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