeros(&pmc->tomb);
 677                                 igmpv3_clear_zeros(&pmc->sources);
 678                         }
 679                 }
 680                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 681                         if (pmc_prev)
 682                                 pmc_prev->next = pmc_next;
 683                         else
 684                                 in_dev->mc_tomb = pmc_next;
 685                         in_dev_put(pmc->interface);
 686                         kfree_pmc(pmc);
 687                 } else
 688                         pmc_prev = pmc;
 689         }
 690         spin_unlock_bh(&in_dev->mc_tomb_lock);
 691
 692         /* change recs */
 693         for_each_pmc_rcu(in_dev, pmc) {
 694                 spin_lock_bh(&pmc->lock);
 695                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 696                         type = IGMPV3_BLOCK_OLD_SOURCES;
 697                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 698                 } else {
 699                         type = IGMPV3_ALLOW_NEW_SOURCES;
 700                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 701                 }
 702                 skb = add_grec(skb, pmc, type, 0, 0);
 703                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 704
 705                 /* filter mode changes */
 706                 if (pmc->crcount) {
 707                         if (pmc->sfmode == MCAST_EXCLUDE)
 708                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 709                         else
 710                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 711                         skb = add_grec(skb, pmc, type, 0, 0);
 712                         pmc->crcount--;
 713                 }
 714                 spin_unlock_bh(&pmc->lock);
 715         }
 716         rcu_read_unlock();
 717
 718         if (!skb)
 719                 return;
 720         (void) igmpv3_sendpack(skb);
 721 }
 722
 723 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 724         int type)
 725 {
 726         struct sk_buff *skb;
 727         struct iphdr *iph;
 728         struct igmphdr *ih;
 729         struct rtable *rt;
 730         struct net_device *dev = in_dev->dev;
 731         struct net *net = dev_net(dev);
 732         __be32  group = pmc ? pmc->multiaddr : 0;
 733         struct flowi4 fl4;
 734         __be32  dst;
 735         int hlen, tlen;
 736
 737         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 738                 return igmpv3_send_report(in_dev, pmc);
 739
 740         if (ipv4_is_local_multicast(group) &&
 741             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 742                 return 0;
 743
 744         if (type == IGMP_HOST_LEAVE_MESSAGE)
 745                 dst = IGMP_ALL_ROUTER;
 746         else
 747                 dst = group;
 748
 749         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 750                                    0, 0,
 751                                    IPPROTO_IGMP, 0, dev->ifindex);
 752         if (IS_ERR(rt))
 753                 return -1;
 754
 755         hlen = LL_RESERVED_SPACE(dev);
 756         tlen = dev->needed_tailroom;
 757         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 758         if (!skb) {
 759                 ip_rt_put(rt);
 760                 return -1;
 761         }
 762         skb->priority = TC_PRIO_CONTROL;
 763
 764         skb_dst_set(skb, &rt->dst);
 765
 766         skb_reserve(skb, hlen);
 767
 768         skb_reset_network_header(skb);
 769         iph = ip_hdr(skb);
 770         skb_put(skb, sizeof(struct iphdr) + 4);
 771
 772         iph->version  = 4;
 773         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 774         iph->tos      = 0xc0;
 775         iph->frag_off = htons(IP_DF);
 776         iph->ttl      = 1;
 777         iph->daddr    = dst;
 778         iph->saddr    = fl4.saddr;
 779         iph->protocol = IPPROTO_IGMP;
 780         ip_select_ident(net, skb, NULL);
 781         ((u8 *)&iph[1])[0] = IPOPT_RA;
 782         ((u8 *)&iph[1])[1] = 4;
 783         ((u8 *)&iph[1])[2] = 0;
 784         ((u8 *)&iph[1])[3] = 0;
 785
 786         ih = skb_put(skb, sizeof(struct igmphdr));
 787         ih->type = type;
 788         ih->code = 0;
 789         ih->csum = 0;
 790         ih->group = group;
 791         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 792
 793         return ip_local_out(net, skb->sk, skb);
 794 }
 795
 796 static void igmp_gq_timer_expire(struct timer_list *t)
 797 {
 798         struct in_device *in_dev = from_timer(in_dev, t, mr_gq_timer);
 799
 800         in_dev->mr_gq_running = 0;
 801         igmpv3_send_report(in_dev, NULL);
 802         in_dev_put(in_dev);
 803 }
 804
 805 static void igmp_ifc_timer_expire(struct timer_list *t)
 806 {
 807         struct in_device *in_dev = from_timer(in_dev, t, mr_ifc_timer);
 808         u32 mr_ifc_count;
 809
 810         igmpv3_send_cr(in_dev);
 811 restart:
 812         mr_ifc_count = READ_ONCE(in_dev->mr_ifc_count);
 813
 814         if (mr_ifc_count) {
 815                 if (cmpxchg(&in_dev->mr_ifc_count,
 816                             mr_ifc_count,
 817                             mr_ifc_count - 1) != mr_ifc_count)
 818                         goto restart;
 819                 igmp_ifc_start_timer(in_dev,
 820                                      unsolicited_report_interval(in_dev));
 821         }
 822         in_dev_put(in_dev);
 823 }
 824
 825 static void igmp_ifc_event(struct in_device *in_dev)
 826 {
 827         struct net *net = dev_net(in_dev->dev);
 828         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 829                 return;
 830         WRITE_ONCE(in_dev->mr_ifc_count, in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv);
 831         igmp_ifc_start_timer(in_dev, 1);
 832 }
 833
 834
 835 static void igmp_timer_expire(struct timer_list *t)
 836 {
 837         struct ip_mc_list *im = from_timer(im, t, timer);
 838         struct in_device *in_dev = im->interface;
 839
 840         spin_lock(&im->lock);
 841         im->tm_running = 0;
 842
 843         if (im->unsolicit_count && --im->unsolicit_count)
 844                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 845
 846         im->reporter = 1;
 847         spin_unlock(&im->lock);
 848
 849         if (IGMP_V1_SEEN(in_dev))
 850                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 851         else if (IGMP_V2_SEEN(in_dev))
 852                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 853         else
 854                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 855
 856         ip_ma_put(im);
 857 }
 858
 859 /* mark EXCLUDE-mode sources */
 860 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 861 {
 862         struct ip_sf_list *psf;
 863         int i, scount;
 864
 865         scount = 0;
 866         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 867                 if (scount == nsrcs)
 868                         break;
 869                 for (i = 0; i < nsrcs; i++) {
 870                         /* skip inactive filters */
 871                         if (psf->sf_count[MCAST_INCLUDE] ||
 872                             pmc->sfcount[MCAST_EXCLUDE] !=
 873                             psf->sf_count[MCAST_EXCLUDE])
 874                                 break;
 875                         if (srcs[i] == psf->sf_inaddr) {
 876                                 scount++;
 877                                 break;
 878                         }
 879                 }
 880         }
 881         pmc->gsquery = 0;
 882         if (scount == nsrcs)    /* all sources excluded */
 883                 return 0;
 884         return 1;
 885 }
 886
 887 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 888 {
 889         struct ip_sf_list *psf;
 890         int i, scount;
 891
 892         if (pmc->sfmode == MCAST_EXCLUDE)
 893                 return igmp_xmarksources(pmc, nsrcs, srcs);
 894
 895         /* mark INCLUDE-mode sources */
 896         scount = 0;
 897         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 898                 if (scount == nsrcs)
 899                         break;
 900                 for (i = 0; i < nsrcs; i++)
 901                         if (srcs[i] == psf->sf_inaddr) {
 902                                 psf->sf_gsresp = 1;
 903                                 scount++;
 904                                 break;
 905                         }
 906         }
 907         if (!scount) {
 908                 pmc->gsquery = 0;
 909                 return 0;
 910         }
 911         pmc->gsquery = 1;
 912         return 1;
 913 }
 914
 915 /* return true if packet was dropped */
 916 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 917 {
 918         struct ip_mc_list *im;
 919         struct net *net = dev_net(in_dev->dev);
 920
 921         /* Timers are only set for non-local groups */
 922
 923         if (group == IGMP_ALL_HOSTS)
 924                 return false;
 925         if (ipv4_is_local_multicast(group) &&
 926             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
 927                 return false;
 928
 929         rcu_read_lock();
 930         for_each_pmc_rcu(in_dev, im) {
 931                 if (im->multiaddr == group) {
 932                         igmp_stop_timer(im);
 933                         break;
 934                 }
 935         }
 936         rcu_read_unlock();
 937         return false;
 938 }
 939
 940 /* return true if packet was dropped */
 941 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 942         int len)
 943 {
 944         struct igmphdr          *ih = igmp_hdr(skb);
 945         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 946         struct ip_mc_list       *im;
 947         __be32                  group = ih->group;
 948         int                     max_delay;
 949         int                     mark = 0;
 950         struct net              *net = dev_net(in_dev->dev);
 951
 952
 953         if (len == 8) {
 954                 if (ih->code == 0) {
 955                         /* Alas, old v1 router presents here. */
 956
 957                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 958                         in_dev->mr_v1_seen = jiffies +
 959                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 960                                 in_dev->mr_qri;
 961                         group = 0;
 962                 } else {
 963                         /* v2 router present */
 964                         max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
 965                         in_dev->mr_v2_seen = jiffies +
 966                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 967                                 in_dev->mr_qri;
 968                 }
 969                 /* cancel the interface change timer */
 970                 WRITE_ONCE(in_dev->mr_ifc_count, 0);
 971                 if (del_timer(&in_dev->mr_ifc_timer))
 972                         __in_dev_put(in_dev);
 973                 /* clear deleted report items */
 974                 igmpv3_clear_delrec(in_dev);
 975         } else if (len < 12) {
 976                 return true;    /* ignore bogus packet; freed by caller */
 977         } else if (IGMP_V1_SEEN(in_dev)) {
 978                 /* This is a v3 query with v1 queriers present */
 979                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 980                 group = 0;
 981         } else if (IGMP_V2_SEEN(in_dev)) {
 982                 /* this is a v3 query with v2 queriers present;
 983                  * Interpretation of the max_delay code is problematic here.
 984                  * A real v2 host would use ih_code directly, while v3 has a
 985                  * different encoding. We use the v3 encoding as more likely
 986                  * to be intended in a v3 query.
 987                  */
 988                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 989                 if (!max_delay)
 990                         max_delay = 1;  /* can't mod w/ 0 */
 991         } else { /* v3 */
 992                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 993                         return true;
 994
 995                 ih3 = igmpv3_query_hdr(skb);
 996                 if (ih3->nsrcs) {
 997                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
 998                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
 999                                 return true;
1000                         ih3 = igmpv3_query_hdr(skb);
1001                 }
1002
1003                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
1004                 if (!max_delay)
1005                         max_delay = 1;  /* can't mod w/ 0 */
1006                 in_dev->mr_maxdelay = max_delay;
1007
1008                 /* RFC3376, 4.1.6. QRV and 4.1.7. QQIC, when the most recently
1009                  * received value was zero, use the default or statically
1010                  * configured value.
1011                  */
1012                 in_dev->mr_qrv = ih3->qrv ?: net->ipv4.sysctl_igmp_qrv;
1013                 in_dev->mr_qi = IGMPV3_QQIC(ih3->qqic)*HZ ?: IGMP_QUERY_INTERVAL;
1014
1015                 /* RFC3376, 8.3. Query Response Interval:
1016                  * The number of seconds represented by the [Query Response
1017                  * Interval] must be less than the [Query Interval].
1018                  */
1019                 if (in_dev->mr_qri >= in_dev->mr_qi)
1020                         in_dev->mr_qri = (in_dev->mr_qi/HZ - 1)*HZ;
1021
1022                 if (!group) { /* general query */
1023                         if (ih3->nsrcs)
1024                                 return true;    /* no sources allowed */
1025                         igmp_gq_start_timer(in_dev);
1026                         return false;
1027                 }
1028                 /* mark sources to include, if group & source-specific */
1029                 mark = ih3->nsrcs != 0;
1030         }
1031
1032         /*
1033          * - Start the timers in all of our membership records
1034          *   that the query applies to for the interface on
1035          *   which the query arrived excl. those that belong
1036          *   to a "local" group (224.0.0.X)
1037          * - For timers already running check if they need to
1038          *   be reset.
1039          * - Use the igmp->igmp_code field as the maximum
1040          *   delay possible
1041          */
1042         rcu_read_lock();
1043         for_each_pmc_rcu(in_dev, im) {
1044                 int changed;
1045
1046                 if (group && group != im->multiaddr)
1047                         continue;
1048                 if (im->multiaddr == IGMP_ALL_HOSTS)
1049                         continue;
1050                 if (ipv4_is_local_multicast(im->multiaddr) &&
1051                     !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1052                         continue;
1053                 spin_lock_bh(&im->lock);
1054                 if (im->tm_running)
1055                         im->gsquery = im->gsquery && mark;
1056                 else
1057                         im->gsquery = mark;
1058                 changed = !im->gsquery ||
1059                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1060                 spin_unlock_bh(&im->lock);
1061                 if (changed)
1062                         igmp_mod_timer(im, max_delay);
1063         }
1064         rcu_read_unlock();
1065         return false;
1066 }
1067
1068 /* called in rcu_read_lock() section */
1069 int igmp_rcv(struct sk_buff *skb)
1070 {
1071         /* This basically follows the spec line by line -- see RFC1112 */
1072         struct igmphdr *ih;
1073         struct net_device *dev = skb->dev;
1074         struct in_device *in_dev;
1075         int len = skb->len;
1076         bool dropped = true;
1077
1078         if (netif_is_l3_master(dev)) {
1079                 dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1080                 if (!dev)
1081                         goto drop;
1082         }
1083
1084         in_dev = __in_dev_get_rcu(dev);
1085         if (!in_dev)
1086                 goto drop;
1087
1088         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1089                 goto drop;
1090
1091         if (skb_checksum_simple_validate(skb))
1092                 goto drop;
1093
1094         ih = igmp_hdr(skb);
1095         switch (ih->type) {
1096         case IGMP_HOST_MEMBERSHIP_QUERY:
1097                 dropped = igmp_heard_query(in_dev, skb, len);
1098                 break;
1099         case IGMP_HOST_MEMBERSHIP_REPORT:
1100         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1101                 /* Is it our report looped back? */
1102                 if (rt_is_output_route(skb_rtable(skb)))
1103                         break;
1104                 /* don't rely on MC router hearing unicast reports */
1105                 if (skb->pkt_type == PACKET_MULTICAST ||
1106                     skb->pkt_type == PACKET_BROADCAST)
1107                         dropped = igmp_heard_report(in_dev, ih->group);
1108                 break;
1109         case IGMP_PIM:
1110 #ifdef CONFIG_IP_PIMSM_V1
1111                 return pim_rcv_v1(skb);
1112 #endif
1113         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1114         case IGMP_DVMRP:
1115         case IGMP_TRACE:
1116         case IGMP_HOST_LEAVE_MESSAGE:
1117         case IGMP_MTRACE:
1118         case IGMP_MTRACE_RESP:
1119                 break;
1120         default:
1121                 break;
1122         }
1123
1124 drop:
1125         if (dropped)
1126                 kfree_skb(skb);
1127         else
1128                 consume_skb(skb);
1129         return 0;
1130 }
1131
1132 #endif
1133
1134
1135 /*
1136  *      Add a filter to a device
1137  */
1138
1139 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1140 {
1141         char buf[MAX_ADDR_LEN];
1142         struct net_device *dev = in_dev->dev;
1143
1144         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1145            We will get multicast token leakage, when IFF_MULTICAST
1146            is changed. This check should be done in ndo_set_rx_mode
1147            routine. Something sort of:
1148            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1149            --ANK
1150            */
1151         if (arp_mc_map(addr, buf, dev, 0) == 0)
1152                 dev_mc_add(dev, buf);
1153 }
1154
1155 /*
1156  *      Remove a filter from a device
1157  */
1158
1159 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1160 {
1161         char buf[MAX_ADDR_LEN];
1162         struct net_device *dev = in_dev->dev;
1163
1164         if (arp_mc_map(addr, buf, dev, 0) == 0)
1165                 dev_mc_del(dev, buf);
1166 }
1167
1168 #ifdef CONFIG_IP_MULTICAST
1169 /*
1170  * deleted ip_mc_list manipulation
1171  */
1172 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im,
1173                               gfp_t gfp)
1174 {
1175         struct ip_mc_list *pmc;
1176         struct net *net = dev_net(in_dev->dev);
1177
1178         /* this is an "ip_mc_list" for convenience; only the fields below
1179          * are actually used. In particular, the refcnt and users are not
1180          * used for management of the delete list. Using the same structure
1181          * for deleted items allows change reports to use common code with
1182          * non-deleted or query-response MCA's.
1183          */
1184         pmc = kzalloc(sizeof(*pmc), gfp);
1185         if (!pmc)
1186                 return;
1187         spin_lock_init(&pmc->lock);
1188         spin_lock_bh(&im->lock);
1189         pmc->interface = im->interface;
1190         in_dev_hold(in_dev);
1191         pmc->multiaddr = im->multiaddr;
1192         pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1193         pmc->sfmode = im->sfmode;
1194         if (pmc->sfmode == MCAST_INCLUDE) {
1195                 struct ip_sf_list *psf;
1196
1197                 pmc->tomb = im->tomb;
1198                 pmc->sources = im->sources;
1199                 im->tomb = im->sources = NULL;
1200                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1201                         psf->sf_crcount = pmc->crcount;
1202         }
1203         spin_unlock_bh(&im->lock);
1204
1205         spin_lock_bh(&in_dev->mc_tomb_lock);
1206         pmc->next = in_dev->mc_tomb;
1207         in_dev->mc_tomb = pmc;
1208         spin_unlock_bh(&in_dev->mc_tomb_lock);
1209 }
1210
1211 /*
1212  * restore ip_mc_list deleted records
1213  */
1214 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1215 {
1216         struct ip_mc_list *pmc, *pmc_prev;
1217         struct ip_sf_list *psf;
1218         struct net *net = dev_net(in_dev->dev);
1219         __be32 multiaddr = im->multiaddr;
1220
1221         spin_lock_bh(&in_dev->mc_tomb_lock);
1222         pmc_prev = NULL;
1223         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1224                 if (pmc->multiaddr == multiaddr)
1225                         break;
1226                 pmc_prev = pmc;
1227         }
1228         if (pmc) {
1229                 if (pmc_prev)
1230                         pmc_prev->next = pmc->next;
1231                 else
1232                         in_dev->mc_tomb = pmc->next;
1233         }
1234         spin_unlock_bh(&in_dev->mc_tomb_lock);
1235
1236         spin_lock_bh(&im->lock);
1237         if (pmc) {
1238                 im->interface = pmc->interface;
1239                 if (im->sfmode == MCAST_INCLUDE) {
1240                         swap(im->tomb, pmc->tomb);
1241                         swap(im->sources, pmc->sources);
1242                         for (psf = im->sources; psf; psf = psf->sf_next)
1243                                 psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1244                 } else {
1245                         im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1246                 }
1247                 in_dev_put(pmc->interface);
1248                 kfree_pmc(pmc);
1249         }
1250         spin_unlock_bh(&im->lock);
1251 }
1252
1253 /*
1254  * flush ip_mc_list deleted records
1255  */
1256 static void igmpv3_clear_delrec(struct in_device *in_dev)
1257 {
1258         struct ip_mc_list *pmc, *nextpmc;
1259
1260         spin_lock_bh(&in_dev->mc_tomb_lock);
1261         pmc = in_dev->mc_tomb;
1262         in_dev->mc_tomb = NULL;
1263         spin_unlock_bh(&in_dev->mc_tomb_lock);
1264
1265         for (; pmc; pmc = nextpmc) {
1266                 nextpmc = pmc->next;
1267                 ip_mc_clear_src(pmc);
1268                 in_dev_put(pmc->interface);
1269                 kfree_pmc(pmc);
1270         }
1271         /* clear dead sources, too */
1272         rcu_read_lock();
1273         for_each_pmc_rcu(in_dev, pmc) {
1274                 struct ip_sf_list *psf;
1275
1276                 spin_lock_bh(&pmc->lock);
1277                 psf = pmc->tomb;
1278                 pmc->tomb = NULL;
1279                 spin_unlock_bh(&pmc->lock);
1280                 ip_sf_list_clear_all(psf);
1281         }
1282         rcu_read_unlock();
1283 }
1284 #endif
1285
1286 static void __igmp_group_dropped(struct ip_mc_list *im, gfp_t gfp)
1287 {
1288         struct in_device *in_dev = im->interface;
1289 #ifdef CONFIG_IP_MULTICAST
1290         struct net *net = dev_net(in_dev->dev);
1291         int reporter;
1292 #endif
1293
1294         if (im->loaded) {
1295                 im->loaded = 0;
1296                 ip_mc_filter_del(in_dev, im->multiaddr);
1297         }
1298
1299 #ifdef CONFIG_IP_MULTICAST
1300         if (im->multiaddr == IGMP_ALL_HOSTS)
1301                 return;
1302         if (ipv4_is_local_multicast(im->multiaddr) &&
1303             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1304                 return;
1305
1306         reporter = im->reporter;
1307         igmp_stop_timer(im);
1308
1309         if (!in_dev->dead) {
1310                 if (IGMP_V1_SEEN(in_dev))
1311                         return;
1312                 if (IGMP_V2_SEEN(in_dev)) {
1313                         if (reporter)
1314                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1315                         return;
1316                 }
1317                 /* IGMPv3 */
1318                 igmpv3_add_delrec(in_dev, im, gfp);
1319
1320                 igmp_ifc_event(in_dev);
1321         }
1322 #endif
1323 }
1324
1325 static void igmp_group_dropped(struct ip_mc_list *im)
1326 {
1327         __igmp_group_dropped(im, GFP_KERNEL);
1328 }
1329
1330 static void igmp_group_added(struct ip_mc_list *im)
1331 {
1332         struct in_device *in_dev = im->interface;
1333 #ifdef CONFIG_IP_MULTICAST
1334         struct net *net = dev_net(in_dev->dev);
1335 #endif
1336
1337         if (im->loaded == 0) {
1338                 im->loaded = 1;
1339                 ip_mc_filter_add(in_dev, im->multiaddr);
1340         }
1341
1342 #ifdef CONFIG_IP_MULTICAST
1343         if (im->multiaddr == IGMP_ALL_HOSTS)
1344                 return;
1345         if (ipv4_is_local_multicast(im->multiaddr) &&
1346             !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1347                 return;
1348
1349         if (in_dev->dead)
1350                 return;
1351
1352         im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1353         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1354                 spin_lock_bh(&im->lock);
1355                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1356                 spin_unlock_bh(&im->lock);
1357                 return;
1358         }
1359         /* else, v3 */
1360
1361         /* Based on RFC3376 5.1, for newly added INCLUDE SSM, we should
1362          * not send filter-mode change record as the mode should be from
1363          * IN() to IN(A).
1364          */
1365         if (im->sfmode == MCAST_EXCLUDE)
1366                 im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1367
1368         igmp_ifc_event(in_dev);
1369 #endif
1370 }
1371
1372
1373 /*
1374  *      Multicast list managers
1375  */
1376
1377 static u32 ip_mc_hash(const struct ip_mc_list *im)
1378 {
1379         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1380 }
1381
1382 static void ip_mc_hash_add(struct in_device *in_dev,
1383                            struct ip_mc_list *im)
1384 {
1385         struct ip_mc_list __rcu **mc_hash;
1386         u32 hash;
1387
1388         mc_hash = rtnl_dereference(in_dev->mc_hash);
1389         if (mc_hash) {
1390                 hash = ip_mc_hash(im);
1391                 im->next_hash = mc_hash[hash];
1392                 rcu_assign_pointer(mc_hash[hash], im);
1393                 return;
1394         }
1395
1396         /* do not use a hash table for small number of items */
1397         if (in_dev->mc_count < 4)
1398                 return;
1399
1400         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1401                           GFP_KERNEL);
1402         if (!mc_hash)
1403                 return;
1404
1405         for_each_pmc_rtnl(in_dev, im) {
1406                 hash = ip_mc_hash(im);
1407                 im->next_hash = mc_hash[hash];
1408                 RCU_INIT_POINTER(mc_hash[hash], im);
1409         }
1410
1411         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1412 }
1413
1414 static void ip_mc_hash_remove(struct in_device *in_dev,
1415                               struct ip_mc_list *im)
1416 {
1417         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1418         struct ip_mc_list *aux;
1419
1420         if (!mc_hash)
1421                 return;
1422         mc_hash += ip_mc_hash(im);
1423         while ((aux = rtnl_dereference(*mc_hash)) != im)
1424                 mc_hash = &aux->next_hash;
1425         *mc_hash = im->next_hash;
1426 }
1427
1428
1429 /*
1430  *      A socket has joined a multicast group on device dev.
1431  */
1432 static void ____ip_mc_inc_group(struct in_device *in_dev, __be32 addr,
1433                                 unsigned int mode, gfp_t gfp)
1434 {
1435         struct ip_mc_list *im;
1436
1437         ASSERT_RTNL();
1438
1439         for_each_pmc_rtnl(in_dev, im) {
1440                 if (im->multiaddr == addr) {
1441                         im->users++;
1442                         ip_mc_add_src(in_dev, &addr, mode, 0, NULL, 0);
1443                         goto out;
1444                 }
1445         }
1446
1447         im = kzalloc(sizeof(*im), gfp);
1448         if (!im)
1449                 goto out;
1450
1451         im->users = 1;
1452         im->interface = in_dev;
1453         in_dev_hold(in_dev);
1454         im->multiaddr = addr;
1455         /* initial mode is (EX, empty) */
1456         im->sfmode = mode;
1457         im->sfcount[mode] = 1;
1458         refcount_set(&im->refcnt, 1);
1459         spin_lock_init(&im->lock);
1460 #ifdef CONFIG_IP_MULTICAST
1461         timer_setup(&im->timer, igmp_timer_expire, 0);
1462 #endif
1463
1464         im->next_rcu = in_dev->mc_list;
1465         in_dev->mc_count++;
1466         rcu_assign_pointer(in_dev->mc_list, im);
1467
1468         ip_mc_hash_add(in_dev, im);
1469
1470 #ifdef CONFIG_IP_MULTICAST
1471         igmpv3_del_delrec(in_dev, im);
1472 #endif
1473         igmp_group_added(im);
1474         if (!in_dev->dead)
1475                 ip_rt_multicast_event(in_dev);
1476 out:
1477         return;
1478 }
1479
1480 void __ip_mc_inc_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1481 {
1482         ____ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE, gfp);
1483 }
1484 EXPORT_SYMBOL(__ip_mc_inc_group);
1485
1486 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1487 {
1488         __ip_mc_inc_group(in_dev, addr, GFP_KERNEL);
1489 }
1490 EXPORT_SYMBOL(ip_mc_inc_group);
1491
1492 static int ip_mc_check_iphdr(struct sk_buff *skb)
1493 {
1494         const struct iphdr *iph;
1495         unsigned int len;
1496         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1497
1498         if (!pskb_may_pull(skb, offset))
1499                 return -EINVAL;
1500
1501         iph = ip_hdr(skb);
1502
1503         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1504                 return -EINVAL;
1505
1506         offset += ip_hdrlen(skb) - sizeof(*iph);
1507
1508         if (!pskb_may_pull(skb, offset))
1509                 return -EINVAL;
1510
1511         iph = ip_hdr(skb);
1512
1513         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1514                 return -EINVAL;
1515
1516         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1517         if (skb->len < len || len < offset)
1518                 return -EINVAL;
1519
1520         skb_set_transport_header(skb, offset);
1521
1522         return 0;
1523 }
1524
1525 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1526 {
1527         unsigned int len = skb_transport_offset(skb);
1528
1529         len += sizeof(struct igmpv3_report);
1530
1531         return ip_mc_may_pull(skb, len) ? 0 : -EINVAL;
1532 }
1533
1534 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1535 {
1536         unsigned int transport_len = ip_transport_len(skb);
1537         unsigned int len;
1538
1539         /* IGMPv{1,2}? */
1540         if (transport_len != sizeof(struct igmphdr)) {
1541                 /* or IGMPv3? */
1542                 if (transport_len < sizeof(struct igmpv3_query))
1543                         return -EINVAL;
1544
1545                 len = skb_transport_offset(skb) + sizeof(struct igmpv3_query);
1546                 if (!ip_mc_may_pull(skb, len))
1547                         return -EINVAL;
1548         }
1549
1550         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1551          * all-systems destination addresses (224.0.0.1) for general queries
1552          */
1553         if (!igmp_hdr(skb)->group &&
1554             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1555                 return -EINVAL;
1556
1557         return 0;
1558 }
1559
1560 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1561 {
1562         switch (igmp_hdr(skb)->type) {
1563         case IGMP_HOST_LEAVE_MESSAGE:
1564         case IGMP_HOST_MEMBERSHIP_REPORT:
1565         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1566                 return 0;
1567         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1568                 return ip_mc_check_igmp_reportv3(skb);
1569         case IGMP_HOST_MEMBERSHIP_QUERY:
1570                 return ip_mc_check_igmp_query(skb);
1571         default:
1572                 return -ENOMSG;
1573         }
1574 }
1575
1576 static __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1577 {
1578         return skb_checksum_simple_validate(skb);
1579 }
1580
1581 static int ip_mc_check_igmp_csum(struct sk_buff *skb)
1582 {
1583         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1584         unsigned int transport_len = ip_transport_len(skb);
1585         struct sk_buff *skb_chk;
1586
1587         if (!ip_mc_may_pull(skb, len))
1588                 return -EINVAL;
1589
1590         skb_chk = skb_checksum_trimmed(skb, transport_len,
1591                                        ip_mc_validate_checksum);
1592         if (!skb_chk)
1593                 return -EINVAL;
1594
1595         if (skb_chk != skb)
1596                 kfree_skb(skb_chk);
1597
1598         return 0;
1599 }
1600
1601 /**
1602  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1603  * @skb: the skb to validate
1604  *
1605  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1606  * skb transport header accordingly and returns zero.
1607  *
1608  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1609  *  standard
1610  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1611  * -ENOMEM: A memory allocation failure happened.
1612  *
1613  * Caller needs to set the skb network header and free any returned skb if it
1614  * differs from the provided skb.
1615  */
1616 int ip_mc_check_igmp(struct sk_buff *skb)
1617 {
1618         int ret = ip_mc_check_iphdr(skb);
1619
1620         if (ret < 0)
1621                 return ret;
1622
1623         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1624                 return -ENOMSG;
1625
1626         ret = ip_mc_check_igmp_csum(skb);
1627         if (ret < 0)
1628                 return ret;
1629
1630         return ip_mc_check_igmp_msg(skb);
1631 }
1632 EXPORT_SYMBOL(ip_mc_check_igmp);
1633
1634 /*
1635  *      Resend IGMP JOIN report; used by netdev notifier.
1636  */
1637 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1638 {
1639 #ifdef CONFIG_IP_MULTICAST
1640         struct ip_mc_list *im;
1641         int type;
1642         struct net *net = dev_net(in_dev->dev);
1643
1644         ASSERT_RTNL();
1645
1646         for_each_pmc_rtnl(in_dev, im) {
1647                 if (im->multiaddr == IGMP_ALL_HOSTS)
1648                         continue;
1649                 if (ipv4_is_local_multicast(im->multiaddr) &&
1650                     !READ_ONCE(net->ipv4.sysctl_igmp_llm_reports))
1651                         continue;
1652
1653                 /* a failover is happening and switches
1654                  * must be notified immediately
1655                  */
1656                 if (IGMP_V1_SEEN(in_dev))
1657                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1658                 else if (IGMP_V2_SEEN(in_dev))
1659                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1660                 else
1661                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1662                 igmp_send_report(in_dev, im, type);
1663         }
1664 #endif
1665 }
1666
1667 /*
1668  *      A socket has left a multicast group on device dev
1669  */
1670
1671 void __ip_mc_dec_group(struct in_device *in_dev, __be32 addr, gfp_t gfp)
1672 {
1673         struct ip_mc_list *i;
1674         struct ip_mc_list __rcu **ip;
1675
1676         ASSERT_RTNL();
1677
1678         for (ip = &in_dev->mc_list;
1679              (i = rtnl_dereference(*ip)) != NULL;
1680              ip = &i->next_rcu) {
1681                 if (i->multiaddr == addr) {
1682                         if (--i->users == 0) {
1683                                 ip_mc_hash_remove(in_dev, i);
1684                                 *ip = i->next_rcu;
1685                                 in_dev->mc_count--;
1686                                 __igmp_group_dropped(i, gfp);
1687                                 ip_mc_clear_src(i);
1688
1689                                 if (!in_dev->dead)
1690                                         ip_rt_multicast_event(in_dev);
1691
1692                                 ip_ma_put(i);
1693                                 return;
1694                         }
1695                         break;
1696                 }
1697         }
1698 }
1699 EXPORT_SYMBOL(__ip_mc_dec_group);
1700
1701 /* Device changing type */
1702
1703 void ip_mc_unmap(struct in_device *in_dev)
1704 {
1705         struct ip_mc_list *pmc;
1706
1707         ASSERT_RTNL();
1708
1709         for_each_pmc_rtnl(in_dev, pmc)
1710                 igmp_group_dropped(pmc);
1711 }
1712
1713 void ip_mc_remap(struct in_device *in_dev)
1714 {
1715         struct ip_mc_list *pmc;
1716
1717         ASSERT_RTNL();
1718
1719         for_each_pmc_rtnl(in_dev, pmc) {
1720 #ifdef CONFIG_IP_MULTICAST
1721                 igmpv3_del_delrec(in_dev, pmc);
1722 #endif
1723                 igmp_group_added(pmc);
1724         }
1725 }
1726
1727 /* Device going down */
1728
1729 void ip_mc_down(struct in_device *in_dev)
1730 {
1731         struct ip_mc_list *pmc;
1732
1733         ASSERT_RTNL();
1734
1735         for_each_pmc_rtnl(in_dev, pmc)
1736                 igmp_group_dropped(pmc);
1737
1738 #ifdef CONFIG_IP_MULTICAST
1739         WRITE_ONCE(in_dev->mr_ifc_count, 0);
1740         if (del_timer(&in_dev->mr_ifc_timer))
1741                 __in_dev_put(in_dev);
1742         in_dev->mr_gq_running = 0;
1743         if (del_timer(&in_dev->mr_gq_timer))
1744                 __in_dev_put(in_dev);
1745 #endif
1746
1747         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1748 }
1749
1750 #ifdef CONFIG_IP_MULTICAST
1751 static void ip_mc_reset(struct in_device *in_dev)
1752 {
1753         struct net *net = dev_net(in_dev->dev);
1754
1755         in_dev->mr_qi = IGMP_QUERY_INTERVAL;
1756         in_dev->mr_qri = IGMP_QUERY_RESPONSE_INTERVAL;
1757         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1758 }
1759 #else
1760 static void ip_mc_reset(struct in_device *in_dev)
1761 {
1762 }
1763 #endif
1764
1765 void ip_mc_init_dev(struct in_device *in_dev)
1766 {
1767         ASSERT_RTNL();
1768
1769 #ifdef CONFIG_IP_MULTICAST
1770         timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1771         timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1772 #endif
1773         ip_mc_reset(in_dev);
1774
1775         spin_lock_init(&in_dev->mc_tomb_lock);
1776 }
1777
1778 /* Device going up */
1779
1780 void ip_mc_up(struct in_device *in_dev)
1781 {
1782         struct ip_mc_list *pmc;
1783
1784         ASSERT_RTNL();
1785
1786         ip_mc_reset(in_dev);
1787         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1788
1789         for_each_pmc_rtnl(in_dev, pmc) {
1790 #ifdef CONFIG_IP_MULTICAST
1791                 igmpv3_del_delrec(in_dev, pmc);
1792 #endif
1793                 igmp_group_added(pmc);
1794         }
1795 }
1796
1797 /*
1798  *      Device is about to be destroyed: clean up.
1799  */
1800
1801 void ip_mc_destroy_dev(struct in_device *in_dev)
1802 {
1803         struct ip_mc_list *i;
1804
1805         ASSERT_RTNL();
1806
1807         /* Deactivate timers */
1808         ip_mc_down(in_dev);
1809 #ifdef CONFIG_IP_MULTICAST
1810         igmpv3_clear_delrec(in_dev);
1811 #endif
1812
1813         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1814                 in_dev->mc_list = i->next_rcu;
1815                 in_dev->mc_count--;
1816                 ip_mc_clear_src(i);
1817                 ip_ma_put(i);
1818         }
1819 }
1820
1821 /* RTNL is locked */
1822 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1823 {
1824         struct net_device *dev = NULL;
1825         struct in_device *idev = NULL;
1826
1827         if (imr->imr_ifindex) {
1828                 idev = inetdev_by_index(net, imr->imr_ifindex);
1829                 return idev;
1830         }
1831         if (imr->imr_address.s_addr) {
1832                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1833                 if (!dev)
1834                         return NULL;
1835         }
1836
1837         if (!dev) {
1838                 struct rtable *rt = ip_route_output(net,
1839                                                     imr->imr_multiaddr.s_addr,
1840                                                     0, 0, 0);
1841                 if (!IS_ERR(rt)) {
1842                         dev = rt->dst.dev;
1843                         ip_rt_put(rt);
1844                 }
1845         }
1846         if (dev) {
1847                 imr->imr_ifindex = dev->ifindex;
1848                 idev = __in_dev_get_rtnl(dev);
1849         }
1850         return idev;
1851 }
1852
1853 /*
1854  *      Join a socket to a group
1855  */
1856
1857 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1858         __be32 *psfsrc)
1859 {
1860         struct ip_sf_list *psf, *psf_prev;
1861         int rv = 0;
1862
1863         psf_prev = NULL;
1864         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1865                 if (psf->sf_inaddr == *psfsrc)
1866                         break;
1867                 psf_prev = psf;
1868         }
1869         if (!psf || psf->sf_count[sfmode] == 0) {
1870                 /* source filter not found, or count wrong =>  bug */
1871                 return -ESRCH;
1872         }
1873         psf->sf_count[sfmode]--;
1874         if (psf->sf_count[sfmode] == 0) {
1875                 ip_rt_multicast_event(pmc->interface);
1876         }
1877         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1878 #ifdef CONFIG_IP_MULTICAST
1879                 struct in_device *in_dev = pmc->interface;
1880                 struct net *net = dev_net(in_dev->dev);
1881 #endif
1882
1883                 /* no more filters for this source */
1884                 if (psf_prev)
1885                         psf_prev->sf_next = psf->sf_next;
1886                 else
1887                         pmc->sources = psf->sf_next;
1888 #ifdef CONFIG_IP_MULTICAST
1889                 if (psf->sf_oldin &&
1890                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1891                         psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1892                         psf->sf_next = pmc->tomb;
1893                         pmc->tomb = psf;
1894                         rv = 1;
1895                 } else
1896 #endif
1897                         kfree(psf);
1898         }
1899         return rv;
1900 }
1901
1902 #ifndef CONFIG_IP_MULTICAST
1903 #define igmp_ifc_event(x)       do { } while (0)
1904 #endif
1905
1906 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1907                          int sfcount, __be32 *psfsrc, int delta)
1908 {
1909         struct ip_mc_list *pmc;
1910         int     changerec = 0;
1911         int     i, err;
1912
1913         if (!in_dev)
1914                 return -ENODEV;
1915         rcu_read_lock();
1916         for_each_pmc_rcu(in_dev, pmc) {
1917                 if (*pmca == pmc->multiaddr)
1918                         break;
1919         }
1920         if (!pmc) {
1921                 /* MCA not found?? bug */
1922                 rcu_read_unlock();
1923                 return -ESRCH;
1924         }
1925         spin_lock_bh(&pmc->lock);
1926         rcu_read_unlock();
1927 #ifdef CONFIG_IP_MULTICAST
1928         sf_markstate(pmc);
1929 #endif
1930         if (!delta) {
1931                 err = -EINVAL;
1932                 if (!pmc->sfcount[sfmode])
1933                         goto out_unlock;
1934                 pmc->sfcount[sfmode]--;
1935         }
1936         err = 0;
1937         for (i = 0; i < sfcount; i++) {
1938                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1939
1940                 changerec |= rv > 0;
1941                 if (!err && rv < 0)
1942                         err = rv;
1943         }
1944         if (pmc->sfmode == MCAST_EXCLUDE &&
1945             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1946             pmc->sfcount[MCAST_INCLUDE]) {
1947 #ifdef CONFIG_IP_MULTICAST
1948                 struct ip_sf_list *psf;
1949                 struct net *net = dev_net(in_dev->dev);
1950 #endif
1951
1952                 /* filter mode change */
1953                 pmc->sfmode = MCAST_INCLUDE;
1954 #ifdef CONFIG_IP_MULTICAST
1955                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1956                 WRITE_ONCE(in_dev->mr_ifc_count, pmc->crcount);
1957                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1958                         psf->sf_crcount = 0;
1959                 igmp_ifc_event(pmc->interface);
1960         } else if (sf_setstate(pmc) || changerec) {
1961                 igmp_ifc_event(pmc->interface);
1962 #endif
1963         }
1964 out_unlock:
1965         spin_unlock_bh(&pmc->lock);
1966         return err;
1967 }
1968
1969 /*
1970  * Add multicast single-source filter to the interface list
1971  */
1972 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1973         __be32 *psfsrc)
1974 {
1975         struct ip_sf_list *psf, *psf_prev;
1976
1977         psf_prev = NULL;
1978         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1979                 if (psf->sf_inaddr == *psfsrc)
1980                         break;
1981                 psf_prev = psf;
1982         }
1983         if (!psf) {
1984                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1985                 if (!psf)
1986                         return -ENOBUFS;
1987                 psf->sf_inaddr = *psfsrc;
1988                 if (psf_prev) {
1989                         psf_prev->sf_next = psf;
1990                 } else
1991                         pmc->sources = psf;
1992         }
1993         psf->sf_count[sfmode]++;
1994         if (psf->sf_count[sfmode] == 1) {
1995                 ip_rt_multicast_event(pmc->interface);
1996         }
1997         return 0;
1998 }
1999
2000 #ifdef CONFIG_IP_MULTICAST
2001 static void sf_markstate(struct ip_mc_list *pmc)
2002 {
2003         struct ip_sf_list *psf;
2004         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2005
2006         for (psf = pmc->sources; psf; psf = psf->sf_next)
2007                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2008                         psf->sf_oldin = mca_xcount ==
2009                                 psf->sf_count[MCAST_EXCLUDE] &&
2010                                 !psf->sf_count[MCAST_INCLUDE];
2011                 } else
2012                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
2013 }
2014
2015 static int sf_setstate(struct ip_mc_list *pmc)
2016 {
2017         struct ip_sf_list *psf, *dpsf;
2018         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2019         int qrv = pmc->interface->mr_qrv;
2020         int new_in, rv;
2021
2022         rv = 0;
2023         for (psf = pmc->sources; psf; psf = psf->sf_next) {
2024                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2025                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
2026                                 !psf->sf_count[MCAST_INCLUDE];
2027                 } else
2028                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
2029                 if (new_in) {
2030                         if (!psf->sf_oldin) {
2031                                 struct ip_sf_list *prev = NULL;
2032
2033                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2034                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2035                                                 break;
2036                                         prev = dpsf;
2037                                 }
2038                                 if (dpsf) {
2039                                         if (prev)
2040                                                 prev->sf_next = dpsf->sf_next;
2041                                         else
2042                                                 pmc->tomb = dpsf->sf_next;
2043                                         kfree(dpsf);
2044                                 }
2045                                 psf->sf_crcount = qrv;
2046                                 rv++;
2047                         }
2048                 } else if (psf->sf_oldin) {
2049
2050                         psf->sf_crcount = 0;
2051                         /*
2052                          * add or update "delete" records if an active filter
2053                          * is now inactive
2054                          */
2055                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2056                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2057                                         break;
2058                         if (!dpsf) {
2059                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2060                                 if (!dpsf)
2061                                         continue;
2062                                 *dpsf = *psf;
2063                                 /* pmc->lock held by callers */
2064                                 dpsf->sf_next = pmc->tomb;
2065                                 pmc->tomb = dpsf;
2066                         }
2067                         dpsf->sf_crcount = qrv;
2068                         rv++;
2069                 }
2070         }
2071         return rv;
2072 }
2073 #endif
2074
2075 /*
2076  * Add multicast source filter list to the interface list
2077  */
2078 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2079                          int sfcount, __be32 *psfsrc, int delta)
2080 {
2081         struct ip_mc_list *pmc;
2082         int     isexclude;
2083         int     i, err;
2084
2085         if (!in_dev)
2086                 return -ENODEV;
2087         rcu_read_lock();
2088         for_each_pmc_rcu(in_dev, pmc) {
2089                 if (*pmca == pmc->multiaddr)
2090                         break;
2091         }
2092         if (!pmc) {
2093                 /* MCA not found?? bug */
2094                 rcu_read_unlock();
2095                 return -ESRCH;
2096         }
2097         spin_lock_bh(&pmc->lock);
2098         rcu_read_unlock();
2099
2100 #ifdef CONFIG_IP_MULTICAST
2101         sf_markstate(pmc);
2102 #endif
2103         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2104         if (!delta)
2105                 pmc->sfcount[sfmode]++;
2106         err = 0;
2107         for (i = 0; i < sfcount; i++) {
2108                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2109                 if (err)
2110                         break;
2111         }
2112         if (err) {
2113                 int j;
2114
2115                 if (!delta)
2116                         pmc->sfcount[sfmode]--;
2117                 for (j = 0; j < i; j++)
2118                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2119         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2120 #ifdef CONFIG_IP_MULTICAST
2121                 struct ip_sf_list *psf;
2122                 struct net *net = dev_net(pmc->interface->dev);
2123                 in_dev = pmc->interface;
2124 #endif
2125
2126                 /* filter mode change */
2127                 if (pmc->sfcount[MCAST_EXCLUDE])
2128                         pmc->sfmode = MCAST_EXCLUDE;
2129                 else if (pmc->sfcount[MCAST_INCLUDE])
2130                         pmc->sfmode = MCAST_INCLUDE;
2131 #ifdef CONFIG_IP_MULTICAST
2132                 /* else no filters; keep old mode for reports */
2133
2134                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2135                 WRITE_ONCE(in_dev->mr_ifc_count, pmc->crcount);
2136                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2137                         psf->sf_crcount = 0;
2138                 igmp_ifc_event(in_dev);
2139         } else if (sf_setstate(pmc)) {
2140                 igmp_ifc_event(in_dev);
2141 #endif
2142         }
2143         spin_unlock_bh(&pmc->lock);
2144         return err;
2145 }
2146
2147 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2148 {
2149         struct ip_sf_list *tomb, *sources;
2150
2151         spin_lock_bh(&pmc->lock);
2152         tomb = pmc->tomb;
2153         pmc->tomb = NULL;
2154         sources = pmc->sources;
2155         pmc->sources = NULL;
2156         pmc->sfmode = MCAST_EXCLUDE;
2157         pmc->sfcount[MCAST_INCLUDE] = 0;
2158         pmc->sfcount[MCAST_EXCLUDE] = 1;
2159         spin_unlock_bh(&pmc->lock);
2160
2161         ip_sf_list_clear_all(tomb);
2162         ip_sf_list_clear_all(sources);
2163 }
2164
2165 /* Join a multicast group
2166  */
2167 static int __ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr,
2168                               unsigned int mode)
2169 {
2170         __be32 addr = imr->imr_multiaddr.s_addr;
2171         struct ip_mc_socklist *iml, *i;
2172         struct in_device *in_dev;
2173         struct inet_sock *inet = inet_sk(sk);
2174         struct net *net = sock_net(sk);
2175         int ifindex;
2176         int count = 0;
2177         int err;
2178
2179         ASSERT_RTNL();
2180
2181         if (!ipv4_is_multicast(addr))
2182                 return -EINVAL;
2183
2184         in_dev = ip_mc_find_dev(net, imr);
2185
2186         if (!in_dev) {
2187                 err = -ENODEV;
2188                 goto done;
2189         }
2190
2191         err = -EADDRINUSE;
2192         ifindex = imr->imr_ifindex;
2193         for_each_pmc_rtnl(inet, i) {
2194                 if (i->multi.imr_multiaddr.s_addr == addr &&
2195                     i->multi.imr_ifindex == ifindex)
2196                         goto done;
2197                 count++;
2198         }
2199         err = -ENOBUFS;
2200         if (count >= net->ipv4.sysctl_igmp_max_memberships)
2201                 goto done;
2202         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2203         if (!iml)
2204                 goto done;
2205
2206         memcpy(&iml->multi, imr, sizeof(*imr));
2207         iml->next_rcu = inet->mc_list;
2208         iml->sflist = NULL;
2209         iml->sfmode = mode;
2210         rcu_assign_pointer(inet->mc_list, iml);
2211         ____ip_mc_inc_group(in_dev, addr, mode, GFP_KERNEL);
2212         err = 0;
2213 done:
2214         return err;
2215 }
2216
2217 /* Join ASM (Any-Source Multicast) group
2218  */
2219 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2220 {
2221         return __ip_mc_join_group(sk, imr, MCAST_EXCLUDE);
2222 }
2223 EXPORT_SYMBOL(ip_mc_join_group);
2224
2225 /* Join SSM (Source-Specific Multicast) group
2226  */
2227 int ip_mc_join_group_ssm(struct sock *sk, struct ip_mreqn *imr,
2228                          unsigned int mode)
2229 {
2230         return __ip_mc_join_group(sk, imr, mode);
2231 }
2232
2233 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2234                            struct in_device *in_dev)
2235 {
2236         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2237         int err;
2238
2239         if (!psf) {
2240                 /* any-source empty exclude case */
2241                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2242                         iml->sfmode, 0, NULL, 0);
2243         }
2244         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2245                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2246         RCU_INIT_POINTER(iml->sflist, NULL);
2247         /* decrease mem now to avoid the memleak warning */
2248         atomic_sub(struct_size(psf, sl_addr, psf->sl_max), &sk->sk_omem_alloc);
2249         kfree_rcu(psf, rcu);
2250         return err;
2251 }
2252
2253 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2254 {
2255         struct inet_sock *inet = inet_sk(sk);
2256         struct ip_mc_socklist *iml;
2257         struct ip_mc_socklist __rcu **imlp;
2258         struct in_device *in_dev;
2259         struct net *net = sock_net(sk);
2260         __be32 group = imr->imr_multiaddr.s_addr;
2261         u32 ifindex;
2262         int ret = -EADDRNOTAVAIL;
2263
2264         ASSERT_RTNL();
2265
2266         in_dev = ip_mc_find_dev(net, imr);
2267         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2268                 ret = -ENODEV;
2269                 goto out;
2270         }
2271         ifindex = imr->imr_ifindex;
2272         for (imlp = &inet->mc_list;
2273              (iml = rtnl_dereference(*imlp)) != NULL;
2274              imlp = &iml->next_rcu) {
2275                 if (iml->multi.imr_multiaddr.s_addr != group)
2276                         continue;
2277                 if (ifindex) {
2278                         if (iml->multi.imr_ifindex != ifindex)
2279                                 continue;
2280                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2281                                 iml->multi.imr_address.s_addr)
2282                         continue;
2283
2284                 (void) ip_mc_leave_src(sk, iml, in_dev);
2285
2286                 *imlp = iml->next_rcu;
2287
2288                 if (in_dev)
2289                         ip_mc_dec_group(in_dev, group);
2290
2291                 /* decrease mem now to avoid the memleak warning */
2292                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2293                 kfree_rcu(iml, rcu);
2294                 return 0;
2295         }
2296 out:
2297         return ret;
2298 }
2299 EXPORT_SYMBOL(ip_mc_leave_group);
2300
2301 int ip_mc_source(int add, int omode, struct sock *sk, struct
2302         ip_mreq_source *mreqs, int ifindex)
2303 {
2304         int err;
2305         struct ip_mreqn imr;
2306         __be32 addr = mreqs->imr_multiaddr;
2307         struct ip_mc_socklist *pmc;
2308         struct in_device *in_dev = NULL;
2309         struct inet_sock *inet = inet_sk(sk);
2310         struct ip_sf_socklist *psl;
2311         struct net *net = sock_net(sk);
2312         int leavegroup = 0;
2313         int i, j, rv;
2314
2315         if (!ipv4_is_multicast(addr))
2316                 return -EINVAL;
2317
2318         ASSERT_RTNL();
2319
2320         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2321         imr.imr_address.s_addr = mreqs->imr_interface;
2322         imr.imr_ifindex = ifindex;
2323         in_dev = ip_mc_find_dev(net, &imr);
2324
2325         if (!in_dev) {
2326                 err = -ENODEV;
2327                 goto done;
2328         }
2329         err = -EADDRNOTAVAIL;
2330
2331         for_each_pmc_rtnl(inet, pmc) {
2332                 if ((pmc->multi.imr_multiaddr.s_addr ==
2333                      imr.imr_multiaddr.s_addr) &&
2334                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2335                         break;
2336         }
2337         if (!pmc) {             /* must have a prior join */
2338                 err = -EINVAL;
2339                 goto done;
2340         }
2341         /* if a source filter was set, must be the same mode as before */
2342         if (pmc->sflist) {
2343                 if (pmc->sfmode != omode) {
2344                         err = -EINVAL;
2345                         goto done;
2346                 }
2347         } else if (pmc->sfmode != omode) {
2348                 /* allow mode switches for empty-set filters */
2349                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2350                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2351                         NULL, 0);
2352                 pmc->sfmode = omode;
2353         }
2354
2355         psl = rtnl_dereference(pmc->sflist);
2356         if (!add) {
2357                 if (!psl)
2358                         goto done;      /* err = -EADDRNOTAVAIL */
2359                 rv = !0;
2360                 for (i = 0; i < psl->sl_count; i++) {
2361                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2362                                 sizeof(__be32));
2363                         if (rv == 0)
2364                                 break;
2365                 }
2366                 if (rv)         /* source not found */
2367                         goto done;      /* err = -EADDRNOTAVAIL */
2368
2369                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2370                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2371                         leavegroup = 1;
2372                         goto done;
2373                 }
2374
2375                 /* update the interface filter */
2376                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2377                         &mreqs->imr_sourceaddr, 1);
2378
2379                 for (j = i+1; j < psl->sl_count; j++)
2380                         psl->sl_addr[j-1] = psl->sl_addr[j];
2381                 psl->sl_count--;
2382                 err = 0;
2383                 goto done;
2384         }
2385         /* else, add a new source to the filter */
2386
2387         if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2388                 err = -ENOBUFS;
2389                 goto done;
2390         }
2391         if (!psl || psl->sl_count == psl->sl_max) {
2392                 struct ip_sf_socklist *newpsl;
2393                 int count = IP_SFBLOCK;
2394
2395                 if (psl)
2396                         count += psl->sl_max;
2397                 newpsl = sock_kmalloc(sk, struct_size(newpsl, sl_addr, count),
2398                                       GFP_KERNEL);
2399                 if (!newpsl) {
2400                         err = -ENOBUFS;
2401                         goto done;
2402                 }
2403                 newpsl->sl_max = count;
2404                 newpsl->sl_count = count - IP_SFBLOCK;
2405                 if (psl) {
2406                         for (i = 0; i < psl->sl_count; i++)
2407                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2408                         /* decrease mem now to avoid the memleak warning */
2409                         atomic_sub(struct_size(psl, sl_addr, psl->sl_max),
2410                                    &sk->sk_omem_alloc);
2411                 }
2412                 rcu_assign_pointer(pmc->sflist, newpsl);
2413                 if (psl)
2414                         kfree_rcu(psl, rcu);
2415                 psl = newpsl;
2416         }
2417         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2418         for (i = 0; i < psl->sl_count; i++) {
2419                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2420                         sizeof(__be32));
2421                 if (rv == 0)
2422                         break;
2423         }
2424         if (rv == 0)            /* address already there is an error */
2425                 goto done;
2426         for (j = psl->sl_count-1; j >= i; j--)
2427                 psl->sl_addr[j+1] = psl->sl_addr[j];
2428         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2429         psl->sl_count++;
2430         err = 0;
2431         /* update the interface list */
2432         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2433                 &mreqs->imr_sourceaddr, 1);
2434 done:
2435         if (leavegroup)
2436                 err = ip_mc_leave_group(sk, &imr);
2437         return err;
2438 }
2439
2440 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2441 {
2442         int err = 0;
2443         struct ip_mreqn imr;
2444         __be32 addr = msf->imsf_multiaddr;
2445         struct ip_mc_socklist *pmc;
2446         struct in_device *in_dev;
2447         struct inet_sock *inet = inet_sk(sk);
2448         struct ip_sf_socklist *newpsl, *psl;
2449         struct net *net = sock_net(sk);
2450         int leavegroup = 0;
2451
2452         if (!ipv4_is_multicast(addr))
2453                 return -EINVAL;
2454         if (msf->imsf_fmode != MCAST_INCLUDE &&
2455             msf->imsf_fmode != MCAST_EXCLUDE)
2456                 return -EINVAL;
2457
2458         ASSERT_RTNL();
2459
2460         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2461         imr.imr_address.s_addr = msf->imsf_interface;
2462         imr.imr_ifindex = ifindex;
2463         in_dev = ip_mc_find_dev(net, &imr);
2464
2465         if (!in_dev) {
2466                 err = -ENODEV;
2467                 goto done;
2468         }
2469
2470         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2471         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2472                 leavegroup = 1;
2473                 goto done;
2474         }
2475
2476         for_each_pmc_rtnl(inet, pmc) {
2477                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2478                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2479                         break;
2480         }
2481         if (!pmc) {             /* must have a prior join */
2482                 err = -EINVAL;
2483                 goto done;
2484         }
2485         if (msf->imsf_numsrc) {
2486                 newpsl = sock_kmalloc(sk, struct_size(newpsl, sl_addr,
2487                                                       msf->imsf_numsrc),
2488                                       GFP_KERNEL);
2489                 if (!newpsl) {
2490                         err = -ENOBUFS;
2491                         goto done;
2492                 }
2493                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2494                 memcpy(newpsl->sl_addr, msf->imsf_slist_flex,
2495                        flex_array_size(msf, imsf_slist_flex, msf->imsf_numsrc));
2496                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2497                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2498                 if (err) {
2499                         sock_kfree_s(sk, newpsl,
2500                                      struct_size(newpsl, sl_addr,
2501                                                  newpsl->sl_max));
2502                         goto done;
2503                 }
2504         } else {
2505                 newpsl = NULL;
2506                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2507                                      msf->imsf_fmode, 0, NULL, 0);
2508         }
2509         psl = rtnl_dereference(pmc->sflist);
2510         if (psl) {
2511                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2512                         psl->sl_count, psl->sl_addr, 0);
2513                 /* decrease mem now to avoid the memleak warning */
2514                 atomic_sub(struct_size(psl, sl_addr, psl->sl_max),
2515                            &sk->sk_omem_alloc);
2516         } else {
2517                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2518                         0, NULL, 0);
2519         }
2520         rcu_assign_pointer(pmc->sflist, newpsl);
2521         if (psl)
2522                 kfree_rcu(psl, rcu);
2523         pmc->sfmode = msf->imsf_fmode;
2524         err = 0;
2525 done:
2526         if (leavegroup)
2527                 err = ip_mc_leave_group(sk, &imr);
2528         return err;
2529 }
2530
2531 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2532         struct ip_msfilter __user *optval, int __user *optlen)
2533 {
2534         int err, len, count, copycount;
2535         struct ip_mreqn imr;
2536         __be32 addr = msf->imsf_multiaddr;
2537         struct ip_mc_socklist *pmc;
2538         struct in_device *in_dev;
2539         struct inet_sock *inet = inet_sk(sk);
2540         struct ip_sf_socklist *psl;
2541         struct net *net = sock_net(sk);
2542
2543         ASSERT_RTNL();
2544
2545         if (!ipv4_is_multicast(addr))
2546                 return -EINVAL;
2547
2548         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2549         imr.imr_address.s_addr = msf->imsf_interface;
2550         imr.imr_ifindex = 0;
2551         in_dev = ip_mc_find_dev(net, &imr);
2552
2553         if (!in_dev) {
2554                 err = -ENODEV;
2555                 goto done;
2556         }
2557         err = -EADDRNOTAVAIL;
2558
2559         for_each_pmc_rtnl(inet, pmc) {
2560                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2561                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2562                         break;
2563         }
2564         if (!pmc)               /* must have a prior join */
2565                 goto done;
2566         msf->imsf_fmode = pmc->sfmode;
2567         psl = rtnl_dereference(pmc->sflist);
2568         if (!psl) {
2569                 count = 0;
2570         } else {
2571                 count = psl->sl_count;
2572         }
2573         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2574         len = flex_array_size(psl, sl_addr, copycount);
2575         msf->imsf_numsrc = count;
2576         if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2577             copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2578                 return -EFAULT;
2579         }
2580         if (len &&
2581             copy_to_user(&optval->imsf_slist_flex[0], psl->sl_addr, len))
2582                 return -EFAULT;
2583         return 0;
2584 done:
2585         return err;
2586 }
2587
2588 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2589         struct sockaddr_storage __user *p)
2590 {
2591         int i, count, copycount;
2592         struct sockaddr_in *psin;
2593         __be32 addr;
2594         struct ip_mc_socklist *pmc;
2595         struct inet_sock *inet = inet_sk(sk);
2596         struct ip_sf_socklist *psl;
2597
2598         ASSERT_RTNL();
2599
2600         psin = (struct sockaddr_in *)&gsf->gf_group;
2601         if (psin->sin_family != AF_INET)
2602                 return -EINVAL;
2603         addr = psin->sin_addr.s_addr;
2604         if (!ipv4_is_multicast(addr))
2605                 return -EINVAL;
2606
2607         for_each_pmc_rtnl(inet, pmc) {
2608                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2609                     pmc->multi.imr_ifindex == gsf->gf_interface)
2610                         break;
2611         }
2612         if (!pmc)               /* must have a prior join */
2613                 return -EADDRNOTAVAIL;
2614         gsf->gf_fmode = pmc->sfmode;
2615         psl = rtnl_dereference(pmc->sflist);
2616         count = psl ? psl->sl_count : 0;
2617         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2618         gsf->gf_numsrc = count;
2619         for (i = 0; i < copycount; i++, p++) {
2620                 struct sockaddr_storage ss;
2621
2622                 psin = (struct sockaddr_in *)&ss;
2623                 memset(&ss, 0, sizeof(ss));
2624                 psin->sin_family = AF_INET;
2625                 psin->sin_addr.s_addr = psl->sl_addr[i];
2626                 if (copy_to_user(p, &ss, sizeof(ss)))
2627                         return -EFAULT;
2628         }
2629         return 0;
2630 }
2631
2632 /*
2633  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2634  */
2635 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2636                    int dif, int sdif)
2637 {
2638         struct inet_sock *inet = inet_sk(sk);
2639         struct ip_mc_socklist *pmc;
2640         struct ip_sf_socklist *psl;
2641         int i;
2642         int ret;
2643
2644         ret = 1;
2645         if (!ipv4_is_multicast(loc_addr))
2646                 goto out;
2647
2648         rcu_read_lock();
2649         for_each_pmc_rcu(inet, pmc) {
2650                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2651                     (pmc->multi.imr_ifindex == dif ||
2652                      (sdif && pmc->multi.imr_ifindex == sdif)))
2653                         break;
2654         }
2655         ret = inet->mc_all;
2656         if (!pmc)
2657                 goto unlock;
2658         psl = rcu_dereference(pmc->sflist);
2659         ret = (pmc->sfmode == MCAST_EXCLUDE);
2660         if (!psl)
2661                 goto unlock;
2662
2663         for (i = 0; i < psl->sl_count; i++) {
2664                 if (psl->sl_addr[i] == rmt_addr)
2665                         break;
2666         }
2667         ret = 0;
2668         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2669                 goto unlock;
2670         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2671                 goto unlock;
2672         ret = 1;
2673 unlock:
2674         rcu_read_unlock();
2675 out:
2676         return ret;
2677 }
2678
2679 /*
2680  *      A socket is closing.
2681  */
2682
2683 void ip_mc_drop_socket(struct sock *sk)
2684 {
2685         struct inet_sock *inet = inet_sk(sk);
2686         struct ip_mc_socklist *iml;
2687         struct net *net = sock_net(sk);
2688
2689         if (!inet->mc_list)
2690                 return;
2691
2692         rtnl_lock();
2693         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2694                 struct in_device *in_dev;
2695
2696                 inet->mc_list = iml->next_rcu;
2697                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2698                 (void) ip_mc_leave_src(sk, iml, in_dev);
2699                 if (in_dev)
2700                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2701                 /* decrease mem now to avoid the memleak warning */
2702                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2703                 kfree_rcu(iml, rcu);
2704         }
2705         rtnl_unlock();
2706 }
2707
2708 /* called with rcu_read_lock() */
2709 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2710 {
2711         struct ip_mc_list *im;
2712         struct ip_mc_list __rcu **mc_hash;
2713         struct ip_sf_list *psf;
2714         int rv = 0;
2715
2716         mc_hash = rcu_dereference(in_dev->mc_hash);
2717         if (mc_hash) {
2718                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2719
2720                 for (im = rcu_dereference(mc_hash[hash]);
2721                      im != NULL;
2722                      im = rcu_dereference(im->next_hash)) {
2723                         if (im->multiaddr == mc_addr)
2724                                 break;
2725                 }
2726         } else {
2727                 for_each_pmc_rcu(in_dev, im) {
2728                         if (im->multiaddr == mc_addr)
2729                                 break;
2730                 }
2731         }
2732         if (im && proto == IPPROTO_IGMP) {
2733                 rv = 1;
2734         } else if (im) {
2735                 if (src_addr) {
2736                         spin_lock_bh(&im->lock);
2737                         for (psf = im->sources; psf; psf = psf->sf_next) {
2738                                 if (psf->sf_inaddr == src_addr)
2739                                         break;
2740                         }
2741                         if (psf)
2742                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2743                                         psf->sf_count[MCAST_EXCLUDE] !=
2744                                         im->sfcount[MCAST_EXCLUDE];
2745                         else
2746                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2747                         spin_unlock_bh(&im->lock);
2748                 } else
2749                         rv = 1; /* unspecified source; tentatively allow */
2750         }
2751         return rv;
2752 }
2753
2754 #if defined(CONFIG_PROC_FS)
2755 struct igmp_mc_iter_state {
2756         struct seq_net_private p;
2757         struct net_device *dev;
2758         struct in_device *in_dev;
2759 };
2760
2761 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2762
2763 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2764 {
2765         struct net *net = seq_file_net(seq);
2766         struct ip_mc_list *im = NULL;
2767         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2768
2769         state->in_dev = NULL;
2770         for_each_netdev_rcu(net, state->dev) {
2771                 struct in_device *in_dev;
2772
2773                 in_dev = __in_dev_get_rcu(state->dev);
2774                 if (!in_dev)
2775                         continue;
2776                 im = rcu_dereference(in_dev->mc_list);
2777                 if (im) {
2778                         state->in_dev = in_dev;
2779                         break;
2780                 }
2781         }
2782         return im;
2783 }
2784
2785 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2786 {
2787         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2788
2789         im = rcu_dereference(im->next_rcu);
2790         while (!im) {
2791                 state->dev = next_net_device_rcu(state->dev);
2792                 if (!state->dev) {
2793                         state->in_dev = NULL;
2794                         break;
2795                 }
2796                 state->in_dev = __in_dev_get_rcu(state->dev);
2797                 if (!state->in_dev)
2798                         continue;
2799                 im = rcu_dereference(state->in_dev->mc_list);
2800         }
2801         return im;
2802 }
2803
2804 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2805 {
2806         struct ip_mc_list *im = igmp_mc_get_first(seq);
2807         if (im)
2808                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2809                         --pos;
2810         return pos ? NULL : im;
2811 }
2812
2813 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2814         __acquires(rcu)
2815 {
2816         rcu_read_lock();
2817         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2818 }
2819
2820 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2821 {
2822         struct ip_mc_list *im;
2823         if (v == SEQ_START_TOKEN)
2824                 im = igmp_mc_get_first(seq);
2825         else
2826                 im = igmp_mc_get_next(seq, v);
2827         ++*pos;
2828         return im;
2829 }
2830
2831 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2832         __releases(rcu)
2833 {
2834         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2835
2836         state->in_dev = NULL;
2837         state->dev = NULL;
2838         rcu_read_unlock();
2839 }
2840
2841 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2842 {
2843         if (v == SEQ_START_TOKEN)
2844                 seq_puts(seq,
2845                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2846         else {
2847                 struct ip_mc_list *im = v;
2848                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2849                 char   *querier;
2850                 long delta;
2851
2852 #ifdef CONFIG_IP_MULTICAST
2853                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2854                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2855                           "V3";
2856 #else
2857                 querier = "NONE";
2858 #endif
2859
2860                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2861                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2862                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2863                 }
2864
2865                 delta = im->timer.expires - jiffies;
2866                 seq_printf(seq,
2867                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2868                            im->multiaddr, im->users,
2869                            im->tm_running,
2870                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2871                            im->reporter);
2872         }
2873         return 0;
2874 }
2875
2876 static const struct seq_operations igmp_mc_seq_ops = {
2877         .start  =       igmp_mc_seq_start,
2878         .next   =       igmp_mc_seq_next,
2879         .stop   =       igmp_mc_seq_stop,
2880         .show   =       igmp_mc_seq_show,
2881 };
2882
2883 struct igmp_mcf_iter_state {
2884         struct seq_net_private p;
2885         struct net_device *dev;
2886         struct in_device *idev;
2887         struct ip_mc_list *im;
2888 };
2889
2890 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2891
2892 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2893 {
2894         struct net *net = seq_file_net(seq);
2895         struct ip_sf_list *psf = NULL;
2896         struct ip_mc_list *im = NULL;
2897         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2898
2899         state->idev = NULL;
2900         state->im = NULL;
2901         for_each_netdev_rcu(net, state->dev) {
2902                 struct in_device *idev;
2903                 idev = __in_dev_get_rcu(state->dev);
2904                 if (unlikely(!idev))
2905                         continue;
2906                 im = rcu_dereference(idev->mc_list);
2907                 if (likely(im)) {
2908                         spin_lock_bh(&im->lock);
2909                         psf = im->sources;
2910                         if (likely(psf)) {
2911                                 state->im = im;
2912                                 state->idev = idev;
2913                                 break;
2914                         }
2915                         spin_unlock_bh(&im->lock);
2916                 }
2917         }
2918         return psf;
2919 }
2920
2921 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2922 {
2923         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2924
2925         psf = psf->sf_next;
2926         while (!psf) {
2927                 spin_unlock_bh(&state->im->lock);
2928                 state->im = state->im->next;
2929                 while (!state->im) {
2930                         state->dev = next_net_device_rcu(state->dev);
2931                         if (!state->dev) {
2932                                 state->idev = NULL;
2933                                 goto out;
2934                         }
2935                         state->idev = __in_dev_get_rcu(state->dev);
2936                         if (!state->idev)
2937                                 continue;
2938                         state->im = rcu_dereference(state->idev->mc_list);
2939                 }
2940                 if (!state->im)
2941                         break;
2942                 spin_lock_bh(&state->im->lock);
2943                 psf = state->im->sources;
2944         }
2945 out:
2946         return psf;
2947 }
2948
2949 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2950 {
2951         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2952         if (psf)
2953                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2954                         --pos;
2955         return pos ? NULL : psf;
2956 }
2957
2958 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2959         __acquires(rcu)
2960 {
2961         rcu_read_lock();
2962         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2963 }
2964
2965 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2966 {
2967         struct ip_sf_list *psf;
2968         if (v == SEQ_START_TOKEN)
2969                 psf = igmp_mcf_get_first(seq);
2970         else
2971                 psf = igmp_mcf_get_next(seq, v);
2972         ++*pos;
2973         return psf;
2974 }
2975
2976 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2977         __releases(rcu)
2978 {
2979         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2980         if (likely(state->im)) {
2981                 spin_unlock_bh(&state->im->lock);
2982                 state->im = NULL;
2983         }
2984         state->idev = NULL;
2985         state->dev = NULL;
2986         rcu_read_unlock();
2987 }
2988
2989 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2990 {
2991         struct ip_sf_list *psf = v;
2992         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2993
2994         if (v == SEQ_START_TOKEN) {
2995                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2996         } else {
2997                 seq_printf(seq,
2998                            "%3d %6.6s 0x%08x "
2999                            "0x%08x %6lu %6lu\n",
3000                            state->dev->ifindex, state->dev->name,
3001                            ntohl(state->im->multiaddr),
3002                            ntohl(psf->sf_inaddr),
3003                            psf->sf_count[MCAST_INCLUDE],
3004                            psf->sf_count[MCAST_EXCLUDE]);
3005         }
3006         return 0;
3007 }
3008
3009 static const struct seq_operations igmp_mcf_seq_ops = {
3010         .start  =       igmp_mcf_seq_start,
3011         .next   =       igmp_mcf_seq_next,
3012         .stop   =       igmp_mcf_seq_stop,
3013         .show   =       igmp_mcf_seq_show,
3014 };
3015
3016 static int __net_init igmp_net_init(struct net *net)
3017 {
3018         struct proc_dir_entry *pde;
3019         int err;
3020
3021         pde = proc_create_net("igmp", 0444, net->proc_net, &igmp_mc_seq_ops,
3022                         sizeof(struct igmp_mc_iter_state));
3023         if (!pde)
3024                 goto out_igmp;
3025         pde = proc_create_net("mcfilter", 0444, net->proc_net,
3026                         &igmp_mcf_seq_ops, sizeof(struct igmp_mcf_iter_state));
3027         if (!pde)
3028                 goto out_mcfilter;
3029         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3030                                    SOCK_DGRAM, 0, net);
3031         if (err < 0) {
3032                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3033                        err);
3034                 goto out_sock;
3035         }
3036
3037         return 0;
3038
3039 out_sock:
3040         remove_proc_entry("mcfilter", net->proc_net);
3041 out_mcfilter:
3042         remove_proc_entry("igmp", net->proc_net);
3043 out_igmp:
3044         return -ENOMEM;
3045 }
3046
3047 static void __net_exit igmp_net_exit(struct net *net)
3048 {
3049         remove_proc_entry("mcfilter", net->proc_net);
3050         remove_proc_entry("igmp", net->proc_net);
3051         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3052 }
3053
3054 static struct pernet_operations igmp_net_ops = {
3055         .init = igmp_net_init,
3056         .exit = igmp_net_exit,
3057 };
3058 #endif
3059
3060 static int igmp_netdev_event(struct notifier_block *this,
3061                              unsigned long event, void *ptr)
3062 {
3063         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3064         struct in_device *in_dev;
3065
3066         switch (event) {
3067         case NETDEV_RESEND_IGMP:
3068                 in_dev = __in_dev_get_rtnl(dev);
3069                 if (in_dev)
3070                         ip_mc_rejoin_groups(in_dev);
3071                 break;
3072         default:
3073                 break;
3074         }
3075         return NOTIFY_DONE;
3076 }
3077
3078 static struct notifier_block igmp_notifier = {
3079         .notifier_call = igmp_netdev_event,
3080 };
3081
3082 int __init igmp_mc_init(void)
3083 {
3084 #if defined(CONFIG_PROC_FS)
3085         int err;
3086
3087         err = register_pernet_subsys(&igmp_net_ops);
3088         if (err)
3089                 return err;
3090         err = register_netdevice_notifier(&igmp_notifier);
3091         if (err)
3092                 goto reg_notif_fail;
3093         return 0;
3094
3095 reg_notif_fail:
3096         unregister_pernet_subsys(&igmp_net_ops);
3097         return err;
3098 #else
3099         return register_netdevice_notifier(&igmp_notifier);
3100 #endif
3101 }